The Phylosophy of Universal Grammar

The Philosophy of Universal Grammar

The Philosophy of Universal Grammar W O L F R A M H I N Z EN A N D MI C H E L L E S H E E H A N

1

3

Great Clarendon Street, Oxford, ox2 6dp, United Kingdom Oxford University Press is a department of the University of Oxford. It furthers the University’s objective of excellence in research, scholarship, and education by publishing worldwide. Oxford is a registered trade mark of Oxford University Press in the UK and in certain other countries # Wolfram Hinzen and Michelle Sheehan 2013 The moral rights of the authors have been asserted First Edition published in 2013 Impression: 1 All rights reserved. No part of this publication may be reproduced, stored in a retrieval system, or transmitted, in any form or by any means, without the prior permission in writing of Oxford University Press, or as expressly permitted by law, by licence or under terms agreed with the appropriate reprographics rights organization. Enquiries concerning reproduction outside the scope of the above should be sent to the Rights Department, Oxford University Press, at the address above You must not circulate this work in any other form and you must impose this same condition on any acquirer Published in the United States of America by Oxford University Press 198 Madison Avenue, New York, NY 10016, United States of America British Library Cataloguing in Publication Data Data available Library of Congress Control Number: 2013940286 ISBN 978–0–19–965483–3 Printed and bound in Great Britain by CPI Group (UK) Ltd, Croydon, cr0 4yy

Contents Acknowledgements List of Abbreviations Introduction 1 The project of a science of language 1.1 1.2 1.3 1.4 1.5

Grammar as a domain of inquiry Cartesian linguistics, 1637–1966 Ancient India The Modistic viewpoint Conclusions

2 Before there was grammar 2.1 2.2 2.3 2.4 2.5 2.6 2.7

Overview Where semantics begins Lexicalization: the creation of the lexical atom Concepts $ intentionality $ intensionality $ grammar $ reference Parts of speech and their grammatical semantics The modes of signifying reloaded Conclusions

3 The content of grammar 3.1 3.2 3.3 3.4 3.5 3.6 3.7

Grammar as relational From parts of speech to the sentence Content from composition? Content from the interface? The phase as the smallest unit of grammatical organization The absence of X-within-X in language (with Boban Arsenijević) Conclusions

4 Deriving the formal ontology of language 4.1 4.2 4.3 4.4

The grammar of reference The Topological Mapping Theory (TMT) Extending the TMT Extending the Extended TMT (with Txuss Martín) 4.4.1 From 3rd-person object reference to pronouns 4.4.2 Reference and Person

viii xi xv 1 1 4 15 23 33 35 35 37 47 53 59 69 73 75 75 80 88 95 101 110 115 117 117 119 130 139 139 144

vi

Contents 4.4.3 Semantic approaches to the essential indexical 4.4.4 The hierarchy of reference revealed through Romance object clitics 4.4.5 Summary 4.5 Conclusions

5 Cross-linguistic variation 5.1 5.2 5.3 5.4 5.5 5.6 5.7

The apparent problem of linguistic variation Linguistic variation vs grammatical variation Diagnosing grammatical variation Revisiting the head parameter Revisiting the null subject parameter Revisiting the alignment parameter Conclusions

6 The rationality of Case 6.1 6.2 6.3 6.4 6.5 6.6

The apparent irrationality of Case What do the Cases mean? Vergnaud’s conjecture and ‘abstract Case’ The Case filter, reference, and Tense Event mereology and Case Conclusions

7 Language and speciation 7.1 7.2 7.3 7.4

Why linguistics is biology Language evolution in a new key The speciation of modern Homo sapiens Conclusions

8 Biolinguistic variation 8.1 Questioning double dissociations in developmental language disorders 8.2 Thought in genetically normal language-less or language-impaired adults 8.3 Thought without the linguistic genotype 8.4 Variation in the linguistic genotype 8.5 Formal thought disorder and the de-integration of grammar 8.6 Grammatical meaning in the brain 8.7 Conclusions

149 153 172 173 175 175 178 181 184 193 197 201 202 202 204 211 218 223 236 238 238 240 250 260 262 263 266 275 279 285 292 295

9 Thought, language, and reality

296

9.1 The nature of semantics

297

Contents 9.2 9.3 9.4 9.5 9.6 9.7

How can it be that truth is a grammatical concept? The grammar of truth: a unified account The limit of grammar and the limits of thought Language and reality Is there a science of thought? Conclusions

References Index

vii 304 311 317 322 328 336 339 375

Acknowledgements WH In 2008, I received a note from Professor Elisabeth Leiss from the University of Munich, arguing that the position I had defended in my 2006 book was not so much a ‘minimalist’ position as a ‘modistic’ one, in the sense of fourteenth-century Modistic universal grammar. This note would prove to have a profound influence on the course of my research and that of many others over the next four years. I suggested applying for a project, and it was awarded under the name ‘Un-Cartesian Linguistics’ in the frame of a DFG-AHRC bi-national agreement (AHRC award AH/ H50009X/1), from 2009 to 2012, to the Durham and Munich departments. Apart from the principal investigators, there were three post-doctoral project members, Dr Martina Werner (2009–11), MS (2010–2011), and Dr Alex Drummond (2011–12), and three doctoral students, Ulrich Reichard, Surachai Payawang, and Jadwiga Piskorz. Elisabeth’s linguistic thought has had a profound influence on me, and I have seen the fundamental difference between lexical and grammatical meaning most clearly through conversations with her. I leave to the historians the question of how close, in the end, this book has come (back) to the position of the Modists—but it is greatly inspired by them (see also Leiss and Abraham, to appear). Ulrich Reichard played a bigger role in this project than I could have ever expected any PhD student to play. An inexhaustible source of energy, inspiration, equilibrium, and constructive criticism, Uli has done wonders to keep the group together, sharpen ideas, and drive it in new directions. His role in this project as a co-author of several papers has been massive, and also his role in the genesis of this book cannot be overstressed: he has commented meticulously on every chapter, and the book wouldn’t be what it is without his help. At Durham, our project was embedded in my research group on Language & Mind, consisting of six further doctoral students: Tom Hughes, David Kirkby, Alex Malt, James Miller, Andrew Woodard, and Pallavi Worah. Thanks to you, David and James, for reading and commenting on major parts of the book. As time passed, the character and dynamics of this group became extremely special to me. It consisted of researchers who couldn’t be more different in their intellectual temperaments, but who nonetheless appeared to speak a common language, with a lot of common ground and intellectual ambition. Every single one of the projects that emerged out of this group seems unique and special to me (see Kirkby, 2013; Miller, 2013; Reichard, 2013; Malt, 2013; Worah, 2013; Hughes, 2013; Woodard, forthcoming). I sincerely hope that

Acknowledgements

ix

cooperation in this group was as rewarding for others as it was for me. In this world of academia, it seemed a sane place. My ideas, that is, seemed less insane than normally. Just a train ride of 15 minutes away, a further connection emerged with language sciences in Newcastle, and I am extremely grateful and honoured by the interest that Anders Holmberg, Noel Burton-Roberts, and many others in Newcastle took in our work. A long stream of visitors, too, deserve special thanks for inspiring and enriching us, particularly Cedric Boeckx, Emma Borg, John Collins, John Mikhail, Paul Pietroski, and Ian Roberts. Many further visitors that have helped and inspired us in numerous workshops and conferences over the past years are simply too numerous to mention, but I have been particularly influenced by Eric Reuland and Johan Rooryck. Farther away, a number of fabulous philosopher-linguists and friends have helped me find my way throughout these years, and some of their fingerprints can be found dotted all over this book: notably Cedric Boeckx, Pino Longobardi, Jaume Mateu, Nirmalangshu Mukherji, Paul Pietroski, Tom Roeper, Anna-Maria di Sciullo, Rajendra Singh, and Tom Stroik. The ear that linguists have lent my concerns has been, professionally speaking, the best thing that ever happened to me. This book would also be absolutely nothing without those that have become my close collaborators and co-authors over many years: apart from my present co-author herself, this particularly applies to Boban Arsenijević and Txuss Martín. On the brain sciences side, Nathalie Tzourio-Mazoyer has been a never-ending source of wisdom over the last two years on any matters concerning the brain; David Poeppel, too, has been friendly and cooperative indeed. But an extremely special thanks goes to Tim Crow, whose groundbreaking views on human speciation and the aetiology of schizophrenia has in part changed the course of my research. Last but not least, I feel a deep gratitude for two linguists who, at one point, understood me better than I did, and brought me on my way: Juan Uriagereka and Noam Chomsky. A few months into our project, John Davey, our editor, who has been an unfailing source of advice, wisdom and support for me for more than a decade, took an immediate interest in our plan for a book. He and his co-editor, Julia Steer, are simply wonderful to work with. But who would we all be if some human beings did not keep us down to earth: Ariadne and Konstantinos, my beloveds, and my non-linguistic friends: Amandine, Roland, Machiel, Robert, Sylvia, and Vroni. I may not deserve it, but please continue keeping me down and close.

MS The project which gave rise to this book took me, in 2010, from the relative comfort of the department of English Language and Linguistics in Newcastle to the less familiar territory of the Department of Philosophy in Durham. This transition, whilst extremely stimulating and thought provoking, could have been difficult had it not

x

Acknowledgements

been for the welcome I received from the people I met in Durham. My co-author was, of course, largely responsible for this, but so too were his PhD students Tom Hughes, David Kirkby, Alex Malt, James Miller, Andrew Woodard, Pallavi Worah, and especially Ulrich (Uli) Reichard, with all of whom I had the pleasure and privilege to interact on a regular basis during my brief time in Durham. They each taught me a great deal about philosophy along the way and introduced me to questions I had never before even considered. Teaching the course ‘Language & Mind’ along with Uli was particularly stimulating in this regard, and I must thank also the undergraduate students who took that course for their probing questions and unfailing interest in the intersection between linguistics and philosophy. During the writing of this book, I had the pleasure of bouncing ideas off many people on many different occasions. Parts of Chapter 4 and the associated Linguistic Analysis article were presented at the universities of Manchester and Cambridge, at ‘Rethinking Parameters’ in Madrid, and at the ‘Workshop on clausal and nominal parallels’ in Aarhus, where the respective audiences provided very useful questions and critique. Thanks go especially to George Walkden, Ian Roberts, Theresa Biberauer, and Adam Ledgeway in Cambridge; Silvio Cruschina, Yuni Kim, and Andrew Koontz-Garboden in Manchester; Olaf Koeneman, Olga Fernández Soriano, and Giuseppe Longobardi in Madrid; and Eva Engels, Guglielmo Cinque, Per Anker Jensen, and especially Sten Vikner in Aarhus. A version of Chapter 5 was presented at the workshop ‘UG: the Minimum’ in Durham in 2011 and my thanks go also to the audience there and at the associated British Academy-funded conference ‘The Past and Future of Universal Grammar’ for critical feedback and discussion, especially Anders Holmberg, Halldór Ármann Sigurðsson, and Aritz Irurtzun. Other people have also been kind enough to discuss some of the ideas presented in chapters 4, 5, and 6 on a more informal basis, notably Jenneke van der Wal, Joe Emonds, Heather Burnett, and Norma Schifano. None of the people mentioned, of course, is responsible for any errors or omissions in what follows. Outside of the academic domain, I also received a great deal of support and encouragement from family and friends whilst writing this book. Jenny, my running partner, Emily and the rest of my football buddies, Lucy, Richard, Sophie, Emma, Imke, and Simon, you, along with José, did a pretty good job of keeping me sane whilst my thoughts were elsewhere. Thanks for putting up with me!

List of Abbreviations ABS

Absolutive Case

ACC

Accusative Case

ASD

Autism Spectrum of Disorders

BPS

Bare Phrase Structure

C-I

Conceptual-Intentional

C-T

Complementizer-Tense

CL

Clitic

CP

complementizer phrase

CRTM

Computational-Representational Theory of Mind

DP

determiner phrase

ECM

exceptional case-marking

ECP

Empty Category Principle

EPP

Extended Projection Principle

ERG

Ergative Case

fMRI

functional magnetic resonance imaging

FOFC

Final-over-Final Constraint

FTD

formal thought disorder

GB

Government and Binding

IM

internal Merge

IP

Inflectional Phrase

I/T

Inflection/Tense

LCA

Linear Correspondence Axiom

LOT

Language of Thought

MEG

magnetoencephalography

MFC

medial frontal cortex

MPU

Main Point of Utterance

NOM

Nominative Case

NP

Noun Phrase

NSL

null subject language

PART

Partitive

PF

Phonetic Form

xii

List of Abbreviations

P&P

Principles and Parameters

PP

prepositional phrase

PSG

Phrase Structure Grammar

SC

Small Clause

SLI

Specific Language Impairment

SMT

Strong Minimalist Thesis

STS

superior temporal sulcus

SUBJ

Subjunctive

TMT

Topological Mapping Theory

ToM

Theory of Mind

TP

tense phrase

UG

Universal Grammar

UTAH

Universal Theta Assignment Hypothesis

VP

verb phrase

WALS

World Atlas of Language Structures

XP

X Phrase

%

variability in acceptability judgements

=

glossing convention for clitics

Like everything metaphysical, the harmony between thought and reality is to be found in the grammar of the language. (Wittgenstein, 1953: }55)

Introduction As we approach a new domain of inquiry, we typically do it with caution. Initially, we will barely know what we are talking about. A shifting fabric of data, assumptions, and predictions will form, until we eventually arrive at something that we feel we might dignify with the term ‘theory’. This theory will give way to others, a process influenced by theoretical interests and concerns reflected in various technical concepts and idealizations, and unpredictable flights of genius or the imagination. Within layers and layers of theoretical descriptions, an ontology may eventually crystallize: a sense of what kind of entities and structures our domain actually involves. Grammar is a case in point. Though commonly viewed as simply a social convention, which in more or less useful ways constrains our use of words, it has been the subject matter of a struggle for properly describing and identifying it as a domain of scientific inquiry for more than two millennia. This began at the very inception of scientific inquiry in India, and it continues today, 400 years after the onset of the scientific revolution in the West. There is little doubt about the existence of grammar as a way of organizing words that is distinct from phonology. But what, exactly, is it that we are talking about here? What domain of scientific inquiry, if any, is grammar? The question could be dismissed if it wasn’t actually of genuine human interest. Grammar is among the things most immediately present as we open our mouths and talk. No word that we utter or that comes to our minds lacks grammatical properties, which are that in virtue of which a word functions in the way it does, in the context of others. No word is a ‘bare symbol’. Where there is one word, others are never far, and grammar is the system within which they find their places, creating ever new meanings as they combine. Grammar is the chemistry of words. So grammar is never not there, and as words combine, sentences form, which are among the most curious objects of nature: they can be true or false. A creature understanding sentences is a creature understanding the notion of a fact. It knows about objects and that they can have properties; about events and that they can happen; about propositions and that can be true, though they need not be. A world known in this way is, we will claim, a grammaticalized world: a world that provides a framework for thought, and for science and art. So grammar may actually matter. But then again, it may not. Perhaps all that it seems to accomplish, as a principle organizing a specifically human form of experience, is really accomplished by something else. Perhaps a creature thinking as we do is not inherently a grammatical creature at all. Thought, as a system, might be

xvi

Introduction

autonomous, and happen in a non-linguistic brain much as it does in ours. And so, if grammar is non-arbitrary, it would be only insofar as it reflects such a system of thought. That there is a choice in the matter shows there is a question, and this is the question we address here. What is grammar really all about? Why does it exist? What does it do? The caution referred to above will typically, as scientific inquiry tries to get going, lead to background such ‘philosophical’ questions—and rightfully so. We might, for example, simply ignore the actual functioning and interpretation of the system, and start by describing its operations purely formally, regarding it as consisting of arbitrary symbols manipulated by formal rules. Yet, we feel that, after two millennia, it is time to address the question head-on, and defend a specific claim about the content of grammar. Since grammar is universal in human populations, severe pathologies or environmental circumstances aside, it is ‘universal grammar’ that we are addressing: the grammatical, as such. The claim will be: grammar, in imposing a curious kind of structure on lexical concepts—grammaticalizing them—enables a novel form of knowledge of the world, which would not exist otherwise or lexically, involving reference and predication. Grammar, as we will put it, is the world as known. It is the world as it appears in the format of knowledge, where this format is a uniquely grammatical one. This format of knowledge is universal in our specific version of humanity, as compared with all other hominins in the prehistory of modern Homo sapiens, and defines this peculiar version of humanity. For this reason, the topic of universal grammar, we claim, is the topic of our speciation. Thinking about human biology, about what defines our species, is to think about grammar, and vice versa. A much more common viewpoint is what we will call the ‘Cartesian’ one (never mind that Descartes himself may never have defended it): that thought is universal and immutable in this species, while language is merely a contingent way of expressing it in a physical medium. According to this view, grammar is not an organizing principle of our specific kind of mind, or of how meaning is organized in it (though it may, at least partially and imperfectly, ‘mirror’ it). So if grammar went missing or disintegrated, much of our mind could stay intact. In a similar way, a printer or fax machine attached to your personal computer won’t much affect its internal computations as driven by its software. Meaning, on this view, might be said to consist in things in the world, viewed as the meanings of words and sentences; in mind- and language-independent ‘propositions’; in internal ‘psychological’ or ‘cognitive’ representations; or in the ‘use’ we make of words. But it doesn’t consist in language. But it is our linguistic mind, by containing and combining lexicalized and grammaticalized meanings, we contend, which creates meaning of a new sort. We defend this alternative, because we claim that grammar explains the species-defining features of the mode of thought in question, and because we have no idea how else to explain them. The kinds of meanings expressed in any human language, at least in part, are

Introduction

xvii

not like the contents of the thoughts of any non-linguistic creature. Grammar explains, in a principled way, why these meanings exist, and how and why they arise when they do at particular moments in our lives—why, in particular, words occur in acts of reference that can take a narrow range of forms; why there is both event and fact reference; why properties are distinguished from objects or states of affairs; and why we refer to persons not as we refer to their bodies. One can state— and formalize—these facts. The challenge is to explain them. Re-thinking the foundations of grammar benefits from putting the contemporary version of universal grammar research into perspective, and we do this in Chapter 1. History helps, both to see continuity in our insight into the nature of grammar over the aeons, and to see quite different takes on it. In contemporary linguistics, historical awareness of the roots of the contemporary version of the enterprise tends to stop with the rationalist Port Royal tradition; but there are at least two other sciences of grammar, which offer quite different visions of what grammar actually is: in particular, some later instances of Ancient Indian grammar, and Modistic grammar. Wedding the enormous descriptive and empirical richness of modern generative grammar to these more ancient and medieval frameworks of thought, leads, we contend, to an ultimately more interesting result. Chapter 2 begins a journey that answers this question: What are the effects of grammar on the organization of meaning? A logically possible answer is: none. That is, the grammaticalization of the hominin brain changed absolutely nothing regarding what meanings we can grasp or thoughts we can think. But this answer seems far too radical. In our species, we see grammar where we see the mode of thought in question. In other species, we see, in essence, neither. Where language breaks down due to brain lesions or other acquired disorders, much thought can be preserved. But in this case, the linguistic genotype is normal, and language has developed normally in the patient’s life. Where the linguistic genotype is not normal, we see an effect of that on our mental phenotype. Prima facie, then, it seems implausible that grammar would have no such effect on the organization of our rational minds. So what is the effect, exactly? Chapter 2 begins from a form of semantics already present in prelinguistic perception. But as percepts become lexicalized as words, and as words then start to develop grammatical functions, the organization of meaning changes again, and with each step. We specifically develop an account of the kind of semantics associated with lexical atoms and with part of speech distinctions: the distinction, say between how the word run functions when read as a noun (‘a run’) or as a verb (‘to run’, ‘runs’). Once we have arrived there, we argue, we have already arrived at grammar, and the semantics in question reflects how words implement grammatical functions: grammatical semantics. These grammatical functions seem to be universal: they depend on relations between words, which as such do not depend on the word classes (parts of speech distinctions, like between nouns, verbs, and adjectives), which we also contend are

xviii

Introduction

ultimately language-specific and should not be part of the vocabulary of universal grammar. And so, in Chapter 3, we develop an account of grammar as purely relational, and as defining a novel semantics in which reference and truth are possible. Moreover, we argue that the recursivity of language can be reconstructed, predicted, and understood on this basis. The chapter also develops an account of the smallest unit of grammaticality, which we identify with the phase of contemporary grammatical theory. This unit of grammatical organization moves to centre stage in Chapter 4, where we claim to derive formal-ontological distinctions in the objects we refer to in language from the grammatical dynamics in which such acts of reference are configured. An example of what we call a formal-ontological distinction is the difference between an action and an object, or an object and a fact. These are not distinctions like those between a man and a woman, or between whisky and water, or between water viewed as H2O or as XYZ. They are formal distinctions, not material ones. They do not concern existence in a metaphysical sense. The availability of a system of reference in which every single act of reference involves such a formalontological distinction is the most astonishing fact: it needs to be explained, and to follow from something. We argue that it follows from grammar, and from nothing else. Grammar is the only known principle that can, and does, organize the forms of reference in question. In this sense, formal-ontological distinctions are, in essence, grammatical distinctions, and the world as characterized by such distinctions is a grammaticalized world. The formal ontology of semantics co-varies with grammar, and viewed as a semantic or metaphysical ontology, it has no independent status. This has an important consequence for the relation between grammar and thought. For the kind of content that comes, in our argument, with grammar, and the formal ontology in question, is a (or even the) prime feature of the kind of thought that philosophers since ancient times have highlighted: thought that can be true or false, that is responsive to reasons, and that can be shared. If the content and formal ontology in question fall out from grammar, there is nothing else for a theory of thought to accomplish: a theory of (such) thought will be a grammatical theory. Hence there is no independent ‘Language of Thought’: it is not needed. The reason that it was taken to be needed is that language was merely taken to be a peripheral organ, or an expressive tool. If this expressive tool is subject to massive cross-linguistic variation, as it plainly seems to be, and thought is not, then there must be a categorial thought-language distinction. And so we need to address the issue of cross-linguistic variation. Does it provide a challenge for our account of universal grammar? Not if the primary dimension of variation is the organization of language at a lexical and morphological level; not, if traditional dimensions of variation, as captured through syntactic parameters, are given their proper place in the grammar, as only affecting their sensory-motor externalization.

Introduction

xix

Yet there are other, massive challenges to our account. Some aspects of linguistic organization traditionally identified as grammatical, such as Case, are said not to feed into the organization of thought or meaning—at least not directly. So why are they there? If they exist in the organization of grammar, they therefore either disconfirm the view that grammar necessarily subserves the organization of a special kind of meaning, or they are wrongly taken not to feed into the organization of meaning. We argue that once morphological Case features are carefully distinguished from the grammatical relations they express, they turn out to be interpretable indirectly in the formal ontology of meaning. This is argued in Chapter 6. In chapters 7 and 8, which form a unity, we look at variation of a very different kind, which we call biolinguistic variation, following Boeckx (2012): variation in the linguistic genotype, and the shifts in our mental phenotypes that such variation can systematically induce, within the genus Homo and within our own species. Comparison of these cognitive types lends support to our ‘Un-Cartesian’ contention: that grammar is the fundamental organizational principle behind a cognitive phenotype that is unique to our species, and defines it. In Chapter 8, we devote particular attention to Formal Thought Disorder in schizophrenia, for which the Un-Cartesian hypothesis makes an obvious prediction: grammar should disintegrate in this condition as well, along with the fragmentation of thought that we observe there. This, we conclude there, turns out to be a real possibility: looking at grammar with UnCartesian eyes, therefore, may throw light, not only on thought, but on mental health as well. That language in its normal use is a condition for mental health seems a natural suggestion: a sane thinker is also a speaker. Yet language has barely been looked at as playing this role, which reflects its status in philosophy and psychiatry today. Having argued all this, we turn into a more philosophical mode in the final chapter. If forms of reference such as truth and object-reference are grammatical in nature, what does this really mean, in terms of the meaning of the notions of reference and truth involved? ‘Truth’, in what sense? This notoriously inscrutable notion intrinsically connects with metaphysical issues: issues of objectivity, in particular. If truth is a grammatical concept, as we centrally contend here, how does this affect our notion of what truth is? Can we really mean by ‘truth’ what contemporary discussions of the notion since Tarski have taken it to mean? Yes, is the answer—it is just that the notion in question has been presupposed in these discussions. The story we tell, by contrast, is an explanatory one. It provides an aetiology for it: a rationale for why it exists, and has the features it does. Seen as a whole, this book reflects the ambition to rethink the place of grammar in human nature, and thereby to rethink its place in philosophy as well. We wanted to weave inquiry into the nature of grammar into a single scheme, in essentially all of its central dimensions: philosophical, linguistic, historical, evolutionary, biological, neurological. We wanted there to be a single book that does all of this, in a coherent and unified, theoretically principled, and empirically sound manner.

xx

Introduction

We hope these are benefits that outweigh the inherent difficulties that mar interdisciplinary work. Ultimately, our ambition is philosophical, and we take this to be a philosophical book. Yet, at the same time, we do not think that essentially any of the points we will be making is independent of the empirical facts of language and how we interpret these. Shielding the philosophy of language from systematic exposure to such facts predicts nothing good, we think. While we are crucially trying to weave a whole picture, then, quite a few parts of this book delve ruthlessly into linguistic details. This is a risk for us, given that we want this to be a philosophical book and to be read by philosophers as well. But we have trusted that the project is worthwhile, and of intrinsic human interest, disciplinary separations aside. In more practical terms, we trust that, where the linguistic details get too gruesome, as for example in Section 4.4 of Chapter 4, in Chapter 5, and in parts of Chapter 6, philosophers will simply choose to read introductions and summaries, and otherwise turn a blind eye. We stress that, apart from its first section, Chapter 5 really is written with linguists in mind, and the same is true for all but the first two sections of Chapter 6. We are no less concerned for our linguistic readers, and have reserved much of the metaphysical dimensions of our story for Chapter 9, leaving our empirical claims about, say, the grammar of reference or Case assessable independently. While this is very much a collaborative book, with both authors working closely together throughout, WH can be considered primarily responsible for the material in all the chapters with the exception of chapters 4 and 5. The first three sections of Chapter 4 are based on Sheehan and Hinzen (2011) and the fourth on Martín and Hinzen (2012). In addition, Section 3.6 and parts of Section 3.5 of Chapter 3 are based on Arsenijević and Hinzen (2012). Last but not least, we emphasize that what we call ‘Un-Cartesian linguistics’ is a programme. Like the Minimalist Program, you can engage with it or not, but it cannot be refuted, in the way a hypothesis can be (though it can be shown fruitless). As in the case of the Minimalist Program, engaging with our programme promises rewards: to make more principled sense of grammar. But there is nothing wrong in not pursuing it, and nothing intrinsically right in doing so. If you think that thought is ‘mental’ and ‘non-linguistic’, as the (or at least some) Cartesian rationalists did, and that it can be investigated in these terms, we find nothing wrong in this: programmes are judged by their fruits. At the same time, we do not want to overstate this point, for we would like to be refuted: we would like a reader to show us that grammar does not organize the formal ontology of semantics, that semantics is autonomous, and that thought is independent of grammar. Therefore, we formulate, over and above an Un-Cartesian programme, an Un-Cartesian hypothesis: that the formal ontology of language, and hence the content of thought, obeys grammatical principles.

1 The project of a science of language 1.1 Grammar as a domain of inquiry If you tell me that your father was six foot tall, something normally happens that is rather surprising, if we think about it. I may have never seen your father, and he isn’t present. Yet I recognize that what you are telling me is a possible item of knowledge. Indeed I will assume, at least more often than not, that you are telling me a truth. And so I will typically come to assume that it is a fact that your father was six foot tall, which as such obtains independently of what we say or believe. That a notion of truth is involved is part of what defines thought itself: as Frege put it, ‘I call a thought something for which the question of truth arises’ (Frege, 1918–19/1956: 292). So somehow it appears as if, quite apart from our perceptual encounters with the non-linguistic world, we can extract knowledge from grammatical strings of expressions, based on thought. While perception and the extraction of adaptively relevant information from sensory data via inference are ubiquitous in the animal world, and perception, too, is normally of objects as existing independently of us, knowledge obtained in this other way is a singularity, and it has a different quality. We cannot perceive that some particular thought is true or false, in some objective sense. Without this peculiar other source of knowledge, then, there would be no science, and humans would have no history. That grammar is a distinctive way of organizing meaning and thought, making knowledge possible, is the central thesis of this book. It is also the basis of the claim we make for the philosophical and epistemological significance of grammar. It gives a novel account of what grammar actually is, or does. Versions of this idea are as old as human scientific inquiry into grammar itself, and they are at the heart of the very first attempt to develop a ‘science’ of language more than 2,500 years ago in Ancient India (Matilal, 1990). In the West, grammar was discovered as a potential domain of scientific inquiry only much later, but the idea of a connection between grammar and a different way of cognizing the world figures prominently in ‘modistic’ universal grammar in the fourteenth century as well (Covington, 2009). The same applies to Edmund Husserl’s (1900/2009, 1929/1969) remarkable project of a ‘pure logical grammar’, the most elaborate as well as ignored

2


universal grammar project in the twentieth century to come from philosophy (Edie, 1977).1 We do not see this idea as clearly, or indeed at all, in the Port Royal tradition (Arnauld and Lancelot, 1660), or in its modern incarnation (Chomsky, 1966). Since we are nowhere near a genuine science of grammar today and the struggle for making sense of it as a domain of enquiry very much continues, we believe that our discussion will benefit from taking a broader historical perspective, within which contemporary assumptions can be properly contextualized.2 The philosophy of grammar we will develop here gives grammar a principled role to play in the genesis of a particular kind of meaning: grammatical meaning, as we will call it, which is to say, meaning intrinsically depending on grammatical principles of organization. In a contemporary setting, such a notion may well strike one as contradictory. Upon opening a standard philosophy of language textbook (e.g. Lycan, 2008), one finds essentially four theoretical conceptions of meaning discussed: (i) meaning as a psychological ‘idea’ (a ‘mental particular’); (ii) words and sentences ‘stand for’ things in the world, which are their meanings (the ‘reference theory’), (iii) meaning as an abstract object outside of both the mental and physical realm (the proposition theory), and (iv) the ‘use theory’ (meaning as deriving from how we use words). The mental particulars posited on the first conception are not linguistic in nature; nor are the physical objects in the world that define the meaning of words and sentences on the reference theory; nor are, evidently, propositions; nor is the use of words on the fourth conception, where this ‘use’ is thought of in more generic terms and analogized with chess playing in Wittgenstein (1953), governed by the ‘rules and conventions of a community’ like many other non-linguistic activities.3 So, somehow, it appears as if language simply need not be discussed in a philosophy of meaning, and that there is no grammatical meaning. In line with this, much of the focus of the philosophy of language is on the word (and indeed, the name) as the basic unit of meaning and reference. In thinking about mind, modern philosophy more generally, and the Port Royal tradition in particular, has favoured the idea that there is a universal system of thought to which all humans somehow have access, defining the contents of their ‘propositional attitudes’ and representing the true locus of meaning, while language is

1 Husserl’s project is different from both the Modistic and the Port Royal one, which Husserl references, and it is crucially meaning-based, involving a notion of grammatical meaning much in our sense below. Unfortunately, for reasons of space, Husserl’s enterprise will receive only scant attention in this book (though see fn. 10). 2 We will, in what follows, write ‘universal grammar’ for the enterprise of a universal science of language as such, while reserving capital letters (Universal Grammar/UG) for the modern technical sense of universal grammar. 3 The early Wittgenstein thought of sentences as of ‘pictures’, in the so-called ‘picture theory’ of meaning (Wittgenstein, 1922: sentence 2.1), which is yet another notion of meaning not invoking linguistic organization in any specific sense.

1. The project of a science of language

3

merely a way of expressing such meaning contents in a public medium. Frege gives a classical formulation of this picture: The thought, in itself immaterial, clothes itself in the material garment of a sentence and thereby becomes comprehensible to us. We say a sentence expresses a thought. A thought is something immaterial and everything material and perceptible is excluded from this sphere of that for which the question of truth arises. (Frege, 1918–19/1956: 292)

As Frege adds on p. 302 in a footnote, comparing a thought to a planet that is independent of us seeing it: When one apprehends or thinks a thought one does not create it but only comes to stand in a certain relation, which is different from seeing a thing or having an idea, to what already existed beforehand.

According to Frege, the thought is thus prior to and independent of its linguistic articulation. As such it is also the locus of semantics or content, and grammar has nothing to do with its existence or intrinsic nature. Accordingly, from this point of view, it is hard to see wherein the epistemological significance of grammar might lie, except for providing us with an arbitrary means to ‘reach out’ to such objects or to convey them. The empirical study of grammar, it would then seem, is just more psychology, which Frege and many others after him axiomatically banned from what would become analytic philosophy. A position of this nature can motivate the consistent lack of a systematic study of grammar in philosophical curricula in the philosophy of language over the last 150 years, where logic and semantics have been the frameworks of choice, and linguistics has had little influence on the philosophy of language.4 The inherent epistemological significance of grammar also disappears on another prominent twentieth century version of the same Cartesian-rationalist view, where language as the expressive system in question is encapsulated in a mental ‘module’ (Pinker, 1994; 1997), separate from general cognition and the ‘beliefs’ and ‘desires’ that constitute human reason (Fodor, 1975; 2000; 2001). We explore a different hypothesis here: that grammar, rather than being merely expressive of thought, is generative of the specific kind of thought that we find expressed in languages across the world, and a rationale for its existence. It provides a thought with a novel structural format that makes it possible, thereby enabling knowledge and forming a crucial element in the genesis of human reason itself, as well as a new species (see Chapter 7). In this opening chapter, we begin by tracing Cartesian rationalist linguistics to its beginnings in the seventeenth century, and identify a core problem in it that we think is responsible for the perennial instability of the universal grammar enterprise and 4 The concepts and premises of the two paradigms seem barely compatible, as a look at Chomsky (2000a) or Antony and Hornstein (2003) suggests.

4


the persistent criticism it has received over the last three centuries. This problem is the very separation of a universal, immutable world of ‘thought’, ‘concepts’, or ‘notions’, on the one side, from an expressive and variable system of ‘language’, on the other. After the demise of the Port Royal tradition and universal grammar itself in the nineteenth century, the reinvention of the universalist project in the mid-twentieth century in linguistics still proceeded on the old Cartesian premises: a notion of grammatical meaning—that some kinds of meaning are intrinsically grammatical— doesn’t surface. Instead, the attempt is to turn grammar into a ‘natural object’ defined by its formal properties only, with no reference to meaning, though perhaps ‘interfacing’ with it, as in contemporary Minimalist grammar. In the Ancient Indian context, as we discuss in Section 1.3, grammarians crucially did work with a notion of grammatical meaning, and they made it the basis of their metaphysics, much as we will do in Chapter 9. The same applies, in other ways, to Modistic grammarians, which we discuss in Section 1.4. Again, the objective of grammar as a science is achieved there, not by abstracting from meaning, but by attempting to delineate the specific kind of meaning that comes with grammar—for the Modists, meaning associated with the parts of speech. We will partially disagree with this particular Modistic conclusion in Chapter 2, where we conclude that parts of speech distinctions, insofar as they exist in a language, already reflect the grammatical word: grammatical functions that words play in sentence contexts. These can be reflected in morphological properties of words, but the latter, unlike the grammatical functions, are variable across languages. Section 1.5 concludes.

1.2 Cartesian linguistics, 1637–1966 The philosophical conception of language as a tool for the generation of knowledge is quite different from the one that underlies modern linguistics. The latter is an instance of the rationalism that characterizes Cartesian linguistics in the tradition of Port Royal, which in turn leaned heavily on Descartes’s Regulae and was revived in Chomsky (1966). Central to this conception is a distinction between grammar and logic, with the former viewed as the ‘art of speaking’ (Arnauld and Lancelot, 1660) and the latter as the ‘art of thinking’ (Arnauld and Nicole, 1662). In this Cartesian universe, speech is squarely located in the ‘material’ world, while thought uncontroversially belongs to the ‘mental’ world—a ‘dualism’ that the quote from Frege above illustrates again, though in his own nineteenth-century terms. Thought, in short, is universal and structured by logic, while grammar is a conventional system for expressing it ‘materially’. This directly predicts that any claim about a ‘universal’ grammar will have to come out as controversial, for it is virtually inconceivable that, given the diversity among the world’s languages in their external appearances (and that they are not all like Latin, which was thought for centuries to codify the canons of logic most closely),


5

linguistic categories will match up uniformly with the categories of logic (or semantics) viewed as given a priori. That is, set against a world that is universal by definition, the science of grammar is bound to fall victim to abundant variation. As we shall see below, this almost inescapable scepticism about universal grammar is indeed a recurring theme where rationalist assumptions about language and thought are made. The following quote from the Grammaire générale et raisonnée illustrates the rationalist viewpoint: La Grammaire est l’Art de parler. / Parler, est expliquer ses pensées par des signes, que les hommes ont inventez à ce dessein. / On a trouvé que les plus commodes de ces signes, estoient les sons & les voix. (Arnauld and Lancelot, 1660: 5)5

A grammar, therefore, is a ‘rational’, ‘scientific’, ‘general’, or ‘philosophical’ grammar, only insofar as it captures the universal structure of thought, while, to the extent that it does not, it reflects the arbitrariness of an expressive tool. Put differently, the nonarbitrary aspects of grammar (the ones amenable to science) are precisely the ones that reflect the ‘manner in which men use [signs] for signifying their thoughts’ (Arnauld and Lancelot, 1660: 41). A grammar will be ‘perfect’ when it is a perfect mirror of or transparent ‘window’ into the world of thought. The contents of this world are ‘thoughts’ and ‘ideas’, which are structured by operations described as ‘mental’, such as conjunction, disjunction, conditioning, affirmation, and denial, which speech is there to externally signify. Characteristically for this conception, the origin of this world of thought—or semantics as such—is not addressed. Against the background of Cartesian mind–body duality, speech is also a function of the ‘body’, and segregated from thought viewed as a function of the ‘mind’. So there will be no unified science of grammar and thought, any more than there can be a unified theory of noise and meaning. Cartesian dualism soon became obsolete through Newton’s destruction of the mechanical philosophy that defined the notion of body involved (Chomsky, 2000a), and even Frege’s nineteenth-century version of it has recently come to be questioned in approaches that regard propositions as a consequence of what kinds of mind we possess (Soames, 2010).6 Yet, in some ways, related dualisms have lived on in the study of language.

5 ‘Grammar is the art of speaking. Speaking is to explicate one’s thoughts through signs, which humans have invented for this purpose. One has found that the most suitable of these signs are sounds and the voice.’ (Our translation.) 6 Soames (2010: 7) develops what he terms a ‘cognitive realist’ account of propositions, according to which propositions are types of ‘cognitive events’ involving predication, which is said to be a ‘primitive cognitive act’. But the notion of predication in question is a grammatical one, it is one kind of denotation that clauses can intuitively have, and the term ‘cognitive’ does not increase understanding of why propositions (or the other types of clausal denotations) exist and have the structures they do (see further Reichard, 2013: ch. 1).

6


Thus, the Strong Minimalist Thesis of Chomsky (2007a) and the comparative approach of Hauser et al. (2002) are both based on the view that we need not account for thought: it is what provides the constraints for language to satisfy, much as in the expressive model above. It is a system that is part of the ‘Conceptual-Intentional (C-I) systems’, which are prelinguistic by definition and merely form an ‘interface’ with grammar. They are thus exactly what the grammarian does not have to account for. There also is the idea that the sensory-motor externalization of language in a physical medium is only a contingent aspect of language that it in principle could do without.7 That internal system, Chomsky (2008: 4) specifically suggests, evolved first, with externalization a later evolutionary development, ‘a secondary process’, which used already existing sensory-motor articulators. If so, grammar in its core aspects— in what remains after we subtract externalization—will be purely ‘internal’, a mental system (yet even as such still distinct from ‘thought’, from which it is divided by the semantic ‘interface’). On the other hand, the Minimalist Program is as such flexible enough to allow for quite different conceptions of what role grammatical structuring might play in regards to the posited internal systems of thought. One view could be that the grammar merely internally articulates them, giving them a linguistic form. Another is that it fundamentally changes and reorganizes these systems, thereby giving us a different mode of thought, if not thought as such. In the latter cases we clearly cannot speak of an ‘interface’ between grammar and thought any more, and this is the view we pursue here. Some statements of Chomsky are consistent with this latter perspective, such as the assertion that ‘UG primarily constrains the “language of thought,” not the details of its external expression’ (Chomsky, 2007b: 22). Here the word ‘constrains’ could be read as entailing that grammar provides a structural format on which a distinctly human system of thought inherently depends (something needs then to be said about what kind of thought this is). On the other hand, little in current Minimalist syntax gives us any insights about what it is about grammar that makes it play that role of constraining the language of thought. This is particularly true for conceptions of narrow syntax that reduce it to the combinatorial operation Merge: defined as the recursive formation of binary sets, it amounts to no more (or less) than discrete infinity, and thus yields no insights into what makes grammar specific, let alone a source of a particular and novel mode of thought. Indeed, Minimalist syntax has been virtually based on a conception of grammar on which it does not change the nature and structure of thought or the C-I systems, reducing grammar to a mere ‘linking system’ between meaning (assumed already given) and sound or sensory-motor articulation (also assumed already given, cf. Chomsky, 2005). 7 And that partially disguises the essential nature of the internal system that such externalization encodes (Boeckx, 2008; 2011). See also Burton-Roberts (2011).


7

Our more specific question in this book then will be how grammar might play a role in regards to a genuine theory of thought (or of semantics), a task that, as noted, was not addressed by the rationalists and has escaped us since. Such a theory is not only of interest in itself, we will argue in Chapter 7, but is also needed for biological reasons, namely in an account of human speciation: the distinctive ‘rationality’ of our thought is something that can be dated, and there is no good evidence for its existence prior to that of modern Homo sapiens, or even in the anatomical beginnings of our species: for in this species itself, there is no such evidence before about 75,000 years ago in Africa and 40,000 years ago in Europe, which is when a novel mode of thought is richly manifest in the unprecedented creativity of the essentially modern cultures arising then. This mode of thought needs to be explained, and we will argue that language explains it: not merely because we have evidence that language was present at these times and not (long) before them, but because the grammatical organization of meaning in language yields the characteristic features of this sapiensspecific mode of thought. The theory of thought we offer is therefore a unified theory of thought and language, and we do not know how to pursue an explanatory theory of thought or semantics—a genuine cognitive theory—otherwise. That theory will not turn out to be consistent with regard to physical externalization—a merely contingent aspect of linguistic organization which language, at least in principle, could do without. If we think of cognition as ‘embodied’, then language is the unique embodiment of the kind of thought we are addressing here, and embodiment requires a physical substance. We return to this matter in chapters 8 and 9. The uncontroversial universality of grammar in our species does not then lie, for us, in the fact that it reflects (or mirrors) a universal system of thought (‘notions’, ‘concepts’). This departure from rationalism is a crucial step, since, as noted earlier, it is precisely the virtually unavoidable divergence between a (presupposed) system of ‘concepts’ and the variation in their external expression that has historically lead linguists to reject Universal Grammar, as the following quote from Jespersen illustrates: With regard to the categories we have to establish in the syntactic part of our grammatical system, we must first raise an extremely important question, namely, are these categories purely logical categories, or are they merely linguistic categories? If the former, then it is evident that they are universal, i.e. belong to all languages in common; if the latter, then they, or at any rate some of them, are peculiar to one or more languages as distinct from the rest. Our question is thus the old one: Can there be such a thing as a universal (or general) grammar? (Jespersen, 1924: 46–7)

In other words, on this view the existence of a universal grammar stands and falls with the alignment of grammar and logic, with logic thought of as structuring thought and regarded as universal and immutable. Similarly, the categories of external reality (‘corresponding’ to ‘notional categories’ in the ‘mind’) are regarded as

8


universal and immutable. A universal grammar is then predictably rejected on the grounds that unlike thought (or reality), grammar appears highly variable and as such does not seem to reflect the immutable logical and semantic structure of thought or reality. As Jespersen specifically observes, citing the examples of ‘Man is mortal’ and ‘Men were deceivers ever’: A logician would have preferred a construction of language in which both sentences were in the same universal number (‘omnial’, as Bréal calls it) and in the same universal or generic tense, but the subject of the former in the common gender and that of the latter in the masculine gender, for then the meaning would have been unmistakable: ‘all human beings have been, are, and always will be mortal,’ and ‘all male human beings have been, are, and always will be deceitful.’ But as a matter of fact, this is not the way of the English language, and grammar has to state facts, not desires. (Jespersen, 1924: 54–5).

The rejection of a universal grammar, then, is the price to pay for maintaining the ethos of grammar as an empirical science, a pervasive attitude throughout the twentieth century, notably in Bloomfield. Jespersen at the time of writing in the early twentieth century already echoes a conclusion widely endorsed by prominent comparative linguists in the first half of the nineteenth century, where the death of Cartesian universal grammar was widely taken for granted: ‘A universal grammar is no more conceivable than a universal form of political Constitution or of religion, or than a universal plant or animal form; the only thing, therefore, that we have to do is to notice what categories the actually existing languages offer us, without starting from a ready-made system of categories’ (Steinthal, Characteristik, 104f.). Similarly, Benfey says that after the results achieved by modern linguistics universal and philosophical grammars have suddenly disappeared so completely that their methods and views are now only to be traced in such books as are unaffected by real science (Geschichte der Sprachwissenschaft, 306). And according to Madvig (1856, p. 20, KL p. 121), grammatical categories have nothing to do with the real relations of things in themselves. (Jespersen, 1924: 48)8

But let us rehearse how this conclusion was arrived at. First, a world of thought or ‘notional categories’ is separated from a world of so-called ‘linguistic categories’, and both of these from the categories of reality itself, where the categories of thought and those of reality are both taken as given (Jespersen says nothing about their origin, any more than rationalists did, or cognitive or functional linguists such as Langacker (1987) do, when they define word classes in terms of semantic ‘notions’ such as ‘object’, ‘property’, or ‘event’). Universal grammar, we then say, only exists if the categories of grammar reflect those of either thought or reality in some tight and

8 Steinthal’s conclusion, cited in the quote below, seems very similar to how universal grammar is rejected in writers such as Haspelmath (2007; 2010) today. It is also interesting how much of science is guided by largely a priori intuitions about what is ‘conceivable’ or not. Steinthal’s biological speculation about the ‘universal animal form’, in particular, is debatable and rejected in Sherman (2007).


9

universal fashion. But this is not the case, due to linguistic variation. Therefore there is no universal grammar. This general pattern of argument has not essentially changed in the last one hundred years. As Haspelmath (2010: 3) notes, Typologists have often observed that cross linguistic comparison of morphosyntactic patterns cannot be based on formal patterns (because these are too diverse), but has to be based on universal conceptual-semantic concepts.

The ‘conceptual-semantic concepts’ in question are effectively Jespersonian ‘notions’, such as ‘agent’, ‘recipient’, and ‘physical transfer’. The concepts meant to allow crosslinguistic comparison are assumed to also include some logical concepts such as ‘argument’, defined as, say, a ‘referential phrase that can fill a verb’s semantic valency position’ (where, crucially, ‘reference’ is not taken to be a grammatical notion, as it will be here). The concepts in question are classified as constructs of the typologist that allow language comparisons, but that do not correspond to anything in either a speaker’s language or mind. Indeed there are no cross-linguistic categories, for nothing in language corresponds one-to-one to the conceptual-semantic constructs. Haspelmath (2010) illustrates with the putative category ‘dative’: no ‘concept’ defined in conceptual-semantic terms like the above is ‘instantiated’ by, say, the Nivkh Dative-Accusative case, which, as it happens, encodes the causee of an act rather than the recipient. Neither is, Haspelmath argues, such a category instantiated by, say, the Russian Dative or the Finnish Allative, both of which, although they do encode the recipient of physical transfer actions, have many properties that such a definition will not capture, and which therefore cannot be equated as instances of the same category.9 Let us now consider how the argument would go if grammatical categories are sui generis, a genuine innovation in hominin evolution, reflecting a new mode or format of thought that did not predate grammar. Then the project of wanting to match such categories with those of non-linguistic thought is forlorn: one cannot ‘translate’ them into any other and non-grammatical—say, semantic—terms. Neither can one translate them into non-grammatically characterized categories of ‘reality’—say, ontological categories viewed as independent from those of language or thought. In fact, even percepts do not reflect external reality as independently (e.g. physically) described, a point to which we return in Chapter 2. Even less we would expect that the independent description of reality would reflect the grammatical relations in which we argue knowledge becomes possible. Now the entire starting point changes. If we do not allow ourselves to presuppose the categories of a Cartesian system of ‘thought’ or of ‘notions’, nor a ready-made categorial structure of external reality as it appears in the format of knowledge to us, then grammatical categories—apart from perceptual ones—are all we have to go by. 9

For our own take on Case see Chapter 6.

10


We can’t reduce them to anything else. If grammatical categorizations across languages are indeed arbitrary and variable, whereas those of thought and reality are immutable, there should be some evidence independent of grammar that the relevant non-grammatical categories exist, and that they can be accessed independently of grammar. But we know of no such evidence: they are inherently grammatical, or linguistic categories and relations in disguise. Even if so, the question now remains whether grammatical categories and relations are indeed arbitrarily variable. But there is little if any disagreement over the fact that, in all human languages, one can tell a truth and also say something false; that one can use a word to refer to an event and distinguish such an act of reference from one referring to an object; one can refer to an object (a ‘he’, ‘she’, or ‘it’) as distinct from the speaker and the hearer; one can express a proposition whose truth value is objective and distinguish it from an emotive or subjective expression, or an intersubjectively valid one; one can distinguish a property from the object that has it; and so on. These are defining features of human rationality. Suppose, secondly, that these very distinctions in reference are unavailable pre- or non-grammatically, and again that they cannot be characterized in independent semantic terms, as long as we do not implicitly invoke grammatical distinctions. Thirdly, suppose that a proper analysis of the parts of speech and their grammatical functions exactly aligns with these very modes of signification. Then the argument from variation would be moot. If things are so easy, why is the rejection of universal grammar so widespread? Again, history gives us some clues. The first half of the nineteenth century was characterized by a revolt in linguistics against the Port Royal model of grammar and the general attempt to align grammar with logic (Graffi, 2001). Instead, the linguists of the day—for example Steinthal, mentioned above, but also the earlier Neogrammarians, for example Hermann Paul, whose Principles of Linguistic History is from 1880, and the psychologist Wilhelm Wundt (1832–1920)—tried to align grammar with psychology, to the extent even that the terms ‘sentence’ and ‘judgement’ (German Urteil) came to be used virtually interchangeably. Grammatical processes, in short, reflect psychological ones involved in the genesis of a judgement viewed as a mental operation. In the later parts of the nineteenth century, however, there was a countermovement, in which notions such as judgement or ‘thought’ (the German Gedanke) were stressed to have non-psychological meanings, giving rise to the ‘anti-psychologism’ defended in various forms by Bolzano, Brentano, Frege, Husserl, Moore, Wittgenstein, and Russell (prior to 1909). Within philosophy, this indicated the need to return to an alignment of language and logic, and to address the issue of meaning in abstract propositional rather than grammatical terms. This alignment would remain in place for much of the twentieth century, where anti-psychologism became an axiomatic premise of ‘analytic philosophy’, and where the nature and structure of ‘content’ is a matter of metaphysics or, on ‘naturalistic’ conceptions, physics and


11

causal relations, but not grammar. Frege’s contention, that content is not a matter of psychological ‘ideas’ internal to the head is thus preserved, as is his idea that content is non-linguistic. The notion of grammatical meaning—that grammar itself could provide or organize a novel form of meaning—does not surface. In our proposal, of course, meaning is not a matter of non-linguistic ‘ideas’, either. But nor is it a matter of metaphysics or causality, being a matter of grammar (and of course the lexicon) instead.10 Nineteenth-century historical-comparative linguists never quite warmed to the question of linguistic meaning in the first place, focusing on phonology and the laws of sound change, in particular. While phonemes are beyond the threshold of human conscious apperception in ordinary language use, on the other hand, meaningful words are not. So, with a move to morphology, the mysterious box of meaning had to be opened. Bloomfield (1933: ch. 9) thus addresses the matter of meaning, yet going barely beyond the word level and treating grammar without reference to meaning. He moreover sticks to a behaviourist methodology where the meaning of a word is assumed to be the situation in which the word is uttered and the response its utterance elicits. A related position in Skinner (1957) would become a central target of Chomsky (1959), running directly counter to the creative aspect of language use that Chomsky adopted from the Cartesian tradition. But words as such are not the locus of human freedom or reference yet, in the sense that they are largely given to us as something we need to acquire. Freedom of self-expression and reference to arbitrary objects of our thought depend on the generativity that only comes with the phrase and the sentence, which have their own meanings. Yet, even in twentieth-century linguistics, this matter remained difficult to address. By 1930, linguistics as a field had become autonomous from both logic and psychology, having become first a comparative-historical discipline and then a descriptive one, insofar as the historical and comparative study of 10 Some qualifications are appropriate here: for example, as we shall argue in Chapter 3, Frege developed crucial insights on the formal structure of meaning as aligned with that of grammar. And Husserl, while engaging in what he called Cartesian Meditations and widely perceived as preoccupied with the intentional fine-structure of consciousness and its phenomenology, in fact regarded language as a constitutive condition of rational knowledge (Bruzina, 1981), like Merleau-Ponty (1945) would do later. Husserl also made a foundational distinction between ‘empirical grammars’ and ‘the grammatical’ itself, or the ‘ideal form of language’, which is the a priori and universal basis for logic, but is crucially conceived, unlike in Chomsky, in non-psychological, and indeed normative (but not social-conventional), terms. Like the Indians and the Modists, Husserl’s conception of universal grammar in this sense is crucially meaningbased. We would call nothing ‘language’ that does not consist of structural wholes that are units of meaning built from other such units, in accordance with their structure. The sentence is formed in accordance with a priori laws of signification and must necessarily have a sense, which moreover is necessarily one (the ‘unity of meaning’). This is thought of as an a priori law of meaning (apriorisches Bedeutungsgesetz), which ‘normatively determines and guarantees the possibility and the unity of a given independent meaning’ (Edie, 1977: 161), with no language failing to be formed on the basis of units of meaning. While being the most elaborated proposal for a universal grammar in the twentieth century, apart from Chomsky’s, Edie notes that it fell ‘upon deaf ears’. I greatly thank Alex Malt and Uli Reichard for discussions of this issue.

12


languages was eventually seen to depend on the prior availability of descriptive sets of data. As transpired in historical-comparative linguistics, even the parts of speech system of the generalized Indo-European language family were by no means universal, a relativistic perspective gained ground on which change and diversity were not accidental but essential aspects of language, which any empirical perspective had to respect.11 As Bloomfield summarized the situation in 1933: The only useful generalizations about language are inductive generalizations. Features which we think ought to be universal may be absent from the very next language that becomes accessible. Some features, such as, for instance, the distinction of verb-like and noun-like words as separate parts of speech, are common to many languages, but lacking in others. The fact that some features are, at any rate, widespread, is worthy of notice and calls for an explanation: when we have adequate data about many languages, we shall have to return to the problem of general grammar and to explain these similarities and divergences, but this study, when it comes, will be not speculative but inductive. (Bloomfield, 1933: 20)

Bloomfield stresses that this study of linguistic variation is autonomous from psychology, leading to a more ‘empirical’ approach to grammar: We have learned, at any rate, what one of our masters suspected 30 years ago, namely, that we can pursue the study of language without reference to any psychologistic doctrine, and that to do so safeguards our results and makes them more significant. (Bloomfield, 1933: vii)

The evidence, he argues, for the putative ‘mental processes’ that speakers were supposed to undergo was no more than the linguistic process itself. The postulation of the former adds nothing to the analysis of the latter. Two decades later, the cognitive revolution and Chomsky’s unashamed ‘mentalism’ arrives, wiping the behaviourist method defended by Bloomfield (1933) off the linguistic scene—though the externalism, functionalism, and empiricism it was founded upon would live on in philosophy in various guises, notably in the work of W. van Orman Quine, as discussed in Hinzen (2006). As Chomsky (1966) made clear, on the other hand, the ‘new’ perspective was at the same time an old one: a return to the Port Royal paradigm, though now enriched by the methods of computational science (such as the theory of recursive functions), which could tackle the generative basis of the creativity of language use, perhaps the most crucial axiom that Chomsky identified in the Cartesian framework (Descartes, 1637). Curiously, though, Cartesian rationalism is resuscitated without the project of aligning grammar and logic—at least initially, before the project of ‘logical form’ arrived on the scene (Hornstein, 1984). Rather, maintaining the Cartesian axiom of the distinctness of language and thought, Chomsky now focuses on grammar alone,

11 Cf. Evans and Levinson (2009: 446): ‘The diversity of language is, from a biological point of view, its most remarkable property.’


13

making it an autonomous ‘mental’ system consisting of arbitrary formal or algebraic rules, distinct in particular from those of semantics or logic. A decade later, Montague (1973) would do the same for semantics, turning it into an autonomous system as well, which did not depend on any particular theory of grammar (cf. Jacobson, 2012, for a similar view today). In the hands of Fodor (1975; 1983), the treatment of grammar as an autonomous system would become more than a temporary methodological decision of linguists, namely an architectural assumption, accompanied by notions of ‘modularity’, ‘encapsulation’, and ‘innateness’ (Pinker, 1994; 1997), wedded to a physicalist metaphysics of mental representations and mind. Linguistic functions such as phonological encoding or the ‘syntax-module’ now morph into domain-specific, ‘informationally encapsulated’ components of the mind, part of a genetically fixed modular architecture. Having each their own distinctive ontologies, they operate serially (with ‘syntax’ e.g. coming ‘before’ semantic interpretation, which is read off from it). Very explicitly, now, thought is independent of grammar—it is a ‘central’ rather than ‘peripheral’ system, domain-general and integrative in character—or else it is an independently generative component operating in ‘parallel’ with syntax and phonology, as in Jackendoff (2002). The idea of an inherent connection between grammar and thought remains lost and becomes hard to even formulate. Fodor’s Cartesian rationalism is crucially distinct from that of the Cartesian rationalists cited in Chomsky (1966), who hardly saw a difference between grammar, insofar as it is a science, and thought. Chomsky (1966) himself, however, citing the rationalist Beauzée saying that ‘the science of language differs hardly at all from that of thought’, particularly stresses the need for what we may call a restrictive and explanatory theory of semantics. As he complains, this very task was effectively trivialized in the way that the Cartesian rationalists carried out their project. While ‘deep’ structures that convey semantic content (thought) were proposed, no explicit and restrictive principles for their generation were formulated, and no independent evidence provided. Deep structures are posited that simply consist of actual sentences in a simpler or more natural organization. Illustrating with a well-known example from the Port Royal grammar (Chomsky, 1966: 80), for the ‘Surface Structure’ Invisible God created the visible world, say, the ‘Deep Structure’ might be ‘God WHO is invisible created the world WHICH is visible’, which combines three simple sentences: 1. God is invisible, 2. He created the world, 3. The world is visible. Chomsky characterizes this as an empirical hypothesis for which effectively no relevant evidence is provided. As he concludes, a ‘general’ theory of language that is not restrictive in regards to what is a possible Deep Structure AND what is a possible Surface Structure is empty. Any science of language—any universal grammar in the Cartesian sense—needs to respect this double constraint. This insight has not proved to be influential in the tradition to follow, however. In fact, the project of a restrictive and explanatory theory of semantics has barely begun

14


today (see Pietroski, 2011, for one formulation). And it is clear that while for Cartesian rationalists the connection between language and thought was, from the viewpoint of ‘scientific’ grammar, extremely close, Chomsky has been substantially more cautious in this respect, and in essence has maintained the independence of thought from language. From the viewpoint of syntactic autonomy and modularity, the aforementioned project in its original Cartesian shape must seem positively misguided. The paradigm of Principles and Parameters (P&P), too, is barely relevant in relation to this aim. Rather than focusing on the question of the (universal) structure of thought, P&P is focused on the question of linguistic variation again, much as nineteenth-century linguistics was: none of the ‘principles’, let alone the parameters, are conceived as having any intrinsic rationale in the organizational principles of universal human thought. We return to ‘syntactic parameters’ in Chapter 5. As late as 1995, by a time when the ‘interfaces’ had become highly relevant to the design of grammar, Chomsky (1995) describes the derivational process as largely one of eliminating (‘checking’) ‘uninterpretable’ (hence purely ‘formal’) features in syntax—a process that continues until representations are generated that only contain information that external semantic and sensory-motor systems can ‘read’. Hence these systems are still constituted independently and ‘wait’ for syntax to deliver something substantively interpretable to them. As noted, the very metaphor of the ‘interface’ of grammar with systems of thought (the ‘Conceptual-Intentional’ or C-I systems of Chomsky, 1995) implies a separation. In short, while the generative enterprise in the twentieth century is substantially less ambitious than the original Cartesian one in regard to the theory of grammar, it adheres to some crucial elements of its conceptual scheme, such as the conception of language as a medium for expressing or signifying thought, viewed as the locus of semantics and logic. Unsurprisingly, therefore, it also inherits some of the problems of the rationalist picture: thus, prima facie it remains hard to imagine how or why an expressive tool should be universal or biologically ‘hardwired’, rather than a mere convention. Variation is still what we would expect, when the rationalization of universality through an intrinsic grammar-thought connection is not available. It is fair, then, to describe the present project as ‘Un-Cartesian’ in spirit.12 It makes no assumptions on a pre-existing system of thought with which grammar interfaces (with qualifications on a system of ‘concepts’ partially lexicalized in a mental lexicon, as we will discuss in Chapter 2). It centrally consists in working out a principled notion of grammatical meaning, reflecting the fact that the grammaticalization of concepts in linguistic constructions itself has a semantics to it—a semantics that, we

12 There is a historically interesting question, though, how ‘Cartesian’ Descartes in fact was. Cottingham (1998) in particular, argues that thought depended on language for Descartes.


15

contend, cannot be replicated in the absence of the relevant principles of grammatical structuring. Grammatical meaning, we will argue, and nothing else, is what gives grammar a universal rationale: it is why grammar is there, and explains how it is built. It can thus also be used to divert criticisms of universal grammar that are based on the rejection of the claim that there is a ‘language-module’ in the sense of an autonomous, genetically based and linguistically specific endowment for language in humans, consisting in arbitrary formal rules (Pinker, 1994). That very rejection unites such writers as Evans and Levinson (2009), Christiansen and Chater (2008), and Tomasello (2008). We react to their criticisms on these grounds in Chapter 7. As unexpected as our claim may sound in a contemporary context, however, it was a central and prominent idea in two universal grammar traditions that predated Port Royal. We turn to the Indian tradition first.

1.3 Ancient India Systematic inquiries into the nature of language in the West do not begin before Plato’s Cratylos, and are confined to some speculations about words or names (onomata) even there. The paradigm of scientific theorizing in the early Greek tradition is mathematics, and geometry in particular. Scientific theorizing about grammar in the shape of a distinct discipline separate from the study of the universal forms of thought or metaphysics, is largely missing in the foundations of Western science. Aristotle in De interpretatione I–IV didn’t fail to leave his stamp in regards to an identification of the parts of speech making up a proposition (see Moro, 1997: appendix), but his treatment is still embedded within larger concerns in semantics and metaphysics, with the latter as the most foundational discipline. His medieval interpreters, in any case, took him to be a logician not a grammarian.13 Basically, a scientific treatment of grammar has to wait until the arrival of Modistic grammar in the fourteenth century, which would set out by discarding much of the Latin (and Greek) grammar of the time (in particular Priscian’s, of the sixth century ad) and attempting to get both the connection between, and the distinctness of, logic and grammar right, an insight documented in Abelard’s Dialectica (twelfth century) (Covington, 2009: 11; and see Section 1.4 below). Yet, Modistic grammar blossomed for barely half a century, when Occam’s nominalism wiped it off the philosophical scene. In the subsequent early modern scientific revolution, grammar does not figure on the agenda of science. Galilean physics is based on the mathematization of nature 13 Later commentary suggests that Aristotle in fact rejected the distinction between semantic and syntactic analysis (Moravcsik, 1968). Aristotle aimed to find the basic elements of sentences of subjectpredicate form as these would be the entities that also fall into only one category of metaphysical existence. This for Aristotle was where the link between language and metaphysics existed. We thank James Miller for this reference.

16


again, and in the scheme of Descartes’s version of mechanistic explanation, language cannot in principle figure: for it cannot be conceived without thought, which it expresses, and to which mechanical principles of explanation do not apply. In this way language is excluded from the Western scientific paradigm. Curiously, the opposite is true of the Indian scientific tradition, where from the inception of human theoretical speculation the very paradigm of scientific theorizing is grammar, the most important so-called ‘limb’ in the curriculum for the study of the Vedas, where it figures along with phonetics, etymology, metrics, astronomy, and the science (or art) of ritual. Of these limbs grammar was the ‘prime mover’ (Matilal, 1990: 7). The study of language is formally identified as Vyākarana, and linguistics as a discipline is joined by the philosophy of language from early on, unlike in the nineteenth or twentieth century, where these two disciplines have proceeded on largely separate paths and with disparate goals, virtually indicating different ‘paradigms’ in thinking about language, as Chomsky (2000a) and Antony and Hornstein (2003) systematically document. The philosophy of language is pursued by sages who write on grammar (the grammar of Sanskrit) and call themselves ‘grammarians’, recognized as having distinct concerns from the ‘logicians’ (Matilal, 1990: 42). We will re-encounter this same important dichotomy in the case of the Modistic grammarians in the next section. At a time when Plato was still reflecting on whether names are conventional or natural devices, Pāṇini completed a monumental oeuvre, the As·t.ādhyāyī, which consists in a system of about 4,000 interacting rules of four different types that together aim to give a maximally economical and complete, as well as theoretically consistent, description of the grammar of Sanskrit, a language initially based on the sacred and ancient texts (collections of hymns) guarded by the Brahmin religion. As Bloomfield (1933: 11) would famously put it, ‘No other language, to this day, has been so perfectly described’, in a way that is ‘complete and accurate’, and based ‘not upon theory but upon observation’. In the context of colonial Europe, this grammar would lay the foundations of historical-comparative linguistics and the systematic study of language change in the Indo-European language family: this was after William Jones, in 1786, announced the hypothesis, based on the method of the comparisons of word (sound) forms, that Greek, Latin, Germanic, and Sanskrit must have had a common origin in some ancient language of ‘Proto-Indo-European’. Pāṇini’s methods would teach nineteenth-century historical linguists how to do morphological analysis in the descriptive characterization of their own languages, effectively contributing to the emergence and foundations of modern structuralist linguistics in the early twentieth century, as it eventually emerged out of the Western encounter with Sanskrit. The indebtedness of both Saussure and Bloomfield to Pāṇini is acknowledged and clear (Singh et al., 1995): it was Pāṇini, after all, who had first delivered a system for structurally analysing each and every word-form of a language, leaving nothing to the informal understanding of the language user, thus


17

giving a proof of concept for a full and scientific account of a language in the structuralist sense. Rather than dealing with a particular corpus of Vedic texts, Pāṇini’s grammar, at a basic level, reflects the insight that Sanskrit exhibits a structural complexity that is subject to formal rules and regularities, which, together with relevant exceptions to their application and a list of primitives (roots) to which they apply, capture an infinity of cases, thereby doing justice to the unboundedness of language. Sentences, the Indians recognized, are ‘innumerable’, requiring some kind of formal structure (Cardona, 2009: 148, 154). This structure is analysed at a number of different ‘levels’— semantics/meaning, morphosyntax, morphology, and phonology, with asymmetric dependencies between these (so that, e.g., semantics can influence morphology, but not the other way around) (Kiparsky, 2002). It is noteworthy in this regard that, although Pāṇinian grammar is no doubt ‘formal’, grammatical description starts from meanings to be signified, serving as conditions for the introduction of affixes after bases, making semantics ‘intimately linked with grammar’ (Cardona, 2009). In this tradition, the epistemological and philosophical significance of the study of grammar runs deep. Pāṇini’s grammar was widely recognized, particularly by Patanjali (2nd century ad) and Bhartrihari (5th century), not to belong to any particular Vedic school but to be common to them all (Cardona, 2009). One might thus characterize it as a general analysis of the structure of Vedic knowledge (the Vedas, literally, are ‘the body of knowledge’) and an organizing principle behind the rituals that were associated with it. In the perception of the time, the Vedas, by laying an ancient foundation for language, lay a foundation for the fundamental distinctions and classifications of the world, and for the rituals that sustain and ground the social and natural order (Halbfass, 1991). Moreover, according to the ‘Advaitic’ position defended by Bhartrihari as well as the Manu-smrti and other dharma texts, the Ultimate Reality itself, or Brahman, is of the nature of word (sabdabrahman, where sabda means ‘word’) (Aklujkar, 2009). As Tripath (2009) puts it, ‘the Ultimate Reality is the Word-principle’, not unlike perhaps in the Christian Genesis. Accordingly, it makes sense that whatever is created can be linguistically analysed, comprehended, and communicated—it has a linguistic nature—and all science must adhere to the principles of grammar. It also makes sense that the study of grammar becomes part of spiritual practice, i.e. yoga, the attempt to realize a supreme state of being. Naturally, Brahman is One, and without beginning and end, as God is in the Christian tradition. But we can only perceive creation in its multiplicity and in space and time. The same is true for the word-principle: whereas meaning is something unitary, we can only study it by looking at verbal activity, which is temporal and sequential. Yet the study of grammar aims beyond this surface. Viewed as a spiritual practice, it has been termed sabdapurvayoga or the yoga of transcending the verbal multiplicity or plurality through the words themselves. All such verbal multiplicity occurs in time; hence, it is

18


an activity, temporal and sequential. Spiritual practice of grammar shows the path to submerge such a sequentiality into the plenitude of the Self, the Consciousness. (Tripath, 2009: 183)

The Indian tradition, then, gave language and hence grammar a role to play in the scientific study of the world and our place therein that was both principled and systematic. Ultimately, this appears to be the reason why linguistic discussions in the West came to fruition only so much later, while in India both grammar and the philosophy of language were cornerstones of science in a consistent fashion for more than a thousand years after the onset of human scientific speculation as such. These considerations also help us to understand why the study of grammar in the Indian tradition has a normative significance that it is taken to lack today. Of course, Pāṇini’s grammar is now widely regarded as ‘descriptive’ rather than ‘prescriptive’ in character (Matilal, 1990: 10), which is to say: it is not prescriptive in the sense in which Latin set a standard for grammaticality of European languages in the eyes of some eighteenth-century linguists, whose moulding of all languages on the Latin model would be ridiculed by nineteenth-century comparativists (recall Jespersen’s comments above). In Pāṇini’s case, however, understanding how complex words arise by strictly rule-based application of such operations as replacement, affixation, augmentation, and compounding seems to, at the same time, serve to better understand how they function, i.e. what they mean or signify when occurring in the ancient language in which the scriptures are written. Indeed, the process of inflecting a given root is being compared to a process of ‘cleansing’ the root in question, making it fit for the purposes of ritual—a ‘raw’ concept (root) cannot be so used or occur as such in a sentence (Matilal, 1990: 9). In this sense, word and sentence formation rules do contribute to the fixation of a norm for the formulation of knowledge and for the practice of ritual. As we shall see in Bhartrihari’s case, this conclusion is strengthened in a different and crucial way. To assert the same of modern generative grammar would be untrue, particularly under the understanding of syntactic autonomy, where formal rules are arbitrary manipulations of ‘symbols’ that are thought not to inherently relate to their use or semantics, which has its independent ontology and formal structure. As Chomsky famously puts it, the computational system of language that is governed by the rules in question could be just what it is now, yet be used for purposes of locomotion (Chomsky, 2000a: 27). It becomes linguistic in character only through its contingent embedding in already existing ‘performance’ systems, such as the C-I systems mentioned above, which have a non-linguistic semantics. In this sense, the connection between grammar and thought is something of an evolutionary accident. Relatedly, Chomsky implies (2000a: 106) that the ordinary term ‘language’ may not actually describe anything in nature, or a ‘natural kind’: a nominalist stance, on which the term ‘language’ is merely an informal label that is eliminated, as scientific inquiry proceeds, by technical abstractions (e.g. ‘I-language’), which need not cohere


19

with how we intuitively describe the world. Perhaps, indeed, no essence is hidden under such common-sense terms as ‘chemical’, ‘physical’, or ‘electrical’. Perhaps, the name ‘language’, too, may just be a name, referring loosely to a scattered range of phenomena that form no unity of which there could be a single science, as the Indian tradition conceived it. But then again, maybe not: maybe the term ‘language’ does depict a natural kind or distinctive principle. In the Indian tradition, it depicts the principle of reality itself. A non-nominalist stance of this sort invites probing deeper into the question of what grammar actually is—a question that might well have some substantive and principled answer in naturalistic inquiry. The connection between language and speciation that we tackle in Chapter 7 will support this perspective in its own ways. Today we assume that there is only one human ‘spirit’—the ‘psychic unity of mankind’—and only one living human species, originating in a Sub-Saharan population that found itself biologically endowed to put its sensory-motor apparatus to a fundamentally novel use: talking, manifesting a mode of thought apparently equally novel, as pre-sapiens hominin culture suggests nothing like it. As a consequence of this, an essentially modern culture erupts, rather abruptly (Tattersall, 2008), defining a novel way of being in the world after millions of years of relative evolutionary stagnation. The biological endowment furthermore must have stabilized genetically in this species by 40,000 years ago, when it reached Australia (Crow, 2002a). The species has remained a single one. Its language capacity is such that any child can learn any language; it unifies this species, drawing a boundary around it. Speciation, language, and thought are thus virtually a single evolutionary event, which asks for a unified principle behind it (Crow, 2008): a principle with the power of reformatting human thought and defining a new species. Species however are real as biological entities. So if language defines a species, it is also real, as a biological kind. It is worth asking, then, what grammar, as such, is or does—not allowing us any nominalist escape routes, and hence taking it as a fundamental and unified element of nature and cognition, which perhaps cannot be decomposed. In this regard, Bhartrihari gave an answer that is particularly interesting in the present context. Bhartrihari considers grammar as the essential guide to the reality of language: unlike pronunciation, it is the inner essence of language. But for Bhartrihari, language also lies at the heart of the universe. If so, the study of grammar is the foundation of all knowledge. But why exactly should this be? He tells us: Words are the sole guide to the truths about the behaviour of objects; and there is no understanding of the truth about words without grammar. (Vakyapadiya I.13)

The first sentence is correct in the perhaps obvious sense that to state or even contemplate a truth about anything, we need words to refer to objects and say something about them. Of course it is true that we may capture the behaviour of objects in the form of another notation that makes no use of words, say a differential equation. Yet, this is using a language too, that of mathematics, which moreover has a

20


formal grammar to it. That mathematical language depends on natural language, moreover, and the mathematician’s thinking about these equations and his regarding them as true will presumably take a propositional format (it is certainly logical). The second sentence in the quote is also correct in the equally obvious sense that words are inherently grammatical entities (they can combine by grammatical rules) and grammar alone tells us about their behaviour. Knowledge of reality is thus made possible by grammar. But why does it follow that language is therefore at the basis of the universe? This, again, is correct—though now not in an obvious way—in the sense that it is part of what it means to say that some proposition P is true (or that it is an item of knowledge), that the world is as P says. If (i) words are needed to state a truth, (ii) grammar is the essential guide to the behaviour of words, and (iii) we can formulate (at least some) truths in terms of such words, grammar is therefore the study of reality.14 One can, of course, redefine what truth means. One could say, perhaps, that we never really quite claim that something is true, but only that it is likely, or well confirmed. Or we could say that whatever truth we can claim, it will be true of the world as appearance, not as reality. But it is undeniable that ‘true’ is not what ‘likely to be confirmed’ means. ‘True’ means true, and it is arguably an analytic claim that P can only be true if, not only language, but also the world is a certain way: indeed, it has to be in the exact way that P says. One could also redefine the term ‘reality’, and say that it is not really reality we mean, but appearance. Yet, again, when we say that something is real, we are not saying that it appears to be real. ‘Real’ does not mean ‘appears/seems to be real’, and to master the grammar of such raising verbs is to master the metaphysical distinctions in question. If truth depends on grammar, therefore, grammar is a guide to the real, which is what Bhartrihari says: by learning about how words function in language, we learn about objects in the world. The argument for the normativity of grammar made above, and for the epistemological significance of grammar, is therefore strengthened. Interestingly, Bhartrihari adds to his argument that ‘all thing-classes depend upon word-classes’ (Vakyapadiya I.15), a claim that will play an important role in the second chapter. We interpret this as saying that in the way we think about, refer to, and know the world, objects fall under certain formal classes: they are ‘objects’, ‘events’, ‘propositions’, and so on, exhibiting what we call a formal ontology. No such ontology can be defined in purely perceptual terms. The question then is whether this formal ontology is independent of the form of language. On Bhartrihari’s view, it is only through words, and in the way these figure in sentences, that things are understood or become known. We know the world through seeing, tasting, touching, and hearing it, too; but this is knowledge in a non-propositional sense. In 14 We return to this argument at greater length in Chapter 9. The three premises of this argument are obviously crucial.


21

this sense, Bhartrihari’s conclusion makes sense that what we know as the world is the creation of language. This does not mean that when we talk about a tiger, it is not a tiger we are talking about, but instead a word. Rather, it means that, as an object of knowledge, the tiger is inherently referred to under a grammatical form. Like Pāṇini’s grammar, Bhartrihari’s is meaning-based. Much of his work revolves around reflections about what grammatical meaning is, for which Bhartrihari reserves the term sphota. Word and meaning, on this view, relate like cause and effect. Strictly, sentence (i.e. grammatical) meanings have no parts, and that sentences are uttered in a temporal sequence of words, he argues, shouldn’t confuse us into thinking that they do. Sentence meanings, in short, are indivisible unities, and the words are no more than ‘instructions’ for calling up a certain sphota. We shall come back to Bhartrihari’s rejection of the principle of compositionality, which is defended in the Indian tradition by the opposing empiricist Nyaya and Mimamsa schools (Pradhan, 2009: 287), in Chapter 3. The same non-compositional view is endorsed by Bhartrihari for the meaning of words: while a word has phonemes as parts, its meaning has no parts, leading to an ‘atomist’ position in contemporary terms (Fodor, 1998; Hinzen, 2012). Bhartrihari’s view of grammar must be centrally meaning-based, for language has a ‘potency to reveal the world’, which depends on its having meaning, which ‘lies in the inner structure of the sentence’ (Pradhan, 2009: 284). Without meaning, that is, language cannot be the source of knowledge it is on this view. In more contemporary terms, we might say, language has a ‘representational power’ or ‘intentionality’. Yet, characteristically, these twentieth-century terms from the philosophy of mind exactly leave out the grammatical nature of the meanings in question, much as language and speech are commonly regarded as arbitrary ways of expressing knowledge rather than constituting it. Again rightly in our opinion, Bhartrihari rejects an empiricist and psychologistic position on the nature of meaning, as was held in the Nyaya and Mimamsa schools. On this empiricist view, meaning depends on the occurring of certain ‘ideas’ or other psychological processes in our minds. Bhartrihari, like Frege, rejects such a view, while also rejecting Frege’s contention that a thought is objective and independent of the mind that grasps it, or of whether any mind grasps it. As noted earlier, Fregean ‘senses’ have no linguistic nature, much as planets don’t, even though, unlike planets, they belong to a ‘Third Realm’.15 Bhartrihari’s sphota is abstract too, but inherently linguistic in nature, and hence grammatical. As we will here interpret Bhartrihari’s view, it treats meaning under the viewpoint of triangulation (Davidson, 1984), though it articulates the latter in a novel (and in a grammarian’s) way, much as we will in Chapter 9: it views the dynamics of language

15

For some qualifications of this claim about Frege, see Section 3.2.

22


as part of a single scheme in which language, reality, and thought are intrinsically connected. Language is not conceived as a system of formal rules or as independent of its use, but as the ‘unity of three distinct elements, i.e., thought, language, and the object meant’ (Patnaik, 2009: 200). We will in this sense call Bhartrihari a ‘substantivist’ as opposed to a ‘formalist’. It is clear that the term ‘rationalism’ does not describe this view in an appropriate way, despite the fact that Bhartrihari anticipated many aspects of modern nativism (Patnaik, 2009: 202): unless a child has some innate ability, it cannot acquire language, he argues. Without such a ‘seed’, language couldn’t be acquired through effort or instruction. Language is an inherent constituent of our awareness, even prior to becoming gradually conscious of it in the course of development. The conception of language as triangulation makes it easier to grasp the position that Bhartrihari takes on the relation between language and thought, illustrated by a quote from Matilal’s seminal book (1990): Language is not the vehicle of meaning or the conveyor-belt of thought. Thought anchors language and language anchors thought. Sabdana, ‘languageing’, is thinking; and thought ‘vibrates’ through language. In this way of looking at things, there cannot be any essential difference between a linguistic unit and its meaning or the thought it conveys. Sphota refers to this non-differentiated language-principle. (Matilal, 1990: 85)

Wittgenstein will later formulate related views.16 Of course it didn’t escape Bhartrihari that sometimes it appears to us that we didn’t say what we thought; or that sometimes, we don’t know how to say what we think. But the idea is that, without language, we wouldn’t have a particular mode of thought in the first place. He specifically distinguishes three stages on the path to an utterance: in the first stage, he metaphorically suggests, the language and what it means is like the yolk of a peahen’s egg, whose eventual rich coloration lies dormant in an undifferentiated state. There is the same potency for forming meanings in every speaker. At the stage of this potency, language and thought are still identical. Then there is an intermediate stage of speech-planning, prior to an utterance, where the sentence/thought is about to be given an articulated shape in a sequential form. Finally, there is the actual speech event, the emission of a sound. Given the generativity of grammar, there is on this model an immediate explanation for the generativity and discrete infinity of thought—for which an explanation in, say, a Gricean framework is missing, as Patnaik (2009: 196, 200) points out. There is also an explanation for the fact that, as Davidson argues at length, the notion of a true belief depends on the notion of a

16 Cf. ‘When I think in language, there aren’t ‘meanings’ going through my mind in addition to the verbal expressions: the language is itself the vehicle of thought.’ (Wittgenstein, Philosophical Investigations, }329). ‘Like everything metaphysical, the harmony between thought and reality is to be found in the grammar of the language’ (Wittgenstein 1967: }55). We thank Pallavi Worah for these quotations.


23

true utterance (Davidson, 1984: essay 11). We return to these points, and to other advantages of Bhartrihari’s view, in Chapter 9. Summing up, what can we extract from this Indian tradition, regarding our purposes here? Despite the richness of more than a thousand years of Ancient Indian philosophy of language, apparently anticipating many of the main paradigms of twentith-century theorizing about meaning, as documented in Matilal (1990) or Chaturvedi (2009), it has barely come to the attention of the Western reader. But the Indian grammatical tradition not only clearly anticipates and positively informs formalist and structuralist thinking in the nineteenth and twentieth centuries, it also goes far beyond it in developing a more ‘substantivist’ viewpoint amenable to inform novel philosophical theorizing. The Indian speculative mind gave the organizational principles of language—grammar, in short—a metaphysical and epistemological significance that is distinctively different from the role that language has played in Western theorizing since the Greeks. It is also crucially different from the metaphysics and epistemology of Port Royal (or of Chomsky, 1966), where language reflects thought but is not a constitutive principle of reality, truth, or knowledge. With both structuralist and generativist theorizing in the domain of language in the twentieth century, the door to such metaphysical and epistemological theorizing was firmly shut—perhaps necessarily so, if the discipline was to be set on a new foundation. Yet, in this way explanatory problems in the foundations of semantics and knowledge disappeared from the agenda, at the price of depriving the study of grammar of a philosophical and normative significance it clearly once had, and perhaps of losing a clear sense of what its essence might be.

1.4 The Modistic viewpoint The major grammarian of antiquity, whose work formed the starting point for the Modistic Universal grammarians, was Priscian (6th century). His Institutiones grammaticae comprise eighteen books and lean heavily on the Greek grammarian Apollonios Dyscolos’ work On syntax (second century ad). A contemporary of Priscian was Boethius, reckoned to be the last great logician of antiquity, who translated works of Aristotle such as the Categories and De interpretatione. Aristotle was perceived more as a logician than as a grammarian as well, since to some extent he had dealt with semantics, which Priscian had ignored, with the consequence that semantics and grammar were divorced. A focal point in Priscian’s work is a system of ‘parts of speech’ (partes orationis), the number of which he identified, like Dionysios Thrax (ad 217–145) in his ‘art of grammar’ before him, as no less than eight. The parts of speech played (and play) a much lesser role in logic, where, as Aristotle’s De interpretatione (I–IV) teaches, the essential two elements of every proposition are the noun and the verb (playing the roles of subject and predicate, respectively), plus a connecting element, which

24


Aristotle identifies as grammatical Tense. In fact, as Covington (2009: 9) discusses, a systematic connection between logic and grammar had been lost by the time Priscian wrote. It was re-established gradually, however, from the eleventh century onwards, through the works of writers such as Anselm of Canterbury (1033–1109) and Peter Abelard (1079–1142), and then in the twelfth century in the Priscian commentaries by William of Conches (1125) and his pupil Petrus Helias, which set the scene for the Modistae. On the one hand, at this point, logic and grammar were re-united and recognized as dealing in some way with the same phenomena. On the other hand, it dawned on theorists that they were nonetheless doing so in quite different ways. It appears as if the deepest and most innovative insights were achieved in the course of a productive conflict between logical and grammatical points of view, with the parts of speech as one important dividing line between these. Thus, the author of the Dialectica Monacensis (early 13th century: see de Rijk, 1967) notes that the logician has no use for more than two parts of speech (subject and predicate) since he is ‘only concerned with relations between whole propositions’ (Covington, 2009: 11) and hence can take them for granted, not analysing their internal composition. The same reduction is noticeable in twentieth-century discussions in philosophical logic, where there is a similar austerity in descriptive vocabulary, as for example in the dualities between ‘singular’ vs ‘general’ propositions, or ‘referential’ vs ‘attribute’ expressions: none of these dualities map onto grammar in any precise way, where, as we will show in Chapter 4, reference is a notion that rather requires a scale. A related difference in the ‘lens’ with which grammar is looked at transpires in the thought that the grammatical difference between subject and predicate maps onto the difference between ‘reference’ and ‘generality’, or ‘objects’ and ‘properties’, understood as semantic notions. But clauses can be subjects (or at least topics), and predicates can lack generality (as in nominal Small Clauses: see Moro, 1997). The first flickering of a novel grammatical interest and insight revolved around the multiplicity of the parts of speech, their exact definitions, and their rationales. Grammarians wondered why they exist, which they thought was a question to be answered by finding out about their function. This is illustrated in Covington (2009: 9–10) with William of Conches (12th century), who complained about the obscurity of Priscian’s definitions of these parts, and in Petrus Helias’s query about what their functional explanations might be. We will raise and answer the same question in Chapter 2. The search for functional explanations illustrates a concern with semantics that grammarians certainly had, despite the fact that ‘semantics’ was primarily associated with logic, in this broadly Aristotelian tradition. However, as just illustrated with Aristotle’s slimmed-down version of the parts of the speech, the grammarian’s interest in meaning, insofar as it depends on its finer articulation within the sentence or the internal organization of speech, is crucially different from the logician’s. In particular, it was clearly recognized that although the


25

grammarian cares for meaning, he does not care for the relation between words and the real-world objects they are used to refer to. For the same reason, truth (in the sense of what facts really obtain in the world) is also not the grammarian’s concern. Identifying the exact sense, therefore, in which grammatical meaning is a crucial concern of the linguistic theorist, while at the same time truth and reference are not, is very important for understanding the novel universal grammar paradigm that evolves in the middle and late 13th century, as well as for re-conceptualizing universal grammar on the lines attempted here. The productive conflict between logicians and grammarians we see in late Medieval Europe is strikingly similar to the one we have found above in the Indian tradition, where the struggle for the right meta-theoretical framework for philosophy and metaphysics divided the Nyayaikas (logicians) from the Vaiyakaranas (grammarians, effectively Pāṇinians). The former are artha-pramanakah: ‘those who regard things and events as authority’: they were, as Matilal (1990: 42) puts it, ‘interested in the way the world is (or is supposed to be), not particularly in how people speak about it’. The grammarians on the other hand, are sabdapramanakah (sabda means ‘word’): they are ‘those who regard speech patterns as authority’. As the post-Pāṇinian linguist Patanjali put it, they ‘accept the authority of the speech. What speech “tells” us is what we depend on (for deciding issues).’ From this point of view, to understand the way in which we know the world it is not irrelevant how we speak about it. On the contrary, it is a source of philosophical insight. For example, why can we say things like sthali pacati (‘the cauldron cooks’), when it is clear semantically that a cauldron is not a cook but that wherein he cooks? Is it just a confusion of grammar that ‘John cooks’ and ‘The cauldron cooks’ look the same, indicating a mismatch between grammar and the logical structure of thought or the ontological structure of reality? The confusion disappears if we realize that being a ‘subject’ engaged in an action is not the same as being an ‘agent’. Hence the cauldron, even in subject position, is not grammatically depicted as the thematic Agent—and nor as the undergoer or Patient (as which it would be the thing cooked). As the grammatical subject, it is rather involved (and in some sense actively) in the action of cooking the food inside it, which is what the speaker is here saying.17 So the grammar gets this right. The identity of a cauldron as that of a physical object defined by, say, shape, size, and weight, is irrelevant to the grammarian—and it does not reflect the way in which we have natural knowledge of the world when we speak about things that cook. Looked at in these terms, Chomsky’s (2000a) attacks on the philosophical notion of reference as applied to natural language is simply an instance of a grammarian’s insight that had dawned in human history twice before. Chomsky notes, for example, 17 Similar remarks apply to British weathermen speaking of ‘many places struggling to see temperatures rising above freezing’, or ‘temperatures struggling to get above freezing’.

26


that a ‘person’ such as Sylvester the baby donkey may well be turned into a rock by a witch, while the little girl who owns him tries to convince the world that it—the rock—is her baby donkey: physical criteria of individuation are irrelevant for how we refer to the world when we use language (Chomsky, 2009). In a similar vein, Chomsky (2000a: 37) notes that a city such as London can be destroyed and rebuilt 200 miles away, with people thinking of it as ‘the same city’, referring to it with the same name. So the way in which we refer to a city is not based on an individuation of such objects under a physical perspective. At the very least, a city is at a place, rather than a place. However, twentieth-century philosophy and semantics has sided with logicians not the grammarians, and for the (mathematical) logician it is irrelevant how the world appears in the format of speech. Frege in particular was interested in the immutable laws of truth, ascribed to propositions in a Third Realm. Chomsky’s views on reference as inapplicable to natural language (see e.g. Ludlow, 2011; Antony and Hornstein, 2003) are the last instance of a continuing struggle to tell logic and grammar apart. The same problem mars work on ‘thematic roles’ (Agent, Theme, Goal, Instrument, etc.), which are in part semantically individuated, without leaving any grammatical reflex.18 If the list gets too long, their identification depends on mixing grammatical and logical or conceptual criteria for them, which is to say: grammatical and logical semantics. It does not follow from the above, on the other hand, that truth and reference have no role to play in the theory of grammar. Indeed, as stressed earlier, both Indian and Modistic grammarians had a systematic concern with meaning, unlike structural or generative linguistics. This illustrates our claim that a clear-cut and principled notion of grammatical meaning remains to be developed, giving a rationale for grammar itself. Crucial to the Modistic attempt to provide such a rationale was a novel theory of the parts of speech. Lying at the heart of Modistic grammar, they are probably one of the least developed aspects of generative linguistic theory, where they play essentially no role in syntax at all and there is little theory of them, as Baker (2003) notes. X-bar theory, Minimalist Merge, and phase theory all abstract from (or generalize over) their distinctions, or postulate them as part of a universal set of innate (perhaps internally decomposed) ‘features’. Generative typology is criticized for taking such features as universally available and instantiated, a priori primitives (Haspelmath, 2010). For the Modists, by contrast, grammar begins with the emergence of a system of parts of speech, and without a theory of the latter, there is no theory of grammar. Such a theory cannot come from Merge/recursion in the Minimalist sense, which is generic and found in all sorts of discretely infinite domains in nature. Nor can it come from the lexicon, if the latter is pregrammatical (see also Boeckx, 2010, and Chapter 2). 18 Hale and Keyser (2002) is an attempt to constrain this number by identifying thematic roles with a narrow range of grammatical configurations.


27

Generative grammar—rightly from a grammarian’s point of view—has distanced itself from (logical) semantics: it is not concerned with what is true or false, like whether water is H2O, or whether peanuts are inanimate. Whether the phonological word sounding ‘water’ is used to refer to H2O depends on historically contingent facts like whether or not we have been abducted overnight and wake up on a planet called Twin Earth where what looks to us like water happens to be XYZ (Putnam, 1973). As for peanuts, it is in the power of grammar to make them animate, if we tell a narrative in which two little peanuts fell in love (Nieuwland and Van Berkum, 2006). But the generative tradition has remained sufficiently attached to the formal viewpoint of logic to relegate semantics to an autonomous ‘semantic component’, and to not perceive parts of speech distinctions as central to the organization of grammar. It rather tried to reduce the last two remaining Aristotelian parts of speech, subject and predicate, to formal or ‘configurational’ notions (phrase structure): not only could logicians thus do without (much of ) grammar, but grammatical theory, or so it appeared, could, too.19 While reducing ‘subject’ and ‘predicate’ to grammatical notions would at least make them grammatical, parts of speech in the sense of nouns and verbs are essentially treated as formal features entering the syntactic process as primitives, hence as lexical. Recent developments in Minimalist syntax have continued the traditional trend. Pietroski (2011), in particular, when identifying the ‘minimal semantic instructions’ that the grammar gives to the semantic component, makes no reference to any parts of speech, including even the Aristotelian template of subject and predicate. As he shows, to obtain a logic or semantics one does not need subjects and (sentential) predicates. All one needs is (monadic) predicates and a way of conjoining them, plus some relevant adjustments for thematic roles, and a closure operation (existential quantification as in the Neo-Davidsonian framework). The implication is that, if these are what the grammar minimally ‘tells’ the semantics, semantics does not need to know about parts of speech: they are semantically inert. Ipso facto, semantics will not know about grammatical meaning, which is reflected in part of speech distinctions, and logic and semantics will not illuminate semantic distinctions, if any, that depend on grammar. It follows that, either, grammatical meaning does not exist—as nominalist logicians in the fourteenth century concluded, bringing down the Modistic enterprise—or it does, in which case semantics as so construed will not touch upon what is linguistically specific and requires grammar (and Bhartrihari would have been wrong when he claimed that word and meaning relate as cause and effect). What, then, did the Modists have to say about the parts of speech that has escaped generative theorizing in the twentieth century? Their innovation is embedded in an Moro (1997) argues against this trend and proposes that the notions of ‘subject’ and ‘predicate’ are not only independent of the notions of ‘noun’ and ‘verb’, thus providing no analysis for them, but are grammatical primitives crucial to syntactic analysis. 19

28


explicit reflection on what it takes for grammar to be a science (scientia) rather than the art (ars) that it had been in the Western world until the twelfth century. As an art, grammar is taught as a means to speak better Latin. As a science, by contrast, it aims at truth, and indeed it has to uncover necessary truths, which obtain universally in a particular domain of inquiry. Yet universality as such is not the essential issue here, insofar as things can be universal without having any force of necessity. Enumerating linguistic universals empirically is thus no contribution to grammar as science in the Modistic sense. They need to follow from first principles, which should apply to all languages. Since languages vary, a need arises to distinguish dimensions wherein variation is inessential from a core where languages are the same. Such a core arguably must exist: the linguistic ability and grammatical organization as such are species-universal. So we need to determine what these consist in. There are not two human species, a linguistic and a non-linguistic one; or one whose grammatical ability linguistic theory captures, while there is another that requires a completely different theory. So whatever grammatical theory we wish to develop, it needs to capture a biological property that defines a species and sets its boundary (Crow, 2002a). The search for necessary truths, too, appears barely controversial in contemporary terms: in a science such as physics we don’t want it to be an accident how objects move (as long as idealized circumstances obtain). The Modistic viewpoint therefore simply means that the internal organization of language is not an accident or an arbitrary convention either. Using the Aristotelian distinction between the practical and theoretical (= ‘speculative’) sciences, grammar thought of in these terms is thus scientia speculativa, which does not aim at action or at changing how people behave, but at truth. On the other hand, precisely the parts of speech do not seem to be universal, and it is even unclear what they are, and how many they are. How then can we turn grammar into a scientific domain? The proposal of Chomsky has been to study language as a ‘natural object’, ‘a real object of the natural world‘ (Chomsky, 1995: 11), identified as a purely formal structure, arbitrarily interpreted, which is then embedded in a number of performance systems, hence viewed much as the formal symbols of a logical language. This is not only what naturalism in the study of grammar has come to mean, but is also in concord with what twentieth-century philosophical naturalists would accept a naturalistic account of grammar to be: for philosophical naturalism precisely struggles with the integration of meaning into the natural world, regarding intensions in particular as ‘creatures of darkness’ (Quine, 1956). No engagement with such creatures is required if grammar consists in the manipulation, by formal rules, of physically individuated symbols encoding various ‘features’ of language. This object of study, as is appropriate for a natural object, is then equally deprived of normative aspects. In line with this, Davidson (2004: 131–3) argues that an account of language cannot be fully naturalistic because it is inherently normative; but this does


29

not affect Chomskyan generative grammar as a naturalistic enterprise, he argues, for that enterprise simply doesn’t bear upon meaning—and it can be naturalistic, as claimed by Chomsky, precisely for this reason. Naturalism in this sense is exactly not the avenue on which the Modists sought to reveal the essential nature of language and turn it into an object of science. On the contrary, the essential move consisted in the claim that every word, in addition to pronunciation and meaning, has a set of properties called modi significandi (modes of signifying), with (speculative) grammar defined as the study of these. A science of language, the idea was, must start from a principled identification and classification of these modi, which can be combined in a way that is again not due to semantic reasons and distinctive of this particular system. For example, Socrates and philosopher combined does not merely yield Socrates, philosopher (except in apposition). Rather, it can yield something where a specific relation between the two is grammatically specified, yielding a new object with a completely different mode of signifying, as in Socrates is a philosopher (Covington, 2009: 26, 32). The modi in question must also define the parts of this peculiar system, which the Modists took to be the parts of speech that much of grammatical theory since antiquity had been about. But now, a principled explanation for these had to be found—one couldn’t just enumerate them, as Priscian had done. They should form an ordo naturalis, a natural order governed by some principle. Now, clearly, words have two kinds of properties, sound (vox) and (lexical) meaning. But sound is arbitrary and hence cannot form the basis of a science of language. Neither can the word-forms and their derivations (morphology), which differ radically across languages, as was clear to the Modists from even a cursory look at Greek and Latin. So there must be some other properties that words have and that form the basis of a ‘universal’ grammar. Since, on the Modistic view, the basis of language is signification, it makes sense to look for universals right there. This is where the modes of signification, as special new properties postulated for words, come in. They define the parts of speech, on this view, and nothing else can. In particular, the parts of speech cannot be defined in terms of either lexical meaning or semantic reference (what is signified, or the significatum). For example, currere (Vrun) and cursus (Nrun) have the same meaning in this sense—they might be used to refer to the exact same environmental event—but they clearly differ in how they signify. Hence they are different parts of speech, and meaning (in a non-grammatical sense), therefore does not account for what parts the system has. The modes of signifying, then, reflect what we may call the first stage in the grammaticalization of meaning, or grammatical semantics. In response to the obvious objection that the parts of speech do not seem to be the same across languages, the Modists replied that some modes of signifying define a part of speech, and make it what it is. Others are ‘accidental’, and some of these are in turn such that the parts of speech only have them by forming a dependency with others (i.e. relationally: they

30


are modi respectivi). Moreover, accidental modes can be missing, like Tense on the verb, which therefore is crucially not part of the ‘essential’ modes defining the notion of ‘verb’). But words also combine and enter grammatical constructions. A principle for ‘construction’ (principium constructionis) therefore has to be found, too. Vox, again, clearly does not provide it, and neither does non-grammatical meaning: indeed, we can again see that currere (Vrun) and cursus (Nrun), although having the same meaning (in the sense of significatum), enter constructions in very different ways. The modes of signification, therefore, must also be that in virtue of which words can enter constructions. This is the second stage of grammaticalization: constructio. They are the rationes consignificandi—the rationale behind words having meaning in context, i.e. in relation to other words. There is then also a well-formedness condition for any construction, which is congruitas (congruence, concord). It applies to every construction and is stated in terms of the modes of signifying of the words joined, depending on their compatibility. Forming all constructions involves resolving grammatical relations of dependency (between a dependens and a terminans). The sentence is the level where all dependencies are resolved, and perfectio (completeness) is achieved. The Modistic theory of the sentence regards it as both partially independent of the theory of constructions and of the theory of the parts of speech: a sentence, in an obvious sense, is not a part of speech (it isn’t a part of anything grammatical)—except when it is embedded, in which case it becomes a part of speech and does require a mode of signifying. As we shall see in Chapter 3, when discussing embedding, and in Chapter 4 when discussing the forms of reference, a clause indeed changes its mode of signification when embedded: it never refers as a matrix clause can, exhibiting always a degree of intensionality and hence not being referentially ‘complete’. Importantly, compatibility of modes of signifying in the process of construction can exist even if there is no compatibility of meaning (significatum), as in ‘cappa categorica’ (‘categorical cape’), which is congruous but not meaningful in a nongrammatical sense, showing that non-grammatical meaning does not interfere with constructio. As Thomas of Erfurt explains: Grammaticality and ungrammaticality are caused by the compatibility or incompatibility of modes of signifying, which are intrinsically the concern of the grammarian. But the semantic well-formedness or deviance of the utterance is caused by the compatibility or incompatibility of the meanings of the words. Hence ‘black cape’ is grammatical and semantically well-formed, and ‘categorical cape’ is semantically deviant; but both are grammatical. [ . . . ] But the compatibility or incompatibility of the specific meanings of the words is not the concern of the grammarian, but rather the logician; hence it is not the cause of grammaticality or ungrammaticality. (Grammatica speculativa, p. 304, cited from Covington, 2009: 34)

Effectively, then, both stages of grammaticalization are independent of meaning or reference in a non-grammatical sense, and they are independent of the notion of


31

truth condition, which the Modists saw as belonging to the province of logic and not to be a principium constructionis. Thomas’s statement completely echoes Chomsky’s statement in Chomsky (1965) that grammar cannot be accounted for in independently given semantic terms. But this crucially did not mean, for the Modists, that grammar has no content of its own. Like the Indian grammarians, Modistic grammarians had a ‘substantivist’ take on grammar—not a formalist or minimalist one. It is not a feature (the ‘edge’-feature in Chomsky, 2008) that allows a word to enter a construction, but a mode of signifying, associated with a lexical item grammaticalized through its part of speech status. Modes of signifying cannot be said to be ‘features’ in the contemporary sense at all, any more than semantic reference or presupposition are such a ‘feature’. The edge-feature is a purely technical and generic device that could be used to formally describe any system that has a combinatorial property (chemistry, say), and it thus does not characterize grammar specifically. It was precisely the claim that words lack the modes of signifying which finally brought the Modistic enterprise down, in the first decades of the fourteenth century. As nominalist logicians would conclude at the time: there is thought, and there is language, and there is the world out there. Words are arbitrary representations of the elements of thought; but these are concepts, in a non-linguistic sense, not words, and these concepts stand in a natural, not a linguistic or conventional, relation to objects out there. No new ways of signifying come with words. Language does not speak to matters of meaning. In fact it is often misleading, and studying the syntax of natural language is in no way to study cognition. Grammar thus ceases to be a cognitive theory and loses its epistemological significance. One could well argue (cf. Leiss, 2009) that this early fourteenth-century nominalist conclusion, for the most part, has not been abandoned since. One strategy, then, is to turn grammar into a science by depriving it of ‘obscure’ notions such as meaning and effectively throw out from it the notions of reference and truth that are taken not to be naturalizable (Pietroski, 2005; Chomsky, 2000a). An appropriate ‘natural object’ of study is thus created, justified by an appeal to the need for idealizations, with a concession that the result may not really capture in any intimate way what we mean by ‘language’, giving it different labels such as ‘Ilanguage’ or ‘competence’ instead. The Modistic move is fundamentally different. Grammar becomes a science by including meaning as well as the normative significance it unavoidably brings with it, and reconstructing grammatical theory around the specific kind of meaning that grammar organizes. The result can now hardly be said to be a ‘natural object’ like others: it is—inherently—a grammatical one. This, however, given the Aristotelian conception of science involved, does not mean it is not a science. Nor is the object of study a purely formal one, and a natural object for this reason. Grammar, then, defines our species and is an aspect of our biology. Yet, at the same time, it is sui generis, not a ‘natural object’ among others, to be studied like

32


them. No natural object exhibits ‘modes of signification’, or so it would seem—which is precisely the reason why from within a Chomskyan point of view one may want to abstract from such obscurities when aiming to naturalize the theory of grammar. But this may deprive grammar of a rationale, which the Minimalist Program centrally seeks. From this point of view the enterprise of biolinguistics—at least under one standard way of construing it—is as problematic as is the search for a ‘biomusic’ or a ‘biomathematics’, if this means that music or mathematics should be studied in biological terms. They should not. In fact, the opposite may be true: (some aspects of ) biology should be studied in musical, grammatical or mathematical terms. Let us summarize, as we have attempted in the Indian case, what we can extract for the purposes of this book from this short episode in the history of linguistic thought. The articulation of the concept of modi significandi allowed for a crucial move: to attempt to give a principled account, lacking and unneeded in logic, of the parts of the system of grammar, thereby distinguishing the tasks of the grammarian and the logician. The move depends on distinguishing the modi essendi—the nature and properties of objects in the world that we refer to—from how we refer to them when we use grammar to do so, or the modi significandi. This gives rise to a distinctive kind of grammatical cognition, which cannot be accounted for in terms of (lexical) meaning or (semantic) reference, and thus points us to the essence of grammar as a domain of scientific inquiry distinct from logic or semantics. What grammar yields, is a system of signification that allows us to represent the world under the perspective of knowledge. The imposition of a parts of speech system on ungrammaticalized lexical concepts (the first step of grammaticalization), and the imposition of the principium constructionis on the parts of speech subject to the constraints of congruence (second step), finally gives rise to a mode of signification that is inherently propositional, and in which the grammatical process terminates. This is the rationale of grammar. This account is also very different from that of contemporary cognitive or functional linguistics, which shares the rationalism of the Port Royal tradition, insofar as language on this account is there to communicate thoughts, but a theory of thought is missing. The proposal is equally different from the Port Royal model itself and its (undeveloped) conception of a ‘Deep Structure’ as ‘reflecting’ an independent world of thought governed by the principles of logic. There is a notion of grammatical meaning in the Modistae, none in the Cartesians: the epistemological and philosophical significance of grammar thus returns. Even metaphysical significance does, if we recognize how the Modists, like the Indians, weaved their model of grammar together with a form of realism that is incompatible with a philosophy of grammar on which there is no intrinsic connection between grammar and knowledge, or grammar and ontology, or on which language is purely ‘internal’ or formal.


33

1.5 Conclusions The twentieth century has seen a revival of the concept of universal grammar after its widespread rejection in the nineteenth century, in the course of a gradual departure from the Port Royal model. This rationalist model aims to align grammar with logic, and hence necessarily falters in its universalist ambitions when this alignment, quite predictably, does not withstand the test of cross-linguistic variation. As a consequence, linguistics in the nineteenth century became comparative-historical, focusing on the study of variation and historical change, particularly in the case of phonology, endorsing an inductive and particularist spirit that seemed the only option if the study of grammar was to be kept empirical, eventually leading to structuralist linguistics in the early twentith century. This spirit retains its repercussions in typology today.20 By the mid-twentieth century, Chomsky (1957; 1965; 1966) had re-introduced deductive theorizing and a search for ‘first principles’ into linguistics, returning to the rationalism of Port Royal in a twentieth-century setting, where computational science and naturalism had changed the theoretical landscape. Moving the creative aspect of language use to centre-stage, the formulation of recursive rules that can formally generate an infinite output eventually became the heart of the grammatical enterprise. This was carried out against a background of architectural assumptions in which the language faculty is a ‘module’ of the brain that ‘interfaces’ with a logical system of thought, the true locus of semantics, described in the autonomous vocabulary of formal semantics and the theory of sets and types. In this way, the task of an explanatory theory of semantics or of thought was not systematically addressed or even formulated. In linguistic practice, much if not most work still focused on crosslinguistic variation, rather than the way in which grammar constrains thought.21 Pre-Port Royal, we see that the project of a science of language is set up in a very different way. It is not premised by the naturalistic context of a ‘biolinguistics’ or the naturalism of the computational theory of mind, in which mental activity is described as the manipulation of formal symbols implemented physically in the brain, connected on standard views to the world via a causal relation of ‘reference’ (Fodor, Haspelmath (2010: 2): ‘It was one of the major insights of structuralist linguistics of the 20th century (especially the first half ) that languages are best described in their own terms (e.g. Boas 1911), rather than in terms of a set of pre-established categories that are assumed to be universal, although in fact they are merely taken from an influential grammatical tradition (e.g. Latin Grammar, or English grammar, or generative grammar, or Basic Linguistic Theory). This alternative, non-aprioristic approach to categories can be called categorial particularism. In this approach, language-particular analyses can be carried out independently of comparative linguistics.’ 21 At the same time, the creative aspect of language use was not really addressed. The idea of a brain implementing recursive rules does not really speak to it. Thus a computer can have recursive output, yet not be creative (or even thoughtful), while a human being can be creative even within a finite (if large) domain (see Pullum and Scholz, 2010, for discussion). 20

34


1990). On the contrary, grammar is a science in its own terms—not in biological or computational ones. It is science because it organizes a domain of knowledge that the Indians first realized has a systematic structure to it that can be described formally and algebraically in terms of rules. At the same time, this is not a meaningless structure. On the contrary, it correlates with a specific mode of thought, which gives rise to a unique form of knowledge that is inherently linguistic in character. Indian linguistics and its technical methods swept into the West in colonial Europe, changing the landscape of linguistics and laying the foundation for the discipline itself as it is taught today. Yet this latter idea was not sent along, and it is only slowly rediscovered today (Chaturvedi, 2009; Ford et al., 1997). In a little-known episode six centuries earlier, grammarians in Paris and across Europe had come up with a related view: that grammar and logic are in fact crucially distinct, and that the former offers a pathway for understanding the genesis of a system of signification centred on the parts of speech that has a uniquely grammatical character. In the twentieth century, the parts of speech as a theoretical concept were barely addressed in generative grammar, while typologists would make their variation across languages the prime argument in their rejection of universal grammar. On both the Indian and the Modistic paths, grammar has a normative significance—a reason for why, as philosophers and linguists, we should actually be interested in the study of grammar as a domain. In correlating with a particular novel mode of thought, it speaks to our nature and rationality. It may not then be a ‘natural object’ studied as other such objects. But this does not prevent its naturalistic study. On the contrary, as we will discuss further in Chapter 7, there is, within our own species, systematic genetic variation in regards to our rationality and language faculty. Where we see human reason fragmented, as in autism or schizophrenia, and grammar fragmenting alongside, the study of such variation in naturalistic (in particular genetic) terms informs both the nature of grammar and of human reason. Kinds of language—linguistic phenotypes—are kind of minds: mental phenotypes.

2 Before there was grammar 2.1 Overview This chapter needs to take us to the beginnings of grammar, and so it needs to be about what came before grammar. So, before there was grammar, what was there? What was the world like? What kind of meaning existed? Of course, the world was perceived, and as we will argue in Section 2.1, there is a kind of semantics in perception alone, long before language arrived in evolution. Perception serves a cognitive function: it filters a mass of environmental stimuli in terms of unified and recurring percepts. Perception is of objects as falling under meaningful categories: we see sunlight beaming into the room or persons walking in the street, hear cars starting their engines or a piano playing, or feel a dog’s skin under our fingers. We do not perceive photons, acoustic waves, or molecules, which physically correspond to such objects of perception. Instead, stimuli that can be otherwise quite different are grouped together into more abstract classes equivalent under perception. Cars’ engines can sound quite differently, yet this need not make them sound like anything other than cars’ engines, qua auditory objects. A more trained ear may hear a Mercedes, or even father’s Mercedes, showing that the genesis of such perceptual objects is not driven by instinct alone, and can differ from individual to individual, and of course species to species. Even so, perception is still largely an involuntary and online process, occurring as and when the stimulus does. Yet, since it has a semantic content not definable in terms of independently describable or physical features of the stimulus, it seems right to say that it is not stimulus-determined, and that it comes with a semantics: a content that does not mirror objects out there as independently described. Eventually, in this pregrammatical world, percepts became lexicalized: they were given a phonological identity and became entities freely available in our minds, manipulable irrespective of any perceptual stimulus present. We call this process lexicalization. But these lexical items, as such, did not have a referential semantics yet. As such, they only had a lexical content, and they were not used to refer. Their lexical content, moreover, was still largely defined in terms of perceptual features: the function of the lexicon remains classification of an experienced world, not reference

36


to it. In order for intentional reference to arise, grammar had to come into play: it is what provides the rules for the manipulation of such entities. In this manipulation, the lexical contents of the items concerned play little role, and it has a sole purpose: to generate acts of reference on the basis of such lexical contents, where these acts can take a number of grammatically distinct forms (see Chapter 4). Intentionality (with a ‘t’) refers, as we will use the term, to our ability to refer to an object in the world deliberately and flexibly, using some meaningful category that identifies the referent. Apart from perceiving cars, persons, musical instruments, cities, and dogs, we clearly also refer to such things, conceived in these very terms (as cars, persons, etc.). When we do so, intensionality (with an ‘s’) arises systematically. This is because, if we refer to an object as a table (as in using the expression ‘this table’ to point to a particular table), the external physical and social environment in which such an act takes place would equally allow us to refer to this same object as a surface, a place to sit, a piece of furniture, and so on. Such meaningful categories, therefore, while all applying to objects in the external world, are not determined by them. Because objects of reference do not determine the category involved in such acts, two different categories can be involved on different occasions with regard to the same external object. In that case, the percept need not change alongside: while staring at a given table, I can think of it, first as a table, then as a surface, then as a place to sit, then as a Bauhaus design, then as a place to sit or to stand on, then as Granny’s heritage, etc. It is true that our attention may well direct itself to different aspects of our percepts as the categories change in terms of which we think of them. Still, it appears that there is also significant continuity: the percept does not keep track of the near infinitude of ways in which we might think of any given object, in addition to perceiving it. This is indeed an indication that it makes sense to say that, over and above perception, thought is involved as we come to refer to objects intentionally and intensionally. Where it is, my thought that this is a table is not the same thought as my thought that this is a Bauhaus design, even if it is the exact same external object that is on this occasion thought to be either a table or a Bauhaus design. If there is such an intensionality effect, this suggests that thought is involved: intensionality, as Davidson (1982) points out, is a criterion of thought. Another such indication is that any act of intentional reference can take place when the object of reference is not present in the speech context. I can, for example, refer to Socrates, who died a long time ago, or to ice-cream, when I am lacking it. While I can freely imagine ice-cream, when it is not there, this is typically not the same as perceiving one. But it is also different from intentionally referring to ice-cream. Percepts have a semantic content, then, but they are not manipulable objects in the way that words are, which can also be used to refer. I perceive something as a table, or as green. But this does not mean or entail that I refer to the table, or to the colour

2. Before there was grammar

37

green, or indeed think of either of these. Equally, as we refer, for example to ‘Plato’s teacher’, we need not perceive. Where words are used to refer, grammar is involved. The moment words do more than perceptually classifying environmental stimuli, and come to exhibit referential properties, they are already grammaticalized, acting as nouns, verbs, or adjectives. With words falling into such classes, their referential properties dissociate: nouns, the common intuition is, refer to objects, verbs to actions, adjectives to properties. Words functioning referentially, therefore, are already parts of speech. Distinctions among these, we will argue, are grammatical distinctions, part of grammatical semantics in our sense. The path to intentional reference that we will describe, thus leads from perception to grammar, via the parts of speech. Along this path, semantics changes multiple times. Hence semantics as such cannot predict or explain that any transition will happen: it is the other way around, that cognitive changes happened, with at least the final one induced by grammar, with consequences for the organization of meaning ensuing from this. Only at the end of this path is there anything in our experience that is an object of intentional reference, and a mode of thought that such acts manifest. Reference is therefore not a base notion of semantics: for it arises only with grammar. Nor is there any intensionality before there is intentionality, and vice versa. We begin by laying out in more detail what we mean by the prelinguistic semantics of perception, in Section 2.2, and argue further that once words come along in evolution or development, the meanings of these words are not plausibly prelinguistic ‘concepts’. We then discuss the semantics that came along with a process of lexicalization, in Section 2.3. Section 2.4 defends the correlation claim, which is that intentionality, intensionality, concepts, and grammar, arise only together. Section 2.5 develops a systematic account of the parts of speech as reflecting a grammatical semantics already. Section 2.6 lays out the connection with this account with that of the Modists. Section 2.7 concludes.

2.2 Where semantics begins Where does ‘semantics’ begin? Let us start with a broad sense of the term, where it covers the emergence of ‘meaning’ long before words exist, and independently of whether there is communication. In any creature that has a brain at all, the analysis of sensory stimuli involves a transduction of such stimuli into neural signals, which transmit the information contained in the receptors to the brain. The brain then extracts information from the sensory signals and at least some brains will form ‘representations’ of environmental variables in some particular domain. These can in turn be structured, with different mental ‘symbols’ representing different environmental variables, and formal computations can be defined over them. In the wellstudied case of bee navigation, for example, the representations can represent such

38


variables as food sources, the time of day, or the compass direction of the sun as a function of the time of day (Gallistel, 1998). A structured formal semantics, mediating a perception-action circuit, is thus in place long before language arrives on the evolutionary scene. Ipso facto, a semantics of this kind will not tell us what is special about language or the organization of meaning in it—our basic question here. Nonetheless, there has been a temptation to think that when words came along, their semantics was already there: this, for example, would be so if human words ‘expressed’ prelinguistic ‘concepts’, which are their meanings. At this stage, therefore, we need to be more precise about what we actually mean by a word. First, there is the notion of the word as a prosodic unit (the ‘phonological word’); then there is the notion of the semantic word or lexeme, which is the word understood as an abstract vocabulary item with a given meaning that can take different forms, such as the verb run, which can take the forms runs, ran, run, etc. Even more abstract is the notion of a lexical root, which involves a semantic core possibly shared across lexemes of different categories, e.g. the root √run as involved in the verb run and the homophonous noun run, which occurs in the expressions a run, many runs, running, Mary runs, etc. Finally, there is the grammatical word: the word as a morphosyntactic unit or as functioning in a sentence context. The second notion of ‘word’ appears to be virtually identical to what are called ‘concepts’ in the philosophical literature (e.g. Fodor, 1998), where indeed there is the same notational convention of referring to such entities with small capitals. Presumably, the words ‘runs’ and ‘ran’ express the same ‘concept’—or at least, the question of whether they do seems barely, if ever, discussed. Since the technical notion of a lexeme is more precise, it is not clear why one should invoke the notoriously unclear notion of ‘concept’ at all. That notion is often applied to non-linguistic beings (Carey, 2009), and if we do not wish to ascribe lexemes to the animals, then the term ‘concept’ has an advantage over ‘lexeme’. It appears, however, that we will lose important distinctions in this process. While meaningful categories in the sense above seem to be involved in both perception and in intentional reference, they are stimulus-controlled in perception, while being freely manipulable in intentional reference. When manipulated for purposes of reference, they are symbolic units that have a phonological identity. Hence lexicalization has taken place: the brain has created items that can be freely called up, partially independent of perception. This is unlike in the case of percepts, where there is no reference and no need for it, and no creative and systematic combinatorics.1 None of this takes the point away that perception in prelinguistic cognitive systems can involve classes of considerable abstraction, as when prelinguistic human infants individuate objects via sortal kinds (e.g. box, car, toy, cat) (Xu, 2005), a process that 1 The imagination is creative, but it is not combinatorial, it appears, in any systematic way, and it is not propositional either (see Reuland, 2010, for discussion).


39

is to some extent independent of the object’s perceptual properties. Objects that are all toys, for example, can be perceptually quite different. One of the most abstract of such sortals is object: whether something counts as an object in perception (or is identical to another) does not solely depend on its spatio-temporal properties. The same is true for sorts such as agent, cause, attention, or plurality, which again are, although not independent of perceptual features, only obscurely related to them and not apparently defined in such terms (Carey, 2009: chs. 5, 6).2 If so, we might call them ‘concepts’, also because they are ‘inferentially rich’ in the sense that how something is sortally analysed enters into the child’s decisions on how to act (e.g. whether to reach for an object). However, although such representations differ from ‘perceptual representations in their abstractness and their conceptual content’, Carey (2009) stresses that this is in part a ‘matter of degree’ (p. 449). Indeed, like the abstract concepts of core cognition, the outputs of perceptual input analysers (whether an object is red, round, and distant, for example) are accessible to so-called ‘central’ systems as well, where they enter into inferential processes and determine actions such as reaching. Moreover, like representations of depth, the representations of objects, agents, and number are the output of evolutionarily ancient, innate, modular input analyzers. Like the perceptual processes that compute depth, those that create representations of objects, agents, and number continue to function continuously throughout the life span. And like representations of depth, their format is most likely iconic. (Carey, 2009: 450)

An iconic representation is an analogue one whose parts correspond to the parts of the object represented: the head of a tiger in a photo, say, to its actual head in reality. No word or sentence has meaning in this analogue fashion. Related to that, no issue of intensionality arises with iconic representations, as Fodor (1998) argues: an iconic representation of a giraffe cannot help being a representation of my favourite animal, if the giraffe is my favourite animal. When we use words, however, intensionality effects arise pervasively and systematically: thinking about animals identified as giraffes is not thinking about my favourite animals, even if they are my favourite animal. Representations in the linguistic format are also not reflexively and causally triggered by a perceptual input, as concepts in Carey’s sense are; and they are not modular, being capable of describing any domain whatsoever (Spelke, 2003), even if perhaps imperfectly, as in the case of emotions; and neither percepts nor core concepts nor iconic representations are normatively constrained, in the sense that

2 As Carey (2009) stresses even for object kind sortals, although these are perceptually based, ‘it is unknown what these bases are’ (p. 273). That I perceive you as a person, rather than as a surface or a physical object, does not appear to be explained by the visual properties of your surfaces, or your physical and spatiotemporal properties—even though, equally clearly, there are correlations between the physical properties of an object and the way we perceive it.

40


there are correctness conditions that they can fail to meet: indeed, we do not control perception. Neither percepts nor concepts based on them require reasons, and the question of truth does not arise for them. For illustration, consider the Heider– Simmel (1944) demonstration that normal subjects endow a movement pattern of animated geometrical shapes as depicted in Figure 1 with rich social meaning, reporting percepts such as the large triangle ‘chasing’ the small triangle, shapes ‘looking for’ one another, ‘hiding’, ‘conferring’, being ‘furious’ or ‘frightened’.

FIGURE 1

Percepts of this nature, which arise without conscious control, systematically deteriorate or disappear when the movement pattern is slowed down or sped up. Subjects then report merely seeing shapes and geometrical paths again. This becomes the general situation with patients suffering from bilateral amygdala damage, who also describe the same animations in asocial and purely geometric terms, despite otherwise normal visual perception (Heberlein and Adolphs, 2004). No questions of truth or correctness arise in either the patients or the normal controls. Yet this perceptual semantics is rich, and explanatory questions arise that are completely unresolved even in this domain (never mind grammatical semantics). In line with this, Mausfeld (2011) stresses the ‘explanatory gap between the information available in the sensory input and the meaningful categories that characterize the output of the perceptual system’ (p. 159). As he illustrates, objects that we perceive are endowed with an abundance of material qualities. In the case of perceived surfaces, for example, these are captured in verbal descriptions such as ‘soft’, ‘wet’, ‘malleable’, ‘silky’, ‘juicy’, ‘edible’, or ‘deformable’. Many such perceptual attributes correspond, in metaphysical terms, to dispositional and causal properties of the objects perceived. Again, the explanatory connection between such attributes and the physical or visual properties of the stimulus that trigger their perception remains largely obscure. We do not know where the semantic contents of our percepts and core concepts come from. These problems deepen as we move up from perception to relational concepts such as cause, which David Hume famously noted have no basis in our sensory-motor ‘impressions’. There is similarly no perceptual basis for our notion of truth, or the relational notion of a ‘predicate’. We perceive things as objects external to ourselves, but this construal is reflex-like and requires no concept of what an object is, nor


41

thoughts involving such a concept. Thinking in this sense is, unlike perception, not an ‘online’ process: as we think, we don’t necessarily externally perceive anything or attend to what we perceive—most of which will typically be irrelevant to the contents of our thought. In turn, most things that are physically present and that we perceive, we don’t also name, when we see them—fortunately so, for there is so much that we see all the time, most of which is of no relevance at all to what we are thinking: for example, the font type of the words you are perceiving as you read this, or your mood, both of which you largely ignore as you try to come to grips with what you read. As you do, perceptual stimuli associated with reading, while processed, are thus switched to offline in conscious processing. The phenomenology of reading is crowded out by the phenomenology of thought. Much of the latter phenomenology is accompanied by silent soliloquy, with sentences or fragments of them flicking through our minds incessantly, to various degrees of (usually silent) articulation—a barely controllable predicament that can annoy and keep us awake at night, requiring extraordinary efforts to be terminated altogether, as in long-term meditational practices. It is as if each of these thoughts cries out for our attention, not leaving us in peace, showing that linguistic thought of this nature tends to be inherently related to consciousness. That relation is very clear in actual speech itself, where we virtually always know that we speak, when we do, and indeed normally know what we say. Most normal utterances appear to be spontaneous rather than carefully planned and executed, yet the moment we make these utterances, we are consciously aware of them and their content (though we may get distracted by something we see or hear while we speak, including the quality of our voice). Both in soliloquy and actual speech, moreover, it is linguistically meaningful units that we are consciously aware of, such as words, phrases, or whole utterances, but not, say, arbitrary stretches of phonation within an utterance lasting exactly 300 milliseconds, or three quarters of a word or phrase. The phenomenology in question thus obeys in part linguistic principles. Not only does semantics systematically change as we move from perception to grammar, therefore, but conscious experience does as well. Consciousness presumably begins with perception—percepts are conscious virtually by definition—but then, perception need not be accompanied by percepts that we actually consciously process (like the letters you perceive as you read, when focusing on the content of what you read). Consciousness thus changes fundamentally as language intrudes and takes over our conscious attention, switching non-linguistic perception to ‘offline’. This in turn changes our non-linguistic experience. If Helen Keller is any source, the sensory-motor experience of water is not the same when we can name it and when we cannot. With words, experience itself becomes a matter of reflection. Even if other species have a ‘mind’ or ‘think’ (in some senses of these words), only we may be aware that we do, which in turn influences what we think about: about thought itself, for example.

42


The source of this transformation in both semantics and consciousness plainly cannot be semantic: for semantics is what is being transformed. If ‘semantics’ means ‘relations to the external world’ (perhaps specified causally), it cannot at present much illuminate or explain the contents even of percepts. The same applies to the phenomenon of intentional reference, in either acts of pointing or the use of words. A young infant pointing declaratively points, say, to an elephant, viewed as an objective entity that the adult also sees and understands in the same way, as falling under the same ‘concept’. As words come along, the infant might mutter ‘elephant’ while pointing to the elephant, and in any act of reference that we make as adults, some meaningful descriptive category is involved in a similar way (for qualifications see below and Chapter 4). None of this follows from anything semantic. Thus, reference in non-linguistic animals takes a completely different form (and may well be explainable causally). Intentional reference, in the human sense, involves language, and it needs to be explained linguistically, not semantically. When, under the causal control of an external trigger, monkeys or chickens call for alarm, the call can be acoustically distinct depending on what predator is perceived to cause the threat (Hauser, 1996; Cheney and Seyfarth, 1990; 1998). But the great apes do not call for alarm as the monkeys do, which we expect if such reference is functional and apes simply are not under the relevant kind of threats. Alarm calls are not for conceptualization or reference to the world, but for ensuring that the right action is performed under the right external circumstances, expedited by a reflex-like, emotional reaction to a given trigger, with perhaps no underlying communicative intention at all. For that reason, the question of which ‘concept’ is being expressed by any of these calls appears to be mute. Is it eagle, or into the bushes, or danger from the air? No concept appears needed because, even without one, the triangle between the caller, the stimulus, and the action is causally and adaptively closed. The calls occur when the stimulus does, which is also when the percept does, which identifies the predator. There is no evidence that the calls in question are ever applied to anything other than the stimulus, in the very adaptive context in question; or that the percept involved in one such call can be targeted as such, and be combined with another percept, a process that we argue requires lexicalization. Every one of the calls is complete in itself, requiring no further specification of what is to be done by the recipient (see also Bickerton, 2009: 44–7, 68–9).3 Lexemes by contrast not only can be used referentially in the physical absence of their referent, but are also very incomplete in their meaning. The word ‘eagle’ by itself does not denote anything in particular: not this eagle or that, not all eagles or some, not a kind 3

We second Bickerton (2009: 69) in his claim that the units of alarm call systems:

‘don’t really translate into human language. We can give an approximate meaning, or several possible meanings, in terms of our language, but the idea that underlying both is the exact same semantic expression is simply baseless.’


43

of bird as opposed to another, not the property of being an eagle, etc.—things that it can denote only once it appears in the right grammatical configurations. It is also used for purposes of reference and predication, in addition to being used as a directive for action, and it again requires a phrasal context, hence grammar, when it is so used. As a lexical item, moreover, it automatically stands in relations of hyponymy (an eagle is an animal, as is a leopard), and hyperonomy (there are several kinds of eagles). The existence of such a system of hierarchical relations is not driven merely by instinct, in the way that animal categorizations of environmental stimuli develop, as the facts suggest that it is subject to significant variation across speakers and languages. The use of a lexeme is also never indexical in the sense in which animal alarm calls are. Not even words standardly classed as ‘indexicals’, such as pronouns or demonstratives, are indexical in the sense of the monkey calls. Thus the 1st-person pronoun need not denote the speaker: ‘I’ denotes the logophoric agent of the embedded clause in (1) below and the sentence subject in (2), hence in neither case the speaker. Moro (1997) makes a case that ‘io’ (‘I’) in (3) plays the role of the predicate (hence not of a referential expression). In (4), ‘he’ refers to whoever enters through this door, not a particular person. (5) can be said intelligibly by an academic speaking in Durham while currently employed as a guest professor in Barcelona: (1)

He said to me: ‘I love you.’

(2) Kon K@gna n@-ññ yl-all John hero be.PF-1sO 3M.say-AUX.3M ‘Johni says that hei is a hero.’ (Lit.: ‘Johni says that Ii am a hero.’) (3) Sono io. ‘It’s me’

(Sigurðsson, 2004) Amharic

(Schlenker, 2003: 68) Italian (Moro, 1997)

(4) He who enters through this door will be crowned. (5)

Yes, I am there now.

Even so-called ‘indexical expressions’, then, exhibit intensionality and allow for displaced, predicative and descriptive readings, apart from their more paradigmatic referential ones. Even in the referential readings, a concept is always involved. Thus if I refer to myself as ‘I’, I am plainly not conceptualizing myself in the same way as when I point to my body and say: (6)

I feel totally happy in this body now.

The referential behaviour of indexicals, moreover, as we shall see in more detail in Section 4.4, co-varies with their grammatical behaviour. What we find again, then, is a correlation between grammar, lexemes, reference, and intensionality: these occur together. Correlatively, intensionality is absent and reference is most ‘direct’, when no

44


content word is involved, as in interjections. In that case, emotional content is critically involved in the utterances, and there is very little grammar: (7)

a. Wow! b. Sh***! c. Hey!

Where there is no language or lexemes at all, any case for either ‘concepts’ or reference is very hard to make. As Davidson (2004) stresses, neither a ‘concept’ nor reference can sensibly be attributed simply because an animal makes an adaptively significant distinction. Sunflowers can direct themselves towards the sun without anything we would be tempted to call thought. To take a more complex case, cephalopod mollusks vary their appearance to suit the surroundings. To find the most adaptive camouflage from within their coloration repertoire, they must use spatial information. The common cuttlefish, Sepia officinalis, uses multiple cues such as spatial scale, contrast, and depth to select between the ‘Mottle’ body pattern (displayed on continuous patterned surfaces) and the ‘Disruptive’ one, used on discrete objects such as pebbles. The nature of the cues can be investigated experimentally. Juvenile cuttlefish classify visual textures according to whether they are uniform or patterned, and whether the pattern includes visual edges. Specifically, they detect edges by sensing the relative spatial phases of two spatial frequency components (Zylinski et al., 2009). There is a function, therefore, from visual features of the external environment, to a certain output (a camouflage), which is internally mediated in the animal by a classification of visual textures. The classification scheme is more abstract than the more detailed visual features that are so classified. Yet the terms ‘uniform’ and ‘patterned’ are clearly merely two labels from within the human conceptual repertoire that are descriptively useful to capture these classifications. That the former is the human concept uniform—or perhaps the meaning of the English phrase ‘uniform surface’ or ‘not a pattern including visual edges’ is a suggestion that the data do not support. Why should the cuttlefish have thoughts about surfaces and edges, when it suffices to simply detect and classify them correctly, and incorrect classification will lead to death rather than falsehood? Referring in our present sense is again not merely representing: this is particularly clear if we view reference on the model of pointing, as we do here. If I point to a man with my index finger, my finger does not represent the man. Neither does the word ‘the’, when I use a definite noun phrase. A pack of hunting wolves must have some shared internal representation of the prey, but they do not use it to refer to it. Infants can manage the distinction between singular and plural nouns between twenty-two and twenty-four months of age, and in parallel with this their capacity develops to distinguish non-linguistically between singular and plural sets, in a manual search paradigm. Li et al. (2009) found no evidence that, in languages such as Japanese or Mandarin where number marking on nouns is impoverished,


45

the non-linguistic ability in question is delayed. If so, there is a sense in which there is a mental representation with the content ‘singular’—as opposed to ‘plural’—set that does not depend on its morphological realization. Yet there is no reference in this case, and the suggestion that these very young infants think about sets with the concept ‘singular’, and that intensionality effects might arise, seems unwarranted. While these infants belong to the linguistic species, representations of distinctions relating to number are well known to have a basis in non-linguistic species. Analogue representations of magnitudes of number by no means depend on language and are not specific to humans, forming a ‘foundational’ abstraction of very ancient origin (Gallistel, 2009). This arguably includes (i) the ability to individuate in parallel the members of 1, 2, 3, and 4-membered sets, (ii) quantifying over such sets with such quantifiers as some, all, singular, dual, or plural, and (iii) representing the analogue magnitude of sets and the ability to compare these on the basis of approximate cardinal values. On the other hand, the representation of large (greater than three) exact numerosities, i.e. the list of integers, does not develop in children unless a language model is available in which such numerosities can be systematically referred to with numerals (Carey, 2009). It also fails to develop in the home sign systems developed by deaf children among one another when they are born to non-signing parents (Spaepen et al., 2011; for similar evidence of the absence of large exact representations of number in cultures whose languages lack a verbal count routine, see Gordon, 2004; Frank et al., 2008; Pica et al., 2004). Again, this illustrates both in what way there is cognitive continuity between humans and non-humans in terms of representations, and discontinuity in regards to reference, once language and lexemes are involved. This suggestion is consistent with what lies at the heart of the continuist position, namely the central assumption of the computational-representational theory of mind, that both the human and nonhuman brain forms ‘symbolic representations’ that are physically realized and subject to computation in the causation of behaviour. Words, on the other hand, are not in any clear sense causally efficacious in the generation of action. Using words is a deliberate action in itself, independent of perception, with no non-circularly identifiable causes (Chomsky, 1959), taking place for purposes of reference rather than action, at least if we disregard imperatives and other performatives. Functions defined over symbolic mental representations, furthermore, are constrained to be mirrored by functions over what they represent (the mapping between entities in the represented domain and the symbols is structure-preserving or ‘homomorphic’: cf. Gallistel and King, 2009: ch. 4). It is unclear how this constraint is even meaningful in the case of words and the way they are connected in grammar. Thus, when grammar categorizes a given lexical root as either a verb or a noun (to run vs a run), it does something that nothing in the external environment, as independently described, needs to reflect: as the Modists noted, it reflects a difference in perspective or reference, not lexical content. When we will look at grammatical

46


relations (see Chapter 3), it will be even less clear how such a homomorphism could be defined. These considerations underscore that the issue of representation is to be distinguished from the phenomenon of intentional reference as described here. The grammatical word appears to be no representation in the sense of the computational theory of mind at all, which thereby falls short of capturing what makes our species unique and our cognitive mode discontinuous with any other. Demoting the cognitive significance of language, Gallistel (2010), takes a stronger position, pitched against Carey’s (2009) ‘discontinuist’ view, which gives language a crucial role to play in the genesis of uniquely human concepts. Yet the evidence provided is again about representational power, not reference. Thus, for example: Both mice and adult human subjects represent the uncertainty in their estimates of elapsing durations (a probability distribution defined over a continuous variable) and discrete probability (the proportion between the number of trials of one kind and the number of trials of a different kind) and combine these two representations multiplicatively to estimate an optimal target time. (Balci et al., 2009)

Yet, how seriously should we take the idea that the mouse can think ‘quite sophisticated thoughts’ about how a given probability should be ‘combined with its representation of its uncertainty regarding the duration of an elapsed interval in order to choose the optimal decision point in a timing task’ (Gallistel, 2010)? If we follow the old Leibnizian idea that thought just is calculation, it follows definitionally that the mouse thinks. In a similar way, twentieth-century functionalist philosophy of mind defines mental states as computational, with computations performed over physically represented symbols in the brain that are causally and homomorphically related to their referents in the world. If we make this definitional move, the question of whether mouse thought is like human thought is begged. What is missing in the mouse’s thought is, we claim, not only concepts, but a formal ontology, by which we mean a system of formal distinctions by which objects of reference are classified as objects and events, propositions and facts, properties or states. In any human language, reference is to objects of a small number of specific formal types, of some of which we can predicate existence or truth: a creature that can form sentences has a concept of being. This formal ontology comes, not with symbolic representation as such, which is pervasive in animal cognition and found in animal communication as well, but grammar, which is unique to our species. It is first manifest in the emergence of a classification of lexical items into parts of speech (Section 2.5). Mice neither lexicalize their ‘concepts’ (= symbolic mental representations), nor do they turn them into parts of speech. Humans seem to be the only creature that, apart from performing computations over symbolic mental representations co-varying with environmental variables, also and additionally refers to these variables, in specific ways that depend on a grammar and a lexicon. This claim entails that what the generative grammar tradition, given its ‘internalist’ stance (Chomsky,


47

2000a; Hinzen, 2006; Pietroski, 2008), has characteristically excluded from the study of grammar is essential to it. But it is not externalism to incorporate notions such as reference and truth into grammar: not if grammar is the only system in which such phenomena arise, in the systematic ways described in Chapter 4, and not if no single grammatical expression is ever completely free of them, making them part of what it is to be grammatical.

2.3 Lexicalization: the creation of the lexical atom Until now, we have sought to distinguish percepts from concepts, which we take to presuppose lexicalization, and distinguished the intentional reference that involves lexemes from mental representation in a much broader sense. Let us now look at lexicalization and its semantic rationale in more detail. What is lexicalized is a selection of perceptual features from classes of percepts that together create a more abstract percept, reducing the sensory complexity of the world even further. Perceptual features themselves are already abstractions from an infinity of data that the sensory world presents us with, creating an initial order in it. The process of lexicalization de-couples the percepts that are selected from their respective visual stimuli, giving us new and more abstract entities, lexical items, which are stimulusfree and independently manipulable, enabling creative thought and reference. For this reason, we may describe this process as one of ‘de-indexicalization’. Any such process will require another process of re-indexicalization: freedom from experience is bought at the cost of having to re-establish a link with experience. It is what we are trying much of our waking lives to achieve (when we are not day-dreaming): saying something true rather than false, seeking evidence, exercising doubt. Where concepts are de-indexicalized, a mechanism is needed to relate them back to the world on occasions of activating them: a system converting concepts into referential expressions. In this reference-system, reference to the world will be a creative (intentional) act subject to conscious control—and of error as well. We identify this system with grammar. Lexemes are the basis of this reference system and anchor it in a world of shared conceptualization. They are the non-creative part in a system that is otherwise the hallmark of individual creativity: which lexical items exist in a language is something we have to learn. They are given to us, as a public way of organizing human experience. In a linguistic community into which a child is born, an inventory of lexical items offers a precategorization of the experienced world that the child can take over without having to create its own. Various experiences will now present themselves as instances of ‘the same’, with lexicalized concepts defining the criteria of (cross-modal) ‘sameness’. As lexemes are manipulated in the place of the perceptual features defining them, we come to live in a symbolic world of culture (a Popperian

48


‘Third World’), an abstract domain of sharable thought, in addition to the world of sense experience, which we only confront individually. In the generation of lexical items on the basis of selections among perceptual features there can be no ‘mistake’ yet: any such selection will still simply correspond to perceived reality. There is no need for a notion of reference. Variation in vocabulary across languages is therefore not harmful and quite expected: colour is perceived, and perception is causally controlled. The possibilities of mistake or disagreement arise later, as a reference system develops in the form of a grammar, and truth and falsehood along with it. Since neither reference nor truth are perceptual categories, and hence external control is lost, we would expect language to be more constraining in its grammatical aspects than in its lexicon: grammar provides the control now needed. Once a particular perceptual class has become a lexeme, the grammatical processor is not sensitive any more to what perceptual features define it, and the lexeme will therefore occur in the grammar as a lexical atom: its lexical content is not visible to the grammatical processor. As that processor builds structures, however, the process of lexicalization can apply again, now to whole configurations. Thus, the lexical item in (8a) might have the structure in (8b): (8)

a. kill b. [cause [to die]]

However, the insight that Fodor articulated in (1970) is that (8a) and (8b) actually differ in meaning. Ipso facto, if (8a) is grammatically complex underneath its mono-morphemic surface (contrary to what Fodor himself suggested), it follows that the lexicalization of its underlying structure makes a semantic difference. Lexicalization is meaningful, whether it takes place before or after the grammar operates: this is what we will have learned from Fodor. By now, lexical decomposition is widely assumed, though the details have changed. Harley (2012) argues that the right way to analyse kill as cause to die is to have a stative predication embedded under the causative predicate, as in The bride [made [SC Bill dead]], where SC stands for a Small Clause denoting the resultant state, and the whole structure in (9a) is spelled out as (9b):

uP

(9) a.

b. u⬘

DP u CAUSE

‘kill’ SC

DP

Pred DEAD


49

As Hinzen (2012) discusses, the lexicalization of (9a) as (9b) is indeed semantically meaningful. We can generate a configuration in the grammar, as in (9a), and then pack it into a single lexical item, but the lexical item will then crucially denote a single event that, unlike the one in (9a), is not decomposed into two parts (or an event and a resultant state). Thus, whereas the decomposed form (10) is naturally interpretable, the same intended meaning is unavailable or unnatural in (11): (10) John caused Bill to die by the time he reached the hospital. (11)

John killed Bill by the time he reached the hospital.

Similarly, (12) is fine, and while (13) describes a rather unusual activity (making a fish dead would be less unusual), it is interpretable. However, (14) is completely out: (12)

John made Mary happy and it surprised me that she was.

(13)

John made Mary dead and it surprised me that she was.

(14)

*John killed Mary and it surprised me that she was.

Moreover, while the j-option in (15) is quite bad but still available, the same interpretation in (16) is completely unavailable: (15)

Maryi made [Johnj sick] by PROi/?j eating too much.

(16) Mary sickened John by eating too much. Moreover, we notice that while certain grammatical configurations such as (9a) can find their way into lexical items, with a consequent change in meaning, this becomes impossible when the configurations become too large. Thus, sentences that carry truth values when asserted cannot be lexicalized, indicating a fundamental dichotomy between grammatical and lexical meaning: no lexical item is, as such, true or false.4 Such complexity does not fit into a lexical atom. The same applies to a referential phrase such as the man: there is no lexical item ‘the-man’. Reference, like truth, is not a lexical property, showing that if concepts, as we have argued, are used referentially, they have to be grammaticalized. Lexicalization—the creation of a lexical address for a set of perceptual features—then, is meaningful, and so is grammaticalization. But the two are crucially distinct. The conclusion that lexical items are atoms for the grammar does not take the point away that these items, as organized lexically, are nonetheless bundles of perceptual features, and hence are internally complex in this sense. This is suggested by a task such as detecting the common features of such lexemes as ‘table’, ‘chair’, ‘fish’, and ‘lamp’, which allow us to identify ‘the odd one out’. The same conclusion is suggested by the fact, noted above, that lexical roots form relations of hyponymy, 4 Pythagoras’ Theorem is a sentence, which is presumably true. But when we refer to it, using a nominal phrase or name, i.e. ‘Pythagoras’ Theorem’, we only refer to it, and do not assert it as true.

50


which atoms, being structureless, cannot do. Roots start acting as atoms when they enter the grammar, and together with that, enter into acts of reference. The way they do so appears to be subject to cross-linguistic variation. Armoskaite (2011) argues that in Blackfoot the specification of a grammatical category is required at the lexical level, while Lithuanian exemplifies the option that roots are precategorial. Vinokurova (2005) similarly argues that, in Sakha, a notion of ‘lexically primitive’ (rather than grammatically derived) Nouns and Verbs is required. Yet other languages such as Tagalog have been argued to have a morpholexically defined word class of verbs, in the absence of morphosyntactic distinctions among content words (Himmelmann, 2007). There appears to be variance, in short, in how roots enter the linguistic system, and the degree to which their lexical encoding reflects grammatical roles they tend to play. This variance in itself suggests that the lexicon represents a distinct way of organizing semantic information, which is independent of grammatical ways of organizing meaning. Meaning changes (roots can be used to refer) as lexical roots become grammaticalized, but this need not be reflected in how such items are stored lexically, as the cross-linguistic variance in question suggests.5 Nonetheless, it is unclear to what extent lexical roots really live an independent mental life, or a life independent of the actual words in which we see them occurring. If we abstract out a lexical root such as √run from English nouns or verbs that contain it, it is not clear what we are really thinking about. In pondering this question, we must be telling ourselves that we are not supposed to be thinking about a run, several runs, running in general, the fact that John runs, etc., but about √run, pure and simple. What kind of meaning is this? Modistic grammarians were well aware of the problem of properly characterizing the difference between ‘mental concepts’ and the words that express them. They carefully reflected on the extent to which the former were actually distinct from the words, as rationalist grammarians in the Port Royal traditions would later assume (and the majority of philosophers in the field of ‘concepts’ do today, cf. Fodor, 1998, and subsequent debates). By the early fourteenth century the contention gained ground that the grammatical features of words are at least partially irrelevant to the concepts they express, and should not be ascribed to these. Even if there is some kind of grammaticality to human thought (as there must be in a Language of Thought), and there are ‘mental sentences’, this ‘mental grammaticality’ need not be the same as ‘spoken grammaticality’. According to the nominalists, the former kind of grammaticality was indeed radically different, making language systematically misleading in the

5 In line with this, according to Levelt et al. (1999), information about ‘grammatical class’, in his terms, although lexically represented, is only activated when necessary, namely when sentences are produced or understood, but not when processing single words. On a related view, the grammatical category of a word does not enter into the brain’s processing of its content when the word appears outside of a sentence context (e.g. in picture naming or quotation contexts) (cf. Vigliocco et al., 2011).


51

study of pure thought (thought cleansed of linguistic distortions). As the Modists noted, however, some of the grammatical specifications are more relevant semantically than others—like the fact that a given word is a Noun. To the extent that this was the case, the (circular) suggestion was made that, in such cases, the mental concept in question is a ‘noun-like concept’. Thus the late fourteenth-century author Pierre d’Ailly writes: The exact reason why a particular word is a noun is that it is the representation of a noun-like concept, or that it is the symbol for a mental noun, which is the same thing. (Cf. Covington, 2009: 125)

This concept, on this view, explains why the word has the grammatical meaning it does, illustrating a semantic account of word classes of exactly the kind we will encounter and reject in Section 2.5. Other grammatical specifications of words in sentence-contexts appeared to traditional and contemporary grammarians as less relevant semantically. Morphological Case marking, for example, is a standard example of what is arbitrary and language-specific, as we will discuss further in Chapter 6. There is less reflection on this methodological problem in analytic philosophy, where the term ‘concept’ is often used virtually equivalently with the notion of ‘mental representation’. Yet the methodological problem in question is obviously not overcome if, as customary in philosophical practice, we effectively take concepts to be lexemes, denoted by capitalizing words. In this case, the ‘concepts’ so created— man, kill, etc.—are trivially mapped from the words, usually with a grammatical category assumed to be specified. At the same time, it is also said to be of the essence of such concepts to be (i) shared and (ii) to compose with one another (Fodor, 1998). Yet it is words, which do not belong to anyone, which are ‘shared’ between speakers, not concepts, which are in individual speakers’ heads. Moreover, it is not clear what evidence there is for a systematic combinatorics of concepts independently of how we see words involved in such a combinatorics. Thus, a typical psychological study of what is called ‘concept combination’ for example may compare the following combinations (Wisniewski and Love, 1998): (17)

a. zebra bag b. zebra trap c. donkey trap

The finding is that when the modifier noun has a particular ‘salient’ feature, as in (17a), and the head noun refers to an entity that can naturally receive this feature, it is more likely that subjects will generate interpretations where the feature is ascribed as a ‘property’ to the referent of the head noun (‘bag with black and white stripes’). This is less likely when the modifier noun has a salient feature, but that feature cannot easily be ascribed to the head noun (17b); or when there is no salient feature (17c). In these cases, a ‘relational’ interpretation is preferentially generated (e.g. ‘trap for a

52


zebra’), where ‘zebra’ plays the role of a thematic argument. The methodological problem is that what is being investigated here is plainly not ‘conceptual’ combination but word combination, more specifically noun–noun combination, with one of the nouns playing the role of head and the other as modifier (in 17a), or argument (in 17b–c). ‘Noun’, ‘head’, ‘modifier’, and ‘argument’ are grammatical notions, and although the lexical concepts involved will trigger different readings depending on the plausibility of the result, these readings are ones that are made available grammatically. Thus it is because ‘bag’ is perceived as the head of the construction in (17a) that this combination is not interpreted as a zebra in the form of a bag (which is a logically and conceptually possible combination, grammatically expressed through the phrase bag zebra), but a bag with the stripes of a zebra. The head determines the referent, the other word is grammaticalized as a modifier. An alternative experiment that would eliminate the language confound, yet tests a productive and systematic conceptual combinatorics, seems hard to fathom. We also cannot simply endow putative non-linguistic concepts with a semantics in which, say, the concept man is interpreted as a ‘predicate’, and a predicate in turn is a set of individual men. For at the pregrammatical stage we are now imagining, there are not yet either predicates or referential expressions, both of which are relational and grammatical notions again. Only words can have these semantic properties, and words have them only when they occur in appropriate syntactic positions (see Chapter 3). The evidence, then, for concepts, their combinatorics, and their semantics, comes from words and how these function grammatically. The world of concepts, insofar as they combine, is a grammatical world. Stockall and Marantz (2006) argue on the basis of magnetoencephalography (MEG) evidence in priming tasks that the brain processes all morphologically complex words so as to initially decompose them into independently meaningful morphemes—not only in the processing of regular allomorphs (like–liked), but in irregular allomorphs (e.g. teach–taught) as well.6 Even when the brain processes the letter string ‘taught’, that is, it separately activates the stem teach and the functional morpheme that denotes the grammatical meaning ‘past tense’. How two words are related in terms of their morphemic decomposition, moreover, is different from them being related in terms of orthography or phonological similarity, and indeed from whether they involve non-grammatical semantic similarities that occur in the absence of any plausible morphological relationship (as in the pairs boil–broil or crinkle– winkle, see Feldman, 2000, and Rastle et al., 2000). If so, there is a level of lexical semantic organization prior to and independent of morphology, and there is a MEG component to index it (the M350). But the semantics of the stems is then still not semantics in a generic, non-linguistic sense: access to or activation of this 6 For obligatory decomposition approaches to polymorphemic words see also Taft (2004) and de Almeida and Libben (2005).


53

semantics is after decomposition of a given morphological structure, in which the relevant stem appears. There is no evidence here for access to pristine, purely lexical and not-yet-categorized roots or non-linguistic ‘concepts’. The problem of whether such lexical roots are psychologically real, independently of their occurrence in a linguistic (morphological) format, thus remains. In summary, in the course of lexicalizing a given percept, a bundle of perceptual features is given a phonological and semantic identity that makes it available to a novel system: grammar. These bundles are internally complex and organized hierarchically in relation to one another, as relations of hyponymy suggest, yet they are single objects or atoms from the viewpoint of this novel system, grammar, in which they function as atoms that cannot be further analysed. Atomization is a process that can apply to (small) chunks of grammatical structure itself, and is in itself a meaningful process, which makes a difference to how we refer to the world, when such atoms show up in grammatical configurations. With lexicalized concepts that can play the role of such atoms, we leave the world of perception, which is still stimulus-bound, and enter a world that can be crucially shared, unlike a stimulus, which only affects us individually. A new semantics arises from this, resting on words, on which all acts of (intentional) reference seem to depend. With a lexicon in place in a language, the world becomes a place that is precategorized. The mental activity of generations is off-loaded into a public resource, a residue of shared experience, in which the world is structured through words and their possible referents, rather than merely through what we individually sense and feel. Through shared words, our minds form a single connected system, harbouring knowledge, by the time a language is mastered. As lexemes, words may be partially independent of the grammatical specifications that words in sentence contexts come to exhibit, since languages seem to differ in the extent to which words come with prespecifications of how they grammatically behave. Moreover, the way they can combine systematically does not depend on the feature specifications that define their lexical content: their status as grammatical words appears to be independent of that. Lexemes are not connected to stimuli as percepts are, but they form a necessary resource for acts of reference that presuppose a shared world interpreted in terms of such lexemes. As acts of reference are enabled, the system thus remains necessarily connected to the world through lexicalized concepts that may in turn be largely based on percepts. But falsehood can now happen, too.

2.4 Concepts ´ intentionality ´ intensionality ´ grammar ´ reference So far we have (i) looked at the transition from percepts to lexical items, (ii) distinguished (mental) representation from reference, and (iii) distinguished the lexical content of lexemes or roots, which is specified in terms of perceptual features, from the grammatical meaning that words acquire when they occur in a grammatical

54


context and are used to refer. We have referred the term ‘concept’, but generally concluded that what we were really talking about were either percepts (or at least entities continuous with percepts), mental representations, or lexemes. If so, the notion of a ‘concept’ is of little use, no case for prelinguistic concepts distinct from percepts remains, and we arrive at what is generally considered a ‘harsh’ stance on the animals. Our stance effectively assumes the existence of concepts only where there are lexemes, and indeed only where such lexemes enter into acts of intentional and intensional reference, which in turn require grammar. It is this correlation that the title of this section expresses, and we will use the term ‘concept’ in what follows with this meaning: there are concepts in this sense only where there are lexemes, there is grammar, and there is intentional and intensional reference. This is our answer to the question of what ‘concepts’ really are, who has them, and what their semantics is. To defend this answer, let us start from a natural suggestion: a concept is always involved when we use a word to refer to a thing. The speaker will always have some ‘concept’ of what it is he is referring to, viewing it as a person, a kiss, an emotion, a conversation, etc. We have suggested that even acts like referring with such a word as ‘this’ or index finger pointing depend on some such descriptive concept of what it is we are pointing to, and they would not work without it. As in the case of perception, the objects of such acts of reference are also never individuated in terms of their physical properties, as noted in Chapter 1 (and cf. Chomsky, 2000a), and this would follow if, instead, their identities are grounded in what concepts we apply to them, or how we think of them: as a person, a city, a river, etc., where usually multiple such choices can be made for any one external object. Thus we can refer to a building as ‘a house’, ‘the university’, or ‘Old Shire Hall’, among an indefinite number of other possibilities. None of these are physical descriptions, or descriptions of an object as a physical one. When the building is destroyed through an earthquake, the university need not be: it may retain its identity. The house by contrast will most likely be taken to be destroyed. As for Old Shire Hall, it may or may not go out of existence: for it may be rebuilt stone by stone or otherwise, retaining its name. What causes these facts? What explains that houses get destroyed by earthquakes, while named buildings need not be? Plainly, the concepts we have of these things, and not any independent fact about the external world as physically described. Concepts in the substantive lexicon have descriptive contents—person, house, kiss, etc.—which determine identity conditions of objects of reference: viewed as persons, they change in other ways than viewed as bodies. Any act of reference, then, contains an identifying description, however reduced it may be, which as such must involve a concept, which is what supplies the descriptive content in question, and hence the identity conditions for the referent (we return to so-called ‘rigid’ or ‘direct’ reference, which is often taken to be unmediated by a description and seemingly contradicts the claim just made, in Chapter 4). Because the act of reference is anchored in the description, the thing referred to therefore need


55

not causally affect us when we refer to it, and it need not be perceivable or characterizable in sensory-motor vocabulary: for it is grounded in the concept involved, as a basis for the act of reference. For the same reason, the referents of the words, as characterized independently of how we refer to them, will never exhaust the words’ meaning, and thus intensionality arises: what we mean can never be gleaned from what we refer to as independently described, since how we describe a thing systematically affects what we can be truthfully said to refer to. Lois Lane may refer to Clark Kent as ‘Superman’, yet it may be false to say of her that she referred to Clark Kent. We may refer to a building as Fountains Abbey or as The Wonder of the North, and then, as in the case of Lois, circumstances may arise in which it might not be right to report us as attributing this property to the ruin in North Yorkshire that we can see as the speech act takes place. So if human linguistic reference is structurally based on concepts, and concepts have a semantics independently of what they are used to refer to (as independently described), we predict that human linguistic reference will systematically exhibit intensionality. If reference depends on concepts, it is sensitive to a description, and hence it need not be transparent in relation to the object referred to. But even if acts of reference are necessarily based on concepts, it is not yet clear that without reference, there will be no concepts. Maybe there are some concepts, but they simply aren’t used to refer. In that case, they might not be lexemes, and independent of language and words. Qua concepts they would have a descriptive content, but maybe there can be description without reference. This, however, does not seem intuitively right: whoever describes, describes something, an object of reference, even if the object is not named or directly referred to.7 Let us explore, then, the opposite thesis, which is entailed by the following claim: The Correlation Claim: There are concepts if and only if there is intensionality. We just argued for the right-to-left direction: Where intensionality exists—there is reference and it is sensitive to a description—concepts must be involved: the ones that provide for the lexical content of the description used. No human act of reference is ever fully free of intensionality in this sense, interjections and indexicality included, a point to which we return. Intensionality is thus a plausible sufficient criterion for concept possession. The left-to-right direction, however, is much less obvious, especially on ‘internalist’ conceptions of concepts such as Chomsky (2000a) or Pietroski (2008). Intensionality, we may grant, requires two things: that words are used referentially, and that different concepts can be involved in such acts that are 7

And even if, as it may later turn out, we make an informed judgement that such objects of reference do not in fact exist. Existence claims arise at the level of full propositions, not acts of reference, where existence is presupposed but not judged (see further pp. 99–100 below).

56


true of the same external referent. A creature to which we ascribe intensionality thus needs both a sense of extensional reference and truth, and it needs to engage in such acts of reference under the kind of perspectives that concepts encode. But why should this link between concepts and reference be made, which will exclude concepts along with intensionality from non-pointing and non-(intentionally-)referring creatures? There would then be no concepts, in particular, in core cognition (Carey, 2009), and they would have to be called by a different name, and maybe be aligned with percepts. Why burden, the internalist might ask, a theory of concepts with a notion of reference that Chomsky (2000a) doesn’t even consider to apply to natural language? Yet it is a virtual triviality to state that to have a concept is to know what it applies to—apply truly, that is, for a concept applies to a thing if it is true that the thing referred to with the concept falls under the concept.8 How can we avoid a connection between concepts and reference/truth if it amounts to a virtual triviality? What is a concept if it is not true of something? How is saying that the concept of a book applies truly to books even different from saying that ‘book’ refers to books?9 How else would we state what the content of a concept is? And how could we do so more minimally than by saying that ‘book’ applies to books? Clearly we couldn’t say: ‘book’ applies to books if it appears to us to so apply. This would not only be incorrect, but the notion of appearance is in fact more complex than that of truth, presupposing it analytically. The truth, however, is independent of what we think—nothing is true because we think it is, and this is constitutive of what truth means for us. Similarly, reference on occasions of language use is to objects that are (taken to be) independent of the speaker and hearer—this again being constitutive of what reference means. So if there is an intrinsic nexus between concepts and reference/truth, then knowing a concept requires a notion of ‘the world’ as independent of (or to be distinguished from) what we think about it. Without such a notion, we wouldn’t be thinking at all. Thinking requires both a notion of the world and of thought; and concepts, as involved in thinking (as opposed to perception), must therefore be used to refer. This is our argument. If, in all acts of reference, concepts are involved, and in order for there to be concepts they have to be used to refer, concepts and intensionality arise together, grounding the correlation claim above.10 Where intensionality cannot be established, a case for non-human concepts (in the same sense of this term) 8

This is to say absolutely nothing about how we know when a concept applies truly. This is again to say absolutely nothing about what books are—and in particular whether they are viewed as abstract (‘Bill and Bob have both read Buddenbrooks’) or concrete objects (‘Bill wrapped Buddenbrooks in Christmas paper’) (see Chomsky, 2000a). These are conceptual distinctions arising when words are used to refer on occasions of language use, not at the level of stating their lexical content: both of the above sentences refer to what is (lexically or descriptively) a book. 10 Crucially, though, the descriptive content of the concept involved can enter into the act of reference in different ways, giving rise to the different forms of reference that we distinguish in Chapter 4. In purely predicative forms of reference, the concept is the main determinant of the referent; in ‘rigid’ forms of reference, it plays a minimal role; definite and quantificational forms are in the middle between these. 9


57

cannot be made. That it cannot be established in the non-human case follows from the well-known observation (Hauser, 1996; Fitch, 2005) that while functional reference is quite ubiquitous in animal communication, reference in the present sense (intentional reference) is present only in Homo sapiens. We return to these data in the next subsection. Concepts, then, along with reference, present an evolutionary novelty, and their connection with words may be intrinsic. Whenever concepts are used to refer, words are normally involved (though they need not always be, as in acts of pointing). In philosophical and psychological practice, concepts effectively are identified as the meanings of words, and there does not appear to be any other way to name or identify concepts than through the words that express them. Concepts identified with word meanings, however, already reflect processes of lexicalization and grammaticalization. The assumption that these are irrelevant to the existence or identity of the concept involved begs the question against the possibility we entertain here: that the grammaticalization of the brain may affect the organization of meaning. By now we have a first indication of how it may do: through the evolution of reference, which, we have argued above, is, in fact, a co-evolution of reference and concepts, for one depends on the other. The intrinsic semantics of concepts, as a novelty in evolution that arose with or in our species, is a referential semantics, where the notion of reference is intentional reference. We could, of course, if we wished, simply abstract the meanings that words have in the human lineage from them—abstracting from the fact that they are linguistic, as most discussion of ‘concepts’ in philosophy and psychology tends to do—call them concepts, and ascribe them to non-human species. But this is to impose our conceptual scheme on the cognitive systems we study. Since non-human animals never respond linguistically to what we say about their minds, it is unclear how to test whether they share this conceptual scheme with us or not. Put differently, if they had concepts, they would necessarily share our conceptual scheme—for there is no other way we could understand them except by mapping their scheme onto ours. But then there is no empirical question of whether they share our scheme or not (Davidson, 2004), and the question of animal concepts is unresolvable. The question, in turn, of whether they have our conceptual scheme, arguably reduces to whether we can interpret them using our scheme: and the fact is we cannot. Pet owners and their pets can without doubt communicate and coordinate on many things, like a ball to be fetched or a game to be played. Nonetheless, the point of owning a pet is not that the pet has a mind like ours, but that it does not. What makes it so charming to have a dog—as opposed to a human pet (or slave)—is that we precisely don’t share a world with him/it: a realm in which thoughts can be true independently of whether we think them, and words are true of things that are

58


independent of these words. Contrary to popular opinion, imposing our mind on non-human animals is not a sign of our respect for them: perhaps it is the opposite, to colonize a territory that we cannot chart. Do we perhaps lack the means to chart it, even in the case of our fellow-humans? Might ‘other minds’ be like a foreign country whose customs are completely different? What if we cannot establish intensionality—shared reference and concepts, even if the chosen concepts are not always the same when reference takes place—even in human communication? Have we not learned that external behavioural and other physical data, no matter how rich, cannot uniquely determine or fix any concept involved in an act of reference? Is it not that nothing a person says or does will ever possibly conclusively determine the concepts she expresses, leaving an infinity of interpretive options (Kripke, 1982)? Not if we are right that reference to a world that is independent of how we think of it is a precondition of having concepts and thought in the first place. Divergence in concepts when faced with a given referent is then based on sameness of reference. Concepts are objectively anchored. For concepts to exist, we must be aware that thoughts that involve them can be wrong, which in turn means that we must have a concept of thought. That concept in turn entails that a thought that I think need not be a thought that you think, and we both can be wrong—at least in particular cases. But then, I have to distinguish the 1st Person (‘me’) from the 2nd Person (‘you’), and both of these from the ‘it’—the object of reference. These are distinctions involving the system of grammatical Person, and hence are linguistic ones.11 If so, in order for there to be thought, I have to have language. This gives an indication of the direction of our answer to the Kripkean sceptic: no, concepts are not inscrutable even in the human case, for it is a condition on having them that we use them to refer, which not only requires a shared world but also an intersubjective dimension that is linguistically configured. Reference underdetermines concepts, leading to intensionality; but intensionality presupposes reference. Intensionality in word reference means that we can distinguish concepts, as something that we share with others, from the referents to which these concepts may or may not apply, independently of what we think. This is why Lois Lane can be confused about who is Superman and who is Clark Kent. But then, her concept when she references Superman is just the one we would have, if we had the beliefs that she does. Moreover, such confusions are typically short-lived. Language, then, we will assume, enables reference and truth, and via reference and truth enables concepts, which can enter into thoughts but in a way that the

11

They are not non-grammatical semantic ones, as the same external object (me, say), can be the referent of ‘I’, ‘you’, or ‘he’. But acts of reference to the same person involving these items differ in grammatical meaning, and hence use (see Section 4.4).


59

truth of these thoughts (or external circumstances) does not determine which concept is chosen. This in itself is an indication of thought taking place.

2.5 Parts of speech and their grammatical semantics Given our account of percepts, lexicalization, and lexemes, what can we say about the ‘grammatical word’ and how semantics changes again when the latter evolves? The grammatical word is the word as occurring in specific grammatical configurations where it carries out grammatical functions, such as that of predicate or modifier. From an Un-Cartesian point of view, we do not expect in that case that the explanation for what the words mean could be non-grammatically semantic: for we expect their semantics to change upon grammaticalization (the alternative would be that grammatical organization makes no difference to meaning at all). It is therefore surprising to see that (lexical or non-grammatical) semantics has virtually universally been the basis upon which linguists and typologists outside of generative grammar have attempted to found a theory of the parts of speech. Almost invariably, the starting point of discussions of the word classes in functional and cognitive linguistics (where most attention has been devoted to this problem) is that the lexicon is partitioned into (notional) ontological classes such as object, action, and property (often called ‘lexical semantic classes’), corresponding to the parts of speech ‘Noun’, ‘Verb’, and ‘Adjective’, respectively (see e.g. Croft, 1991; 2001; 2007; Langacker, 1987).12 The most crucial two equations appear in (18) (see e.g. Bisang, 2011): (18)

The parts of speech Action-denoting lexemes (= verbs) Object-denoting lexemes (= nouns)

It is then commonly recognized that such a characterization of the parts of speech is clearly not sufficient. Yet, discussions nonetheless often start out from the intuition

While I will take opposition below to either ‘cognitive’ or ‘semantic’ accounts of part of speech distinctions, Langacker (1999: section 1.3) makes an important distinction within the framework of cognitive grammar between ‘semantic construal’ and ‘conceptual content’. The former involves a particular perspective on reality enabled by grammar, in line with a conception of grammar that regards semantics as inherently present in syntactic form (syntax is meaningful, but the kind of meaning is different from lexical content). Semantic construal is used to define the parts of speech, so that ‘run’ as a noun and as a verb involve different construals of the same conceptual content. Langacker’s distinction is clearly related to the Modistic distinction between structured perspectives on reality and what they termed the modus essendi, and it is also compatible with the present approach. All three are radically incompatible with an approach such as Jackendoff ’s (2002), who deprives syntax of meaning altogether and sees no distinctive abstract semantics in parts of speech distinctions. The present approach is different from Langacker’s insofar as his starting point is cognition, whereas ours is grammar, so that we are coming from different directions. However, the heuristic ideas that grammar reduces to ‘semantic construal’ is shared between his and our approach. We are extremely thankful for Jaume Mateu for raising these interesting points. 12

60


that they should be, in a perfect system. That is, if grammatical categories do not mirror notional ones, things are ‘imperfect’ somehow. As Himmelmann (2005) points out, however, if word classes were directly predictable from the notionalontological class of their lexical meanings, there would be no need for establishing word classes on the basis of grammatical criteria. Grammatical categories, on this view, are only useful insofar as there is no perfect match between lexical meaning and grammatical category: in other words, when grammar adds something new. It can do this by giving words particular morpholexical characteristics that make them fall into equivalence classes defined by these characteristics—e.g. affix sets with which a given lexical root occurs, distinct conjugation or declension classes, etc.—and/or morphosyntactic characteristics—like occurring in particular syntactic positions in a phrase structure tree, forming the complement of a definite article, playing the role of predicate, etc. What we propose here is that the emergence of parts of speech classes among content words reflects a distinct kind of grammatical semantics as well. According to this idea, grammar itself has an abstract semantics to it. This view appears to be virtually absent from discussions of parts of speech, which, apart from lexical semantics, largely seem to focus on morphosyntax and pragmatics (though see fn. 12). We would expect this on the view that we here dispute: that grammar never adds anything to meaning. Morphology, however, which is one of the prime loci of crosslinguistic variation and also largely absent in many languages, can hardly ground an account of parts of speech that is meant to have universal validity. To the extent that there is anything universal about parts of speech distinctions, therefore, it would seem better to ground it in a semantics specifically reflected in such distinctions. If and to the extent that this semantics reflects grammatical functions that such parts play, it would be an aspect of grammatical semantics in our sense, which in general is the most likely candidate for cross-linguistic universals. We will contend in this subsection that categories like ‘Noun’ and ‘Verb’ indeed are ultimately language-specific and morphological entities, making them slightly different categories in different languages, though family resemblances can be seen. But they are one way in which a language can morpholexically and morphosyntactically manifest a grammatical semantics (grammatical functions that words play) that is unavailable through lexical content (feature specifications) alone. If this semantics is universal, it would provide a rationale for why a Noun–Verb distinction is so widely attested among languages. Arguably, this grammatical semantics relates to the possibility of making formal-ontological distinctions in how words are used to refer, or what perspective we take on reality. In particular, the distinction between actiondenoting and object-denoting lexemes in (18) above, we will argue, is already a grammatical one, which cannot be explicated in (non-grammatical) semantic terms: it is not entailed by the classes of lexical-perceptual features defining the contents of particular lexemes.


61

In short, there is a universal formal distinction between referring, on an occasion of language use, to something as an ongoing action or as an object, which is grounded grammatically. And there are partially language-specific ways in which parts of speech classes can reflect the availability of such grammatically configured forms of reference.13 A content word used to refer to a specific object, for example, will cooccur with a determiner in a language that has determiners, like English (‘this man’). The referential expression as a whole will then typically also be an argument. Moreover, such words will then typically also be nouns. But this does not show that the same category of ‘noun’ is found in all languages where such referential functions can be enacted. Nor does it show that the connection between the notion of ‘argument’, which is a purely grammatical notion, and the notion of ‘noun’, is intrinsic. Distinctions of formal ontological category and grammatical functioning within acts of reference can be coded at the level of the word itself, turning it from a lexeme into a grammatical word. But they need not. Even when coded at the level of the word, they prove inherently unstable: in English, for example, we can insert the same lexeme into several different grammatical configurations and thereby change the formal ontology associated with it, as when re-categorizing the word laugh, which normally is used to denote an action, to refer to something viewed formally as an object (a laugh, laughter). This process is very systematic and does not in this sense seem to be lexical. To whatever extent ontology is lexically coded, therefore, it is not fixed at this level, whereas the ontology of a particular word used referentially within a grammatical context is necessarily determined. This of course is not to deny that perceptual aspects of the object–action distinction are available pregrammatically. Rocks, water, and persons have different and distinctive perceptual features from things that we perceive as ongoing actions, such as flying, kicking, or swallowing. It is also not to deny, as discussed in Section 2.2, that babies parse the world into perceptual objects in many ways like we do when we refer to them, applying abstract criteria of identity to them and viewing them as instances of basic-level kinds such as dog or hammer, or even global kinds such as animal, vehicle, or furniture (Mandler, 2004; Carey, 2009). None of this, however, gives rise, in our terminology, to a formal ontology, which rather arises with (intentional as much as intensional) reference to such entities, which is absent in the babies in question. Reference in this sense is to entities falling under a narrow range of formal ontological types that systematically relate to one another (with objects, e.g. being inherent parts of events, and events parts of propositions), in ways that we will see grammar, and no other system, governs and restricts. To properly develop this account, we will proceed contrastively, by discussing an approach to the parts of speech that deprives grammar of the epistemological 13 ‘N and V-features’, then, are not part of universal grammar, as contended in Boeckx (2010). They are not coded into the linguistic genotype.

62


significance that we here want to credit to it: Croft’s (2007) cognitive-functional approach to the word classes. It assumes that we need to start our explanations from what Croft calls ‘conceptual frameworks’. These are defined as: inventories of the types of semantic structures that are expressed in language and the semantic features or properties that must be instantiated in utterances, and therefore must be expressed in grammatical inflections and constructions. (Croft, 2007: 340)

Of course, this stance fully entails what we call a Cartesian rationalist point of view: the ‘semantic structures’ in question—contents of possible thoughts—are viewed as non-linguistic in nature, given to language from the outside, as something to be ‘expressed’ in it. Croft (2007) interestingly aims to expand this general stance into a principled explanation of the elements of grammar, in an effort to answer the question of why there is such a thing as grammar in the first place, with a specific organization into parts of the sort that we find in it. The answer given, following the lead of Chafe (1994), is that the origin and nature of grammar lie in the ‘verbalization of experience in the process of communication’ (Croft, 2007: 341). Precommunication, that is, there is assumed to be an ‘experience’, called a ‘unified whole’, which is crucially not ‘propositional’ in character. Indeed we cannot circularly presuppose any grammatical structuring principles in characterizing what such an experience is like (p. 348). In the process of communicating it, this ‘undivided whole’ is then first broken into parts (called ‘concepts’ and ‘relations’, which seem non-linguistic in nature), by non-grammatical operations such as ‘sub-chunking’, which decompose the whole in question, and ‘construal’ or ‘framing’ (called ‘propositionalizing’). The result of these operations is a structured propositional representation that has individuals and predicates in it, foregrounded and backgrounded information, and an ontology (events, situations, etc.). The third basic operation is ‘categorization’, which yields the word classes. In the verbalization process, these parts of speech are then put together again, so as to reconstruct the original ‘unified whole’, which after all is what the speaker wanted to communicate. Again, none of these operations are taken to be grammatically specific. They are called ‘cohering’, ‘structuring’, and ‘particularizing’. Together, they provide for the unity of a sentence. Insofar as the grammar of a sentence instantiates the results of all of these various analytic and synthetic operations (which are not formally defined), a principled explanation for grammar is found. The crucial question in the present context must be whether the operations in question are independently explanatory or whether they merely re-describe independently attested grammatical operations in non-grammatical terms, with no added explanatory value. An immediate problem for the account is that ‘experience’ is a species-specific notion co-varying with the cognitive resources available to a given creature. How we experience and are in the world changes radically, in particular, as language intrudes into our mind. Only a grammaticalized world is a world with a past


63

and future—a history—or with facts that can be items of propositional knowledge. When characterized in non-linguistic terms, therefore, the notion of experience does not yield (or somehow latently contains) the primitives and operations of grammar, and certainly cannot fully explain them. Experience as such even fails to entail the existence of ‘concepts’, as we have seen: a creature can experience long before it can conceptualize. Perceptual experience is richly structured, but does not reflect the categories of grammar, and ‘sentence’ is not an experiential category in any sensory-motor sense, similarly to an experience of pain, redness, or sadness. As we think, judge, and articulate, no sensory-motor experience of the world needs to be involved.14 Croft’s example sentence, Next the boy comes upon a slide, invokes a perceptual content. Even so, it is structured by grammatical relations that have no experiential content, such as subject, predication and modification. The word ‘next’ is not clearly a perceptual category, as it denotes a sequential relation between a number of events that is given as part of a grammatical process of intentionally referring to a number of objects viewed formally as events. The specific definiteness of ‘the boy’ is not an experiential category either. There is also no known characterization of predication in non-grammatical terms. One could define predicates in formal-semantic or settheoretical terms, but as Davidson (2005) discusses, the original grammatical predicates enter into the definition of the sets that then represent the predicates, which means that the definition explains nothing. Croft (2007) points to the ‘difference between individuals (arguments) and events (predicates) in propositionalizing’ (p. 363), yet ‘argument’ is a grammatical notion that cannot be captured by a non-grammatical distinction between individuals and events. Thus, subjects that do not denote individuals (e.g. phrases such as all men, mass nouns such as beef, or clauses such as that John is silly) can be grammatical arguments (or at least topics); individuals can be parts of grammatical predicates (e.g. John is the man, John [killed Bill]; events can be subjects and arguments (e.g. John wanted to kiss Mary, Kissing Mary is a delight) and they are interpreted as individuals in the Davidsonian paradigm (Davidson, 1970). Generality, referentiality, and subtypes of the latter, such as definiteness and specificity, are also grammatical properties, insofar as words have them when and only when they are introduced by relevant functional heads and occur in relevant syntactic positions in the sentence. The same applies to aspectual distinctions at the level of the verb, and corresponding distinctions in the noun (mass, count, Number, Person, etc.). Like Bisang (2011), Croft further analyses reference and predication as pragmatic discourse functions that can co-define the parts of speech, but again there is no evidence that these exist

14 Of the sort that clearly is involved in the storage and neural processing of lexical items, which do connect with sensory-motor representations (Barsalou, 2008; Pulvermueller, 2012).

64


independently of grammar, and as we have seen, reference takes a very different format when grammar is absent.15 Croft acknowledges some of the problems with the idea that the part-whole organization of the sentence is determined or explained by non-linguistic processes of composing and decomposing an experience, in the following passage: One cannot simply equate the predicate-argument contrast with the syntactic noun-verb distinction, or even the semantic classes that are prototypically taken to represent the nounverb distinction. Verbs and/or predicates are usually equated with actions, nouns and/or arguments with objects (persons or things), and adjectives with properties. But actions and qualities can be nouns (explosion, length) and as such, function as arguments in a clause (The explosion reverberated through the valley). Conversely, both objects and properties can be predicated. In English, nonverbal predication is a syntactic construction with an independent copula and no morphological alteration of the object/property word (She is tall, She is a student); but in other languages, the object/property word is directly inflected with ‘‘verbal’’ affixes and its category status is therefore often disputed ( . . . ). Finally, both objects and actions can function as modifiers (a doglike creature, industrial policy, a sleeping fox). On top of this, whole clauses can function as arguments or modifiers. Clauses functioning as arguments are complement clauses, as in She believes that he’ll call back this afternoon. Clauses functioning as modifiers are relative clauses, as in the house that sold last week. (Croft, 2007: 364)

The natural conclusion to draw from these observations is that the predicateargument distinction, which is purely grammatical, cannot be explicated in terms of word classes: adjectives and nouns are equally good predicates as verbs, yielding truth-value bearing propositions in exactly the way that verbs do, after Tense is specified; and word classes in turn cannot be explicated in formal-ontological terms. Grammar, that is, rather than predictable in its workings from the way that words can be grouped into classes by lexical semantic criteria, is substantively innovative in regards to the organization of meaning. It adds meaningful structure

15 As Baker (2003: 15, fn. 3) points out, the attempt to ‘externally ground’ the tripartite Noun/Verb/ Adjective distinction in a THING/ACTION/PROPERTY distinction and pragmatic functions between referring, predicating, and modifying, makes us wonder why there are these three distinctions and functions. ‘These “external groundings” look like different labels for the language-internal noun/verb/ adjective distinction’. Elsewhere (p. 293) he points out that ‘what we really mean by a thing is “whatever can be referred to by a noun” ’. Our crucial departure from Baker is that, for him (Baker, 2003: 15), referring is something inherent in nouns just as predicating is inherent in verbs. He is in this sense a lexicalist about what we consider inherently grammatical functions, and which we think nothing in the organization of the mental lexicon as described above predicts. The organization of meaning at the level of the lexicon as well as the parts of speech have different functions. On the other hand, in an interesting passage Baker agrees ‘with the medieval grammarians ( . . . ) and with Langacker (1987): the lexical category distinctions correspond not so much to ontological distinctions in the kinds of things that are out there in the world, but rather to the different perspectives we can take on those things, the different ways that our linguistic capacities give us of describing them’ (p. 294), with the slight qualification that we would like to replace the word ‘describing’ in the end by ‘referring to’.


65

rather than recapitulating one that is already given, or else merely messing it up in an arbitrary way. The same conclusion follows from the consideration that the terms ‘argument’ and ‘predicate’ are relational or correlative ones. Something can only be an argument if there is a predicate of which it is an argument, and vice versa. This is not clearly true of either the distinction between nouns and verbs or between objects and events. Thus, in a metaphysical ontology, ontological categories might simply be listed, in which case they are not relational and co-dependent.16 The same applies to nouns and verbs, if claims are true that some languages have word classes of nouns but not of verbs (e.g. Tongan: Broschart, 1997) or of verbs but not of nouns (Cayuga: Helmbrecht, 2005). If so, there can be arguments and predicates without any necessary reflection of this distinction in an either semantically, morpholexically, or morphosyntactically based classification of lexical items. As Himmelmann (2005: 87) points out: it is widely agreed that the members of the major lexical categories NOUN and VERB have prototypical discourse (or pragmatic) functions, i.e. reference and predication, respectively. It is also widely, if not unanimously, agreed that these major discourse functions are universal in the sense that all languages have to provide means for performing these functions. However, it would be wrong to conclude from these two widely held views that the distinction between the two word classes noun and verb is also universal [our emphasis]. That is, from the fact that all languages have to provide means for performing the acts of reference and predication it does not follow that the linguistic units used as referring and predicating expressions are formally distinguishable (in terms of phonological, morphological or syntactic properties). And even if one were to grant that in most, if not all languages referring expressions in some way can be distinguished from predicating expressions with regard to at least one formal property, it does not necessarily follow that such formal differences are found on the lexical level. It is very well possible that one may distinguish between referring phrases (marked, for example, by some kind of article) and predicating phrases (which may be unmarked or have a specific marker of their own) without being able to claim that the content words of the language fall into two classes with regard to their propensity to occur in one or the other phrase type.

Tagalog arguably illustrates how a language may set up its basic grammatical functions without any distinction among content words in regards to the syntactic positions they can occur in: reference and predication—the relational core of grammar and the essence of its significative function—arise without a classification of words into nouns and verbs. In Tagalog this applies, as Himmelmann argues, more specifically to four of the major grammatical functions of words: predicate, subject, non-subject argument or adjunct, and modifier. That a word or phrase performs any one of these functions is not predictable either semantically (on the basis of 16 Some ontological categories, however, are commonly recognized to form relations of ontological dependence with other categories (Fine, 1995).

66


notional-ontological class of the head involved), morpholexically, or morphosyntactically. This is expressed as: The syntactic uniformity claim for Tagalog: ‘All content words may occur, without further derivation or conversion, in the same kind of phrase-structural positions.’ (Himmelmann, 2007; for an older tradition supporting various versions of the same claim, see e.g. Müller, 1882: 99 ff., Bloomfield, 1917; Lemaréchal, 1989; Gil, 1993; Naylor, 1995).

For example, content words that can semantically (notionally-ontologically) be classed as actions and that are fully inflected for voice and mood can occur in all the positions where, in English, one would expect a noun to occur, for example as the complement of the specific article ang in (19), or with quantifiers such as the existential quantifier may as in (20) (from Himmelmann, 2007:266):17 (19) iuuwi nya ang àalagaan i-RED1-uwí' niyá ang RED1-alaga'-an CV-RED1-returned_home 3.SG.POSS SPEC RED1-cared_for-LV nya niyá 3.SG.POSS ‘He would return the ones he was going to care for.’ (20) may ipàpakita ako sa iyo may i-RED1-pa -kita akó sa iyo EXIST CV-RED1-CAUS-visible 1.SG LOC 2.SG.DAT ‘I have something to show you.’ ‘There are things to be shown.’ At the same time, Tagalog roots arguably do form classes defined by morpholexical criteria—a classification which is neither reflected in what grammatical functions a particular word takes up nor in the ontological classes into which un-affixed uses of the roots in question fall (such as action/object). Himmelmann (2007) calls one such class ‘V-words’, where V stands for ‘voice’-markers with which the relevant roots occur. V-verbs may denote the actual performance of an action, which he argues roots cannot do, yielding an interesting insight into the nature of grammatical

17 Abbreviations: CAUS causative CV conveyance voice DAT dative LOC locative LV locative voice POSS possessive RED reduplication (numbers indicate different formal types of reduplication) SPEC specific article


67

meaning. While this makes V-words similar to English verbs in some respects, Himmelmann stresses that they are crucially different in others. In particular, the classification of an English word as a verb constraints its morphosyntactic behaviour, i.e. its position in the phrase-structure tree. But a Tagalog V-word can, like all other content words in this language, occur in any position, picking up a grammatical function accordingly, in line with Syntactic Uniformity. If so, definitions of parts of speech—such as ‘V-word’ or ‘verb’—will always and to some extent be language-specific: for they necessarily depend on the formal resources of a language, and these resources will typically differ. So different languages will have different morpholexical and morphosyntactic classes. These are defined by various ontological, morpholexical, and morphosyntactic criteria to different degrees, yielding word classes of various overlapping sorts. Does that create a problem for a universal grammarian? Not if we set our sights on the grammatical functions and relations that hold, in particular languages, between the words that fall into such classes. As long as it is agreed that what basic grammatical functions and relations there are (such as reference, predication, and modification) is not affected by what word classes there are, and that they are universal, such cross-linguistic variation in the parts of speech systems does not create a problem for our view. On the contrary, it enhances the conclusion that grammar is more than an ‘expansion of the lexicon’. It is the relational and referential character of grammar that is of the essence in the reformatting of cognition that we are investigating here. The functions and relations in question, and the new cognitive format they describe, may be partially independent of morphology, but they are not merely semantically given: they only arise in grammatical configurations, and are grammatical in this sense. They cannot be predicted from or are determined by what kinds of words there are. Thus, performing the referential function is not the same as being a noun, though nouns can perform this function, as can adjectives, in relevant configurations. Performing the predicative function is not the same as being a verb, though verbs tend to function as predicates; and performing the modificational function is not the same as being an adjective. This stance makes some of Croft’s observations in the long quote above completely predictable: it explains without further ado why ‘clause-type’ syntactic objects can function as both modifiers and arguments. Similarly, it explains why the lexicalsemantic status of a root such as √explode does not determine an ontology, allowing for both verbal (explode) or nominal (explosion) uses. The way Croft puts this fact is by saying that ‘actions and qualities can be nouns (explosion, length) and as such, function as arguments in a clause (The explosion reverberated through the valley)’. But there is no necessary correlation—not cross-linguistically, or even in English— between the grammatical function ‘argument’ and the word-class ‘Noun’. If an action-root is grammaticalized as a noun, this does not mean that an action can be a noun: for when we grammaticalize the root √explode as the noun explosion, ontology changes: we refer to it as if to an object. The way in which we refer to this

68


root differs when we say (The bomb) exploded and The explosion (happened). What is different in these two acts of reference is not lexical content but grammatical perspective: whether we take the referent in view as something that happened at some point prior to the point of speech, or whether it is an object that can as such be referred to without any signification of it as an event. If we connect ontology to what we refer, an object or an event, there is therefore no partial mismatch between formal-ontological and grammatical category, indicating an imperfection, but rather a perfect match. Similarly, the fact that adjectives and nouns can be predicates entails no such mismatch, contrary to Croft’s claim that ‘both objects and properties can be predicated’. If we look at a sentence like She is a student, the fact is not that an ‘object is predicated’, which in our terms is nonsensical, but that, grammatically, there is a predicate here, like in any sentence, which, like any other sentential predicate, expresses (in ontological terms) a property (namely, being a student). As it happens, the syntactic constituent playing this grammatical functional role on this occasion can also function referentially on other occasions (as in A student came in). The same phrase can thus be both referential and predicative, depending on which grammatical relations hold between parts of speech involved. In short, if ‘nominal’ stops to mean ‘refers to an object’, which rather is a grammatical function, there is no mismatch when the nominal does not refer to an object but to a property. For in that case, its grammatical function is that of predicate, and its semantics is in accordance with that. A mismatch only arises by misidentifying the relevant notion of meaning as a lexical one, when it must be grammatical. Analogous observations apply to the alleged mismatch leading Croft to the assertion that ‘both objects and actions can function as modifiers’. The problem that forms the starting point of most attempts to define the parts of speech (Bisang, 2011), then, arguably does not exist. It arises from a blind spot with regards to how grammar itself organizes semantic information: there is an abstract semantics that goes with grammatical classifications of words and the relations in which these stand. One cannot, therefore, relegate semantics to an autonomous realm of ‘conceptual structure’ as in Jackendoff (2002), which cannot by its nature and definition capture the reformatting that concepts undergo when they become parts of speech, and when parts of speech enter grammatical relations. The result of this reformatting is what we are calling grammatical semantics here. A systematic and determinate formal ontology of reference is the most crucial aspect of this semantics, and there are no mismatches between ontology and grammar if the present perspective is adopted. Since reference and truth is of the essence of thought as viewed in the philosophical tradition for several thousands of years, our claim that thought with such properties is of its essence grammatical, begins to make sense. In the last section of this chapter we will see how the Modistic account of the parts of speech anticipates many of these conclusions and is directly woven into an account of grammar as having a crucial epistemological function that nothing else can


69

substitute: it is a mode of cognizing the world that is not available non-grammatically. This element is manifestly missing in Croft’s account of the origin of grammar as arising from the verbalization of experience, and points us to a conclusion that we argue any philosophy of grammar should move to centre-stage.

2.6 The modes of signifying reloaded According to Bursill-Hall (1971: 102), the Modists were well-aware that ‘syntax is based on the relations between words’ (our emphasis), yet they insisted—probably more than cross-linguistic data such as the above warrant—that such syntactic relationships can be expressed only by means of the properties of words viewed as parts of speech, characterized by particular modes of signifying. These modi organize meaning at the word level—they are an ens rationis simplex—whereas sentences crucially are complex entities, and therefore a theory of words yields no theory of the sentence. This is consistent with the above, as is their insight that, as Bursill-Hall (1971: 103) puts it, ‘notional meaning is not enough’: ‘if a grammatical system is to be set up in terms of meaning, it has to be done in terms of syntactic meaning, not root (or primary) meaning’ (p. 103). In other words, the parts of speech cannot be defined in terms of conceptual or lexical content. Indeed, as noted, the Modists also saw clearly that ‘semantic content’ in the contemporary philosophical sense cannot be of any interest to the grammarian. The intellect, for the Modists, is a ‘passive’ power, indeterminate in itself. In the process of mentally representing reality, language does not enter, and the same applies to the external and metaphysical properties of the res (thing) of which such an organism forms conceptions. By contrast to the passivity of the intellect, grammatical meaning—our understanding of the external res as it is expressed in speech—is an active and dynamic process. It changes the passive relations in which we stand to reality when we are causally exposed to it. Speech reflecting this mode of cognition requires a physical medium, say the voice (vox), which signifies the thing (res). But it is only when speech (dictio) is organized into parts of speech (partes orationis) by means of modes of signifying that it comes to reflect the understanding of reality in a form that is expressed in language. At this stage, words never merely signify an external content. They always also ‘con-signify’ a property of the referent, like its number or Tense. Con-signified meaning can be different when both the voice and signification (what is signified) are the same or similar, as in the examples of dolor (pain) and doleo (I feel pain) and dolenter (painful), or of currere (to run) and cursus (run), which have the same primary (notional) meaning but reflect different modes of signifying (cursus is used to refer per modum permanentis, while currere is used per modum fluxus). This is clearly not a difference in their perceptual features, and indeed the external reality (perceived or not) can be exactly the same, no matter whether we refer to it using the word dolor or

70


the word doleo. Nonetheless there is a clear difference in meaning, and this difference goes with grammar. Simplifying the Modistic theory greatly, we can say that nouns con-signify stability, permanence, and substance (modus substantiae seu entis), whereas verbs con-signify becoming (modus esse et fieri) (Bursill-Hall, 1971: 117). Other modi distinguish between subclasses of nominals, such as nouns and pronouns, or verbs and participles. The former pair share the ‘static’ mode of con-signifying, but we obtain pronouns from nouns by dropping the feature of ‘determinate understanding’ (modus apprehensionis determinatae): by referring to a thing with a pronoun, we do not so much refer to it as having some particular descriptive property (apart from the ç–features of the pronouns, in contemporary terms), but simply as an object. For some of the Modists, and Siger de Courtrai in particular (Bursill-Hall, 1971: 121), but interestingly not all of them, all parts of speech were also rigorously ordered in mereological relations, rather than merely forming an arbitrary assorted set. As a consequence they stand in relations of presupposition, with e.g. verbs presupposing nouns (see similarly Kayne, 2011), and we obtain one part from another by adding relevant features, preserving all previous ones. In terms of the phasal architecture that we will describe in the next chapter, nominals are ‘first phase’ denotations, which are presupposed as parts in ‘second phase’ denotations, i.e. the verbal phase (see p. 104 below). We take over from the Modists the insight that the meaning difference that arises when the same precategorial root appears with different categories, does not have to do with perceptual features, semantic content, or external reality as such. For, again, external reality can be exactly the same whether we use doleo or dolor, and perception cannot distinguish between these. Neither is the ‘permanence’ of the latter properly characterized as an aspect of its semantic content: how long the pain lasts makes no prediction whatsoever in regards to the question of whether the noun or the verb will be used. An explosion may last a millisecond, a minute, or an hour, but that makes no difference to whether we say The bomb exploded or refer to The explosion of the bomb. There are events (such as subliminal perception) which do not last long enough to cross the threshold of consciousness, yet when we refer to one such event, it can be permanently available as that object of nominal reference. By contrast, we can only use a verb in relation to a speech event, and when we say The bomb explodes, it is ipso facto said to do so at the very time that the speech act takes place, and in an unbounded (non-telic) fashion. Yet again this is not predicted by any inherent features of the event as such, and says nothing about how permanent or nonpermanent it is. A verb used as a grammatical predicate can depict a change of state such as blush, which is fast, or a change of state such as decay, which can be very slow (such as the decay of uranium), or indeed be a state, which never changes at all, according to some, such as God lives. None of these features, which are distinctions of conceptual content rather than grammar, make any difference whatsoever to the status of any of these words as verbs or as predicates.


71

Just as grammatical distinctions of the kind just illustrated need not make any difference to the perceptual features of the external event or be reflected in them, differences in the formal ontology of objects need not make any such difference either. We may have a stain of wine on our shirt, yet this makes no predictions about whether we will refer to it as a stain, this Bordeaux, a glass of Bordeaux, or wine. However, once we refer to it grammatically as this glass of wine on your shirt, we are necessarily bound by the mereological structure that the choice of part of speech and its internal organization induces: a glass of wine is, necessarily, also wine (or involves this substance), while this is evidently not true the other way around. By the time we have reached the parts of speech, therefore, truth does not any more reflect what causally affects us (either internally or externally). It reflects, rather, how we grammaticalize reality. If parts of speech are ordered in part-whole relations, and aspectual distinctions and analogous distinctions in the noun reflect mereological relations, then to grammaticalize reality is to mereologize it—again, not merely in thought or in our concepts, but crucially in the ways that we refer to and make claims of truth about it, using words. This is a process that is not merely ‘active’ in the Modistic sense, but creative: it doesn’t passively reflect properties that the experiential world independently has. In adding a systematic ontology that yields a mereological structure for its elements, grammar sets a perspective on what we perceive, rather than adding more perceptual features or conceptual ones. This perspective does not reflect reality but gives it a different format: the way it appears to us when we think and know it. Perspectives are not to be found in reality. They are what reality becomes when it is known. The difference between the Modistic account, which gives grammar an epistemological significance, and Croft’s functionalist account, now becomes stark. If anything, on Croft’s account, grammar destabilizes a proper functioning of language in the representation of knowledge, by letting different lexical semantic classes perform any given propositional act function, and reference and predication in particular. This, the claim was, induces mismatches, like when dog, belonging to the lexical semantic class object, functions as a modifier as in a doglike creature, which only property words are meant to do. The puzzle disappears when we credit grammar rather than the lexicon with setting the formal ontology of reference. ‘Modifier’ is a grammatical relation, not a morphological or semantic category. Where a content word such as dog functions as a modifier, we do not construe its denotation as an entity (we do not refer per modum entis), nor do we simply refer to it as a property (as we would in saying something about doglikeness). Indeed, doglike is not in a referential position at all, but adds a further predicate to a referent determined by a given head noun, a grammatical fact that the morphology marks by making it an adjective. Grammar and formal ontology thus move hand in hand, and it is descriptively incorrect to say that a word such as dog, by belonging to a particular lexical semantic

72


class, can perform the role of a predicate: it cannot, unless it becomes an adjective or verb. How, Thomas of Erfurt asks, is liber philosophiae (book of philosophy) different from liber philosophicus (philosophical book) (Covington, 2009: 37)? Why does grammar distinguish these, given that the significatum (semantic content) seems to be identical? Because, in the second case, we do not think of philosophy as a distinct and separate entity from the book, but rather as an attribute: the mode of signifying changes. This change does not track any change in external reality or content. What partially ‘corrects’ for mismatches induced by grammar for Croft, by contrast, is precisely said to be the ‘nature of reality’. Despite the fact that on his account any lexical semantic class can be used to perform any propositional act function, the nature of reality makes one lexical semantic class a better fit for each propositional act function than others. The nature of reality is that entities in the world differ as to their temporal duration, their stability over time, and their relationality. This reality leads certain types of entities to be more likely to be used for certain propositional act functions in the act of communication (Croft 1991: 123). Objects are more lasting, more stable, and non-relational, and hence are best suited for reference. Actions are transitory, involve change over time, and are relational, and hence are best suited for predication. Properties are lasting and stable, but relational, and hence are best suited for modification. ( . . . ) The facts of reality, and the way they guide speaker choices, are manifested in the typologically unmarked grammar of the prototypical nouns, verbs and adjectives. (Croft, 2007: 368)

However, as was pointed out above, objects that last only a few milliseconds make for perfect objects of reference, as do all actions when we refer to them through nominals, a decision that has nothing to do with how long they last. It seems to be precisely our grammatical ability to use nominals in referential expressions that stabilizes an object of reference, irrespective of its real-time permanence, making this object available for reference and predication over an arbitrary period of time. In turn, actions such as atomic decay or cosmic expansion are among the most stable phenomena in the universe, more stable than chairs and persons. Yet this makes them neither less nor more suited for purposes of either reference or predication. It is thus tempting to suspect that the nature of reality precisely imposes no constraints on what becomes an ‘object’ of reference and what is predicated of it. Again, this does not take the point away that all human languages, in different ways, group perceptual stimuli into equivalence classes of perceptual features, leading to a first layer of semantic structure or lexemes, prior to the point where different perspectives on aspects of reality that are so selected become possible through a grammatical system.18 But that the match between grammar and ontology becomes perfect the 18 According to Davidson (1984) and Neo-Davidsonian semanticists (Pietroski, 2005), actions are objects/individuals, and verbs are predicates of such objects. This is taken to be a lexical fact true of verbs. However, a predicate such as explodes becomes an object of reference (rather than a predicate) precisely when we refer to it with a nominal such as explosion or the exploding of the bomb. This fact does


73

moment that we accord grammar the function of reference under a perspective, and hence the generation of a formal ontology, is in itself an argument for the view that grammar has this very function. Other things being equal, an account of grammar that has this consequence is to be preferred over an account where grammar induces mismatches and its epistemological function is unclear.

2.7 Conclusions In this chapter we have traced back a grammatical mode of signifying to the origins of meaning as such in perception, where a rich and largely internally induced semantics exists long before language arrives. The same applies to computational systems in the brains of insects, birds, and mammals with which such animals perform on tasks that require the computations of values for variables in such a way that a homomorphism exists between the internal computations and the perceived structure of external reality. All of this is long before animals obtain internal address systems (roots) for their representations of environmental features, which can be used to symbolically refer to the world. From the point that they do, tapping into the ‘concepts’ of non-verbal animals confronts us with considerable methodological obstacles. However we may overcome these, a distinction between the ‘concepts’ expressed as lexemes and whatever ‘concepts’ may exist without these seems appropriate, with the possibility of an extensionality-intensionality distinction as a marked dividing line between the two. The moment that lexical semantic classes defined by perceptual features obtain a grammatical layer of structure, or grammatical words form, the organization of meaning changes again, crucially without any new perceptual features being added to their content: it would make no sense to add an ‘object’-feature to the lexeme run, when we decide to use it to refer to an object (the great North run) as opposed to an action, as in run fast. Grammar does here not add more content but a new mode of cognizing, in which a structured formal ontology emerges with the modes of reference that grammar provides us with. It yields perspectives on a reality already perceptually analysed, which are themselves not a part of this reality, but correspond to the specific grammatical way in which it is known by us. We give credit to the Modists for exactly this point. The availability of these perspectives appears to be universal, while morphologically defined classes of lexical items with which particular languages implement them,

not entail that actions (or events) are objects before we have grammatical means of referring to them as such. Indeed, nouns need not denote objects either when they are used as predicates or parts of predicates. Ontology is not fixed at a lexical level, and Neo-Davidsonianism mixes up grammatical functions (including those encoded in the parts of speech) and lexical semantic classes.

74


do not appear to be. In this sense, we cannot build a theory of grammar on a theory of the parts of speech, as the Modists proposed. We have to do it the other way around: building a theory of the parts of speech, where they exist in language, from a theory of grammar and the relational structures it provides within its configurations. We thus devote the next chapter to an account of the relations in which such parts stand when they occur in a sentence, with a view to reconstructing the recursive nature of language on this novel basis.

3 The content of grammar 3.1 Grammar as relational The previous chapter dealt with the imposition of a first grammatical layer on words, which thereby become ‘parts of speech’. The emergence of part of speech distinctions cannot be the key to grammar, however. An adjective (e.g. ‘male’) may function as a predicate (‘[is] male’), but need not do so (‘the male and the female’). A verb (e.g. ‘fly’) can be a predicate (‘flies’), but need not be (‘Flying is beautiful’). A nominal like ‘a man’ can be used to refer to an object, but can be a predicate (‘[is] a man’), resulting in a truth value when applied to a subject, much as a predicate realized through an adjective or verb. Predication, therefore, is a grammatical category independent of the parts of speech, though it can be realized by them. A verb is also statistically more often a predicate than a nominal, but this does not show that the notion of ‘predicate’ can be defined as a ‘verb’, or as ‘either an adjective or a verb’, or as ‘either an adjective or a verb or sometimes a noun’. On the contrary, understanding what an adjective or verb is will in part involve understanding what grammatical roles they play in relation to other parts of speech. ‘Predicate’, we argue, is a term indicating a grammatical relation or function, and it cannot be accounted for by any morpholexical distinctions. We stress that, in what follows, we will use the term ‘predication’ in a purely grammatical sense, not a broader logical one: namely, to denote the sentential predicate that is applied to a sentential subject, as ‘flies’ is applied to ‘Socrates’ in ‘Socrates flies’, yielding a truth value as a result. In this sense of predication, a sentence only ever has one predicate (which again can be realized through different parts of speech, with language-specific differences in which parts of speech can play which roles), exactly as it only ever has one truth value. In the phrase kill Bill, the verb ‘kill’ is also predicated of the internal argument ‘Bill’, but the result is crucially not a truth value: a verb phrase is not evaluated for truth, in the way a sentence is. Predication, in this sense of a grammatical relation between the subject and predicate of a sentence, is not a part of speech: it relates different parts of speech playing distinct grammatical roles, and hence is not present in either of them. Predication in this sense is not found in morphology, let alone the lexicon. In

76


morphology, as Di Sciullo (2005) points out, a complex word can, say, encode thematic structure (bank-robber, shop-lifter, etc.), but neither predication, nor quantification, reference, and truth can be found within a single word, encoded in its internal structure. Relatedly, a sentence is not a part of speech, although clauses can be arguments, a different matter, and although the temptation has arisen to turn the sentence into a part of speech, say an ‘inflectional’ or ‘extended’ projection of the verb. Of course, predication is not a lexical concept either. There is a word ‘predication’, which we can use to refer to the topic under discussion, namely predication: but this topic is the grammatical relation of predication, not a lexical concept. This chapter, then, is about the transition from parts of speech to grammar proper, which is a far better candidate for a genuine invariance in the species than morphology-based distinctions. But what is grammar, so that it can be invariant? To answer this question, we will ask about its content. What is the ‘content of grammar’, as distinct from the content encoded in lexical concepts or parts of speech distinctions? To explain this question, consider how one might answer it after looking at a contemporary textbook on Minimalist syntax. The content of grammar, one would then conclude, consists of a ‘computational system’ made up of the following: A. The local relation of (recursive) Merge, where Merge(A,B) = {A,B} and A can be contained in B or vice versa. B. The long-distance relation Agree. C. Constraints on computation. D. Interface constraints. (A) gives an arguably minimal account of the combinatoriality of language. (B) accounts for the fact that elements in a grammar can enter long-distance relations with other such elements elsewhere in a given structure. (C) prominently includes principles of computational efficiency that are argued in Minimalism to be expected to be operative in any well-designed computational system in nature, and to be indeed factually operative in language (Chomsky, 1995). (D) encodes the assumption that what computations take place is constrained by demands imposed by interfacing performance systems in the mind that access the computational system. As we would conclude from this list, the content of grammar is purely formal. Merge in particular, as noted, is a generic operation that will not tell us anything about what is specific to language. Language-specific content will thus rather come from what Merge applies to or from what enters Agree relations: lexical features, said to make up lexical items, which are conceived as sets of such features in standard Minimalism. These features are morphologically visible in language-specific ways, for example in the form of Case and ç-features (Person, Number, Gender), which are taken to play a crucial role in driving the grammatical derivation forward in this framework. But some of these features, too, in some or all of their occurrences, are

3. The content of grammar

77

commonly taken to be ‘uninterpretable’, and grammar is said to be designed so as to ‘check’ and eliminate them through Agree and Merge, before the semantic component is accessed which is said not to be able to ‘read’ them. Constraints on computation (C), finally, are commonly regarded as ‘third factor’ conditions: constraints not due to either experience or UG (Chomsky, 2005). Hence they aren’t language-specific and certainly not semantic in nature either. In short, interpretable content and language-specific information comes either from the lexicon or the interfaces—it is external to grammar, and grammar is completely content-free. Feature-specifications in the lexicon, however, do not seem the right place to account for grammaticality. Moreover, a theory is lacking of which ‘features’ the lexicon contains or fails to contain. Against the above picture, therefore, and consistent with Boeckx’s (2010; 2012) critique of the role of features in Minimalist syntax, we will argue in this chapter against a lexico-centric focus in the theory of grammar, and the implicit contention in Minimalism that grammar is not—or not directly—involved in the generation of content. Note that if semantics is distinguished from syntax as ‘content’ is from ‘form’, we may expect that the programme of a universal grammar will become harder to formulate, not easier. This is because semantics is commonly regarded—in typology as well as (and essentially uncontroversially) in the philosophy of language, where language-specific variation in general is barely discussed—as the invariant part of the organization of human language. So if semantics is independent of syntax in the way that content is of form, then we have no particular reason to expect invariance in syntax: for syntax will now be arbitrary and ungrounded in semantics. Why should such an arbitrary system be invariant? It would seem like an unmotivated evolutionary accident (see also Christiansen and Chater, 2008). We could maximize the explanatory force of third factor conditions, so as to make the syntactic system less arbitrary, but this, as noted, will not lead to insight specific to language, and will divorce us from the functions of language. The field of comparative syntax has sought to model variation in syntax through a system of (arbitrary) ‘Principles and Parameters’, but as we will discuss in Chapter 5, this programme faces significant challenges today (Newmeyer, 2005; Boeckx, 2011), which has resulted in the suggestion to shift the locus of variation solely to the lexicon and morphophonology (Berwick and Chomsky, 2011). Yet, the claim that the essence of universal syntax is recursion or hierarchical organization (or A and perhaps C above) is too weak to be an inroad into the language-specific content of grammar; and it also leaves the principles of semantic content unaddressed, giving grammar no intrinsic semantic rationale and semantics no intrinsic grammatical rationale. A foundational—even if initially maybe unsurprising—claim in what follows will be that the content of grammar is essentially relational. If grammar is relational, its content cannot be accounted for by any content that is intrinsic to any word or ‘concept’. Indeed, we stress throughout this book that how words function is—apart

78


from their substantive lexical content or the concept they encode—itself a matter of what grammatical relations they enter into: these change how a given word signifies. Grammar is about relations between words, and about what novel content arises from this: this is grammatical meaning, in our sense. We claim that this content cannot be reconstructed from the contents of words. It does not reside in ‘features’, e.g. ‘Person’, or ‘ACC’ (for Accusative) either. Instead, theoretical descriptive terms of traditional grammar such as ‘Person’ or ‘Case’ refer to grammatical relations or dependencies into which nominal arguments enter. These hold in the case of Person between determiner phrases (DPs) and T/C in the left periphery of the clause, and they show up on the surface of languages as a morphological feature marked on (pro-)nouns, verbs, both, or neither. Similarly, we will argue in Chapter 6 that ‘ACC’ is a morphological feature that, where it exists in a language, encodes syntactic dependencies between v and an argument DP. The dependencies are part of grammar; the feature itself is not. Grammar appears to contain no uninterpretable features at all (cf. also Sigurðsson, 2009; 2012; Kayne, 2008). Consider an example. It makes no difference to the interpretation of German (1) that the 1st Person singular is also marked on the verb, which it isn’t in its English translation (‘I go’): (1)

Ich I

geh-e go-1p

Hence the Person feature on the verb might be said to be ‘uninterpreted’. However, if grammar is relational and what is interpreted is not words or features but relations, it makes perfect sense for there to be a relation of ‘Person’ expressing that a given referent (a person) carrying a certain theta-role is the subject of a given sentential predicate, resulting in a proposition involving an event, whose only participant is also the speaker in the speech event, and hence the one asserting the proposition that this event obtains. This (highly complex) relation holding between the DP in subject position and the verbal predicate then shows up in the form of morphological features on the subject, the predicate, both, or neither, in language-specific ways that make no difference to the interpretation just depicted. While there can be no question that Person, as a relation, is interpretable, the same question, however, arises for other syntactic dependencies such as Case-relations, where, as we shall see in Chapter 6, the answer is more complex. A long tradition resists the contention that the content of grammar is relational and that it cannot be accounted for in terms of the contents of words. For example, if the relations in question are denoted by (or the ‘referents’ of ) words, as has been widely assumed since Russell (1903), all content can be reconstructed lexically, and no content needs to be grammatical—if there are relational aspects of grammar that matter to meaning, there are lexical items to denote them. They exhibit these relational aspects as part of their meaning. For example, instead of saying that


79

‘being a predicate of a subject’ is a grammatical relation, we could say that this is an intrinsic property of certain lexical items, say verbs, or even certain ‘concepts’, say blue. We think this is a mistake and will rather follow Frege in this regard, who, as we will argue, did importantly not hold such a lexicalist view. He rightly saw that terms denoting the grammatical functioning of a given word, and associated formal-ontological distinctions in their denotation, co-vary with grammar and its intrinsic modes of signification.1 In particular, the distinction between ‘object’ and ‘concept’ is, at heart, a grammatical one—part of the intrinsic content of grammar—not a metaphysical one. It arises on occasions of language use, where acts of reference are configured. We develop this claim in Section 3.2, which contains our basic case for grammar having a content of its own. Even Frege himself, however, sticks to a word-based conception of semantics, as his puzzlement about so-called identity sentences (and intensionality more generally) illustrates. Thus, (2) is supposed to have the logical form ‘a = b’, where ‘a’ and ‘b’ are names, whose referent is an object: (2)

Hesperus is Phosphorus.

If the content of such a sentence is ‘composed’ of the meanings of its parts (ultimately, words), then of course the content should be the same as that of ‘Hesperus is Hesperus’—which it intuitively isn’t. Hence the puzzle that Frege’s theory of ‘senses’—which are equally word-based as well as non-linguistic entities—was meant to solve. But the puzzle disappears if what a word contributes to the meaning of a sentence is its grammatical meaning as well, not merely its lexical meaning: in particular, whether it functions as a predicate or subject, properties that make the relation of the two nominals in a clause always and necessarily asymmetric (see Moro, 1997; and Woodard, in preparation). If the content of grammar cannot be reconstructed lexically, or by somehow putting together (‘composing’) the intrinsic contents that the lexical items have that it contains, the standard claim that meaning in language is ‘compositional’ has to be qualified, and we turn to this issue in Section 3.3. In essence, the principle of compositionality as standardly viewed deprives grammar of a content again, at the cost of having to misinterpret it as lexical: for compositional meaning is ultimately solely word-based again. But neither reference, predication, nor intensionality belong to either lexical or compositional semantics. Section 3.4 revisits another claim, that the content of grammar could come from an ‘interface’ with non-linguistic systems of thought, which of course we are committed to reject here: such systems could never interpret the outputs of grammar, if grammar has a content of its own. If

1

In this point we are indebted to Uli Reichard.

80


grammar is productive of a new kind of meaning—reference in its various forms, we contend—then the architecture assuming an ‘interface’ is not correct. The crucial grammatical relations transpire within grammatical templates that we will here identify with the phases of contemporary Minimalist syntax, reinterpreted in Section 3.5 as the smallest units of grammatical organization and as units of (intentional) reference, which can take a number of different forms in human language, ordered on a scale. The root meaning (and earliest developmental manifestation) of this notion of reference we take to be (e.g. index finger) pointing, which is deictic. By vastly expanding the range of possible reference beyond where our index finger can point, grammar becomes a device for ‘extended’ deixis. Its organization reflects this function. This notion of deictic reference has to be carefully distinguished from a notion of (i) lexical reference or content, which content words and concepts have intrinsically to their lexical specifications and without mediation by grammar; and (ii) functional reference that we find in other species. Crucially, the invariance of such a system of (extended) deictic reference in the human species is not in doubt, and hence it is a candidate for grounding the invariance that grammar represents. Finally, this relational reconstruction of grammar is, in Sections 3.5 and 3.6, the basis of reconstructing the recursivity of grammar, which is a derived notion rather than a primitive here (Arsenijević and Hinzen, 2012). This reconstruction will explain why grammatical recursion cross-linguistically takes on a specific form and is subject to restrictions, which we argue current Merge-based accounts of ‘narrow syntax’, which leave the matter to the ‘interfaces’, neither explain nor predict. The restrictions also do not seem to be due to lexical specifications of the lexical items merged. We suggest that they need to follow from the organization of the grammar itself, or its intrinsic content, and that they do indeed fall out freely from the phasal dynamics as interpreted here.

3.2 From parts of speech to the sentence The sentence, as structured by a predication, is traditionally the core unit of grammar. But it has remained unclear how to explain the sentence for many decades. The traditional Aristotelian notion of the sentence conceives it as consisting of a subject, a predicate, and a sign of Tense. But generative grammar has resisted regarding putatively ‘semantic’ notions such as subject or predicate as theoretical primitives of grammatical description, attempting to reduce them to formal notions, and the notion of sentence has not fared better. This is manifest in how early generative grammar in the era of Phrase Structure Grammar (PSG) registers the puzzle that the sentence causes in a rule such as (3), where the category ‘S’, unlike the ‘lexical’ categories N or V, stands in no inherent relation to the categories of its parts. Neither the noun nor the verb ‘projects’:

3. The content of grammar (3)

81

S!NP VP

While stipulative, this rule brings out the fact that a sentence is not a part of speech, or projected from the lexicon. This same insight is later brought out in Moro’s (1997; 2000) configurational analysis of predication, which maintains that ‘all and only bare small clauses instantiate predicative linking’ (Moro, 2000: 71), where the form of such a Small Clause is as in (4), where X and Y are categories and XP and YP are their respective maximal projections, none of which projects further or lends its label to the structure as a whole: (4)

[SC XP YP]

If predication is not lexical, it should have no label within PSG. On this view, no predicative linking exists where a head takes a complement (5), or a head combines with another head (6)2: (5)

[X YP]

(6)

[X Y]

Moro’s hypothesis is highly restrictive: it gives us a specific grammatical correlate of predication, before one element of the Small Clause moves out of it and a sentencelevel predication is generated. If Moro is right, we may therefore conclude that predication does indeed have a grammatical signature to it.3 We might as well conclude that predication is the relation that is realized in the [XP YP] configuration, since no other system appears to exhibit a similarly specific signature for this relation. Logic, in particular, presupposes a classification of primitive symbols into terms and predicates in its lexicon. Thus, in a logic class on first order predicate logic, a student might be given a sentence like ‘Socrates flies’. He is then told to look for the argument and the predicate. The sentence is viewed as containing two lexical items. As long as proper names are regarded as referential, the choice is clear: ‘Socrates’ is the argument and ‘flies’ is the predicate. So the ‘logical form’ is flies (Socrates). The lexical entries for these two words tell us that Socrates is mapped to an object, flies to a set or property. This approach not only takes predication for granted, but it misinterprets it as lexical: as the denotation of a designated set of lexical items. This is the logical version of the lexicalist misunderstanding of parts of speech distinctions: ‘predicate’ does not mean the same as ‘verb’. No word is a predicate lexically; nor is any part of speech.

2 A form of combination that is the grammatical origin of the noun–verb distinction, hence the parts of speech, according to Kayne (2011). 3 Whether Moro’s analysis is ultimately successful in reducing the notion of predication configurationally is another question, discussed in Rosselló (2008), who however agrees with the need for a configurational analysis.

82


Moro’s analysis brings out this latter point formally as well: being the predicate in a predicational relation is not an intrinsic property of YP. For YP as such also occurs in, say, (5), where it functions as an argument. YP thus is the predicate in question only relationally, relative to a subject. The same then applies to the notion of subject. Nothing is a subject because of its phrasal status, say as a DP, or its lexical status, say as a nominal, or its semantic status as being ‘about some object’. Thus, a subject need not be a DP, it need not be a nominal, and it need not pick out some object: rather, it is a grammatical notion. The same is true again for the logical opposite of predicativity, namely referentiality. Referentiality is not a property determined by a phrase in isolation, let alone a lexical item in isolation. ‘Man’ does not refer to any particular man; ‘the man’ can. The former is a lexical item; the latter is not. Even a definite description need not be singularly referential. ‘The bus’ can be used to refer to a specific bus (‘the bus is here’), but it need not be (‘I always take the bus after work’). The point is reinforced once more, and in a more dramatic way, if we realize that the same phrase can play different grammatical roles at different stages of the derivation. Thus, a DP like ‘this man’, looked at in isolation, certainly seems to be both referential and a paradigmatic argument. But as an argument it is nonetheless interpreted as a thematic role, which is technically a predicate. Similarly, a CP like ‘this is the man’, looked at non-relationally, is referential in the sense of denoting a truth value when asserted. But embedded as an argument, it will act as a predicate applying to a mental event, as in ‘believes [this is the man]’, and it will not be interpreted as a truth value. The combination of a subject with a predicate resulting in a predication with a truth value moreover does not have the nature of either a subject or a predicate, any more than it is a noun or a verb. This is also what the traditional semantic intuition suggests that a YP (e.g. ‘loves Mary’) predicated of an XP (‘John’) denotes neither what X denotes nor what Y denotes, but a truth, which is neither X nor Y and formalontologically distinct from either of them. A grammatical relation, then, again, cannot be captured lexically. A predication, rather, has no conceptual-lexical content at all, and it is no lexical category, in the sense of reflecting processes of categorization of environmental features. If so, we would not expect predication to be illuminated by ‘mapping’ such a configuration to any kind of external and non-relational and non-grammatical object, such as some ‘relation of predication’ (perhaps a set-theoretical construct) or a metaphysical object, such as a state of affairs. In fact, the search for objectual or conceptual correlates for the structural relation in question—for something out there in the world that corresponds to subjects, predicates, grammatical Tense, and syntactic configurations—will seem misguided. Such an attempt can be found in Russell’s (1903) attempt to map ontology from the parts of speech rather than grammar: in this framework, ‘substantives’ denote objects, adjectives denote properties, and verbs denote relations. For example, in


83

‘A is different [differs] from B’, the verb denotes the relation of ‘difference’ (Russell, 1903, }54). The verb, Russell suggests, accounts for the character of the sentence as truth-evaluable. This leads Russell to consider (and leave unresolved) the immediate puzzle of why, in a list consisting of A, the relation of difference, and B, no truth value arises, and of why, in a nominalization such as ‘A’s being different from B’, a verb should be present and yet evaluability for truth is absent. If predication and truth evaluability is due to a lexical fact, it should be present when the relevant lexical item is present. This puzzle illustrates the problems of a lexicalist and intrinsicalist approach, which regards grammatical function as intrinsic to lexical items or phrases. Such an approach, which is maintained throughout much of the history of analytic philosophy and in formal compositional semantics, is also illustrated in Davidson’s (1967) suggestion to replace the Russellian ‘verb-as-relation’ view with the ‘verb-asobject’ view, where the object in question is said to be an ‘event’. We come back to semantic compositionality more systematically in the next subsection. As noted, when X-bar theory replaced PSG, it ‘corrected’ the ‘defect’ noted in (3) by making sentences come out as projections of a lexical inflectional feature after all, namely Inflection/Tense (I/T), with subjects as specifiers of the node that these head. But as we would expect from the above discussion, the X-bar theoretic conception of the sentence as a projection of I/T has never seemed particularly illuminating, inviting the suspicion of some contemporary thinkers that perhaps grammatical theory should not regard the sentence as an explanandum at all. Pietroski (2011), in particular, on the basis of a Neo-Davidsonian semantic motivation, suggests that the sentence is not the central organizing principle of grammar, and that notions such as singular reference and truth should be relegated to post-syntactic components. Recent Merge-based syntax has even less to say about the notion of sentence than PSG or X-bar theory, continuing the traditional tendency and bringing it to a peak: if we deplete the structure-building algorithm of any category-specific elements, reducing it to recursion, what PSG left out can be accounted for even less. The insight from (3) and from Moro that we want to maintain is that predication is not a relation between a lexical category label and itself: it is not the result of projection. It is, rather, a relation between two lexically specified and functionally fully expanded/projected categories, the combination of which, in the right configuration, takes us somewhere else: the grammatical. The question of the content of grammar now arises again: What cognitive transition does it mark? With grammar at our disposal, we now can, as Frege famously noted, refer to a given concept (as in ‘the concept horse’), in which case it will be an object (of reference), and it will then be correct to say, as Frege famously did, that ‘the concept horse is not a concept’. But we can also use it as a predicate (as in ‘Ellie is a horse’ or ‘every horse’), in which case it will be a function, in Frege’s terminology. So is ‘horse’ a function or an object? Is ‘Vienna’ an object (a city)? It certainly can be, as in ‘Vienna is a city’, but it need not, as in ‘Trieste is no Vienna’, where it, in ontological parlance, ‘denotes a property’ (the property of being Vienna). So does ‘Vienna’ denote a property, then, as on the

84


‘predicate’ view of proper names (Burge, 1973)? Does a verb denote a relation, as on Russell’s view, or an object, as on Davidson’s? These questions, we argue, are meaningless, since they involve a category mistake: questions of denotation are answered by an analysis of grammatical function on an occasion of language use, not lexical content. Grammar enables the forms of reference in question, depending on what structures it generates. A percept lexicalized as a lexeme makes as such no predictions about how it will be grammaticalized—whether it becomes a noun or a verb, a predicate or an argument, etc. Imagine a creature that perceives and can form equivalence classes of perceptual inputs, applying them adaptively. One such class—let us call it water—will be applied to things that we would call water when it perceives this substance, and in these moments it will be connected causally to H2O, as a chemical analysis reveals. With another such concept, horse, it will be connected causally to a certain natural kind, a species. But no such creature need thereby be able to refer to something as a substance as opposed to an object, say, as we can when we use the bare noun water as opposed to the phrase a drop of water; or to refer to an object as opposed to a property, as when using a proper name such as ‘Ellie’ or a referential DP such as ‘this horse’, as opposed to using a verb phrase predicatively, such as ‘BE a horse’; or to put the two together to obtain a judgement with a truth value (‘Ellie is a horse’), which could be asserted or denied and will reflect the way the world is, if asserted and indeed true. Unlike the question of whether a particular substance we call ‘water’ is indeed H2O, or whether H2O exists, the availability of a system of such formalontological distinctions is not a question of material existence or empirical inquiry: we cannot investigate the non-linguistic material world so as to find out. The question of ‘whether there are properties’ out there, along with tables, water, and horses, cannot be decided by empirical inquiry. Grammar, in inducing such distinctions, is foundational for experience, not itself given in experience. With the notion of truth, a notion of ‘world’ arrives, and the same considerations apply here. The world, as Wittgenstein (1922) put it in the opening proposition of the Tractatus, ‘ist alles, was der Fall ist’ (all that is the case). What is the case is what is true. But truth requires predication (in the present sense), and thus the sentence (viewed as a grammatical object). It follows logically that there is no ‘world’ without the sentence (though, obviously, there always is an ‘environment’, in which the animal behaves adaptively in its niche). Whatever a chimpanzee might think, he doesn’t think about ‘the world’ as such, as philosophizing hominins tend to do. The arrival of the sentence, for these reasons, or the transition from a perceptually and conceptually given world to one that is grammaticalized, is a truly momentous one. Only in such a world can there be such a thing as a metaphysics, or science. Nothing we have got so far accounts for this transition. Predication has no conceptual content; there is nothing to visualize it; and it is not a part of speech. We could say that it is ‘purely structural’, but then again it is not, for it is not purely structural, in the way that a fractal geometric pattern is, say. And it is crucial to the


85

approach above that any predication has content and is unthinkable without it: a predication at the sentential level gives rise to a unit of structure that can necessarily—or ipso facto—be evaluated for truth. A person uttering: ‘John left’, with normal intonation, would be contradicted by someone else replying: ‘That is not true’. Hence he did assert something as true, even if no lexical concept of truth was used (a point to which we return in Chapter 9).4 But this means that it will in some sense co-vary with the world: if the world is as the sentence says, then the sentence is ipso facto true. Predication as a grammatical relation therefore comes with content, and indeed with the most paradigmatic form of content that there is: truth conditional content. Ultimately, any account of predication will have to respect this double constraint: it is structural and relational, and hence not a part of speech or a concept, but it is contentful, and indeed it has a content that no part of speech has, and no lexical item. One could, of course, simply abstract from the intrinsic content of grammar—the whole and vast expansion of the space of meaning that it comes with—and restrict oneself to a ‘formal’ analysis of it. One could talk of grammar solely, say, in terms of Merge, recursion, uninterpretable features, their elimination under Agree, and locality domains. In a similar way, one could analyse a cooking recipe purely formally in terms of a computation of symbols embedding others in a hierarchy of steps, perhaps leaving the connection of this analysis to the actual recipe and the ingredients to an ‘interface’. Foundational problems in our understanding of semantics may well make such a formal treatment of grammar advisable, and they did so in the 1960s, as Chomsky (1965) convincingly argued. The increasing importance, however, of the C-I interface in Minimalist grammar leads to more probing questions about its nature, and the search for a more ‘principled’ explanation of grammar in the sense of Chomsky (2004) leads to the re-emergence of this issue. And as Chomsky (1965) suggested, the formal treatment of grammar at the expense of semantics was never meant to be principled or axiomatic: instead, the ‘boundary separating syntax and semantics (if there is one)’ should ‘remain open until these fields are much better understood’ (Chomsky, 1965: 77, 159). Ultimately, the formal strategy will only work if the ‘content’ that we locate on the ‘other’ (non-linguistic) side of the interface in question does not itself depend on the ‘form’ that we posit on the linguistic side. It cannot work, if the form 4 A sentence fragment lacking the relation of predication will not be truth bearing in the same sense. A large literature on ‘fragments’ maintains that even sub-sentential units can be evaluated for truth, either because they contain elided material (Merchant, 2004) or because such material is unneeded (Stainton, 2006). Hinzen (2013a) argues that both views can only be maintained by ignoring semantic differences between the fragment and the grammatical structure that would be truth-evaluable. For example, the content of ‘Nice shirt!’ is not that of ‘This is a nice shirt’ or ‘You are wearing a nice shirt’, which are both grammatically distinct from the fragment in question, and differ in meaning. In particular, they can both be embedded, while the fragment cannot be; and the fragment carries affect, which its propositional renderings do not convey. Of course, one can develop notions of meaning that abstract from such differences.

86


of language is the form of a content that does not exist without language—as of course we argue here. The opposite has been the traditional assumption, where semantic content has been taken to be independent of linguistic form—the very rationale for adding a ‘semantic component’ to grammar that is in charge of content and generates it via non-grammatical operations such as functional application or predicate modification (see e.g. Heim and Kratzer, 1998; see also Jackendoff, 2002). Historically, the assumption has been so influential that it is by now firmly institutionalized, with philosophers solely focused on ‘content’ to the near-exclusion of syntax (‘form’), while syntacticians are involved with ‘pure form’, leaving semantics—matters of truth and reference—to the philosophers and logicians.5 Accounting for the content that grammar has in a separate ‘semantic component’ seems to be impossible, however, if this content is relational. Suppose we deny this, and we postulate an ontology of ‘properties’ qua entities ‘denoted’ by the relational or predicative part of a sentence or thought. If there were such entities, then expressions functioning predicatively become ‘names’ of such objects, eliminating the relational character of predication in the sense above and seemingly solving our problem—all content is now intrinsic to the lexical items involved, nothing is relational, and insofar as meanings are complex and propositional, what puts lexical meanings together is nothing in the grammar but composition operations in our semantic component. But, of course, the problem is then not solved, for if one now asks what these new ‘objects’—properties—are, the answer is that they are in every respect like objects, except in that they do not behave like objects precisely insofar as they behave as predicates that apply to objects.6 We would then have come full circle: the existence of predicative units in linguistic expressions led us to postulate a new entity, a property, but in answering the question what kind of ‘object’ this is, the best we could do is to point to the relation of predication (or a relation such as ‘instantiation’). Not only is this circular, but the theoretical advance is purchased at the cost of a well-known paradox known since antiquity: If a predicate denotes some kind of object as well, what connects the subject of the predication to this new object? A new relation is now required, creating an infinite regress and confirming our initial impression that the objectification of relations was a questionable move to begin with. This is to question a non-linguistic ontology of properties, but this doubt should not be equated with a rejection of ‘realism’ in regards to properties (or traditional socalled ‘universals’). The denial of the existence of properties as referents of predicates Thus treatises in philosophy on the ‘nature and structure of content’ abound (e.g. Fodor, 1990; King, 2007), but treatises on ‘form’ are completely absent, illustrating the view that content can or must be treated independently of form. 6 The same applies to the attempt to specify the objects in question by saying that one of the objects is ‘incomplete’ or ‘unsaturated’ or needs to be ‘applied’ to the other; this is to pinpoint the problem more than a solution (see further Davidson, 2005). 5


87

is not analogous to the rejection of universals or to be equated with ‘nominalism’ (which nothing in the present book will support). It is not nominalism to assert, as we just have, that stipulating an ontology of properties is no advance over understanding predication. On the contrary: it points to the need to understand grammar in substantive as opposed to merely formal terms, and to the insight that something other than grammar, such as ontology (if regarded as independent of grammar) or semantics will not solve the problem of predication. Grammar, in short, may itself be foundational, rather than being founded upon something else. The standard view, of course, is that the most foundational of disciplines are metaphysics and ontology. Grammar, if motivated at all, would thus only be motivated if founded upon independently given distinctions in these other disciplines. Linguistics also reflects these priorities, for example in its attempt to found part of speech distinctions, which we have argued are inherently grammatical, on independent semantic distinctions such as object vs event; or in the rationalist contention that linguistic universals, if there are any, must be based on semantics or the universal structure of thought (Arnauld and Lancelot, 1662; Haspelmath, 2007); or in standard classifications of inherently grammatical and relational elements such as structural as opposed to ‘inherent’ (= lexical) Case as ‘uninterpretable’ and as to be ‘eliminated’ before the interface with the semantic component is reached. Grammar as founded upon Merge and the need to eliminate such ‘features’ has no inherent motivation or foundational significance: it solves a problem it has created for itself, through the existence of purely formal features that need to be eliminated. On this picture the lexicon is inherently motivated, insofar as it interfaces with the semantic component directly. Grammatical representations, on the other hand, have to be ‘cleansed’ before such an interface can be accessed (the problem of grammar, as it were, is grammar itself ). This architecture makes sense on the assumption of an autonomous syntax and the independence of thought (C-I systems) from grammar. But it becomes unclear, on this view, why there should be a lexicon–grammar distinction in the first place—or why we ever leave the realm of the interpretable, and enter the realm of the grammatical. This, of course, might in theory simply be an accident. But on the Un-Cartesian view, grammar yields a formal semantics, and can be given a principled rationale in these terms, explaining why it exists and is distinct from the lexicon. For this to work out, grammar must have a content of its own, as we argue here it does. The difference that it makes in relation to the organization of meaning, if there is one, should clearly relate to truth and propositional forms of reference, since these fall on the grammatical side. This possibility raises interesting and novel foundational questions. Thus, predication does not seem to have any physical correlates at all. On the linguistic surface, we can see words functioning as predicates, but we cannot see predication, which is purely relational. We can observe the world in its physical structure, but we will not find

88


predication in it, any more than we will find truth, in the sense of an empirical given or an object of nature. Although the joining of a subject and a predicate presumably happens at particular points in time and space when human agents use language, and a brain can clearly perform a predication that involves a grasp of the notion of a truth value, no particular part of brain space or brain time is a predication. In line with this, the form of the sentence does not pinpoint any physical correlate of predication. Instead, it presents us with a pattern that is grammatical rather than physical in nature. But does this pattern really illuminate how, out of it, a truth value emerges? In one sense it doesn’t, except for indicating that the phenomenon does seem to depend on a certain grammatical, i.e. a sentential, configuration, giving it a grammatical signature. But then, is there any other discipline that takes us further? The more we think the phenomenon is grammatical, grammar may well be both the beginning and the end of an answer to the question of the origin and nature of truth, hence be foundational in a novel fashion. Where a metaphysics arises with grammar, there is no extragrammatical reality against which we can assess the accuracy of claims about grammar, any more than we can test mathematics through physics. We couldn’t meaningfully ask, for example, whether relations exist in the world out there (or else in an abstract, non-physical world). Where relations are grammatical, nothing but grammar will be the right place to look for them. There will be no independent metaphysics in which we can ground grammar. We develop this view, that predication and hence truth are inherently grammatical, further in the next section, by contrasting it with the approach of semantic compositionality.

3.3 Content from composition? The basic formula of modern compositional semantics might be put as (7): (7)

Lexical meaning + syntax = propositional meaning.

But doesn’t this express the very message above, or indeed of this book at large, namely that it is grammar that mediates between the lexicon and propositional meaning? If so, the framework for addressing these issues is already in place and we need to look no further. Addressing this objection will help to bring the claim of this chapter into better focus. The problem, we will now argue, is that how compositional semantics implements the suggestion formulated in (7) recapitulates the original problem, by misidentifying the locus of its solution. What syntax does on this conception is to assign a part–whole (constituent) structure to an expression. In contemporary syntax, this structure is modelled by Merge, hence through a binary set potentially containing other such sets. Take the sentence (8) and its (vastly simplified) formal generation in (9):


John swims

(9)

Merge (John, swims) = {John, swims}

89

There is nothing in the set {John, swims} that makes it in any way intelligible why the meaning of (8) arises from the independent meanings assigned to ‘John’ and ‘swims’—except that we are now told that the meaning of ‘swims’ is such that when combined with the meaning of ‘John’ it yields the meaning of (8). This is not false but it is the problem we started with, and compositional semantics is in this sense explanatorily vacuous. There is not necessarily an ontology of ‘properties’ now, of course, but only of ‘semantic values’, but this does not change the basic explanatory problem. If no independent meaning or ontology of the parts is assumed, on the other hand, and their meanings are taken to be abstracted from the meaning of (8) taken as a whole, the explanatory problem becomes even more obvious: now we are deprived, in principle, of an explanation of how the meaning of (8) is strictly a function of its independently meaningful parts (Collins, 2011: ch. 3). Either way we run into the foundational problem noted above: the distinction between an object and a Fregean ‘concept’, or an object and a property, we argued, is not an ontological or semantic but a grammatical and relational one. It depends on how a concept or lexical item grammatically functions, which no concept or word as such can predict. We have to wait for grammar to rule the matter one way or the other. There is therefore no semantic value that ‘horse’ as such has and that we could insert into the compositional computation of grammatical content. And when it is decided which of these things the word does stand for, this cannot be explained non-relationally by going back to ontology or semantics and positing some ontological or semantic entity that it stands for, in order to then explain what meaning results. This explanatory problem, which compositional semantics faces when added to a formal syntax as an independent component, is general. Zimmermann (2012) gives the following general account of the practice of compositionality in linguistic semantics. Given a linguistic expression C, its syntactic construction type ˜ is identified, of which two immediate constituents A and B are parts (possibly after changing C’s surface appearance into what is taken to be its ‘logical form’). With that construction type a function ˆ is then uniquely associated that assigns the semantic value of C to the pair consisting of the semantic values of its immediate parts. One of these values will usually be regarded as unproblematic—as when ‘John’ is simply assigned John as its semantic value, an object (a problem, if object-reference is grammatically configured: see Chapter 4). The other value is then defined to be the mapping of this value to the value of the whole. In the case of (8), ˆ will assign a meaning to ‘swims’ defined to be the very meaning that maps John to the meaning of (8). But if this function is viewed as a set, say—the set of those objects that are mapped by the function to True if they swim—the definition of the set requires an

90


understanding of the very predicate in question, depriving the move of independent explanatory value (cf. Davidson, 2005). It is true, then, on both the present and the standard compositional view, that grammar mediates between lexicon/concepts and propositional meaning. But as compositional semantics spells out this intuition, the idea becomes dependent on a notion of ‘syntax’ that precisely deprives it of playing a constitutive role in the genesis of propositional meaning, with the consequence that the explication of how propositional meanings are generated compositionally is explanatorily vacuous, and formalontological distinctions are misidentified as lexical-semantic ones. Consider, in this regard, again the semantic difference between (10) and (11). As described in Chapter 2, taking inspiration from the Modists, this is a difference in the perspective with which we refer to an object, not in what it is that we refer to (the external significatum): (10)

John laughs

(11)

John’s laugh

The difference between (10) and (11) is not lexical, in the sense that it cannot be explained by what content the lexeme laugh contributes to the expression, since that root is identically present in both. The function of the lexicon, on the account of Chapter 2, is to collect perceptual features of environmental stimuli into equivalence classes: but in terms of perception, nothing needs to change when we move between using (10) and (11) on a given occasion of language use. So the difference between them is not lexical. One could always make it lexical, by ‘composing’ the lexeme laugh with an added lexical ‘action-feature’ in (10) and an ‘object-feature’ to (11). But this seems non-explanatory, especially as the process leading to the duality in (10)–(11) is productive, and we would have to distribute such new ‘lexical features’ to an endless set of items in the substantive lexicon, missing the systematicity of the process and that it rather seems to depend on how a given lexeme is grammaticalized on an occasion, in the course of a given syntactic derivation. Similarly, we observe a difference between the clause Superman left, when it is used in isolation to make an assertion of truth, as in (12), and when it occurs embedded, as in (13), in which case no assertion of truth is made about it by the speaker: (12)

Superman left

(13)

Lois believes Superman left.

Instead, (12) as occurring in (13) is interpreted intensionally, namely as referring, in terms of intensional semantics, to a function from possible worlds to truth values, rather than to a truth in the actual world. This semantic difference again does not seem to be due to the words involved, nor to how this lexical material is ‘composed’: the string of words in (12) and of the same clause as occurring in (13) at least seems to


91

be exactly the same. It follows that the difference in meaning just noted must either be due to differences in hidden lexical material (a covert functional head, say: cf. de Cuba, 2007), or else a difference in grammatical meaning in our sense, reflecting the content of grammar, which is what we propose here.7 The grammar of ‘Superman left’ clearly is different in the two cases: for the clause is a grammatical argument in (13), but not in (12). The challenge is to give an explanatory account of this grammatical difference that can yield the intensionality effect. We sketch such an account in Section 3.5 (see further Hinzen, Sheehan, and Reichard, 2013). The claim, then, is that grammar and the lexicon contribute to meaning in different ways, inducing different kinds of effects. Intensionality falls on the grammatical side: it is not an effect of the lexicon, nor can it be explained by composing the contents of lexical items—which would explain why it has caused puzzlement and consternation in over 100 years of compositional semantics. The same is true for reference as well (as we would expect, if the claim applies to intensional interpretations, which in the terms of Chapter 4 is one weak way in which clauses can refer). Thus, consider that no (substantive) lexical item is as such referential. As noted, a common noun such as ‘man’ cannot be used to refer to a particular object, such as a particular man, unless it is uttered together with an act of pointing or a demonstrative, in which case the object-reference is contributed by the gesture or the deictic element, not the content word, which rather contributes the NP-restriction: (14)

[Pointing:] man

(15)

this man

Where it is clear enough which man is meant, ‘this’ in (15) can be reduced to the weaker form ‘the’, which lacks the locative/deictic element contained in ‘this’ or ‘that’, and referentiality to an individual object will then still be maintained. Referentiality in this case, too, is an aspect of the phrase as a whole, and the act of (either stronger or weaker) deixis (or grammatical pointing) that it encodes within its grammatical context. Reference thus comes from grammar, not the lexical item or its substantive content, which doesn’t refer, but merely supplies concepts that become part of an identifying description. Reference to particular objects, as in the paradigmatic use of (15), is thus not due to composing the referents of lexical items: ‘man’, as such, lacks such reference (taken in isolation, it doesn’t tell us even whether it will be used as a noun or a verb); and the word ‘this’ has no substantive lexical content to begin with, and equally lacks object-specific reference when used in isolation (as when uttered out of the blue when entering a room, with no identifying description).

7 We argue extensively against the former approach in Sheehan and Hinzen (2011), and return to such evidence in Chapter 4, Section 4.3, where we pursue the second approach in more detail.

92


Grammar provides what ‘man’ and ‘this’, each taken in isolation, lack: reference to objects, based on a description. It thus doesn’t add to the contents of lexical items, but to their grammatical semantics: how they can be used to refer. Grammar maps lexical concepts/contents, which as such lack reference, to objects of reference. Even where it combines two content-words, say table and top, so as to obtain table-top, we see reference decided in some initial way: whatever a person uses this compound to refer to, it will be a top (of a table), not a table (with a top). This crucial difference is nothing that is specified in either ‘top’ or ‘table’: it comes from grammar (namely, headedness). Even the grammatical operation of modification is thus not symmetric, and it is certainly not free of grammatical meaning. That meaning, however, adds nothing to the lexical features or content of ‘table’ or ‘top’. Reference is not more lexical content, it is grammar. Even a nominal compound, though, still lacks individual-specific reference: tabletop, as such, cannot refer to any particular table-top. Modification is not enough. The compound must enter the grammatical template illustrated in (15), and combine with a lexically meaningless item, such as ‘the’, rather than any other content word. This remains true for kill Bill, where again two contentful lexical items seem to be combined directly, as in compounds, and again without the mediation of a lexically meaningless word, but where the grammar now generates an argument instead of a modifier. Here, however, Bill stands in for the determiner in the nominal projection, and this is from where its object-referentiality derives (see Chapter 4). This also illustrates that the referential effect generated from adding ‘the’ is not due to whatever residual lexical content ‘the’ might have, for this putative content is absent in the case of Bill, when it stands in for the determiner. As a lexical item, ‘man’ also does not refer to a particular set of men which is rather what ‘these men’ in (16) refers to: (16)

I don’t like these men.

Clearly, we cannot take the less complex item ‘man’ to denote something that a grammatically more complex item denotes, namely ‘these men’, when the latter inherently contains the former as a proper part. If ‘these’ is dropped, ‘men’ comes to stand on its own; but then it still does not pick out a set of men, for which the determiner ‘the’ or ‘these’ would have to be added. Instead it refers generically to anything qualifying as a man as in (17–18): (17)

I don’t like men.

(18)

Men are brutes.

In (19), in turn, ‘men’ is quantificationally interpreted (meaning whether you have seen some men): (19)

Have you seen men?


93

We see, then, that in (14–19), formally different entities, all constrained by the same identifying description, are being referred to: a single given man, a set of men, arbitrary instances of the kind man, and an indefinite number of such instances. The meaning of the lexical item ‘men’ cannot account for any of these facts about reference, which depend on how this item functions grammatically: whether it is inflected for Number, the complement of a determiner, the determiner is overt or covert, and the noun is in a referential or predicative position. It follows that reference, too, like intensionality, is an instance of grammatical semantics, and that combining the lexical contents of words does not account for these differences in reference. The difference between a property and an object, which arises in the mapping of a lexical content, which as such provides an identifying description, to an object of reference, is not a lexical difference, any more than the difference between an object and an action is in (10–11). The difference is the contribution of grammar. Content does not only come from the lexicon, but the grammar, too, though it is one of a very different nature in the latter case. This is the second problem we see with semantic compositionality: assuming a meaningless syntax, it has all content coming from the lexicon. But the lexicon as such never entails reference. It only entails what we have called classification in Chapter 2. And just as parts of speech distinctions reflect the functioning of words in grammar, some items in the lexicon, mostly small (‘the’, ‘a’, ‘that’, etc.) reflect grammar as well, and its inherent task: regulating reference. But these items do not account for reference, which depends on whole phrases and how these function in relation to others. The same applies to even the putatively most ‘directly’ object-referring lexical items, namely proper names, pronouns, indexicals, and clitics, as anticipated in Section 2.2. (20) illustrates that proper names can function as predicates: (20)

Was the already centralized French state more likely to produce a Napoleon than the decentralized United States?8

The same is easily possible even for personal pronouns when they bear Accusative case as in (21), or for 3rd-person pronouns, which can marginally take a determiner: (21)

I am me. ‘I am the person I am.’

(22) He is a she. In short, forced in the role of predicate by grammar, commonly referential elements such as proper names and pronouns resort to whatever little descriptive content they have, leaving it to the interpreter to figure out which salient property precisely ‘being me’, or a pronoun of feminine Gender, evokes. The same is even true, though now

8 N. Ferguson, Civilization, Penguin: 2011 (157), speculating about better prospects for democracy in North America than in France in the late 1800s.

94


much more marginally, for demonstratives. Thus we can just about interpret (23), where a person might be commenting on what is going on in the here and now (or perhaps a Hegelian philosopher is creating abstractions from acts of deixis): (23)

the this and the that

And we get (24), which might be said by a patient who lost his deictic sense through brain damage but then recovered: (24)

There was no this and no that.

Clitics, too, can stand for predicative nouns, as long as they are not specified for grammatical Person, such as in the Italian (25) (from Longobardi, 2008): (25)

a. Gianni è monarchico e anticlericale; se anche Maria lo/*la/*li/*le fosse . . . Gianni is monarchical and anticlerical; if also Maria CL(3SgM)/*CL(other forms)-were . . . b. Se Gianni fosse te o anche solo se Maria lo/*ti fosse . . . If Gianni were you or even just if Maria CL(3SgM)/*CL(2Sg)-were . . .

Literally every lexical item, then, including deictic elements or traditional ‘indexicals’, can find itself in the role of a predicate. The same applies to the more grammatical elements such as Tense, which are arguably referential and deictic elements as well, and which are also inflectional and not proper lexical items to start with. The overall conclusion must be that the lexicon has no control over the forms of reference. Rather, the process of reference begins when lexically common nouns, which are always endowed with a descriptive content, become the complement of grammatical elements without descriptive content, creating a unit of grammar that can be referential depending on its grammatical position and the closed-class items involved. With some nominals—demonstratives, pronouns, clitics—the descriptive content is next to nil but not zero, and hence it can be exploited when the grammar is such that descriptive readings are forced. In such cases these items become complements of a determiner. Where they are referentially used, the determiner must remain absent. The compositional model is thus based on a problematic idea: that the basic primitive of semantics is word meaning, and that word meaning in turn consists in reference to external objects. In the case of such reference, the idea is, nothing really is to be explained—this kind of semantics, as it were, comes for free, and can be assigned to lexical items directly. Words are simply names for things, and names are labels.9 Confronted with the question of where sentence meanings come from, which do not

9 Thus, Frege (1892) effectively defined the class of ‘proper names’ in purely semantic terms, comprising everything that has reference to a ‘definite object’.


95

seem to be objects like tables and chairs, this lexicalist bias, in the absence of any notion of grammatical meaning, then makes a prediction: sentence meaning must somehow arise from word meanings, by ‘composing’ the latter. Since the base notion of meaning is that of ‘standing for’ or externally referring, sentences will now be taken to also ‘stand for’ or refer to external objects—just to objects of a different kind: a Russellian compound, say, thought of as a ‘chunk of the world’, consisting of the referents of the words involved, one mapped to a property, the other to an object, say, which are both making up the compound. Meaning, then, is totally non-linguistic: words are simply symbols, and sentences are symbols of other kinds. Their meanings are external objects. A puzzle with the pervasive intensionality of reference in natural language now ensues, which has defined the philosophy of language for more than 100 years. But if we are right, the alleged external meanings simply recapitulate grammatical distinctions, misinterpreting them lexically. Compositional semantics, given its lexical bias, bypasses grammar, failing to recognize the categorically different kind of meaning it generates. This different kind of meaning has nothing to do with external objects, as non-linguistically described. Indeed, if the lexicalist model worked, then all meaning is already there, just as the world itself is already there, before we ‘label’ its objects. All that language contributes is a system of signs for referring to it. The point of language, however, is infinity: we can refer to ever new objects, whatever the fancy may be that strikes us. One wouldn’t suppose that, before such a fancy strikes, all of these possible objects of reference already pre-exist our reference to them. The range of different types of external objects posited, moreover, needs to be explained: semantics does not come for free, and while the types in question define our metaphysics, metaphysics does not seem to explain their existence. If the formal ontology is not identifiable irrespective of grammatical function, the explanation may well be grammatical. If this is the case, grammar will ipso facto not merely be a formal system, which as such only ‘interfaces’ with a substantive or interpretive system.

3.4 Content from the interface? That a particular domain of inquiry be studied formally is obviously desirable from a scientific point of view, if that domain lends itself to a formal analysis at all—indeed it is a condition for scientific inquiry. But any such decision is independent of the domain’s particular ontology, and it implies nothing about what this ontology is: the formal analysis of a domain says nothing about the kind of entities it contains. The formal study of grammar, too, is therefore not the same thing as the study of form (if this is taken to be ontologically distinct from ‘content’). In a similar way, the formal study of patterns of planetary motion implies nothing about whether planets are

96


‘formal’ objects or not. Grammar can be a fully material domain of inquiry, much like planetary motion, even though it exhibits formal patterns that can be studied as such, as in the case of planetary motion. ‘Form’ and ‘content’ are then not names for things that exist out there, that can be studied as such, and that are separate from one another as domains of inquiry. Everything, rather, can be studied in terms of both its form and its content (insofar as it has formal properties): that is a matter of choice (and reflection), not of ontology. This contention raises some potential problems about the study of what, in frameworks such as Minimalism, is taken to be a ‘real biological object’: namely, the ‘computational system’ of language. That a particular domain is studied formally or computationally does not mean that there is a ‘computational system’ that is being studied—if this means that our object of inquiry is a purely formal system consisting of arbitrary manipulations of symbols in abstraction from semantic content. Our object of inquiry, rather, in this biolinguistic domain, is grammar. The petals of the sunflower exhibit a Fibonacci pattern. But it is not ultimately this pattern, but the flower—through the pattern—that biology studies. Grammar, too, has a form as well as a content. That content is, we will ultimately argue in Chapter 9, the world— namely, as it appears to us when a grammatical pattern is laid upon it. The Minimalist Program has brought the attempt to describe grammar in abstraction from its content to a peak, thereby sharpening the question we are asking. It also instantiates but also radicalizes the long-standing attempt to eliminate stipulations in the formalization of grammar. The PSGs of the 1950s showed how to characterize, in a descriptively adequate fashion, the order and projection properties of contiguous relations in grammar. Non-contiguous relations were taken care of by the transformational component. The latter was later unified with a PSG with a recursive base, which was then in turn simplified by X-bar theory, which stated the laws of phrase structure in non-category specific terms. Transformations were restricted in their descriptive power through general conditions on transformations. Each of these steps implied a radical reduction in the options of variation that can exist in the syntactic system of language as so described, and it already exemplifies what only later came to be called a ‘minimalist inquiry’. At the end of this process, Minimalism stripped PSG down to its bare bones: Bare Phrase Structure (BPS), viewed as built by the simplest conceivable computational operation, now in charge of both contiguous and non-contiguous relations: Merge. The grammatical relations it creates are sisterhood—two syntactic objects are merged into the same set—and containment—a set generated by Merge contains another set generated by Merge. None of these come with any grammatically specific content. Nor does the operation Merge itself, which is insensitive to any grammatically specific material such as part of speech distinctions or grammatical relations in the sense above. Even if the Minimalist claim that this system gives a minimal account of both contiguous and non-contiguous relations in grammar is true, therefore,


97

grammar has no content on this description that goes beyond the statement of discrete infinity and the analysis of the latter through Merge.10 This is why the content has to come from elsewhere on this account: narrow syntax, which is an abstraction from content, ‘interfaces’ with the semantic component, the ‘conceptual-intentional’ (C-I) systems. This embedding in the rest of our cognitive architecture is what gives it a linguistic character, in other words, an interface, which divides form and content; content, perhaps in the form of a compositional semantics in the sense discussed above, is added to the Merge-based system viewed as something that is independent as an object of scientific inquiry. And perhaps, the thought is, the language faculty is ‘perfectly designed’ in light of constraints imposed by this interface, at least the semantic one, since it appears to take priority over the other, the sensory-motor one (‘interface asymmetry’: Chomsky, 2008). The constraints imposed by the interfaces on the semantic side however are barely investigated and have tended to be of the most generic kind, like the principle of Full Interpretation. This principle maintains that every element of a syntactic representation arriving at the interface must be interpreted by the C-I systems. Since these systems know nothing about grammar, however, any meaning generated by grammar must be unreadable by these systems and hence cannot exist. The absence of a notion of grammatical meaning in this framework is thus not a contingent aspect of it, but built into its architecture. If grammar is uninterpretable, its rationale can only be to eliminate uninterpretable (‘formal’) features contained in it before points of access to the interface, which is regarded as the basic principle driving its dynamics (Gallego, 2012; Richards, 2012; Chomsky, 2007a; 2008). But since there is no point to these uninterpretable features in the first place, there is no rationale, really, to grammar at all. Results in comparative cognition have not, on our view, come to lend support to an architecture where a grammar, although minimal in formal terms, is arbitrary and unprincipled in substantive terms, and interfaces with a semantic component, which is non-minimal in formal terms but principled in substantive ones.11 This predicament is what we predict if even positing ‘concepts’, let alone propositions, on the non-linguistic side of the interface, is subject to severe doubts, if ‘concepts’ means

10 That language is ‘discretely infinite’ is a property it shares with myriad other systems in nature that can be modelled in this fashion, including arithmetic, music, phyllotaxis, and DNA (see e.g. Katz and Pesetsky, 2009; Mukherji, 2009: ch. 6; Gimona, 2006; Searls, 2002). Merge, which amounts to discrete infinity, therefore, gives us a minimalist analysis of what we knew before, that human language is recursive. 11 Chomsky (2008: 141) in particular suggests that ‘C-I incorporates a dual semantics, with generalized argument structure as one component, the other one being discourse-related and scopal properties. Language seeks to satisfy the duality in the optimal way ( . . . )’. No evidence for this hypothesis is provided, and prospects for finding evidence that such a semantics—in essence, all of thought as it is manifest in the semantics of human languages—is present in non-linguistic systems of thought appears slim.

98


anything like the meanings of human words. Even the most optimistic accounts of ‘animal thoughts’ acknowledge that a gap with the kind of thought expressed in language remains: a chimpanzee, or a zebra finch, do not think like us. There is no common metric by which we can compare non-linguistic thought with linguistic thought in any systematic fashion, however, so as to assess how much of the latter there is in the former. No one has studied whether non-human animals can process the contents of verb phrases, say, but not sentences; or of non-finite clauses, but not finite ones (which would mean to use a linguistic metric, given the absence of a nonlinguistic one). For all we can tell, there appears to be no phrasal organization of meaning at all, which also is arguably not found in the communications of the ‘language-trained’ apes (Tomasello, 2008). Nor is there evidence for a semantics of this nature in animal thought rather than communication (Fitch, 2005). This is again not to deny that the formal-ontological notion of an ‘object’ is partially given in perception—and again not to affirm that, thereby, a creature with such perception has thoughts about objects involving the concept of ‘object’. Consider in this regard the processing of a voice, an auditory object that is the perceptual correlate of some acoustical pattern that reaches our ears. This pattern is cognitively analysed and different perceptual categories are imposed on it, which might involve the recognition of the voice as human, or as that of a particular person (a source), in which case certain structural invariants would be used for categorization corresponding to characteristics of the speaker’s vocal tract. It might also involve the recognition of a vowel as associated with a particular pattern of change that can be produced by different possible sources and representing an invariance across different auditory events and sources. What comes to be perceived in each of these cases is an auditory object involving an extraction of abstract properties from a given auditory input. These are meaningful for us and required for the perception of the ‘same’ voice at different times and in different states of vocal health, or of the same vowel or melody produced by different persons at different times. It will also typically involve sensory integration, as when the perception of a voice is integrated with the visual processing of a face. The emergence of a unified auditory object, from low-level temporal and spectral (frequency) analyses, is eventually helped by the attribution of ‘meaning’ in the sense of triggering a particular concept of what is being perceived.12 Perceptual objects in this sense, however, even as analysed by meaningful categories, never involve a relation of reference, since none is required, given how the representations generated in this domain are rooted ipso facto in the sensory context. Moreover, before we apply the notion of object to

12 Stages before and after attribution of meaning appear to be neuro-anatomically distinct (Griffiths and Warren, 2004).


99

perception naively, we should be aware that the very notion of ‘perceptual object’ is also quite problematic, with some frameworks explicitly rejecting this ontology.13 We will continue to assume, therefore, that perceptually experiencing the world as structured into objects falling into various more abstract classes is very different from an ability to deliberately refer to such objects symbolically. As we suggested in Chapter 2, the reason that we can refer to objects to which we are not exposed in ‘online’ perception is that we have concepts—the meaningful categories that we apply to perceptual objects—which as such characterize possible objects of reference. So acts of reference are anchored in concepts, and such acts can succeed even when they are based on little more than that, as when I refer generically to kings of France: nothing, in this case, anchors the act of reference except the descriptive content of the phrase kings of France (and how it functions grammatically). Reference can be less generic and more specific than that, yet some residue of descriptive content is never completely absent, as noted even for the case of pronouns and indexicals above. This appears to be the crucial difference to functional reference in animal alarm calls, which do not tap into abstract conceptual resources in addition to perceptual classifications. No concept is needed. When an amodal concept evolves from a modality-specific or multi-modal percept, the concept is not ‘indexed’ to the context of immediate sensory-motor processing anymore. The connection thus has to be reinstated (cf. p. 47 above). But perception cannot do this, since grammar, when accessing concepts and grammaticalizing them, can establish acts of reference that cannot be linked to any percept, even in principle. Grammar, therefore, needs to contain elements and procedures that re-indexicalize the novel contents that it generates. This is a sensible rationale for the existence of traditionally so-called ‘indexical’ expressions in grammar, such as pronouns and demonstratives. It also rationalizes the grammatical processes that license these items in appropriate positions where they take on interpretable grammatical functions (for example movement into subject position to become the subject of a predication in the present sense, i.e. a truth-value bearing expression), which involves Case or Person relations. Much of the machinery of grammar, in short, serves to implement the re-indexicalization that is required in acts of reference based on concepts as distinct from percepts. The transition to grammar is thus the transition from sensory exposure to the world and conceptualization to reference to this world, itself an indication of thought. Our thoughts can now be wrong. That again is an indication of thought, as opposed to dealing with the world perceptually and practically. This possibility of falsehood, however, only arises at the end of the grammatical process: it is not an 13

In the psychophysical model of Bregman (1990), for example, the fundamental elements of the auditory world are auditory ‘streams’, rather than objects, and the concept of an auditory object is explicitly rejected. For discussion, see Griffiths and Warren (2004).

100


option yet at the level of object reference or the nominal phrase, where all such reference comes with a presupposition of existence, as crucially distinct from assertions of existence (as in ‘there exists a king of France’), which of course can be wrong. Acts of reference to objects as existing (‘the king of France’), by contrast, cannot be: no truth value can be assigned to ‘the king of France’. In this sense there is, in the case of reference to objects in grammar, ‘immunity from error’: error, like falsehood, requires the sentence. Someone who knew that there wasn’t a king of France couldn’t point this out to a person who didn’t know it by using a noun phrase. So questions of existence are not decided or resolved at the level of nominal reference to objects, and no such resolution is a precondition for symbolic reference in grammar. But that Superman, by some non-grammatical (such as physical) criterion, does not exist, is a consideration irrelevant to the Superman aficionado who is infatuated with his (or her) hero and refers to him, as something existing, by the name of ‘Superman’. In a less affected mood, he may agree with us that ‘Superman does not exist’, which would be contradictory if using this proper name implied existence or a claim to existence. Language moulds our sense of reality, creates a world, rather than merely responding to one. ‘Reference to non-existents’ is no anomaly of language, but its hallmark, and no motive for re-interpreting expressions as non-referential ones when a presupposition of existence fails. The absence of the phenomena of both reference and truth in perceptual systems or animal cognition confirms our stance, assuming the essential absence of substantive constraints on grammar as coming from postulated C-I systems, and the absence of independent predictions from a ‘semantic component’ for what we will find in grammar. Given the emergence, with grammar, of a novel referential-deictic system distinct from both perception and classification, the conclusion should be that there is no system there on the C-I side of the interface, equivalent in semantic power to a grammar-based system, which could then constrain the grammatical system. In fact, as noted we predict that whatever non-linguistic system exists on the other side of the posited interface, couldn’t even systematically interpret the generations that arrive on the grammatical side (Hinzen, 2006; 2008; 2009). C-I systems should look at the deliverances of grammatical derivations with bafflement or stupor, failing to interpret them altogether, much as we would be when confronted with communications from aliens in a never-encountered code. This in turn engenders another conclusion: that the form-content dichotomy must be given up. Content has to lie on the side where the form is: the grammatical one. This, of course, will hold exactly insofar as meaning is grammatical. Insofar, that is, as there is a difference that grammar makes to the organization of meaning. The specific content of grammar, we here suggested, lies in grammatical relations, since we know it does not lie in the atoms of the lexicon. Yet Merge applies to lexical atoms generating binary sets, and the members of a set stand in no inherent relations to one another. Lexical items in turn are collections of lexical features, and some of these


101

(Case, ç-features) will trigger relations of Agree and potentially applications of ‘internal Merge’ (IM) on the standard model: the merging of a syntactic object A contained in another one, B, to that object B. IM creates a discontinuous constituent, and hence grammatical relations, with interpretive consequences as in the case of operator-variable dependencies or binding. But there is no theory of grammatical relations in grammar so construed, viewed as primitives or fundamental elements of grammar. If they exist, they are coded featurally. Feature-based coding does not of course exclude grammatical relations, but by coding these uniformly through features and generic operations such as Agree and Merge, a restrictive and principled theory of grammatical relations and hence of the content of grammar does not come into view. Since formal semantics, in turn, tends to code relations by assigning them as denotations of predicative expressions, hence effectively lexically, the very term ‘grammatical relation’ lacks a status within the Minimalist framework. It will be a theoretical primitive in the present one, and our basis for reconstructing the recursivity of grammar, following Arsenijević and Hinzen (2012), in the rest of this chapter. This position contrasts with the standard Minimalist position, but is not, on our view, inconsistent with Minimalism as such. The difference is that recursion as described through Merge is a purely formal universal of human language that does not say anything about grammar specifically. It is the culmination of a formal tradition in linguistic theory that abstracts from semantic interpretation and is, given its bare minimality, the peak to which such a perspective can be driven. That is the strength of the Merge-based viewpoint—but also, as we will argue, its weakness. For as we may already conclude from the above, Merge leaves us without a handle or substantive insight on grammatical relations, or on the content of grammar. Grammar is recursive; but its substantive content is not recursion.14

3.5 The phase as the smallest unit of grammatical organization The above account entails that the smallest unit of grammar is a unit of reference; that any such unit is based on a lexical concept that is as such non-referential but has a descriptive content that will be the basis of its referential use; and that the possible forms of reference for a given lexical concept depend on how it is grammaticalized: what grammatical role it plays and which head takes its lexical 14 Relegating predication, reference, and truth, to ‘semantics’ or C-I, which we have argued against above, interestingly unites Minimalist grammar and its rivals. Thus, for Jackendoff (2011), the inertness of grammar in relation to the genesis of a propositional semantics is even more pronounced, as is the form– content distinction, with syntax in charge of the former and semantics in charge of the latter. In line with this, he stresses the existence of relational cognition in non-human primates, as Boeckx (2010) does, who suggests regarding Merge as completely unconstrained, with relevant constraints on grammar coming from the way in which derivations based on Merge hook up with extra-linguistic concepts such as thing or location at phase-boundaries. Our scepticism with regard to calling these ‘concepts’ is clear from the above.

102


projection as the complement. Putting these observations together, we obtain (26), where the ‘interior’ houses the lexical concept and its lexical projection (e.g. Noun Phrase (NP), as in (27)), and the edge houses grammatical elements that govern how it is used to refer (e.g. determiners): (26)

[EDGE [INTERIOR]]

(27)

[the [man]]

This unit of grammaticality cannot be reduced further. If there was no edge governing reference, we would only have a unit of conceptualization, not reference. If there is an edge but it is ‘weak’, in a sense that we will make precise in Chapter 4, we get a unit of grammatical organization and hence reference, but reference will be established via the descriptive content of the lexical material in the interior only, as in I like kings of France, where the referent of the internal argument is no objects in particular, but objects only insofar as they instantiate the generic description ‘being a king of France’. If we only had an edge, we might have functional reference, not an act of human grammatical reference, which is always based on a lexical concept with a descriptive content used predicatively. Note that there is no recursion in this unit of grammar: nothing recurs. There is a single edge and a single interior. In line with this, there is a single referent, established at the edge, whatever it might be; and there is a single lexical concept. If another substantive lexical concept is added, it thus has the status of a modifier, which as such is not required and goes beyond the minimality exhibited in (26): (28) [DP the [NP good [NP man]]] As we would predict from this, modifiers exhibit recursivity in a much clearer way than units of reference: they can be stacked unboundedly (subject to conceptual constraints, though, as studied in Cinque, 1999): (29)

[the [good old wise [man]]]

Crucially, the recursion, insofar as there is one, is part of the interior, not the edge (but we return to determiner and clitic doubling below and in Chapter 4). Moreover, none of it disturbs the essential generalization above: that there will be a single referent at the edge—even if described by multiple predicates—and a single lexical head that the unit in question turns into a referential expression. By not exhibiting recursion, the smallest unit of grammar exactly resembles, as it should, an act of (e.g. index-finger) pointing: no such act is ‘recursive’ either, in the sense that it would ever contain two referents, with one referent embedded in the other. The question thus arises what the status of recursion in grammar is, which we address in the next subsection. (26) resembles the recent notion of a phase as conceptualized in Chomsky (2007a and b; 2008; 2013), and we will thus suggest that the phase is the smallest unit of


103

grammatical organization. From the non-recursivity of this unit, at any given point of the computation, it then follows that there will only ever be a single phase in a derivation, resulting in a ‘single-phase’ architecture. At any one moment in the derivation, that is, only one referent is ever computed. This is true of units such as (31) as well, where ‘that’ is the edge element governing reference and the TP, which corresponds to NP in the nominal case (30), is the interior specifying the descriptive content of a particular state of affairs: (30)

[the [NP kings of France]]

(31)

[that [TP there are kings of France]]

The same template repeats again in (32), where the embedded CP functions as a descriptive predicate by means of which we pick out a referent at the edge of v*, in this case a mental event of believing: (32)

[v* believes [CP that [TP there are kings of France]]]

We see it one final time in (33): (33)

[CP C [TP Bill T [v* believes [that there are kings of France]]]]

Here C is again the edge—(33) as a whole refers to a truth value—and TP provides the descriptive content, as in (31). In each of (30–3), there is a single referent and a single identifying description, illustrating the presence of the same, single, non-recursive template in (26). No sentence has two subjects or two predicates; no sentence has two truth-value referring expressions, as illustrated in (33), whose meaning is such that no truth value is assigned to the embedded clause ‘that there are kings of France’. Instead there is only one truth value and it is assigned to (33) as a whole. The embedded clause is therefore not evaluated for reference/truth, in the way it would be if it was not embedded as an argument. This is the exact prediction that our template makes, and the intensionality of clauses that are embedded arguments is thereby deduced: for given the constraints of the template, the embedded clause must function as a predicate within the higher phase in which it enters, rather than carrying a truth value into the process of semantic composition. Far from being the anomaly that intensionality has been, causing puzzlement and consternation from a viewpoint of semantic compositionality for more than a century, it now follows from the architecture of grammar, being a core aspect of grammatical semantics (Hinzen, Sheehan, and Reichard, 2013). The referents of the phases are formal-ontologically distinct: if we wished to use ontological language, the nominal phase is mapped onto an ‘object’, the verbal to an ‘event’, the clause to a ‘proposition’. Crucially, these are asymmetrically ordered with respect to one another, with propositions always (or architecturally) containing events as part of their structure, forming their thematic core, and events in turn

104


containing objects. We can thus think of these three phases as the ‘first’, ‘second’, and ‘third’ phases. The non-recursivity of the phasal template entails that none of these formal-ontologically distinct objects of reference will ever embed another object, within the same phase. Instead, when an event embeds an object, say, as all events do, we expect that the object in question must first be mapped from a predicate, in order to be licensed to the event. This is what we find: arguments are relations (Bowers, 2011). An object-referential expression such as ‘John’ can only function as an object within an event if it is interpreted as bearing a grammatical relation to it, which is what the term ‘thematic role’ denotes. The form of a verb–noun embedding is therefore not of the form (34) but (35): (34)

[VP [DP]]

(35)

[VP [Ł DP]]

That is, the object is an Agent, Patient, Goal, etc. Crucially, neither of these reflect properties intrinsic to the referents in question, in which case they would be lexicalsemantic, rather than grammatical relations. With theta-roles, therefore, we are firmly in the domain of grammatical meaning. As Bowers (2011: 8) stresses, ‘Agent’ viewed as a theta-role is a grammatical category, not a (lexical-)semantic one. As he explains, following Emonds (1976), a pseudo-passive as in (37) is possible just in case the subject in (36) bears the argument-relation of Agent rather than Theme: (36)

The bird/the book flew across the room

(37)

The room was flown across by the bird/*the book

The issue here is not animacy, a semantic notion, since even the animacy of the subject ‘the bird’ does not prevent it from being construed as a Theme-argument in (36), in which case the bird is merely involved as an inert projectile and the pseudopassive is now also impossible for it. Similarly, a fairy tale context could allow the construal of ‘the book’ as an Agent-phrase, showing that nothing in its intrinsic lexical feature specifications rules the construal of a referring expression as exhibiting a particular grammatical semantics. The structural nature of the ambiguity in question is illustrated in (38), where ‘John’ can again have the grammatical meaning of either Agent or of Theme: (38)

John appealed to Mary.

In the second case, (38) means that John was attractive to Mary, in which case the pseudo-passive is again impossible: (39)

*Mary was appealed to by John.

Non-grammatical notions such as ‘intentional agency’ do not seem to succeed in explaining these facts. As Bowers notes, (40), with ‘unintentionally’, only has one interpretation:


105

The room was deliberately/unintentionally flown across by the bird.

It uniquely means that the bird crossed the room intending to fly somewhere else; it does not mean that the bird was sent across the room in the manner of an inert projectile. In other words, the grammar of the argument as an Agent requires construing the world in particular terms, namely with the bird as an intentional agent, no matter whether lack of intentionality is specified lexically through the adjunct. A world seen through the lenses of grammatical meaning becomes different indeed. Crucially, there is no recursion in the theta-domain either: theta-roles, like argument-structure itself, form a strictly finite, and indeed very limited, template. The grammar stretches to generating structures of three syntactic arguments, but not more: all other dependents must be adjuncts. An ‘Agent-of ’ phrase embeds no other Agent-of phrase recursively, but a Theme-phrase. So the ‘intransitive’ structure in (35) can be transitivized, in which case the additional head v introduces another argument, and we get two (necessarily different) theta-roles: (41) [[Ł2 DP] v [VP [Ł1 DP]]] We then obtain, formal-ontologically, two events, but crucially not two separate ones, and not two of the same kind: one will be embedded or contained in the other as a proper part: an event of killing Bill, say, will involve an event of ‘making dead’ performed by an Agent, which is followed by its causal consequence, the state of the Theme’s being dead. So event embedding comes at the price of the embedded event becoming a proper part of the embedding one. The V-phrase that is embedded in the v-phrase is again not referentially independent, but integrated with the higher head via a grammatical relation, described mereologically as that of ‘part’ (see Section 2.5 above). And again, the process is sharply limited: after transitivizing a phrase [Bill [dead]] so as to obtain The bride made Bill dead, which can be spelled out as The bride killed Bill, we cannot transitivize again, with (44) meaning that Elle made the bride kill Bill: (42)

[Bill [dead]]

(43)

The bride made [Bill [dead]] = The bride killed Bill

(44)

*Elle killed the bride Bill.

A general principle, then, seems to veto the embedding of the same in the same. Embedding in grammar is constrained by having the embedded element play a relational role, specifically becoming a part in whatever the embedding phrase refers to.15

15 Our claim here goes counter to Ramchand (2008), who argues that argument structure in syntax is ‘constructed systematically on the basis of primitive recursive syntactic relationships’ (p. 16), which on her view involve three primitive ingredients in the human conceptualization of event-structure: a causing or initiating sub-event, a process-denoting sub-event, and a sub-event corresponding to a resulting state.

106


In this sense, embedding has an inherent content, a grammatical meaning, and is not ‘blind’ or ‘formal’. The same is true in clausal embedding, where the embedded clause necessarily plays the role of a content of a mental event referred to by the matrix verb, or in NP–NP embedding, where the embedded NP plays a role as a Possessed, a Theme, or a location: (45)

John’s mother

(46)

The bride’s purse16

(47)

The vase on the table

All of these instantiate our template, and they are single referential expressions, with (47) clearly referring to one thing, a vase, which is identified through a descriptor encoding the predicate ‘being on the table’, which is related to the higher DP via the quasi-thematic role of location. This is also clear from Agreement facts: (48)

The vase on the table is/*are broken

What is not possible is a juxtaposition of two referential DPs, without one of these playing a role as part in relation to the other: (49)

*John mother

(50)

*the bride purse

(51)

*the vase the table

This is so unless the two DPs are part of an enumeration, as in (52), where the grammar computes two referents, and the Agreement facts change: (52)

The vase, the table are/*is broken.

This architecture may explain why the basic pattern is that V embeds N and not the other way around. Suppose that ‘V’ and ‘N’ are lexical ‘features’, which are intrinsic to the specifications of a lexical item. Then they are atomic and independent, and we predict nothing about how they will relate. That is, we will have to stipulate how they relate: through relations of semantic or categorial ‘selection’. Relatedly, the same problem arises if we take a lexical and intrinsicalist view of events and objects: if these are what is denoted by V and N, respectively, they will again be independent of one another, and their relation, if any, will have to be stipulated. On the present model,

Although these components can be combined (‘merged’, on her model), however, and the result has compositionally determined readings, the combinatorics is strictly templatic rather than unboundedly recursive: these primitives only compose in a highly limited and predetermined fashion. Similar remarks apply to the notion of ‘composition’ of the basic elements of argument structure in Hale and Keyser (2002: ch. 1). 16

This of course has another reading, where ‘bride’ is a modifier, hence not independently referential.


107

the mystery of why V embeds N and not the other way around dissipates and in fact becomes unformulable. The explanation of the generalization is that event-denotation is not lexical: it is grammatically derived. Unlike object-reference, which is mono-phasal, event-reference is necessarily bi-phasal. So any object-phase will be embedded in an event-phase, and it is a general constraint on embedding that embedding-as-argument in grammar creates a part-whole relation between embedder and embedded. So objects will be parts of events, not the other way around. Put differently, V contains N, necessarily, exactly as events contain objects. In a similar way, T contains v/V as part of its internal architecture; non-finite T will be part of finite T; and finite T will be a part of C. None of these, then, quite simply, will be lexical notions. Interpreting them as ‘features’ fails to recognize their nature as grammatical and derived. Similarly, putative relations of ‘selection’ holding between one feature and others are a misleading way of stating what is in effect a grammatical dependency: intuitively, an event has whatever parts it does, but it doesn’t ‘select’ its parts, any more than a human body selects an arm. Nor does it make sense that within a phasal template, D selects N, or C selects T. Rather, every referential expression is based on the lexical predicate it is based on, and the two pairs necessarily form a unit (though the edges can be missing, as discussed in more detail in Chapter 4). Where edges are differentiated in a more fine-grained way, as in Rizzi (1997), a selection relation between functional projections again makes no sense, as the relevant hierarchy of projections is hierarchically ordered. In line with our stress on the finiteness of the process of creating part-whole relations in the formal ontology, the process as a whole finds an ultimate limit in the assignment of the truth value, which contains literally everything else as a part and hence cannot itself be a part. More precisely, there is nothing, at the point of reaching the truth value, that the grammar can do so as to change the formal ontology. Anything that can happen, such as ‘slifting’ (53), assertions of veridicality (54), adding tags (55), or evidentials (56), leaves the formal ontology and the truth value itself unchanged: (53)

The bride was killed, I believe.

(54)

It is true that the bride was killed.

(55)

The bride killed Bill, didn’t she?

(56)

The bride killed Bill—I saw it.

The essential non-recursivity of grammar as depicted above, and the reinterpretation of the recursivity of grammar in terms of a generation of an ontology structured by a part–whole relation, now predicts that there will be no recursivity in the elements added in (53–6) after the truth value is assigned: and indeed, there is no such recursivity:

108


(57)

*The bride was killed, I believe I hope.17

(58)

It is true that it is true that the bride was killed.18

(59)

*The bride killed Bill, didn’t she did she?

And where languages encode evidentiality morphemically, recursive stacking of such morphemes is unavailable. Why then is it that the grammatical process should come to an end with the truth value assigned? Why is there no further ontological category that grammaticalizes? We return to this question in Chapter 9. Let us now come back to the operation Merge and to the standard claim that grammar is recursive, with Merge being its one most essential element. One could point out that our phasal template contains a combination of elements, and hence presupposes something like Merge. Indeed this is the standard justification of Merge as the basic operation of grammar: where grammar is not said to be based on it, the operation is somewhere hidden in whatever operations are assumed. Now, we agree of course that the template contains a combination of elements—but then, saying this about our template adds no insight to it. Indeed, one might call it an abstraction from grammar, given that the symmetries that Merge builds precisely arise by abstracting from the fact that no symmetric structures ever arise when two syntactic objects merge. Merge thus throws no light on the observation that symmetric structures in grammar are banned (Kayne, 1994; Moro, 2000; diSciullo, 2005). If recursion in grammar is modeled through Merge, moreover, embedding will typically be interpreted ‘formally’, in line with the standard assumption that there is no grammatical meaning and form and content are distinct. This means that semantic effects of embedding as described above, relating to the emergence of a formal ontology structured by part–whole relations, have to come from something other than grammar. But where should we go to derive the effects? The lexicon, we have argued, as much as the putative non-linguistic C-I systems, are the wrong place to look (even if they were to support some initial stages of the formal-ontological hierarchy, before we come to the part that depends on grammar). Nor would it help, we have argued, to go down the standard route of adding a compositional semantics to a Merge-based syntax. In fact, we now see stronger evidence against this conclusion: for if V contains N as a part, necessarily (and similarly for all other categories

17 Intended is the reading where the ‘slifted’ clause ‘The bride was killed’ is asserted, while the assertion is then qualified twice (cf. ‘the bride was killed is what I believe and hope’. It is possible of course to say ‘The bride was killed, I think I said’, where there is a single such qualification, and ‘I said’ is a predicate of ‘I think’. 18 While (58) is not ungrammatical, the second ‘it is true’ makes no contribution to meaning, unless the two ‘it is true’ phrases depict epistemic assessments of different speakers (in which case recursivity is broken and the verdict against embedding of the same in the same is adhered to).


109

that arise in the grammatical process), then the denotational content of V (its grammatical meaning) cannot be derived compositionally by first assigning a denotation to N: the way N is interpreted partially derives from its embedding in V. Of course, N also has a grammatical meaning on its own. This is whatever grammatical meaning is assigned phase-internally in a mono-phasal derivation. But much of its grammatical meaning is not assigned internally to its own phase. This is not only true for the thematic role it plays in relation to an event, but also for, say, its referential specificity, which may depend on its entering relations marked morphologically as Cases, which are not phase-internal relations. For example, Megerdoomian (2008: 79–81) shows that in Eastern Armenian internal arguments with a definite specific interpretation (such as proper names) need to carry accusative Case, and that indefinite existential DPs, which may be specific or not, carry the indefinite specific interpretation only when they carry accusative Case: (60) Ara-n Siran-in Ara-NOM Siran-ACC ‘Ara is kissing Siran.’ (61) *Ara-n Siran-in Ara-NOM Siran ‘Ara is kissing Siran.’

hamp'uyr-um kiss-IMPF

hamp'uyr-um kiss-IMPF

e be-PRES/3SG

E Armenian

e be-PRES/3SG

(62) Ara-n ašxat-um e mi hat dzi-an b@Rni Ara-NOM try-IMPF be- PRES/3SG one CL horse-ACC catch-Subj/3SG ‘There is a specific horse such that Aran is trying to catch it.’ If accusative Case is determined at the level of vP, it follows that DPs acquire some aspects of their interpretation only later in the derivation, in the next phase, not at the edge of their own phase. Compositional semantic interpretation by contrast is local. It does not capture the relational content of grammar. We conclude that although it remains true that a grammatical system of the kind that we are depicting in this chapter—in essence, a system of reference that sets up a formal-ontological space in which such reference happens—can be modelled formally through Merge, this fact adds no grammatical insight and in fact makes apparent generalizations (which go to the heart of what grammar is and does) harder to state. Plainly, a system that is based on the unbounded generation of arbitrary structures through Merge (‘wild-type Merge’ in the sense of Boeckx, 2008; 2010), has to be restricted in such a fashion that the grammatical patterns that we actually see come out. Plainly, we cannot assume that a Merge-based system is more ‘minimal’ if the minimality of the description is bought at the cost of outsourcing the complexity elsewhere—especially if it is outsourced to a component that we barely know how to study in non-linguistic terms, such as the C-I systems, and that is certainly harder to

110


study than grammar itself. Moreover, apparently universal restrictions on recursion, as discussed above and below, are not predicted on such a model.19 We close this chapter with a review of apparent restrictions on recursion in human language, other than those discussed above, and apparent exceptions to them, which a Merge-based account does not predict. This subsection is largely based on updating Arsenijević and Hinzen (2012).

3.6 The absence of X-within-X in language (with Boban Arsenijević) The primary focus of attention in linguistics when ‘recursion’ is being debated is not recursion as a mathematical property, nor hierarchy or Merge as such, nor computability, but the apparent occurrence of ‘self-similar’ structures involving the recurrence of some syntactic category X in itself.20 In other words, it is recursion in a linguistically meaningful sense. We have above suggested that because embedding is grammatically meaningful and creates a relation in which the embedder and embedded category are always asymmetrically related as part and whole, there is no embedding of the same in the same. Moreover, there strictly is always only a single referent per phase, thus deriving the intensionality effects arising when a clause embeds another. We also noted that Merge as such is silent on whether there is such a restriction or not. It therefore doesn’t speak to the issue of why or whether there is such (non-)recurrence. In this subsection we discuss data on self-similarity in language that fall out from the account of the previous subsection. Why, for example, should it be the case that a sequence in which elements traditionally categorized as C embed other Cs really is a [C- v . . . [C- v . . . [C- v]]] sequence, as seen in (63), or even a [C- v-D . . . [C- v-D . . . [C- v-D . . . ]]] sequence, as seen in (64): (63)

[CP Allegedly, [TP John will [vP deny [CP that [TP Bill has ever [vP said [CP that . . . ]]]]]]]

(64)

[CP Allegedly, [TP John will [vP deny [DP the very possibility [CP that [TP Bill has ever [vP defended [DP the claim [CP that . . . ]]]]]]]]]

The ‘recursion’ of C in C, that is, is mediated by a rigid, linear, and finite sequence of non-Cs (linear in the sense that one has to come after the other in a relatively 19 One of the few writers who link recursion to semantics is Fitch (2010), who argues that only if we ‘combine intuitions about whether a string is grammatical or not, and whether the units are of the same type of referent’ (our emphasis) can we approach some sort of empirical test for whether recursion in the sense of self-embedding is really present in language. Fitch suggests we ‘use the correct [semantic] interpretation to probe the underlying rule system’, seeing this as an explication of the Hauser et al. (2002) stance, according to which it is not recursion that is specific to language, but ‘recursion and the mapping to the interfaces’. This is the methodological suggestion we implement here. 20 For example, Everett’s claim against recursivity in Pirahã is based on the observation that Pirahã is ‘the only language known without embedding (putting one phrase inside another of the same type or lower level, e.g. noun phrases in noun phrases, sentences in sentences, etc.)’. (Everett, 2005: 622)


111

rigid hierarchical order, before a new C can be inserted).21 This basic observation can be put by saying: paradigmatic forms of recursion in language such as the above are cyclic. If cycles are phases, what we see is that recursion is mediated by phase boundaries, turning phases into the units of recursion. Why should recursion be cyclic (‘recursion-by-phase’)? An answer to this question on currently standard assumptions is not easy to obtain. On the above approach, on the other hand, the question doesn’t even arise. For the basic unit of grammar is the (single) phase, our basic template. A bi-phasal structure can thus only recapitulate this template, with the lower phase figuring in the interior of this structure, hence receiving a predicative or relational interpretation. Even cyclic recursion of C-in-C need not be productive beyond a single round. Consider CPs that are specifiers of another CP, as with conditional clauses and other ‘high’ modal modifiers: (65) [CP [CP If John buys apples] Mary can make a pie] A first thing to note here is that both CPs function very differently: one to make an assertion, the other to restrict its scope. In this sense, this is not a recursive structure in the above sense of self-similarity. In any case, recursion of this form of embedding is sharply limited. Adding a conditional modifier to the conditional clause yields an ungrammatical structure: (66)

*[CP1 [CP2 [CP3 If the shop opens] if John buys apples] Mary can make a pie] (intended: ‘If the shop opens, John will buy apples. If he does it, Mary can make a pie.’—i.e. CP3 is a conditional clause restricting CP2, which in turn restricts CP1)

This again instantiates a restriction we found above in (45–7), where D-within-D shows that not even a single such round of ‘recursion’ can happen without material intervening between the embedding and embedded occurrence of D, and hence without its happening cyclically. On the other hand, it might be argued that we find unbounded recursion in the domain of verbs, in the cases of verbs of several closed classes: (67)

a. John [wanted [to start [to cause Bill [to die]]]] b. John [wanted [to want [to want [to want [to want . . . ]]]]]

Does (67a) illustrate verbs directly embedding other verbs, without running through other, non-verbal categories in between? This would not hold in cartographic views, where aspectual, modal, and other verbs with this type of behaviour are treated as functional (or semi-lexical), rather than lexical categories. This implies that they all 21 Haegemann (2009) and Arsenijević (2008) argue that (64) is in fact a universal structure, even where the structure appears to be (63).

112


belong to different categories, and hence that these examples do not exhibit a case of recursion of the same in the same. Even for examples like (67b), at least the category of the non-finite inflection, call it TP, intervenes between the instances of the modal category involved. If there is also vP below each TP, or if TP itself is a cycle, the putative recursion is indirect and mediated by cycle boundaries. Moving on to the domain of verbs, consider serial verb constructions. But only a very limited number of verbs may appear in one serial verb construction. Moreover, it is of a templatic nature: not by accident, the term used for it includes the notion of a ‘construction’. Two different types of serial verb constructions can be recognized on the background of our discussion so far. One is the type in which the serialization involves the interpretation of coordination between the serialized verbs (which often include restructuring verbs such as aspectuals and modals). (68) Sira la'o hakru'uk deit. Tetun-Dili they walk bend.over just ‘They walked just bent over.’ lit., according to most analyses: ‘They walked and bent over (simultaneously)’ [Williams-van Klinken et al. (2001)] These constructions are similar to the English type He tried and opened the bottle or She went and read the book. The analysis of this type of serialization is most likely to involve a segment of a functional sequence (for the restructuring members of the series), and/or coordination, thus avoiding a direct embedding of elements of the category V in other elements of the same category. There are, however, more complex serial verb constructions, with more than one, and potentially infinitely many, lexical (i.e. non-restructuring) verbal roots. Pawley (2006) proposes a classification with three ways in which verbal roots can combine into complex verbs. The first is compact serialization, in which verbal roots are mutually causally related, illustrated in (69a). The second is what he calls narrative serialization, in which different verbs, which may be primitive or complex, form a flat sequence on the temporal line; an example is given in (69b). The third is what he calls verb-adjunct construction, in which although apparently two verbs combine, one of the verbs acts as a modifier, as in (69c); this asymmetry can be verified by testing how the verbs interact with aspect marking, negation and other functional categories of the verbal domain. (69)

a. pk sugstrike extinguished ‘put out (a fire)’

Kalam

b. . . . kuñp ognap tb d am katp-at okok l-l, . . . kuñp(leaves) some cut get go house-area about put-SS:PRIOR ‘ . . . having gathered kuñp leaves and taken to the house and put (them), . . . ’

3. The content of grammar c. Sawan [guglum ag-ig] Sawan snoring say-SS.SIM ‘Sawan is asleep, snoring.’ [Pawley (2006: 8–11)]

113

k-jp. sleep-PRES.PROG<3SG>

Compact serialization is clearly templatic: verbs compose under the template of telicity, or causation. While it is possible that compact serial verbs take part in deriving other compact serial verbs, this has to be mediated by lexicalization, in the same way as argued for noun compounds below. This is supported by the general tendency of this type of serialization to lead to lexicalization of the resulting complex verbs, whereby they acquire idiosyncratic meanings. The verb adjunct construction is not even a proper case of serialization, as it forms asymmetric pairs in which one member is phrasal. Narrative serialization is close to a flat multiple coordination: series of temporally ordered independent events are realized, crucially without a structural way to mark a closer relation between certain subseries. Pawley himself notes that, with respect to the analysis of all three types of serial verb constructions in Kalam, ‘there are certain advantages in a construction-based treatment’ (Pawley, 2006: 3)—hinting again at the bounded rather than infinitely recursive nature of their generation. What about the appearance of two projections, both labelled P, within an expanded functional sequence of a prepositional phrase (PP)? Does this not indicate a form of recursion qua embedding of the same in the same, within a single phase? An instance is the recursion of den Dikken’s (2009) locational and directional PPs: [PPDir . . . [PPLoc . . . ]]. But the arguments from above apply here as well. There are still only exactly two instances of the PP category here, and moreover, the two instances are not really of the same category, as they are sub-specified for the kind of spatial meaning they contribute: one is locational, the other directional. Labelling them LocP and DirP would in fact be more appropriate—with the category P linked to the entire domain. A clear case of self-embedding however appears to be (70), in the domain of nouns: (70) [N[N[N[NwarN filmN] studioN] committeeN] sessionN] Perhaps, here there simply is no complex underlying structure (for instance with reduced predicative relatives), there is no functional structure intervening before N embeds in N, and there is simply a set of bare nominals that enters a strictly recursive structure. The prediction that our model here makes, however, is that in this case we should get no referential meanings: the form of generation that we see is not the one that goes with category changes at phase boundaries and concomitant part–whole relations. And that seems to be correct: the form of embedding that we see is compounding in the service of generating new quasi-lexical concepts, rather than grammatical expressions that can be used to refer. This is why every round of

114


embedding applies between two elements of the same category, and produces another entity of the same category, effectively deriving a binary version of the arithmetical successor function with which Chomsky (2008) analogizes recursion/ Merge in language. Relatedly, in this case, semantic compositionality applies. While the case just discussed involves seemingly direct embedding of one lexical category in another, there are also well-known cases of seemingly direct embedding of one functional (or edge) category in another of the same kind. A case in point is demonstrative doubling and determiner spreading in Greek (Panagiotidis, 2008: 451):22 (71) Afto to vivlio. this the book ‘This book.’ (72)

to vivlio to kokino. the book the red ‘The red book.’

Greek

Greek

However, note that (73), which doubles the same determiner, is out, as is (74), which reverses the order of the two determiners in (71): (73) *to to vivlio (74)

*to afto vivlio

We are looking, then, at the option in Greek of ‘doubling the edge’, with the demonstrative extending the ‘extended’ projection of the head noun. The more deictic item is more at the edge, taking the more weakly deictic element as a complement. This order, which we would expect on the present model, cannot be reversed. The order, in short, must be such that the deictic effect increases as we move to the edge, and the extended projection has again an essentially fixed, irreversible, and templatic structure, serving to make reference more specific and rule out error through misidentification. Moreover, a single referent is being computed, and thus for the phenomenon of determiner doubling there is no corresponding phenomenon of ‘multiple definiteness’ or ‘reference-doubling’. Overall, then, nothing changes in the principles we have discussed. That a single referent is being computed is equally true in (71), where again we see that the effect of determiner spreading is to make a (single) referent available twice, with a second deictic act using a more narrowed-down description. The intensionality of XPs embedded in other XPs is illustrated here, since ‘the red’ has no independent referent, showing that the two DPs are not on a par. While the Greek edge is certainly extended further, in short, it is still not the case that it is recursion of the same in that edge. No different conclusion

22

We thank for Andreas Pantazatos for data and discussion.


115

emerges from Complementizer doubling in Spanish with ‘way of speaking/asking’ verbs, where for example we find (75): (75)

Me preguntó que qué quería. CL ask.3SG that what want.1/3SG ‘He/she asked me (*that) what did I/he/she want.’ [Demonte and Fernández-Soriano (2009: 30)]

Peninsular Spanish

Crucially, the two Complementizers are here of different kinds, one being interrogative/wh, the other not, breaking the symmetry. Moreover, such sequences are again non-recursive, and Brucart (1993) puts one of the Cs in Spec-C, the other in C, showing they are not on a par. Demonte and Fernández-Soriano (2009) place the whelement in FocP and the declarative que ‘that’ in ForceP. In summary, a wide range of examples including potential counterexamples to our argument suggest that within the domain of grammar, there simply is no embedding of the same within the same, and hence no recursion in the sense of self-embedding, which is what our approach to the phasal template predicts, but a Merge-based approach does not. So far, the facts of recursion follow directly from the basic idea that the grammar builds referents from predicates at phasal edges, and that all structure-building and interpretation is constrained by the single-cycle constraint. This model predicts that there can be no recursion in the sense of self-similarity internal to the phase, and nothing like this can happen when phases embed in others, either. This is what the above data show, and the stipulation of an unbounded recursive operation Merge adds nothing to these insights.

3.7 Conclusions This chapter has aimed to depict the content of grammar as distinct in kind from the content of the lexicon and as a better candidate for a human invariance than an autonomous component of ‘syntax’. An invariance in the latter would be a surprising fact, and comparative syntax, focusing on such a component, has struggled to obtain a principled account of such variation (see Chapter 5). Although the assumption that grammar has no content of its own has never been axiomatic in generative grammar, it has exerted a powerful force on theorizing, with recursion in a purely formal sense moving to centre stage especially in Minimalism, where much of the explanatory burden of grammar is shifted to the interfaces. Of these, the semantic one is now widely thought to have pride of place. The more forcefully we believe in a fully equipped system of thought on the non-linguistic side of a semantic ‘interface’, the more a conception of grammar reduced to Merge, Agree, and generic principles such as computational efficiency will seem warranted and a form-content dichotomy viable. Yet, the relevant comparative data for the existence of the relevant systems of non-linguistic thought are not there, and it is unclear how conceptually coherent it is

116


to even search for them, when the distinctions we need seem to be of an intrinsically grammatical nature, as is suggested by the fact that the formal ontology we have discussed arises by phase, under the constraints of a particular linguistic architecture (and see the next chapter for a more specific grammatical analysis of this formal ontology). For content that depends on grammar, the form-content dichotomy becomes conceptually incoherent. Rather than distinguishing form and content, then, an appropriate research programme is to distinguish content that depends on grammatical form, from content that does not. This is nothing other than the attempt to delineate the notion of grammatical meaning. That the arrival of grammatical organization in the hominin brain makes no difference at all to what meanings we can grasp as a species, seems unlikely. Lexical content in particular, obeys fundamentally different principles than grammar, with lexical units representing pockets of memorized information about patterns of experience. Human language, by contrast, is a reference system: a device for carrying out at will, and comprehending, acts targeting referents in an infinite space of possible thought that exhibits a formal ontology. We have argued that the formal classes involved do not exist in a metaphysical realm where they can be inspected sub specie aeterni. Rather, they are an aspect of the creativity of language use, in which acts of reference become possible that are not causally or perceptually controlled, but instead controlled internally, by grammar, on an occasion of language use. It makes sense, therefore, to regard as the smallest unit of grammatical organization the unit of deictic significance or of reference, and to reconstruct the recursivity of grammar on this basis. If we do so, a number of contingent facts that recursion exhibits in human language are predicted and find a rationale. Where the content of grammar is shifted to the interfaces, by contrast, we are left with a formal abstraction of grammar in which the universality of semantics is simply assumed, while universals of grammar now become either empty (reducing e.g. to ‘recursion’ or ‘hierarchy’) or hard to see, since they have been separated from the ways in which they are used, becoming arbitrary. The semantic universals in turn become hard to explain. Again, the key to this problem is to distinguish those aspects of semantics that find a grammatical explanation from those others that do not. To the former, we have argued, belongs the availability of a reference system, which a Merge-based account of grammar together with the principle of semantic compositionality cannot, on our view, explain.

4 Deriving the formal ontology of language 4.1 The grammar of reference We have argued that any content word acquires its forms of referentiality on an occasion only as a function of the grammatical relations into which it comes to enter, rather than in virtue of its lexical specifications, which do not predict its referential functioning (with ‘compositionality’ as no solution to this problem, we have suggested). That reference is not lexical is shown by the fact alone that any word that can be used to refer can also be used to predicate, an insight that, as we have illustrated above, even applies to indexicals and the personal pronouns, which are reckoned the most deictic of devices in language. Reference is in this sense a grammatical category for us, part of what we have called the ‘content’ of grammar, as is its inverse, predicativity. Reference moreover appears as the most basic grammatical category, around which all other grammatical functions revolve, with the grammatical process itself terminating in the assignment of a truth value. But we haven’t shown yet how exactly reference arises from grammar. None of the above would be of much interest if, after a word enters the grammatical process, it could become referential in just any way, or in ways we cannot correlate with its specific grammatical behaviour. Instead, however, what we will see in this chapter is that the grammar narrowly constrains the ways in which words can be used to refer, making available a small number of discrete options in which this can happen, organized in the form of a hierarchy, ranging from purely predicative nominals to quantificational, to referential, to deictic, and finally to personal ones. Higher forms of referentiality in this sequence entail lower ones, and none are lexical. Again, nothing in semantics per se predicts either of the constraints in question or their hierarchical organization. We will describe these options and how they arise, as induced by a simple principle: the gradual movement of lexical material to the ‘edge’ of the phase. In this sense, we say that reference is induced ‘topologically’. The term ‘topology’ indicates the possible transformations (or deformations) of a given geometrical

118


object, while leaving its basic identity intact (invariant under the transformations in question). In the case of grammatical reference, the geometrical object is the phase. Its internal composition and the ways in which this composition can change induce, we shall see, distinctions that we can then describe formal-ontologically, and that we find formalized in one way or another in virtually any formal semantic theory of natural language, like the distinction between a proposition and a property, or an event and a truth value. We will conclude that the grammar of human language organizes a formal ontology of this kind, which as such provides a framework of thought: a system in which thought can arise. This topological approach to reference builds on and reformulates pioneering work of Longobardi (1994; 2005), which we here expand and implement in a way that directly connects with the phasal approach we have taken in Chapter 3. How the grammatically available forms of reference for a given word are induced is also independent of its so-called ‘lexical category’, i.e. its part-of-speech status (in particular, whether it is a noun or a verb). Noun-headed structures are surely different from verb-headed ones—for example in that the former are necessarily contained in the latter, and only the latter are truth-evaluable. However, this is not a consequence of a lexical ‘N-feature’ or some lexically specified ‘reference feature’, but of the grammatical fact that the nominal phase is normally the first in the grammatical process and the clausal one is the last. Moreover, as we shall see, the way in which different kinds of referentiality are induced grammatically for both kinds of lexical heads is the same. Reference is a consequence of the phasal topology, not the categorical label ‘Noun’, which we have characterized in Chapter 2 as a morpholexical manifestation of grammatical relations in which a particular lexical item tends to stand. In a similar way, the traditional term ‘adjective’ encodes words primarily functioning predicatively, being confined to the ‘interior’ of the phase; and ‘verb’ is a traditional term denoting words that can enter thematic, aspectual, and propositional relations. A consequence of the present approach, thus, is that no other category than the one basic category of referentiality (and its inverse, predicativity) is in fact needed: the so-called ‘lexical’ categories, N(oun), V(erb), and A(djective) can all be reduced to language-specific morpholexical markings of universally available underlying grammatical relations needed to establish forms of reference. Speaking of formally distinct types of referents such as ‘objects’ or ‘propositions’ is to use the traditional philosophical and ontological terms, and talk of ‘reference to’ such ‘entities’ has traditionally been taken to induce ‘ontological commitments’ to entities of these various formal types. Given the present grammatical approach, the situation is different: the burden of this chapter is to illustrate how acts of reference involving the formal ontology in question are governed by narrowly grammatical principles. They therefore induce ontological commitments only within a formal frame given by grammar. Once this system is in place, existence of entities falling

4. Deriving the formal ontology of language

119

under the formal types in question can be discussed, and existence—say, of the Eiffel tower, or Bolt’s winning in the Olympics—is not then a question of grammar any more. The system of formal categories itself, however, into which these existents fit, is not independent of grammar. The logic of the chapter is to firstly embed Longobardi’s (2005) Topological Mapping Theory (TMT) into our phasal architecture in Section 4.2, and then to expand it to the referential possibilities that are configured grammatically in the domain of clauses, showing that the principles for establishing forms of reference are the exact same ones there, following Sheehan and Hinzen (2011), in Section 4.3. We then propose a further expansion to the domain of pronouns and clitics, which we here, based on Martín and Hinzen (2012), argue in Section 4.4 to go strictly beyond the ontology that we see in the domain of 3rd-person object reference tackled in the previous section. In line with our approach above, also the forms of reference exhibited by indexicals crucially cannot be understood or explained (nongrammatically) semantically. In fact, with the personal pronouns, the referential potential of grammar comes to a peak, and the same topological mapping principles are exploited here again. In this way, the chapter traces the genesis of referentiality from its weakest form—purely predicative and non-scope taking nominals—all the way to its strongest form—the referentiality of the 1st-person pronoun ‘I’. Since the semantics of the personal pronouns connects very directly to foundational issues in the philosophy of mind and epistemology, notably the existence of ‘selves’ as objects of reference, the extension of the TMT into this domain goes a long way towards establishing the goal of this book at large: to demonstrate the epistemological significance of grammar.1

4.2 The Topological Mapping Theory (TMT) If the grammar didn’t operate on a given lexical concept, the concept would never become referential. It could be activated by sensory exposure to a certain object; but the whole point about the emergence of lexical items is that they can be activated in the absence of a sensory trigger. Yet, in this case, they would be useless if no mechanism was available that turned them into referential expressions. This mechanism, we have suggested, is grammar. But it is a matter of observation that in any human language, the grammar can make a given lexical concept referential in different ways and to different degrees. Thus, take the lexical concept ‘man’. In (1), this concept is grammaticalized in such a way that no more than an abstraction, manhood, is involved, while a natural interpretation of (2) is that John is a cannibal, liking man-meat (cf. John likes beef ): 1 We are deeply grateful to Pino Longobardi for inspiring this project and for continued conversations on it.

120


(1)

a. John is man enough to solve this problem. b. John is more man than boy.

(2)

John likes man.

In (3), in turn, singular reference is to a particular person under the name of ‘John’, but ‘man’ comes to denote no more than a property ascribed to this individual, and the indefinite ‘a’ has no ‘specific’ reading: (3)

John is a man.

This is different in the non-copular construction in (4), where ‘a man’ is an argument rather than a predicate, succeeds in taking scope, and denotes a particular individual classified as a man, though the speaker may not be familiar with him and no specific man needs to be referred to: (4)

John likes a man.

If we add marking for grammatical Number instead of the indefinite article, as in (5) below, ‘man’ comes to denote a class of individual men, though no particular individual men are being referred to, making the expression generic: (5)

John likes men.

In (6a), the generic interpretation remains available, but a predicative one now becomes additionally available, where John is man-hunting, as it were. While the nominal is specified for Number in the morphology in this case, this morphology is not interpreted (thus, (6a) may still be true when John finds that one man is enough). If we add a numeral as in (6b), a generic interpretation, according to which John is looking for groups consisting of three men each in general, remains available, and reference is still not to specific men, in the way it is when we add ‘the’ or ‘these’ to the numeral, but rather to an arbitrary set of three men: (6)

a. John is looking for men. b. John is looking for three men.

As we replace the weak determiner ‘a’ in (3)–(4) with ‘the’, which is ‘strong’ in the terms of Zamparelli (2000: ch. 2), it again may be that merely the property of ‘being the man’ is denoted, as in (7): (7)

John is the man.2

2

There is an idiomatic reading here, ‘John is cool’, which is not intended.


121

This predicative reading becomes impossible in the non-copular constructions in (8), where the nominal occurs in an argument-position (object or subject), showing that ‘the’ imposes stronger constraints on referentiality: (8)

a. John likes the man/the men/the three men. b. The man/men/three men arrived.

That ‘the man’ truly is still predicative in (7) is demonstrated by pronominalization and cliticization patterns. Thus, to the question ‘who is the man?’, it is natural to answer in English using the neuter pronoun: ‘it’s John’, but not: ‘he’s John’ (see Mikkelsen, 2005). In a language like Catalan, too, a DP of this nature is cliticized by the neuter clitic ‘ho’, which is impossible in referential contexts, a point to which we return in Section 4.4: (9)

Cliticization of non-referential nominals: a. En Pere és el Porter The Peter is the Porter ‘Peter is the porter’

Cat

b. En Pere ho és The Peter neu is ‘Peter is (the porter)’ If ‘man’ appears without any definite article, but still in a referential position such as that of subject as in (10), predicative or generic readings are also completely impossible, and the only possible reading is an evolutionary one (as in ‘man arrived 150k years ago’), where ‘man’ denotes a kind (and there is a more bizarre one according to which man-meat arrived—after the beef, say, at a cannibal party): (10)

Man arrived.

Where ‘the’ in the copular construction in (7) is strengthened to ‘this’, the predicative reading is out in this position as well, and the speaker will necessarily make deictic reference to a particular individual in the context: (11)

John is this man.

How can we make sense of this bewildering variety of ways in which a given, completely standard nominal such as ‘man’ can be used to refer? We will list four basic conclusions from the above few examples. First, the very same lexical nominal (‘content word’), as it enters the grammatical process, can end up referring in very different ways, where each of these ways is characterized, not by any change in the lexical concept involved, but a change in the formal ontology of the referent. In particular, this referent can formally be:

122 (i) (ii) (iii) (iv) (v) (vi) (vii) (viii) (ix)

The Philosophy of Universal Grammar an abstraction (1) an arbitrary amount of a kind (2) an arbitrary number of individual instances of a kind (6a) a set of sets with a particular cardinality each (6b) a kind (5, 10) a property (3, 7) a particular individual instance of a kind potentially unknown to the speaker (4) a particular individual instance of a kind known to the speaker (8, 11) a particular set within the set of sets with a particular cardinality containing instances of a kind (8).

The second observational fact is that which of these options is chosen co-varies with grammatical factors, such as whether a determiner is present, how strong the determiner is, and which grammatical position in the sentence the nominal is in. By contrast, there are no semantic or other non-grammatical factors that govern these various uses. The third observational fact is that in terms of referentiality, some of them are ‘stronger’ than others, with reference to abstractions, non-individuated substances, and properties being weak. The fourth observation is that as reference gets stronger, we also see grammar recruiting more resources in addition to the nominal (NP) projection itself, such as grammatical Number, a numeral attached to the nominal, a determiner attached to the nominal in addition to a potential numeral, a deictic element as present in a demonstrative, and so on, with each new layer containing the previous expansion as proper parts (cf. Zamparelli, 2000; Pereltsvaig, 2006): (12)

DP D⬘ D

NumP NumP⬘

the

NumP

NP ...

three men These observations raise an immediate question about proper names, which, although they normally occur without a determiner in languages like English, have the semantics of individually specific referential DPs, and hence are strongly referential. Moreover, how can proper names be paradigmatic arguments, if arguments require a determiner position (Stowell, 1989; Szabolsci, 1994)? The exact same


123

question arises for pronouns, demonstratives, and clitics, all of which normally occur without a determiner, but of course standardly have individually specific referential uses, and for kind-referring nominals, as in (10). The first to have tackled this important puzzle systematically seems to have been Longobardi (1994; 2001; 2005), and we believe that the significance of these results is even deeper in the context of an approach to the phase as a unit of deictic reference in grammar. In essence, the proposal is that the forms of reference that exist in grammar can be read off topologically from how the phase is internally organized and transformed. How this happens is, as anticipated in the overview above, moreover independent of ‘lexical category’ in the traditional sense of, say, Baker (2003), opening up the way to a purely grammatical rather than lexical account of reference. Longobardi’s (2005: 9) account of object reference states the following: Topological Mapping Theory (TMT): Object reference if and only if N-to-D movement. Longobardi argues that, in Italian, object-reference of proper names can occur in two ways. Either the proper name surfaces after nominal modifiers such as possessors, in which case there is also an article in D (13a), forming a CHAIN with the nominal; or there is no article, in which case the TMT applies and the name must move to D (by substitution) overtly, explaining why it now precedes what would normally be a prenominal modifier in (13b). (13)

a. [DP Il mio the my

[NP Gianni ]] . . . Gianni

b. [DP Gianni mio [NP tGianni ]] . . . c. *[DP D mio [NP Gianni ]] . . .

(expletive-associate CHAIN) (overt movement chain) (no PF-visible chain)

As Longobardi (2005: 26–7) shows, ‘il’ in (13a) functions differently from the definite article and can differ in morphology from the latter in other languages such as Catalan. For English, the principle of N-to-D movement when a determiner is absent remains valid in the case of object-reference, yet now the movement is obligatorily covert. This explains why (14a) is grammatical and is object-referential, why (14b), in which the movement is overt, is ungrammatical, and why (14c) derives a descriptive reading: (14)

a. [DP D old [NP John ]] came in b. *[DP John old [NP John ]] came in

(covert movement) (overt movement)

c. #[DP The old [NP John ]] came in These data are the inverse of the Italian ones. D must remain phonologically empty for object reference in English to occur, so that it can be targeted by covert N-substitution, making it inconsistent with a descriptive reading. In terms of the organization of the phase as described in the previous chapter, repeated in (15):

124 (15)

The Philosophy of Universal Grammar [edge [interior]]

this means that (i) the edge needs to be projected for object-reference to occur, (ii) the nominal head of the lexical projection NP in the ‘interior’ has to be copied to the edge, thereby emptying the interior of lexical content. This ‘movement to the edge’ involves no lexical change: the exact same lexical item is involved whether or not the movement takes place. Yet, if it does, reference becomes objectual; if it doesn’t, and functional material is projected in the edge, such as grammatical number or a determiner and a restriction, then object reference disappears, as in (16–19), proving again that this particular ontology is not lexical but grammatical: (16)

The Gianni I fell in love with in Rome isn’t the one you know.

(17)

I have many Giannis in my class this year.

(18)

Ci sono due Gianni nelle there are two Giannis in.the ‘There are two Giannis in my class.’

(19)

Conosco tre Marie know.1s three Marys ‘I know three Marys.’

mia classe. my class

Ital

The same data show that names need not be of ‘type ’, any more than pronouns or clitics do. This would in turn explain an empirical fact about proper names, which has enthralled philosophers since the landmark set by Kripke (1972): in their most typical uses, the descriptive properties that we associate with the referent of a particular name need not co-determine the identity of the referent as picked out by its proper name. Thus, for example, if we think of Goedel under the description ‘the discoverer of the Incompleteness Theorems’, then Goedel, if referred to as ‘Goedel’, remains the exact same individual, even if it turns out that it was Schmidt who discovered the theorems and Goedel stole them. The explanation for this crucial fact is that if we have no more to play with than the single phase with its basic bipartite structure, and the nominal head of the lexical projection in the interior of the phase is copied to the edge, then there is nothing in N that could still enter into the determination of the identity of the referent: there is no nominal restriction in this position anymore. In fact, we predict that not even the fact that Mary is called ‘Mary’ needs to enter. In other words, we predict that if Mary is referred to as ‘Mary’, and ‘Mary’ is in the edge, the referent will be what it is even if it turns out that the person in question is not called ‘Mary’. In Kripkean terms, the reference is thus predicted to be ‘rigidified’. If, say, Mary turns out to really be called Billy, or she is called Mary but is in fact not a woman, then the identity of the referent will not change. Rigidity, in this sense, is a grammatical phenomenon.


125

In turn, if the name ‘Mary’ stays in N, we predict that an interpretation will result that exploits whatever little descriptive information a nominal head that is a proper name entails: namely, that the person is called ‘Mary’. ‘Mary’ in (19), say, therefore, denotes ‘being called “Mary” ’. Such readings for proper names get easier, the more a given name triggers descriptive associations (as when the person denoted by the name is famous). These associations then allow us to form a new predicate on the spot, which will then be lexicalized through the relevant name that stays in N. One could, obviously, if one wanted, spell out the rigidity of proper names in their object-referential uses as ‘constancy of reference across all possible worlds in which the referent exists’, yet this won’t change the basic claim above: that the relevant causal mechanism that gives rise to rigidity and the special referential properties of proper names is solely grammatical. Constancy across possible worlds expresses a grammatical fact in (naturalistically controversial) metaphysical terms. Despite these benefits, Longobardi’s topological hypothesis faces a serious problem. This is because Longobardi (1994) explicitly analyses DPs of the form [the [NP]] as ‘quantificational’, i.e. as having the basic Russellian (1905) semantics in (20b,c), where ‘∃:’ denotes the second-order property ‘there is a unique x’: (20)

a. the man b. the x (man x) c. ∃:x (man x)

The problem with this proposal is that, as has been widely discussed since Donnellan (1966), all definite descriptions such as (20a) can have referential uses. This is not true for indefinite descriptions, as we see when comparing (21) with (22) and (23): (21)

I like a man

(22)

I like the man

(23)

The man is insane.

In (21), ‘a man’ is a scope-taking nominal, which entails that it allows both for readings in which there is a particular man that I like, and in which no definite man is referenced, and reference is merely to an x such that x is a man. Although, in the first case, a specific individual is contextually involved, it is not being referred to in any direct way (which can be the very point of such an utterance, as when a woman happens to talk to the actual man she likes, but wants to hide it). Such reference, on the other hand, does take place in both (22) and (23), where the nonspecific or attributive reading becomes unavailable in (23) and is only marginally possible in (22). For example, in a context where the choice is between a man and a woman (say during a dance), you go for the former, although you have no idea who he might be. (22) can then also trigger a generic reading, according to which you always go for the man, especially if there is intonational stress on ‘man’. The

126


attributive reading re-appears, on the other hand, when we make the descriptive content of the NP richer, as in (24), uttered when standing near the scene of a horrendous murder crime: (24) The man who did this is insane. On the basis of these data one could argue, contrary to a long-standing tradition, that the referential use of definite descriptions is in fact the primary one, while the ‘attributive’ one is secondary: for the latter is available only where some property describing a certain referent is sufficiently salient, as in (24), or where a nominal is in an predicative position, as in (25a), or where it co-varies with the value assigned to a variable (25b): (25)

a. I am the man. b. In any ordinary household, when the woman cooks, the man does the dishes.

Sometimes, the very same definite description can occur twice in a sentence, one time interpreted referentially, the other predicatively, depending on its grammatical position: (26)

I wished her husband wasn’t her husband.

(26), from Lycan (2008), is coherent as an assertion only if, in the first occurrence, the description ‘being her husband’ is used to pick out a particular referent, and if this description is not a condition on the identity of the referent. For, in this case, the speaker can coherently use the occurrence of the same definite description in the predicative position to point out that it would be nicer if the individual referred to in the subject position was free of the descriptive property in question. This option disappears again, where either the descriptive property is too involved and the descriptive reading is forced in both positions (*I wished the man who committed this horrible crime wasn’t the man who committed this terrible crime), or the property is not properly ascribed, as in epithets (cf. *I wished the idiot wasn’t the idiot). In sum, indefinite descriptions can, apart from being predicative, be specific, but they cannot yet be object-referential. Definite descriptions can be object-referential, but can remain predicative or quantificational as well, if relevant conditions obtain. If they do, the descriptive content of the nominal that is the complement of the definite determiner enters into reference-fixation, but still not the identity of the referent (a person picked out as a husband need not, in every possible world, be a husband). While we observe a scale of increased referentiality in the transition from indefinite to definite descriptions, it would thus be wrong to conclude that definite descriptions, in and of themselves, have a fixed semantics: for their referential properties depend both on the NP-restriction and their grammatical and non-grammatical context, exactly as in the case of proper names. The question whether ‘definite descriptions’ are referential or attributive therefore cannot be answered: it is meaningless. For it is a question


127

about a phrase; but the question concerns how that phrase functions, which is a matter of the grammatical relations in which it occurs. While definite descriptions, unlike indefinite ones, are primarily referential, but can have predicative uses, complex demonstratives start to resist descriptive or predicative uses, explaining why (27) has a completely different meaning from (26): (27)

I wished her husband wasn’t this husband.

Nonetheless, as has been noted in King (2001) and Elbourne (2008), even complex demonstratives still have non-referential uses under certain restricted conditions. Consider the narrative context in (28) and the bound-variable context in (29): (28)

So I am in this bar, and there is this guy who tells me about his wife . . .

(29)

Whenever Mary makes a knot in her handkerchief, this knot is forgotten when it would be most important.

The speaker in (28) is not exactly pointing to any particular individual, as when we meet a person on the street and the same speaker tells me: ‘I met this guy in the bar’.3 Note, however, that none of the occurrences of ‘this’ in (28–9) is consistent with intonational stress on ‘this’; and in none of them, the NP-complement of the demonstrative determiner can be dropped or be replaced by a deictic pronoun like ‘him’, showing that the NP-restriction is the primary reference-fixer in these cases. Where the determiner is stressed in a complex demonstrative, as in (30), or where the NP-complement is dropped, descriptive/predicative and bound-variable readings disappear: (30)

I met this guy in a bar.

(31)

I met him in a bar.

These data suggest that we also cannot speak of a uniform Russellian semantics for definite descriptions and proper names (and perhaps encompassing demonstratives), or of a dual or ‘ambiguous’ semantics.4 In all the cases we have considered—bare nominals, names, definite determiners, demonstratives, and pronouns—referentiality is scalar: it can be purely predicative, quantificational/scope-taking, referential, or deictic, depending on the grammatical context. These options are ordered hierarchically on a scale, and none is lexically determined. Moreover, there is a clear topological mapping principle for each of these four options. Thus, in the first case, purely predicative nominals, no edge is projected, as in (32), which is equivalent to (33): (32)

I am hunting unicorns.

(33)

I am unicorn-hunting.

3

In line with this, complex demonstratives of the former type appear in the existential construction ‘There is this guy I know’. 4 Here we are indebted to conversations with Tom Hughes.

128


For this reason, referentiality is determined purely via the descriptive content of the nominal. Since no object is being referred to, (32) and (33) can be true even if there are no unicorns: there is no existence presupposition, the intended reading is merely generic, and both expressions are non-specific regarding Number, although Number is morphologically present in (32). If, in turn, an edge is projected for a nominal argument, as arguably in (34) (which is equivalent to ‘I am eating some lamb’), but it is left empty, the nominal retains obligatory narrow scope, no individual ontology is projected, and no individual-specific readings are possible. Yet an existence presupposition now arises: (34) cannot be true if there is no lamb: (34)

I am eating [x [lamb]].

If the edge is projected and filled, but it is weak, as in (35), the nominal can take both narrow and wide scope, making the sentence false in the latter case, if there are no unicorns: (35)

a. I am hunting [a [unicorn]]. b. I saw [NUM [unicorn-s]]

If the edge is projected and filled, and it gets stronger, scope is necessarily wide, making the sentence always false: (36)

I am hunting [the [unicorn]]

This becomes even clearer where the edge gets even stronger, through the addition of a deictic element, assuming with Leu (2008) that ‘this’ morphologically decomposes into the definite determiner and a deictic element (‘the’+‘here’): (37)

I am hunting [this [unicorn]]

And if (38) was ever asserted, and there are no unicorns, we would have to conclude that the referent was identified under the wrong description: (38) This unicorn is dead. Finally, where the edge becomes so strong that the NP-complement and the descriptive content encoded in it can be dropped, a maximally referential and minimally descriptive form of referentiality is reached: (39)

This is dead.

(40) It/he is dead. In summary, the hierarchy of nominal reference in question is that depicted in (41), where ‘<’ means ‘is referentially weaker than’ or (equivalently) ‘is more dependent on the lexical content of the descriptive predicate’: (41)

(*the) *(NP) < *(a) *(NP) < *(the) *(NP) < *(this) (NP) < *(he) (*NP)


129

Proper names represent a case exactly intermediate between stage 3 and 4 of the progression in (41): there is no lexical determiner, yet the nominal vacates the interior to occupy the edge. The general principle that increased referentiality corresponds to the phase becoming more ‘edge-heavy’ is thus vindicated for this case, too, along with the fact that increased referentiality correlates with increased grammatical resources. The same follows from any approach to pronouns on which these are inserted directly in the edge, as opposed to moving there from N (see Longobardi, 2008, and below, Section 4.4). The same follows for the case of rigid reference to kinds. Thus, consider that in (42) and (43), the (a) options simply cannot mean what the (b) options do: (42)

a. Man comes from Africa. b. The man comes from Africa.

(43)

a. Men are brutal. b. The men are brutal.

In other words, reference to kinds is inconsistent with the definite article, except in cases such as the generic one in (44), where English allows an expletive article deprived of its usual referential meaning: (44) The whale is a mammal. Again this follows if, in (42), ‘man’ is base-generated in the phase-interior, and then moves to the edge—except if the expletive option is available, as in (44), which kicks in with ‘whale’, where the lack of an article would normally trigger mass-referential readings (cf. ‘I ate whale’). The availability of this expletive article appears to be subject to an essentially arbitrary cross-linguistic variation, as we would expect from a lexical idiosyncrasy. Thus, either sentence in (45) is ungrammatical in English: (45) a. *The man is a mammal. b. *Whale comes from Africa. And while we can say (46) in German, where ‘die Callas’ functions virtually as a proper name and ‘die’ ceases to function as a normal definite or demonstrative article, ‘der’ in (47) retains a deictic dimension, pointing to the singer called ‘Domingo’ (who is not quite as famous as La Callas): (46)

Die Callas ist grossartig. ‘the Callas is great’

(47)

Der Domingo ist grossartig. ‘the Domingo is great’

130


Where the lexical determiner is not recycled for the expletive option, obligatorily covert N-to-D movement explains why kind-denoting NPs must be bare, as in (42) above or (48) below, whereas in Italian, where for lexically idiosyncratic reasons an expletive determiner is more generally available, kind-denoting NPs cannot be bare: (48)

(*the) apples are/milk is good for you (* with a kind reading)

(49)

a. I love (*the) good wine. b. Amo *(il) buon love.1s the good ‘I love good wine’

vino wine

Ita [based on Longobardi (1994: 631)]

While the bare NP in (49a) can receive a kind reading, its Italian equivalent in (49b) cannot. In Italian, bare NPs receive only an existential reading, which corresponds to a DP configuration in which the null determiner binds a variable restricted by NP. In the present context, this confirms a general and very important point: the form of referentiality that is topologically derived within the nominal phase has nothing to do with the lexical status of proper names, or the semantic fact that names are given to individual persons/objects. The kinds rigidly referred to in (42), (48), or (49) are not individual persons or objects. Yet the principle, by which ‘object-reference’ is established in these cases, according to the TMT, is exactly the same as in the case of lexical proper names. The point is simply that, since the lexical common noun involved in these cases is not lexically attached to any particular individual, the ‘object-reference’ in question comes out as reference to a (natural) kind. A kind, therefore, as referred to in this fashion, is nothing other than the objectual correlate that the grammar can compute for a lexical common noun. As we shall now see in the next section, this insight is in fact broader: for in the case of clauses, the grammar can generate a clausal analogue of ‘object-referential’ expressions as well. But due to the verb-based nature of clauses, the ‘objects’ in question will come out as quite different in nature: namely, as what we call (talking in semantic mode) truth values or facts. This would confirm once more that what we are targeting here—the nature and origin of language-specific forms of reference—is strictly independent of lexical category. We turn to this issue in the next section, which is based on Sheehan and Hinzen (2011) (henceforth, Shee&H).

4.3 Extending the TMT Shee&H propose (50): (50) The Extended TMT: Object-reference in clauses is derived by V-to-C movement.


131

If so, what really matters for object-reference is again not any lexical labels or features such as ‘N’ or ‘D’, but simply the naked topology of the phase, in which there is a basic bi-partite division into a predicative core (the interior) and a referential periphery (the edge), irrespective of lexical category specifications: (51)

[referential edge [predicative interior]]

The ‘interior’ provides for the descriptive content of a lexical concept that, in the course of the grammatical structure-building process, is converted into an expression that is made referential in one of the grammatically available ways. In (52), this basic template is instantiated in two grammatically equivalent ways: (52)

a. [referential the [predicative kings of France]] b. [referential that [predicative there are kings of France]]

Just as the NP provides a descriptive condition for the identity of the referent mapped from ‘the’ at the edge of (52a), TP provides such a description for the referent mapped from ‘that’ at the edge of (52b). This makes a direct prediction: if ‘that’ in the edge of the clause regulates the reference of clauses, there should again be two extremes: a weakest case, in which the edge is not projected at all; a stronger case in which it is projected, but remains empty; and a yet stronger case, in which it must be filled— either, we expect, through a clausal equivalent of the Italian ‘expletive’ definite article, or through T-to-C movement. Shee&H present a broad range of data in support of this prediction. For clauses, the strongest form of reference that is available is reference to a truth value: there is nothing more fundamental or extensional than truth. Truer than true, nothing is, and the truth of a sentence, if it obtains, holds no matter what: in particular, all referential terms in it can be replaced by co-extensional ones salva veritate. Since nothing is ever asserted to be false, moreover,5 this truthvalue is always truth. Reference to the truth in this sense is achieved only at the root, i.e. in matrix clauses. In that case, we see that the Complementizer is obligatorily absent in a language like English (53), exactly as the determiner in the case of object (as well as kind) reference (54): (53)

(*That) John left.

(54)

(*The) John

As noted in Shee&H (2011: 27), this absence cannot be due to a general ban on Complementizers in matrix assertions, since V2 languages provide clear evidence that matrix clauses can be CPs. Rather, (53) would be explained on more principled grounds if (53) patterned with (54) in that, in English, the head in the interior of the 5 ‘It is false that John left’ asserts, of course, that it is true that it is false that John left, i.e. that John didn’t leave.

132


phase (V/T) must move to the edge (C) covertly in the case of truth-reference, a movement that any overt Complementizer would block. In line with this hypothesis, we observe that if a non-expletive Complementizer is overt in Catalan as in (55), or in French, as in (56), readings are derived in which no truth value is asserted (both examples have subjunctive morphology usually found in embedded clauses):6 (55)

Que sempre hagi de fer el dinar jo That always have.3SG.SUBJ of do.INF lunch I.NOM ‘That I have to always do lunch!’

Catalan

(56) Que Jean soit malade de la tuberculose en 2003! French That Jean be.SUBJ ill of the tuberculosis in 2003 ‘For Jean to be sick with tuberculosis in 2003!’ [Schlenker (2005: 10)] We expect these clauses not to be truth denoting, unless an expletive Complementizer is lexically available that forms the clausal equivalent of the ‘expletive’ definite article in Italian nominals, as in Gascon (57): (57) *(Que) soi that be.1S ‘I am Gascon’

Gascon Gascon

Gascon [Campos (1992)]

‘Que’ in this language appears to be precisely restricted to finite matrix clauses in which a truth is asserted. V-initial and V2 languages provide other evidence, to which we return shortly, that V-to-C movement indeed can take place overtly as well in truth-denoting clauses, paralleling the Italian nominal case. This derives, via the phasal model, an old Fregean intuition: that (assertoric) sentences are proper names, just derived ones. They, too, are object-denoting, with truth as the traditional name for the ‘object’ denoted. The weakest form of reference available for a finite clause is where the clause is merely a proposition: a possible truth. That is, the speaker takes no attitude towards the truth-value of the clause. We predict that in this case, the edge can be empty, as it indeed can be in English: (58)

I believe/say/suppose (that) John left.

The further prediction is that where clauses lack a CP layer altogether, such as clauses occurring as complements of raising, exceptional case-marking (ECM), and restructuring verbs, they fail to gain independent referential status. No properly bi-phasal structure is generated, allowing Case to be assigned from beyond the clause boundary in ECM (see further Chapter 6).

6 For a range of other examples from other non-V2 languages, including Spanish, Catalan, Romanian, and Portuguese, see Shee&H (2011: 29–31).


133

Finally, just as definite descriptions have always oscillated in philosophers’ intuitions in regards to whether they are referential or quantificational, we obtain an intermediate case in the clausal instance too. Clauses differ from nominals in that the latter are normally embedded as arguments. They can be vocatives, of course, but nonetheless, a nominal such as ‘the man’ cannot normally stand on its own, as a selfstanding utterance. Clauses, however, can be embedded as arguments as much as they can be unembedded. A clause that has been evaluated for truth by the speaker and occurs as an argument is factive (fact-referring). We predict from the above that, in this case, the edge must be filled. Vindicating this prediction is complicated by a number of factors, and in particular the heterogeneity in the class of factive verbs. As Shee&H argue, however, on the basis of a typology of such clauses, the purest case of factive verbs is that of the so-called ‘emotive factives’. In this case, the complementizer is obligatory, as predicted, exactly as the determiner is obligatory in the case of definite but not indefinite DPs: (59)

I regret/resent/ignore/neglect *(that) John left

Factives, thus, are the clausal analogue of definite descriptions: these, as illustrated above, can be referential; but they necessarily have a predicative core (NP-complement), which enters the act of reference. In a similar way, factive clauses can be viewed as referential expressions, but they nonetheless never rigidly refer to truth values, as only matrix clauses do: the speaker asserting (59) does not assert that John left.7 Rather, the embedded clause is still a grammatical predicate, like all other arguments are, when they are embedded in higher phases. Thus, in kill Bill, Bill is a predicate (the Theme) of an object, an event, which is derived at the edge of v. Similarly, in (59), that John left is a description of a content identifying a particular mental event that is referenced by the matrix verb. Expectedly, factive clauses also constitute an intermediate case with regards to intensionality effects: if Lois Lane resents that Superman left, she need not resent that Clark Kent left, even if these two are, on some intuitions, the same fact. But if Superman left, then Clark Kent left, showing that reference is rigid and maximally extensional in the matrix case. Factive clauses are as truth-denoting as things can get, in human language, in an argument position, given the constraint argued for in Chapter 3 that only whole sentences can denote truth-values. Factive clauses share other properties in common with definite DPs, such as a presupposition of existence. This presupposition cannot be cancelled: (60)

John cares/is glad that the world is flat, *but it isn’t.

(61)

Jeanne d’Arc talked to the king of France, *but he didn’t exist.

7 This is to share with Melvold (1991) and Haegeman and Ürögdi (2010) the intuition that referentiality, not factivity per se, is the crucial explanatory notion underlying the phenomena traditionally discussed under the heading of ‘factivity’.

134


This proposal must crucially not be confused with an older one, according to which factive CPs are in fact nominals. According to this proposal, C is a nominal head in these instances, and the clauses in question contain a silent ‘fact’-nominal (Kiparsky and Kiparsky, 1970). This suggestion is natural where factive clauses are (rightly) diagnosed as referential, but referentiality is (wrongly) diagnosed as a consequence of lexical category (‘nominal’), which we have argued in Chapter 2 is a confusion of lexical and grammatical semantics. As we have seen, there is nothing necessarily referential about nominals. Moreover, what speaks against the older proposal is that the distribution of factive clauses is not the same as that of definite DPs, as also discussed in Picallo (2002), Ambridge and Goldberg (2008), Haegeman and Ürögdi (2010), and Betti (2011). Consider, for example, (62) and (63): (62)

I am surprised (*at)/aware (*of ) that you feel that way.

(63)

I am surprised *(at)/aware *(of ) that.

Moreover, factive clauses behave differently regarding extraction: they are only weak islands, whereas complex NPs are strong islands. We return to extraction below. Let us summarize, then, the ‘Extended TMT’ as proposed in Shee&H, which associates a threefold formal ontology with the clause: (64)

The Extended Topological Mapping Theory: Object-reference iff C is substituted by V/T.

This entails that, in the absence of such movement—or a relevant PF-visible (Phonetic Form-visible) CHAIN—clausal reference will be to a mere proposition: an object that can be possibly true and possible false, but is not evaluated yet. Where there is movement to the edge (or CHAIN-formation), this, by contrast, yields reference to a truth (a derived proper name), or fact (subordinate case). Again, the hypothesis is that this movement can happen overtly or covertly depending on arbitrary externalization factors, as in the nominal case: in the clausal case, the movement is covert in English but overt in V2 languages, yielding in both cases a rigid interpretation in which the descriptive condition expressed by TP does not fix the identity of the referent (even if it helps to determine or identify it). This proposal meshes well with the well-established fact that in V2 languages such as German, Dutch, Icelandic, Yiddish, Swedish, Norwegian, Old French, and Kashmiri the V/T-to-C chain is required to be established overtly in matrix truth-denoting clauses (cf. Den Besten, 1981; Vikner, 1995; Diesing, 1990 on Yiddish; Bhatt and Yoon, 1999, on Kashmiri; and Holmberg, 2010, for an overview). The fact that this involves movement of V/T into the C-domain is suggested by the position of adverbials/ negation (65b), and the fact that the presence of another XP in preverbal position licenses subject-verb inversion (65c):

4. Deriving the formal ontology of language (65) a. laRkan boy.ERG

por read.PAST

135

Kashmiri

akhbaar newspaper

b. az por laRkan akhbaar today read.PAST boy.ERG newspaper c. akhbaar newspaper

por laRkan az read.PAST boy.ERG today

[Bhatt and Yoon (1999: 48)]

While a number of influential approaches to V2 (Weerman, 1988; Truckenbrodt, 2006; Julien, 2010), have already taken this movement to be semantically motivated, Shee&H go beyond previous accounts of V2 by providing not only a plausible semantic correlate for the head-movement involved, but also by assimilating it to head-movement in the nominal domain, with parallel semantic implications: constant reference irrespective of descriptive conditions. As noted earlier, the analysis of truth-evaluated clauses in subordinated positions—intermediate between truth-reference in the matrix case and proposition-reference in the subordinate case—faces a number of empirical problems due to the heterogeneity of the domain of factive clauses. Common to all of these types of clauses, however, is that no truth is literally asserted in these subordinate positions: the proposition encoded in the subordinate clause is not asserted as true by the speaker, though it may be asserted by the matrix subject and is presupposed as true by the speaker; but equally, that no mere proposition is denoted either, in the sense of something that can still be either true or false, i.e. has not been evaluated yet by the speaker. But the class of factives is usefully split into a narrower one compromising the so-called ‘emotive factives’ (e.g. regret, resent, mind, care) (Melvold, 1991), which exemplify the phenomenon in the purest terms, and a broader class of ‘semi-factives’, which can lose their factivity in some contexts (Karttunen, 1971), and include the sometimes so-called ‘cognitive semi-factives’ (e.g. know, discover, find out, forget) and the ‘communication semi-factives’ (e.g. disclose, divulge, confess, reveal). As Shee&H show, careful attention to syntactic differences in the behaviour of the complements of true factives, semi-factives, and non-factives shows that true factivity is a distinct syntactic type. The emotive factive verbs are distinctive precisely in selecting a single type of complement, the ‘definite CP’ representing fact-reference. An interesting phenomenon that helps to tease factive and semi-factive subordinate clauses apart is ‘slifting’, in the sense of Ross (1973). Consider (66) vs (67):8 (66)

the class is cancelled, he said/affirmed/revealed/I assume/believe/discovered/ found out.

(67)

*The class is cancelled, he regrets/resents/cares. 8

Note that (66), with ‘that’, is possible if read as a topicalization structure.

136


While subordinated (cf. Grimshaw, 2010), ‘slifted’ clauses form a kind of intermediate case, carrying some assertoric force indexed to the speaker, which is why the speaker cannot felicitously cancel the assertion involved: (68)

The class is cancelled, he said/affirmed/revealed, #but it’s not.

In other words, slifted clauses are (at least partially) asserted (hence truth-denoting) clauses, and it is revealing that the most paradigmatic factive verbs (i.e. the ‘emotive’ ones) are impossible in this case (cf. (67)). This confirms a crucial commitment of the above mapping hypothesis for clauses: factive clauses are fact denoting, not truthdenoting; their truth is presupposed or referred to, but not asserted. At the same time, slifting allows us to tease factives and semi-factives apart, since the latter are possible in this instance (cf. (66)). The complementizer must be dropped in instances of slifting, as it often can be with semi-factives, unlike factives: (69)

He discovered/knew/realized the man had left.

(70)

?*He resented/cared the man had left.

Again, this is predicted by the present approach, since the Complementizer should be obligatory only in the true factive case. If, by contrast, in the case of an assertion of truth, V moves to C, this movement would in turn explain why, in the instance of slifting, an empty C-position turns out not to be subject to Empty Category Principle (ECP) effects, which, in traditional terms, would ban the occurrence of phases with an empty edge from occurring in ‘ungoverned’ positions. As we have seen, English bare common nouns, unlike Italian ones, can circumvent this problem, as (71) illustrates: (71)

Apples are good for you.

This possibility follows if covert N-to-D movement takes place in these instances, voiding the ECP effect and yielding the rigid-referential reading that, in Italian, requires the expletive determiner to be overtly present. Why, then, can a slifted clause occur in preverbal position, as in (66), when preverbal CP-subjects usually require an overt Complementizer, which however is banned in (66)? Again, this would naturally follow if, due to the assertive nature of the clauses in question, V-to-C has taken place, voiding the ECP effect in this instance as well, and yielding truth-denoting readings. Slifting, in being restricted to asserted clauses, further illustrates that if a clause becomes truth-referring (rather than fact-referring), it cannot be extracted from, illustrating a more widely attested difficulty for phases with stronger forms of referentiality to be ‘penetrable’ from the outside: (72)

*What kind of car Fred would buy was Mary saying? [Grimshaw, (2010: 6)]


137

Embedded assertive clauses are widely agreed to be strong islands for extraction (Holmberg, 2010).9 By contrast, fact-referring embedded clauses are generally assumed to only be weak islands, allowing object extraction (to a degree) but not adjunct extraction, as illustrated in (73–4): (73)

?What car do you mind that Fred bought?

(74)

*When do you mind that he arrived?

Proposition-referring clauses (indefinites), in turn, the most weakly referential case, are most generous in what extractions they allow: (75)

When do you believe that he arrived?

These facts are natural under the phasal model. As has long been recognized, referentiality is a crucial factor constraining extraction. In nominals, specific definites are hardest to extract from. If referentiality is computed in the phase—and there is nowhere else it could be computed—and greater referentiality correlates with a more expanded or edge-heavy phase, then it is natural to expect that the heavier the edge and the more expanded the phase, the harder sub-extraction would be. Hence, in the clausal case, we expect that assertoric or truth-referring clauses are hardest to extract from, and proposition-referring indefinites should be easiest to extract from.10 A second way of teasing semi-factives and factives apart is via embedded root phenomena, which, in the terms of Bentzen et al. (2007: 14), are governed by the ‘Assertion Hypothesis’: ‘the more asserted (the less presupposed) the complement is, the more compatible it is with V2 (and other root phenomena)’. Consider in this light embedded V2 in (76): (76)

9

a. Han sa/innså at I universitetstida hadde han Norwegian he said/realized that in university-time.the had he vært veldig ambisiøs. been very ambitious ‘He said/discovered that, in his university days, he had been very ambitious.’

But not in German, according to Vikner (1995), which is a mystery. Nonetheless, the parallel is not perfect, as indefinite DPs are still well known to be weak islands, rather than non-islands (Chomsky, 1977), as Shee&H (p. 45) note. We still have no better explanation than that, although referentiality is a purely grammatical category and is established topologically, it will be harder to establish object-referentiality, the more grammatical complexity is built. Clauses are intrinsically more complex than nominals, and it might be that, cognitively, computing an objectual correlate from a proposition—a truth value or fact—is intrinsically more complex than computing one from a property, allowing sub-extraction even where it ceases to be possible in nominals. 10

138

The Philosophy of Universal Grammar b. *Han he hadde had

angret regretted han he

på/benektet on/denied vært veldig been very

at i universitetstida that in university-time.the ambisiøs. ambitious. [Bentzen (2010: 169)]

Again these data follow if verb-movement to the edge yields assertive (truth referring) clauses, whereas the factive complement clause of ‘regret’ in (76b) is fact referential and hence excludes embedded V2, which rather correlates with assertivity. The semi-factive ‘realize’ is compatible with V2, but not the true factive ‘regret’. The same is found in regards to gerundive complements. Emotive factive predicates like non-assertive predicates take these (77), whereas assertive predicates including semifactives do not (78): (77)

I resent/regret/avoid/deny [PRO being wrong].

(78)

*I assume/disclose/know/suppose/say [PRO being right].

This simplifies the data considerably (see Shee&H, pp. 36–7), yet the crucial point remains that gerundive complements are never assertive: to assert a truth, a gerundive complement, which can never form a matrix clause, is insufficient, whereas it is quite sufficient to refer to a fact (such as ‘me being wrong’). Thus, in (77–8), we find factives and semi-factives parting company accordingly. A third observation is that true factives cannot form the ‘Main Point of Utterance’ (MPU), in the sense of Simons (2007), whereas the complements of semi-factive and non-factive verbs can. In (79), corner brackets indicate possible answers to the question posed: (79)

Q. A1. A2. A3.

What’s up with Mary? I think/guess/know [she’s not feeling well]. #[I care/mind that she’s not feeling well]. It’s possible/likely [she’s not feeling well].

Why can the complements of true factives not form the MPU? This follows if, for something to be the MPU, it has to be asserted. If ‘asserted’ means truth-referring, and true factives are fact-referring, it follows that true factive complements cannot form the MPU. Slifting, embedded V2, sub-extraction, gerundive complements, and MPU all exemplify the distinctive behaviour of clauses that become referential in subordinate positions, corresponding to rigidly object-referring nominals in the nominal case, even if, in the case of clauses, reference to truth is ultimately only found in matrix positions. We again conclude that referentiality is a grammatical category, not a lexical one, and it arises from progression from the phase interior to the phase edge, with a number of identical strategies available in both the nominal and clausal domain for


139

how a number of formal-ontologically distinct referents can arise as the grammatical computation proceeds. Reference begins where ‘lexical projections’ start to have an edge in which various functional projections regulate how the words are made to refer. Where this edge is projected insufficiently far (e.g. there is no determiner), we retain a predicative reading. Where the edge starts to fill, quantificational and referential readings are derived. Where it is filled, but not extended far enough— e.g. a deictic element is missing in the nominal case—readings will oscillate between referential and predicative, depending on a number of grammatical factors.11 The culmination of the process is where the descriptive core of the phase is emptied and only the edge remains. This gives us ‘rigid’ readings. We can, as we have, express this progression in the traditional intuitive ontology of metaphysics or semantics—speaking about a progression from ‘properties’ to ‘objects’ to ‘events’ to ‘propositions’ to ‘facts’ to ‘truth-values’, instead. Yet, it appears that nothing would be gained from this move, and reinterpreting what is essentially a grammatical process in metaphysical terms seems to deprive us of the actual explanatory principles involved. Ignoring these principles can be harmless in theoretical endeavours, and the metaphysical one in particular, but it is harmful the moment that we seek to understand how the narrow range of formal-ontological distinctions, which unfailingly accompanies the building of grammatical structures, arises. If we want to naturalize the ontology of semantics, metaphysics is no way out.

4.4 Extending the Extended TMT (with Txuss Martín) But we are not in fact at the end of the hierarchy of reference yet. All we have talked about so far is the formal ontology of 3rd-person object reference. The grammar, as we shall see in this section, crucially goes beyond this ontology. This is immediately surprising from a traditional nominalist point of view, where language is an arbitrary representational system: if it is, how could there be something beyond object reference? Isn’t a world consisting of objects all there is? As we will show, it isn’t, and nothing other than grammar takes us beyond what we have got so far. 4.4.1 From 3rd-person object reference to pronouns The new realm we have to enter is traditionally characterized under the label of ‘indexicality’. When it became clear, in the 1970s, that proper names have a special, ‘direct’ form of referentiality (Kripke, 1972), which the semantics of definite descriptions cannot emulate, the real challenge was still waiting in the wings: how do demonstratives such as ‘this’ or the personal pronouns refer? Do they really refer 11

Consistent with this approach, it has recently been argued that Classifier-NP constructions in Mandarin, Swahili, or Nguni are not referential (Ndayiragije, 2011; Wu and Bodomo, 2009), contrary to claims by Cheng and Sybesma (1999) according to which Cl substitutes for D in these languages.

140


in no different way than proper names do—by ‘directly referring’ to their referents? Are they object-referential in the same way as proper names, the only difference being that context is needed to figure out what their respective objects are, on an occasion of language use? What, however, could the semantics of indexicals be, if it does not reduce to some sort of direct, though perhaps contextually mediated, relation between word and object, which we already see in the case of proper names? Since demonstratives and pronouns are grammatically quite distinct from other nominals, our approach in this book makes us expect that their semantics will be different as well, and that it falls under the rubric of grammatical not lexical semantics. So the distinction between indexicals and proper names, in particular, should be a semantically and epistemologically significant distinction. Moreover, the semantic differences in question should cohere with our nominal mapping topology so far. If 3rd-person object-reference (as found in proper names) is not the key to the semantics of deictic pronouns and demonstratives, this should show in the topology of pronominals. A natural expectation is that the further extended the nominal edge is, and the higher an element is merged or moved up in the left periphery of the nominal phase, the less descriptive content will play a role, making the languageworld connection even tighter. We pursue this intuition in the rest of this chapter and focus particularly on clitics. The search for a distinctive character in the semantics of indexicals makes us recall the phenomenon of ‘essential indexicality’ (Perry, 1993; Lewis, 1983), discussed by philosophers for some decades. As the name says, it consists in the suggestion that there is something ‘essential’—or irreplaceable—about indexicals in their most ordinary uses. To take a classical example, for Perry’s (1993) famous amnesiac Rudolf Lingens, who is lost in the Stanford library and does not know who he is, the sentences (80) and (81) do not mean the same thing: (80)

Lingens is at Stanford

(81) I am at Stanford. Lingens, having read a biography of Lingens while in the library, knows that (80) is true. But he does not therefore know (81). Nor does he know (82): (82) This is the Stanford library. Similarly, if I look at myself in the mirror, without realizing it is me, then (83) could be mildly amusing, while (84) would not be, and might instead represent a genuine discovery (Kaplan, 1977): (83)

His/that guy’s pants are on fire.

(84)

My pants are on fire.


141

Indexicals such as deictic pronouns or demonstratives, then, seem to make a distinctive difference to the meaning of a sentence, and they couldn’t be replaced by definite descriptions or proper names. If so, this would rationalize the important fact that such devices apparently exist in all human languages. The rationale would be that they could not be eliminated without semantic loss. If they cannot be eliminated without semantic loss, it would be more than a slight pragmatic inconvenience if we lacked these devices. Rather, if we resorted to names and descriptions instead, the organization of meaning would change. But in fact, it is indeed rather unclear in what way eliminating indexicals would really be a pragmatic inconvenience: although it is true that proper names are not always known or remembered, a logically possible human language could be such that ‘temporary’ proper names would be used in this case, like ‘dude/bloke’, or ‘dude1’, ‘dude2’, etc. It is thus doubtful that the rationale for indexicals is pragmatic. We also see pronouns and clitics to be particularly sensitive to grammatical formatives such as Case and Person-features, which again suggests they are an inherent part of the grammatical system, rather than forming a curious set of lexical items that only make a pragmatic difference. While essential indexicality provides a nice rationale for the existence of indexicals, however, it needs to be explained. Why, or in virtue of what, does it obtain? Again, even if the phenomenon in question is properly characterized as a semantic rather than pragmatic one, semantics as such makes no predictions for why it should exist. The core meaning of ‘semantic’ is that it denotes relations to the world: content. However, in terms of relations to the world, nothing changes when we compare pronouns with any other nominals: there is a word/phrase, on the one side of the semantic relation, and an object/thing, on the other, in both cases. A person is the same external object, whether I refer to it as ‘you’, as ‘the reader of this book’, or by using her proper name. In the latter two cases, different lexical concepts are involved, but this has no impact on what object it is semantically or externally. This is a problem for the view of indexicals as ‘semantically underspecified’ or ‘contextdependent’ expressions. This view suggests that underspecification and contextdependence could be eliminable by simply always having relevant proper names or uniquely identifying descriptions ready to hand: that is, by simply fully specifying the reference. Ensuring this would be merely pragmatic inconvenience. If indexicality is semantically essential, therefore, the underspecification or context-dependence view does not seem to be correct. It also does not seem to help with explanation if, as in Heim and Kratzer (1998), pronouns are formalized as indices, which do nothing other than encode (nongrammatical) reference and have no semantic content besides. Insofar as pronouns have ç-features (and hence are not like indices, which presumably lack them), these features are taken to define partial identity functions on the individuals of the domain of quantification, encoding presuppositions for reference (cf. Heim, 2008): if the individual satisfies the ç-feature in question (e.g. by being male), the individual is

142


mapped onto itself; otherwise the function is undefined, and reference fails. This is to treat ç-features as not different in principle from any other nominal restriction on a determiner. In line with that, a long tradition maintains that 3rd-person pronouns such as ‘he’ project phrases such as (85) that are essentially similar to DPs, containing a D-head taking an NP-restriction, with an additional restriction provided by ç-features (cf. Elbourne, 2005, 2008; Roelofson, 2008; Johnson, 2011): (85)

DP DP D he

NP

{Person, Number, Gender}

This structure again makes us expect no semantic difference between such pronouns and definite descriptions, as on the semantic underspecification view above. The solution to this dilemma in the present context is obvious: if there is a semantic difference, but semantics as such does not predict it and standard semantic models do not deliver one, the difference must be one of grammatical semantics: how meaningful structures function in grammar, viewed as a reference system.12 Indeed, the most obvious empirical difference between pairs like (80–1) is a grammatical one, to do with grammatical Person: while proper names are always grammatically 3rd Person, ‘I’ is 1st Person. This same difference is found in (83)–(84). In (82), on the other hand, the most obvious difference is the presence of a (proximal) deictic element in ‘this’ that is absent in ‘the Stanford library’. But the problem is that, traditionally, the semantics of reference seems to abstract from the notion of grammatical Person and from the deictic element in demonstratives. For example, in Kaplan’s (1989a, b) classical approach to ‘directly referential expressions’, these are singular terms whose only semantic contribution to the proposition expressed is said to be the object referred to. This object is a referent. Distinctions of grammatical Person or in how it is being referred to make no difference to it or to the propositional content that it determines. The same is true where a 1st- or 2nd-person feature is said to be a ‘presupposition’ on reference (Heim, 2008): again, this doesn’t change the object qua external object of reference, if the presupposition is satisfied. The same is true again where demonstratives and pronouns are identified with definite descriptions. Thus, Elbourne (2008)

12 Indeed, note that strictly speaking, the semantic term ‘indexical’, if viewed as denoting a particular class of words, is itself a conceptual error. The term ‘indexicality’ indicates a way in which a given word grammatically functions. Indexicals viewed as words do not need to function in this fashion (given for example predicative or bound uses that they also allow).


143

suggests that a complex demonstrative such as this cat expresses an ‘individual concept’. Such an entity is presumably not marked for grammatical Person: it is an abstract semantic or model-theoretic entity. This individual concept is then analysed as containing an ‘index’, i, and a relational concept, R: [[[that i] R] cat]. But the index is said to be, say, Felix himself, if Felix is the cat referred to it (Elbourne, 2008: 433). It is irrelevant to this referent, qua referent, that it is the referent of a deictic act involving a demonstrative. If language didn’t exist, it would be the exact same referent that it is. There is, then, an obvious grammatical difference between our example sentences above, but it does not connect in any natural way with standard accounts of pronominal or deictic reference in the semantic literature. But can it be plausible that language makes no difference to what we can refer to? Might its structures not be constitutive for some of the kinds of objects we refer to, so that these would not exist without grammar as a device for referring to them? Might reference to ‘selves’—our ‘Egos’—be a case in point? Selves, in this case, would inherently be linguistic objects: without being language users and mastering grammar, we would not have (or be) selves (though we could have or be animal selves). Can it be accidental to the existence of rational selves that they are language users and need to master the system of grammatical Person?13 Zahavi (2006: 27–9) stresses that self-awareness involved in conscious states cannot be construed along ‘subject-object’ lines, yet again, it appears to be exactly the grammar of self-reference that tells us how this is the case: the grammar of 3rd-person object reference (‘the/my self ’, ‘my body/brain’) is fundamentally different from that of 1st-person reference (‘I’; see e.g. Bianchi, 2006). We would worry about an English-speaking child who never referred to itself in the grammatical 1st person. A real possibility, then, arises on the horizon: that the formal ontology of grammatical reference, which begins with predicative forms of reference, ends with deictic and personal such forms, which are constitutive for the semantics involved. Summing up the situation, the existence of indexicals as a class of expressions poses an explanatory puzzle: (i) it seems to relate to the organization of meaning rather than being merely pragmatic; (ii) clear grammatical differences accompany the elements of this class; yet, (iii), these grammatical differences must be meaningfully related to the apparent semantic differences, ideally explaining them in a principled

13 While the idea that rational selves are, inherently, language-users, seems reasonable enough, the role of language for the existence of what philosophical discussions refer to as ‘the self ’ is barely considered in current discussions of the notion of ‘self ’, which are striking in their near-complete omission of any consideration of the significance of language or a linguistic perspective (Sass and Parnas, 2003; Chung et al., 2007; Kircher and David, 2003; Zahavi, 2006; Lowe, 2008). The self is said to be the referent of the 1stperson pronoun, but there is virtually no reflection of the fact that this is, in fact, a grammatical category, defined through its contrast with the grammatical 2nd and 3rd Person. Clearly, ‘first’, ‘second’, etc. have no non-grammatical meanings here. We return to this issue in Section 8.5.

144


fashion. In the rest of this chapter we will therefore cast a fresh look at the grammar and meaning of Person and deixis. As we shall see in the next sub-section, the personal pronouns are, in fact, referential in a distinctly different and stronger way than 3rd-person pronouns. In Subsection 4.4.3, we argue that this difference is not a non-grammatical semantic (e.g. lexical) one. In Subsection 4.4.4, we show, through a detailed study of Romance object clitics, that it also involves an increase in grammatical complexity relative to all other nominals, involving an expansion of the nominal left edge, as well as, crucially, Case. Case and ç-features encode a relational semantics, as mirrored morphologically in Agreement relations. This is not reflected in standard semantic approaches to pronouns, in the sense that there is absolutely nothing relational about deictic pronouns, since there is nothing relational about a word referring directly to an object. Nor does the standard presuppositional semantics of ç-features make any predictions for such relations. In line with such a semantics, even standard Minimalist syntax takes Case and ç-features as paradigmatically ‘uninterpretable’. But they are only uninterpretable if we take ‘semantics’ in the traditional core sense of dealing with relations between words and things. They are not, if we take seriously what Agreement patterns in language tell us: that Case and Person features express relations between event-participants and speech-participants (Sigurðsson, 2004). The semantics of Person and Case is a grammatical semantics, not a lexical one. 4.4.2 Reference and Person Let us adopt the traditional view that 3rd-person nominals are in fact no-person nominals (Benveniste, 1966). In other words, these are nominals that have not yet entered the Person system in grammar. What, then, does grammatical Person add, when it comes into play? Let us recall our above hierarchy of increasing referentiality as the phasal edge becomes more ‘heavy’, repeated here as (86): (86)

(*the) *(NP) < *(a) *(NP) < *(the) *(NP) < *(this) (NP) < *(he) (*NP)

Pronouns specified for Person (the ‘personal pronouns’) clearly behave like 3rdperson pronouns at the top of this scale, insofar as they, too, do not require a nominal restriction (87a–c), unlike the definite determiner, and indeed cannot co-occur with one. Only plural personal person pronouns behave differently in this respect (87d), but as has long since been argued, at least the so-called 1st-person plural is not really a 1st Person:14

14 Thus, ‘we’ certainly does not refer to multiple ‘I’s (1st Persons), something that seems inherently contradictory. Cf. Benveniste (1966), or Boas (1911: 39): a ‘true first person plural is impossible, because there can never be more than one self ’. It could be that an analogous observation holds for the 2nd Person plural.

4. Deriving the formal ontology of language (87)

145

a. *he man b. *I boy c. *you boy d. we linguists

However, while 3rd-person pronouns take a relative modifier (and acquire a purely quantificational rather than referential reading in this case), the personal pronouns don’t: (88)

a. He who enters this room will be shot. b. *I who enter(s) this room will be shot.

Moreover, while it is a paradigmatic property of 3rd-person pronouns that, in appropriate contexts, they can act as bound variables, this is not true for personal pronouns, which resist binding. That they can nonetheless allow it, is often asserted on the basis of examples such as (89), where ‘my’ is said to have a bound reading to the effect that nobody else around here can take care of his or her own children. Another example is (90), which is said to have a bound reading to the effect that no one else would admit that he is wrong: (89)

I’m the only one around here who can take care of my children. (Kratzer, 2009)

(90)

I’m the only one around here who will admit that I could be wrong. (Partee, 1989)

What we notice here, however, is that, on the intended reading of (89), if it is felicitous at all, ‘my’, can be exchanged for the 3rd-person pronoun ‘her’, without any change in meaning. Similarly, in (90), the second occurrence of ‘I’ could simply be exchanged for ‘he’. This logically entails that pronouns, where they seem to be bound, fail to be personal pronouns except morphologically. Even morphology itself points us to this conclusion. Thus, Kratzer notes that, in English, in order for a variable reading to be available at all in a dependent clause, the agreement morphology of its subject must reflect that of the clause it depends on. Thus, in (91), where this is not the case, the bound reading disappears: (91)

I am the only one around here who is willing to admit that I am wrong.

In German, due to its different agreement morphology, bound variable readings of structures in a translation of (90) are unavailable in the first place (Heim, 2005): (92) Ich bin hier die einzige, die zuzugeben bereit ist, dass ich unrecht habe. I am here the only one who to admit ready is that I wrong am ‘I am the only one here who is willing to admit that I am wrong.’ Just as in (87d) above, bound variables become available again when we switch from 1st person singular to 1st person plural, which again is expected if the 1st Person plural is not really a 1st Person:

146


(93) Wir sind hier die einzigen, die zuzugeben bereit sind, dass wir unrecht haben. We are here the only ones who to admit ready are that we wrong are But for one of us (the German one), no bound variable reading really emerges either in this case: the second ‘we’ only has a ‘collective’ interpretation. Consider in this regard also the ‘partial binding’ example in (94), from Rullmann (2004): (94)

Only you prepared a handout for our first appointment.

If we imagine Kratzer using (94) to address a former student of hers at a reunion, there is a reading that she could paraphrase as: ‘you were the only x such that x prepared a handout for the first conversation of x and me’ (Kratzer, 2009: 190). Here, while the plural our indeed has a bound variable interpretation, its 1st-person feature obviously cannot come from an agreement chain originating with you, and its interpretation is not that of a variable. In other words, insofar as we can dissociate the Number and the Person feature of bound plural (so-called) 1st-person pronouns, the Person features are not bound in these instances, as shown by the fact that there is no antecedent in (94) with appropriate features that could act as a binder. The same observation can be made in relation to Romance clitics and their capacity to express bound readings: (95)

Bound readings of Catalan and Spanish clitics: a. *Un / el parisenci compra plaça de parking a the parisian buy.3s spot of parking si ell#i/j proi/j té cotxe if he pro have.3s car INTENDED: ‘A Parisian buys a parking spot if he has a car’

Cat

b. *[Cada uno de nosotros]i dijo que mei gusta each one of we said.3s that cl.1s like.1s INTENDED: ‘Each of us said: “I liked” ’

Sp

c. [Cada uno de nosotros]i dijo que lei gusta each one of we said.3s that cl.3s like.1s ‘Each of us said he liked it’

Sp

d. *?[Cada uno de nosotros]i dijimos que nosi gusta each one of we said.1p that cl.1p like.1s INTENDED: ‘Each of us said he liked it’ (6¼ ‘Each of us said WE like it’) e. Nosotrosi dijimos que nosi We said.1p that cl.1p INTENDED: ‘We said we like it’

gusta like.1s

In (95a), from Picallo (2007), the bound reading can only be provided by the null pronoun. (95b) shows once more that the 1st-person singular pronoun resists


147

binding, while in (95c), the 1st-person features of the pronoun in the antecedent are interpreted there, while the bound pronoun is a 3rd-person clitic. In (95d), the 1stperson features of the plural clitic resist binding again: there is no different interpretation for each assignment of a value to the variable bound by the quantifier. Although there is a real quantifier in subject position and we might expect a bound reading, we do not obtain one: the sentence is ungrammatical, or at least very marked. (95e), in turn, is grammatical, but the 1st-person plural clitic is not bound, because it refers collectively to the whole plurality ‘nosotros’: ‘we’. The generalization we extract from these data is that singular personal pronouns, unlike 3rd-person pronouns, do not allow an antecedent: they cannot be bound. Insofar as, marginally, it appears as if they can be bound in some languages, their person-features are not interpreted, turning the pronoun effectively into a 3rd-person (= non-person) one. Insofar as Person-features are interpreted, they require referential readings. This is, as we have seen, not so for the other pronouns, nor for demonstratives, proper names, or definite descriptions: none of these require referential readings. This gives us the beginnings of a crucial semantic difference in the personal pronouns, which we would like to relate meaningfully to their grammar. Let us first consider evidence apparently pointing in the opposite direction, however. In some languages such as Dutch, the personal pronouns, unlike the nonpersonal ones, can be freely locally bound in environments where the Binding Theory (Condition B) should rule out such binding: (96) Ik/Jij voelde me/je wegglijden ‘I/You felt myself/yourself slide away’

Dutch

They also seem to allow the ‘sloppy’ reading of (97), rendered explicitly in (98): (97)

Wij voelden ons wegglijden en jullie ook We felt ourselves slide away and you too

(98) We felt ourselves slide away and you felt yourselves slide away too Does this show that, contrary to what we just claimed, personal pronouns can have bound readings? Reuland (2011: 160) claims just that (cf. also Adger, 2011). In the case of (97), Reuland’s conclusion immediately follows, if indeed, as he claims, (99) is the correct formalization of the sloppy reading in (98): (99)

Wij (ºx (x voelden (x wegglijden))) & jullie (ºx (x voelden (x wegglijden)))

There is a completely featureless, ‘bare’ variable involved in the structure of (97), then, which is interpreted as 1st Person in the first conjunct and as 2nd Person in the second. It is not clear, however, how this can be, if the putative ‘bound’ personal pronouns have any feature-specifications, and hence are truly personal pronouns. It is also not clear what the grammatical formative, in (96) or (97), is that might be the equivalent

148


of this ‘variable’, which is a logical notion, not a grammatical one: (99) is a logical formula. That this bare variable would be 1st Person in the first conjunct and 2nd Person in the second would suggest that the two pronouns need to have the Person features of their respective clausal antecedents, where these features are being interpreted. This, however, would seem to be a fact about feature sharing between two pronouns, not necessarily variable binding: sharing a feature entails no grammatical dependency. In that case, there is no variable binding of the pronouns in either (96) or (97). The idea would be that the two pairs of pronouns in (96) share a Person feature value, but there is no binding. This would make sense of the grammatical semantics of (96). The person grammaticalized as feeling the sliding and the person grammaticalized as doing the sliding are distinct in terms of their relational interpretation as event participants, and playing the role of the person doing the speaking is something different again. A Person feature is shared, yes, but it is not clear why this entails binding, or should entail it. (96) may thus not be properly described as a case of ‘co-reference’ or ‘identity’, as is suggested by the traditional notational device of co-indexation. Identity may be better approximated grammatically through the Control-relation, as in ‘I like going to the movies’, or in (100), where grammatically, the subject of ‘like’ and ‘see’ are the same: (100)

I like seeing me in the mirror.

But there is crucially no identity between ‘I’ and ‘me’, in this example, which play fundamentally different roles in order for such a sentence to be coherent at all. Again, potentially, an instance of feature-sharing without feature-binding. Personal pronouns, we conclude, truly resist binding. The cases that have come close are ones in which the person features are those of an antecedent, where they are interpreted. Where there is no antecedent, as e.g. in the partial binding case (94), the variable readings disappear. If there is an antecedent, this is not a case of binding, or there are no person features, except morphologically. Direct referentiality, if so, is an intrinsic aspect of the semantics of the personal pronouns. The same conclusion is suggested by the fact that where a predicative nominal is cliticized in Romance, the clitics involved are not specified for Person, as mentioned in the previous chapter (cf. p. 94 above, and Longobardi, 2008); and by the fact that while all 3rd-person pronominals allow predicative readings, the situation is different for personal pronouns, a point to which we return in the next subsection. Our hierarchy of referentiality in (86) therefore must be expanded by a new layer, thereby coming to a new peak. This layer corresponds to the most referential/deictic form of denotation we have encountered so far. In line with this and our topological mapping principle, we shall see shortly (Section 4.4.4) that the personal pronouns are also grammatically the most complex in terms of the topology of the phase. This again confirms, for the domain of pronominals, the intrinsic connection that we have claimed exists between grammar and referentiality. It would also explain essential


149

indexicality: if the relevant indexicals are grammatically the most complex, and their grammar is meaningful, it is expected that they cannot be replaced by anything that is strictly less complex grammatically, without semantic loss. Before we show this, and in order to motivate such a novel approach to the essential indexical, we will consider a more standard semantic take on the essential indexical and argue that it fails—as it must, if indeed essential indexicality is part of grammatical rather than lexical semantics. 4.4.3 Semantic approaches to the essential indexical Suppose it is a lexical semantic fact about the word ‘I’ that it is to be evaluated in regards to both a ‘world’ w and a more fine-grained part of it, a ‘context’, c, which models the speech situation and in particular contains a function ca that returns the agent of the context (see e.g. Schlenker, 2011). The semantic value of ‘I’ is thus (101): (101) 〚I〛c, w = ca On this basis, the following semantic explanation can be offered for Lingens’ predicament in (80)–(81), repeated here as (102)–(103): (102)

Lingens is at Stanford

(103)

I am at Stanford.

These are now said to be true in exactly the same possible worlds, which means that they have, in this sense, the same content. Lingens believes the proposition expressed by the former because this proposition is true in all worlds compatible with what Lingens believes. However, the proposition expressed by the latter contains the semantic value in (101), and hence has to be evaluated in regards to a context as well. Since Lingens does not know that the agent of the context is Lingens, he does not believe (103). In line with this, he also does not believe (104): (104)

I am Lingens.

This explanation is lexicalist in the sense that it is taken to be a fact about the lexical item ‘I’ that it has the context-sensitive semantic value in (101). Note again that the difference in grammatical Person that distinguishes (102) and (103), as noted above, does not enter into this explanation. If we are right, then, this semantics cannot be correct. And in fact, we will now see that several doubts arise about it. Firstly, arguably the semantic explanation just given simply shifts the explanatory problem. For the next question is now: why do indexicals have to be evaluated in relation to the ‘context’, in a way that other referential expressions do not? Coding this fact through the appropriate semantic value (assuming it is appropriate) is coding (or formalizing) the fact to be explained, not the explanation. Consider the

150


analogous case of proper names. These, arguably, have special semantic properties in comparison to definite descriptions. Surely this is not simply a brute (perhaps conventional) fact about proper names as a particular class of symbols, but asks for some deeper explanation. Again, we cannot obviously say that the explanation is that names are constant in their reference across all possible worlds, for this is the special semantics to be explained. Secondly, the semantic notion of ‘context’ used in the semantic explanation above seems to be non-explanatory: we argue now that it ultimately cannot be characterized non-circularly, i.e. without reference to the indexical elements in question. Consider again that the knowledge that Lingens is said to lack on the above account is that Lingens is the agent of the context: (105)

ºc ºw [ca = Lingens]

Suppose, however, that Lingens’ amnesia involves a further mental disorder: while he is able to make extensive observations of the context he is in, about who acts in it, who speaks, and so on, he nonetheless fails to recognize (106): (106)

I am the agent of the context.

Evidently, most of us do not have this mental disorder, in much the same way that most of us are not amnesic. Some schizophrenics, however, do: they hear voices speaking (auditory linguistic hallucinations), but may fail to recognize which voice is their own, i.e. who it is that is speaking. The fact that we lack this disorder shows that the ability to recognize (106) as correct is, if it obtains, a non-trivial fact about mental health. The point then is that an object known or referred to under a 3rd-Person definite description such as ‘the agent of the context’ or ‘the speaker’ need not be known under the 1st-Person perspective encoded in the word ‘I’, any more than the referent of ‘I’ and the referent of ‘Lingens’ in (104) need to be recognized as the same, even if they extensionally are. It follows that evaluating (104) in relation to an external ‘context’, in addition to a ‘possible world’, is simply not going to do the work required. The context, viewed non-linguistically, does not contain the information needed to recognize (103) as true. The information needed to recognize it as true is not that Lingens is the agent of the context. Rather, it is (104). What is needed is not the semantics of context, but the right grammar. It is both a necessary and a sufficient condition for an agent to recognize that a certain proper name or description holds of himself that he, when assessing whether they hold, refers to himself under a 1stPerson perspective. Reference to oneself under any other perspective is a sign of mental ill-health, as the case of schizophrenia suggests, where the use of pronouns, which unlike nouns in English are systematically marked for Person, is distinctively disturbed (cf. Watson et al., 2012). Going to a context will reveal much about Lingens and much about agents and speakers, but no Self. The referent of the word ‘I’ is not to be found in the context. The 1st-Person pronoun is the irreducible access to this


151

object, and there is no other road to it. The personal pronouns thus are ‘essential’ in a stronger sense than the above semantic model assumes, and they are more essentially grammatical than contextual. This directly connects with our third doubt, anticipated in the beginning of this Section 4.4. It relates to the very notion of a ‘context-sensitive expression’. Is it really true that an essential indexical refers like a proper name does, just in a more contextdependent fashion? Consider that, since a name such as ‘John’ occurs rather frequently, there can always be confusion as to who it applies to: any given context might have several Johns in it that could be being referred to. Moreover, the person referred to might be Bill, but he was referred to as ‘John’. Things get worse for common nouns. If somebody comes into a room and says ‘man’ (the exclamation meaning ‘damn’ is not intended here), we have no idea what or who this refers to. Similarly, in any context, there could be misunderstanding and confusion in regards to what the phrase ‘the book on the table’ refers to. There might be more than one book or table, it might be unclear what qualifies as a book, or something might be mistaken for a book and yet be referred to under this description. The exact opposite is true of the personal pronouns and deictic expressions, where the relevant problems simply do not arise: where the word ‘you’ is used in the place of ‘John’, there is simply no issue of having used the wrong name; or having referenced the wrong person. If a person says ‘I’ rather than ‘man’, we immediately know what (or who) he refers to, and there can be no mistake. Clearly, this is due to the fact that no descriptive lexical content is involved— not even information about gender, which the 1st- and 2nd-Person pronouns seem to lack in most of the world’s languages, and which in the 3rd-Person case could still lead to referencing a person mistakenly as ‘she’ when in fact it is a ‘him’. Since the referent of ‘I’ or ‘you’ is (essentially) not described, no description can fail to apply either.15 The referent of ‘I’ is crucially not the same as that of ‘my body’ (or perhaps ‘my brain’), which picks out a 3rd-Person referent that anyone other than me can refer to as well. A speaker can be mistaken in calling something ‘my body’, but not in calling himself ‘I’. That a descriptive predicate is involved in the one case but not the other explains why my body can change, but we do not take this to entail a change in me. Analogous comments are appropriate for the indexical ‘this’. Therefore, where these are used, there is never a question about who or what in the context these words refer to, and in which there can be no non-trivial criterion for the use of these words, in the sense that we have to check that the criterion applies before using them.16 Nor can the 1st-Person entity referred to as ‘I’ change, for it could only change 15

For a contemporary discussion of this old philosophical point, which we here reduce to a grammatical truth, see e.g. Lowe (2008). Wittgenstein (Tractatus Logico-Philosophicus, 5.632) expressed it by saying: ‘Das Subjekt gehört nicht zur Welt, sondern es ist eine Grenze der Welt.’ (‘The subject doesn’t belong to the world, it is a boundary of the world.’) 16 I don’t first have to check that I am called ‘John’, that I am human, that I am in my usual mood, etc., in order to know what ‘I’ refers to. Neither ‘I’ nor ‘this’ abbreviates a description. This is almost but perhaps

152


if its properties changed; but if no identifying properties are involved in picking it out, none can change either. Even in the case of proper names, let alone definite descriptions, by contrast, there always can be a question of who or what is being referenced: we have to consult the context in order to figure out the referent. In this sense, contextsensitivity arises as soon as—and the more that—a lexical description is involved (if only through the choice of a name), and it systematically disappears when the description is eliminated and reference becomes grammatically more constrained. A fourth and final doubt about the semantic account above is the general fact that, as pointed out throughout this chapter, referentiality is simply never a lexical fact. In every single kind of nominal, referentiality (and attaining specific forms of it) depends on the grammatical relations in which these expressions enter and the topology of their phases. This casts a general doubt on any lexical-semantic explanation of essential indexicality, especially because the generalization just rehearsed is equally true for pronouns, which can all occur in predicative and quantified positions as well, as noted: (107)

a. That’s him. b. He is a she. c. I am not a better person: I am me.17 d. I am not myself today. e. These are the we inside of me.18 f. Sono io. be.1s I ‘It’s me’ [Moro (1997)]

Ita

(107c) is of particular interest. Here the 1st-Person pronoun ‘me’ occurs in a predicative position, and the reading is descriptive (‘I am the person I am, no other (better) person’). But it is crucial to the description ‘the person I am’, which as a whole is of course 3rd-personal, that the 1st-Person feature occurs within it, and that it is purely deictic rather than descriptive. Thus the sentence ‘I am him’, where the 1st-Person feature is not repeated in the predicate, is unsuited to make the claim that is made in (107c). The same applies to the slightly different ‘I am this guy’, which again has a not quite true for ‘you’, which one might argue depends on the notion ‘not me’. If this dependence is asymmetric, ‘I’ does not in turn depend on the notion ‘not you/him’. The absence of criterial properties includes that of existence: while a speaker using ‘I’ in a referential position presupposes his own existence (as with any other referential expression), he does not assert it. This is why prime minister Ahmed Shafik could respond to a question of his whereabouts during a critical period in the Egyptian revolution: ‘I didn’t exist then, it was when I was wiped out.’ (The New York Review of Books LIX: 13 (2012), 25). 17

Hugh Grant, playing a TV producer, speaking to his (soon former) girlfriend in American Dreamz. Brain scientist Jill Bolte Taylor on the different ‘personalities’ processed in the two brain hemispheres: http://www.ted.com/talks/jill_bolte_taylor_s_powerful_stroke_of_insight.html, accessed April 2013. 18


153

different meaning. Neither is the 1st-Person feature in the predicate equivalent in meaning to the 3rd-Person phrase ‘the speaker’, which is a 3rd-Person referent, too. Note also that while the 1st-Person feature can occur within a 3rd-Person expression, as in (108), this is not true the other way around: (108)

[3rd [1st my] body]

(109) *[1st [3rd my body’s] me] This shows an asymmetry between my body and me: I have a body; but my body doesn’t have me. Within grammatical semantics, the referent of the 1st-Person pronoun is not anything near a 3rd-Person referent. In fact, it is no ‘object’ at all, as has been noted in criticisms of Descartes’s substance ontology of the self for three centuries. The philosophical terms ‘the I’, or ‘the Self ’ are therefore misleading and unsuited, as they are 3rd-Person referents encoding reference to an object. The same conclusion follows on Buehler’s (1934) account of the ‘origo’ as it functions within the grammar viewed as a reference system: like the point O in a coordinate system, it is the point from which all object reference originates and relative to which it is determined, while not itself being an object of reference at all. We conclude from these four considerations that no lexical-semantic value and no semantics of ‘context’ will be the key to the problem of the essential indexical: that problem is not one of lexical meaning but of grammatical semantics. Next, and in line with this observation, we will show that the grammar of these expressions is indeed special, and exceeds all other nominals we have encountered so far in grammatical complexity. Moreover, in line with the general approach of this chapter, what we will see is an expanded phasal edge, and a complete independence from lexical semantic meaning as mediating the act of reference. 4.4.4 The hierarchy of reference revealed through Romance object clitics We will now provide evidence that an extended topology applies to indexicals. Above the edge filled by a determiner or by N-to-D movement in the case of referential nominals, indexicals involve a deictic layer, allowing the NP-complement to be absent: the edges of the DPs they head are by this point ‘heavy’ enough to stand on their own, supporting a referential function that does not depend on the presence of any lexical layer with a substantive descriptive content – not even one that has been moved to D, as in the case of proper names, which provides a clue for why indexicals go beyond proper names. Like demonstratives, the personal pronouns engage this deictic layer as well, and they, too, can crucially occur without an NP-restriction, which we argue, against Elbourne (2008), to be lacking completely by this point in the referential hierarchy, rather than merely being deleted in morphophonology but present grammatically or semantically. Unlike demonstratives, however, the personal pronouns also go beyond this deictic layer by engaging the system of grammatical Person— necessarily so, as we have argued above. Where this happens, D-to-D movement

154


within the extended edge of the phase that includes the deictic layer can be observed. We show this increase in grammatical complexity in the nominal phasal edge, with consequences for the need to license the relevant nominals sententially, through a close analysis of Romance object clitics. We will see that some of these clitics have purely predicative interpretations, other (accusative clitics) can be referential, yet others (dative clitics) must be referential and have personal interpretations. In line with this progression towards referentiality within the clitic system, we see an increase in the internal complexity of the clitics concerned—exactly as we must assume if our model of reference as an aspect of grammatical meaning is correct. Our analysis will be consistent with rich typologies developed for different types of pronouns that have already demonstrated that these do not form a unified lexical class (Roca, 1992; 1996; Cardinaletti and Starke, 1999; Déchaine and Wiltschko, 2002; Ormazábal and Romero, 2010; or Martín, 2012; among others). It will also be consistent with the broad topological approach of Longobardi (2008), who already noted that pronouns are ‘edge-elements’ par excellence. In fact, Italian and English cease to differ in regards to the obligatory placement of pronouns in D: (110)

Noi ricchi/*I ricchi noi stiamo trascurando certi problemi We rich/*the rich/*Rich we are neglecting certain issues I poveri imitano noi ricchi/*i ricchi noi The poor imitate us rich/*the rich us/*rich us Noi due/*I due noi . . . We two/*The two we/*Two we . . .

Ita

As Longobardi also notes, this remains so even in languages without an overt article as in Serbo-Croatian (Progovac, 1998) or Polish (Rutkowski, 2002), where nouns and proper names pattern alike, but pronouns do not. He suggests that while common and proper nouns are found optionally at the left edge of the phase and move there, pronouns are found obligatorily and may be inserted there, a fact that Longobardi links with the fact that personal pronouns, unlike nouns, are specified for Person. Person, in other words, is checked ‘high’ in the nominal phase, with consequences for cross-phasal licensing relations that involve personal pronouns. Gender, by contrast, is checked ‘low’, and it tends to be absent when Person is present. Correlatively with pronouns being paradigmatic edge elements, they have, as noted, very little descriptive or lexical content, which they shed in favor of stronger referential import. That import is at its peak with the 1st-Person pronoun ‘I’, probably the closest candidate to what Russell called a ‘logically proper name’:19 it cannot fail in its reference and is not mediated by any description whatsoever, as discussed

19 At least when it occurs in the nominative, as opposed to accusative as in ‘I am me’, where ‘me’ has a predicative reading, as discussed.


155

above. In line with that, the personal pronouns make the highest demands on grammatical structure, occupying a higher position than 3rd-Person pronouns (Shlonsky, 2000; Poletto, 2000; Bianchi, 2006) and engaging the upper layers in the C-field associated with logophoric features, as argued in Sigurðsson (2004). This hierarchical relation between the personal and non-personal pronouns is moreover an asymmetric one, in the sense that any personal-pronominal referent can also be referred to in the 3rd Person (‘he’, ‘John’, ‘that person’, ‘the object having genetic code XYZ’), whereas the reverse is not true: 3rd-Person objects of reference need not even be language-users. This illustrates that as we progress into the Person system, reference becomes more language-dependent, making increasing demands on grammar. Along with the lack of dependence on the descriptive content of substantive lexical items, dependence on the non-linguistic context decreases. In none of the pronouns we will discuss are their referential properties a function of their lexical specifications. Instead, they are a function of their topology and where in the hierarchical structure of the clause they are located (cf. Diesing, 1992; Mahajan, 1992; de Hoop, 1996; Reinhart, 1997; Winter, 2000; Zamparelli, 2000; Bartos, 2001; Borer, 2005; Danon, 2006; or Pereltsvaig, 2006; among others). We will start with purely predicative clitics, move on two grammatically more complex accusative ones, of which we distinguish two types, one with more referential import than the other, then to the still more complex dative clitics, and then the personal pronouns. 4.4.4.1 Predicative clitics Bare nominals in languages like Catalan or Spanish are standardly considered predicative, lacking referentiality, definiteness, and exhibiting lowest scope (Picallo, 2007; Dechaine and Wiltschko, 2002): (111)

Bare predicative nominals in Catalan and Spanish a. El president necessita escorta The president needs bodyguard ‘The presidents needs bodyguards’ b. En Pere sempre porta The Peter always wears ‘Peter always wear a jacket’

Cat

jaqueta jacket

c. Hay sillas para todos there-is chairs for everybody ‘There are chairs for everybody’

Sp

They are also banned from argumental positions, such as those of subjects, indirect objects (dative), or ECM environments, as seen in (112a–c). Plural bare nominals do occur in direct object position, though (see (112d)), which reveals differences between accusative and dative nominals, to which we will return below:

156


(112)

Restrictions on predicative nominals: a. *Amigo de María telefoneó friend of Mary telephoned.3s INTENDED: ‘A friend of Mary telephoned’

Sp

b. *(Le) dan libros a niño dat.3p give.3p books to child INTENDED: ‘They give books to a child’ c. *Considero libros aburridos consider.1s books boring INTENDED: ‘I consider (some) books (to be) boring’ d. Vi estudiantes en el saw.1s students in the ‘I saw students at the bar’

bar bar

Now, the clitic that resumes the Catalan nominals in (111a–b) is the so-called partitive ‘en’ (113a–b), which in these contexts is used predicatively, and thus has the same properties as its phrasal counterpart: it lacks referentiality, cannot express definiteness, and has lowest scope. In Catalan no other clitic can do this job. Since Spanish lacks a counterpart of the partitive clitic, it either uses nothing at all in this case, or else resorts to recycling some other clitic, as we see in (113c) (see Longa, Lorenzo and Rigau, 1998, on the recycling strategy): (113)

Cliticization of predicative nominals a. El president en/ *la necessita The president part acc.3fs need.3s ‘The president needs it’ (a set of bodyguards) b. En Pere sempre en/ The Peter always part ‘Peter always wears it’ c. / Las hay para The acc.3fs there-is para ‘There is one for everybody’

*la acc.3fs

Cat

porta wear.3s

todos everybody

Sp

Our claim was that pronouns are not specified for reference lexically. Rather, their interpretation with respect to our scale—as either predicative, quantificational, referential, or deictic—is determined grammatically. Clitics interpreted predicatively should be least in grammatical complexity. That is indeed the case with partitive ‘en’, which doesn’t have any ç-features whatsoever.20 For Spanish, the recycling strategy 20 Arguably the ‘e’ in ‘en’ (and many other Catalan and Romance clitic pronouns) is purely epenthetic, i.e. needed for syllabification.


157

selects either no clitic or a very underspecified one. Another predicative clitic is the Catalan neuter clitic ‘ho’.21 ‘Ho’ is a proform for predicates in general, whether clausal, adjectival, or nominal: (114)

a. El cotxe sembla bo the car seems good ‘The car seems good’

Cat

b. El cotxe ho sembla the car neut seems ‘The car seems (good)’ c. En Pere es un the Peter is a ‘Peter is an idiot’ d. En Pere ho/ the Peter neut ‘Peter is (an idiot)’

idiota idiot *el acc.3m

és is

Interestingly, as we saw above, even definite descriptions can be pronominalized by means of ‘ho’, when they are in predicative position (cf. (9), repeated below as (115)), something that can now be taken as independent confirmation of our claim that those DPs do indeed act as predicates in these contexts, rather than as arguments: (115)

Cliticizing predicative nominals: a. En Pere és el porter The Peter is the Porter ‘Peter is the porter’ b. En Pere ho/ *el The Peter neut acc.3m ‘John is (the porter)’

Cat

És Is

Catalan ‘ho’ is special in that it doesn’t have clear counterparts in other major Romance languages, where the neuter clitic is normally homophonous with the accusative masculine singular clitic, (‘lo’ in Spanish). That this clitic is not accusative, however, is shown by its lack of gender agreement (on the role of accusative in the build-up of referential meaning in clitics, see below): 21

(i) a. La chica parece the girl looks ‘The girl looks nice’

simpática nice.fem

b. La chica lo / *la the girl neut / acc.3sf ‘The girl looks so’

parece looks

Spanish

158


Embedded CPs, which always function as predicates, also cliticize through the neuter clitic, unlike what happens to argumental noun phrases (116b vs 116d): (116)

Cliticizing clauses: a. No entenc el tema neg understand.1s the topic ‘I don’t understand the topic’ b. No el/ *ho neg acc.3sm neut ‘I don't understand it’

entenc22 understand

c. No Entenc el que vols neg understand.1s the that want.2s ‘I don’t understand what you mean’ d. No *el/ ho neg acc.3sm neut ‘I don’t understand it’

Cat

dir to-say

entenc understand

Clausal complements, then, unlike referential DP complements, do not get pronominalized by means of accusative clitics, but rather only by means of the neuter clitic, which we take to lack both morphological Case and ç-features. The lack of Case is particularly interesting, as the presence of Case has been independently linked to an increase of referentiality along our scale, which we will also see evidence for below (for precedents of this claim see Enç 1991; de Hoop, 1996; Torrego, 1998; Cardinaletti and Starke, 1999; Danon, 2006; Lidz, 2006; among others; see also Section 6.5). Accusative case, however, comes in two forms, which gives further structure to our progression. 4.4.4.2 Accusative clitics, weak and strong Accusative clitics are different to the rest of the Romance pronominal system. Morphologically, they are virtually identical to definite determiners (Postal, 1966, and subsequent literature), and like them, and unlike the rest of the clitic paradigm of most of Romance, they have Gender features. Syntactically, they participate in a number of irregularities like opacity, deletability, ordering variation, or the so-called Person Case Constraint (Perlmutter, 1971; Kayne, 1975; Bonet, 1991, 1995, 2008; Anagnostopoulou, 2005; Adger and Harbour 2007; Ormazábal and Romero 2007; among many others; see Martín, 2012 for a review). With respect to their interpretation, the idea that accusative clitics are linked to referential specificity is quite widespread (cf. Suñer, 1988; Uriagereka, 1995; Roca, 1992, 1996; Sportiche, 1996; Fernández-Soriano, 1993; or Ormazábal 22

To avoid further complication, we skip over the phonological reduction processes of clitics, as they don’t have any effect whatsoever on the argument we’re making. For instance, in (116b) we should have: ‘No l’entenc’ rather than ‘No el entenc’.


159

and Romero, 2010). Suñer (1988) suggested that accusative clitics are inherently marked for specificity. On this view, accusatives cannot refer to negative phrases (117a), nonspecific indefinites (117b–c), or interrogative elements (117d): (117)

Accusative clitics and specificity: a. *[A ningún bedel]i loi veo to no janitor acc.3ms see.1s intended: ‘I see no janitor working’

trabajando working

Sp

b. *[Algún escritor famoso]i no loi he visto Some writer famous neg acc.3m have.1s seen ‘I haven’t seen a famous writer’ (non-specific interpretation) c. [A un famoso escritor]i no loi he to a famous writer neg acc.3m have.1s ‘I haven’t seen a famous writer’ (specific interpretation)

visto seen

d. *¿A quii eli veus? to who acc.3m see.2s intended: ‘Who do you see?’

Cat

In addition, as discussed by Suñer (1988), in the dialects of Spanish where accusative doubling is allowed, there is a specificity restriction on the phrases that can be doubled: (118)

Specificity constraint in accusative doubling: a. Lai vimos a Mafaldai acc.3sf see.1p to Mafalda ‘We saw Mafalda’ b. Lai vimos a acc.3sf see.1p to ‘We saw the girl’

la the

Rioplatense Sp

chicai girl

c. *Lai vimos a alguna chicai acc.3sf see.1p to some girl ‘We saw some girl’ (non-specific interpretation) However, the link of accusative pronouns to specificity holds only to a degree: it cannot be regarded as intrinsic to them as lexical items, in line with our general grammatical approach to reference. In fact, the so-called direct object nominals can have referential, quantificational, or predicative interpretations. Consider (119): (119)

Non-referential accusative nominals: a. Todos buscan a una secretaria All.pl look-for to a secretary ‘There is a secretary everybody is looking for’ (referential)

Sp

160

The Philosophy of Universal Grammar b. Todos buscan una secretaria All.pl look-for a secretary ‘They all look for a secretary’ (quantificational) c. Todos buscan secretaria All.pl look-for secretary ‘They all are secretary-hunting’ (predicate modifier)

The same applies to the pronominal version of these sentences, as Clitic LeftDislocation paraphrases suggest: (120)

Non-referential accusative pronominals: a. *(A) una secretariai todos lai buscan to a secretary all.pl acc.3sf look-for ‘There’s a secretary everybody is looking for’ (referential)

Sp

b. (*A) una secretariai todos lai buscan to a secretary all.pl acc.3sf look-for ‘They all look for a secretary (quantificational) c. (*A) secretariai todos (*lai) buscan to secretary all.pl acc.3sf look-for ‘They all are secretary-hunting’ (predicate modifier) This is even clearer in languages with a richer clitic paradigm, like Catalan, where the accusative clitics get the referential nominals, and the partitives get the predicative ones (with quantificational cases somewhere in between): (121)

Partitive vs. accusative clitics: a. (De) dones en veiem (tres) cada dia of women part see.1p three each day ‘We see (three) women every day’ (non-referential readings)

Cat

? b. Tres dones en / (les) veiem cada dia Three women part acc.3pf see.1p each day ‘We see three women every day’ (quantificational reading)

c. *(A) tres dones *en / *(les) veiem cada dia to three women part acc.3pf see.1p each day ‘We see (three) women every day’ (referential reading ONLY) Additional evidence for different interpretations of the accusative clitics, with different referential import, comes from backward anaphora. Neither quantificational (123b) nor predicative interpretations (123c) of the accusative clitic support these:


Differential availability of backward anaphora: a. Lai vi ayer, a Maríai acc.3fs saw.1s yesterday to Mary ‘I saw Mary yesterday’ b. *Lai he visto, (a) [la madre de cada acc.3f have.1s seen to the mother of each INTENDED: ‘I’ve seen the mother of each student’

161

Sp

estudiante]i student

c. *Como lai he arreglado, podemos conservar as acc.3fs have.1s fixed can.1p to-preserve el vino en barricai the wine in cask INTEND: ‘As I’ve already fixed iti, we can keep the wine in caski’ There is evidence, then, that while there is a clear link between accusativity and specificity, the accusative clitic can receive different interpretations in different grammatical contexts, some with more referential import than others. Moreover, there is a progression, in terms of referential import, from the predicative, neuter, and partitive to the accusative clitics. While, unfortunately, the difference between the two types of accusative clitics, which differ in their referential import, is not morphologically marked in the pronominal systems of the languages we have considered, extra grammatical marking for more strongly referential direct objects is visible in a subset of Spanish or Catalan nominals (Differential Object Marking). Also, we can see some other languages where accusative case is linked to different interpretations, including Turkish (Enç, 1991), Kannada (Lidz, 2006), and Hebrew (Danon, 2006). 4.4.4.3 Dative clitics The progression continues with dative clitics, which we now argue are structurally more complex than either predicative or accusative clitics and contain the latter as parts of their structure. They are, we argue, like the strong accusative clitics above in including a deictic layer, but they also show stronger behaviour, aligning them with the personal clitics, with which they are known to pattern in a number of domains.23 In line with Martín (2012), we specifically suggest that dative clitics are internally complex, built out of the accusative clitic plus a locative (deictic) element, and involve D-to-D movement in the phasal edge. The deictic layer is also shared with the personal pronouns. Ever since Kayne’s (1975) seminal work, it had been uncritically assumed that clitics form a unified class morphologically, syntactically, and semantically. This led to debates on, for example, the generation site of those elements. According to the 23 In fact, Martín (2012) suggests that the personal clitics (1st and 2nd Person) are in fact dative, regardless of their apparent syntactic function.

162


Movement Hypothesis, clitics are generated in argument positions within VP and subsequently move up toward the inflectional layer of the sentence (cf. Kayne, 1975, 1991, 1994). In contrast, the Base-generation Hypothesis argues that clitics are externally merged in inflectional positions (Borer, 1984; Suñer 1988; Fernández-Soriano 1993; among others). In recent years, however, an intermediate hypothesis has been defended since Roca, 1992, 1996, or Sportiche, 1996, among others. On this view, which we assume here, accusative clitics are base-generated within VP and move subsequently to inflectional positions, with consequences for both their interpretation and their morphosyntactic properties, different to the rest of the paradigm (Gender features, regular number; see Martín 2012 for review). Datives and personal clitics, however, are directly base-generated in inflectional positions, and hence, in our terms, in an edge position. This hypothesis immediately explains why clitic doubling is a phenomenon mostly related to the personal and dative clitics.24 It also suggests a natural topological explanation of the following crucial fact concerning their semantics. Dative phrases cannot host bare nominals, a constraint similar to the ones we saw in subject positions in (112), repeated here as (123c):25 (123)

Dative cliticization of bare nominals: a. *Lei di un libro a niñoi dat.3s gave.1s a book to child INTENDED: ‘I gave a book to a child’

Sp

b. *Lesi di libros a niñosi dat.3p gave.1s books to children INTENDED: ‘I gave books to children’ c. *Amigo de María telefoneó friend of Mary telephoned.3s INTENDED: ‘A friend of Mary telephoned’ If the dative clitic (present because of dative doubling) is dropped, the sentence’s grammaticality improves, at least in the case of the bare plural (124b), especially if the direct object libros ‘books’ is also a bare plural: (124)

Bare plurals: a. *Di libros a niñoi gave.1s books to child INTENDED: ‘I gave books to a child’

Sp

24 Accusative doubling is absent in most of Romance, except for a few dialects of Latin American Spanish, while dative (or personal) doubling is present in many Romance languages, and all the Spanish dialects. 25 Most examples in this section are from Spanish, because of the clearer status of clitic doubling in this language.


163

b. ?Di libros a niñosi gave.1s books to children INTENDED: ‘I gave books to children’ It seems, therefore, that the presence of the dative clitic, which is the only obligatory element in the dative doubling construction, imposes extra demands on the referential interpretation of the dative phrase. To explain that, and following Strozer (1976), Masullo (1992), Demonte (1995), or Cuervo (2003), Martín (2012) suggests that the presence or absence of the clitic in fact corresponds to two different constructions, with only the real dative construction requiring clitic doubling. Without the clitic we have a pure prepositional phrase, with rather different properties. For instance, in the prepositional case, if a dative clitic is present it can’t be co-referential with the nominal complement of the preposition:26 (125)

Datives and prepositions: a. Pablo *(lei) puso Paul dat.3s put.3s

a to

el the

caféi coffee

b. Pablo (lei) puso azúcar en Paul dat.3s put.3s sugar to BOTH: ‘Pablo put sugar in the coffee’

el the

café*i coffee

azúcar sugar

a. Pablo *(lei) cocinó una tarta a Paul dat.3s baked.3s a cake to

Andreai Andrea

Sp

Sp

b. Pablo (lei) cocinó una tarta para Andrea*i Paul dat.3s baked.3s a cake for coffee BOTH: ‘Pablo baked a cake for Andrea’ It again seems that the dative clitic imposes requirements on the referential interpretation of the dative phrase it co-refers with, which can’t be predicative. According to Roca (1996), there is a difference in the kind of phrases that can be doubled by accusative and dative clitics, with accusative, but not dative, being restricted by specificity considerations. Dative clitics, but not accusative ones, can be doubled by negative phrases (126a), nonspecific indefinites (126b), or interrogative elements (126c): (126)

Constraints on Dative doubling: a. No lei doy nada neg dat.3s give.1s nothing ‘I give nothing to nobody’

a to

nadiei nobody

Sp

26 This indeed entails that the particle ‘a’ that appears in the dative construction is not a real preposition, but rather a case marker, as argued by Masullo (1992).

164

The Philosophy of Universal Grammar b. A un escritori nunca lei to a writer never dat.3s ‘I never give anything to writers’

doy give.1s

nada nothing

c. ¿A quiéni lei das dinero? to who dat.3s give.2s money ‘To whom do you give money?’ In (126a), where the meaning of the sentence is not predicative but quantificational, the dative clitic is fully optional: No (le) debo nada a nadie, (Le) doy un libro a tres personas. When the phrase is maximally referential, as in rigid reference, the clitic becomes obligatory: No *(le) debo nada a ella / a María. Most interestingly, in predicative interpretations the clitic is totally out, as we saw above: (*Les) doy libros a niños. So the presence of the dative clitic is strongly linked to referential (pro-) nominals, and less so to predicative ones. It is also interesting that the quantificational ones are not so bad without the clitic, forming an intermediate case. With respect to (126b), this sentence doesn’t have a non-specific (cardinal) reading, that is to say, it never has the reading (or it is rather marginal) where I never give assignments to just one writer (I only give assignments to sets of more than one writer). This sentence just has strong readings, either specific (there’s a particular writer I never give any assignment to, because he’s always late in turning them in), or generic (I never give anything to anybody belonging in the class of writers).27 This is what we expect if the presence of the dative clitic correlates with strong readings. Finally, with respect to (126c), the same quantificational considerations we raised with respect to (126a) apply, and the clitic is again optional, although maybe not as optional as in the (126a) quantificational reading. We do not have at this point a full account for the apparently bigger optionality of the clitic in these weak quantificational contexts. Why do we need doubling information about the dative argument of the verb, i.e. the clitic plus the full phrase, and why is accusative doubling cross-linguistically less widespread? What we see is that in quantificational interpretations (bound variable), the dative pronoun is at least not as obligatory as in the referential cases. Prima facie, these data are consistent with the claim that dative clitic doubling or standing-in for noun phrases is a window into (or expresses) a purely grammatical property of these phrases, irrespective of the lexical content of the noun phrases concerned: their referentiality, which we saw is not yet stably found at the level of the accusative. If the dative structurally contains the accusative (as argued in Martín 2012), what we see is a progression towards referentiality starting from a lexical core and ending with a clitic whose content is now essentially exhausted by the referentiality of the phrase it doubles, without which it can now appear. 27

Generic or kind reference patterns with strong readings, according to Longobardi (1994, 2005).


165

4.4.4.4 Additional a-marking in strong accusative and dative nominals Before we move on to the personal pronouns proper, let us dwell on the existence of extra [a]-marking of strong accusative and dative phrases, indicating an increase in grammatical complexity correlating with an increase in referentiality, as predicted on the present model. In Spanish, that extra element [a] is either a (locative or directional) preposition, or an accusative or dative case marker. We do not go into the details of this here, as it would take us too far afield. The presence of [a] is not directly (or lexically) linked to a specific Case, rather the link depends on the grammatical context. All dative 3rd-Person phrases, as well as a specific subset of 3rd-Person accusative phrases, get [a]-marking. Notice, however, that all the personal clitics, with no exception, also get the [a]-marking in case of doubling: (127)

[a]-marking with personal clitics: a. Me vieron *(a) mi cl.1s see.3p to me ‘They saw me’ b. Me dijeron eso *(a) cl.1s said.3p that to ‘They said that to me’ c. Te vieron *(a) cl.2s see.3p to ‘They saw you’

Sp

mi me

ti you

d. Te dijeron eso *(a) cl.2s said.3p that to ‘They said that to you’

ti you

Personal pronouns thus behave like the 3rd-Person dative clitics, and unlike the accusative clitics, in requiring the presence of [a]-marking, in line with the link we claim between an increase in referentiality and extra grammatical structure.28 Interestingly, the presence of [a]-marking with 3rd-Person pronouns is also obligatory (128a), but in that case the interpretation of the 3rd-Person pronoun can only be referential, and it cannot be a bound variable (128b): (128)

[a]-marking with referential 3rd-Person pronouns: a. Le vieron *(a) Él cl.3s see.3p to He ‘They saw him’

Sp

28 Notice that in the full phrase with a-marking (‘a mi’, ‘a ti’), the 1st and 2nd pronouns change form with respect to their clitic form (‘me’ and ‘te’). The presence of the extra ‘I’ might be related to the presence of the deictic feature as well, which we are about to posit in all personal and dative clitics. We leave the investigation of this topic for future work.

166

The Philosophy of Universal Grammar b. *[A cada hombre]i lei dijeron eso to each man cl.3s said.3p that INTENDED: ‘They told that to each man’

*(a) to

éli he

To make sense of this pattern of [a]-marking, Martín (2012) adopts the arguments in Jayaseelan and Hariprasad (J&H) (2001), according to whom it holds universally that referring nominal expressions, as opposed to non-referring ones, contain place deixis. As a consequence, J&H postulate a DeixP in the extended nominal expression whose presence is crucial to the semantics of reference, because it provides a deictic index to the nominal that allows a referential interpretation, as either a strong accusative (*acc) or a personal clitic including 3rd-Person dative clitics.29, 30 The [a] element has clitic counterparts in some Romance languages, like Catalan ‘hi’, French ‘y’, Italian ‘ci’, Paduan ‘ghe’, etc. Although these elements have usually been considered locatives, this is a simplification, as argued in Kayne (2008). That element is used in clearly non-locative contexts (129a, b), and this suggests that the locative uses are but particular cases of a more general meaning, which we consider deictic in the sense of Kayne, and which are in complementary distribution (129c, d): (129)

[a] as deixis a. Jean y pense John cl thinks ‘John thinks of it’

Fr

It

b. Gianni ci pensa John cl thinks ‘John thinks of it’ c. Jean pense à Marie ! Jean (*a) John thinks to Mary John to ‘John thinks of Mary / John thinks of her’ d. Il va à Paris ! Il (*a) He goes to Paris He to ‘John goes to Paris / John goes there’

y cl

y cl

pense thinks

Fr

va goes

This complementary distribution seems to suggest that [a] and [y] are contextual allomorphs of the same underlying element. In Martín (2012), it is suggested that both elements are the realization J&H call ‘DeixP’, an element which makes those elements closer to essential indexicals, rather than merely referential in the sense that 29

The distinction between strong *acc and weak acc is inspired by the proposals in Enç (1991), Torrego (1998), Danon (2006), or Lidz (2006), among others, according to which accusative case morphology is optional in many languages with inanimate objects, and when it is present it indicates a specific (i.e. referential) reading. 30 See den Dikken (2010) for independent arguments in favour of the incorporation of a DeixP in the extended projection of nominals.


167

3rd-Person nominals, including proper names, are. That deictic element is then part of the structure of both datives and strong accusatives, as depicted in (130). Under this tree we see examples of dative clitics in some Romance languages, with capital letters indicating silent elements: (130)

The hierarchical structure of dative clitics DP

⇒ dative clitic (deictic) ⇒ strong accusative (referential)

DxP D DP31

DX D a. CATALAN: b. PADUAN: c. SARDINIAN: d. FRENCH: e. SPANISH:

[l(s)] D D [l] [l]-e

[i] [ge] [bi] [ui] DX

NP L(s) D [li(s)] D D

⇒ weak accusative (quantificational) ⇒ partitive (predicative)

[-ui as a deictic in French] [-e indicates a epenthetic vowel]

The tree depicts four hierarchically ordered layers, in which grammatically complex clitics contain as subparts grammatically simpler ones, with the dative clitics, for example, amounting to the structure [D + deix] (Martín, 2012, following Kayne, 2008; Caha, 2009). This layered hierarchy is mirrored in the morphological structure and syntactic behaviour of clitics, as shown in (a)–(e), and it entails the four interpretive classes we have distinguished in (pro-)nominals as a consequence. Thus, as noted, partitive clitics are entirely devoid of extended structure. They are pro-forms for empty noun phrases, and can only be interpreted predicatively, as they occupy the core of the nominal phase. Climbing up the phase, we find weak accusative clitics next. These clitics project a D layer that endows them with Gender and Number features, corresponding to the ‘lower’ region of D viewed as a ‘field’ of functional projections, with Person in the highest position. Such features allow weak referentiality properties (cardinal interpretations), like for example getting narrow scope or bound readings. Then there is the deictic layer, and the referential interpretation of the clitic becomes stronger, while at the same time retaining its Gender features. D stays in place, and this allows Gender features, but at the same time, D is bound by the deictic head, which imposes a referential (3rd-Person) reading.

31 The lower DP may very well correspond to the projection NumP argued for since at least Ritter (1991). We are solving this merely terminological issue by calling it DP, in order to emphasize the fact that the elements that can occupy that position (numerals, quantifiers, (in)definite determiners) are all determiner-like, but nothing we say depends on this particular choice.

168


Dative clitics enter the realm of personal clitics, as they are dependent for interpretation on the system of participants in the discourse. Because of that, they add an additional D layer on top of the deictic head. That head, which is overtly visible in Catalan (130a) or Sardinian (130c), gives them their deictic interpretation, which is exactly the same as what we see in personal clitics, as we will see below. Because the dative can be lexicalized by any part of the complex dative phrase in (130), with all other parts remaining silent, it is quite expected that the dative can have the overt form of an accusative (standard Catalan 3rd-Person plural dative: els), the form of a locative (Paduan ghe), the form of a locative plus an accusative (Sardinian [bi+lis]), or the form of a dative plus a locative (Catalan [els+hi]). That extra D layer—an extension of the phase edge—provides these clitics with a number of morphological and syntactic properties: (i) Dative clitics don’t get Gender features (in virtually all of Romance), as they are blocked by the Person features, exactly as happens with the personal clitics, which also lack Gender. In the next section, we argue that Gender and Person features are in complementary distribution. (ii) The [D + deix] configuration accounts for the intriguing morphological form of dative clitics in some Romance languages, like for instance the Catalan ‘els hi’, with ‘hi’ a locative/deictic clitic. It also accounts for the formal syncretism of dative and locative clitics in Northern Italian languages such as Paduan (130b). (iii) The fact that dative clitics include accusatives also gives a principled explanation to many syntactic puzzles of these clitics, including opacity in clitic clusters, or the famous Person Case Constraint. For full details, we refer the reader to Martín (2012). The crucial point for our purposes here is that the structure in (130) suggests the availability of D-to-D movement, where the [l]-head merged in the lower D crosses over the deictic layer to reach into the higher regions of D. This, we suggest, is responsible for the extension of the referential hierarchy proposed in Longobardi (1994, 2005) and Shee&H (2011) to deictic and ultimately personal interpretations. The personal pronouns, to which we turn in the next section, thus conform to a structure of an extended left edge. 4.4.4.5 Personal clitics The above proposal entails that 1st- and 2nd-Person clitics should always come with a deictic layer, and include a higher D-head containing this layer, thereby going beyond the structural complexity of proper names. As we claimed in the previous section, the 3rd-Person dative pronouns include a deictic layer, which is visible in languages like Catalan or Sardinian. As we shall see below, that layer is also visible in the Latin personal pronouns. Personal clitics, including as they do the deictic layer but also activating the Person layer, thus have more structure than any other referential (pro-) nominals, and may be directly merged in the higher D-layer, unlike the head of dative clitics, which may only move there from the lower D-head. Interesting in itself is the fact that Latin lacks 3rd-Person pronouns, for which there are demonstratives instead, illustrating again the split we have noted here


169

between merely object-referential (3rd or no Person) expressions and those carrying person-reference in discourse: (131)

Latin personal pronouns: 1st

2nd

NOMINATIVE

ego

tu

ACCUSATIVE

me-

te-

GENITIVE

mei-

tui-

DATIVE

mihi

tibi

me-

te-

ABLATIVE

Of particular relevance for our case in point here are the dative pronouns. It is clear, given the other pronouns in the paradigm, that [m] and [t] are 1st- and 2nd-Person morphemes, respectively. As a result, the dative pronouns seem to be composed of those Person features plus something else, i.e. [m + ihi] and [t + ibi] respectively. The question then is: What is the semantic import of [ihi] and [ibi]? Starting with the latter, Latin ‘ibi’ is a distal locative adverb equivalent to English ‘there’. The proximal equivalent of ‘ibi’ is ‘hic’, often translated to English as here (Panhuis, 2006). So the 2nd-Person pronoun ‘tibi’ seems to have the structure [t + ibi], i.e. [2nd Person + locative]. We would expect then that the 1st-Person pronoun had the same structure, i.e. Person feature [m] + locative [ihi]. Now, in itself, [ihi] does not correspond to any locative element in Latin. The closest element we can find is the above-mentioned 1st-Person demonstrative or adverb ‘hic’. So let us assume that [ihi] contains a reduced version of the proximal element. It would make sense if the 1st-Person pronoun contains a proximal deictic like ‘hic’, just like the 2nd-Person contains the distal deictic ‘ibi’. This is close, but we still have the problem of the 1st ‘I’ in ‘ihi’. To examine it, let us go back to the 2nd-Person dative ‘tibi’. We just saw that ‘tibi’ seems to have the structure [t + ibi], i.e. [2nd Person + locative], and that ‘ibi’ is a distal locative element. Now, interestingly, ‘ibi’ is part of the following paradigm: (132)

The paradigm of ‘ibi’ demonstrative interrogative

relative

indefinite

ibi there

ubi where

alicubi somewhere

ubi where?

170


So it is apparent that ‘ibi’ and ‘ubi’ have a relationship in Latin similar to that of English there and where (where this a referential element, and wh- is an interrogative element, something also visible in that and what, or then and when, with that, then, and there all being deictic). That is to say, the class of words in (132) suggests that ‘ibi’ could be considered bimorphemic, forming a minimal pair with the interrogative and relative pronoun ‘ubi’. The latter, in turn, should be bimorphemic too, and this prediction seems to hold: [u-] is an interrogative particle in Latin (equivalent to English wh-), present in other interrogative words like ‘unde’: ‘whence?’ or ‘from where?’ (opposed to ‘inde’: ‘thence’ or ‘from there’), ‘utrō’: ‘which way?’, or ‘utrōbī’: ‘in which place?’. That [u-] is also part of the qu- element of many other interrogative words in Latin, like ‘quis’: ‘who’, or ‘quid’: ‘what’, where quis the base for many of the interrogative words of Romance languages. Therefore, it seems that we can decompose ‘ubi’ into [u + bi], i.e. [interrogative + place] with the meaning ‘what place’, or ‘where’. But then, analogously, ‘ibi’ should also be decomposable and have the structure [i + bi]. As we have just said, [-bi] is likely to mean ‘place’, something confirmed by two facts: (i) Firstly, by the fact that ‘utrō’ means ‘which way?’ in Latin, but ‘utrōbī’ means ‘in which place?’ So ‘utrōbī’ can clearly be analysed as [utrō + bī], where [bī] means ‘place’. (ii) Secondly, by the fact that ‘bi’ does mean ‘place’ in Sardinian: (133)

Place deixis in Sardinian: Bi nke nd' at issitu tres place loc part at exited.3p three ‘There came three of them out of there’ [from Jones (1993)]

Sard

The question that remains, then, is that initial ‘i’ of ‘ibi’ (and ‘-ihi’). Interestingly, ‘i’ is (at least part of ) the so-called locative pronouns of some Romance languages (French y, Catalan hi, Italian ci). However, since, as we have seen, these elements are deictic in a more general sense rather than merely locative, let us assume that ‘i’ is really that, a deictic.32 So [i + bi] would have the structure [deictic + place]. As a result of all this, then, the structure of ‘tibi’ is [t + i + bi], that is to say [2nd Person + deictic + place]. Analogously, the structure of mihi should be [m + i + hi], i.e. [1st Person + deictic + (proximal) place]. Both of them have the structure in (134), which as the reader will note is essentially the same as we saw in (130) above (again, capital letters mean silent or moved elements):

32 Additional evidence in support of ‘i’ as a deictic element, is that according to Leu (2008), a demonstrative like ‘this’ has the underlying structure [the + here], with ‘here’ a clear deictic element. A similar structure seems to occur with the Latin distal demonstrative ‘ille’ (that), which could also be decomposed in parallel to the English case as [i + lle], with ‘i’ a deictic, and ‘lle’ as the element that later became the definite article (which Latin lacked) in most Romance languages.


171

The structure of tibi: DP t (Person)

DxP i-bii (deictic)

DP D

NP BIi

Compare in this regard also the Spanish and Catalan 3rd-Person strong pronouns. We give the paradigm and the argumentation for Catalan, but the case of Spanish is essentially similar: (135)

Catalan 3rd-Person strong pronouns number gender strong acc singular masc ell el fem ella la plural masc ells els fem elles les

These pronouns are interesting because they also have more grammatical structure than their accusative clitic counterparts, and this yields interesting differences both morphosyntactically and semantically, exactly along the lines we are suggesting in this section (see also Cardinaletti and Starke, 1999). In particular, the fact that they have more structure explains why these pronouns don’t undergo processes of phonological reduction and are stressed, unlike their clitic counterparts. Thus, compare the accusative clitics (or definite determiners) in (136a) and (136d), with their strong pronoun counterparts in (136b) and (136e). In line with our account, the dative clitic in (136c) patterns with the strong pronouns: (136)

Cat

a. L’/ *El ha corregut acc.3sm acc.3m have.3s run.part ‘He has run it (e.g. the marathon)’ b. Ell ha He have.3s ‘He has run’

corregut run.part

c. Li/ *L ha dit dat.3s dat.3s have.3s told ‘He has told him the truth’ d. Per + el (cotxe) For + the (car)

= Pel (cotxe) = For-the (car)

la the

veritat truth

172

The Philosophy of Universal Grammar e. Per + ell For + he

= Per ell (but not *pell) = For him

This follows if strong pronouns conform to the structure we have suggested for dative clitics and the Latin pronouns: (137)

DP (e)-l (no gender)

DxP DX (deictic)

DP l(a)(s)

NP

(NOTE: The ‘(e)’ in the higher instance of ‘el’ is merely epenthetic). This extra structure provides strong pronouns with their special kind of grammatical semantics, explaining why strong and personal pronouns cannot be bound, as noted earlier in relation to (95): a bound pronoun, apparently, must have an empty or (weak enough) edge. This further also relates to another important aspect of personal clitics: their lack of Gender.33 4.4.5 Summary In this section we have gone along with the traditional philosophical diagnosis of ‘essential indexicality’ and asked about the nature and explanation of this curious phenomenon, which is apparently irreducible in language and of foundational significance for a number of important problems in the philosophy of mind including how we think about the nature of the self. We have suggested that, rather than a semantic phenomenon that consists in some special relation between a particular class of words and the non-linguistic ‘context’, the phenomenon is not a lexical or non-grammatically semantic one at all. The phenomenon, rather, belongs to the expanded domain of grammatical semantics. As a consequence of this, the relevant 33 In Harley and Ritter (2002), Picallo (2007), Boeckx (2011), and Martín (2012), among others, it is suggested that grammatical Gender and Person are in complementary distribution, and hence may be contextual realizations of the same underlying morpheme (which Boeckx, 2011 suggests might be little n). Thus, in most of the languages of the world (with a few exceptions notably within the Semitic family), 1stand 2nd-Person pronouns have Person but not Gender features. Like Benveniste (1966) and subsequent literature we have considered the 3rd Person not to be a real person, that is to say, a participant in the discourse (cf. Harley and Ritter, 2002). If Person and Gender are in complementary distribution, Gender seems to occupy the place of the deictic feature in the non-deictic pronouns. The pronouns that do have Gender are all 3rd Person, and all are either referential or quantificational. Predicative pronouns (like ‘en’, ‘ho’, ‘lo’), do not even have Gender. This clearly suggests that Gender is lowest in the grammatical genesis of a referential expression, and that the path to reference mediated by the Person system, which is at the peak of the grammatical process, leads via deixis.


173

notion of context cannot be identified in non-linguistic terms. Decisive, moreover, are not these linguistic items themselves, qua lexical items, but the grammar of their uses. Not only do the so-called ‘indexicals’ not form a natural (lexical) class, but their referential properties can systematically differ depending on their grammar on an occasion of language use. Nor is the class of so-called ‘pronouns’ homogenous. It transpires against the background of a general ‘topological’ mapping from the phasal dynamics of grammar to the forms of reference that there is, in fact, a hierarchy of referentiality. It ranges from purely predicative to quantificational forms of reference involving scope, to rigid forms of 3rd-Person object-reference, and from there all the way to deictic and personal forms of reference that, at the end of the scale, we only find in the domain of the personal pronouns and the object clitics that pattern with them (dative clitics). Our account rationalizes the existence of pronouns in all human languages, which was our initial question: these, we can now conclude, express forms of grammatical reference and a relational semantics that no nominal can emulate and that is unmediated by lexical content. Indexicality is essential, because (i) it is grammatical, and (ii) because its grammar is such that other forms of reference as found in proper names, definite descriptions, or 3rd-Person accusative pronouns, are strictly less complex than the grammar of pronominals in their essentially indexical uses. A corollary of our account above is the apparent interpretability of Case, to which we return in Chapter 6.

4.5 Conclusions As ‘concepts’ are formed from perceptual features of the environment, and concepts become lexemes, meaning gets encapsulated in units that live a life independent of environmental stimulation. It thus has to be related back to the environment, so as to be useful in directing one’s thought to it. Concepts/words thus have to be made referential again or be reconnected to the environment, which is not the case for perceptual features. Grammar is the mechanism that, within the single phase, converts concepts/words into referential expressions. As lexical content enters the phase, and moves towards the edge, referentiality increases and becomes increasingly independent of descriptive content that the concepts/words, qua lexical items, encode. The exact form in which a given concept/word is made referential depends on a number of options, all of which are essentially grammatical in nature, and more specifically, topological: it co-varies strictly with such factors as whether an edge is projected, the edge is light or heavy, whether the content of the interior becomes the content of the edge, or whether the edge is expanded by extra layers. With any of these options, which become available as the topology does, rather than being predictable from or depending on ‘lexical category’, a formal ontology emerges:

174


reference is never to ‘bare’ referents, but based on a concept, and to entities of different formal types. We conclude, on the basis of the evidence provided, that metaphysical or semantic intuitions about what formal types these are, which are almost universally taken to be non-grammatical in nature by metaphysicians and semanticists, are in fact grammatical in disguise: no other, non-grammatical theory is available that explains, in a systematic and naturalistic fashion, why these types emerge, and no others; why they co-vary with grammatical complexity in such narrow ways; or how the types systematically relate to one another. We have now reached the core claim of this book, and provided a substantial part of its evidential base: for, if the above empirical claims about how the grammar configures referential expressions in the nominal, pronominal, and clausal domains, are correct, then the epistemological significance of grammar is evident.

5 Cross-linguistic variation 5.1 The apparent problem of linguistic variation One crucial challenge facing any universal grammar project is the obvious problem of cross-linguistic diversity.1 While it is uncontroversial that all humans, pathologies aside, share an ability to acquire human languages and that these same humans plausibly also share a system of reference of the kind proposed in Chapter 4, what is much more controversial is whether the linguistic systems they use for this purpose can be claimed to be grammatically non-distinct in any meaningful sense. In short, a significant challenge facing the Un-Cartesian view is not that grammatical semantics might fail to be universal, but rather that it might fail to be grammatically specified. Cross-linguistic variation may support such a conclusion and so it has to be scrutinized, a task to which we turn in this chapter. Chapters 1–4 have proposed that core principles of grammar effectively organize a particular form of meaning—grammatical meaning—which, we argue, is an evolutionary novelty that does not appear to be available without grammar. But linguistic variation is standardly taken to affect grammar as well: grammar, indeed, is what has been the primary focus of the field of comparative syntax in the Principles and Parameters (P&P) tradition. Those who reject the Chomskyan UG approach often do so on the basis that it cannot accommodate the full range of linguistic diversity (see Evans and Levinson, 2009; Levinson and Evans, 2010; and the replies to Evans and Levinson, 2009). Rejecting any notion of universal grammar then comes at a price: a radical thought–language divide, in the light of a mode of thought that clearly is universal. Thus, for example, Everett (2005) claims that the Pirahã language lacks the putative universal of recursion. To avoid the claim that Pirahã speakers cannot embed one thought in another, he is forced to conclude that recursion, while absent in Pirahã ‘language’, is present in Pirahã ‘thought’. More generally, the uniformity of thought is preserved in discussions of cross-linguistic variation, and is indeed the basis for mapping out such variation, a notion that is only meaningful against a 1 This chapter is focused on cross-linguistic variation without (essential) variation in the linguistic genotype. We will call variation of the latter kind ‘biolinguistic variation’, following Boeckx (2011) and discuss it at length in Chapter 8.

176


standard specifying what is constant. This constancy is also not questioned where thought is said to be ‘relative’ to language to some extent (Majid et al., 2004), for such relativity primarily affects the lexical-conceptual resources on which thought is based, not the formal ontology or grammatical semantics as discussed here. If, as has been proposed in the preceding chapters, this divide is given up, and grammar inherently constructs the semantics of a specific kind of thought, then variation across languages becomes highly relevant, for two distinct reasons. Firstly, superficial (non-grammatical) variation may serve to obfuscate more abstract grammatical patterns, rendering the structure of grammatical semantics opaque (see for example the covert/overt movement distinction discussed in Chapter 4, which when viewed in abstraction reveals an underlying uniformity). For this reason a significantly constrained and explanatory model of the mapping from grammar to surface forms is crucial to the Un-Cartesian enterprise. Only with the service of such a model can language-specific forms provide evidence for the grammatical basis of semantics.2 The second reason concerns the potential existence of ‘syntactic parameters’ of the traditional Government and Binding (GB) kind. From a Cartesian perspective such variation is ultimately unproblematic, though it is rejected in some recent Minimalist work on independent grounds (cf. Boeckx, 2001; 2012). In such a perspective, syntax is an autonomous system that simply translates a pre-existing system of propositional thought into a linguistic format in arbitrary and variable ways.3 Syntactic differences between languages, while requiring explanation in the context of UG as defined in this tradition, will have no effect on the structure of thought. From an Un-Cartesian perspective, however, grammatical variation, if it exists, raises some interesting challenges: we would not expect the organizational principles giving rise to grammatical semantics to be parameterized. Indeed, the basic deictic function of language appears not to vary across populations. On the strongest sense of the hypothesis, then, the prediction is that there should be no grammatical variation at all, without semantic variation. Let us consider a concrete, if hypothetical, example of how it might not be true that grammar, insofar as it configures its own distinct kind of grammatical semantics, is universal in the species. Ontologists in some population might posit a formalontological category crucial to the content of thought as intuited and formally analysed by their best logicians, which they call a ‘pact’. It is claimed to be universally available in human thought and defined to be systematically related to the familiar

2 Abstract grammatical patterns cannot merely be assumed to exist in spite of this surface variation, as this would make the Un-Cartesian approach essentially Cartesian (positing an abstract semantic representation which is only arbitrarily related to the surface patterns attested in actual natural languages). 3 Though in the Minimalist Program (Chomsky, 1995 onwards) this system is taken to be highly minimal, a point to which we return below.

5. Cross-linguistic variation

177

category of a ‘fact’, in the following way. Facts are commonly conceptualized as propositions that are established as true: a fact is a given proposition converted into a truth. Now suppose that facts could also be converted into something else— something even more fundamental, just as a fact is something ontologically more fundamental than a mere proposition, which as such can still be both true and false. This then is what a pact would be: a fact converted into something that guarantees its factivity somehow, or makes it a fact, just as a fact is what makes a true proposition true. So we have ‘truth-makers’ (= facts) and ‘fact-makers’ (= pacts). A human population with professional ontologists in it that posit pacts would refute the Un-Cartesian perspective, exactly insofar as there would be no basis in grammar for the ontological distinction in question. But no ontologist has ever posited pacts, to our knowledge. Basically, human thought seems not to reach that far, meeting its limit in facts/truths. If there is controversy about the truth of proposition P, we can assert that ‘it is a fact that P’. But if this is challenged again, what can we do but re-establish the fact or otherwise give up on the view that it is one? So ontologists pose truth-makers, which are facts, but no fact-makers. Why should this be? Our answer should be obvious by now: pacts would exceed the reach of grammar. The limits of grammar are the limits of thought. We will revisit this curious conclusion in Chapter 9. The arguments for universality, however, cannot come purely from the apparent lack of variation at the semantic level. It must also be shown that this universality has a grammatical basis. Thus, in another hypothetical population, thought might not exceed grammar, but could fail to co-vary with it. There might be a way of classifying objects of thought or of reference, independently of what grammatical categories and relations are used in linguistic acts of reference to such objects, and without any correspondence between grammatical patterns and the ontology. Or, as noted above, the topological mapping principles that we have outlined in the previous chapter might be reversed, in the sense that more projected nominals are consistently less extensional; grammatical predicates are mapped to objects and grammatical subjects to properties; sentences are mapped to objects and lexical items to truth values; and so on. Surely, in this case, we would conclude from such variation that another system must organize meanings into properties, objects, facts and so on, which would then allow grammar to vary: for that other system would account for what is constant, while grammar is merely a language-specific idiosyncrasy or tool, with no epistemic effects. Yet again, we know of no such patterns or reversals. They not only appear contingently absent, moreover, but conceptually impossible to the extent that there simply is no way to conceptualize a metaphysical notion such as ‘property’, without bringing in, directly or indirectly, the grammatical notion of a ‘predicate’. All that said, such considerations still do not address the basic fact of linguistic diversity, much of which has previously been analysed as grammatical variation. It must be shown, therefore, on some independent basis, that much of this variation

178


is not in fact grammatical. The present chapter focuses on this task, providing an explicit diagnostic for ‘grammatical variation’ and reconsidering three of the classic parameters of the GB tradition in its light. It is argued that at least the parameters considered here do not constitute points of major grammatical variation. Indeed, it will be argued that parameters such as the head-parameter and the null subject parameter are best located in the post-syntactic component dealing with morphophonological ‘externalization’, in the sense of Berwick and Chomsky, 2011 (or ‘PF’ in the sense of Chomsky, 1995). As for the ‘alignment parameter’, things are more complex, but again no grammatical macro-parameter is involved. The structure of the remainder of this chapter is as follows. Section 5.2 begins by defining what is meant by grammatical variation and how this is distinct from linguistic variation in a broader sense. Section 5.3 considers what should be considered part of grammar and suggests a means to diagnose instances of genuine grammatical variation. Section 5.4 considers word order variation in the light of the Un-Cartesian hypothesis and argues that the order of head-complement pairs is best located in the PF-component. Section 5.5 turns to the null subject parameter and argues that it too should be located in the domain of externalization. Section 5.6 turns to the various instantiations of the ‘Alignment parameter’ separating ergative from accusative languages and argues that there do appear to be minor grammatical differences between (some) ergative and accusative systems. Rather than stemming from a parameterization of thematic structure (Marantz, 1984; Levin, 1983; Dowty, 1991) or Agree (Bobaljik, 1992; 1993), however, these are better modelled via an interconnected series of morphological and syntactic micro-parameters. Moreover, as these micro-parameters appear to have micro-semantic effects, they may not pose a serious problem for the Un-Cartesian position. Section 5.7 concludes.

5.2 Linguistic variation vs. grammatical variation Languages clearly differ from each other at the lexical, phonetic, phonological, and morphological level as well as in terms of word order and the (overt) realization of arguments. As Ramchand and Svenonius (2008: 220) note, the unassailable fact of this surface-linguistic variation coupled with the assumption that all humans share the same basic system of thought provides two options regarding the connection between syntactic representations and the C-I interface that is assumed in standard Minimalist syntax: [E]ither the mapping from syntax to C-I is trivial, in which case all syntaxes have to be the same (and highly abstract) at the interface, or else the mapping from syntax to C-I is nontrivial, in which case two languages might express ‘the same thought’ while feeding off different syntax/semantic representations.


179

Ramchand and Svenonius advocate the second of these possibilities and their choice is consistent with the standard Cartesian approach to variation. Traditionally, Principles and Parameters Theory (P&P) accounts for variation by way of ‘parameters’. On the Government and Binding understanding of parameters, they are effectively the underspecified principles of a richly specified UG, which are filled in on the basis of experience (Chomsky, 1981). In their ideal macro-parametric incarnation, a single parameter gives rise to a number of seemingly unrelated surface effects (Rizzi, 1982; Baker, 1996). Such a model, if correct, implies that all linguistic variation is effectively built into UG. Meanwhile, however, macro-parameters have become controversial even within the generative tradition itself (cf. Newmeyer, 2005: 77–103). What has generally replaced them in the Minimalist Program is micro-parameters, low-level parameters whose effects can be local and independent.4 As such, Minimalist parameters are merely taken to denote differences in the formal specification of functional lexical items, making them effectively lexical (cf. the ‘Borer–Chomsky conjecture’). As Boeckx (2011) points out, this change of perspective is no accident. The Minimalist take on syntax makes it hard to see how the basic operations of syntax, abstract and minimal as they are, could be parameterized at all. Agree, for example, does not become a different operation depending on which feature acts as a probe, or depending on whether a feature is grammaticalized in one language but not another; and similarly for Merge. The same applies to classical Minimalist ‘third factor’ conditions like ‘Shortest Move’ or ‘Attract closest’, which again it seems to make little sense to assume to be operative in one language but not another (Boeckx, 2011). In positing a computational system ‘perfected’ in line with the Strong Minimalist Thesis, Chomsky has raised the degree of abstraction in linguistic descriptions, thereby potentially overcoming the problem of identifying neutral cross-linguistic descriptors. This problem arises in any universal grammar approach that uses theoretical terms used in the description of particular languages in order to make Universalist claims. But if the genetic core of language that remains after abstracting cross-linguistic variation is just (internal and external) Merge/recursion, it appears that no language could seemingly lack this core. Hence the problem of languageneutral descriptors is overcome. At the same time, this solution comes at the price that no genetic linguistic core in fact remains: nothing is said about language specifically, if we assert that it is generable by Merge or is recursive (or is recursive and exhibits principles of computational efficiency leading to ‘phases’). Moreover, and crucially for our purposes, taken to its logical conclusion, the study of UG will now be divorced from the study of cross-linguistic variation: it primarily constrains the Language of Thought, following deductively from the Strong Minimalist Thesis, with language variation demoted to a matter of the accidents of ‘externalization’ 4 See Biberauer (2008) for an in depth discussion of macro- and micro-parameters and Roberts (2012) and Section 5.6 below for an attempt to resolve the tension between micro- and macro-parameters.

180


(Chomsky, 2010: 61). In this way, Chomsky has effectively come to pursue the opposite course from GB theory: rather than building variation into UG, it is now excluded from it, in an effort to maximize the effects of non-language-specific factors such as third factor principles or effects of acquisition. The view implied by the Un-Cartesian view is basically the first view above, that is, the one that Ramchand and Svenonius reject: the mapping between syntax and the C-I interface is direct, in fact there is no C-I ‘interface’ at all. When grammar has established its relational structure, we do not have to go to another component so as to determine the semantics there, using independent principles. It is the mapping between syntax and PF, in this model, which must, then, be indirect, in order to capture the fact that languages differ from each other. The implications of such a view are very strong, and thus worth exploring for methodological reasons alone. It is important to be clear, moreover, what it does not imply. Firstly, languages clearly differ with respect to their lexical inventories. Which lexical items are available in a given language is partly determined by cultural practices. Thus many languages famously lack a term for the relative spatial terms ‘left/right’ because they use an absolute (North/South based) spatial reference system (Majid et al., 2004).5 It is also partly due to historical accident. Thus French lacks a word for ‘shallow’ and must express this concept via periphrasis peu profond ‘little deep’ (cf. Kayne, 2006). Likewise, Brown (2011) notes that while most languages have distinct lexical items for finger and hand, many do not and this appears to be an areal feature of parts of Australia and North America. Differences of this kind are purely lexical and while they necessarily affect the exact thoughts that can be expressed in a given language (simply because the thoughts in question will contain different concepts), they seem to have no obvious impact on the grammatical semantics available in a given language as characterized here. In this much, the claim must be that any human language can create the same kinds of thoughts, and not the exact same set of thoughts. Lexical variation, then, is non-negotiable. What about variation in the functional lexicon? Ramchand and Svenonius note that some languages have a first person plural inclusive/exclusive distinction whereas others do not. This difference also is essentially lexical in character, akin, for example, to the different division of the colour spectrum in different languages (Berlin and Kay, 1969). The difference is that the features 1st and 2nd Person are independently required in English, and arguably universally. English simply lacks this particular lexical distinction and the lexical item we can realise either the features 1st + 2nd Person or 1st + 3rd Person, leading to a semantic ambiguity that does not affect, say, Quechua. This difference does not appear to affect the grammatical semantics of

5 The causality could, of course, go in the other direction, but the point stands that this kind of variation is purely lexical and does not seem to affect grammatical semantics.


181

these two languages either, though, again, it may affect the precise thoughts that can be (unambiguously) expressed. Perhaps more challenging is the fact that some languages have complex noun class systems which appear to participate in grammatical relations, and which are impoverished or simply lacking in other languages (cf. Contini-Morava, 1997, on Swahili). These are arguably more complex versions of the gender classes in some IndoEuropean languages, membership of which differs from language to language in semantically arbitrary ways.6 Yet again, this appears to be essentially a lexical difference between languages, with minor effects on the exact thoughts expressed (though cf. Kayne, 2006). Thus, psycholinguistic experiments have revealed that gender class membership in many Indo-European languages has minor effects on conceptual semantics. Sera, Berge, and Castillo (1994) report that speakers of Spanish unconsciously attribute masculine/feminine properties to inanimate objects based on their grammatical gender. The implication is again that languages can create the same kinds of thoughts, but that lexical differences between languages may affect the particular thoughts that they can express. In sum, the linguistic variation discussed thus far is thus not grammatical in nature but rather lexical. In the next section, we turn to the difficult task of diagnosing grammatical variation.

5.3 Diagnosing grammatical variation Chapters 1–4 mention core aspects of grammar, all feeding into grammatical semantics: the phase, the predicate/argument distinction, modification, part-whole relations between grammatical categories, reference. But this is clearly non-exhaustive. Something else which presumably falls within the remit of ‘grammar’ is the so-called functional sequence, which determines the universal hierarchical order of functional projections in the nominal and clausal peripheries. While still controversial, this idea has gained currency in recent years following much work by Cinque (1999, 2005, 2010) and others in the cartographic tradition. Cinque (1999) argues, for example, that adverbs in all languages are externally merged subject to the universal functional sequence in (1), which also determines scope: (1)

[Moodspeech-act frankly [Moodevaluative fortunately [Moodevidential allegedly [Modepistemic probably [Tpast once [Tfuture then [Mod irrealis perhaps [Modnecessity necessarily [Modpossibility possibly [Asphabitual usually [Asprepetetive again [Aspfrequentative(I) often [Modvolitional intentionally [Aspcelerative(I) quickly [Tanterior already [Asp terminative no longer [Asp continuative still [Asp perfect(?) always [Asp

6 Though classes 1 and 2 are very often made up largely of +human nouns, as Jenneke van der Wal reminds us.

182

The Philosophy of Universal Grammar just [Asp proximative soon [Aspdurative briefly [Aspgeneric/progressive characteristically [Aspprospective almost [Aspsg.completive(I) completely [Asppl.completive tutto [Voice well [Aspcelerative(II) fast/early [Asprepetetive(II) again [Aspfrequentative (II) often [Aspsg.completive(II) completely ]]]]]]]]]]]]]]]]]]]]]]]]]]]]]] retrospective

The evidence for (1) comes from certain regularities and gaps in the surface ordering of adverbs and verbal suffixes across languages. On Cinque’s view, these surface orders, which differ in certain constrained ways, essentially disguise the universal base-generated structure.7 One can tentatively conclude, then, that there is no grammatical variation in the base generation of functional heads. Cartographic hierarchies provide a fine structure to the more basic phasal structure [C-T . . . [vV . . . [D-N]]] that we adopt elsewhere in this book. Another potential component of grammatical semantics is thematic structure. While the strongest version of Baker’s (1988) Universal Theta Assignment Hypothesis (UTAH) is difficult to maintain, the relative version of it pursued by Dowty (1991), Baker (1997) and Platzack (2011) appears to be empirically supported as well as conceptually elegant. The crucial idea behind these proposals is that the thematic role that a given DP receives depends upon the position in which it enters the derivation and that the way this is determined is cross-linguistically constant.8 According to Baker (1996, 1997), for example ‘themes’ universally originate lower than ‘agents’ as reflected by universal facts about anaphor binding and incorporation. If thematic structure is a component of grammar, and indeed grammatical semantics, as we have assumed here (see also Bowers, 2011), then the prediction is that it will not be subject to parametric variation. We return to this issue in section 5.6 in relation to ergativity. Beyond thematic structure, there is reference and predication. Is there variation regarding the principle of ‘movement towards the edge’ as characterized in Chapter 4? As we noted, what most obviously varies in the languages we have considered is how this movement is realized, for example through lexical insertion of a determiner where it is available, or through head-movement, where such movement can be overt or covert. This kind of variation does not affect the basic grammar of reference we have posited, and the variation in question is most naturally described as lexical and morphophonological. A ‘semantic parameter’ for how nominals are mapped to a pre-given set of assumed semantic ‘types’ was posited by Chierchia (1998), but Longobardi (2001), consistent with our proposal in Chapter 4, has disputed the need for any such parameter. Operator-variable interpretations, too,

While Cinque (1999, 2005) proposes that the differences in surface order across languages result from subsequent differences in movement operations, Abels and Neeleman (2012) have shown that the same facts can be modeled by parameterizing the way in which such structures are linearized. We return to this issue below. 8 In fact, even those approaches which permit movement into theta-positions, retain the idea that thetarole assignment is configurationally determined (cf. Hornstein, 1999; Sheehan, 2012). 7


183

are part of the dynamics that create particular kinds of (in this case, bound) interpretations for nominals, and are thus part of grammatical semantics in our sense. Internal Merge arguably provides a minimalist account of the dependencies involved. Again, such movement can be overt and covert, as is well known. Inasmuch as overt and covert movements pattern alike, they can be considered merely as alternative realizations of the same grammatical structure. If Case is part of grammatical semantics, variation in Case-assignment, too, would be relevant as a potential aspect of syntactic variation affecting meaning. Case, however, is widely taken not to affect semantic interpretation. In Chapter 6 we will argue that Case and Agree are morphological features and relations, respectively, which reflect grammatical relations that are crucial to grammatical semantics, while not being themselves a part of grammar as distinct from morphophonology. We therefore expect no variation in regards to the relations that the relevant features/ relations reflect—licensing relations, in the terms of Chapter 6—but there would be no problem with variation in regards to different language-specific resources (e.g. Case-inventories) in which these relations are reflected. We revisit this conclusion at length in Chapter 6, but return to potentially the best candidate for a syntactic parameter, the existence of both accusative and ergative alignments, below. One crucial question from an Un-Cartesian perspective, then, is whether these grammatical operations/properties are subject to parameterization across languages. There are still at least two distinct ways of interpreting this question, however. On the one hand, there is a question whether the same operations exist in all languages, and are not themselves subject to parameterization. On the other hand, the issue is whether actual applications of these operations remain constant across languages, where the semantic representation remains constant, so that all languages use the same applications of Agree and wh-movement, for example to construct equivalent questions. Again, the second, stronger version of this conjecture is much more controversial than the first. Let us assume, in the absence of strong evidence to the contrary, that these basic operations of syntax in the Minimalist sense are not subject to parametric variation (though cf. Baker, 2010). The question remains, however, whether there are other grammatical parameters that might constitute a problem for the Un-Cartesian hypothesis. There are many points of variation that in the GB era were analysed as grammatical parameters. While these points of variation often concern aspects of language that appear to have nothing to do with the kinds of issues we have been discussing thus far (for example, word order, the realization of arguments, or Case), it would nonetheless be relevant for the Un-Cartesian view if they could be shown to constitute genuine instances of grammatical variation. The selection of what is part of ‘grammar’ cannot be made purely on the basis of what is clearly connected to grammatical semantics, as this would render the Un-Cartesian position unfalsifiable.

184


In order to test the hypothesis that grammar is invariant, we therefore need an independent diagnostic such as (2): (2)

If a parameter P feeds/bleeds some grammatical operation, then P must be grammatical.

That is, if a point of variation interacts systematically with one of the operations which we have identified as core components of grammar (e.g. certain instances of head-movement, wh-movement, phases), then said point of variation must be grammatical. Conversely, as long as there is no evidence that a given parameter feeds/ bleeds any narrow grammatical operation, it remains possible that said parameter is non-grammatical in nature (though not necessarily so). Let us then, in the remaining sections, revisit in more detail some of the classic ‘syntactic parameters’ of the GB era in the light of the claim in (2). We note that with the possible exception of the alignment parameter discussed in Section 5.6, these parameters do seem to concern matters of externalization: for example, the order in which morphemes are pronounced, or whether arguments are pronounced at all. Nonetheless, it is perfectly possible that syntax could itself encode such aspects of linguistic organization. It will be argued below, however, that these classical ‘syntactic parameters’ are best interpreted as different externalization options in a given morphological channel. In that sense, they are not properly characterized as syntactic parameters. As things stand, then, ‘syntactic variation’ is not the seemingly obvious fact that it has been taken to be (though other aspects of variation may, of course, prove more problematic than those considered here).

5.4 Revisiting the head parameter One of the core parameters of the GB era is the ‘head-parameter’, a version of which is given in (3): (3)

X precedes/follows its complement.

This states a basic parameterisation in the X-bar template (a core component of UG in the GB sense). Simply put, languages like English set the parameter in (3) to ‘precedes’, whereas languages like Japanese set it to ‘follows’, yielding head-initiality/ finality respectively: (4) [John has [read [the letter]]]

[T [V [O]]]

(5) [John-ga [[tegami-o] John-SUBJ letter-OBJ ‘John read the letter.’

[[[O]V]T]

yon]-da] read-PAST

Japanese


185

Such an approach to word order variation was attractive for several reasons. Firstly, it appeared to provide an explanatory account of the tendency towards ‘harmony’ in natural languages, whereby the various head-complements have a tendency to be ordered alike within a given language. Given the general nature of the X-bar template, the default prediction is that all head-complement pairs will pattern alike. Indeed, Dryer’s (1992) large-scale survey found the following head-complement pairs tended to pattern with the order of verb and object across languages:9 (6) Head-complement correlation pairs (taken from Dryer, 1992: Table 39, 108) (i) (ii) (iii) (iv) (v) (vi) (vii) (viii) (ix) (x) (xi) (xii) (xiii)

verb-object adposition-DP copula verb-predicate want-VP auxiliary verb-VP negative auxiliary-VP complementizer-S(entence) article-noun plural word-noun noun-genitive adjective-standard of comparison verb-PP adverbial subordinator-S(entence)

The parameter in (3) also had the apparent advantage of reducing the differences between Japanese and English to a difference in order, whilst maintaining constant their constituent structure. This version of the head parameter was syntactic, then, in that it involved a manipulation of the core phrasal template. In the context of Bare Phrase Structure (Chomsky, 1995), on the other hand, a syntactic parameter of this kind is impossible to formulate. According to the definition in Chomsky (1995), Merge yields an unordered set, in terms of which the notions precede/follow are meaningless. While it has been proposed that Merge should actually yield ordered pairs (Zwart, 2009), the mainstream view is now that the order between head and complement should be imposed only at the mapping to PF. One influential proposal in this vein has been to reinterpret Kayne’s (1994) influential Antisymmetry hypothesis as a linearization mechanism (cf. Chomsky, 1995; Moro, 2000). Kayne (1994) proposed a formalization of the connection between linear order and hierarchical structure:

9 This is not to say, however, that complete harmony across all head-complement pairs is the norm. See the papers in Biberauer and Sheehan (forthcoming) for discussion of disharmonic word orders.

186


(7)

Linear Correspondence Axiom (LCA, Kayne, 1994: 6) [For a given phrase marker P, where d is the non-terminal to terminal dominance relation, T the set of terminals and A the set of ordered pairs such that for each j, Xj asymmetrically c-commands Yj], d(A) is a linear ordering of T.

(8)

Informal implication of the LCA Categories higher in a given syntactic structure will precede those lower in said structure.

The LCA in (7) basically makes it impossible that Japanese and English share the same constituent structure and implies, rather, that the hierarchical relations between categories must be reversed in Japanese as compared with English. The basic corollary of Kayne’s approach is thus the need to encode all word order variation in the syntactic component. If the LCA-based view of word order is correct, in short, the existence of head-final languages like Japanese provides immediate support for the existence of syntactic variation (cf. Kayne, 1998; 2004; Cinque, 1999; 2005; Biberauer, Holmberg, and Roberts, 2008, forthcoming, for a decidedly syntactic view of word order, as compared with the PF-approaches in Richards, 2004; 2008; Abels and Neeleman, 2012; Sheehan, forthcoming a, b). The advantage of this proposal is that variation reduces to parametric differences in the application of internal Merge (movement) across languages. The Un-Cartesian position, like the recent Minimalist one of Berwick and Chomsky (2011), however, naturally favours the idea that word order variation is determined by more superficial PF-parameters, possibly relative to individual categories (e.g. head-parameter, spec-parameter, pronounce head/foot, etc.). It is worth considering, then, to what extent the head-parameter passes our diagnostic in (2), repeated here as (9): (9)

If a parameter P feeds/bleeds some grammatical operation then P must be grammatical.

To what extent, then, does head-complement order feed/bleed head-movement, whmovement or Agree, or have an impact on the phasal architecture of a given language? This is a question of some scope and all we can offer here is an impressionistic view of the cross-linguistic picture. It has been argued that head-final languages like German lack V-to-T-movement (cf. the evidence in Vikner, 2005), and this has sometimes been taken as evidence that head-movement to a phrase-final head is banned. Note, in this regard, that head-movement to C is possible in German, as discussed in Chapter 4, but that CP is a head-initial phrase in German. Kayne (1994) further observes that there do not seem to be any verb-penultimate languages, suggesting that head-movement to the right is universally banned. However, given that, in most cases, head movement in a head-final language will be string vacuous, it


187

is difficult to rule out its existence across the board (cf. also Richards, 2008, for critical discussion). This is doubly the case because, as Sheehan (forthcoming c) shows, adverbs cannot generally intervene in head-final sequences. The matter thus remains unresolved and we leave it to future investigation. Kayne (1994) further noted that very many head-final languages seem to lack whmovement and provided an elegant account of this based on the single specifier condition. In simple terms, if spec CP is already occupied by a TP complement then this blocks wh-movement. A combination of the features of the World Atlas of Language Structures shows, however, that this initial impression may not hold up to a broader cross-linguistic sample. Disregarding those languages listed as having ‘no dominant order’ on WALS, we get the following results for OV and VO languages: TABLE 1. Order of verb and object (Dryer 2011a) and position of interrogative phrases (Dryer 2011b) Initial interrogative phrase (264)

Not initial interrogative phrase (614)

Mixed (23)

VO (705)

156

253

15

OV (713)

76

317

6

Although the vast majority of OV languages for which there is data lack whmovement, there are nonetheless 76 OV languages which have wh-movement. Strictly speaking, though, what is relevant to Kayne’s (1994) proposal is whether the head whose specifier would host a wh-phrase is phrase-final, as it is well known that many OV languages have a phrase-initial complementizer layer (Biberauer, Holmberg, and Roberts, 2008). At a rough approximation, we might consider this head to be that which surfaces as a question particle in polar questions. If this is the case then we can also test for instances of Q-final languages with wh-movement. Disregarding those languages which either lack question particles or in which the question particle is neither final nor initial, we get the following results: TABLE 2. Position of polar question particle (Dryer, 2011c) and position of interrogative phrases (Dryer, 2011b) Initial interrogative phrase (264)

Not initial interrogative phrase (614)

Mixed (23)

Initial Q-particle (130)

65

35

4

Final Q-particle (313)

31

215

6

188


Here again we see a strong skewing away from wh-movement in Q-final languages, but there are nonetheless 31 Q-final languages which permit wh-movement. Assuming that at least some of these are genuine counterexamples, it would appear that head-complement order does not bleed wh-movement, contrary to Kayne’s original hypothesis, though there is an undeniable preference for OV and Q-final languages to be wh-in-situ. The fact that this tendency is not absolute is perhaps not surprising given the fact that Kayne’s original explanation of the effect faces certain technical problems.10 Finally, in support of the claim that head-finality results from syntactic movement, Kayne (1994) observes that while many postpositions display agreement morphology, prepositions generally do not. He attributes this to the fact that Agree takes place in spec-head configurations, and postpositions involve leftwards movement of a DP to the specifier of P. Once again, however, this generalization may not hold up to empirical scrutiny on a broader sample. According to the data on WALS (Bakker 2011; Dryer 2011d), the instances of person agreement on prepositions and postpositions are broadly comparable. Disregarding those languages with no dominant order of adposition and noun, we get the following results:

TABLE 3. Adposition-noun order (Dryer, 2011d) and person marking on adpositions (Bakker, 2011) No adpositions No person Person marking (63) marking (209) with pronouns only (83)

Person marking with pronouns and nouns (23)

Postpositions (577)

12

95

35

13

Prepositions (512)

9

76

32

9

Table 3 shows that the agreement patterns for prepositions and postpositions are broadly similar so there is no evidence that head-complement order feeds/bleeds Agree in this context.11 Finally, can head-complement order affect the phasal architecture of a given language? One contrast which is potentially relevant to this issue is the fact that head-final languages such as Japanese appear to permit subextraction from specifiers, unlike head-initial languages like English (cf. Lasnik and Saito, 1992). To the extent 10

The most obvious being that subsequent LCA-based approaches to word order have either retained the single specifier condition and rejected the possibility of local comp-to-spec movement (cf. Kayne, 1998; Aboh, 2004; Cinque, 2005) or rejected the single specifier condition (Biberauer, Holmberg, and Roberts, forthcoming). 11 Note also that Kayne’s proposed analysis relies crucially on spec-head agreement, which is also difficult to state in Minimalist terms.


189

that islandhood is connected to phasehood, this could be taken to indicate that the two types of languages differ with respect to whether specifiers are necessarily phases or not. Sheehan (forthcoming a) provides a PF-account of this difference, however, whereby specifiers are phases in both cases and the difference stems from the requirement of PF to impose an unambiguous order of terminals. In this case, then, there may be a PF-explanation for an apparent interaction between word order and a syntactic operation. As such, this initial cursory investigation suggests that the head-parameter does not immediately qualify as a syntactic parameter.12 Head-complement order does not feed/bleed other narrow syntactic operations, and as such, while a salient and important property of language, it may not be syntactically encoded (either in the X-bar template or as a movement operation). It is worth considering, then, whether PF approaches to word order can account for the wealth of empirical evidence cited in favour of the LCA (cf. Abels and Neeleman, 2012, for related discussion). Kayne (2009) mentions the following (cf. also Kayne, 1994; 2004): a. Specifiers surface to the left, not the right;13 b. There are no V-penultimate or clitic-penultimate languages, but there are V2 and clitic-second languages; c. AuxV and VAux differ in that VAux order typically requires adjacency between V and Aux; and d. There are AuxOVX languages (e.g. many Mande languages), but there seem to be no XVOAux languages. As Biberauer and Sheehan (2012) note, most such asymmetries basically reduce to (a) or the Final-over-Final Constraint, a descriptive constraint over disharmonic headcomplement orders: (10)

12

The Final-over-Final Constraint (FOFC—Biberauer, Holmberg, and Roberts 2008; 2009; forthcoming):14 If Æ is a head-initial phrase and is a phrase immediately dominating Æ, then must be head-initial. If Æ is a head-final phrase, and is a phrase immediately dominating Æ, then can be head-initial or head-final.

There are other interesting connections between word order and apparent parameters of variation which also deserve further consideration, such as the apparent lack of verb-medial ergative languages (Mahajan, 1994) and the fact that virtually all object-initial languages are ergative (Levin, 1983; Manning, 1996). Roberts (2010) takes the first of these gaps as evidence that the head-parameter is encoded via movement in the narrow syntax. 13 It has been suggested that Sign Languages may constitute an exception to this generalization (cf. Cecchetto, Geraci, and Zucchi, 2006, but see Abner, 2010 for counter-arguments). 14 The version of FOFC given here is the first version offered by BHR and it subsequently undergoes revisions, to accommodate apparent counterexamples. I abstract away from such complications here but cf. Biberauer, Holmberg, Roberts, and Sheehan (in progress) for a broad empirical assessment of the constraint.

190


FOFC amounts to the observation that of the four potential combinations of pairs of head-complement dyads, (11a)–(11c) are fairly well attested, whereas (11d) is either unattested or exceedingly rare (cf. Bazalgette, Roberts, Sheehan, and van der Wal, 2012 for statistical evidence): (11)

Harmonic and disharmonic combinations:

β⬘

(a) αP γP

β⬘

(b) β

α

β

β

αP α

β⬘

(c)

γP

(d)

∗

αP

αP γP

β⬘

α

α

β γP

The evidence for an asymmetry of this kind is discussed in the above papers as well as Biberauer, Newton, and Sheehan (2009), Biberauer, Sheehan, and Newton (2010), Sheehan (forthcoming b), and Biberauer, Holmberg, Roberts, and Sheehan (in progress). One example, discussed by Biberauer, Holmberg, and Roberts (2008) is the lack of V-O-Aux surface orders which holds in all synchronic and diachronic varieties of Germanic, as well as Finnish and Basque, languages which permit all other possible permutations of V, O and Aux: (12)

Lack of *V-O-Aux a. O-V-Aux German and dialects of German, Dutch and its dialects, Afrikaans; Old English, Old Norse; b. O-Aux-V or so-called verb-raising/VR structures: Swiss German dialects, Dutch and its dialects, Afrikaans; Old English, Old Norse; c. Aux-O-V or so-called verb-projection raising/VPR structures which involves a head-initial TP and a head-final VP: Swiss German dialects, Dutch dialects, spoken Afrikaans; Middle Dutch, Old High German, Old English, Old Norse; d. V-Aux-O required for CP-complements in German, Dutch, Afrikaans and their dialects; possible with PP-complements in Dutch and Afrikaans and, to a lesser extent, German; possible with DPs in Old English and Old Norse; e. Aux-V-O English, Mainland Scandinavian, Icelandic; Old English f. *V-O-Aux unattested.15 [summary based on BHR (2007: 97)]

The same gap appears to surface in other syntactic contexts: *[C-TP]-V, *[Pol-TP]-C, *[Asp-V]-T, *[V-O] . . . C, suggesting that this is a general word order restriction.

15

(i)

A-bar VP-fronting gives rise to a superficial exception to this otherwise robust gap: I asked him to pay the bill, and pay the bill he did.

Given that these structures involve non-local A-bar movement of VP (past the subject) they fall outside the constraint as described in (10) because VP is not dominated by AuxP in its derived position (cf. Biberauer, Holmberg, and Roberts, 2008).


191

Work by Kayne, Cinque, and collectively by Biberauer, Holmberg, and Roberts has shown that it is possible to provide an analysis of these asymmetries in an LCAbased model by restricting the application of movement operations. Proposals which attempt to relocate word order variation in the PF-component face the challenge of providing an alternative explanation for such asymmetries.16 As FOFC is the only asymmetry concerning head-complement order, rather than specifier-phrase order, a crucial question is whether it can be captured in a PF-approach to the headparameter. In previous work, one of us (Sheehan) has argued that all of the asymmetries discussed by Kayne (2009) actually reduce to one: (13)

If no order is determined between two categories A and B by other means, then A precedes B if A asymmetrically c-commands B.

In a sense, (13) can be considered a very weak version of the LCA, which stems from processing pressures. Crucially, though, (13) retains the explanatory power of the LCA without forcing PF-motivated movements into the syntax. As long as there is a systematic syntax-PF connection, the kinds of asymmetries discussed by Kayne (1994, 2004) and Biberauer, Holmberg, and Roberts (forthcoming) can also be explained as PF effects.17 Let us assume that while phrase structure is uniform across languages, heads are specified in language specific ways to either precedep or followF the head which they select. The first stage of linearization orders heads based on this very local information. Head-head orderings are efficient in that they do not overspecify linear order (they are maximally local). They are inefficient, however, in that they often underspecify linear order (where word order is disharmonic, rather than consistently head-initial/head-final). In such cases asymmetric c-command is used to resolve the under-specification. (14)

Revised LCA: (i) If a category A c-selects (and c-commands) a category B, then A precedes/ follows B at PF. (ii) If no order is specified between A and B by the sum of all precedence pairs defined by (i), then A precedes B at PF if A asymmetrically c-commands B.18

16 That is assuming that these asymmetries have a formal rather than a diachronic or functional explanation. Cf. Sheehan forthcoming b and Bazalgette, Roberts, Sheehan, and van der Wal (2012) for discussion of one potential functional approach to FOFC and some of the problems it faces. 17 More generally, it is presumably always the case that linearization is always parasitic on narrow syntactic relations and/or constituent structure, given the well-known fact that transformations are always structure-sensitive. 18 One question is why (ii) is not parameterized in the same way that (i) is. Biberauer, Roberts, and Sheehan (2013) propose that for parsing reasons, (ii) is a no-choice parameter, always set to precedence.

192


This revised LCA explains the Final-over-Final Constraint as a PF effect. The basic explanation of FOFC works as follows. With harmonic combinations (specifiers omitted), the sum of local head-parameters yields a single unambiguous linear order because of transitivity: (15)

AuxP Aux>V, V>O AuxP

VP VP

(16)

O

=Aux>V>O

AuxF V>Aux, O>V AuxF

VF VF

O

=O>V>Aux

Crucially, note that (16) has a harmonic head-final order without the need for movement, and with the same c-command relations as the head-initial order in (15). The revised LCA is not violated in such cases because c-command is not invoked: the local ordering of heads taken transitively, yields an unambiguous ordering of all categories. Now consider the disharmonic combinations: Disharmonic combination 1—initial-over-final (with linear order represented for exposition purposes only): (17)

Aux P Aux P

Aux>V, O>V

VF O

VF

Here, the sum of locally determined PF-parameters leaves the order of terminals underspecified: no order is specified between Aux and V. By (14ii), Aux asymmetrically c-commands O, and so must precede it giving the linear order Aux >O>V. This ‘inverse FOFC’ order is therefore expected to be possible in the languages of the world, as appears to be the case. Now consider the other disharmonic (FOFC-violating) final-over-initial order: (18)

AuxF V>Aux, V>O VP VP

AuxF O


193

This time it is known that both Aux and O must follow V because of the head parameters in (18), but again no order is specified between Aux and O themselves. By (14), Aux again asymmetrically c-commands O, and so must precede it. This gives the unambiguous linear order V>Aux>O rather than the FOFC-violating order *V>O>Aux. Note that V-Aux-O is an attested word order (cf. (12) above).19 The crucial idea behind this account of FOFC is that there is no syntactic ban on the FOFC structure per se, it is just that a [[head-complement]-head] structure will never be linearized in such a way as to give rise to the FOFC surface order. In this section, we have considered whether head-complement order should be syntactically encoded or not. In the absence of evidence that head-complement order interacts with syntactic operations, we have tentatively concluded that the head parameter is a PF matter. We have further argued that a PF head parameter can account for the word order asymmetries and gaps which are often take as evidence for the claim that word order is syntactically encoded (cf. Richards, 2004; 2008 for further examples of this).20 In the following section, we consider another classic GB parameter in a similar way.

5.5 Revisiting the null subject parameter Another of the classic GB parameters is the ‘Null Subject Parameter’, which regulates the (non)-pronunciation of subject pronouns (Chomsky 1981: 240; Rizzi 1982: 117). Thus English contrasts with Italian in that subject pronouns must generally be overtly realized in finite clauses in the former but not the latter: (19)

Canto spesso sing.1SG often ‘I sing often.’

(20)

*(I) sing often.

Italian

This point of variation is of particular interest as it has been argued that a large number of seemingly unrelated properties correlate with the non-pronunciation of subject pronouns. This has been taken as evidence that the realization of subjects is not merely a PF affair but rather the result of a syntactic parameter with complex surface effects.

19

Note that the revised version of the LCA provides an explanatory account of FOFC as a PF-effect, even where phrasal movement is concerned (cf. Sheehan, 2010b, forthcoming a). 20 Clearly linearization is much more complex than this simple picture, however. Verb initial orders cannot be derived simply via a head-parameter, for example, and would rather have to involve realizational/PF-parameterization of the kind discussed in Chapter 4. We leave a full investigation of these issues to future work.

194


Neeleman and Szendrői (2007) offer an explicitly PF-only analysis of radical prodrop in languages like Chinese and Japanese, which often lack agreement morphology on the verb and unlike Italian, permit the omission of any pronominal argument of a verb or noun: (21)

a. siken-ni otita. exam-DAT failed ‘pro failed the exam.’

Japanese

b. Bill-ga settokusuru. Bill-NOM persuades ‘Bill persuades pro.’ c. [mimi-ga] nagai. [Japanese, from Neeleman and Szendrői (2007: 672–3)] ear-NOM long ‘pro’s ears are long.’ In their terms, these languages permit any argument to be dropped because of a morphological zero spell-out rule (which applies subject to pragmatic principles). (22) [KP +p, a] , zero

[Neeleman and Szendrői (2007: 682)]

Indirect evidence for such a rule comes from the observation that only languages with agglutinating nominal morphology appear to permit radical pro-drop of the Japanese kind. In Neeleman and Szendrői’s terms this is because in languages with fusional morphology, the application of (22) is blocked by the elsewhere principle (Kiparsky, 1973). In languages with agglutinating morphology, on the other hand, (22) applies freely as it spells out an entire KP and other morphological rules target proper constituent subsets of KP (i.e. K and DP). In fusional languages, however, more specific morphological rules for KPs block the application of the zero spell-out rule. If this analysis is along the right lines, then, radical pro-drop may not be the result of a syntactic parameterization, but rather of a fairly superficial morphological effect: the rule in (22). This raises the question of the status of (22). It is clearly morphological rather than syntactic but is it a language-specific rule induced on the basis of data or a universal principle which is only visible in non-fusional languages? Neeleman and Szendrői discuss the possibility that (22) is a universal morphological rule but dismiss this idea in favour of the proposal that it is simply acquired as a rule. The pervasive nature of (22) in unrelated languages could be due to 3rd factor pressures, however, whereby non-pronunciation is preferred over pronunciation. The frequent occurrence of (22) only in languages with agglutinating morphology yields the appearance of systematic parametric variation but if Neeleman and Szendrői are right this is simply an effect of the elsewhere principle. So in reality all we have is a third factor constraint (the Elsewhere Principle) which conditions morphological variation in a principled manner. That said, it remains to be seen, as in


195

the case of word order, how the radical pro-drop parameter interacts with core processes in grammar and their interpretive effects. To the best of our knowledge, no such interaction has been reported and so we tentatively conclude that radical pro-drop relates only to externalization. This raises the question of the status of null subjects in languages like Spanish and Italian, which do not permit generalized argument drop.21 Neeleman and Szendrői (2005, 2007) explicitly claim that agreement-licensed null subjects instantiate a partially separate phenomenon, whereby zero spell-out of fusional pronouns is dependent on the presence of agreement morphology. This is in keeping with work by Holmberg (2005), Sheehan (2006), and Roberts (2010) who all argue that null pronominals in many null subject languages (NSLs—Finnish, Spanish, Italian, European Portuguese) are deleted pronouns. Combining Neeleman and Szendrői (2005) and Müller (2005), we can write this nullness into the morphological (externalization) component of the relevant languages:22 (23)

a. [KP +p, a, çi] , zero / T[çi] b. T[çi] wherever T fails to be subject to a morphological impoverishment operation that leads to a neutralization of ç-features.23

Once again, according to (23), the availability of null subjects in L is dependent only upon morphological properties of L, this time the lack of impoverishment of T. As to why so many languages with rich subject agreement permit null subjects, this could again be due to the fact that third factor economy pressures favour a rule such as (23) to exist in languages with non-impoverished phi-features on T. If this approach is along the right lines and the null subject parameter is also effectively a PF (morphological) parameter, then the prediction, again, is that it will not be encoded in syntax or feed/bleed syntactic operations. In this regard it has been noted that many NSLs appear to require verb-movement into the inflectional domain. Barbosa (1995) and Alexiadou and Anagnostopoulou (1998) go as far as to claim that it is verb movement in these languages that satisfies the subject requirement. There are reasons to disfavour such a view, however, and to prefer a view whereby null subjects are independent of verb movement. One such reason is that there are non-NSLs with head-movement (e.g. French: Pollock, 1989) and so headmovement appears to be independent of a positive setting of the null subject 21 For reasons of space we do not discuss semi- and partial pro-drop languages, which require finer grained distinctions (cf. Biberauer, Holmberg, Roberts, and Sheehan 2010). 22 As Neeleman and Szendrői (2005) note, here the elsewhere condition will not block (23) in a language with fusional nominal morphology. 23 Cf. Müller (2005) for a discussion of impoverishment in relation to a number of languages. He assumes that impoverishment takes place presyntactically, because he takes pro to be licensed narrow syntactically. If null pronouns are simply deleted then impoverishment can be post-syntactic, as is more standardly assumed in Distributed Morphology.

196


parameter. Schifano (2012) finds a strong correlation between the position in which the verb is spelled-out morphophonologically and the tense/mood/aspect distinctions displayed on the verb, suggesting that verb movement within the inflectional domain is morphological in nature (though this is not her conclusion). Other apparent interactions appear to exist between subject pro-drop and both Agree and wh-movement. A closer inspection of the facts reveals that the connection is to rich agreement morphology, however, rather than the operation Agree per se. Non-NSLs like English and French also have agreement with the subject but lack null subjects. The connection between morphological richness and subject pro-drop is captured by (23). The connection to wh-movement is more complex. Rizzi (1982) noted that NSLs like Italian differ from some non-NSLs like English with respect to wh-extraction of subjects. Whereas English and French display that-trace effects, Italian (like Spanish and Portuguese) does not: (24)

Chi credi che partirà? who think.2s that leave.FUT.3s ‘Who do you think will leave?’

(25)

Who do you think (*that) will leave?

(26) Qui est-ce que tu crois who is-this that you believe ‘Who do you think will leave?’

Italian

qui/*que who/that

partira? leave.FUT.3s

French

At first sight then, this appears to be an instance of the null subject parameter interacting with a syntactic operation. The availability of subject extraction in NSLs across a complementizer is attributed by Rizzi (1982) to the fact that (i) NSLs also permit post-verbal subjects and (ii) wh-extraction from the post-verbal position is unconstrained, unlike wh-extraction from the preverbal subject position: (27) Ha telefonato has telephoned ‘Gianni rang’.

Gianni Gianni

Italian

The problem with this account is that it is not clear what actually prevents subject extraction across a that-type complementizer in English and French. Pesetsky and Torrego (2001) and Rizzi and Shlonsky (2007) propose syntactic analyses of the effect, but Kandybowicz (2006) argues that a PF account is superior. The first reason for this is that, as Kandybowicz (2006: 221) notes, even some dialects of English are reported to lack the that-trace effect (cf. Sobin, 1987). Moreover, prosodic factors seem to mitigate the effect, notably the presence of an adverb (Kandybowicz, 2006: 221, citing Bresnan, 1977: 194, Barss and Déprez, 1986, and Culicover, 1993): (28)

Who do you think that after years and years of cheating death __ finally died?


197

Moreover, a nuclear pitch on or contraction of the auxiliary following the complementizer also yields a grammatical result for many speakers (Kandybowicz, 2006: 221): (29)

A: I didn’t think that John would survive. B: Well then, who did you think that __ WOULD?

(30)

?Who do you suppose that’ll leave early?

Given these facts, it seems unlikely that the that-trace effect is a syntactic effect after all (cf. Kandybowicz, 2006 for a PF-analysis). Of course, there are other differences between NSLs and non-NSLs which require explanation, notably differences in the scope of indefinite subjects and the availability of free inversion (cf. Sheehan, 2010a, for an analysis of inversion in NSLs compatible with a PF account). Pending a full investigation of such effects, we tentatively conclude that the null subject parameter is better modelled as a PF-parameter built into the morphological component of language.

5.6 Revisiting the alignment parameter Thus far we have considered two of the parameters of the GB era and tentatively concluded that they may not be syntactic parameters. This is perhaps not surprising given that both points of variation concern manifestly morphophonological properties of language: word order and the overt/covert realization of pronouns. In this section we consider a parameter which appears, at the outset, to be more morphosyntactic in nature, and argue that the result is slightly different (a more foundational discussion of Case must wait until the next chapter). Some languages display accusative alignment in that the subjects of intransitive and transitive clauses pattern alike with respect to Case and/or agreement: (31)

She is kissing her

(32)

She is eating

Other languages, however, display ergative alignment in that the object of a transitive and the subject of an intransitive pattern alike with respect to Case and/or agreement and the transitive subject patterns differently:24 (33) Angute-m qusngiq ner-aa. man-ERG reindeer.ABS eat-TR.3s/3s ‘The man is eating (the) reindeer.’

Yup’ik

24 Given space restrictions, we will not address the problem of split-ergativity here, despite its obvious relevance. The reason for this is that recent research suggests that it may be illusory. Coon and Preminger (2012), building on work by Laka (2006) suggest that where subjects in imperfective aspects fail to receive ERG it is because they are not actually transitive subjects. See Sheehan (2013) for a discussion of other kinds of splits in this context.

198


(34) Qusngiq reindeer.ABS ‘The reindeer is eating.’

ner' -uq. eat-INTR.3s [Yup’ik, Eskimo-Aleut, Bobaljik (1993: 3)]

Early approaches to ergativity took it to result from a thematic parameter. Thus Marantz (1984) and Levin (1983) pursued the idea that thematic roles in syntactically ergative languages are assigned in different configurations than in accusative languages (cf. also Dowty, 1991). Updating the machinery slightly, this would give the following: (35) Accusative:

[vP agent v [VP V theme]]

(36) Syntactically ergative:

[vP theme v [VP V agent]]

Since thematic structure is part of grammatical semantics on the present account, this is precisely the kind of variation that is problematic for the Un-Cartesian hypothesis. It is interesting to note, therefore, that this thematic view of syntactic ergativity has now almost universally been rejected, with the consensus being rather that the differences between ergative and accusative systems concern only differences in Case/Agreement rather than thematic structure. The reason for this is that there are certain syntactic operations which are sensitive to thematic structure and these function the same way in both ergative and accusative systems (e.g. incorporation, idiom formation, anaphor binding: Dixon, 1994). Subsequent approaches to alignment, on the other hand, posit non-thematic but nonetheless syntactic parameters in order to capture the basic ergative/accusative contrast. Thus, Bobaljik (1992, 1993) proposed the Obligatory Case Parameter, which regulates which Agree relation is obligatory in intransitive clauses in a given language. In accusative languages, the relevant head is ‘AgrS’, in ergative systems it is ‘AgrO’ (in a model where Agree relations are encoded as functional heads): (37) accusative trans: [AgrSP Agent AgrS [AgrOP Theme AgrO [vP . . . ]]] accusative intrans [AgrSP Theme/Agent AgrS [vP . . . ]] (38) ergative trans: [AgrSP Agent AgrS [AgrOP Theme AgrO [vP . . . ]]] ergative intrans [AgrOP Theme/Agent AgrO [vP . . . ]] Because in ergative languages the AgrO head is always present/active, accusative Case is always assigned and so the subjects of intransitive predicates get the same Case as the objects of transitives. In accusative languages, AgrS is always present/active and so the subject of an intransitive receives the same Case as the subject of a transitive predicate. While this approach does not involve the parameterisation of thematic structure, it nonetheless involves the parameterisation of basic Agree relations. Once again, though, this kind of approach faces certain empirical problems. The prediction of (37)–(38) is that Ergative Case (ERG) should pattern like Nominative Case (NOM), whereas the Absolutive Case (ABS) on Themes/Agents in intransitive


199

clauses in ergative languages should pattern with Accusative Case (ACC). We know of no language where this prediction holds, however (cf. Manning, 1996, Aldridge, 2008, for discussion). Rather, ergative languages seem to pattern in one of two ways: Pattern 1: ERG and SABS pattern like NOM, OABS patterns like ACC (e.g. Basque, West Greenlandic—Anderson, 1976; Manning, 1996) Pattern 2: ERG patterns like an inherent Case and ABS patterns like NOM (e.g. Dyirbal, Mayan High ABS languages—Dixon, 1994; Coon et al., 2011) Pattern 1 languages appear to have an accusative syntax with surface ergative morphology, whereas Pattern 2 languages appear to have an ABS which functions like NOM. No language of which we are aware has the properties predicted by (37)–(38). Any account of ergativity needs to capture these two distinct patterns, and the Obligatory Case Parameter does not appear to be able to do so. Note, nonetheless, that Pattern 2 strongly suggests that ergativity is not only a morphological phenomenon. Further indication of this comes from the fact that apparently all Pattern 2 languages as well as some Pattern 1 languages display a further syntactic effect of ergativity: the ban on wh-extraction of ERG DPs: (39) *Maktxel max-ach who ASP-ABS2 intended: ‘Who saw you?’

y-il-a’? ERG3-see-TV [Q’anjob’al, Mayan, Coon et al. (2011)]

(40) *angutii [ti aallaat tigu-sima-sa-a] man.ABS gun.ABS take-perf-rel.tr-3s ‘the man who took the gun’ [West Greenlandic, Eskimo-Aleut, Manning (1996: 84)] This very pervasive effect appears to suggest that alignment cannot simply be a matter of morphology, rather there must be a syntactic difference between accusative and what we will term ‘syntactically ergative systems’. This is different from the thattrace effect observed in accusative systems as it affects only transitive subjects and does not appear to be mitigated by prosodic factors. In fact syntactically ergative languages use a valency-changing strategy to alter the Case of the Agent in order that it can be extracted, usually an Antipassive: (41) yabu [bural-ŋa-ŋu mother.ABS see-ANTIPASS-REL ‘Mother, who saw father, was returning.’

ŋuma-gu] banaga-nyu father-DAT return-NONFUT [Dyirbal, Dixon (1994: 170)]

All of this shows that ergativity is not a uniform phenomenon; languages differ as to how ergative their syntax is. Some languages seem to be only morphologically ergative, as has often been noted in the literature (cf. Anderson, 1976), others display a basically accusative syntax but show the restriction on the extraction of ERG DPs. Finally, a small number of languages appear to have a high ABS, equivalent to NOM

200


in an accusative language and also display the ban on wh-extraction of ERG DPs. This kind of variation is what we would expect if Case is partially a reflection of syntax, but also of the ‘translation’ of syntax into partially independent, morphophonological terms (Sigurðsson, 2012). We return to this view in Chapter 6. One way to capture these (and other) patterns is via an interconnected set of ‘micro-parameters’, following the format in Roberts (2012) (see Sheehan, 2013): (42)

Alignment parameter hierarchy:

Basic alignment parameter: Does transitive ‘v’ assign theta-related ERG to its specifier in L? N Accusative (Russian...)

Y Split-S parameter: Do all ‘v’ s in L assign ERG?

Y Morphologically Split-S (Chol, Basque)

N Syntactic ergativity parameter: Does VERG bear an EPP feature in L?

N Morphologically ergative (Walpiri)

Y High/low ABS parameter: Does VERG assign structural Case in L?

Y Low ABS (West Greenlandic, Tagalog)

N High ABS (Dyirbal, Q’anjob’ al)

The two parameters at the top of this parameter hierarchy are morphological in nature and concern the language-specific marking of a universal theta-relation. The lower two parameters are, however, ‘syntactic’, in the sense that they concern (i) the presence/absence of a movement trigger associated with the head in question, and (ii) whether this head enters into a grammatical relation with a lower DP marked by Agree. It is, on this account, the presence of the Extended Projection Princicple (EPP) (the movement trigger) which gives rise to the ban on ERG DPs, as the presence of the ABS DP in spec vP acts as an intervener (cf. Aldridge, 2004; 2008; Coon et al., 2011). It is in those cases where this v head fails to assign a structural Case that ABS is equivalent to NOM (Legate, 2008). Is the existence of syntactically ergative languages problematic for the UnCartesian hypothesis? This depends on whether there is any evidence that the two syntactic parameters in (42) have semantic import. Interestingly, in the case of the EPP feature, such evidence does exist. It has been noted in a number of unrelated ergative languages displaying the ban on wh-extraction of ERG DPs, that the DP which receives ABS must be obligatorily definite/specific:


201

(43) B-in-ili ng babae ang isda. TR.PERF-buy ERG woman ABS fish ‘The woman bought the/*a fish.’ [Tagalog, Austronesian, Aldridge (2008: 984)] This follows if movement of the ABS DP past the ERG DP forces a wide scope reading on the former (cf. Aldridge, 2004; 2008 for discussion). It is not clear, at present, how general this pattern is, though it is also attested in West Greenlandic (cf. Manning, 1996). If it holds more generally, then this small syntactic difference between accusative and (syntactically) ergative systems also has a minor semantic impact, as the Un-Cartesian hypothesis would predict. In sum, this brief consideration of ergativity has argued, in line with much recent work, that there is no deep thematic parameter or parameterization of Agree relations. There do appear to be fairly low-level syntactic differences between (some) ergative and accusative systems. This does not undermine the Un-Cartesian approach, however, inasmuch as these minor syntactic differences appear to correlate with minor semantic differences.

5.7 Conclusions This chapter has considered linguistic variation and whether it affects grammar as described here. Grammatical semantics is rather uncontroversially universal; nor do the principles of reference and predication as described here form natural dimensions of expected variation, since the grammatical semantics in question is arguably conceptually non-distinct from the grammatical patterns in which it arises. Crosslinguistic variation could suggest that an alternative, non-grammatical system is what accounts for the formal ontology of semantics, but this does not seem to be the case— that is, the syntactic variation most scrutinized in the generative tradition is not of that kind. Indeed, it turns out that many classic GB syntactic parameters might well plausibly be reformulated as PF-parameters affecting only ‘externalization’ (morphological operations, vocabulary insertion, linearization). In particular, this may well be the case with the head parameter and the null subject parameter. In the case of alignment, we have seen evidence of a syntactic parameterization based on the fact that ERG DPs cannot be wh-extracted or relativized in a diverse range of ergative languages. Following Sheehan (2013), there may be a ‘hierarchy of micro-parameters’ to explain the attested classes of ergative languages and one-way implications. Crucially, the syntactic parameters in said hierarchy actually appear to have some semantic import, in which case a change in grammar does entail a change in meaning, as the Un-Cartesian view would predict.

6 The rationality of Case 6.1 The apparent irrationality of Case The interpretation of Chomsky’s (2007a) ‘radical’ conception of the semantic interface explored here as the Un-Cartesian hypothesis maintains that there effectively is no such interface: there is a grammatical and a non-grammatical (lexical, perceptual) way of organizing semantic information, but there is no autonomous semantic or syntactic component. We therefore do not expect autonomous syntactic principles or operations. With this in mind, this chapter turns directly to a property of natural language that is often taken to provide some of the strongest evidence for syntactic autonomy, described under the traditional descriptive category of ‘Case’: in many languages, argument-nominals (DPs) are marked morphologically as being ‘accusative’, ‘nominative’, ‘dative’, and so on. But what on earth, we might wonder, does the morphological case-marking in the English word him add to its meaning, namely a particular person we refer to? What does it add to the word Peter, if we assume that such nominals, though not morphologically case-marked, can be ‘abstractly’ marked for it? In these questions, however, we see standard assumptions about meaning at work, which we have questioned here: meaning is based on the word, words refer to things in the world. Case marking, then, seems like an irrelevant flourish. It is, we conclude, something merely ‘syntactic’. Since a notion of grammatical meaning is missing, we further conclude that it is an arbitrary constraint, a ‘filter’ imposed by grammar as an autonomous system on nominals in argument positions—though we might concede that, occasionally and perhaps not very systematically, Case marking relates to thematic roles, or to event-semantics. Case, then, is (largely) irrational. It makes no deeper sense. In line with this conclusion, we probably cannot find a single Introduction to the Philosophy of Language that even discusses the phenomenon in question. To illustrate the situation insofar as linguistics is concerned, Svenonius (2007: 20) notes that ‘attempts to [ . . . ] reduce structural cases to semantically based feature systems [ . . . ] go back hundreds of years and have invariably failed’. Pesetsky and Torrego (2011: 52) state that the principles of Case assignment ‘look quite specific to syntax and morphology, with little apparent connection to external cognitive systems’. This

6. The rationality of Case

203

partially echoes the claim by Chomsky (2000b: 127) that ‘[s]omething like Ł-theory is a property of any language-like system, whereas [Case] checking theory is specific to human language . . . ’. If so, in a Minimalist frame, where interface explanations are a prime rationalization of narrow syntactic elements and operations, Case raises particularly severe problems of rationalization: why does it exist? In Chomsky’s (2000b; 2001) Agree-based framework, Case is the ‘uninterpretable feature’ par excellence, to be deleted in syntactic representations before ‘transfer’ to the semantic interface (i.e. semantic interpretation) takes place. It is said to be ‘motivated . . . by legibility conditions’, in the sense that Case exists in order to render DPs active as goals in the narrow syntax, but this proposal fails to give Case any rationale independent of the autonomous narrow syntax, and it faces other problems, which we discuss below. We think that the Un-Cartesian programme provides a perfect opportunity to rethink Case. In this chapter, we will contend that Case is utterly rational: it morphologically reflects the grammatical process of licensing a referential nominal to the heads of two higher phases, v and C, of which the nominal becomes a part. The correlate of this is that the nominal will receive relational interpretations: its reference is now resolved in relation to the referents of the higher phases, becoming a participant in the events and propositions they denote as speech acts take place. This process does not happen with adjuncts, or with clauses. What Cases mark, therefore, are the cross-phasal grammatical relations arising in the build-up of referential meaning and the inherent formal ontology that the latter has in grammar. The Cases and the formal ontology of semantics, therefore, are correlative. Without the one, the other wouldn’t exist. Case is rational in this sense. It would be irrational only if grammar was. But it isn’t, as we have argued here. The idea that Case is uninterpretable arises against a lexical, non-relational, and non-grammatical conception of meaning. Nothing in this account entails that morphological Case marking must exist in a language. ACC, NOM, etc. are morphological features, which as such cannot be the same as the grammatical relations they morphologically manifest. The latter are improperly described as Cases, we will argue, for how these relations map onto the morphological features in question is a many-to-many, and both intra- and crosslinguistically highly variable affair. A feature such as ‘ACC’, therefore, doesn’t have a single grammatical meaning—though important tendencies can be seen, as seems already clear from Section 4.4 above. In line with this, we make a terminological decision: Since, in any case ACC, NOM, etc. are morphological features and we think that the grammatical relations that such features manifest are not properly said to be ‘Cases’ for the reason just given, we reserve the term ‘Case’ for the basic, and unquestionably morphological phenomenon we are concerned with here. This is also in line with our general practice in this book to write terms such as ‘Person’ with a capital when we mean the technical grammatical category rather than the

204


common noun or a potential semantic category. In short, Case as the term is used here is a morphological phenomenon that translates syntactic patterns into morphological terms. Before proceeding, we stress what this chapter is about, and what it is not about. The chapter is about the interpretation, and the possible rationalization of a phenomenon in light of its currently known empirical features. It is not presenting a new theory of Case, or expanding the empirical coverage of given theories. So it is, if you like, a philosophy of Case. We will proceed as follows. In Section 6.1, we articulate the above proposal in more detail, putting it squarely on the table. Section 6.2 tracks the evolution of the notion of the Case Filter in generative grammar and the notion of ‘abstract Case’, concluding that no such autonomous constraint exists as part of grammar viewed as distinct from morphophonology, and that it is only problematically assumed even within more standard assumptions in the Minimalist framework. Section 6.3 argues that, fundamentally, the issue is not ‘abstract Case’ but the need to license referential expressions against the referents of higher phases into which they enter, which includes a crucial connection between Case and Tense. Finally, in Section 6.4, we discuss evidence that has been taken to suggest a rather direct connection between Case features such as ACC and semantics, and more specifically the mereological structure of events. We argue against the idea that ACC has such a direct semantic rationale. Its rationale, rather, lies in grammar and how the latter organizes referentiality and the formal ontology of meaning. The sensitivity of Case assignment to the formal-ontological state-event distinction suggests a similar conclusion. Overall, we conclude that the challenge of Case to the rationality of grammar, if grammar is rethought on Un-Cartesian lines, disappears.1

6.2 What do the Cases mean? In Chapter 4, we have mainly considered how the phasal architecture serves to construct units of referentiality within both the clausal and nominal domains. But as described in Chapter 3, such phasal units also compose in a specific way, 1 This does not mean that there are no other challenges. Indeed, there are plenty, and the Un-Cartesian programme will need to address them one by one. An obvious other challenge is islands, which are virtually routinely taken to be irrational, on the grounds that a sentence like ‘*Who did Susan ask why Sam was waiting for __?’ expresses a perfectly well-formed thought, namely: which person was such that Susan asked Sam why he, Sam, was waiting for him/her. But all that this observation seems to amount to is that thoughts need to be expressed by the sentences by which they are indeed expressed—the second one, in this instance. Why, that is, should we have the expectation that a given sentence could express what another sentence expresses? As for why sentences like the first are ungrammatical, this needs to follow from how adjuncts are structuralized relative to arguments, and which grammatical relations can be established across which phasal boundaries. Since phases and the adjunct-argument distinction are interpretable in grammatical semantics, it is not clear why islands, if and to the extent that they are an epiphenomenon of this phasal dynamics, should not arise.


205

generating a dialectic of reference and predication that ends with the assignment of the truth value. In the course of this dialectic, a complex formal ontology arises by phase, which crucially has a part-whole structure to it: thus, as reference to individuals is made, these will necessarily contain substances as part of their constitution; objects of reference that are formally events, in turn, will necessarily contain objects (event participants); causative events will contain states; propositions will contain events; and propositions will be parts of speech acts. Since CP necessarily contains vP, which necessarily contains DP, we may think, in line with Chapter 3, of DP as the ‘first phase’, of vP as the ‘second’, and CP as the ‘third’. As a DP transgresses its first phase boundary and enters the second and third phases, until it is licensed as a sentential argument, cross-phasal dependencies form. These correlate with the relational interpretations that all referential argument-nominals have. To illustrate these interpretations, consider (1): (1)

I read.

Here the personal, nominative DP is simultaneously a speech act participant (the logophoric agent), the grammatical subject of a predication that is part of the proposition that I read (whose truth is claimed in this speech act), and an event participant (the agent of the reading). That is, the grammatical argument ‘I’ expresses a set of cross-phasal relations into which it has entered. These relations are established between the nominal and the heads of the higher phases, with v and Voice being of particular importance. Thus, construction types with different such heads, such as the active/passive, the causative, the unaccusative, and the anticausative constructions, differ in their argument structure, which in turn shows up in differences in Case-assignment. For example, when we start from the transitive construction in (2), then the ‘passive’ Voice in (3a) and the unaccusatives construction in (3b) changes Case-assignment leading to ACC-to-NOM conversion: (2)

We [voice-v killed himACC]

(3)

a. HeNOM [voice-v was killed]. b. HeNOM [voice-v died]

As is clear from these examples, the Voice-v complex affects Case-assignment: different such heads instruct the morphology in different ways. What, then, do such markers as ‘NOM’ and ‘ACC’ in (2) and (3) mean? They indicate that the DP has undergone relational licensing with respect to the higher phase(s) of which it has become a part, the interpretive correlate of which is that they have become parts of referential expressions with a different formal ontology. The Cases, where they are present morphologically in a language, therefore mark the cross-phasal relational structure of grammar, which is interpretable in grammatical semantics. DP denotations need to become interpreted relationally in addition to referentially, as event- and speech act participants.

206


Chomsky (2001: 6) suggests that in regular accusative systems, ACC is assigned in the domain headed by the phasal head v, when that head is in a technical sense ‘complete’ or ‘non-deficient’ (= ‘v*’), while NOM demarcates the next phase, where it is assigned a value under Agreement by the Tense head (T), again when this head is relevantly ‘complete’. If so, the Cases are immediately rationalized, as long as these phases themselves are interpretable, which they of course are in the present framework. These ‘structural’ Cases, that is, will demarcate phasal domains—more precisely, those phasal domains incorporating another phase with a different formal ontology, as when an event-denoting phase contains an object-denoting one (this qualification would explain why D as such is not a Case domain). Thus formulated, however, the rationalization of Case still assumes, with much GB and Minimalist work, that ‘NOM’ and ‘ACC’ are names of entities that figure in ‘narrow syntax’ and UG, corresponding to ‘syntactic features’ that require ‘checking’ under ‘Agree’ in the course of the syntactic derivation. We will argue against this below, in line with Sigurðsson (2012), who also argues (p. 206) that NOM-assignment does not involve T, contrary to the assumptions of a long tradition in generative grammar (Lasnik, 2008), but is regulated by a complex of functional heads in the expanded domain of v, including ‘Voice’ and applicative heads. In this approach, NOM is an ‘elsewhere case’, assigned to a nominal in PF whenever syntax has not transferred any specific Case instructions for it within the domain of v (NOM is thus ‘no Case’). ‘High’ movement of a nominal into the C-domain is then not ‘for Case’, nor for reasons of the Extended Projection Principle (EPP), but for reasons of establishing interpretable ç-feature relations, in particular Person. Sigurðsson (2012) also reinterprets Chomsky’s principle of ‘star-augmentation’ as encoding an instruction for the morphology to interpret syntax, and expands it to include assignment of Dative and Genitive. Our proposal will significantly cohere with this proposal and its implication that ‘syntax does not operate with case features’ (Sigurðsson, 2012: 193), which instead are partially arbitrary and language-specific morphological expressions of a range of underlying grammatical relations. Even if the core Cases do not correspond in a one-to-one fashion to the phasal domains v and C, however, as Chomsky’s model entails, they would be semantically interpretable in grammatical semantics. For they reflect the cyclic licensing of arguments in different domains, which is inherently an aspect of grammatical semantics. This is crucially not because the Cases inherently relate to thematic roles. The latter, too, we have argued, are an aspect of grammatical semantics and an aspect of how arguments are interpreted relationally (and see Bowers, 2011). Thus, in (3) above, the thematic role of the internal argument remains identical as its Case changes, and NOM is marked on both ‘we’ and ‘he’, which carry distinct thematic roles. Rather than to thematic roles, the Cases relate to argumenthood, more precisely the argumenthood of canonical DP-arguments, assuming that adjuncts and clausal or PP-arguments are exempt from structural Case-marking, and need not be licensed in


207

the way that referential nominals need to be. Since adjunction involves a different (grammatical) semantics than argument taking, it follows that Case, if involved in licensing referential arguments, is interpretable not thematically or as adjuncts are, but as arguments denoting objects viewed as having entered events denoted by the heads of higher phases. In (3a) and (3b), ‘him’ and ‘he’ both carry Case, as expected from the fact that they are both arguments. This means, in terms of their grammatical semantics, that there are not two referents, an event and an object (a kicking and a male person), in these constructions, but one referent, an event of kicking, of which the object has become an inherent part. In other respects, however, the interpretations are different, as again the grammar predicts: thus, in (3a), the NP is part of the predicate; in (3b), it is the subject. Subject and predicate are neither thematic notions nor non-grammatical semantic ones (but of course both are interpretable as grammatical semantic ones, since in combination they give rise to a truth value). Neither is the notion of ‘argument’ a merely thematic or non-grammatically semantic one. In fact, to understand the mereological notion of ‘part’ just invoked, which as such is rather vague and metaphysical, we need to invoke grammatical relations. Thus, when I say (4), then both Jim and Dan are, purely semantically speaking, ‘parts’ of the event of seeing that is being referred to: (4)

I saw Jim with Dan.

However, ‘Jim’, the argument, is a part of it in a different way from Dan. For example, (5) is an entailment of (4), while (6) is not: (5)

I saw Jim.

(6)

*I saw with Dan.

This entailment obtains for no semantic reasons, but grammatical ones: the way in which Jim, the grammatical argument, is connected to the event of seeing, is different from the way in which Dan is, the adjunct. It is in the nature of adjuncts that they can be dropped, leaving the reference of the object to which they attach intact. On these grammatical grounds, (5) follows from (4).2 Thematic roles are not sufficient to license the formal ontology in question, predicting that the relations marked by the morphological Cases, where these exist in a language, take us further into the realm of grammatical meaning than thematic relations do, which adjuncts can carry as well. Equipped with thematic roles alone, there wouldn’t need to be propositions, or subjects in a language, and there couldn’t be. There might be ‘arguments’ in a purely 2 This is the essence of the argument presented in Reichard and Hinzen (2011) according to which Davidson (1967) is wrong that positing a (non-linguistic) semantic ontology of ‘events’ as objects explains why the entailment obtains. This semantic ontology is explanatorily inert. The actual explanation is purely grammatical, as seen in the text, where no reference to any extra-linguistic objects is made.

208


logico-semantic sense, but not in a grammatical one. There could be a complete language with thematic roles that only had a capacity to conjoin predicates, combining adjuncts with adjuncts only, without ever yielding propositional information or truth values.3 Such a language would not need Case. But it wouldn’t encode propositional information either. We will technically assume, with much linguistic literature (Kratzer, 1996; Pylkkänen, 2008; Sigurðsson, 2012), that different arguments are licensed in the extended edge of the verbal domain (second phase) by different kinds of functional heads: specifically, external arguments/subjects are licensed by (e.g. active or passive) Voice, direct objects by v-type heads, and indirect objects by an ‘applicative’ head (Appl). So as we go beyond the first phase and enter the second, (definite) nominals are brought into matching relations with at least these three distinct kinds of head: (7)

[Voice . . . NP3 . . . [Appl . . . NP2 . . . [v-V . . . NP1]]]

There is no a priori reason why we should expect that such relations would show up in the morphology. Thus it is not surprising that in many languages (100 of 261 languages in total studied by Iggesen, 2008), there is no morphological reflex of such relations in Case assignment. That there can be such a reflex is equally unsurprising, however. Where it exists, it will make perfect sense that the relevant markers will show up in two places, like on a noun and a verb, given that grammar is relational and that such markers are interpreted in terms of grammatical relations. This then has nothing to do with any such marker being ‘uninterpretable’. The relations in question are as such fully interpretable in terms of (speech-)event licensing. There is not only no grammatical or communicative requirement for overtly marking the relational structure of grammar in morphology, i.e. for Cases, but we expect the problem of mapping grammatical relations to a sensory-motor surface to be solved in many languages in many different ways. That is, a one-to-one correlation between morphological Case-markers and the underlying relations they manifest is not expected and would come as a surprise. Nor does such a correlation seem available either cross-linguistically or even within a single language, as Sigurðsson (2012) stresses (and see further on in this section). That said, morphology does interpret syntax in the instance of Case, as when a particular Case interprets the (non-thematic) grammatical relations between a direct object and an event-denoting verbal head. We have also noted evidence from Romance object clitics and differential Case marking in Chapter 4 that ACC-marking rather systematically correlates with the dimension of grammar that we have identified as referentiality.4 Also interesting

3 Truth values are an optional ingredient in the Neo-Davidsonian ‘minimal semantics’ of natural semantics, based on the operation of adjunction and thematic roles, sketched in Pietroski (2011). 4 Differential subject marking also exists. It is not clear how similar it is to differential object marking, but interestingly it is sensitive to the Person/no Person distinction in Dyirbal (Dixon, 1994).


209

in this regard are languages that make Case distinctions amongst pronouns but not full DPs (e.g. English, French, Italian, Spanish: Asbury, 2008); and relatedly, Case marking can show up on determiners but not nouns, as in German. Both of these facts make sense: pronouns, which are closely related to determiners, embody forms of grammatical reference in a purer form than content nominals. Hence the languages in question again illustrate a correlation between grammar, Case-marking, and reference. In Section 6.4, we will discuss further evidence from event semantics below that a morphological category such as ACC can, at least in some languages (e.g. Finnish, German, and English), track aspects of event reference quite consistently (cf. van Hout, 1996; 2004, Kratzer, 2004; Borer, 2005; Travis, 2010). There can be little doubt, then, that, to a significant extent, Case morphology tracks grammar and the genesis of referential meaning in it: it reflects the licensing of referential arguments interpreted as relations to events, including speech events.5 We come back to this in Sections 6.4 and 6.5. Despite the fact that such markers track grammatical relations, it remains true, at the same time, that any single morphological Case-marker such as ACC can express a great number of different underlying relations. For example, ACC in Icelandic, Sigurðsson (2012: 199) lists, can mark: ‘some subjects; most direct objects; some indirect (‘first’) objects ( . . . ); regular ECM subjects; most ECM predicative NPs; subjects in some experiencer ECM complements; some raised NPs; complements of certain prepositions; some dislocated NPs; certain adverbial NPs’. In turn, different Case-markers can indicate the presence of the exact same syntactic relation. Thus in languages with ‘inherent’ cases (dative, genitive, instrumental, ablative, partitive, comitative, etc.), these inherent Cases can all be used to mark the relation between the verb and the direct object, as in hundreds of ordinary verbs in Icelandic (Thráinsson, 2007: 208f.). Moreover, two closely related languages such as German and Icelandic can map the equivalent syntactic relation onto different Cases, even with semantically equivalent predicates: (8)

a. Hún she

kastaði threw

steininum/*steininn. stone.the.DAT/stone.the.ACC

b. Sie hat den Stein/*dem Stein she has the.ACC stone/the.DAT stone ‘She threw the stone.’

geworfen. thrown

Icelandic German

5 According to Caha (2009), also, the morphological syncretisms which languages display in their morphological Case marking paradigms are not arbitrary, but are rather subject to universal principles. This raises challenges for the claim that morphological Case marking is independent of syntax. McFadden (2012) claims that only the ‘inherent’ Cases behave as Caha claims. We leave a full exploration of this idea to one side here.

210 (9)

The Philosophy of Universal Grammar a. Hún she

stýrði steered

skipinu/*skipið. ship.the.DAT/ship.the.ACC

Icelandic

b. Sie hat das Schiff/*dem Schiff gesteuert. German she has the.ACC ship /the.DAT ship steered ‘She steered the ship.’ [Sigurðsson (2012: 10–11)] Even in the same language, two verbs very close in meaning can surface with objects marked with different morphological Case markers (cf. German gratulieren (DAT) and beglückwünschen (ACC), both meaning ‘congratulate’, McFadden, 2004: 96). The following Icelandic example from Andrews (1982) is another famous example: (10)

a. Mig me.ACC

velgir am.nauseated

við by

setningafræði syntax

b. Mér býður við me.DAT am.nauseated by ‘I am nauseated by syntax.’

setningafræði syntax

Icelandic

The possibility of sociolinguistically determined variability in Case assignment is also attested (see Sigurðsson, 2009: 23, for an Icelandic example). This is arguably also the explanation for the fact that morphological markers of Case can clearly vary over time (cf. Allen, 1995, on the loss of dative subjects in the history of English). A great multiplicity of factors, then, can influence how Case morphology will end up interpreting grammatical relations: Case-assignment, which proves to be sensitive (to varying degrees) to (i) lexical idiosyncrasies, (ii) the DP in question’s thematic role, (iii) sociolinguistic factors, (iv) the syntactic position/dependencies of said DP, and (v) the presence and status of other DPs in its local domain (cf. Marantz, 1991; Baker, 2010). The same bewildering multiplicity shows up in Case inventories themselves. According to Calabrese (1996: 86), not counting vocative, Kabardian has only two morphological Cases (NOM/ACC), Modern Greek has three (NOM/ ACC/GEN), Ancient Greek had an additional morphological case (DAT), Latin and Old High German both had (different) five-way distinctions, Turkish and the Slavic languages have six morphological cases, and Central Armenian has seven. Many languages lack morphological Case altogether as noted (e.g. Mandarin, Thai) (Iggesen, 2008; 2011). Some languages assign such markers based on loosely semantic criteria (possibly subtypes of thematic-roles), but these patterns are not semantically based in any universal sense (McFadden, 2004). This ‘Case chaos’ makes it seemingly impossible to regard a morphological category such as ‘ACC’ as a ‘syntactic’ category or a category of UG, rather than a descriptive category for how different languages reflect syntax in language-specific and variable ways. It also raises questions for the proposal that there should be something like ‘abstract ACC’ (or perhaps ‘abstract ERG’), a term that amounts to carrying a morphological category into syntax, except that in this case, no such


211

morphological category is present. Consider (11), where ACC-marking tracks the syntactic dependency between the internal argument and the active verb: (11)

I saw him/*he.

In (12), there is no such ACC-marking: (12)

I saw Peter.

We would not be inclined to conclude from this that the syntactic dependency between the verb and the object is therefore a different one (11) and (12). But this example is not therefore evidence for the presence of an ‘abstract ACC-feature’ in the syntax underlying (12). It just shows us that a given syntactic dependency need not be morphologically marked in the same way in the pronominal and nominal case. A dependency is not a marker. A marker is a marker of a dependency. In other languages, it is only marked on the verb, or on both the verb and the DP (Baker, 1988). We will discuss the notion of ‘abstract’ or ‘structural’ Case in the next section in more detail. For now we summarize the proposal. Case is rational, for morphological Case markers, if and where they exist in a language, are a manifestation of the relationality of grammar and cross-phasal dependencies that are essential to the relational interpretation of NPs as arguments relative to the denotations of higher heads. The distribution of such markers moreover provides us with an independent pointer to what we regard as the essence of grammar, namely reference. The moment we deprive grammar of meaning intrinsic to it, and locate meaning at the level of the content of thought, the problem of the rationality of Case appears: for Case has no meaning that can be spelled out in terms of lexical-semantic features or else thetaroles. But it is this absence of a systematic notion of grammatical meaning that creates this problem. The problem disappears, when such a notion is adopted. In this sense, the irrationality of Case is based on the idea of an autonomous, non-semantic syntax, rather than independent evidence for it. It is as if, where a rationalization of Case has been sought, the only candidates for meaning associated with it were either lexical or (at best) thematic. Where Case turns out to be assigned non-thematically, and to have no lexical meaning, the conclusion is that it is uninterpretable. But the blind spot in this argument is the notion of grammatical meaning. In fact, the existence of Case provides us with an independent pointer to its existence: for without this crucial dimension of natural language, Case wouldn’t exist.

6.3 Vergnaud’s conjecture and ‘abstract Case’ As noted, there appear to be mismatches between Case assignment and thematic roles as well as between Case assignment and grammatical functions. Thus, subjects

212


that are nominatives can be Themes or Agents (13–14), and in certain contexts, apparent subjects can also be accusatives (15): (13) He arrived

NOM = subject and theme

(14) He was kicked by John

NOM = subject and patient

(15) I believe [him to have killed her].

ACC = agent and subject

These mismatches have been taken as evidence that such markers are a purely ‘formal’ aspect of grammar with no semantic import—reflecting an apparently arbitrary constraint imposed on grammaticality, independent of semantic interpretation. Somehow, the idea is, Case is connected to the licensing of DPs in an autonomous syntactic component. Curious in itself in this argument, however, is that it is DPs that would require such licensing. DPs are nominals with a potential for referentiality, unlike predicative nominals, which may require no structural Casemarking at all. It is natural to suggest, therefore, that what is formally licensed is not some autonomous phrasal type, but an expression that carries out a particular referential function. In the same way, it can hardly be an accident that the ‘exceptional’ accusatives in (15) show up in cases where the subject in question is part of a clause that is referentially weak in terms of our hierarchy, lacking Finiteness and hence deictic anchoring in the speech context. One could suspect from this alone that referentiality will have to be at least a part of what rationalizes Case, as we will indeed conclude. Evidence that (overt) DPs, unlike other phrases, need to be especially licensed comes from observations that DPs appear to be banned in certain syntactic positions, while clausal arguments can surface in some such positions (with minor complications such as extraposition in (17a)): (16)

a. *It seems [John to be right]. b. *I have a feeling [my nephew] c. *John is proud [his son]

(17)

a. It seems [ti to be right [that John is invited]i] b. I have a feeling [that my nephew is in danger] c. John is proud [that his son won]

However, these data again appear to follow without further ado under the natural assumption that DP is the paradigmatically referential category, whereas clauses cannot refer to objects (at least not objects like tables and people). In other words, referential arguments make particular demands on the grammar, which needs to incorporate them as predicates of the referents of higher heads that take them as arguments, as per the theory developed in Chapter 3. This is also what we saw with


213

objects clitics in Chapter 4: demands on the relational interpretation of strongly referential nominals are higher than in the case of more weakly referential or predicative nominals, or embedded clauses, which are always predicates anyhow when integrating into a higher phase (with qualifications for the factive case). The grammar meets these demands by harnessing increasingly complex machinery, as we saw when moving from neutral or predicative clitics to ACC-marked ones, which can be weak or strong, and then to Datives, which we argued contain the accusatives. The same happens in languages with richer morphological Case paradigms, where the embedded referential DPs in examples like (16b–c) require a ‘thematic’ or ‘inherent’ Case such as Genitive (Chomsky, 1986) to be licensed. Otherwise, as in English, a preposition has to be introduced. In the progression towards referentiality, clauses are close to their maximal projection when the C-T system is activated and Tense becomes finite. Such domains are islands for Case-assignment, and if Case is not assigned within them, the derivation collapses. (15), then, repeated here as (18), is the flip-side of this fact: (18)

I v believe [TP him to have killed her].

Where the clause is less projected and Finiteness is lacking, the phase boundary remains penetrable from the outside, and an exceptional Case, assigned by v from within the next phase, shows up (for non-finite clauses that are tensed, see Section 6.4). Where, on the other hand, the embedded clause is projected further and includes a C-layer, such marking is impossible:6 (19)

a. I v hope [CP C (*John) to be at the door]

(20)

b. John tries [CP C (*Mary) to like French toast]

What is ‘licensed’, then, is the referentiality of the DP within the phase of which it becomes a part; and the difference in referentiality in the embedded phases shows up rationally in a difference in Case-assignment. What morphological Case markers mark is not, then, ‘abstract Case’ in narrow syntax, but cross-phasal dependencies that we knew independently exist, to do with the establishment of a formal ontology within a dialectic of reference and predication that proceeds phase-by-phase. (16)– (18) again provide evidence that Case interprets grammar, in our sense; but they are not evidence for anything like ‘syntactic’ or ‘abstract’ Case. A Case marker, which is a morphological feature, manifests underlying syntactic relations, which are not features. Talking of ‘Case-relations in syntax’ is thus to read language-specific morphology into syntax—with the potentially disastrous consequence that we now have to

This is what is technically formalized in Minimalist syntax as C having ç-feature specifications that ‘intervene’ with ECM across a CP ‘barrier’ (Nunes, 2010). This amounts to saying that ECM is a result of an insufficiently projected second phase. 6

214


make sense of how a bewildering variety of Cases, as noted in the previous section, map onto the posited ‘Case-relations’ in syntax. Vergnaud’s (1977) classic idea was that, whether a language has Case morphology or not, there is an abstract universal mechanism which serves to license DPs that occur in certain syntactic positions. In Vergnaud’s original formulation, taken up by Chomsky (1981), the ‘Case Filter’ was taken to apply only to overt DPs: (21)

Case Filter: *NP if NP has phonetic content and has no Case [Chomsky (1981: 49)]

This seemed to explain—without any reference to what we are calling grammatical semantics—why overt subjects in non-finite clauses in English have ‘governed’ (objective) Case, rather than nominative (subject) Case (Vergnaud, 1977: 22). For example, him in a sentence like (18) above, or in It is illegal for him to leave, fails to receive NOM-Case within its own clause and so is only licit if it receives Case from an external governing preposition or verb. Covert nominals were exonerated from this requirement in order to allow for the special behaviour of PRO, the covert subject of non-finite control clauses, whose distribution was regulated via the PRO-Theorem (Chomsky, 1981), and which was taken to not only lack but resist Case.7 Yet again, no such theorem may be required, if we assume that controlled PRO in non-tensed embedded clauses is never independently referential (the point of being ‘controlled’ by another DP), implying that the higher demands on the grammar imposed by referential arguments, mentioned above, will not arise. The Case Filter as expressed in (20) posits a universal constraint within syntax that regulates the appearance of overt DPs in certain positions, irrespective of whether they are marked morphologically for Case or not. It is a constraint on DPlicensing in narrow syntax, interpreted as linked to the morphological Cases, except that these are taken to be possibly invisible. It raises clear problems for rationalization. In GB theory, it was at one point stipulated to hold in order to render NP chains ‘visible’ for theta-role assignment (Chomsky, 1986), but this rationale ceases to be explanatory outside the context of GB theory as chains are no longer taken to exist, and there is no obvious reason why theta role assignment should be parasitic on Case: indeed, Case goes beyond theta-assignment, as noted, reflecting grammatical processes that do not exist in the theta-system. This would be predicted if there is a correlation between Case and the formal ontology of semantics, for thematic roles do not require any formal ontology in this sense. PPs and CPs can receive thetaroles, without receiving Case, showing that the dimension that the Cases mark is a different one. Nor is argumenthood per se the point, for PPs and CPs can be 7

Subsequent research raised problems for this claim. See e.g. Chomsky and Lasnik’s (1993) claim that PRO requires a special ‘Null Case’, assigned by non-finite T, criticized by Hornstein (1999), who assumes that a DP needs to move as many times as necessary to get Case, thus reviving the Case-filter.


215

arguments, too. The reason that DPs need Case, but not PPs or CPs, must lie in the special form of referentiality that we find in the DP domain. Vergnaud’s conjecture that there is ‘abstract’ (‘syntactic’) Case would be very plausible, if indeed the behaviour of referential nominals in languages with poor Case morphology such as English was as if morphological Cases were—invisibly— present, and if, more generally, nominals in a language predictably occurred in positions where such markers, if they existed, would predict them to occur. Thus, as Lasnik (2008: 18) succinctly states Vergnaud’s conjecture: Vergnaud’s now very familiar basic idea was that even languages like English with very little case morphology pattern with richly inflected languages in providing characteristic positions in which NPs with particular cases occur.

The problem with this idea, as Sigurðsson (2012: 191) notes, is that much research has broken the correlation between morphological Cases and ‘characteristic positions’. Icelandic in particular exemplifies a language with ‘rich inflection’, in which there is no direct correlation between ‘particular cases’ and ‘characteristic positions in which NPs ( . . . ) occur’ (Zaenen, Maling, and Thráinsson, 1985; Yip, Maling, and Jackendoff 1987; Sigurðsson, 1989; Marantz, 1991). As Sigurðsson concludes, ‘overt NP-placement is largely independent of the morphological marking of NPs’ and ‘A-licensing is unrelated to morphological case’ (and cannot be accounted for either through some formal substitute, such as the EPP [extended projection principle], either, as Marantz (1991) claimed). The relation between morphological Case markers and grammatical dependencies is rather indirect, reflecting the multiplicity of possible solutions to the problem of mapping underlying relations to features on a sensory-motor surface. With this indirectness goes the naturalness of the idea that syntax might have morphological features in it, except that they are abstract. Yes, a grammatical dependency between v and a c-commanded DP can exist independently of thematic relations (this may be true even in the majority of ergative languages: cf. Section 5.5, and Legate, 2008; 2011). But while this is evidence that DPs enter licensing relations beyond thematic roles, and that Case morphology is not arbitrary, it is no evidence for an ‘abstract Case feature’ in narrow syntax. In a Minimalist context, the Case filter has been re-thought. While maintaining the existence of abstract Case features operative in narrow syntax, Chomsky (2000b) notes a shift from the Case filter as traditionally stated to an Agree-based system based on the need to check ç-features: With this shift of perspective, structural Case is demoted in significance. The Case Filter still functions indirectly in the manner of Vergnaud’s original proposal, to determine the distribution of noun phrases. But what matters primarily are the probes, including ç-features of T, v. That reverses much of the recent history of inquiry into these topics and also brings out more clearly the question of why Case exists at all [ . . . ] [U]ninterpetable features activate the goal of a probe, allowing it to implement some operation (Agree or Move). (Chomsky, 2000b: 127)

216


The new perspective in question is implemented in the context of the Minimalist Program via the obligatory presence of ‘uninterpretable Case features’ on D. Such features, being uninterpretable, must then be ‘deleted’ in the course of the syntactic derivation to avoid a crash upon transfer to the semantic interface. Agree is the narrow syntactic dependency that is established in order to value/delete uninterpretable features of this kind, under c-command. However, the direct connection between NP-licensing and Case morphology is broken. Rather, what triggers movement of NPs into particular positions is the need to check uninterpretable ç-features (Person, Number, Gender) on C-T and v, respectively. Case and ç-feature agreement are taken to be two sides of the same coin, deleted in a single application of Agree (cf. Leland and Kornfilt, 1981; Pesetsky and Torrego, 2001; 2004; Müller, 2009; Zeijlstra, 2012; for critique). So there are no ‘Case-positions’, a result prefigured by the earlier one that Case is not assigned in particular dedicated structural positions via Spec-Head agreement, by the special heads AGR-O and AGR-S, as still assumed in Chomsky (1995: ch. 3) but not Chomsky (1995: ch. 4), where such heads are said to be non-existent. Rather, DP-licensing is governed by the two major probes, v and T, whose feature specifications are assumed to require matching with those of a suitable local DP in their respective domains. But Case-features, like ç-features, are only ‘uninterpretable’ if we ignore the relational nature of grammar. A ç-feature on a verb will come out as uninterpretable on the verb, if the verb or predicate is looked at in isolation, or non-relationally. But this is to think lexically, and that meaning created in the grammatical process would be word-based. Grammatical meaning is not, being purely relational. As illustrated in (1), repeated below as (21), the 1st Person morphologically marked on the pronoun is a morphological feature, which marks a syntactic relation of identity, holding between a reader (the agent of the event of reading denoted by the vP) and the speaker: (22)

I read.

It is therefore rational and expected that in many languages with richer Agreement morphology than English, features such as these will show up in two places, on the verb and the noun. Thus there is no ‘uninterpretable feature’ in syntax here, and nothing needs to ‘checked’ before an ‘interface’ is reached. Everything is as it should be, and no cleansing of syntax is required. The relation ‘Agree’, viewed as a primary syntactic relation, is problematic, then, too: what ‘agrees’ are two morphological features. Such Agreement reflects a syntactic relation, but this does not mean that, in syntax, two ‘features’ exist that are related via Agree. The features exist in morphology, the relation in grammar. Agree is problematic on other, conceptual grounds: usually taken to be a purely formal aspect of grammar, it is designed precisely to eliminate uninterpretable features. So it lacks an independent rationale. Its rationale is to eliminate what has no rationale. It also yields


217

no account of grammar: for there will then only be one single, generic grammatical relation, Agree, viewed as a cover term for whatever specific grammatical relations there might be, depending on which features have to be matched. Linguistic specificity will come from lexical feature specifications only. In this regard, Agree is a descriptive term only. What we need is a principled account of what grammatical relations exist, viewed as primitive elements of grammar.8 On this perspective, Chomsky’s rationale for Case features in syntax, that they render a goal ‘active’ when it needs to ‘agree’ with a given probe, disappears: there is no such need. While we disagree with the standard Minimalist model of Case as an uninterpretable feature, the step from this model to ours is nonetheless not far, if we interpret the phasal dynamics that is described in Minimalism in our present terms. We assume, with Chomsky, that where a star-augmented v is involved, it will license both the relation of the verb with its direct object and the external argument, as per Burzio’s (1981; 1986) generalization and Chomsky’s interpretation of it in Chomsky (1995: ch. 4), giving rise to a complex event containing another as a part.9 With star deletion, ACC fails to be assigned by v, leading to unaccusativity; and in the same way, T can occur deficiently when deprived of C, in which case DPs are not licensed in their internal domains either, leading to the ECM phenomenon or raising. As for ‘high’ movement of arguments into C, when the rationale cannot be the existence of uninterpretable Case-features in syntax, we follow Sigurðsson (2012) in the suggestion that the need to compute values of ç-features is an independent rationale for such movement: NPs need to be evaluated, not only as event participants, but also as to whether they are Persons, and if so, whether they are one of the speech event participants. This process involves the high left periphery of the C-phase. Only in this way, a propositional claim will arise, and hence such high movement is rationalized as well. There are Person, Number, and Gender relations marked morphologically by ç-features, then, but no ‘abstract Nominative Case’ that as such provides problems for rationalization. It is thus only a small step from the current Minimalist take on Case expressed in Chomsky’s quote above, to the present more ‘radical’ model, where the assumption of ‘purely formal noise’ in syntax that is ‘up for deletion’ before the tribunal of the semantic interface is given up: the relevant notion of ‘semantic’ here is not grammatical, by definition and on architectural grounds. Where a grammatical notion of meaning is employed, the focus is on the irreducibly relational structure of grammar, not the feature specifications of particular lexical items. As noted, we do not further our understanding of such relations by positing only one such relation, the generic relation Agree, as holding between whatever features we might lexically 8

This point was raised by Alex Drummond in conversation. See Pesetsky and Torrego (2011: 58) for further evidence based on Wurmbrand (2001) that it is not the lexical verb but v that is involved in the ACC dependency. 9

218


find. This, effectively, would deprive us of a theory of grammar, or of what the grammar does. For all the work would be done by the lexicon. Our viewpoint is therefore different: morphological Cases, where they exist in a language, reflect underlying grammatical relations that are universal and responsible for part-whole relations between the referents of three phases that are ordered in part-whole relations.10 In the next subsection, we reinforce this conclusion, (i) by providing further arguments against its interpretation as implying a ‘Case-filter’, and (ii) by reinforcing an apparent correlation between Case, reference and Tense.

6.4 The Case filter, reference, and Tense There is a crucial difference between ç-features and Case-features on our account: the former reflect a core process of grammar that is at the heart of grammatical semantics, while Case morphology only indirectly relates to (various) grammatical relations, interpreted both in terms of licensing and the formal ontology of semantics. Case assignment in morphology reflects the result of an externalization process that is, like the marking of the Agree relation itself, variable and multi-factorial: whether it is marked (externalized) on DP as ACC (dependent-marking) or the verbal head, as ç-feature agreement (head-marking) or both or neither appears to be a purely a morphological matter with all possible combinations attested in the languages of the world (cf. Bickel, 2011 for a recent study of so-called ‘head’ and ‘dependent’ marking).11 In Chomsky’s (2000b; 2001) system, too, there is a crucial asymmetry between ç-features found on verbal heads and those located on D. Firstly, defective T/v heads can lack ç-features (and hence lose the ability to ‘assign Case’), so there is a sense in which these ç-features are not a necessary property of verbal heads. Secondly, as Preminger (2011) argues at length, Agree is actually a fallible operation. Even where T/v bear ç-features, this simply means that Agree must take place if possible. In the absence of any visible/active goal to agree with, Preminger argues, the ç-features on T/v simply receive a default value. Case features, as they are generally thought of, appear to be different, however. If the Case Filter holds, then the implication is that DPs must always form grammatical dependencies with some head in the verbal projection. Where no such dependency can be established, the derivation will crash, making Case the only feature which is a true ‘derivational timebomb’ (cf. Preminger, 2011). This makes sense if the grammatical dependencies

This echoes a conclusion we earlier reached in Chapter 2 regarding the parts of speech: these, too, reflect grammar, which provides their rationale; but they are not grammar, and remain language-specific. The rationale for Case, though language-specific, lies in the relationality of grammar, too. 11 Latin has both Case (‘dependent-marking’) and agreement morphology (‘head-marking’, in the sense of Nichols, 1986). Dyirbal has Case but no agreement morphology, Swahili has agreement morphology without Case and Mandarin has neither. 10


219

marked by Case are an integral part of grammar, but morphological Agreement is not. It also confirms that Case features and ç-features are not on a par. Moreover, within Chomsky’s own system evidence exists that the rationale of Case cannot be to license Agree. It is now well documented that ‘inactive’ DPs lacking unvalued Case features can act as defective interveners, suggesting that a kind of Agree can take place even where a goal is inactive. Thus, an experiencer contained in a PP blocks raising because it intervenes in the Agree relation between matrix T and the embedded subject (cf. McGuinness, 1998; Hartman, 2011):12 (23) Gianni sembra (??a Piero) fare il suo dovere Italian Gianni seems.3SG to Piero do.INF the his duty Lit. ‘Gianni doesn’t seem to himself to do his duty.’ (adapted from McGuiness, 1998: 92) (24) Jean semble (??à Marie) avoir du talent French Jean seems.3SG to Marie have.INF some talent ‘Jean seems to Mary to have talent.’ (adapted from McGuiness, 1998: 90) (25)

a. Cholesterol is important (*to Mary) to avoid. b. John was claimed (*to Bill) to have stolen the art. c. The hurricane threatened (*me) to destroy my house. [Hartman, 2011]

This, combined with the above fact that T/v’s ç-features can receive a default value, suggests that a system with uninterpretable ç-features could actually function without ‘abstract Case’. Again, therefore, the need to Agree does not justify the existence of Case, in conformity with our account above. Other doubts against the universality of the Case Filter within standard syntactic assumptions have come in two forms. On the one hand, it has been suggested that the Case Filter ought to be ‘parameterized’ to hold only in some languages (cf. Diercks, 2012, for a recent proposal of this kind in relation to some Bantu languages, but Halpert, 2012, for counterarguments). Secondly, a number of people have recently argued against the existence of the Case Filter in any language. On one version of this view, Case becomes akin to all other ‘uninterpretable features’: Case assignment is allowed to fail. If so, Case features do not appear to be different from other syntactic features that are fallible in Preminger’s (2011) sense. Licensing relations between DPs and v and C, however, insofar as they give rise to the formal ontology of semantics, seem to be universal. If so, it makes sense to distinguish Case assignment from the relations in question, which seem to have an independent status. The radical alternative Hartman (2011) offers a potential explanation as to why the English translations of the first two examples are grammatical. He suggests that the constraint nonetheless holds universally, as suggested by the fact that other raising verbs are elsewhere subject to the constraint (e.g. (34c)). 12

220


would be that such licensing relations are radically different across languages, giving rise to varying formal ontologies as a result. Marantz (1991), McFadden (2001; 2004; 2012), and Landau (2006) have all attempted to ‘reduce’ ‘abstract’ Case to morphological Case.13 But as Marantz acknowledges, there is still no explanation for the distribution of PRO in this attempt. In recognition of this fact, Marantz is left with what he calls ‘the residue of structural Case’, which he formulates as a stipulation:14 an NP argument is PRO iff not governed at S-structure by a lexical item or [+ tense] INFL. These perennial impasses in the theory of PRO and the Case Filter may be addressable in a more principled way if we start from the basic idea that grammar governs referentiality, and referential arguments need to be licensed as such, which cannot happen in phases that contain them if these phases are insufficiently projected. This is a constraint that naturally would not apply to PRO, insofar as it lacks referentiality of its own, as noted in the previous section. Licensing of overt referential DPs, by contrast, will be naturally obligatory. Moreover, DPs enter several licensing relations cyclically, so as to give rise to the formal ontology of language, with v-DP relations encoding event licensing and C/T-DP relations capturing speech event licensing. In this architecture, everything is interpretable, there is no ‘abstract Case’, and issues about motivating apparently arbitrary ‘formal constraints’ do not arise. If the Case filter held, on the other hand, there would be no reason to expect that it was a ‘PF’- or externalization requirement holding of overt DPs only. It is thus interesting to see that the claim that PRO lacks Case has come to be seen as empirically problematic.15 There are languages such as Russian, Icelandic, and Ancient Greek suggesting through Case concord that the subject position of a nonfinite clause can in fact be licensed by the head of the clause. Consider the following data from Russian, from Landau (2008): (26) Ivan pokljalsja druzjam [PRODAT sdelat' eto sam/samomu zavtra]. Ivan.NOM vowed friends to.do it himself.NOM/DAT tomorrow ‘Ivan vowed to his friends to do it alone tomorrow.’ NOM OK—73%; DAT OK—45% (27) Ona poprosila ego [PRODAT ne ezdit' tuda odnogo/odnomu zavtra]. she.NOM asked him.ACC not to.go there alone.ACC/DAT tomorrow ‘She asked him not to go there alone tomorrow.’ ACC OK—60%; DAT OK—90% Russian, [Landau (2008: 888–90)] 13 Landau proposes a more elaborate control calculus, whereby clauses are specified as [+/ T] and [+/ Agr] and only clauses which are [+ Agr] [+ T] can host referential subjects, which may or may not be overt. 14 McFadden also admits in his conclusion that he has no viable alternative explanation for the distribution of PRO and the distribution of for in non-finite clauses. 15 It could be claimed that PF needs to know which form of a given DP to pronounce, but this does not seem to be a deep explanation of the existence of ‘abstract Case’. It sheds no light, for example on why multiple case forms exist in the first place (where they indeed do).


221

The fact that the embedded modifier in both examples can optionally get DAT marking strongly suggests that PRO in Russian has Case and that this Case might be licensed by a dependency with non-finite T. However, as Landau shows at length, ‘Case independence’ of this kind is limited to tensed infinitival contexts, namely contexts where an embedded clause can be modified by a temporal adverb which indicates a different temporal anchoring from the matrix clause. Finiteness therefore does not come as a binary contrast: there are finitely tensed clauses, and there are non-finite ones, but these can be tensed or not. Only where they are tensed is PRO ‘Case-licensed’ within the non-finite clause. This again coheres with our above suggestion that referentiality is a notion crucially involved in data usually described as illustrating purely formal ‘Case licensing’. The progression towards referentiality posited in the phasal model of Chapter 3 proceeds from non-finite clauses to finite ones. PRO is licensed in the former indicating a correlation between referentiality, finiteness, and morphological Case. Through the intermediate case of tensed non-finite clauses, in which PRO turns out to be Case-bearing as well, the picture now differentiates further, and is reinforced. The difference between the two types of non-finite clauses is illustrated in the following English examples: (28)

a. *Yesterday Johni managed/began [[ tomorrow].

T]

PROi to solve the problem

b. Yesterday Johni proposed/preferred [[+T] PROi to solve the problem tomorrow]. Verbs like manage/begin and other implicative, aspectual, and modal predicates select tenseless infinitival complements, whereas propose/prefer and other factive, propositional, desiderative, interrogative verbs (at least) optionally select for tensed infinitival complements. Interestingly, even in languages like English, which lack Case concord, there is some evidence that tensed infinitivals differ from tenseless infinitivals in the licensing of PRO. This evidence comes from the existence of partial control. Example (29a) shows that predicates such as meet/kiss/part company require a plural subject: (29)

a. *John met/kissed/parted company early this morning. b. %Johni proposed/preferred [PROi+ to meet/kiss/part company]. c. *Johni managed/began [PROi+ to meet/kiss/part company].

(29b) shows that when such verbs occur in tensed infinitival contexts, such verbs are legitimate even where their controller is singular. (29c) shows that this is not the case where these verbs occur in an untensed infinitival clause. A tensed infinitival clause, in other words, being more referential, licenses its own subject, which in these cases is only partially controlled by the antecedent. Sheehan (2012, building on Raposo, 1989; Sitaridou, 2007; Sheehan and Parafita Couto, 2011, on European Portuguese; and Rabelo, 2004; 2010; Modesto, 2007; 2010, on Brazilian Portuguese) shows that the

222


distribution of inflected infinitives in European Portuguese in Control contexts is also sensitive to the tensed/untensed distinction: inflected infinitivals as in (30–32) are only possible in tensed non-restructuring environments: (30) O professor obrigou os alunosi [a PROi ler(em) o livro]. the teacher obliged the students to read.INF.3PL the book ‘The teacher obliged the students to read the book.’

ep

(31) proi prometemos à Ana [PROi comprar(%mos)=lhe um livro] promised.1PL to.the Ana buy.INF.1PL=her a book ‘We promised Ana to buy her a book.’ [Sitaridou (2007: 195)] (32) %O Pedroi pensa [PROi+ reunir(em)=se the Pedro thinks.3SG meet.INF.3PL=SE.3 Lit. ‘Pedro thinks to meet tomorrow.’

amanhã]. tomorrow

It has been independently established that the inflected infinitive can establish the kind of dependency marked morphologically normally by NOM, hence the fact that overt subjects are possible where inflected infinitives surface as adjuncts, subjects, or as the complements of factive/epistemic verbs: (33) pro escrevi a carta [para (eles) perceberem]. wrote.1SG the letter for them understand.INF.3PL ‘I wrote the letter in order for them to understand.’

ep

(34) Será difícil [eles aprovar*(em) a proposta]. be.FUT.3SG difficult they approve.INF.3PL the proposal ‘It will be difficult for them to approve-Agr the proposal.’ [Raposo (1987: 86)] Moreover, raising is possible only from uninflected infinitival clauses (Raposo, 1989: 297; Quicoli, 1996: 59): (35) Os meninos the boys

parecem seem.PRES.3PL

[ti

ter(*em) have.INF.3PL

razão] reason

ep

As predicted, partial control in European Portuguese is also limited to tensed infinitival contexts and true instances of partial control require inflection (cf. Sheehan, 2012, on instances of ‘fake’ partial control). Taken together, the data from partial control, Case independence and inflected infinitival control all suggest that tensed infinitival clauses differ from untensed infinitival clauses in that the former but not the latter license PRO (cf. Cinque, 2006; van Urk, 2010; Grano, 2012, for the same conclusion based on various subsets of the data). The Case and Agreement data from morphologically rich languages such as Portuguese, Russian and Icelandic further suggest that PRO enters the kind of dependencies morphologically marked through


223

Cases, something which Landau (2000; 2004; 2006; 2008) has argued for at length. These facts make it impossible to attribute all instances of Control to movement, which was one of the crucial functions of the Case filter discussed in the previous section, therefore again weakening its empirical support. The explanatory factors, we suggest again, are different, crucially relating to Tense instead. Only tensed T, it transpires more clearly now, can assign Case. Why is Tense relevant? Tense is essential to the referential anchoring of objects, events, and propositions. Whenever a clause is tensed, in that it has its own temporal reference, it is to that degree referentially complete and independent, and it needs and can take a subject: its phase can in this sense ‘close’. Again this is not evidence, however, for a NOM-feature in syntax; nor is it evidence that it is such a feature that licenses PRO in the tensed case. It is rather evidence that as a phase completes and referentiality is established, argument DPs need to enter licensing relations with the heads that denote entities of which they become parts. Earlier we have assigned ‘high’ or NOM-case with ç-features in the high left field of C, where argument DPs that are persons are checked for their status as speech participants. Such checking requires finite Tense: for without it, there is no proposition, and there is no speech event. Plausibly, the connection we see in the data between the licensing of subjects and finiteness, or Tense in the embedded case, is arguably an aspect of ç-feature checking as well, or a presupposition for it.

6.5 Event mereology and Case Part-whole relations are ubiquitous in language. On the account of the DP we have defended in Chapter 4, the predicates that define the referential potential of predicative nominals form necessarily a part of quantificational nominals, which in turn are necessarily parts of referential ones. Thus, any object of reference qualifying as the man with the Martini would also qualify as a man with a Martini. Similarly, if I refer to John, using the proper name referentially, then my object of reference is necessarily a man called ‘John’, where the name is used predicatively. In turn, if something qualifies as a man with a Martini, then it will also qualify as an instance of man or manhood. Analogously in the verbal case: I can refer to something as an event or a state: Mary’s getting ill, for example, or Mary’s being ill. Again, in this case, Mary cannot get ill without eventually coming to be ill. In both cases, whether event or state, there will be the same property involved as a part: the one encoded in the adjective ill. In both the nominal and the verbal case, these part-whole relations arise as the grammatical complexity increases. The grammar and the mereology thus co-vary. Mereology, however, cannot have an independent status in the present framework, and it is not a primitive descriptor—any more than we claim other semantic notions are, such as ‘event’ and ‘object’. We cannot, therefore, rationalize

224


the existence of any particular Case on the basis of some independently assumed system of mereology that is part of our metalanguage. It would also not fit our proposal in this chapter so far that there is some direct semantic connection between a Case such as ACC and some semantic primitive, such as telicity. Indeed, as we will argue in this section, although a morphological Case such as ACC often does correlate with particular mereological properties of events, the latter arise from other primitives involved in the process of licensing arguments to events, notably the referentiality of the nominal arguments, but also encyclopaedic (nongrammatical) knowledge. As such, the often-claimed connection between accusative Case and telicity arises as a confluence of various factors, it is not direct, and Case cannot be rationalized directly in (non-grammatical) semantic terms, in line with our proposal so far. A connection between Case-marking and event mereology has been long-noted in the semantic literature. In a comprehensive recent discussion, Travis (2010) argues at length that (inner) ASP(ect) is closely involved in ACC. She bases this claim on a wealth of evidence from unrelated languages suggesting that morphological Case is sensitive to the inner aspect of events. We can interpret this claim in different ways, however. One way is that we assume that there is such a thing as ‘inner aspect’ viewed as a semantic phenomenon, and then argue that there is a morphological Case reflecting or representing it. But in the present context, we cannot allow ourselves this starting point. This is because we seek to explain the formal ontology of semantics rather than taking our semantic intuitions for granted and then formalizing them in one or the other semantic metalanguage. Grammar, we contend, doesn’t merely ‘represent’ the independently given aspectual semantics of certain verbs. So we need to scrutinize the claim in question in another way: whether there is something about the grammar of the constructions in question that explains their interpretive properties. If so, these are part of grammatical semantics (and if not, not). This applies to telicity and boundedness in particular. The most famous case study comes from Finnish, a morphologically rich language in which morphological Cases clearly seem sensitive to aspects of semantics that can be described mereologically (Kiparsky, 1998). In Finnish, two object Cases are permitted: Accusative (ACC) and Partitive (PART). Interestingly, the choice between these two does not seem to be idiosyncratic or lexically determined, but is argued to be due to either (i) the inner aspect of the predicate or (ii) the status of the DP-object in relation to the animacy hierarchy. Kratzer (2004) argues that (i) and (ii) should be considered separately, and we focus here only on the aspectual distinctions. It seems that the alternation between PART and ACC on objects is directly determined by the mereological properties of the predicate/DP. As Kiparksy (1998: 1) notes, ‘in its aspectual function, partitive case is assigned to the objects of verbs which denote an unbounded event’:


225

(36) a. Ammu-i-n karhu-a / kah-ta karhu-a/ karhu-j-a Finnish shoot-PAST-1Sg bear-PART / two-PART bear-PART/bear-Pl-PART ‘I shot at the (a) bear / at (the) two bears / at (the) bears.’ (activity) b. Ammu-i-n karhu-n / kaksi karhu-a / karhu-t shoot–PAST-1Sg bear-ACC/ two.ACC bear-PART/ bear-Pl.ACC ‘I shot the (a) bear / two bears / the bears.’ (accomplishment) [Finnish, Kiparsky (1998: 3)] Predicates which are inherently unbounded almost always assign PART to their complements (e.g. states, intensional, motion/contact (touch)). As noted above, this is part of a more general typological trend (Travis, 2010: 136, citing Noonan, 1992). This has been taken as good reason to believe that ACC may be directly involved in the calculation of event semantics.16 However, prima facie the difference between (36a) and (36b) is actually one to do with reference: a specific single bear or several specific bears are said to be shot and killed in the latter. As a result, a specific event is being referenced, involving one or more specific bears, all of which end up shot (and possibly dead), reflecting a state in which the event in question pans out. In the former case, no specific event or state is referenced, and any bear involved may have escaped unscathed. What the data suggest, then, is that ACC-marking in this case reflects a shift in reference: an object is licensed to an event in both cases, as part of the v-DP dependency we have discussed, and this dependency is reflected in Case assignment in both cases. But the progression from PART to ACC suggests an increase in referentiality, in a similar way as we found in event participants denoted by clitics in Chapter 4. One can describe this difference in a mereological language and involve the semantic term ‘telicity’, but this does not appear to be a relevant primitive of the description. Indeed it is not clear at all that telicity is a notion entering into grammatical semantics as an independent primitive. It is true that perceiving reality under the above part-whole ontology reflects what we have called, in Chapter 1, a grammatical ‘perspective’ on reality, rather than an independent, nonlinguistic feature of the external world; but this does not entail the claim that telicity is a primitive of grammatical semantics. Keeping this in mind, let us consider the old claim that certain direct objects serve the function of ‘measuring out events’ in the sense of Tenny (1992; 1994). Thus the activity verb ‘read’ is atelic in its intransitive function, ruling out (37a), but it can be telic if it surfaces with the right kind of direct object ‘measure’ (37b): (37)

a. *John read in an hour b. John read the book in an hour.

16

Kiparsky (1998: 1) argues that Finnish instantiates a ‘semantically conditioned structural [C]ase’.

226


This has led to the question of what determines the class of ‘measures’. While Dowty (1991) argued that only DPs which bear a certain thematic role (incremental theme) can function as measures, more recently Borer (2005) argues that such a position is untenable for reasons we return to shortly. Instead, she proposes that it is DPs bearing ACC that function as measures (cf. Borer, 2005: 49), again making Case interpretable in seemingly directly semantic terms. Kratzer (2004) similarly proposes, building on work by Ramchand (1997) and Kiparsky (1998), that accusative Case in many languages depends crucially on the presence of a semantic ‘telicity operator’ (cf. also van Hout, 2004). Finally, Travis (2010) provides a comprehensive overview of the ways in which inner aspect (telicity) and Case interact, again implying that it is Case which serves to turn a DP into a measure. In none of these approaches, however, is it explicitly stated that the narrow syntactic dependency between v (or some other functional head in the extended verbal projection) and a c-commanded DP is itself interpretable. For Kratzer, for example, Accusative Case is rethought as an uninterpretable [uTelic] feature, which is deleted/valued by an interpretable ‘telicity head’. The telic interpretation, in such cases, comes from the presence of an interpretable telic operator/feature and not from the dependency itself.17 In Borer’s terms, conversely, it is the (derived) position of a DP in spec-AspQ which serves to turn it into a measure. It is not clear, then, to what extent we can really speak about interpretable Case-features in these approaches, and it is not clear whether they really rationalize Case. What, if any, is the status of the semantic notion of ‘measure’ in grammar? Let us recall, to begin with, that there are close semantic parallels between homogeneous/ heterogeneous DPs and vPs. Thus, as Bach (1986) notes, DPs and vPs react similarly to certain kinds of modification: (38) A lot of mud/*dog was in evidence.18 (39)

John slept/*found a unicorn a lot last night.

[adapted from Bach (1986: 6)]

These examples show that mud, a mass noun, like sleep, an atelic verb, can be modified by the quantifier a lot, whereas the count noun dog and the telic VP find a unicorn cannot. Facts such as these have been taken to suggest a primitive Although the paper in which Kratzer discusses this idea is entitled ‘Telicity and the meaning of objective case’, that is, a strong sense remains in which accusative Case and indeed the grammatical dependency it necessitates is still itself uninterpretable, in line with the standard Chomskyan position. The same is true for Pesetsky and Torrego’s (2001; 2004) idea that ACC is a relational property of grammar, established in the course of the derivation, rather than present at the level of the numeration. For these authors, the narrow syntactic Agree relations forced by Case (a uT feature on nominals) are purely syntactic in nature, as they are for Chomsky (2000b; 2001). For us, ACC should properly be thought of, not as an uninterpretable feature, but rather as the grammatical dependency between v and a local DP interpretable in terms of the formal ontology of semantics. 18 It is of course possible to coerce a mass reading of dog, but this is irrelevant to the observation that only unbounded nominals and vPs can be modified by ‘a lot’. 17


227

mereological basis jointly underlying the mass-count and telic-atelic distinctions. Yet, again, what is the theoretical and explanatory status of these mereological notions? For us, facts like the above suggest that when we come to the level of masses/substances and unbounded activities, respectively, grammatical complexity has not progressed enough to license a formal ontology of objects/bounded events. As Hinzen and Uriagereka (2006) show, grammatical complexity of masses/activities is systematically lower than that of objects/events, in parallel ways. The moment a part-whole structure arises from this, as when a single object like an event entails a state as a proper part, the denotation ceases to be ‘homogeneous’, in semantic terms. However, the process of forming the part-whole structure just described, while mereologically describable, is purely grammatically driven. It depends on the left edges of relevant phasal projections, rather than the contents of lexical items and their lexical semantic properties. There does not appear to be a particular connection with Case, in this instance. Following Zwarts (2005: 751), the relevant notions of homogeneity/heterogeneity are commonly taken to be cumulativity rather than divisibility, based on his observation that mass nouns, which are canonically homogeneous, may be non-divisible but must be cumulative (e.g. furniture): (40)

X is homogeneous iff: (i) there are p, q ∈ X such that p + q exist (ii) for all p, q ∈ X, if p + q exists, then p + q ∈ X

As Zwarts notes, this definition has the advantage of applying equally well to different lexical categories (he himself focuses on adpositions). Thus lexical mass nouns and bare plurals are homogeneous in that milk + milk ∈ milk and cars + cars ∈ cars, as are activity verbs (run + run ∈ run). As has long been noted, however, the denotation of a verb as telic or non-telic depends on its grammatical context (cf. Verkuyl, 1972; Jackendoff, 1996; Tenny, 1992; 1994). Thus, unergative ‘activity’ vPs are generally taken to be homogeneous (atelic): (41)

Mary ran for an hour/#in an hour.

But where they surface with what Tenny calls a ‘terminus’, the natural reading becomes heterogeneous/atelic: (42)

Mary ran [PP to the shops] #for an hour/in an hour.19

Heather Burnett (personal comment) notes that there is a reading under which ‘for an hour’ is fine here, namely a reading whereby John spent one hour at the shops. This reading is irrelevant to the point at hand. 19

228


The natural explanation for this fact is the PP involved, which functions as an argument (see e.g. Folli and Ramchand, 2005), especially if we notice that the atelic reading re-appears when we change the preposition:20 (43)

Mary ran towards the shops for an hour/#in an hour.

At the same time, neither (42) with ‘for’ nor (43) with ‘in’ seems to be strictly ungrammatical: while we may break our heads over what kind of event a person uttering such expressions might be referring to, contexts can be fathomed in which either reading would be grammatical, no matter how unexpected the utterance may be: for example, in the case of (42), one in which Mary ran to different shops for one hour in total, or, if ‘shop’ is in the singular, to the same shop repeatedly, also for an hour, or to the same shop only one time, but then in an effort that led to success in precisely one hour, while others might have achieved this in much less time. In (43), in turn, the task may have been vaguely set for Mary to run in the direction of the shops; and comparatively to others, she did it in an hour (maybe it took her long to get started). This suggests that the data in question are not part of grammatical semantics, or are so only insofar as different lexical items trigger different readings with different degrees of statistical prevalence. The trigger for the shift described in semantic terms as one in ‘telicity’, thus appears to be multi-factorial: it is in part lexical, residing in the lexical semantics of the preposition and the verb, in part semantically compositional (given the additional PP-constituent involved), and in part contextual. Again, these facts do not particularly relate to Case. The PPs in question are standardly taken to differ from DPs in that they do not require Case, and as such there is no additional grammatical dependency between the PP and the verb involved. Folli and Ramchand (2005: 14) moreover discuss complex PPs that are adjuncts rather than arguments, but can still trigger telicity. The same conclusion seems appropriate where particles are involved: The presence of a telic particle gives rise to a telic event, irrespective of Case: (44)

John ate up an apple #for a minute/in a minute.

On the other hand, with the following classes of transitive verbs, it appears that the mereological status of the event as homogeneous/heterogeneous does depend partially on the status of the direct object argument in English: Verbs of consumption/creation (‘incremental theme’ verbs: Dowty, 1991). a. Mary ate apples for an hour/#in an hour.

(45)

b. Mary ate the apple for an hour/in an hour. c. Mary painted pictures for an hour/#in an hour. d. Mary painted the picture for an hour/in an hour.

20

Where to the shops is heterogeneous in path-based terms, whereas towards the shops is not.


229

Again, however, there are readings where in an hour is permitted in (45a): in particular, one in which there were separate events, in each of which an apple was eaten up, always in an hour; and another, more natural one, in which the apple-eating took an hour, while the eating of the oranges took longer (maybe because they had to be peeled). While it is true, then, that the different direct objects preferentially trigger telic or atelic readings of the activity involved, this effect is not strict, and moreover, Case cannot be the relevant factor in explaining when telic readings are preferentially triggered. Nor are telic readings unavailable, where such verbs surface without an overt internal argument: (46)

Mary ate/painted for an hour/in an hour.

Confronted with the PP ‘in an hour’, the interpretive system will simply conclude that Mary ate something or other, her meal say, in an hour. Similarly, with verbs of directed motion, a referential internal argument such as ‘the trolley’ makes reference to a bounded event possible (even if dispreferred) (47b), but the same effect can, at a stretch, be obtained when a tacit internal argument is taken to be present in (47a): (47)

a. Mary pushed for an hour/??in an hour. b. Mary pushed the trolley for an hour/?in an hour.

It is true that the cases with ‘in’ are odd on a telic reading, and the a-example surely more so than the b-example, but Borer (2005: 128) provides other examples which are fully felicitous, at least if the button in (48) is hard to push, suggesting that it is in part encyclopaedic world knowledge that makes these examples preferentially atelic: (48)

Kim pushed the button for ten minutes/in ten minutes.

In short, it appears as if the same grammatical dependency between v and an internal DP-argument, marked in many languages as ACC, can be involved, and yet telicity may be both absent or present, being the overall effect of a number of different factors, as suggested above. Borer (2005: 49), citing Rosen (1999), nonetheless argues that Case determines whether a DP is a measure (rather than thematic considerations). One piece of evidence cited is the fact that arguments contained in Small Clauses which receive accusative Case under ECM can also serve as measures, yielding telic readings: (49)

Terry ran [us ragged] in an hour/?for an hour.

Here the DP us is not thematically dependent on the main verb but does receive ACC under ECM. Borer takes this as evidence that measures are those DPs which receive ACC, i.e. DPs which enter into an Agree-based dependency with v—against the above conclusion that such a dependency leaves telicity open. However, the primary point for her here is that Case in the sense of this dependency is independent of thematic roles. Moreover, Rosen’s claim that (49) is necessarily telic does not seem to

230


be quite correct. Rather, (49) is ambiguous. The conative alternation in (50) has also been taken as support for Borer’s position: (50)

a. John shot the bear #for an hour/in an hour b. John shot at the bear for an hour/#in an hour

The idea is that elimination of the ACC-relation via the introduction of a PP layer serves to make an ambiguous event atelic, again suggesting that Case is crucial to event mereology (cf. also Travis, 2010). However, as suggested above, this could equally well be explained by describing the difference between the two above examples as one of reference: embedding the referential DP ‘the bear’ in a preposition takes away its referential force, as per the account of embedding in Chapter 3. A stronger reason perhaps to believe that ACC directly contributes to event semantics comes from the behaviour of unaccusative verbs. These fall basically into two classes: change of state verbs and verbs of motion. The former class appears to be ambiguous between telic and atelic meanings (though this is context-sensitive), while the latter seem to be inherently telic. In both cases, crucially, the lexical mereology of the internal argument fails to affect the mereology of the event. Consider first change of state verbs: for example, cool, melt, or dissolve. In the absence of any telic adverbial, these are always ambiguous, as Hay, Kennedy, and Levin (1999) note:21 (51)

a. The soup cooled for ten minutes/in ten minutes. b. Some soup cooled in ten minutes/for ten minutes.

Leaving the source of this ambiguity aside for the moment, note that under standard assumptions, the unique argument of an unaccusative verb is a theme and so the facts in (51) are highly unexpected where incremental themes are taken to necessarily bound events. Apparently, it is not thematic structure, after all, which determines whether a given DP can function as the ‘measure’ of an event. Of course it is possible to simply stipulate that these derived subjects are not incremental Themes but rather ‘patients’, but then the diagnostics which show them to pattern with other underlying Themes are somewhat undermined (cf. Burzio, 1986). Moreover, and crucially, where these verbs surface in a causative alternation and assign ACC, they behave like verbs of consumption: 21

Jackendoff (1996) claims, on the contrary, that unaccusative verbs display the same alternations as verbs of consumption, giving the following examples: (i) A pile of dirt accumulated in/*for an hour. (ii) Dirt accumulated for/*in an hour. But we disagree with these judgments: (i) is perfectly consistent with an atelic reading and (ii) with a telic reading. These readings become more felicitous where the PP modifiers are more plausible: (iii) I cleaned my house but dirt accumulated (again) in only a week (iv) ?I sat there watching while a pile of dirt accumulated for an hour.

6. The rationality of Case (52)

231

a. John melted the chocolate for ten minutes/in ten minutes. b. John melted chocolate for ten minutes/?in ten minutes.

In an account which takes measures to be determined on a thematic basis, these data are problematic, and the position that Case determines telicity is to that extent supported: the Themes in (52) involve the dependency marked by ACC, whereas those in (51) do not. Yet again, that the formal ontology of reference is not determined thematically is something we have already argued; and once we accept that a formal event ontology depends on v-DP relations, we can also be open to the further fact that the boundedness of an event can depend on a number of factors other than Case. Indeed, on the approach where ACC uniformly triggers boundedness, the question would remain why telic readings are available in (51) and preferred in (53): (53)

The delegates entered/arrived/left ?for ten minutes/in ten minutes22

As for the change of state verbs, Borer (2005: ch.2) argues, based on cross-linguistic evidence, that these verbs can function either as unergative or unaccusative verbs (in her terms, the same listeme can be inserted in either type of clausal skeleton). We can suggest, then, that where these verbs function as unergatives, they receive an atelic reading in line with what we saw above for unergative ‘activities’ like ‘run’ in (41). There is some support for Borer’s view from Italian where verbs like impallidire ‘to go pale’ and ghiacciare ‘to freeze’ behave like unaccusative verbs in that they always select essere ‘to be’ in the perfect tense (Kayne, 2009). Interestingly, these verbs have a strong tendency to be telic in Italian, by contrast to English.23 But this again leaves open the question why unaccusative verbs give rise to a telic event.24 Overall, this discussion suggests that we need to abandon the view of both Tenny (1992) and Dowty (1991) who take the relevant distinction between measures and non-measures to be purely thematic, but also that we cannot assume Case will settle this semantic matter. It is simply not true that, regardless of its Case, an incremental Theme will measure out an event, whereas other kinds of internal arguments will not.

22 There is a reading which is felicitous here whereby the length of time which the delegates spent inside was ten minutes. This reading is irrelevant. 23 As illustrated by the following Italian examples (Norma Schifano, personal comment):

(i) Il lago si è ghiacciato in poche ore/??per qualche giorno. the lake is frozen in few hours/for some days ‘The lake froze in a few hours/??for some days.’

Italian

(ii) La signora è impallidita in pochi secondi /??per qualche minuto. The woman is paled in few seconds/for some minutes ‘The woman went pale in a few seconds/??for a some minutes.’ 24 In the case of verbs of coming and going, the presence of a null locative might be involved, but the same is not true with change of state verbs.

232


Alternating ditransitive verbs can be used to make a similar point. Consider first the spray/load alternation, which Borer (2005) again takes as evidence that Case is crucial to event mereology. Consider, for example, (53), based on the discussion in Jackendoff (1996: 309): (54)

a. Bill sprayed paint on the wall ?in/for an hour b. Bill sprayed the paint on the wall in/for an hour. c. Bill sprayed the wall with paint in an hour/for an hour. d. Bill sprayed the wall with the paint in an hour/for an hour.

Here, the argument that (the) paint only serves to measure out the event when it surfaces as the direct object and receives ACC, rather than where it surfaces inside a PP, simply doesn’t work. Thus there is a v-DP dependency involved in (54a, b) as well, and the relevant factor is here the referential semantics of the internal argument, not Case. Where the internal argument lacks the definite determiner, as in (54a), reference can only be determined via the descriptive properties of the noun, but we still obtain telic readings such as that it took Bill an hour to spray paint on the wall, while doing the same with chocolate, expectedly, took him much longer. Again, this reading is unusual, but not ungrammatical, suggesting that, in fact, all of the examples in (54) are ambiguous. Another problem is that, as Baker (1997) shows, there is independent evidence from incorporation to suggest that thematic structure actually varies in examples (54a–d). Thus, either internal argument can form a synthetic compound with the verb: (55)

a. wall-spraying b. paint-spraying

Baker argues at length that of the arguments of a verb, only Themes can surface in synthetic compounds, and so the pattern in (55) is support for two distinct base generated structures for spray/load verbs.25 As such it is impossible to say here whether Case or thematic structure is the relevant factor in whatever distinctions we observe in (54). This situation contrasts sharply with the ditransitive alternation, which, Baker argues, is non-thematic in nature and instead Case-related, as witnessed by the fact that only one of the two internal arguments can ever incorporate: (56)

a. I read books to the children b. I read the children books

(57)

25

(prepositional dative) (double object construction)

a. book-reading b. *children-reading

As Borer (2012) notes, however, adverbs can also form synthetic compounds: pan-frying, hill-walking


233

These facts suggest that books is the (incremental) Theme in both (56a) and (56b), with the different surface patterns being due to different Case-assignment patterns. The ditransitive alternation therefore provides a potential testing ground in English for whether it is Case or thematic roles which determine whether DPs function as measures. If Case was the relevant factor, then we might expect that the Theme books should only function as a measure in the prepositional dative (56a). If, conversely, thematic roles are key, then we might expect that the Theme should function as a measure in both constructions. Baker (1997) argues that it is the second prediction that holds, which speaks against the idea that telicity is uniquely associated with Case, rather than multi-factorial, as assumed here. Consider the following examples, where John works in a museum and has to read out information to customers as part of his job. In both patterns, it is the Theme ‘(the) information’ that leads to preferences in regards to event telicity, with the definite-referential Theme again leading to ambiguity, as seen previously: (58)

a. John read information to customers for an hour/?in an hour b. John read customers information for an hour /?in an hour. c. John read the information to customers for an hour /in an hour. d. John read customers the information for an hour /in an hour.

(59)

a. John read information to the customers for an hour/?in an hour b. John read the customers information for an hour /?in an hour. c. John read the information to the customers for an hour /in an hour. d. John read the customers the information for an hour /in an hour.

These examples, then, on the one hand, are more pertinent than the spray/load data because there is independent evidence that this is a Case-based alternation.26 On the other hand, the same observations we made earlier in relation to (43) hold again: we do not consider the PP with ‘in’ to be ungrammatical in (58a, b) and (59a, b), unusual as it may be. Moreover, the crucial factor again appears to be the definite article, not Case. In the ditransitive construction, the Theme receives ACC in both of its variants. The fact that it can only undergo passivisation from the prepositional dative construction for most speakers may follow from locality (cf. Haddican and Holmberg, 2012, on the connection between scrambling and passivization from double object constructions). It is thus quite expected, on the Case-based approach to boundedness, that the Theme will function as a measure in both constructions. Nonetheless, it remains mysterious, for the Case-based approach, why the goal, which also receives ACC in the double object construction, should not become a ‘measure’. 26 As Baker (1997) notes, these facts are problematic for approaches which take the ditransitive alternation to be thematic (Harley, 2002; Ramchand, 2008).

234


In sum, Baker’s data do not support the claim that Case relations provide a way for DP-arguments to measure out events. They also do not refute this position—yet, at a more basic level, as noted, our doubts remain as to what the basic data are, to what domain they really belong, and whether it is the domain of grammar. In particular, even with a Theme to which ACC is assigned, what remains is an ambiguity between telic and non-telic readings; and as noted, we find telic readings triggered compositionally through prepositions in the absence of ACC. The so-called ‘secundative’ alternations provides a better test still, as according to Baker’s diagnostic, it is a Case-based alternation and only one argument receives ACC in each alternation: (60)

a. The Queen presented the medals to the soldiers. b. The Queen presented the soldiers with the medals.

(61)

a. medal-presenting b. *soldier-presenting

Now consider contrasts in regards to judgements about the following: (62)

a. The Queen presented soldiers with medals for an hour/?in an hour. b. The Queen presented the fifth battalion with medals for an hour/in an hour. c. The Queen presented soldiers with her awards for an hour/?in an hour. d. The Queen presented the fifth battalion with her awards for an hour/in an hour.

Such contrasts do seem to exist, and they are sensitive to the status of the ACC (recipient) object and not that of the Theme contained in the PP. Now consider the pattern in the prepositional dative construction: (63)

a. The Queen presented medals to soldiers for an hour/?in an hour b. The Queen presented her awards to soldiers for an hour/in an hour c. The Queen presented medals to the fifth battalion for an hour/in an hour d. The Queen presented her awards to the fifth battalion ?for an hour/in an hour

Here, there is sensitivity to the status of the ACC (theme) object. However, the status of the recipient in the PP also seems to have an effect, making (63c) ambiguous between a telic and an atelic reading. The facts thus remain far from clear-cut, and it is not clear where, in the theory of grammar, the notion of ‘measure’ really belongs. Persistent ambiguities, contextuality, and uncertain judgements are indicative for a phenomenon that, unlike the referential ones of grammatical semantics in Chapter 4, cannot be uniquely traced to grammatical factors.


235

In other cases, a connection between Case and telicity/bounded is broken completely, while nonetheless the Cases track a crucial difference in the formal ontology: whether an event or a state is denoted. Judgements seem more unambiguous in this instance, as we would expect. Thus, stative verbs appear to be inherently ‘homogeneous’ and to remain so regardless of the properties of their internal arguments: (64)

John loved the play/plays/milk for many years/#in many years.

In Tenny’s terms, these internal objects are not ‘measures’ of the event. They nonetheless appear to get ACC, at least in English: the relevant v-DP dependency licensing the argument to the event is fully in place. That dependency, as such, then, is again independent of telicity. Nonetheless, it is interesting to note potentially relatively consistent morphological variation in regards to Case-assignment in this instance: Japanese stative predicates assign NOM to their complements (Travis, 2010: 3, citing Uesaka, 1996: 102; Hirakawa, 1994: 4): (65) Takeshi-wa Keiko-ni-kimochi-ga/*o waka –ru Japanese T-Top K-Gen-feelings-Nom/*Acc understand –Imp ‘Takeshi understands Keiko’s feelings.’ [Fukuda (2006: 102)] Perhaps, then, in statives, a different v is involved, which cannot license objects to events but only to states, which is then reflected in such morphological variation. In fact, in many unrelated languages, stative verbs fail to assign ACC. Consider the case in Irish, from Noonan (1992: 355), also discussed by Travis: (66) Tá eagla roimh an bpúca is fear before the Puca ‘Ailill fears the Puca.’

ag Ailill at Ailill

Irish

In Irish, stative verbs with two arguments are never realized as transitive. Based on data of this kind, Noonan argues that stative verbs are inherently unaccusative. In fact, even in English there is suggestive evidence that ACC is not assigned, as passive formation is somewhat marked (cf. Tenny, 1994): (67)

The tyre is touching the pavement.

(68)

?The pavement is being touched by the tyre.

Where a verb is ambiguous between a stative and an eventive reading, the eventive reading seems to be favoured in the passive (controlling for adjectival passives): (69)

The table was touched by John.

While the case of telicity in events can only inconclusively be traced to Case, then, as discussed, the state-event distinction may well be so traced. The latter, but not likely the former, belongs to grammatical semantics. That said, the variation in

236


Case-assignment even in the case of the state-event alternation illustrates our basic point: that, while the Cases are the morphological reflex of the relationality of grammar, they do not figure in grammar itself, where what figures instead is licensing relations between objects and various formally distinct objects, such as states, events, and propositions. In conclusion, the proposal that ACC tracks event mereology is not only empirically problematic, but is conceptually problematic from the Un-Cartesian perspective as well, where no such thing as an independent event mereology is assumed (and we would also want a principled account of ‘Case’ that covers both NOM and ACC). In our terms, such an account is possible if we abandon the intrusion of morphological concepts into syntax, and develop a principled account of grammar viewed as organized around a number of relations which license referential arguments cyclically relative to different phasal heads, yielding the formal ontology of semantics. In this ontology, as noted in the beginning of this section, part-whole relations are ubiquitous: they lie at the heart of grammatical semantics in our sense. But mereological notions as employed in the event-semantic literature have no independent status, and they have been used to formalize facts that can vary with context and seem to follow from a number of different factors, including lexical semantics, the referentiality of the internal argument, and semantic compositionality, rather than following from the grammar directly.

6.6 Conclusions Something regulates the distribution of overt DPs: their distribution is clearly limited. Referentiality at the level of clauses, verb phrases, and nominals plays a role in this process. The morphological Cases, however, do not. They are language-specific reflexes of grammatical relations that license referential arguments to verbal and clausal heads cyclically, generating a formal ontology of reference alongside. The existence of such relations, as well as their potential morphological reflex in languages that have relevant resources, is the most rational fact we can expect in the organization of grammar: for it amounts to the recognition that grammar, as a relational system, exists. If it does, then the new semantics it gives rise to should be one assigned to the relations it consists in. Relations are interpreted, not features. Case features are marked in morphology on different constituents, where these are both parts of the same grammatical relation. The relations in question, reflected by morphological Case features, are uncontroversially interpretable in terms of arguments doing more than playing the role of Agents, Goals, or Themes: they are also parts of events, which in turn are parts of propositions, which in turn figure in speech events. Nothing in thematic structure, nor anything in compositional semantics mapped from lexical items as such, predicts this deeper move into the realm of grammatical semantics. Several of the relations in question however can be mapped


237

onto the same morphological feature, and the same such relation can be mapped onto different such features. Cross-linguistically, some consistencies arise, but overall they are limited. There is no science of Case then to be expected. But this does not disturb the prospects of a science of grammar. In that science, it is often the case that morphology will provide hints for the relations present. But this should not lead us to posit ‘Case-features in syntax’, interpreted as part of a ‘formal licensing’ mechanism for arguments. Viewed as features in morphology reflecting grammatical relations that are not features, the Cases are interpretable through licensing relations. But these relations are not ‘formal’. Nor are they easily associated with any such predetermined semantic notions as ‘measure’ or ‘telicity’. Instead they lie at the very heart of what we call grammatical semantics.

7 Language and speciation 7.1 Why linguistics is biology Just as most working linguists see themselves as concerned with language rather than thought, discussions of language evolution today largely concern the evolution of language, not the evolution of thought (and much in these discussions concerns the evolution of communication, i.e. the externalization of language as a communicative medium). Discussions of the evolution of language, in turn, are often centred on the question of the existence of a language-specific mental ‘module’ and its possible adaptationist or exaptationist evolutionary explanation. So enshrined is the Cartesian assumption that there simply is no field called ‘the evolution of thought’, just and for the same reason as there is no field called ‘the acquisition of thought’, analogously to the ‘acquisition of language’—as if there was no such issue, and prelinguistic infants always already thought in the ways that beings do who have acquired a language. This latter assumption becomes an axiomatic one when such infants, or indeed nonhuman beings as well, are said to think in a ‘Language of Thought’ (LOT). Assuming such a ‘language’, Fodor and Pylyshyn (1988: 28) could still assert as recently as twenty-five years ago: ‘that infraverbal cognition is pretty generally systematic seems, in short, to be about as secure as any empirical premise in this area can be’. All cognitive organisms, that is, verbal or not, think in much the same way. They are symbol-users (in their LOT) and their ‘mental representations’ have a systematic and compositional semantics. But much of this, of course, becomes partially false or even incoherent, if we are right (Hinzen, 2013b; Burton-Roberts, 2011). More specifically, it becomes problematic the moment we realize three facts in conjunction: The specific symbolic mode of thought that we see manifest in all of the world’s languages and human cultures did not predate the evolution of language, as far as archaeological evidence from hominin material cultures suggests, as well as evidence from comparative cognition, as reviewed below. This mode of thought, too, needs to be explained: it does not come for free. The organization of the lexicon and the grammar directly yield the speciesspecific features of this mode of thought, in particular its propositionality, as argued in Chapters 3 and 4.

7. Language and speciation

239

These three considerations suggest that the evolution of language is inseparable from the evolution of a particular mental phenotype. We will argue that this changes the case for the biology of language, i.e. the biolinguistic programme as such. The connection between universal grammar and species-specificity has always been an important part of the biolinguistic programme from the outset, as is clear in Chomsky (1957; 1959; 1965) and Lenneberg (1967) (and see Bickerton, 1990). But it is only in the last two or so decades that archaeological evidence for this claim has strengthened, showing how tight, actually, the connection between language and speciation is. One could still have assumed, in the 1970s, that language evolved in several continents and in several interbreeding hominin species. Not so today, we will argue, where evidence has accumulated that language is a species-defining feature: a new species, emerging in sub-Saharan Africa, defined by language and endowed with a genuinely different mental phenotype, brought language along with it when migrating over much of the rest of the surface of the earth, after its exodus from Africa. To a significant and surprising extent, then, the evolution of language is the evolution of our species, whose genotype is a linguistic one. Variations in this genotype give rise to different linguistic phenotypes, which are in turn kinds of mind: the mind of an autistic child, a child with Specific Language Impairment, or a schizophrenic adult. A modularist perspective would predict that kinds of mind need not be kinds of language, and that the two can vary independently. By and large, however, we will not find evidence for this in this and the next chapter. The kind of variation in the linguistic phenotype tends to be a mirror of the mental phenotype. The picture is similar in adults with no variation in the linguistic genotype, namely deaf, language-less adults or adults with acquired language disorders. The relative normality of the thinking of such individuals is a reflection of their intact linguistic genotypes. In Section 7.2, we begin by rethinking the biolinguistic programme on this basis, assessing the claim that the topic of UG is the topic of the speciation of modern Homo sapiens and its specific mental phenotype: it is not merely the topic of the evolution of the ‘language module’. Modularity has been the primary target of criticisms of the very idea of a biology of language (Tomasello, 2008; Christiansen and Chater, 2008). But these criticisms, we argue, depend on Cartesian assumptions as much as the modularist perspective itself does, and do not apply to our account. In Section 7.3, we summarize the evidence from the origin of our species that we think supports our stance. Section 7.4 concludes. Before starting we want to note that the account of grammar we have developed so far, of course, is just meant to be that: an account (a philosophy, if you will) of grammar. It is thus made true or wrong by the facts of human grammar, and it competes against other such accounts. Nonetheless, it also needs to be able to re-assert itself against known facts from evolution and biology. It should aspire to innovating these fields as well, speaking to the ways in which we are searching for the biological foundations of language. Our attempt in this chapter is to live up to these expectations.

240


7.2 Language evolution in a new key In classical, twentieth-century terms, ‘Universal Grammar’ (UG) technically describes the theory of the genetic part of a theory of human language: a theory of aspects of language that are genetically based and specific to language, separate from general intelligence. Using a Chomskyan metaphor, the language faculty is viewed as the ‘language organ’: a subsystem distinctive in its function and with sufficient internal integrity to study it in partial abstraction from its interactions with a number of other such (‘mental’ and ‘bodily’) organs, with which it forms ‘interfaces’. This ‘organological’ conception is as such general enough to allow for a large range of more specific views. Only one of them is that of language as a ‘module’ viewed under the prism of Fodorian faculty psychology: as an ‘encapsulated’ system, dedicated and innate, hardwired into the architecture of the mind/brain, separate from both general intelligence/cognition and peripheral organs, operating ‘fast’ and autonomously, by its own computational principles, impenetrable from outside by general cognition or ‘the cognitive mind’. Nonetheless, much discussion on the evolution of language has targeted the evolution of the language faculty in this exact modular sense, with classical language-‘nativists’ (Pinker, 1994; Pinker and Bloom, 1994; Pinker and Jackendoff, 2005) giving it a classical adaptationist explanation and minimalists suggesting an exaptationist one (Uriagereka, 1998; Chomsky, 2005; 2010). ‘Antinativists’ deny the need for any such module altogether. Prominent reasons for this denial are that linguistic communication evolves as part of a biological adaptation for collaborative activities and cultural life more generally, which is non-linguistically specific (Tomasello, 2003; 2008); that no such module can have evolved (Christiansen and Chater, 2008); or no such module exists in any interesting sense given the absence of relevant cross-linguistic universals (Evans and Levinson, 2009). Although the Fodorian conception does not sit well with the Strong Minimalist Thesis, it still accords with the working assumption of many, perhaps most generative linguists, with regards to what they are studying: not thought, but an autonomous structure that can (or must) be studied independently from general ‘cognition’. Moreover, it is studied formally, or as a computational system, without regard to its semantic interpretation or content, which is the content of thought on the modular conception: the content of a sentence is that of the thought expressed in the sentence (Fodor, 2001). To be sure, this stance is historically well motivated, since the principles of syntax as posited since the 1950s seemingly lacked an independent semantic motivation, turning syntax into an ‘arbitrary’ (or unmotivated) aspect of linguistic organization. The ‘principles and parameters’ of linguistic computation as unearthed in the 1980s were—and remain, on standard views—‘functionally arbitrary’ in the same sense: they are not motivated by independent functional considerations or derivable from principles of learning, general cognitive principles,


241

or communicative efficiency.1 Evidence for this conclusion classically motivated the ‘autonomy’ of syntax and that of linguistics as a discipline separate from psychology more generally, and it is the reason why philosophy of language textbooks still deal with ‘content’ almost exclusively, as if ‘form’, i.e. syntax, did not matter to it. We also see the influence of this modular conception where a generative perspective is expanded to other domains of cognition, such as morality, where the ‘moral faculty’ has effectively become a ‘module’ as well, viewed analogously to the ‘language module’ (Mikhail, 2007).2 Our account, while consistent with a broadly ‘organological’ conception, is clearly not consistent with a classical modularist view in either Fodor’s or Pinker’s above sense, and we have noted how it is different from subsequent minimalist theorizing motivated by the Strong Minimalist Thesis (SMT) as well. In Fodor’s own words, the modular view advocates a ‘principled distinction between perception and cognition’ (Fodor, 1985: 3), with language falling on the side of perception. The latter is thought of as continuous with reflex-like (‘dump’ and ‘fast’) responses to the environment. The account assumes, by contrast to the behaviourist conception of operant conditioning, which discarded any internalist determinants of behaviour, that such responses are modularized and innate structure in different domains contributes to the stimulus. The ‘cognitive’ mind, by contrast, is not reflex-like: it is ‘smart’ and ‘slow’, comprising ‘higher’ cognitive processes such as thought or reasoning, which exhibit creativity, domain-generality, and non-encapsulation. So, on this conception, the evolution of the language organ is not and cannot be the evolution of the cognitive mind. Indeed, the mind is said to have its own ‘language’, the LOT (Fodor, 1975), which is distinct from both particular languages and UG. In this regard our argument has been that the term ‘cognitive’ is empty or at least does not make the relevant distinction: the question is what language (and grammar in particular) adds, and the Un-Cartesian hypothesis is that it changes or re-formats cognition, giving us different kinds of thoughts to think. So the evolution of language significantly is the evolution of (a particular kind of cognitive) mind, and Fodorian modularity is not for us: the classical adaptationist approach of Pinker and Bloom (1990) or the ‘parallel’ architecture of Jackendoff (2002) cannot be how we approach the evolution and biology of language here. The benefit of this move is that the most common current criticisms of the UG project simply do not apply to our project. For what precisely unites the three broad criticisms of UG mentioned above is their opposition to the existence of a Fodorian UG-module—a biologically based, linguistically specific, encapsulated, and functionally arbitrary system:3

But see Roberts (2012) for an attempt to ground parameters in extra-linguistic ‘third factors’. For more recent work questioning the modular perspective in line with the more recent minimalist perspective, see Kirkby and Mikhail (2013). 3 Evans and Levinson’s (2009) criticism of a minimalist, underspecified UG as ‘empty’ clearly also doesn’t apply to our proposal. 1 2

242


there is no Faculty of Language in the Fodorean sense of Faculty Psychology (no module or modules wholly specific to language processing), and UG is an unfortunate misnomer, because there is nothing essentially grammatical about the capacities an infant uses to acquire language’ (Levinson and Evans, 2010: 2742); It is not that the evolution of some kind of innate syntactic template such as universal grammar is impossible, it is just that currently there is no evidence for it empirically, no precise formulation of it theoretically, and no need for it at all ( . . . ). Our conclusion is thus that although many aspects of human linguistic competence have indeed evolved biologically, specific grammatical principles and constructions have not (Tomasello, 2008: 313); It is implausible that such an [language-specific biological] endowment could evolve through adaptation ( . . . ). Instead, we have suggested that some apparently arbitrary aspects of language structure may arise from the interaction of a range of factors, from general constraints on learning, to impacts of semantic and pragmatic factors, and concomitant processes of grammaticalization and other aspects of language change. But, intriguingly, it [is] also possible that many apparently arbitrary aspects of language can be explained by relatively natural cognitive constraints—and hence, that language may be rather less arbitrary than at first supposed (Christiansen and Chater, 2008: 507).

There is, in short, on all three views, no biology specific to language—no UG rooted in the human genome, no biolinguistics in the original sense of Chomsky (1957) and Lenneberg (1967). But is it therefore true that language simply doesn’t matter to biology? That there is no such thing as biolinguistics altogether? Central in our argument for the opposite conclusion will be precisely that language is not a Fodorian module, and that, instead, the evolution of language is the evolution of a mode of thought defining and demarcating a new species in nature and brain (Dubois et al., 2010; Clowry et al., 2010; Vigneau et al., 2011; Miyata et al., 2012). It is therefore not something that evolved after a new mode of thought was already in place in one or in several species, in the way that cultural innovations arise in historical time, in the absence of a biological change. If so, there is biolinguistics even if there is no language module, and we expect to find a Cartesian rather than Un-Cartesian philosophy of grammar in those who deny the existence of biolinguistics. And indeed, the Cartesian assumption is very clear in Christiansen and Chater’s suggestion that core properties of language are explained and predicted by the properties of thought, assumed to be independently given. They endorse the: standard assumption is that thought is largely prior to, and independent of, linguistic communication. Accordingly, fundamental properties of language, such as compositionality, functionargument structure, quantification, aspect, and modality, may arise from the structure of the thoughts language is required to express. (Christiansen and Chater, 2008: 501)4 4 Cf. also: ‘The cultural evolution of language does not, of course, take place in a biological or social vacuum, but rather, is shaped by multiple constraints. One type of constraint derives from the nature of the thoughts that language expresses. For example, whatever the nature of our mental representations, they apparently afford an infinite range of different thoughts, promoting the likely emergence of compositionality in language’ (Christiansen and Chater, 2008: 452). But a discretely finite system requires a generative


243

That thought would exhibit these properties independently of language, and hence language would exhibit them only derivatively (because it expresses thought), is of course exactly the assumption we call Cartesian.5 It is particularly implausible in this case, since the aspects of language in question are uncontroversially linguistic ones that have specific grammatical correlates and cannot, we contend, be defined without reference to these. One might formalize them as semantic or logical ones in some available formal notation—but this won’t change the fact that they have their basis in the structure of human language, where they are empirically manifest (Crain and Khlentzos, 2010). In Tomasello’s case, the Cartesian assumption is enshrined in the dialectic of his (2008) narrative for how linguistic communication evolved. The driving force behind language evolution on this account is the evolution of a new kind of cooperative form of shared intentionality that evolved in hominins but no other primates and engenders a novel form of communicative intentions. As these intentions appear, novel grammatical devices became necessary to meet the demands for communicating them. Eventually the devices conventionalize as the arbitrary grammaticality patterns of a particular language, eventually becoming enshrined as the normative commitments that a person must shoulder in order to count as a speaker of this language. While, initially, hominins might only have wanted to communicate desires, in particular, leading to what Tomasello terms a ‘grammar of requests’ also found in the ‘language-trained’ apes, eventually they would want to communicate to share information, leading to a more complex ‘grammar of informing’, exhibiting a ‘serious syntax’. Finally, the desire would evolve to share whole narratives involving sequences of connected events with shared participants, requiring what Tomasello (2008: 294) calls a ‘grammar of sharing’ that comes with a ‘fancy syntax’ (namely the actual syntax of all human languages). ‘The main point’, he summarizes: is simply that communicative function drives the process, such that all linguistic communities who wish to tell narratives and engage in other forms of extended discourse must create grammatical conventions of the general type described here for getting these things done. (Tomasello, 2008: 286)

The Cartesian implication is that the more complex intentions with their distinctive contents evolved first, independently of language and grammar, which are merely arbitrary conventionalizations of them. There simply arose new kinds of thoughts that wanted to be expressed; and hence the grammar had to be invented that would get them across. Language evolution, therefore, boils down to solving a problem of

system, which is ipso facto given on standard accounts of grammar, and obscure if independent from grammar. 5 More generally, constructionist approaches to language acquisition seem to crucially require a rich system of thought which can then be linked to arbitrary linguistic templates by the child.

244


externalization (or communication). Grammar has no other epistemological significance and is not involved in the aetiology of the thoughts themselves—the Cartesian position. Again, however, to evolve language in the human sense, we have to evolve the mind that is expressed in it, and that mind doesn’t come for free and it evolves over evolutionary time. Intentions, by contrast, do come for free, in the context of our inquiry: for they, and their recognition by others, are prelinguistically given. Intentionality is therefore not the mystery we have to solve here. The mystery we have to solve is why there are intentions of different, partially species-specific kinds. In this regard Tomasello agrees, and the biological change that he postulates in the genus Homo is one that leads to intentions that are of a uniquely social and cooperative kind. Intentions, however, also have a content. And here the problem arises: cooperativity and communicative intentions on their own make insufficient predictions for what is being communicated, and hence the kind of mind that we are here interested in: a mind that can have contents that we recognize can be true or false, that we can believe even when they are false, or disbelieve when they are true; a mind that engages in different forms of nominal and clausal reference; and that can grasp algebraic rules holding between values of variables (Marcus, 2001). No social and cooperative mind needs to be productive or systematic;6 or construct theories and reasons from unobservable causes.7 In making the communicative needs and desires of a mind capable of generating contents of this nature basic to our account of grammar, we are presupposing the basic structure of a linguistic mind already. Even if we posit special, non-linguistic mind-reading (‘theory of mind’) abilities in it, the problem remains. A mind can only, in the relevant theoretical sense, ‘read’ the contents of the mental states of others if it knows what a belief is, and if it can generate the same propositions that the other mind thinks or takes an attitude towards. This requires a grammar of some sort, assuming that propositions are different from lexical atoms, i.e. structured and compositional; while it is logically conceivable that this grammar is fundamentally different from that of natural language, it is not clear why we should make this assumption if we know that mind-reading matures with the maturation of specific aspects of recursive syntax (de Villiers, 2007). Tomasello (2008: 321) calls ‘recursive mindreading’ a ‘skill’, which is key to the new social-cognitive capacities in question. Yet, we do not see such a 6 Indeed, according to Tomasello’s account of language acquisition, even early linguistic children, while being fundamentally cooperative, do not show signs of such productivity (though see Kowalski and Yang, 2012, for a critique of the evidence). 7 As Penn et al. (2008: 121) note, ‘it is very hard to see how a discontinuity in social-cognitive abilities alone could explain the profound differences between human and nonhuman animals’ abilities to reason about causal relations in the physical world or nonhuman animals’ inability to reason about higher-order spatial relations’, a point echoed by Carpenter and Tomasello (2000) himself, when he acknowledges that human language learners manifest cognitive capacities—such as analogical reasoning and abstract rule learning—that are both independent from mind-reading abilities and absent in non-human animals.


245

recursive mind-reading capacity in non-grammatical creatures (Fitch, 2010), or indeed on Tomasello’s own account of the fundamental discontinuity between human and non-human minds. Moreover, such mind-reading has inherent structural properties—like the occurrence of an (unendorsed or unasserted) proposition as an argument within an asserted one, as in [He believes [I am here]]—which seems conceptually non-distinct from the hierarchical configurations that standard syntax posits in the linguistic expressions encoding them. In short, whatever prelinguistic mind-reading capacities we posit, we would need to be careful not to implicitly characterize them as already exhibiting what we can only describe as a form of grammatical organization. If the mind-reading is based on perception more than conceptualization, no genuine concept of thought nor propositional thoughts about thoughts need to be as involved at all, given distinctions between perception and concepts as involved in propositions that we stressed in Chapter 2. Where explicit reasoning about thought takes place and thoughts about thoughts are generated, it is hard to imagine them not to exhibit a part-whole organization that is recognizably and inherently grammatical, with each thought corresponding to one constituent and one the grammatical argument of the other, and only the matrix constituent evaluated for truth. (Declarative) pointing, too, like recursive mind-reading, is only found in linguistic creatures. It is distinct from imperative pointing, which is there to request or demand things, and evolves later, by the first birthday, cross-culturally (Butterworth, 2003). It is not found in autistic children (Tomasello, 2008: 142–3, citing Carpenter, Pennington, and Rogers, 2001; Mundy and Siegman, 2006; Liebal et al., 2008). One kind of declarative pointing can be, in Tomasello’s technical term, ‘expressive’—used to share an attitude about a common referent—as when the infant points to an animal in the distance, say an elephant, expresses emotions, and alternately gazes at the adult. The infant, we might gloss this, wants to share its sheer excitement at the fact that there is an elephant there. Plainly, the elephant is not requested, and there is nothing that the infant wants for himself: rather, the elephant is viewed as an aspect of the external world—a common referent that is shared and objective. That referent is individuated as falling under a particular concept, which comes from an equally shared conceptual space of which the infant and adult are both parts, who will thus individuate their experience and react to stimuli in similar ways. This crucially triadic structure is nothing other than the ‘triangulation’ that Davidson (2004) argues is the necessary infrastructure of all language, rationality, and thought. Declarative pointing, that is, already manifests a propositional competence in the present sense—that is, a grasp of reference and predication, as based on a shared concept, which yields an objective state of affairs if applying to an object referred to, and can be accompanied by an attitude towards the fact that the state of affairs obtains (the attitude presupposes the proposition/state of affairs in question, as what it is directed at). This appears even clearer in the case of

246


what Tomasello terms ‘informative’ declaratives, which are enacted by the infant to provide the adult with information about some referent that the adult currently lacks but needs or wants (for himself, not necessarily for the infant). Tomasello takes this as evidence for a ‘prelinguistic’ infrastructure of shared intentionality on which linguistic abilities can then be grafted—no further biological adaptation for language is needed. For this to be an argument against an innate language-specific faculty, further evidence is required that what shows itself in infant gestures such as pointing and pantomime is not effectively the language faculty at work. Infants are linguistic creatures, whose brain exhibits language-specific asymmetries from its earliest developmental stages (Glasel et al., 2011; Dubois et al., 2010), and whose language faculty starts developing in utero. It would be no big surprise, then, if, even before grammar is morphophonologically expressed, the behaviour we find is that expected of a creature that appropriates the world in grammatical terms and has the kind of brain needed to do so. Strong evidence in favour of our opponent here would be that we find the gestures in question in non-linguistic beings. But that is exactly not the case, on Tomasello’s own account. Even the ‘linguistic’ gestures of human-reared and language-trained apes, on his account, do not approximate infant declarative pointing and are not propositional in nature (see also Terrace, 2005). We conclude that infant pointing may be most appropriately understood as the first manifestation of what we call the grammatical mind, rather than as a prelinguistic adaptation on which language can be built. That acts of infant pointing communicate pretty much what adult pointing also does, is predicted on this view: an infant at age 11.5 months, for example, just like an adult, might point to his glass at the dinner table, after someone else had poured water into his glass a few minutes earlier (Tomasello, 2008: 114). The infant points to a fact, that the glass is empty, informing the adult of this fact, in the context of another fact, that the infant desires to drink more. A creature that needs a ‘fancy’ grammar to tell the narratives it invents, equally is, even more so than a creature that points declaratively or a creature that reasons about other minds recursively, already a linguistic creature: its thought is (implicitly assumed to be) grammaticalized already. That we have a creature of this kind therefore does not explain grammar. At least partially, such a creature is explained by grammar—an observation that does not in any way diminish the crucial importance of the evolution of cooperation and shared intentionality for the evolution of communication or the externalization of language. Evans and Levinson, finally, concur with Tomasello’s focus on the ‘intentionattributing cognitive infrastructure that provides a universal base for language evolution’ (Evans and Levinson, 2009: 472), yet further construct an argument against UG as a module from the pervasiveness of cross-linguistic variation at every level of linguistic organization. Stating variation, however, does not absolve us from the task of stating what grammar is or does, universal in human populations as it is. Indeed,


247

any statement about variation in a universal trait is only meaningful against an account of what does not vary in this trait. What is commonly taken not to vary across languages is of course thought, or semantics, a position widely maintained in linguistic typology: Haspelmath (2010), for example, contends that concepts used for cross-linguistic comparisons must be universally applicable—and ‘conceptualsemantic concepts’ are one kind of concept that fits this bill. Semantics, in short, is one of the anchor points on which cross-linguistic comparisons must be based. Evans and Levinson differ from this consensus in stressing that cross-linguistic variation does in fact affect thought or semantics. They explicitly reject the idea that ‘languages directly encode the categories we think in, and moreover that these constitute an innate, universal “language of thought” or “mentalese” ’ (Evans and Levinson, 2009: 435). If and to the extent that languages are not merely the local ‘cloths’ for universal thoughts expressed in them, language will have an epistemological significance and the position appears at first to reject a Cartesian assumption. However, a look at the list of the Neo-Whorfian semantic variation that the authors list shows that none of this variation is variation in what we have here analysed as grammatical semantics or the formal ontology of language. The list is (Evans and Levinson, 2009: 435–6): languages may lack words for logical connectives such as ‘if ’ or ‘or’; specific colour words such as ‘blue’ or ‘green’ (see also Franklin et al., 2008); words for specific body parts such as ‘leg’; words to mark Tense, Aspect, or numbers; plural markers meaning roughly ‘of unexpected number’, said to be found in Kiowa; words or affixes indicating the sources of evidence as available in Korean or Central Pomo; classifiers specifying an object for whether it is liquid, ropelike, containers, etc., as in Athabaskan languages; words coding intricate kinship relations, as in Australian languages; or expressions for ‘left of ’, ‘right of ’, ‘back of ’, ‘front of ’; or languages where a term like ‘banana’, in the absence of a classifier, will only denote a mass or substance. But clearly, all of this is fundamentally lexical variation: it so happens that different languages, at the level of their compendia of words, choose to encode different concepts. None of this affects what speakers of any of these languages will find true or false; nor will it deprive them of any sense of what truth or falsehood are; that truth is distinct from belief; that objects are different from properties; or that possibilities are different from facts—though cultural variation in values and environmental conditions will make them interested in partially different facts and these will be expressed in different ways, depending on lexical resources available (though such resources can always be changed). It transpires from this discussion that while Evans and Levinson reject the Fodorian and Pinkerian commitment to a language-independent LOT and its universal semantics, this is chiefly because the lexical resources of different languages simply do not map onto the same concepts, and because these concepts are assumed to enter the LOT. There is, in this sense, linguistic relativity. But that very notion—in that a non-trivial relation can only relate two different things—encodes a Cartesian

248


assumption: that thought and language are distinct, though the latter can influence the former. The crucial task, which is a theory of the format of sapiens-specific thought itself, is not addressed (beyond the stress on social-cognitive infrastructure discussed above); and no distinction is made between effects on the organization of meaning arising at a lexical and at a grammatical level. As we have seen above and in previous chapters, a Cartesian stance is also essentially adopted in the major traditional alternative to the adaptationist view of language evolution, namely the early and middle Minimalist paradigm, in which a core of language—narrow syntax, or the ‘computational system’ of language—rather than primarily reflecting adaptive constraints, is said to ‘crystallize’ in evolution, snow-flake like, reflecting a minimal genetic change effectively reducing to the evolution of Merge within the given cognitive infrastructure of the hominin brain, together with generic and physicochemical factors (Chomsky, 2005; 2010; Hinzen, 2006; Berwick and Chomsky, 2011). We thus conclude that UG in the generative tradition and today’s major anti-UG paradigms share Cartesian assumptions. The expectation should thus be that the dialectic in regards to the biology of language changes with an Un-Cartesian view. It is not a ‘modular’ one in the sense above, hence escapes from the primary target of traditional arguments leveled against UG: this book defends nothing like the functionally arbitrary ‘language faculty’ of classical generative linguistics and Fodorian faculty psychology. Much of our foregoing discussion has been about the wonders of grammar and what it does to cognition, reformatting our mind and what it means to know something. An ‘interface’ with an independent system of thought or a ‘cognitive’ module is one that we are committed to reject. The ‘language organ’, if this is still the right metaphor, is thus also not here one mental ‘organ’ among others that could be separately lesioned while leaving others intact (as classical arguments for modularity entail). The transformation that we suggest the grammaticalization of the mind represents, moreover, does not seem to be based on some novel biological adaptation of a non-linguistic nature, if we are right that core aspects of grammar simply do not have non-grammatical analyses or explanations. But despite this absence of non-grammatical explanations of the facts about cognition that we are here interested in, grammar in our sense is nonetheless not an ‘arbitrary’ system: most, and perhaps all of grammar, if we are right, is rationalizable in terms of how it mediates reference, when pointing gives out. What we are talking about, then, when we talk about UG or the evolution of language is, effectively, the evolution of our mind and the thinking that takes place in it, insofar as that thinking is unique to us. There is such a thing as biolinguistics because (i) there is no non-linguistic biological adaptation that can substitute for language, and (ii) the grammatical organization of language reflects a new mode of thought defining a new species, because it defines its mind. This is partially consistent with the original arguments for UG, and in particular the domain-specificity of the biology of language: it is still the case that general intelligence—in the sense of smart


249

adaptive behaviour involving inference or ‘reasoning’ (‘thought’ in a generic sense)— is not our topic here. Many non-human animals are fabulously smart, not to mention hominins other than Homo sapiens. And as Descartes (1637) was the first to note, human beings that have normal language can be extremely stupid, if not outright incapable of finding their way around life. Whatever explains smartness, in short, or adaptivity, we can ignore it here.8 Nor does it appear as if social-cognitive capacities can entirely substitute for linguistic ones, as discussed above. Moreover, this difference in thought is not a difference within one species, nor is it scattered across species. Instead, it defines one single species. This connection between language and speciation, in itself, is a pointer to a connection between language and genetics, and a much stronger one than was available in the 1950s, when much of the evidence for the speciation of modern Homo sapiens was not yet in, as we will discuss below. It is thus expected that those who deny any genetic endowment specific to language will tend to downplay the specificity of modern Homo sapiens as well, and locate the origin of language outside of the context of our speciation, some hundreds of thousands of years earlier in hominin evolution (see for example Levinson and Evans, 2010: 2742, discussed in the next section). Evidence against this early evolution is therefore evidence in favour of a biological change for language. Thought content defining the new species is found expressed in the world’s languages, in cultural practices such as art and science, and in our, again universal, ‘moral nature’. A parsimonious explanation for this evolutionary discontinuity, which draws a line around one species, would want to relate this human invariance—(rational) thought—with the major other human-specific invariance—grammar—especially if we are right in our take on grammar, for then the latter can be turned into the generative mechanism behind the former. Our reinforcing argument in this regard has been that grammar, looked at in a new way, exactly provides the relevant forms of rational organization, allowing the grasp of a unique kind of grammatical meaning that is unavailable lexically or non-linguistically. Since the grammaticalization of hominin thought further depends on a genetically based change in cortical organization that took place in the genus Homo and reflects the organization of language, and the de-grammaticalization of the sapiens brain arguably correlates with the breakdown of the relevant mode of thought as well as these forms of cortical organization (Sections 8.4 and 8.5), a plausible case for a biologically specific endowment for language in humans not only remains, but is strengthened—without any ‘autonomous’ language-module being invoked.

8

Indeed, whereas art, culture, science, and morality are presumably all adaptive traits, much discussion on ‘progress’ in biology suggests that there is no biological meaning to the assertion that such capacities are, in some absolute sense, ‘more adaptive’ than those of a Neanderthal or chimpanzee or mouse mind.

250


7.3 The speciation of modern Homo sapiens The heuristic rejection of the need for special explanatory principles for the case of language as well as our species as such was an important ingredient in the behaviourist research programme that predated modern linguistics. As John B. Watson put it in 1913: ‘The behaviorist, in his efforts to get a unitary scheme of animal response, recognizes no dividing line between man and brute.’ Skinner (1953), too, suggested, quite reasonably: ‘It would be rash to assert at this point that there is no essential difference between human behaviour and the behaviour of lower species; but until an attempt has been made to deal with both in the same terms, it would be equally rash to assert that there is’ (Skinner, 1953: 38). Skinner (1957) precisely is an attempt to integrate language in such a unitary scheme. Like empiricists before and after him, he regarded human linguistic behaviour as conditioned by the contingencies of experience and the forces of an external environment acting on an organism—like everything else. No determinants internal to the organism are called for. One could put Chomsky’s central (1959) claim against Skinner (1957) as saying that this externalist commitment is absurdly radical. It is natural to assume that, apart from externalist determinants, internalist ones are called for in the case of language. Put like this, the case against Skinner is consistent with the position that Chomsky (2000a) and Hinzen (2006; 2007) describe as a form of ‘internalism’, as well as the current broad ‘three factors’ framework of Chomsky (2005), which maintains that language design is shaped by (i) language-specific innate endowment (= UG), (ii) experience, and (iii) third factor conditions not specific to language or humans or indeed organisms. The claim that (i) is non-empty is equivalent to the claim that, insofar as Skinner’s alternative scheme of explanation is a failure, special explanatory problems are posed in the case of language, pointing to organism-internal principles and an apparently species-specific and human-unique adaptation. This position deserves to be qualified as the null hypothesis (for internal determinants cannot obviously be excluded). The criticisms of UG mentioned earlier, which again maintain the absence of (i) and are in this sense continuous with Skinner’s programme, should thus be regarded as quite radical. They can only be maintained if linguistic behaviour is continuous, after all, with non-linguistic behaviour, and explainable on similar grounds. For no genetic change specific to the evolution of language and this species, is said to be required: nothing is special to language and nothing, therefore, calls for special explanatory principles, exactly as was assumed until the mid-twentieth century. Despite an emphasis on internal factors, discontinuity, and species-specificity, the connection between language and species is not as explicit in Chomsky (1959; 1965) as we will argue now it should be today. Moreover, species-specificity could only have meant human specificity at that point, as evidence on speciation patterns within the genus


251

Homo was simply not yet in at this point. Fifty years later, generative grammar has come to systematically incorporate considerations of language evolution, but the situation is not very different in regards to the connection between language and species. UG is certainly not proposed as a theory of speciation. Emphasis since the 1990s to ‘minimize’ the contribution of the first factor may even not be conducive to systematically addressing this connection. The proposal, in particular, that the evolution of language, in its grammatical aspects, will ‘reduce to the emergence of the operation Merge’ (Berwick and Chomsky, 2011), leaves us with no clear clues on the nature of language or what makes it specific, given the genericity of Merge/recursion. The stress on domain-generality, minimality, and genericity, and on the prelinguistic richness of the so-called ‘Conceptual-Intentional’ systems, thus comes with the danger of leaving us without a grasp of what the effects of grammar on cognition actually are. Stress on minimal analysis is uncontentious, of course, wherever we start from some descriptively rich theory of some domain that we seek to make explanatorily stronger. But the position we end up with may also yield an account of grammar so minimal that we fail to recognize the phenomenon in it with which we started. A formal abstraction may be created that doesn’t connect to the natural phenomenon any more that we started from. Picasso’s depiction of the abstract form of a bull still captures its essence. Depicting the bull as an extension in space would not. Does Merge still capture the essence of grammar? Can it explain a reformatting of mind or the emergence of a new mode of thought and a species? Would a ‘minimalist theory of gravitation’, say, even make sense? However serious we take this concern, it has remained the case, with the arrival of Minimalism, that UG is still not part of a general theory of speciation. As noted, in the 1950s, we should have expected this. This was the high time of the Modern Synthesis, which, in exactly the same way as Skinner in his study of animal behaviour, was near-exclusively focused on external factors in the evolutionary origins of organismic form. Although Darwin himself stressed the existence of ‘two laws’ of evolution—natural selection and the ‘laws of form’ (evolution by natural law)—the focus of the Modern Synthesis on natural selection as the main force of evolutionary change is undeniable (Gould, 2002; Hinzen, 2006: ch. 2). Like empiricism, it remains externalist in its core commitments: evolutionary change is conditioned by random genetic variation within populations and the changes in gene frequencies caused by environmental forces acting on the variants and affecting their differential reproduction. No independent internal determinants of evolutionary change coming from the constraints imposed by developmental systems are called for. This is precisely the assumption that the ‘Evo-Devo’ research programme set out to address and discard, taking: the contributions of the generative processes into account, as entrenched properties of the organism promote particular forms of change rather than others. On this view, natural

252


selection becomes a constantly operating background condition, but the specificity of its phenotypic outcome is provided by the developmental systems it operates on. Hence the organisms themselves represent the determinants of selectable variation and innovation. At the theoretical level, this shifts a significant portion of the explanatory weight from the external conditions of selection to the internal generative properties of evolving phenotypes. (Pigliucci and Müller, 2010: 13–14; our emphasis)

Hinzen (2006) proposes to view the theory of UG as an instance of ‘internalist’ or ‘generative’ biology in this sense, applied to language: an account of the generative principles that give rise to new behaviours in humans. Yet Hinzen also leaves the connection with speciation largely out of account, and in a way, Evo-Devo does not make it a central topic either: it is continuous, in particular, with the stress of the ‘rational morphologists’ of the nineteenth century (Gould, 2002) on what they called the ‘unity of type’—the existence of abstract structural templates (Baupläne) that cut across different species operating under different adaptive regimes. In this sense, the reality, or at least the centrality, of species is denied here as much as it is in Neo-Darwinism. In addition to externalism, the Modern Synthesis was committed to gradualism in evolutionary change, whereby most adaptive genetic variants have slight phenotypic effects and diversification comes about through speciation viewed as the gradual evolution of reproductive isolation among populations (Futuyma, 1986: 12). Darwin himself had been gradualist on the topic of speciation, and much of the beauty and simplicity of his theory rested on the absence of a special mechanism of speciation introducing discontinuities into the smooth flow of natural selection. In the inexorable action of natural selection in modifying gene pools, species lost their individuality as they became no more than arbitrarily defined segments of steadily evolving lineages. (Tattersall, 2008: 7)

Speciation thus wasn’t viewed as a discrete genetic event, and the idea of ‘saltation’ became a priori suspect, sidelining early champions of such evolutionary change such as De Vries (1901), Bateson (1894), and Goldschmidt (1940). By the 1970s, on the other hand, it was clear that this a priori commitment simply did not fit the empirically given pattern of evolutionary change in the fossil record, which is rather one of ‘punctuated equilibria’ (Eldredge and Gould, 1972), with long periods of stasis alternating with sudden bursts of change in which new species arise. Rather than slowly morphing into other species, that is, with a range of intermediate forms, species evolve quite suddenly, in events that Eldredge and Gould speculated might last no more than 5,000 to 50,000 years. They then often persist relatively unchanged over long periods of time, until they disappear again and are replaced by other species (J. H. Schwartz, 1999). The genus Homo, in particular, is not a story of gradual fine-tuning in which a major central lineage is increasingly acquiring a human shape over the eons. It is a history of experimentation with multiple hominin species


253

originating and partially co-inhabiting the same environments, doing battle in them and going extinct more often than not. If anything, then, species are for real; they are individuals that exist as such on the biological scene, being ‘fully individuated entities, playing evolutionary roles that go beyond simply being intermediates between their ancestors and descendants’ (Tattersall, 2008: 17). What individuates a species, therefore, is a question of genuine theoretical interest, and very much an open one. Reading even current major textbooks and anthologies in the philosophy of biology (Sterelny and Griffiths, 1999; Hull and Ruse, 1998) yields a false impression in this regard, as they virtually start from the demise of ‘human nature’, given that humans are a species, and species are not real given a Neo-Darwinian perspective, or only as real as the ‘isolationist’ or ‘biological’ species concept of Mayr (1942) allows. Against this consensus stands the fact, which we discuss in more detail below, that ‘humans’ are many species, that ours is only one of these, which evolved in a discontinuous fashion, and around which language draws a line, suggesting that the evolution of language and our species may not only have been evolutionary ‘events’, but the same event. In this light, the persistent absence of essentially any systematic discussion of language in the philosophy of biology is noteworthy: as of now, language is here simply not regarded as part of the solution to the riddle of the speciation of modern Homo sapiens and the genesis of our mind. The focus of discussions has been, and remains, on the individual organism, if not its genes, as the fundamental entities of evolution. But language is a population-level phenomenon; though processed in individual brains, it does not exist on an individual level (cf. Stainton, 2011). Factually speaking, there are no private languages. Moreover, language defines a species, if current thinking on palaeoanthropology is on the right track. This species is regarded as a single one today: there are no different species of humans alive. This is a non-trivial fact, which asks for an explanation: something holds this species together, as there appears to be no relevant genetic variation with regards to this species’ basic cognitive traits. As noted it is also, and crucially, now regarded to be a distinct species from any now extinct precursor hominin, an insight that did not exist in the late 1950s and 1960s, where Neanderthals were still widely regarded as simply an early variant of modern humans. These were also the formational years for the major post-behaviorist philosophical framework in thinking about the mind, namely functionalism and the ‘computational-representational theory of mind’ (Fodor, 1975; Lycan, 2003; Rey, 2003), where ‘multiple realizability’—and hence the irrelevance of species boundaries—became a defining property of mentality as such. Chomsky himself, to our knowledge, never endorsed functionalism in these formational years of ‘cognitive science’, and indeed rejects it in Chomsky (2003) as a complete misunderstanding of the computer analogy and Turing (1950) in particular. Generative linguistics was nonetheless widely perceived as consistent with it, given its focus on the abstract study of a ‘computational system’ irrespective of its

254


use as well as its organic basis. The embodiment of linguistic cognition eventually becomes a crucial aspect of the Minimalist research programme, in which the study of the ‘interface’ of language with sensory-motor systems moves to centre stage, and considerations of externalization of language in a particular sensory-motor modality have sometimes been taken to carry crucial insights in regards to central design features of language (see e.g. Carstairs-McCarthy, 1999). Nonetheless, in practice, linguistic theory has remained focused on the computational study of language, and the more recent focus on ‘interfaces’ has in fact only led to the formulation of an even higher abstraction: ‘narrow syntax’ as reduced to Merge, which retains the nature of an essentially disembodied and multiply realizable process. The more ‘third factor’ conditions become relevant, which relate to abstract organizational principles of a physico-chemical sort, the less the organism, and the species, will figure in linguistic explanations. Let us then re-raise the question of why language and species, as a matter of empirical and palaeoanthropological fact, do intrinsically relate. As Stringer (2011) recalls, in the 1970s: the origin of modern humans was hardly recognized as a specific topic worthy of scientific study. The standard classification of humans had living people, the Neanderthals, and diverse remains from sites like Broken Hill in Africa and Solo in Java all classified as members of our species. With such different-looking fossil members within Homo sapiens, the origins of features like a chin, a small brow ridge and a globular skull were, not surprisingly, lost in a morass of diversity. Moreover, with the predominance of multiregional or Neanderthal-phase models, the origins of those features were apparently scattered among many different ancestors living right across the Old World, so modern human evolution was not so much an event as a tendency ( . . . ). For human behaviour too, there was an emphasis on gradual evolutionary trends ( . . . ). It looks very different forty years on—for most scientists, Africa has been established as the centre for both our physical and our cultural origins. The evolution of ‘modern’ Homo sapiens can be viewed as an important physical and biological event, backed up by fossil and genetic evidence. (Stringer, 2011: 201)

Events need causes. They are the results of these causes. In the case of events that are speciations, the result involves the eventual establishment of a reproductive barrier with other groups of organisms, which establishes the species as a biological individual in its own right. Such a barrier can come about through various causes, and the development of new anatomical features per se, which allow us to recognize a subgroup of a given species, is not the same as a speciation event. This is relevant to the fact that the crucial kind of behavioural innovations that we associate with our species—such as symbolism, personal adornment, elaborate burials and weaponry, and art—are manifest in the archaeological record only some tens of thousands of years later than anatomically modern humans appear on the African scene, as marked by a distinctively shaped brow, chin, and thorax. Fossils with essentially


255

these distinctively modern human features—despite some remaining residual differences—are found from about 195,000 years ago, which is the dating of the Omo skull from Ethiopia.9 Tattersall (2008: 94–5) lists four further fossils: a skull found at the site of Herto, also in Ethiopia, dated at 160,000 years ago; some fragmentary 115,000-year-old remains from the Klasies River in South Africa; a partial cranium from Singa, Sudan, dated at 100,000 years ago; and an apparently very clear example from the burial site of Jebel Qafzeh in the Levant, dated at 92,000 years ago. In none of these cases is anatomical modernity accompanied by behavioural modernity, and the archaeological record associated with any of these types of humans at their respective sites shows no signs of symbolic behaviour of the sort that we can associate with language. The record of the humans from Jebel Qafzeh, in particular, is said to be ‘indistinguishable’ (Tattersall, 2008: 96,103) from that of the Neanderthals coinhabiting the region at the time. The fossil evidence for an African origin of modern humans is corroborated by evidence from mitochondrial DNA passed down through females, which is traceable to a single variant in Africa that belonged to a population emerging between 290,000 and 140,000 years ago; and Y-chromosome DNA passed down through the male line, which again can be traced to Africa. Emigrants of this African population arrived in Europe between 39,000 and 51,000 years ago, where the ‘human revolution’ would play itself out more fully than it did in Africa, given the absence of ‘any strong evidence for figurative or clearly representational art anywhere before the European material dated at about 40,000 years’ (Stringer, 2011: 209). Evidence for symbolism is present in Africa much earlier, as in flat pieces of ochre engraved with geometric designs and pierced shells used as beads from Blombos cave in South Africa near Klasies, dated at around 75,000 years ago (Henshilwood, 2006), and similar shells in North Africa that are 82,000 years old (Taforalt, Morocco: Bouzouggar et al., 2007), and in the Near East from about 90,000 years ago (Qafzeh, d’Errico et al., 2009: 34). However, iconographic representations are not actually found in Africa before 26,000—28,000 years ago (Apollo 11 site, Namibia: McBrearty and Brooks, 2000). Even personal ornaments disappear from the African scene again after their appearance around 80,000 years ago, with no reliably dated findings dated at between 70,000 and 40,000 years ago (d’Errico et al., 2009: 34–5). At around the latter of these times, beads re-appear in Africa and the Near East, and appear for the first time in Europe, Asia, and Australia. Clearly, it took well over 100,000 years before the first anatomically modern humans started to behave as members of our species do today. Evidence until recently moreover has favoured the assumption that the transition in Africa from archaic humans (Homo heidelbergensis) that formed the predecessors

9

More fragmentary remains from Guomde, Kenya, are perhaps 250,000 years old (Stringer, 2011: 202).

256


of both Neanderthals and modern Homo sapiens has been a relatively smooth one, taking several hundreds of thousands of years, from about 500,000 years ago. New evidence, however, suggests that the famous Broken Hill cranium (‘Rhodesian Man’), the first important human fossil ever found in Africa and initially dated at around the same time, is only between 200,000 and 300,000 years old (Stringer, 2011: 252–3). It follows that the transition even to anatomical modernity was, in fact, anything but gradual. Given the datings for the earliest anatomically modern human fossils mentioned above, there was barely time at all. If no very rapid evolutionary transition took place, then it appears that we must assume that a greater variation in human populations existed at the time, with possibly several human species co-existing, and some interbreeding. We know that there is admixture in modern human DNA of genetic material from archaic humans, including Neanderthals and other descendants of heidelbergensis (Denisovans), showing that the picture is not as clear-cut as original versions of the ‘Recent African Origin’ (RAO) model suggested. Stringer (2011: 261–3) concludes that even though interbreeding of modern Homo sapiens with archaic humans did happen (in and out of Africa, but more likely mostly in Africa) during periods of co-existence, and hence our genes come from multiple regions, ‘mostly out of Africa’ or ‘predominantly of recent African origin’ remain accurate descriptions of modern human origins. That is, no ‘large-scale integration’ of different gene pools took place (Tattersall, 2008: 104). There is little reason to lump all human fossils of the last 1 million years or so together under a single species name, if separate species with long separate evolutionary histories can be discerned, large-scale physical differences in their anatomy, not to mention their archaeological records, remain, and even in our own species things took a long time before we see its essential features. It may be that the pre-eminence of Africa in the human story is not solely a direct reflection of superior cognitive or biological endowment of the species inhabiting that continent, but also a matter of greater opportunities for morphological and behavioural variation and innovation, as perhaps provided by larger geographical and human population sizes within which they could be developed and conserved. Perhaps, biologically and cognitively, the resources we needed were in place by the time anatomically modern human developed, but their fully human expression and behavioural manifestation had to await a ‘releasing factor’—some cultural rather than biological innovation (Tattersall, 2008: 103). As d’Errico et al. (2009) put this concern: the cognitive prerequisites of modern human behaviour were in place earlier in time amongst both ‘archaic’ and ‘modern’ populations, and we must invoke, among other factors, historical contingencies triggered by climate and demographic factors rather than a single speciation event or mutation to explain the emergence, disappearance, and re-emergence in the archaeological records of symbolic traditions. (d’Errico et al., 2009: 50)


257

However, although it seems likely that such contingencies did play an important role in the history of our species after its earliest anatomical beginnings after 200,000 years ago, earlier African hominins and Neanderthals were anatomically and behaviourally very different. Why should they have been cognitively the same, especially if their cultures suggest no such thing? Why would, for hundreds of thousands of years, cultural opportunities not have arisen for these humans to exercise their cognitive modernity and leave some traces of it behind in the archaeological record? In any case, the following facts remain: The first flickering of behaviourally modern humanity involving creative symbolic thought, scattered as it is initially, is not seen anywhere earlier than at most 100,000 years ago, on current evidence. Language, being symbolic in the same sense, is therefore not likely to have been around much earlier (for complications, see Behar et al., 2008). A system exhibiting discrete infinity cannot evolve gradually: the step from oneword or two-word sentences to sentences infinitely long is as large as the step from three- or four-word sentences. A modern human culture erupts and stabilizes in a relatively discontinuous fashion in Cro-Magnon Europe around 40,000 years ago. Even the Châtelperronian culture, whose authorship remains unclear but which could have been created by Neanderthals, remains distinct from the new Aurignacian culture in the short period where these co-existed, and is replaced by the latter in an abrupt fashion, as is the previous, Mousterian one, and as are those of the Homo erectus East Asians (Tattersall, 2008: 106–7). After that, there is a phase change in the pace of technological change, with four major cultural traditions evolving in Cro-Magnon Europe between about 40,000 and 10,000 years ago. The difference that gave the Cro-Magnons an edge over the Neanderthal Europeans was surely not physical, but must have been largely cognitive (Tattersall, 2008: 104). The African record for similar modernity visible from 75,000 years ago is scarce, with ‘no continuity nor real accretion of innovations’ after its first occurrence (d’Errico et al., 2009: 48–9). For most of the six million years after the split from the chimpanzee, evolutionary change was mind-bogglingly slow and often stagnating, with nothing even remotely suggesting either this eruption or showing a momentum building up to it, indicating a lack of relevant biological and cognitive capacities; These capacities had stabilized biologically in our species at the very latest by the time of the Aurignacian revolution, which is also the time of the arrival of our species in Australia (Thorne et al., 1999), where we see no other linguistic or cognitive abilities in humans than elsewhere.

258


A case therefore remains for biological and cognitive capacities distinctive of different human species, and ours in particular, even if, importantly, these appear to need a releasing, perhaps cultural factor. The facts are not consistent with a new mode of cognition evolving largely culturally and simultaneously in different species. Such evolution also doesn’t significantly continue in our species, given the last point above. In short it looks as if the brain has to be right, though the external conditions do too. If they are not adverse, biology will break through. Not any hominin brain will do, and we cannot explain language evolution through language ‘adapting to the brain’ (Christiansen and Chater, 2009): the brain had to change first. The question we will address in the rest of this section is whether the releasing factor for our new mode of cognition to break through was most likely language. As Tattersall (2008: 103) puts it, ‘the fundamental innovation that we see with the Cro-Magnons and their African precursors is that of symbolic thought, and this is something with which language is virtually synonymous’. A creature with symbolic thought in Tattersall’s sense has stopped orienting its mental processing towards external stimuli, but lives in a mental world whose primitive elements are ‘symbols’ that it manipulates instead of the external objects, breaking down the external and recreating it in their own heads (Tattersall, 2008: 101). Can such a species be different from a linguistic species? What good would a new, symbolic mode of thought be, if it couldn’t be expressed and (to some relevant extent) shared, given the sad lack of telepathy in our species? The appropriate mode of expression for such a mode of thought, however, is language and nothing else: the ‘embodiment’ of thought is language, not some other system. A mode of thought inexpressible in language would be a mystery to us: we wouldn’t know it exists. In the same way, language not used to express thought would not be called by the name of language. We cannot, say, re-define the term ‘language’ so as to only denote speech, since a creature can speak yet be a parody of a human otherwise. In this sense, the evolution of language and the evolution of thought are conceptually non-distinct. Evidence for double dissociations traditionally cited in defenses of Fodorian modularity, if not outright circular through the way ‘language’ and ‘thought’ are defined, does not suggest the opposite, as we argue in Chapter 8. In short, we would not call a form of speech or sign ‘language’, if the speech or sign were not accompanied by thought of the kind that we normally express in language; and similarly, a form of thought that wasn’t expressible in speech or sign, would not be thought of the same kind, and certainly not be known or recognizable as such by others, and it could not be shared as a public property, leading to a pace change in cultural evolution. It would also not have the properties of objectivity or intersubjectivity that so centrally marks the propositional contents of the sentences we utter, which can be grasped by two speakers in the exact same way, even if they disagree on their truth value (indeed they must agree on the contents, if such disagreement is to


259

be logically possible). Language, then, unlike speech, forms a unity with thought, and Tattersall is right to see them as a package deal. This conclusion, which we will argue, in the final chapter, to be a necessary one, would make sense of the archaeological record, for if the earliest anatomically modern humans had a capacity for the relevant mode of thought, there is currently little evidence, as we just saw, that they could use it: language in the full sense, which includes externalization, was both missing and needed, explaining the relative poverty of their culture. Put differently: these humans had a mind, of a new and spectacular kind; but they didn’t yet know it; and thus it isn’t manifest in the material cultures they created. The earliest cave artists knew they had a mind. They had concepts of both an outer and an inner world, and of experience as a relation between the two; cave art, atmospheric as it is, seems to address and intentionally re-create such an experience. Language, it seems, is needed to know our minds; and it reflects such knowledge, as art does. Thought in non-linguistic species, if it exists, is thought with no awareness of it; thought that has no concept of itself, a concept that requires, and is a reflection of, having language. What we have talked about here is language in the full sense of an externalized medium. Unlike Humean impressions and ‘ideas’, concepts are shared, and they apply, if they apply, independently of whether we take them to apply: objectively. Ideas are formed in response to stimuli that are given online in experience. Like percepts, they are confined to individual agents’ minds. Concepts constitute a world of objective (speaker-independent) thoughts or contents, which can be shared, and the truth value of which is not a matter of what sense impressions reach our sensory organs: truth is a matter of thought, not of perception. We could identify concepts with ideas, and argue that the latter are objective and shared since they are ‘innate’. But it is hard to tell such ‘ideas’ apart from the words that lexicalize them, and the concepts that we identify as the meanings of these words. If these ideas are to do the job of such lexicalized concepts, they need to apply to the world in an objective fashion, so that thinkers take it to be independent of their judgements whether they apply or not to a given object. When such thinkers become speakers, we can see that with language as a shared medium, and an external world shared as well in which words are used, a baseline of objective meaning will arise against which judgements can be made. Without language, it is not clear what will provide the baseline, and our contention remains that no creature without language has ideas in the sense of concepts that we express in words. We return to this Davidsonian line of reasoning (Davidson, 2004) in the final chapter. Thought in this sense is not required for survival or adaptive action: ants compute paths of motion and calculators compute numbers, without giving it any thought that they do—which would require concepts of what these things are. The modularist tradition aligns language with computational machinery of the mind that operates reflex-like and unconsciously, separate from thought; yet every act of language use is

260


a conscious one: we know what we say, when we do. This begins to make sense if we are right that the connection with thought is essential to language, and language makes us conscious of the fact that we think, giving us a concept of thought, which nothing else can. While the externalization of thought in language is essential on this picture, appropriate vocal tract anatomy, which is widely taken to have long predated language, is not and can be present where language is not, as in seals or parrots. Evidence for such anatomy in Neanderthals, then, such as a lowered larynx is not evidence for language, pace d’Errico et al. (2009: 44), who infer normal ‘language abilities’ from evidence of abilities for speech (p. 44), and the genetics of language from the genetics of speech (p. 50), given evidence since Krause et al. (2007) that the FOXP2 critical for normal speech is present in the Neanderthal genome. In whatever sense humans before, say, 100,000 years ago were ‘talking’, it seems, this was ‘talk’ of a kind that wasn’t, as far as present evidence goes, paired with anything like a modern human culture of the kind that we see in Cro-Magnon Europe after 45,000 years ago, and hence not evidence for a mode of thought that is manifest in such a culture. Statements such as the following, which dissociate the species boundary from the language boundary, are rather remarkable in this light: We now know that genetically we were nearly identical to Neanderthals (only half a dozen genes separate us, mostly associated with skin and bone development). We thus shared essential features with the common ancestor who goes back at least 400,000 years ( . . . ) there can be no doubt that pre-modern humans were talking. (Levinson and Evans, 2010: 2742)

Clearly, in whatever way they were talking, they didn’t have the things to tell to one another that we know have richly structured the variable life world of a Cro-Magnon. As the evidence base stands, the connection between language and the species boundary remains crucial, and even more so if our own species wasn’t quite the species it would become, before it invented language as the sole appropriate medium of expression for a biologically novel mode of cognition.

7.4 Conclusions The story of language evolution under the Un-Cartesian perspective is not the story of a Fodorian module, and it’s not the story of a minimalist computational system satisfying the interface demands imposed by pre-existing systems of thought. It is the story of a system that is inseparable from a mode of thought that goes with it: the evolution of language and the evolution of this mode of thought are one and the same. Language defines the mental phenotype of that species: its kind of language is its kind of mind. None of this is to question the significance of the new sociocognitive infrastructure that characterizes our species and the kind of communicative intentions it may well uniquely have. But what needs to be accounted for is the


261

kind of mind that it has and what kind of contents it can grasp and intend to communicate. Grammar looked at in the Un-Cartesian way provides the cognitive infrastructure we need to obtain, with no extra-linguistic system of thought required. Language is thus not peripheral or an afterthought to the species problem, or how we evolved. The story of language is the story of that species, and the theory of its speciation, which we have not attempted to develop, should be a linguistic theory. Our species is not uniform, however, with regards to its genetic endowment for higher forms of cognition. There is genetic variation in our species, some of which, we will now argue, contains further pointers to the essential unity of language and thought when viewed from an evolutionary point of view. Our linguistic genotype is variable, but our mind varies along with it.

8 Biolinguistic variation In Chapter 5 we distinguished ‘cross-linguistic’ from ‘biolinguistic’ variation. Speakers of different human languages do not differ with regards to their linguistic genotype, which is the subject matter of UG in the twentieth-century sense. A comparative analysis of human languages will thus not compare humans with the linguistic genotype and without, or compare different such genotypes. In theory, one could attempt to distill a common core from the cross-linguistic variation that we observe, and then ascribe this core to UG. Yet, capturing the seemingly enormous variation among languages will require a rich descriptive vocabulary that may not lend itself to distilling such a biological, language-specific core. Nor may we want to ascribe aspects of language that such vocabulary captures to UG, which would then have to be very richly specified, as it was in the GB era. The minimalist alternative, to keep this core maximally underspecified, is also problematic, as it will then be hard to capture the cross-linguistic variation in a principled way. It is a promising perspective, then, to rather focus on variation in the linguistic genotype or UG itself, and in the mental phenotypes that go along with that. In this chapter, we do just that, and start by very briefly reviewing the Un-Cartesian hypothesis regarding language and thought against findings from developmental language disorders, questioning the notion of modularity as applied to these disorders (Section 8.1). We then discuss some evidence about what type of thought can prevail when the linguistic genotype is present but there is no or a radically reduced linguistic phenotype (Section 8.2). Next we review the notion of thought as applied to beings without the linguistic genotype (Section 8.3). Section 8.4 looks at co-variation between the mode of thought and that linguistic genotype, when it is present, with particular regard to the case of schizophrenia. The case of Formal Thought Disorder is particularly important for our enterprise here, as it is arguably characterizable as a ‘logopathy’—a disturbance of reason—as opposed to a ‘thymopathy’—a disorder of mood, such as depression (Dörr, 2010). The Un-Cartesian hypothesis therefore predicts that this disorder should involve a fragmentation of grammar. We make a case that there is considerable evidence for this, in Section 8.5. On the basis of this, finally, Section 8.6 makes some suggestions for where to look for grammatical meaning in the brain.

8. Biolinguistic variation

263

By and large, we believe that the findings we discuss in this chapter are consistent with the perspective we have pursued in this book, according to which the linguistic genotype defines a mental phenotype as well, rather than varying independently of it. There is considerable evidence that where there is variation in UG, thought varies alongside, and that thought fragments where and to the extent that grammar does, as the Un-Cartesian perspective predicts.

8.1 Questioning double dissociations in developmental language disorders A traditional modularity-based perspective may superficially contradict the idea that variation in the linguistic genotype will be variation in the mental phenotype. However, calling instances of language dissociating from thought by the name of ‘language’ has often meant to beg the question of modularity. Pinker (1994: 50–1) tells the story of Denyse, who comes across as a ‘loquacious, sophisticated conversationalist’, telling lively tales in flawless English about receiving statements from her bank account in the mail, losing her bank book twice; or about her sister’s wedding. Yet, the stories she so earnestly tells are figments of her imagination. As it happens, Denyse has no bank account, cannot really handle this concept, and her sister is unmarried. So her conceptual system is not like ours, and she doesn’t understand when sentences are true or false. This is not a deficit of language only if we exclude these features from what we call ‘language’. Or consider Williams syndrome, a rare genetic disorder found in 1 in 7,500 live births (Strmme, Bjrnstad, and Ramstad, 2002), which has been one of the classical cases supporting Fodorian modularity (Moscovitch and Umilta, 1990; Pinker, 1994), given a seeming pattern of cognitive impairment in conjunction with relatively spared language abilities (Bellugi et al., 1988). Yet again, patients with this syndrome have ‘normal language’, only if we decide to ignore aspects of language that are not normal, excluding these from what the label ‘language’ covers, which begs the question of modularity. Thus, in Williams syndrome, we could restrict the term ‘language’ to morphosyntactic and phonological processes as viewed under the perspective of autonomy, and find these processes relatively unimpaired. In line with this, individuals with this syndrome have been described as having ‘perfect’ syntax and morphology (PiatelliPalmarini, 2001). Yet a recent review (Brock, 2007) finds that: individuals with Williams syndrome have never been found to perform better than typically developing controls of comparable mental age on any tests of syntax or morphology, nor is there any convincing evidence for a dissociation between lexical and rule-based grammar. (Brock, 2007: 119)

It is noteworthy that individuals with Williams syndrome even outperform controls on tests of receptive vocabulary as opposed to grammatical abilities, which ‘are no

264


better than one would predict on the basis of overall cognitive abilities at any stage of development’ (Brock, 2007: 116). In turn, spatial language is distinctively impaired, and pragmatics is unusual as well, as is early language development. Semantics is not addressed in Brock’s review, following the traditional tendency of simply not including it under the term ‘language’, as if meaning was not regulated grammatically and located instead at some level of ‘thought’. As things stand, more evidence is needed to draw firm conclusions from this syndrome, and as of now it does not seem to support a modular perspective: it does not clearly exemplify a mind with impaired thought or cognition but spared language. Classical single-case studies like that of Laura (Yamada, 1990), who was 16 at the time of Yamada’s study, do not yield a different conclusion. Laura as described by Yamada exhibits a more severe and pervasive cognitive deficit than Williams syndrome children, and a discrepancy between her linguistic and non-linguistic cognitive abilities was found to be especially marked in the case of ‘syntax’. Yamada specifically and convincingly argued that her case study provided good evidence against a Piagetian model of language acquisition and for a modularity-based model, which, in her words, posits a ‘highly evolved, specialist human ability’ at the heart of the language-acquisition process, as opposed to ‘cognitive, social-interactive, and perceptual’ capacities (Yamada, 1990: 6). Laura’s language is described to be productive and structurally complex, involving complex patterns such as subordination, passivization, and complementation. At the same time, she had conceptual difficulties, and her use of ‘syntactically and semantically well formed utterances’ often turned out to be ‘factually incorrect’, indicating a difficulty with what Yamada describes as ‘semantic knowledge’. Thus, when asked about who the president was, she would name a president, or asked when an event took place she would name a point in time (Yamada, 1990: 62; see also 67). What this may suggest is that she could sort out lexical semantic categories, while being unable to encode propositional claims in grammar. To the extent that reference or understanding topics are also regulated by grammatical principles, her difficulties do suggest a difficulty with grammatical meaning in our sense. Describing her ‘syntax’ as relatively impaired thus raises an issue in the present dialectical context, when it is clear that part of understanding sentential grammar is to understand when (and in part also whether) particular sentences are true or false, or when something is a topic of which something else is a predicate. Yamada’s case against pregenerative conceptions of language is well made, and her defence of modularity should be seen in this historical context. Yet again, Laura’s cognitive impairments only fail to show in language if we do not include under ‘language’ those aspects of it that reflect the cognitive impairment in question. It does not seem implausible to conclude that Laura’s language is relatively normal insofar as it connects with normal thoughts that we would express in language, and that it is abnormal insofar as it does not. As for dissociation in the opposite direction, Specific Language Impairment (SLI) is found in about 7% of 5-year-olds (Tomblin et al., 1997) and has a likely genetic


265

basis. Again, it initially seemed to present a seemingly clear case of modularity: a deficit specific to the language faculty, but not other cognitive domains. But even today, SLI remains defined so as to make it an instance of modularity: it is defined negatively, or by exclusion: SLI comprises impairment of language production and/or comprehension, in the absence of a diagnosis of autism, hearing impairments, a general developmental delay, and neurological disorders such as seizures or perinatal bleeds (see e.g. Schwartz, 2009). In seeming contradiction to his own definition, Schwartz (2009: 3) notes that ‘SLI may occur at the same rate in other populations of children with language disorders. If this is true, subgroups of children with autism, children with genetic syndromes, and children with hearing impairments may have SLI’. Hence SLI may not be an impairment specific to the domain of language after all. Much research on the range of language impairments in SLI has focused on morpho-syntax and finite verb morphology in particular, where deficits are distinctly visible and highly characteristic. Yet, the question of whether these affect grammar or primarily its externalization in overt morphophonology is an open question. Eisenbeiss, Bartke, and Clahsen (2006), in a comparative study of children with SLI and typically developing (TD) children matched for mean length of utterances, report high accuracy scores for structural Case marking in both groups, as well as overgeneralization of structural Cases where the adult language would require lexical case marking (e.g. lexical Datives). This suggests sensitivity of both TD and SLI children to abstract, structure-based regularities, and is evidence against broad syntactic, as opposed to morphological, deficits in the latter group. Investigation of linguistic meaning in SLI seems to have focused, as is often the case, on word meaning, leaving everything above the word level to ‘morphosyntax’ without regard to the fact that meaning at a grammatical level is distinctly different from meaning at a lexical level, and syntax has an organizational function with regards to meaning as well. If grammatical meaning is unimpaired, this might suggest that deficits in SLI are not so much deficits in grammar or language proper as deficits in its externalization. ‘SLI’ would then be something of a misnomer. Further support for this perspective comes from evidence from a recent study of Donlan, Cowan, Newton, and Lloyd (2007), who explored the performance of children with SLI and age- and language-matched controls on mathematical tasks. While predictably underperforming in tasks that were linguistically loaded, as well as in tasks comparing number magnitudes that did make lesser demands on speech perception, comprehension, and production, the children in the SLI group did not perform differently from their typically developing peers on a test of abstract mathematical understanding that required invoking principles of commutativity and non-commutativity, respectively, across addition and subtraction operations. Algebra, however, involves principles that are not different, if not, in part, indistinguishable, from those invoked in contemporary minimalist syntax, for example the hierarchical and recursive formation of sets leading to discrete infinity in both

266


domains. Indeed, traditionally, language competence has been a prime illustration for what has been dubbed our ‘algebraic’ mind (Marcus, 2001), indicating that our linguistic and mathematical abilities may involve a common computational substrate, with arithmetic, whose ‘lexicon’ consists of numbers, ‘abstracted’ out from the richer and more embodied structures of language. Data showing that arithmetic and linguistic processing seem to rely on partially common neural substrates (Baldo and Dronkers, 2007) also support the idea that arithmetic and language, in part, are nondistinct as cognitive domains. On this line of reasoning, the data above, together with the general pattern of an impairment primarily of inflectional morphology in SLI, again suggest that it may not be language as such, as a cognitive-computational system, which is impaired in SLI, but rather its mapping to the sensory-motor interface: its externalization in a morphophonological surface. Once again, the cognitive tasks in the above study do not involve ‘language’, only if we exclude from the domain of language its abstract computational principles. There is other evidence, however, as noted, that the language phenotypes of individuals with SLI and with autism can overlap, and that cognitive phenotypes in these two disorders may do, too, though the data remain conflicting (Tager-Flusberg, 2004; Taylor, Maybery, and Whitehouse, 2012). If evidence that individuals with SLI exhibit cognitive characteristics reminiscent of autism was corroborated, SLI would be connected to a ‘cognitive’ phenotype as much as a linguistic one. There is also strong evidence, surveyed in Schwartz (2009), that children with SLI have problems with perception, with working memory, and with executive functions. This does not suggest a deficit specific to language. Overall, then, SLI, too, remains an unclear case for modularity, and it gives little support for the view that ‘thought’ may be preserved when ‘language’ is not. Thought, we maintain, is preserved to the extent that language is (and to the extent that it doesn’t depend on language). Again, for any such view to be supported, we need to avoid circularly excluding from ‘language’ what are core properties found operative in its computational system, which we have strong independent reason to separate from the principles governing its externalization in a physical medium (morphophonology). Externalization, though a crucial part of language, as we contend (see also next chapter), may be impaired when ‘language’ is not (or less so). We will now review cases where externalization is either missing altogether or severely impaired.

8.2 Thought in genetically normal language-less or language-impaired adults Adults that have developed no language at all, in the absence of any genetic abnormality leading to a developmental language disorder, have by definition normal specifications for UG—the linguistic genotype. A famous case is Ildefonso, the hero


267

of Susan Schaller’s deeply insightful book ‘A man without words’ (Schaller, 2012, first published 1991).1 This man was 27-years-old at the time Schaller met him in a language class for deaf adults: an intelligent man who, as Schaller makes very plausible, lacked the idea of what language or words were. The book describes the journey he undertook to ‘break the code’, the first of which was to understand that there was a code: that things had names, which meant something. After an arduous journey, Ildefonso eventually did develop a sign language, yet one that, the book suggests, remained different and more basic, though no details are provided. Raised in poverty-stricken circumstances in Southern Mexico, he had had no exposure to sign language, and time spent with a deaf sibling brother appears not to have been sufficient for the creation of a home sign system. Particularly noteworthy are the persistent difficulties that Ildefonso would come to have with grammar, an unfathomable mystery when the mystery of words has only just revealed itself. Even parts of speech distinctions proved extraordinarily hard to teach, particularly verbs, as did combinations of them (Schaller, 2012: 63, 72), and the related insight that words do not indiscriminately contain both the thing named and everything attached to it (p. 97). It is very clear from Schaller’s nuanced and personal description that this man, even if his initial attempts to communicate reminded her of a ‘mimicking chimpanzee’, had a sense of self, and a human nature: ‘I knew he had human intelligence, a human personality, and an awareness of himself as human’ (Schaller, 2012: 103). One thing that struck her most was ‘his strong desire to relate to others, to communicate even without tools’ (p. 157). She also perceived him as a moral agent (p. 192), and certainly as one with an ‘inner life’ (p. 105), though he ‘knew next to nothing about the world’ (p. 110), and maybe even lacked ‘beliefs, ideas, and assumptions’ (p. 105), collecting information primarily through the record of his senses (p. 92). Though he was by no means a tabula rasa, that is, ‘his mind was empty of all information that needs language as its conduit’ (p. 74). The question, of course—ours and very much Schaller’s as well—is how much information that is. Throughout the book, the central question haunting Schaller was ‘what his thinking had been like without language’. Schaller, it appears from reading this book, never found out: somehow, the matter remained shrouded in mystery. Maybe, where words are missing in a mental life, it may prove hard to describe such mental life, once words are finally there. Where does this striking case leave us? Much of Schaller’s book supports a case that the intrusion of externalized, public language into the mind is a seismic shift, remoulding much of our mental life and thought, consistent with the present perspective, notwithstanding a rich mental life based on experience and featuring practical reasoning independently of language. It also underlines that this was not a human being using language in the privacy of his own mind, with externalization being an 1 We thank Paul Elbourne, who brought this remarkable case to our attention (and raised it as a potential objection to our programme).

268


afterthought that didn’t change what was going on inside: a linguistic being, which Ildefonso was in one sense clearly not, requires the use of language as a shared communicative medium. At the same time, he was a normal linguistic being genetically, and whatever thinking did take place in his solitary mind prior to grasping the idea of language was that of a being with a language faculty that starts operating and maturing as we are born, and indeed earlier. We don’t know the degree to which it did in Ildefonso’s case, and how much of the idea of a word and of sentences had dawned in him. Overall, the case shows little about the power of genuinely nonlinguistic thought or its independence from language, beyond what we have already assumed: that perception and practical reasoning does not depend on language. The same, with qualifications, is true for another population of humans who equally have the linguistic genotype, but who have an acquired language disorder, often as the result of a stroke: individuals suffering from Broca’s (‘agrammatic’), Wernicke’s (‘fluent’), or ‘global’ aphasia (in which all linguistic abilities can be obliterated).2 Considerable capacities for thought and practical reasoning can be maintained in profound aphasic impairment, indicating a seeming autonomy of reasoning from language within an established cognitive architecture, allowing reasoning to be maintained when language is inaccessible—at least ‘language’ involving the principles and elements of morphophonology. The qualification noted above—that this happens when and after a linguistic mind has fully evolved and has been operative for much of a lifetime—is crucial, of course: none of this is evidence that such thought capacities can exist when language has never got into place. Nonetheless, we can ask: does aphasia show that language can be a mere ‘scaffold’ in development, which is tossed away once the cognitive architecture is in place? The thoughts of patients with severe aphasia at least appear rather normal: they drive cars, make elaborate travel arrangements, can plan their finances and travels, hypothesize, and develop elaborate strategies to circumvent their communication difficulty. And yet, compared with SLI, where language, even if delayed, is still in place, this is a case that can result in the radical excision of linguistic performance. Since, in SLI, a basic system of linguistic competence is maintained that can support abstract thought, our earlier question can now be raised about severe aphasia, and becomes more stark: if and where cognitive performance is still intact, is that because relevant aspects of language—though not of course morphophonological processing—have remained in place? At least initially, the answer now would seem to be much less likely, not least because the patients in question often have large lefthemisphere lesions covering the most canonical language regions in the brain. In a recent experiment, Willems et al. (2011) tested how patients with severe aphasia, given the opportunity to convey their communicative intentions non-verbally, perform

2

We are very grateful to Rosemary Varley for conversations on issues of the following section.


269

on a non-linguistic communication game in which the task was to convey the position and orientation of a geometrical shape on a grid to a naive receiver of the message. The expectation was that a patient whose grasp of meaning and whose communicative intentions and general problem-solving capacities were in place, would perform well on the task of conveying them. The same game was also played with the patients in the receiving role, which required them to figure out the communicative intentions of the sender of the messages. Patients performed well on this task, clearly suggesting that the problem of severe aphasia is not one of communication per se, but with language. This is as expected, if aphasia is a disorder of language, not of communication. Uncontroversially, communication is not the same as language: all species, including our own, communicate without language (though humans, in addition, communicate with it). Just as communication does not depend on language, solving a communication problem—conveying messages and understanding the content of messages— does not, either. What Willems et al.’s experiment does not address is the question of whether agrammatical aphasics have communicative intentions whose content requires a sentence to express. If they could have communicative intentions with contents of a kind that uniquely require a sentence to express them, such intentions and contents would prove to be independent of whether or not there is a shared linguistic medium in which they can be conveyed. There are two possibilities, in that case: the internal generative system powering or enabling the genesis of such intentions and contents is relevantly like language in an internalist sense, which then must be still sufficiently intact to generate them internally, even if they cannot be externalized (mapped to the sensory-motor interface); or the system is a different one. In the latter case the UnCartesian hypothesis would face a serious objection: thought is independent of language, even in an internalist, generative sense of ‘language’. In the former case, the problem that patients with severe aphasia face is not one of language but merely one of externalization: it’s a communication problem that their linguistic minds cannot solve any more. We don’t know how to decide between these options, even in principle. Since thoughts in severe aphasia cannot be externalized in language, and some thoughts cannot be expressed in anything other than a sentence, we simply don’t know how to find out whether there are such thoughts in severe aphasia. The thoughts we can find evidence for through non-linguistic forms of communication will necessarily not be of the right kind. Some evidence nonetheless supports the idea, exactly as in the case of SLI, that insofar as certain abstract forms of thought are still intact, relevant aspects of linguistic organization are, too. Varley and Siegal (2000), Varley, Siegal, and Want (2001), and Varley, Klessinger, Romanowski, and Siegal (2005) tested the ability of patients with severe and longlasting language impairments on causal reasoning, ‘Theory of Mind’ (ToM), and mathematical tasks. Participants proved capable of generating causes of both possible

270


and improbable events; they appeared to reason about false beliefs; and they successfully completed calculation tasks such as 59 13; 13 59; 60 12; 12 60, as well as tasks requiring sensitivity to structural features of mathematical expressions such as brackets and embedding (e.g. (5 x (6 + 8))), and the notion of infinity (e.g. real numbers between 1 and 2). One patient, SA, in particular, had severe impairments on grammar comprehension and production, and on word production, despite relatively spared word comprehension. He solved a switched container task in which he was shown containers with unexpected contents—for example a pill bottle containing buttons. In the test phase, he learned to associate cue cards with judgements that a naive person would make about the contents of a given box: cards with ‘X thinks’ written on them for judgements about belief, cards with ‘really’ for judgements about reality. The suggestion was that when the box contained unexpected items, person X would have a false belief. In the test phase, cue cards were presented together with questions like: ‘What does X think is in the [container]?’ However, all that this task seems to require is sensitivity to statistical regularities between the appearance of a box and its contents. Given the relatively intact word comprehension, it would then not be so surprising if some of the lexical contents of the words involved would be represented as well. But even so, this experiment seems to show nothing beyond what we would expect: that infants with immature grammar or adult patients with impaired grammar that are otherwise normal will not fail tests that do not require sophisticated grammar. Concepts like ‘belief ’ may be involved, but that lexical semantics is independent of grammar and that, to a significant extent, reasoning is so as well, is something we have already assumed. As is well known, 3-year-olds solve tasks of telling where an agent will look for some object first when that object has been displaced in his absence (Siegal and Beattie, 1991). Other tasks in the ToM paradigm are solved even at 15 months, when more automatic/implicit reasoning tasks are posed (Onishi and Baillargeon, 2005). So to the extent that we are willing to describe all of these tasks as ‘reasoning about beliefs’, it is already known that such reasoning does not require mature language. At the same time, it is also widely agreed that while there are perhaps some precursors to ToM in non-humans including chimpanzees and birds, and in prelinguistic infants, full-blown ToM is unique to humans (Cheney and Seyfarth, 1998; Tomasello, 2008; Penn et al., 2008) and developmentally matures only together with full access to the language faculty (deVilliers and Pyers, 2002; deVilliers, 2007). Astington and Jenkins (1999), in a longitudinal study of 3-year-olds conducted over seven months, found that scores on first-order false belief tasks were highly correlated with scores on language tests, with earlier language scores longitudinally predicting later ToM scores, and syntactic subtests as the better predictors as compared with lexical-semantic abilities. Confirming and expanding these findings concerning a close link between ToM abilities and grammar, a later study (Paynter and Peterson, 2010) showed for all three subgroups of their study, namely normally developing


271

preschool children, children with Asperger’s, and children with high functioning autism, that syntactic skills (as assessed with Bishop’s TROG-2 measure, see Bishop, 2003) ‘contributed uniquely and significantly over and above both vocabulary (lexical language skill) and other non-linguistic ToM correlates to the predicting of individual variability in children’s false belief understanding’ (Bishop, 2003: 383). Autism moreover is linked to a much more serious delay in the growth of ToM than Asperger’s, with delayed language in autism as the most likely explanatory factor behind delayed ToM. Of particular significance with regard to predicting ToM deficits is the understanding of the grammar of clausal complementation, i.e. recursion (Lind and Bowler, 2009). In adults, finally, there is evidence that performance on a non-verbal ToM task is disrupted in adults who satisfy a simultaneous task demand for verbal shadowing, which ties up their language faculty (Newton and de Villiers, 2007). It is very unclear whether the data from severe aphasia cast doubt on the conclusion that full ToM and language form a unity. So far they do not, and in fact the profile of accomplishment of patients appears to exactly mirror the available linguistic resources, co-varying with them: tasks are mastered to the degree that the grammatical requirements they impose are known to be independently mastered by the patients. A recent study (Apperly et al., 2006) investigates patient PH who suffers from severe aphasia with a linguistic performance comparable to that of the two patients SA and MR tested in the studies conducted by Varley and Siegal (2000) and Varley, Siegal, and Want (2001). It mirrored the pattern observed among 3- to 4year-old children who still fail standard false belief tasks (Smith et al., 2003). PH solved non-verbal first-order as well as second-order false belief tasks (involving false beliefs about false beliefs). The latter, in their verbal versions, include grammatically complex stories and questions, making failure on the task hard to assess in terms of whether it reflects a failure of language, executive function, or ToM (Perner and Wimmer, 1985). In the non-verbal version designed for this study, PH watched videos involving three participants, one of which (‘the pointer’) is understood to have the task of pointing to a box in which an object is hidden. Another participant, ‘the helper’, was understood to assist in this task by placing a marker on one of the boxes. A third person, ‘the man’, shows the helper where the object is located. Neither PH nor the pointer knows in which box the object is hidden. Then the helper pretends to leave the room, while in fact staying in it, though neither the pointer nor the man see this. The man swaps the boxes. The helper pretends to re-enter the room. PH needs to predict: Where will the pointer point? PH solved this task, which the authors describe as one of ‘second-order false belief ’: for the pointer will have a false belief to the effect that the helper has a false belief, and thus she will point to the box opposite to the one to which the helper will point. But in fact, the task need not be described involving the vocabulary of belief and truth at all. It is enough to have seen what others have seen or not seen, to have kept track of this, and to engage in practical reasoning with regard to how agents behave in

272


light of what they have seen. This is a capacity so ancient in evolutionary terms (Hare et al., 2001) that it doesn’t seem to tell us much in the present context. Specifically, the algorithm could be: .3 From this perspective it is interesting that PH performed at chance in versions of false belief that did require generating the propositional meaning of sentences like ‘He thinks there is a red ball in the box but really there is a blue ball in the box’, which require a grasp of the syntax of clausal complementation, and to then assess which of two pictures was truly described by such a sentence (Apperly et al., 2006: 344). On the other hand, PH did master verbal false belief tasks whose grammar was simpler, and such that the ‘meaning of the false belief question was not critically dependent upon its grammar’ (Apperly et al., 2008: 348). This illustrates that performance on ToM tasks where grammar is required aligns with grammatical abilities, as expected. The appropriate conclusion therefore seems to be: False belief tasks that require the processing of language and propositional meaning cannot be solved when the language faculty is not available; tasks are solved to the degree that they impose this requirement and the relevant demand can be met through a residual grammar capacity that is still in place. PH’s performance on ToM then exactly matches what we know and could have predicted from his linguistic phenotype, together with the intactness of his non-linguistic cognition. Overall, no evidence from severe aphasia reviewed above therefore supports a Cartesian conclusion. Apperly et al.’s (2008: 346) conclusion that ‘belief reasoning seems to be possible in the absence of the specific grammatical construction—embedded complement clauses—that is used for representing beliefs in language’ is correct, but what it means is what we should have assumed before: that belief reasoning remains possible only insofar as no such requirements on grammar are made, and that it becomes impossible otherwise. In fact, the evidence positively supports the view that insofar as systematic thought about beliefs, under this description, takes place, a lexical concept of belief is needed, jointly with a specific grammar that generates propositions from such lexical concepts by phase. Eventually, this grammar will produce bi- or multiple-phasal structures, in which two or more phases express a propositional content but only the highest phase is evaluated for reference/truth, reflecting the viewpoint of the ‘I’ that generates the thought in question. On this view, the theoretical construct of ‘ToM’, originally designed in the age of modularity as one ‘module’ among others, appears to be simply unneeded and is misleading: to achieve full ToM, all that a super-engineer would have needed to do is to start from the lexicalization of certain concepts, a process that is independently needed, and a capacity to generate phasally structured, recursive hierarchies in which embedded phases will ipso facto express propositional 3 Penn et al. (2008: 119–20) suggest a similar deflationary analysis for cases where both chimps and scrub-jays have been said to ‘engage in ToM’ (cf. Hare et al., 2001; Dally et al., 2006).


273

thought contents that are the syntactic arguments of verbs lexically denoting mental events. A creature of this sort, with lexicalized concepts and grammar and a capacity to theorize, will ipso facto have full-blown ToM, which will simply fall out from these capacities, assuming an otherwise normal cognitive architecture. Below the level of such full ToM, of course, are then lesser abilities that are usually lumped together under the term ‘ToM’ but are completely different and may not involve thought at all: ‘mentalizing’ in this broad sense, in particular, can simply include the monitoring and reading of intentions and states of information of agents in strategic interactions, taking into account what they have seen and therefore ‘know’. Such monitoring may be permanently ‘online’ in social interaction, involving the ceaseless scanning of the social environment and extraction of information on who wants and knows what (Bruene, 2005). In the present frame of things it is a form of social reasoning and cognition that is not human-specific and of no direct interest to the different question we are asking here. If decomposed into this ability, on the one hand, and an independent capacity to conceptualize, grammaticalize and theorize, on the other, ToM becomes redundant as an independent theoretical construct. In its full form, we could say, it does not merely depend on language, as Newton and de Villiers (2007) claimed: it is language. Suppose, however, if only purely for the sake of the argument, that there was a hypothetical patient XY with severe aphasia who could solve ToM tasks equivalent to ones for which normal grammar comprehension is required (we do not know of such a case, or even what it might look like). In that case, such patients, with a brain manifestly lesioned in what are taken to be ‘language’ regions, would solve tasks in their ‘thought’ without the assistance of language. It could then be proposed that language merely ‘scaffolds’ mental development and can, later in life, leave higher mental functions (‘thought’) in place when language is lesioned. Since XY’s reasoning, cannot, by assumption, have taken place in language, his reasoning will have taken place in another language-like medium, ‘mentalese’, which by assumption is a kind of language, and hence has a grammar, yet is not governed by the principles of language (UG). This would then seem to establish a Cartesian conclusion with regards to the language-thought divide—except that it would have presupposed this conclusion, which follows only if we begin from a thought-language distinction, with language as a module distinct from thought, and exclude from what we denominate as ‘language’ the kinds of meanings and structures that we will take to be processed in the posited mental language of ‘thought’. But it is not clear which independent evidence supports the view that language in the ordinary sense and in normal minds does not in fact compute such meanings and structures, so that they would instead have to be included under the category of ‘thought’. We would, that is, effectively define language as reducing to what appears on its sensory-motor interface, that is in morphophonology, thereby excluding from it aspects of grammar that yield the meanings in question. If so, where morphophonology has broken down,

274


only thought will be left—the Cartesian conclusion. But this conclusion is viciously circular in the face of positive evidence that grammar is the crucial generator of the meanings in question, and no other such generator is known. The more appropriate conclusion, therefore, in the hypothetical case in question, would be that XY has core computations of grammar still intact, even if the mapping to the interfaces is lesioned. This conclusion would be reinforced by independent reasons from standard linguistic theory that we need to distinguish grammar from morphophonology anyhow. There is some other evidence, however, that we will not, in fact, likely encounter patient XY in reality. A long-standing hypothesis in aphasia research is the Tense Underspecification Hypothesis (Wenzlaff and Clahsen, 2004; 2005), according to which a specific hierarchical layer in the Inflectional Phrase (IP) cannot be processed in agrammatic aphasia: finite Tense, performance on which is significantly more impaired than on Mood ( Realis). Impairments on grammarmediated temporal reference, rather than (subjunctive) mood or agreement marking, are seen in both production and comprehension, for both present and past tense, in both regular and irregular verb forms, and may have analogues in pronominal reference. This suggests that they affect grammatical computations proper, rather than merely their morphophonological expression (Clahsen and Ali, 2009). The problem of agrammatism, this evidence therefore suggests, is not merely one of externalization: it affects how semantic information is grammatically encoded in the brain—even if, as we have seen, this leaves much thought intact. If nonetheless we did find a patient like XY, the alternative to the scaffolding view would again be to say that precisely insofar as some mentalizing function requires structures of a kind that UG generates, the mind in question is still not de-grammaticalized—even if the mapping to the sensory-motor interface is severely impaired. This position would not face the problem that the scaffolding view faces: to provide evidence that the kind of representations involved in reasoning in mentalese are conceptually distinct from ones we know to exist in language. The dialectic is similar with regard to the mathematical problems that the patients studied in Varley et al. (2005) solved: structure-building in arithmetic, if recursive, minimally involves Merge and hierarchy. Good performance on the arithmetical tasks is therefore evidence that these aspects of grammar—which are, uncontroversially, core elements of the computational system of grammar (‘narrow syntax’) as described today—are still in place. Overall, then, the case of aphasia provides again no clear evidence against the Un-Cartesian programme; and in some ways, it supports it. So far, we have not encountered serious evidence that thought dissociates from language. The picture, rather, that emerges in this section, is that language emerges robustly in this species, and thought along with it. Neither the developmental disorders we have looked at, nor the genetically normal adults without language or with language impairments support the view that thought can emerge without language: rather, the former co-varies with the latter, as slight genetic abnormalities disturb its normal course


275

of development, or adverse circumstances prevent it from governing thought in normal ways.

8.3 Thought without the linguistic genotype At the beginning of this chapter, we mentioned the position of Fodor and Pylyshyn, who, back in 1988, could still confidently assert that non-verbal cognition was of essentially the same type as verbal cognition, instantiating ‘physical symbol systems’ with ‘mental representations’ that are compositional and exhibit constituent structure, allowing systematicity and productivity in thought and reasoning even independent of language. If so, a ‘Computational-Representational Theory of Mind’ (CRTM) (Rey, 2003) can be applied to both human and non-human organisms, and linguistic cognition will simply be an instance of a cognitive architecture described in the generic terms of the CRTM. Our restrictive goal in this short subsection is to provide our evidence that no such thing can be assumed, partially rehearsing arguments already put forward in Chapter 2. The central intuition of Fodor (1975; 2008), that our mind is not an associative engine but a computational one, is correct: to explain its workings, we need to appeal to the part-whole organization of mental representations. But this says essentially nothing about what makes a linguistic mind distinctive and distinguishes it from an insect mind, where the CRTM is equally required to make sense of complex computations mediating adaptive behaviour (Gallistel and King, 2009). A linguistic mind in our sense is not a computer but a concept user, where the notion of concept is that of a lexeme and contrasts with that of ‘percept’: it is something that can apply or fail to apply to an object. In the former case, we say it will apply truly: the proposition saying that it holds, as a property, of that object, will be true. In the latter case, we say that the concept user in question has made a mistake. Davidson (2004) argues that, without the possibility of a mistake in this sense, there is no such thing as a concept. We think the point is well made. It is difficult to find the same notion of a possibility of mistake in the realm of perception. There can be a form of misrepresentation there, but by and large, mental representations in this domain are still stimulus-controlled (even if not stimulus-determined, as noted in Chapter 2). The same applies in the realm of ‘core cognition’ (Carey, 2009; Spelke, 2003), where representations, while of considerable abstraction, are still continuous with perception, being dedicated input-analysers in specific domains, and probably, as Carey suggests, iconic in their format. Iconic representations do not clearly allow for intensionality, a hallmark of both conception and thought: it arises when the question of what thought is being thought by an agent is systematically sensitive to what concept, with its descriptive content, is involved in it. This is always true of linguistic representations, all of which can switch from truth to falsehood as lexical concepts involved in them are exchanged for others

276


that apply to the same objects. My thought that three giraffes are being depicted is not the same as my thought that three of my favourite animals are depicted, even if giraffes are my favourite animals. Thinking, or saying, the one thing and the other are not the same.4 But a picture of three giraffes, as Fodor (2008) points out, cannot in this case help being a picture of three of my favourite animals. Do we find thought that allows for mistakes and intensionality in organisms that lack the linguistic genotype? Organisms, Davidson (2004) notes, make adaptive discriminations of environmental features: the sunflower orients towards the sun, the earthworm towards food, the bee computes values of algebraic variables denoting angles or the solar ephemeris. Adaptive behaviours based on discriminations of environmental features and computations over environmental variables need not apply any concept of the things discriminated, and they do not require any insight on the part of the plant or animal into what they are doing. The earthworm may take poison for food and die, yet we lose a distinction by assimilating this to the making of a mistake. Death doesn’t prove us wrong, and survival doesn’t prove us right (notwithstanding possible correlations between the two in the human case). In one of Davidson’s examples (2004:141), an octopod is trained to distinguish between two letters, ‘V’ and ‘W’, leading it to be disposed to react to them differentially. What if it reacts to ‘V’ as we have trained it to react to a ‘W’? Has it made a mistake? Has it misapplied a concept, or indeed our concept of a ‘V’? Suppose we answer that what provides such evidence is that the animal does not act in accordance with our idea of what makes the two stimuli similar. But then, this idea is our idea. The animal’s action is then still perfectly consistent with it acting according to some rule—and indeed, whatever it does could be consistent with a rule, though possibly not ours. And so we have no basis to conclude that the creature followed a rule at all: If behaviour is always consistent with some rule, there is no rule or concept. There is a rule or concept if there is evidence that a creature understands the idea of an error: that, objectively, a given concept can fail to apply. There is evidence that every normal human does so, or that we possess the concept of truth that is required for it. The evidence is a specific pattern of social and communicative interaction: our concept of what some object is, is connected to a range of other concepts, which are parts of beliefs. Many of these must be true, if doubt about some of them is to make sense. We assume this when we interpret other human beings: we assume that, largely, our beliefs are true, and those of the agents are, too. Assuming this, we can figure out what they say and belief in a particular case. This answer to why we grant concepts to ourselves and not the octopod depends on our sharing a conceptual framework mediated by linguistic communication, which in turns depends on what we are calling a grammatical mind. Evidence that we share 4 Even if this is denied by some philosophers, the very fact that there is disagreement on how to treat such cases in formal semantics shows that an intensionality effect exists.


277

causal encounters with external objects is not enough. We share these with the octopod. The content of a perceptual belief such as that a lion is present, as Davidson (2004: 142) argues against Burge (1986), is not the usual or normal cause of that belief. His reason for this claim is that if past correlations of lions with stimuli similar to the present stimulus were the cause of my perceptual belief, then an infinity of other causes would equally qualify as causes of my belief: for example, past stimulations of my sense organs, or past photons streaming from lion to eye. But then, if the causal theory held, my belief would be about stimulations or photons, when in fact it is about lions. The content would co-vary with endless other causal explanations we could give; and hence the content is not explained causally. The causes are too many, but our concept, that of a lion, is only one. What our beliefs are beliefs about, rather, is what our concepts determine. A shared concept will narrow down the possible causes to only one. What counts as ‘similar stimuli’ or ‘similar responses’ is determined by how we think of the world: our concepts determine classes of possible objects of reference. ‘When it is conceptualization that is to be explained’, Davidson points out, ‘it begs the question to project our classifications onto nature’ (2004: 142). Concepts, in order to play this role, must be able to apply or not to apply, which is to say that they are parts of propositions that we assume can be true or false, forming the content of judgements. Judgements that something is green, or a ‘V’, thus have external promptings, but they are not explained by what is externally or causally there. The truth of propositions is distinct from beliefs, and can be taken to be distinct only if a concept of belief is available. As noted earlier in relation to ToM, evidence that chimpanzees track what others see and want (Hare et al., 2001), is not evidence that they have such a concept. For they need not embed any such concept in propositions that they judge to be true or false, when practical strategies for adaptive responses in competitive interactions that do not involve any such concept, suffice (Penn et al., 2008). Adaptive behaviours in non-linguistic beings may be symbolic in the sense that algebraic variables are represented that take a potential infinity of values, entering formal computations (Gallistel and King, 2009). Some such symbols may also represent particulars, and hence are functionally discrete, giving rise to a form of ‘compositionality’, as when scrub-jays remember where they have hidden what and when, for thousands of independent cache sites (Clayton et al., 2003). Each such system of mental representation comes with a ‘syntax’, perhaps one for each species (cf. Bloom, 2000: 517: ‘every species gets the syntax it deserves’). But none comes with grammar, in our sense: a system converting lexicalized concepts into acts of reference that have a systematic and objective formal ontology, involving the concept of truth. Syntax, therefore, in the generic sense of the CRTM, is not the point. Grammar, if we are right, may be. Davidson’s notorious verdict (1982) that non-human animals do not think may seem harsh, and it has seemed circular to some, in the sense that it first defines

278


thought as propositional, and propositionality as sentential, and then concludes that non-linguistic beings do not think and are not rational. But this would be circular only if no argument is given that propositionality depends on language. But there is such an argument, and hence the step from propositionality to human uniqueness is not definitional. The argument is that thinking requires knowing that thoughts can be false. Knowing that, in turn, requires knowledge that truth is not belief, and hence concepts of these. Neither perception, nor cooperation, nor causation are a basis for these. Language, arguably, is. Also the step from the existence of thought to its propositionality is not definitional: rather, it is rationally indubitable that we think. Thinking, the argument then proposes, does not come in isolation, but is only what it is in the context of other attitudes such as beliefs and desires. These must form a rational pattern, for the logical consequences of a particular belief, e.g., are part of what makes it the belief in question. Hence, if the beliefs would be radically irrational or illogical, they wouldn’t be the beliefs in question. All of these attitudes furthermore share an element: a particular kind of content, to which all of such attitudes are directed (this does not imply that this is some kind of ‘entity’ to which they are directed). A crucial property of such content is that it can be true or false, independently of what we believe it is. Grasping such contents thus requires possession of a concept of truth. Thought thus turns out to be both logical and propositional, in this sense. Could there be animals with a non-propositional kind of thought? Imagery, say? Or emotions? But what would it be for there to be such thoughts? It would, obviously, be for us to be able to interpret these animals as having such thoughts. Interpretation would happen according to the constraints under which all interpretation takes place, and all interpretation, arguably, is such that it maximizes rationality and agreement, against a background of shared concepts and beliefs: we wouldn’t ever get far, in interpreting others, by assuming that they do not take the same things to be true or false, that they are systematically irrational, or that they have radically different concepts. There is no room, here, for an alternative conceptual scheme, as concepts, where they are ascribed, will come out as ours. Thoughts, therefore, where they exist, must be thoughts we could have. Nothing, here, rules out that, one day, we might be able to interpret dogs or missiles or computers as thinking. It is just that, so far, we are not. There might be thinking dogs, one day. But a thinking dog, like Brian in Family Guy, would be a talking one: this is the argument. Language is a measure of mind, and none other is so far known.5

5 Penn et al. (2008), providing a comprehensive discussion of comparative cognition research over several decades, do claim to compare human and non-human minds, assuming there to be such things as the latter. This contention depends on shortcutting from mental representation to thought, whereas we have been at pains to stress the distinction between representation and (intentional) reference in Chapter 2.


279

8.4 Variation in the linguistic genotype In the cases of language-less adults, the aphasic patients, Laura, and persons with Williams syndrome, the claim can be defended that thought is essentially normal, despite the difficulties that all of these patients or individuals face. It may not be available in its full richness, and there may be low intelligence, but thought is not ‘abnormal’ in the sense that it would be disordered, incoherent, or illogical. In line with that, there is a linguistic genotype (UG) in all of these cases, and a linguistic phenotype has either developed normally, with delays, or late, or it breaks down after it has normally developed. Abnormalities in both the linguistic genotype and phenotype exist in some of these cases, but they may seem to affect linguistic processing more at the periphery than in its core components, leaving, along with some aspects of grammar, much thought intact. More interesting cases to consider in this regard are therefore ones where there is variation in the linguistic genotype, of a kind that goes along with a severe disruption of the very organization of thought itself: Autism Spectrum of Disorders (ASD) and schizophrenia, in both of which there is no question that thought is comparably disturbed. From the present perspective, thought and language, of course, should stand and fall together. ASD affects four in every ten thousand individuals (Baron-Cohen et al., 1985), and it is, as its name suggests, a widely diverse spectrum of disorders with a range of partially heterogeneous symptoms and great variation in these. It is perhaps most commonly regarded primarily as a ‘cognitive’ and ‘social’ deficit, rather than a ‘linguistic’ one. Nonetheless, apart from impairments in social interaction and inappropriately restrictive behaviour, impairments in verbal and non-verbal communication are a part of a basic triad of symptoms and a principal early manifestation of the disorder. Pragmatic language disorders are found in all autistic children (Young et al., 2005). As for non-pragmatic ones, it is noteworthy that there are still relatively ‘few studies [which] have examined grammatical abilities in autism, with mixed findings’ (Eigsti et al., 2007). Most commonly, however, autism is not regarded as a disorder of the language faculty. Rather, language and communication abnormalities in this disorder are regarded as consequences of a more general deficiency in ‘mentalizing’ or ‘ToM’ (Baron-Cohen, 1995; Baron-Cohen, Leslie, and Frith, 1985). Above we have tried to splash doubt onto the very coherence of ToM as a theoretical construct or natural kind. What is called ToM, as if it existed, we argued, arises in part from a disregard of

Even if the equation of thought with mental representation is endorsed, Penn et al. (2008) show that principled differences in human and non-human forms of thought remain. The authors reject language as the principle explanatory factor for the discontinuity in evolution in question. This rejection is based, in our terms, on modular or minimalist (‘recursion-only’) conceptions of language that we here question, as well as on notions of ‘ToM’ or data from aphasia that we discuss in this chapter.

280


the fact that grammar is constitutively involved in the genesis of particular kinds of meaning, and in part reduces to capacities of perception, conceptualization, and practical reasoning. ToM, taken as a fundamental element of what is taken to be ‘thought’ rather than ‘language’, is nonetheless still very much alive within psychology and neuroscience in discussions of both autism and schizophrenia (BaronCohen, Leslie, and Frith, 1985; Frith, 1992; Bruene, 2005). Only more recently, language abnormalities in autism have received more detailed attention and have been looked at in their own right. In this regard it has been suggested that subgroups with both normal language skills and impairments at all levels of linguistic organization exist within the autistic population, suggesting that impaired ToM, if universal in autism, can leave language abilities unaffected (Kjelgaard and Tager-Flusberg, 2004; Tager-Flusberg and Joseph, 2003). Maybe, then, the subgroup with significant language disorganization is a distinctive phenotype of autism that has to be thematized in its own right, and resembles the phenotypes of Down syndrome or SLI (Tager-Flusberg, 2004). Down syndrome children have normal ToM abilities (at least to the extent that these are languageindependent) (Baron-Cohen, Leslie, and Frith, 1985), showing again that deficits affecting the language faculty, viewed as a ‘module’ independent of the ‘ToM module’ (Belkadi, 2006), may define an independent subtype of autism. On the other hand, findings mentioned earlier from Paynter and Peterson (2010) concerning the close link between mastery of ToM and linguistic abilities, particularly syntactic ones, strongly question the separation of language from ToM as distinct modules. Insofar as ToM goes beyond the perceptual and practical reasoning abilities in non-humans and prelinguistic infants, it forms a close unity with language and may be conceptually indistinguishable from it. The characterization of autism of the relevant subtype as a deficiency of the language module may thus again presuppose the modular view of language: if we decide to regard the language faculty as an autonomous module, independent of cognition and ToM, with syntax being an arbitrary formal complexity that plays no role for thought, then of course it can be distinctly affected, leaving cognition intact. But this view circularly depends on characterizing syntax so as for it not to affect thought or cognition. Nonetheless, we will not pursue the autism case further here, fascinating as it is, since schizophrenia is of much greater interest to our project, as a candidate for variation in UG within our species. Found in about 1% of the world’s population (Jablensky et al., 1992; Mowry et al., 1994), and first manifest in early adulthood, schizophrenia is diagnosed through a constellation of symptoms and a degree of functional impairment. Among the ‘positive’ symptoms—involving an excess or distortion of normal function—are auditory hallucinations (particularly verbal), delusions (distorted perception and interpretation of the external world), and problems in expressing thought through language leading to discourse that is disorganized and difficult to follow. Positive thought disorder in this latter sense in particular


281

includes ‘derailment’—in spontaneous speech, the speaker tends to slip off track, juxtaposing ideas that normal hearers do not perceive as related—and ‘tangentiality’—answering questions in an oblique or irrelevant manner. Both are aspects of what Bleuler (1911/1950) terms the ‘loosening of associations’: the apparent following up of associations between words that, together, do not produce a coherent discourse (propositional meaning, in our sense). Maher et al. (1983) give the following example: If you think you are being wise to send me a bill for money I have already paid, I am in nowise going to do so unless I get the whys and wherefores from you to me. But where the fours have been, then fives will be, and other numbers and calculations and accounts to your noaccount . . .

In rare and severe cases, speech can be unintelligible, with neither individual words nor sentences together yielding an overall discernable meaning: ‘word salad’ or schizophasia. Andreasen (1986) gives the following example: They’re destroying too many cattle and oil just to make soap. If we need soap when you can jump into a pool of water, and then when you go to buy your gasoline, my folks always thought they should, get pop but the best thing to get, is motor oil, and, money. May may as well go there and, trade in some, pop caps and, uh, tires, and tractors to grup, car garages, so they can pull cars away from wrecks, is what I believe in. So I didn’t go there to get no more pop when my folks said it. I just went there to get a ice- cream cone [sic], and some pop, in cans, or we can go over there to get a cigarette.

Severity of thought disorder is assessed in a number of clinical rating scales, for example the Scale for Assessment of Thought, Language and Communication (TLC) (Andreasen, 1986). When Kraepelin first dubbed schizophrenia as ‘dementia praecox’ (Kraepelin, 1896/1919), he considered ‘derailments in the train of thought’ to be fundamental to this disease (p. 72), including loosening of associations (‘the most different ideas follow one another with most bewildering want of connection’, p. 56) and incoherence (‘bewilderingly nonsensical utterances . . . apparently represent a senseless jiggling of words’, p. 71) (cited form Kuperberg, 2010a: 177). Although thought disorder is relatively rare in the schizophrenia syndrome, Bleuler (1911/1950), too, considered ‘disorders of association’ of primary importance for diagnosis, demoting delusions and hallucinations to secondary and derivative status as ‘accessories’. But why are patients where language abnormalities like the above are found, said to be ‘thought disordered’? As Kuperberg (2010a) notes, this reflects the perspective of the founders of psychopathology in the late nineteenth century, who regarded the abnormalities in question, not primarily as language disorders (aphasias), but as a manifestation of an underlying disorder in thought (Bleuler 1911/1950; Kraepelin, 1904; Kraepelin 1906). This conclusion is telling, from the present point of view. Why should a disorder that is manifest in language be a disorder of our mental nature, which then is only

282


‘expressed’ in language? Clearly this follows again only if we exclude from what we call ‘language’ the aspects of illogicality, incoherence, and meaninglessness, that we perceive in thought disorder. Language is then simply an arbitrary medium used for communication, while our mind is structured by independent and non-linguistic principles, which break down in this case. This would be plausible if we knew what these independent principles were—and implausible if the actual pattern of disorganization we find in the language of schizophrenic formal thought disorder stands in an explanatory relationship to, or illuminates, what is described as a thought disorder. From an Un-Cartesian perspective, we will naturally explore the latter alternative. As this perspective would also predict, the boundaries between thought and language disorder in this symptom have, for more than a hundred years, proved difficult to draw. As has often been the case in discussions of thought disorder, a principled boundary between thought disturbance and language disturbance is difficult if not impossible to draw in assessing behaviour that largely depends on verbal performance: ‘thought disorder is when talk is incoherent. And talk is incoherent when the thought is disordered’ (Rochester and Martin, 1979: 6). Andreasen (1986: 473), too, notes that most of the phenomena captured under the label ‘formal thought disorder’, from an empirical perspective, are actually disorders of (linguistic) communication. As Levy et al. (2010) document, although both Kraepelin (Kraepelin, 1896/1919) and Bleuler (Bleuler, 1911/1950) considered the symptom in question as primarily reflecting a disorder of thought rather than speech, the point that thought cannot be extricated from language wasn’t lost to either of them: even from a Cartesian point of view, verbal communication is the primary locus where thought is manifest and externalized. Yet it is characteristic that Bleuler describes this as a deplorable limitation, forcing the theorist through a detour: ‘we are forced to deduce most of the anomalies from the oral and written productions of the patients.’ (Bleuler, 1911/1950: 39) Kraepelin, too, referred to ‘derailments in the train of thought’ (Kraepelin, 1896/ 1919: 72) as much as to ‘derailments in the expression of thought in speech’ (Kraepelin: 70) and to ‘disorders in connected speech’ (p. 67) or ‘confused syntax’ (p. 71). This stance, which assumes a duality of thought and language and a disorder of the former as prior to a disorder in the latter, recurs to this day. Thus, Titone et al. (2007: 93) write that ‘spoken or written language is the most common medium for conveying disturbances in thinking’. Yet, in their view, ‘the thinking anomalies associated with psychotic conditions are not, fundamentally, speech or language disorders ( . . . ). Rather, when language is used in an idiosyncratic way, it represents the outcome of a deviant thought process’. Clearly, whether this is coherent completely depends on what theoretical view we take of language. Another tradition starting from Kleist (1914) disputes that formal thought disorder (FTD) is an illness of the mind (Geisteskrankheit) and claims it to rather reflect a


283

disorder of language (aphasia). But our Un-Cartesian stance entails that this is a false dichotomy, which fails to recognize that language and (at least one particular kind) of thought may be integrally related. But to this day, studies of language in schizophrenia with systematic involvement of linguists remain rare. An early exception is Chaika (1974) who, apart from echoing the classic clinical observations that sentences were produced ‘according to the semantic features of previously uttered words, rather than according to a topic’, found other abnormalities, in particular syntactic ones. The observation just cited, too, amounts to a grammatical one, if we recognize that ‘topic’ itself is also a grammatical notion, related to the notion of ‘subject’ and contrasting with the notion of ‘predicate’ (as what is said about the topic). Loss of topic is thus possibly an epiphenomenon of loss of grammar, which in turn would seem to predict more reliance on word-word associations. It would also predict that hearers notice a discrepancy between word meanings and discourse-level meaning: where grammar is lacking as an organizational principle of meaning, individual words cannot merge smoothly into a coherent grammatical meaning corresponding to a communicative intention, and words will show up in spontaneous speech that will seem to a hearer, who is in the process of reconstructing communicative intentions online, to not possibly be the intended ones. In normal speakers, by contrast, speech errors such as failures to utter intended words that are coherent with the discourse meaning are ‘momentary deviations from a discourse plan that is immediately resumed’ (Covington et al., 2005: 78). The assumption that there is no discourse plan in the schizophrenic case, because of a fragmentation of grammar that does not allow such a plan to be computed and upheld, would make sense of this difference. One could, obviously, maintain that discourse levels only require ‘discourse-level intentions’, with these being interpreted in non-linguistic terms and as having their contents language-independently. That, of course, is the Cartesian view that we encountered before in connection with Tomasello’s (2008) account of communicative intentions. Yet again, some mechanism for generating the contents in question must then be assumed, and it is not clear what this could be. Positive FTD seems like a perfect case to test the Un-Cartesian hypothesis. The Un-Cartesian model of grammar argues that the intrusion of grammar into the hominin brain specifically correlates with a fundamental shift in what we have called deictic reference, which we take to be essentially grammatically mediated. It begins with index finger pointing in infancy, but grammar is required for most of our acts of reference, including reference to propositions, facts, truths, or objects located in the past or future. We therefore predict, on the basis of this conception of grammar, that, with grammar fragmenting, reference and propositional thinking will fragment as well and a rational connection between thought and world will be lost. Thought will not be ‘about’ anything anymore: there won’t be topics, which are grammatically the subjects of predications that are true or false. Words may well be understood, consistent with lexical knowledge and semantic memory being relatively spared in

284


schizophrenia, but their use in coherent acts of reference to the world will be impaired. These predictions clearly speak to a range of schizophrenic symptoms, and a range of empirical evidence supports them. Thus there is over-proportional use in FTD, relative to non-FTD schizophrenics and normals, of ‘generic’ and indefinite noun phrases (‘the Chinese’, ‘religion’, ‘stuff ’, ‘ones’, ‘mankind’, ‘wives’, ‘a man’), which are low in the hierarchy of reference described in Chapter 4. Schizophrenics, in general, seem ‘more concerned with universal issues than normal speakers’ (Rochester and Martin, 1979: 125). Generics do not refer to specific objects or discourse participants, hence do not require setting up reference chains involving specific objects referred to before, which are partially governed by grammatical principles. Patients with FTD, more so than those without, use definite NPs with unclear or indirect referents (Rochester and Martin, 1979: 148, 167); they prefer to point non-linguistically, and have difficulties with deixis (p. 160). Lack of referential cohesion across stretches of discourse is said to particularly affect pronouns (McKenna and Oh, 2005: 112), which are the most paradigmatically deictic devices. Watson et al. (2012) found, among a group of people at high genetic risk of schizophrenia, that those who would subsequently develop schizophrenia used significantly more second-person pronouns, as revealed before diagnosis at two separate assessments approximately 18 months apart. This supports and gives a specific linguistic-grammatical signature to the view that schizophrenia, beginning from preclinical stages, involves the breakdown of the ‘deictic frame’ (Buehler, 1934) of interpersonal communication—that is, the perspective of a ‘self ’ referring to itself in the first person, who talks with a second person about a shared, objective world distinct from them both (Crow, 2007). Put differently, and in line with Section 4.4, it involves the breakdown of the system of grammatical Person. Pursuing this line of thought, then, language in schizophrenia may not only ‘provide a window into relationships between the language system and non-verbal higher-order thought’, as Kuperberg (2010b: 8) conjectures, but speak to a stronger philosophical view: that there is no non-linguistic (rational) thought. Where language fragments, thought does, too: it becomes illogical. This also gives us the right contrast with the aphasic case discussed above. If thought disorder is related specifically to how meaning is organized and integrated in language at different levels, it must be the case that in agrammatism (Broca’s aphasia), where thought can be largely intact, the pattern of language disintegration is different. Indeed there is evidence that it is, and that standard aphasia test-batteries do not detect FTD (Oh et al., 2002; Covington et al., 2005). Morphophonology in particular is largely intact in schizophrenia but not aphasia. As Covington (2009: 87) notes: ‘Patients with aphasia have normal thoughts and express them with difficulty; those with schizophrenia have unusual thoughts (or disorganized discourse plans) and express them with comparative ease’. Where the linguistic genotype is normal, in short, as in aphasia, thoughts are normal; where it is severely shaken, as it may be in


285

schizophrenia, thought disintegrates. In the next section, we discuss this conjecture in the context of current linguistic and psychiatric perspectives on FTD.

8.5 Formal thought disorder and the de-integration of grammar There is little discussion of thought disorder in the philosophy of psychiatry, where discussions of schizophrenia have standardly focused on monothematic delusions such as the Capgras delusion, and very little reference is made to language altogether. Philosophical discussions have often focused on the status of delusional beliefs viewed as ‘propositional attitudes’ and as non-linguistic theoretical constructs (Campbell, 2001). Delusions appear to challenge ‘empiricist’ conceptions of thought and belief, and hence speak to the issue of empiricism and rationalism. Yet research on language has for many decades questioned this dichotomy. Utterances express, in a pretheoretical sense, attitudes, and these attitudes can have a propositional content, but such utterances are not typically ‘caused’, in any specified sense of ‘cause’, by anything in our experience. Utterances are typically creative, and, insofar as we know, uncaused. Propositional contents in turn need not be ‘objects’ of any ‘propositional attitudes’, when it is grammar that generates such contents from given lexical resources, and the meaning of such mental constructs is intrinsic to them. Campbell (2001) has interestingly broached the question of whether monothematic delusions involve a loss of an understanding of linguistic meaning. Yet Campbell’s notion of meaning is linguistically unspecific, and he makes no distinction between lexical and grammatical meaning. In landmark volumes such as Chung et al. (2007) or Kircher and David (2003), a linguistic perspective barely figures in a systematic fashion. It appears that a perspective on language as a condition on mental health is systematically missing. It is also absent more generally in phenomenological approaches to the mind, despite a strong emphasis on the distinctness of the kind of phenomenology that is language-mediated in Merleau-Ponty (2006: 206– 11), who essentially defended what we would call an Un-Cartesian position. One important theoretical perspective on schizophrenia (Sass and Parnass, 2003) characterizes its spectrum of symptoms under the unifying perspective of a ‘disturbance of the self ’ (ipseity disorder). Much evidence supports this perspective, yet language is absent in this discussion, too, where traditional notions from philosophical epistemology (‘subject’ and ‘object’, ‘self ’ and ‘other’, ‘subjective experience’, ‘first person perspective’, ‘meaning’, etc.) prevail and are used in non-linguistic senses. In philosophy, the spectrum of views taken on the nature of the self is remarkable for its systematic omission of language more generally—as if language simply didn’t matter to having the kind of self that we do, and was merely a way of talking about it. Discussions of the nature of the self usually begin as a departure from a view ascribed to Descartes, according to which selves are simple, immaterial substances. A second possibility is taken to be that we are indeed such a substance, but that we also have

286


physical properties (Lowe, 2006). A third is the alternative, famously proposed by Hume, that we are bundles of thoughts. Each of us is ‘a bundle or collection of different perceptions, which succeed each other with an inconceivable rapidity, and are in a perpetual flux and movement’ (Hume, 1737/1888: 252). A fourth is that we are our bodies. A fifth is that we are computer programmes, hence abstract objects, which can be ‘multiply realized’. A sixth is that we simply don’t exist: there are no selves (see Olson, 2007, for an overview). It is remarkable that in each and every one of these proposals language does not seem to matter to our existence: the existence of selves. Despite emphasis on the importance of the ‘first-person perspective’ (Zahavi, 2006; Lowe, 2006), the assumption appears to be the Cartesian one that such a perspective is somehow available independently of language—despite the fact that a grammatical distinction (‘1st Person’) is effectively written into them. Zahavi (2006: 27–9) stresses that self-awareness involved in conscious states cannot be construed along ‘subject-object’ lines, yet, as we have seen in Chapter 4, it is the grammar of self-reference that tells us exactly how this is the case: the grammar of 3rd-Person object reference (‘the/my self ’, ‘my body/brain’) is different from that of 1st-Person reference (‘I’; Bianchi, 2006). The grammatical distinction is essential: we would worry about an English-speaking child who never referred to itself in the grammatical 1st Person. That grammatical complexity increases with personal forms of reference, as discussed in Section 4.4, is not likely an accident, making grammar essential to the phenomenon itself. If schizophrenia, (i), is a disturbance of the self (Sass and Parnass, 2003), (ii) it involves a breakdown of grammatical principles in the organization of meaning, and (iii) self-reference is mediated grammatically, then a non-linguistic perspective on the self is challenged. Another non-linguistic approach to FTD is what we may call the ‘pragmatic’ theory. It explains FTD as a failure to master principles of rational communication, more specifically (i) to make proper use of ‘context’ (Kuperberg et al., 2006), (ii) to take the needs of a listener (knowledge, beliefs, and intentions) properly into account (Frith, 1992; Abu-Akel, 1999; Langdon et al., 2002; Tenyi et al., 2002), or (iii) to master ‘cohesive ties’ in discourse (Rochester and Martin, 1979). What must be intended here, however, is largely linguistic communication, as opposed to communication as such. Communication in a linguistic and specifically grammatical format makes a fundamental difference (Tomasello, 2008), entailing that the term ‘communication disturbance’ as such makes no predictions. If ‘context’ is used so as to include linguistic context (‘co-text’), grammar is systematically involved in the generation of context and is likely to be impaired when context processing is. Indeed, grammar will be the mechanism that defines context: only in virtue of grammar organizing semantic information, will there be aspects of meaning influenced by context (as opposed to grammar). Where grammar is impaired, the very distinction between what belongs to language and what belongs to context may cease to be meaningful for subjects: everything becomes context. The ‘pragmatic’ explanation is therefore


287

premised on grammar and semantics being intact, which however is the crucial question at issue. Do thought disordered schizophrenics really think normal thoughts and extract linguistic meaning in normal ways, while only a failure to communicate makes them ill? Successful linguistic communication also surely involves taking the needs of listeners into account and mentalizing about what they know and want. But ToM deficits appear to be very different in autism and schizophrenia (Bruene, 2005), and in the latter they do not seem to correlate with FTD in particular (McKenna and Oh, 2005: 121–2). To the extent that ToM, which, we have argued, comprises a number of very different phenomena, depends on language, it will fail to represent an alternative to grammar: explicit thinking about the mental states of other people (independently of perception/visual stimuli) requires a certain format, which may have to recapitulate linguistic distinctions, as in the representation [S1 He believes [S2 the puppet is in the box]], where a propositional thought embeds another, recapitulating the pattern of sentential recursion. S1 here moreover expresses the specific thought that some third person (‘he’) thinks a thought S2, which the 1st Person (the speaker) also thinks, while at the same time the truth values assigned to it by both persons may differ. These person distinctions are not merely semantic in nature, as argued in Section 4.4, but inherently grammatical: a person referred to as ‘I’ (1st Person) and as ‘you’ (2nd Person) may be the same person, semantically. But when that person stands in front of a mirror and mutters to himself: ‘I hate you’, this person would typically not say ‘I hate me’, let alone ‘I hate him’, illustrating that the grammatical Person system is crucial to understanding the thought expressed. Deictically anchoring properties and thought contents is critical to the identity of the recursive thought above in a similar fashion. This raises questions for how a generic ‘ToM-module’ can make the right predictions, as the relevant grammatical distinctions are not involved in this case. Where concepts of mental states are mastered, but they cannot be indexed to particular grammatical persons at particular times, we predict what we find in FTD: an over-attribution of mental states (‘hyper-ToM’: Bruene, 2005: 4). Schneiderian verbal auditory hallucinations of thoughts spoken out loud or thought commentary, could then be analysed as a failure to ‘index’ thoughts and voices to the right ‘persons’, where the notion of ‘person’ is grammatical, and inherently makes reference to speech acts. It is crucial in this regard that the use of pronouns, which are paradigmatically marked for Person, is error-prone in schizophrenia, as noted. Evidence from functional MRI also suggests a close relationship between grammar and ToM by showing that brain areas for language comprehension and mental state understanding partially overlap. For example, a functional connectivity analysis of the left posterior middle temporal region shows that this core area for sentence comprehension is strongly connected with posterior superior temporal sulcus (STS) and medial frontal cortex (MFC), considered as core ToM areas (Turken and

288


Dronkers, 2011; Amodio and Frith, 2006). Ferstl and von Cramon (2002) also underline that the MFC is involved when two different sentences are related—an aspect of grammatical meaning—as well as during ToM tasks, suggesting a connection between grammatical processing in a discourse context and the representation of mental states. The MF region shows decreased recruitment in schizophrenic patients during story comprehension (Dollfus et al., 2008). ‘Cohesion in discourse’, too, systematically depends on understanding referentiality in noun phrases, which imposes grammatical constraints on such cohesion, and clausal reference, as well as topics, in line with grammatical principles (Erteschik-Shir, 2007). Unlike the pragmatic theory just discussed, the ‘expressive semantic’ theory of McKenna and Oh (2005) maintains that, if one wished to construe FTD as a language abnormality, ‘it would be one which was characterized chiefly by semantic errors in expressed speech’. Consistent with a modular conception, it is by contrast often contended that ‘at the level of syntactic processing, schizophrenic patients’ speech is usually normal, with no relevant aberrations’ (Marini et al., 2008: 145). Yet we again need to ask what exactly ‘syntactic’ and ‘semantic’ are here taken to mean. We have argued that meaning as organized at a grammatical level is different in kind from meaning organized at a lexical level: for example, judgements of truth and falsehood cannot be encoded structurally in single words, even if these are internally complex (Di Sciullo, 2005); neither can the different types of nominal reference we have discussed in Chapter 4. The label ‘semantics’ thus captures a heterogeneous domain, with grammar crucial to one kind of meaning but not another. If so, syntax and semantics can only partially be independent. As Chomsky noted in 1965: the ‘boundary separating syntax and semantics (if there is one)’ should ‘remain open until these fields are much better understood’ (Chomsky, 1965: 77, 159). A sentence like ‘The man drank the guitar’, typically said to contain a ‘semantic’ violation, clearly comes with a fully determinate and unambiguous meaning, even if no one would have or accept such a thought. This very meaning is what creates a conflict with non-linguistic expectations in healthy individuals and is needed to explain the conflict. As Nieuwland and van Berkum (2006) show, moreover, the right context will make a putative ‘semantic’ violation such as ‘The peanut was in love’ perfectly well formed, showing that semantics is not at fault and grammar in conjunction with the lexicon rigorously rules what the meaning is. In a similar way, a violation typically said to be ‘syntactic’, such as ‘The man drank the with’, is always also a semantic violation, since no coherent meaning results when the grammar is not right. Where a so-called ‘syntactically’ disorganized outcome is found in schizophrenic speech, then (e.g. ‘People God to ought, God . . . heaven my will point no all’ in response to ‘Why do you think some people believe in God?’), then this ipso facto entails that no coherent meaning is produced either, indicating a potential semantic problem. In turn, where a speech outcome looks syntactically normal, at least at the single-sentence level (‘Oh, it [life in a hospital] was superb, you


289

know, the trains broke, and the pond fell in the front doorway’: cf. Oh et al., 2002: 235), the question arises: ‘do these words mean anything?’ (Covington et al., 2005: 92). Maybe the words do, however, but the way they are connected grammatically does not: there is meaning at a sentence-level, too, and the meaning arises from how words are connected in grammar, in a process that yields subjects, predicates and specific configurations required to encode a coherent thought about the world that can be properly assessed for truth and falsehood. Therefore, if these sentences do not mean for the patient what they mean for us—which seems plausible indeed—what may well be impaired in them is how meaning is organized by grammar. Indeed, the expressive ‘semantic’ anomalies that are found to be strongly associated with the presence of thought disorder are taken to primarily ‘affect the assembly of meanings at a level above that of the single word’ (Oh et al., 2002: 243), representing a form of what the authors call ‘structural semantics’. If so, they precisely affect meaning generated at a syntactic level—grammatical meanings, in our terms—and could therefore inform our grasp of the organization of grammar as such, raising interesting and novel questions of how the grammar affects meaning and what kind of meaning this is: propositional meaning, we hypothesize here. This conclusion would be consistent with a characteristic reduction of syntactic complexity (recursion) often reported for schizophrenia (e.g. DeLisi, 2001); with the fact that FTD does not affect naming (McKenna and Oh, 2005: 181), with Walenski et al.’s (2010) finding, in a study of grammatical Tense, ‘that thought disorder may be linked with grammatical impairments’ (Walenski et al., 2010: 262), ‘unrelated to general cognitive dysfunction’; and with a recent hypothesis according to which ‘building a gestalt meaning through language requires us not only to match input with prior-stored associations, but also to combine words together (through syntactic and semantic mechanisms) in order to construct the meaning of individual clauses’ (Ditman and Kuperberg, 2010). These two processes are fundamentally different in character: the former is associative, the other is computational, and requires grammar. Indeed they can conflict, as in the sentence ‘Every day at breakfast the eggs would eat’, where all words are lexico-semantically well related, but the result of a combinatorial analysis based on grammar yields a very implausible propositional meaning. Kuperberg argues that language abnormalities in schizophrenia arise from an imbalance between these two processes: where grammar-based computation of meaning is impaired, lexico-semantic associative processes win the upper hand, yielding the ‘loosening of associations’ so characteristic of schizophrenia. As we might put this claim: the grammatical straightjacket, insofar as it governs what is claimed to be true about a particular topic in discourse, or what is being referred to by a particular nominal phrase, is thrown off. Meaning based on the grammatical analysis of clauses is effectively ‘grammatical meaning’ in our present sense. An interesting recent study of Titone et al. (2007), using a paradigm already mentioned in Section 2.5, can serve to illustrate this perspective (though it is not

290


Titone et al.’s). The authors studied noun-noun combinations such as ‘zebra bag’, a process that they characterize as ‘conceptual combination’. In fact, as noted in Chapter 2, it is words that are being combined here, and what is, most specifically, combined, is nouns. Two kinds of interpretations were studied for these combinations: the ‘property interpretation’, where the modifying noun has a salient property (such as having stripes), which is then attributed to the head noun, resulting in the concept of a bag with zebra stripes; and the ‘relational interpretation’, where ‘zebra’ would signify a functional role of the head noun, like its being carried by some particular animate agent. The interpretation generated is then: ‘bag for/carried by a zebra’. It is known since Estes and Glucksberg (2000) that property interpretations are most frequent when the modifying noun has a salient dimension (like having stripes) that fit one dimension of the head noun, as in the case of the ‘zebra bag’. They become more rare where the modifying noun has such a dimension but it fits no dimension of the head noun naturally (e.g. ‘zebra trap’). They become even rarer when the modifying has no salient dimension relevant to the head noun (e.g. ‘donkey trap’). Now, it is clear that property interpretations are easier to obtain, via a basically semantic strategy: ‘property interpretations elicited by these types of combinations require only semantic access to the meanings of the modifier and head nouns’ (Titone et al., 2007: 95). Put differently, property interpretations come close to what would indeed be purely conceptual combinations, without much mediation by grammar (adjuncts, indeed, exhibit little grammar). Relational interpretations are different in this respect. Titone et al. assume they require a more intricate ‘postretrieval selection process’: ‘semantic access of word meanings is necessary but not sufficient for generating relational interpretations’ (Titone et al., 2007: 96). However, it appears more appropriate to express this fact by saying that they require more grammar. For the control mechanism behind what the authors term the ‘controlled use of semantic memory representations’ is just that: grammar (at least in this case). Indeed this is clear in the very grammatical complexity of ways of expressing these meanings: ‘bag carried by a donkey’, ‘bag for a donkey’, ‘bag with a donkey in it’, etc. Here the head noun is not a modifier but an argument, playing a thematic role, for example as Agent or Recipient. A broad notion of ‘conceptual combination’ does not respect this difference. Titone et al.’s experiment, thus interpreted, is an experiment in grammatical meaning, not merely in conceptual combination. The aim of the study was to test whether patients with schizophrenia have intact lexical-semantic retrieval, while only post-retrieval selection, as required for relational interpretations, is impaired. This would predict that these patients could generate property interpretations, but have difficulties generating relational interpretations. Three groups were tested: patients with schizophrenia exhibiting high levels of thought disorder on a clinical scale, patients with low levels of thought disorder, and controls. The results were: the integrity of and initial access to lexical-semantic


291

memory is spared in schizophrenia. None of the groups differed in how noun-noun combinations designed to elicit property interpretations were processed. By contrast, the controlled processing needed to generate relational interpretations is impaired in both groups diagnosed with schizophrenia, with patients with high levels of FTD producing significantly fewer such interpretations than patients with low levels. Unlike the authors, we interpret this as a strong pointer to the view of FTD as a disintegration of grammar, in which grammatical meanings get gradually out of reach of the patient’s mind. The grammatical perspective on FTD emerging here also makes sense of many independent findings about the schizophrenic brain. Studies in structural and functional MRI have shown that severity of thought disorder is ‘integrally related to the abnormal neural activity observed in language-related networks’ (Borofsky et al., 2010; Shenton et al., 1992). Aberrant functional connectivity and activations within cortical language circuitry are a consistent finding in both patients and subjects at genetic high-risk (Kircher et al., 2001; 2005; Li et al., 2010). Of particular significance is the left-lateralization of language in the normal human brain (Vigneau et al., 2011), and the reduction of cerebral asymmetries (de-lateralization) in patients with schizophrenia compared to controls (Dollfus et al., 2005; Li et al., 2007; Angrilli et al., 2009). This provides an independent argument that schizophrenia and language are related. An argument from migration patterns of modern Homo sapiens further suggests that both language and schizophrenia must have biologically stabilized in our species by around 40,000 years ago, when our species reached Australia, where we do not see any further development of our basic cognitive or linguistic abilities. Neither has language, as we have concluded in Chapter 7, originated long before the speciation of Homo sapiens (Crow, 2008). Language and schizophrenia, which are both confined to our species and universal in it, may thus be intrinsically related, like two sides of one coin: schizophrenia may reflect a failure of left-hemispheric dominance for language (Miyata et al., 2012), turning it into a by-product of the evolution of language itself— or its ‘price’ (DeLisi, 2001; Crow, 2000; 2002a, b; 2008). With this possibility, we can now pull all strands of discussion in this and the previous chapters together, against the background of the Un-Cartesian project. We need to account for a cognitive discontinuity in human evolution, which is speciesdefining and linked to language. Only at a point in evolutionary time when it is widely assumed that full language had arrived, do we also see a new mode of cognition and thought manifest that generates a recognizably modern culture unprecedented in the evolutionary record or in any other species at this point. The UnCartesian view is that language and thought, in our species, form a unity, with the specific mode of thought expressed in language inextricable from language. We now have an additional argument that this may be the right conclusion: language may be more central, not merely to thought, but also to mental health, than is commonly assumed. In defining a new mode of thought, the language faculty also defines a

292


new notion of what it is to be rational, distinct from notions of rationality that we can apply to animals, where such notions will typically connect with a notion of adaptive behaviour. This possibility is less likely in the human case, where the adaptive advantage of a notion of truth is not, to put it mildly, obvious. Why, in fact, should we be thinking in the first place? We turn to rationality, and how linguistics is linked with it, in the next and final chapter. We close this chapter with some notes on what the above suggests in terms of the implementation of language in the brain.

8.6 Grammatical meaning in the brain6 The most pressing question must be where grammatical meaning is located in the brain. As noted in Hinzen and Poeppel (2011), several decades of neuroimaging the human brain have not only taken modular conceptions of language on board, but also modular conceptions of the internal architecture of language. Thus decades of investigations of looked for ‘syntax’ and ‘semantics’ in the brain, as if these are separable constructs. As Pylkkänen (2011) notes, cases where sentence-level semantics (grammatical meaning) is considered and hence semantics is not simply equated with lexical semantics, are rare. Pylkkänen herself reconstructs the notion of sentence-meaning compositionally, hence as a composition of given lexical meanings, but this is to assume the descriptive framework of standard formal semantics, which was designed as a formal (nonpsychological or neurophysiological, perhaps even non-empirical) account of language-world relations in the tradition of formal logic, where language and world relate as ‘form’ and ‘content’ do (or ‘syntax’ and ‘semantics’): hence language is not involved in the genesis of new content, or a new kind of meaning. None of this is surprising when it is clear that linguistic theory itself has, equally for decades, stressed that ‘syntax’ and ‘semantics’ are axiomatically distinct, and the boundary between them has defined kinds of linguists and even drawn disciplinary boundaries, with philosophers of language almost exclusively focused on ‘semantics’ and logic, as if grammar simply didn’t much matter to it. In addition to this, early studies of language in the brain were strongly relying on the lesion method, which suggested particular loci in the brain as regions in which particular linguistic functions can be localized, with Broca’s region (Brodmann’s areas 44 and 45) associated with ‘syntax’ and Wernicke’s area with ‘semantics’. As we have argued above, the modular separation of ‘syntax’ and ‘semantics’ borders on incoherence: every single sentence containing a syntactic violation is semantically deviant; and every grammatical expression is ipso facto also meaningful, as the famous sentence Colourless green ideas sleep furiously illustrates, which contrasts with Furiously sleep ideas green colourless not merely in grammaticality but also in meaning. The interpretation of such a stimulus as carrying a particular meaning, moreover, is a reflexlike operation, which we cannot consciously control. We cannot decide to perceive a language that we speak as ‘pure form’. That the brain 6 The following section would not have been possible without the help of Nathalie Tzourio-Mazoyer, who has been an unfailing source of advice (but not to blame for any errors in what follows).


293

should compute form autonomously from meaning would thus a priori not be expected. The same problem in severing grammar from meaning can be illustrated with thought disorder, if we are right above that it has an important grammatical dimension. If there is such a thing as grammatical meaning—meaning intrinsically depending on grammatical organization—it couldn’t logically be that there is a ‘syntax’ region that was not also a ‘semantics’ region (though the reverse is not true, to the extent that lexical meaning is grammar-free). Consistent with this, 14 out of 16 neuroimaging studies on semantic aspects of sentence comprehension reviewed in Hagoort et al. (2009) found reliable activation of left inferior frontal cortex including BA 44 and 45. A study of Rodd et al. showed that the same areas are sensitive to ambiguous words processed in sentential contexts (e.g. ‘There were dates and pears in the kitchen bowl’ vs ‘There was beer and cider on the kitchen shelf ’). Willems et al. (2009) argue that the regions in question as well as the posterior part of the superior temporal sulcus are multimodal areas involved in the integration of semantic information from language, gesture, pantomime, and pictures, during sentence processing. Aboitiz et al. (2006) relegate the syntactic functions of Broca’s region to working memory and not to language-specific mechanisms, as do Fadiga et al. (2009), who see Broca’s region as subserving complex hierarchical dependencies in general, as involved in language, music, and the execution and understanding of action. If so, Broca’s region may inform us about the neural basis of ‘syntax’ in the generic sense that we have tried to avoid here, but tell us little about grammar in the language-specific sense that we target here. Against such views as Aboitiz et al.’s, Grodzinsky and Santi (2008) argue that two distinct intra-sentential dependency relations activate different brain regions, and only one activates Broca’s area: syntactic movement (e.g. Which boy does the girl think t examined Steven?). The other dependency is antecedent-reflexive binding (e.g. Jill thinks the boy examined himself). But while these may indeed differ syntactically, they differ semantically as well. Grodzinsky and Santi’s point about syntactic specificity in Broca’s region is also not meant to be inconsistent with semantic processes being located there too. A ‘syntacto-centric’ perspective dissociating semantics and syntax in the brain has also come under critical scrutiny since Kuperberg (2007), who argues that the brain does not strictly ‘respect a distinction between semantic and syntactic representation and processes’. Vigneau et al. (2006), in a large-scale meta-analysis of a decade of neuroimaging studies of language processing, found, unlike what was observed for phonology and semantics, that semantic and syntactic brain areas appeared intermingled, supporting a theoretical shift to viewing ‘cognition in the brain as supported by overlapping and interactive large-scale cognitive networks rather than modules’. Another study (Röder et al., 2002) comparing sentences composed of words and pseudo-words also underlined a substantial spatial functional overlap of semantic

294


and syntactic language areas, questioning the functional–anatomic link between syntax/Broca’s area and semantics/Wernicke’s area. Recent functional connectivity studies of brain activity in the conscious, resting state (‘day-dreaming’) have also provided evidence for a language network that goes beyond the traditional superior temporal (Wernicke’s) regions (for comprehension) and inferior frontal (Broca’s) regions (for production), including also adjoining prefrontal, temporal and parietal regions as well as bilateral caudate and left putamen/globus pallidus and subthalamic nucleus (Tomasi and Volkow, 2012). Brain activity in this state involves ceaseless conscious activity that is interrupted and inhibited by task demands (e.g. perception, which ‘interrupts the train of thought’), and appears to in part involve the same left-lateralized conceptual processes that are mediated by language: in the words of Binder et al. (1999), ‘semantic knowledge retrieval, representation in awareness, and directed manipulation of represented knowledge for organization, problem-solving, and planning’ (see also Ferstl et al., 2002; 2008; and Turken and Dronkers, 2011). While the networks subserving grammatical meaning remain to be isolated, it may be that the networks involved will include in particular the frontal medial area, which is part of the extended language network. It would be the neural correlate of the interface between grammar, the expression of thoughts, their understanding, and the notion of self. Following the hypothesis of Crow that language symptoms in schizophrenia are related to decrease in leftward lateralization (Crow, 2002b; Angrilli et al., 2009) and investigating the reproducibility of fMRI (funcational magnetic resonance imaging) in patients compared to healthy volunteers two years apart, Mazoyer et al. (1993) applied a paradigm comparing story listening in the mother tongue (involving deception and ToM-related material) to story listening in an unknown, foreign language, known to be sensitive in detecting higher order language comprehension areas. In healthy controls, the task activated a large left-hemispheric network for the meaningful stories, which is reduced to a sparse and weakly lateralized set when meaningless sentences are processed. In support of Crow’s hypothesis, lower leftward asymmetries were observed in schizophrenic patients (Dollfus et al., 2005), while lower reproducibility was present in the left posterior part of the superior temporal sulcus, left inferior frontal gyrus, and left posterior middle frontal gyrus (Maïza et al., 2010). This increase in variability was dependent on performances in patients only, who were paired with controls for age, gender, handedness and educational level. The regions showing more variation in activity and reproducibility when patients processed the stories had been shown earlier to increase their activity from word to sentence and text, independently of the modality, in healthy volunteers (Jobard et al., 2007). It is possible, therefore, that these areas provide neural support of grammatical meaning, and that modifications of activity in these areas in patients that have lower comprehension of the texts correspond to a deficit in grammatical meaning in these patients. This illustrates that, in the search for the biological foundations of language


295

in the brain, the notion of grammatical meaning, contradicting a traditional syntaxsemantics split, may well be able to play an important role, and enough pointers suggest that this will be a fruitful avenue to explore.

8.7 Conclusions As we vary the linguistic genotype, different linguistic phenotypes arise that define different kinds of mind. In all of the cases we have considered, language remains a plausible ‘measure of mind’—mirroring and allowing us to map its capabilities and incapabilities. The basis of grammar on our account is referring, but referring is not merely something semantic. It is the expression of the deictic intentions of a creature that is also rational. We refer for a reason. Our species is thus not only rational in the sense that it thinks, and that it has a concept of thought, and hence has a concept of truth. It is also rational in the sense that its linguistic capacity defines a standard of mental health. Language, that is, may not merely organize our mind. Rather, when this mind becomes disorganized or irrational, that disorganization appears inseparable from a disorganization of language. This is what the Un-Cartesian hypothesis directly predicts. The disorganization of language particularly in its referentialdeictic aspects may well account for core features of the thought disorder syndrome. One could explore alternatives, such as a disturbed ToM module, yet as a theoretical construct, the latter seems ill-defined, comprising a range of different phenomena that form no natural unity: some are perceptual and undoubtedly language-independent and not uniquely human; others involve thought and concepts and seem to depend on language. And viewed in isolation, ToM makes no predictions for propositionality and truth, which is what we need to explain, and which we argue we can explain if we regard them as grammatical in nature. The evolution of language, then, may well be the evolution of thought, and thought and the linguistic genotype may fragment together. This brings us to our next and final chapter, the most philosophical one of this book: grammar cannot define thought, if it does not also define truth. It is one thing to argue, as we have, that the forms of (pro-)nominal and clausal reference available in this species are grammatically configured: this, in traditional terms, is a ‘syntactic’ story. It is another thing to argue that beyond this syntactic story, no additional semantic story is required: that grammatical meaning is just that, grammatical, and nothing other than grammar grounds it.

9 Thought, language, and reality This chapter addresses the philosophical challenge just posed—the challenge of how, having taken traditional semantic notions out of the context of ontology and metaphysics and integrated them into grammar, to tell in what way they still connect with the intuitions associated with these semantic notions. Thought, grammar, and reality have to be woven into a single coherent scheme. As we saw in Chapter 1, they have very much been part of such a scheme in Ancient India, in Aristotle, and throughout the Middle Ages, with the Modists adding a particularly sophisticated version of this idea, in which grammar is the world as known: experience converted into knowledge, or knowledge in a uniquely linguistic format. In Cartesian rationalism, this single scheme got broken (Leiss, 2009): language, more or less deficiently, mirrors the independently given structure of rational thought, but not of reality. How close can the conception we have now developed be to the Modistic and ancient ones? We will begin with some notes about the status of semantics within philosophy and linguistics, and the traditional question of whether semantics is ‘mathematics or psychology’ (Partee, 1980)—an alternative that leaves out grammar, or the possibility that meaning may, in part, be inherently linguistic. We then address the question head-on of how it can be that truth comes out as a notion of grammatical semantics. Our answer, in Section 9.2, will appeal to Tarski’s Equivalence Scheme, which we will claim to grammatically derive. Section 9.3 provides a unified account of truth-attribution that comprises both lexical and non-lexical versions of this phenomenon, and explains their relationship. Section 9.4 locates the phenomenon of truth-attribution within the overall architecture of grammar, and discusses evidence that it is, in fact, the limit of grammar, which, at the same time, also appears to be the limit of thought. With this account of the relation of grammar and truth in place, we return to the metaphysical question just raised, in Section 9.5: If truth is a grammatical notion, what should we conclude for the connection between language and reality? Finally, in Section 9.6, we close this book with a reflection on where this leaves us with regards to prospects for a science of thought, and whether Friedrich Max Müller (1887/1909) was right when he proclaimed that any such science is identical to the science of language. As we will see, a prominent tradition in philosophy maintains that meaning and the mental are not subject to a natural science, which is said to be in part because they are inherently

9. Thought, language, and reality

297

normative, and no naturalistic investigation can bring out what makes this so. We will by contrast conclude that, in essence, our linguistic mind as naturalistically described is the only thing that grounds reason and norms in this sense.

9.1 The nature of semantics How we think of the discipline of semantics today is, unavoidably, in part a consequence of its contingent intellectual history. Meaning, perhaps more than any other aspect of natural language, has always caused philosophical problems, and semantics got shaped in part in response to these problems. Much of contemporary linguistic semantics has its intellectual roots in the nineteenth-century analytical philosophical tradition, although the interests of the latter have been quite different from the interests of most linguists today: they primarily related to concerns in philosophical logic, logicism, epistemology, and the philosophy of science. In the first half of the twentieth century, many of these concerns remained, but they also got transformed. A chief concern was the epistemological and metaphysical problem of realism, or with how our minds really ‘connect’ with the world. This is one rationale for why the primary issue in semantics was reference, in the sense of a semantic relation to a mind- and language-independent world, and why questions about putative ‘reference to non-existents’ as in the case of ‘the king of France’ or ‘Pegasus’ have been incessantly discussed in philosophy ever since Russell (1903) (see Lycan, 2008, for an overview). In addition to a concern with realism, however, another core commitment emerged, which was equally metaphysical in nature: naturalism. Spearheaded by Willem van Ormen Quine, ‘naturalizing philosophy’ would become one of the most central philosophical projects in the second half of the twentieth century. It in turn entailed naturalizing semantics. The relevant notion of naturalism, here, is crucially that of metaphysical naturalism, which is, in essence, the same as contemporary physicalism in the philosophy of mind, and is crucially distinct from the methodological naturalism of Chomsky (2000a). Metaphysical naturalism is an ontological view, based on an a priori and fixed distinction between ‘physical’ entities and ‘non-physical’ ones. A further commitment to physicalism then leads to a problem with regard to meanings, since these are thought to fall on the non-physical side, and non-physical entities cannot be ‘first-class’ citizens of the world when a physicalist ontology is assumed. This in turn leads to a number of familiar metaphysical solutions, such as that non-physical entities (including meanings) must be non-existent or fictional, that they need to be reduced to physical entities, must supervene on them, or be identical with them. The problem this has created for a naturalized semantics has again motivated a primary concern with reference in the semantic sense, and a theoretical bias in favour of viewing meaning extensionally in terms of causal or perhaps conventional ‘symbol-world’ relation unmediated by concepts or the mind. Symbols, the thought

298


was (and is), are physically individuated objects, and so are the objects that such symbols ‘stand for’. If meaning is reference, and reference is causal, the metaphysical problem of meanings seems to disappear (at least until intensionality effects start causing problems again). It is clear from the above that the notion of reference in question is a technical one, defined by specific theoretical concerns in a given philosophical context. It is not primarily rooted in the empirical study of language, or motivated by insights from the latter. Nonetheless, it is thought to apply to ordinary language use, and hence it is vulnerable to empirical scrutiny. Evoking empirical features of such language use, Chomsky (2000a) argues that the technical notion of reference in question does not, in fact, apply to language viewed as a natural object and from a biolinguistic point of view. Whether this is right or not, it illustrates that an agenda motivated by metaphysical naturalism does not necessarily reap empirical benefits. A referencetheoretic theory of meaning may make sense from the viewpoint of physicalism as a metaphysical doctrine, but not actually yield an object of naturalistic inquiry that can define a research programme. In this book, we have assumed that how humans use words to refer to the world is a crucial object of naturalistic inquiry. But this inquiry has made no use of a semantic relation of reference between words and things: indeed, we have emphatically stated that there is no such relation, at a lexical level (although there is lexical content at that level), and that, at the grammatical level, the organizational principles behind reference are not semantic but grammatical. In general, metaphysical naturalism has made it hard to see how meanings could be at least in part mediated by language or the mind, and how meanings of this kind could be subjected to naturalistic inquiry. In the wake of the cognitive revolution, philosophers developed the framework of the Computational-Representational Theory of Mind (CRTM: Lycan, 2003; Rey, 2003), which made room, in the head, for ‘mental representations’. But these are not linguistically specific entities: they are representations, with the meanings they represent again outside both language and the head or mind. The move in question is consistent with metaphysical naturalism, and indeed motivated by it: such representations would again only be symbols, viewed in turn as physical entities, which means that they can enter causally into the genesis of behaviour. If we further define computations over such physical symbols, and let these track environmental relations between contents of such symbols, thinkers will become ‘physical symbol systems’, which pose no metaphysical problems. There can be thinking in physical nature, and such thought can be studied computationally (see Fodor, 2000: ch. 1). Yet again, it is not clear how an actual object of naturalistic inquiry arises from this move: there is nothing in it that makes us understand why some of these mental representations have propositional meanings, for example, and others not; or why we seem to think different kinds of thoughts, at least in part, from other species. It doesn’t answer the former question to say: mental representations that are propositional are those that denote extra-linguistic


299

propositions; or to say: because thought is propositional, some of these mental representations must have a propositional format. The CRTM was one philosophical account of how psychology could be science, following from an earlier such account: in the first half of the twentieth century, Watson (1913) and Skinner (1953) vowed to make ‘psychology scientific’, just as Bloomfield (1933) vowed to make ‘linguistics scientific’ (and free of psychology altogether). But none of these programmes led to what we would consider as naturalistic inquiry into language or the mind today (Chomsky, 1959). Neither has the ‘naturalized philosophy’ of Skinner’s close colleague Quine led to naturalistic inquiry into the mental: in its shape as eliminative materialism, there evidently can be no such inquiry; in its shape as functionalism (Lycan, 2003), such inquiry would be radically different from that of any other aspects of human biology; and in its shape as anomalous monism (Davidson, 2004), the mental is an ‘anomaly’ in physical nature and no natural object of inquiry either (Hinzen, 2006). Not only need a metaphysical, physicalist stance not yield a naturalistic research programme, however, but an explanatory approach to a particular kind of meaning pursued in the spirit of methodological naturalism may not be sensitive to such metaphysical constraints. Thus, Chomsky (2000a) denies even the coherence or formulability of physicalism. Contrary to the behaviourists, generative grammar, endorsing a methodological naturalism only, did not engage in the rhetoric of making psychology scientific and instead insisted that the proof of ‘naturalistic’ inquiry can only consist in pursuing it. Such inquiry, in order to proceed at all, will naturally define new objects of inquiry that lend themselves to naturalistic inquiry in a productive fashion (e.g. ‘I-language’). But it is not premised by a prior metaphysical view of what a natural object is or can be, or whether there are meanings or not. We might compare this way of proceeding with the philosophical worry expressed by Fodor (1990: 51) that, if intensional and semantic predicates were to ‘form a closed circle’, this would: preclude a physicalistic ontology for psychology since if psychological states were physical then there would surely be physicalistically specifiable sufficient conditions for their instantiation. But it’s arguable that if the ontology of psychology is not physicalistic, then there is no such science.

Intensional and semantic predicates, therefore, should not form a closed circle: they should lend themselves to physicalist reduction. If so, it makes sense, with regards to the discipline of semantics, to distinguish a physicalist ontology of form, posited internally to the organism—a system of structured mental representations defined through their formal-computational properties—from an ontology of content, which involves causal relations of such representations to things in the world (Fodor, 1990; 1998). Nothing different, effectively, was assumed in modern generative linguistics itself, which in practice focused on uninterpreted properties of form as well, viewed

300


as internal to the organism and subject to computational study, while leaving content to ‘semantics’ or denying that it would lend itself to naturalistic inquiry at all. Fodor’s above concern is shared by Davidson (2004: 147), who agrees that the ontology of psychology is physicalistic, and that psychological events can, one by one, be physically describable, yet who contends that no definitional or nomological reductions of mentalistic vocabulary to physicalistic vocabulary is on the cards. Unlike Fodor, therefore, Davidson concludes that the mental is only an issue of choice of vocabulary: insofar as nature is described in mentalistic terms, it is not up for reduction, and the mental is an ‘anomalous’ aspect of nature in this sense. At the level of ontology, however, it plays no role.1 In this sense, there cannot really be a ‘science of the mental’ viewed as an object of nature. Psychological descriptions will forever remain inherently normative, in a way that physicalistic vocabulary is said not to be (Davidson, 2004: 115). Even if we expel the mind from the naturalistic study of objects of nature, however, it will then again be true that we can study it formally: through formal semantics, decision theory, and logic, which do not study either the mind or language as natural objects. These indeed are the prime ingredients of Davidson’s ‘Unified Theory of Thought, Meaning and Action’ (Davidson, 2004: 151 ff.). We revisit this Unified Theory in the end of this chapter, arguing that it systematically eliminates grammar in the present sense from its theoretical scheme. A formal approach to semantics is thus consistent with both the reductionist perspective and that of anomalous monism, since it is defined so as to assume a categorical form-content distinction in place, and along with it a conception of ‘syntax’ where syntax is solely on the side of ‘form’, leaving out meaning. Syntax or form are what’s in the head on the reductionist conception or the CRTM more generally, while, from the viewpoint of the anomaly of the mental, they are merely part of a formal framework used to describe the mental. The categorical formcontent dichotomy is also behind the familiar contention that ‘syntax is not semantics’, used classically by Searle to reject the computer model of mind, and more recently by Davidson (2004: 132–3) to reject the philosophical significance of cognitive and linguistic theory. It is also consistent with the neo-Fregean contention that ‘semantics is not psychology’: semantics is outside the head/mind, which cannot have any influence on meaning. As these paradigms developed in the second half of the twentieth century, generative linguists of course did all the while regard their emerging field as a branch of psychology. The goal of linguistic theory, on this view, is to describe the state of knowledge in which native speakers are when they know the meanings of, and the well-formedness conditions on, linguistic expressions. Something in those speakers’ Cf. Davidson (2004: 114): ‘[i]n my view the mental is not an ontological but a conceptual category. ( . . . ) To say of an event, for example an intentional action, that it is mental is simply to say that we can describe it in a certain vocabulary—and the mark of that vocabulary is semantic intentionality.’ 1


301

heads is taken to determine what this knowledge is like. Generative linguistics is internalist inquiry in this sense. But then a conceptual tension with the status of semantics will automatically arise: for semantics, by Frege’s definition, is about relations between mind (or language) and world, and it is not psychology; and even for psychology in the sense of the CRTM, both extensions and intensions are mind-external. Semantics thus pulls the internalist linguist in an externalist direction, and a conceptual problem arises with what semantics could even mean, if it is to be a matter of psychology. One predictable reply has been to pull semantics back into the internalist domain (Jackendoff, 2002), leaving reference and truth either out of it (see also Horwich, 2010) or turning them into optional ingredients of a semantic theory (Pietroski, 2011). A predictable response of the philosophical semanticist has been that ‘semantics without reference’ is not semantics (Borg, 2012): even a ‘minimal’ semantics needs to include the technical philosophical notion of reference. In this book, of course, we have tried something different, circumventing this whole issue: we have just studied a phenomenon, intentional reference in ordinary language use, and we have done so grammatically, since it seems to arise in grammatical patterns and to be governed by grammatical principles. We have neither regarded grammar as a part of psychology or as ‘internal’, nor as ‘external’. We have engaged in no ontological reflections, in any metaphysical or extra-linguistic sense of ontology, at all. Another response to the problem was the initial one of formal semantics itself: semantics, much like syntax and pragmatics, was simply a formal science, a branch of mathematics (Montague, 1974). What it is, ontologically speaking, that we are formalizing, and how it relates to obscure matters of ‘mind’, then just doesn’t matter. In practice, however, linguists have not been content in taking this stance, trying rather to fit formal semantic models into a speaker’s head somehow, aligning the new formal discipline of semantics with their prior view of linguistics as a branch of psychology.2 But as Partee (1980) argues in a classical discussion of whether semantics is ‘mathematics or psychology’, the attempt to interpret the formal models of semantics as describing an aspect of psychology falters when it comes to the semantics of the propositional attitudes: what speakers mean and believe is not in line with what a model predicts that assumes propositions are functions from possible worlds to truth values. Any semantic theory that wants to be psychologically sensitive has to question the notion of a proposition involved in the semantic models concerned. The tension between the attempts to regard semantics as a formal science and as a part of psychology, analogously with linguistics, has thus remained. Partee’s problem, and the solutions later developed for it, assumes that the meanings of sentences are propositions and that there are ‘propositional attitudes’ More recently, several philosophers have made a similar move (King, 2007; Soames, 2010), like Hinzen (2006; 2007), while still not invoking grammar in an explanatory role, or indeed engaging in an explanatory project for what form the organization of meaning in the grammatical brain actually takes. 2

302


that are defined as ‘relations’ to them. These assumptions reflect a categorical thought-language divide, with the meaning of language conceived as the content of thought (Fodor, 2001), and a Cartesian ontology in place on which a proposition is a non-linguistic entity, while a ‘sentence’ is either deprived of a content or has it ‘derivatively’. Again, we do not see how propositions so conceived will lend themselves to naturalistic inquiry; nor does it become clear why they relate so intimately to language. So we have endorsed a different heuristic here, first suggested in Hinzen (2006): that what has been metaphysically characterized as ‘propositions’ is simply one element in a formal ontology that comprises other formal distinctions such as that between propositions and facts or that between facts and truths, and is systematically regulated grammatically. These ‘entities’, then, are not primitives of some posited semantic or metaphysical ontology (indeed they are formal categories, and hence not entities at all), nor do they belong to psychology. On this approach, a standard argument for propositions that we find in all textbooks of the philosophy of languages doesn’t get off the ground. The argument is that two sentences, from the same or from different languages, can have the same meaning or ‘express the same thought’; hence meaning must be something non-linguistic. Yet, clearly, to whatever (limited!) extent two sentences of the same language can have the same meaning, grammar will explain that: for example, theta-theory, a subcomponent of grammar, explains why John killed Bill and Bill was killed by John have the same meaning—to the (not complete!) extent that they do. Similarly, that a French and an English sentence can mean the same is explained by the independently established fact that grammar is independent of morphophonology: grammatical relations do not sound a particular way. The argument for propositions as metaphysical entities follows only if language is reduced to the morphophonological surface, deprived of meaning, which is circular when the view that language is deprived of meaning is to be established and a non-linguistic ontology of meanings is to be motivated (Hinzen, 2006; 2007). This heuristic is, as noted in Chapter 5, refutable, through evidence that the formal ontology of semantics does not co-vary with grammatical distinctions or map onto independently given grammatical patterns.3 With this grammatical perspective, the scene now changes: the question is not any more whether semantics is psychology or mathematics, or whether it is defined over a 3 See Reichard and Hinzen (2012) for some further case studies in the grammatical essence of what are claimed to be ontological of metaphysical distinctions. For example, Soames (2010: 102) makes a metaphysical distinction between ‘act-types’ and ‘propositions’. The latter are claimed to be ‘a certain kind of property’, unlike propositions. Yet, in all examples given, the two categories correspond closely to the grammatical one between Complementizer Phrases (CPs) (propositions) and gerundive constructions lacking an agent (act-types). Similar remarks hold for Moltmann’s (2007) distinction between ‘tropes’, which correspond to adjective nominalizations (e.g. ‘wisdom’), and ‘states’ (‘John’s being tired’), which are gerundive constructions. Grammar systematically co-varies with the distinctions made, and it arguably explains why we have the different intuitions about them that we do, which means that they do not have to remain metaphysical primitives in our semantic ontology—a clear advantage of this approach.


303

structure of mental representations or not. The question, rather, becomes whether semantics as a field has an independent formal ontology, with respect to the ontology that grammar yields. Such an ontology can, of course, be formalized within semantics, but formalizing it will not yield the answer to the question just raised. Positing a reference relation, too, is to formulate the problem rather than provide its solution: the problem of why humans, beginning from infant pointing, refer, in a way that no other species does. Our causal encounters with an environment make no predictions in this regard. Nor does it help to simply posit an ontology of thought that language is then said to ‘reflect’. Nor will it help to presuppose some formal meta-language and write up ‘logical forms’ in it that help to encode intuitions we have about the truth conditions of expressions. Since the logical forms are simply designed to formalize the intuitions in question, they do not answer the question of where these truth conditions come from—whether grammar induces them, or they have an independent origin. We may observe that representing proper names have a distinctive form of rigid or direct reference (Kripke, 1972); but representing them as constants in ‘logical form’ will not answer the question of why they exist or refer in the way they do: such a formal representation will be our problem, not the answer to our problem. Partee’s question of whether semantics is mathematics or psychology is thus characteristic for a possibility that it omits: that semantics might be linguistics, and part of grammar in particular. In part, of course, we have argued, it isn’t, for even percepts have a distinctive semantics, which we already do not know how to explain (Section 2.1). Lexicalization viewed as the creation of a lexical atom then comes with a semantics, too, as do lexemes and their inherent hierarchical organization in terms of hyponymy and hyperonymy. Grammatical patterns, too, come with a distinctive semantics of their own; and here is where things end. The ontology of our theory of the propositional forms of reference is grammatical. Calling it psychological adds nothing to our problem, and we do not see how we can go deeper than grammar, laying a foundation for grammar itself. Indeed, also with regards to calling it ‘internal’, we surely cannot disregard the fact that the sole place where grammar is found is in humans who communicate linguistically and at a population level. Grammar does not mature in the individual—it is, factually, only a population-level phenomenon, even if the members of such populations, as is surely the case, individually need a sapiensspecific brain and the linguistic genotype to partake in this phenomenon. For this reason, too, our account endorses no internalist ontology. It is language in the socialinteractive and population-level sense, where issues of reference arise: we do not point in the privacy of our own heads. Truth, too, is not a subjective phenomenon, or even an intersubjective one. All that said, there is a problem, with which the rest of this chapter will be concerned. It is one thing to provide evidence that it is a consequence of grammar that, as lexical concepts are inserted into a grammatical derivation, a formal ontology arises and the concepts in question are converted phase by phase into referential

304


expressions. Another thing is to say that when we talk about the formal ontology of truth, what we are talking about is indeed what the philosopher or semanticist takes truth to be: a semantic notion that intrinsically connects with issues of objectivity or the reality of the external world. Arguably, an account of truth that does not make sense of the objectivity of truth, is not an account of truth at all. Nor is it an account of thought. The notion of objectivity here need not be a controversial metaphysical one. The fact is much simpler: as we understand the notion of truth, truth is not appearance or likelihood; not belief or assertion; not presupposition. Truth is just that: truth, and being objective is part of what truth means. It is true that the inputs to the grammatical process are lexical items that we have assumed are ultimately grounded in percepts. Percepts are rooted in the external world that we act in. But perception, we have also assumed, cannot ground objectivity. So how can grammar do so? This is not the problem that grammar is ‘pure form’, and cannot speak to matters of ‘content’—for precisely such a distinction, which can be metaphysically or methodologically motivated, has not been made here. If a particular kind of form is intrinsic to a particular kind of content, no form-content dichotomy is coherent any more. So it is not as if we posited some forms, and the problem of how to connect these to independent and external matters of content is now still completely open. But the concern remains. For we have developed an account of sapiens-specific thought that makes such thought come out as intrinsically grammatical, and which also gives a grammatical explanation of the fact that such thought can be true or false. The question that remains is how it can be that truth involves a relation to the objective world, and yet be grammatical? Indeed, come to think of it, might we have cut the connection between thought and reality, supporting some kind of idealism in the end, on which truth is ‘mind-dependent’? Would this not be to contradict the very content of truth, in which case we have achieved little and a semantic story is yet to be added to our account?

9.2 How can it be that truth is a grammatical concept? Anyone can agree that the connection between truth and grammar is very close in the following sense: lexemes cannot be evaluated yet as true or false, nor can even many grammatical configurations: the grammatical correlates of truth are in fact quite specific, as we will see again below. The insight was first formulated by Aristotle, who noted that ‘falsity and truth have to do with combination and separation’ (De Interpretatione 1, 16a, 13). But the combination and separation in question is not simply a generic one: they have to do with grammar—which is what combination and separation means in this case.4 Truth attribution, in this sense, is a structural 4

For evidence that this was, indeed, Aristotle’s view, see Reichard (2013), Section 3.1.


305

phenomenon, and more specifically, a grammatical one, as the prime ingredients that have to be combined and separated in appropriate ways are grammatical ones: for Aristotle, a subject, a predicate, and finite Tense, occurring in the right configuration.5 Where Tense is missing, in particular, as in the small clause (SC) in John considers [SC him [a man]], no truth is attributed by either speaker or hearer to the predication encoded in this SC (i.e. him being a man). Moreover, it is ungrammatical to say: ‘him a man is true’; or ‘it is true that him a man’. So the notion of truth cannot be applied. Where Tense is added, but remains non-finite, as in John considers [SC him to be [a man]], the same remains true. Even where Tense is present and finite, no truth needs to be attributed to the proposition of which it is a part, as seen in John believes [CP he is [a man]], where no such truth as that he is a man is asserted by the speaker. Yet, specific as the grammatical preconditions for truth may be, this says absolutely nothing about what being true means; or when a given proposition is true; or what conditions have to be satisfied for this to be so. For millennia, philosophers and metaphysicians have tried to specify these conditions. Thus they ventured the proposals like that the world has to be a certain way, namely so as the proposition says it is: what the proposition says is the case, also has to be the case; the proposition has to mirror or correspond to reality. This all sounds innocuous enough, if only we knew how to tell what ‘being the case’, ‘mirroring’, or ‘corresponding’ means, without circularly using the notion of truth to elucidate these notions. This asks for a genuine ‘theory’ of truth—a theory that tells us, in terms other than truth itself, what truth is. But a widespread consensus today maintains that there probably is no such theory (Horwich, 2010). It wouldn’t follow from this that truth does not exist. But it would suggest that it is a primitive: it cannot be analysed, or broken down into constituents on the basis of which we could construct a theory of it. Even so, this would not be the end of the theory of truth, since Alfred Tarski had a profound insight in this regard. Rather than further pondering the content of truth, he wondered, why not reduce our statement of the conditions that have to hold for a given proposition to be true to instances of the following schema, where ‘P’ denotes an arbitrary sentence that is referred to on the left-hand side (LHS) of the equivalence in (1) and used on the right-hand side (RHS): (1)

The Equivalence Schema: ‘P’ is true iff P.

For example:

Which is not to say that the function of finite Tense in English, which is the deictic anchoring of sentences, is implemented in the same way in all languages. Arguably, such anchoring in Halkomelem (Salish) is on location rather than time, and it is on Person in Blackfoot (Ritter and Wiltschko, 2009). 5

306 (2)

The Philosophy of Universal Grammar ‘Snow is white’ is true iff snow is white.

The profundity of (1) lies in the fact that (as long as ‘P’ does not contain any lexical truth predicate itself ), there is a lexical truth predicate present on the LHS, but none on the RHS, while, at the same time, the two sides appear logically equivalent. Ordinary speakers, that is, will typically agree that (2) is the case: the LHS and RHS seem to have the same semantic content, whether we spell this out in terms of both sides having the same truth conditional content, or inferentially, through two seemingly valid (truth-preserving) inferences: ‘Snow is white’ is true ∴ Snow is white Snow is white ∴ ‘Snow is white’ is true The natural conclusion from these observations is what we independently know: the lexical truth predicate (as present on the LHS) is not in fact required to state a truth, or to predicate truth of a proposition.6 Thus, a speaker who utters the sentence (3) with normal intonation in the course of a conversation, will normally be taken to have claimed the truth of the proposition that John left:7 (3)

John left.

Such a speaker would normally be taken to be contradicted by another person uttering any of the sentences in (4), but crucially not by a speaker uttering either sentence in (5): (4)

a. No, this is not true. b. That John left is not true.

(5)

a. That John left is unlikely. b. You don’t believe that John left.

It is possible, therefore, to state a truth without using the lexical concept of truth (the lexeme true). (5) shows that, while the speaker who utters (3) puts forward the proposition that John left as true, he does not assert anything about what he believes or finds likely, say. As he might comment later: ‘I said that John left, but I didn’t believe it’, and he would then not have contradicted what he said (though he will have eradicated its conversational force). As we will put this point technically in what

6 Strictly speaking, the very term ‘truth-predicate’ is a misnomer: ‘predicate’ is a grammatical notion, not a lexical one. ‘True’ is lexically an adjective, which tends to (but of course need not) function grammatically as a predicate. 7 If it is a conjecture or supposition that John left, not an assertion, intonation and/or grammar will typically have to change (‘if John left . . . ’, ‘Suppose John left’, ‘John left?’, etc.).


307

follows: without using the lexical truth predicate, the speaker effects a truth-attribution. Somehow, that is, truth is predicated when (3) is asserted, whereas the predicates ‘likely’ or ‘being believed’ are not involved: this, at least, would seem to be a natural explanation of why we obtain a contradiction with (4) but not (5). How exactly the truth predicate is ‘involved’, even if it is not present as a lexeme, is the central question we address in the rest of this subsection. The truth and significance of (2) are quite unobjectionable. (2) is not trivial, in particular. Rather it is a surprising fact that a sentence of which truth is predicated through a lexical truth predicate should be equivalent to one where there is no such truth predicate. (2) is also unobjectionable in the sense that it does absolutely nothing to dispute the objective force that the notion of truth has. Ordinary speakers including children would agree, on reflection, that truth is objective in this sense: it is not the same as belief, or assertion, or even agreement. Truth can hold even where our beliefs are wrong; or when we both are wrong. (1) does not contradict this insight in any way, and indeed partially tells us why it holds: the condition stated in (1) for ‘P’ to be true does not constrain this condition in any way, and not in particular by any epistemic clauses (such as are involved in statements like “P” is true iff P is/can be known/verified’). Above that, (1) tells us that the reason truth has this objective force cannot lie in the lexical content of the word ‘true’: for the statement that the speaker makes in (3) comes with the same objective force as when the lexical truth predicate is used, and it is equally taken to state a fact that holds objectively, or independently of whether the speaker happens to assert it or not. We are learning from (1), in short, how to talk about objective conditions for propositions to hold, without bringing in any semantic ‘relations to the world’ as expressed by certain words or predicates. All we need, it appears, is an appropriate sentence, like the RHS of (1) or (3), and an act of using it in which the objective force of truth will be implicit (the exact notion of ‘implicit’, again, needs to be clarified). This clearly points in the direction in which we want to go: grammar suffices. Maybe truth does not require us to go beyond grammar. (1) does not take us beyond language, to any non-linguistic ontology. It simply invokes the equivalence of two sentences, the explanation of which is, we will argue below, purely linguistic. But let us first look at the grammar of truth-attribution in more detail. As noted in Chapter 3, the objective force in question disappears when we change the grammar of a sentence like (3), as when we let it occur as a syntactic argument in (6): (6)

Bill said John left.

There is here no truth predicated of the proposition that John left. Since the reason why this is so cannot be lexical—the lexical elements in the proposition in question are identical—the reason must be grammatical. The grammatical difference involved, namely that ‘John left’ is a syntactic argument in the one case but not the other,

308


moreover explains the difference in interpretation, since, as we have seen in Chapters 3 and 4, it is a completely general fact about syntactic arguments that they never effect an attribution of truth (by the speaker): truth is asserted (by speakers) in matrix contexts only. The lexical truth predicate, then, is not necessary to effect a truth attribution. It turns out that it is also not sufficient. Thus, trivially, if the speaker had uttered (7), the lexical truth predicate would have occurred and it would have been attributed to the very same proposition (that John left), but the speaker would not of course then have asserted that very proposition: (7)

Bill considers it true that John left.

Had the speaker uttered (8), on the other hand, he would have attributed truth to (or asserted) the proposition that it is true that John left, but not the proposition that John left (though his implicit commitment to asserting this latter proposition would have followed logically, by a familiar reasoning): (8)

It’s true that John left.

To effect an attribution of truth to the exact proposition that John left, therefore, rather than another proposition that entails it, (3) is exactly what is needed. In line with this, the question whether John left is answered naturally by (3), and only bizarrely by (8), which normally presupposes that (3) has been asserted in the same discourse before. Truth-attributions in which the lexical truth predicate occurs thus asymmetrically depend on truth-attributions exemplified in (1), where it does not occur. Uttering (3) with normal intonation is a matter of choice, but effecting a truth attribution, when doing so, is not: context-independently, a truth-attribution to a particular proposition will be effected.8 Whatever the context and whatever the speaker’s intention, if he said (1) with normal intonation, it isn’t optional then to take him to have said that (1) is a probability, that it is what he believes, or his life’s greatest wish. In this sense, the effect is not a ‘pragmatic’ one. This is also the moment to fend off another contrarian intuition: that the truth attributed in (3) is part of an ‘implicit commitment’ that a speaker who asserts (3) shoulders. On this view, the commitment to the fact that the proposition that John left is true would be part of what can be derived from (3) using various auxiliary assumptions from logic and speech act theory—but it would not be an inherent aspect of the grammar of (3), and so truth would not be a grammatical notion, any more than the notion of belief is. A speaker asserting (3), the thought is, will also typically believe that John left. 8

Though not by an actor on stage of course, whose assertions are made on behalf of a character (who then is committed to the assertion as much as any person not on stage would be, and may indeed die as a result, contrary to the actor). We thank Uli Reichard for this clarification.


309

But surely, the predicate ‘believes’ is not somehow involved in the grammar of (3)? So why should this be more plausible in the case of truth predication? But we have already addressed this objection: ‘This is not true’, said in response to an assertion of (3), is in contradiction with it. ‘You don’t believe this’, by contrast, is not: it is rather a complaint about why the assertion took place. Relatedly, (1) tells us that a sentence in which the lexical truth predicate occurs is equivalent to one where it does not occur: no such equivalence is available for the belief-predicate. The two predicates are therefore crucially distinct: in our assertoric use of ‘snow is white’, the truth predicate must somehow already be implicit, which is why we get away with the equivalence; but the belief-predicate is not, and therefore we don’t, in that case.9 Truth-theoretic force not only co-varies with grammar, as illustrated by (6) above, but such co-variation is highly systematic, as we saw in Chapter 4 (and see the next subsection below). This co-variation leads us to the following answer to the question we have raised in this subsection: all we need to understand, in order to understand truth, is the grammar of truth attribution, disregarding the lexical truth predicate or the lexical concept of truth: the phenomenon is no deeper than that (or, to put it differently, grammar is deep enough, and maybe there isn’t anything deeper). This answer depends on deriving (1). Before we do that, note that what accounts for the truth attribution effected in (3) also cannot be the lexical predicate that we see overtly occurring in it, namely ‘left’. The truth predicate that we are claiming is ‘involved’ is not somehow ‘absorbed’ by this lexical predicate that is overtly present. This is because this same predicate is also present when (3) occurs in (6), where no truth is predicated of the embedded clause by the speaker, as noted. Crucially, therefore, the truth predicate involved in the truth attribution encoded in (3) is purely structural or configurational.10 The same is true when a lexical truth predicate does occur, and a truth attribution is effected, as in (8). But what is asserted there is not that John left but that it is true that he did. This, (8) says, is true. Therefore, there is another truth predicate involved in (8) as used assertorically that is not lexically expressed, but is purely structural and has the lexical one in its scope. Indeed, (8) only needs to occur embedded as an argument, and no truth will be attributed by the speaker to (8) at all, showing that the lexical truth predicate is not decisive. The crucial elements, rather, which are both necessary and sufficient for a truthattribution to occur, are (in English) a subject, a predicate, and finite Tense, as 9 Of course, the view we are developing does not exclude modelling truth like belief, as an implicit commitment of speakers, and formalizing such commitments in a logical meta-language. But we are after the source of the phenomenon of truth, and the more intrinsic and intricate the co-variation with grammatical patterns is, the less explanatory an account based on implicit commitments will become. 10 This possibility should not be surprising, as many concepts that are also lexicalized can, in other instances, be present purely structurally: identity, for example, is expressed in the lexical concept ‘identical’, but also in the sentence ‘John is him [pointing]’ or ‘Hesperus is Phosphorus’. The same applies to thematic role concepts: ‘Agent’, qua thematic role, is a grammatical notion, as we have stressed, and is expressed purely configurationally; but ‘is the agent of ’ is a way of denoting the agent of an action lexically.

310


occurring in a matrix or root configuration. As noted, all of these elements are crucially grammatical notions: they relate to grammatical meaning in our sense. Consider now how this configurational analysis explains why the Equivalence Schema intuitively holds, which in contemporary discussions (cf. Horwich, 2010) has the status of an unexplained or primitive fact about how the notion of truth happens to be used. Consider (9): (9) ‘John left’ is true iff John left. Why is this true? The answer is that although (3) identically occurs as a constituent in (9), as much as it does in (6), the grammar of (6) and of (9) are again not the same, just as was the case with (3) and (6). If the grammar of the right-hand side (RHS) of (9) contained what we have called the structural truth predicate, in exactly the way that (3) does when (3) is asserted, then it would not be puzzling at all that the RHS of (9), if asserted, could be equivalent to the LHS, where a truth predicate is lexically present. The question therefore is: Is the RHS indeed grammatically relevantly identical to (3), and relevantly different from that of (3) as occurring in (6)? The answer is of course yes: for whereas the grammar of (3) as occurring in (6) is that of a syntactic argument, the grammar of (3) as occurring in (9), is not: (9) is a paratactic, not a hypotactic construction, and (3) occurs in a root (or ‘unembedded’) context in (9) (Heycock, 2006). (9), then, holds because the structural truth predicate is present on both sides. This explanation, however, is not complete. For on the LHS there is an additional lexical predicate. So the explanation is only complete if we can also explain why this lexical predicate does not disturb the effect of the structural predicate present there. This lexical predicate on the other hand has no substantive descriptive content: it doesn’t add any content to the compositional semantics of an utterance. So we do not expect it will affect the truth-attribution effected on the LHS. Nonetheless, we need to explain why this predicate, lexically empty as it may be, does crucially not induce an opacity effect for the embedded clause on the LHS: curiously, in the sentence ‘That John left is true’ as much as in the sentence ‘It’s true that John left’, the full truth of John’s having left is assumed by the speaker. Why is this? This remaining riddle will be solved in the next subsection. Consider what alternative explanation we might have tried: it so happens, we might have said, that, in the language of the person we are interpreting, the sentence ‘John left’ expresses the belief that John left. ‘True’ as occurring on the LHS therefore is a predicate that, as applied to the language of this person, has the sentence ‘John left’ in its extension. This sentence is therefore true in that language iff John left. This explanation invokes no grammar (except that one word is isolated as a predicate, which is a grammatical fact, but often mistakenly taken to be a lexical one), and it does invoke a non-linguistic ontology of ‘belief ’. What meanings are associated with an utterance is then a matter of what beliefs are expressed in it. The two explanations


311

couldn’t be more different, therefore. All we invoked in our explanation was grammar, and there was no need for an ontology of belief at all. We will revisit this curious conclusion below, when arguing that the ontology in question only arises when we fail to give grammar its due.

9.3 The grammar of truth: a unified account Let us now turn to what we just identified as the remaining riddle, and also expand the coverage of our account to the full range of constructions in which truth attribution can be effected: (10)

It seems that John left.

(11)

It’s that John left.

(12)

It’s true that John left.

(13)

That John left is true.

(Moro, 1997)

All of these constructions are similar in that they come with some assertoric or truththeoretic force assigned to the embedded proposition. This could be questioned for (10), which, apart from the reading of interest here, has another reading of which (14) is a paraphrase (see Ruwet, 1991): (14)

It appears that John left.

The reading that we are interested in, however, is paraphrased as in (15), where the lexical truth predicate appears: (15)

It seems true that John left. / That John left seems true.

While (14) expresses the conjecture or impression that John left, in (15) the same proposition is put forward with somewhat more certainty, involving more truththeoretic force. To be sure, the speaker of (10), under the intended reading, falls short of asserting that John left—and indeed, the grammar of (10) itself suggests no such thing, according to our own analysis, since ‘John left’ is not a root clause in it. But it is also not that the speaker asserting (10), under the intended reading, considers the truth value of the proposition that John left still totally open. Rather, the moment that doubt was to creep back in, the reading would shift to (14) (and ‘seems’ would be intonationally stressed). The truth-theoretic force in question is also clearly stronger than in the case of ‘It seems possible/probable/likely that John left’, which does not capture the intuitive meaning and force of (10) at all, under any reading, and which therefore demonstrates that ‘true’ is quite the right notion to use in (15) to bring out the force of (10) under the reading in question. The force is also of course stronger than in (6), where,

312


as we noted, essentially no truth-theoretic force is attached to the embedded proposition. In sum, the embedded clause in (10), while technically in a subordinated position (‘John left’ is no matrix clause), is not in a position as opaque as the one where it occurs in (6), but not as transparent as that of a root clause either. The important conclusion to draw from all this, of course, is that the lexical truth predicate used in (15) to bring out the relevant reading of (10) is again not lexically overt in (10). And again, the truth predicate lexically expressed in (15) and structurally present in (10), under the relevant reading, is again, in either case, within the scope of another and structurally expressed truth predicate. For this very reason, what is asserted to be true in (15) is that it seems that John left, not that John left. Put differently, the structurally present truth predicate again overrides the effect of the lexical one, and it always has maximal scope. Against this background, let us now look at (11), the second construction type above, repeated here as (16): (16)

It’s that John left.

If we first compare ‘said’ in (6) with ‘appears’ in (14), then ‘appears’ with ‘seems’ in (8), and then ‘seems’ with the cliticized copula ‘’s’ in (16), we see the lexical content of the matrix verb gradually fading away, until eventually it is bleached altogether, even morphologically. As the lexical content of the matrix predicate fades away, the truththeoretic force of the embedded proposition increases. (10), under the relevant reading, when asserted, involves some truth-theoretic force attributed to the proposition in question. In (16), curiously, the truth of the proposition that John left is presupposed: it is taken as a fact that John left. (16) might for example be an answer to the question ‘what’s the matter?’ or ‘what is it?’ (see Moro, 1997), and when answering this question with (16), the answer could roughly be paraphrased with (17): (17) It’s the truth/the fact that John left.11 The point about example (16), then, is again that, unlike in (12) or (13), no lexical truth predicate is present. Yet truth is attributed to the proposition that John left. Ipso facto, once again, the phenomenon under investigation is not lexical: it must follow from the grammar. The effect is achieved by voiding the matrix verb of lexical content completely, and by thereby avoiding the trap of syntactic embedding. The difference in the degree of opacity of the embedded clause in the above constructions and in relation to (3) and (6) can now be traced to a purely grammatical fact: the clause occurring embedded in either (10) or (16) is not the syntactically 11

(16) is not fully grammatical for some speakers of English (including one of the co-authors of this book), but it improves with ‘just’ inserted before ‘that’. It is also fine in German (Es ist, dass sie geweint hat) or Italian (C’è que Gianni e partito). For some, (17) with ‘the truth’ is out. This correlates with the fact that the proposition embedded in (16) is factive (presupposed as true and referenced as a fact) rather than asserted as a truth. This is what we expect given the general fact that no proposition, as long as it is embedded at all, is asserted as true (see Chapter 3).


313

subordinated argument of the matrix verb ‘seems’ or ‘is’, respectively, in the way that it is the syntactic argument of ‘says’ in (6). Instead, ‘seems’ in (10) and ‘is’ in (16) take a Small Clause as their complement, of which ‘John left’ is the subject. The general rule is that the more we approximate a root context, the more a clause becomes transparent. Consider clausal adjuncts in this regard: since adjuncts are not syntactic arguments, lacking Case and thus exhibiting reduced connectivity to the matrix clause (see Chomsky, 2004: 116–18), they, too, approximate a root context or a form of parataxis. In line with that, truth-theoretic force reappears in them: (18)

Bill said John left, even though John didn’t.

Here, the clause ‘even though John didn’t’, being an adjunct, has syntactic connectivity with the syntactic object to which it attaches that is looser than in the case of arguments. The above account thus predicts that it should also be transparent for truth-evaluation. And of course it is: the speaker of (18) does assert it as true that John didn’t leave. This again confirms our earlier conclusion that when, and when not, a truth-attribution occurs, and to which degree it occurs, is regulated by grammar (‘it ‘supervenes’ on grammar, in philosophical terms), not by non-grammatical kinds of entailment or implicit commitments. Let us now turn to (12) and (13), repeated here as (19) and (20): (19)

It is true that John left

(20)

That John left is true.

These are the most canonical constructions discussed in philosophy when the ‘truth predicate’ is being discussed. The first observation to make here is that these are copular constructions. This is by now exactly what we expect, and what can now be explained grammatically: for whoever asserts (19) or (20), does assign truth-theoretical force to the proposition that John left. Hence the sentence denoting this proposition cannot occur as a syntactic argument, predicting that we find a copular construction with a Small Clause complement instead. The copular verb is not even ‘seems’, but the one completely bleached of lexical content, namely ‘is’. The derivations we take to be as follows. In (20), That John left is clearly the subject of the predicate ‘true’: (21)

is [SC [CP that John left] true]

This whole subject moves to the specifier of T(ense), yielding (22): (22) [[CP that John left] T is [SC t true]] Also (19) is derived from a SC, but this time the subject is the expletive ‘it’: (23)

is [SC it true]

314


Again the subject ‘it’ moves to the sentential subject position: (24) it is [SC t true]] The that-clause, however, which functions as the ‘associate’ of the expletive, forms an adjunct that is extraposed, as the fact suggests that it can be omitted (yielding the fully grammatical sentence It is true): (25)

[[it is [SC t true]] [SC that [John left]]]

This yields an explicit and unified account of the syntax of (19) and (20), which also explains their semantic properties. The embedded clause in these instances can have no greater assertoric force than in these very constructions, short of occurring as a root clause itself. This is because in neither case is the clause a syntactic argument of the matrix verb. Lack of opacity follows. We now also expect why ‘seems’—called the ‘quasi-copula’—forms an intermediate case. Given how weak the copula and quasi-copula are in terms of their substantive lexical content, neither of them can refer to an event that is properly independent of the one referred to by the embedded verb ‘left’. In other words, they have no event position of their own, which could take another event or proposition as argument, creating a structure semantically in which one event is part of another. In this sense, (19) and (20) are more like a mono-eventive (or mono-clausal) structures, predicting the empirical facts we observed: that its embedded clause will be less opaque than in a clearly bi-clausal and bi-eventive structure such as (6), where there are two events, one of saying and one of leaving. The embedded clause can therefore carry some truth-theoretic force, as indeed it does. While truth predication is paradigmatically a matrix (root, or whole-sentence) phenomenon, raising constructions approximate this phenomenon to some (though not a full) extent. As for (13), it is grammatically exactly like (19), except of course that there is no lexical truth predicate in it at all, despite the fact that a truth-presupposition occurs. In an assertion of (19), which is ungrammatical when used out of a relevant context, there is nonetheless no assertion that John left. It’s being taken as old information, rather, that he did, turning this into an established fact, which is referenced in the paraphrase ‘It’s the fact/truth that John left’. If we want to have this structure as patterning with the previous ones, and with copular constructions in general, we have essentially only two options: we will begin with a SC forming the complement of ‘is’, and that can logically either be the one in (26) or the one in (27), in which the roles of subject and predicate have simply reversed: (26)

is [SC it [CP that John left]]

(27)

is [SC [CP that John left] it]


315

In short, ‘it’ is either the subject of the SC or its predicate. There do not seem to be other options. As Moro (1997) argues at length, several empirical facts suggest the latter option, and we will adopt this analysis here. For example, it is quite implausible for the clause to be a predicate in such constructions, as the ungrammaticality of (28) suggests: (28)

*Peteri is [[SC ti [CP that [IP John left]]]

Also, there is no particular reason why an NP such as ‘it’ should not to play the role of (or to stand in for) a predicate, since definite nominals, for example, can be perfectly good predicates, as in (29), and can then invert, making perfectly good sentential subjects: (29)

John is the mayor.

(30)

The mayor is John.

In line with this, the analysis of ‘it’ as a predicate as in (27) would reflect an ‘inverse’ option generally available in copular sentences, with the SC-predicate moving over the SC-subject into sentence subject position, to yield the overt form (11/16). Further evidence for this hypothesis is that where the predicate does ‘invert’ over the SC-subject in this fashion, we see that there is generally a copula required, as (32) shows where such inversion has taken place: (31)

I consider [SC John [NP the problem]]

(32)

I consider [SC the problem *(to be) [NP John]]

In line with that, (33) is ungrammatical: (33)

*I believe/consider it that John left.

But why, then, is (34) ungrammatical, which, if ‘it’ is the SC-predicate, would simply correspond to the ‘canonical’ (‘non-inverted’) case? (34)

*That John left is it.

However, there is a compelling independent reason for why (34) should be ungrammatical: ‘it’, as a placeholder for a predicate, is a pro-form, which as such cannot be left stranded (see Moro, 1997: 179–80). Inversion of the predicate over the subject is therefore obligatory, and (30) is predicted to be out. The analysis of (11/16), then, (35): (35)

iti is [SC [that [John left]] proi]

In this construction, there is no lexical predicate in the SC, and the copula does not encode an independent event, as a normal matrix verb does (it lacks an independent event position, in Neo-Davidsonian terms). Hence it seems plausible that the truthattribution cannot be delayed to the matrix clause in this case, and the grammar can

316


realize the required structure for an attribution of truth only by resorting to a ‘dummy’ predicate, whose only function is to assemble the configuration required for a truth-judgement to occur. For considerable further evidence in favour of this analysis, see Moro (1997) and Hinzen (2013a). (35) now suggests that the structure for (36) should be (37): (36)

it seems that John left

(37) iti seems [SC [that John left] proi] By contrast, when the lexical truth predicate is present, the structure is (38), where ‘that John left’ is again an extraposed adjunct that is the associate of the expletive: (38)

a. [it seems [SC __ true]] [that John left]

In support of the presence of a ‘dummy’ predicate in (37), consider that (39), where ‘John’ is raised from the position of the subject of the embedded clause to sentencesubject position, is fully grammatical, while (40), from Moro (1997: 195), is out: (39)

John seems [ t to be sad]

(40)

*it seems [for John to be sad]

‘John seems to be sad’

This data follow on the above analysis. For if the SC in (37) assembles the configuration for a truth-attribution (though finite Tense is still missing), the predicate obviously requires a truth-evaluable constituent to apply to. But the non-finite clause in (39) doesn’t qualify as such, as we independently know: no truth can be grammatically predicated of it: (41)

*[for John to be sad] is true

Why, in turn, can we say It seems that John left, but not *That John left seems? Again, because the underlying structure would be (42), and pro must move, parallel to (34): (42)

*[That John left] seems [SC t pro]

By contrast, we expect That John left seems true to be fine, as it indeed is. In summary, we have now provided a unified analysis of four constructions in which truth-theoretic force is present, yet not in a root context. It has transpired that in all of these cases, a root context, and the relevant configuration required for a truth-attribution to occur, is closely approximated. This is possible because in none of the cases is the clause in question syntactically subordinated as a syntactic argument. Opacity is maximal when the clause occurs as a properly hypotactically embedded syntactic argument. As we move to more mono-eventive structures, where the embedded clause is not a syntactic argument of the matrix verb, opacity is slightly lifted, and a truth predicate can be inserted, as we have just seen. As we move to


317

matrix clauses, finally, or to coordinative or paratactic constructions, it is lifted completely, and transparency is maximal. Truth-theoretic force, in short, is regulated grammatically. At the same time, the most basic fact about knowledge of truth is (1). And to appreciate (1), all we needed to understand was the grammar of truth attribution. Our understanding of truth reduces to our understanding of grammar, and this is the answer to our question what it means that truth is a grammatical notion.

9.4 The limit of grammar and the limits of thought Any root context in grammar has a special, extensional semantics. If I shout (43), I directly refer to John, who must be a part of the speech context (or must be taken to be): (43)

John!

Reference is not left open or undecided here, any more than I have argued truth is in an utterance of (3). Both are root phenomena, illustrating, now in the nominal case, the correlation we have argued for. When proper names occur in embedded positions, on the other hand, we again see that referential force is lost, and reference becomes less direct. Thus, in (44), the speaker clearly doesn’t ipso facto refer to John, in the sense that there is some direct or causal connection to him within the speech context: (44)

I called John yesterday.

This is even more obvious in (45) and (46): (45)

I have no idea who John is.

(46)

Are you John?

Root uses of proper names thus connect with the external world in ways that embedded uses of them do not, paralleling what we observed for clauses. Only the former uses can be described as ‘indexical’. Similar observations apply to interjections, as in (47), where it will be immediate from the circumstance of use what ‘Sh***’ refers to: (47)

Sh***!

The same is true of the actions that imperatives such as ‘Come!’ refer to: nothing here can drive a wedge between the utterance and the action it refers to. The same applies to the intended subject of the action, which is the always silent YOU that is part of such a speech act.

318


In line with these general observations on the role of the grammatical root, we have argued above that as the grammatical derivation reaches the root (and nothing takes place to pervert the normal course of events, for example wh-movement to form a question), a potential truth is assembled: language, if you allow the metaphor, can now ‘touch’ the world. At this point, nothing lexical mediates the act of reference any more: all concepts have entered the derivation and are processed for reference, having been converted into referential acts or predicates in the course of the derivation. In (43), by contrast, there is still at least the minimal lexical content of ‘John’. In an assertion, however, we are, as the derivation stops, alone with grammar, and truth. No lexical concept with descriptive content stands between us and the world, and reference is maximally ‘direct’, yielding the most extensional form of reference that exists. In that form of reference, all concepts can now be substituted for co-referential ones, leaving the truth in question untouched: truth is ‘extensional’. Beginning from conceptualization, then, we are, with truth, at grammar’s other end. And this, indeed, is the limit of grammar: nothing happens, grammatically, beyond this point. It follows that grammar and truth intrinsically relate: the organization of grammar inherently reflects the organization of truth. And the limit in question, it turns out, is also the limit of human thought. To arrive at this conclusion, consider what else can happen in grammar after a truth-attribution has been effected. We cannot, of course, add a completely new clause I said that, to derive (48), for this would deprive us of the truth-theoretic force again: (48)

I said that John left.

Moreover, as noted in Chapter 3, such ‘recursion’ will anyhow only reassemble the basic Aristotelian template that a statement of truth exploits, with ‘I’ as the subject, ‘SAY that John left’ as the complex predicate, and Tense marked as an inflection on the matrix verb (leading to the spell-out of SAY as ‘said’). Not only does the added clause not have any new structure itself. The newly arising complex (or recursive) sentence also does not have any different structure from that of a single sentence: it still only has a single subject, a single (complex) predicate, a single Tense (PAST),12 and (most crucially) a single truth value, as we have seen. In this sense, truthattribution remains the ‘maximal’ phenomenon ever computed in grammar: all that ever happens (unless an interrogative, optative, or imperative is derived, of course), is that a single truth value is assigned, with truth value assignment being crucially an undivided phenomenon.

12 Of course there is an embedded Tense (specified on ‘left’). However, at the matrix level, Tense is evaluated as indexical in relation to the point of speech (the Now of the speaker), and embedded Tenses are evaluated differently, namely at least partially in relation to the temporal reference point established by the matrix Tense. In particular, (6) has a reading where what Bill said is ‘John left’, hence where the leaving took place prior to the saying (see Giorgi, 2010).


319

After the truth is assembled, we could on the other hand add a clausal adjunct. In principle, an arbitrary number of such adjuncts can be added to a fully configured assertion (John left, without any doubt, before noon . . . ). Every such adjunct functions as a further specification of the truth claimed, but none is grammatically required, and none of this changes this truth as such, or the fact that it is being attributed to a given proposition. In languages where assertions require a specification of evidentiality (the source of information on the basis of which the assertion is made), on the other hand, we would have to add a relevant verbal inflection or clitic to the assertion. These are arguably not the same as modals, and where an assertion is specified for evidentiality in such languages, the truth value of the proposition is not wiped out, as it is in (48) (cf. Aikhenvald, 2004: 2): the assertion stands, and what is added is an indication of its epistemic source. This difference interestingly correlates with a difference in the grammar of evidentiality markers as compared with that of modals (or adjuncts): while modals are recursive to some extent (e.g. ‘Evidently, John possibly left’), evidentials lack this property, in a similar way as English tag-questions do, which similarly are consistent with an assertion taking place (cf. ‘John left, didn’t he?’, but not ‘*John left, didn’t he, did he?’).13 At the point in a syntactic derivation where modals are added, then, the derivation is not yet ‘halted’ and the truth value is not yet assigned. By contrast, where evidentials or tags are added, the truth value has been assigned, and this is why structure-building is essentially stopped: the process of adding the evidential or tag therefore won’t be recursive, and it also won’t change the truth-conditional (or compositional) semantic content of the assertion made. Modals and adjuncts, which are recursive and have compositional propositional content, are not grammatically required. Evidentials, where they exist in a language, are grammatically required (and tags are at least similar and unlike modals/adjuncts in exhibiting no lexical content). Non-recursivity in the latter case thus points us to an inherent limit to structure building in grammar. Truth, then, does indeed have a signature as a ‘maximal’ phenomenon in the design of grammar: when the truth value is assigned, both recursion and compositionality are halted. Of course, when (3) is uttered, another person may hear this utterance and also hear it doubted. Thus she may want to re-confirm it. This moment, where the epistemic subject changes, is the moment where the lexical truth predicate acquires its crucial function. For the person may now utter (49): (49)

13

It’s true that John left

As Uli Reichard notes in personal communication, even though tags cannot deprive a sentence of assertoric force, they can weaken the force. On our account here this is expected, given that they intervene between force and the root.

320


Although this adds nothing to the compositional semantic content of (3), it does involve a change in indexicality, in the sense that the predication of truth will typically be indexed to a different speaker (or epistemic subject) than the assertion of (3) to which (49) reacts. If so, however, there is then still a single structural truth predicate present, necessarily indexed to the speaker in the actual act of utterance. Where more lexical truth predicates are recursively added, and the epistemic Self remains the same, only a reading is derived in which, with every new round of iteration, the message gets ever more intense (until one feels this must be the speech of a maniac): (50)

It’s true that it’s true that it’s true that John left.14

Let us assume, then, that the predication of truth is a ‘maximal’ structural complexity towards which a syntactic derivation gravitates, and which stops the derivation when it is reached. This makes a prediction for those constructions in which some assumption or presupposition of truth is in place, although there is no lexical truth predicate: namely (apart from of course (3)) the copular constructions in (10–11). If, in these constructions, with the help of the pro-predicates that we argued they contain, the basic skeleton of a truth-attribution is assembled, an iteration of these constructions should be difficult or out. Consider, then, the attempt to perform such iterations in (51) and (52): (51)

*?it’s that it’s that John left

(52)

?*it seems that it seems that John left

Judgements on these constructions differ. For us, the second only has the reading where ‘seems’ has the reading of ‘appears’, consistent with our prediction; and (51) is essentially uninterpretable. As we noted, the meaning of (51) is similar to ‘it’s a fact that John left’. And the above predicts why we cannot go further than that: having made reference to a fact in this construction, we can’t make reference to this reference, as it were. We have reached the limits of grammar. And hence, on the above hypothesis, we should not be able to iterate. Where, on the other hand, the lexical truth predicate occurs, idle or even non-idle (as in the case of an indexical shift) iterations are possible; but, as we have argued, truth as structurally expressed is the more fundamental phenomenon. And that phenomenon does not iterate. Let us connect this point to a curious fact about an apparent limit imposed on what we may call the ‘semantic ontology of natural language’. This ontology happens to be 14 One could attach ‘it is false’ to an assertion of ‘it is true that John left’, and thus change the semantic predicate. This does yield a compositional semantic effect—which is exactly what we predict from the fact that negation unlike truth is always lexically overt and a paradigmatic example of an element that yields a compositional semantic effect. But any such negative element will necessarily be in the scope of a sentential truth predicate (we never assert a falsehood).


321

such that metaphysicians commonly distinguish between ‘propositions’ and ‘facts’. For example, we might wonder whether John left; once we have established that, we might utter (53): (53)

It’s a fact that John left.

For a fact is semantically similar to a proposition that is (or has been) evaluated as true. If we say (53), however, we usually consider the matter settled. In no human language does there seem to exist a third ontological category—call this fictional category ‘pact’—such that, once it was challenged that (3) is indeed a fact, as claimed in (53), and we concluded that (53) was indeed correct, we could comment as in (54), with the semantics that the fact has now been converted into a pact, just as we previously converted a proposition into a fact: (54) It’s a pact that John left More fundamental than facts, in other words, things don’t get. Truths can, at best, be facts, and that is it. Truth, in the grammatical sense, is not ‘recursive’ in this sense, and doesn’t iterate for this reason. This is not simply a metaphysician’s theoretical decision or choice: rather, we find ourselves confronted, in thought, with an absolute limit. Why should there be this curious restriction on the ontological categories involved in the semantics of natural languages? If we can move from propositions to facts (with the latter category being ipso facto truth-involving in a way that propositions are not), why can’t we move further? Given the above, one answer is now natural: this restriction reflects an analytic truth of grammar. Anything more complex than truths, the grammar cannot compute. As we have argued, the sentential truth predicate we have posited brings the process of generating the complexities of grammatical semantics to a close. In terms of categorial complexity, nothing happens after that. There is thus a hierarchy of semantic evaluations: we can evaluate an object for its colour or beauty, an act for its morality, or proposition for its possibility or desirability. But evaluations like the former two will happen inside of clauses, and evaluations like the latter will be in the scope of evaluations for truth. But a more basic or fundamental level of semantic assessment—more abstract even than truth—does not seem to exist. In this section we have considered clausal embedding, modals, adjuncts, evidentials, tags, and lexical truth predicates: once a configuration is assembled that can effect a truth-attribution, any of these can be added. Sometimes, this is possible in an iterative, recursive fashion; sometimes not. Where not, this illustrates that the grammatical process has come to an end with the configuration obtained; where it is possible, the recursion reassembles the same template in the next step, but the template never changes; and the structural truth predicate will never iterate or become recursive. The conclusion is that the point in a syntactic derivation where an expression is evaluated for truth and put to an assertoric use in discourse does

322


indeed mark an upper boundary in the build-up of structural complexity in grammar. This is why, when the lexical truth predicate is now added, nothing in the compositional semantic content changes, and it can be dropped again. The root of a (declarative) clause is not merely the moment where truth is attached to the proposition that the clause expresses, it is also the moment where the grammatical process itself comes to an end. Grammar, in this sense, is bounded by truth. This conclusion derives a central contention of a ‘deflationist’ approach to the metaphysics of truth: that the truth predicate has no substantive empirical content, in the way that ‘blue’ does, or ‘water’. Why is this so? It simply follows, if truth is a grammatical notion. Being structural in nature, we do not expect it to have a substantive lexical content, and we have a positive reason to conclude that the search for an external relation of truth—in the way that we may want to assume a relation between ‘water’ and the substance H2O that it denotes—is misguided, exactly as the deflationist contends. The function of truth in grammar is to bound, rather than extend, the compositional process applying to lexical meanings. When this boundary is reached, semantic valuation for truth takes place and no compositional and truthconditionally relevant content can be added. The function of the truth predicate is not to contribute substantive semantic content but to stop the process of composing it. Of course, we have also derived the Tarskian Equivalence scheme, the centrepiece of the deflationist edifice (at least for sentences not themselves containing the truth predicate). The explanation is: two truth predicates are involved, one lexical and one structurally expressed, and the second is present even where the first is not, in a root context. Overall, our account then vindicates and derives the deflationist intuition that the truth predicate has no real-world, empirical, or substantive content, and that therefore no naturalistic or reductive analysis is appropriate for it (Horwich, 2010). For its content is purely grammatical, and our understanding of the notion of truth reduces to our competence in handling grammar. At the same time, there is of course another way of reading our result as well, which we think is ultimately more appropriate. Rather than saying we have ‘deflated’ truth, reducing it to grammar, we could say that we have directed attention to something curious about grammar: that it lays the foundation for truth. While we have, in a sense, deflated truth, we have, at the same time, inflated grammar. In this sense, the mystery of truth remains the same it has been all along: it has just been transferred to another domain—grammar, where we argue it has its home. This leads us, finally, to a metaphysical epilogue of our book.

9.5 Language and reality Our perspective involves an account of thought, and no such account would be complete from a philosophical point of view if it came without an account of truth, a notion to which thought is taken to intrinsically relate. We have now said, using (1),


323

what it is for a sentence to be true, and our answer has closely linked truth with the practice of assertion. But there is no argument here that what we assert to be true is in fact true. Nor should there be such an argument, for no such fact obtains: assertion does not entail truth. Nonetheless, might it be the case that what we think is true and choose to assert, is in fact largely false? This is not a metaphysical question, concerning the objectivity of truth, which we have addressed above. It is rather an epistemological question, asking whether the way in which we think and use language might be largely out of touch with reality. Maybe we do indeed regard assertions sometimes as ‘gambles’, perhaps especially in a scientific context: we never really know, we think, whether what we say is true is, in fact, true. Our assertions, we feel, are just our ‘best bet’. Nonetheless, the more an assertion approximates a gamble, the more we sense that its status converts from an assertion to a conjecture, and that intonation may indicate that. We think that there is something to be said, especially outside of a scientific context, for the claim that we are, in choosing to assert what we do, also mostly right, and that the idea that our thought might be out of touch with reality is really mistaken. Assertion, in other words, is not a gamble. An immediate objection to this claim is that an agent who thinks truth conditional thoughts must have an ability to be surprised—as when his beliefs unexpectedly turn out to be mistaken (Davidson, 1982). But notice that our claim wasn’t actually one about beliefs, but rather one about assertion. An assertion—putting forward a proposition as true, which in turn has now transpired to simply mean: using the proposition in the right configuration—is, on our account, a deictic act, like pointing, though a more grammaticalized version of it. There is a sense in which we cannot make a mistake in pointing, not even in declarative pointing. An infant silently pointing to an airplane in order to draw an adult’s attention to it may well be wrong that it is an airplane. But this doesn’t make the act of pointing mistaken or unsuccessful. The child may also be pointing to the airplane, and I can’t figure out, from his extended index finger, and from the context, what it is he is pointing to. Again, while the act does not succeed in this case, it is not mistaken. But an assertion in our framework is nothing other than an act of pointing: to a ‘truth value’, if you will. This suggests that assuming that such acts can be ‘false’ is to make a category mistake: propositions are true or false, not acts of asserting them. In that case, however, it is also a category mistake to say that such acts will typically be right. The claim under discussion, therefore, must rather be the claim that the proposition that is put forward as true in an assertion will, typically, also be a true proposition. There is a proposition involved, in any act of assertion, and the proposition can be false even if the act of assertion says it is true. This allows for the possibility of mistake, even if, as our claim goes, such mistakes are rare. Why then should we endorse this claim? Our answer appeals to the fact that the practice of assertion, on our analysis, has the notion of truth built-in, so to speak. Truth, when asserted, is not marked, but falsehood is: asserting ‘John left’ suffices to assert a truth,

324


but asserting a falsehood does require a lexical marker: negation. So truth is the basis, falsehood the marked exception. Even if we assert the falsehood of some proposition P, it’s still a truth that we assert: that P is false. Truth is in this sense an inescapable default, and falsehood is marked as the exception, as the negation of truth. Truth and falsehood are in this sense not on an equal footing, and this is because truth is grammatical, and falsehood is lexical. We cannot reverse this asymmetry, making falsehood the default or the configurational case: uttering ‘John left’ when we think this is false, and adding an assertion marker when we exceptionally think it is true. For such a reversal is not simply a matter of changing the meaning of one conventional lexical marker for another: it amounts to de-grammaticalizing our mind. The impossibility of a reversal is not logical or conceptual, for falsehood is compositional: it is the negation of truth. Falsehood, in this sense, presupposes truth, and we couldn’t make an empirical discovery to the effect that, in some culture, it did not. The minds of the speakers of such a culture would be ones that we couldn’t interpret. A recent film directed by Ricky Gervais depicts the ‘invention of lying’, in a culture where there is no word for truth, and everything people ever say is what they take to be true. There is a scene where the mutant protagonist, having just told the first lie ever, tries to express the thought that what he said on this occasion, amazingly, wasn’t true. Yet, as there is no word for truth, the thought cannot be expressed, leading to complete consternation in the faces of his friends. This is a culture in which the configurational analysis of truth holds to the strongest extent: there is no lexical item ‘true’, and hence there are no thoughts about truth. If, in our terms, speakers in this culture say something false, it can only be by accident. The culture could barely function otherwise—just as perception could not work in the way it does, if hallucinations were the rule rather than the exception. It is hard to imagine a culture, by contrast, in which, not lying, and not the lexeme true, but truth itself was invented. If we are right, it would be a world without grammar. It couldn’t be one in which all people tend to always lie, for, conceptually, lying presupposes truth. People might only be making bets, yet bets, again, as understood by us, are bets that something is true. A world with grammar in which there is no lexeme ‘true’, and hence no reflection on (or reference to) truth, is a world in which falsehood must be rare. Our world has both grammar and the lexeme true, and even a lexeme false, which could lexicalize the grammatical pattern ‘not true’. Further lexemes could then be lexicalized, like the concept of possibility, which again could arise grammatically, as the disjunction ‘true or false’. In a sceptical mood, we might then decide to prefix everything we normally simply assert with the qualifier ‘it is possible that’. But we couldn’t then escape from truth here, and it would be the truth of a possibility we would assert. Scepticism only makes sense against a background of truth that is assumed. As the conceptual edifice is built, in short, the essential phasal template that gives us truth in the first place is presupposed, as its grammatical foundation. We can then lexicalize grammatical


325

patterns, and among the lexical concepts that we then construct, true will be the most fundamental. For if any concepts at all apply to an object, they do so truly. For a concept to apply to an object is for it to be true that it applies to the object. Truth, and hence grammar, are thus the foundation on which all thought rests. If conceptualization and falsehood both presuppose truth, we cannot be largely wrong: falsehood can always arise, and hence truth be absent—but not at large. For that would deprive thought of a foundation. It would suggest that there is something more fundamental than truth, perhaps divine whim, which is the foundation of thought. But this does not seem to be the case. And we would see why it is the case, if, ontologically, there was nothing more fundamental than grammar, and truth was a grammatical concept. If there was something more fundamental, it would have to encompass grammar, but somehow go beyond it. Such a system would be very easy to imagine, if grammar is just an arbitrary formal system. For then, immediately, there is a deeper system, which provides the ultimate foundation: the world itself, and the formal ontology in it that metaphysicians posit. But the formal ontology of the world, we have argued, falls out from grammar: with grammar, we reach rock bottom. It sets the framework for thought. For the sceptic to be right, the framework would have to change. This line of argument connects with an epistemological and pragmatist one, proposed in the domain of the dynamics of belief revision (Levi, 1980; Rott, 2001): confronted with a doubt in a given instance, we must assume that what we do otherwise believe is also largely right. Doubt can arise coherently only against a background of beliefs that we hold on to; possibility is coherent against a background of truth. Falsehood is a drop, or a current, in an ocean of truth. Any particular belief we hold may be up for grabs as new evidence we receive throws doubt on it. But even if we give up a particular belief, converting it back into a mere hypothesis, we will still leave much of our other beliefs intact. While this argument appears intuitively not implausible, the above is slightly different: it does not depend on an ontology of belief, and it is more naturalistic: rather than positing a pragmatic norm of rational belief change, it exploits the hypothesis that grammar, as a matter of naturalistic fact, configures our mind. Our mind then can only move in this format, and it deprives itself of a foundation if it attempted to move in a world of mere possibility, say, rather than a world of truth in which possibilities arise against a background of truth. The configurational analysis of truth adds an element, then, and perhaps power, to an argument that its epistemological version lacks. While there is a conceptual connection between belief and truth, there is no independent and empirically based configurational analysis that supports it, if belief is, as commonly supposed, a non-linguistic ‘propositional attitude’ that is largely defined in logical terms, namely as occupying a position in logical space or corresponding to a ‘set of possible worlds’ (Stalnaker, 1984). Here is a second argument for the conclusion that what we assert to be true also is, largely, true. Before an act of assertion takes place, a proposition involved in the act is

326


true or false, we assume. After we make a decision and assert it, it becomes part of an act of ‘pointing’: to a truth value, in this instance. The tense of the event that the proposition in question encodes is then deictically anchored relative to the speech time and speech location, as determined through the act of speech itself, a physical event taking place in time, space and discourse. The assertion itself, as an act, is anchored in the agency of the speaker, as is the assertion’s assertoric force, which is anchored in the same agent viewed as an epistemic subject, rather than merely a speech agent. This Ego is also ultimately responsible for which object is denoted by any nominal in the sentence: each such object is located in space in relation to this subject viewed as the ‘deictic centre’ or ‘origo’ of a deictic space of possible reference in the sense of Buehler (1934). All of these specifications are grammatically necessary in every single act of assertion: no assertion can be assessed by a hearer without them. To understand an assertion, then, we need to ‘bind’ all temporal and spatial reference that takes place, as well as illocutionary force, to the Origo. Beyond that, there is nothing in which to ground a speech act: the external world as the context in which the act takes place, the epistemic subject that enacts it, the hearer to whom the act of reference is directed, and the language that the speaker uses, are all part of a single scheme, which nothing else grounds. It is, in a sense, the sum-total of things. The question of a mistake, we have also noted, only arises for the propositions that are part of assertoric acts. But if we look at the complexity just depicted, then, what exactly is the proposition that we can extricate from its web of relations, so as to obtain a theoretical construct that is a potential candidate for what can be systematically false, even when we assert it to be true? In early analytic philosophy, where the issue is the semantics of mathematics, the issue does not really arise. Whether the proposition that 2+2=4 is true or false makes perfect sense as a question: we do not need to consider any speech act. The question whether ‘John left’ is true, by contrast, taken as such, doesn’t make any such sense. We have no idea who ‘John’ refers to, for example: only embedding the proposition in the speaker’s deictic space will tell. We also have no idea where the event of leaving marked grammatically as ‘past’ is located, for to know this we need to locate it in relation to a time of an actual act of speech. To evaluate a proposition as simple as ‘John left’ for truth, then, we have to abandon the world of propositions: we have to look at the linguistic act of which it is a part, and in which it is asserted as true. The ‘proposition’, then, only exists insofar as we extricate it from an act of language use at a specific moment in time. If what is said to be true or false can only be determined from the act of assertion, the actual truth of what is said to be true cannot be determined in abstraction from the act of assertion. This is our second argument: there is ultimately no locus for truth, except in the act of assertion itself, as a concrete event in time and space. And in that act itself, truth is foundational and inescapable. It is well true that, in order to agree or disagree,


327

speakers need to extricate propositions from the speech acts in which they are configured, and to make them sufficiently independent of these acts, so the proposition can remain the same even if speakers disagree on their truth value. Nonetheless, mathematics is special in its relative lack of sensitivity to the deictic anchoring of propositions in speech acts; in ordinary discourse, such independence is harder to achieve. We might call this a linguistic approach to realism. It says that in virtue of having thoughts that have been grammaticalized, we have minds of a kind that can, using language, assert things that are, by and large, also true, for there is nothing that grounds this practice, nothing that we can evaluate it against: the notion of truth against which we would evaluate truths that we assert, is the same old notion of truth, and no more foundational than it. The truth that is an inherent aspect of language use already sets the standard of what counts as real. Minds of such a kind are propositional and logical ones: they are rational. Perhaps there is no option yet, biologically, for the opposite. The opposite—disorganization, systematic irrationality, and systematic falsehood—is seen only very rarely and when the biology and structure of the brain have radically changed: for example, in the case of thought disorder in the schizophrenia syndrome. If so, a theory of the curious rationality of our species is a biological fact: its basis can be investigated in biological terms. It begins with the emergence of lexemes, which, if we are right about them in Chapter 2, are ultimately selections from features describing percepts, which are causally controlled, even if attention is not. Whatever referential act we may construct in grammar, the same lexemes, as ultimately tied to their experiential origins, will be involved. Lexemes in this sense are one safeguard against human thought evolving into a mere flight of fancy, freeing itself from perceptual reality. Grammar is the other safeguard, as it is designed to configure units of referentiality, and no single derivation can fail to derive, or does ever fail to derive, an act of reference. Even thought taking place in a grammaticalized brain, then, can ultimately only be about the world, or reality, and it defines what reality is, as truth exists in no other domain. So thought is about reality, necessarily: the truth just is reality. In this sense, if truth is a linguistic notion, reality is a linguistic notion as well: Bhartrihari’s conclusion, which we encountered in Chapter 1. This conclusion nonetheless seems to meet the opposition of a long tradition described as supporting ‘realism’, in which the truth is not taken to be ‘languagedependent’, and in which making it language-dependent is said to be an expression of ‘anti-realism’. The fear is a traditional, Fregean one: if the contents of thought were ‘psychological’, how could they be objective? Coming back to the beginning of this chapter, however, we have not been talking psychology, but grammar, and we have endorsed no such view as that grammar is ‘internal’. The existence of grammar requires a sapiens-specific brain, but this does not make grammar internal or nonobjective. Nor is there anything in our account of truth that makes it languagedependent. We have taken the Equivalence Schema as our guide, and that schema

328


says nothing about the language-dependence of truth, and as we have noted, it seems to lend no support to any other non-realist notion of truth either.15 That truth, turns out to be a grammatical notion also does not make it a non-semantic notion: not if we accept the notion of grammatical meaning, and not if we accept that a grammaticalized mind will also develop a model-theoretic competence, in the sense of a logical ability that can take given grammatical meanings as inputs, define model-theoretic equivalence classes over them, and define a relation of logical consequence for them. For example, for such a mind, ‘If a man smokes cocaine, he will get mad’ is made true by the same models as ‘A man who smokes cocaine will get mad’, though these two sentences are grammatically different. A grammatical creature will thus naturally be a logical one, and nothing in our account contradicts the Goedelian insight that a grammaticalized mind will eventually also grasp truths that it cannot deductively derive. All that said, the contents that this logical mind operates with, whose truthconditions are stated schematically in (1), are, on our account, ultimately linguistic in nature. What does this entail for the objectivity of the notion of truth used in characterizing them? Does it entail that the truth of such contents can fall short of reality, driving a wedge between the true and the real? Not if ‘true’ means ‘real’, and the content that we are asserting to be true, when we assert that John left, is that John left. For we are not asserting, then, that it seems that he did, that he may have done so, or that he hasn’t really left. We are not leaving open that reality might be different. We are saying it isn’t different. The notion of reality, therefore, adds nothing to our practice of assertion. For that practice to be out of tune with reality, in turn, as argued above, the conceptual framework that supports it would have to change—impossibly so, if it is biologically rooted rather than merely a conventional aspect of how language happens to be regulated. If no nominalist conception of language is endorsed, nothing then follows for the objectivity of truth if the contents that are said to be true are linguistic in nature.

9.6 Is there a science of thought? We develop our answer to this final question as a critique of an alternative answer, due to the philosopher Donald Davidson, who has developed the project of a ‘science of rationality’, also identified by him under the heading of the ‘Unified Theory of 15 Tarski defined the truth predicate for sentences rather than propositions, and his definitions of such predicates are language-specific. But the insight we have taken from the Equivalence scheme is independent of that feature, as we have not attempted to define truth at all (see also Horwich, 2010). Even in Tarski’s case, the condition for any such predicate to hold of a sentence is not language-relative. Moreover, as Davidson (2005) points out, any attempt to empirically test a given ‘T-theory’ (which derives instances of the Equivalence Scheme for a given language) invokes a general truth predicate (which Tarski did not define).


329

Thought, Meaning and Action’ (Davidson, 2004: 151 ff.). This Davidsonian theory does not identify the science of thought and the science of language, as we effectively do, following Müller’s (1887) early similar proposal. It is also not naturalistic: Davidson, as noted in Section 9.1 above, argues for strong limits on the extent to which thought can be investigated biologically, or as part of natural science. The account of thought that he offers is said not to be ‘in competition with any natural science’ (Davidson, 2004: 134). As noted in the beginning, the ‘normativity’ of the mental is primarily responsible for this conclusion: ‘The reason mental concepts cannot be reduced to physical concepts is the normative character of mental concepts’ (Davidson, 2004: 115). That is, the mental is intrinsically described in terms that, by their very nature, will not lend that domain to naturalistic study. In this section we will argue that this conclusion is far too strong: our mind can remain an object of nature, as can language, even if (and maybe especially if ) we expand the scope of grammatical theory beyond the confines traditionally imposed on it, so as to view it as an instrument of thought. Davidson’s theory has an ontology that figures propositional attitudes, on the one hand, and language, on the other. The two are crucially distinct, although they are, on his account, closely linked. Above that, there is an ontology of actions viewed as ‘rational’ by the standards of formal decision theory. Beliefs, Davidson proposes, never occur in isolation. They are what they are only in the context of other beliefs, and of other propositional attitudes such as desires and intentions. They must form a rational pattern: in particular, it is in the nature of beliefs or thoughts to form logical relations among one another, and a given belief wouldn’t be the belief it is, if it wasn’t part of the same logical network. A radically irrational agent, therefore, wouldn’t have the belief in question. Having a belief at all, then, is to be a rational agent. Irrationality arises against a background of rationality, as a departure from a norm. Just as beliefs we hold must be largely rational, the account maintains that they must also largely be true—in addition to being assumed or taken to be true. What makes our belief that a snake is approaching a belief about a snake approaching, and hence the belief it is, for example, is that what we believe about snakes is largely true. If we believed that snakes were evil spirits or branches of trees with a motor built in to allow movement, we wouldn’t have the belief that we have just described as the belief that a snake is approaching. It wouldn’t be a belief about snakes. Concepts we have of things are, in this sense, a warrant for a degree of truth: for they necessarily form a network, a conceptual structure, and they are parts of beliefs, which in turn must be largely true for these concepts to be what they are. Thought in the human format requires a grasp of the notion of truth, which is mastered only if it is distinguished from belief, which in turn means that a notion of belief (or thought) is required as well. A creature that has the kind of thought in question, therefore, also has notions of belief and truth—as manifest, for example, in any such creature’s capacity for surprise (Davidson, 1982). Surprise that the sun does

330


not set, when I thought it would, shows that I realize that I thought it would set, and yet it didn’t set, making my thought false. Such thought, Davidson argues, requiring as it does a notion of truth, also requires language—an argumentative move that illustrates the ontological distinctness of the propositional attitudes and of language in Davidson’s account, as much as their connectedness. Language as shared with a second person is, on this account, though maybe not the only way in which such a notion can arise, a context in which it arguably can arise. But what is it about language on this account that makes such a peculiar notion as the notion of truth possible? Clearly, it cannot merely be communication with a second person, which will enable such a notion, for we communicate with our pets, too, and on Davidson’s own account, these pets do not think, or not in the same way. Moreover, much of our communication is non-linguistic. So linguistic communication must be the issue—yet again, what is it about language, on this account, which gives it this special role of providing the ontological framework in which thought can arise? It simply doesn’t suffice to think of it as a shared commodity, or a symbol system—indeed symbols (but not thoughts, on this account) are what we share with the ‘language-trained’ apes. The answer, within Davidson’s framework, relates to the fact that language has content, just as beliefs do. These two, in fact, inherently connect on this view: working out the beliefs of a person depends on understanding what he says; but understanding what he says systematically depends on assumptions about what he believes. This again illustrates the distinctness, yet connectedness, of language and the propositional attitudes in Davidson’s theoretical ontology. So words have meanings, and these intrinsically connect with the beliefs that the speaker of the language expresses. Using this assumption in a situation of ‘radical interpretation’, where we know nothing about the meanings of the words that the speaker uses and attempt to figure them out, we will make observations about his preferences among (for us) uninterpreted sentences of his language, in given circumstances that we observe and where we assume he will hold particular beliefs. From patterns of preferences of some sentences over others, interpreted by Davidson as preferences for one sentence to be true in a given context rather than another, we will then deduce further structure among his beliefs and desires. These beliefs and desires are the basis for assigning meanings to his words and utterances: for interpreting him. We assume, in short, that the meaning of what he says is a reflection of the contents of his beliefs; these beliefs come with a notion of truth; and behaviour is interpreted as reflecting preferences for the truth of one sentence over another. Davidson’s position, in this regard, despite his insistence on the language-dependence of thought, is a conservative and Cartesian one: although thought is said to depend on language, truth is a property primarily of the contents of thought viewed as a propositional attitude, not of language. Language may enable, but it cannot explain, why we think thoughts with a truth-theoretic content. In line with that, the notion of truth remains a


331

primitive in Davidson’s theory of rationality, and it becomes the most fundamental notion in his theory of meaning, which in turn takes it for granted. That theory takes the form of a systematic theory of truth for the speaker’s utterances, which we develop as we progress in our attempt to figure out what he believes as he chooses expressions. In this attempt, we will necessarily have to assume—rather than waiting for evidence that this is the case—that these beliefs form a logical pattern, taking our own standards of logicality and rationality for granted, as we must. This necessary assumption—which Davidson calls charity—is one crucial limit that he sees with regard to the extent to which a theory of thought can be naturalistic: our subject matter is regarded as satisfying a norm of rationality from the start, which is automatically factored into the process of radical interpretation as just described. In line with the assumption, the sentences held true by the speaker will be regarded as forming a pattern, which allows us to start ascribing a composite structure to his utterances (called ‘logical form’), which the theory of truth for these utterances will take as a basis to state truth conditions for them recursively. Eventually, his language will thus reflect a grammar (though it is, essentially, a logic), and along with it, it will exhibit productivity, compositionality, and discrete infinity. As we add quantificational apparatus, it will eventually come to involve what Davidson calls an ontology, which, he suggests, provides a ‘framework for thought’ (Davidson, 2004: 140). The mature form that the formal theory of the meanings of the speaker’s utterances takes is that of a Tarskian ‘T-theory’, which, based on axioms about word meaning, grammatical constituent structure, and logical apparatus, deductively derives T-conditions in the format of (1) recursively, with ‘T’ interpreted as truth, for every given expression of the person’s language. We adjust this theory as needed, sensitive to how our insights into the person’s beliefs grow. The theory at any given stage is tested against our decision-theoretic account of the speaker, assuming him to be a rational agent, and it has empirical content in this sense. Time and again, Davidson has returned to the question of whether, and in what sense, this ‘Unified Theory of Thought, Meaning and Action’ is naturalistic, or whether the science of rationality he proposes is or can be science: ‘Is thought, as I have described it, amenable to scientific study?’ (Davidson, 2004: 145). One limit that he sees, as mentioned above, is the necessary assumption that the agent is rational, where what we consider rational will set the standard for what he does: What makes the empirical application of decision theory or formal semantics possible is that the norms of rationality apply to the subject matter. In deciding what a subject wants or thinks or means, we need to see their mental workings as more or less coherent if we are to assign contents to them. ( . . . ) We cannot compare our standards with those of others without employing the very standards in question. (Davidson, 2004: 148) People are as publicly observable as anything else in nature, but the entities we use to construct a picture of someone else’s thoughts must be our own sentences, as understood by us

332


( . . . ). The meanings of our sentences are indeed dependent on our relations to the world which those sentences are about, and our linguistic interactions with others. But there is no escape from the fact that we cannot check up on the objective credentials of the measure we are using ( . . . ); we cannot check up on the objective correctness of our own norms by checking with others, since to do this would be to make basic use of our own norms once more. (Davidson, 2004: 134)

Rationality, therefore, is never an objective fact of the world. It is a standard against which interpretation, as described above, proceeds. Hence there is no objective science of rationality. Does the Un-Cartesian view improve on this predicament? There is one resounding omission in Davidson’s scheme of thought, whose core ingredients are: (i) decision theory; (ii) a Tarskian T-theory interpreted as a theory of meaning; (iii) logic, including a formal-semantic notion of ‘logical form’. What is missing here is nothing other than language, and grammar more specifically. Viewed as ‘syntax’ in the autonomist and internalist tradition, Davidson can be forgiven for not regarding the investigation of what we have called the linguistic genotype and its expression in the human brain as having any intrinsic significance to his project or philosophy altogether (cf. Davidson, 2004: 132–3). Such ‘internalist’ investigations, he contends, and in line with the form-content dichotomy discussed in Section 9.1, do not address the crucial philosophical conundrum: the nature of ‘content’, and what it is to be able to make sense of other beings in linguistic interactions. But this, of course, is from our point of view simply the reflection of a Cartesian perspective: it reflects the view that linguistic form as studied in generative linguistics is just that, form, and content is separate from and external to it. If thought is not merely language-dependent (Davidson’s contention), but language transforms thought, the picture changes: language can (and must) inform content, and the empirical investigation of language does speak to the philosopher’s concern. The omission of what grammar does in relation to the human mind is, we submit, why the question of how language turns the trick of yielding a sense of truth is ultimately left unanswered by Davidson. Although, crucially, thought is said to be language-dependent by him, and he brands himself as ‘almost totally non-Cartesian’ (Davidson, 2004: 17), Davidson crucially assumes a dual ontology invoking both a domain of thought and of language, as noted, with different interlinked components of his overall theory of rationality targeting these, respectively: a version of Ramsey’s decision theory for the propositional attitudes, and a Tarskian T-theory for the meanings of linguistic expressions. Grammar, as an independent explanatory factor, does not come into view. It is because of that, we contend, that his unified theory offers no account of truth, no account of belief, and no account of language. Belief is presupposed insofar as it is a condition for interpretation. In this role they must be assumed to be coherent, largely shared, and largely correct. Language is presupposed insofar as it is a necessary condition for thought. But as noted, it is not clear what it is


333

about language that allows for propositional thought, for we cannot say that truth arises with language, as it doesn’t do on this account, since the account lacks a notion of grammatical meaning. Truth, rather, is simply written into the semantic component as its most basic and primitive notion, and it is presupposed to be a property of the contents of propositional attitudes that a rational language user expresses in what he says. Clearly, this account is not consistent with the contention that language, by intruding into the hominin brain, transforms the cognitive mind and yields a novel format for thought, opening a new world. The consequence of this Un-Cartesian move is to radically diminish the role of the decision-theoretic component of Davidson’s account of rationality, which is based on a belief-desire psychology that has played essentially no role in this book: this component is where, on his account, all content effectively comes from, and from which linguistic meaning is derivative. This way of setting things up has not only the disadvantage of overlooking that an account of grammar is an account of (one kind of) content (unless the existence of grammar is completely irrelevant to what kinds of meanings exist in this species, contrary to evidence suggesting the opposite). It also overlooks that no account of belief can be an account of grammar and grammarconditioned effects on meaning, since how grammar organizes meaning largely does not exploit decision-theoretic principles, or what we consider rational in light of what we believe and desire. Grammatical meaning is not a matter of rational choice. We may find it hard to believe that ‘The hiker who lost was kept walking in circles’ is what somebody expresses in an utterance—it would be more rational to hear ‘The hiker who was lost kept walking in circles’—yet the decision isn’t ours, but grammar’s (Berwick et al., 2011). If an expression means what it does in virtue of its grammar, this is what is the case and we cannot change that. Any assertion does express an attitude, but which attitude it expresses is not a matter of a hearer’s rational choice, suggesting that the rational theory of decision is not the right framework to understand the nature of linguistic meaning (though choices among expressions can no doubt illuminate or reveal, and in this sense allow us to study, what meanings we take a linguistic expression to have). We choose what we say, not what grammatical expressions mean, or how grammar organizes meaning in our brains. By organizing meaning in our brain in novel and propositional ways, however, grammar does make us rational: it makes us thinkers, in very much Davidson’s sense. We obtain, therefore, an account of a new cognitive phenotype from an account of the linguistic genotype, on an empirical basis. In this sense, our account of thought is naturalistic—it takes our cognitive phenotype as an object in nature that we can study as an expression of its linguistic genotype. It is only in this way, we contend, that the fine structure of normativity can be understood and grounded. Far from the cognitive mind failing to be a source of normativity, it is, ultimately, the only such source. Normativity comes in several varieties, not all of which are language-dependent. One form of normativity is

334


conventional, like driving on the left in Britain or saying ‘rain’ for rain rather than ‘Regen’, as in German. Such conventional arrangements lead to normativity in the sense that they restrict our behaviour and we accept such restrictions for the sake of avoiding sanctions. But we nonetheless perceive such conventions as arbitrary, and in exactly this sense, community norms such as these have no normative force: there is nothing right about driving on the left, and nothing right about calling rain ‘rain’ rather than ‘Regen’ (Germans are not wrong for making the opposite decisions). Normativity in this first sense thus poses no problems, and other species are capable of developing norms in this sense as well. The same applies to social norms such as are imposed by dominance hierarchies in baboons based on kinship relations (Wittig et al., 2007): it is not humanly specific, grounded in biology and perception of social reality, and does not pose any particular philosophical problems.16 A third sense of normativity is again perceptual, but is more interesting by reflecting constraints unconsciously imposed by the organization of our mind on things in various cognitive domains. Thus, we hear a melody played backwards as being ‘wrong’ somehow, and the same applies to when we hear a grammatical sentence played backwards. Some cross-cultural data suggests universals of music perception across otherwise fundamentally different musical genres (Hauser and McDermott, 2003), and some studies suggest that aspects of this are innate, with 16 week old babies turning away from speakers playing dissonant chords showing signs of distress, while looking towards speakers smilingly in the consonant case (Trainor and Heinmiller, 1998). In line with such findings, Broca’s area ‘lights up’ on MEG scans when subjects hear a dischordant musical sequence ‘in much the same way as it does when they hear an ungrammatical utterance’ (Maess et al., 2001). And in the domain of language, human infants from a very early age distinguish sentences from two non-native languages in different rhythmic classes—as long as the sentences are played forwards; they are unable to do so if the sentences are played backwards (Jusczyk, 1997). In adults, the brain, confronted with patterns of a new language violating principles of UG, will first turn on Broca’s area, which then however deactivates, reflecting a response to the perception of a ‘wrong’ grammatical pattern (Musso et al., 2003). And where infants perceive languages signed or spoken that do not cohere with principles of full-blown languages, as in pidgins, infants create a richer language, imposing it on the impoverished language present in their perceptual experience, suggesting they perceive the Pidgin they hear as ‘wrong’ somehow: it doesn’t meet a linguistic standard, which their minds impose. Clearly, normativity in these senses, found in a number of cognitive domains, is grounded purely in biology again. A fourth and the most relevant kind of normativity is arguably only found in language, since it depends on a notion of objectivity and truth, and these notions are

16

In this and the following paragraphs I am obliged to Rhiannon Bull.


335

only reflected upon in linguistic creatures. Normativity in this sense, then, is again not due to environmental influences or conventional community norms, but depends on a specific cognitive system, language. It is the result of having a linguistic mind, not what creates such a mind. If we include language within biology—viewing language as a natural object, a subject matter of biology—then this final normativity is again grounded in biology. A grammaticalized mind will distinguish claims from truths. It will reason about why something is false if its truth was assumed or believed. It will hypothesize and test, invoking causes and probabilities. It will thus reason logically, and it will reason morally as well, if it has moral concepts that can be inserted into an emerging logic machine. One can see how a community structure generated by linguistic minds in this sense will exhibit morality and standards of truth. Normativity, then, pace Davidson, is no principled limit to understanding the mind as a natural object. If we begin from a metaphysical premise, and exclude musical and grammatical cognition from the domain of the ‘biological’, talking about biology will not ground any normativity. If we don’t, and the biological, at least in the case of our species, includes the musical and the linguistic, biology not only can ground normativity, but it is not clear what else might do so. The normativity in question is crucially different from conventional ways of restricting behaviour, which by contrast to forms of normativity by the nature of our minds, do not in fact come, we have argued, with genuine normative force. Nonetheless, if we let this conclusion stand like this, it would not be all there is to be said in terms of the possibility of a science of thought. Prelinguistic ‘thought’, if this is the right word, involves cognitive processing that is adaptive—proceeding in tune with the adaptive constraints imposed by the environmental niches that an organism inhabits. If we wanted to develop a prelinguistic notion of truth that applied to organisms in general, it would capture this adaptivity in some way or another. With the grammaticalization of truth, however, things fundamentally change, creating a new framework for thought in which truth itself eventually becomes a subject of reflection. This yields a metaphysics, a concern with the world as such, and what things exist in it. Such thought takes place in the single scheme described above, involving language, the way it is anchored in the speaker’s deictic space, and the world itself in which deictic acts supported by grammar take place. We can, it turns out, explore the grammar of this single scheme, in which thought, language, and world, come together. Needless to say, however, this exploration is not independent of the scheme itself. Thus it cannot be foundational, in the sense that it would tell us how grammar itself, or the framework for thought, arises from a deeper foundation. Curiously, then, nothing prevents us from investigating grammar naturalistically, and advancing on the riddle of what changed the hominin brain, creating the neural infrastructure for language, and a new species defining an unprecedented form of

336


culture. But there is no reason to expect, for the reason just given, that the investigation of language at the level of the biology and evolution of the brain will yield concepts that are foundational for grammatical ones. Ontologically, they will not inform the theory of grammar, which they presuppose, even if, empirically, they will.

9.7 Conclusions No account of thought can be that: an account of thought, if it does not address properties that, for millennia, have been associated with thought in philosophical reflections: thought, apart from being in some sense a ‘psychological’ process, also relates to reality, or has content. For this reason, no account of thought can be ‘internalist’, or purely psychological. But this does not mean it cannot be grammatical. It hasn’t been, in much of modern philosophy, because grammar was regarded as extraneous to it. In the previous eight chapters, we have argued that empirical evidence does not support this claim, which we call Cartesian. The organization of grammar systematically reflects the organization of thought, and no system independent of grammar needs to be invoked to explain how thought comes to involve the curious formal distinctions that much of this book has been about: distinctions such as those between substances and individuals, objects and properties, events and propositions, facts and truths, selves and their bodies, appearances and reality. Grammar, looked at with fresh eyes, yields these distinctions, and it should be given credit for generating them rather than merely lending them physical expression—which is what morphophonology does. Yet, for all that, our question about the emergence of a framework of thought, cannot be merely one about some formalontological distinctions: we need an account of how thought involving such distinctions relates to the real world, the world in which grammar is processed. Our response in this chapter has been that grammar, in defining a new framework for thought, comes with truth and the Self factored in. In the case of truth, this is because the use of the word ‘true’ is, by a common assumption in the contemporary philosophy of truth, governed by a certain equation: the Equivalence Scheme. As we have interpreted this scheme, it shows that our sense of truth is ultimately not lexical but grammatical. Our use of the lexeme true just points us back to our use of grammar. In using grammar, the notion of truth, and hence reality, is factored in. It is not as if we make a choice in endorsing it or not: we find ourselves having and using it, as we start using language, and the denial of truth would require a different conceptual framework altogether. As language arises, truth is present, and it will be distinct from appearance, and from belief. The question whether truth as configured in grammar really is truth presupposes the idea of truth, but that idea is no different from the notion of truth that is present in the system whose foundations are here questioned.


337

With a notion of truth and objectivity, minds will operate with a different norm. There is a different standard now for what is true, different from the one that is set by what is taken to be true. So the structure of normativity, which begins with perception and does not depend on language, will change, and a different foundation is laid for human behaviour. Due to its foundational nature, no external grounding can be given for it. Yet it can be investigated in naturalistic terms, if we shed metaphysical presuppositions on what does, and what does not, belong to the realm of the biological, recognizing the mental not only as real, but as setting a normative standard in nature for what else is real.

References Abels, Klaus and Neeleman, Ad. 2012. Linear asymmetries and the LCA. Syntax 15: 25–74. Abner, Natasha. 2010. WH-words that go bump in the right. In M. B. Washburn, K. McKinney-Bock, E. Varis, A. Sawyer, and B. Tomaszewicz (eds.), Proceedings of the 28th West Coast Conference on Formal Linguistics. Somerville, MA, Cascadilla Proceedings Project, 24–32. Aboh, Enoch. 2004. The Morphosyntax of Complement-Head Sequences. Oxford: Oxford University Press. Aboitiz F., García, R. R., Bosman, C., and Brunetti, E. 2006. Cortical memory mechanisms and language origins. Brain & Language 98: 40–56. Abu-Akel, A. 1999. Impaired theory of mind in schizophrenia. Pragmatics & Cognition 7: 247–82. Adger, David. 2011. Bare Resumptives. In A. Rouveret, (ed.), Resumptives at the Interfaces. Elsevier: 343–66. Adger, David and Harbour, Daniel. 2007. Syntax and syncretisms of the Person Case Constraint. Syntax 10: 2–37. Aikhenvald, Alexandra. 2004. Evidentials. Oxford: Oxford University Press. Aklujkar, Ashok. 2009. Veda revelation according to Bhartrhari. In Chaturvedi (ed.), 1–98. Aldridge, Edith. 2004. Ergativity and word order in Austronesian languages. PhD thesis, Cornell University. Aldridge, Edith. 2008. Generative approaches to ergativity. Language and Linguistics Compass 2(5): 966–95. Alexiadou, Artemis and Anagnostopoulou, Elena. 1998. Parametrizing AGR: word order, V-movement and EPP-checking. Natural Language and Linguistic Theory 16: 491–539. Allen, Cynthia L. 1995. Case Marking and Reanalysis: Grammatical Relations from Old to Early Modern English. Oxford: Clarendon Press. de Almeida, R. G. and Libben, G. 2005. Changing morphological structures: The effect of sentence context on the interpretation of structurally ambiguous English trimorphemic words. Language and Cognitive Processes 20: 373–94. Ambridge, B. and Goldberg, A. 2008. The island status of clausal complements: evidence in favor of an information structure explanation. Cognitive Linguistics 19: 349–81. Amodio D. and Frith, C. 2006. Meeting of minds. Nature Reviews Neuroscience 7: 268–77. Anagnostopoulou, E. 2005. Strong and weak person restrictions: A feature checking analysis. In L. Heggie and P. Ordónez (eds.), Clitic and Affix Combinations: Theoretical Perspectives. Amsterdam/Philadelphia, PA: John Benjamins, 199–235. Anderson, Stephen. 1976. On the notion of subject in ergative languages. In C. N. Li (ed.), Subject and Topic. New York: Academic Press, 1–24. Andreasen, N. C. 1986. The scale for assessment of thought, language, and communication (TLC). Schizophrenia Bulletin 12: 473–82.

340

References

Andrews, Avery. 1982. The representation of case in Modern Icelandic. In J. Bresnan (ed.), The mental representation of grammatical relations. Cambridge, MA: MIT Press, 427–503. Angrilli, A., Spironelli, C., Elbert, T., Crow, T. J., Marano, G., and Stegagno, L. 2009. Schizophrenia as failure of left hemispheric dominance for the phonological component of language. PLoS One 4(2): e4507. Antony, Louise and Hornstein, Norbert (eds.). 2003. Chomsky and his critics. Oxford: Blackwell. Apperly, I. A., Samson, D., Carroll, N., Hussain, S., and Humphreys, G. 2006. Intact first- and second-order false belief reasoning in a patient with severely impaired grammar. Social Neuroscience 1(3–4): 334–48. Armoskaite, Solveiga. 2011. The destiny of roots in Blackfoot and Lithuanian. PhD thesis, University of British Columbia. Arnauld, Antoine and Lancelot, Claude. 1660/1676/1966. Grammaire générale et raisonnée ou La Grammaire de Port-Royal, Grammatica Universalis, I. Stuttgart Bad Cannstatt: Frommann-Holzboog. Arnauld, Antoine and Nicole, Pierre. 1662/1861. The Port-Royal logic. Tr. by Thomas Spencer Baynes. Edinburgh: James Gordon. Arsenijević, Boban. 2007. Phases as complete referential expressions. In C. de Cuba and I. Mitrovic (eds.), Novi Sad Generative Linguistics Workshop 2006. Novi Sad: Filozofski fakultet u Novom Sadu. Arsenijević, Boban. 2008. From spatial cognition to language. Biolinguistics 2: 3–23. Arsenijević, Boban. 2009. Clausal complementation as relativization. Lingua 119: 39–50. Arsenijević, Boban and Hinzen, Wolfram. 2012. On the Absence of X-within-X Recursion in Human Grammar. Linguistic Inquiry 43(3): 423–40. Asbury, Anna. 2008. The Morphosyntax of Case and Adpositions. PhD thesis, University of Utrecht. Astington, J. W., and Jenkins, J. M. 1999. A longitudinal study of the relation between language and theory of mind development. Developmental Psychology 35: 1311–1320. Bach, Emmon. 1986. The Algebra of Events. Linguistics and Philosophy 9: 5–16. Balci, F., Freestone, D., and Gallistel, C. R. 2009. Risk assessment in man and mouse. Proceedings of the National Academy of Science USA, 106(7): 2459–2463. Baldo, J. V. and Dronkers, Nina. 2007. Neural correlates of arithmetic and language comprehension: A common substrate? Neuropsychologia 45: 229–35. Baker, Mark C. 1988. Incorporation: a Theory of Grammatical Function Changing. Chicago: University of Chicago Press. Baker, Mark C. 1996. On the structural position of Themes and Goals. In J. Rooryck and L. Zaring (eds.), Phrase Structure and the Lexicon. Dordrecht: Kluwer, 7–34. Baker, Mark C. 1997. Thematic roles and syntactic structure. In Liliane Haegeman (ed.), Elements of Grammar. Dordrecht: Kluwer, 73–137. Baker, Mark C. 2003. Lexical Categories. Cambridge: Cambridge University Press. Baker, Mark C. 2008. The macroparameter in a microparametric world. In T. Biberauer (ed.), The Limits of Syntactic Variation. Amsterdam: John Benjamins, 351–74.

References

341

Baker, Mark C. 2010. Types of Cross-linguistic Variation in Case Assignment. Ms., Rutgers University. Paper presented at Workshop on Variation in the Minimalist Program, and scheduled to appear in a volume edited by Silvia Martinez Ferreirro and Carme Picallo. Bakker, Dik. 2011. Person marking on adpositions. In M. S. Dryer and M. Haspelmath (eds.), The World Atlas of Language Structures Online. Munich: Max Planck Digital Library, feature 48A. Available online at http://wals.info/feature/48A. Accessed on 3 Oct 2012. Barbosa, Pilar. 1995. Null Subjects. PhD thesis, Massachusetts Institute of Technology. Baron-Cohen, Simon. 1995. Mindblindness: an essay on autism and theory of mind. Cambridge, MA: MIT Press. Baron-Cohen, Simon, Leslie, A. M., and Frith, U. 1985. Does the autistic child have a ‘theory of mind’? Cognition 21: 37–46. Barsalou, Larry. 2008. Grounded cognition. Annual Review of Psychology 59: 617–45. Barss, Andrew and Déprez, Viviane. 1986. On extractions and affective triggered inversion. Ms., Massachusetts Institute of Technology. Bartos, H. 2001. Object Agreement in Hungarian: a Case for Minimalism. In G. M. Alexandrova and O. Arnaudova (eds.), The Minimalist Parameter. Amsterdam/ Philadelphia, PA: John Benjamins, 311–24. Bateson, William. 1894. Materials for the study of variation, New York: Macmillian. Bazalgette, Tim, Roberts, Ian G., Sheehan, Michelle, and van der Wal, Jenneke. 2012. Two statistical typological generalisations and their consequences. Paper presented at The Syntax of the World’s Languages (SWL), Dubrovnik, September 2012. Behar, D. M., Villems, R., Soodyall, H., Blue-Smith, J., Pereira, L., Metspalu, E., Scozzari, R., Makkan, H., Tzur, S., Comas, D., Bertranpetit, J., Quintana-Murci, L., Tyler-Smith, C., Spencer Wells, R., and Rosset, S. The Genographic Consortium. 2008. The dawn of human matrilineal diversity. The American Journal of Human Genetics 82(5): 1130–1140. Bejar, Susana and Massam, Diane. 1999. Multiple Case checking. Syntax 2(2): 65–79. Belkadi, Aïcha. 2006. Language impairments in autism: evidence against mind-blindness. SOAS Working Papers in Linguistics 14: 3–13. Bellugi, Ursula, Marks, S., Bihrle, A., and Sabo, H. 1988. Dissociation between language and cognitive function in Williams Syndrome. In D. Bishop and K. Mogford (eds.), Language development in exceptional circumstances. Edinburgh: Churchill Livingstone. Bentzen, K. 2010. Exploring embedded main clause phenomena: The irrelevance of factivity and some challenges from V2 languages. Theoretical Linguistics 36: 163–72. Bentzen, K., G. H. Hrafnbjargarson, Þ. Hróarsdóttir, and A.-L. Wiklund. 2007. The Tromsø guide to the Force behind V2. Working Papers in Scandinavian Syntax 79: 93–118. Benveniste, Emil. 1966. Être et avoir dans leur fonction linguistique. In E. Benveniste: Problèmes de linguistique générale. Paris: Gallimard. Bergman, T. J., Beehner, J. C., Cheney, D. L., and Seyfarth, R. M. 2003. Hierarchical classification by rank and kinship in baboons. Science 302: 1234–1236. Berlin, Brent and Kay, Paul. 1969. Basic Color Terms: Their Universality and Evolution: Berkeley and Los Angeles: University of California Press. Berwick, Robert C. and Chomsky, Noam. 2011. The biolinguistic program: The current state of its development. In Di Sciullo and Boeckx (eds.), 19–41.

342

References

Berwick, Robert C., Pietroski, Paul, Yankama, Berach, and Chomsky, Noam. 2011. Poverty of the Stimulus Revisited. Cognitive Science 35: 1207–1242. Betti, Arianna. 2011. Against Facts. Ms., Vrije Universiteit Amsterdam. Bhatt, R. 2010. Comments on ‘Referential CPs and DPs: An operator movement account’. Theoretical Linguistics 36: 173–7. Bhatt, R. and Yoon, J. 1999. On the composition of comp and parameters of V2. In The proceedings of the tenth West Coast conference on formal linguistics, Stanford, 41–51. Bianchi, Valentina. 2006. On the syntax of personal arguments. Lingua 116: 2023–2067. Biberauer, Theresa. 2008. The Limits of Syntactic Variation. Amsterdam/Philadelphia, PA: John Benjamins. Biberauer, Theresa, Holmberg, Anders, and Roberts, Ian G. 2008. Disharmonic word-order systems and the Final-over-Final-Constraint (FOFC). In A. Bisetto and F. Barbieri (eds.), Proceedings of XXXIII Incontro di Grammatica Generativa (http://amsacta.unibo.it/2397/1/ PROCEEDINGS_IGG33.pdf; accessed 5 March 2012). Biberauer, Theresa, Holmberg, Anders, and Roberts, Ian G. 2009. Linearization and the architecture of grammar: a view from the Final-over-Final Constraint. In V. Moscati and E. Servidio (eds.), StiL—Studies in Linguistics (Proceedings of XXXV Incontro di Grammatica Generativa), 78–91. Biberauer, Theresa, Holmberg, Anders, and Roberts, Ian G. Forthcoming. A syntactic universal and its consequences. Linguistic Inquiry. Biberauer, Theresa, Holmberg, Anders, Roberts, Ian G., and Sheehan, Michelle. 2010. Parametric Variation: Null Subjects in Minimalist Theory. Cambridge: Cambridge University Press. Biberauer, Theresa, Holmberg, Anders, Roberts, Ian G., and Sheehan, Michelle. In progress. The Final over Final Constraint: a Word-Order Universal and its Implications for Linguistic Theory. MIT Press. Biberauer, Theresa, Newton, Glenda, and Sheehan, Michelle. 2009. Limiting synchronic and diachronic variation and change: the Final-Over-Final Constraint, Language and Linguistics 10(4): 701–43. Biberauer, Theresa, Roberts, Ian G., and Sheehan, Michelle. 2013. No choice parameters and the limits of syntactic variation. To appear in Proceedings of WCCFL 31, Cascadilla Press. Biberauer, Theresa and Sheehan, Michelle. 2012. Disharmony, Antisymmetry, and the Finalover-Final Constraint. In M. Uribe-Etxebarria and V. Valmala (eds.), Ways of Structure Building. Oxford: Oxford University Press, 206–44. Biberauer, Theresa and Sheehan, Michelle. Forthcoming. Theoretical Approaches to Disharmonic Word Orders. Oxford: Oxford University Press. Biberauer, Theresa, Sheehan, Michelle, and Newton, Glenda. 2010. Impossible changes and impossible borrowings: the Final-over-Final Constraint. In A. Breitbarth, C. Lucas, S. Watts, and D. Willis (eds.), Continuity and Change in Grammar. Amsterdam: John Benjamins, 35–60. Bickel, Balthazar. 2011. Government and agreement: what’s where why? Paper presented at Explorations in Syntactic Government and Subcategorisation, Cambridge, 1 Sept 2011. Bickerton, Derek. 1990. Language and species. Chicago: University of Chicago Press. Bickerton, Derek. 2009. Adam’s tongue. New York: Hill and Wang.

References

343

Binder, J. R., Frost, J. A., Hammeke, T. A., Bellgowan, P. S., Rao, S. M., and Cox, R. W. 1999. Conceptual processing during the conscious resting state. A functional MRI study. Journal of Cognitive Neuroscience 11(1): 80–95. Bisang, Walter. 2011. Word Classes. In: Jae Jong Song (ed.), The Oxford Handbook of Language Typology. Oxford: Oxford University Press. Bishop, D. V. M. 2003. Test for reception of grammar (Second edition). London: Psychological Corporation. Bleuler, Ernst. 1911/1950. Dementia praecox or the group of schizophrenias. New York: International Universities Press. Bloom, Paul. 2000. Language and thought: Does grammar makes us smart? Current Biology 10 (14): R516–17. Bloomfield, Leonard. 1917. Tagalog Texts with Grammatical Analysis. 3 vols. Urbana, IL: University of Illinois. Bloomfield, Leonard. 1933. Language. New York: Henry Holt. Boas, Franz. 1911. Introduction. In F. Boas (ed.), Handbook of American Indian languages, 1–83. Bureau of American Ethnology Bulletin, 40. Bobaljik, Jonathon. 1992. Nominally absolutive is not absolutely nominative. In J. Mead (ed.), The Proceedings of the Eleventh West Coast Conference on Formal Linguistics. Stanford: Stanford Linguistics Association, 44–60. Bobaljik, Jonathon. 1993. On ergativity and ergative unergatives. In C. Phillips (ed.), Papers on Case and Agreement II. MITWPL, 45–88. Bobaljik, Jonathon. 2008. Where’s Ö? Agreement as a post-syntactic operation. In D. Adger, S. Béjar, and D. Harbour (eds.), Phi-Theory: Phi Features Across Interfaces and Modules. Oxford: Oxford University Press, 295–328. Bobaljik, Jonathan and Landau, Idan. 2009. Icelandic Control Is Not A-Movement: The Case from Case. Linguistic Inquiry, 40(1): 113–54. Boeckx, Cedric. 2008. Bare syntax, Oxford: Oxford University Press. Boeckx, Cedric. 2010. Defeating Lexicocentrism. Ms. To appear as Elementary Syntactic Structures: Prospects of a Feature-Free Syntax. Cambridge, Cambridge University Press. Boeckx, Cedric. 2011. Approaching parameters from below. In Di Sciullo and Boeckx (eds.), 205–21. Boeckx, Cedric (ed.). 2011. The Oxford handbook of linguistic minimalism. Oxford: Oxford University Press. Boeckx, Cedric 2012. Considerations pertaining to the nature of logodiversity, or How to construct a parametric space without parameters. http://ling.auf.net/lingbuzz/001453. Accessed 25 April 2013. Bonet, E. 1991. Morphology after syntax: Pronominal clitics in Romance. PhD thesis, Massachusetts Institute of Technology. Bonet, E. 1995. The feature structure of Romance Clitics. Natural Language and Linguistic Theory 13: 607–47. Bonet, E. 2008. The Person Case Constraint and repair strategies. In R. D’Alessandro, S. Fischer, and G. H. Hrafnbjargarson (eds.), Person Restrictions. Berlin: de Gruyter, 103–28. Borer, H. 1984. Parametric Syntax: Case Studies in Semitic and Romance Languages. Dordrecht: Foris.

344

References

Borer, Hagit. 2005. The Normal Course of Events: Structuring Sense, Vol. II. Oxford: Oxford University Press. Borer, Hagit. 2012. In the Event of a Nominal. In M. Everaert, M. Marelj, and T. Siloni (eds.), The Theta System: Argument Structure at the Interface. Oxford: Oxford University Press. Borg, Emma. 2012. Pursuing meaning. Oxford: Oxford University Press. Borofsky, J. A., McNealy, K., Siddarth, P., Wu, K. N., Dapretto, M., and Caplan, R. 2010. Semantic processing and thought disorder in childhood-onset schizophrenia. Journal of Neurolinguistics 23: 204–22. Bouzouggar, A., Barton, N., Vanhaeren, M., d'Errico, F., Collcutt, S., Higham, T., Hodge, E., Parfitt, S., Rhodes, E., Schwenninger, J.-L., Stringer, C., Turner, E., Ward, S., Moutmir, A., and Stambouli, A. 2007. 82,000-year-old shell beads from North Africa and implications for the origins of modern human behavior. PNAS 104(24): 9964–9969. Bowers, John. 2011. Arguments as relations. Cambridge, MA: MIT Press. Bregman, A. S. 1990. Auditory Scene Analysis. Cambridge, Massachusetts: MIT Press. Bresnan, Joan. 1977. Variables in the theory of transformations. In P. Culicover, T. Wasow, and A. Akmajian (eds.), Formal Syntax. New York: Academic Press, 157–96. Brock, Jon. 2007. Language abilities in Williams syndrome: A critical review. Development and Psychopathology 19: 97–127. Broschart, Jürgen. 1997. Why Tongan does it differently: Categorial distinctions in a language without nouns and verbs. Linguistic Typology 1: 123–65. Brown, Cecil H. 2011. Finger and Hand. In M. S. Dryer and M. Haspelmath (eds.), The World Atlas of Language Structures Online. Munich: Max Planck Digital Library, chapter 130. Available online at http://wals.info/chapter/130. Accessed on 20 Dec 2012. Bruene, Martin. 2005. ‘Theory of Mind’ in Schizophrenia. Schizophrenia Bulletin 31(1): 1–22. Brunet, E., Sarfati, Y., Hardy-Bayle, M. C., and Decety, J. 2003. Abnormalities of brain function during a nonverbal theory of mind task in schizophrenia. Neuropsychologia 41: 1574–1582. Brucart, Josep M. 1993. Sobre la estructura de SComp en español. In Amadeu Viana (ed.), Sintaxi: Teoria i perspectives. Lleida: Pagès, 59–102. Bruzina, Ronald. 1981. Dependence on language and the autonomy of reason: an Husserlian perspective. Man and World 14: 355–68. Buehler, Karl. 1934. Sprachtheorie. Jena: G. Fischer. Burge, Tyler. 1973. Reference and Proper Names. Journal of Philosophy 70(14): 425–39. Bursill-Hall, G. L. 1971. Speculative grammars of the Middle Ages. The doctrine of partes orationis of the Modistae. The Hague and Paris: Mouton. Burton-Roberts Noel. 2011. On the grounding of syntax and the role of phonology in human cognition. Lingua 121(14): 2089–2102. Burzio, Luigi. 1981. Intransitive verbs and Italian auxiliaries. PhD thesis, Massachusetts Institute of Technology. Burzio, Luigi. 1986. Italian Syntax: a Government-Binding Approach. Dordrecht: D. Reidel Pub. Co. Butterworth, G. 2003. Pointing is the royal road to language for babies. In S. Kita (ed.), Pointing: Where language, culture, and cognition meet. Hillsdale, NJ: Lawrence Erlbaum, 9–33. Caha, P. 2009. The nanosyntax of case. PhD dissertation, Universitetet Tromsø.

References

345

Calabrese, Andrea. 1996. Some remarks on the Latin case system and its development in Romance. In J. Lema and E. Trevino (eds.), Theoretical analyses on Romance languages: Selected papers from the 26th linguistic symposium on Romance languages. Amsterdam: John Benjamins, PP. Campbell, John. 2001. Rationality, meaning, and the analysis of delusion. Philosophy, Psychiatry, and Psychology 8(2–3); 89–100. Campos, H. 1992. Enunciative elements in Gascon. Linguistics 30: 911–40. Cardona, George. 2009. Bhartrhari and Patanjali: Traditions preserved. In Chaturvedi (ed.), 119–62. Cardinaletti, Anna and Starke, Michal. 1999. The typology of structural deficiency: A case study of the three classes of pronouns. In H. van Riemsdijk (ed.), Clitics in the Languages of Europe. Berlin: Mouton, 145–233. Carey, Susan. 2009. The origin of concepts. Oxford: Oxford University Press. Carnie, Andrew. 2005. Some remarks on markedness Hierarchies: A reply to Aissen 1999 and 2003. Coyote Working Papers in Linguistics 14: 37–50. Carpenter, M., Pennington, B. F., and Rogers, S. J. 2001. Understanding of others’ intentions in children with autism and children with developmental delays. Journal of Autism and Developmental Disorders 31: 589–99. Carpenter, M., and Tomasello, M. 2000. Joint attention, cultural learning, and language acquisition: Implications for children with autism. In A. M. Wetherby and B. M. Prizant (eds.), Communication and language issues in autism and pervasive developmental disorder: A transactional developmental perspective. Baltimore, MD: Brookes, 31–54. Carstairs-McCarthy, Andrew. 1999. The Origins of Complex Language. Oxford: Oxford University Press. Carstens, Vicky. 2012. Delayed Valuation: a reanalysis of ‘upwards’ complementizer agreement and the mechanics of Case. Ms. http://ling.auf.net/lingbuzz/001432. Accessed 1 Nov 2012. Cecchetto, Carlo, Geraci, Carlo, and Zucchi, Sandro. 2006. Strategies of relativization in Italian Sign Language. Natural Language and Linguistic Theory 24: 945–75. Chafe, Wallace. 1994. Discourse, Consciousness and Time: The Flow and Displacement of Conscious Experience in Speaking and Writing. Chicago: University of Chicago Press. Chaika, E. 1974. A linguist looks at schizophrenic language. Brain and Language 1: 257–76. Chaturvedi, M. (ed.). 2009. Bhartrhari: Language, thought and reality. Delhi: Motilal Banarsidass Publishers. Cheney, Dorothy L. and Seyfarth, Robert M. 1990. How Monkeys See The World: Inside The Mind Of Another Species. Chicago: University of Chicago Press. Cheney, Dorothy L. and Seyfarth, Robert M. 1998. Why monkeys don’t have language. In: G. Petersen (ed.), The Tanner Lectures on Human Values, 19. Salt Lake City: University of Utah Press. Cheng, Lisa and Sybesma, Rint. 1999. Bare and not so bare nouns and the structure of NP. Linguistic Inquiry 30: 509–42. Chierchia, Gennaro. 1998. Reference to Kinds across Languages. Natural Language Semantics 6: 339–405. Chomsky, Noam. 1956. Three models for the description of language. IRE Transactions on Information Theory IT-2 2: 113–24.

346

References

Chomsky, Noam. 1957. Syntactic structures. Gravenhage: Mouton. Chomsky, Noam. 1959. A review of B. F. Skinner’s Verbal Behaviour. Language 35: 26–58. Chomsky, Noam. 1965. Aspects of the Theory of Syntax. Cambridge, MA: MIT Press. Chomsky, Noam. 1966. Cartesian Linguistics: A Chapter in the History of Rationalist Thought. New York: Harper & Row. Chomsky, Noam. 1977. On Wh-Movement. In P. W. Culicover, T. Wasow, and A. Akmajian (eds.), Formal Syntax. New York: Academic Press, 71–132. Chomsky, Noam. 1981. Lectures on Government and Binding. Dordrecht and Cinnaminson: Foris Publications. Chomsky, Noam. 1986. Knowledge of language: its nature, origins, and use. New York: Praeger. Chomsky, Noam. 1995. The Minimalist Program. Cambridge, MA: MIT Press. Chomsky, Noam. 2000a. New horizons in the study of language and mind. Cambridge: Cambridge University Press. Chomsky, Noam. 2000b. Minimalist Inquiries. In R. Martin, D. Michaels, and J. Uriagereka (eds.), Step by Step: Essays on Minimalist Syntax in Honor of Howard Lasnik. Cambridge, MA, MIT Press, 89–155. Chomsky, Noam. 2001. Derivation by phase. In Michael Kenstowicz (ed.), Ken Hale: A Life in Language. Cambridge, MA: MIT Press, 1–52. Chomsky, Noam. 2003. Reply to Lycan. In Antony and Hornstein (eds.), 255–62. Chomsky, Noam. 2004. Beyond Explanatory Adequacy. In A. Belletti (ed.), Structures and Beyond. Oxford: Oxford University Press, 104–31. Chomsky, Noam. 2005. Three factors in language design. Linguistic Inquiry 36: 1–22. Chomsky, Noam. 2007a. Approaching UG from below. In U. Sauerland and H.-M. Gärtner (eds.), Interfaces + Recursion = Language? Chomsky’s Minimalism and the view from syntaxsemantics. Berlin and New York: Mouton de Gruyter, 1–29. Chomsky, Noam. 2007b. Of minds and language. Biolinguistics 1: 1009–1027. Chomsky, Noam. 2008. On Phases. In Freidin et al. (eds.), 133–66. Chomsky, Noam. 2009. Conclusion. In Piattelli-Palmarini (ed.), 379–410. Chomsky, Noam. 2010. Some simple evo devo theses. How true might they be for language? In: Larson et al. (eds.), 45–62. Chomsky, Noam. 2013. Problems of projection. Ms. Massachusetts Imstitute of Technology. Chomsky, Noam and Lasnik, Howard. 1993. The theory of principles and parameters. In J. Jacobs, A. von Stechow, W. Sternefeld, and T. Vennemann (eds.), Syntax: An international handbook of contemporary research, vol. 1. Berlin: Walter de Gruyter, 506–69. Christiansen, Morten and Chater, Nick. 2008. Language as shaped by the brain. Behavioral and Brain Sciences 31: 489–558. Chung, M., Fulford, K., Graham, G. (eds.). 2007. Reconceiving schizophrenia. Oxford: Oxford University Press. Cinque, Guglielmo. 1999. Adverbs and Functional Heads: a Cross-Linguistic Perspective. New York: Oxford University Press. Cinque, Guglielmo. 2005. Deriving Greenberg’s Universal 20 and its exceptions. Linguistic Inquiry 36(3): 315–32. Cinque, Guglielmo. 2006. Restructuring and functional heads. Oxford: Oxford University Press.

References

347

Cinque, Guglielmo. 2010. The Syntax of Adjectives: a Comparative Study. Cambridge, MA: MIT Press. Clahsen, Harald and Ali, Mohammad. 2009. Formal Features in Aphasia: Tense, Agreement, and Mood in English agrammatism. Journal of Neurolinguistics 22: 436–50. Clayton, N. S., Bussey, T. J., and Dickinson, A. 2003. Can animals recall the past and plan for the future? Nature Reviews Neuroscience 4(8): 685–91. Clowry, Gavin, Molnar, Z., and Rakic, P. 2010. Renewed focus on the developing human neocortex. Journal of Anatomy 217: 276–88. Collins, John. 2011. The unity of linguistic meaning. Oxford: Oxford University Press. Contini-Morava, Ellen. 1997. Noun classification in Swahili: a cognitive-semantic analysis using a computer database. In R. K. Herbert (ed.), African Linguistics at the Crossroads. Cologne, Rüdiger Köppe, 599–628. Coon, Jessica, Mateo Pedro, Pedro, and Preminger, Omer. 2011. The role of Case in A-Bar extraction asymmetries: evidence from Mayan. Ms. http://ling.auf.net/lingbuzz/001401. Accessed 5 Feb 2013. Coon, Jessica and Preminger, Omer. 2011. Transitivity in Chol: A new argument for the split VP hypothesis. In L. Fainleib, N. LaCara, and Y. Park (eds.), Proceedings of NELS 41. Coon, Jessica and Preminger, Omer. 2012. Taking ‘ergativity’ out of split ergativity: A structural account of aspect and person splits. Ms. http://ling.auf.net/lingbuzz/001556. Accessed 30 June 2012. Cottingham, John. 1998. ‘The only sure sign . . . ’: Thought and language in Descartes. In J. Preston (ed.), Thought and language, Cambridge: Cambridge University Press, 29–50. Covington, Michael A. 2009. Syntactic theory in the high middle ages. Cambridge: Cambridge University Press. Covington, Michael A., He, C., Brown, C., Nac, L., McClain, J., Fjordbak, B., Semple, J., and Brown, J. 2005. Schizophrenia and the structure of language. Schizophrenia Research 77: 85–98. Crain, Stephen and Khlentzos, Drew. 2010. The Logic Instinct. Mind and Language 25(1): 30–65. Croft, William. 1991. Syntactic Categories and Grammatical Relations: The Cognitive Organization of Information. Chicago: University of Chicago Press. Croft, William. 2001 Radical Construction Grammar: Syntactic Theory in Typological Perspective. Oxford: Oxford University Press. Croft, William. 2007. The origins of grammar in the verbalization of experience. Cognitive Linguistics 18(3): 339–82. Crow Tim J. 2000. Schizophrenia as the price that Homo sapiens pays for language. Brain Research Reviews 31: 118–29. Crow, Tim J. (ed.). 2002a. The speciation of modern Homo sapiens. Oxford: Oxford University Press. Crow, Tim J. 2002b. Handedness, language lateralisation and anatomical asymmetry: relevance of protocadherinXY to hominid speciation and the aetiology of psychosis. British Journal of Psychiatry 181: 295–7. Crow, Tim J. 2007. Schneiderʼs Nuclear Symptoms as the Key to the Structure of Language. Neurology, Psychiatry and Brain Research 14: 87–94.

348

References

Crow, Tim J. 2008. The ‘big bang’ theory of the origin of psychosis and the faculty of language. Schizophrenia Research 102(1–3): 31–52. de Cuba, C. 2007. On (Non) Factivity, Clausal Complementation and the CP-Field. PhD dissertation, Stony Brook University. Cuervo, C. 2003. Datives at large. PhD dissertation, Massachusetts Institute of Technology. Culicover, Peter. 1993. Evidence Against ECP Accounts of That-t Effect. Linguistic Inquiry 24: 557–61. Dally, J. M., Emery, N. J., and Clayton, N. S. 2006. Food-caching western scrub-jays keep track of who was watching when. Science 312(5780): 1662–1665. Danon, g. 2006. Caseless nominals and the projection of DP. Natural Language and Linguistic Theory 24: 977–1008. Davidson, Donald. 1967. The Logical Form of Action Sentences. In Davidson (2001), 105–21. Davidson, Donald. 1970/2001. Events as Particulars. Reprinted in Davidson (2001), 181–8. Davidson, Donald. 1982. Rational Animals. Dialectica 36: 318–27. Davidson, Donald. 1984/1991. Inquiries into Truth and Interpretation. Oxford: Clarendon Press. Davidson, Donald. 2001. Essays in actions and events. Second edition. Oxford: Oxford University Press. Davidson, Donald. 2004. Problems of rationality. Oxford: Oxford University Press. Davidson, Donald. 2005. Truth and predication. Cambridge, MA: Belknap Press of Harvard University. Déchaine, R. M. and Wiltschko, M. 2002. Decomposing pronouns. Linguistic Inquiry 33: 409–42. DeLisi, Lynn E. 2001. Speech disorder in schizophrenia: review of the literature and exploration of its relation to the uniquely human capacity for language. Schizophrenia Bulletin 27: 481–96. Demirdache, Hamida and Uribe-Etxebarria, M. 1998. The syntax of temporal relations: a uniform approach to tense and aspect. WCCFL 16: Proceedings of the Sixteenth West Coast Conference on Formal Linguistics, 145–59. Demonte, Violeta 1995. Dative alternation in Spanish. Probus 7: 5–30. Demonte, Violeta and Fernández-Soriano, Olga. 2009. Force and finiteness in the Spanish complementizer system. Probus 21: 23–49. Den Besten, Hans. 1981. On the interaction of Root Transformations and lexical deletive rules. Groninger Arbeiten zur germanistischen Linguistik 20: 1–78. den Dikken, Marcel. 2010. On the functional structure of locative and directional PPs. In Guglielmo Cinque and Luigi Rizzi (eds.), Mapping Spatial PPs: The Cartography of Syntactic Structures, Volume 6. Oxford: Oxford University Press, 74–126. d’Errico, Franceso, Vanhaeren, Marian, Henshilwood, Christopher, Lawson, Graeme, Maureille, Bruno, Gambier, Dominique, Tillier, Anne-Marie, Soressi, Marie, and van Niekerk, Karen. 2009. From the origin of language to the diversification of languages. In F. d’Errico and Jean-Marie Hombert (eds.), Becoming eloquent. Amsterdam and Philadelphia, PA: John Benjamins, 13–68. Descartes, René. 1637. Discours de la méthode pour bien conduire sa raison, et chercher la vérité dans les sciences. Leiden.

References

349

de Villiers, Jill. 2007. The interface of language and theory of mind. Lingua 117(11): 1858–1878. de Villiers, Jill and Pyers, J. 2002. Complements to cognition: a longitudinal study of the relationship between complex syntax and false-belief understanding. Cognitive Development, 17: 1037–1060. De Vries, Hans. 1901. Die Mutationstheorie. Leipzig: Verlag von Veit. Diercks, Michael. 2012. Parameterizing Case: Evidence from Bantu. Syntax 15(3): 253–86. Diesing, Molly. 1990. Word order and the subject position in Yiddish. Natural Language & Linguistic Theory 8: 41–79. Diesing, Molly. 1992. Indefinites. Cambridge, MA: MIT Press. Di Sciullo, Anna-Maria. 2005. Asymmetry in Morphology. Cambridge, MA: MIT Press. Di Sciullo, Anna Maria and Cedric Boeckx (eds.). 2011. The Biolinguistic Enterprise: New Perspectives on the Evolution and Nature of the Human Language Faculty. Oxford: Oxford University Press. Ditman, T. and Kuperberg, G. R. 2010. Building coherence: A framework for exploring the breakdown of links across clause boundaries in schizophrenia. Journal of Neurolinguistics 23(3): 254–69. Dixon, Robert M. W. 1994. Ergativity. Cambridge: Cambridge University Press. Dörr, Otto. 2010. Schizophrenia, language and evolution. Actas Esp Psiquiatr 38(1): 1–7. Dollfus Sonia, Razafimandimby, A., Delamillieure, P., Brazo, P., Joliot, M., Mazoyer, B., Tzourio-Mazoyer, N. 2005. Atypical hemispheric specialization for language in righthanded schizophrenia patients. Biological Psychiatry 57: 1020–1028. Dollfus Sonia, Razafimandimby, A., Maiza, O., Lebain, P., Brazo, P., Beaucousin, V., Lecardeur, L., Delamillieure, P., Mazoyer, B., Tzourio-Mazoyer, N. 2008. Functional deficit in the medial prefrontal cortex during a language comprehension task in patients with schizophrenia. Schizophrenia Research 99: 304–11. Donlan, C., Cowan, R., Newton, E. J., and Lloyd, D. 2007. The role of language in mathematical development: evidence from children with specific language impairments. Cognition 103: 23–33. Donnellan, Keith S. 1966. Reference and definite descriptions. The Philosophical Review 75: 281–304. Dowty, David. 1991. Thematic proto-roles and argument selection. Language 67(3): 547–619. Dryer, Matthew S. 1992. The Greenbergian word order correlations. Language 68(1): 81–138. Dryer, Matthew S. 2011a. Order of Object and Verb. In M. S. Dryer and M. Haspelmath (eds.), The World Atlas of Language Structures Online. Munich: Max Planck Digital Library, chapter 83. Available online at http://wals.info/chapter/83. Accessed on 20 Dec 2012. Dryer, Matthew S. 2011b. Position of Interrogative Phrases in Content Questions. In M. S. Dryer and M. Haspelmath (eds.), The World Atlas of Language Structures Online. Munich: Max Planck Digital Library, chapter 93. Available online at http://wals.info/chapter/93. Accessed on 21 Dec 2012. Dryer, Matthew S. 2011c. Position of Polar Question Particles. In M. S. Dryer and M. Haspelmath (eds.), The World Atlas of Language Structures Online. Munich: Max Planck Digital Library, chapter 92. Available online at http://wals.info/chapter/92. Accessed on 21 Dec 2012. Dryer, Matthew S. 2011d. Order of Adposition and Noun Phrase. In M. S. Dryer and M. Haspelmath (eds.), The World Atlas of Language Structures Online. Munich: Max Planck

350

References

Digital Library, chapter 85. Available online at http://wals.info/chapter/85. Accessed on 3 Oct 2012. Dubois, Jacques, Benders, M., and Lazeyras, F. 2010. Structural asymmetries of perisylvian regions in the preterm newborn. Neuroimage 52: 32–42. Edie, James M. 1977. Husserl’s conception of ‘the grammatical’ and contemporary linguistics. In J. N. Mohanty (ed.), Readings in Husserl’s Logical Investigations. The Hague: Martinus Nijhoff, 137–62. Eigsti, I.-M., Bennetto, L., and Dadlani, M. B. 2007. Beyond pragmatics: morphosyntactic development in autism. Journal of autism and developemental disorders 37: 107–1023. Eisenbeiss, Sonja, Bartke, Susanne, and Clahsen, Harald. 2006. Structural and lexical case in child German: Evidence from language-impaired and typically-developing children, Language Acquisition 13(1): 3–32. Elbourne, Paul. 2005. Situations and Individuals. Cambridge, MA: MIT Press. Elbourne, Paul. 2008. Demonstratives as individual concepts. Linguistics and Philosophy 31: 409–66. Eldredge, Niles and Gould, Stephen J. 1972. Punctuated equilibria: an alternative to phyletic gradualism. In T. M. Schopf (ed.), Models in Palaeobiology. San Francisco: Freeman Cooper, 82–115. Emonds, Joseph. 1976. A transformational approach to English syntax. New York: Academic Press. Enç, M. 1991. The semantics of specificity. Linguistic inquiry 22: 1–25. Erteschik-Shir, N. 2007. Information Structure: The Syntax-Discourse Interface. Oxford: Oxford University Press. Estes, Z. and Glucksberg, S. 2000. Interactive property attribution in concept combination. Memory & Cognition 28(1): 28–34. Evans, Nicholas and Levinson, Stephen. 2009. The myth of language universals: language diversity and its importance for cognitive science. Behavioral and Brain Sciences 32: 429–92. Everett, Daniel L. 2005. Cultural constraints on grammar and cognition in Pirahã: Another look at the design features of human language. Current Anthropology 46: 621–46. Fadiga, L., Craighero, L., and D’Ausilio, A. 2009. Broca’s Area in Language, Action, and Music. Annals of the New York Academy of Sciences 1169: 448–58. Feldman, L. B. 2000. Are morphological effects distinguishable from the effects of shared meaning and shared form? Journal of Experimental Psychology: Learning, Memory and Cognition 26(6): 1431–1444. Fernández-Soriano, O. 1993. Los pronombres átonos. Madrid: Taurus. Ferstl, Evelyn and von Cramon, Yves. 2002. What Does the Frontomedian Cortex Contribute to Language Processing: Coherence or Theory of Mind? NeuroImage 17: 1599–1612. Ferst, Evelyn C., Neumann, Jane, Bogler, Carsten, and von Cramon, D. Yves. 2008. The Extended Language Network: A Meta-Analysis of Neuroimaging Studies on Text Comprehension. Human Brain Mapping 29(5): 581–93. Fine, Kit. 1995. Ontological Dependence. Proceedings of the Aristotelian Society, 269–90. Fitch, Tecumseh W. 2005. The evolution of language: a comparative review. Biology and Philosophy 20: 193–230

References

351

Fitch, Tecumseh W. 2010. Three meanings of ‘recursion’: key distinctions for biolinguistics. In Larson et al. (eds.), 73–90. Fodor, Jerry A. 1970. Three reasons for not deriving ‘kill’ from ‘cause to die’. Linguistic Inquiry 1: 429–38. Fodor, Jerry A. 1975. The language of thought. New York: Crowell. Fodor, Jerry A. 1983. Precis of The Modularity of Mind. Behavioral and Brain Sciences 8: 1–5. Fodor, Jerry A. 1990. A theory of content. Cambridge, MA: MIT Press. Fodor, Jerry A. 1998. Concepts. Where cognitive science went wrong. Oxford: Clarendon Press. Fodor, Jerry A. 2000. The mind doesn’t work that way. Cambridge, MA: The MIT Press. Fodor, Jerry A. 2001. Language, thought, and compositionality. Mind and Language 16(1): 1–15. Fodor, Jerry A. 2008. LOT2. The language of thought revisited. Oxford: Oxford University Press. Fodor, Jerry A. and Lepore, Ernest. 2002. The compositionality papers. Oxford: Oxford University Press. Fodor, Jerry A. and Pylyshyn, Zenon W. 1988. Connectionism and cognitive architecture: A critical analysis. Cognition 28: 3–71. Folli, Rafaella and Ramchand, Gillian. 2005. Prepositions and Results in Italian and English: An Analysis from Event Decomposition. In H. Verkuyl, H. De Swart, and A. van Hout (eds.), Perspectives on Aspect. Dordrecht: Kluwer Academic Publishers, 81–105. Ford, A., Singh, R., and Martohardjono, G. 1997. Pace Panini. New York: Peter Lang. Frank, Michael C., Everett, Dan, Fedorenko, E., and Gibson, E. 2008. Number as a cognitive technology: Evidence from Piraha language and cognition. Cognition 108: 819–24. Franklin, A., Drivonikou, G. V., Clifford, A., Kay, P., Regier, T., and Davies, I. R. L. 2008. Lateralization of categorical perception of color changes with color term acquisition. Proceedings of the National Academy of Sciences 47: 18221–18225. Freidin, Robert, Otero, Carlos P., and Zubizarreta, Maria Luisa (eds.). 2008. Foundational Issues in Linguistic Theory: Essays in Honor of Jean-Roger Vergnaud. Cambridge, MA: MIT Press. Frith, Chris D. 1992. The Cognitive Neuropsychology of Schizophrenia. Hove: Lawrence Erlbaum Associates. Frege, Gottlob. 1892. On Sense and Reference, tr. by P. T. Geach. In P. T. Geach and M. Black (eds.), Translations from the Philosophical Writings of Gottlob Frege. Oxford: Blackwell (1966), 56–78. Frege, Gottlob. 1918–19/1956. Der Gedanke: Eine logische Untersuchung. Beiträge zur Philosophie des Deutschen Idealismus I: 58–77. The translation ‘Thought: A logical Inquiry’ by P. Geach appeared in Mind 65(259): 289–311. Fukuda, Shin. 2006. Japanese passives, external arguments and structural Case. San Diego Linguistic Papers 2: 86–133. Futuyma, Douglas J. 1986. Evolutionary Biology. Sunderland, MA: Sinauer. Gallego, Angel. 2012. Introduction. In Gallego (ed.), 9–44. Gallego, Angel (ed.). 2012. Phases—Developing the framework. Berlin and Boston: Mouton De Gruyter.

352

References

Gallistel, Randy. 1998. Symbolic processes in the brain: The case of insect navigation. In D. Scarborough and S. Sternberg (eds.), Methods, models, and conceptual issues. Vol. 4 of Invitation to cognitive science, series ed. D. Osherson. Cambridge, MA: MIT Press, 1–51. Gallistel, C. Randy. 2009. The foundational abstractions. In Massimo Piattelli-Palmarini, Juan Uriagereka, and Pello Salaburu (eds.), Of minds and language. Oxford: Oxford University Press, 58–73. Gallistel, C. Randy. 2010. Review of Carey’s ‘The Origin of Concepts’. Lingua 121(2): 308–12. Gallistel, C. Randy and King, A. P. 2009. Memory and the computational brain. New York: Wiley-Blackwell. Gil, David. 1993. Tagalog semantics. Berkeley Linguistics Society 19: 390–403. Gimona, M. 2006. Protein linguistics—a grammar for modular protein assembly? Nature 7: 68–73. Giorgi, A. 2010. About the speaker. Towards a syntax of indexicality. Oxford: Oxford University Press. Glasel, H., Leroy, F., Dubois, J., Hertz-Pannier, L., Mangin, J. F., Dehaene-Lambertz, G. 2011. A robust cerebral asymmetry in the infant brain: the rightward superior temporal sulcus. Neuroimage 58(3): 716–23. Goldschmidt, R. 1940. The material basis of evolution. New Haven: Yale University Press (repr. 1982). Gordon, P. 2004. Numerical cognition without words: Evidence from Amazonia. Science 306 (5695): 496–9. Gould, Stephen J. 2002. The structure of evolutionary theory. Cambridge, MA: Belknap Press. Grano, Thomas. 2012. Control and restructuring at the syntax-semantics interface. PhD thesis, University of Chicago, Illinois. Graffi, Giorgio. 2001. 200 Years of Syntax. Amsterdam: Benjamins. Griffiths, Timothy D. and Warren, Jason D. 2004. What is an auditory object? Nature Reviews Neuroscience 5: 887–92. Grimshaw, Jane. 1991. Extended projection. Ms., Rutgers University. Grimshaw, Jane. 2010. Main Clauses as Arguments. Ms. Rutgers University. Grodzinsky, Yosef and Santi, A. 2008 The battle for Broca’s region. Trends Cognitive Sciences 12: 474–80. Haddican, William and Holmberg, Anders. 2012. Object Movement Symmetries in British English Dialects: Experimental evidence for a case-based approach. Ms. http://ling.auf.net/ lingBuzz/001408. Accessed 4 Oct 2012. Haegeman, Liliane. 2009. Everything is relative: main clause phenomena and adverbial clauses. A talk presented at the UCL seminar, 6. 5. 2009. Haegeman, Liliane, and Ürögdi, B. 2010. Referential CPs and DPs: An operator movement account. Theoretical Linguistics 36: 111–52. Hagoort, Peter, Baggio, G., and Willems, Roel. 2009. Semantic unification. In M. S. Gazzaniga (ed.), The cognitive neurosciences, 4th ed. Cambridge, MA: MIT Press, 819–36. Halbfass Wilhelm. 1991. Tradition and reflection: Explorations in Indian thought. Albany: State University of New York Press. Hale, Kenneth and Keyser, Samuel J. 2002. Prolegomena to a theory of argument structure. Cambridge, MA: MIT Press.

References

353

Halpert, Clare. 2012. Structural Case and the nature of vP in Zulu. In Proceedings of NELS42. Amherst, MA: GLSA. Hare, Brian, Call, J., and Tomasello, M. 2001. Do chimpanzees know what conspecifics know? Animal Behaviour 61(1): 771–85. Harley, Heidi. 2012. Lexical decomposition in modern syntactic theory. In Werning et al. (eds.), 328–50. Harley, H. 2002. Possession and the double object construction. Source: Linguistic Variation Yearbook 2(1): 31–70. Harley, Heide and Ritter, Elisabeth. 2002. Person and number in pronouns: A feature-geometric analysis. Language 78: 482–526. Hartman, Jeremy, 2011. (Non-)Intervention in A-movement: Some cross-constructional and cross-linguistic considerations. Linguistic Variation 112: 121–48. Haspelmath, Martin. 2007. Pre-established categories don’t exist—consequences for language description and typology. Linguistic Typology 11: 119–32. Haspelmath, Martin. 2010. Comparative concepts and descriptive categories in cross-linguistic studies. Language 86(3): 663–87. Hauser, Marc D. 1996. The evolution of communication. Cambridge, MA: MIT Press. Hauser, Marc D. 2000. Wild minds. New York: Holt. Hauser, Marc D., Chomsky, Noam, and Tecumseh Fitch, W. 2002. The faculty of language: What is it, who has it, and how did it evolve? Science 298: 1569–1579. Hauser, Marc D. and McDermott, J. 2003. The evolution of the music faculty: a comparative perspective. Nature Neuroscience, 6(7): 663–8. Hay, Jennifer, Kennedy, Christopher, and Levin, Beth. 1999. Scalar Structure Underlies Telicity in Degree Achievements. In T. Mathews and D. Strolovitch (eds.), SALT IX. Ithaca: CLC Publications, 127–44. Heberlein, A. S. and Adolphs, R. 2004. Impaired spontaneous anthropomorphizing despite intact perception and social knowledge. PNAS 11 May 2004, 101(19): 7487–7491. Heider, F. and Simmel, M. 1944. An experimental study of apparent behavior. American Journal of Psychology 57: 243–49. Heim, Irene. 1982. The Semantics of Definite and Indefinite Noun Phrases. PhD dissertation, University of Massachusetts; published 1989, New York: Garland. Heim, Irene. 2005. Features on bound pronouns. Ms., Massachusetts Institute of Technology, Department of Linguistics. Heim, Irene. 2008. Features on Bound Pronouns. In D. Adger, S. Bejar, and D. Harbour (eds.), Phi Theory: Phi Features across Interfaces and Modules. Oxford: Oxford University Press. Heim, Irene and Kratzer, Angelika. 1998. Semantics in Generative Grammar. Oxford: Blackwell. Helmbrecht, Johannes. 2005. Das Problem der Universalität der Nomen/Verb-Distinktion. ASSIDUE Arbeitspapiere des Seminars für Sprachwissenschaft der Universität Erfurt Nr. 16. Henshilwood, C. S. 2006. Modern humans and symbolic behaviour: Evidence from Blombos Cave, South Africa. In G. Blundell (ed.), Origins. Cape Town: Double Storey, 78–83. Heycock, Caroline. 2006. Embedded Root Phenomena. In M. Everaert and H. van Riemsdijk (eds.), The Blackwell Companion to Syntax, Vol. II. Oxford: Basil Blackwell, 174–209.

354

References

Himmelmann, Nikolaus P. 2005. Gram, construction, and class formation. In Clemens Knobloch and Burkhard Schaeder (eds.), Wortarten und Grammatikalisierung. Berlin and New York: Walter de Gruyter, 79–94. Himmelmann, Nikolaus P. 2007. Lexical categories and voice in Tagalog. To appear in Peter Austin and Simon Musgrave (eds.), Voice and Grammatical Functions in Austronesian Languages. Stanford: CSLI. Hinzen, Wolfram. 2006. Mind design and minimal syntax. Oxford: Oxford University Press. Hinzen, Wolfram. 2007. An essay on names and truth. Oxford: Oxford University Press. Hinzen, Wolfram. 2008. The Succesor Function+Lex=Human Language? In Grohmann, K. (ed.), InterPhases: Phase-Theoretic Investigations of Linguistic Interfaces. Oxford: Oxford University Press, 25–47. Hinzen, Wolfram. 2009. Hierarchy, Merge, and Truth. In Piattelli-Palmarini et al. (eds.), 123–41. Hinzen, Wolfram. 2012. Syntax in the Atom. In Werning et al. (eds.), 351–70. Hinzen, Wolfram. 2013a. On truth-attribution without the lexical truth-predicate. Ms., University of Durham. Hinzen, Wolfram. 2013b. Narrow syntax and the Language of Thought. Philosophical Psychology 26(1): 1–23. Hinzen, Wolfram and Poeppel, David. 2011. Semantics between cognitive neuroscience and linguistic theory. Language & Cognitive Processes 26(9): 1297–1316. Hinzen, Wolfram, Sheehan, Michelle, and Reichard, Uli. 2013. Intensionality, Grammar, and the Sententialist Hypothesis. Ms., University of Durham. Hinzen, Wolfram and Uriagereka, Juan. 2006. On the Metaphysics of Linguistics. Erkenntnis 65(1): 71–96. Hirakawa, Yahiro. 1994. Case checking and a double nominative construction in Japanese. Ms. McGill University. Holmberg, Anders. 1986. Word Order and Syntactic Features. PhD thesis, Stockholm University. Holmberg, Anders. 2005. Is there a little pro? Evidence from Finnish. Linguistic Inquiry 36(4): 533–64. Holmberg, Anders. 2010. Verb second. To appear in T. Kiss and A. Alexiadou (eds.), Syntax: An International Handbook of Contemporary Syntactic Research. Berlin: Walter de Gruyter. Hooper, J. and Thompson, S. 1973. On the applicability of root transformations, Linguistic Inquiry 4: 465–97. Hornstein, Norbert. 1984. Logic as Grammar: An Approach to Meaning in Natural Language. Cambridge, MA: Bradford/MIT Press. Hornstein, Norbert. 1999. Movement and Control. Linguistic Inquiry 30(1): 69–96. de Hoop, Helen. 1996. Case Configuration and Noun Phrase Interpretation. New York: Garland. Horwich, Paul. 2010. Truth, meaning, reality. Oxford: Oxford University Press. van Hout, Angeliek. 1996. Event semantics of verb frame alternations. PhD thesis, Tilburg University.

References

355

van Hout, Angeliek. 2004. Unaccusativity as telicity checking. In A. Alexiadou, E. Anagnostopoulou, and M. Everaert (eds.), The unaccusativity puzzle: explorations of the syntax-lexicon interface. Oxford: Oxford University Press, 60–83. Hughes, Tom. 2013. Reference, description, and ambiguity: a grammatical account. PhD thesis, University of Durham. Hull, David L. and Ruse, Michael. 1998. The philosophy of biology. Oxford: Oxford University Press. Hume, D. 1739/1888. Treatise of Human Nature, ed. by L. A. Selby-Bigge. Oxford: Clarendon Press. Husserl, Edmund 1900–1/2009. Logische Untersuchungen. Frankfurt: Meiner. Husserl, Edmund 1929/1969. Formale und transzendentale Logik. Tuebingen: Niemeyer. Huxley, Thomas H. 1863. Evidence as to Man’s place in nature. London: Williams & Norgate. Iggesen, Oliver A. 2008. Number of cases. In M. Haspelmath, M. S. Dryer, D. Gil, and B. Comrie (eds.), World atlas of language structures online, chap. 49. Available at http:// wals.info/chapter/49. Accessed 9 Nov 2010. Munich: Max Planck Digital Library. Iggesen, Oliver. A. 2011. Person Marking on Adpositions. In M. S. Dryer and M. Haspelmath (eds.), The World Atlas of Language Structures Online. Munich: Max Planck Digital Library. Jablensky, A., Sartorius, N., Emberg G., Anker, M., Korten, A., Cooper, J. E., Day, R., and Bertelsen, A. 1992. Schizophrenia: manifestations, incidence and course in different cultures. Psychological Medicine 20: 1–97. Jackendoff, Ray. 1991. Parts and boundaries. Cognition 41: 9–45. Jackendoff, Ray. 1996. The Proper Treatment of Measuring Out, Telicity, and Perhaps Even Quantification in English. Natural Language & Linguistic Theory 14(2): 305–54. Jackendoff, Ray. 2002. Foundations of language. Oxford: Oxford University Press. Jackendoff, Ray. 2011. What is the human language faculty?: Two views. Language 87(3): 586–624. Jacobson, P. 2012. Direct compositionality. In Werning et al. (eds.), 109–28. Jayaseelan, K. A. 2008. Topic, focus, and adverb positions in clause structure. Nanzan Linguistics 4: 43–68. Jayaseelan, K. A and Hariprasad, M. 2001. Deixis in pronouns and noun phrases. Linguistic Analysis 31: 132–47. Jespersen, Otto. 1924. Philosophy of grammar. London: Allen and Unwin. Jobard, G., Vigneau, M., Mazoyer, B., and Tzourio-Mazoyer, N. 2007. Impact of modality and linguistic complexity during reading and listening tasks. NeuroImage 34: 784–800. Jones, M. A. 1993. Sardinian Syntax. London: Routledge Johnson, Kyle. 2011. Pronouns vs. definite descriptions. Ms., University of Massachusetts at Amherst. Julien, M. 2010. Embedded clauses with main clause word order in Mainland Scandinavian. Ms., Lund University. Jusczyk, Peter. 1997. The Discovery of Spoken Language. Cambridge, MA: MIT Press. Kandybowicz, Jason. 2006. Comp-trace effects explained away. In D. Baumer, D. Montero, and M. Scanlon (eds.), WCCFL 25: Proceedings of the 25th West Coast Conference on Formal Linguistics. Somerville, MA, Cascadilla Proceedings Project, 220–8.

356

References

Katz, J. and Pesetsky, D. 2009. The recursive syntax and prosody of tonal music, Ms., Massachusetts Institute of Technology. Kaplan, David. 1977/1989a. Demonstratives. In J. Almog, J. Perry, and H. Wettstein (eds.), Themes from Kaplan. Oxford: Oxford University Press, 481–563. Kaplan, David. 1989b. Afterthoughts. In J. Almog, J. Perry, and H. Wettstein (eds.), Themes from Kaplan. Oxford: Oxford University Press, 565–614. Karttunen, Lauri. 1971. Some observations on factivity. Papers in Linguistics 4: 55–69. Kayne, Richard. 1975. French Syntax: The transformational cycle. Cambridge, MA: MIT Press. Kayne, Richard. 1998. Overt vs. covert movement. Syntax 1(2): 128–91. Kayne, Richard. 1991. Romance clitics, verb movement, and PRO. Linguistic Inquiry 22: 646–87. Kayne, Richard. 1994. The Antisymmetry of Syntax. Cambridge, MA: MIT Press. Kayne, Richard. 2000. Parameters and Universals. Oxford: Oxford University Press. Kayne, Richard. 2004. Antisymmetry and Japanese. In L. Jenkins (ed.), Variation and Universals in Biolinguistics. Amsterdam: Elsevier, 3–37. Kayne, Richard. 2005. Movement and Silence. Oxford: Oxford University Press. Kayne, Richard. 2006. On parameters and on principles of pronunciation. In H. Broekhuis, et al. (eds.), Organizing Grammar: Linguistic Studies in Honor of Henk van Riemsdijk. Berlin: Mouton de Gruyter, 289–99. Kayne, Richard. 2008. Expletives, Datives, and the Tension between Morphology and Syntax. In T. Biberauer (ed.), The Limits of Syntactic Variation. Amsterdam/Philadelphia, PA: John Benjamins, 175–217. Kayne, Richard S. 2009a. Why is Syntax Antisymmetric? Paper presented at Theoretical Approaches to Disharmonic Word Orders. Newcastle, June 2009. Kayne, Richard S. 2009b. A note on auxiliary selection and silent causation. In L. Baronian and F. Martineau (eds.), Le français d’un continent à l’autre. Mélanges offerts à Yves Charles Morin. Québec: Presses de l’Université Laval, 211–35. Kayne, Richard S. 2011. Antisymmetry and the lexicon. In Anna Maria Di Sciullo and Cedric Boeckx (eds.), 329–53. King, Jeff. 2001. Complex Demonstratives: A Quantificational Account. Cambridge, MA: MIT Press. King, Jeff. 2007. The Nature and Structure of Content. Oxford: Oxford University Press. Kiparsky, Carol and Kiparsky, Paul. 1970. Fact. In M. Bierwisch and K. E. Heidolph (eds.), Progress in Linguistics. The Hague: Mouton, 143–73. Kiparsky, Paul. 1973. ‘Elsewhere’ in Phonology. In S. Anderson and P. Kiparsky (eds.), A Festshrift for Morris Halle. New York: Holt, Rinehart and Winston, 93–106. Kiparsky, Paul. 1994. Paninian Linguistics. In R.E. Asher (ed.), Encyclopedia of Language and Linguistics. Oxford and New York: Pergamon Press. Kiparsky, Paul. 1998. Partitive Case and Aspect. In M. Butt and W. Geuder (eds.), The Projection of Arguments. Stanford, CA: CSLI, 265–307. Kiparsky, Paul. 2002. On the architecture of Panini’s grammar. Ms., Stanford University. Kircher, Tilo T., Bulimore, E. T., Brammer, M. J., Williams, S. C., Broome, M. R., Murray, R. M., and McGuire, P. K. 2001. Differential activation of temporal cortex during sentence completion in schizophrenic patients with and without formal thought disorder. Schizophrenia Research 50(1–2): 27–40.

References

357

Kircher, Tilo, Oh, Tomasina, Brahmer, M. J., and McGuire, P. K. 2005. Neural correlates of syntax production in schizophrenia. British Journal of Psychiatry 186: 209–14. Kircher, Tilo and David, Anthony. 2003. The Self in Neuroscience and Psychiatry. Cambridge: Cambridge University Press. Kirkby, David. 2013. On Judgement: Psychological Genesis, Intentionality and Grammar. PhD thesis, Durham University. Kirkby, David and Mikhail, John. 2013. The linguistic analogy. Philosophy Compass. Forthcoming Kjelgaard, M. M. and Tager-Flusberg, H. 2001. An investigation of language impairment in autism: Implications for genetic subgroups. Language and Cognitive Processes 16: 287–308. Kleist, K. 1914. Aphasie und Geisteskrankheit. Münchener Medizinische Wochenschrift 61: 8–12. Kowalski, Alix and Yang, Charles. 2012. Verb Islands in Child and Adult Language. BUCLD Proceedings. Forthcoming Kraepelin, Ernst 1896/1919. Dementia praecox and paraphrenia. Chicago: Chicago Medical Book Company. Kraepelin, E. 1904. Lectures on Clinical Psychiatry. New York: Hafner. Kraepelin, E. 1906. Über Sprachstörungen im Traume. Leipzig: Verlag von Wilhelm Engelmann. Kratzer, Angelika. 1996. Severing the external argument from the verb. In Johann Rooryck and Laurie Zaring (eds.), Phrase structure and the lexicon. Dordrecht: Kluwer, 109–37. Kratzer, Angelika. 2004. Telicity and the meaning of objective case. In J. Guéron and J. Lecarme (eds.), The Syntax of Tense. Cambridge, MA: MIT Press, 389–423. Kratzer, Angelika. 2009. Making a pronoun: Fake indexicals as windows into the properties of pronouns. Linguistic Inquiry 40: 187–237. Krause, J., Lalueza-Fox, C., Orlando, L., Enard, W., Green, R. E., Burbano, H. A., Hublin, J.-J., Hänni, C., Fortea, J., de la Rasilla, M., Bertranpetit, J., Rosas, A., and Pääbo, S. 2007. The derived FOXP2 variant of modern humans was shared with Neandertals. Current Biology 17: 1908–1912. Krifka, Manfred. 1998. The Origins of Telicity. In S. Rothstein (ed.), Events and grammar. Dordrecht: Kluwer, 197–235. Kripke, Saul. 1972. Naming and Necessity. In Davidson, Donald and Harman, Gilbert (eds.), Semantics of Natural Language. Dordrecht: Reidel, 253–355, 763–9. Kripke, Saul. 1982. Wittgenstein on Rules and Private Language. Oxford: Blackwell and Cambridge: Harvard University Press. Kuperberg Gina. 2007. Neural mechanisms of language comprehension: Challenges to syntax. Brain Research 1146: 23–49. Kuperberg, Gina. 2010a. Language in schizophrenia, Part 1. Language and Linguistics Compass 4(8): 576–89. Kuperberg, Gina. 2010b. Language in schizophrenia, Part 2. Language and Linguistics Compass 4(8): 590–604. Kuperberg, Gina R., Kreher, D. A., Goff, D., McGuire, P. K., and David, A. S. 2006. Building up linguistic context in schizophrenia. Neuropsychology 20: 442–52.

358

References

Kuperberg, Gina, McGuire, P., and David, A. 2000. Sensitivity to linguistic anomalies in spoken sentences: a case study approach to understanding thought disorder in schizophrenia. Psychological Medicine 30: 345–57. Laka, Itziar. 2006. Deriving Split-ergativity in the progressive: the case of Basque. In Alana Johns, Diane Massam, and Ndayiragije Juvenal (eds.), Ergativity. Emerging Issues. Dordrecht: Springer, 173–96. Landau, Idan. 2000. Elements of Control: Structure and Meaning in Infinitival Constructions. Dordrecht and Boston: Kluwer Academic Publishers. Landau, Idan. 2004. The Scale of Finiteness and the Calculus of Control. Natural Language and Linguistic Theory 22(4): 811–77. Landau, Idan. 2006. Severing the distribution of PRO from Case. Syntax 9(2): 153–70. Landau, Idan. 2007a. Movement Resistant Aspects of Control. Natural Language and Linguistic Theory 71: 293–325. Landau, Idan. 2007b. EPP extensions. Linguistic Inquiry 38(3): 485–523. Landau, Idan. 2008. Two Routes of Control: Evidence from Case Transmission in Russian. Natural Language and Linguistic Theory 26: 877–924. Langacker, Ronald W. 1987. Foundations of Cognitive Grammar, vol. I: Theoretical Prerequisites. Stanford: Stanford University Press. Langacker, Ronald W. 1999. Grammar and conceptualization. Berlin: Mouton de Gruyter. Langdon, R., Coltheart, M., Ward, P. B., and Catts, S. V. 2002. Disturbed communication in schizophrenia: the role of poor pragmatics and poor mind-reading. Psychological Medicine 32: 1273–1284. Larson, Richard K., Déprez, V. M., and Yamakido, H. (eds.). 2010. Approaches to the Evolution of Language. Cambridge: Cambridge University Press. Lasnik, Howard. 2008. On the development of Case theory: Triumphs and challenges. In Freidin (ed.), 17–41. Lasnik, Howard and Saito, Mamoru. 1992. Move [alpha]: conditions on its applications and output. Cambridge, MA: MIT Press. Legate, Julie A. 2008. Morphological and Abstract Case. Linguistic Inquiry 39(1): 55–101. Legate, Julie A. 2011. Types of ergativity. Lingua 122: 181–91. Leland, George and Kornfilt, Jaklin. 1981. Finiteness and boundedness in Turkish. In F. Heny (ed.), Binding and filtering. Cambridge, MA: MIT Press, 105–27. Leiss, Elisabeth. 2009. Sprachphilosophie. Berlin and New York: de Gruyter. Leiss, Elisabeth and Abraham, Werner, to appear: Language and memory: How language creates the human mind. Amsterdam: John Benjamins. Lemaréchal, Alain. 1989. Les parties du discours. Sémantique et syntaxe. Paris: P.U.F. Lenneberg, Eric H. 1967. Biological Foundations of Language. New York: Wiley. Leu, Tom. 2008. The internal syntax of determiners. PhD dissertation, New York University. Levelt, Willem J. M., Roelofs, A., and Meyer, A. S. 1999. A theory of lexical access in speech production. Behavioral and Brain Sciences 22: 1–38. Levi, Isaac. 1980. The enterprise of knowledge. Cambridge, MA: MIT Press. Levin, Beth. 1983. On the nature of ergativity. PhD thesis, Massachusetts Institute of Technology.

References

359

Levinson, Stephen C. and Evans, Nicholas. 2010. Time for a sea-change in linguistics: response to comments on ‘The Myth of Language Universals’. Lingua 120: 2733–2758. Levy D. L., Coleman, M. J., Sung, H., Ji, F., Matthysse, S., Mendell, N. R., and Titone, D. (2010). The Genetic Basis of Thought Disorder and Language and Communication Disturbances in Schizophrenia. Journal of Neurolinguistics 23(3): 176. Lewis, David. 1983. Attitudes de dicto and de se. In D. Lewis (ed.), Philosophical Papers, Vol. 1. Oxford: Oxford University Press. Li, P., Ogura, T., Barner, D., Yang, S.-Y., and Carey, S. 2009. Does the conceptual distinction between singular and plural sets depend on language? Developmental Psychology 45(6): 1644–1653. Li, X., Branch, C. A., Ardekani, B. A., Betisch, H., Hicks, C., and DeLisi, L. 2007. fMRI study of language activation in schizophrenia, schizoaffective disorder and in individuals genetically at high risk. Schizophrenia Resesarch 96: 14–24. Li, X., Branch, C. A., Nierenberg, J., and DeLisi, L. E. 2010. Disturbed Functional Connectivity of Cortical Activation during Semantic Discrimination in Patients with Schizophrenia and Subjects at Genetic High-risk. Brain Imaging and Behavior 4: 109–20. Lidz, Jeffrey. 2006. The Grammar of Accusative Case in Kannada. Language 82: 10–32. Liebal, K., Columbi, C., Rogers, S., Warneken, F., and Tomasello, M. 2008. Cooperative activities in children with autism. Journal of Autism and Developmental Disorders 38: 224–38. Lind, S. E. and Bowler, D. M. 2009. Language and theory of mind in autism spectrum disorder: The relationship between complement syntax and false belief task performance. Journal of Autism and Developmental Disorders 39(6): 929–37. Longa, Victor M., Lorenzo, G., and Rigau, G. 1998. Subject clitics and clitic recycling: locative sentences in some Iberian Romance languages. Journal of linguistics 34: 125–64. Longobardi, Giuseppe. 1994. Reference and proper names. Linguistic Inquiry 25: 609–65. Longobardi, Giuseppe. 2001. How Comparative Is Semantics? Natural Language Semantics 9: 335–69. Longobardi, Giuseppe. 2005. Toward a unified grammar of reference. Zeitschrift für Sprachwissenschaft 24: 5–44. Longobardi, Giuseppe. 2008. Reference to individuals, person, and the variety of mapping parameters. In H. H. Mueller and A. Klinge (eds.), Essays on nominal determination. Amsterdam and Philadelphia, PA: John Benjamins, 189–211. Lowe, E. Jonathan. 2006. Non-Cartesian substance dualism. Erkenntnis 65(5): 5–23. Lowe, E. Jonathan. 2008. Personal Agency. Oxford: Oxford University Press. Ludlow, Peter. 2011. The philosophy of generative grammar. New York: Oxford University Press. Lycan, William. 2003. Chomsky on the mind-body problem. In Antony and Hornstein (eds.), 11–28. Lycan, William G. 2008. Philosophy of Language. Oxford: Routledge. Maess B., Koelsch S., Gunter T. C., and Friederici, Angela D. 2001. Musical syntax is processed in Broca’s area: an MEG study. Nature Neuroscience 4(5): 540–5. Mahajan, Anoop. 1992. The specificity condition and the CED. Linguistic Inquiry 23: 510–16. Mahajan, Anoop. 1994. The ergativity parameter: have-be alternation word order and split ergativity. In Proceedings of NELS 24. Amherst, MA: GLSA, 317–31.

360

References

Majid, A., Bowerman, M., Kita, S., Haun, D. B. M., and Levinson, S. C. 2004. Can language restructure cognition? The case for space. Trends in Cognitive Sciences 8(3): 108–14. Maher, Brendan A., Manschreck, T. C., and Molino, M. A. 1983. Redundancy, pause distributions and thought disorder in schizophrenia. Language and Speech 26: 191–9. Maïza, O., Mazoyer, B., Hervé, P.-Y., Razafimandimby, A., Dollfus, S., Tzourio-Mazoyer, N. 2010. Impact of cognitive performance on the reproducibility of fMRI activation in schizophrenia. Journal of Psychiatry and Neuroscience 35(6): 378–89. Malt, Alex. 2013. The Embodiment of Language: Grammar and the Extended Mind. PhD thesis, University of Durham. Mandler, J. 2004. The foundations of mind: The origins of the conceptual system. New York: Oxford University Press. Manning, Christopher. 1996. Ergativity: Argument Structure and Grammatical Relations. Stanford, CA: CSLI Publications. Marantz, Alec. 1984. On the Nature of Grammatical Relations. Cambridge, MA: MIT Press. Marantz, Alec. 1991. Case and licensing. In G. F. Westphal, B. Ao, and H.-R. Chae (eds.), ESCOL ’91: Proceedings of the Eighth Eastern States Conference on Linguistics, 234–53. Marcus, Gary F. 2001. The Algebraic Mind. Cambridge, MA: MIT Press. Marini, A., Spoletini, I., Rubino, I. A., Ciuffa, M., Bria, P., Martinotti, G., Banfi, G., Boccascino, R., Strom, P., Siracusan, A., Caltagirone, C., and Spalletta, G. 2008. The language of schizophrenia. Schizophrenia Research 105: 144–55. Martín, Txuss. 2009. Case interpretation and noun phrase configuration. Ms., New York University. Martín, Txuss. 2012. Deconstructing Catalan Object Clitics. PhD dissertation: New York University. Martín, Txuss and Wolfram Hinzen. 2012. The grammar of the essential indexical. Ms., University of Durham. Masullo, P. 1992. Incorporation and Case Theory in Spanish: A crosslinguistic perspective. PhD dissertation, University of Washington. Matilal, Bimal Krishna. 1990. The word and the world. India’s contribution to the study of language. Oxford: Oxford University Press. Mausfeld, Rainer. 2011. The perception of material qualities and the internal semantics of the perceptual system. In L. Albertazzi, G. van Tonder, and D. Vishwanath (eds.), Perception beyond Inference. The Information Content of Visual Processes. Cambridge, MA: MIT Press, 159–200. Mayr, Ernst. 1942. Systematics and the Origin of Species. New York: Columbia University Press. Mazoyer, Bernard, Tzourio, N., Frak, V., Syrota, A., Murayama, N., Levrier, O., Salamon, G., Dehaene, S., Cohen, L., and Mehler, J. 1993. The cortical representation of speech. Journal of Cognitive Neuroscience 5(4): 467–79. Mazoyer, Bernard and Tzourio-Mazoyer, N. 2008. Functional deficit in the medial prefrontal cortex during a language comprehension task in patients with schizophrenia. Schizophrenia Research 99: 304–11. McBrearty, Sally and Brooks, Alison. 2000. The revolution that wasn’t: a new interpretation of the origin of modern human behavior. Journal of Human Evolution 39: 453–563.

References

361

McCloskey, James. 1996. Subjects and Subject Positions in Modern Irish. In B. Borsley and I. Roberts (eds.), The Syntax of the Celtic Languages. Cambridge: Cambridge University Press, 241–83. McFadden, Thomas. 2001. The morphosyntax of case and adpositions: a research program. Ms., University of Pennsylvania. McFadden, Thomas. 2004. The position of morphological case in the derivation: a study on the syntax-morphology interface. PhD thesis, University of Pennsylvania. McFadden, Thomas. 2012. Deriving the distinction between structural and inherent case. Paper presented at the Linguistics Association of Great Britain’s Annual Meeting, Salford, UK. McGuinness, Martha. 1998. Locality in A-movement. PhD thesis, Massachusetts Institute of Technology. McKenna, Peter and Oh, Tomasina. 2005. Schizophrenic speech. Cambridge: Cambridge University Press. Megerdoomian, Karine. 2008. Parallel nominal and verbal projections. In Freidin et al. (eds.), 73–104. Melvold, J. 1991. Factivity and definiteness. MIT Working Papers in Linguistics 15: 97–117. Merchant, Jason. 2004. Fragments and ellipsis. Linguistics & Philosophy 27: 661–738. Merleau-Ponty, Maurice. 1945/2006. Phenomenology of Perception. London: Routledge. Mikhail, John. 2007. Universal Moral Grammar: Theory, Evidence and the Future. Trends in Cognitive Sciences 11: 143–52. Mikkelsen, Line. 2005. Copular Clauses: Specification, predication and equation. Amsterdam: John Benjamins. Miller, James. 2013. Realism, Truthmakers and Language. PhD thesis, University of Durham. Miyata, Jun, Sasamoto, Akihiko, Koelkebeck, Katja, Hirao, Kazuyuki, Ueda, Keita, Kawada, Ryosaku, Fujimoto, Shinsuke, Tanaka, Yusuke, Kubota, Manabu, Fukuyama, Hidenao, Sawamoto, Nobukatsu, Takahashi, Hidehiko, and Murai, Toshiya. 2012. Abnormal Asymmetry of White Matter Integrity in Schizophrenia Revealed by Voxelwise Diffusion Tensor Imaging. Human Brain Mapping 33: 1741–1749 Modesto, Marcello. 2007. Inflected infinitives in Brazilian Portuguese as an argument both contra and in favor of a movement analysis of control. Revista letras, Curitiba 72: 297–309. Modesto, Marcello. 2010. What Brazilian Portuguese Says about Control: Remarks on Boeckx and Hornstein. Syntax 13: 78–96. Moltmann, Friederike. 2007. Events, Tropes, and Truthmaking. Philosophical Studies 134: 363–403. Moltmann, Friederike. 2012. ‘Truth Predicates’ in Natural Language. Forthcoming in D. Achourioti, H. Galinon, and J. Martinez (eds.), Unifying the Philosophy of Truth. Dordrecht: Springer. Montague, Richard. 1973. The Proper Treatment of Quantification in Ordinary English. In Jaakko Hintikka, Julius Moravcsik, and Patrick Suppes (eds.), Approaches to Natural Language. Dordrecht: D. Reidel, 221–42. Moravcsik, Julius M. E. 1968. Aristotle’s Theory of Categories. In Moravcsik, J. M. E. (ed.), Aristotle: A Collection of Critical Essays. London: Macmillan, 125–45 Moro, Andrea. 1997. The Raising of Predicates: Predicative Noun Phrases and the Theory of Clause Structure. Cambridge: Cambridge University Press.

362

References

Moro, Andrea. 2000. Dynamic anti-symmetry. Cambridge, MA: MIT Press. Moscovitch, M. and Umilta, C. 1990. Modularity and neuropsychology: Modules and central processes in attention and memory. In M. Schwartz (ed.), Modular deficits in Alzheimer-type dementia. Cambridge, MA: MIT Press, 1–59. Mowry B., Lennon, D. P., and De Felice, C. M. 1994. Diagnosis of schizophrenia in a matched sample of Australian aborigines. Acta Psychiatrica Scandinavica 90: 337–41. Müeller, Friedrich Max. 1887. The science of thought. London: Longmans, Green, and Co. Müller, Friedrich Max. 1882. Grundriss der Sprachwissenschaft. Vol. II, Pt. 2. Vienna: Alfred Hölder. Müller, Gereon. 2006. Pro-Drop and Impoverishment. In P. Brandt and E. Fuβ (eds.), Form, Structure, and Grammar. A Festschrift Presented to Günther Grewendorf on Occasion of his 60th Birthday. Akademie Verlag, Berlin, 93–115. Müller, Gereon. 2009. Ergativity, accusativity, and the order of Merge and Agree. In K. K. Grohmann (ed.), Explorations of phase theory: Features and arguments. Berlin: Mouton de Gruyter, 269–308. Mukherji, Nirmalangshu. 2009. The primacy of grammar Cambridge, MA: MIT Press. Mundy, P. and Siegman, M. 2006. Joint attention, social competence and developmental psychopathology. In D. Cicchetti and D. Cohen (eds.), Developmental psychopathology, vol. 1: Theory and methods. Hoboken, NJ: John Wiley. Musso, Mariacristina, Moro, Andrea, Glauche, Volkmar, Rijntjes, Michel, Reichenbach, Jürgen, Büchel, Christian, and Weiller, Cornelius. 2003. Broca’s area and the language instinct. Nature Neuroscience 6: 774–81. Naylor, Paz B. 1995. Subject, Topic, and Tagalog syntax. In Benett, David, Bynon, Theodora, and Hewitt, George B. (eds.), Subject, Voice and Ergativity. London: SOAS, 161–201. Ndayiragije, J. and Nikiema, E. 2011. Why classifiers are not determiners: From Chinese to Bantu. Presentations at ‘Formal Linguistics’ conference, Guangdong, December 2011. Neeleman, Ad and Szendrői, Krista. 2005. Pro Drop and Pronouns. In J. Alderete et al. (eds.), Proceedings of the 24th West Coast Conference on Formal Linguistics. Somerville, MA: Cascadilla Proceedings Project, 299–307. Neeleman, Ad and Szendrői, Krista. 2007. Radical pro drop and the morphology of pronouns. Linguistic Inquiry 38(4): 671–714. Neeleman, A. and van de Koot, H. 2008. Dutch Scrambling and the Nature of Discourse Templates. Journal of Comparative Germanic Linguistics 11(2): 137–89. Newmeyer, Frederik J. 2005. Possible and Probable Languages: A Generative Perspective on Linguistic Typology. Oxford: Oxford University Press. Newton, A. and de Villiers, J. 2007. Thinking while talking. Psychological Science 18(7): 574–9. Nichols, Johanna. 1986. Head-marking and dependent-marking grammar. Language 62: 56–119. Nieuwland, Mante S. and Van Berkum, Jos J. A. 2006. When Peanuts Fall in Love: N400 Evidence for the Power of Discourse. Journal of Cognitive Neuroscience 18(7): 1098–1111. Noonan, Maire B. 1992. Case and syntactic geometry. PhD thesis, McGill University. Nunes, Jairo. 2010. Relativizing Minimality for A-movement: phi and theta relations. Probus 22: 1–25.

References

363

Oh, Tomasina M., McCarthy, R. A., and McKenna, Peter. 2002. Is there a schizophasia? Neurocase 8: 233–44. Olson, Eric T. 2007. What Are We? A Study in Personal Ontology. Oxford: Oxford University Press. Onishi, K. and Baillargeon, R. 2005. Do 15-month-old infants understand false beliefs? Science 308: 255–8. Ormazábal, J. and Romero, J. 2007. The Object Agreement Constraint. Natural Language and Linguistic Theory 25: 315–47. Ormazábal, J. and Romero, J. 2010. Object clitics and agreement. Ms., Euskal Herriko Unibertsitatea/University of the Basque Country and Universidad de Extremadura. Panagiotidis, L. 1997. The fine structure of the left periphery. In L. Haegeman (ed.), Elements of Grammar: Handbook in Generative Syntax. Dordrecht: Kluwer, 281. Panagiotidis, Phoevos. 2008. Diachronic stability and feature interpretability. In Theresa Biberauer (ed.), The Limits of Syntactic Variation. Amsterdam: John Benjamins, 441–56. Panhuis, D. 2006. Latin Grammar. Ann Arbor, MI: University of Michigan Press. Partee, Barbara. 1980. Semantics—Mathematics or Psychology? In Rainer Bäuerle, Urs Egli, and Arnim von Stechow (eds.), Semantics from Different Points of View Berlin: SpringerVerlag, 1–14. Partee, Barbara. 1989. Binding Implicit Variables in Quantified Contexts. In C. Wiltshire, R. Graczyk, and B. Music (eds.), Papers from CLS 25. Chicago Linguistic Society, 342–65. Patnaik, T. 2009. Thought and language: The Bhartrharian perspective. In Chaturvedi (ed.), 185–204. Pawley, A. 2006. On the argument structure of phrasal predicates in Kalam, a language of the Trans New Guinea family. The 32nd Annual Meeting of the Berkeley Linguistics Society, Berkeley, California, USA. Paynter, Jessica and Peterson, Candida. 2010. Language and ToM development in autism versus Asperger syndrome: Contrasting influences of syntactic versus lexical/semantic maturity. Research in Autism Spectrum Disorders 4: 377–85. Penn, D. C., Holyoak, K. J., and Povinelli, D. J. 2008. Darwin’s mistake: Explaining the discontinuity between human and nonhuman minds. Behavioral and Brain Sciences 31: 109–78. Pereltsvaig, A. 2006. Small nominals. Natural Language and Linguistic Theory 24: 433–500. Perlmutter, D. M. 1971. Deep and Surface Structure Constraints in Syntax. New York: Holt, Rinehart and Winston. Perner, Josef and Wimmer, Heinz. 1985. ‘John thinks that Mary thinks that’: Attribution of second-order beliefs by 5- to 10-year-old children. Journal of Experimental Child Psychology, 39: 437–71. Pesetsky, David and Torrego, Esther. 2001. T-to-C movement: causes and consequences. In Michael Kenstowicz (ed.), Ken Hale: a Life in Language. Cambridge, MA: MIT Press, 355–426. Pesetsky, David and Torrego, Esther. 2004. Tense, Case, and the Nature of Syntactic Categories. In J. Guéron and J. Lecarme (eds.), The Syntax of Time. Cambridge, MA: MIT Press, 495–537. Pesetsky, David and Torrego, Esther. 2011. Case. In C. Boeckx (ed.), 52–72.

364

References

Perry, John. 1993. The Problem of the Essential Indexical and Other Essays. Oxford: Oxford University Press. Piatelli-Palmarini, Massimo. 2001. Speaking of learning: How do we acquire our marvellous facility for expressing ourselves in words? Nature. 411: 887–88. Piattelli-Palmarini, Massimo, Salaburu, Pello, and Uriagereka, Juan (eds.). 2011. Of minds and language: an encounter with Noam Chomsky in the Basque Country. Oxford: Oxford University Press. Picallo, Carme. 2002. Abstract agreement and clausal arguments. Syntax 5(2): 116–47. Picallo, Carme. 2007. On gender and number. Report de Recerca GGT-07-08, Universitat Autònoma de Barcelona. [Published partially as Picallo, C. 2008. On gender and number in Romance. Lingue e Linguaggio 1: 44–66. Pica, Pierre, Lemer, C., and Izard, V. 2004. Exact and approximate arithmetic in an Amazonian indigene group. Science 306(5695): 499–503. Pietroski, Paul M. 2005. Events and Semantic Architecture. Oxford: Oxford University Press. Pietroski, Paul M. 2008. Minimalist Meaning, Internalist Interpretation. Biolinguistics 4: 317–41. Pietroski, Paul M. 2011. Minimal semantic instructions. In C. Boeckx (ed.), 472–98. Pigliucci, Massimo and Müller, Gerd B. 2010. Elements of an Extended Evolutionary Synthesis. In Pigliucci and Müller (eds.), Evolution—The Extended synthesis. Cambridge, MA: MIT Press. Pinborg, Jan. 1967. Die Entwicklung der Sprachtheorie im Mittelalter. Münster: Aschendorff. Pinker, Steven. 1994. The language instinct. London: Penguin. Pinker, Steven. 1997. How the mind works. New York: W. W. Norton & Company. Pinker, Steven. 1999. Words and rules. London: Weidenfeld & Nicolson. Pinker, Steven and Bloom, Paul. 1990. Natural Language and Natural Selection. The Behavioral and Brain Sciences 13: 704–84. Pinker, Steven and Jackendoff, Ray. 2005. The Faculty of Language: What’s Special About It? Cognition 95: 201–36. Platzack, Christer. 2011. Towards a Minimal Argument Structure. Ms., Lund. Available at: http://ling.auf.net/lingBuzz/001409. Accessed 12 Jan 2012. Poletto, C. 2000. The Higher Functional Field in the Northern Italian Dialects. Oxford: Oxford University Press. Pollock, Jean-Yves. 1989. Verb Movement, Universal Grammar and the Structure of IP. Linguistic Inquiry 20: 365–424. Postal, Paul. 1966. On so-called ‘pronouns’ in English. In D. Reibel and S. Schane (eds.), Modern Studies in English. New York: Prentice Hall, 201–24. Pradhan, R. C. 2009. Bhartrhari as a philosopher of language. In Chaturvedi (ed.), 281–90. Preminger, Omer. 2011. Agreement as a fallible operation. PhD thesis, Massachusetts Institute of Technology. Progovac, Liliane. 1998. Determiner phrase in a language without determiners. Journal of Linguistics 34: 164–79. Pullum, Geoffrey K. and Scholz, Barbara C. 2010. Recursion and the infinitude claim. In Harry van der Hulst (ed.), Recursion in Human Language. Berlin: Mouton de Gruyter, 113–38.

References

365

Pulvermueller, Friedemann. 2012. Meaning and the brain: The neurosemantics of referential, interactive, and combinatorial knowledge. Journal of Neurolinguistics 25: 423–59. Putnam, Hilary. 1973. Meaning and Reference. Journal of Philosophy 70: 699–711. Pylkkänen, Liina. 2008. Introducing arguments. Cambridge, MA: MIT Press. Pylkkänen, L., Brennan, J., and Bemis, D. K. 2011. Grounding the Cognitive Neuroscience of Semantics in Linguistic Theory. Language and Cognitive Processes 26(9): 1317–37. Quicoli, A. Carlos. 1996. Inflection and Parametric Variation: Portuguese vs. Spanish. In Robert Freidin (ed.), Current Issues in Comparative Grammar. Dordrecht: Kluwer Academic Publishers, 46–80. Quine, Willem van Orman. 1960. Word and Object. Cambridge, MA: MIT Press. Rabelo, Poliana Camargo. 2004. Sobre a questão do controle com o infinitivo flexionado português, Universidade de Brasília. MA dissertation. Rabelo, Poliana Camargo. 2010. Argumentos (EPP) nulos no português do Brasil em contextos oracionais finitos e infinitivos, Universidade de Brasília. PhD thesis. Ramchand, Gillian C. 1997. Aspect and Predication. Oxford: Clarendon Press. Ramchand, Gillian C. 2008. Verb Meaning and the Lexicon: A First-Phase Syntax. Cambridge: Cambridge University Press. Ramchand, Gillian and Svenonius, Peter. 2008. Mapping a parochial lexicon onto a Universal Semantics. In T. Biberauer (ed.), The Limits of Syntactic Variation. Amsterdam: John Benjamins, 219–45. Raposo, Eduardo. 1989. Prepositional Infinitival constructions in European Portuguese. In O. Jaeggli and K. Safir (eds.), The Null Subject Parameter. Dordrecht: Kluwer. Rastle, K., Davis, M. H., Marslen-Wilson, W. D., and Tyler, L. K. 2000. Morphological and semantic effects in visual word recognition: A time-course study. Language and Cognitive Processes 15(4–5): 507–37. Reichard, Uli. 2013. Grammar, ontology, and the unity of meaning. PhD thesis, Durham University. Reichard, Uli and Hinzen, Wolfram. 2011. The event-argument: in search for evidence. Ms., University of Durham. Reichard, Uli and Hinzen, Wolfram. 2012. Grammar and ontology: What is the connection? Ms., University of Durham. Reinhart, Tanya. 1997. Quantifier scope: How labor is divided between QR and choice functions. Linguistics and Philosophy 20: 335–97. Reuland, Eric J. 2010. Imagination, Planning, and Working Memory: The Emergence of Language. Current Anthropology 51(S1), S99–S110. Reuland, E. J. 2011. Anaphora and Language Design. Cambridge, Mass: MIT Press. Rey, Georges. 2003. Chomsky, intentionality, and a CRTT. In Antony and Hornstein (eds.), Chomsky and His Critics. Oxford: Blackwell, 105–39. Richards, Marc. 2004. Object Shift and Scrambling in North and West Germanic: A Case Study in Symmetrical Syntax. PhD thesis, Cambridge University. Richards, Marc. 2007. On Feature Inheritance: An Argument from the Phase Impenetrability Condition. Linguistic Inquiry 38: 563–72.

366

References

Richards, Marc. 2008. Desymmetrization: Parametric Variation at the PF-Interface. The Canadian Journal of Linguistics/La revue canadienne de linguistique 52(2–3): 275–300. Richards, Marc. 2012. In Gallego (ed.), 195–232. de Rijk, L. M. 1967. Logica Modernorum: A contribution to the history of early terminist logic. Assen: Van Gorcum. Rijkhoff, Jan. 2002. Verbs and nouns from a cross-linguistic perspective. Rivista di Linguistica 14(1): 115–47. Ritter, Elizabeth. 1991. Two functional categories in noun phrases: evidence from Modern Hebrew. In S. Rothstein (ed.), Syntax & Semantics 26: 37–62. Ritter, Elizabeth and Wiltschko, Martina. 2009. Varieties of INFL: TENSE, LOCATION, and PERSON. In Hans Broekhuis, Jeroen Craenenbroeck, and Henk van Riemsdijk (eds.), Alternatives to Cartography. Berlin: Mouton de Gruyter, 153–201. Rizzi, Luigi. 1982. Issues in Italian syntax. Dordrecht: Foris. Rizzi, Luigi. 1997. The fine-structure of the left periphery. In L. Haegemann (ed.), Elements of Grammar. Dordrecht: Kluwer, 281–337. Rizzi, Luigi. 2004 The structure of CP and IP: The cartography of syntactic structures, vol. 2, New York: Oxford University Press. Rizzi, Luigi and Shlonsky, Ur. 2007. Strategies of Subject Extraction. In H.-M. Gärtner and U. Sauerland (eds.), Interfaces + recursion = language? Chomsky's minimalism and the view from syntax-semantics. Berlin: Mouton de Gruyter, 115–60. Roberts, Ian G. 2010. A Deletion Analysis of Null Subjects. In T. Biberauer, A. Holmberg, I. G. Roberts, and M. Sheehan (eds.), Parametric Variation: Null Subjects in Minimalist Theory. Cambridge: Cambridge University Press, 58–87. Roberts, Ian G. 2012. Macroparameters and Minimalism: A Programme for Comparative Research. In C. Galves, S. Cyrino, R. Lopes, F. Sândalo, and J. Avelar (eds.), Parameter Theory and Linguistic Change. Oxford: Oxford University Press. Roca, F. 1992. On the licensing of pronominal clitics: The properties of object clitics in Spanish and Catalan. MA thesis, Universitat Autònoma de Barcelona. Roca, F. 1996. Morfemas objetivos y determinantes: los clíticos del español. Verba 23: 83–119. Rochester, S. and J. Martin. 1979. Crazy talk. New York: Plenum Press. Röder, B., Stock, O., Neville. H. Bien, S., and Rösler, F. 2002. Brain Activation Modulated by the Comprehension of Normal and Pseudo-word Sentences of Different Processing Demands. NeuroImage 15: 1003–1014. Rodd, J. M. et al. 2005. The neural mechanisms of speech comprehension: fMRI studies of semantic ambiguity. Cerebral Cortex 15: 1261–1269. Rodriguez-Ferrera, S., McCarthy, R. A., and McKenna, P. J. 2001. Language in schizophrenia and its relationship to formal thought disorder. Psychological Medicine 31: 197–205. Roelofson, Floris. 2008. Anaphora resolved. PhD dissertation, ILLC, Universiteit van Amsterdam. Roeper, Tom. 2009. The prism of grammar. Cambridge, MA: MIT Press. Roeper, Tom and Snyder, William. 2005. Language learnability and the forms of recursion. In A. M. Di Sciullo and R. Delmonte (eds.), UG and External Systems. Amsterdam: John Benjamins, 155–69. Rosen, Sara T. 1999. The Syntactic Representation of Linguistic Events. GLOT International 4(2): 3–11.

References

367

Ross, J. R. 1973. Slifting. In M. Gross, M. Halle, and M.-P. Schutzenberger (eds.), The Formal Analysis of Natural Languages. Hague: Mouton de Gruyter, 133–69. Rosselló, Joanna. 2008. Some arguments against some prevalent ideas on specificational sentences. Journal of Portuguese Linguistics 7(2): 109–50. Rott, Hans. 2001. Change, Choice and Inference. Oxford: Oxford University Press. Rullmann, H. 2004. First and second person pronouns as bound variables. Linguistic Inquiry 35: 159–68. Russell, Bertrand. 1903. Principles of Mathematics. London: Routledge (2010). Russell, Bertrand. 1905. On Denoting. Mind 14: 479–93. Rutkowski, P. 2002. Noun/pronoun asymmetries: evidence in support of the DP hypothesis in Polish. Jezikoslovlje 3(1–2): 159–70. Ruwet, N. 1991. Syntax and Human Experience. Chicago: University of Chicago Press. Sass, Louis and Parnass. J. 2003. Schizophrenia, consciousness and the self. Schizophrenia Bulletin 28(3): 427–44. Sauerland, Uli and Elbourne, Paul. 2002. Total Reconstruction, PF Movement, and Derivational Order. Linguistic Inquiry. 33(2): 283–319. Schaller, Susan. 2012 (first published 1991). A man without words. Berkeley: University of California Press. Schlenker, Phillipe. 2003. A Plea for Monsters. Linguistics and Philosophy 26: 29–120. Schlenker, Phillipe. 2005. The Lazy Frenchman’s Approach to the Subjunctive. In T. Geerts, I. van Ginneken, and H. Jacobs (eds.), Proceedings of Going Romance XVII. Amsterdam and Philadelphia, PA: John Benjamins, 269–309. Schlenker, Phillipe. 2011. Indexicality and De Se Reports. In C. Maienborn, K. von Heusinger, and P. Portner (eds.), Semantics: An International Handbook of Natural Language Meaning, vol. 2. Berlin: Mouton de Gruyter, 1561–1604. Schwartz, Jeffrey H. 1999. Sudden origins: Fossils, Genes, and the Origin of Species. New York: Wiley. Schwartz, Richard G. 2009. Specific language impairment. In R. G. Schwartz (ed.), Handbook of Child Language Disorders. Hove: Psychology Press, 3–43. Searls, David. 2002. The language of genes. Nature 420: 211–17. Sera, M. D., Berge, C., and del Castillo Pintado, J. 1994. Grammatical and conceptual forces in the attribution of gender by English and Spanish speakers. Cognitive Development 9: 261–92. Sheehan, Michelle. 2006. The EPP and Null Subjects in Romance. PhD thesis, Newcastle University. Sheehan, Michelle. 2010a. ‘Free’ inversion in Romance and the Null Subject Parameter. In T. Biberauer, A. Holmberg, I. G. Roberts, and M. Sheehan (eds.), Parametric Variation: Null Subjects in Minimalist Theory. Cambridge: Cambridge University Press, 231–62. Sheehan, Michelle. 2010b. Extraposition and Antisymmetry. In J. Van Craenenbroeck (ed.), Linguistic Variation Yearbook 10. Amsterdam, John Benjamins, 203–54. Sheehan, Michelle. 2012. A new take on partial Control: defective thematic intervention. Cambridge Occasional Papers in Linguistics 6: 1–47. Available online at http://ling.auf.net/ lingbuzz/001500. Accessed 5 May 2012. Sheehan, Michelle. 2013. Towards a parameter hierarchy for alignment. To appear in Proceedings of WCCFL 31, Cascadilla Press.

368

References

Sheehan, Michelle. Forthcoming a. Some implications of a copy theory of labeling. Syntax. Sheehan, Michelle. Forthcoming b. Explaining the Final-over-Final Constraint: formal and functional approaches. In T. Biberauer and M. Sheehan (eds.), Theoretical Approaches to Disharmonic Word Orders. Sheehan, Michelle. Forthcoming c. The Final-over-Final Constraint and adverbs. To appear in T. Biberauer, A. Holmberg, I. Roberts and M. Sheehan, The Final-over-Final Constraint: A Word-Order Universal and its Implications for Linguistic Theory. Cambridge, Massachusetts: MIT Press. Sheehan, Michelle and Hinzen, Wolfram. 2011. Moving towards the edge. Linguistic Analysis 37: 405–58. Sheehan, Michelle and Parafita Couto, Maria Carmen. 2011. Optional inflected infinitives in Portuguese and Galician. Paper presented at Poznań Linguistics Meeting (special session on the syntax of gerunds and infinitives), Poznań, Poland, May 2011. Shenton, M., Kikinis, R., and Jolesz, F. 1992. Abnormalities of the left temporal lobe and thought disorder in schizophrenia. New England Journal of Medicine 327: 604–12. Sherman, Michael. 2007. Universal Genome in the Origin of Metazoa. Cell Cycle 6(15): 1873–1877. Shlonsky, Ur. 2000. Subject position and copular constructions. In H. Bennis, M. Everaert, E. Reuland (eds.), Interface strategies. Amsterdam: Royal Netherlands Academy of Arts and Sciences, 325–47. Siegal, M. and Beattie, K. 1991. Where to look first for children’s knowledge of false beliefs. Cognition 38: 1–12. Siewierska, Anna. 2011. Alignment of Verbal Person Marking. In M. S. Dryer and M. Haspelmath (eds.), The World Atlas of Language Structures Online. Munich: Max Planck Digital Library. Sigurðsson, Halldór Á. 1989. Verbal syntax and case in Icelandic. Lund. [Republished in Reykjavik: University of Iceland, Institute of Linguistics (1992).] Sigurðsson, Halldor Á. 2004. The Syntax of Person, Tense and Speech Features. Italian Journal of Linguistics 16: 219–51. Sigurðsson, Halldor. Á. 2009. Remarks on features. In K. K. Grohmann (ed.), Explorations of phase theory: features and arguments. New York: Mouton de Gruyter, 21–52. Sigurðsson, Halldor. Á. 2012. Minimalist C/case. Linguistic Inquiry 43: 191–227. Singh, Rajendra, Dasgupta, Probal, and Lele, Jayant (eds.). 1995. Explorations in Indian Sociolinguistics. London: Sage. Sitaridou, Ioanna. 2007. Romance infinitives with subjects, subjunctive obviation and Control Theory. In Luis Eguren and Olga Fernández Soriano (eds.), Coreference, modality, and focus: studies on the syntax-semantics interface. Amsterdam and Philadelphia, PA: John Benjamins, 191–236. Smith, M., Apperly, I. A., and White, V. 2003. False belief reasoning and the acquisition of relative clause sentences. Child Development 74: 1709–1719. Skinner, B. F. 1953. Science and Human Behavior. Englewood Cliffs, NJ: Pearson. Skinner, B. F. 1957. Verbal behavior. New York: Appleton-Century-Crofts. Soames, Scott. 2010. What is Meaning? Princeton: Princeton University Press.

References

369

Sobin, Nicholas. 1987. The Variable Status of Comp-trace Phenomena. Natural Language and Linguistic Theory 5: 33–60. Spaepen, E., Coppola, M., Spelke, E., Carey, S., and Goldin-Meadow, S. 2011. Number without a language model. Proceedings of the National Academy of Sciences 108(3): 163–8. Spelke, Elizabeth. 2003. What makes humans smart? Core knowledge and natural language. In D. Gentner and S. Goldin- Meadow (eds.), Language in mind. Cambridge, MA: MIT Press, 277–311. Sportiche, Dominique. 1996. Clitic Constructions. In J. Rooeyck and L. Zaring (eds.), Phrase Structure and the Lexicon. Dordrecht: Kluwer, 213–77. Stainton, Robert. 2006. Words and Thoughts: Subsentences, Ellipsis, and the Philosophy of Language. Oxford: Oxford University Press. Stainton, Robert. 2011. In Defense of Public Languages. Linguistics and Philosophy 34: 479–88. Stalnaker, Robert. 1984. Inquiry. Cambridge, MA: MIT Press. Sterelny, Kim and Griffiths, Paul. 1999. Sex and death. An introduction to philosophy of biology. Chicago and London: University of Chicago Press. Stringer, Chris. 2011. The origin of our species. London: Penguin. Stockall, Linnaea and Marantz, Alec. 2006. A single route, full decomposition model of morphological complexity: MEG evidence. The Mental Lexicon 1(1): 85–123. Stowell, Tim. 1989. Subjects, specifiers, and X-bar theory. In M. Baltin and A. Kroch (eds.), Alternative conceptions of phrase structure. Chicago: University of Chicago Press, 232–62. Strømme, P., Bjørnstad, P.G., and Ramstad, K. 2002. Prevalence estimation of Williams syndrome. Journal of Child Neurology 17: 269–271. Strozer, Judith. 1976. Clitics in Spanish. PhD dissertation, University of California, Los Angeles. Suñer, M. 1988. The role of agreement in clitic-doubled constructions. Natural Language and Linguistic Theory 6: 391–434. Svenonius, Peter. 2002. Icelandic Case and the structure of events. Journal of Comparative Germanic Linguistics 5: 197–225. Svenonius, Peter. 2007. Interpreting uninterpretable features. Linguistic Analysis 33(3–4): 375–413. Szabolcsi, Anna. 1994. The Noun Phrase. In F. Kiefer and K. É. Kiss (eds.), The Syntactic Structure of Hungarian 27, Syntax and Semantics series. New York: Academic Press, 179–274. Tager-Flusberg, H., and Joseph, M. J. 2003. Identifying neurocognitive phenotypes in autism. Philosophical Transactions of the Royal Society of London B 358(1430): 303–14. Tager-Flusberg, H. 2004. Do autism and Specific language impairment represent overlapping language disorders? In M. L. Rice and S. F. Warren (eds.), Developmental Language Disorders. London: Lawrence Erlbaum Associates Publishers, 31–52. Taft, Marcus. 2004. Morphological decomposition and the reverse base frequency effect. Quarterly Journal of Experimental Psychology 57A: 745–65. Tarski, Alfred. 1944. The Semantic Conception of Truth and the Foundations of Semantics. Philosophy and Phenomenological Research 4: 341–76. Tattersall, Ian. 2002. The case of saltational events in human evolution. In Tim Crow (ed.), 49–60. Tattersall, Ian. 2008. The world from beginnings to 4000 BCE. Oxford: Oxford University Press.

370

References

Taylor, L. J., Maybery, M. T., and Whitehouse, A. J. 2012. Do Children with Specific Language Impairment have a Cognitive Profile Reminiscent of Autism? A Review of the Literature. Journal of Autism and Developmental Disorders 2 Feb 2012. Tenny, Carol. 1992. The Aspectual Interface Hypothesis. In I. Sag and A. Szabolsci (eds.), Lexical Matters. Stanford: Center for the Study of Language and Information, 1–27. Tenny, Carol. 1994. Aspectual Roles and the Syntax-Semantics Interface. Dordrecht: Kluwer. Tenyi, T., Herold, R., Szili, M., and Trixler, M. 2002. Schizophrenics show a failure in the decoding of violations of conversational implicatures. Psychopathology 35: 25–7. Terrace, Herbert. 2005. Metacognition and the evolution of language. In H. Terrace and P. Metcalfe (eds.), The Missing Link in Cognition. Oxford: Oxford University Press, 84–115. Thorne, A., Grün, R., Mortimer, G., Spooner, N. A., Simpson, J. J., McCulloch, M., Taylor, L., and Curnoe, D. 1999. Australia’s oldest human remains: age of the Lake Mungo 3 skeleton. Journal of Human Evolution 36: 591–612. Titone, D., Libbena, M., Niman, M., Ranbomb, L., and Levy, D. L. 2007. Conceptual combination in schizophrenia: Contrasting property and relational interpretations. Journal of Neurolinguistics 20: 92–110. Tomasello, Michael. 2003. Constructing a Language: A Usage-Based Theory of Language Acquisition. Cambridge, MA: Harvard University Press. Tomasello, Michael. 2008. Origins of human communication. Cambridge, MA: MIT Press. Tomasi, D. and Volkow, N. D. 2012. Resting functional connectivity of language networks: characterization and reproducibility. Molecular Psychiatry 17(8): 841–54. Tomblin, Bruce J., Records, Nancy L., Buckwalter, Paula, Zhang, Xuyang, Smith, Elaine, and O’Brien, Marlea. 1997. Prevalence of Specific Language Impairment in Kindergarten Children. Journal of Speech, Language, and Hearing Research 40: 1245–1260. Torrego, Esther. 1998. The Dependencies of Objects. Cambridge, MA: MIT Press. Trainor, Laural J. and Heinmiller, Becky M. 1998. Infants prefer to listen to consonance over dissonance. Infant Behavior and Development 21: 77–88. Thráinsson, Höskuldur. 1994. Icelandic. In E. König and J. van der Auwera (eds.), The Germanic Languages. London: Routledge, 142–89. Thráinsson, Höskuldur. 2007. The syntax of Icelandic. Cambridge: Cambridge University Press. Travis, Lisa deMena. 2010. Inner aspect: the articulation of VP. Dordrecht and London, Springer. Tripath, K. D. 2009. Thought, language and consciousness: Bhartrhari’s View of language. In Chaturvedi (ed.), 177–84. Truckenbrodt, H. 2006. On the semantic motivation of syntactic verb movement to C in German. Theoretical Linguistics 32(3): 257–306. Turing, Alan M. 1950. Computing machinery and intelligence. Mind 49: 433–60. Turken, A. U. and Dronkers, Nina. 2011. The neural architecture of the language comprehension network. Frontiers in Systems Neuroscience 5: 1–20. Uesaka, Miwako. 1996. The ‘te-i-ru’ Construction in Japanese: Interaction between Aspect and Syntax. MA thesis, McGill University. Uriagereka, Juan. 1995. Aspects of the Syntax of Clitic Placement in Western Romance. Linguistic Inquiry 26: 1–79.

References

371

Uriagereka, Juan. 1998. Rhyme and reason. Cambridge, MA: MIT Press. Uriagereka, Juan. 2008. Syntactic anchors. Cambridge: Cambridge University Press. van Urk, Coppe. 2010. On obligatory control: A movement and PRO approach. Ms. Available online at http://web.mit.edu/cvanurk/www/papers.html. Accessed 18 July 2012. Vangsnes, ystein Alexander. 2002. Icelandic Expletive Constructions and the Distribution of Subject Types. In P. Svenonius (ed.), Subjects, Expletives, and the EPP. New York: Oxford University Press, 43–70. Varley, Rosemary and Siegal, M. 2000. Evidence for cognition without grammar from causal reasoning and ‘theory of mind’ in an agrammatic reasoning and ‘theory of mind’ in an agrammatic aphasic patient. Current Biology 10: 723–6. Varley, Rosemary, Siegal, M., and Want, S. 2001. Severe impairment in grammar does not preclude theory of mind. Neurocase 7: 489–93. Varley, Rosemary, Klessinger, N., Romanowski, C., and Siegal, M. 2005. Agrammatic but numerate. Proceedings of the National Academy of Sciences, USA. 102: 3519–3524. Vergnaud, Jean-Roger. 1977/2008. Letter to Noam Chomsky and Howard Lasnik. April 17, 1977. Published in Freidin et al. (eds.), 3–15. Verkuyl, Henk J. 1972. On the compositional nature of the aspects. Dordrecht: D. Reidel Publishing Company. Vigliocco Gabriella, Vinsona, David P., Druksa, Judit, Barberb, Horacio, Cappa, Stefano F. 2011. Nouns and verbs in the brain: A review of behavioural, electrophysiological, neuropsychological and imaging studies. Neuroscience and Biobehavioral Reviews 35: 407–26. Vigneau, M., Beaucousin, V., Hervé, P.-Y., Duffau, H., Crivello, F., Houde, O., Mazoyer, B., and Tzourio- Mazoyer, N. 2006. Meta-analyzing left hemisphere language areas. Neuroimage 30: 1414–1432. Vigneau, M., Beaucousin, V., Hervé, P.-Y., Jobard, G., Petit, L., Crivello, F., Mellet, E., Zago, L., Mazoyer, B., and Tzourio-Mazoyer, N. 2011. What is right-hemisphere contribution to phonological, lexico-semantic, and sentence processing? NeuroImage 54: 577–93. Vikner, S. 1995. Verb movement and expletive subjects in the Germanic languages. Oxford studies in comparative syntax. Oxford: Oxford University Press. Vikner, Sten. 2005. Immobile complex verbs in Germanic. Journal of Comparative Germanic Linguistics 8: 83–115. Vinokurova, N. 2005. Lexical Categories and Argument Structure: A Study with Reference to Sakha. Doctoral Dissertation. LOT series, Utrecht University. Walenski, M., Weickert, T., Maloof, C., and Ullman, M. 2010. Grammatical processing in schizophrenia. Neupsychologia 48: 262–9. Ward, P. B. and Catts, S. V. 2002. Disturbed communication in schizophrenia: the role of poor pragmatics and poor mind-reading. Psychological Medicine 32: 1273–1284. Watson, A. R., Defterali, Ç., Bak, T. H., Sorace, A., McIntosh, A. M., Owens, D. G. C., Johnstone, E. C., and Lawrie, S. M. 2012. Use of second person pronouns and schizophrenia. British Journal of Psychiatry 200: 342–3. Watson, John B. 1913. Psychology as the Behaviorist Views It. Psychological Review 20: 158–77. Weerman, Fred. 1988. The V2 conspiracy: a synchronic and a diachronic analysis of verbal positions in Germanic languages. Dordrecht: Foris Publications, 1989.

372

References

Wenzlaff, M. and Clahsen, Harald. 2004. Tense and agreement in German agrammatism. Brain and Language 89: 57–68. Wenzlaff, M. and Clahsen, Harald. 2005. Finiteness and verb-second in German agrammatism. Brain and Language 92: 33–44. Werning, Marcus, Hinzen, Wolfram, and Machery, Edouard (eds.). 2012. The Oxford Handbook of Compositionality. Oxford: Oxford University Press. Westergaard, M. 2007. English as a Mixed V2 Grammar, Poznań Studies in Contemporary Linguistics 43(2): 107–31. Willems, Roel M., Benn, Y., Hagoort, P., Toni, I., and Varley, R. A. 2011. Communication without a functioning language system. Neuropsychologia 49: 3130–3135. Willems, Roel M., Özyürek, A. I., Hagoort, P. 2009. Differential roles for left inferior frontal and superior temporal cortex in multimodal integration of action and language. Neuroimage 47: 1992–2004. Williams-van Klinken, C., Hajek, J., and Nordlinger, R. 2001. Serial Verbs in Tetun-Dili: a preliminary account. In G. Hull and L. Eccles (eds.), Studies in the Languages and Cultures of East Timor, Vol. 4. Sydney: Academy of East Timor Studies, 48–60. Winter, Yoad. 2000. Distributivity and dependency. Natural Language Semantics 8: 27–69. Wittgenstein, Ludwig. 1922. Tractatus Logico-Philosophicus. London: Kegan Paul. Wittgenstein, Ludwig. 1953. Philosophical Investigations, tr. G. E. M. Anscombe. Oxford: Blackwell. Wisniewski, E. J. and Love, B. C. 1998. Relations vs. properties in conceptual combination. Journal of Memory and Language 38: 177–202. Wittig, R. M., Crockford, C., Wikberg, E., Seyfarth, R. M., and Cheney, D. L. 2007. Kinmediated reconciliation substitutes for direct reconciliation in female baboons. Proceedings of the Royal Society of London, Series B 274: 1109–1115. Woodard, Andrew. Forthcoming. Identity in grammar. PhD thesis, University of Durham. Worah, Pallavi. 2013. Minimalism in Semantics. PhD thesis. University of Durham. Wu, Y. and Bodomo, A. 2009. Classifiers 6¼ Determiners. Linguistic Inquiry 40: 487–503. Wurmbrand, Susanne. 2001. Infinitives: restructuring and clause structure, Berlin and New York: Mouton de Gruyter. Xu, F. 2005. Categories, kinds and object individuation in infancy. In L. Gershkoff-Stowe and D. Rakison (eds.), Building object categories in developmental time. Papers from the 32nd Carnegie Symposium on Cognition. Hillsdale, NJ: Erlbaum, 63–89. Yamada, Jeni E. 1990. Laura: a case for the modularity of language. Cambridge, MA: MIT Press. Yip, Moira, Maling, Joan, and Jackendoff, Ray. 1987. Case in tiers. Language 63: 217–50. Young, E.C., Diehl, J. J., Morris, D., Hyman, S. L., and Bennetto, L. 2005. The Use of Two Language Tests to Identify Pragmatic Language Problems in Children with Autism Spectrum Disorders. Language, Speech and Hearing Services in Schools 36: 62–72. Zaenen, Annie, Maling, Joan, and Thráinsson, Höskuldur. 1985. Case and grammatical functions: the Icelandic passive. Natural Language and Linguistic Theory 3: 441–83. Zahavi, Dan. 2006. Subjectivity and Selfhood: Investigating the First-Person Perspective. Cambridge, Massachusetts: MIT Press. Zamparelli, Roberto. 2000. Layers in the determiner phrase: Outstanding dissertations in linguistics. New York: Garland Pub.

References

373

Zeijlstra, Hedde. 2012. There is only one way to agree. Ms., http://ling.auf.net/lingbuzz/001435. Accessed 1 March 2012. Zimmermann, Thomas Ede. 2012. Compositionality problems and how to solve them. In Werning et al. (eds.), 81–108. Zubizarreta, Maria Luisa. 1998. Prosody, focus, and word order. Cambridge, MA: MIT Press. Zwart, Jan-Wouter. 2009. Structure and order: asymmetric merge. In Boeckx (ed.), 96–118. Zwarts, Joost. 2005. Prepositional aspect and the algebra of paths. Linguistics and Philosophy 28 (6): 739–79 Zylinski, Sarah, Osorio, D., and Shohet, A. J. 2009. Feature and texture perception in the cuttlefish Sepia officinalis. Journal of Vision 9(13): 1–10.

Index Agree 101, 216–19 Ailly, Pierred’ 51 alignment parameter hierarchy 200 aphasia 268–74 and communication 269 assertion 323 as pointing 323, 325–6 atom, atomism (lexical) 17, 21, 47–53, 100, 106, 244 Autism Spectrum of Disorders (ASD) 279–80 and schizophrenia 287 autonomy 13, 87, 115, 202, 240–1, 263 unreal at the level of the brain 293 Aristotle 15, 304–5 Bhartrihari 19–23, 327 behaviourism 250 and species-uniqueness 250 Bickerton, D. 42, 239 biolinguistics 238–9, 242, 248–9, 258, 335 Blackfoot 50 Bleuler, E. 281 Bloom, P. 277 Bloomfield, L. 11–12 Boeckx, C. 26, 77, 109, 176, 179 Borer, H. 226, 229, 232 Burge, T. 277 Campbell, J. 285 Carey, S. 39, 45–6 Cartesian see linguistics Case 78, 132, 144, 183, 313 and Agree 218–19 as cross-linguistically variable 209–10 exceptional 213 Filter 214–15, 218–19 meaning of 205–7 as morphological 203

and phase domains 206 as (ir)rational 202–4, 206, 211, 224, 226 and referentiality 212 as reflecting the formal ontology of language 207, 235 as reflecting the relational interpretation of nominals 203 as relational 203–4, 211, 213 and telicity/boundedness 224–36 and Tense 218, 221–4 and thematic roles 206–7, 211–12, 232 no uninterpretable features 208, 211, 216, 220, 226 Chaika, E. 283 charity 331 Chomsky, N. 6, 12–14, 18, 23, 26, 31, 46, 54, 76, 85, 97, 179–80, 202–3, 206, 214–15, 217–18, 240, 248, 250, 253, 288 Christiansen, M. and N. Chater 242, 258 Cinque-hierarchy 181–2 clitics accusative 158–61 and Case 213 dative 161–6 hierarchical structure of 167 mapping to morpho-phonology 215, 219 partitive and neuter 156–8 personal 168–70 and strong pronouns 171–2 cognition 241 cognitive linguistics, 59 Colourless green ideas sleep furiously 292 composition(ality) 51, 79, 88–95, 97, 108–9 as locality 109 Computational-Representational Theory of Mind 253, 275, 277, 298–9 communication 238, 243–4, 246, 269, 330

376

Index

concepts 38–40, 47, 50–1, 53–7, 275–6, 259 concept combination 51 ‘Conceptual-Intentional’ (C-I) systems 6, 178 consciousness 260–1 content and form 85–6, 95, 241, 292, 299–300, 304, 332 of grammar 76–8 see also grammar; meaning of propositions as opposed to sentences 301 context 151, 173, 286 convention 334 cooking 85 copular constructions 313–16 core cognition 275 Croft, W. 59, 62–5, 69, 71–2 Crow, T. 19, 294 D-to-D movement 168 Davidson, D. 63, 83, 245, 259, 275–7, 299–300, 323, 328–35 d’Errico, F. 256, 260 definite descriptions 126–7, 133, 141 deixis 80, 94, 144 and assertion 323 in Dative clitics 166 and rationality 295 and thought disorder 283 Descartes, R. xvi, 249 description 54 ditransitive alternation 232–3 ECM 213 Elbourne, P. 142–3, 153, 267 embedding as meaningful 106 embodiment 254, 258 empiricism 21, 250–1 equivalence scheme (Tarski) 305–7, 310, 322 ergative 197–201 Evans, N. 242, 246–7, 260 Everett, D. 175 events 83

and states 235–6 evo-devo 251–2 experience 62 externalism 250–1, 300 externalization (of language) 6–7, 18, 134, 177–8, 179, 184, 195, 238, 243–6, 303 in agrammatic aphasia 269 (not) an afterthought or accident 6, 254, 259, 267–8, 260, 267, 327 of Case relations 218–20 with regard to SLI 265 extraction 136–7 factivity 133–4 Final-over-Final Constraint (FOFC) 189–93 Finnish 224 Fodor, J. A. 13, 39, 48, 50, 240–2, 253, 260, 263, 275, 298–9 form see content Formal Thought Disorder (FTD) 283 see also schizophrenia concept combination in 289 expressive semantic theory 288 in functional MRI 291 and grammatical meaning 288–9 and monothematic delusions 285 pragmatic theory of 286 Frege, G. 1, 3, 79, 132 Gallistel, R. 45–6, 275 gradualism 252, 256 grammar in the brain 292–3 content of 76–7, 83, 85, 91–2, 96 defining truth 295 device for extended deixis 80 of deixis 144 epistemological significance of 1, 119, 174 as foundational 84, 87, 174, 295, 303, 322 as an invariant 249 limits of 296, 317–20 and logic 7–8, 16, 328, 331, 335 as organizing a formal ontology 72–3, 174 see also ontology

Index as a reference system 47, 148, 220, 248 as relational 75, 77, 82, 101 and thought, and reality 296 vs. ‘syntax’ 277, 288, 292, 293, 295 units of 102 Harley, H. 48 Haspelmath, M. 9, 33 headedness 52, 92 Heider-Simmel 40 Himmelmann, N. 60, 66–7 Homo Heidelbergensis 255–6 Horwich, P. 301, 305, 310 human nature 253, 267 Hume D. 40, 286 Husserl, E. 1, 11 Ildefonso 266–8 indexicals 43, 93–4, 99, 123, 139–43, 148–9, 150–1 as ‘essential’ 172–3 inserted in the edge 154 (not) as lexically semantic 149–53, 173 not a unified lexical class 153 indexicalization 47 inflected infinitives 221–2 interfaces 76, 79, 85, 97, 100 ‘radical’ conception 202 intensionality 55–6, 90–1, 133, 275–6 intentionality 36, 246 internalism 56, 250–2, 301, 303, 327, 332 Jackendoff, R. 68, 301 Jespersen, O. 7–9 Kraepelin, E. 281 Kuperberg, G. 284, 293 Langacker, R. 59 Language of Thought 6, 50, 238, 241, 247 not required 261 Laura 264 Leiss, E. 31, 296 Levinson, J. 242, 246–7, 260

377

lexemes 47, 60 as bundles of perceptual features 47–50 and concepts 54 lexicalization 35, 47–53 Linear Correspondence Axiom 186 linguistic genotype 261, 239, 261–3, 266, 275–80 linguistic relativity 248 see also Whorf, B. linguistic variation see variation linguistics Cartesian xvi, 8, 62, 175, 239, 242–3, 247–9, 272–3, 283, 296, 331–2, 336 Un-Cartesian 14, 175, 202, 241, 248, 260–1, 282–3 Lithuanian 50 logic 81 Longobardi, G. 118–19, 123, 154, 168 McKenna, P. 288 Main Point of Utterance 137 Mausfeld, R. 40 meaning in the brain 292–5 grammatical 66–7, 69–70, 91–2, 106, 116, 217, 288–9, 293, 309–10 vs. associative cognition 289 in the philosophy of language 2 not a matter of rational choice 333 measure of mind 295 mental as an ‘anomaly’ 299–300 science of 300 mereology 71, 105–7, 223, 227, 232, 235–6 Merge 76–7, 80, 88, 109–10, 251 Merleau-Ponty, M. 285 Minimalism 6, 76, 85, 96, 217, 240, 248, 262 mistakes 48 Modern Synthesis 252 modes of signifying 69–71 Modists 15, 23–32, 50–1, 69, 74, 296 modularity 3, 238–9, 241–2, 259–60, 263–6, 280, 292 as circular 263–5 and Theory of Mind 272–3

378

Index

Moro, A. 43, 79–83, 311–16 morphological decomposition 52 Müller, F. M. 296, 329 names (proper) 122–4, 129, 141, 150, 152, 317 nativism 240 Neanderthals 253–6, 257, 260 naturalism 10, 28–9, 33–4, 174, 297, 331 metaphysical and methodological 297–9 nominalists 50, 87 normativity 334–5 ontology (formal) 46, 59, 64, 68–71, 73, 79, 86–7, 89, 107, 121–2, 139, 173, 176, 224, 235–6, 247 commitments 118–19 as grammatical 83–4, 95, 139, 302–3 limits of 321–2 and phases 103, 107, 205 other minds 57–8 Out of Africa 254–5 Pāṇini 16–18 Partee, B. 296, 301, 303 partial control 221 parts of speech 37, 59–69, 75 perception 35–6, 245 semantics of 38 Person 78, 155, 217 complementary with Gender/Number 168 in Dative clitics 167–8 first Person 142–9, 150–2 semantics of 142–4 perspectives 56, 68, 71, 73 phases 80, 102, 131, 137, 153, 173, 204, 206 and cross-phasal dependencies 205 Pietroski, P. 31, 301 Pinker, S. 13, 263 pointing 80, 245, 303 and grammar 248 Port Royal xvii, 2, 5, 10, 32 predication 63, 75–6, 82, 87–8 not a part of speech 75 Preminger, O. 218

principles and parameters 14, 77, 176, 179 alignment parameter 197–201 head parameter 184–93 null subject parameter 193–7 PRO 220–3 propositions (and propositionality) 62–3, 246, 278 belief 325, 330 propositional attitudes 301–2, 310–11, 325, 329–30 Pylkkänen, L. 292 quasi-copula (seems) 314 Quine, W. v. O. 297, 299 rationality 143, 245, 249, 278, 283–4, 292, 332 realism 86, 322–7 recursion 80, 101–2, 107–8, 319 as cyclic 111 not direct 110–15 in mind-reading 245 reference 36–7, 57, 91–4, 99, 117–18, 128, 297–9 and event ontology 225 and extraction 137 to facts 133–8, 143 functional 42, 57, 99 as grammatical 37, 82, 91–3, 123, 128, 130, 138–9, 148, 298, 303 to kinds 129–30 organized hierarchically 117, 148, 156, 167 as the problem rather than solution 303 and reason/rationality 295 as scalar and hierarchical 126–8, 173 strong in Dative clitics 164 topological 117–18, 130 as tracked by Case 209, 212 to truth 131–2 Reuland, E. 147 rigidity of clauses 139 as an explanatory problem 303 of proper names 124 roots 50 Russell 78–9, 82–3

Index Sakha 50 Schaller, S. 266–8 schizophrenia 280–92 and aphasia 284 and autism 287 as a disturbance of the Self 286 and laterality 291 as a logopathy 262 and reference 283–4 selection 106–7 Self (‘first person’) 143, 173, 285–6 in D. Hume 286 disturbance of 286 semantics 77, 296–304 see also content as explanatory 150 grammatical 60, 116, 140, 153, 301–2 see also meaning outside of the head 300 as (not) psychology 301–3 sentence 80 not a part of speech 76, 81 serial verbs 111–12 Sigurðsson, H. 206, 208–9, 215, 217 Skinner, B. F. 250–1, 299 slifting 135–7 speciation 239, 249–50 event 253–4 and language 258, 260 species and language 254, 260 reality of 252–3 -specificity 62, 239, 244, 250, 303, 328 Specific Language Impairment (SLI) 265–6 and autism 266 Stringer, C. 254–5 Strong Minimalist Thesis 6 Tagalog 50, 65 Tarski, A. 305, 322 Tattersall, I. 19, 252–3, 255–6, 258 Tense Underspecification Hypothesis 274 telicity 224–35 thematic structure 76, 207, 232 Theory of Mind 244–5, 270–4

379

in autism and schizophrenia 279–80, 287 in the brain 294 dependent on language 271 and grammar 287–8 many different things 273 as a module 272–3, 287, 295 no predictions for propositionality 295 thought 1, 7, 58, 99, 240, 242, 247–8 in animals 98, 276–7 as conscious 261 as co-varying with the linguistic genotype 261–3 dependent on language 259 disordered 279, 281–5 without conceptual scheme 278 ‘embodied’ in language 258 evolution of 238–9 invariant, like grammar 249 Language of see Language of Thought limits of 318–22 as natural and biological 335 non-propositional 278 science or theory of 7, 328–30, 335–6 in severe aphasia 268–72 and truth 329–30 Titone, D. 289–90 Tomasello, M. 242–6, 283 Topological Mapping Theory 119, 130, 148, 154, 168, 173 extended 130, 134, 139, 153 triangulation 21–2, 245 truth 85, 132, 304–14 and assertion 323 -conditions 305 and copular constructions 313–16 deflation of 322 epistemology of 323 grammar of 311–21 as grammatical 304–17, 328 as independent from belief 277–8 as the limit of grammar 318 as objective 304, 307 and realism 323–7 theories of 305 vs. survival 276

380

Index

unaccusatives 230 uninterpretable features 78 universal grammar 77, 240 in India 15–23 genetic variation in 261, 262, 279 organological conception 240 Un-Cartesian hypothesis xx, 14, 175, 180, 183, 202, 241, 248, 260, 262, 269, 274, 282–3, 285, 291–2, 295, 332–3, 336 see also linguistics Unified Theory of Thought, Meaning and Action (Davidson) 300, 328–9, 331 Universal Theta Assignment Hypothesis (UTAH) 182

V2 134–5 Varley, R. 270–2, 274 variation biolinguistic 262–95 cross-linguistic 195–201 fundamentally lexical 181, 247 in semantics 177 Vergnaud, J. R. 214 Williams syndrome 263–4 Wittgenstein, L. 84 Whorf, B. 247 word grammatical 61

The Phylosophy of Universal Grammar

Recommend Documents