The Syntax of Agreement and Concord - Mark C. Baker

This page intentionally left blank

T H E S Y N TA X O F AG R E E M E N T A N D C O N C O R D

In this series 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115

l i l i a n e h a e g e m a n: The syntax of negation p a u l g o r r e l: Syntax and parsing g u g l i e l m o c i n q u e: Italian syntax and universal grammar h e n r y s m i t h: Restrictiveness in case theory d . r o b e r t l a d d: Intonational morphology a n d r e a m o r o: The raising of predicates: predicative noun phrases and the theory of clause structure r o g e r l a s s: Historical linguistics and language change j o h n m . a n d e r s o n: A notional theory of syntactic categories b e r n d h e i n e: Possession: cognitive sources, forces and grammaticalization n o m t e r t e s c h i k - s h i r: The dynamics of focus structure j o h n c o l e m a n: Phonological representations: their names, forms and powers c h r i s t i n a y. b e t h i n: Slavic prosody: language change and phonological theory b a r b a r a d a n c y g i e r: Conditionals and prediction c l a i r e l e f e b v r e: Creole genesis and the acquisition of grammar: the case of Haitian creole h e i n z g i e g e r i c h: Lexical strata in English k e r e n r i c e: Morpheme order and semantic scope a p r i l m c m a h o n: Lexical phonology and the history of English m at t h e w y. c h e n: Tone Sandhi: patterns across Chinese dialects g r e g o r y t. s t u m p: Inflectional morphology: a theory of paradigm structure j o a n b y b e e: Phonology and language use l a u r i e b a u e r: Morphological productivity t h o m a s e r n s t: The syntax of adjuncts e l i z a b e t h c l o s s t r a u g o t t and r i c h a r d b. d a s h e r: Regularity in semantic change m aya h i c k m a n n: Children’s discourse: Person, space and time across languages d i a n e b l a k e m o r e: Relevance and linguistic meaning: the semantics and pragmatics of discourse markers i a n r o b e r t s and a n n a r o u s s o u: Syntactic change: a minimalist approach to grammaticalization d o n k a m i n k o va: Alliteration and sound change in early English m a r k c . b a k e r: Lexical categories: verbs, nouns and adjectives c a r l o ta s . s m i t h: Modes of discourse: the local structure of texts r o c h e l l e l i e b e r: Morphology and lexical semantics h o l g e r d i e s s e l: The acquisition of complex sentences s h a r o n i n k e l a s and c h e r y l z o l l: Reduplication: doubling in morphology s u s a n e d wa r d s: Fluent aphasia b a r b a r a d a n c y g i e r and e v e s w e e t s e r: Mental spaces in grammar: conditional constructions h e w b a e r m a n, d u n s ta n b r ow n and g r e v i l l e g . c o r b e t t: The syntax-morphology interface: a study of syncretism m a r c u s t o m a l i n: Linguistics and the formal sciences: the origins of generative grammar s a u m u e l d . e p s t e i n and t. d a n i e l s e e ly: Derivations in minimalism pa u l d e l a c y: Markedness: reduction and preservation in phonology y e h u d a n. f a l k: Subjects and their properties p. h . m at t h e w s: Syntactic relations: a critical survey m a r k c . b a k e r: The syntax of agreement and concord Earlier issues not listed are also available

CAMBRIDGE STUDIES IN LINGUISTICS General editors: p. a u s t i n, j . b r e s n a n, b. c o m r i e , s . c r a i n, w. d r e s s l e r , c . j . e w e n, r . l a s s , d . l i g h t f o o t, k . r i c e , i . r o b e r t s , s . r o m a i n e , n. v. s m i t h

The Syntax of Agreement and Concord

T H E S Y N TA X O F AG R E E M E N T A N D CONCORD M A R K C . BA K E R Rutgers University

CAMBRIDGE UNIVERSITY PRESS

Cambridge, New York, Melbourne, Madrid, Cape Town, Singapore, São Paulo Cambridge University Press The Edinburgh Building, Cambridge CB2 8RU, UK Published in the United States of America by Cambridge University Press, New York www.cambridge.org Information on this title: www.cambridge.org/9780521855471 © Mark Baker 2008 This publication is in copyright. Subject to statutory exception and to the provision of relevant collective licensing agreements, no reproduction of any part may take place without the written permission of Cambridge University Press. First published in print format 2008

ISBN-13 978-0-511-38884-2

eBook (NetLibrary)

ISBN-13 978-0-521-85547-1

hardback

ISBN-13 978-0-521-67156-9

paperback

Cambridge University Press has no responsibility for the persistence or accuracy of urls for external or third-party internet websites referred to in this publication, and does not guarantee that any content on such websites is, or will remain, accurate or appropriate.

This book is dedicated to the three wonderful children – Catherine, Nicholas, and Julia – that God has given me to enliven my journey through this life. I only ask that they not fight about who gets to be the noun, who the verb, and who the adjective.

Contents

Acknowledgments List of abbreviations and conventions

page xi xiv

1

Introduction: category distinctions as a window on the theory of agreement

1

1.1 1.2 1.3 1.4 1.5

A generalization to be explained The incompleteness of previous discussions What a better theory could look like What is in this book What is not in this book

1 3 6 6 7

2

Basic agreement and category distinctions

12

2.1 2.2 2.3 2.4 2.5 2.6

The generality of the categorical asymmetries in agreement The category-theoretic infrastructure The agreement-theoretic contribution Explaining the basic categorical asymmetries in agreement Issues arising Conclusion

13 27 40 48 56 64

3

The unity of verbal and adjectival agreement

65

3.1 3.2 3.3 3.4

Downward agreement on adjectives Upward agreement on verbs Verbs that cannot agree in person, like adjectives Conclusion

67 74 85 107

4

Explaining the restriction on person agreement

111

4.1 4.2 4.3 4.4 4.5 4.6

Person agreement and other categories Operator-variable agreement and Agree A locality condition on first and second person variables On the strictness of locality conditions involving heads Deriving the SCOPA Conclusion

112 121 124 138 142 148 ix

x

Contents

5

Parameters of agreement

153

5.1 5.2 5.3 5.4 5.5 5.6 5.7 5.8 5.9 5.10 5.11 5.12

Introduction: parameters and other kinds of variation Agreement on tense Agreement on FA and the formulation of the parameters Agreement on complementizers Agreement on determiners Agreement on adpositions Agreement on v Agreement on the linker head Agreement in auxiliary constructions A third value for the Direction of Agreement Parameter Many little parameters or two big parameters? General conclusion

153 157 171 178 184 191 196 206 207 214 219 244

Appendix: Table of languages and their agreement properties References Index

246 254 264

Acknowledgments

As I get older, I seem to be losing my sense of history. As in many families, I have fewer pictures documenting every developmental stage of my youngest child than I do of my oldest child. In the same way, I feel like I have less to say by way of preface and acknowledgments with each book that I write. But there are still plenty of people to thank, so here goes. I do not think that I would have started another book so soon after finishing Lexical Categories (Baker 2003a) if Andrew Winnard of Cambridge University Press had not approached me about the possibility of publishing some manuscripts he had seen on my website. Talking to him made me realize that there was in fact a common theme underlying many of those studies of particular languages, which I had not fully realized, and that common theme had to do with agreement. I further realized that my theory of lexical categories raised some huge unanswered questions about agreement – questions that had not been answered by other people’s theories either. So I decided to delve into this topic with some gusto. As I sought to write a capstone essay that would draw together my little discoveries about agreement in particular languages (Mapudungun, Kinande, Lokaa, Icelandic), the outlines of a broader theory began to emerge, and the capstone essay took over the book as a whole. I fear that some of the original papers are still unpublished (or published in less-visible venues), but I hope that the final book is more consistent and unified, and paints a bigger picture than it otherwise would have. It was also personally rewarding for me to dive back fully into pure linguistic research after some time spent as department chair and being involved in some cognitive science projects. I thank Andrew and Cambridge University Press for the excuse to do this. Of course, one needs not only a project worth working on, but also some time to work on it. For meeting that need, I thank Rutgers University for providing a sabbatical leave and a competitive leave fellowship, and the American Philosophical Society for awarding me a sabbatical leave fellowship (funded by the Mellon Foundation) for 2005–2006. I would not have had the freedom

xi

xii

Acknowledgments

to do the survey of 108 languages reported in chapter 5 if I had not received this special support. One also needs some facts to work with, and it is nice when some of them are new ones that people have not had access to before. In this most recent period of my career, I have been privileged to work with native-speaker linguists on their fascinating languages in ways that have been inspiring and helpful to me. This group includes Elisa Loncon (Mapudungun), Alexander Iwara (Lokaa), Willie Udo Willie (Ibibio), and – of special significance to this particular study – Philip Mutaka (Kinande). Kinande gets the most press in what follows, but meditating on the challenges and wonders of all of these languages has provided the impetus to do this work. I also wish to thank many linguists who have generously answered email questions arriving from me “out of the blue” about the languages that they know when I felt some small but crucial piece of a puzzle was missing. These are acknowledged individually at the relevant points in the text, but here I single out Halldór Sigurð sson for service beyond the call of duty, since Icelandic turned out to be particularly important at several points. And one needs some inspiring colleagues, who can stimulate you, challenge you, and help you put your ideas to the test. I have been lucky enough to present this material in a number of colloquiums and more extended forums, including graduate seminars at Rutgers University in spring 2005 and fall 2006, an LSA summer institute class at MIT in 2005, extended colloquia at UCLA and Georgetown University in 2006, and week-long classes at the University of the Basque Lands and the LOT summer school at the University of Amsterdam in 2006. I thank all of these audiences for their encouragement and input, including José Camacho, Liliana Sanchez, Roger Schwarzchild, David Pesetsky, Esther Torrego, Seth Cable, Hilda Koopman, Anoop Mahajan, Philippe Schlenker, Carson Schütze, Raffaella Zanuttini, Bob Franks, Michael Diercks, Itziar Laka, Javier Ormazabal, Myriam Uribe-Extebarria, Hans Broekenhuis, Jenneke van der Wal, and others that I forgot – or whose names I never even learned. I thank Jessica Rett and Cedric Boeckx for sending me written comments on parts of the manuscript. I thank my students who have also been interested in issues of case and agreement, including Vita Markman, Natalia Kariaeva, Jessica Rett, and Carlos Fasola. Three people stand out for special thanks in this category. First, I thank Chris Collins for early collaborative work on what we originally called “the Bantu Parameter.” This provided the seeds for what became chapter 5, once I was finally able to investigate how that parameter might apply to other languages. Second, I thank Ken Safir for his enormous influence on chapter 4, helping me to get up to speed (or at least closer to speed) on the issues of binding, person, and pronoun interpretation that are crucial there, and showing me how to

Acknowledgments

xiii

say what I wanted to say in a much cleaner and more straightforward way. Third, I thank Carlos Fasola, who in the guise of being my research assistant helped me to discover the properties of many of the 108 languages discussed in chapter 5, and helped to nurture in us both a common pleasure in grammar-reading. One might not literally need a loving and supportive family in order to write a book like this, but I certainly would not want to do it any other way. Many thanks to my wife Linda and my three children for much help, support, prayers, and companionship along the way. Finally, I am convinced that I needed the grace of the Lord Jesus Christ, the love of God, and the fellowship of the Holy Spirit to do something like this. I thank my God for providing all of the people and resources listed above, and everything else besides.

Abbreviations and conventions

In this book, I cite examples from a large number of languages, some of which I do not know well. This presents certain challenges for effective glossing. The safer course would be to simply follow the glossing practice of the source that the example is taken from. The problem with this is that it multiplies greatly the number of abbreviations used, and can obscure comparison by giving similar morphemes very different glosses. This problem could be addressed by trying to impose a uniform system of glossing on all of the languages considered. But that creates other problems: in particular, languages might have morphemes that are similar in their usage but not identical in all respects, and I might not know enough to do it accurately. I have tried to strike a middle path between these two courses, making the glosses more uniform when I thought I could do it with reasonable accuracy and when the morphemes are relatively important to my topic – in particular, when they are agreement morphemes. I am not fully satisfied with the results, and experts on the relevant languages may be even less so. But that is what I did. Agreement morphemes (particularly those on verbs) are glossed by a complex symbol that begins with a number indicating the person of the agreed-with phrase (1, 2, or 3), then has a lower-case letter indicating the number of the agreed-with phrase (s, singular; d, dual; p, plural), and then a capital letter indicating the grammatical function of the agreed-with phrase (S, subject; O, object; P, possessor; A, absolutive; D, dative/goal; E, ergative). Thus, 1pS means first person plural subject agreement, 3sO means third singular object agreement, and so on. Sometimes one member of this triple is missing when the corresponding category is not marked – for example, when the agreement indicates person but not number, or vice versa. When two agreement factors are expressed with a single portmanteau morpheme, their features are separated with a slash. The reader should also note that 1, 2, and 3 have two meanings in agreement morphemes: they can mean first, second, or third person (all languages), or they can mean a third person noun phrase in class 1 (human singular), class 2 (human plural), or class 3 (singular) in a Bantu language or Lokaa. Thus 1sS xiv


xv

always means first person singular subject, but 1S in the gloss of a Niger Congo language means subject agreement with a noun of class 1 (third person human singular). (1S in the gloss of a non-Bantu language could mean first person subject agreement, with number unspecified.) I hope this will not be unduly confusing. Other abbreviations used in the glosses of linguistic examples are as follows. Readers should consult the original sources for more on what these categories amount to in particular languages. 1–19 a,b,c abs acc act adj adv aff agr an aor appl art asp assoc aug aux ben caus cl comp compl cond conj cont contr cs dat decl dem

noun class (Bantu) gender/number categories (Southern Tiwa) absolutive case accusative case actor adjectival adverbial affirmative agreement animate aorist tense applicative morpheme article aspect associative marker augmented auxiliary benefactive applicative marker causative noun class (Bantu), classifier (Tariana) complementizer completive conditional conjunction continuous aspect contrastive construct state (Berber) dative case declarative demonstrative

xvi


desid det dir disj dr dur dyn erg eu ext f fam foc fut fv gen ger hab i imp impf inan ind indf inf instr int intrans inv irr lk loc log m n neg ni nom noml np

desiderative determiner direct disjunctive prefix directional durative dynamic ergative case euphonic extended aspect (Bantu) feminine gender familiar focus future final vowel (Bantu, indicative mood marker?) genitive case gerund habitual aspect irrational imperative imperfective inanimate indicative indefinite infinitive instrumental intentional intransitive inverse irrealis linker locative logophoric masculine gender neuter gender negative noun incorporation nominative case nominalizer nonpast tense

Abbreviations and conventions npst nsf obj obl opt pass past perf pl poss pp pred pres prev prog ptpl q r rcp refl rel rem rep rpst sbjn sg ss stat sub subj t ta th thsy top tr unposs val vbzr veg

nonpast tense noun suffix (determiner?) objective case marker oblique optative passive past tense perfective aspect plural number possessor past perfective predicative head present tense preverb progressive aspect participle question particle rational recent past reflexive relative remote past reported recent past subjunctive singular number same subject stative aspect subordinate subject Tense, unspecified tense marker Transitive animate (Algonquian) thematic hearsay particle topic transitive unpossessed validator verbalizer vegetable gender (Mayali)

xvii

xviii


vis x

visible special gender class in Burushaski

The following are abbreviations of linguistic terms: names of principles, grammatical categories, language families, and construction types. a, ap c, cp d, dp deg ecm ecp epp fp ie n, np nc p, pp pf plc plc(h) scopa spec, xp t, tp v, vp v, vp vso, sov, etc.

wals

adjective, adjective phrase complementizer, complementizer phrase determiner, determiner phrase degree head Exceptional Case Marking Empty Category Principle “Extended Projection Principle” feature (triggers the movement of a phrase to the category that bears it) functional phrase Indo-European languages noun, noun phrase Niger-Congo languages adposition (preposition or postposition), adpositional phrase Phonological Form Person Licensing Condition Person Licensing Condition applied to Heads Structural Condition on Person Agreement Specifier of XP tense head, tense phrase light verb (abstract verbal element, assigner of external argument), light verb phrase verb, verb phrase Verb-subject-object word order; subject-object-verb order, etc. World Atlas of Language Structures

Finally, the following are some conventions used in presenting examples: (x) (*x) *(x)

The example has the same grammatical status with or without X included. The example is good without X, but bad when it is included. The example is bad unless X is included.

In some cases, an agreement morpheme and the NP that it agrees with are both underlined.

1

Introduction: category distinctions as a window on the theory of agreement

1.1

A generalization to be explained

Even though agreement phenomena are some of the most familiar and wellstudied aspects of grammar, there are certain basic questions that have rarely been asked, let alone answered. One such question concerns the fact that, in many languages, the three major lexical categories – noun, verb, and adjective – behave quite differently with respect to agreement. Words of all three categories can bear similar, even cognate, inflectional affixes. But verbs are consistently the most prolific agreers, often agreeing with their subjects in person, number, and gender features, and sometimes agreeing with their objects in these features as well. Adjectives clearly participate in agreement, but they do so more modestly: they rarely or never agree with more than a single nominal, and they can agree in number and gender but they typically do not show first or second person agreement forms. In light of these differences, traditional grammar often says that adjectives participate in concord, a distinct phenomenon from the agreement that involves verbs. Nouns are more conservative still. Although they are often inflected, sometimes with the same affixes that adjectives take, nouns do not need to agree with another NP in their environment the way that an adjective in a similar structural configuration does. My thesis in this book is that the true theory of agreement should be able to explain these robust cross-categorial differences, and that much can be learned about agreement by seeking a theory that meets this condition. To illustrate the fundamental contrasts in agreement behavior that I have in mind, consider (1) from Swahili. (1)

a. Ni-li-kuwa ni-ki-som-a. (Ashton 1949) 1sS-past-be 1sS-cont-read-fv ‘I was reading.’ b. Ni-Ø m-refu. 1sS-be cl1-tall ‘I am tall.’ 1

2

Introduction c. Ni-li-po-kuwa ki-jana . . . sasa ni-li-po m-tu m-zima, . . . 1sS-past-when-be cl7-child now 1sS-be-when cl1-man cl1-whole ‘When I was a child . . . Now that I am a man . . .’

The copular verbs in all three examples manifest agreement with an understood subject that is first person, singular, and animate; let us put that aside.1 The interesting differences are found on the postcopular lexical category. The main verb in (1a) also shows agreement that is first person, singular, and animate; its prefix is thus identical to the one on the copula. (2) shows a transitive verb in Swahili, in which the verb agrees with its understood object in person, number, and gender as well as with its subject. (2)

Juma a-li-ni-ambia kwamba . . . Juma 1S-past-1sO-tell that ‘Juma told me that . . .’

(Vitale 1981:62)

In contrast to the main verb in (1a), the predicate adjective in (1b) cannot bear the first person singular prefix ni. Rather it must bear the class 1 prefix m. This morpheme expresses singular number and human/animate gender, in partial agreement with the subject, but it does not express first person. The same prefix is thus used on the adjective when the subject is third person, as in (3a). (3b) shows an adjective with a different prefix, manifesting a different combination of number and gender in agreement with its subject. This confirms there is partial agreement on adjectives in Swahili, but not complete agreement with first and second person subjects. (3)

a. Hamisi yu-Ø m-refu (Ashton 1949) Hamisi cl1-be cl1-tall ‘Hamisi is tall.’ b. Mi-zigo hii mi-zito. cl4-loads these cl4-heavy ‘These loads are heavy.’

Finally, the first predicate nominal ki-jana ‘child’ in (1c) is different from both the verb and the adjective in that it does not even agree with the subject in gender. It bears the class 7 agreement prefix ki, reflecting the diminutive meaning of the predicate noun, not the class 1 agreement prefix m that the 1 In contemporary Swahili, the stem of the verbal copula is normally omitted in the simple present, when it bears no morphology other than the subject agreement marker. (The stem li does show up in some archaic expressions, such as proverbs, suggesting that this is an innovation.) Swahili also has an uninflected copular particle ni. It is not clear whether the ni in (1b) is an agreement marker or this copular particle, but this is orthogonal to the point I want to make using this example.

The incompleteness of previous discussions

3

first person singular subject triggers on the predicate adjective in (1b) (which also appears on the second predicate nominal m-tu m-zima ‘full-grown man’ in (1c)). A more striking example of a predicate nominal that does not agree with its subject is shown in (4), where the predicate noun ‘clouds’ fails to agree with the subject ‘sign of rain’ in gender or in number. (4)

Dalili y-a mvua ni ma-wingu. (Ashton 1949) cl9.sign cl9-assoc cl9.rain pred cl8-clouds ‘Clouds are a sign of rain.’

So verbs in Swahili show full agreement, adjectives show partial agreement, and nouns are inflected but do not agree. This is also true for many other, unrelated languages, as we shall see in chapter 2. 1.2


Although this pattern of facts is reasonably well known, at least for IndoEuropean languages and Semitic languages, linguists have rarely attempted to explain it. Consider, for example, Stassen’s (1997) extensive typological study. Stassen clearly recognizes the empirical connection between being a verb and bearing person agreement, stating it as follows (p. 38): (5)

The Agreement Universal If a language has person agreement in intransitive main clauses, this agreement will at least be used in sentences with event predicates [i.e. with prototypical verbs – MCB].

In connection with this universal, he writes: As far as I am aware, no known principle of linguistic theory prevents us from imagining a possible natural language with some form of person agreement where this agreement can be used only with predicate adjectives or only with predicate nouns, but not with predicate verbs. However, in my sample [of 410 languages – MCB] I have not found a single instance of a language in which such a situation can be observed. In this book, no attempt at an explanation for the Agreement Universal will be made.

In contrast, this book will make an attempt at explaining the Agreement Universal and related facts, using the tools of formal generative linguistics.2 2 Stassen sees this as only a one-way implication, from verbiness to person agreement, not a bidirectional implication as I argue. For a discussion of languages in which adjectives and/or nouns bear person agreement as well as verbs, see section 2.5.1 below.

4

Introduction

Within the Chomskian tradition, the (often tacit) state of the art has been simply to stipulate which feature slots are present but unvalued on a particular lexical item, thereby specifying explicitly its agreement potential (Chomsky 2000, Chomsky 2001). Fleshing out this background assumption, the Swahili pattern in (1) could be described by saying that verbs in Swahili have one or two empty slots for person, number, and gender features, depending on their transitivity. Values for these slots are then filled in by a process of agreement. In contrast, adjectives in Swahili would have only a single set of feature slots, and they happen to lack a slot for a person feature. Nouns in Swahili would have no empty feature slots: either they have no slots at all, or (better) they have slots for gender and number, but those features are already valued and hence are not open for agreement. On this baseline view, the Swahili words used in (1) could have representations like the following: (6)

soma ‘read’ verb Person: Number: Gender: (Person: (Number: (Gender:

refu ‘tall’ adjective Number: Gender:

jana ‘child’ noun (Person: 3rd) (Number: sg) (Gender: 7)

) ) )

A variant of this view would put the feature slots, valued or unvalued, not on the lexical category itself, but rather on some functional category associated with it. For example, the subject agreement features might be attributed not to the verb itself, but to the Tense/Infl node that selects VP. Similarly, object features might be attributed to a v projection distinct from verb, the valued number feature might be on a Number head rather than on the noun itself, and so on. Although it may be descriptively accurate, this baseline generative view is not very satisfying theoretically. If the lexical categories varied randomly in their behavior with respect to agreement within and across languages, then this sort of theory is the best one could hope for. But they do not. The pattern shown in (1) and described in (6) is in fact the norm for languages that use the relevant features as part of their grammatical system. Given this, some deeper theoretical relationship should be found between saying what lexical category a word belongs to and saying what its behavior with respect to agreement is.


5

It might seem tempting to take the agreement potential of a word as being basic, and define its lexical category in terms of that. For instance, one could define an adjective in Swahili as a word that can agree in number and gender, but not person, whereas verbs are defined as words that can show person agreement and nouns as words that do not undergo agreement at all. (Ouhalla (2005b) makes a proposal along these lines, defining verbs as categories that have person features and nouns as categories that have a class (gender) feature.) To the extent that there are other morphological and syntactic differences between nouns, verbs, and adjectives in Swahili, one might attempt to derive them from this inflectional difference. For example, one might derive the fact that adjectives can function as attributive modifiers but nouns generally cannot from these definitions plus the stipulation that attributive modifiers must agree with the noun that they modify. This approach has some intuitive appeal, because the inflectional differences among the lexical categories are often very salient in richly inflected languages. Indeed, awareness of the noun/verb distinction in Western grammar comes historically from the observation of Ancient Greek and Latin grammarians that some words inflect for number, gender, and case whereas others inflect for tense and subject agreement (Robins 1989). This is often how the noun/verb/adjective distinctions are presented in modern descriptive grammars as well. If, however, one’s linguistic theory aspires to give a general account of the human language capacity, and if it seeks to have explanatory depth, it becomes clear that this temptation should be resisted. One reason is that some languages clearly distinguish nouns, verbs, and adjectives even though they have little or no agreement. Modern English is very nearly a language of this type: agreement has been lost entirely on adjectives, and only survives on third singular verbs in the present tense. Japanese and Edo are even better examples (see Baker 2003a for discussion, e.g. pp. 240–5 on Japanese). A definition of the lexical categories that is rooted in their differing agreement potentials seems artificial and unenlightening for these languages. A second reason is that, even in languages that have a substantial amount of agreement, it is difficult to see how some of the purely syntactic differences between nouns, verbs, and adjectives can plausibly be derived from the inflectional differences. For example, only adjective phrases can be the complements of dedicated degree heads like so, as, too, and how in English (Chris is too hungry versus *Chris too likes wine, *Chris is too (a) fool). It is hard to see any plausible direct connection between a fact like this and the agreement asymmetry in (1), such that the former could be explained in terms of the latter.

6 1.3

Introduction What a better theory could look like

A more promising approach is to work the other way around. Suppose that we had a syntactically oriented theory of the lexical categories that was designed to explain why nouns, verbs, and adjectives occur in systematically different structures. That theory could be combined with a structural theory of agreement that did not explicitly mention the lexical categories at all. If everything worked properly, the agreement differences among the lexical categories might follow from the independently established syntactic differences together with general principles of agreement that are blind to category labels. Such a theory could explain why the noun/verb/adjective distinction does not depend on a language having agreement, but shows up in systematic ways in languages that do have agreement. The goal of this book is to develop this sort of approach, and to learn what can be learned about the general theory of agreement by doing so. In fact, the needed subtheories already exist, for the most part. In Baker 2003a, I developed a theory of the noun/verb/adjective distinctions designed to account for their syntactic differences across languages. That work did not consider agreement at all, however. Meanwhile, Chomsky (2000, 2001) and other minimalist theorists have developed a sophisticated syntactic theory of agreement, looking primarily at the agreement properties of verbs. This book considers how these theories can be combined – with a few adjustments – to achieve a unified theory of agreement that applies to all three lexical categories and explains their differences in agreement behavior in terms of their more basic syntactic differences. 1.4

What is in this book

I pursue this goal as follows. Chapter 2 is concerned with documenting more fully the agreement asymmetries sketched in (1), and showing how the core cases of agreement on simple predicates and modifiers can be explained by the conjunction of my theory of lexical categories and the Chomskian theory of Agree. Two key additions to the theory of agreement are found to be necessary. The first is the idea that a head can search upward through the syntactic structure for something to agree with as well as downward. The second is the idea that agreement in first and second person features is subject to a more stringent locality condition than other sorts of agreement are (the Structural Condition on Person Agreement, or SCOPA). Chapter 3 extends the account to other, less canonical constructions, including agreement on unaccusative and raising adjectives, agreement with wh-phrases,

What is not in this book

7

agreement in oblique subject constructions, agreement in double object constructions, and long distance agreement. In addition to proving that the ideas introduced in chapter 2 are general and robust, this chapter proves that adjectives and verbs should fall under the same theory of agreement, rather than having distinct theories of agreement (for verbs) and concord (for adjectives). For example, I show that agreement on raising adjectives is sensitive to all the same factors as agreement on raising verbs, and that verbs lose the ability to agree in first and second person when they happen to appear in syntactic configurations that are similar to those that adjectives always appear in. Chapter 4 turns to a closer examination of the SCOPA, considering in a deeper way why agreement in first and second person should be subject to a tighter locality condition than other sorts of agreement. I claim that this is the projection onto agreement of a general condition that applies also to first and second person pronouns as opposed to third person pronouns: first and second person pronouns must be bound by the closest relevant operator, whereas third person pronouns need not be. Thus, the SCOPA is not an ad hoc stipulation, but rather something rooted in the fundamentals of what it is for any linguistic expression to be first or second person. Chapter 5 then turns from the universal aspects of the theory of agreement to those aspects that vary parametrically. I claim that there are two such aspects. First, in some languages agreement depends on there being a unidirectional ccommand relationship, whereas in other languages it does not. Second, in some languages agreement is dependent on the case features of the two expressions being co-valued whereas in other languages it is not. These parameters are motivated by a close comparison of agreement in Niger-Congo languages with agreement in Indo-European languages. Their validity is then tested against a sample of 108 languages taken from around the world. I conclude that agreement is a domain of grammar in which there are both substantive universal constraints and a degree of highly patterned and systematic crosslinguistic variation. 1.5


Almost as important as saying what this book is about is being explicit at the beginning about what this book is not about. This book is almost entirely about the syntax of agreement – not about the morphology of agreement or about the semantics of agreement. This is not to say that agreement is a purely syntactic phenomenon; it certainly is not. Nor is it always easy to tease apart the syntactic aspects of agreement from its morphological or semantic aspects. But I believe that there is a syntactic aspect to this topic, which can and should be

8

Introduction

distinguished from the purely morphological and the primarily semantic. Here is a brief sketch of why. Consider first the distinction between morphology and syntax. It is a fact about Spanish and other Western European languages that verbal agreement shows number and person but not gender. There is, then, no difference in the verbal agreements in (7a) and (7b), although there is in the adjectival agreements in (7c) and (7d). (7)

a. Nosotros com-emos las manzanas. we.m.pl eat-1pS the apples b. Nosotras com-emos las manzanas. we.f.pl eat-1pS the apples c. Nosotros estamos list-o-s. we.m.pl are.1pS ready-m-pl d. Nosotras estamos list-a-s. we.f.pl are.1pS ready-f-pl

If one looked only at a single language or language family, one might consider this fact about verb agreement to be on a par with the fact that adjectives inflect for number and gender but not person. But crosslinguistic comparison reveals a clear difference. The inability of verbs to manifest agreement in gender is a property of certain IE languages, but it is clearly not universal. Many languages do have verbs that agree with their subjects in gender as well as in number and person; examples include Swahili, Arabic, Yimas, and Mohawk. In contrast, the failure of adjectives to agree in person is much more general, arguably universal. Furthermore, as we shall see, the ability of verbs to agree in person is contingent on the kind of syntactic configuration that holds between the verb and the agreed-with argument (see chapter 3). In contrast, whether gender agreement is present on verbs does not (as far as I know) interact with syntactic configurations in any interesting way. Taken together, these facts show that whether or not person agreement is possible is often a syntactic fact, whereas the presence versus absence of gender agreement on verbs is not syntactic in the same way. I therefore take the latter fact to be purely morphological in nature: the gender features are present on the verb in Spanish, but the morphemes that happen to spell out a [1st person plural feminine] feature bundle are no different from the morphemes that spell out a [1st person plural masculine] feature bundle in this language. This is not meant to deny that the absence of gender in verb agreement in some languages is an interesting and grammatically significant feature of the


9

language. For example, there could be a general rule of impoverishment of the sort studied by Bonet (1991) and other Distributed Morphologists, which systematically removes the gender feature from the feature bundles created by agreement prior to the insertion of actual morphemes. Such rules of impoverishment may obey laws of their own and have a major impact on what agreement looks like in particular languages. But whatever principles govern these phenomena, they are very likely to be different from those that underlie the contrasts in (1).3 More generally, I assume that the apparatus of Distributed Morphology is present to interpret in nontrivial ways the feature bundles that are placed on syntactic heads by the principles of agreement that I discuss, but I do not consider that aspect of agreement here in any depth. Bobaljik (to appear) has constructed an argument that all of agreement takes place after the syntax, in the morphological part of the PF component (see also Marantz 1991, Halle and Marantz 1993). Whereas I follow a fairly standard Minimalist division of labor between morphology and syntax, Bobaljik claims that even the matter of which NP a given word agrees with is determined in the postsyntactic morphological component. I do not follow Bobaljik in this for a variety of reasons: (a) I am not convinced that agreement relationships never feed other syntactic and semantic processes, (b) I think they are sensitive to more details of syntactic structure than other PF processes are, and (c) I am too conservative.4 But the difference between the two conceptions is not as large or important as it might seem. Bobaljik acknowledges that agreement is sensitive to syntactic relations, including a notion of which of two NPs is the higher one (c-command) and some notion of local domains (clausal constituents, phrases). He simply claims that this information is still available at PF, inherited from the syntax. As a result, the difference in framework does not matter much in practice: people who are attracted to Bobaljik’s Distributed Morphology architecture can interpret my title The Syntax of Agreement and Concord as meaning “those 3 There is probably nothing special about gender such that it alone can undergo impoverishment within the verbal inflectional system. There are a number of New World languages (e.g., Diegeño, Zoque, Hixkaryana) in which verbs inflect for person but not for number, suggesting that the number feature is deleted from the PF representation in those languages. Tsez is a language in which verbs agree with subjects or objects in number and gender but not in person (Polinsky and Potsdam 2001); a possible analysis could be that the person feature is deleted by a rule of impoverishment in this language. I assume then that any feature can in principle be involved in postsyntactic adjustments of this kind. 4 It is also not clear to me how to generalize Bobaljik’s approach from the one type of agreement he considers in detail (single agreement on the finite verb), to agreement on the full range of functional heads I am concerned with here. It would undoubtedly be interesting to try to do this and then compare the results with my theory, but I leave at least the first step of this process to others.

10

Introduction

aspects of the syntactic representation that agreement and concord make use of,” rather than “what happens in the syntactic component with regard to agreement and concord.” Consider now the distinction between syntax and semantics. There are instances of “agreement” that are rather clearly semantic in nature. The most famous is the possibility of collective nouns to trigger plural agreement on verbs (but not demonstratives) in many British varieties of English. An example is: (8)

a. (*) This band are brilliant live. b. This band is brilliant live. c. *These band are brilliant live.

(Wechsler and Zlatić 2003:76–9)

This sort of agreement is semantic in that it depends on semantic properties of the subject – in particular, the ability of a singular noun to denote a group consisting of several members. British speakers are also reported to detect a semantic distinction between using the plural form are and the singular form is in a sentence like (8): is goes with a collective reading of the predicate and are with a distributive reading. Somehow, then, a semantically licensed form can replace a morphosyntactically licensed form under certain conditions in British English. Nothing in this book will tell you why.5 More directly relevant to the range of data I have been focusing on, there is a kind of semantic agreement that can be seen in (9c–d). (9)

a. b. c. d.

Those women are a committee. The committee on diversity is three women and two men. Those women are theoretical linguists. *Those women are a theoretical linguist.

(9a–b) are like the Swahili examples (1c) and (4), which I have been using to show that there is no agreement between a predicate nominal and its subject; one is singular and the other plural. Nevertheless, a kind of agreement is enforced in (9c–d), such that (9d) is unacceptable. How do we resolve the paradox? What seems to be going on here is that nouns like committee and woman individuate things in very different ways. What counts as a single committee also counts as several distinct women. In contrast, nouns like woman and linguist individuate things in a similar way. If all the women are linguists and all the linguists are women, then there could never be a different number of linguists and women. 5 The issue of semantic agreement can arise for gender features as well as for number. Thus in Swahili (but not in Chichewa) nouns that denote humans but belong to some gender class other than class 1 nevertheless trigger class 1 agreement on verbs and other heads (Ashton 1949). This case also has the earmarks of semantic agreement. (I thank an audience at Georgetown University, especially Michael Diercks, for discussion.)


11

Given this, I take the appearance of agreement in (9c–d) to be an epiphenomenon of there being two NPs that pick out the same set of people independently, but which happen to individuate the members of that set in the same way. The two noun phrases thus both turn out to be plural, each by its own independent calculation. In contrast, (9a–b) show that when the two NPs individuate the set in question differently, they can have different number values. I take the less-common sentences in (4) and (9a–b) to be telling the truth that there is no grammatical agreement between predicate nouns and their subjects the way that there is between predicate adjectives and their subjects. (In contrast to (9a–b), one never finds a predicate adjective that does not agree with the subject in languages like Spanish, where adjectives are inflected.) I concede that (9c–d) shows a kind of agreement, but claim that it is semantic agreement rather than syntactic agreement, and the notions needed to explicate it are different. It is certainly an incompleteness and perhaps a deficiency of this work that it does not provide a theory of the morphological and semantic aspects of agreement, complete with a battery of diagnostic tests for discerning which is which, and an overarching theory of how all three types of agreement are integrated together. But if this work attempted those achievements, it would have another serious deficiency: it would be way over its page limit. There is plenty to say for one book about the narrower topic of those aspects of agreement that are arguably syntactic, as will soon be evident.

2


In chapter 1, I presented the Swahili data in (1) to illustrate that verbs can show agreement in person, number, and gender, adjectives show agreement (concord) in number and gender but not person, and nouns do not undergo syntactic agreement at all. (1)

a. Ni-li-kuwa ni-ki-som-a. (Ashton 1949) 1sS-past-be 1sS-cont-read-fv ‘I was reading.’ b. Ni-Ø m-refu. 1sS-be cl1-tall ‘I am tall.’ c. Ni-li-po-kuwa ki-jana . . . 1sS-past-when-be cl7-child ‘When I was a child . . .’

I also claimed that these differences in the agreement properties of the lexical categories are found in a wide variety of languages and deserve a general explanation in terms of more basic principles of syntactic theory. And I held out the hope that the effort to do so would help to refine and elaborate the theory of agreement, which has been a topic of great interest and attention in recent Chomskian work, especially since Chomsky 2000 posited Agree as one of the most basic syntactic processes. It is the burden of this chapter to develop and support these claims. I begin by showing as briefly as possible that the pattern in (1) is also found in other agreement-rich languages from around the world, providing empirical justification for the goal of deriving it from general principles (section 2.1). I then introduce and briefly motivate the necessary parts of my (2003a) theory of lexical categories, culminating in a precise understanding of the different syntactic structures that the lexical categories appear in (section 2.2). The next step is to present the Minimalist theory of agreement, together with a generalization of that theory that is needed to apply it to structures containing adjectives 12

The generality of the categorical asymmetries in agreement

13

(section 2.3). The crucial adjustment is that agreement-bearing heads must be able to search upward through the syntactic structure to find something to agree with as well as downward. I then apply the theory of agreement to the differing syntactic structures that contain instances of lexical categories, to show how the agreement asymmetries are accounted for (section 2.4). For the noun–adjective contrast, nothing new is required, but the adjective–verb contrast requires a novel principle that expresses what is special about first and second person agreement as opposed to agreement in other features. Finally, I discuss some apparent counterexamples to the claim that nouns and adjectives cannot show agreement in first or second person, and some cases where adjectives agree in some syntactic contexts but not others (section 2.5). Throughout this chapter, I concentrate on the most familiar syntactic categories (nouns, verbs, and adjectives) used in the most canonical ways (as predicates and as modifiers) and present the simplest versions of the principles needed to account for them. Subsequent chapters extend this core account empirically to other syntactic structures and other parts of speech, as well as deepening the principles theoretically and showing how they vary parametrically. For now, though, I focus on telling the basic story for the basic facts as directly as possible. 2.1


Suppose that agreement in Swahili were the way I have described it, but other languages had very different patterns of agreement. Suppose, for example, that it were just as common to find languages in which predicate adjectives agree in person with their subjects, predicate nouns must agree syntactically in number and gender, and verbs do not agree at all. Still other languages might have other imaginable categorical asymmetries in agreement, or they might treat all categories in the same way. If that were the case, then the right theory would be to stipulate on a language-by-language basis which words or word classes can agree with which features. In the limit, it could be that there is no general theory of agreement, and this is a topic for the lexicographer of a particular language rather than for the comparative syntactician. But that is not what a look at other languages reveals. On the contrary, the Swahili pattern proves to be robust and widespread, found in agreement-rich languages from different language families and different parts of the world. To show this, I put alongside the Swahili facts facts drawn from Mayali (Australian), Syrian Arabic (Afro-Asiatic), Tariana (Arawakan, from South America), and Spanish (Indo-European), showing that they are all essentially the same at this level of description. Furthermore, I believe the patterns to be

14


presented are consistent across the language families that these languages represent to the extent that the related languages show agreement at all. Since some of these language families are quite large, the pattern in question is indeed a widespread one.1 Note that I am not (yet) arguing that the pattern is universal – only that it is much more common than one would expect to arise by mere chance, and hence worthy of being explained. Once we have gained some sophistication in treating these matters theoretically, I return to the question of whether these categorical asymmetries in agreement are truly universal (section 2.5). How to present this data is something of a practical problem. Given that I need to cover quite a bit of ground in order to make my point, it would become tedious to most readers for me to present a sketch of each language, describing the essentials of its gender and number system, the morphological structure of its words, and other basic information needed to fully understand the examples presented. Furthermore, even if I did this, it would not eliminate readers’ need to either trust my presentation or check the facts for themselves. Therefore, I opt for a minimal presentation, noting that published and generally available grammars exist for all of these languages. It thus should be perfectly feasible for most skeptical or curious readers to check whether I got it right. 2.1.1 Adjectives versus nouns: predicative structures The first asymmetry to consider is the fact that predicate adjectives must agree with their subjects in number and gender, but predicate nouns need not. Predicate nouns often seem to agree with their subjects in number and gender, but that is a semantic effect that arises when their number or gender features are semantically interpretable and both nouns happen to carve up the world in the same way, as discussed in section 1.5. When two nouns have different inherent genders or individuate entities differently, then mismatches in gender and number are tolerated in predicate nominal constructions. This is shown in the English and Swahili examples in (2), repeated from chapter 1. (2)

a. Dalili y-a mvua ni ma-wingu. cl9(sg).sign cl9-assoc cl9.rain pred cl6(pl)-clouds ‘Clouds are a sign of rain.’ b. Those women are a committee

1 The generalizations also hold for Yimas, a New Guinean language with a three-way number distinction and more than ten genders; see Foley 1991 for data and description. I omit citing examples for reasons of space.


15

Such mismatches of number and gender are not tolerated in Swahili when the predicate is an adjective, as shown in (3) (also from chapter 1). (3)

Mi-zigo hii mi-zito (*ki-zito). cl4(pl)-loads these cl4(pl)-heavy (*cl7(sg)-heavy) ‘These loads are heavy.’

Example (4) illustrates the same contrast in Spanish, a language that distinguishes two grammatical genders and has a two-way number system. (4)

a. Est-as mujeres son gord-as. (*gord-o) These-f.pl women(f.pl) are fat-f.pl (*m.sg) (Predicate A, agreement) b. Est-as mujeres son un grup-o. These-f.pl women(f.pl) are a m.sg group-m.sg (Predicate N, no agreement)

Examples (5) and (6) illustrate the contrast in Syrian Arabic, which also has two grammatical genders and distinguishes singular from plural (Cowell 1964: 420–1). Predicate adjectives: (5)

(*ˇzc a¯ n-¯ın). (number agreement) a. L-walad zˇ u¯ c a¯ n the-child.m.sg hungry.m.sg hungry-pl ‘The child is hungry.’ b. L-bə nt zˇ u¯ c a¯ n-e (*ˇzu¯ c a¯ n). (gender agreement) the-girl.f.sg hungry-f.sg hungry(m.sg) ‘The girl is hungry.’

Predicate nouns: (6)

(no number agreement) a. Mac b¯ud-ak ə l-mas¯ari. idol.m.sg -2sP the-money.pl ‘Your idol is money.’ b. Hayy mod¯el ə zˇ d¯ıd. (no gender agreement) this.f.sg model.m.sg new.m.sg ‘This one (referring to a car) is a new model.’

The Australian language Mayali has four genders (masculine, feminine, vegetable, and neuter), although little or no number is represented on adjectives and nouns. (7) shows agreement on predicate adjectives for gender; (8) shows an absence of agreement on predicate nouns for number or gender.

16 (7)

Basic agreement and category distinctions a. Na-meke bininj na-kimuk. (Evans 2003:557) m-that man.m m-big ‘That man is big.’ b. Ngal-eke daluk ngal-kimuk. f-that woman.f f-big ‘That woman is big.’

(8)

a. Ngaleh ngal-yuhyungki ngurrurdu ngaleng bininj, Bulanjdjan. f.dem f-ancestor emu.f she human.m subsection ‘The female ancestor emu, she was a human of Bulanjdjan subsection.’ b. Ngad kun-kanj bedberre bininj. (Evans 2003:555–6) we.pl n-meat.sg 3p.obl Aborigine ‘We are meat for Aborigines.’

The South American language Tariana has an unusual gender system. Aikhenvald (2003) presents it as an open class system, with predicate adjectives taking any of more than one hundred distinct “classifier” suffixes in order to show agreement with the subject noun phrase along some semantic dimension. (9a) versus (9b) shows agreement with the subject in number in Tariana; (9a) versus (9c) shows how agreement with a female-human-denoting subject is different from agreement with a plate-denoting subject. (9)

(Aikhenvald 2003:499) a. Wyume-ma-se matSa-ma-pidana. last-cl:f-contr good-cl:f-rem. p.rep ‘The last one (a female) was beautiful.’ b. Keru-ma-pe na-pidana. (Aikhenvald 2003:252) angry-cl:f-pl 3pS.become-rem. p.rep ‘They (women) became angry.’ c. Nuha karapi hala-karapi-niki. (Aikhenvald 2003:119) I plate open-cl:plate-compl ‘My plate is completely full of holes.’

In contrast, (10) shows that predicate nouns in Tariana do not need to have a classifier suffix that agrees with their subject in gender or number. In (10a), the predicate nominal ‘rat’ does not bear the feminine and plural markers ma and pe in agreement with its subject, whereas the adjectives ‘good’ and ‘angry’ in (9a) and (9b) do. In (10b), the predicate nominal does not bear the animate plural classifier peni that an adjective like ‘good’ would in a similar context (see (19c)).

The generality of the categorical asymmetries in agreement (10)

17

a. Pedalie-ma-pe inari na-matSi-ka old-cl:f-pl mucura.rat 3pS-transform-th na-pidana. (Aikhenvald 2003:252) 3pS.go-rem. r.rep ‘The old women transformed into mucura rats.’ b. AtSa ihya alia-ka-naka (Aikhenvald 2003:491) men 2.pl exist-decl-pres.vis ‘You are real men.’

Notice that there are significant differences in the systems of grammatical features that are used in this range of languages, to say nothing of the obvious differences in how those ϕ-features are spelled out as affixes. Some languages distinguish three values of a number feature (Yimas; see note 1), some two values (Swahili), and some do not spell out a number feature at all (Mayali). Some languages have two distinct genders (Arabic), some three (Greek), some four (Mayali), some more than five (Swahili), some more than ten (Yimas), and some more than a hundred (Tariana). These important differences should be attributed to a rich and interesting theory of which features are present in a language, and how those features are spelled out in the morphology. (On the latter, see Halle and Marantz 1993 and subsequent work in Distributed Morphology; for steps toward the former, see Harley and Ritter 2002.) But at the same time, there is a consistent difference between nouns and adjectives when it comes to agreement that crosscuts these differences in feature systems and how they are manifested in morphemes. Inasmuch as the lexical category distinctions are primarily syntactic in nature (see Baker 2003a:ch. 5, for discussion), this consistent difference should probably be explained in terms of the syntax of agreement. 2.1.2 Adjectives versus nouns: modifying structures The same difference between nouns and adjectives can be seen in a different syntactic environment, when the noun or adjective is used as an attributive modifier, adjoined directly to a noun phrase. In this environment too, adjectives show gender and number agreement with the head noun, whereas nominal modifiers do not. (11) and (12) show this in Spanish; (13) and (14) show it in Swahili; (15) and (16) are from Arabic.2 2 Attributive adjectives are also said to agree with the noun that they modify in definiteness in Arabic. Predicate adjectives, however, do not agree with their subjects in definiteness. This suggests to me that it is wrong to fully subsume the feature ± definite into the other ϕ–features in this language. A plausible alternative is to say that the modificational structure is really one of

18 (11)

(12)

(13)

Basic agreement and category distinctions a. el hombre gord-o (Spanish, agreeing adjectives) the.m.sg man.m.sg fat-m.sg ‘the fat man’ b. las mujer-es gord-as the.f.pl women-f.pl fat-f.pl ‘the fat women’ a. la mujer-vampiro (Spanish, nonagreeing nouns)3 the.f.sg woman.f.sg-vampire.m.sg ‘the vampire woman’ b. el hombre-araña the.m.sg man.m.sg-spider.f.sg ‘the spider-man’ a. m-tu m-zuri (Swahili, agreeing adjectives) cl1-man cl1-good (Ashton 1949:11) ‘a fine man’ b. wa-toto wa-zuri cl2-child cl2-good ‘beautiful children’

(14)

(15)

c. mi-ti mi-zuri cl4-tree cl4-good ‘fine trees’ a. m-tu Ø-tajiri (Swahili, nonagreeing nouns) cl1-man cl9-wealth/merchant (Ashton 1949:47) ‘a rich man, merchant’ b. Ø-bata ma-ji cl9-fowl cl6-water ‘water fowl’ a. ?ə bno zˇ -ˇzu¯ c a¯ n (Arabic, agreeing adjectives) son.his det-hungry.m.sg (Cowell 1964:500–1) ‘His hungry son’ b. wə rt-e zg¯ır-e inheritance-f.sg small-f.sg ‘a small inheritance’ c. l-c arab ə s-s¯uriyy-¯ın det-arabs.pl det-syrian-pl ‘the Syrian Arabs’

3

apposition between two DPs: [DP D NP] [DP D A ØN ]. The second DP in apposition to the first naturally has the same definiteness value. I thank José Camacho (personal communication) for calling these examples to my attention.

The generality of the categorical asymmetries in agreement (16)

19

ə l-fə dd-a (Arabic, nonagreeing noun modifier) s-sakak-¯ın det-knife-pl det-silver-f (Cowell 1964:507) ‘the silver knives’

(17) shows attributive adjectives agreeing with the head noun in Mayali; (18) shows comparable noun-noun structures in which there is no agreement in gender. (17)

a. na-rangem na-mak ‘good boy’ (Evans 2003:182) m-boy m-good b. man-me man-mak ‘good food’ veg-food veg-good

(18)

c. kun-wardde kun-mak ‘good rock’ n-rock n-good a. an-bornde gun-yarl (Evans 2003:247–8) veg-banyan n-string ‘banyan (bark) string’ b. gunj an-djomborl kangaroo.m veg-pad ‘pad made by kangaroos’ c. gun-mogen ngan-gorle n-bundle veg-spears ‘bundle of bamboo spears’

Only in Tariana there is a complication. Tariana’s attributive adjectives do agree with the noun they modify through the medium of a “classifier” suffix, as shown in (19). (19)

a. heku-na hanu-na (Aikhenvald 2003:73) wood-cl:vertical big-cl:vertical ‘a big tree’ b. kule-kha matSa-kha (Aikhenvald 2003:88) fishing.tool-cl:curved good-cl:curved ‘a good fishing line’ c. tSãri matS-i:te; a˜ tSa matSa-peni (Aikhenvald 2003:174) man good-cl:animate men good-cl:animate.pl ‘a good man’ ‘good men’

The surprise is that when a noun modifies another noun in Tariana, it too seems to have a classifier suffix:

20 (20)

Basic agreement and category distinctions a. heku-na [tsuli i-tape]-na (Aikhenvald 2003:85) tree-cl:vertical diarrhea indf-medicine-cl:vertical ‘a tree, (which is) a medicine for diarrhea’ b. heku-na [pana-phe matSa-phe]-na tree-cl:vertical leaf-cl:leaf.like good-cl:leaf.like-cl:vertical ‘a tree which has beautiful leaves; a good-leafed tree’

This seems like a counterexample to my otherwise robust generalization. But (20b) illustrates two important features of this structure: first, the agreeing classifier is peripheral to the NP, not necessarily attached to the noun itself, and second, the agreeing classifier does not replace the normal classifier associated with the noun but rather is added to it. There is no such stacking of classifiers on the agreeing adjectives seen in (19). I interpret this as being consistent with my generalization, as follows. The fact that we see pana-phe and not pana-na (leaf-CL:vertical) in (20b) shows that nouns do not themselves agree with other nouns in Tariana. The second na in (20a–b) is a realization of agreement with the head noun on a higher functional head that appears in this sort of modificational structure. The agreement properties of this sort of functional head are interesting in themselves, but are not directly relevant to the generalization now under consideration. What is this mysterious additional functional category? I claim that it is an instance of the head that linguists like den Dikken and Singhapreecha (2004) have studied under the name “linker.” In many languages, a noun (phrase) usually cannot modify another noun (phrase) directly; some additional head is required to make the structure work. This is true in at least Spanish and Swahili among the languages under consideration. Some N-N structures like (12) and (14) are possible, but this is not the most common or productive pattern (they may be listed compounds, in fact). Rather, the particle de (Spanish) or a (Swahili) must usually come between the head noun and the nominal modifier: (21)

a. una pulsera *(de) oro a.f.sg bracelet.f.sg of gold.m.sg ‘a bracelet of gold, a gold bracelet’ b. ki-ti *(ch-a) m-ti cl7-chair cl7-assoc cl3-wood ‘a chair of wood, a wood chair’

In Baker (2003a) I discussed the badness of these examples without the extra head. I attributed this to the Noun Licensing Condition (a generalization of part of the θ -Criterion, see the discussion of (51) for a brief review); in intuitive terms, the modifying noun cannot receive a θ -role when it is adjoined directly


21

to another noun. Thus a preposition or similar particle is needed to θ -mark the nominal modifier. From this perspective, it is not at all surprising that an additional functional particle would be necessary in Tariana structures like (20); if anything, the surprise is that no such particle is present in the Arabic and Mayali examples cited. (I have no information about the productivity of these constructions in those languages; I cherish the hope that, like Spanish and Swahili, they are only possible in a restricted set of cases.)4 Consider now the agreement properties of the examples in (21). The modifying noun clearly does not agree with the modified noun: we do not see *ki-ti ch-a ki-ti for ‘chair of wood’ for example. But in Swahili the linking particle does agree in gender and number with the head noun – a fact to be explained when I generalize the theory of agreement to a wider range of categories (see section 4.1.4). I claim that the Tariana structure in (20b) is identical to (21b) except that Tariana is a head-final language, so the linking particle is postpositional to the NP (not prepositional), and the linking particle has no phonological content beyond that of the agreement. Overall, then, the generalization that adjectives undergo agreement and comparable nouns do not holds for attributive structures as well as for predicative structures. This is significant because it shows that the theoretical explanation for the difference should not be tied too closely to the syntax of predication. The correct analysis must be general enough to apply to attributive modification structures as well. (There are some languages in which adjectives seem to agree in predicative environments but not in attributive ones; I return to this in section 2.5.2.) 2.1.3 Adjectives versus verbs Next, I document the generality of the agreement asymmetry between verbs and adjectives – the fact that verbs can show agreement in first and second person as well as in number and gender but adjectives cannot.5 Many linguists already realize that this is true for many languages, and it is reflected to some degree in typological generalizations such as Stassen’s (1997) Agreement Universal, quoted in section 1.3. Here I illustrate the breadth of this observation 4 Another possibility is that linking particles are required for nominal modification in these languages too, but the particles happen to be phonologically null. It is not immediately obvious to me how to tell if that is true or not, but I leave the matter aside; it is an issue more for my 2003a book than for the current project. 5 It is also true, a fortiori, that nouns do not agree in first or second person any more than they agree in number and gender. I do not illustrate this separately, but see examples (1c), (8b), and (10b) above. See section 2.5.1 for discussion of apparent counterexamples to this generalization.

22


using the same range of languages as I used in the previous two subsections. In the interests of brevity, I only present verbs showing distinctive first and second person agreement with their subjects, but the reader should bear in mind that several of these languages also have object agreement, and this also has distinctive first and second person forms. (There is object agreement in at least Swahili, Mayali, and Yimas – and maybe also in Spanish and Arabic depending on whether object “clitics” in these languages are really instances of agreement (see Shlonsky 1997 for Arabic and Ormazabal and Romero 2006 for Spanish).) Examples (1a–b) already showed that this is true for Swahili. (22) shows a similar contrast with a first person plural subject; notice that the adjective agrees with the subject in number and gender (animacy), but not in person. (22)

a. Tu-li-(vi)-pot-ez-a vi-tabu vy-ote. (Vitale 1981:17) 1pS-past-8O-lose-caus-fv cl8-book cl8-all ‘We lost all of the books.’ b. Tu-Ø wa-refu. 1pS-be cl2-tall ‘We are tall.’

This difference is also familiar from Indo-European languages. For example, verbs inflect for first and second person in Spanish, but adjectives do not: (23)

a. (Nosotras) com-emos las manzanas. we.f.pl eat-1pS the apples b. (Nosotras) somos gord-as. (*gord-amos) we.f.pl are.1pS fat-f.pl fat-1p

In Syrian Arabic, verbs are like adjectives in agreeing with their subjects in gender and number: (24)

bə rn¯ett-i? (Cowell 1964:420) a. Wə sl-et arrived-3sS.f hat.f.sg-my ‘Has my hat arrived?’ b. Wə sl-u l-ə kt¯ab-¯en taba-c i? arrived-3pS det-book-pl two-my ‘Have my two books arrived?’

But verbs also have endings that agree with subjects in first or second person: (25)

smə c -tu, smə c -na (Cowell 1964:193) smə c -t, smə c -ti, hear-1sS hear-2sS.f hear-2pS hear-1pS ‘I heard’ ‘you(F) heard’ ‘you(PL) heard’ ‘we heard’


23

Adjectives cannot take these endings, nor is there any equivalent in the adjectival paradigm. For example, the adjectival (participial) form ‘dined’ in (26) takes the suffix ¯ın, expressing the plurality of its first person plural subject, but there is no expression of the first person feature. This ¯ın suffix is the same as would be used with a third person plural subject, and a form like *mətc aˇssˇ-na with a distinctive first person affix is apparently impossible. (26)

Katter x¯erak, nə hna mə tc aˇssˇ-y¯ın ya b¯ek. (Cowell 1964:266) thank you we dined.ptpl-pl sir ‘Thank you, but we have already dined, sir.’

A similar difference between verbs and adjectives exists also in Tariana. Recall that adjectives in Tariana bear classifier suffixes that agree with nouns in various semantic features, as well as agreeing in number. In general, verbs do not take these classifier affixes; rather, most verbs take prefixes that agree with the subject argument.6 Included in this prefixal agreement paradigm are special forms for first and second person subjects: (27)

a. Nuha siruri nu-walita nu-a. (Aikhenvald 2003:236–7) I cumatá.leaves 1sS-offer 1sS-go ‘I am making an offering of cumatá leaves.’ b. Wa-na i-walita na:-pidana. 1.pl-obj 2pS-offer 3pS.say-rem.p.rep ‘You make an offering to us, they say.’

(28) shows that no comparable first or second person morphology appears on predicate adjectives, even when their subjects bear these features. (28)

a. Ha-ehkwapi-nuku ma:tS-ite hiku nhua. (Aikhenvald 2003:495) dem:an-world-top bad-cl:an appear I ‘I look bad in this world.’ b. Ma:tSi-pu-naka pi-rena phia. (Aikhenvald 2003:249) bad-aug-pres.vis 2sS-feel you ‘You are feeling bad (suffering).’

In (28a), the predicate adjective partially agrees with the subject ‘I’ in that it bears the animate singular classifier ite. But it clearly does not bear the first person prefix nu seen in (27a), or any equivalent. Aikhenvald (2003:76) states explicitly that “Underived adjectives never take cross-referencing prefixes.” 6 Some stative nonagentive verbs in Tariana do not bear agreement at all, neither verb-like nor adjective-like. I analyze this interesting phenomenon in section 3.2.2.2. Classifier suffixes do show up on verbs in relative clauses and cleft constructions. I take this to be an instance of adjective-like agreement on a complementizer element; compare section 4.1.4.

24


Attributive adjectives are also incapable of agreeing in first or second person. Not all languages allow attributive adjectives to modify first and second person pronouns, but some do, including Tariana and Kinande (Bantu, related to Swahili). When such a structure occurs, the adjective again agrees with the pronoun in number and animacy, but not in person: (29)

a. [Pi-na inasu-ite-nuku] nuhua-wya-ne ma:-kasu nu-ñha. 2s-obj lazy-cl:an-top.obj 1s-ext-foc.subj neg-let 1sS-eat ‘I am the only one who will not let [you the lazy one] eat.’ (Tariana, Aikhenvald 2003:188–9) b. Itwe ba-kuhi mo-tu-a-gend-ire. (Kinande) we cl2-short aff-1pS-t-go-ext ‘We short ones went.’ (*tu-kuhi 1sS-short)

This sort of fact is not so frequently reported in grammars, but I assume that it is general. This time Mayali is the language that presents an instructive complication. Verbs clearly agree with subjects (and objects) in person and number, as shown in (30). (30)

Ngayi nga-wurlebme, la ngudda wanjh ngune-bo-rro! I 1sS-swim.np conj you then 2dS-water-strike.imp ‘I’m going to swim, and you two strike the water.’ (Evans 2003:692)

The complication is that adjectives – particularly adjectives that are given temporary, stage level readings – can take the verbal prefixes (which do not show gender features) instead of the adjectival prefixes (which do). (31a) and (31b) are thus both possible, with a difference in meaning. (31)

a. Alekke daluk ngal-warre (Ø-dowi-men). f.dem woman f[Adj]-bad 3sS-die-imp ‘That woman is bad/ugly (and must die).’ b. Alekke daluk ga-warre. f.dem woman 3sS[Verb]-bad ‘That woman’s no good (right now, she is upset, sick, etc.).’

In similar semantic situations, adjectival roots can even bear first and second person agreement prefixes: (32)

a. Ngayih nga-mak. (Evans 2003:354) I 1sS-good ‘I am healthy.’


25

b. Yi-keb-mak. 2sS-face-good ‘You are good looking.’

Taken by themselves, the examples in (32) look like counterexamples to the generalization that adjectives cannot be inflected for person. But the alternation between (31a) and (31b) points to a different interpretation. Mayali allows adjectival roots to be transformed into verbs by a process of zero-derivation, the verbal version having a more temporary, stage-level meaning (see also Heath 1984 on related Nunggubuyu). Another sign that a process of zero-derivation exists in Mayali is the fact that “adjectives” can host noun incorporation (see (32b)), which is otherwise a property of verbs not adjectives in Mayali and other languages (see (42) and Baker 2003a:70–2). Furthermore, “adjectives” – and even nouns referring to life stages, but not others – can bear tense/aspect suffixes that are normally attached to verbs: (33)

Ngaye nga-wurdurd-ni galuk ngaban-na-ng. (Evans 2003:357) I 1sS-child-past bye.and.bye 1sS/3pO-see-pp ‘When I was a child, I would see them.’

The agreement in (32) thus falls into place. It is not at all surprising that “adjectives” can show first and second person agreement when they are really verbs created from adjectival roots by zero-derivation. The data in (32) are thus no more problematic for my generalization about agreement than a sentence like The sky usually clears at night is for the generalization that adjectives like clear in English are not inflected for tense or agreement. This point is important because it helps us to interpret the typological record. Typologists with a functionalist orientation, such as Stassen (1997), have argued for implicational universals concerning agreement, stateable as follows (see also Croft 1991:82): (34)

a. If class-membership predicates agree in person, then so do property predicates. b. If property predicates agree in person, then so do event predicates.

If one equates “class-membership predicates” with nouns, “property predicates” with adjectives, and “event predicates” with verbs, this research purports to find the following range of languages and no others: (35)

a. b. c. d.

languages without agreement (in person) on any predicates languages with person agreement on verbs but not adjectives or nouns languages with person agreement on verbs and adjectives but not nouns languages with person agreement on all three lexical categories

26


The languages described in (35b) are very common, and are the type that I have focused on in this section. Languages like (35a) are also common – East Asian languages and some West African languages are obvious cases in point – but they are of only negative interest for a theory of agreement. Languages that superficially fit the description in (35d) are not so common, but they do exist: Turkish, Lango, Nahuatl, Abaza, Guarani, and Jakaltek are some examples; I return to these in section 2.5.1 below. The crucial question now is whether typological studies like Croft’s and Stassen’s give us any reason to think that languages of the type described in (35c) exist. I claim that these studies do not in fact support the existence of such languages. Whereas it is fairly safe to equate “class-membership predicates” with the syntactic category of noun and “event predicates” with the syntactic category of verb, equating “property predicates” with the syntactic category of adjective is notoriously unreliable. Many languages have property-denoting words that are syntactically verbs or nouns rather than adjectives (Dixon 1982). Furthermore, it is common for verbs to be derived from adjectival roots, derivations that may or may not be morphologically marked (see Baker 2003a:159–69 for examples and discussion). Mayali is an excellent example. Given this, imagine a language that does not have adjectives as a distinct class (at the relevant level), and that has person agreement on verbs but not on nouns (as usual). Stassen and other functionalist typologists would interpret this as a language in which the antecedent clause of (34a) is false, but the antecedent clause of (34b) is true. Hence it supports their implicational universals. Such a language does not fit the description in (35c), however. It would not be a language in which adjectives agree with their subjects in person; rather it would be a language that has verbs where other languages have adjectives. One can in principle tell the difference between the two categories whenever there is independent evidence as to whether a property-denoting predicate is an adjective or a verb – for example, when there is evidence from the distribution of copulas, or from the attachment of tense morphology, or from the form of negation. (These are Stassen’s (1997) other tests for a “verbal encoding strategy.”) Suppose that we remove from the sample all those cases where “property denoting” words are verbs as opposed to adjectives by criteria that do not involve agreement. Is it still true of this residue that it contains languages in which “property denoting words” (adjectives) agree in person but nouns do not? As far as I can tell, there is no evidence that it is.7 So the statement that adjectives never show person 7 Stassen (1997) gives three distinct criteria for identifying a “predication strategy” as being nonverbal: a predicate is nonverbal, (i) if it does not have person agreement in a language

The category-theoretic infrastructure

27

agreement could very well be universal (with a qualification to be given in section 2.5.1). 2.2


Section 2.1 documented that the differences in how the major lexical categories behave with respect to agreement are too consistent across languages for us to be content with stipulating what feature slots are present on particular categories on a language-by-language basis. Rather, the pattern should be derived from a general theory of the lexical categories, in interaction with a general theory of agreement. In this section, I present the first of these two subtheories, giving a precise characterization of the syntactic structures over which the observed agreement relationships hold. In fact, I believe that my (2003a) category theory serves the need well, so I review the basics of that theory and some of its motivation here. The only change that I propose is the addition of a layer of functional structure above the lexical heads, where agreement morphemes can be housed. 2.2.1 Essentials of category theory It has been realized since the Middle Ages that the lexical categories have different syntactic distributions. For example, in English and many other languages, only verbs can be used alone as the main predicate of a matrix clause ((36)), only nouns can be used as arguments in subject and object positions ((37)), and only adjectives can be used as bare adnominal modifiers ((38)): (36)

a. Chris sings. b. *Chris talls. c. *Chris mans.

(37)

a. Chris respects men. b. *Chris respects tall. c. *Chris respects sing.

where event predicates do have person agreement, (ii) if it needs a copular/auxiliary particle in a language where event predicates do not need separate copular/auxiliary particles, and (iii) if it is negated using a different “negation strategy” from that used with event predicates. Stassen does not mention cases in which these criteria give conflicting results. In particular, he does not mention any cases in which a language has predicates that show person agreement but are nonverbal by the auxiliary criterion or the negation criterion. I interpret this as evidence in favor of the claim that true adjectives never show person agreement. See also Croft 1991:131, where the author observes that it is languages in which “adjectives” bear the tense-aspect morphology of verbs that allow adjectives to agree like verbs, for example in person.

28 (38)

Basic agreement and category distinctions a. The tall person arrived. b. *The man person arrived. c. *The sing person arrived.

Until recently, most generative work on the lexical categories has concentrated on capturing their similarities, through X-bar theory and the positing of parallel systems of functional categories. In Baker 2003a, however, I sought to explain these and other, more subtle morphosyntactic differences in a unified way. My theory is based on the definitions in (39), together with the basic principles in (40). (39)

a. Verbs are lexical categories that license a specifier. b. Nouns are lexical categories that have a referential index. c. Adjectives are lexical categories that have neither a specifier nor a referential index.

(40)

a. All the θ -roles of a head must be coindexed with a maximal projection immediately dominated by a projection of that head. b. The Noun Licensing Condition (NLC): A referential index must be coindexed with a dependent element that it c-commands (a θ -role, a bound pronoun, or a trace of movement). c. The Reference-Predication Constraint: No syntactic node can have both a referential index and a specifier (cf. the logical incompatibility of predication and reference, Geach 1962).

Rather than merely saying that verbs are +V – a meaningless feature that no principle of grammar refers to – I say that verbs are the only lexical category that has a specifier as well as a complement. This is seen most clearly in paradigms like (41), from the Nigerian language Edo. Even in a small clause environment, where Tense is not a factor, verbs can take subjects directly, but adjectives and nouns can only be predicated of a subject if they are the complements of a copular particle: (41)

´ ı yá a. Uy` [èmátòn pèrhé]. Uyi made metal be.flatV ‘Uyi made the metal be flat.’ ´ ı yá b. Uy` [èmátòn *(yé) pèrhè]. Uyi made metal pred flatA ‘Uyi made the metal flat.’ ` o yá ´ ı *(rè) o` khaèmwèn]. c. Oz´ [Uy` Ozo made Uyi pred chief ‘Ozo made Uyi a chief.’


29

This basic difference between verbs and other categories is harder to see in Indo-European languages and others in which the copular particle Pred is phonologically null. The three lexical categories look structurally parallel in (42), for example. (42)

a. Chris made [the metal shine]. b. Chris made [the metal smooth]. c. Chris made [them leaders].

Nevertheless, one can use indirect means to tell that there is an extra structural projection in sentences like (42b) and (42c) that is not needed in (42a). For example, a noun that expresses the sole argument of an intransitive verb can incorporate into that verb in polysynthetic languages like Wichita and Mohawk ((43a)), but the sole argument of a predicate adjective or a predicate noun cannot incorporate into the adjective or noun ((43b–c)). (43)

a. Hánnhirh ta:c-ehe:k-ʔ irhawi. (Wichita: NI into V) ground.loc t/agr-cloth-be.lying (Rood 1976:5) ‘The cloth is lying on the ground.’ b. *né:rhir?as-tac ti-ʔ i. (OK: tac ti-ʔ i né:rhirʔ a) buffalo-fat t/agr-be fat t/agr-be buffalo ‘The buffalo is fat.’ (Wichita: no NI into A; Rood 1976:13) (Also OK: tac ti-ré:rhir?as-?i ‘The buffalo is fat.’) c. *Ka-nerohkw-a-nuhs-a’ (OK: Ka-nuhs-a’ ne o-nerohkwa-kvha.) Ns-box-Ø-house-nsf Ns-house-nsf ne Ns-box-former ‘That box is a house.’ (Mohawk: no NI into N) (e.g., a child’s playhouse, or a homeless person’s shelter)

This contrast follows from the assumption that only the subject of the verb is generated inside the maximal projection of the verb, together with the fact that head movement can never lower a word to attach it to a word that did not c-command it in the first place. The same difference between verbs and the other lexical categories can often be detected using unaccusativity diagnostics – phenomena in which the sole argument of an unaccusative (nonagentive) verb behaves like a direct object. A well-known case is (44a) from Italian, in which the sole argument of a verb like ‘sink’ (but not one of an agentive verb like ‘telephone’) can be expressed as a partitive clitic ne attached to the finite verb. In the Government-Binding era, this was taken as evidence that the theme argument of ‘sink’ was generated inside the VP, where its trace is properly governed by the verb (Burzio 1986).

30 (44)

Basic agreement and category distinctions a. Ne sono affondate due. V: OK of.them are sunk two (Burzio 1986, Cinque 1990) ‘Two of them sank.’ b. *Ne sono buoni pochi (dei suoi articoli). A: bad of.them are good few (of his articles) (Cinque 1990:7) ‘Few of them are good.’ c. ?*Ne sono professori molti. N: bad of.them are professors many (Baker 2003a) ‘Many of them are professors.’

The point of interest here is that the sole argument of thematically similar adjectives and nouns cannot be expressed as a ne clitic attached to the finite verb, as shown in (44b–c).8 I interpret this as showing that the subject of the nonverbal predication is not inside the AP or NP where it would be lexically governed and thus licensed by the adjective or noun. Rather, it is in the specifier of a nonlexical projection, the PredP, and moving from there incurs a violation (see Baker 2003a:62–9 for a full discussion, couched in more current theoretical terms). There is also indirect evidence of the structural distinction in the simple paradigm in (36) from English. The agreement-bearing Tense node can merge with the verb under adjacency, but it cannot merge with the predicate noun or adjective. I take the inability of the predicate noun or adjective to inflect for tense in English to be due to the presence of a null Pred head: this null head blocks Tense from merging with the lexical head at PF. If analyses like these are correct, then there is a systematic difference between the phrase structures that verbs appear in and the phrase structures that nouns and adjectives appear in. This structural difference can be used to explain the special agreement properties of verbs, given that agreement is defined over syntactic structure. The defining property of nouns that distinguishes them from verbs and adjectives is not in the kinds of phrase structures that can be built from them, but rather in the fact that noun projections are associated with a referential index. This can be seen most clearly in paradigms like (45), originally discussed by Kayne (1984). (45)

a. Italy{i} ’s invasion of Albania (grieved the expatriate community). b. The Italian invasion of Albania . . . c. Italy{i} ’s destruction of itself{i} . . . .

8 Note that there are a few adjectives in Italian that do permit ne-cliticization; see section 3.1.1 for discussion.


31

d. *The Italian destruction of itself{i} . . . . (compare The Italian self-destruction . . .)

Inside a derived nominal, the name of a country and an adjective derived from it can be nearly synonymous, either one serving to express the agent of the event named by the derived nominal, as shown in (45a) and (45b) (see also Grimshaw 1990, Giorgi and Longobardi 1991, and many others). But when the object of the derived nominal is a reflexive anaphor, a difference emerges: the genitive noun phrase is fine ((45c)), but the nationality adjective is strongly degraded ((45d)). My interpretation of this fact is that the noun Italy has a referential index, and hence can be the local syntactic binder that the anaphor requires. The adjective Italian, in contrast, has no referential index, and hence cannot be the local binder that itself requires. (45d) is thus ruled out as a violation of Chomsky’s Binding Principle A. This kind of account extends to the more elementary fact that nouns and their projections can serve as the subjects and direct objects of verbs and other θ marking heads, given Williams’s (1989) view that θ -roles are a kind of anaphor. Under this assumption, (46b) is ruled out for the same reason as (45d): the agent θ -role of the verb destroy is an anaphor that has no syntactic binder. (46)

a. Italy{i} destroyed Albania{k} b. *Italian destroyed Albania{k}

In the same way, (37b) and (37c) are ruled out because (unlike the noun men) the adjective tall and the verb sing do not bear a referential index that can bind the anaphoric internal θ -role of the verb respect. The primary significance of nouns being the only lexical categories that have referential indices for the theory of agreement comes from (47), a principle that is implicit in Chomskian work and explicit in HPSG work, including Pollard and Sag (1994) and Wechsler and Zlatić (2003:11). (47)

XP can have intrinsic ϕ-features (pre-specified values for person, number, and gender) only if XP has a referential index.

One may speculate as to why (47) should be true. I suspect that the ultimate answer has to do with an even more fundamental property of nouns than their bearing a referential index. In asking why only common nouns can be the restrictors of quantifiers (No dog barked vs. *No big barked and *No run barked), Geach (1962) and Gupta (1980) proposed that only nouns have criteria of identity. By this, they meant that only nouns have the kind of lexical semantics that supports judgments of whether two objects are the same according to

32


some standard. The examples in (48a) are thus meaningful, but the examples in (48b–c) are not. (48)

a. i. ii. iii. b. i. ii. c. i. ii.

That is the same man as you saw yesterday. That is the same water as was in the cup this morning. The Chinese want to have the same liberty as the Americans have. #That is the same long as this. #She is the same intelligent as he is. #I saw Julia the same sing as Mary did. #I watched Nicholas the same perform a stunt as Kate performed.

This lexical semantic property underlies the fact that only nouns have referential indices, because referential indices are used as a grammatical expression of (approximately) presupposed coreference, and coreference is ultimately a judgment of sameness: X and Y are coreferential if the entity that X refers to is judged to be the same entity that Y refers to according to a suitable standard. Common noun meanings give the standard that makes this judgment possible, so the linguistic expressions involved in the anaphoric relationship must be nouns (or pronouns – functional categories that also bear indices). It is reasonable to think that the same criterion of identity that underwrites the referential index also underwrites the presence of ϕ-features. This is clearest for number features like singular, dual, and plural, because they concern enumeration, just like the quantifiers that Geach and Gupta were interested in. In an observation that goes back to Frege, Geach and Gupta observe that counting – deciding whether there is one of a thing, or two of it, or more – presupposes being able to individuate those things. This in turn requires being able to decide which constitute the same thing. Since only common noun meanings permit this, only nouns can be intrinsically singular or dual or plural. This reasoning also applies to person features such as first person and second person. These features relate to the referential properties of a linguistic expression, so it makes sense that they too are dependent on a criterion of identity. For example, all of the first person pronouns in a sentence like (49) include the same designated person (the speaker) in their reference. (49)

I{i} told my{i} sister that our{i+k} mother would mail me{i} the document.

Therefore it makes sense that these person features are associated with the same class of linguistic entities as referential indices are. (I return in detail to the special qualities of first and second person elements in section 4.3.)


33

The connection between ϕ-features and the criterion of identity/referential index that defines the category of noun is least obvious for gender features. In part, this is because these do not have any stable semantic interpretation the way that number and person features do (see Harley and Ritter 2002). Nevertheless, it seems plausible to say that gender features have at least the function of placing the referent of the expression within the natural folk taxonomy that is imposed by the language. The gender 10 marking on the demonstrative pronoun izi in (50) from Chichewa (Bantu), for example, has a semantic interpretation to the extent that it constrains the demonstrative to refer to something that could be described by a class 10 noun in the language – to bees or zebras or elephants, say, but not to people, or lions, or knives. (50)

Awa zi-na-wá-lúm-a izi. (Mchombo 2004:52) cl2.these 10S-past-2O-bite-fv cl10.these ‘These (e.g. bees) bit these (e.g. hunters).’

It follows that two pronouns of the same gender can potentially refer to the same entity, whereas two pronouns of different genders generally cannot. Gender can thus act as a kind of surrogate criterion of identity, inherited from the set of criteria of identity that are associated with common nouns that have that gender. From this perspective, it seems natural that the linguistic entities that have criteria of identity are the same as those that have grammatical gender. At least, that is what I assume, pending further inquiry into the syntax and semantics of gender across languages. The third and final lexical category is the adjective.9 Adjectives already differ from verbs in not having a specifier, and from nouns in not having a referential index. I claim that that is all one needs to say about adjectives. The specific morphosyntactic properties of adjectives follow simply from this definition in interaction with general principles such as the ones in (40). Consider, for example, the fact that only adjectives can be used as bare attributive modifiers in a noun phrase, as shown again in (51). (51)

a. The tall person arrived. b. *The man person arrived. c. *The sing person arrived.

9 Unlike Jackendoff (1977), I consider P to be a functional category, for reasons discussed in Baker 2003a:303–11. Note that the Reference-Predication Constraint in (40c) insures that, from a theoretical point of view, there is no fourth lexical category that has the positive qualities of both a noun and a verb.

34


(51b) is ruled out by the Noun Licensing Condition in (40b) (essentially a generalization of one half of Chomsky’s (1981) θ -Criterion): there is no θ -role or other dependent element that man can be coindexed with in this structure. In essence, man as an attributive modifier fails to receive a θ -role. (51c) is ruled out by (40a), the other half of the traditional θ -Criterion: the verb sing has a θ -role to assign to a subject, but there is no nominal phrase within its maximal projection for it to assign that θ -role to. The adjective tall, in contrast, has no θ -role to assign and no index that needs to be licensed. Therefore, the structure in (51a) violates neither (40a) nor (40b), and is grammatical. This, then, is the kind of explanation I give for why adjectives appear in a different range of syntactic positions than other categories. Similar accounts can be given for why only adjectives are used as resultative secondary predicates (e.g., I pounded the metal flat/*shine/*a sword) and as the complements of dedicated degree heads (e.g., Chris is too smart/*hunger/*a genius). 2.2.2 Functional heads as the loci of agreement There is one additional assumption about structure, not made in Baker 2003a, that is useful in explaining why the different lexical categories have different behaviors with respect to agreement. Let us assume that it is not technically the lexical category that agrees with something in its environment, but rather a functional category that immediately dominates the lexical category, as stated in (52). (52)

Any lexical category can be immediately dominated by the projection of a functional head that matches it in gross categorical features. Functional heads, unlike lexical heads, can manifest agreement.

For verbs, this is a familiar assumption. There are strong and well-known reasons to say that “verbal agreement” is not an inherent property of V nodes per se, but rather of Tense/Infl, a functional category that selects the verb phrase and often fuses with the verb to form a single phonological word. For example, there is no agreement on verbs in English when the Tense is nonfinite rather than finite ((53a–b)). Furthermore, when Tense is kept separate from the main verb by an intervening negation, agreement shows up on the Tense position, not on the verb itself ((53c)). (53)

a. Chris likes swordfish. b. For Chris to like(*s) swordfish (would be unfortunate). c. Chris does not like(*s) swordfish


35

This is often generalized to object agreement as well: object agreement is on the functional head v (or Aspect, or AgrO), which selects VP and fuses with the verb. (52) is the assertion that this holds true for nouns and adjectives as well as for verbs. Empirical support for (52) comes from incorporation constructions, in which one lexical category undergoes head movement to adjoin to another lexical category (Baker 1988). Li (1990) shows that it is not possible for a lexical head to move through a functional head position on its way to another lexical head position. Standard effects of Li’s Generalization are the fact that only verbs that are not marked for tense incorporate into higher verbs in causative constructions and other complex verbal predications. Now, if the agreement that is often realized on a verb is really a property of Tense or some other functional head that dominates VP, Li’s Generalization predicts that verbs incorporated into other verbs should be stripped of their agreement affixes as well as of their tense/aspect markers. This is true; the minimal pair in (54) from Southern Tiwa provides an illustration. (54)

(Frantz 1993) a. Ow-t’am-ban hliawra-de u-napir-hi-’i. 2sS/aO/cO-help-past lady-sg 3sS/cO-sew-fut-sub ‘You helped the lady sew.’ b. Ow-napir-t’am-ban hliawra-de. 2sS/aO/cO-sew-help-past lady-sg ‘You helped the lady sew.’

The unincorporated verb ‘sew’ in (54a) bears its own agreement prefix, as well as tense and subordination suffixes. These tense and subordination suffixes are predictably absent when ‘sew’ is incorporated into the higher verb ‘help’ in (54b), in accordance with Li’s Generalization. The agreement prefix u seen in (54a) also disappears in (54b), suggesting that it too is attributable to a functional category that dominates VP. As a result, the complex verb is ow-napir-t’am-ban, not something like *ow-[u-napir-hi-(‘i)]-t’am-ban. Agreement affixes are also stripped of adjectives in the Hebrew adjective incorporation construction in (55b), discussed by Borer (1991). (55)

a. Sney ha-sir-im sˇxor-im. two the-pot-pl black-pl ‘The two pots are black.’ b. Sney sir-im hi-ˇsxir-u [AP t haki sˇe epˇsar]. two pot-pl V-black-m.pl most possible ‘Two pots blackened as much as possible.’

36


The predicate adjective in (55a) bears a masculine plural suffix im in agreement with the subject ‘two pots’, but this affix does not appear in (55b) when the adjective is incorporated into the inchoative verb (although the derived verb as a whole does agree with its subject in the usual way for Hebrew). This also follows from Li’s Generalization together with (52), the assumption that agreement is not a property of the adjective proper, but rather of a functional head that dominates AP. This functional head must be omitted when A-to-V incorporation takes place, or the head-movement would be blocked. When the functional head is omitted, so too is the number–gender agreement with the subject that is otherwise characteristic of adjectives in Hebrew.10 Although nouns do not agree the way that verbs and adjectives do, they often do have inflectional affixes that display the number and/or gender of the noun itself. The incorporation test shows that this inflection too should be attributed to a functional head distinct from the noun proper. For example, (56) from Mayali shows that the prefix gun, which expresses the neuter gender of the direct object ganj ‘meat’ in (56a), is necessarily absent when the noun incorporates into the verb as in (56b). (56)

a. Aban-yawoih-warrgah-marne-ginje-ng gun-ganj. (Evans 2003:1) 1sS/3pO-again-wrong-ben-cook-pp n-meat ‘I cooked the wrong meat for them again.’ b. Aban-yawoih-warrgah-marne-(*gun)-ganj-ginje-ng. 1sS/3pO-again-wrong-ben-(n)-meat-cook-pp ‘I cooked the wrong meat for them again.’

This range of evidence thus supports the view that any lexical category can be dominated by a functional category. Features of person, number, and gender are spelled out on the functional category, not on the lexical category proper. In informal statements throughout this book, I continue to use phrases like “lexical category X agrees with Y,” but these expressions are always to be understood as shorthand for the more accurate “the functional head that immediately dominates X and fuses with it to form a single word at PF agrees with Y.” Putting these ideas together, I posit the partial structures in (58) for basic verbal, adjectival, and nominal predications like those in (57).

10 It is true quite generally that inchoative and causative verbs derived from adjectival roots do not contain adjectival agreement morphology inside them; the verbal morphology replaces the adjectival inflection, rather than being built on top of it. What is not so clear is which of these inchoative and causative formations are derived by adjective incorporation in the syntax, and which (if any) are formed by some kind of lexical derivation.

The category-theoretic infrastructure (57)

37

a. Ni-li-anguka. (Swahili) 1sS-past-fall ‘I fell.’ b. Ni-Ø m-refu. 1sS-be cl1-tall ‘I am tall.’ c. Ni-li-po-kuwa ki-jana . . . 1sS-past-when-be cl7-child ‘When I was a child . . .

(58)

a.

b.

FVP VP

FV NP

NP I

V´ V

I

PredP

(PP)

Pred´ FAP

Pred Ø

FA

AP A

fall

tall PredP

c. NP I

Pred´ Pred Ø

FNP FN

NP{i} N{i} child

Not shown in (58b–c) is the projection of the copular/auxiliary verb and any verbal functional heads that dominate that. Any such structure is higher than PredP and largely irrelevant to the agreement dynamics currently under study. The verbal clause in (58a) may well have additional higher functional heads as well, of course. Notice that there are differences in these structures that the differences in agreement can plausibly be pinned to. First, since nouns are the only category that has intrinsic ϕ-features, the complement of FN in (58c) has ϕ-features of its own, whereas the complement of FA in (58b) does not. This could affect the ability of the functional head to agree with the subject in Spec, PredP (known to be possible in (58b) but not in (58c)). Second, since verbs are the only category that take subjects (specifiers) directly, FV c-commands the subject of predication

38


in (58a) whereas FA in (58b) does not. The next step, then, will be to articulate a general theory of agreement that is indeed sensitive to structural distinctions such as these. If FV in (58a) is another name for Tense/Infl, as I assumed above, what are FA and FN ? An initially plausible answer might be that FA is Degree and FN is Determiner, these being the most familiar functional categories that are associated with AP and NP, respectively. But facts from Spanish suggest that this is wrong. Spanish has independent words that are of category Degree and Determiner, both of which can be seen in (59). (59)

L-as chic-as estab-an tan enferm-as que no pod-´ıan habl-ar. the-f.pl girl-f.pl were-3pS so sick-f.pl that not could-3pS talk-inf ‘The girls were so sick that they couldn’t talk.’

These phonologically independent and semantically meaningful functional heads are distinct from the inflectional affixes as that express gender and number on the noun and the adjective. (There is an additional manifestation of feminine plural as on the determiner; I return to this in section 4.1.2.) This suggests that FN and FA are distinct from Determiner and Degree, and lower than them in the functional structure. FN might be Number, in the sense of Ritter (1991) and others. FA does not correspond to any other, independently known functional category that I am aware of. I continue to refer to these heads simply as FN and FA , leaving their exact nature open.11 The statement in (52) does not necessarily require that a functional head be generated above each lexical head; it merely admits of the possibility. I thus leave open the possibility that whether such heads are generated or not may vary from language to language, and even from lexical item to lexical item within a single language. For example, Swahili has both agreeing and nonagreeing adjectives. The latter are said to come from Arabic, but Ashton (1949) reports no syntactic difference between them. (60) compares the nonagreeing adjective safi ‘clean’ with the agreeing adjective zuri in predicative and attributive environments.

11 One might think that FN and FA are the same thing as the n and a heads that take bare roots as complements in Marantz 2000 and related work in Distributed Morphology. That is a conceivable identification, but unlike Marantz I assume that the complements of these heads are fully and intrinsically specified for syntactic category (see Baker 2003:265–75). I also permit there to be instances of AP and NP that are not dominated by FN P and FA P, as discussed in the next paragraph.

The category-theoretic infrastructure (60)

a. Sahani (zi-0) safi. plates cl10-be clean ‘The plates are clean.’ b. ch-umba safi cl7-room clean ‘a clean room’

compare:

compare:

39

Ch-akula hiki si ki-zuri. cl7-food this not cl7-nice ‘This food is not good.’

ch-akula ki-zuri cl7-food cl7-nice ‘good food’

Similarly in Mayali, there are nouns that bear an overt gender prefix, and there are other nouns from each gender class that do not. (61) displays examples of both kinds from Evans 2003:182. (Agreeing adjectives are included in these examples to make visible the gender that each noun has for purposes of agreement.) (61)

a. na-rangem m-boy

na-mak m-good

b. ngal-kohbanj ngal-mak f-old.woman f-good

vs.

bininj man

na-mak m-good

vs.

daluk ngal-mak woman f-good

One way to model this type of variation is to assume that each lexical category may or may not be immediately dominated by a functional category which is a potential target for agreement. Whether a given functional head appears or not can be regulated by the grammar of a language as a whole – plausibly there are none of these functional projections in an agreementless language like Chinese – or by individual lexical items within the language in question. Whenever the functional head is absent, there is no agreement either, by (52). This cannot be the whole story about why we observe variation in the presence of agreement, however. It is also necessary to stipulate that an existing functional head may or may not be designated as a probe (agreement-bearing head) that seeks for features in its environment to agree with. To see this, consider the fact that finite verbs agree with their subjects in Spanish but not in Japanese. One cannot simply say that there is no Tense node above the verb phrase in Japanese, because one can observe the presence of tense in Japanese both semantically and morphologically. Rather, we must say that the Tense head is present but not active for agreement (not a probe) in Japanese, whereas it is active in Spanish. Similarly, both English and Spanish have phonologically overt articles, but these are probes for agreement in Spanish but not in English. An important moral of (60) and (61) is that it is not possible to predict in a principled way whether or not there will be agreement in a given configuration in a given language. Rather, the emphasis for syntactic theory must be on what a given head agrees with and in what features it can agree, given that the head is present and can agree at all.

40 2.3

Basic agreement and category distinctions The agreement-theoretic contribution

Now that we have a handle on the different structural configurations that the lexical categories appear in, the next task is to introduce a general theory of agreement that combines with these structural differences to give the systematic differences observed in section 2.1. As my starting point, I adopt the theory of Agree developed by Chomsky (2000, 2001), who had facts about agreement in expletive and quirky subject constructions in English and other Western European languages primarily in mind. The essential components of this theory are summarized in (62), together with convenient names that I can use to refer to each subcondition. Also included is a page number from Minimalist inquiries (MI, Chomsky 2000) where the condition is discussed. (62)

A functional head F agrees with XP, XP a maximal projection, only if: a. F c-commands XP (the c-command condition, MI:122).12 b. There is no YP such that F c-commands YP, YP c-commands XP, and YP has ϕ-features (the intervention condition, MI:122). c. F and XP are contained in all the same phases (e.g., full CPs) (the phase condition, MI:108).13 d. XP is made active for agreement by having an unchecked case feature (the activity condition, MI:123).

I first review briefly the original motivation for each clause in this condition. I then propose a modification that allows it to be applied to constructions that contain adjectives. 2.3.1 Characteristic properties of agreement The requirement that the agreeing head be in a c-command relationship with the agreed-with nominal in (62a) is largely taken for granted, c-command being a requirement for many syntactic relationships, including binding and movement. Perhaps the most striking evidence in its favor is the contrast in Hindi shown in (63), as analyzed by Bhatt (2005:775) (see also Boeckx 2004:26). 12 An expression X c-commands another expression Y if and only if X does not dominate Y and every phrase that dominates X dominates Y (Chomsky 1986:8; see also Reinhart 1983 and much subsequent work). I further follow Chomksy 1986:7 in assuming that X dominates Y only if every segment of X dominates Y. In a structure like [X Y [X . . .]] where Y is adjoined to X, there are two segments of X, only one of which dominates Y, so the category X as a whole does not dominate Y in the relevant sense. 13 Bobaljik and Wurmbrand (2005) argue that even smaller verbal projections than CPs are opaque to agreement when they are the complements of lexical verbs. Their argument is based on the claim that there are two kinds of restructuring, one that allows agreement at a distance and one that does not. I do not go into the matter of agreement on restructuring predicates in this work.

The agreement-theoretic contribution (63)

41

chaah-ii. a. Vivek-ne kitaab parh-nii Vivek-erg book.f.sg read-inf. f.sg want-perf. f.sg ‘Vivek wanted to read the book.’ b. Mona kuttõ-ko dekh-naa/*nii chaah-tii thii. Mona dog.m.pl-acc see-inf/inf. f.sg want-hab.f.sg be-past. f.sg ‘Mona wanted to see the dogs.’

Bhatt shows that when the subjects of certain verbs that take infinitival complements are marked by a postposition (the ergative ne) and hence ineligible for agreement, the matrix verb can agree downward with the object of the lower verb, as in (63a). But the situation is not symmetrical: (63b) shows that when the argument of the lower verb is marked by a postposition (dative/accusative ko), it is nevertheless impossible for the lower verb to agree with the subject of the higher verb, even though the higher verb does. This looks like evidence that the agreeing head must c-command the agreed-with phrase at some point in the derivation. Nevertheless, it is this uncontroversial and little-examined assumption about agreement that I end up modifying most dramatically. Example (62b) expresses the notion that, although agreement can be a fairly long distance relationship, it is blocked if there is an intermediate phrase between the agreeing head and the agreed-with phrase that bears the same kinds of features as are involved in the agreement relationship. Perhaps the most striking example of this condition at work in constraining agreement is the following, from Icelandic (Schütze 1997:108–9, Boeckx 2000, Bobaljik to appear). (64)

virð -ast/?*virð -ist vera taldir l´ıka hestarnir. a. Jóni John.dat seem-3pS/seem-3sS to.be believed.pl to.like horses.nom.pl ‘John seems to be believed to like horses.’ b. Mér ?*virð -ast/virð -ist Jóni vera taldir l´ıka Me.dat seem-3pS/seem-3sS John.dat to.be believed.pl to.like hestarnir. horses.nom.pl ‘John seems to me to be believed to like horses.’

(64a) is an example of the matrix verb ‘seem’, when it has a dative subject, agreeing long distance with the plural nominative object ‘horses’ across three other verbal heads (‘to be’, ‘believed’, and ‘to like’). (64b) is identical, except that an extra dative argument has been added to the structure, the optional experiencer of ‘seem’. This experiencer functions as the subject of ‘seem’, causing the dative NP ‘John’ (thematically the subject of ‘like’) not to raise as high, so that it remains inside the c-command domain of ‘seem’. In this structure, it is impossible for ‘seem’ to agree with the nominative object ‘horses’, the

42


agreement apparently blocked by the presence of another NP between them. Such facts motivate the intervention condition in (62b).14 The intervention condition has also been applied to double object constructions by authors like Boeckx and Niinuma (2004) to explain why in many cases the verb can show object agreement with the NP bearing the goal/benefactive θ -role, but not with the NP bearing the theme θ -role. (65) is a typical example of this widespread pattern, taken from Chichewa (Mchombo 2004:80–3). (65)

a. A-lenje a-ku-ph´ık-´ıl-á anyan´ı z´ı-túmbûwa. 2-hunters 2S-pres-cook-appl-fv 2-baboons 8-pancakes ‘The hunters are cooking the baboons some pancakes.’ b. A-lenje a-ku-wá-ph´ık-´ıl-á z´ı-túmbûwa (anyáni). 2-hunters 2S-pres-2O-cook-appl-fv 8-pancakes 2-baboons ‘The hunters are cooking them (the baboons) some pancakes.’ c. *A-lenje a-ku-z´ı-ph´ık-´ıl-á anyáni (z´ı-túmbûwa). 2-hunters 2S-pres-8O-cook-appl-fv 2-baboons 8-pancakes ‘The hunters are cooking them (some pancakes) for the baboons.’

The asymmetry in (65) follows from (62b) given the assumptions that both objects of the verb are inside the c-command domain of the agreementbearing functional head (here v), and that the goal argument asymmetrically c-commands the theme argument. The second assumption is consistent with Barss-Lasnik (1986) c-command tests applied to the base structure in (65a), which generally show that the benefactive asymmetrically c-commands the theme (see also Marantz 1993). Thus, the structure of a sentence like (65) is

14 There has been some controversy in the recent literature about the so-called “defective intervention” effect in (64b), for both empirical and conceptual reasons. First, it is not always true that a dative NP prevents Tense from agreeing with a lower nominative in Icelandic; there is no intervention effect when Tense and both NPs are all in the same clause, for example (Bobaljik, to appear). Second, dative arguments seem never to prevent Tense from agreeing with a lower nominative in related languages, such as Dutch (Hans Broekhuis, personal communication). Therefore, it is not clear that this intervention effect is general enough to motivate a fundamental principle. Finally, from a conceptual viewpoint, it is not clear why a phrase that cannot itself be agreed with should count as an intervener that can block agreement with another NP. For purposes of this work, I need not worry much about these puzzles, because none of the cases that are crucial to my theory involve defective intervention. Rather, they all involve simple intervention, in which the phrase X that prevents a head F from agreeing with a more remote phrase Y is itself agreed with by F (see, for example, (73) and (76b) below). In this respect, they are like (65), which is a crosslinguistically robust effect. Nor is this sort of intervention effect conceptually problematic. So I take (62b) – with perhaps refinements about exactly what features YP needs to have to count as an intervener – to be a true principle of agreement even if one of the examples that initially motivated it should turn out to be dubious.

The agreement-theoretic contribution

43

approximately (66) (not shown is the verb ‘cook’ moving through the applicative head to land in v).15 (66)

[ v(=FV) [ApplP baboons Appl [VP cook pancakes ]]]

Agree OK

Agree blocked

The intervention condition plays an important role in what follows, explaining why the FN associated with a predicate noun cannot agree with the subject, whereas the FA associated with a predicate adjective can. I discuss conditions (62c) and (62d) together, because they have some similar effects and there may be some redundancy between them. They account for data like the following, from expletive constructions in English: (67)

a. There seem [to be three unicorns in the garden]. b. It seems/*there seem [that three unicorns are in the garden]. c. It seems/*there seem to two maidens [that three unicorns are in the garden].

(67a) shows that agreement on the Tense node associated with seem can see into the nonfinite, raising-type clause in order to agree with the plural NP three unicorns contained in that clause. In contrast, (67b) shows that this Tense cannot see into the corresponding finite clause, and (67c) shows that it cannot see into the PP in order to agree with the plural NP two maidens. The contrast between (67a) and (67b) follows from (62c), a special case of Chomsky’s (2000, 2001) Phase Impenetrability Condition, which holds that full CPs (and also some vPs) are “chunks” for interpretation. Once formed, these tensed CPs are shipped off to the interpretive components immediately, with the result that only a constituent at the edge of one of these phases remains available for further computation (in this case, for agreement). Nonfinite clauses of the raising type are not phases, however. The phase condition could also account for (67c), if one held that PPs also count as phases. Alternatively, (67c) could be ruled out by the activity condition in (62d): once the case feature of two maidens is checked by the preposition to, it becomes inactive with respect to agreement, and the Tense associated with seem cannot agree with it. The activity condition could also rule out (67b): the finite Tense of the embedded clause checks the nominative case of three unicorns, making it ineligible to enter into agreement with the matrix Tense. Given that both (67b) and (67c) can be ruled out by both (62c) and (62d) under plausible assumptions, I am not always sure which is crucially violated in a particular example. This need not concern us greatly, however, 15 See section 3.3.3 for more on agreement in double object constructions.

44


because for now I am primarily interested in the contrasts in (67) because they illustrate a characteristic pattern of agreement. In practice, where either condition will do I appeal primarily to the phase condition, because it is easier to state and may be more fundamental.16 Chomsky assumes one further condition on the Agree relation that I suppress here, which is that the agreeing head (the probe) and the agreed-with NP must match in features. The idea behind this is that the probe has certain predefined feature slots that need to receive a value from some other phrase in the structure that has specified values for those same features. It makes sense that (for example) a verb that needs a value for its person feature is not going to receive it by agreeing with a PP or a CP, since those categories do not bear person features. But one goal of this work is to eliminate, as much as possible, the idea that probe heads can be arbitrarily specified as having certain feature slots as opposed to others. Stipulating what unvalued features a given head has on a case-by-case basis does not capture the systematic differences in how verbs, adjectives, and nouns behave with respect to agreement. Instead, I am foregrounding the idea that all Fs are potential agreers and they agree with whatever features they can find in their environment according to structural principles. If agreeing functional heads are not prespecified as agreeing in particular features, it follows that there cannot be any matching condition of the kind Chomsky envisions. This condition should become superfluous – at least if agreement in ϕ-features is the only relevant kind of agreement in natural languages. 2.3.2 Generalizing agreement theory to adjectives All of the paradigm-forming examples of agreement in the previous subsection were instances of agreement on verbs. That is characteristic of the recent Chomskian literature on these matters, which has focused on this most prolific of agreeing categories. Consider then how this existing theory of agreement can be adapted to structures with other lexical categories. We run into a problem immediately when we consider the fact that predicate adjectives agree with their subjects in many languages. If my (2003a) approach to lexical categories is at all correct, then the partial structure of an adjectival predication is something like (68) (=(58b)).

16 In chapter 5, I show that the activity condition in (62d) is parameterized in a significant way, holding in some languages but not in others. Languages in which (62d) does not apply allow double agreement in structures like (67b) (see section 5.9), but they do not allow agreement into a PP as in (67c) (section 5.6). This suggests to me that both tensed CPs and PPs are phases, and that the phase condition is absolute and invariable across languages.

The agreement-theoretic contribution (68)

45

PredP NP

Pred´

girls Pred FAP [F,pl] AP FA Ø [F,pl] A tall

In some languages, this PredP structure can appear by itself as a matrix clause (Edo, Arabic, Hebrew); in others it must be selected by a nonthematic auxiliary verb (e.g., The girls are tall) or a small-clause-selecting verb (I consider the girls tall, Good nutrition made the girls tall). The crucial thing to notice is that the subject ‘girls’ is never in the c-command domain of the agreeing head FA . This subject could move higher still (in a raising construction, like The girls seem tall), but it cannot move lower, by the ban on downward movement. Thus, either the structural analysis of adjectival predication needs to be changed, or the conditions on agreement need to be changed. I propose to change the conditions on agreement.17 In particular, I suggest that the c-command condition in (62a) should be revised as follows: (69)

F agrees with XP, XP a maximal projection, only if: a. F c-commands XP or XP c-commands F (the c-command condition).

(69) permits agreement in (68), since the subject of the predication c-commands the functional head FA associated with the adjective.18 17 The more traditional view, which preserves the idea that heads always look downward for NPs to agree with, is to say that the subject of predication is generated inside the maximal projection of the adjective, as in (i) (see Chomsky 1993:8): (i)

[TP The girlsi are [VP [ FA [AP ti tall ]]]]

In this structure, FA can probe downward, agreeing with the trace/copy of the subject inside AP. The cost of this view is that it has no explanation for the fact that verbs behave like unaccusative predicates whereas thematically similar adjectives do not: see the discussion of (43) and (44) above, and Baker 2003a:ch. 2. More generally, the Baker 2003a system for explaining the differences among the lexical categories would unravel. Note that agreement with the subject could appear on Pred in (68) (see section 2.5.1). But that would not be the right analysis for the kind of agreement that appears on both predicate adjectives and attributive adjectives, since there is no Pred in structures of attributive modification. 18 A more radical and elegant possibility, pointed out to me by Itziar Laka (personal communication), would be to do away with the c-command condition on agreement altogether. It is not clear that any examples are uniquely ruled out by this condition, and Laka observes that

46


Does Bhatt’s evidence from long distance agreement in Hindi, repeated here as (70a–b), prove that (69) is a nonstarter? The answer is no, on two counts. First, Bhatt himself points out that infinitival morphology in Hindi never initiates an agreement relationship; it only shows agreement if it appears on the path between the matrix verb and the NP that the matrix verb ultimately agrees with. As a result, there is no agreement between the infinitival verb and its object in (70c), even though agreement here would be downward. (70)

chaah-ii. (=(63a)) a. Vivek-ne kitaab parh-nii Vivek-erg book.f.sg read-inf. f.sg want-perf. f.sg ‘Vivek wanted to read the book.’ b. Mona kuttõ-ko dekh-naa/*nii chaah-tii thii. Mona dog.m.pl-acc see-inf/inf. f.sg want-hab. f.sg be-past. f.sg ‘Mona wanted to see the dogs.’ c. [Imlii khaa-naa] achchh-aa hai. (Bhat 2005:771) tamarind.f.sg eat-inf. m.sg good-m.sg be.pres.3sS ‘To eat tamarind is good.’

The infinitive in (70b) also does not intervene structurally between the matrix verb and the NP that it agrees with (the matrix subject), so one would not expect it to agree any more than the infinitive in (70c) does. Furthermore, embedded verbs clearly do agree with matrix subjects in restructuring constructions in other languages, such as the Serbian example given in (71) (Sandra Stjepanovic, personal communication). (71)

Mi ga [VP zel-imo [VP da posjet-imo]]. we him want-1pS that visit-pres.1pS ‘We want to visit him.’

Of course there could be differences in the structure of complements of wanttype verbs in Hindi versus Serbian that one could relate the difference in agreement to. But at least (71) shows that there is no obvious up-versus-down asymmetry in the agreement properties of want-type constructions crosslinguistically, of the kind that would support a down-only theory of agreement prior to in-depth analysis of a wide range of relevant cases.19 sentences such as A group of children walk to school together every day, although considered substandard, are often produced in many languages, even though children does not c-command walk. But despite the thinness of the evidence that it is required, I continue to use the c-command condition, at least for expository purposes. 19 Some minimalists might declare the possibility of upward agreement out of bounds on theoretical grounds. Chomsky (2000) considers the c-command condition a “perfection” of the language faculty because it restricts the space in which a head can find something to agree with,

The agreement-theoretic contribution

47

Having generalized the c-command condition on agreement as in (69), it behooves me to make corresponding changes to the other conditions on agreement in (62), so that they properly regulate upward-probing agreement as well as downward-probing agreement. The intervention condition can be generalized as follows: (72)

F agrees with XP only if: b. There is no YP such that YP comes between XP and F and YP has ϕ-features (the intervention condition).

The notion “comes between” can be defined in terms of c-command: A comes between B and C if and only if either (i) B c-commands A and A c-commands C, or (ii) C c-commands A and A c-commands B.20 An application of this condition to upward agreement could be Italian examples like those in (73) (from Cinque 1990:28), which have a structure roughly like [NP1 verb [NP2 Pred [FA [Adjective]]]]. (73)

a. Ne ha resi infelici (*infelice) i of.it has.3sS rendered unhappy.m.pl (*unhappy.m.sg) the sostenitori. supporters.m.pl ‘He has rendered its supporters unhappy.’ b. Ne ritenevo opportuna la riapertura. of.it consider.1sS appropriate.f.sg the opening.f.sg ‘I consider its opening appropriate.’

reducing computational load. Also, some forms of cyclicity require that the uninterpretable features of a head be dealt with immediately upon introducing that head into the structure by Merge (e.g., Chomsky 2000:132). This would imply that agreement must happen before higher phrases have been included in the structure. I assume that the phase condition is adequate to narrow the probe’s search space and avoid computational explosion, and that cyclicity requires only that all features must be satisfied by the time that the phase is complete, not necessarily immediately upon introducing a feature-bearing category. 20 An alternative formulation could be stated in terms of the m-command relation of Chomsky 1986: X m-commands Y if and only if X does not contain Y and the smallest maximal projection that contains X also contains Y. Then A comes between F and NP if and only if either (i) F mcommands A and A asymmetrically m-commands NP, or (ii) NP asymmetrically m-commands A and A m-commands F. This alternative definition in terms of m-command implements a form of equidistance, in which the specifier and complement of a single projection count as equally close to a potentially agreeing head (because they m-command each other). In most cases the difference does not matter, but this alternative may have some advantages when it comes to agreement on Pred (section 2.5.1) and agreement on possessive determiners (section 4.1.2).

48


In such examples, the adjective clearly agrees with NP2 (the subject of the small clause), which is the closest c-commanding noun phrase, and not with NP1 (the subject of the matrix clause), which is farther away.21 Finally, the phase condition works as stated, given the understanding that agreement is blocked when the functional head is in a phase that does not contain the NP it might agree with, as well as when the NP is in a phase that does not contain the functional head (see (67b)). The phase condition thus also rules out upward agreement in configurations like the following: (74)

*Three women said [that there are likely [that it will rain]].

Putting the pieces together, the revised syntactic condition on agreement is given in (75). (75)

F agrees with XP, XP a maximal projection, only if: a. F c-commands XP or XP c-commands F (the c-command condition). b. There is no YP such that YP comes between XP and F and YP has ϕ-features (the intervention condition). c. F and XP are contained in all the same phases (the phase condition). d. XP is made active for agreement by having an unchecked case feature (the activity condition).

I adopt this as my first significant modification to the theory of agreement. 2.4

Explaining the basic categorical asymmetries in agreement

Now that we have a theory that can account for the agreement between a predicate adjective and its subject, we are ready to explain the crosscategorical differences in agreement: the fact that predicate adjectives differ from verbs in not showing first or second person agreement, and the fact that nouns differ from adjectives in not agreeing at all. I begin with the second of these contrasts, since it follows from the principles already given. 2.4.1 Explaining the noun–adjective contrast In fact, the agreement differences between nouns and adjectives follow immediately from the intervention condition in (75b), given the structures in (76) (=(58)). 21 In similar sentences in Icelandic, an adjective can never agree with an NP that is more remote than the subject of its predication, even if the subject of its predication has quirky case and hence does not trigger agreement on the adjective itself. Thus, in the Icelandic version of ‘The women(NOM) consider Mary(DAT) to be cold,’ ‘cold’ must be default masculine singular, not feminine plural in agreement with ‘the women’ (Sigurð sson, personal communication). This is an even closer analog to the “defective intervention” example in (64).

Explaining the basic categorical asymmetries in agreement (76)

a.

PredP NP

PredP

b.

Pred´

NP

they Pred FAP [F,pl] Ø FA AP [F,pl] A tall

49

Pred´

they Pred FNP [F,pl] FN NP[M,sg] Ø *[F,pl] N group[M,sg]

The one significant difference between these structures is that the complement of the potentially agreeing F head, namely ‘group’, has a referential index and therefore has ϕ-features of its own in (76b). Since this phrase c-commands FN and is c-commanded by the subject of predication ‘they’, it counts as an intervener, blocking an agreement relationship between FN and ‘they’. In contrast, the complement of FA in (76a) – the AP headed by ‘tall’ – does not have ϕ-features. It thus does not count as an intervener, and it does not block agreement between ‘they’ and FA . The asymmetry between A and N with respect to agreement thus reduces to the independently motivated intervention condition on agreement in (75b). Of course, nothing prevents FN in (76b) from agreeing with its NP complement, thereby manifesting the number and gender features that are intrinsically associated with that NP. This I claim is the source of the inflectional affixes that appear on nouns in many languages. These are often cognate with those that appear on adjectives, but are not used in the same way for agreement. In intuitive terms, predicate nouns do not agree with their subjects because they have to agree with themselves. There can be agreement in a predicate nominal construction, in fact, but it is internal to what is usually considered to be the noun itself. Now imagine what would happen if a second FN were generated immediately above the first FN in (76b), giving a structure like [‘they’ Pred [FN [FN [NP ‘group’]]]]. Could this higher FN agree with the subject? If so, one might expect there to be languages like Swahili in which predicate nominals bear two inflectional prefixes, an inner one that shows the inherent gender and number of the predicate nominal, and an outer one that agrees with the subject of predication. But we did not observe that in any of the languages surveyed in section 2.1.1. The structure itself should not be out of the question; after all, there can be two agreement-bearing functional heads above VP, one the locus of subject agreement and one the locus of object agreement (see (82b) below). If it is not possible to have two such heads above an NP, that is something

50


we want to explain, not stipulate. But recall from (52) that the functional head FN is, by definition, a functional head that shares the same essential categorical properties as its NP complement – an idea borrowed from Grimshaw’s (1991) notion of an extended projection. Now within my approach to lexical categories, the essential categorical property of an NP is having a referential index, and with that referential index come intrinsic ϕ-features. The lower FN P in the structure under consideration must thus be nominal in the same sense of having a referential index and ϕ-features; presumably it inherits the index and ϕ-features of its NP complement. Consider then the effect of this on the higher FN . Just like the lower FN , its complement is the closest phrase to it that has ϕ-features; therefore its FN P complement prevents the higher FN from agreeing with a more remote NP, such as the subject of predication. I conclude that even iterating functional heads do not make it possible for a predicate nominal to agree with its subject. Such iteration would only generate multiple redundant agreements with the predicate nominal, and this is presumably ruled out for economy reasons. There is an advance here over the baseline view, mentioned in section 1.2, which says that the ϕ-feature slots of a noun phrase are already valued and hence not available for agreement. This standard view does not explain why a noun could not have a second set of ϕ-feature slots which are unvalued and hence open for agreement. In contrast, the theory just sketched does rule out the analog of this possibility without additional machinery. Finally, recall that the agreement difference between nouns and adjectives is found in attributive contexts as well as predicative ones. The explanation given for (76) generalizes easily to this case. Following Baker 2003a, I assume that attributive modifiers are simply adjoined to the phrase that they modify, giving the structures in (77), enriched by the presence of an agreement-bearing F head that dominates the attributive modifier. (77)

zg¯ır-e a. wə rt-e inheritance-f.sg small-f.sg ‘small inheritance’

b. s-sakak-¯ın ə l-fə dd-a (Arabic) the-knife-pl the-silver-f.sg ‘the silver knives’

NP NP inheritance FA [F,sg] [F,sg]

NP FAP

NP AP A

small

knives FN [pl] *[pl]

FNP NP[F,sg] N silver[F,sg]


51

Here again, the presence of ϕ-features on the complement of FN makes the complement a closer goal for agreement than the NP that FN P is adjoined to. Therefore, if an attributive N bears inflectional morphology at all, it manifests its own features, not those of the noun that it modifies. In contrast, there are no ϕ-features on the complement of FA , so there is no intervener to prevent FA from agreeing with the NP that it adjoins to, which c-commands it.22 The reasoning is the same as in (76), except that no Pred head is present. 2.4.2 Explaining the adjective-verb contrast Next let us turn to the difference between adjectives and verbs when it comes to agreement in person. This requires a second addition to the basic theory of agreement, one that draws a distinction among the various ϕ-features. The empirical difference is that the functional heads associated with verbs can show agreement in first and second person as well as in number and gender, whereas FA is never first or second person. The most basic theoretical difference between verbs and adjectives is that verbs license a specifier and adjectives do not. This means that an FV typically c-commands the agreed-with subject inside VP, whereas an FA does not c-command the subject, as shown in (78). (78)

a.

b.

PredP NP

Pred´

we Pred FAP [1,Fpl] AP Ø FA [F,pl] [*1] A

FVP (=TP) FV VP [1,F,pl] NP V´ we V [1,F,pl] fall

(PP)

tall

22 Technicalities involving the c-command relationship in adjoined structures, introduced in Chomsky 1986 and mentioned in note 12, are relevant here. In particular, the NP headed by ‘inheritance’ in (77a) does not dominate the FA P that is adjoined to it, since only one segment of the NP contains FA P. Therefore, the NP does c-command FA , and so agreement between FA and the NP it is adjoined to is consistent with the c-command condition. (There is a tension, however, between this definition and my explanation of the lack of first and second person agreement in attributive constructions given in chapter 4. In the end, the significance of the segment-category distinction might be different for Agree than for binding; see section 4.5 for discussion.)

52


Given this, one might reasonably entertain a stipulation that downward agreement is required for agreement in first or second person, whereas upward agreement is possible for other features. But this would not turn out to be right, for reasons I discuss at length in the next chapter: there are structures in which FA also agrees downward, but it still cannot bear first or second person features. Instead, I develop a version of the rather traditional idea that there is something special about the specifier–head relationship when it comes to agreement. Inspired by Chomsky 1986, work around the early 1990s often assumed that agreement took place only if there was a specifier–head relationship between the agreed-with NP and the agreeing head (see Kayne 1989, Chomsky 1991, Kinyalolo 1991, Koopman and Sportiche 1991; see also Koopman 2006 for a recent defense of this idea). The pendulum has since swung the other way, with the specifier–head relationship playing no role at all in agreement in Chomsky 2000. I propose to take an intermediate position, claiming that a particularly rich form of agreement becomes possible if and only if there is a direct relationship of Merge between the controller of agreement and its target: (79)

The Structural Condition on Person Agreement (SCOPA) A functional category F can bear the features +1 or +2 if and only if a projection of F merges with an NP that has that feature, and F is taken as the label for the resulting phrase.

Notice that, in accordance with minimalist views about phrase structure, (79) does not actually grant any special status to the specifier configuration: F can agree with NP in +1 (first person) or +2 (second person) if NP is the specifier of FP, but it can also agree in this richer way if NP is the complement of F. In addition to being consistent with Chomsky’s (1995) Bare Phrase Structure, this formulation pays off when we consider agreement on adpositions and determiners in section 4.1. The SCOPA makes possible an account of the difference between adjectival agreement and verbal agreement as follows. The verbal head T, like its VP complement, is verbal in the sense of taking a specifier (another application of (52)). Thus T generally has an “EPP” feature that causes one NP inside its complement (typically the highest one, by the Minimal Link Condition) to move to become its specifier, as shown in (80). Once that movement takes place, T can agree with NP in all features, including +1 and +2:

Explaining the basic categorical asymmetries in agreement (80)

53

FVP (=TP) NP

FV´

VP we FV [1Fpl][1Fpl] NP V´ V

(PP)

fall

In contrast, FA cannot trigger movement of the subject of an adjectival predication to its specifier position. Probably it cannot have an EPP feature (by (52)), but even if it did, this would constitute an instance of downward movement; the NP would originate in a position that c-commands its destination, in violation of the fundamental laws of movement: (81)

PredP NP

Pred´

we Pred FAP [1Fpl] AP Ø FA [F,pl] [*1] A tall

FA can thus agree with the subject of predication, but only at a distance. Hence, it can only agree in features other than +1 and +2, by the SCOPA. This then gives us a way of thinking about the difference between adjectives and verbs when it comes to agreement. Transitive verbs in some languages can agree with an object as well as a subject, and such agreement can be for first and second person features as well as for number and gender. Let me now make explicit how this fits in. The first step is to say that there can be two functional projections above VP, the heads of which can potentially bear agreement. For concreteness, I follow much current Chomskian literature in assuming that these heads are finite Tense (as assumed above), and v – the light verb which θ -marks the subject and (in many languages) licenses accusative case on the object. Tense is the usual locus of subject agreement and v the usual locus of object agreement. The structure of a

54


transitive clause with double agreement on the verb, such as (82a) from Swahili, is thus roughly as shown in (82b). (82)

a. Juma a-li-ni-ambia kwamba . . . (Vitale 1981:62) Juma 1S-past-1sO-tell that ‘Juma told me that . . .’ b.

TP NP

T´

Juma T [3,sg, [3,sg, an] an] NP

vP v’

NP

v’

me v VP [1,sg] [1,sg] NP V’ V

CP

tell that . . .

In this implementation, v has the perhaps rather special property of being able to license two specifiers: the thematic subject, which is base-generated there, and an object which can arrive there by movement.23 Given that movement of an object to Spec, vP is possible, a projection of v merges with this object, so v can agree with the object in first or second person as well as in other features, in keeping with the SCOPA. Note that the subject agreement in (82a) 23 Other assumptions are possible. For example, one could distinguish two separate heads, one that creates the base position for the subject, and another immediately above or below it that provides the landing site for the accusative object. I leave the details open, adopting (82) for purposes of presentation, because of its relative simplicity and familiarity. I also note that the arrangement in (82) is not necessarily universal. It could be that in other languages the agreement-bearing heads are C, or Aspect, or Mood, instead of or in addition to T and v. In general, it is not the label of the category that determines what sort of agreement that category will manifest, but rather the geometry of the configuration that the category appears in. See also section 3.2.2.1 for discussion of how T and v are involved in languages with ergative agreement systems. (For evidence that T agrees with the subject and v agrees with the object in the normal way in a language with an active agreement system – a language in which some intransitive verbs bear “subject” agreement morphemes and others bear “object” agreement morphemes – see Baker 1996:ch. 5 on Mohawk. Other languages with active agreement (Choctaw, Guaran´ı) seem similar to Mohawk in this respect, but I have not studied the matter carefully.)


55

is left-adjacent to the tense morpheme li, whereas the object agreement is leftadjacent to the transitive verb root ambia ‘tell’. This morphological structure fits naturally with the syntactic structure in (82b), where the two agreements pertain to two distinct functional heads. The SCOPA raises many questions. For example, it is natural to ask what happens to agreement on FV if something other than the agreed-with NP satisfies the EPP feature of FV . The prediction is that in any such environment the verb should become more adjective-like in its agreement properties, being able to agree with an NP in gender and number but not in first or second person. Chapter 3 explores this prediction in detail, claiming that it is confirmed. However, the SCOPA definitely does not look like a plausible candidate for a basic principle of Universal Grammar. Rather, it is the kind of highly particular statement that one wants to derive from more general principles. Chapter 4 explores how this can be done in terms of the fundamental properties that distinguish first and second person pronouns from other nominal categories. Why does the SCOPA contain the additional qualification that F must be taken as the label of the phrase that results from merging it with NP? This is added to capture the observation made in section 2.1.3 that even attributive adjectival modifiers do not show first or second person agreement when they adjoin to a first or second person pronoun. One example of this is (83) from Tariana, repeated from (29a) above. (83)

Pi-na inasu-ite-nuku nuhua-wya-ne ma:-kasu nu-ñha. 2s-obj lazy-cl:an-top.obj 1s-ext-foc.subj neg-let 1sS-eat ‘I am the only one (who) will not let you the lazy one eat.’

In examples like these, one cannot complain that the FA is too far from the first or second person element, for the two do undergo Merge. The difference is that in these attributive constructions the first or second person pronoun counts as the head of the newly formed phrase, as shown in (84). This is unlike the structure in (80), in which the agreement-bearing FV head provides the label for the newly formed phrase. (84)

DP DP

FAP

you FA [2,sg,an] [sg,an] [*2]

AP A lazy

56


This minor-looking difference in structure apparently makes all the difference when it comes to agreement, first and second person morphology being licensed in (80) but not in (84). This shows that it matters which of the merged categories projects in cases of first and second person agreement. It will be an important virtue of the deeper theory that underlies the SCOPA that this odd qualification makes more sense; see section 4.5 for discussion. 2.5

Issues arising

Before closing this chapter, I consider two kinds of languages that seem to depart from the norm that I have described in this chapter when it comes to the agreement properties of lexical heads. The first is a set of languages in which all lexical heads – nouns and adjectives as well as verbs – seem to show person agreement with the subject of the clause. The second kind is a set of languages in which predicate adjectives agree with their subject in number and gender, but attributive adjectives do not agree with the noun they modify. Neither turns out to be problematic for my proposals once one looks into the details. 2.5.1 Person agreement on adjectives and nouns My theory is designed to explain why verbs often agree in first and second person features, whereas adjectives and nouns do not. As mentioned at the end of section 2.1.3, the typological literature recognizes this as being a strong tendency, but claims that it is not a universal feature of language. In particular, a number of languages have been reported as allowing the same person agreement affixes to attach to all three lexical categories. For example, in (85) from Turkish, the same first person singular suffix attaches to the predicate nominal ‘teacher’, the adjective ‘dirty’, and the verb ‘read’. (85)

a. Ben oku-r-um. (Kornfilt 1997:78–83) I read-aor-1sS ‘I read.’ b. Ben temiz-im. I dirty-1sS ‘I am dirty.’ c. Ben o¨ gretmen-im. I teacher-1sS ‘I am a teacher.’

Other language that fit this description include the Salish and Wakashan languages, Lango, Abaza, Nahautl, Guaran´ı, and the Mayan languages.

Issues arising

57

The account of these languages that I offer is simply that the person agreement in these languages is not on FA/N , the functional category that can be generated immediately above all uses of A and N. Rather it is agreement on some more verbal functional category higher in the structure of the clause, a functional category that is generated above the subject and can trigger movement of the subject to its specifier. For Turkish, this interpretation is supported by the fact that when examples like those in (85) appear in past tense rather than present tense, an overt past tense morpheme di appears attached to the predicate. It is significant that the person agreement marker appears outside of this tense marker, not inside of it, attached directly to the nominal or adjectival root: (86)

a. Ben kitab-i oku-du-m. (Kornfilt 1997:78–83) I book-acc read-past-1sS ‘I read.’ b. Ben temiz-di-m. I dirty-past-1sS ‘I was dirty.’ c. Ben o¨ g˘ retmen-di-m. I teacher-past-1sS ‘I was a teacher.’

This suggests that it is the Tense node that agrees with the subject in person in Turkish, not the noun or adjective (i.e., not FN or FA ). This unproblematic fact is merely obscured somewhat by the fact that Tense and the noun or adjective sometimes form a single word on the surface in Turkish as a result of head movement, PF merger, or cliticization. This makes it look like the noun or adjective agrees with the subject – especially in the present tense, when the tense marker is phonologically null. That person agreement in Turkish is really a property of Tense, not FN or FA , is even clearer in the future tense. This tense does not merge with a nonverbal root; rather an auxiliary root ol ‘be’ needs to be included to support it when the predicate of the clause is a noun or adjective, as shown in (87). In this case, when tense and the nonverbal predicate show up in different words, it is perfectly clear that the person agreement belongs to the word that contains the tense and not to the word that contains the nominal or adjectival root: (87)

a. gel-ecek-sin (Wetzer 1996) come-fut-2sS ‘you will come’

58

Basic agreement and category distinctions b. Ben temiz ol-aca˘g-im. (Kornfilt 1997:78–83) I dirty be-fut-1sS ‘I will be dirty.’ c. Ben o¨ g˘ retmen ol-acag-im. I teacher be-fut-1sS ‘I will be a teacher.’

There is no logical necessity that this be so. It is easily imaginable that the person agreement could have appeared on the noun or adjective instead of or in addition to appearing on the auxiliary. But it does not, supporting my claim that adjectives and nouns in themselves never show person agreement. The structure for examples like (86b) and (87b) is shown in (88), where Tense affixes to the adjacent adjective when it is present and auxiliary insertion takes place when Tense is future.24 TP

(88)

DP

T´

(I) [1,sg]

PredP DP AP

T

Pred´ FAP

Pred

Ø -PRES dI-PAST EcEk-FUT [1,sg]

(FA) [sg, *1]

rich

Essentially the same analysis can be applied to Lango, Wakashan, Salish, Abaza, Tzotzil, and similar languages. For example, Noonan (1992:144–6) shows that person agreement prefixes attach directly to nouns, adjectives, and verbs in the unmarked present tense in Lango, but crucially they do not attach to nouns and adjectives in the past tense; then they appear only on a finite auxiliary. These languages, then, do not count as real counterexamples to the claim that adjectives and nouns themselves do not agree with their subjects in person but only in number and gender. One language that might call for a slightly different treatment is Classical Nahuatl, as described by Launey (1981). First and second person agreement prefixes can attach directly to nouns in this language, as shown in (89). (Launey 24 Indeed, there is evidence that an auxiliary root i is inserted in the present and past tenses as well, prior to T cliticizing to the previous word. This form of the auxiliary, however, is elided after consonant-final adjectives or nouns, for phonological reasons; see Kornfilt 1997:77–83 for discussion.

Issues arising

59

does not clearly distinguish adjectives from nouns in Nahuatl, so presumably adjectival predicates like cualli ‘good’ behave similarly.) (89)

a. Ni-mex`ıca-tl. (Launey 1981:26) 1sS-mexican-nsf ‘I am a Mexican.’ b. Am-mex`ıca-’. 2pS-mexican-nsf.pl ‘You (pl.) are Mexicans.’

The presence of the determiner-like noun suffixes tl and ’ in these examples shows that they are not verbs derived from nominal roots, for which agreement in person would be expected; tl in particular shows up on most singular nouns but not on verbs in Nahuatl. Nor is there any sign of tense marking in these examples or others that involve predicate nominals. Tense suffixes never show up on nouns in Nahuatl the way they do in Turkish and Abaza, nor do overt auxiliaries appear in nonpresent tenses to bear the tense and agreement. This seems like a clearer case of person agreement attaching directly to a noun, then. Two interpretations of this data are possible within the theory I am developing, and I do not have clear facts to tell which is true. One is that the agreement is really housed syntactically on a null Tense node in Nahuatl, which then cliticizes onto the adjacent noun word, just as in Turkish. On this view, the only difference between Nahuatl and Turkish would be that Nahuatl does not have auxiliary verbs that can be used to support other types of Tense. The second possibility is that the examples in (89) are bare PredPs, not embedded under a TP at all, as seems to be possible in Edo, Hebrew, and other languages. In that case, I would claim that the first and second person prefixes are actually realizations of agreement on the Pred head itself, but not on FN . Since Pred does merge directly with the first or second person subject of the predication (see (88)), having Pred bear person agreement features is a theoretical possibility, consistent with the SCOPA.25 Which of these analyses is correct should be settled by doing a detailed study of the temporal properties of examples like (89) to determine if a Tense node is present semantically or not (see Benmamoun 2000 for such a study in Arabic). Even though we cannot inspect the location of agreement relative to Tense directly in Nahuatl, there is a bit of confirming evidence that the person agreement in question is on a higher, clause-like functional head and not on the lower, purely nominal head that I call FN . This involves the location of person 25 For predicate nominals, taking this option might require adopting the alternative formulation of intervention mentioned in note 20; on that definition, the ϕ-features on the FN P complement of Pred do not prevent Pred from agreeing with its specifier by the intervention condition.

60


agreement with the subject on possessed nouns that are used predicatively. In fact, the person agreement appears outside of the possessor agreement prefix in Nahuatl, as shown in (90). (90)

(Launey 1981:91) a. Ti-no-cihu¯a-uh 2sS-1sP-woman-poss ‘You are my wife.’ b. An-to-pil-hu¯an 2pS-1pP-child-poss ‘You (pl) are our children.’

Assuming that the morphological structure of the predicate typically reflects the syntactic structure and derivation of the clause (the Mirror Principle of Baker 1985), this morpheme order indicates that the functional head that agrees with the subject of predication is higher in the phrase structure than the functional head that agrees with the possessor of the noun phrase. Now the functional head that agrees with the possessor is presumably D (Determiner); see section 4.1.2 for examples, references, and discussion. FN is (by definition) the functional head immediately above the NP projection. FN is thus lower in the phrasal architecture than D – which is just what we need for Spanish noun phrases like l-as chic-as (the-F.PL girls-F.PL) (see the discussion of (59)). (90) thus suggests that FN does not agree in person with the subject even in Nahuatl. In contrast, Tense and Pred are both above the DP projection in a predicational sentence. Attributing the agreement in person to one of these heads, rather than to FN , thus gives a better account of the morphological structure of predicate nominals in Nahuatl, in addition to being consistent with the SCOPA. I conclude that, although there are clear counterexamples to a superficial statement such as “nouns and adjectives are never inflected for person,” there seem to be no true counterexamples to my actual theoretical claim, which is that the functional heads most intimately linked to adjectives and nouns cannot be the loci for agreement in first or second person. Higher heads in a clause containing a nominal or adjectival predicate clearly can, including Tense and perhaps Pred. But that is not surprising, since those categories are intrinsically verbal in the sense of licensing a specifier. 2.5.2 Agreement differences among adjectives In presenting my theory above, I emphasized that predicate adjectives and attributive adjectives typically have the same agreement behavior, even though they are found in somewhat different syntactic environments. On the one hand, both types of adjectives generally show number and gender agreement with a nearby noun, whereas nouns in similar syntactic positions do not. On the

Issues arising

61

other hand, neither type of adjective can bear first or second person agreement, whereas verbs can. I took it to be an important sign of the adequacy of my theory that it was able to give a unified analysis of agreement on both predicate adjectives and attributive adjectives. And indeed adjectives in both contexts do show the same agreement in all of the test languages that I surveyed. This is not always the case, however. There are also a reasonable number of languages in which predicative adjectives agree with their subjects but attributive adjectives do not. Kannada (Dravidian) provides one example: (91) shows that there is gender and number agreement on predicative adjectives; (92) shows that the same adjective is invariant when it appears in attributive position (Sridhar 1990:249–50). (91)

a. Avanu tuNTa(nu). he naughty.m.sg ‘He is naughty.’ b. AvaLu tuNTa-Lu she naughty-f.sg ‘She is naughty.’ c. Avaru tuNTa-ru they naughty-pl ‘They are naughty.’

(92)

a. a: tuNTa huDug-a (Sridhar 1990) that naughty boy-m ‘that naughty boy’ b. a: tuNTa huDug-i that naughty girl-f ‘that naughty girl c. a: tuNTa huDug-a-ru that naughty boy-m.pl ‘those naughty boys’

In fact, an analysis of this is readily available. I claim that adjectives in themselves never agree in Kannada; in more technical terms, FA is either absent or not a probe. The facts in (92) follow immediately. What then is the source of the agreement in (91)? I claim that these are not really predicate adjective constructions, but predicate nominal constructions in which the adjective modifies a pronominal. Thus, a more literal gloss for (91c) would be ‘They are naughty ones.’ The “agreement” in these examples is not a realization of FA , but rather a realization of some part of a nominal structure that contains the adjective. This analysis actually is fairly transparent, and similar to what Sridhar himself says. He says that predicative adjectives in Kannada “behave syntactically like

62


nouns” and “appear in their nominal form.” Note that the “agreement” on predicate adjectives is basically a pronominal form. The adjectives in (91) do not end in a, i, and aru, the normal masculine, feminine, and plural endings in Kannada (compare the nouns in (92)). Rather, they end in n(u), Lu, and aru – the same endings that the demonstrative pronouns take in (91). So these are not technically predicative adjectives, they are attributive adjectives that modify a pronominal element. The adjectival part of the predicate does not agree, but the pronoun part does agree with the subject, as pronouns normally do with their antecedents. That sort of agreement is quite a different matter from the Agree process under study here (see section 4.2 for a comparison). To complete the analysis, we need to say why predicative adjectives must be incorporated into a larger DP in Kannada but not in many other languages. Put another way, we must say why this language allows predicate nominals (which may contain an adjective) but not simple predicate adjectives. Baker 2003a:164–5, 210 provides an answer. There I showed that predicate nominals and predicate adjectives require slightly different copulas: the copula associated with a predicate nominal must θ -mark the predicate NP, whereas the copula associated with a predicate adjective must not θ -mark its complement. Most languages have both kinds of copula, but some do not: some have only the θ -marking copula, and hence have only predicate nominal constructions (my principal example was Vata). Kannada is another language of this less common sort: it has only a copula that selects NP, not one that selects AP. That selectional property is what prevents it from having a simple adjectival predicate, with no agreement. The Basque language illustrates a second reason why a language might have agreement on predicative adjectives but not on attributive adjectives. (93) shows in a single sentence that an attributive adjective cannot agree with a plural noun phrase, whereas agreement in number is strongly preferred on a predicate adjective (Maia Duguine, personal communication; cf. Saltarelli 1988:248). (93)

Katu lodi(*ak) hori-ek oso gaizto-ak dira. cat fat-abs.pl those-abs.pl very bad-abs.pl 3pA.be ‘Those fat cats are very mischievous.’

The contrast in (94) suggests that the analysis proposed for Kannada is not appropriate for Basque. (94a) shows that Basque is like English in that adjectives cannot take PP complements when used in attributive position. However, (94b) shows that a predicative adjective can have PP complement and still show number agreement with the subject of the clause.

Issues arising (94)

63

a. liburu (*nire klase-rako) interesgarri hori-ek book my class-for interesting those-abs.pl ‘those interesting (*to my class) books’ b. Liburu hori-ek nire klase-rako interesgarri-ak dira. book those-abs.pl my class-for interesting-abs.pl 3pA.be ‘Those books are interesting to my class.’

If the plural suffix ak in (94b) were the realization of some nominal head in a predicate nominal that contained ‘interesting’ as an attributive adjective, we would expect the phrase ‘for my class’ to be ruled out, just as it is in (94a). Basque thus seems to be a purer case of predicate adjectives agreeing and attributive adjectives not. The key to understanding these Basque facts is, I claim, observing that the head nouns of the subjects in (93) and (94b) are themselves not marked for number; there is no suffix like ak on these expressions either. This is a general property of Basque: number (and case) morphology is spelled out exactly once in the nominal, on the very last word of the phrase (the demonstrative in (93) and (94b); see Saltarelli 1988:75–82 for other possibilities). This suggests that the Basque noun does not in fact bear ϕ-features; rather intrinsic ϕ-features (at least number) appear only on D, the highest head in the Basque nominal structure. On this assumption, the structure for (93) would be (95). PredP

(95)

DP[pl] NP NP[-] cat

D[pl] FAP

AP

Pred´

those

FA[-]

FAP AP

Pred

FA[pl]

be

bad agree OK

fat

* agree, no c-command

*agree, no ϕ -features

The attributive adjective could in principle agree with the NP that it adjoins to, but agreement fails because that NP has no features to agree with. The entire subject DP does have ϕ-features, but the FA associated with the attributive adjective cannot agree with this DP, because DP dominates FA and therefore does not c-command FA . (Recall also that the goal of an agreement relationship must be a maximal projection, so FA cannot simply agree with D itself.) The predicate FA P, in contrast, is not contained in the plural DP. FA

64


is c-commanded by the plural DP, and can therefore agree with it in plurality. For Basque, then, the agreement difference between predicate adjectives and attributive adjectives follows from an independently observable property of where ϕ-features are generated in the complex nominal. This is a parametric difference between Basque and (say) Spanish, where ϕ-features are generated on the N itself and therefore can spread to other phrases inside the DP by agreement with NP. A perusal of the 108 languages surveyed in chapter 5 suggests that Basque is not alone in this respect; there are several other languages in which ϕ-features are marked only once in a DP, on a peripheral functional category rather than on the noun head. Possible languages of this type include Khoekhoe (Africa), Alamblak (New Guinea), and various languages of Mesoamerica and South America (Otomi, Jacaltec, Sanuma, Yagua, Canela-Krahô).26 2.6

Conclusion

The first part of this chapter documented evidence from a variety of languages that attests to consistent differences in how the different lexical categories participate in agreement: verbs can agree in person, number, and gender; adjectives agree in number and gender only; nouns do not agree with anything outside of themselves. The second part showed how these asymmetries can be explained by combining my (2003a) theory of lexical categories with Chomsky’s (2000, 2001) theory of agreement. Once we say that functional heads can probe upward as well as downward for something to agree with, the fact that adjectives agree with their subjects but nouns do not follows from the intervention condition on agreement. The fact that verbs agree in person but adjectives do not follows from the fact that only verbal projections license specifiers, together with the stipulation that a head can agree with an NP in first or second person features only if it has merged directly with that NP (the SCOPA). Two novel assumptions about agreement were thus needed to make the account work. It is the task of the next chapter to provide independent support for those assumptions, proving that the theory has a desirable degree of generality. 26 Dutch and German are said to have the opposite asymmetry, in which attributive adjectives agree but predicative adjectives do not. Such cases are very rare; I do not know of any others. Moreover, the “agreement” in question is an extremely impoverished kind: it does not vary with the number and gender of the noun. (In Dutch, the “agreement” is an invariant schwa; in German, it shows case but not ϕ-features.) This leads me to conjecture that these morphemes are not really agreements, but rather “linker” morphemes of the kind that come between A and N in Tagalog and other languages. I leave full consideration of this conjecture to linguists who know more about Germanic languages than I do.

3


In chapter 2, I developed the theory of agreement so as to explain a robust crosslinguistic pattern: verbs agree with their subjects and objects in person, number, and gender, adjectives agree in number and gender but not person, and nouns do not agree at all. My goal was not to stipulate these differences directly. Rather, I derived them from a general, structurally based theory of agreement based on Chomsky 2000, 2001, together with an independently motivated theory of the syntactic structures that nouns, verbs, and adjectives appear in. The same theory of agreement thus applies to all the lexical categories, with the differences falling out from differences in the configurations that the categories appear in. The central principles of this approach are repeated in (1) and (2), with the major innovations in italics. (1)

F agrees with XP, XP a maximal projection, only if: a. F c-commands XP or XP c-commands F (the c-command condition). b. There is no YP such that YP comes between XP and F and YP has ϕ-features (the intervention condition). c. F and XP are contained in all the same phases (the phase condition). d. XP is made active for agreement by having an unchecked case feature (the activity condition).

(2)

The Structural Condition on Person Agreement (SCOPA) A functional category F can bear the features +1 or +2 if and only if a projection of F merges with a phrase that has that feature, and F is taken as the label of the resulting phrase.

One can, however, still wonder whether having a unified theory of these matters is truly an accomplishment, bringing us closer to the truth of the matter. There is a long tradition of treating agreement on verbs and agreement on adjectives as two quite different phenomena. Indeed, the two are sometimes given different names: concord for the phenomenon of adjectives agreeing with the nouns they modify, as opposed to agreement proper for the relation verbs have with their subjects and objects. For example, Chomsky (2001:34n.5)

65

66


writes “There is presumably a similar but distinct agreement relation, concord, involving Merge alone.” And there are some good reasons for this traditional distinction. First, the features manifested on the agreeing head are different in the two cases: person features show up on verbs but not on adjectives. Second, the typical syntactic configurations that trigger agreement are different in the two cases: verbs search downward for something to agree with (Chomsky 2000, 2001), but adjectives search upward (see section 2.3.2). Third, adjectives and verbs agreeing with the same NP occasionally show different values for number and gender (see Wechsler and Zlatić 2003 on Croatian/Serbian). Given these apparent differences, one can question whether it is really a good idea to account for agreement and concord under the same formal theory. The fact that it can be done does not necessarily imply that it is right to do so. It could be that the erstwhile unification in (1) and (2) is a combination of formal tricks and spurious generalizations, rather than the uncovering of a fundamental unity in grammar. This chapter argues in favor of the unified theory of agreement by looking at some less-standard instances of agreement on adjectives and verbs. My goal is to show that in some of these cases agreement on adjectives is more verb-like, whereas in others agreement on verbs is more adjective-like. These examples are intermediate between the most canonical instances of adjective agreement and the most canonical instances of verb agreement, and they support the claim that there is no intrinsic difference in how these categories agree. I consider three kinds of noncanonical agreement configurations in particular. First, I look at instances of adjectives searching downward for something to agree with, rather than upward (section 3.1). This happens with a small class of ergative and raising adjectives, which do not have a thematic subject. In these instances it is clear that downward adjectival agreement is sensitive to the same factors as downward verbal agreement is. Second, I look for instances of verbs searching upward for something to agree with, in the manner of adjectives (section 3.2). Such configurations are predictably rare, but I find two possible examples: verbs that agree with a moved wh-phrase in Spec, CP, and agreement in ergative languages. Third, I discuss instances of verbs agreeing with something that is not in the specifier of their associated functional category at any level of analysis (section 3.3). In these contexts, the verb is not able to agree with the NP in person features; it shows agreement for number and gender only, in a manner reminiscent of adjectives. The last kind of example that could connect the two domains would be special environments in which an adjective can show person agreement with a

Downward agreement on adjectives

67

nominal, the way verbs often do. This situation, however, can never arise (apart from the kinds of examples discussed in section 2.5.1), for very simple reasons. Despite this gap in the pattern, the big picture is clearly that agreement on verbs and agreement on adjectives are not distinct phenomena. On the contrary, there is ample evidence that the two should fall under the same general theory of agreement. The fact that my unified theory generalizes from the simple structures considered in chapter 2 to the less obvious structures surveyed in this chapter provides strong support in its favor.

3.1


Perhaps my most startling theoretical innovation is the claim that a potentially agreeing head can either probe downward or upward to find an NP with ϕfeatures to agree with ((1a)). The idea that agreement can be upward is necessary to handle the agreement between a predicate adjective and its subject, since the subject is generated in Spec, PredP and is never inside the c-command domain of the adjective or its associated FA head. But (1a) does not specifically tie upward probing to adjectival heads, nor does it limit downward probing to verbal heads. Therefore, adjectival heads should also in principle be able to probe downward for something to agree with. When they do, the agreement on adjectives should be more obviously parallel to the agreement found on verbs. We will see that this is true.

3.1.1 Adjectives with NP complements One class of adjectives that is relevant to testing this theory is the so-called ergative adjectives discovered by Cinque (1990). Cinque argues that Italian has two distinct classes of adjectives, one that selects a subject argument and the other that selects an object. This distinction is parallel to the much more familiar distinction between unergative and unaccusative verbs, and it is revealed by some of the same syntactic tests. For example, the sole argument of an unergative (agentive) verb like telefonare ‘to telephone’ is like the subject of a transitive verb in that it cannot be expressed as the clitic ne, which gets attracted to the finite verb (compare (3c) with (3b)). In contrast, the sole argument of an unaccusative (nonagentive) verb like arrivare ‘arrive’ is like the object of a transitive verb in that it can undergo ne-cliticization (compare (3d) with (3a); see Burzio 1986 and Belletti and Rizzi 1981 for extensive discussion).

68 (3)

The unity of verbal and adjectival agreement a. Ne ha affondato due –. (object of transitive verb) of.them have sunk two (Cinque 1990:5) ‘He/she has sunk two of them.’ b. *Ne hanno avuto successo due –. (subject of transitive verb) of.them have had success two (Cinque 1990:5) ‘Two of them had success.’ c. *Ne telefonano molti –. (sole argument of agentive verb) of.them telephone.3pS many (Burzio 1986:20) ‘Many of them telephone.’ d. Ne arrivano molti –. (sole argument of nonagentive verb) of.them arrive.3pS many (Burzio 1986:20) ‘Many of them arrive.’

In a similar way, the sole argument of adjectives like ‘good’ and ‘dangerous’ cannot undergo ne-cliticization, whereas the sole argument of adjectives like ‘well-known’ and ‘obscure’ can (Cinque 1990:7): (4)

a. *Ne sono buoni pochi (dei suoi articoli). of.them are good.m.pl few of his articles.m.pl ‘Few of them (his articles) are good.’ b. *Ne sono pericolosi molti (di viaggi). of.them are dangerous.m.pl many of journeys.m.pl ‘Many of them (journeys) are dangerous.’

(5)

a. Ne sono note solo alcune (delle sue poesie). of.them are well.known.f.pl only some of his poems.f.pl ‘Only some of them (his poems) are well known.’ b. Ne e` oscuro più d’ uno (di motivo). of.them is obscure.m.sg more than one of reason.m.sg ‘More than one of them (reasons) is obscure.’

Cinque (1990) also gives five other syntactic tests that reveal the same distinction between ‘good’-class adjectives and ‘well-known’-class adjectives. For convenience, I’ll refer to adjectives like ‘good’ as normal adjectives and adjectives like ‘well-known’ as Cinque adjectives, after their discoverer.1 Updating Cinque’s structures in accordance with my (2003a) claim that adjectives never have specifiers, the contrasting structures are as shown in (6).

1 Cinque himself calls adjectives like ‘well-known’ “ergative adjectives,” parallel to Burzio’s (1986) “ergative verbs,” but I avoid Burzio’s terminology because it is too easily confused with ergative case and agreement marking.

Downward agreement on adjectives (6)

69

a. [TP – be+TENSE [PredP few+ne PRED [FA [AP good ]]]] b. [TP – be+TENSE [PredP – PRED [FA [AP well-known some+ne ]]]]

The clitic ne can move to adjoin to the tensed auxiliary in (6b) but not in (6a) because only in (6b) is its trace properly governed by a lexical head (the adjective ‘well-known’; see Baker 2003a:63–9 for specific semi-Minimalist formulations of the relevant principles). The question of interest for us is what are the implications of this structural difference for agreement on FA ? We can infer from Cinque’s failure to use agreement as evidence for his distinction that normal adjectives and Cinque adjectives do not differ in so salient a property. In fact, both classes of adjectives agree in number and gender with their sole argument. (7) shows that there is agreement with the sole argument of a Cinque adjective, even when necliticization proves that that argument has not moved higher, to Spec, TP (this can also be observed in (5)). (7)

a. Ne sono ormai probabili le dimissioni. of.them are already likely.m.pl the resignations.m.pl ‘Their resignations are already likely.’ b. Ne sono oscuri i motivi of.them are obscure.m.pl the reasons.m.pl ‘Their reasons are obscure.’

The adjectival agreement in the normal adjective structure in (6a) is the kind that motivated upward probing in chapter 2: FA does not c-command ‘few of them’ at any level of structure, but ‘few of them’ c-commands FA , and that is sufficient for agreement to take place. Agreement in the Cinque adjective construction in (6b) is an instance of the more familiar downward agreement: FA c-commands NP but not vice versa in this structure. (It has been known since Burzio 1986 that ne-cliticization is only possible with postverbal NPs in Italian; movement of the argument to subject position renders the sentence ungrammatical. Thus, the agreed-with argument in (6b) must remain in the c-command domain of FA at all syntactic levels for ne-cliticization to be possible.) Moreover, the morphological realization of this downward agreement is identical to that of upward adjectival agreement. This shows that both upward and downward agreement are possible for the FA associated with adjectives, and the two are equivalent from the viewpoint of morphology. 3.1.2 Raising adjectives Another relevant case to consider is raising adjectives such as likely. In English, these can appear in expletive constructions in which their local subject has no

70


ϕ-features. These are precisely the kinds of constructions in which verbal predicates like seem manifest longer distance, unambiguously downward agreement, as shown in (8a). (8)

a. There seem [– to be three unicorns in the garden]. b. There are likely [– to be three unicorns in the garden].

Unfortunately, we cannot observe agreement on the adjective likely in (8b) in English, for the obvious reason that adjectives never inflect for agreement in English. (Either FA is not generated in English, or it is generated but is always spelled out as Ø.) Fortunately, we can check similar structures in Icelandic, which is perhaps the only IE language that has both a raising adjective comparable to likely and robust inflection of adjectives.2 (9) is a relevant example. (9)

til að verð a valdir einhverjir komúnistar fiað eru v´ıst l´ıklegir there are thsy likely.m.pl to be elected some communists.m.pl ´ı stjórnina. to board.the ‘There are likely to be some communists elected to the board.’

Note that the adjective l´ıklegir ‘likely’ does agree with the masculine plural NP ‘some communists’ that is embedded inside its clausal complement. (I thank Halldór Sigurð sson and Kjartan Ottosson for judgments and much valuable discussion relevant to this section and the next.) This then is a second clear instance of FA probing downward for an NP to agree with, rather than upward. 3.1.3 Comparing adjectival and verbal agreement If downward adjectival agreement is the same phenomenon as verbal agreement, we expect it to be subject to the same restrictions as verbal agreement. That seems to be true. For example, a Cinque adjective in Italian can agree with its bare NP complement ((7)), but it cannot agree with an NP embedded in a PP, as shown in (10). (10)

a. Era oscura (*oscuri) a tutti la propria origine. was obscure.f.sg (obscure.m.pl) to all.m.pl the their.own origin.f.sg. ‘Their own origin was obscure to everybody.’ (Cinque 1990:13)

2 Raising adjectives seem to be quite rare crosslinguistically, for unknown reasons. For example, there are no such adjectives in the Romance languages or in Russian, even though those languages have raising verbs comparable to seem.


71

b. Era chiaro (*chiare) alle donne che avrebbe was.3sS clear.m.sg (*clear.f.pl) to.the women.f.pl that had piovuto. rained ‘It was clear to the women that it had rained.’

This shows that adjectival agreement is subject to the activity condition: there is no agreement with an NP whose case has already been checked by P.3 (10) is parallel to the absence of agreement between seem and the object of the preposition in (11b). (11)

a. There arrive three new women each day. b. *There seem to three new women that it will rain.

Similarly, the raising adjective ‘likely’ in Icelandic can agree with an NP inside a nonfinite clause, as in (9), but it cannot agree with a similar NP inside a finite clause: (12)

(*eru l´ıklegar) að þrjár konur séu ´ı herberginu. fiað er l´ıklegt there is likely.m.sg (*are likely.f.pl) that three woman are in room.the ‘It is likely that three women are in the room.’

The contrast between (9) and (12) is parallel to the contrast between (13a), where the raising verb seem agrees into a nonfinite clause, and (13b), where the raising verb seem fails to agree into a finite clause. (13)

a. There seem to be three women in the garden. b. *There seem that three women are in the garden.

This shows that downward adjectival agreement is subject to the phase condition in (1c), since the tensed clause is a distinct phase from the matrix clause but this type of infinitival clause is not. Finally, we can consider the intervention condition in (1b). The canonical evidence for this condition comes from examples like (14a) in Icelandic, where agreement on the matrix raising verb ‘seem’ with the embedded nominative NP ‘horses’ is blocked by the presence of the dative NP Jóni that appears between the two. (14b) is a similar example with a raising adjective; here too the presence of the intervening dative NP (the subject of the embedded predicate) prevents the adjective from agreeing with the nominative object of the lower predicate (Sigurð sson, personal communication).4 3 Alternatively, (10) could illustrate the phase condition on adjectival agreement, if PP counts as a phase, as suggested in section 2.3.1. 4 A complication is that Sigurð sson reports that (14b) “cannot even be constructed.” As far as any theory that I know would have it, examples like this ought to be possible with default masculine

72 (14)

The unity of verbal and adjectival agreement Jóni vera taldir l´ıka hestarnir. a. ?*Mér virð ast me.dat seemed.3pS John.dat to.be believed.pl like horses.nom.pl ‘I perceive John to be believed to like horses.’ (Schütze 1997:108) b. fiað eru l´ıklegar sumum mánnum til að leið ast there are likely.f.pl some men.dat toward to find.boring sumar konur. some women.f.pl ‘Some men are likely to be bored with some women.’

I conclude that agreement on predicate adjectives can indeed probe downward, the way that verbs do. When it does, it is transparently subject to exactly the same factors that verbal agreement is subject to. This can even be seen internal to each example I have given, because the tensed copular verb and the predicate adjective always agree with the same noun phrase, or, if no NP is accessible, they both show default agreement. I take this to be strong evidence in favor of a unified account of agreement on adjectives and verbs. If adjectives can take part in either upward or downward agreement, what decides which way it will go in a given sentence? Are there ever cases of optionality, in which the adjective can probe upward and agree with one NP or probe downward and agree with another in free variation? Or does one kind of agreement take priority over the other? In fact, it is not clear that this possibility ever arises, because of case theory. To investigate the question, we would need a configuration in which an adjective has both a thematic subject and an NP complement. Near-candidates would be adjectives like ‘certain,’ which take part in the alternation in (15), discussed by Cinque (1990:2–3). (15)

a. Gianni e` certo che verrò. Gianni is certain that I.will.come b. Gianni e` certo di questo. Gianni is certain of this c. Che verrò e` certo. that I.will.come is certain d. Questo e` certo. this is certain

When the thematic subject is not projected and the complement of the adjective is nominal, as in (15d), the adjective agrees with its complement, as expected. singular morphology on the copula and adjective. If they are not, some additional explanation is required.


73

When the subject is projected, as in (15b), the adjective apparently agrees only with its subject, never with its complement. But this does not prove that upward agreement takes precedence over downward agreement when both are possible. The reason is that the NP internal argument of the adjective in (15b) (unlike the parallel clause in (15a)) actually appears inside a PP headed by de, and we know that agreement never looks inside PPs. So agreement with the internal argument of ‘certain’ is blocked in (15b) for the same reason as it is in (10a–b). The question then is why de must appear in (15b) but not in (15a, c–d). The answer (as anticipated by Cinque) presumably comes from case theory: adjectives do not check/assign structural case. In (15d), the adjective’s internal argument can have its case licensing by Tense, but in (15b) this case-licensing relationship is established with the subject of predication. Therefore, the preposition must be inserted to license (or spell out) case on the internal argument. This preposition then prevents the adjective from agreeing with the internal argument. Therefore, there is no simple adjectival construction in which both upward and downward agreement are expected to be possible. Finally, I need to make something explicit in my theory in order to ensure that there is no person agreement even on Cinque adjectives in Italian examples like ‘We are well known’ or on raising adjectives in Icelandic examples like ‘You are likely to win.’ In these examples, the surface subject is generated inside the AP and moves to the matrix Spec, TP. The principles of movement would allow the NP to move through Spec, FA P on its way to Spec, TP, resulting in the following syntactic representations: (16)

a. [TP We[1,pl] be+T [PredP t[1,pl] Pred [ t[1,pl] FA [AP well-known t ]]]]

b. [TP You[2,sg] be+T [PredP t[2,sg] Pred [ t[2,sg] F A [AP likely [TP t to win]]]]

If these representations were possible, then FA could manifest first or second person features in these configurations, in keeping with the SCOPA. Cinquetype and raising-type adjectives could thus be exceptions to the general rule that adjectives cannot agree in person. But Cinque adjectives and raising adjectives are not exceptional in this respect; they support the same range of inflections as other adjectives in the languages in question. In fact, an answer to this problem was implicit in the convention stated in (52) of chapter 2, repeated here as (17).

74 (17)

The unity of verbal and adjectival agreement Any lexical category can be immediately dominated by the projection of a functional head that matches it in gross categorical features.

This is a version of the standard assumption that the functional heads that appear with VPs are somehow verbal, those that appear with NPs are nominal, and those that appear with APs are adjectival (compare Grimshaw’s (1991) notion of an extended projection). Within my (2003a) theory of categories, however, being adjectival as opposed to verbal has a precise syntactic meaning. Verbal phrases have the defining property of having a specifier, whereas adjectival phrases cannot have a specifier. (17) projects this property onto the functional heads above VP or AP as well. Therefore, the FA that dominates AP in structures like (16) is itself adjectival – meaning it must not have a specifier. Thus, there cannot be a trace of the moved NP in Spec, FA P, and FA cannot have +1 or +2 features (although it may agree at a distance with the moved NP in other features). In more familiar terminology, FA never has an EPP feature, whereas the functional heads associated with verbs do. With this clarification, we can maintain the generalization that true adjectives never agree with an NP in first and second person features, despite the fact that adjectives appear in a wider range of syntactic structures than were considered in chapter 2.

3.2

Upward agreement on verbs

Next I consider whether it is possible for verbs, like adjectives, to agree upward with something that they (and, more precisely, the functional heads associated with them) do not c-command. Recall from section 2.4.2 that verbs can have at least two distinct functional heads generated above them: Tense, the normal locus of subject agreement, and v, the normal locus of object agreement. Either of these functional heads could conceivably probe upward for something to agree with, so I discuss them separately, beginning with Tense. 3.2.1 Upward agreement on tense The prospects for finding agreement on Tense that is unambiguously upward are very limited. Under standard assumptions, there are no θ -positions where an NP can originate that are higher than Tense but still in the same clause. The θ -positions of the clause are all internal to the vP complement of the Tense head. It often happens that one of these NPs moves to Spec, TP, landing in a position that c-commands Tense. Such NPs often agree with Tense, resulting in canonical subject agreement. However, it is possible to say that the agreement relationship


75

was established prior to the NP movement, when Tense c-commanded the NP, as assumed in Chomsky 2000, 2001:5 (18)

[TP The three womeni do+T [not [vP all ti V [VP like eggplant]]]] AGREE

There could of course be NPs that are generated in a higher clause than the Tense node, which would c-command Tense at all levels. We could thus contemplate the possibility of upward agreement on the embedded verb in sentences like those in (19). (19)

a. *Three women said that there seem that it will rain. b. *I told three women that there seem that it will rain.

This sort of agreement is quite impossible. However, this does not prove that verbal functional heads cannot search upward for something to agree with. Agreement in examples like (19) is already ruled out by the phase condition in (1c), since the CP complement of say or tell counts as a phase which contains the Tense associated with seem but not the putative controller of the agreement.6 Virtually the only chance for T to manifest agreement that is unambiguously upward, then, would be in structures that involve wh-movement. Wh-movement might move (say) a plural NP from a lower phase (where T cannot agree with it) to place it in Spec, CP. Once it reaches Spec, CP, the +wh NP would asymmetrically c-command the agreeing T, and it would be contained in the same phase as that T. Therefore, if upward verbal agreement is allowed by Universal Grammar, it might show up in this configuration. Kayne (2000:190) discusses examples like (20a) (originally from Kimball and Aissen 1971), which are possible in certain dialects of English. (20)

a. the people who Clark think are in the garden b. the person who Clark think*(s) is in the garden

In (20a), the verb think appears in its plural form, despite the fact that its subject is third person singular. It apparently agrees instead with the plural relative 5 In chapter 5, I argue that Kinande has English-like subject–verb agreement, but never allows agreeing heads to probe downward. If this analysis is accepted, then that language has many instances of T (and v) probing unambiguously upward for something to agree with in a postmovement structure (see section 5.2). 6 Occasionally TPs are not contained in CPs that constitute phases: this is true of raising and ECM infinitives in Chomsky’s theory. But these TPs have Ts that do not manifest agreement, at least in familiar languages like English. Hence, upward agreement with a phrase in the matrix clause is not expected even here.

76


pronoun who in Spec, CP (which in turn agrees with the head of the relative the people). This relative pronoun clearly c-commands the Tense associated with the verb, and not vice versa. In the spirit of Kimball and Aissen’s original analysis, I conjecture that, because a subject like Clark has only unmarked ϕ-features (third person, singular), its features can be left unspecified in some dialects. When this happens, T cannot agree with the subject. The wh-phrase who is thus the phrase with explicit ϕ-features that is closest to T searching upward, and it governs agreement on T.7 This account can explain why the opposite of (20), upward agreement with a singular who over a plural subject, is not possible in these dialects (Kimball and Aissen 1971; Kayne 2000:190): (21)

*the man who the girls thinks is in the garden

Here the NP the girls has a marked ϕ-feature, namely plural. This cannot be left underspecified in the syntactic representation. Therefore the subject counts as an intervener that is closer to the finite T than the singular who. Agreement between Tense and who is thus blocked by the intervention condition ((1b) in this example). One might wonder if the agreement between T and who in (20a) could instead be analyzed as T agreeing downward with who before who moves to the highest Spec, CP. This alternative seems unlikely. Kimball and Aissen (1971:246) state that “only verbs between the position of the relativized NP and the head of the relative clause are subject to nonstandard agreement.” Agreement is thus not found in examples like those in (22). (22)

a. The boy think*(s) the people are in the garden. (Kayne 2000:208) b. Lucine know*(s) which people John think(s) are crazy. (Linda Baker, personal communication; cf. Kimball and Aissen 1971:246)

(22a) shows that the matrix T cannot agree with a plural NP in the argument position that the wh-phrase in (20a) moves from. This is expected, since the 7 Hans Broekhuis (personal communication) points out that something needs to be said about why this dialect does not allow T to agree with a plural direct object – the closest NP with ϕ-features searching downward – when the subject is third person singular. This would wrongly allow sentences like *Clark like the girls. I assume that this sort of agreement violates the activity condition in (1d), the object already having its accusative case licensed by v. This in turn raises the question of how the agreement with the wh-phrase in (20a) can be squared with the activity condition. One possibility is that the uninterpretable features involved in wh-movement render the wh-phrase eligible for agreement. For another conjecture, stated in terms of the revised and parameterized version of the activity condition that I propose in chapter 5, see note 6 in chapter 5.


77

plural NP is in a phase (the embedded CP) that does not contain T. (22b) shows that even when a plural NP moves to the specifier of the embedded CP, agreement is not possible in the relevant dialect. Restating Kimball and Aissen’s generalization in current terms, agreement takes place between T and the whphrase only when the wh-phrase crosses over T to land in a higher position.8 This, then, is an instance of upward verbal agreement showing up in exactly the situation where we predicted it might. See also Bruening 2001:sec. 4.3 on socalled participle agreement in Passamaquoddy for another, more robust instance of verbs agreeing in ϕ-features with an operator in Spec, CP. (It is not entirely clear, however, whether the relevant agreement morphemes in Passamaquoddy are realizations of agreement on the T node, or some other head, such as v or C.) 3.2.2 Upward agreement on v Consider next the possibility of upward agreement on v, the other agreementbearing functional head that can dominate VP. The prospects for this happening are somewhat brighter, inasmuch as there is an NP that is regularly found above v – the agent noun phrase, generated in Spec, vP in Chomsky 1995 and related work. The question, then, is whether v ever agrees with the agentive subject, or whether subject agreement is necessarily in Tense or some other higher functional node. 3.2.2.1 Basque and other languages with ergative agreement Bejar (2003) and Rezac (2003) have argued that agreement between v and the subject happens in a class of languages with complex agreement systems, including Basque, Georgian, and Nishnaabemwin. The Basque case is perhaps the most transparent, so it is the one that I review here. In Basque, there is a certain agreement slot, the prefix to the auxiliary root, which normally registers agreement with the object of a transitive verb. Agreement with the subject of the transitive verb is distinct from this, both in terms of its position (it suffixes to the auxiliary) and in terms of phonological features, as shown in (23).9 8 (20a) might not count as evidence for upward agreement if wh-movement has intermediate landing sites between the lower Spec, CP and the higher Spec, CP, as Kayne (2000:208) assumes, citing evidence from quantifier stranding. This theoretical device is not necessary to account for agreement with a wh-phrase in Spec, CP in my theory, and I take this to be a good thing, given that the evidence for other landing sites is weak and theory-internal at best. (I find Kayne’s example quite bad, for example.) 9 For all but a small number of Basque verbs, tense and agreement appear on an auxiliary verb rather than on the main verb. I assume that this is a phenomenon akin to do-support: perhaps most

78 (23)

The unity of verbal and adjectival agreement ni ikusi n-au. (Salaburu Etxeberria 1981:138) a. Andoni-k Anthony-erg me.abs see 1sA-aux ‘Anthony sees me.’ b. Andonik zu ikusi z-aitu. (Salaburu Etxeberria 1981:138) Anthony-erg you.abs see 2A-aux ‘Anthony sees you.’ c. Nik zu ikusi z-intu-da-n. (Saltarelli 1988:306) I.erg you.abs see 2A-aux -1sE-past ‘I saw you.’ d. Ikusi n-indu-zu-n. (Saltarelli 1988:306) see 1sA-aux-2E-past ‘You saw me.’

In these examples, prefixal n is object agreement with first person singular, as seen in (23a, d), whereas suffixed da is subject agreement with first person singular ((23c)). Similarly, the prefix z is object agreement with second person singular polite ((23b–c)), and the suffix zu is subject agreement with second person singular polite ((23d)). Rezac (2003) thus assumes that forms like n and z are morphological realizations of agreement on v, and da and zu are realizations of agreement on T, in a standard structure like (24) for (23c). (I suppress the head-finality of Basque for expository convenience.) (24)

TP NP

T´

T vP I [1,sg] [1,sg] NP v’ t

NP

v’

you v VP [2,sg] [2,sg] NP V’ t

V

XP

see verbs cannot raise up to v, and the auxiliary is inserted into v and raises to T, thereby spelling out tense, transitivity, subject agreement, and object agreement, but not the lexical content of the verb or some aspect distinctions. In any case, I abstract away from this. Note that the auxiliary stem itself also shows complex allomorphy conditioned by various factors as well as bearing the fairly regular and easily segmentable agreement affixes.


79

Based on this partial paradigm, one would expect to find forms like AUXda-n for ‘I verbed him’ and AUX-zu-n for ‘you verbed him’. But instead one finds the forms in (25). (25)

(Salaburu Etxeberria 1981:142) a. Nik liburua erosi n-u-en. I.erg book.abs read 1sA-aux-past ‘I read a book.’ b. Zuk liburua erosi z-en-u-en. you.erg book.abs read 2sA.fam-aux-past ‘You read a book.’

In these examples, the thematic subject triggers the “object” agreement prefixes n and z rather than the “subject” agreement suffixes da and zu. The agreement affixes in (25a–b) are therefore identical to those in (23a–b) (although the suppletive form of the auxiliary root is different; see note 9). This is known in the Basque literature as ergative displacement. Building on Bejar’s work on Georgian, Rezac analyzes this as follows. He assumes that third person arguments are not marked for person features in Basque, but rather are unspecified for this feature, much as I assumed for the subject Clark in (20). As a result, when v probes downward for something to agree with in sentences like (25), it finds no suitable goal. In this case, v probes upward instead, and agrees with the agent NP in its specifier position.10 This agreement relationship makes the NP inactive, presumably by valuing its case feature. As a result, T is not able to agree with the subject, by the activity condition. The thematic subject thus triggers “object” agreement but not subject agreement in these examples: (26)

TP T

vP

[---]

NP

v’

VP I v [1,sg][1,sg] NP V’ AGREE

book V XP [--] read 10 Rezac (2003) himself proposes a Minimalist interpretation under which the agreement of a head with its specifier does not really count as upward agreement in my sense (see also Bejar 2003). He suggests that the copy of a head that is taken as the label of the projection in Chomsky’s Bare Phrase Structure theory counts as a full-fledged instance of that head. Everything contained in the maximal projection, including the specifier, then counts as being in the c-command domain of that copy of the head. Taking his view would decrease the amount of evidence that counts as unambiguous support for my theory of agreement. On my theory, however, no such radical view about the nature of labels and their role in a syntactic derivation is required.

80


This then is a second apparent instance of upward agreement in the verbal domain.11 Given the analysis in (26), we would also expect v to agree upward with the thematic subject in intransitive clauses in Basque. These would be like the transitive clauses just discussed inasmuch as there is no NP with ϕ-features within the c-command domain of v that it can agree with. Therefore, v is forced to search upward in these clauses too. This is correct: the sole argument of an intransitive verb in Basque triggers “object” agreement forms such as prefixal n and z, as shown in (27). (27)

(Ortiz de Urbina 1989:9) a. Ibil-tzen n-intz-en. walk-hab 1sA-aux-past ‘I used to walk.’ b. Jaus-i-ko z-ara. fall-fut 2sA.fam-aux.npst ‘You will fall.’

Since this type of agreement is triggered by intransitive subjects as well as by transitive objects, it is usually – and more accurately – called absolutive agreement rather than object agreement. But it can be nicely analyzed as agreement on v, which probes downward when it can (in transitive sentences) and upward when it cannot (in intransitive sentences). This is also one possible analysis of ergative-absolutive agreement paradigms in other ergative languages, even if they have no equivalent of ergative displacement. These languages could be treated just like Basque, except that the object NP is never underspecified for ϕfeatures. As a result, v always probes downward to find the object in transitive clauses and probes upward to find the subject in intransitive clauses, resulting in an ergative-absolutive pattern of agreement. There are other languages of the world that have ergative agreement patterns, and there is no consensus about how they should be analyzed theoretically. Nevertheless, it is likely that many of them also involve v agreeing upward with the agent in Spec, vP. Rezac’s view that absolutive agreement is agreement on the lower functional head (here v) is a descendant of one proposed by Chomsky (1993:9) and argued for by Bobaljik (1993) for Inuit (see also Levin and Massam 1984). In contrast, various authors have argued that absolutive agreement is agreement on the higher functional head (here T) and ergative 11 Number agreement does not displace in Basque the way that person agreement does; see Rezac 2003 and references there for details. I abstract away from this here. Also ergative displacement happens in some tenses (e.g., past) but not others (present), a fact that needs to be taken into account in a fuller treatment.


81

agreement is agreement on the lower functional head (v), including Murasugi (1992), Campana (1992), Bittner (1994), and O’Herin (2002) (see also Bittner and Hale 1996). According to this second view, an ordinary transitive sentence like ‘you see me’ would have a structure like the following: (28)

TP NP

T´

T vP [1,sg] [1,sg] NP

v’

you v VP [2,sg] [2,sg] NP V’ V [1,sg] see

XP

This structure presents at least one theoretical difficulty as compared to Rezac’s (26): the Tense node either has to look past the agent NP in order to agree with the object, or the object needs to move up past the agent NP to be agreed with by T. The first option violates the intervention condition in (1b), whereas the second might violate the Shortest Move condition, plus it is often not borne out in the surface word order (for example, the word order in (29b) is Agent-ThemeVerb, not Theme-Agent-Verb). But these theoretical problems might well be solveable,12 and (28) has empirical advantages in some languages. For example, consider the following range of data from Abaza, which O’Herin (2002) uses to argue for a structure like (28): 12 It is interesting to note in this regard that ergative agreement patterns are not particularly common, suggesting that there is something special and marked about them. For example, the World Atlas of Language Structures (Haspelmath et al. 2005:map 100) identifies only 19 languages with ergative agreement in its sample of 390 languages, as compared to 212 languages that have a “normal” accusative pattern of agreement. One marked possibility might be that the realization of ergative case on the agent NP makes its ϕ-features invisible to T by something akin to P-insertion. Then the agent would not have the features that would make it count as an intervener between T and the object. Alternatively, it is possible that the object does move above the subject to Spec, TP, but the higher copy of the chain deletes rather than the lower one, so one still observes Agent-Theme-Verb order in (29b). (This is equivalent to O’Herin’s own view; see section 5.10.2.1 for evidence that something like this happens in Berber.)

82 (29)

The unity of verbal and adjectival agreement d-ʕ ay-d. (p. 64) a. A-phas def-woman 3sA.r-come-dyn ‘The woman came.’ b. Zaw ə a-phas də -y-ʕ a-y-d. (p. 67) someone def-woman 3sA.r-3sE.m-hear-pres-dyn ‘Someone hears the woman.’ c. D-qac’a-b. (p. 76) 3sA.r-man-stat.pres ‘He is a man.’ d. s-ba-ra (p. 28) 1sE-see-noml ‘seeing me’

(29a) is a simple intransitive verbal clause and (29b) is a simple transitive clause; they show that the transitive object triggers the same sort of agreement on the verb as the intransitive subject does (here the third singular human prefix d). The transitive subject is indicated by a different sort of agreement (for example, third singular masculine y in (29b)). (29c) shows a predicate nominal construction, which clearly has a T node (realized as the stative present suffix b) but presumably not a v node. This form bears an absolutive agreement d but not an ergative agreement y – evidence that absolutive agreement is agreement on T. Conversely, (29d) is a nonfinite verbal construction formed from a transitive verb; this example presumably contains a transitive v but no finite T. This form can bear an ergative agreement but no absolutive agreement, suggesting that ergative agreement is agreement on v (O’Herin 2002:76). Finally, O’Herin points out that absolutive agreement is the outermost prefix on the verb in examples like (29b), whereas ergative agreement appears closer to the verb stem. Mirror Principle-style reasoning thus suggests that absolutive agreement is on the higher functional head, the last one to unite with the verb – namely T. These arguments also carry over quite well to Mayan languages, such as Tzotzil (see Aissen 1987 for relevant data). So (28) is a serious analytic possibility. It is not clear to me whether (28) is the correct analysis for languages with ergative agreement, or whether (26) is, or whether Universal Grammar allows two distinct kinds of ergative agreement. But for my current interests it is not essential to have the last word on this topic, because both theories share the assumption that v sometimes agrees upward rather than downward. For Rezac the intransitive v agrees upward to give absolutive agreement, whereas for O’Herin the transitive v agrees upward to give ergative agreement. Either way,


83

adjective-like upward agreement plays a role in the analysis of ergativity, so upward agreement must be theoretically possible.13 3.2.2.2 Subject agreement in Tariana Verbal agreement in Tariana provides an interesting variation on the theme of v agreeing upward with the agentive subject. In transitive clauses, Tariana verbs agree with their subjects but not with their objects, as shown by the examples in (30) (all Tariana data come from Aikhenvald 2003). (30)

di-walita. a. Na-na kuphe-nuku 3.pl-obj fish-top.non.a/s 3sS-offer ‘He offered them fish.’

(p. 236)

b. Nuha siɾ uɾ i nu-walita nu-a. I cumatá.leaves 1sS-offer 1sS-go ‘I am making an offering of cumatá leaves.’

Many intransitive verbs show the same agreement with their sole argument: (31)

nu-a. (p. 239) a. Nu-ɾ uku 1sS-go.down 1sS-go ‘I am going downstream.’ b. Di-thuka-kha di-ɾ uku di-a diha-na-ne. (p. 235) 3sS-break-intrans 3sS-fall 3sS-go art-cl:vertical-foc.a/s ‘It (the penis of an evil spirit) was breaking and falling off.’

But there is a subclass of intransitive verbs that do not show any agreement with their sole argument. These are stative/nonagentive verbs, presumably a subset of the unaccusative verbs of Tariana: (32)

Lama-sina diha-dapana. (not: *di-lama-sina . . .) burn-rep.nonvis art-cl:habitation ‘This house was burning.’

(p. 239)

Why should the presence of subject agreement depend on the argument structure and lexical semantics of the verb in Tariana? This makes sense if the one agreeing head in this language is v rather than T. Unlike all transitive verbs 13 There is one viable theory of ergative agreement systems that makes no use of upward agreement: the view that both ergative and absolutive agreement are realizations of agreement on T (see Bok-Bennama 1991 on Inuit and Bobaljik and Branigan 2006 on Chukchi). I argue that this is true for Nez Perce (section 5.10.2.2), and tentatively accept it for Inuit and Chukchi as well. This approach does not seem so plausible for Abaza or Tzotzil, however, given the existence of syntactic environments like (29c–d) which have one sort of agreement but not the other.

84


and some intransitive verbs, stative intransitive verbs like ‘be burning’ plausibly lack a v head. This would account for their inability to show agreement, given the assumption that v is the only functional head with agreeing power in Tariana. Support for this idea comes from certain other details about agreement in Tariana. First, agreement is not morphologically related to Tense or Mood morphology in Tariana, as it is in many languages. Agreement is realized as a prefix on the verb, whereas Tense is a suffix. Nor do variations in Tense condition allomorphy of the agreement prefixes in this language. Second, Tariana has nonfinite verb forms that are comparable to gerunds in other languages. Gerundive verbs lack the ability to agree with their subjects in languages where subject agreement is associated with the Tense node, including English, Italian, Kinande, Lokaa, and Mapudungun, among others. But Tariana is different in this respect: nonfinite gerundive verbs show exactly the same subject agreement as other verbs in the language. Aikhenvald (2003:461) thus states that “all nominalizations formed on prefixed verbs cross reference the A/Sa [i.e. subject] constituent and can be derived from any verb.” Some representative examples are:14 (33)

na-ɾ apa-nipe] pi-na tuki nu-kalite-de. a. [H`ı maɾ u dem:an dance.master 3pS-dance-noml 2.sg-obj a.little 1sS-tell-fut ‘I will tell you a little about the dance master’s dancing.’ (p. 462) b. hinipuku w-ehpani-nipe; pe:the na-ni-nipe (p. 461) garden 1pS-work-noml manioc.bread 3pS-do-noml ‘our working in gardens’ ‘their making manioc bread’

This fits very well with the idea that subject agreement is not a property of finite Tense, which is missing in (33), but rather of v, which is present to assign the agent θ -role to the subject in each example. Third, passive verbs in Tariana systematically lack subject agreement. Rather, the prefix slot where this agreement normally appears is filled with the invariant prefix ka. Thus there is no 3sS prefix di on the passive verb in (34a), and no first person nu in (34b). (34)

a. Ha-ne ka-ñha-kana-mhade di-a. (p. 259) dem:inan rel-eat-pass-fut 3sS-aux ‘This one will be in the process of being eaten up by the jaguar.’

14 I do not know why ‘the dance master’ is translated as a singular noun but triggers plural agreement on the verb here. Perhaps this is a plural of honorification.

Verbs that cannot agree in person, like adjectives b. tʃ o! Nhua-sini ka-ñha-kana-kasu. Oh I-too:act rel-eat-pass-int ‘I, too, am about to be eaten!’

85

(p. 259)

Here the active v is missing and a finite Tense is present, and agreement is also missing. Overall, subject agreement in Tariana is contingent not on the verb being finite as opposed to nonfinite, but rather on the voice and lexical semantics of the verb – exactly the distribution we expect if the sole agreement factor in this language is attributed to v rather than to T. But even though this agreement has the earmarks of being associated with the v node, it is nevertheless agreement with the thematic subject. This then is another instance of v agreeing upward with the agentive subject generated as its specifier. Comparing Tariana with Basque gives us another chance to consider what governs the choice of upward versus downward agreement when both are possible, a matter that I was unable to resolve in the domain of adjectives. There seems to be a basic difference between the two languages on this point. On Rezac’s analysis, the v in Basque prefers to agree downward with the thematic object; it only agrees upward when there is no object that is marked for ϕ-features. In contrast, the v in Tariana can only agree upward with the thematic subject. This difference could be parametric in nature. In chapter 5, I argue that Kinande and other Bantu languages have a more restrictive version of the c-command condition in (1a) than the Indo-European languages do: Kinande allows agreement on a head F only if the agreed-with NP asymmetrically c-commands F. Given such a parameter, much of the difference between Tariana and Basque can be captured by saying that Tariana is like Kinande in allowing only upward agreement. (See section 5.6.2 for some confirming evidence for this from the syntax of quantifiers in Tariana.) I conclude that v not only may but actually must agree upward in some languages.

3.3

Verbs that cannot agree in person, like adjectives

So far I have shown that in certain configurations verbs can be like adjectives in agreeing with a higher NP rather than a lower one. Now I turn to the other salient difference between adjectival agreement and verbal agreement that could stand in the way of a unified theory: the fact that verbs agree with their arguments in person as well as number and gender, whereas adjectives never do. The SCOPA stated in (2) says that this is because FA never merges directly with an NP that has +1 or +2 features, whereas an FV can. For adjectives, the consequence is a firm one: the complement of FA is AP (by definition) which

86


does not have ϕ-features, and inasmuch as FA is intrinsically adjectival it never licenses a specifier (see (17)). Therefore, the functional head most immediately associated with adjectives can never agree in person for fundamental reasons.15 But for verbs there should be chances of finding intermediate cases. There might very well be instances in which a functional head dominating VP agrees with something that it does not merge with, either because it exceptionally lacks an EPP feature, or because its EPP feature is discharged by something that does not have ϕ-features. In these situations, the expectation of my theory is that the verbal head can still agree with a nearby NP in number and gender, but should lose the ability to agree in person. Such instances of verbal agreement would be adjective-like in this respect, and would support the notion that the same category-neutral theory of agreement applies to both. In this section I document a series of such cases. 3.3.1 Tense agreeing with Spec, CP One case that should be relevant is the agreement between Tense and a whphrase in the Spec, CP above it, found in the nonstandard dialect of English discussed by Kimball and Aissen (1971) and Kayne (2000) (see section 3.2.1). The crucial paradigm is in (35). (35)

a. b. c. d.

?the people who Clark think are in the garden ?The people who Clark are hoping are in the garden I who am (*is) in the garden *I who Clark am hoping am in the garden

(35a) repeats a standard example of the effect, where the verb think shows plural agreement with the wh-word who rather than singular agreement with its subject Clark. (35b) is a similar example where the crucial verb is be, which shows suppletive plural agreement with the moved wh-word. Two informants for this dialect say that they do not find this sentence appreciably worse than (35a). (35c) shows that a wh-phrase can count as first person singular when it is anteceded by a first person singular pronoun in a nonrestrictive relative clause. (35d) is the crucial test sentence. Here the verb be shows suppletive first person singular agreement with the wh-phrase in Spec, CP, despite its grammatical subject being third person singular. My informants agree that this is completely impossible, clearly worse than the example with only number agreement in 15 But recall that a head higher than FA – T or Pred – in a sentence containing adjectival predication can agree in person with the subject, and in some languages these higher heads appear morphologically joined with the adjective. See section 2.5.1 for discussion.


87

(35b). Apparently, then, T can agree in number with something in Spec, CP, but it cannot agree in person with something in that position. This matches the expectation of my theory: the wh-phrase never merges directly with a projection of T, so the specific environment for agreement in first or second person does not hold. T agreeing with Spec, CP is more adjective-like in this respect than T agreeing with its own specifier. Similarly, when verbs agree with an operator in Spec, CP in Passamaquoddy, they agree in the third person features singular versus plural, animate versus inanimate, and proximate versus obviative – not in first or second person (Bruening 2001:207). 3.3.2 Agreement in quirky subject constructions Another relevant case to consider – and one that has been much discussed in the recent literature – concerns quirky subject constructions in Icelandic. Icelandic has many dyadic predicates in which the thematically higher argument (typically the experiencer of a psych predicate) has a lexically specified case, often dative. Since this argument is the highest, it is raised to Spec, TP to satisfy the EPP feature of T. However, T is unable to agree with its quirky-case marked subject. Under these conditions, T can (for many speakers, must) show a degree of agreement with its nominative case object. (36a) is an example, where the plural agreement on ‘was bored by’ is controlled by the plural nominative object, and not by the dative subject (compare the singular agreement in the parallel (36b)). (36)

þeir. (Taraldsen 1995:307) a. Henni leiddust her.dat was.bored.by.3pS they.nom ‘She was bored with them.’

b. Henni leið ist bókin s´ın. (Boeckx 2000:356) her.dat was.bored.by.3sS book self’s ‘She finds her own book boring.’

This is another situation where T, a verbal functional head, agrees with something that is not its specifier, something that never merges with it. And in this case too T can only agree with the nominative object in number, not in person. Thus, examples like (37), which have first or second person nominative pronouns, are ungrammatical.16 16 Bobaljik (to appear: note 27) notes that first and second person pronouns cannot be used as objects of dative subject verbs even when the verb is used in a control infinitive. He therefore suggests that the restriction in (37) has nothing to do with morphological agreement per se. Instead, I tentatively follow Schütze (1997) and say that the same kind of syntactic agreement takes place in infinitives as in tensed clauses in Icelandic, but the agreement is not spelled out morphologically on the nonfinite verb.

88 (37)

The unity of verbal and adjectival agreement við . (Taraldsen 1995:309) a. *Henni leiddumst her.dat was.bored.by.1pS we.nom ‘She was bored with us.’ ´ veit að honum l´ıkið b. *Eg þið . (Sigurdsson 2002:720) I know that him.dat like.2pS you.nom.pl ‘I know that he likes you all.’

Again, this is just what my theory – which was designed primarily to capture the fact that adjectives agree in number but not person – predicts should be the case. My account of these Icelandic facts is not radically different from theories that have been proposed in the past. I submit, however, that it is a bit simpler and more elegant. Many previous researchers, including Taraldsen (1995), Sigurð sson (2000, 2002), Boeckx (2000), and Anagnostopoulou (2003) have developed the intuition that the Icelandic verbs in (37) cannot undergo full agreement with their objects because they have previously undergone a type of agreement with the quirky case subject. One way to express this idea is that the dative subjects are not completely devoid of ϕ-features; rather they bear a third person feature and are unspecified for number only (Anagnostopoulou 2003:269–70). The first time T searches, it finds the quirky case subject, establishes an Agree relationship with it, and moves it to the Spec, TP position. Because the dative subject is only partially specified for ϕ-features, however, only the person feature of T is valued (as −1, −2); the number feature of T is left unspecified by this first Agree relationship. Since it still has an unvalued feature, T can probe again, this time locating the nominative object. Agreement with that object then values the number feature of T, but cannot change the already valued person feature. This explains why examples like (36) are grammatical, but examples like (37) are not, according to this line of reasoning. Although this line of research and mine account for the same facts, the alternative theory rests squarely on one not very plausible assumption: that all quirky-case-marked arguments are third person and unspecified for number. It is far from obvious that this is true. If one considers dative-case arguments to be simple NPs, then they seem able to have every possible value of person and number. It is odd then to say that mér in (38) is third person and unmarked for number when its interpretation is first person and singular. (38)

Mér l´ık-ar (*l´ık-a) bók-in. (Taraldsen 1995:310) I.dat like-3sS like-1sS book-the.nom ‘I like the book.’


89

The alternative would be to analyze dative case expressions as PPs that contain an NP with ordinary ϕ-features as their complement (see Markman 2005, among others). On this second view, it makes sense that the PP as a whole is unmarked for number, even though the NP it contains is singular or plural. But it does not make sense that the PP is marked for third person. One would expect such PPs to have no ϕ-features whatsoever. So there is no very plausible analysis of phrases in dative case that makes sense of the partial ϕ-feature specification that the Boeckx–Sigurð sson–Anagnostopoulou analysis depends on. In contrast, my account does not depend in any crucial way on the ϕ-features of the dative subject; all that is required is that this have the minimal feature needed to satisfy the EPP property of T in Icelandic. For me, person agreement with the nominative object in (37) is not blocked because there is partial agreement with the dative subject. Rather there is a direct configurational explanation: the dative subject fills the specifier position but cannot be agreed with (probably because of the activity condition; see chapter 5 for discussion). Therefore, T agrees with a more remote NP. This precludes it from showing person agreement, by the SCOPA. My theory is simpler in that it avoids making an uncomfortable stipulation about the ϕ-features of dative expressions. It also captures a similarity between these special verbal constructions in Icelandic and the robust fact that adjectives do not show person agreement, which alternative theories do not capture.17 These Icelandic examples vindicate the choice made in section 2.4.2 to say that it is nonlocal agreement that cannot pick up +1 or +2 features. A plausiblelooking alternative was to stipulate that downward-probing agreement can include agreement in first or second person, but upward-probing agreement cannot.18 That would have been sufficient to explain why predicate adjectives 17 Sigurð sson (2000) mentions an empirical consideration that could weigh against an analysis like mine. He shows that T can agree in person with a postverbal pronoun when the subject position is occupied not by a dative subject, but by the expletive þ að in special copular constructions like (i). (i)

það erum

bara við . It are.1pS only we.nom ‘It’s only us.’

Following the second suggestion in Sigurð sson’s note 25, I tentatively assume that þ að in (i) is introduced higher than the T space, in a position like Spec, CP. This opens up the possibility that the pronoun in (i) is really in Spec, TP, where it agrees in person with T, the T+verb combination then moving to C. (This might not extend to Sigurð sson’s example (67), however, so I leave a full analysis to future research.) 18 This alternative theory would also wrongly allow Cinque adjectives and raising adjectives to agree with an NP in person.

90


do not agree in person, whereas verbs canonically do. The upward versus downward proposal would also extend to the agreement with wh-phrases in Spec, CP, shown in (35). But it does not extend to this Icelandic paradigm. Here T is agreeing downward, with an object that is always within its c-command domain, yet agreement is restricted to number. This shows that it is not whether agreement is upward or downward that is crucial, but rather whether it is strictly local, with F and its controller being merged together directly. Other examples below confirm this. The literature observes that this pattern of facts seems to be rather peculiar to Icelandic. Other languages have similar-looking dative subject constructions, but the verb shows full agreement with its nominative object. For instance, (39) shows that a dative-subject verb like ‘like’ can show first or second person agreement with its nominative object in Russian and Greek. (39)

a. Ja znaju, emu nravimsja my. (Russsian, Sigurð sson 2002:720) I know him.dat like.1pS we.nom ‘I know that he likes us.’ b. Tu Petru tu aresis esi. (Greek, Anagnostopoulou 2003:91) the Peter.gen cl.gen please.2sS you.nom ‘Peter likes you; you appeal to Peter.’

German and Kannada are also like Russian and Greek in this respect (on Kannada, see Sridhar 1979, 1990). To the extent that my theory naturally explains the Icelandic pattern, the existence of this full-agreement pattern is potentially threatening to it. Sigurð sson (2002), however, makes the crucial observation that dative subject constructions in Icelandic differ from their Russian and German counterparts in ways other than agreement. In particular, Icelandic quirky case subjects are famous for being especially similar to canonical subjects in nominative case when it comes to a range of syntactic properties, including word order, raising, control, conjunction, and anaphora, as originally shown by Zaenen, Maling, and Thráinsson (1985). In contrast, dative “subjects” are not nearly as subject-like in German, but behave more like fronted topics. Even in Russian and Greek, the oblique arguments have only a subset of the normal syntactic properties of nominative subjects. For example, the dative experiencer can be elided in a conjunction when it is parallel to a nominative subject in Icelandic, but not in Greek or German.19 19 Oblique subjects do show some binding theoretic properties that are similar to those of ordinary subjects in Greek and Russian. But see Franks 1995:254–5 for data showing that anteceding

Verbs that cannot agree in person, like adjectives (40)

91

a. Hann segist vera saklaus en – hefur v´ıst verið he.nom says.refl to.be innocent but [dat] has apparently been hjálpað ´ı prófinu. helped on the.exam ‘He claims that he is innocent, but apparently has been helped during the exam.’ (Icelandic, Zaenen et al. 1985:456–67) b. *O Petros latrevi tin musiki ala dhen – the Peter.nom adores.3sS the music.acc but not [gen] aresun ta mathimatika. please.3pS the mathematics.nom ‘Peter adores music but doesn’t like math.’ (Greek, Anagnostopoulou 2003:91)

This makes it very reasonable to suppose that the preverbal oblique expressions in Greek, Russian, and German do not actually sit in Spec, TP. Rather, they could sit in some higher position, such as Spec, TopicP, or they could be the specifier of some decomposed functional category that has some of the properties of T in English and Icelandic but not others. Now if subject properties show that oblique NPs are not in the normal Spec, TP position in Greek, Russian, and German, then it is perfectly possible that the nominative pronouns in examples like (39) are in Spec, TP. If they are, then it is expected that T can bear +1 or +2 agreement. After this agreement is established, the V+T combination could move on to a higher head position, such as C in verb-second clauses in German, without destroying the agreement. One would certainly like to refine this analysis, to learn more about exactly where fronted dative expressions appear in different languages and why they appear there. But there is no strong reason to fear that agreement in languages like Greek, German, and Russian will be particularly problematic for my theory, even though they differ somewhat from the Icelandic facts that support that theory so nicely.20 Another language that works like Icelandic – and so supports my theory – is Gujarati, discussed by Bhatt (2005:801). (41a) shows that in perfective aspects when the subject is marked ergative, the Gujarati verb can agree with the object, anaphors and controlling participial clauses are not reliable diagnostics of being a subject in Russian. 20 Similarly, the Indo-Aryan languages Maithili and Nepali allow full person agreement with nominative “objects” in dative experiencer constructions, but Bickel and Yadava 2000 show that the dative experiencers lack important subject properties in these languages (properties that their Icelandic analogs have). Taraldsen (1995) points out that transitive impersonal constructions in Italian are like quirky subject constructions in Icelandic in not allowing the verb to agree with a first or second person nominative object. Examples like this with featureless clitics or expletives in Spec, TP are another possible source of evidence for the SCOPA.

92


even though it is marked accusative (unlike the better-known facts in Hindi). On my theory, this suggests that the accusative particle ne is a pure case marker, not an adposition which creates a barrier for agreement (and that agreement in Gujarati is not dependent on case; see chapter 5). Crucially, though, this verb is a participial form that agrees only in gender. Auxiliary verbs in Gujarati do show person agreement as well as gender and number agreement; thus, ‘be’ has a special second person form in agreement with the second person nominative subject in (41b). The key example is (41c), in which there is an auxiliary along with the perfective verb and the object is second person plural. Here the second person form cho, seen in (41b), is impossible, and a default third person form che is used instead. (41)

a. mEN tehmahri behEn-one bolawi. (Bhatt 2005:774) I.erg your sisters.f-acc invited.f ‘I invited your sisters.’ b. tEhme a¯ w-ya cho. (Bhatt 2005:801) you.pl come-perf. m.pl be.pres.2pS ‘You have come.’ c. Mãi tam-ne mar-y¯a che. (Bhatt 2005:801) I.erg you.pl-acc strike-perf. m.pl be.pres.3S ‘I have struck you.’

Gujarati is thus like Icelandic in allowing verbs to agree in gender but not in person when the agreed-with NP is something other than the phrase in Spec, TP.21 A non-Indo-European language in which this effect of the SCOPA can be seen is the Muskogean language Chicasaw. Monroe and Gordon (1982) discuss “dative subject” constructions like the following:

21 Bhatt took (41a) and (41c) from different sources, which apparently use different transliteration systems for short and long vowels. This accounts for the different forms of the ergative first person pronoun. (41) leads Bhatt (2005:800) to posit what he calls the Person Generalization: “Dissociated agreement does not involve Person,” where by “dissociated agreement” he means agreement where the head does not license case on the agreed-with NP. Like my SCOPA, Bhatt’s Person Generalization would cover the fact that adjectives do not agree in person, since adjectives are not structural case licensers. The two generalizations are not identical, however. For example, there is some reason to think that T in Icelandic licenses nominative case on the object in examples like (36) (Taraldsen 1995, Schütze 1997), but it still cannot agree with the object in person. Conversely, there are Cs in Kinande that agree with an operator in person even though they do not assign case to those operators (see section 5.4.1).


a. Hattak-at talowa’ [i-m]-alhkaniya-tok. man-subj song 3-dat-forget-past ‘The man forgot the song.’ b. Talowa’ [a-m]-alhkaniya-tok. song 1s-dat-forget-past ‘I forgot the song.’

93

(p. 90)

(p. 85)

c. Doris-at Claire-at [i-m]-ala-tok. (p. 99) Doris-subj Claire-subj 3-dat-arrive-past ‘Doris gave birth to Claire.’

The overt subjects in (42a, c) do not bear overt dative case, but the “agreement” prefix attached to the verb does. I tentatively assume that this prefix is actually an incorporated oblique pronoun; when there is an overt “subject,” it is really a topic licensed by binding this dative-case pronoun, which is the true argument. Munro and Gordon give evidence from the switch reference system of Chicasaw that the dative argument does count as the syntactic subject in examples of this type (see also Davies 1986 on related Choctaw). Suppose then that the dative pronoun is in Spec, TP before it cliticizes onto the verb. (42b) shows that there is no additional agreement on T with this dative subject (the verb is not a-m-alhkaniya-li-tok ‘1-DAT-forget-1sS-past’). The question, then, is whether T agrees with the internal argument of the verb in these dative subject constructions in Chicasaw, the way it does in Icelandic. One cannot tell simply by inspecting the morphology on the verb in (42), because third person subject agreement is always Ø in Chicasaw. But evidence that it can agree is the fact that the internal argument can bear nominative case in dative subject constructions – and only in dative subject constructions – as shown in (42c). I take this to be indirect evidence that this internal argument agrees with T, and hence can get case from it. (This assumes that Chicasaw is a language in which agreement is case-related in the sense of chapter 5. It also entails that there are two sources for the SUBJ suffix in Chicasaw: it can be a topic marker, or it can be the realization of case licensed by T; compare Japanese “nominative” ga.) Imagine, then, how one would say ‘I forgot you’ in Chicasaw. The dative subject would be in Spec, TP, and T would be left to search downward, inside VP for something to agree with. But the only argument inside VP is second person, and T cannot agree at a distance with such an argument, by the SCOPA. If these assumptions are correct, one might expect ‘I forgot you’ to be ungrammatical in Chicasaw, on a par with the badness of (37) in Icelandic. That is correct: Munro and Gordon (1982) report that dative subject constructions are impossible if the

94


non-dative argument is first or second person. Speakers are thus driven to offer complex paraphrases when asked how to say ‘I forgot you.’22 I conclude that there are languages both inside and outside the Indo-European family in which the impossibility of nonlocal verbal agreement with first and second person arguments can be observed in oblique subject constructions – although further typological study of the matter would be most welcome. 3.3.3 Agreement in double object constructions So far the focus has been on the agreement properties of T when something other than what it agrees with occupies Spec, TP. Do similar considerations apply to v, the other functional head with agreement-bearing potential associated with verbs? Are there times when the Spec-head relationship between v and the direct object is disrupted by some other syntactic element, with the result that the verb can only show limited agreement with the direct object? In fact, such cases exist and have been discussed in the recent literature, often under the name of the Person Case Constraint (PCC), after Bonet 1991. The effect shows up in double object and applicative constructions, in which there is a benefactive/goal argument in the domain of v as well as the normal direct object. In such configurations, v (in languages where v agrees at all) normally shows full person and number agreement with the benefactive/goal argument. This is expected, assuming that the benefactive/goal argument is structurally the highest NP in the complement of vP (at the relevant level), hence the NP that can most readily be attracted to Spec, vP (see section 2.3.1). When the goal merges with the projection of v, agreement in first or second person as well as number and gender is permitted by the SCOPA. The striking fact is that, in some such languages, the verb can also show a degree of agreement with the second internal argument, the theme/direct object. Crucially, however, this agreement is for number and gender only, not for person.23 One language that fits this characterization is Classical Nahuatl (Launey 1981). (43) displays examples with a simple double object verb ‘give’. (43b) 22 The facts that Davies (1986) reports for closely related Choctaw are a little bit different, but Choctaw also shows a difference between third person and non-third person arguments that can plausibly be attributed to the SCOPA. In Choctaw, ‘I forgot you’ is sayable, but ‘you’ cannot get nominative case and it triggers object agreement on the verb (Davies 1986:87, 92–93). This shows that the second person argument cannot agree with T in the relevant construction, but only with v. 23 For previous attempts to give a unified explanation for the agreement properties of dative subject constructions in Icelandic and double object constructions, see Boeckx 2000 and Anagnostopoulou 2003. For differences between their theories and mine, see discussion in the previous section.


95

contains the third person object marker /k/ (orthographic c or qu) and the animate plural object marker im. This form implies that both the goal and the theme arguments are third person and that one or the other of them is plural. (43a) contains the first person singular object marker n¯ech and the animate marker plural im. This form can only be interpreted as having a first person singular goal and an animate plural theme. (43)

a. Xi-n¯ech-im-maca hu¯ehuèx¯olô. (Launey 1981:174) 2sS.imp-1sO-pl-give turkeys ‘Give me some turkeys.’ (Not: ‘Give me to some turkeys.’) b. Ni-qu-im-maca hu¯ehuèx¯olô in n-ocn¯ı-uh. (Launey 1981:174) 1sS-3O-pl-give turkeys in 1sP-friend-poss ‘I gave some turkeys to my friend.’ (Also: Ni-qu-im-maca c¯e huèx¯olotl in n-ocn¯ı-hu¯an. ‘I gave a turkey to my friends’)

The constitutes a fairly standard PCC effect. The constraint has been formulated as follows (Anagnostopoulou 2003:251, based on Bonet 1991:182; see also Ormazabal and Romero 2007 for important empirical refinements). (44)

Person Case Constraint Context: Ditransitives with phonologically weak direct and indirect objects. Observation: If a direct and an indirect object co-occur, the direct object has to be 3rd person (i.e., cannot be 1st or 2nd person).

This constraint as stated applies to weak pronouns, clitics, and null pronouns related to agreement on the verb; I will be concerned only with agreement. My theory of agreement contributes to an explanation of this pattern as follows. In addition to assuming that the goal argument is the highest NP in the complement of v (at the relevant level),24 let us also assume that v in Nahuatl has exactly one EPP feature. This means that it can trigger the movement of one and only one NP to remerge with it as a specifier. Since the goal NP is the higher of the two internal arguments, it is the one that is attracted by v. The SCOPA then implies that v can agree with the goal but not the theme in first or second person. But the SCOPA does not constrain number and gender agreement; it is permitted for v to agree at a distance with the theme NP that remains in its complement for these features. The structure of (43a) then is (45).

24 I leave open the possibility that the goal starts lower in the VP and becomes the highest NP as the result of some kind of movement process (“dative shift”), as in Larson 1988 and Baker 1997. Whether this prior movement occurs or not is orthogonal to the matter at hand.

96


(45)

vP NP

vP

(you) NP

v´

me v VP [1,sg] [1,sg] [an, pl] NP

V´

V give

NP turkeys [an, pl]

I call this the two-and-a-half agreement phenomenon. This moniker expresses the observation that Classical Nahuatl cannot show full agreement with three distinct arguments (it cannot express things like ‘He gave you to me’), but it can do a bit more than just agreeing with two arguments. One can also observe a very similar two-and-a-half agreement effect in causative verbs derived from transitive verb roots in Nahuatl (Launey 1981:191). Another language with two-and-a-half agreement is Southern Tiwa, whose complex agreement system is described in detail in Allen, Gardiner, Frantz, and Perlmutter 1990. In this language, there are synthetic agreement forms for any combination of subject and goal arguments. The relevant agreement forms also vary with the gender and number of the theme, as shown in (46) (p. 335). (46)

a. Ka-‘u’u-wia-ban. 1sS/2sO/aO-baby-give-past ‘I gave you the baby.’ b. Kam-‘u’u-wia-ban. 1sS/2sO/bO-baby-give-past ‘I gave you the babies.’ c. Kow-keuap-wia-ban. 1sS/2sO/cO-shoe-give-past ‘I gave you (the) shoes.’

These particular examples all have a first person singular subject and a second person singular goal, both expressed in the agreement prefix. In addition, (46b) has a plural theme in contrast to the singular theme of (46a), and this also affects the verbal prefix. Similarly, (46c) has a plural inanimate theme whereas (46b) has a plural animate theme, and this difference too is registered in the verb


97

prefix. So there is a degree of agreement with the theme argument of ditransitive verbs in Southern Tiwa. However, just as in Nahuatl, one can never have first or second person agreement with the theme argument of a triadic verb: there is no prefix form for ‘I gave you to him’ or ‘He gave me to you,’ for example. The explanation is the same as for Nahuatl: the higher goal argument can move to Spec, vP and trigger +1 or +2 agreement, but the lower theme argument cannot. The v head can agree with the theme argument at a distance, but only in number and gender – the same limited repertoire of features as adjectives agree in. Some Bantu languages may also provide instances of two-and-a-half agreement, such as Shambala, whose object prefixes are described by Duranti (1979). Duranti mentions that reflexive verb forms in Shambala cannot bear any (other) object prefix (p. 36). I take this to be tentative evidence that the object prefixes in this language are true instances of agreement on the active v head, not cliticized pronouns derived by some kind of movement (see note 25, and also section 5.7.1 on object agreement in Bantu more generally). Nevertheless, the presence of two internal arguments can be registered on the verb in some instances, as in (47) (Duranti 1979:36). (47)

a. A-za-m-ni-et-e-a. she-past-him-me-bring-appl-fv ‘She has brought him to me.’ Not: ‘She has brought me to him.’ b. A-i-wa-mw-et-e-e. she-past-them-him-bring-appl-fv ‘She brought them to him.’ (marked number for theme) b. Na-i-mw-itang-i-a. I-it-him-call-appl-fv ‘I call it (the meeting) for him.’ (marked gender on theme)

Shambala thus permits agreement for more than just the subject and one internal argument. But agreement with two internal arguments is not unrestricted. For example, (47a) cannot be interpreted as having first person agreement with the verb’s theme argument and third person agreement with the goal argument, but only the other way around. The second imaginable interpretation is also ruled out when the order of the object prefixes is reversed (*a-za-ni-mw-et-e-a, intended to mean ‘she has brought me to him’ (Duranti 1979:36)). The only way to render this meaning in Shambala is to project the goal argument as a PP, the ϕ-features of which are inaccessible to v. Only then can v show first person agreement with the theme argument in the presence of a goal:

98 (48)

The unity of verbal and adjectival agreement A-za-ni-et-a kwa yeye. she-past-me-bring-fv to him ‘She has brought me to him.’

(Duranti 1979:36)

This is a classic PCC effect, where the theme object must be third person when the verb is ditransitive ((47)), but not when it is monotransitive ((48)). Put another way, Shambala allows full agreement with the higher NP in the verb phrase (the goal), and partial agreement – agreement in number and gender but not in person – with the lower NP. Under this description, Shambala is very much like Nahuatl and Southern Tiwa. Duranti (1979:36) also mentions that first and second person object prefixes can never appear together on the same verb in Shambala. This too is just what we would expect, given that v can only show person agreement with the one argument that it attracts to its specifier position. Unlike restricted agreement in dative subject constructions, the PCC phenomenon has been studied typologically and is known to be very general. Only a handful of languages have come to light in which first or second person theme arguments can be expressed on the verb in the presence of a goal NP (for example, Haspelmath (2004) presents four possible counterexamples: Kabardian, Lakhota, Haya, and Noon). Moreover, some of the known counterexamples seem to involve clitics or incorporated elements rather than true agreement.25 I thus make the bold claim, which I do not know to be false, that the PCC effect 25 One heuristic for telling whether a morpheme that expresses the object on a verb is a moved clitic pronoun or a true agreement might be to see if the morpheme in question can appear on verbs in the passive or in other nonactive voices (detransitivized reflexive, reciprocal, or anticausative verbs). Typically it is only the active version of v – instances of v that assign an agent role to an NP in their specifier – that is a legitimate bearer of object agreement (the agreement-oriented version of Burzio’s Generalization). A verb in a nonactive voice is thus expected not to be able to agree with a theme or goal object, even when the base verb is a ditransitive, so there is still one internal argument that could in principle be agreed with. Indeed, in languages with fairly clear instances of object agreement, such as Mohawk, Mayali, Mapudungun, and Nahuatl, there is no object agreement with the nonsubject argument of a nonactive ditransitive verb (e.g., Nahuatl: Ti-(*c)-mac-o in x¯ochi-tl (2sS-(*3sO)-give-PASS DET flower) ‘you were given the flowers’; Launey 1981:175). In contrast, in languages that clearly have object clitics, such as the Romance languages and Greek, an object clitic can perfectly well attach to the passive of a ditransitive verb (e.g., Spanish Este libro le fue dado (a Juan) ‘This book him.DAT-was given to Juan’). This heuristic can be applied to the minimal contrast between Shambala, which respects the PCC, and Haya, which does not (Duranti 1979, Hyman and Duranti 1982). Duranti states that Haya allows object prefixes to attach to passive and reflexive verb forms (Duranti 1979:43), but Shambala does not (Duranti 1979:36). That suggests that object prefixes have the status of clitics in Haya, but that of object agreement morphemes in Shambala. This encourages me in my belief that the PCC is exceptionless for true agreement, though not for cliticization (perhaps it holds in that domain only for the functionalist reasons that Haspelmath 2004 discusses).


99

is exceptionless in the area of agreement – although good tests to distinguish instances of true agreement from instances of cliticization and weak pronouns are clearly needed to substantiate this. This analysis of two-and-a-half agreement involves an enrichment of the theory. For examples like (43), (46), and (47), I must say that a single functional head (v) can have multiple specifications for the same ϕ-feature. For example, the v in the Nahuatl example diagrammed in (45) is both singular and plural. (Perhaps it is also simultaneously first person and third person, if third person is a real feature and not just a default value.) It contrasts with the form in (49), which is doubly plural, expressing the plurality of both internal arguments explicitly. (49)

An-tech-im-maca’. (Launey 1981:391) 2pS-1pO-pl-give. ‘You all give them to us.’

The active v in Nahuatl thus probes the phase twice in search for something to agree with, finding the goal the first time and the theme NP the second time. Not all languages allow v to probe twice in this way. For example, Mohawk and Mayali are typologically similar to Nahuatl and Southern Tiwa (Baker 1996) and they have systems of gender and number marking. Yet in ditransitive constructions they show full agreement with the agent and the goal, but no additional gender/number agreement with the theme. Similarly, Chichewa is a Bantu language that has the same sort of gender/number/noun-class system as Shambala, but Chichewa only allows a single object marker that expresses the goal, not a second one that expresses an additional theme argument. (50) gives a minimal pair illustrating this difference. (50)

a. *Alenje a-ku-z´ı-wá-phik-il-a (z´ıtúmbûwa anyâni). 2.hunters 2S-pres-8O-2O-cook-appl-fv 8.pancakes 2.baboons ‘The hunters are cooking them for them.’ (Chichewa, Mchombo 2004:82) (also * . . .wa-zi . . .) b. A-ya-i-dik-i-a. (Shambala, Duranti 1979:37) she-them-it-cook-appl-fv ‘She cooks them for it.’

It is conceivable that these differences could be derived somehow from differences in the ϕ-feature systems of the languages in question or from generalizations about how they are spelled out morphologically – but that does not

100


seem likely given (50).26 Probably there is a parametric difference at work that determines whether v can search for something to agree with more than once or not. This can be expressed as follows: (51)

A particular functional head F can search the phase for something to agree with: i. zero times, or ii. one time, or iii. two (or more?) times.

In essence, (51) says that a grammar can stipulate whether a given functional head is a probe or not, although not (I assume) the particular feature structure of the probe (contra Bejar 2003, Rezac 2003). Nahuatl, Southern Tiwa, and Shambala are languages that adopt the (51iii) option; Mohawk, Mayali, and Chichewa adopt the (51ii) option. The option in (51i) is for languages like English and Chinese, which have transitive v heads, but where those heads do not participate in agreement (see section 2.2.2). Note that even with this enrichment my theory falls well short of stipulating for each grammatical category which features it can and cannot agree with, the brute-force approach that is unexplanatory and allows for more crosslinguistic variation than we actually observe (see section 1.2). It is still predictable in this revised system which NPs v will agree with and in which features. There is an arbitrary component to the theory (does a language have two or two-and-a-half agreement?) but also a principled component (all languages have two-and-a-half agreement rather than three agreement).27 The general formulation of (51) suggests that it should apply to T too, allowing T sometimes to agree twice as opposed to once or not at all. Suppose for simplicity that this were to happen in a language or construction in which there was no agreement-bearing v head. Then we would expect to see instances of one-and-a-half agreement, where the finite verb agrees fully with the highest argument and in number and gender only with a lower argument (assuming T has only one EPP feature, so only the highest argument can remerge with the 26 Variation like that seen between Chichewa and Shambala in (50) is even found internal to the Lakhota language (another language that Haspelmath (2004) says contains PCC violations). Rood and Taylor (1996:sec. 9.3.8.3) state that “Some speakers inflect for both direct and benefactive objects, but others reject these forms as meaningless.” 27 Note that saying whether v is a single probe or a double probe is not the same as saying whether v has one EPP feature or two. Nahuatl and Southern Tiwa are languages in which v is a double probe but has only one EPP feature, resulting in two-and-a-half agreement. A language in which v was a double probe and had two EPP features could be a language that showed full person agreement with three arguments. I do not know that there are any such languages.


101

projection of T). In fact, this situation is attested in Southern Tiwa. Southern Tiwa has a distinct one-and-a-half agreement paradigm that it uses both for the passives of ‘give’-type verbs and for unaccusative motion verbs like ‘come’ when they take a goal argument. (52) shows that, in addition to full person agreement with the goal subject, a passive verb also varies with the number and gender of the theme argument. (52)

a. Hliawrade-ba in-khwian-wia-che-ban. (Allen et al. 1990:334–6) lady-instr 1sS/A-dog-give-pass-past ‘I was given a dog by the lady.’ b. Im-khwian-wia-che-ban. 1sS/b-dog-give-pass-past ‘I was given (the) dogs.’ c. Iw-keuap-wia-che-ban. 1sS/c-shoe-give-pass-past ‘I was given (the) shoes.’

Exactly the same agreement pattern is found on an intransitive verb of the unaccusative class when a goal argument is included. Thus, a sentence like ‘The dog came to me’ has the same prefix in as (52a), showing agreement with the theme ‘dog’ in number and gender as well as with the goal ‘me’ in person, number, and gender (Allen et al. 1990:356). These forms are, however, distinct from the prefix found in a simple transitive like ‘I saw the dog’ (ti-mu-ban) or the ‘The dog saw me’ (no prefix; passive required). Moreover, there is no form for ‘You came to me’ or ‘I came to you.’ In these respects, the one-and-ahalf agreement seen on nonactive (v-less) clauses in Southern Tiwa is perfectly parallel to the two-and-a-half agreement found in active clauses in the same language.28 28 There is a theory of agreement in ergative languages that says that T agrees twice and v does not agree at all (Bok-Bennema 1991, Bobaljik and Branigan 2006). In section 5.10.2.2, I adopt such a theory for Nez Perce. Combining that approach with the reasoning outlined here, one might expect that ergative languages of this type would also show a kind of one-and-a-half agreement, agreeing with their objects in number and gender only. In fact, person agreement is rather impoverished in Nez Perce in a way that might be attributable to the SCOPA: the language shows overt agreement with third person arguments only, and does not encode a contrast between first and second person agreement (see section 5.10.2.2 for examples and references). In Inuit, on the other hand, there is full person agreement for both the subject and the object of a transitive clause, so nothing akin to a PCC effect holds of T in that language. But Bittner (1994) shows that agreed-with objects in Inuit must take wide scope with respect to sentential operators (see also Bok-Bennema 1991). That could mean that T in Inuit has a second EPP feature, so that objects as well as subjects are attracted to Spec, TP. In that case, full person agreement with both arguments is expected (compare note 27).

102


In contrast, the passives of ditransitive verbs do not show agreement with a plural theme object in Nahuatl. At least, Launey (1981) does not mention the possibility, and there are no forms like (53) in his texts:29 (53)

*T-im-mac-o hu¯ehuèx¯olô. (Launey 1981:175) 2sS-pl-give-pass turkeys ‘You were given some turkeys.’ (OK is Ti-mac-o in x¯ochitl ‘You were given the flowers’, without plural im.)

(53) shows that im is not used to spell out an extra plural feature on T, nor is there any other morpheme that does so. Apparently T in Nahuatl is specified as probing once ((51ii)), whereas T in Southern Tiwa and v in Nahuatl are specified for probing twice ((51iii)). How many times a given functional head can agree thus varies independently both within and across languages. Of course we already knew that this was true with respect to whether a functional head agrees zero or one times: for example, v in Russian agrees zero times, in contrast to T in Russian and v in Chichewa, which agree once each. So (51) correctly captures the kind of head-by-head variation that one sees in this domain.30 Summing up this material, we see that, in a variety of languages, triadic verbs allow a limited degree of object agreement with their third arguments. Specifically, such verbs can agree with a third person object, sometimes reflecting differences in gender and number for that argument, but they cannot agree with a first or second person object. My theory offers an explanation for this phenomenon that relates it to the fact that adjectives can agree with an NP in number and gender but not in first or second person. The common theme is that first and second person agreement requires that the agreed-with NP merge directly with the agreement-bearing functional head. In ditransitive constructions, this sort of agreement is ruled out for the theme object because the specifier position of the functional head is already occupied by the higher goal object; in adjectival constructions it is ruled out by the inherent nature of adjectival projections, which do not take specifiers by definition. The reason why the agreed-with 29 This was evaluated by typing Launey’s prose Nahuatl texts into a computer file, searching it electronically for the relevant combinations of letters, and then doing a morphological analysis of the hits to see if any of them had the morphological structure in question. 30 The ungrammaticality of (53) also confirms that im really is an object agreement morpheme that spells out active v, not an agreement on some distinct third head like Appl (as in Adger and Harbour’s 2005 analysis of Kiowa). If im were agreement on Appl, one would expect to find it in passivized applicatives as well, contrary to fact. More generally, in Adger and Harbour’s view there is no reason to expect that agreement on the Appl head can only be agreement for number and gender but not for person.


103

argument does not merge with the functional category is different in the two structures, but the consequences for agreement are the same. 3.3.4 Agreement into an embedded clause So far the examples of nonlocal agreement with verbs have been examples in which something else occupies the specifier of the functional category that the agreed-with NP might otherwise have moved to. Thus, quirky dative subjects in Icelandic and Gujarati block lower arguments from moving to Spec, TP; goals, applied objects, and causees block lower arguments from moving to Spec, vP in Nahuatl, Southern Tiwa, and Shambala. In all such cases, the verb can agree with the lower argument in number and gender only, not in person. Another relevant kind of case is the long distance agreement (LDA) that verbs sometimes show with arguments of an embedded clause. A set of such cases has come to light in the recent literature, in which the matrix verb agrees with a constituent of the embedded clause which moves to the periphery of the embedded clause (either overtly or covertly) but cannot move out of that clause because of language-specific conditions on movement; see, for example, Polinsky and Potsdam 2001 on LDA in Tsez. Since there is clear evidence that the agreed-with NP is trapped in the embedded clause, we know that it does not move to the specifier position of the relevant FV associated with the matrix verb (although the clause that contains it might). The structures in question can be schematized abstractly as in (54), where FV can stand for either T or v. (54) a. FV verb b. FV verb c. *NPi FV

[CP C [TP . . . . NP . . . . ]] Agreement out by phase condition [CP NPi C [TP . . . ti . . .]] LDA agreement configuration verb [CP (ti ) C [TP . . . ti . . .]] * by conditions on movement

The SCOPA now predicts that the configuration in (54b) might allow agreement in number and gender, but should not allow agreement in first or second person. This prediction is consistent with the known facts. The well-known case of Tsez is not directly relevant to this prediction, for the simple reason that verbs in Tsez do not agree with an NP in person even in simple monoclausal structures. As a result, no difference between ordinary verbal agreement and long-distance verbal agreement would be observable in this language. Comparable data from LDA in the Nigerian language Lokaa are relevant, however (taken from my own research with Alex Iwara). Lokaa usually has subject-verb-object word order, but in gerundive constructions the object fronts to a left-peripheral specifier position (see Baker 2005 for detailed analysis). When this happens, the matrix verb agrees in gender with the fronted object of the gerundive verb, not with

104


the gerundive phrase as a whole. (55a) shows an example where the gerundive verb is intransitive, and the matrix verbs take the same noun class prefix as the gerundive verb, in agreement with it. (55b) is a contrasting example in which the gerundive verb is transitive. Its object ‘fish’ must front to come before the gerund, and this time the matrix verbs take the same noun class prefix as this object, not the noun class prefix of the nominalized verb: (55)

a. Ke-pa.a.la. ke-tum ke.-tawa. ger/5-fly 5S-be.very 5S-be.difficult ‘Flying is very difficult.’ b. [E.-sau ke.-de.i] e-tum e.-tawa. 7-fish ger/5-buy 7S-be.very 7S-be.difficult ‘Buying fish is very difficult.’ (* . . .ke-tum ke-tawa)

Now Lokaa has distinct first and second person subject agreement prefixes (see Iwara 1982 for a complete paradigm); (56) gives an example with the first person singular form. (56)

Ami n-tum n-dam. I 1sS-be.very 1sS-be.big ‘I am very big.’

The crucial question, then, is what happens when the object of a gerundive verb is first person singular and the gerund as a whole is in subject position. The distinctive prediction of my theory of agreement is that first person singular agreement on the matrix verbs should be ruled out, because the first person NP does not merge directly with T; only a larger phrase that contains it does. The ungrammaticality of (57) shows that this prediction is correct.31 (57)

*[Min ke-funna] n-tum n-tawa. me 5-surprise 1sS-be.very 1sS-be.difficult ‘Surprising me is very difficult.’

Another language in which this effect can be seen clearly is Basque. Etxepare (2006) documents the existence of LDA in number in certain substandard varieties of Basque. In particular, he shows that certain verbs that select for nonfinite clauses with a controlled subject permit the matrix auxiliary verb to show object agreement in number with the object of the lower verb. (58) reproduces two examples of this: 31 The partial agreement structure in which the matrix verbs show singular animate agreement o also seems to be out (see discussion below), as is agreement with the gerund as a whole. Alex Iwara (personal communication) says that speakers would simply avoid using a gerund in this case, using a subjunctive clause instead (‘It would be very difficult that one would surprise me.’).


105

gusta-tzen za-(izk)-io. a. Nobela erromantiko-ak irakur-tze-a novel romantic-abs.pl read-noml-abs.sg like-hab 3(p)A-aux-3sD ‘He/she likes to read romantic novels.’ (p. 1) b. Liburu-ak eros-te-a erabaki d-(it)-u. book-abs.pl buy-noml-abs.sg decide aux-3sE-3(p)A ‘He decided to buy books.’ (p. 33)

However, Etxepare also observes that this sort of LDA can only add extra number features to the auxiliary in the matrix clause; it cannot add person features. The examples in (59) contrast minimally with those in (58), but they are ungrammatical. (59)

bertan ikus-te-a gusta-tzen zatzaizkio. (p. 39) a. *Zu you.abs.sg there see-noml-abs.sg like-hab aux.2A.3sD ‘He/she likes to see you there.’ b. *Zu gonbida-tze-a baztertu zaituzte. (p. 39) you-abs.sg invite-noml-abs.sg refused aux.3pE.2sA ‘He/she refused to invite you.’

Basque then is another language that confirms the prediction that true LDA can only be agreement in number and gender. Along with Tsez, the other well-known cases of LDA come from Algonquian languages, including Bruening’s (2001) discussion of Passamaquoddy. At first glance, this language seems problematic for my prediction, because the argument of the embedded clause appears to trigger any sort of agreement on the matrix verb: (60)

nuhuw-ok muwinuw-ok keq kis-temu-htit. a. N-kosiciy-a-k 1-know.ta-dir-3.pl three-3.pl bear-3.pl what perf-eat-3.pl.conj ‘I know (of them) what the three bears ate.’ (p. 259) (number agreement) b. Kosiciy-ul kis-ankuweht-uwon atomupil-ol Piyel naka Susehp know.ta-1/2 perf-sell-2.conj car-inan.pl Piyel and Susehp mil-osk-opon-il. give-2.conj.inv-pret-part.pl (p. 266) (person agreement) ‘I know (of you) that you sold the cars Piyel and Sushep gave you.’

In (60a), the matrix verb ‘know’ shows plural object agreement with ‘three bears,’ the topicalized subject of its clausal complement. That example is consistent with my theory. (60b), however, looks problematic: here the matrix verb ‘know’ shows second person singular agreement with the subject of the embedded clause, contrary to my prediction.

106


Bruening (2001) shows, however, that there is an important difference between instances of LDA with third person, number, and animacy features such as (60a) and putative instances that involve first and second person, such as (60b). LDA with third person arguments is sensitive to islands. For example, the matrix verb in (61) cannot agree with ‘Piyel and Sushep’ in the embedded clause, because that NP is further embedded inside a relative clause. (61)

kis-ankuweht-uwon atomupil-ol Piyel naka Susehp *N-kosiciy-a-k 1-know.ta-dir-3.pl perf-sell-2.conj car-inan.pl Piyel and Susehp mil-osk-opon-il. give-2.conj.inv-pret-part.pl (p. 266) ‘I know (of them) that you sold the cars Piyel and Sushep gave you.’

This is more or less expected. Since ‘Piyel and Sushep’ is contained inside a complex NP it cannot move to the edge of the clausal complement of ‘know’ via topicalization (even covertly). Therefore the v associated with the matrix verb cannot access it, by the phase condition on agreement. The surprise is that the matrix verb can show +2 agreement with an argument that is generated inside a relative clause: (62)

Kosiciy-ul eli Susehp kis-ankuweht-aq atomupil mil-ot-opon. know.ta-1/2 C Susehp perf-sell-3.conj car give-2.conj-pret ‘I know of you that Susehp sold the car you gave him.’ (p. 278)

Bruening concludes from this sort of data that “LDA” with first and second person arguments is not the same phenomenon as LDA with third person arguments. The third person cases are true instances of LDA, analyzed just like Polinsky and Potsdam’s analysis of Tsez in (54), which I have been following. But the first and second person cases are instances of proleptic objects, in which a second internal argument is generated in the domain of the matrix verb ‘know’ and stands in a binding relationship (not a movement relationship) with some NP in the embedded clause. This extra argument of the matrix verb is very much like the phrase of you in the English gloss of (62). Since binding relationships are not restricted by island conditions, the example in (62) is possible. Since (61) is not possible, third person nominals cannot be base generated as objects in the matrix clause, for some reason. Bruening himself gives a rather complex explanation for this person asymmetry in Passamaquoddy. His account hinges on the person-sensitive voice system of Algonquian languages, in which first and second person objects trigger inverse forms. Bruening analyzes this inverse system as a voice alternation

Conclusion

107

that is similar to the passive. Then if LDA with first and second person nominals followed exactly the same derivation as third person nominals, it would require the first or second person NP to move from the Topic position at the edge of the lower clause to the Spec, TP position of the matrix clause, an instance of improper movement (A-bar movement feeding A-movement). My theory offers a simpler and more general account of the difference, not dependent on controversial properties of the Algonquian voice-and-agreement system. My theory of agreement states that an Agree relationship between T or v and something stuck in the lower Spec, CP can only be agreement for number and gender (which includes animacy and obviation in Algonquian), not for person. If one has full agreement with a +1 or +2 expression, that NP must be a true argument of the matrix clause, and hence able to move to Spec vP (or Spec TP).32 Therefore, Passamoquoddy also supports my theory, initial appearances notwithstanding. I predict that there cannot be full LDA with a first or second person argument, and in fact putative cases of LDA with such arguments turn out to have a different syntactic structure than those with third person arguments. 3.4

Conclusion

In the course of this chapter, we have found many reasons to believe that adjectival agreement and verbal agreement are two instances of fundamentally the 32 Bruening does, however, have an explanation for a fact that does not follow from my account – the fact that generating a third person argument as a proleptic object of the matrix verb seems not to be possible in Passamaquoddy (so (61) is bad under any analysis). I have no formal account for this; I can only appeal to a kind of economy principle, which says that the true LDA derivation blocks the proleptic object derivation when both are consistent with the superficial morphological facts. It is interesting to compare Bruening’s analysis of Passamaquoddy with Branigan and MacKenzie’s (2002) analysis of Innu-Aimun (IA). Branigan and MacKenzie’s description differs from Bruening’s on two points. First, they claim that constituents of the lower clause can move into the matrix clause in IA, unlike in Passamaquoddy. Second, they do not present any syntactic differences between LDA with third person NPs and LDA with first and second person NPs. If they are correct and IA really differs from Passamaquoddy in these respects, the two Algonquian languages constitute an interesting minimal pair that supports my theory. Given that IA allows NPs to move into the matrix clause, movement of a +1 or +2 NP into Spec, vP is possible, not blocked by language-specific constraints. When movement to this position takes place, object agreement in +1 or +2 with an NP that originated in the embedded clause is expected. (See also Bobaljik and Wurmbrand 2005:845–50, who show that (i) Itelmen matrix verbs can sometimes agree in person with an argument of an embedded infinitive, but (ii) this only happens when there is scopal evidence that the agreed-with argument has moved into the matrix clause.)

108


same phenomenon. Both adjectives and verbs can agree with a c-commanding NP, both can agree with a c-commanded NP, and both are restricted to gendernumber agreement when forced to agree at a distance. The only major difference is that verbal heads can also agree with NPs in first and second person when they merge directly with the NP, whereas this possibility is not available to adjectival heads. But the fact that this is not possible follows from the general theory of categories: adjectival projections by definition do not permit specifiers whereas verbal projections do.33 The comparison between verbs and adjectives also gives strong reason to think that the restrictions on person agreement are syntactic rather than morphological in nature. If it were only adjectives that could not agree in first or second person, one might think that some person feature was missing in their lexical entry or happened not to be spelled out by the morphological rules of the language. But in this chapter we have seen that verbs also sometimes fail to show person agreement, and for verbs this clearly cannot be attributed to general properties of the lexical category or its morphological paradigm. Rather, whether there is person agreement or not clearly depends on the 33 A very different sort of argument that adjectival agreement is distinct from verbal agreement can be found in Wechsler and Zlatić 2003:ch. 3. They point out that there are a few nouns in Serbian/Croatian that trigger feminine singular agreement on attributive adjectives and demonstratives but trigger neuter plural agreement on auxiliaries and participial verbs. One such noun is deca ‘children’: (i)

Ta dobr-a deca su doˇsl-a. (p. 5) that.f.sg good-f.sg children aux.3pS come.past.ptpl-n.pl ‘Those good children came.’

Wechsler and Zlatić conclude, following Pollard and Sag (1994), that nouns have two separate sets of agreement features: concord features, which are accessed by a relation of concord agreement that involves adjectives, and index features, which are accessed by a distinct relation of index agreement that involves verbs (and pronouns in discourse). If they are right about this, then there are two distinct kinds of agreement and not only one. However, I find two things striking about this case. The first is how very rare are nouns that have putative mismatches between concord features and index features; Wechsler and Zlatić (2003:50) list only four. The second is how similar all of these nouns look to another class of collective nouns in Serbian/Croatian (nouns that end in the morpheme ad), which Wechsler and Zlatić treat as feminine singular nouns that have collective meanings, and thus can trigger semantic/pragmatic agreement on the predicate (and on subsequent pronouns), just as nouns like committee, government and band do in some varieties of English (see section 1.5). The only difference reported is that the collective noun unuˇc-ad ‘grandchildren’ can trigger singular or plural agreement on the finite auxiliary, whereas deca ‘children’ can apparently only trigger plural agreement – a difference that Wechsler and Zlatić (2003:76) acknowledge to be mysterious even in their account. I therefore conclude that nouns like deca in (i) also illustrate a type of semantic/pragmatic agreement, and do not motivate a distinction between two kinds of grammatical agreement, one that involves verbs and another that involves adjectives.

Conclusion

109

syntactic structure of the clause – whether there is a dative subject or not, whether it is a monotransitive clause or a ditransitive clause, whether the agreedwith phrase is trapped in a lower clause or not. It is a strength of my theory that the syntactic characterization of when person agreement fails on verbs also explains the general lack of person agreement on adjectives and nouns. Before closing this chapter, I must mention an important residual issue that this material raises, first articulated to me by David Pesetsky (personal communication). The question is what happens to a structure when agreement in person with a +1 or +2 nominal is ruled out by the SCOPA? Descriptively speaking, there are two possible outcomes. One outcome is that the grammar might make do with partial agreement: the functional head in question agrees with the first or second person nominal as best it can and leaves it at that. This is always the result when the agreeing head is adjectival, as far as I know. Thus, Swahili, Hindi, Spanish, Arabic, Mayali, and Tariana all allow sentences like ‘I am tall’ and ‘You are good’, where the adjective agrees with the subject in gender and number but not person. Partial agreement also seems to be good enough for agreement with wh-expressions in Spec, CP in nonstandard English. Full person agreement with I is ruled out in this dialect, as discussed in section 3.3.1, but number agreement with we seems better, on a par with third person plural agreement, as shown in (63). (63)

a. ?the people who Clark are hoping will come b. I, who Clark is/*am hoping will come c. ?we who Clark are hoping will come

The second possible outcome when full agreement with a first or second person argument is impossible is that the syntax crashes, the structure in question being rendered ineffable. That seems to be the case for the double object constructions discussed in section 3.3.3. Thus, structures like ‘She sent me to him’ in Nahuatl, Southern Tiwa, and Shambala do not give rise to agreement in number and gender/animacy with the first person theme; rather they are unsayable (Mohawk), forcing some other structure to be used (e.g., one with the goal expressed as an oblique PP, as in (48) from Shambala). Long distance agreement with a first person object in a gerund construction also leads to ineffability in Lokaa (note 31). Finally, dative subject constructions with a first or second person object are reported to be ineffable in Chicasaw and Icelandic. An interesting theoretical question then arises as to why partial agreement with a first or second person is tolerated in some of these constructions and not others. Unfortunately, I have no definitive answer to offer at this point. I simply

110


note that the difference does not seem to correlate neatly with the verb–adjective distinction. While awaiting further insight into the matter, I assume that partial agreement is generally allowed as far as the syntax of agreement is concerned, and that the structures that crash do so because of other factors.34 34 The licensing needs of the null pronominal pro might be a relevant additional factor. It could be significant that the types of agreement that tolerate partial agreement (adjectives and verbs agreeing with a wh-phrase in Spec, CP) also do not license pro-drop. In contrast, the types of agreement that do not tolerate partial agreement typically do license pro-drop. This is true for agreement with subjects in Chicasaw and Lokaa, and for agreement with objects in Nahuatl, Southern Tiwa, Shambala, Basque, and the Algonquian languages. So the following generalization might be valid: (i)

F can show agreement with a +1 or +2 NP in number and gender but not +1 or +2 if and only if F would not license pro-drop of NP.

If this generalization holds up, it suggests that full agreement is required in some configurations not for the sake of agreement per se, but rather for the sake of the agreed-with NP, so that its features will be recoverable when pro-drop occurs. Of course, working this out more fully would require a better understanding of how agreement licenses pro-drop, and what the relationship is between an overt pronoun and its pro-dropped cousin. But the direction seems promising.

4


Most of the ingredients of the theory of agreement in the previous two chapters are either fairly standard or generalizations of notions that are fairly standard. The notable exception is the Structural Condition on Person Agreement (SCOPA), repeated in (1). (1)

F can agree with XP in +1 or +2 only if a projection of F merges with a +1 or +2 element and F projects.

This highly specific condition is not part of standard theory. Section 3.3 gave a range of evidence showing that this condition correctly characterizes when verbs can and cannot agree in person with their arguments, in addition to ruling out first and second person agreement on adjectives quite generally. But as it stands, (1) is merely a descriptive generalization. The purpose of this chapter is to explore the SCOPA more thoroughly. To accomplish this, I begin by investigating when person agreement is possible on categories other than nouns, verbs, and adjectives, confirming that the SCOPA is descriptively correct (section 4.1). I then go on to derive the SCOPA from more fundamental principles which go to the heart of what it is for any linguistic expression to be first or second person. The first step in this derivation is to distinguish the relationship of Agree that holds between a functional head and a nearby NP from the sort of agreement that holds between operators and the variables that they bind (section 4.2). I claim that first and second person agreement belongs properly to operator-variable agreement and not to Agree per se. It is not particularly surprising, then, that the two are subject to different formal conditions. The second step is to argue that all first and second person elements – including pronouns – must be bound by the closest operator of the relevant type (section 4.3). In this respect, first and second person elements differ from third person elements, which are not subject to any such local binding condition. The final step is to show that locality conditions in general apply to heads in a particularly strict fashion, such that any intervening head blocks a local relationship, regardless of its features (section 4.4). Putting these 111

112


intermediate results together gives the overall result that any head that comes between a first or second person head and the first or second person NP that it is dependent on blocks the person-licensing relationship – a condition that is equivalent to (1). The SCOPA is thus derived from a principle that regulates the distribution of first and second person pronouns plus a general law concerning how locality relationships apply to heads.

4.1

Person agreement and other categories

In previous chapters, I considered only the possibility of person agreement on the lexical categories – or, more accurately, on the functional categories that most immediately dominate those lexical categories. The primary result from chapter 2 was that the functional categories directly associated with verbs, Tense and v, can bear person agreement with subjects and objects, whereas the functional categories associated with nouns and adjectives cannot bear person agreement. Chapter 3 added the nuance that even Tense and v cannot agree with an NP in person if their projection does not merge directly with that NP. One thus finds only number and gender agreement with wh-phrases in Spec, CP in an English dialect, with lower NPs in oblique subject constructions in some languages, with second objects in double object constructions, and with phrases trapped inside an embedded clause in LDA constructions. Overall, person agreement is relatively fragile, and conditions need to be just right for it to take place, as expressed in the SCOPA. When we widen the picture to consider a fuller range of categories, we find some other instances of person agreement. These arise in the range of environments that the SCOPA predicts they should – with one very instructive exception. 4.1.1 Adpositions Consider first the category of adposition. The traditional generative view since Jackendoff (1977) has been that this is a lexical category, but Baker (2003a:appendix) argues that it is actually a functional category. As such, it is a possible host for agreement. Moreover, adpositions normally select NP/DP complements, which can bear ϕ-features, including the features +1 and +2. Since adpositions merge directly with such complements, the SCOPA predicts that adpositions should be like verbs and unlike the other lexical categories in manifesting person agreement. Quite a few languages confirm this expectation, including the examples in (2) from the Mayan language Jacaltec. (2a)


113

shows agreement with the third person object of the preposition; (2b) shows contrasting first person agreement on a similar preposition. (2)

a. y-ul te’ n¨ ah (Craig 1977:110) 3E-in the house ‘in the house’ b. X-Ø-colwa ix w-i¨n. (Craig 1977:102) asp-3A-help she 1sE-to ‘She gave me a hand.’

(3) shows similar examples from postpositions in Abaza (O’Herin 2002:54): (3)

a. awə y a-mʃj taxj that 3s.i-after ‘after that’ b. (sara) s-pngə I 1s-at ‘at my house, by me’

The SCOPA does, however, predict that Ps will not be able to agree in person more remotely with (for example) the subject of its predication. This follows as long as PPs do not have nominal specifiers, and only become predicates by adjoining to a verbal projection (see Baker 2003a:appendix). Since the P does not merge directly with the subject of the predication, P cannot agree in person with that subject (see also Baker 1996:401–4). (4) shows that this is true for Nahuatl, a language in which predicates of all other categories do seem to agree in person with their subjects (see section 2.5.1). (4)

Petla-pan icpal-pan ni-ca’. (Launey 1981:384) mat-on chair-on 1sS-be ‘I am on a mat, on a chair.’ (i.e. I govern.) Not: *Ni-petla-pan, n-icpal-pan ni-ca’.

Adpositions are interesting because they provide our first examples of a category agreeing in person with its complement rather than its specifier. Tense, v, and FA all take nonnominal complements that do not have ϕ-features. Therefore if they also merge with something that does have ϕ-features – as Tense and v often do – that is a second merge, creating a specifier. As a result, person agreement is in practice usually agreement with a specifier. In part because of facts like these, generative theories of the early 1990s attributed a special role to the specifier–head relationship in the theory of agreement, claiming that full agreement occurred only in specifier–head relationships (see Koopman 2006 for a recent defense). But the existence of person agreement inside PPs

114


weighs against this, showing that one does find full person agreement between heads and complements whenever conditions are right for it. This suggests that Minimalist sensibilities are correct, in that there is no fundamental theoretical difference between complements and specifiers (Chomsky 1995). It is only a matter of what merges with what that is important for the theory of agreement, as in other domains.1 4.1.2 Determiners Next, let us consider determiners (Ds) and similar functional heads in the nominal system. The expectations of my theory for this category are similar to those just reviewed for the category P: since D is a functional category that merges directly with a complement that has ϕ-features, it should be able to agree with that complement. And of course articles, demonstratives, and quantifiers do agree with their NP complements in many languages, including Spanish: (5)

a. b. c. d.

el libro los libros la mesa las mesas

(the book.m.sg) (the books.m.pl) (the table.f.sg) (the tables.f.pl)

The definite article agrees with its complement only in number and gender, whereas the SCOPA predicts that person agreement should be possible as well. But this simply reflects the fact that common nouns in Spanish are always third person, never first or second person. Therefore, the complement of this kind of determiner never has a +1 or +2 feature for the determiner to agree with. I assume that (5) is a case of total agreement, not partial agreement, but the person feature is trivial, because articles and quantifiers do not usually take pronouns as arguments (*the we, *every I, *some you). Some Bantu languages have D-like heads that can bear person agreement as well as number and gender agreement. An example is onke ‘all’ in Zulu:2 1 A proponent of Spec–head agreement can, of course, insist that the first person pronouns move to Spec, PP or some other specifier in the extended projection of P in examples like (2b) and (3b). I do not attempt to construct arguments that this is not the case. I simply point out that my theory does not need this extra assumption, for which there is no obvious evidence in the languages cited. 2 Wechsler and Zlatić (2003:15) mention the similar case of ote ‘all’ in Swahili, pointed out to them by Manfred Krifka. They must treat these cases as exceptional: they are instances of concord agreement (as opposed to index agreement) that nevertheless show person, a feature not normally included in the concord feature bundle. Since my theory does not distinguish concord from true agreement, these examples are not exceptional but expected. On the fact that the agreed-with complement of onke ‘all’ comes before it in (6), see section 5.5.1.

Person agreement and other categories (6)

115

Thina s-onke si-fik-ile. (Doke 1963:94) we 1p-all 1pS-arrive-perf ‘We have all arrived.’ (compare: b-onke (cl2-all) ‘all of them’; izinkomo z-onke (cattle.5.pl 5.pl-all) ‘all the cattle’)

This makes sense given that onke ‘all’ is different from most simple determiners in that it selects DP complements, including pronouns, as well as NP complements. Selectional properties like these need to be stated in the grammar anyway; compare English all (the) men vs. every (*the) man. Once it is stipulated that ‘all’ can select a first person pronoun then it follows immediately from the SCOPA that ‘all’ can agree in person with that pronoun. Other determinerlike elements that can inflect for person as well as gender and number in Zulu are dwa ‘only, alone’ and bili ‘both’ (Doke 1963:94–6). Another sort of determiner that clearly shows person agreement in many languages is the possessive determiner, which merges with a specifier (the semantic possessor) as well as a complement to form structures like [DP John DPoss [NP book]] (Abney 1987). There are in general no selectional restrictions on the features of the possessor; in particular, first and second person pronouns can perfectly well function as possessors. Given this, the SCOPA permits D to show person agreement with the possessor. And indeed possessive determiners do agree with the possessor in many languages. (7) shows particularly clear examples from Mapudungun, where the possessive determiner is realized as a separate word from the noun (note that the possessor itself is often pro-dropped).3 (7)

a. mi füchá ruka (Smeets 1989:129–30) 2s.poss big house ‘your big house’ b. yiñ füchá ruka 1p.poss big house ‘our big house’

3 A question that arises here is why doesn’t the complement NP of the possessive determiner block it from agreeing with its specifier by the intervention condition, given that the complement also has ϕ-features. There are two possibilities. One is that the possessive determiner actually agrees with its complement as well as with its specifier in (7), but this is not spelled out since Mapudungun does not have gender distinctions and rarely spells out number agreement. The second possibility is that the intervention condition needs to be defined in terms of m-command rather than c-command, so that the specifier and the complement come out as equally close to the head of the projection (see note 20 to chapter 2). I leave open which of these is correct.

116

Explaining the restriction on person agreement c. (inché) n˜ i ruka I 1s.poss house ‘my house’

In many languages, nouns themselves are said to agree with their possessors. I am committed to the view that this commonplace description is not quite accurate, and it is really always the possessive determiner that agrees with the possessor. This can, however, be disguised on the surface if the noun incorporates into or merges with the determiner to form a single word at PF, as is often the case. Nahuatl is an instructive case in point. Possessive agreement is realized as a prefix attached to the noun root in this language: (8)

a. no-te-uh (compare: te-tl) (Launey 1981:90) 1sP-rock-poss rock-nsf ‘my rock(s)’ ‘a/the rock’ b. am-e-uh (compare: e-tl) 2pP-bean-poss bean-nsf ‘your bean(s)’ ‘(the) bean(s)’

However, there is also a different suffix on the noun when it is possessed as compared to when it is not possessed (uh/Ø vs. tl and its allomorphs). I take this suffix to be the possessive determiner, with the noun root having incorporated into it. As a result, the agreement with the possessor – which is clearly associated with the D node, not the N node in languages like Mapudungun where the two remain separate – comes to appear in the same word as the noun in Nahuatl, where N and D are not separate. This is perfectly parallel to the familiar fact that subject agreement is on Tense rather than the verb, but it often comes to appear on the verb at PF because Tense and the verb combine either by head movement or by PF merger. Many other languages, like Mohawk, do not show any overt difference between possessor suffixes and absolute suffixes (see (9)), but I take them to be syntactically like Nahuatl in these respects. (9)

a. ake-núhs-a’ (cf. ka-núhs-a’) 1sP-house-nsf Ns-house-nsf ‘my house’ ‘a/the house.’ b. sa-sahét-a’ (cf. o-sahét-a’) 2sP-bean-nsf Ns-bean-nsf ‘your beans’ ‘(the) beans’

4.1.3 Degree heads So far, we have seen that adpositions and determiners can show agreement in first and second person features, much as the verbal heads T and v can. In contrast,


117

the SCOPA makes the opposite prediction for degree heads, the functional heads that appear with adjectives. Degree heads appear above AP (and FA P) and below PredP to form predicative structures like [PredP NP Pred [DegP Deg [(FA ) [AP]]]] (see Baker 2003a:214). Since the complement of a degree head is an adjective phrase, and adjective phrases do not have ϕ-features, they will not trigger +1 or +2 agreement on a Degree head. Moreover, inasmuch as Degree is an adjectival functional head, it is like FA and A in not licensing a specifier. Therefore, Degree heads cannot show +1 or +2 agreement with a specifier either. Degree heads do not merge directly with any ϕ-feature bearing element; therefore the SCOPA implies that they should never show first or second person agreement. I do not have much crosslinguistic evidence that bears on this prediction. Most of the languages that I know do not have clear instances of the Degree category, and in some of those that do, the degree heads do not happen to be probes for agreement (e.g., Spanish muy ‘very’ and tan ‘so,’ which do not inflect at all). But what little evidence I do have supports the prediction. In Russian, for example, the degree head tak ‘so’ does manifest gender and number agreement with the subject, as shown in (10a). This shows that it is a probe for agreement. But tak does not have a distinct first person form, as shown in (10b) (Nerea Madariaga, personal communication). (10)

a. Ivan tak-oj xoroˇs-ij; Nadja tak-aja xoroˇs-aja. Ivan.m.sg so-m.sg good-m.sg Nadia so-f.sg good-f.sg ‘Ivan is so good; Nadia is so good.’ b. Ja tak-oj xoroˇs-ij. I so-m.sg good-m.sg ‘I am so good (male speaker).’

This confirms that the degree head cannot agree in person at a distance, as expected. The SCOPA thus correctly predicts that P and D can bear person agreement morphology, but Degree cannot. 4.1.4 Complementizers: a favorable case The last well-established type of functional category to discuss is complementizer and complementizer-like heads. The complement of C is TP, which does not have ϕ-features. Thus, C is expected to show first or second person agreement if and only if there is an operator in Spec, CP that has these features. In all other cases, the C should either agree in number and gender only, or not agree at all. Complementizers in English, the Romance languages, and most others take the no-agreement option, which is trivially consistent with my theory.

118


Some other languages do have inflected complementizer-like heads, however. One example is the particle that introduces a relative clause in the Niger-Congo language Lokaa. This agrees in number and gender with the NP that the relative clause modifies: (11)

a. Y`ı-sòwá. f-á. u´ b`ı o´. o´. -dè.?i (my field notes) 12-pot 12-rel Ubi 1S-buy ‘the pot that Ubi bought’ b. kà-kó.o` k-á. e´ -fém e´ -j´ı´ı 14-pig 14-rel 7-crocodile 7S-eat ‘the pig that the crocodile ate’ c. l`ı-kó.o` j-á. m-fém n-j´ı´ı 11-pig 11-rel 8-crocodile 8S-eat ‘the pigs that the crocodiles ate’

Relative clauses can also be adjoined to pronouns in Lokaa. When this happens, the complementizer agrees with the pronoun in number and gender, but not in person. Thus, the first person pronoun in (12c) triggers the same class 1 (animate, singular) agreement on the relative complementizer as the third person pronoun in (12a). (12b) contains a first person plural pronoun; this triggers plural marking on the complementizer (class 2), but not person marking. (12)

a. Si: eá. w-á. o-bi liman o:-yeni. (Iwara 1982) despise him 1-rel 1S-not money 1S.neg-have ‘Despise him who has no money.’ b. Si: moon b-á. yo-bi liman yo:-yeni. despise us 2-rel 1pS-not money 1pS.neg-have ‘Despise us who have no money.’ c. ami w-á. m-biila I 1-rel 1sS-be.dark ‘I who am dark skinned’

These facts follow from the SCOPA as long as we can convince ourselves that there is no relative operator that bears first person features in Spec, RelP in (12b) or (12c). If there were, we would expect first person agreement on C with that operator to be possible. In fact, there probably is a relative operator in these constructions, for standard reasons having to do with island phenomena and other diagnostics of A-bar movement relationships. Moreover, the relative operator probably does count as being first person: notice that it triggers first person subject agreement on the verbs inside the relative clause in (12b) and (12c). However, the relative operator is probably not in Spec, RelP. The category


119

I have marked as REL in (11) and (12) is really an associative head, also found in noun-noun modification constructions in which there is no operator, such as (13). (13)

o´. :-bòl; l`ı-kó.o` j-á o´. :-bòl (fieldnotes) kà-kóo` k-á 14-pig 14-rel 1-chief 11-pigs 11-rel 1-chief ‘the chief’s pig’ ‘the chief’s pigs’

(13) suggests that the relative particle does not have a +wh-feature that triggers operator movement. Assuming that the very same particle is in the relative clauses in (11) and (12), then there is presumably no operator movement to Spec, RelP there either. The operator movement must be triggered by a distinct head, which we may call C, giving a structure like (14) for the example in (12c). DP

(14)

DPi

RelP

I Rel CP [1,sg,cl1] [sg,cl1] [*1] Opi [1,sg,cl1] C

C’ TP

T’ Ø ti [1,sg,cl1] T VP [1,sg,cl1] be.dark

Given (14), the SCOPA correctly describes the fact that the relative particle can agree with the head of the relative clause (or the operator below it) in number and gender but not in person. Overall, then, we see that there is much evidence for the SCOPA as a descriptive generalization that characterizes when agreement in person features is possible. The categories T, v, P, and D can show person agreement with their complements or specifiers; the categories FA , Degree, and certain complementizer-like particles cannot show person agreement because they do not have specifiers and their complements do not have ϕ-features. 4.1.5 Complementizers: problematic cases Before I go on to derive the SCOPA from more general considerations, I must acknowledge that there are other configurations in which complementizers show

120


a surprising kind of agreement, even in first and second person features. I know of at least two. One well-known case is the complementizer agreement found with subjects in some varieties of Continental West Germanic. (15) shows an unproblematic case where the complementizer agrees with the lower subject in number (from Carstens 2003:393). (15)

die boeken te diere zyn. (West Flemish) Kvinden dan I.find that.pl the books too expensive are ‘I find that those books are too expensive.’

But the complementizer can also agree with first or second person subjects in some of these varieties, even though the subjects do not raise to Spec, CP (Carstens 2003:393): (16)

Kpeinzen dank (ik) morgen goa-n. (West Flemish) I.think that.1s I tomorrow go-1sS ‘I think that I’ll go tomorrow.’

This agreement is unexpected given the SCOPA. An even more striking problem for the SCOPA is posed by another kind of complementizer agreement in the Niger-Congo languages Lokaa and Kinande. In these languages, the complement of a verb of speech is headed by a complementizer that seems to agree with the matrix subject. The complementizer can even agree with the matrix subject in person, as shown by (18b) from Kinande. (17)

(18)

a. Ub`ı o´. -jà.y`ı o` -b´ı nè. yá-yàa` máa` . (Lokaa) Ubi 1S-say 1-that fut 2S-come here ‘Ubi said that they will come here.’ b. Yá-jà.y yá-b´ı/*ó.-b´ı Ub`ı nè. o´. -yàa` máa` . 2S-say 2-that/*1-that Ubi fut 1S-come here ‘They said that Ubi will come here.’ a. Mo-ba-nyi-bw-ire ba-ti Kambale mo-a-gul-ire eritunda. (Kinande) aff-2S-1sO-tell-ext 2-that Kambale aff-1S-buy-ext fruit.5 ‘They told me that Kambale bought fruit.’ Kambale in-di a-gul-e amatunda. b. Mo-n-a-layir-ire aff-1sS-t-convince-ext Kambale.1 1s-that 1S-buy-sbjn fruits.6 ‘I convinced Kambale that he should buy fruits.’

So there are attestations of apparently nonlocal agreement in person on complementizers, both with the complementizer probing downward (West Flemish) and with the complementizer probing upward (Kinande). These examples cease to be problematic if there is a first person operator in Spec, CP in (16) and (18b). In section 4.3, I present evidence that there are such operators, not only in West Flemish and Kinande, but more generally. In

Operator-variable agreement and Agree

121

fact, these operators turn out to be essential to defining what it means for any linguistic element – pronoun or agreeing head – to be first or second person, providing the key to a deeper understanding of the SCOPA. Let us then turn to the project of understanding the fundamental nature of first and second person elements, returning to these cases of unexpected person agreement on C in the course of the discussion. 4.2


The first step is to acknowledge that there is at least one other type of agreement at work in natural language, apart from the relationship between functional heads and nearby NPs that I have been studying throughout this book. This second sort of agreement is the not-so-often-discussed agreement between an operator and the variable that it binds.4 Operator–variable agreement is illustrated in (19). (19)

a. Every boyk hopes that hek (*shek , *theyk , *Ik ) will pass the test. b. Every girlk hopes that someone will hire herk (*himk , *themk , *mek ). c. Only the Yankeesk think that theyk (*hek , *wek ) will win the championship.

When a pronoun is interpreted as being bound by a masculine singular quantifier like every boy, the pronoun must be realized as a masculine singular form he, him, or his, as in (19a). In contrast, when a pronoun is bound by a feminine singular quantifier like every girl, it must be a feminine singular form (her in (19b)), and when it is bound by a plural quantifier like only the Yankees, it must be third person plural (they in (19c)). This then is also a kind of agreement in a broad sense. Operator–variable agreement is clearly a somewhat different phenomenon from the agreement that applies to functional heads, however, because it does not obey the same formal restrictions. For example, we know that a functional head can agree with an NP only if the two are contained in the same tensed CP (the phase condition). That constraint does not apply to operator–variable agreement: in all of the examples in (19) the bound pronoun is properly contained in the finite embedded CP, a phase that does not contain the quantified 4 I thank Ken Safir (personal communication) for calling my attention to the phenomenon of operator–variable agreement, and for suggesting that I couch the ideas developed in this chapter in those terms. He is not to be blamed, however, for how I have acted on his suggestion. The agreement between an operator and a bound pronoun may need to be distinguished from the agreement that holds between any pronoun and its antecedent in the discourse. The latter is probably an instance of semantic/pragmatic agreement of the type put aside in this book (see section 1.5).

122


antecedent. Nevertheless, agreement between the pronoun and its binder is not only possible, but required. Second, we know that in English and many other languages a functional head can agree with an NP only if a case–valuation relationship holds between them (the activity condition; see section 5.3 for a new formulation). But no such constraint restricts operator–variable agreement. For example, the variable in (19b) is in accusative case and the quantified expression that binds it is in nominative case. Nevertheless, ϕ-feature agreement takes place. This kind of agreement, then, is not dependent on case-valuation the way that other forms of agreement are. Third, a functional head can agree with a given NP Y only if there is no other NP Z that is closer to the functional head than Y is (the intervention condition). (20) shows that this condition also does not apply to operatorvariable agreement. (20)

Every girlk told every boy about herk troubles with herk parents.

The two pronouns her in (20) agree with the subject every girl in the feature feminine despite the fact that there is a closer quantified NP every boy in the object position that has a different value for the same feature (masculine). Agreement is clearly with the NP that binds the pronoun, regardless of its syntactic position, for this type of agreement. That is quite different from agreement on functional heads, where relative syntactic position is paramount. Even the c-command condition does not apply to operator–variable agreement, since there are examples like (21) in which the quantified NP does not c-command the pronoun that is referentially dependent on it (see Safir 2004:34). (21)

Someone in every cityk loves itsk (*theirk , *hisk ) weather.

In short, none of the syntactic conditions that are characteristic of the Agree relation between a functional head and an agreed-with NP apply to the agreement relation between a variable and the operator that binds it. The two must be fundamentally different relationships, although both result in the features that are inherent on one expression also being realized on another expression. I state the second kind of agreement as in (22). (22)

If variable X is referentially dependent on operator Y (directly or indirectly), then X has the same ϕ-features as Y.

The qualification “directly or indirectly” in (22) is motivated by examples like (23). Here the first pronoun in the embedded clause is directly dependent on the matrix subject and agrees with it in ϕ-features (singular, masculine) in accordance with (22). The second pronoun his also agrees in ϕ-features. It is probably


123

not directly dependent on the matrix subject, but rather takes the embedded subject he as its immediate antecedent. But it is dependent on something that is itself dependent on the quantified matrix subject. This indirect dependency also involves agreement in ϕ-features. (23)

Every boyk says that hek finished hisk (*herk , *theirk , *ourk ) homework.

More precisely, the term “referentially dependent” in (22) can be defined inductively: X is referentially dependent on Y if (i) X is immediately referentially dependent on Y, or (ii) there is some Z such that X is immediately referentially dependent on Z and Z is referentially dependent on Y. (The equivalence of direct and indirect dependencies plays a role in what follows.) Exactly which ϕ-features participate in operator–variable agreement? The examples that I have used so far illustrate that number features (it vs. them) and gender features (he vs. she) are involved. But first and second person features also participate in this kind of agreement, as shown by the examples in (24), a type of example that has been discussed in unpublished work by Irene Heim (see also Kratzer 1998, von Stechow 2003, Schlenker 2005, and others). (24)

a. Only Ik finished myk (*hisk , *herk , *ourk ) homework. (For x = I, x finished x’s homework; For all x, x = I, not: x finished x’s homework.) b. All of youk are hoping that youk (*Ik , *shek , *theyk ) will win the single’s title. (For all x, x one of you, x hopes x will win the single’s title.)

In these examples, the pronoun can be interpreted as a variable bound by the highest subject if and only if it agrees with that subject in person as well as in number. So person features are full participants in the system of operator– variable agreement. Given this, let us adopt the conjecture in (25). (25)

Agree never puts the features +1 or +2 on a head; +1 and +2 features are always the result of operator–variable agreement.

In other words, I claim that first and second person features are only involved in the operator–variable agreement system. The sort of agreement that applies to ordinary functional heads like T and v and FA and P is always “partial” agreement, initially copying number and gender values only. How then can there ever be first or second person features on T or v? The answer, I claim, is that Agree creates a context where operator–variable agreement can apply, as expressed in (26).

124 (26)

Explaining the restriction on person agreement If F agrees with XP, then F counts as a variable that referentially depends on XP.

Therefore, in addition to copying number and gender features from XP to F, Agree makes F dependent on XP. Many linguists have had the intuition that there is something pronominal about agreement morphemes. For example, there is no doubt that agreement often makes use of the same stock of features as independent pronouns do, and the two are often historically related (see, for example, Givó.n 1976, Bresnan and Mchombo 1987, and references cited there). (26) is a way of capturing formally an aspect of this intuition. Then if an agreeing functional head F is dependent on the XP it agrees with, and if that XP is a pronoun that is dependent on some higher operator Z, then F is (indirectly) dependent on Z and must agree with it by (22). If Z is first or second person, then F will also be first or second person. That is how an agreeing head can come to have a first or second person feature, not directly as a result of Agree, but indirectly as a result of (22) and (26). What is gained by this shift in perspective? Only this: now we have the beginnings of an idea as to why agreement in first and second person might pattern differently from agreement in other features, since it is really part of another kind of agreement. Furthermore, attributing first and second person agreement to operator–variable agreement gives us an idea of what we need to look at next – the special operators that bind first and second person pronouns.

4.3

A locality condition on first and second person variables

A deep answer to the question of why first and second person features are subject to a stricter condition on agreement than other features would be one that related this fact to other, more basic truths about first and second person elements. Let us consider then the behavior of first and second person pronouns, taking them to be more basic bearers of the features +1 and +2 than agreeing functional heads. 4.3.1 The Person Licensing Condition What is the most basic truth about first and second person pronouns? At first glance, it is the idea that (putting aside direct quotation) first person pronouns always include the person who spoke (or wrote) the sentence in their reference, and second person pronouns always include the person to whom the sentence was addressed. This received wisdom has been influentially articulated in the work of the philosopher David Kaplan (1989). But the recent literature has


125

brought to light two kinds of counterexamples to this simple characterization, which suggest that there is a more specific syntactic licensing condition at work. First, Kayne’s (2000:154) example in (27) shows a situation in which a first person pronoun cannot refer to the speaker; this is ruled out when the subject of the sentence is a third person noun phrase that itself refers to the speaker. The pronoun me cannot have any interpretation in this example, and is simply ungrammatical. (27)

The man who is talking to you wants you to give him/*me some money.

Conversely, Schlenker’s (2003) example in (28) from Amharic shows a situation in which a first person pronoun (the pro-dropped subject of the embedded clause) can be interpreted as someone other than the speaker – namely the subject of the matrix clause. (28)

ˇon

ˇə gna nə -ñn˜ y l-all. John hero be.pf-1sO 3m.say-aux.3m ‘John says that he is a hero.’ (lit. ‘John says that I be a hero.’)

(Note that Schlenker reviews evidence that the embedded clause in (28) is not a direct quotation.) However, a first person pronoun in Amharic can only refer to someone other than the speaker in a very specific situation, when it is embedded in the complement of the verb ‘to say.’ Thus, there is need for a nontrivial grammatical condition that regulates the occurrence and interpretation of first and second person pronouns. Building on important insights of Schlenker (2003, 2005), Sigurð sson (2005), and Anand and Nevins (2004), I propose that there are special empty categories introduced at the CP level which designate the speaker and the addressee of the sentence, as stated in (29).5 (29)

a. All matrix clauses and certain embedded clauses have two special null arguments generated within the CP projection, one designated S (for speaker) and the other A (for addressee).

5 The specific syntax-oriented implementation that I adopt is most like “Theory II” of Schlenker 2005, which Schlenker attributes particularly to his dissertation, unpublished work by Irene Heim, and von Stechow 2003. Another precedent for this idea is the logophoric operators associated with Spec, CP, which lie at the heart of syntactic explanations of logophoricity in West African languages in Koopman and Sportiche 1989, Adesola 2004, and related work. See Safir 2005 for a dissenting view, on which first and second person pronouns are not bound by logophoric-like operators. For clarification of the relationship between first and second person licensing and the licensing of logophoric pronouns, see section 4.3.3.

126

Explaining the restriction on person agreement b. In the absence of an overriding control relationship, S designates the person who produced the CP and A designates the person who the CP was addressed to.

The idea is that certain key aspects of the “point of view” from which the clause is interpreted are expressed by syntactically represented elements.6 All uses of a first person pronoun must then be interpreted by being bound by the S operator, and all uses of a second person pronoun must be interpreted by being bound by the A operator, in accordance with (30). (30)

The Person Licensing Condition (PLC) a. A DP/NP is first person only if it is locally bound by the closest c-commanding S or by another element that is first person. b. A DP/NP is second person only if it is locally bound by the closest c-commanding A or by another element that is itself second person. c. Otherwise, a DP/NP is third person.

One minor pay-off of the PLC is that it explains the fact that ordinary nonpronominal NPs are never first or second person, even when they refer to the speaker or hearer. This can be seen in (31), where ordinary DP subjects cannot trigger distinctive first or second person agreement on the verb be. (31)

a. The man who is talking to you is/*am hoping to get some money. b. Sorry honey, but Daddy is/*am too tired to play with you tonight. c. Your honor was/*were misinformed by the counsel for the defense.

It is a general fact that pronouns can be bound by operators in, for example, resumptive pronoun constructions, but lexical NPs generally cannot be (unless they contain a pronominal element; see Aoun and Choueiri 2000 for data and discussion): (32)

a. ?Johnk , whok we all wonder whether hek will actually show up, . . . b. *Johnk , whok we all wonder whether the boyk will actually show up, . . .

Now if lexical NPs cannot be bound by operators ((32b)) and expressions can only be first or second person if they are bound by a special operator ((30a–b)), then it follows that lexical NPs cannot be first or second person. They always fall under the elsewhere case in (30c). This explains the facts in (31).7 6 There may be other such elements too, related to the location where the CP was spoken and the time at which it was spoken, relevant to the interpretation of deictic expressions like now and here. I leave this matter open. Unlike Schlenker (2003), I do not package all of these parameters together as a single semantic operator, because there are contexts in which first person shifts but second person does not in Slave (see the discussion of (44)). 7 Unlike English and French, Spanish allows sentences like Las mujeres trabajamos mucho (the women work.1pS much) ‘We women work a lot.’ I assume that the possibility of such a sentence


127

The PLC might also be used to explain the fact that a first person pronoun cannot refer to the speaker in examples like Kayne’s (27), repeated as (33). (33)

The man who is talking to you wants you to give him/*me some money.

As a matrix clause, the sentence as a whole has an S operator that refers to the speaker. What is special about this example is that it also has a matrix subject that is a definite description that refers to the speaker without being dependent on S. Suppose that we represent this fact by giving both S and the matrix subject the same index i. Now suppose that we want an interpretation in which a pronoun in the indirect object position of the embedded clause also refers to the same individual i. If it is assigned this index, then it is bound by S, but it is not locally bound by S; the matrix subject is a closer NP that bears the same index. Therefore, the pronoun in question cannot be first person according to (30a); rather, it must be third person by default:8 (34)

[CP Si [TP [NPi The man who is talking to you] wants you to give himi /*mei money]]

Of course if the matrix subject refers not to the speaker, but rather to some other guy who happens to be talking to the addressee at the same time, then using me in the embedded clause to refer to the speaker is grammatical and using him would be at best very strange. In that case, the matrix subject has a different index, say k, and S is the closest binder of the pronoun. Then first person features are licensed on the pronoun by (30a). Moreover, if the indexing remains as in (34) but the matrix subject is replaced by I, then too a first person pronoun is licensed in the lower clause: (35)

[CP Si [TP [NPi I] want you to give mei /*himi some money]]

is related to the fact that Spanish is a pro-drop language whereas English and French are not. Thus, one need not say that the verb ‘work’ is agreeing with a first person feature borne by the DP ‘the women’ in Spanish. Rather, we can say that the verb ‘work’ agrees with a null first person pronoun ‘we’, and ‘the women’ is a kind of topic or adjunct loosely related to the sentence by some kind of aboutness relation. 8 Raffaella Zanuttini (personal communication) points out to me that imperatives are different in this respect: (i) is fine, with a proper name that refers to the addressee apparently binding a second person pronoun. (i)

Nicholas Baker wash your (*his) hands!

A possible analysis of this is that imperatives, as a consequence of their special semantics and pragmatics, license an A operator that is lower in the clause, in the TP space rather than at the CP level. If that is correct, then the subject in (i) does not intervene between the A operator and the genitive pronoun, and the pronoun can have second person features.

128


(35) is like (34) in that the embedded pronoun cannot be anchored directly to the S at the head of the matrix clause, there being a closer antecedent. But unlike in (34), the closer antecedent in (35) is a first person pronoun which is itself locally bound to the S in the proper way. This is enough to license a first person feature on the embedded pronoun. The intuition is that first and second person pronouns need to be bound by the designated operator, either directly, or else indirectly through a sequence of local binding relationships to a pronoun that is so bound. On this analysis, (34) illustrates a kind of strong crossover effect that arises with first person pronouns. The ungrammaticality of (34) is thus akin to the ungrammaticality of (36), with the indexing indicated. (36)

*[CP Whoi does [TP [NPi he] want you to give ti some money]] (compare: [CP Whoi C [TP ti wants you to give [NPi him] some money]])

The trace in (36) must be locally bound by the operator in Spec, CP, and this is violated if the matrix subject is also interpreted as having this index. This, then, is a significant similarity between first and second person pronouns and other, less-controversial operator-variable structures, supporting the treatment of first and second person pronouns given in (29) and (30). This analysis of (33) raises some serious theoretical issues concerning indices and their interpretations, which are largely irrelevant to problems of agreement, so I do not push it strongly here. But I find the facts suggestive.9

9 In particular, one has to say that the DP in matrix subject position is not referentially dependent on the S operator (since it is not first person; see (31a)), but it is nevertheless coindexed with it in whatever sense is needed to create the strong crossover effect in (34). This is a very delicate distinction – perhaps ultimately not coherent. My diagnosis of (34) as a type of strong crossover predicts that the phenomenon should be sensitive to c-command: using a third person DP to refer to the speaker should only prevent a first person pronoun in the same sentence from referring to the speaker when the third person DP c-commands the pronoun, therefore counting as its local binder. I believe that this prediction is correct, although natural-sounding sentences are somewhat difficult to construct, and not all speakers agree. I consider (iia), in which Daddy does not c-command I, to be more or less possible, whereas (iib), where Daddy c-commands I is as bad as (34). (ii)

a. (Sk ) Because Daddyk forgot something at the office, hek /I?k have to go back there. b. (Sk ) Daddyk has to go back to the office because hek /I*k forgot something there.

Similarly, I find the first person pronoun to be marginally possible in (iiia) but completely out in (iiib). (iii)

a. (Sk ) This old picture of Daddyk shows himk /me?k with long hair. b. (Sk ) Daddyk finally showed hisk /my*k boss the new contract today.


129

4.3.2

Locality and shifted interpretations of first and second person pronouns Consider next how (29) and (30) account for the fact that a first person pronoun can refer to someone other than the speaker of the sentence as a whole in Schlenker’s Amharic example shown in (37) (repeated from (28)). (37)

ˇon

ˇə gna nə -ñn˜ y l-all. John hero be.pf-1sS 3m.say.aux.3m ‘John says that he is a hero.’ (lit. ‘John says that I be a hero’)

The shifted interpretation of the first person pronoun in the embedded clause in (37), in which it refers to the subject of the matrix clause, can be accounted for by saying that the embedded CP in (37) has an S operator of its own. Moreover, that S operator is controlled by the subject of the matrix clause, a possibility alluded to in (29b). (This is a syntactic version of Schlenker’s own view; see also Koopman and Sportiche 1989 and Adesola 2004 for control of logophoric operators in Abe and Yoruba.) The representation of (37) is thus (38). (38)

[S1k Johni [S2i Ii hero be+T] say+T]

Here the pronoun in the embedded clause is locally bound by the closest S operator, namely S2. Hence it qualifies as being first person, by (30a). But S2 does not designate the same individual as S1, namely the speaker. Rather, it designates John, by virtue of the control relationship. In this way, shifted and unshifted instances of first and second person pronouns can be accounted for within the same formal system for licensing first and second person features.10 The full potential of this treatment of first and second person features can be seen in Slave, as described by Rice (1989) (see Anand and Nevins 2004 for a similar analysis). (39) gives examples that show that in the complement of certain verbs, first person pronouns shift to refer to the matrix subject rather than to the speaker. The matrix subject itself can be any person – first, second ((39c)) or third ((39a–b)). (39)

a. Tony lue gho shéoht´ı en´ıdhe . (Rice 1989:1273) Tony fish 1sS.opt.eat 3sS.wants ‘Tony wants to eat fish.’ (lit. ‘Tony wants I eat fish,’ 1s=matrix subject, not speaker)

10 (37) can also have an unshifted reading, in which it means ‘John said that I am a hero’; see (48) below for discussion.

130

Explaining the restriction on person agreement b. Judóné ri nurse Teddy gho beghárayuhdá sudeli? when q nurse Teddy about 1sS.opt.see.3sO 3sS.want.1sO ‘When does the nurse want to see me again about Teddy?’ (lit. ‘When does the nurse want I see him about Teddy?’ 1s=matrix subject, not speaker; 3s=speaker) (Rice 1989:1274) c. Se-ts’e náod´ı yenenewe . (Rice 1989:1282) 1s-to 3sS.opt.help 2sS.want ‘Do you want him/me to help you?’ (lit. ‘Do you want he help me?’ 1s=matrix subject; 3s= speaker or other)

Rice shows that these complements are not direct quotations, which are syntactically independent of the matrix clause. For example, (39b) shows that it is possible for wh-phrases to move out of the embedded clause in Slave even when first person pronouns have the shifted reading. (In contrast, direct quotations are islands for extraction in English: compare What did Mary say that she will buy? with *What did Mary say “I will buy”?) Just as in Amharic, these embedded clauses have an S operator that is controlled by the matrix subject. The representation for (39c) would be as in (40) (presented with English word order for convenience). (40)

[Si , Ak [youk want [Sk [hei,n help mek,*i ]]]]

Rice (1989) shows that whether a verb induces this shift in the interpretation of first person pronouns is a matter of semi-idiosyncratic lexical selection. Alongside verbs like those meaning ‘want’ in (39) is another class of verbs that do not induce a shift in the interpretation of pronouns inside their clausal complements. (41) is an example; here the first person subject of the embedded clause refers to the speaker of the sentence, not the matrix subject. (41)

gú kod´ıhsho . (Rice 1989:1273) John ʔ erákeʔ e´ e wihs´ı John parka 1sS.made comp 3sS.know.area ‘John knows that I made a parka.’ (1s=speaker) [Si , Ak [Johnn know [(Si ) [Ii make parka]]]]

Slave verbs that do not induce pronoun shift have English glosses such as ‘know,’ ‘hear,’ ‘teach,’ ‘see,’ ‘think, worry about,’ ‘say,’ ‘remember,’ and ‘find out’ (Rice 1989:1274–5). Verbs that do induce pronoun shift have English glosses such as ‘say,’ ‘tell,’ ‘ask,’ ‘want,’ and ‘think’ (Rice 1989:1276). Note that near-synonyms can appear on different lists. This supports a syntax-oriented approach to the phenomenon like the one I have given. In these terms, the difference reduces to whether the S operator in the embedded clause is controlled by the matrix subject or not. The CP complement of ‘know’ in (41) either has


131

no S operator at all, or it has one that is indexed to the speaker, just as the S in a matrix clause invariably is. As a result, any first person pronoun in this complement is locally bound by an S that designates the speaker. This difference between the two kinds of verbs in Slave is not significantly different from the familiar fact that some English verbs take finite complements and others take nonfinite control complements – a difference that also does not seem to be entirely predictable from the meanings of the verbs involved:11 (42)

a. I insist that I teach syntax next year. *?I insist to teach syntax next year. b. I demand that I teach syntax next year. I demand to teach syntax next year. c. *I want that I teach syntax next year. I want to teach syntax next year.

Rice (1989) further shows that some verbs – notably the verb meaning ‘tell’ – induce a shift in the interpretation of second person pronouns as well as first person pronouns. A first person pronoun in a clause embedded under this verb refers to the matrix subject rather than the speaker, and a second person pronoun in the embedded clause refers to the matrix object rather than the hearer: (43)

séhdi. (Rice 1989:1277) John ʔ aran´ıla John 2sS.go.home 3sS.tell.1sO ‘John told me to go home.’ (lit. ‘John told me you go home’ 2s=matrix object)

In contrast, verbs like intransitive ‘say’ induce a shift of first person pronouns in their complement clause, but not a shift of second person pronouns. Thus, the first person object in (44) refers to the matrix subject (not the speaker) but the second person subject refers to the addressee of the whole sentence, not to some covert argument of the matrix verb. (44)

Simon náseneineht’u hadi. (Rice 1989:1279) Simon 2sS.hit.1sO 3sS.say ‘Simon said that you hit him.’ (lit. ‘Simon said you hit me’ 1s=matrix subject, 2s=addressee)

11 One regularity that Rice mentions is that the complements of verbs that do not shift pronouns can have overt complementizers; gú in (41) is a case in point. In contrast, the complements of verbs that do shift pronouns never have an overt complementizer. This relationship between the presence of S and A operators and the type of C that is selected fits well with the idea that these operators occur at the CP level of the clause.

132


Note that the pronouns in the embedded clause in (44) are not interpreted as they would be in normal indirect discourse, nor as they would be in normal direct quotation; rather they are a mixture of the two. This confirms that we need a richer typology of embedded clauses than this simple binary distinction. I account for this difference between ‘tell’ and ‘say’ by saying that the complement of ‘tell’ has an A operator that is controlled by the matrix object, as shown in (45a). In contrast, the complement of ‘say’ has an A operator that is not controlled, there being no suitable controller in the matrix clause. This A operator designates the addressee, as usual whenever there is no control ((29b)). The representations in (45) show that the observed facts follow from these assumptions about structure plus the PLC in (30). (45)

a. [Si , Ak [Johnn tell mei [Sn , Ai [youi,*k /*Ii go home.]]]] b. [Si , Ak [Simonn say [Sn (Ak ) [youk hit men/*i /him*n ]]]]

These examples make several significant points. First, they show that the system generalizes properly from first person to second person pronouns, as we would hope. Second, they show that there are similar but distinct operators for second person and first person pronouns, since second person and first person do not shift in exactly the same environments.12 Third, they give further support for a syntactic approach modeled on control, in that object control exists in both domains, and similar sorts of verbs trigger object control of both kinds. Now we come to the aspect of the Person Licensing Condition that is the most crucial for the project of deriving the SCOPA. Notice that the second person pronoun in (43)/(45a) cannot refer to the addressee of the sentence as a whole, but can only refer to the object of the matrix clause. This shows that it is not enough for a second person pronoun to be bound by any old A operator; rather it must be bound by the closest A operator – the one in the embedded CP, not the one in the matrix CP. Similarly, the first person pronoun in (44)/(45b) cannot refer to the speaker of the sentence as a whole, but only to the matrix subject (Simon). This shows that the first person pronoun can only be bound by the closest S operator. Rice’s extensive Slave data is entirely consistent on this point. Indeed, the same thing could be seen back in example (39c)/(40). There too the first person pronoun in the embedded clause can only be bound by the closest S; therefore it ends up designating the addressee rather than the speaker, given that the embedded S is controlled by the matrix subject. Note 12 This is crucially different from Schlenker’s (2003) implementation, in which all the index parameters are packaged together in a single operator.


133

also that the third person subject ‘he’ of the embedded clause in (40) can refer to the speaker: even though this pronoun is assigned the index i associated with the matrix S, it is not locally bound by the closest S, so first person features are not licensed on it. It is therefore third person, by default, in accordance with the PLC. Rice (1989:1289) also briefly discusses the interpretation of pronouns in sentences with two levels of clausal embedding, such as (46). This provides further evidence for the minimality clause in the PLC: (46)

ʔ adi. en´ıwe John Susan tle go l´ı ʔ aohde John Susan Norman Wells 1sS.opt.go 3sS.want 3sS.say ‘John said that Susan wants (Susan/*John/*me) to go to Norman Wells.’

The notable feature of this example is that the first person pronoun in the most deeply embedded clause must refer to the subject of the immediately higher clause, not to the matrix subject, and not to the speaker. This is additional evidence that a pronoun is only first person if it is bound by the closest c-commanding S operator. The representation of (46) in (47) should make this clear. (47)

[Si Johnk said [Sk Susann wants [Sn In,*k,*i go.to Norman Wells]]]

At first glance, the other languages discussed in the literature on the shifting of first and second person pronouns seem to be different from Slave in this respect. The first person pronoun in Schlenker’s Amharic example, repeated again in (48), can refer to either the matrix subject or the speaker. (48)

ˇon

ˇə gna nə -ñn˜ y l-all. John hero be.pf-1so 3m.say.aux.3m ‘John says that he is a hero’ or ‘John says that I am a hero.’

The same ambiguity is found in Anand and Nevins’s (2004) examples from Zazaki, including (49). (49)

va kε εz dεwletia. Hεseni Hesen.obl said that I rich.be ‘Hesen said that he (Hesen) is rich’ or ‘Hesen said that I am a rich.’

But at least for Zazaki the problem is only apparent. The source of the ambiguity in (49) is not the ability of a first person pronoun to be bound by a more remote S operator; rather it comes from the fact that the Zazaki verb ‘say’ only optionally selects for a controlled S operator in its CP complement. Thus, (49) can have either of the two representations in (50).

134 (50)

Explaining the restriction on person agreement a. [Si Hesenk say [Sk that [Ik rich.be]]] b. [Si Hesenk say [that [Ii rich.be]]]

In both representations, the first person pronoun is bound by the closest S, in accordance with the PLC. Slave is different only in that its verb meaning ‘say’ selects only the equivalent of the CP found in (50a). Crucial evidence that the same minimality condition holds in Zazaki as in Slave comes from examples like (51), which Anand and Nevins discuss as illustrating their “shift together” generalization. (51)

Hεsen va kε pyaay kε m -ra hes kene Hesen says That people that me-obl like do pyaay kε m -ra hes ne kene ame zuja. people that me-obl like neg do came together *‘Hesen said that the people who like him/me and the people who don’t like him/me met.’

Like (49), this sentence is two-ways ambiguous: the first person pronouns in the embedded clause can refer to the speaker or to the subject of the matrix clause, Hesen. But Anand and Nevin point out that the sentence is not four-ways ambiguous: it is impossible for one of the first person pronouns to designate the matrix subject Hesen while the other designates the speaker. The minimality condition in the PLC explains why this is so. Suppose one of the first person pronouns is interpreted as referring to Hesen. This is only possible when the verb ‘say’ exercises its option of selecting a CP with a controlled S operator, so the representation is one like (50a), not one like (50b). Given this, consider the options for the second first person pronoun. It must be bound by the controlled embedded S, not by the matrix S, according to the minimality clause of the PLC. Hence it too is interpreted as referring to Hesen. The representation is given schematically in (52a). (The other, less-relevant representation is given in (52b), for comparison.) (52)

a. [Si Hesenk say [Sk that [people who like mek,*i ] & [people who don’t like mek,*i ] met]] b. [Si Hesenk say [that [people who like mei,*k ] & [people who don’t like mei,*k ] met]]

Therefore, (30) applies as written to Zazaki and as well to Slave. I assume that Amharic is to be analyzed similarly to Zazaki, although the crucial evidence is not available. Indeed, I assume that the PLC is universal.


135

4.3.3 Comparison with logophoric pronouns To conclude this background investigation into first and second person pronouns, it is instructive to compare what we have learned about them with the phenomenon of logophoric pronouns found in West African languages. Most people working on this topic have seen similarities between the two phenomena, and for some pronoun shifting is analyzed as a type of logophoricity (e.g., Safir 2005). I agree that there are many important parallels – but there is also one crucial difference that is highly relevant to my topic. Logophoricity in African languages is the use of a special pronoun in an embedded clause to refer to the subject of a main clause containing a verb of speaking or cognition. I illustrate the phenomenon from Edo (data from O. T. Stewart, personal communication); see also Koopman and Sportiche 1989 on Abe and Adesola 2005 on Yoruba. For example, the ordinary third person ` o., whereas pronoun o.´ in (53a) may not be coreferential with the matrix subject Oz´ the distinctive logophoric pronoun ´ırè.n in (53b) must be. (53)

` o Miànmián wè.e´. o´. k`ıe´ a. Oz´ e` khù. Ozo forgot that he opened door ‘Ozoi forgot that he*i,k (someone else) opened the door.’ ` o. miànmián wè.e´. ´ırè.n b. Oz´ k`ıe´ e` khù. Ozo forgot that he.log opened door ‘Ozoi forgot that hei,*k (Ozo) opened the door.’

The logophoric pronoun cannot, however, be used in argument position of an ordinary matrix clause, nor is it licensed in infinitival embedded clauses: (54)

´ e.n mié.n e` n´ı. a. *Ir` he.log see elephant ‘He saw an elephant.’ ` o miànmián ya t`ıe´ e` bé e´. rè /*´ırè.n b. Oz´ Ozo forget to read book his/his.log ‘Ozo forgot to read his book.’

Following Koopman and Sportiche (1989), this range of data can be captured by saying that logophoric pronouns need to be locally bound by a designated logophoric operator, which is found in the specifier of certain CPs. This is stated in (55) (essentially the same as the Strong Pronouns Licensing Principle of Adesola 2005:190). (55)

A logophoric pronoun must be locally bound by a logophoric operator, or by another logophoric pronoun.

136


Verbs of speaking and cognition select for CPs that have such an operator, and one of their arguments controls that operator. This gives a representation like (56) for (53b). (56)

[Ozok forgot [LOGk that [hek opened the door]]]

In contrast, matrix clauses do not have a logophoric operator, nor do nonfinite clauses like the one in (54b). Therefore the examples in (54) are ruled out by the condition in (55). The parallels between logophoric pronouns and shifted first person pronouns in Slave are evident once one realizes that the logophoric pronoun in Edo is used in the same contexts in which a first person pronoun in Slave has a shifted interpretation. Both logophoric pronouns and first person pronouns are used to designate the subject of the higher speech verb, but not the subject of other sorts of verbs. Moreover, Adesola (2005) shows that logophoric pronouns are subject to a kind of strong crossover effect, much like first person pronouns are in my analysis of (34) in section 4.3.1. In Yoruba (but not in Edo) a normal pronoun can be interpreted as referring to the matrix subject in an example like (57).13 (57)

pé o´ i,k n´ bò. ló.la. (Adesola 2005:184) Olúi ti kéde Olu asp announce that he prog come tomorrow ‘Olu announced that he (Olu or someone else) is coming tomorrow.’

But the existence of such a pronoun in the embedded clause can prevent a logophoric pronoun from designating the matrix subject the way that it normally would: (58)

Olúi so. [LOGi pé [ó*i,k r´ı bàbá o` uni ]] (Adesola 2004:185) Olu say that he see father him.log ‘Olu said that he (someone else) saw his (Olu’s) father.’

The deviance of (58) when o´ is understood as bound by the matrix subject can be attributed to (55): although the logophoric pronoun is bound by the logophoric operator, it is not locally bound by that operator; the nonlogophoric pronoun is a closer binder. The badness of (58) is thus comparable to the badness of [Si Daddyi lost myi wallet] in English. If the closer binder is itself a logophoric 13 This difference between Yoruba and Edo can be modeled by saying that ‘say’-type verbs in Edo require a logophoric operator in their complement, whereas ‘say’-type verbs in Yoruba allow one but do not require one. Note that this is exactly parallel to the difference between Slave and Zazaki when it comes to the shifting of first person pronouns: some Slave verbs require a controlled S operator in their complement, whereas ‘say’ in Zazaki allows one but does not require it.


137

pronoun, however, the sentence is fine with the relevant interpretation (Adesola 2005:185, 192). (59)

Olúi so. [LOGi pé [óuni r´ı bàbá o` uni ]] Olu say that he.log see father him.log ‘Olu said that he (Olu) saw his (Olu’s) father.’

The acceptability of examples like this are the reason why (55) states that a logophoric pronoun can be locally bound by another logophoric pronoun as well as by a logophoric operator. This is parallel to the fact that first person pronouns can be locally bound by an S operator or by another first person pronoun, as shown by the goodness of [Si Ii lost myi wallet]. In all these respects, logophoric pronouns are like first and second person pronouns – similarities that are captured by saying that both must be locally bound by a particular sort of operator (the PLC and (55)). There is one crucial difference between the PLC and (55), however: (55) does not contain the same minimality condition that the PLC does. (55) does not stipulate that a logophoric pronoun must be bound by the closest logophoric operator. The difference is empirically motivated. In a doubly embedded structure like (60) from Edo, the logophoric pronoun in the lowest clause is ambiguous; it can refer to either the subject of the highest clause or to the subject of the intermediate clause. (60)

` o ròró wè.e´. Uy` ´ ı tá wè.e´. Adesuwa bàa´ Oz´ ´ırè.n o` hó! ghé. Ozo thinks that Uyi say that Adesuwa accuse him.log of.lying ‘Ozo thinks that Uyi said that Adesuwa accused him (Ozo or Uyi) of lying.’

The same is true in other West African languages, including Yoruba, Abe, and Ibibio. In contrast, we have seen that shifted first person pronouns in similar contexts in Slave are unambiguous: a first person pronoun in the lowest clause can only refer to the subject in the clause immediately above it, not to the subject of the highest clause, as shown in (46), repeated here as (61). (61)

ʔ aohde ʔ adi. en´ıwe John Susan tle go l´ı John Susan Norman Wells 1sS.opt.go 3sS.want 3sS.say ‘John said that Susan wants (Susan/*John) to go to Norman Wells.’

It follows that when there are two logophoric pronouns in the same clause, they need not refer to the same person; (62) is an example in which they do not (see also Koopman and Sportiche 1989:571–2).

138 (62)

Explaining the restriction on person agreement ` o tá wè.e´. Ad´ ` esúwà ròró wè.e´. ´ırè.n Oz´ hoè.mwé.n ´ırè.n. Ozo say that Adesuwa think that he.log like her.log ‘Ozo said that Adesuwa thinks that he (Ozo) likes her (Adesuwa).’

This contrasts with the Zazaki example in (51), where two first person pronouns in the same clause must refer to the same person. This contrast, assuming that it proves systematic across languages, is problematic for Safir’s (2005) view that “shifted” first person pronouns in languages like Amharic are simply logophoric pronouns that happen to be homophonous with ordinary first person pronouns. I conclude that logophoric pronouns are very similar to shifted first and second person pronouns, but they are not identical to them. All three must be locally bound by a designated operator. The difference is that first and second person pronouns must be bound by the closest operator of the relevant kind, whereas third person pronouns (including logophoric ones) need not be. I do not have any deep insight into why first and second person pronouns should differ from third person pronouns in this way. But apparently they do – and it is this difference that I use to derive the differences between first/second person agreement and other kinds of agreement.

4.4

On the strictness of locality conditions involving heads

So far I have established that a pronoun is first or second person only if it is locally bound by the closest particular element that is also first or second person, as stated in the Person Licensing Condition (PLC) in (30). Moreover, agreeing functional heads are like pronouns in that they count as bound variables dependent on the agreed-with NP, according to (26). Therefore it is natural to think that the PLC applies to these agreeing functional heads as well. Hopefully this is the source of the difference in first and second person agreement as opposed to agreement in other features, the descriptive generalization expressed in the SCOPA. But there is still something that needs to be added before this works. While it may be true that agreement markers associated with first or second person pronouns are subject to the local binding condition in (30), it seems like this condition will always be satisfied in a rather trivial way. By hypothesis, a functional head F has first or second person features only if it enters into an Agree relationship with some pronoun in the same phase that is first or second person. The agreed-with pronoun is itself only first or second person if it is locally bound by an S or A operator, or by some other first or second person


139

pronoun, given the PLC. Then either the S or A operator that binds the pronoun also locally binds the agreement-bearing head (if the head agreed downwards, with a pronoun in its c-command domain), or else the agreed-with pronoun itself locally binds the agreement-bearing head (if the head agreed upwards, with a pronoun that c-commands it). Either way, any configuration that has a first or second person pronoun to agree with will also be a configuration in which the PLC is satisfied for the agreeing head. Therefore, the PLC as stated puts no new restrictions on first and second person agreement that go beyond those built into the Agree relationship itself. In particular, it does not require that the first or second person pronoun merge directly with the agreeing head. So it cannot be used to explain the SCOPA unless something is added. Fortunately there is a new assumption that can be added to this picture. When one considers other sorts of locality conditions, there is evidence that, even when recognizably the same condition applies to both phrases and heads, the locality holds in a stronger form in the case of heads. One case in point is the binding condition that regulates anaphors. All anaphors must be bound by an antecedent within a suitable local domain. For full NP anaphors, this local domain can be a rather large one, such as the smallest indicative clause that contains the anaphor. Thus, the anaphor in object position in (63a) from Icelandic need not be bound by the closest subject, but can be bound by the subject of the matrix clause, given that the embedded clause is subjunctive. This kind of latitude is never granted to anaphors that are bare heads, such as the clitic anaphors that attach to Tense in the Romance languages. These can only take as their antecedent the most local NP imaginable: the specifier of the very T that they attach to. (63b) thus differs from (63a) in this respect (Pica 1991). (63)

sig. (Icelandic, p. 119) a. Jón sagð i þeim að Mar´ıa elski Jon told them that Maria love.sbjn self ‘Johnk told them that Mariai loves him/herselfi,k .’ b. Paul souhaite que Jean se photographie. (French, p. 130) Paul wishes that Jean self photograph.sbjn ‘Paulk wishes that Jeani would photograph himselfi,*k .’

Both sig and se are anaphors, distinguished from comparable pronominals by their need to have a binder within a local domain. But there is also a difference: the relevant domain is much smaller for the anaphoric head than for the anaphoric DP. Another illustration of this observation comes from the theory of movement. All sorts of movement are subject to locality conditions of one kind or another. Moved wh-phrases can only move out of one clause at a time, and only if there

140


is an intermediate landing site at the edge of the clause; this results in the familiar range of island phenomena (Chomsky 1973). Similarly NPs undergoing A-movement for case-and-agreement reasons usually stay within the same clause, and can only move into the next higher clause when very specific conditions are met. Both of these sorts of movements are local to something like a clause. Moved heads are also subject to a locality condition, but a much stricter one. Staying within the same clause/phase is not good enough for a moving head; rather it must land in the immediately superordinate phrase, whether it is a phase or not (the Head Movement Constraint; see Travis 1984, Baker 1988). This results in contrasts like the following: (64)

a. Such opportunities must be taken advantage of –. NP movement b. Which car do you like the color of –? Wh-movement c. *K-ather-a-nuhwe’-s [ne ohsohkwa –]. Noun incorporation 1sS-basket-Ø-like-hab ne color (Mohawk (Baker 1991:371)) ‘I like the color of the basket.’

In (64a), a kind of pseudopassive, the NP apparently raises out of PP, NP, and VP into the subject position. In (64b), the wh-phrase moves out of PP, NP, VP, and TP into Spec, CP. But there is no noun incorporation structure like (64c) in which a noun moves out of a larger NP to adjoin to V. The movement in (64c) is, if anything, shorter than the movements in (64a) and (64b), but yet it is too long. Similar contrasts can be found in complex verbal constructions, as shown in (65) and (66). (65)

(66)

a. Le nuove case furono iniziate a costruire negli anni ’20. The new houses were started to build in.the 20s ‘They started to build these houses in the 20s.’ (Italian, Wurmbrand 2003:20) b. Which book did you start to read – ? a. ə tlə g-e lə waw-ə rkə -nen rə -maraw-at-ə k ekə k. father-erg cannot-pres-3sS/3sO caus-scold-caus-inf son.abs ‘The father cannot scold the son.’ (Chukchi, Baker 1999:369) b. Ga-nae l-in-au payitt-o-k. past2-start-past2–3pS blackberry-eat-inf ‘They started to eat blackberries.’ Not: *Ga-payitt-nae l-in-au o-k ‘(They) blackberry-started to-eat’

In these restructuring cases, the object of the lower verb can move out of the lower VP and out of the higher VP, into the matrix Spec, TP, as in (65a) (the so-called “long passive”). Also in such cases, the +wh object of the lower verb can easily move out of the lower VP and the higher VP, into the matrix Spec, CP,


141

as in (65b). But restructuring seems never to allow “long noun incorporation.” Chukchi provides a good illustration: clitics/agreements that express the object of the lower verb can appear attached to the higher verb, as in (66a), but an incorporated noun root that expresses the object of the lower verb cannot ((66b)). Again, the movement in (66b) is, if anything, shorter than the movements in (65), and yet it is too long. It is helpful to see how this difference between head movement and phrasal movement was captured in Rizzi’s (1990) theory of Relativized Minimality. Rizzi’s idea was that movement relationships were blocked if (among other things) the movement skipped over a position of the same type as the landing site of the movement. Crucial building blocks toward achieving this result are quoted in (67) and (68) (Rizzi 1990:7). (67)

Relativized Minimality: X α-governs Y only if there is no Z such that (i) Z is a typical potential α-governor for Y, (ii) Z c-commands Y and does not c-command X.

(68)

a. Z is a typical potential antecedent governor for Y, Y in an A chain = Z is an A specifier c-commanding Y. b. Z is a typical potential antecedent governor for Y, Y in an A chain = Z is an A specifier c-commanding Y. c. Z is a typical potential antecedent governor for Y, Y in an Xo chain = Z is a head c-commanding Y.

Notice that the movement of phrases divides into two distinct kinds, Amovement and A-bar movement. The difference between the two is defined not only by the syntactic configuration (whether there is an intervening specifier) but also by something about the intrinsic feature content of the position – whether it is an A-bar specifier or an A-specifier, a specifier with quantificational/operator features or one with case/agreement features. Thus NP raising can cross a floated quantifier but not a nominative subject, whereas wh-movement can cross over a nominative subject but not a floated quantifier. Subsequent work identified still more types; for example, Müller and Sternefeld (1993) identify scrambling and topicalization as additional types of movement, which do not interact with each other, or with wh-movement and NP-movement. In contrast, the movement of heads does not divide up into two or more distinct subtypes. There is little or no evidence that some kinds of head movement can skip one type of head position whereas other kinds can skip other types of head positions. This implies that the intrinsic features of heads do not matter when evaluating the locality of a head movement, the way that the features of specifiers matter when evaluating the locality of a phrasal movement. Only a specifier of the same type blocks movement to specifier position, whereas

142


heads of all types block movement to head positions. The upshot is that head movement obeys a more restrictive form of locality than phrasal movement, as observed in (64)–(66). The movement of phrases is local to (roughly) the clause, but the movement of heads is bounded by a single phrase. Suppose that we generalize over the case of anaphor binding and the case of movement, to posit a general law, stated roughly as in (69). (69)

For a locality relationship involving a head, all intervening heads count as interveners, regardless of their intrinsic features.

(69) could be the underlying reason why there are multiple clauses dealing with phrasal movement in (68), but only one clause dealing with head movement. 4.5

Deriving the SCOPA

We are now ready to put the pieces together by applying the general law in (69) to the PLC, the locality condition on the licensing of first and second person features motivated by the data in section 4.3. The PLC is repeated here as (70), with the crucial locality condition unpacked and italicized. (70)

Person Licensing Condition a. A phrase X is first person only if it is locally bound by an S or another element that is first person, and there is no other S that c-commands X but not its local binder. b. A phrase X is second person only if it is locally bound by an A or another element that is second person, and there is no other A that c-commands X but not its local binder. c. Otherwise, X is third person.

(70a–b) is similar to (68a–b) in that the intrinsic features of the operators matter for whether they count as interveners as well as their syntactic positions. Thus, the S operator blocks the more remote binding of first person pronouns but not that of second person pronouns, whereas the A operator blocks the binding of second person pronouns but not first person pronouns. This is seen most clearly in an example like (44) from Slave, repeated here as (71). (71)

Simon náseneineht’u hadi. Simon 2sS.hit.1sO 3sS.say ‘Simon said that you hit him.’ (1s=matrix subject, 2s=hearer) [Si , Ak [Simonn say [Sn [youk hit men/*i ]]]]

Here the lower S operator prevents the pronoun ‘me’ from referring to the matrix S (i.e., to the speaker), but it does not block the pronoun ‘you’ from referring

Deriving the SCOPA

143

to the matrix A (the addressee). So intrinsic features and relative positions both matter in these relations involving pronouns. But according to (69), intrinsic features should not matter for relations involving heads. This means that any intervening head, regardless of its features, should block the licensing of first and second person on a head, much as any head, regardless of its features, blocks head movement. The version of (70) that applies to heads is thus (72), with the italicized locality condition replaced by a non-relativized one. (72)

Person Licensing Condition (Head Version): a. A head X is first person only if it is locally bound by an S or another element that is first person, and there is no other head that c-commands X but not its local binder. b. A head X is second person only if it is locally bound by an A or another element that is second person, and there is no other head that c-commands X but not its local binder. c. Otherwise, X is third person.

I now claim success, inasmuch as (72), the PLC(H), is equivalent to the SCOPA in (1), repeated here as (73). (73)

F can agree with XP in +1, +2, only if a projection of F merges with a +1, +2 element and F projects.

To see the equivalence of (72) and (73), consider first a typical configuration that satisfies the SCOPA, such as a finite T agreeing in person with the nominative subject in Spec, TP in Icelandic or some other language. The syntactic configuration would be approximately (74). (74)

. . . [CP Si Ak that [TP Ii [1,sg] Tense[1,sg]i [vP v . . .]]]

The SCOPA is satisfied because a projection of T merges with a copy of the first person subject that T agrees with. But (72) is also satisfied. The pronoun ‘I’ is a first person element that locally binds the first person head Tense as a result of Agree (see (26)). Moreover, there is no head Y, distinct from Tense and I, that c-commands Tense but not I. (The first head that c-commands Tense – other than perhaps I itself – is the complementizer C. But C also c-commands the subject I, since I is contained in the TP complement of C.) The same is true for a configuration in which P merges with (say) a first or second person pronoun and agrees with it; the pronoun locally binds the agreeing P, and there is no distinct head that c-commands P without also c-commanding that pronoun. More generally, suppose that any agreeing head F merges with a +1 or +2 NP. Then the +1 or +2 NP locally binds the agreeing F. Moreover, there could never be a head G that c-commands F but not the +1 or +2 NP. If

144


G c-commands F, then G merges with something that contains FP, the maximal projection of F. But F merged with the pronoun, by hypothesis, so the pronoun is inside FP and hence also inside the complement of G. So G c-commands the pronoun as well as the head, and does not count as an intervener. Thus, configurations that satisfy the SCOPA also satisfy the head version of PLC in (72).14 Next, consider configurations which do not satisfy the SCOPA because the pronoun is generated higher than the agreement bearing functional head F. The canonical case is predicate adjective agreement, where the first or second person pronoun is generated in Spec, PredP, and FA is the complement of Pred, as shown in (75). (75)

[CP Si Ak C [TP Wei [1,pl] be+T[1,pl]i [PredP Pred [FAP FA [*1,pl] [AP good]]]]]

The condition in (72) is also violated in (75). The closest binder of the first person FA is the copy of ‘we’ in Spec, PredP. But Pred is a head distinct from either FA and ‘we’ that c-commands FA and not ‘we’. Therefore, it prevents ‘we’ from licensing a first person feature on FA (even though it has no relevant feature of the same type). The same result is true more generally. Suppose that the first or second person pronoun does not merge with FP, but also is not c-commanded by F. Then the pronoun must c-command F, or even basic gender-number agreement would fail. As a maximal projection, the pronoun must merge with some other head G, by hypothesis distinct from F, where G projects. That the pronoun c-commands F implies that F is contained in the smallest projection that contains the pronoun G*, a projection of G. More particularly, if the pronoun is the specifier of G*, then F must be contained in the complement of G.15 If F is in the complement of G, then G c-commands F but not the pronoun. Therefore, G is an intervening head that prevents the pronoun from licensing +1 or +2 features on the functional head. These cases that violate the SCOPA thus violate (72) as well.

14 A possible complication could arise if G merges with F, but F projects. Then the higher projection of F merges with the pronoun. Then G might c-command F, but not the pronoun, and count as an intervener. The relationship of projection might need to be stated in a certain way to render this unproblematic (or one could state intervention in terms of asymmetrical m-command, as in note 20 to chapter 2). 15 As usual, a little more care is needed if the pronoun is adjoined to G*. Then F could be in a specifier of G*, or a lower adjunct to G*, and G would m-command F but not c-command it. One might also have to take into account the possibility of multiple specifier constructions; I leave this open.

Deriving the SCOPA

145

The other broad class of cases to consider is configurations that do not satisfy the SCOPA in which the agreeing functional head c-commands the agreed-with DP. Typical cases in point are the downward adjectival agreement found on Cinque adjectives in Italian (section 3.1.1) and the downward verbal agreement found in oblique subject constructions in Icelandic (section 3.3.2). The relevant syntactic configurations are sketched in (76). (76)

a. [CP Si Ak C [TP be+T [ Pred [FAP FA [*1,sg] [AP well.known [I, 1sg]]]]]] b. [CP Si Ak C [TP Mary-DAT Tense[*1,sg]i [vP v [VP bored.with [I,1sg]i ]]]]

These configurations also clearly violate (72). This time the agreed-with pronouns do not bind the agreeing heads because they do not c-command them (by hypothesis). Therefore, the pronouns cannot license first and second person features on FA or Tense, even apart from the “closer head” clause of (72). This will be true for any configuration that meets the description of this case: if F c-commands the pronoun and F does not merge with the pronoun, then the pronoun does not c-command F and hence cannot license +1 or +2 features on F. Putting this together with the previous two cases, we find that all syntactic configurations that satisfy the SCOPA satisfy (72) as well, and all syntactic configurations that violate the SCOPA violate (72) as well. The two are thus linguistically equivalent conditions. Inasmuch as (72) is the result of independently motivated conditions ((30) and (69)), we have explained the SCOPA. There is one other matter to consider with regard to structures like (76), however: could first or second person agreement be licensed on FA or T by virtue of FA or T being locally bound by the S or A operators that have scope over the whole clause? In principle, this could be possible: we know that S and A can license +1 and +2 on pronouns that they bind, and an important assumption of this whole line of analysis is that the agreement on F is essentially a bound variable, subject to versions of the same conditions. For the specific configurations in (76), this is not an issue: S cannot license +1 features on FA or T in (76) because the complementizer C is an intervening head that c-commands FA and T but not the S and A operators. Therefore the super-strict style locality that heads require does not hold. It is, however, perfectly imaginable that the S and A operators could license first or second person morphology on the complementizer itself, there being no other head that intervenes between the operators and the complementizer. This is a good result. Recall from section 4.1.5 that complementizers are the one category on which we found some apparent exceptions to the SCOPA.

146


Exceptions came to light in at least two language families: the Continental West Germanic languages, and the Niger-Congo languages. Consider first the West Germanic languages. A typical case of complementizer agreement in these languages is repeated in (77), from West Flemish. (77)

Kpeinzen dan-k (ik) morgen goa-n. I-think that-1s I tomorrow go-1s ‘I think that I’ll go tomorrow.’

Here the complementizer appears to be agreeing downward with the first person pronoun in Spec, TP, in violation of the SCOPA. But our investigation into the details of first and second person pronouns has revealed that CPs of this sort can have S and A operators generated within them. I claim that the agreement on the complementizer in (77) is made possible by the S operator generated inside CP, as shown in (78). (78)

. . . think [CP Si that[1sg]i [TP I[1sg]i [vP tomorrow ] go+T[1sg]i ]]

This is compatible with (72) (and the SCOPA): the pronoun in Spec, TP locally licenses +1 features on T, and the S operator in CP locally licenses the +1 on C. The Kinande example repeated in (79) is similar: there is an operator in the projection of the embedded speech complementizer ati, and it locally triggers first person agreement on that complementizer. (79)

Mo-n-a-layir-ire Kambale in-di a-gul-e amatunda. aff-1sS-t-convince-ext Kambale.1 1sS-that 1S-buy-sbjn fruits.6 ‘I convinced Kambale that he should buy fruits.’ [Si I i [1sg] T convinced . . . [LOGi [1sg] that[1sg]i [TP he buy fruits]]]

At first, this looked like nonlocal upward agreement on C with the matrix subject – an NP that is not even in the same phase as the C. Now it can be analyzed as local upward agreement with an operator in C, that operator itself being controlled by the matrix subject.16 These examples thus cease to be problematic; on the contrary, they constitute additional evidence in favor of the syntactic reality of S and A operators in the CP. That S and A operators can license agreement on C shows that they are indeed part of the syntactic representation. The fact that they can license agreement only on C – not on T or FA in examples 16 For Kinande, it is better to say that the operator in Spec, CP is a logophoric operator controlled by the matrix subject than to say it is an S or A operator. The reason is because first and second person pronouns in the embedded CP do not shift their interpretation in Kinande the way they do in Slave. It is not the exact nature of the operator that is crucial here, but rather its location and the ϕ-features that it bears.

Deriving the SCOPA

147

like (76) – both confirms precisely where these operators are in the syntactic representation and bears further witness to the extra-tight locality condition on first and second person agreement. (Note that the agreement on C in Kinande agrees with the matrix subject, whereas in West Flemish it agrees with the embedded subject. This difference relates to a very general parametric difference that distinguishes agreement in Niger-Congo languages from agreement in Indo-European languages – the topic of the next chapter.) Finally, a syntactic configuration of special interest is one in which the projection of an agreeing functional head is adjoined to a first or second person pronoun. We saw in chapter 2 that in such a configuration the functional head can agree with the pronoun it is adjoined to in number and gender, but not in person. The canonical case is an adjective adjoined to a pronoun in Tariana or Zulu, as in (80). (80)

a. Thin’ aba-khulu si-ya-khuluma. (*si-khulu) (Doke 1955:51) we cl1.pl-big 1pS-disj-be.speaking 1p-big ‘We big ones are speaking.’ b. [DP [DP we[1pl]k ] [FAP FA [pl,*1]k [AP big]]]

Another case is the projection of the associative head/relative marker in NigerCongo languages (see section 4.1.4). Here person agreement fails, even though the target and the goal are as close to each other as one could ask for; they merge with each other, and no other head or phrase intervenes. For the SCOPA, this configuration motivated the extra condition that when the agreeing head F and the agreed-with pronoun merge, it must be F that labels the resulting projection for person agreement to take place. Does the shift from the SCOPA to the PLC(H) in (72) shed any additional light on this matter? I claim that the answer is yes. When framed in terms of (72), it is clear that the real condition on person agreement is not just a locality condition: the agreed-with phrase must also bind the agreeing head. Here binding is to be understood in the way it is understood in binding theory and the theory of operator–variable relationships, as involving c-command plus coindexing of the sort that represents a referential dependency. This is pressed upon us by the view that (72) is not an independent principle of agreement; rather it is a version of a general operator-binding condition that applies to pronouns/variables more generally. Now, does the pronoun in (80) bind the agreeing head FA in the required sense? The answer is arguably no: it does not c-command FA , rather only a segment of it does. The maximal projection of the pronoun is the ultimate bearer of the pronoun’s index, and this category contains FA P, and hence does not c-command it. The distinction between c-command and containment is

148


known to be a crucial one in binding theory: a phrase cannot be the antecedent of an anaphor or pronoun that is contained within it (the i-within-i condition of Chomsky 1981). Thus, (81a) is acceptable, but (81b) is bad. (81)

a. [Johnk ’s picture of himselfk ] hangs in the front hall. b. *[NPk M. C. Escher’s picture of itselfk ] hangs in the front hall.

Since the NP in brackets in (81b) contains the anaphor, it cannot be the antecedent of that anaphor; it does not bind the anaphor. Taking this back to (80), it is fair to say that the first person plural pronoun does not bind the agreeing FA here. Therefore, it cannot license first person features on FA via (72), a binding theoretic principle. That aspect of agreement that is pure Agree can succeed in (80), but the aspect of agreement that is ultimately part of operator-variable binding cannot – which is the part of agreement that licenses first and second person features.17 I claim, then, that no special stipulation (only a clarification) is needed to account for structures like (80) within the new theory of the SCOPA. This is an encouraging sign that the quest to derive the SCOPA from more fundamental principles was worth the effort and has been successful. 4.6

Conclusion

In the previous section, I showed that the configurations that satisfy the SCOPA also satisfy the Person Licensing Condition for Heads in (72), and the configurations that violate the SCOPA also violate this PLC(H). Thus, the two are 17 Note that the c-command condition on Agree has to be understood a little differently: the first person pronoun does count as c-commanding FA for purposes of pure Agree. This is a delicate distinction, but not too unnatural, I think. It is equivalent to saying that all segments of a category (including the lower one) have ϕ-features needed for agreement, but only the highest one has the index relevant to binding. Technically, we can say that segments participate in c-command, but X binds Y only if X c-commands Y, X is coindexed with Y, and X is the largest phrase that bears the relevant index. (And one can hope that these choices will make sense in an even deeper, more minimalist understanding of these matters.) Notice that an anaphor can seem to depend on the head of a noun phrase that it is contained in in relative clauses, e.g., The man who is proud of himself will fall into error. In such examples, the anaphor is actually directly dependent on a relative operator that c-commands it. In exactly parallel circumstances, a verb or other head inside a relative clause can appear to agree in person with the head of the relative clause, as in the Lokaa example repeated here: (i)

amon b-˙a yo-bi:la we 2-rel 1pS-be.dark ‘we who are darkskinned’

Here, too, agreement is really with the trace of the relative operator, which properly binds the agreeing T. (This reasoning also implies that the relationship between the head of a relative clause and the operator of the relative clause cannot be binding-theoretic in nature.)

Conclusion

149

linguistically (nearly) equivalent, and we can replace the SCOPA in the theory of agreement with the PLC(H). But the PLC(H) is the coming together of two independently motivated principles, namely the PLC in (30) – the general linguistic rule of what it means for a grammatical element to be first or second person – and (69), a general law about how locality conditions apply to heads. In short, I have successfully derived the special behavior of agreement in first and second person from the fundamental defining property of first and second person features. What this ultimately amounts to (beyond theoretical and rhetorical tricks) is a claim to have uncovered a very abstract but significant parallel between the special syntactic behavior of first and second person pronouns and the special behavior of first and second person in agreement. When it comes to pronouns, first and second person pronouns are marked forms. They obey a very specific licensing condition, such that they must be locally bound by the closest operator of a specific type. Third person pronouns are unmarked forms, which are not subject to any such licensing condition; they simply appear whenever a first or second person pronoun is not licensed. The locality condition on first and second person pronouns is seen clearly in examples like (82) from Slave, where the first person pronoun in the most deeply embedded clause can only take the subject of the immediately higher clause as its antecedent. (82)

ʔ aohde ʔ adi. (p. 1289) (=(46)) en´ıwe John Susan tle go l´ı John Susan Norman Wells 1sS.opt.go 3sS.want 3sS.say ‘John said that Susan wants (Susan/*John) to go to Norman Wells.’

In contrast, third person pronouns – even logophoric ones that need to be bound by an operator – do not need to be bound by the closest operator. Thus, examples like (83) from Edo are possible with the indicated interpretation (see also Adesola 2004:207 for Yoruba). (83)

` o ròró wè.e´. Uy` ´ ı tá wè.e´. Adesuwa bàa´ Oz´ ´ırè.n o` hó! ghé. Ozo thinks that Uyi say that Adesuwa accuse him.log of.lying ‘Ozo thinks that Uyi said that Adesuwa accused him (Ozo or Uyi) of lying.’

So there is no condition on third person pronouns that they must be bound by the closest operator of a particular kind. When taken into the domain of heads that have first, second, and third person features (agreeing heads), this means that first and second person heads are subject to a strengthened form of the locality condition, in which every head blocks feature licensing regardless of its own features. However, third person heads are not subject to a strengthened form of the locality condition, because there is no locality condition on the licensing of

150


third person features in the first place; they are simply the default values. Thus, the agreement asymmetry in (84) is found to be of a piece with the binding theoretic asymmetry in (82) versus (83). (84)

a. Hamisi yu-Ø m-refu. (Swahili) Hamisi cl1-be cl1-tall ‘Hamisi is tall.’ b. *Ni-Ø ni-refu. (OK: ni-Ø m-refu) 1sS-be 1sS-tall ‘I am tall.’

To the extent that you believe that there is a real parallelism here – that it is no coincidence that first and second person elements differ from third person elements in both of these respects – I have succeeded in explaining the special properties of first and second person agreement in terms of more general facts. Appendix: a note on anaphoric agreeement and the SCOPA Some languages seem to have a special form of object agreement that is triggered by anaphoric pronouns as opposed to ordinary pronouns. The Swahili example in (85) is a case in point (see Woolford 1999 for discussion). (85)

Ahmed a-na-ji-penda (mwenyewe). (Vitale 1981:137) Ahmed 1S-pres-refl.O-love himself. ‘Ahmed loves himself.’

A possible approach to this phenomenon is to say that the +anaphoric feature of the (overt or covert) object in (85) can count as a ϕ-feature, and v agrees with the object in that feature. If this is the correct analysis of at least some instances of anaphoric agreement, we can then ask what sort of ϕ-feature +anaphoric is. Is it akin to the features +1 and +2, and thus subject to the SCOPA, or is it akin to number and gender features and not subject to the SCOPA? The existing empirical evidence suggests that agreement in +anaphoric is restricted by the SCOPA. For example, the Bantu languages have special reflexive forms of object agreement, but they do not have special reflexive forms of adjectival agreement. This is shown in (86b) from Chichewa. (86)

a. Ndi-na-i-khal-its-a pro[CL4] y-a-i-kali. 1sS-past-4O-become-caus-fv cl4-assoc-cl4-fierce. ‘I made them (e.g. lions) fierce.’

Appendix: a note on anaphoric agreeement and the SCOPA

151

b. Ndi-na-dzi-khal-its-a pro[+ana] w-a-m-kali. 1sS-past-refl-become-caus-fv cl1-assoc-cl1-fierce (*dz-a-dzi-kali) refl-assoc-refl-fierce. ‘I made myself fierce.’

(86a) shows an example of a nonanaphoric null pronoun that functions as both the object of the causativized verb ‘become’ and the subject of the predicate adjective ‘fierce’. (Probably ‘become’ selects a small clause complement that has ‘fierce’ as its predicate, and the subject of the small clause is exceptionally case-marked by the causative morpheme. The details of the structure are not crucial, however.) (86b) displays a minimally different example in which the null object is anaphoric, understood as being coreferential with the subject of the sentence. This null anaphor triggers special anaphoric agreement on the verb, but not on the predicate adjective. Rather, the agreement on the adjective can only show the number and gender of the null anaphor (here singular and human). This looks like another instance of partial agreement: just as verbs can agree with NPs in the feature +1 but adjectives cannot, so verbs can agree with NPs in the feature +anaphoric but adjectives cannot. This parallelism suggests that the SCOPA can be generalized to read as follows: (87)

F can agree with XP in +1, +2, or +anaphoric only if a projection of F merges with a +1, +2, or +anaphoric element and F projects.

A wider range of data also supports the generalization in (87). For example, it is known in the literature on the Person Case Constraint that just as the first object of a double object construction can trigger person agreement on the verb and the second object cannot, so the first object of a double object construction can trigger a reflexive form but the second one cannot (e.g., Bonet 1991:192, Anagnostopoulou 2003:254). Thus, Southern Tiwa has inflected verbs that mean ‘I gave myself a shirt’, but it does not have inflected forms that can mean ‘I gave myself to John’ (Allen et al. 1990). Similarly, Bruening (2001) shows that true long distance agreement is not possible with anaphoric arguments in Passamoquoddy, any more than it is possible with first and second person arguments. Thus, agreement in +anaphoric is absent in the same range of syntactic structures that agreement in +1 and +2 is absent in. In contrast, possessive determiners can show special +anaphoric forms in languages like Greenlandic (Bittner 1994) and adpositions can show special +anaphoric forms in languages like Slave (Rice 1989). These are functional categories that can manifest first and second person agreement as well (see sections 4.1.1 and 4.1.2).

152


So the evidence that +anaphoric agreement has the same limited distribution as first and second person agreement seems quite good. Given this, it is natural to ask whether the SCOPA as it applies to the +anaphoric feature can be derived by techniques similar to the ones I used to derive the SCOPA for the features +1 and +2. The answer is “I hope so.” The extension is far from trivial, because the licensing conditions in (30) clearly do not extend to anaphoric pronouns. In other words, there is no condition exactly like (88) in the grammar: (88)

A phrase X is +anaphoric only if it is locally bound by a designated operator or by another element that is itself +anaphoric, and there is no other operator of that sort that c-commands X but not its local binder.

In general, anaphors need to be bound by a nearby NP in argument position, not by an operator, and the antecedent NP itself does not need to be anaphoric in any sense. The good news is that the general condition on anaphors does not have to be exactly (88) in order to have the desired effect. Any condition of the form in (89a) should turn into the SCOPA-equivalent statement in (89b) when the law of strict locality for heads ((69)) applies to it.18 (89)

a. An element X is +anaphoric only if it is locally bound by SOMETHING or by an another element that is itself +anaphoric, and there is no other THING OF THAT SORT that c-commands X but not its local binder. b. A head X is +anaphoric only if it is locally bound by another element that is itself +anaphoric, and there is no other head that c-commands X but not its local binder.

I am optimistic that a suitable condition like (89a) can be found to achieve this result, since anaphors are by definition elements that need nearby antecedents. But completing the account is complex and leads into the deep waters of principle A of the binding theory – in part because natural languages have different kinds of anaphors (e.g. the SE, SELF, and SE-SELF anaphors of Reinhart and Reuland 1991), each of which is subject to a somewhat different locality condition. Taking this on properly would take us far astray, into topics that I am not well qualified to discuss. Therefore, I leave the task of explaining this aspect of the SCOPA to future research. 18 Another challenge for deriving (89b) would be to make sure that the “something” mentioned in (89a) is never close enough to the functional head to trigger +anaphoric agreement on the functional head directly, without there being a +anaphoric NP nearby for the head to agree with.

5


5.1

Introduction: parameters and other kinds of variation

In the preceding chapters of this book, I concentrated on the putatively universal conditions on agreement – those conditions that shape agreement and concord in all natural human languages. But agreement is a relatively superficial phenomenon, where one expects to see variation as well as uniformity. It is perfectly clear that there is significant variation at some levels. The most obvious type is the variation between languages with a great deal of agreement, where agreement is central to the expression of grammatical relationships, and languages where agreement is either absent or peripheral to the expression of grammatical relationships. The first class of languages are the head-marking languages of Nichols 1986, including the polysynthetic languages of Baker 1996; the second class of languages includes the isolating languages and pure dependent-marking languages. I have modeled this dimension of variation in part by saying that languages can vary as to whether or not functional heads like FA or FN are generated above APs and NPs, and in part by saying that functional heads may or may not be specified as being probes, which look for features to agree with.1 For example, Ps, Ts, and vs are functional heads with semantic content that agree with nearby DPs/NPs in some languages but not others. A language has lots of agreement if it includes suitable functional heads in its syntactic structures and those functional heads are designated as being probes. In contrast, a language has little or no agreement if either functional heads are not projected, or if the functional heads that are present do not count as probes. We might then expect to find a continuum in how much agreement behavior languages manifest, and we more or less do. Of the 108 languages surveyed below, 26 have no agreement, 9 have agreement on one functional head (T or in one case D), 30 have agreement on two heads (T and v or D), 26 1 Functional heads can also be specified as probing more than once, as discussed in section 3.3.3 in the context of double object constructions. The possibility of T agreeing more than once plays an important role in what follows, especially in the analysis of Nez Perce in section 5.11.2.2.

153

154


have agreement on three heads (T, v, and one other), 12 have agreement on four functional heads, and 4 have agreement on all five functional heads considered (T, v, D, P, and C). This looks like a normal distribution, centered on languages in which about 50% of the functional heads are agreers, plus an unexpectedly high number of languages that have no agreement at all.2 There is also a great deal of variation when it comes to how agreement is realized morphologically in particular languages. Languages obviously vary as to which phonological features are used to represent agreement in a particular category. For example, first person plural subject agreement on verbs is realized as mos in Spanish, yakwa in Mohawk, tu in Kinande, iñ in Mapudungun, and so on. Other superficial aspects of morphological variation include whether agreement is realized as a prefix or a suffix, details (or at least idiosyncracies) as to where in the morphological word it appears, which forms have zero exponence, and instances of syncretism, which may be patterned or idiosyncratic. I assume that these matters are handled by a postsyntactic realizational morphology, like that developed by the Distributed Morphologists (Halle and Marantz 1993). I thus continue to assume that the syntax decides which head agrees with which NP in which features, whereas the postsyntactic morphology decides how the features that a head acquires by agreement are spelled in morphemes.3 There may be cases in which it is somewhat unclear which component is responsible for handling a particular aspect of agreement. For example, one might debate whether a head fails to agree with a particular noun phrase, or whether it agrees with that noun phrase but the agreement is spelled out morphologically as zero (see Baker 2006 for one case study). One might also debate whether there is a syntactic basis for certain patterns of syncretism (for example, if a voice alternation is at work in a language with animacy hierarchy effects), or whether it is purely a matter of morphology (see, for example, Baker 2003b on the animacy hierarchy and agreement in Mapudungun). Relevant cases have been discussed in the literature, and I do not intend to add to the discussion here. I simply assume that there is both a syntactic and a morphological component 2 The relatively large number of languages with no agreement at all suggests that there is also a grammatical parameter that specifies whether or not Agree is operative in the syntax of a given language. I do not consider this parameter in any detail here, however. Although the amount of agreement present in a given language may be randomly distributed, which heads bear the agreement clearly is not. As is well known, T is by far the most common agreeing head, v and D are intermediate, and P and especially C are uncommon agreement bearers. I have no explanation as to why this is so. 3 See section 1.5 for a brief comparison of this view with that of Marantz (1991) and Bobaljik (to appear), who hold that agreement is entirely a PF phenomenon.

Introduction

155

to agreement phenomena, that the distinction between the two is clear enough in many cases, and that boundary issues will tend to be resolved as we learn more about both the syntax of agreement and the structure of the morphological component. In this work, I continue to pursue this overall vision by focusing on the syntactic aspect of agreement. The question at hand, then, is whether there are any significant parameters within the syntax of Agree, apart from issues of whether a particular head agrees at all and of how any agreement it has is realized morphologically. Something deserves to be called a grammatical parameter if it concerns what syntactic configurations undergo agreement, and if it is a relatively general feature of the language, not one that is tied to a particular head or construction. I argue that there are at least two such parameters, stated in (1) and (2). (1)

The Direction of Agreement Parameter (preliminary) F agrees with DP/NP only if DP/NP asymmetrically c-commands F.

(2)

The Case-Dependency of Agreement Parameter F agrees with DP/NP only if F values the case feature of DP/NP or vice versa.

These parameters are language-particular narrowings of the universal conditions on agreement. (1) is a more specific version of the c-command condition, which states that F may probe upward or downward for an NP to agree with. Some languages restrict this further, saying that agreement can only be upward. See section 5.10 for a refinement of parameter (1); there I show that there is also at least one language (Burushaski) that requires agreement to be downward, not upward. (2) can be thought of as a version of Chomsky’s activity condition, which says that an NP is visible to agreement only if it has an unvalued case feature. Both conditions are general parameters in that I take them to hold of all agreement relationships, not just those that involve a particular functional head. (1) is also syntactic inasmuch as it depends on the syntactic relationship of c-command, defined over phrase structures. (2) is syntactic inasmuch as the assignment of case is a syntactic matter, although this can be debated (see Bobaljik, to appear). It is very controversial whether grammatical parameters of this sort exist at all. Much current research in the field has been guided by the idea that all parameters are fundamentally lexical in nature. More specifically, many researchers hold that all syntactic variation is attributable to variation in the features of individual lexical items, the syntactic principles themselves being invariant, as stated in (3).

156 (3)

Parameters of agreement The Borer-Chomsky Conjecture All parameters of variation are attributable to differences in the features of particular items (e.g., the functional heads) in the lexicon.

I refer to this view as the Borer-Chomsky Conjecture for the following reasons. It was first suggested, so far as I know, by Borer (1984), shortly after the introduction of the original GB notion of parameter (Chomsky 1981), when cracks in the classical Pro-drop Parameter were already beginning to show (see also Fukui and Speas 1986, Webelhuth 1992). Chomsky (1995) then adopted it, incorporating it into his Minimalist Program, and many learned of it from there. Richard Kayne has also done much to promote it, clarify what it amounts to, and find interesting examples that illustrate it (see, for example, Kayne 2005). I refer to (3) as a conjecture because it clearly has that honorable status: it was posited in advance of the wealth of research in comparative syntax done over the last 10–20 years, and has guided how much of that work has been done. It is an intriguing generalization over a few suggestive examples that were available in the 1980s and early 1990s, which makes sense and has some attractive conceptual properties. Part of my interest in defending (1) and (2) is that they raise the very real possibility that the Borer-Chomsky Conjecture might turn out to be false – that there is variation in the principles of syntax as well as in the lexicon. My discussion proceeds as follows. First I consider the major functional categories one at a time, to see how the parameters in (1) and (2) apply to each. For each category, I begin by contrasting agreement in two large and relatively well-studied language families, Bantu (Niger Congo, NC) and Indo-European (IE). I show that many Bantu languages systematically obey (1) and not (2), whereas many Indo-European languages obey (2) and not (1).4 By first looking at examples from these two families in some depth, it is possible to see the full range of consequences that the choice of (1) versus (2) can have. Then for 4 Neither the Bantu family nor the Indo-European family is completely homogeneous with respect to these parameters, however. While most NC languages that I have checked obey parameter (1), Makhuwa on the eastern periphery of the Bantu area does not; see section 5.11.3.1 for discussion. Similarly, while most IE languages obey parameter (2), Nepali and Maithili on the eastern periphery of the IE area seem not to, based on the data discussed by Bickel and Yadava (2000). Subject agreement on T in Nepali can agree with an ergative case subject as well as with a nominative subject (p. 348). “Object” (nonsubject) agreement in Maithili can agree with a direct object, an experiencer, an indirect object, or even an NP in ablative case (p. 349). Finally, Maithili has raising constructions in which the raised nominal triggers agreement in both the embedded clause and in the matrix clause (p. 362). Nepali and Maithili are different in all these ways from Hindi, not to mention the European IE languages. In the text I often say for simplicity that NC languages work one way and IE languages work another way, and that is largely true, but is a bit of an idealization.

Agreement on tense

157

each functional category, I discuss how the same parameter values show up in other relevant languages from the core 100-language sample of the World Atlas of Language Structures (WALS) (Haspelmath et al. 2005). This accomplishes several goals. First, it shows that the issues raised are not parochial to the NC and IE languages. Second, it gives a sense of how the same agreement parameters can be realized in otherwise typologically distinct languages, which have different basic word orders and case-marking systems. Third, the additional languages may show combinations of parameter settings that are not attested in the NC languages and the IE languages – languages in which both parameters are set as yes, or both are set as no. Once my tour of the functional categories is complete, I come to the crucial question for the theory of parameters: the question of whether languages have consistent settings for the parameters in (1) and (2), or whether different functional categories can have different parameter settings within the same language. To evaluate this, I present statistical information from a study of the 100 languages of the WALS core language sample, showing that languages with consistent parameter settings are much more common than languages with inconsistent parameter settings. From this I conclude that the parameters in (1) and (2) are classical GB-style grammatical parameters, which regulate languages as wholes, and should not be reduced to stipulations about the feature content of individual lexical items the way that the Borer-Chomsky Conjecture would have it. 5.2

Agreement on tense

5.2.1 Tense agreement in NC versus IE languages It is attractive to compare NC languages and IE languages for several reasons. Both are large language families, so one can be sure that the properties in question are robust if they survive or are replicated in many daughter languages of the original stock. Both are relatively well studied (although the generative literature on IE languages is huge compared to that on Bantu languages). Finally, the languages are not wildly different from each other when it comes to syntactic properties other than agreement. The Bantu languages have the same general head-initial phrase structure as IE languages like English, Scandinavian, Celtic, and the Romance languages, they have roughly the same range of syntactic categories, and they have many of the same core syntactic processes (passive, wh-movement, control, raising). These baseline similarities make the differences that exist with respect to agreement stand out all the more clearly. Coming then to agreement, another similarity between the Bantu languages and the IE languages is that in both the finite verb agrees with the preverbal

158


thematic subject in person, number, and (in Bantu) gender. Simple illustrative examples from Kinande are:5 (4)

olukwi (lw’-omo-mbasa). a. Omukali mo-a-seny-ire woman.1 aff-1S/t-chop-ext wood.11 lk11-loc.18-axe.9 ‘The woman chopped wood (with an axe).’ b. Abakali ba-[a]-gul-a amatunda. woman.2 2S-t-buy-fv fruit.6 ‘The women bought fruits.’

The differences in agreement appear when something other than the thematic subject moves to Spec, TP, as is allowed to varying degrees in both Bantu languages and IE languages. For example, some Bantu languages allow locative inversion, where a locative PP rather than the theme NP moves to the subject position of a passive or unaccusative verb (Bresnan and Kanerva 1989). When this happens, T agrees with the fronted locative: (5)

(6)

a. Ku-mu-dzi ku-li chi-tsˆıme. (Chichewa, Bresnan and Kanerva 1989:2) 17-3-village 17-be 7-well ‘In the village is a well.’ (compare: Chi-tsˆıme chi-le ku-mu-dzi ‘The well is in the village.’) b. Pa-m-sikˇa-pa pá-bádw-a 16-3-market-16.this 16/fut-be.born-ind nkhonya. (Bresnan and Kanerva 1989:9) 10.fist ‘At this market will break out a fight.’ a. Omo-mulongo mw-a-hik-a mukali. (Kinande, Baker 2003c) loc.18-village.3 18S-t-arrive-fv woman.1 ‘At the village arrived a woman.’ b. Oko-mesa kw-a-hir-aw-a ehilanga. loc.17-table 17S-t-put-pass-fv peanuts.19 ‘On the table were put peanuts.’

5 My strategy for picking examples from Bantu languages is the following. Wherever possible, I illustrate each property of agreement from two languages. One of these is a language on which some important article has been published on the phenomenon. The other is always Kinande, a language that I worked on myself and on which I have data for the whole range of agreement configurations within a single language. I have not studied the NC family in detail to determine what are the exact boundaries of each phenomenon within that family. The claim is only that these phenomena are found in multiple Bantu languages and seem to be characteristic of how many Bantu languages work as opposed to how IE languages normally work. Similar remarks apply to how I have chosen examples from IE languages: I illustrate with English examples where possible, and otherwise from the published literature on some other language. I am not as concerned about showing that constructions in IE languages are widespread, because this is for the most part well known and uncontroversial.

Agreement on tense

159

Also like Chichewa and Kinande in this respect are Swahili and Kilega (Kinyalolo 1991:17–30).6 Locative inversion also exists in English, and has syntactic properties that are similar to those of the Bantu construction in many respects (Bresnan 1994). But there is an obvious difference in agreement: in locative inversion structures in English, the finite verb agrees with the postverbal nominal, not with the preverbal locative expression: (7)

a. On the table were/*was (put) some peanuts. b. On the table was/*were (put) a peanut.

The oblique subject constructions found in many IE languages are similar, with the dative or ergative subject acting like a locative PP in agreement-relevant respects. The oblique subject does not trigger agreement on T, whereas the theme argument bearing nominative case does, as shown in (8) from Icelandic. (8)

þeir. (Taraldsen 1995:307) a. Henni leiddust her.dat was.bored.by.3pS they.nom ‘She was bored with them.’

b. Henni leið ist bókin s´ın. (Boeckx 2000:356) her.dat was.bored.by.3sS book self’s ‘She finds her own book boring.’

Oblique subject constructions with these agreement properties also exist in Italian, Russian (Franks 1995, Sigurdsson 2002), and Hindi (Bhatt 2005), among others. There are various ways in which one might state this difference theoretically, but the parameters in (1) and (2) are one straightforward way. Agreement in Bantu languages is sensitive to c-command but not case. The moved PP in Spec, TP c-commands T, but the theme argument that remains in the verb phrase does not. Therefore, T in Bantu can agree with the PP, but not with the theme NP. IE languages are not subject to (1), so T can probe downward and agree with the 6 Southern Bantu languages like Sesotho and Zulu are consistent with this, but do not show it as clearly. In these languages, all locative expressions trigger a single kind of agreement (ho) and this is also the expletive agreement found in subjectless constructions. As a result, in an example like (i) it is not clear if ho is T agreeing with the preposed locative, or if it is dummy/default agreement. (i) Setófó-ng hó-phéh-el-o-a 7.stove-loc 17S-cook-appl-pass-fv ‘On the stove is cooked meat.’

nama. 9.meat (Sesotho, Machobane 1993:15)

It is clear, however, that the verb does not agree with the postverbal NP nama, as it does in IE languages. This confirms that the Southern Bantu languages also obey (1) and not (2).

160


theme NP, even when it remains in the verb phrase. However, T in IE languages can only agree with a phrase if it values the case feature of that phrase as nominative. It does this for the theme NP, but not for the fronted PP or quirkycase-marked subject. Therefore in IE languages agreement can only be with the theme, not with the PP or oblique NP.7 Another way to think about the contrast between (5)–(6) and (7)–(8) is to say that locative expressions bear ϕ-features in (some) Bantu languages but not in IE languages; this is the view of Bresnan and Kanerva 1989 and related work. But this alternative approach does not generalize to a second kind of inversion construction, found to varying degrees in some central Bantu languages. These languages allow the object to move to Spec, TP over the subject under certain discourse conditions, often to express contrastive focus on the subject. When this happens, T agrees with the fronted object, not with the thematic subject. Well-studied examples of this subject-object reversal construction exist in Kinyarwanda (Kimenyi 1980) and Kirundi (Ndayiragije 1999); it is also found in Swahili and to a limited extent in Kilega (Kinyalolo 1991) and Kinande (Baker 2003c). (9)

amatá (Kirundi, Ndayiragije 1999) a. Abâna ba-á-ra-nyôye children 2S-past-foc-drink.perf milk ‘Children drank milk?’ b. Amatá y-á-nyôye abâna. milk 3S-past-drink.perf children ‘Children (not parents) drank milk.’

(10)

Olukwi si-lu-li-seny-a bakali (omo-mbasa). (Kinande) wood.11 neg-11S-pres-chop-fv women.2 loc.18-axe.9 ‘WOMEN do not chop wood (with an axe).’ (compare (4a))

It is quite foreign to IE grammatical intuitions that there could be “subject” agreement with the direct object. Direct objects usually cannot move to Spec, TP in IE languages, but there are some plausible cases. For example, Icelandic and Yiddish are Germanic languages that show verb second phenomena even in embedded clauses with an overt complementizer. Diesing (1990) (for Yiddish) and Rögnvaldsson and Thráinsson (1990) (for Icelandic) argue that the object 7 Note that I do not assume that Agree is a prerequisite for movement, as proposed in Chomsky 2000 and related work. In Bantu examples like (4)–(6), movement to Spec, TP is a prerequisite for agreement, rather than the other way around in my implementation (see Carstens 2005 for a different view). In English examples like (7), movement is clearly independent of morphological agreement. I assume that movement is triggered by a distinct EPP feature of heads like T, and has nothing directly to do with agreement (on this point, see also Bobaljik and Wurmbrand 2005).

Agreement on tense

161

has moved to Spec, TP in sentences that have C–Object–Verb[finite]–Subject order, such as (11). (11)

. . . az vayn ken men makhn fun troybn oykh. (Yiddish, Diesing 1990:44) that wine can one make from grapes also ‘(You should know) . . . that one can make wine from grapes also.’

In this respect, then, (11) is syntactically comparable to (9) and (10) in Bantu languages.8 But the agreement properties are different: in (11), the finite verb must agree with the postverbal subject, whereas in (9) and (10) the finite verb must agree with the preverbal thematic object. This difference between Bantu and IE is parallel to the difference seen in locative inversion sentences. In both domains, the Bantu languages agree with the fronted, preverbal phrase whereas the IE languages agree with the postverbal thematic subject that bears nominative case. But it is not plausible to say that direct objects in Bantu languages have ϕ-features that can be agreed with, whereas direct objects in IE languages do not. The NPs that can be in object position in IE languages clearly do have ϕ-features: these features are active in NP-internal concord, and they can trigger agreement whenever they appear in a nominative subject position. In contrast, my parametric proposal based on (1) and (2) correctly captures the patterns. Whatever lands in the Spec, TP position in Bantu governs the agreement on T, be it the thematic subject ((4)), the thematic object ((9) and (10)), or a locative expression ((5) and (6)), because that phrase alone c-commands T. In contrast, T in IE agrees with the NP which has nominative case, regardless of whether that NP is above T or below it. These examples also show the importance of the case condition in (2) in IE languages. Why does the IE finite verb agree with the subject and not the object in examples like (11)? The standard answer within Minimalist syntax involves locality: the thematic subject is closer to T than the object is before anything moves to Spec, TP. But the Bantu languages show that agreement can be sensitive to postmovement configurations: movement to Spec, TP clearly feeds agreement in the Bantu languages, because no phrase meets the condition of c-commanding T prior to movement. After the direct object moves to Spec, TP in sentences like (11) in IE (triggered perhaps by topic-focus features), the subject is no closer to T than the object is. The subject is the closest NP to T 8 There may be a difference in the type of movement undergone: A-bar movement in the Germanic languages, as opposed to A-movement in the Bantu languages. But it is not entirely clear what this difference amounts to apart from the differences in case and agreement under discussion. In any case, the A/A-bar distinction has no direct relevance to agreement in the theory I am developing.

162


searching downward, but the object is the closest NP to T searching upward. So some condition other than simple minimality is needed to determine that T agrees with the subject and not the direct object in this structure. One could stipulate that T needs to agree downward in IE languages, but that turns out to be inadequately general; upward agreement is perfectly fine in predicate adjective constructions, for example. The case-dependence condition in (2) does the trick, and proves useful in other domains as well.9 These object-preposing structures also confirm that the case condition in (2) does not apply in the Bantu languages. How case is assigned in the Bantu inversion constructions is a recalcitrant problem for P&P/Minimalist style theories, especially in the subject-object reversal construction. Suppose that one carries over from IE languages the assumption that the NP that agrees with T is the one that has nominative case, as Ndayiragije (1999:422–4) does. Then the thematic object gets nominative case in (9) and (10). But what case does the subject get? There is no comfortable answer. On the one hand, there is no reason to say that there is a second nominative case. On the other hand, it also seems wrong to say that it has accusative case. This would lead to a theoretically undesirable under-restriction of syntactic possibilities. Nor is there any empirical motivation 9 This does not answer the question of why it is the thematic subject that gets nominative case from T. That could be a straightforward minimality effect: the thematic subject is the closest NP to T prior to movement. Alternatively, it could be the result of a case-marking algorithm like the one sketched in Marantz 1991. I do not undertake a full investigation of how case is assigned here. In chapter 3, I discussed examples like (i), in which the T node in some nonstandard dialects of English agrees with the wh-phrase in Spec, CP examples that Kayne (2000) makes much of. (i)

?the people who Clark think – are in the garden

Now we have to ask in the light of parameter (2) how the matrix T realized on think can agree with the plural who in Spec, CP, even though that T does not assign nominative case to this wh-phrase (but rather to the subject Clark). Note, however, that the wh-phrase who in (i) does bear nominative case, assigned by the finite T in the embedded clause. This seems to be important to the relative acceptability of (i), since even speakers that accept (i) find (ii) worse (see Kayne 2000:210n.4). (ii)

*the people who(m) Clark think he saw – in the garden

In (ii) the wh-phrase bears accusative case, a different case from the one that the matrix T would assign, and agreement with this T is bad. (ii) is thus correctly ruled out by the parameter in (2). So case is relevant even to these marked examples in English. I tentatively assume that, because the wh-phrase in (i) is nominative, the same case that the matrix T assigns, this T can mistake the wh-phrase for being its own case assignee, allowing agreement. The fact that the actual assigner of nominative case is different from the reckoned assigner of nominative case might help account for why (i) is marginal and limited to particular dialects. (In the standard dialect, (i) is ruled out, its ungrammaticality perhaps attributable to the parameter in (2), interpreted strictly.)

Agreement on tense

163

for saying that the postverbal subject has accusative case (it cannot be realized as an “accusative” object clitic attached to the verb, for example). These problems are not so severe if we decouple case from agreement in the Bantu languages (following in part Ura (1996)) by saying that these languages are not subject to (2). Two possibilities can now be explored. The first is that the postverbal subject gets nominative case and the fronted object gets accusative case in Bantu, just as in the Yiddish example in (11). Case assignment would then be the same in both language families, although its relationship to agreement varies. The second possibility is that there is simply no requirement that NPs get case in Bantu languages. Either path avoids the artificial difficulties created by assuming that case and agreement are intimately related in all languages simply because they are related in some. A third difference between agreement on T in NC versus IE languages is seen in sentences in which no argument of the verb moves to Spec, TP, that position either being left empty or occupied by a (possibly null) expletive. In IE languages, the finite verb in such constructions still agrees with the nominative postverbal subject: (12)

a. There is/*are a peanut on the table. b. There are/*is some peanuts on the table.

In contrast, many Bantu languages show only a pleonastic locative agreement in similar constructions. Zulu seems to be particularly rich in these expletive subject constructions. In general, when the thematic subject is indefinite it can remain in Spec, vP, while the verb moves past it to T. When this happens, the verb does not agree with its subject, but rather displays pleonastic ho agreement (Doke 1955, 1963). In contrast, definite NPs do move to Spec, TP, and therefore come before the finite verb. These preverbal NPs do trigger agreement on the verb: (13)

a. Ku-khuluma aba-khulu. (Doke 1955:13–14) 10S-speak 1.pl-elders ‘There are speaking elders.’ b. Aba-khulu ba-ya-khuluma. 1.pl-elders 1.pl.S-disj-speak ‘(The) elders are speaking.’

Kinande has similar verb-subject sentences, although they are relatively restricted, being found only in negative or emphatic affirmative sentences. As in Zulu, the verb agrees with the preverbal subject but not the postverbal one:

164 (14)

Parameters of agreement mukali (omo-soko). a. Mo-ha-teta-sat-a aff-there-neg/past-dance-fv woman.1 loc.18-market ‘No woman danced in the market.’ b. Omukali mo-a-sat-ire (omo-soko). woman aff-1S/t-dance-ext loc.18-market ‘The woman danced in the market.’

See also Collins 2004 on existential constructions in Swahili, Horton 1949 on Luvale, and Ndayiragije 1999 on transitive expletive constructions in Kirundi, which have VOS order and no agreement on V with either S or O. Overall, the frequency and pragmatics of (expletive)-verb-subject constructions seems to vary across the Bantu languages, but the agreement pattern stays quite stable, consistently different from the familiar IE pattern. Examples like (13) and (14) show clearly that the position of the subject relative to T is crucial to whether T agrees with the subject in Bantu languages in a way that it is not in IE languages. There is of course no necessity that a language have any or all of these inversion constructions. The agreement parameters do not determine whether these constructions exist, but only what their agreement properties are when they do exist. The possibility of moving an object to Spec, TP, for example, seems to be quite restricted in both IE and NC languages: one cannot move the object to Spec, TP in the presence of a syntactic subject in English, nor as far as I know in Zulu, Luvale, or Chichewa. Similarly, expletive subject constructions are allowed with unergative verbs in some IE languages but not others, and the same holds true for NC languages. A limiting case is Lokaa, a non-Bantu NC language of Nigeria. Lokaa is typologically similar to Bantu languages in many ways, including the overall nature of its agreement system (Iwara 1982). But Lokaa happens not to allow locative inversion, object preposing, or expletive subject constructions, regardless of what agreement is used. Thus, there are close equivalents in Lokaa for ‘A fish is in the water’ and ‘Women buy cups,’ but not for ‘In the water is a fish’ or ‘There is a fish in the water’ or ‘Cups buy WOMEN’ (meaning ‘only women buy cups’) (Ijaja Eno and Alexander Iwara, personal communication). It is thus impossible to find evidence about the setting of the agreement parameters from this range of constructions in Lokaa. This point is of practical importance as we turn to broader typological investigation of agreement on T. On the one hand, inversion constructions may not be so common (or so commonly described), limiting the diagnostic utility of this test. On the other hand, when inversion constructions do exist, they provide some of the clearest evidence about how the agreement parameters are set.

Agreement on tense

165

5.2.2 Agreement on tense in other languages Now that we have solid motivation for the parameters in (1) and (2), let us consider how these parameters can be seen in a wider range of languages. Many languages are like the IE languages in that there is no clear relationship between word order and subject agreement. Alternative word orders might be possible, but they do not correlate with changes in what the finite verb agrees with. In head-initial languages, clauses in which the thematic subject has not moved to Spec, TP can often be recognized by the presence of verb-subject order, with or without something else appearing before the finite verb. Such orders are often the result of either the verb itself moving to T (see, for example, McCloskey 1996 on VSO order in Irish) or a projection of the verb phrase moving to Spec, TP (see Massam 2001 on verb-initial order in Niuean), while the subject stays in a position lower than Spec, TP. When the verb agrees with the subject in such configurations, it is an indication that downward agreement is allowed, the Direction of Agreement Parameter in (1) being set to “no.” Languages of this sort include Hebrew, Arabic, Finnish, Makah, Halkomelem, some Mesoamerican languages (Nahuatl, Otomi, Jakaltek), and some South American languages (Yagua, Guaran´ı, Wari, Wich´ı). In head-final languages, the order of subject and finite verb will not be so instructive. One can, however, compare agreement in Subject-XP-Verb clauses, where the subject comes before everything but scene-setting adverbs, with XPSubject-Verb clauses, where the subject is closer to the verb and something else has scrambled before it. At least some clauses of the latter sort could be instances of moving XP to Spec, TP, while the subject stays inside vP. If so, and if the finite verb still agrees with the thematic subject in the same way that it does when the order is Subject-XP-Verb, then downward agreement is possible in this language. A likely example of this sort is Amele (New Guinean): (15)

a. Uqa jo ceh-ade-i-a. (Roberts 1987:162) he house build-3pO-3sS-rpst ‘He built houses.’ b. Jo mel age doda cehe-(Ø)-gi-na. (Roberts 1987:142) house boy 3p self build-(3sO)-3pS-pres ‘The boys built the house themselves.’ c. Ija wen Ø-te-i-a (Roberts 1987:146) I hunger give-1sO-3sS-rpst ‘I am hungry.’

(15a) illustrates normal SOV order, with the verb showing subject agreement with the agentive subject (as well as object agreement with the direct object).

166


(15b) shows an alternative OSV order, a possible analysis of which is that the object moved to Spec, TP instead of the subject, as in Kinande and Yiddish. Nevertheless, the thematic subject still triggers third person plural subject agreement on the verb in this example. (15c) shows a psych predication, where the reversal of the thematic subject (the state-denoting noun ‘hunger’) and the thematic object (the experiencer ‘I’) is normal; here too the thematic subject controls subject agreement related to T rather than the thematic object, despite the surface positions of these NPs. Other SOV languages in which scrambling does not affect subject agreement include Khoekhoe, Basque, Abkhaz, Burushaski, Kannada, Choctaw, Ika, Quechua, and many New Guinean languages. So in many languages changes in word order do not induce changes in agreement, suggesting that downward agreement is permitted. There are other languages that do show a Bantu-like dependency of agreement on word order, however. The tense markers in Canela-Krahô are a case in point (Popjes and Popjes 1986). Canela-Krahô is largely head final, but T is a head-initial particle that remains separate from the verb (compare Vata on the analysis of Koopman (1984)). Some Ts apparently have an EPP feature that triggers movement of the subject to Spec, TP, whereas others do not. As a result, in some tenses one sees Subject-T-VP order (the simple past in (16)), and in others one sees T-Subject-VP order (the remote past in (17)). (16)

a. Wa i-te po pupun. (p. 176) I 1sS-past deer see ‘I saw a deer.’ b. Ca a-te ton. (p. 192) you 2sS-past make ‘You made it.’

(17)

a. Pê wa rop cakw˜ı. (p. 180) rem I dog beat ‘I beat the dog long ago.’ b. Pê ca to. (p. 192) rem you make ‘You made it long ago.’

(16) and (17) also show that there is a difference in agreement that goes along with the difference in word order. Some of the tenses that trigger NP movement to Spec, TP also agree with the moved subject, including the past marker te in (16). In contrast, the tenses that do not trigger NP movement to Spec, TP never show agreement with the subject, including the remote past marker pê in

Agreement on tense

167

(17). This suggests that T can only probe upward for something to agree with in Canela-Krahô, as in Bantu.10 Jarawara (Amazonian) is like Bantu in that T affixes to the verb, although it is like Canela-Krahô in being head final. Dixon (2004) describes the difference between actor constructions and object constructions in this language. In the actor construction, the agentive subject is the topic of the sentence, it is usually the first phrase in the clause (85% of the time), and the tense/mood marker agrees with it in gender. In the object construction, the object is the topic of the sentence, the object is usually the first phrase in the clause (73% of the time), and the tense/mood marker agrees with the object in gender.11 (18) shows one sentence of each type (Dixon 2004:418–19). (18)

a. Mioto Watati awa-ka. Mioto.m Watati.f see-decl.m ‘Mioto saw Watati.’ b. Watati Mioto hi-wa hi-ke. Watati.f Mioto.m 3/3.inv-see 3/3.inv-decl.f ‘Watati, Mioto saw.’

I thus claim that whether the object or the subject moves to the specifier of the highest functional head (Tense or Mood) in Jarawara determines whether the object or the subject triggers agreement on that head. Jarawara is minimally different from Amele in this respect, the alternation between SOV and OSV having no consequences for agreement in Amele (compare (15)). Apurinã provides another glimpse of how essentially the same distinction can show up in different ways depending on other syntactic properties of the language. Abstracting away from right dislocation of the object, Apurinã clauses alternate between the two very unusual word orders OSV and OVS, as shown in (19) (Facundes 2000:549–50). (19)

a. Oposo kema pitxi unawa muna txa. Then tapir penis they bring aux ‘Then they brought the tapir’s penis.’

10 Canela-Krahô also has Ts that trigger NP movement to Spec, TP but do not agree with the moved NP. This simply shows again that some members of a category might show agreement and others not in the same language. The point is that agreement is a live option if the NP c-commands T, but not if the T c-commands the NP. 11 This statement is slightly oversimplified: the higher head Mood always agrees with the “topic” in Jarawara, but the lower functional head Tense occasionally agrees with the thematic subject even when the object is fronted. See Dixon 2004 for a full description.

168

Parameters of agreement b. Txipoko-ru o-txima-ãpo-ta-pe hãtako-ro. Fruit-unposs 3S.f-eat.fruit-random-vbzr-perf youth-f ‘The girl would go eating fruit.’

Concomitant with this difference in word order is a difference in agreement: the verb bears a prefix that agrees with the subject in person, number, and gender in the OVS order in (19b), but not in the OSV order in (19a). I take this language to be the mirror image of Bantu. Heads come after their complements in Apurinã, and (unusually) specifiers are on the right edge of the phrase instead of on the left edge. Like Bantu, the verb moves to Tense. If the verb moves but the subject stays in Spec, vP, the word order is OSV as in (19a), just as moving V to T and leaving the subject in vP derives VSO order in various IE languages. In addition, there is the option of moving the subject from Spec, vP to Spec, TP, thereby placing the subject after the finite verb. This derives the OVS order in (19b), just as moving the subject to Spec, TP derives SVO order in Bantu and many IE languages. Finally, the moved subject in (19b) triggers agreement on T and the unmoved subject in (19a) does not; this shows that parameter (1) is set as “yes” in Apurinã. The Apurinã contrast in (19) is exactly like the Zulu contrast in (13), once the differences in word order are recognized. So far I have only considered languages in which one of the agreement parameters is active and the other is not. But one would expect the two parameters to be logically independent. For example, one can readily imagine a language in which T cannot agree with NP unless NP c-commands T and T values the case of NP. With this possibility in mind, consider (20) from Turkish (Kornfilt 1997:385). (20)

a. Haydut-lar köy-ü bas-mi¸s-lar. robber-pl village-acc raid-rep.past-3pS ‘They say that the robbers raided the village.’ b. Köy-ü haydut-lar bas-mi¸s. village-acc robber-pl raid-rep.past ‘They say that the robbers raided the village.’

In the normal SOV order shown in (20a), the definite subject triggers third person plural agreement on the verb. But nonspecific indefinite subjects stay inside the verb phrase in Turkish, showing up left-adjacent to the verb, as shown in (20b). Along with this word order constraint is the fact that these “low” subjects do not trigger agreement on the verb.12 Turkish is different in this respect from Amele, where the low subject continues to trigger agreement on the verb (see (15)). 12 Sentences like (21b) are usually called “subject incorporation” in Turkish. But there is little evidence that there is any real syntactic incorporation in these examples (although there is

Agreement on tense

169

This suggests that parameter (1) is set “yes” in Turkish. But Turkish is also different from Jarawara and Kinande in that the verb does not agree with the fronted object in the inverted OSV structures either. Moving the object to Spec, TP bleeds T-agreement with the thematic subject, but does not feed agreement with the thematic object. Why not? This can be attributed to parameter (2). The fronted object in (20b) clearly still has accusative case. The case feature of this NP is thus not valued by T. If the Case-Dependency of Agreement Parameter in (2) is also set “yes” in Turkish (but not in Jarawara), then it follows that T cannot agree with the object in this structure either. T thus has to be a default form in these “subject incorporation” sentences in Turkish. This then is a plausible instance of a language in which both agreement parameters are set positively. The typology of T-agreement would be complete if we found languages in which both agreement parameters were set “no”. The “no” value of parameter (1) would imply that agreement does not change with word order. A “no” value of parameter (2) would imply that agreement does not change with changes in case marking either. A language that fits this description nicely is Burushaski, an isolate spoken in the Himalayas (Lorimer 1935). Burushaski has an ergative case system, in which the subject of a transitive verb has a different case marking than the subject of an intransitive verb. However, subject agreement on T is not sensitive to this variation in case; both the nominative subject in (21a) and the ergative subject in (21b) trigger the same a form of agreement on the auxiliary and main verb. (21)

u:¸nε xidmt ε cˇ -a b-a. (p. 317) a. Jε I.nom your service do-1sS be-1sS ‘(For these many years) I have been at your service.’ b. Ja be.dpi.ε n ε t-a b-a. (p. 321) I.erg discourtesy do-1sS be-1sS ‘I have committed a discourtesy.’

Burushaski is markedly different from Hindi in this respect, where verbs agree with nominative subjects but not with ergative ones, as shown in (22) (Mohanan 1995:100–1).13 “semantic incorporation” in the sense of van Geenhoven 1998). Even if indefinite, vP-internal subjects in Turkish are incorporated in some sense, that does not by itself explain why the verb cannot agree with them; indefinite objects in Hindi undergo a similar kind of (pseudo) incorporation, but the verb can nevertheless agree with them (Mohanan 1995). 13 Also like Burushaski in this respect is the IE language Nepali (Bickel and Yadava 2000:348; Bobaljik to appear). Bobaljik emphasizes the fact that T in Nepali can agree with an ergative case subject but not a dative case subject. This is important motivation for his claim that agreement is sensitive to

170 (22)

Parameters of agreement baalak-ko ut.h aa-eg-ii. a. Niina Nina.f.nom boy.m-acc lift-fut-f.sg ‘Nina will lift up the boy.’ b. Niinaa-ne baalak-ko ut.h aa-y-aa. (*uthaa-y-ii) Nina.f-erg boy.m-acc lift-perf-m.sg lift-perf-f.sg ‘Nina lifted up the boy.’

Hindi is a typical IE language in which agreement is dependent on case valuation in the way described in (2). The fact that Burushaski behaves differently suggests that its parameter (2) is set “no.” But agreement is also not dependent on word order in Burushaski. (23) shows an example with OSV order, where the object not the subject has moved to the left-periphery; nevertheless, the agreement suffix still registers plural agreement with the thematic subject. (23)

ε -sqnu-mn. Ja au.u Pisn-kuts-ε My father Pisan-people-erg 3sO-kill-3pS ‘The people of Pisan slew my father.’

Hence, agreement is not dependent on either structural position or case-sharing in Burushaski, the fourth logical possibility.14 Other languages of this type include Georgian and Warlpiri, both of which are languages with free word order, in which ergative subjects and nominative/absolutive subjects trigger the same agreement on the verb: (24)

ˇ a. Sina.ber-a jagl-s jval-s mi-s-c-em-s. (Georgian) spinster-nom dog-dat bone-dat(acc) prev-3sO-give-t-3sS.impf ‘The spinster will give the dog a bone.’ (AgrS with nominative ubject) ˇ b. Sina.ber-am jagl-s jval-i mi-s-c-a. (Hewitt 1995:549) spinster-erg dog-dat bone-nom prev-3sO-give-3sS.perf ‘The spinster gave the dog a bone.’ (AgrS with ergative subject)

a three-way distinction among unmarked case, dependent case, and oblique case (cf. Marantz 1991). I interpret the facts differently. According to my typology, languages that agree with ergative as well as nominative should be the same as those that agree with dative as well as accusative, and there is some evidence that this is true (see sections 5.7.2 and 5.11.1). Rather I accept Bickel and Yadava’s arguments that dative experiencers in Nepali are not subjects for purposes of control and raising. T fails to agree with them not because they have the wrong kind of case, but because they are in the wrong syntactic position (consistent with them being PPs, and hence phase barriers to agreement). 14 This discussion leaves open what does determine which NP T agrees with in a “no-no” language like Burushaski. Agreement in such languages is not random and unconstrained. The easiest answer would be to say that T simply probes downward in the pre-movement structure, agreeing with the first NP it finds – the thematic subject in Spec, vP – regardless of how it gets case or whether it moves. I do not know any of these languages well enough to know if there are problems with this simple hypothesis.

Agreement on FA and the formulation of the parameters (25)

171

wangka-ja. (Warlpiri, Simpson 1991:150) a. Ngaju Ø-rna I.abs past-1sS speak-past ‘I spoke.’ (AgrS with nominative subject) b. Ngajulu-rlu ka-rna nya-nyi kurdu. (Simpson 1991:100) I-erg pres-1sS see-npst child ‘I see the child.’ (AgrS with ergative subject)

I conclude that the two parameters in (1) and (2) give us a four-way typology of the agreement properties of Tense, and that this matches well with the kind of variation that we observe in the behavior of “subject” agreement in the languages of the world. The next task is to see whether the same parameters apply to other functional heads. 5.3

Agreement on FA and the formulation of the parameters

Next let us consider FA , the functional head that realizes agreement on adjectives. This is something of an interlude to the main line of discussion, because in fact there is little or no observable difference in adjectival agreement between the NC languages and the IE languages.15 In both families, predicate adjectives agree with the subject of predication in number and gender (but not person) and attributive adjectives agree with the noun that they modify in number and gender (but not person), as seen by comparing (26) in Kinande with (27) in Spanish. This has been discussed at length in the previous chapters. (26)

a. Aba-kali ni ba-kuhi. 2-women pred 2-short ‘The women are short.’ b. aba-kali ba-kuhi 2-women 2-short ‘the short women’

(27)

a. L-as mujer-es son alt-a-s. the-f.pl women.f-pl are.3pS tall-f-pl ‘The women are tall.’ b. l-as mujer-es alt-a-s the-f.pl women.f-pl tall-f-pl ‘the tall women’

15 It is conceivable that more data might reveal some second-order differences in adjectival agreement between NC languages and IE languages. For example, I would predict that a raising adjective could not agree with an NP that stays within its CP complement in a Bantu language, the way it can in Icelandic (section 3.1.2). But raising adjectives are rare crosslinguistically, so there may well never be one in this family, making it impossible to check the prediction.

172


Even though FA does not reveal the existence of the parametric differences in (1) and (2), these structures do have important implications for how the parameters in (1) and (2) are stated. It is worth interrupting the presentation of the parametric differences, then, to make sure that we have an accurate formulation of these parameters. First consider the Bantu languages. The facts surveyed in the previous section have been known for some time, and others have discussed them. The most common interpretation has been that all agreement in Bantu languages is agreement between a specifer and the functional category that it is the specifier of. The theoretical context and exact way of formulating this have varied over time. For Kinyalolo (1991) and others in the early 1990s, agreement was always the effect of a specifier–head relationship holding at some level. For him, there was no parameter of agreement per se; Kilega simply showed more perspicuously what is fundamentally true of all languages. IE languages were assimilated to the Bantu case by positing a process of expletive replacement via subject raising to Spec, TP at LF, so that an agreement configuration between the nominative NP and T is held by LF in IE sentences like (12) (Chomsky 1986b). More recent work cast in terms of Chomsky’s (2000, 2001) theory of Agree puts this another way. Baker (2003c), Collins (2004), and Cartsens (2005) have all stated parameters roughly like the following: (28)

X agrees (downward) with Y only if Y satisfies an EPP feature of X.

On this view, functional heads agree with NPs inside their c-command domain in Bantu languages, just as in IE languages. However, whenever there is ϕ-feature checking between a head and a nominal, there must also be EPP checking. This means that the agreed-with NP raises to become a specifier of the agreeing head subsequent to the agreement relationship being established. The end result is much the same as the earlier theory: agreement holds only between a specifer and its head in Bantu. The adjectival constructions in (26) show that this cannot be quite right, at least if Baker (2003a) and the current work are at all right about the structure of adjectival phrases. The subject of predication in (26a) is not in a specifier– head relationship with FA (or A) at any stage of the derivation; rather it is base-generated higher, in Spec, PredP. In my view, this is crucial to explaining the fact that most predicate adjectives do not behave like thematically similar unaccusative verbs with respect to unaccusativity diagnostics. I have also used it to explain the fact that adjectives cannot agree with their subjects in first and second person, although thematically similar verbs can. These and other syntactic differences between adjectives and verbs are unexplained if we say


173

that the subject of predicate adjectives can be the specifier of FA P/AP. So (28) cannot be correct, within this framework of assumptions. Instead, predicate adjective constructions show that functional heads can probe upward for something to agree with as well as downward. The more general way to state the parameter that distinguishes Bantu and IE languages, then, is to say that functional heads always look upward for something to agree with in Bantu, never downward. This covers the canonical examples in the previous section, because the specifier of a phrase c-commands the head of that phrase (and not vice versa). The formulation in (1) is also a little broader than the one in (28), in that it allows agreement in adjectival constructions such as (26). This is why I take (1) to be superior to alternatives that have been proposed in the literature. Consider next adjectival agreement in IE languages, and the proper formulation of the Case-Dependency of Agreement Parameter in (2). To succeed in my goal of accounting for concord within the same theory as subject–verb agreement, the Activity Condition of Chomsky (2000, 2001) needs to be considered more carefully. The core case of agreement for generative theories has always been agreement between the subject and the finite verb. A robust fact about IE languages from Icelandic to Hindi (but not Nepali) is that the finite verb only agrees with a nominative case noun phrase. This generalization is captured by Chomsky’s Activity Condition: X agrees with Y only if Y has an unchecked case feature – which is then checked/valued (as nominative) by X (here Tense). This insures that the finite verb only agrees with the subject, regardless of details about where the subject appears. This can also be extended smoothly to certain other situations: for example, the transitive verb (technically v) might only show object agreement with the NP that it licenses accusative case on, the possessed noun (technically D) might only show agreement with the NP that it licenses genitive case on, and an adposition might only show agreement with the NP that it licenses oblique case on. But the situation is somewhat different when it comes to agreement on adjectives. The predicate adjective in (27a) agrees with the subject, even though it does not determine the case of the subject (rather the inflected copula does). In another environment, the same A could perfectly well agree with an NP that gets accusative case – for example in the exceptional case-marking structure shown in (29b) from Icelandic (Thráinsson 1979:361). (29)

a. Mar´ıa er góð . Maria.nom is good.f.sg.nom ‘Maria is good.’

174

Parameters of agreement ´ tel b. Eg Mar´ıu vera góð a/ * góð . I believe Maria.acc to.be good.f.sg.acc good.f.sg.nom ‘I believe Maria to be good.’

Why is case assignment required for agreement with T in IE languages, but not for agreement with FA ? Previous work has tended to say that there is something defective about the agreement on adjectives, picking up on the fact that they do not show person agreement. For example, Chomsky (2000, 2001) claims that an NP must have an unvalued case feature to be visible for adjectival agreement as well as for verbal agreement. The difference is that because adjectival agreement is incomplete, lacking a person feature, it does not value the case feature of the associated NP, whereas T is “ϕ-complete” and does value the case feature of NP. This proposal is designed to relate two differences between agreement on adjectives (and participles) and agreement on verbs: the fact that adjectives do not determine the case of the NP they agree with, and the fact that agreement between an adjective and an NP does not prevent another head from agreeing with that same NP (see (27a), where there is agreement also on the verbal copula). On this view, then, adjective agreement is seen as a kind of degenerate case, where case-checking is attempted by Agree but does not in fact take place. But this view about the distinction between adjectival and verbal agreement unravels when one attempts to derive rather than stipulate the fact that verbal heads agree in person but adjectival ones do not – a desirable move for the reasons laid out in chapters 1 and 2.16 In contrast, I want to foreground an obvious fact about adjectival agreement and case in IE languages that is at best peripheral to Chomsky’s account.17 There is clearly an important case relationship between the NP and the adjective that 16 See also Carstens 2001 for criticism of Chomsky’s idea that adjectives and participles do not assign case because they are defective agreers, based on data from Bantu languages similar to that considered in this chapter. Carstens shows that participles in Bantu languages show ϕ-complete agreement (see section 5.9.1); nevertheless, they do not assign case to the subject NP or bleed further agreement between the NP and higher auxiliary verbs any more than ϕ-incomplete participles in IE languages do. Conversely, she shows that finite Tense is not really ϕ-complete in IE languages, because it does not show gender agreement. Nevertheless, finite Tense does assign case and bleeds the establishment of further agreement relationships. Carstens’s conclusion is that the case-assigning properties of functional heads are not predictable from the type of agreement that those heads enter into. She thus urges a return to the older view that some functional heads are simply stipulated as being case assigners (finite T, transitive v) whereas others are not (FA , participial inflection). I follow Carstens in this respect. 17 In one of the few passages in which Chomsky talks explicitly about case concord (Chomsky 2001:17–18), it works as follows (for participle constructions in Icelandic). First the participle agrees with the NP in ϕ-features, but there is no case valuation either way. Then a higher functional head (say T) probes its domain to find something with ϕ-features it can agree with. The participle head has a partial set of ϕ-features (number and gender, but not person) so the T


175

agrees with it in (29). Although it is true that the adjective does not assign case to the noun, the adjective does agree with the noun in case. Intuitively speaking, the case relationship here has the features flow the other way. With T–NP agreement, the inherent nominative feature of finite T becomes a feature of the NP as well; we call this case assignment. With FA –NP agreement, the nominative or accusative feature already associated with the NP (assigned by T or v) becomes a feature of the FA P as well; we call this case concord. But the two processes can both be considered instances of a more general process of case co-valuation (or case feature unification), in which a category that did not have a case value acquires one. This is why (2) was stated as it was, repeated here as (30). (30)

F agrees with DP/NP only if F values the case feature of DP/NP or vice versa.

Adding “or vice versa” to the end of this condition generalizes it from ordinary instances of subject–verb agreement to cover adjective–noun concord as well. This innovation is parallel to the innovation I proposed in section 2.3.2 with respect to the c-command condition on agreement. In standard treatments, F must c-command NP to agree with it, but I argued that NP c-commanding F is also sufficient. What is crucial is that the two stand in a c-command relationship, not the direction of that relationship. Similarly, the standard view has been that F must value the case feature of NP in order to agree with it. Now I am claiming that NP valuing the case feature of F is also sufficient. Once again, the essential requirement (for IE languages) is that there be a case-sharing relationship, not which way the determination of case goes. This idea extends easily to attributive adjectives. Why can an attributive adjective agree with the NP that it is adjoined to in an IE language, even though it does not determine the case of that NP? The answer is that this is possible because the modified NP determines the case of the adjoined adjective. The

agrees with it, but it also keeps probing to find a more perfect match. Eventually it finds the NP, and agrees with that as well. Since T agrees with both the participle and the NP, it assigns the same case (nominative) to both. On this conception, there is no direct case concord between NP and the adjective or participle; both get the same case by agreeing with the same verbal head. A technical problem arises in extending Chomsky’s account to predicate adjective constructions like (29), which also show case concord. This time when the T associated with the copular verb probes downward, the first ϕ-feature-bearer it encounters is the NP subject of the predication. Since this has a complete set of ϕ-features, it is not clear why the T should continue to probe downward to find the additional ϕ-features on FA . Thus, it is not clear why there should be case concord in this construction. Nor do Chomsky’s remarks about case concord have anything to offer in explaining the difference between primary and secondary predicates shown in (31).

176


idea, then, is that agreement in ϕ-features depends on sharing of case features in a certain class of languages. Support for this view comes from some details about agreement with quirky case subjects in IE languages. Sigurð sson (2002) points out that some adjectives can agree with such subjects, whereas others cannot. More specifically, when the adjective is the main predicate of the sentence, it cannot agree with a dative case subject. In contrast, adjectival secondary predicates can agree with the dative case subject in case, number, and gender. Agreeing and nonagreeing adjectives appear side by side in the same sentences in (31) (Sigurð sson 2002:709-10). (31)

a. Strákunum var mjög kalt, svona fáklæddum. boys.the.dat. m.pl was.3sS very cold.n.sg so few.clad.dat.pl ‘The boys were very cold so scantily dressed.’ b. Henni verð ur kalt svona fáklæddri. her.dat. f.sg becomes.3sS cold.n.sg so few.clad.dat. f.sg ‘She will be cold so scantily dressed.’

Why can ‘few-clad’ agree with the subject in (31), whereas ‘cold’ cannot? Part of the answer I adopt from Sigurð sson’s own discussion. He suggests that a predicate X cannot agree in case with Y if X determines quirky case on Y. He writes (p. 710): The reason for this difference has to do with case-assignment or case-matching. A primary predicate with a non-nominative subject is itself an assigner or matcher of the inherent case, whereas a secondary predicate of an argument so case-marked is not its case-assigner. That is, a predicate cannot agree with its own case assignee.

It is a lexical property of the adjective kalt in Icelandic that it selects a dative case subject. Since kalt is responsible for the dative case on the subject, it cannot turn around and agree in case with that subject. Case can flow either direction, from the adjective to the NP or from the NP to the adjective, but it cannot flow both directions at once, from the A to the NP and back to the same A. In contrast, the secondary predicate ‘few-clad’ in (31) is not responsible for the dative case on the subject; it can equally well be predicated of a nominative subject. Dative case does not flow from the secondary predicate to the subject; therefore it can flow from the subject to the secondary predicate. Suppose that we take this generalization of Sigurð sson’s to be a theorem of the true case theory. My primary interest here is to demonstrate the nature of (2), which expresses the dependence of ϕ-feature agreement on case in IE languages. The examples in (31) show this dependency rather clearly. When case


177

theory allows the adjective to agree with the dative subject in case, the adjective also agrees with the subject in number and gender. In contrast, when case theory forbids the adjective to agree with the dative subject in case, agreement in number and gender also fails. Here we see the intimate relationship between case and agreement in IE languages in a very different way from the familiar relationship between subject agreement and nominative case. The parameter in (2) captures the fact that these two types of case dependence occur in the same class of languages, a pay-off of the quest to unify the theories of agreement and concord. Vita Markman (personal communication) and Natalia Kariaeva (personal communication) tell me that the same distinction between primary predicates and secondary predicates is found in Russian and Ukrainian as well. This suggests that (31) is not caused by some of the rather unusual properties of quirky subjects in Icelandic, but may illustrate principles of agreement that hold more generally of IE-type languages.18 With these clarifications and justifications of (1) and (2) in place, let us return to the claim that they hold parametrically, not universally. If that is correct, then we should not see the Icelandic contrast between primary predicates and secondary predicates in languages that are not subject to (2). The Bantu languages are not directly relevant to this issue, because they do not have any overt case marking. But Georgian does have morphologically marked case, and it is like Bantu in not obeying (2). In particular, verbs agree with their subjects in Georgian regardless of whether they are in nominative case or ergative case (see (24)). So agreement does not depend on the sharing of a certain kind of case in Georgian, as it does in IE languages.19 Now predicate adjectives agree with their antecedents in case and optionally in number in Georgian, both when they are primary predicates ((32a)) and when they are not ((32b–c)).

18 However, Natalia Kariaeva (personal communication) points out that quirky dative subjects are less unambiguously subjects in Russian and Ukrainian than in Icelandic (see section 3.3.2 for some discussion). Given this, it could be that primary predicates agree with a null expletive subject, and the dative expression has simply been topicalized to clause initial position. If so, the Slavic analog of (31) does not give crucial support for (2). See Baker and Kariaeva 2006 for an analysis of instrumental case predicates in the East Slavic languages within this theory. These instrumental predicates seem to show ϕ-feature agreement with their subjects even though they do not agree with them in case. The analysis (due to Kariaeva) is that the instrumental predicates really have a PRO subject, and they agree with this subject in default case (spelled out as instrumental on As) and hence also in ϕ-features. 19 Other signs of Georgian’s “no” value for parameter (2) are its having full person agreement on both the auxiliary and the main verb in some complex tenses (section 5.9.2) and its object agreement not being sensitive to whether the object has dative or nominative case (section 5.7.2).

178 (32)

Parameters of agreement ˇ a. Cven uk’ve did-(eb)-i vart. (Hewitt 1995:50) we.nom already big-(pl)-nom be.1pS ‘We are already big.’ b. K’ar-s nu da-t’ov-eb gia-s. (Hewitt 1995:562) door-dat neg prev-leave-fut open-dat ‘Don’t leave the door open.’ c. Me, sˇe-ˇsin-eb-ul-ma, mo-v-Ø-k’al-i. (Hewitt 1995:50) I.erg alarmed-erg prev-1sS-3sO-kill-aor ‘It was I, alarmed, who killed it.’

This is no different from Icelandic. Georgian also has dative subject constructions that are induced by lexical properties of the primary predicate. But unlike IE languages, primary predicates can agree with dative subjects in Georgian, as shown in (33).20 (33)

es saˇcukar-i. a. Bavˇsv-eb-s mo-(Ø)-e-c’on-eb-a-t child-pl-dat prev-3O-appl-like-fut-3sS-pO this present-nom ‘The children will like this present.’ (Hewitt 1995:556) b. g-civ-a; g-cx-el-a (Hewitt 1995:382) 2sO-cold-3sS 2sO-hot-3sS ‘you are cold’ ‘you are hot’ (Literally: ‘it hots to you’)

This is what we expect given that (2) does not hold in Georgian: predicates can agree with NPs in person and number, whether they agree with them in case (as in (32)) or not (as in (33)). 5.4

Agreement on complementizers

Next let us survey the effect that the parameters in (1) and (2) have on other types of agreement, beginning with the relatively rare case of agreement on C. 5.4.1 C-agreement in NC and IE languages We already saw in sections 4.1.5 and 4.5 that there is person agreement on C both in certain NC languages and in certain IE languages. But what is agreed with is different in the two language families. In the NC languages, agreement 20 The details of agreement in Georgian – including theoretical questions about how T and v contribute to that agreement – are notoriously complex. Note in this regard that the agreement with dative subjects in (33) is formally “object” agreement, not “subject” agreement. This interesting fact needs to be explained in a fuller treatment, but should be orthogonal to the point at hand. Also the primary predicates in (33) are apparently verbs, not adjectives, but that too should also be irrelevant; verbs that take quirky case subjects do not agree with those subjects any more than adjectives do in (2)=“yes” languages like Icelandic.


179

seems to be with the matrix subject of a verb of speech (Kinande; see also section 4.1.5 for Lokaa examples):21 (34)

ba-ti Kambale mo-a-gul-ire eritunda. a. Mo-ba-nyi-bw-ire aff-2S-1sO-tell-ext 2S-that Kambale aff-1S-buy-ext fruit ‘They told me that Kambale bought fruit.’ b. Mo-n-a-layir-ire Kambale in-di a-gul-e amatunda. aff-1sS-t-convince-ext Kambale.1 1sS-that 1S-buy-sbjn fruits.6 ‘I convinced Kambale that he should buy fruits.’

In the West Germanic languages, agreement is with the subject of the embedded clause – the same NP that the lower verb agrees with: (35)

die boeken te diere zyn. (West Flemish) a. Kvinden dan I.find that.pl the books too expensive are ‘I find that those books are too expensive.’ (Carstens 2003:393) b. datt-e wiej noar ’t park loop-t (Hellendoorn) that-pl we to the park walk-1pS ‘that we are walking to the park’ (Carstens 2003:397)

It is appropriate, then, to ask if this difference in the agreement behavior of complementizers is related to the known difference in agreement on T between the two language families. To answer this question, we need to add in more details about the structures. Given the analysis in chapter 4, the complementizer in Kinande actually agrees with a logophoric operator in Spec, CP. This logophoric operator is controlled by the subject of the matrix verb, and therefore inherits the person, number, 21 As in many languages of Africa and elsewhere, the complementizer in (34) is cognate with the verb meaning ‘say.’ One could thus question whether (34) really shows complementizer agreement, or whether these are merely instances of subject agreement on the second verb of some kind of complex verbal construction. The problem with this alternative is that it is very unclear what the syntactic relationship between the second VP and the first could be. One can extract wh-phrases out of the embedded clause in Kinande, just as one can from complement clauses in English: (i) Eki-hi ky-o Kambale a-ku-bwir-a a-ti n-a-gul-a? 7-what 7-foc Kambale 1S/t-2sO-tell-fv 1-that/say 1sS-t-buy-fv ‘What did Kambale tell you that I bought?’ This rules out the possibility that ti heads a VP that is conjoined with or adjoined to the VP headed by ‘tell’; if so, (i) would violate the Coordinate Structure Constraint or the Adjunct Island Condition. Conceivably (34) could be some kind of serial verb construction, but Kinande does not otherwise have such constructions. I therefore assume that the most plausible analysis is that ti heads a CP and its relationship to the verb ‘say’ is purely historical. (The same issues arise with respect to the Arapesh example in (40) below.)

180


and gender features of that subject. The syntactic representation of (34b) is thus approximately (36). (36)

[Ii convinced Kambalek [CP Logi C [TP hek T-buy fruits]]]

In contrast, the complementizer in West Flemish agrees with the subject below it. This by itself is enough for simple number–gender agreement. When first or second person agreement is seen, I claimed in section 4.5 that the agreement gets enriched with first and second person features by virtue of C’s proximity to the S (speaker) or A (addressee) operator generated in CP. The C counts as a variable indirectly dependent on this operator, given that Agree makes it a variable dependent on the lower subject and that subject is first person in (35b), hence bound by S. So schematic structures for (35) are as in (37). (37)

a. [Ii find [CP Si that [TP the booksk too expensive are-T]]] b. [(Theyk think) [CP Si that [TP wei to park walk-T]]]

The first parametric question that these structures raise is why can’t C agree with S or A in IE languages when the lower subject is something other than a first or second person pronoun? A plausible answer is that this is a reflex of the case dependence of agreement in IE languages. Many researchers have proposed that finite that-type complementizers participate with finite Tense in the assignment of nominative case (Stowell 1981, Watanabe 1996, Pesetsky and Torrego 2001). Without settling on a particular implementation, suppose that some idea in this family is correct. Then, given (2), C agrees with X only if C assigns (or, better, contributes to the assignment of) nominative case to X. That is true for C and the lower subject. But it is not true for C and the S, A, or Log operators in CP. As null elements with no thematic role, these operators presumably have no need for case. Therefore, complementizers in IE languages cannot agree with these operators directly; they can only agree with them indirectly, if they bind the NP that C is in a case relationship with. The impossibility of examples in IE that are perfectly analogous to (34) is thus another instance of the dependence of agreement on case valuation in IE languages. Consider now Kinande. The complementizer in this language can agree with an operator in CP without there being any case relationship, because (2) does not hold in Bantu languages. Rather, the agreement parameter that is relevant to Kinande is (1), which implies that the C must be asymmetrically c-commanded by what it agrees with. This means that the Bantu C could only agree with an operator in Spec, CP or some similar CP-peripheral position. It could not agree with (say) a third person subject that is internal to the IP complement of C and


181

distinct from the matrix subject. Such a noun phrase would not c-command C, and it would have no relationship to any operator immediately contained in CP. So a structure like (36) is possible in Bantu, but one like (37a). In summary, we see that IE complementizers can agree with nominative subjects but not (directly) with operators in Spec, CP, and Bantu languages can agree with operators in Spec, CP but not (directly) with subjects. On my analysis, this follows from the same parameters that force T to agree with its specifier in Bantu but with the nominative subject in IE languages. T and C thus undergo the same parametric variation. This suggests that the agreement parameters are not keyed just to one particular functional head, but influence whole languages. Kinande also has a second type of complementizer agreement. In focus/whmovement constructions there is a C-like particle that comes after the fronted NP and before the subject of the clause and the finite verb. This particle agrees in number and gender with the fronted NP: (38)

omukali. a. Eritunda ry-o n-a-h-a fruit.5 5.foc 1sS-t-give-fv woman.1 ‘It’s a fruit that I gave to a woman.’ b. Amatunda w’ omukali a-gul-a. fruits.6 6.foc 1.woman 1S/t-buy-fv ‘It’s fruits that the woman bought. c. Ebi-hi by-o Kambale a-gul-a? 8-what 8-foc Kambale 1S/t-buy-fv ‘What did Kambale buy?’

This is entirely consistent with what we know about agreement parameters in the Bantu languages. Given that it is sitting in Spec, CP (or Spec, FocP), the focused NP c-commands the C/Focus head, making agreement with that head possible. In contrast, the Direction of Agreement Parameter predicts that a complementizer in Kinande could not agree downward with (say) a whphrase that remained in situ. (39) shows that this is correct (compare with the synonymous (38c)).22

22 See Carstens 2005 for discussion of the same asymmetry in Kilega, where C agrees with fronted wh-phrases but not with unfronted ones. The Kilega examples are less transparent, however, because the finite verb also moves to C, with the result that agreement on C is hard to distinguish from subject agreement on the finite verb (Kinyalolo 1991, Carstens 2005). If anything, the question particle in (39) agrees with the A operator in CP, since uti comes historically from u-ti (2sS-say) ‘you say.’ This could be just a frozen form, however, not a true instance of agreement.

182 (39)

Parameters of agreement Uti/*bi-ti Kambale a-gul-a ebi-hi? q/8-q Kambale.1 1S/t-buy-fv 8-what ‘What did Kambale buy?’

There is presumably no case relationship between the Focus head and the focused NP in its specifier, that NP getting case in the ordinary object position prior to wh-movement (if case assignment happens in Bantu at all). But case is not an issue, because we know that agreement is not contingent on case in Bantu languages. In contrast, +wh or +Focus complementizers in IE languages do not agree in ϕ-features with the moved phrase. That follows from (2), the stipulation that agreement is contingent on a case-valuation relationship in the IE languages.23 Again it must be pointed out that languages can perfectly well be subject to the parameters in (1) and (2) without this showing up on their complementizers. In fact, agreeing complementizers seem to be fairly rare. In the whole IE family, C-agreement is only known to occur in a few continental West Germanic languages. C-agreement also seems to be fairly rare in NC languages: Chichewa, Zulu, Swahili, and Luvale do not have the C-agreement found in Kinande, for example. Like Kinande, the complementizer in these languages is historically related to the verb ‘to say,’ but in these languages it is the infinitival form of the verb that is used, a form that does not vary with the ϕ-features of operators in CP. This shows us once again that the grammatical parameters do not determine whether a complementizer agrees or not, but rather what it agrees with if it does agree. 5.4.2 C-agreement in other languages A look at C-agreement outside of the IE and NC languages provides some reinforcement for these themes. Most languages simply do not have C-agreement, and this gives no information about how the agreement parameters are set in that language. I found agreement on C in at most 9 of the 108 languages I surveyed. Among the few languages that have C-agreement, the one that most clearly has Bantu-like upward agreement with an operator in CP is the New Guinean language Arapesh (Fortune 1977:98): 23 In some approaches to complementizer-trace phenomena, it is crucial that IE Cs agree with their specifiers so that they can license a trace in the subject position immediately below them (Rizzi 1990). Such an approach is not necessarily incompatible with my parameter. The instances of C that need to have agreement in order to ameliorate complementizer-trace effects are always in structures that have local subject extraction. The finite C participates in assigning nominative case to the extracted subject, so the IE agreement condition in (2) is satisfied. Other wh-phrases could not trigger agreement on C in an IE language, if I am right, but their traces are also not at risk of creating a complementizer-trace violation.

Agreement on complementizers (40)

183

akir urur re-igure. De ja-kalipw-e ja-ka now 1sS-tell-2sO 1sS-that(say) your vulva 3sS-cry.out ‘I told you that your vulva cried out.’

Agreement in Arapesh is like agreement in the Bantu languages in other respects as well, as I show below. Another possible instance is the so-called allocative agreement in Basque, where matrix verbs can show a kind of second person agreement with the addressee of the sentence, in addition to standard agreement with the subject, object, and direct object (Ortiz de Urbina 1989:62n.6). (41)

Z-e-ki-na-t (compare: d-a-ki-t) 3A-past-know-2s.dat-1sE 3A-pres-know-1sE ‘He knows me, y’know.’ ‘He knows me.’

This could be analyzed as agreement between the matrix C and the A operator at the head of the sentence, which is realized on the inflected verb as a result of V-to-C movement. In certain other languages, agreement on C seems to be downward, with the subject of the TP complement of C. The New Guinean language Daga has no C with indirect speech complements, but in relative clauses there is a second suffix on the verb, in addition to T, that (like T) agrees with the subject of the relative clause in person and number. Some examples are (Murane 1974): (42)

Dani tu-n-i a. pa house Dani build-3sS.past-3s.that ‘the house that Dani built’ b. at gega gat yon-an-a place you just stand-2sS.past-2s.that ‘the place where you stood’

Another language with downward agreement on C could be Jakaltek (Craig 1977), assuming that the very high “aspect” head that bears agreement with the absolutive argument in this language is a type of C, a head higher than T. Next let us consider what patterns of C-agreement we would expect to see in languages that set both agreement parameters as “yes” or both as “no.” In a “yes-yes” language, a C would agree with an NP only if that NP was in Spec, CP and if C was in a case-valuation relationship with that NP. But Cs don’t generally enter into case relationships of their own: they neither assign case (apart from T) nor manifest it in relationships of case concord. To the extent that this observation is universal, a “yes-yes” language could not have agreement on Cs. That might be a mildly positive result in some cases. For example, I claimed in section 5.2.2 that Turkish was a “yes-yes” language, based on the behavior of T agreement. It is also true that Turkish does not have

184


agreement on C, as predicted. But this is hardly a stunning confirmation, given that C agreement is rare in languages of all types. What about a “no-no” language, in which agreement can be downward and is not sensitive to case? Cs in that sort of language could agree with NPs inside their CP complement, but they would not necessarily be restricted to agreeing with the nominative subject, the way Cs in IE languages are. One language that might take advantage of this freedom is Nez Perce. In addition to the agreement for subject and object found on finite verbs (see section 5.11.2.2), this language has a variety of initial particles that mark negation, questions, and relative clauses (Aoki 1973:128-9). These C-like particles agree with something inside the clause that they introduce. The agreed-with NP is often the subject, as in (43a) and (43b), but question and relative particles can agree with the object instead of the subject when the subject is third person and the object is first or second person, as shown in (43c). (43)

a. M´ıʔ s-e:x ʔ´ı:n manmaʔ´ı Ø-ki-yúʔ . not-1s I anyway 1sS-go-fut ‘I cannot go in any way.’ b. We:t-e:x ʔ e-né:s-tiw´ıkce? q-1s 1S/3O-pl-accompany ‘Shall I go with them?’ c. We:t-e:x hi-tw´ı:kc-ix? q-1s 3S/1O-accompany-pS ‘Would they go with me?’

Thus, agreement on Cs in Nez Perce is not restricted by case or by direction. Overall, then, we find the same four types of agreement with complementizers that we do with Tenses, even though C-agreement is much rarer, with the caveat that agreement on C cannot arise in a “yes-yes” language. 5.5

Agreement on determiners

Next I consider the possibility of agreement on determiners and similar functional heads in the extended projection of NP, beginning as usual with NC and IE languages. 5.5.1 D agreement in Bantu and IE languages A widespread pattern in Bantu languages is for words inside the noun phrase to follow the noun and agree with it in noun class (gender and number). This is what happens with adjectives, for example, in Kinande, Swahili, Zulu, Chichewa, Luvale, and others:

Agreement on determiners (44)

185

aba-kali ba-kuhi (Kinande) 2-woman 2-short ‘a/the short woman’

I have analyzed these adjectives as adjuncts to NP. A sign of their adjunct status within my theory is that they cannot agree in person with a first or second person pronoun, although they do agree with such pronouns in number and gender (see sections 2.4.2 and 4.5): (45)

Thin’ aba-khulu si-ya-khuluma. (*si-khulu) (Zulu, Doke 1955:51) we cl1.pl-big 1pS-disj-speak 1p-big ‘We big ones are speaking.’

Many other NP-internal elements in Bantu languages show the same basic word order and agreement properties. This is consistent with the view that they may simply be adjectives syntactically, even though they correspond semantically to determiners or other functional heads in other languages. This class typically includes numerals, demonstratives, and possessive pronouns. I tentatively put all these categories aside as not being relevant to the question of agreement on D-like heads. Some Bantu languages also have a small number of elements that have a syntax that is distinct from adjectives. One such element is ‘all’ in Swahili, Kinande, and Zulu. Like adjectives, ‘all’ follows the noun and agrees with it: (46)

a. vi-tabu vy-ote; wa-tu w-ote (Swahili, Ashton 1949) 8-book 8-all 2-person 2-all ‘all the books’ ‘all the people’ b. aba-kali b-osi; ama-tunda w-osi (Kinande) 2-woman 2-all 6-fruit 6-all ‘all the women’ ‘all the fruits’

The difference is that ‘all’ can also agree with a pronoun in first or second person features, at least in Swahili and Zulu. (47)

a. Thina s-onke si-fik-ile. (Zulu, Doke 1963:94) we 1p-all 1pS-arrive-perf ‘We have all arrived.’ (compare: b-onke ‘all of them’) b. sisi s-ote; ninyi ny-ote (Swahili, Ashton 1949) we 1p-all you 2p-all ‘we all’ ‘all of you’

These elements were discussed in section 4.1.2. Within my theory, the difference in agreement shows that ‘all’ is a true functional head, with the DP ‘we’ in its

186


specifier position. So the structure of (47a) is (48a), as compared to the structure for adjectival modifications like (45) shown in (48b). (48)

a. [DP [ProP we] b. [ProP [ProP we]

[D all X]] [AP big]]

(48a) is also consistent with what we now know about agreement parameters in the Bantu languages. The nominal in Spec, DP c-commands the D head ‘all’, so agreement is allowed by (1). ‘All’ presumably does not assign case to the NP in its Spec, but that is not required in a Bantu language. Other words like ‘all’ in Zulu include dwa ‘only, alone’ and bili ‘both’. There is a third class of noun phrase elements that contrasts with both of the above two. Examples of this third class are kila ‘every’ in Swahili and obuli ‘every’ in Kinande. Although similar in meaning, kila and obuli differ from ote and osi ‘all’ in two ways, both illustrated in (49). First, kila and obuli come before the principal noun of the nominal, not after it. Second, kila and obuli stand out among the NP internal elements in Swahili and Kinande in that they do not show agreement with the head noun: (49)

a. kila ki-tabu ; kila m-tu (Swahili, Collins 2004) every 7-book every 1-person ‘every book’ ‘every person’ b. obuli mu-kali; obuli ri-tunda (Kinande) every 1-woman every 5-fruit ‘every woman’ ‘every fruit’

The difference in word order (as well as their meanings) suggests that kila and obuli are not adjectival modifiers; nor are they D-like heads with the NP in their specifier. The simplest analysis given that these languages have specifierhead-complement word order is that kila and obuli are functional heads and the co-occurring NPs are their complements, as shown in (50). (50)

[DP every

[NP woman]]

What does this have to do with the agreement difference? Why do NPs before the D agree with D, but NPs after the D do not in these languages? There is a clear parallel between this and the fact that subjects agree with the tensed verb when they come before it in Bantu languages, but not when they come after it (right dislocation aside). There is also a parallel between this and the fact that complementizers in Kinande agree with moved NPs that appear to their left, but not with in-situ wh-phrases that remain to their right. I claim that this is a third reflection of the parameter in (1), which says that agreement requires


187

c-command in Bantu. When NP is in Spec, DP, it c-commands D and agreement is possible. When NP is the complement of D, it does not asymmetrically ccommand D and agreement is not permitted in most Bantu languages. To get this symmetry to follow from the same principles as the others that we have considered, some fine print becomes relevant. The background to (1) is that Universal Grammar allows a functional head to search for something to agree with in two directions: upward through the structure for something that c-commands it, or downward for something that it c-commands. The idea of (1) is that some languages restrict agreement to the first case only, permitting only an upward search. The question of detail that has not arisen yet is what happens when the agreed-with nominal is the complement of the agreeing functional head and hence both c-commands it and is c-commanded by it. Does this count as an upward search, or a downward search, or can it count as either? We now see that there is an empirical answer to this question. What we observe in these Bantu languages is that agreement seems not to be possible in a relevant configuration like (50). So we need a head agreeing with its complement to be an instance of downward search, but not an instance of upward search. A simple way to do this, already stated in (1) but not needed until now, is to require that the agreed-with NP asymmetrically c-command the agreeing functional head in this type of language. Now compare Swahili and Kinande with the Indo-European languages. In the IE languages, there is no apparent correlation between word order inside DP and whether the D agrees with the noun or not. In particular, many IE languages have D-NP order, in which the D agrees with NP – the forbidden structure in Bantu. This is found even in English with demonstratives (this book vs. these books), and more robustly with all sorts of determiners in the Romance languages: (51)

a. b. c. d.

el muchacho la muchacha los muchachos las muchachas

‘the boy’ (M, sg) (Spanish) ‘the girl’ (F, sg) ‘the boys’ (M, pl) ‘the girls’ (F, pl)

This is what we expect given the agreement parameters in (1) and (2). There is no requirement that the agreement trigger asymmetrically c-commands the agreement bearer in IE, so agreement is not ruled out on these grounds. There is a requirement that one of the members of the Agree relationship values the case feature of the other. But this is satisfied in IE languages, as seen by the fact that determiners agree in case with their NP complements in IE languages that have both overt determiners and overt case marking, such as Greek:

188 (52)

Parameters of agreement a. I Maria aghapuse ton the.f.sg.nom Maria.f.sg.nom loved.3sS the.m.sg.acc Petro . . . Peter.m.sg.acc ‘Mary loved Peter . . .’ (Anagnostopoulou 2003:90) b. O Giani-s estile tis the.m.sg.nom John-m.sg.nom sent.3sS the.f.sg.gen Maria-s to grama. Maria-f.sg.gen the.acc letter.acc ‘John sent Mary the letter.’ (Anagnostopoulou 2003:9)

Case agreement thus makes ϕ-feature agreement possible on determiners in IE, much as it does with adjectives. The statement of the parameter in (1) creates one complication elsewhere in my analysis of Bantu. The problem arises in the arena of predicate nominals. In section 2.4.1, I assumed that the FN head associated with a predicate nominal can agree with its NP complement. Given our current understanding of these principles, this agreement should not happen in Bantu languages, although it could in IE languages. But this sort of agreement does take place in Bantu: I assume that it is the source of the noun class prefixes that manifest the gender and number features inherent in the noun root. Fortunately, a mechanical solution is available. We can say that the predicate nominal moves to adjoin to FN P in Bantu, much as the pronoun might move to Spec, DP in (48a). The structure of a sentence like (53a) would thus have to be (53b) in the Bantu languages. (53)

a. Dalili y-a mvua ni ma-wingu. (Ashton 1949) cl9.sign cl9-assoc cl9.rain pred cl8-clouds ‘Clouds are a sign of rain.’

b.

PredP NP sign of rain [9,sg]

Pred´ Pred ni

FNP NP

FNP

cloud FN NP [8,pl] [8,pl] [*9,sg] cloud AGREE

MOV’T


189

Since FN is realized as an affix on NP, not as an independent word,24 there is no direct evidence for or against this proposal, so I adopt it to save the phenomenon. If, however, languages turn up which are subject to (1) and have FN realized as a separate particle, this derivation will not necessarily work. It will be important to check whether in such a language word order and agreement covary for FN the way they do for determiners in Kinande and Swahili. 5.5.2 D agreement in other languages What do we find when we look at D heads in languages outside the IE and NC families? It is relatively common for a language to have agreement on at least some Ds; this is found in about 40 of my 108 languages. The most common type is the unremarkable agreement of D with its complement NP in number and perhaps gender and case (e.g., Yagua, Jakaltek, Nahuatl, Ojibwa, Halkomelom, Maung, Amele, Georgian, Khoekhoe, Finnish, Arabic). Languages with this sort of agreement need to be analyzed as allowing downward agreement, as in Indo-European. A few languages, however, seem to show an interaction between word order and agreement similar to what we observed in Bantu. One possible case is quantificational heads in Tariana. (54a) shows that the quantifier ‘all’ comes before the NP that it quantifies over and does not agree with it, while (54b) shows that the quantifier ‘many’ comes after the relevant NP and can bear number and classifier agreement with it. (54)

a. thuya nawiki (Aikhenvald 2003:219) all people ‘all (the) people’ b. nu-ita-kanape hanupe-(ma-pe) (Aikhenvald 2003:181) 1sP-daughter-pl many-f-pl ‘my many daughters’

This contrast looks very much like the one in (46) and (49) and should be subject to the same analysis. Somewhat similar is Slave. (55a) shows a quantifier that can come before the associated NP and is invariant; (55b–c) show quantifiers that follow the associated NP and agree with it. The quantifier in (55c) can even agree in 24 What must be assumed about the morphological union between FN and the noun in Bantu? One possibility is that FN is a clitic that cliticizes to the left edge of its specifier at PF. Another is that the head of NP incorporates into FN , and then the NP as a whole undergoes remnant movement to Spec, FN P. Either could be true, as far as I know, although both raise interesting further questions.

190


person, showing that it must be analyzed as a functional head, not as an adjectival modifier. (55)

a. hoyí t’áhsi ; hoyí i dene (Rice 1989:263) any thing any person ‘anything’ ‘anybody’ b. dene naáne; tue ndahe (Rice 1989:262) people some fish some ‘some people’ ‘some fish’ c. dene ʔo´ hla; ʔo´ l a-d´ıt’e (Rice 1989:264) people both both-1pS.number ‘both people’ ‘both of us’

Overall, most D-like words in Slave do not show agreement, regardless of the word order, but the few that do always come after the agreed-with NP. This suggests that agreement requires upward search, as in Bantu. Greenlandic Eskimo provides a third example: most nominal-internal functional heads come after the associated NP and agree with it in number and case, but there are a few “emphatic words” that can come before the associated NP, and these do not agree with the NP (Fortescue 1984:110). (See also section 5.11.2.1 on Berber.) Less striking but potentially relevant are languages in which only one order of D and NP is attested, but there is agreement and the observed order is consistent with NP being in Spec, DP. These languages can also be analyzed as obeying parameter (1), even though language-internal minimal pairs like (54) and (55) cannot be produced, because of lexical factors. Some of the Bantu languages fit this description, lacking a word comparable to ‘every’ in Swahili (Zulu, Luvale); so too do Jarawara, Apurinã, and Arapesh.25 Finally, there are languages which allow both D-NP order and NP-D order in apparently free variation, the D agreeing with the NP in both orders (e.g., Warao, Zoque, Yimas). I tentatively assume that these putative Ds are really adjectival modifiers, with the syntactic status of adjuncts rather than that of heads that select NP complements, and thus fall outside the purview of the headedness parameter. They have a structure like (48b), and are not directly relevant to the topic of this section. 25 A further complication is that in languages that are generally head final and that have Ds that follow NP it is not clear from the word order whether the NP is the complement of D or its specifier. Agreement could thus be analyzed as being either upward or downward in languages like Wari, Lakhota, Gooniyandi, and Alamblak. I tentatively assume that agreement is downward unless there is evidence to the contrary, figuring that the NP would only be in Spec if a marked feature forced it to move there.

Agreement on adpositions

191

Throughout this section, the focus has been on the implications of parameter (1) for agreement on Ds. In principle, one should consider parameter (2) as well. In point of fact, however, I do not know of languages that have overt case marking on both nouns and determiners in which the two ever fail to agree in case. I thus tentatively assume that D and the NP it selects always share a single case feature, quite apart from the agreement parameters (perhaps because they are part of the same extended projection). If this holds universally, then the “only if” clause in (2) is always satisfied, and the Case-Dependency of Agreement Parameter has no detectable consequences in this domain.

5.6


Next let us consider the possibility of agreement on adpositions. My prediction is that agreement on adpositions should be very similar to agreement on determiners. Like determiners, adpositions normally take NP/DP complements, so they will generally agree with that NP only in languages where the Direction of Agreement Parameter is set to “no.” Moreover, since adpositions always casemark their NP complements, the Case-Dependency of Agreement Parameter is always satisfied, so (2) will not give rise to observable crosslinguistic variation. I show that these expectations are borne out. 5.6.1 P agreement in Bantu and IE Beginning with Bantu languages, the expectation is that they should not have Ps that agree with their NP/DP complements in person, number, or gender, because the complement does not asymmetrically c-command the head. In fact, they typically do not. Bantu languages have relatively few prepositions, but most have an instrumental/comitative preposition, such as na in Kinande, and its cognates in other languages. This P does not agree with its NP complement in noun class (see also Kinyalolo 1991:111 for Kilega): (56)

Kambale a-ka-kanay-a na-(*bo) aba-syakulu. Kambale 1S-pres-speak-fv with-2 2-old.people ‘Kambale is speaking with the old people.’

Some, like Chichewa, also have a dative preposition kwa. This too fails to agree with its complement: (57)

Mbidzi zi-na-perek-a msampha kwa nkhandwe. (Baker 1988:229) zebras 10S-past-hand-fv trap to fox ‘The zebras handed the trap to the fox.’

192


The absence of agreement on Ps is rather conspicuous, inasmuch as agreement is otherwise so pervasive in Bantu languages; almost every other argument relationship is either obligatorily or optionally marked by agreement. In the absence of a theory, it might seem odd that Bantu languages do not generalize agreement to the relationship between a P and its argument. But within a true understanding of the syntax of agreement in Bantu, based on (1) and (2), this gap makes perfect sense. Kinyalolo (1991:106–30) in his discussion of agreement on Ps in Kilega makes an important observation that increases the interest of this case considerably. He points out that Ps in Kilega do agree with their complements if and only if the complement has been moved out of PP, either by focus movement or by NP movement in a passive construction. (58) replicates his observation in Kinande. (58)

Kambale a-ka-kanay-a na-bo. a. Aba-syakulu b-o 2-old.people 2-foc Kambale 1S-pres-speak-fv with-2 ‘It’s old people that Kambale is speaking with.’ b. Abasyakulu si-ba-li-kan-ibaw-a na-bo. 2.old.people neg-2S-pres-speak-pass-fv with-2 ‘Old people are not spoken with.’

This shows that it is not enough to rule out agreement in (56) simply by saying that P is not a head that initiates agreement, or that there is no realization for third person animate plural agreement on P in Kinande and Kilega. In fact, there are agreeing forms of the P. The issue is simply that P cannot agree with NP when the NP is its complement, but only when it has moved to a higher position in the clause. The Direction of Agreement Parameter in (1) explains why: agreement requires asymmetric c-command in these languages. Note also that the contrast between (56) and (58) is parallel to what we have seen with other agreeing heads in the Bantu languages: like T, C, and D, a P in Kinande can agree with a phrase that precedes it, but not with a phrase that follows it in the surface word order of the sentence. In (58b), the P apparently agrees with the surface subject in Spec, TP. This raises the question of why the P in an active sentence like (56) cannot show class 1, human singular agreement with its subject in Kinande, resulting in the PP *na-go abasyakulu. The subject certainly c-commands the P, so the parameter in (1) is satisfied. It is true that P does not assign case to the subject, but we know that agreement is not dependent on there being a case-valuation relationship in Bantu. The most promising explanation for this is probably to say that PPs, like CPs and some vPs, count as phases in Chomsky’s (2000, 2001) terms. Chomsky


193

mentions this in passing as a possible way of explaining the fact that movement out of PPs is banned in many languages. I have also entertained this possibility as the most straightforward way of explaining the robust fact that T and v never agree with the NP complement of a P (section 2.3.1). (56) then illustrates the converse case: just as no outside functional head can agree with an NP inside PP, so P cannot agree with an NP that is outside PP. Both sorts of agreement across a PP boundary are universally ruled out by the phase condition on agreement, a corollary of the Phase Impenetrability Condition. Why then is agreement allowed in (58)? Inspired by van Riemsdijk’s classic (1978) study of extraction from PPs in Dutch, the obvious answer is that the complement of P adjoins to PP on its way to Spec, TP or Spec, FocusP.26 This intermediate stop is needed anyway to satisfy the Phase Impenetrability Condition, given that PP is a phase. The intermediate position then is one in which the agreed-with NP asymmetrically c-commands P and is still in the same phase as P. It is only NPs that occupy this position at some point in the derivation that can trigger agreement on P in a Bantu language. The crucial structures are compared in (59). (59)

a. [Kambale T [VP speak [PP with [NP old people]]]] agreement out by parameter (1) b. [Kambale T [VP speak [PP with [NP old people]]]] agreement out by phase condition c. [old.peoplei T [VP are-spoken [PP t i [PP with t i ]]]] agreement allowed

In contrast with what we have seen in Bantu languages, we expect that IE languages could allow a P to agree with its unmoved NP complement. Asymmetrical c-command is not required in these languages, since (1) is set to “no.” Moreover, Ps in IE languages assign case to their NP/DP complements, so (2) should be satisfied. In fact, not many IE languages take advantage of this

26 The reader might wonder why I say that the intermediate position that NPs move through in pseudopassives and focus constructions is adjoined to PP rather than Spec, PP. Theoretical motivation for this is my claim that prepositions are functional heads that have the same major category properties as adjectives, and adjectives by definition do not have specifier positions (Baker 2003a). Empirical motivation for this is Kinyalolo’s (1991) observation that Ps in Kilega agree with moved NPs in number and gender but not in person. This suggests that the NP moves through an adjoined position rather than a Spec position, since agreement in person is possible with specifiers.

194


opportunity, but a few do. Welsh in particular has many Ps that agree with unmoved pronominal complements:27 (60)

a. Soniais i amdan-o ef. (Harlow 1981:220) talked I about-3s.m him ‘I talked about him.’ b. amdanaf i; amdanat (ti); amdani (hi); amdanynt (hwy) about-1s me about-2s you about-3s.f her about-3p them (Harlow 1981:249)

This is yet another manifestation of the systematic difference between Bantu and IE when it comes to agreement. 5.6.2 P agreement in other languages Turning to other languages, we find that a significant minority of the 108 languages have some form of agreement on Ps – between 20 and 30, depending on what one makes of languages in which Ps agree with pronouns only (see note 27). Agreement on P is thus significantly more common than agreement on C, but not as common as agreement on D. Languages in which P agrees with its NP argument and there is nothing unusual about the word order that suggests that the NP has moved must generally be analyzed as permitting downward agreement. (61) and (62) give some examples. (61)

a. y-ul te’ n¨ ah (Jakaltek, Craig 1977:110) 3-in the house ‘in the house’ b. j-iba¨n, w-et 1p-on 1s-to ‘on top of us’ ‘to me’

(62)

a. awə y a-mʃj taxj (Abaza, O’Herin 2002:54) that 3s.i-after ‘after that’ b. (sara) s-pnə I 1s-at ‘at my house, by me’

27 A complication is that Ps do not agree with nonpronominal objects in Welsh: there is no third person masculine agreement on the P in soniais i am y dyn ‘talked I about the man,’ for example. Languages that allow agreement with pronouns but not full NPs are not uncommon (see WALS:map 48); other languages like this in my sample include Persian, Canela-Krahô, Hixkaryana, Guarani, Choctaw, and Lakhota. In some cases this can be explained by reanalyzing the putative agreement morpheme on P as a pronoun that cliticizes to the P, but that does not seem right for all the relevant languages, including Welsh. I leave open the explanation for this phenomenon.


195

Other languages that have agreement on P with in-situ objects include Nahuatl, Arabic, Finnish, Georgian, Burushaski, Tauya, Dani, Lavukaleve, Daga, Yagua, Pirahã, Wari, and perhaps Greenlandic, Quechua, Wich´ı, and Yimas.28 The most compelling evidence for a “yes” setting to the Direction of Agreement Parameter comes from languages that permit an NP to appear outside of PP, the P agreeing with NP only when this happens. This is not common in my sample, but is attested. One such language is Slave (Rice 1990:1197): (63)

e ] gondeh. a. Yahti [ts’éli a ts’´ priest girl to 3.talks ‘The priest talks to the girl.’ b. Ts’éli ai yahti [ei ye-ts’´ e ] gondeh. girl priest 3-to 3.talks ‘The priest talks to the girl.’

Another is Fijian (Dixon 1988:42): (64)

a. ‘Eimami saa qaaqaa a ‘ai-Boumaa [i-na drano]. we asp victorious art native-place about-art lake ‘We, the natives of Boumaa, were victorious concerning the lake.’ b. A drano ‘eimami saa qaaqaa [‘i-na --] a ‘ai-Boumaa. art lake we asp victorious about+3.sg art native-place ‘The lake, we the Boumaa people were victorious concerning it.’

These are the languages that are most clearly like the Bantu languages when it comes to agreement on P. But it is not too surprising that we do not observe this pattern very often, because relatively few languages allow adposition-stranding in the first place. Within the IE languages, for example, P-stranding is found mainly in some languages of the Germanic branch. A language that does not allow P-stranding but has the Direction of Agreement Parameter set “yes” will simply be a language in which Ps do not agree with their NP complements. Such languages are common. It is difficult in practice to distinguish languages in which Ps fail to agree for systematic grammatical reasons (parameter (1) is set as “yes”) from languages in which Ps fail to agree for accidental lexical reasons (the Ps of that language happen not to be agreement bearers). Nevertheless, we can conclude 28 Another issue (motivating the “perhaps” in this statement) is the challenge of distinguishing true adpositions from locative nouns. In quite a few languages, words which are used functionally as locative adpositions are at least cognate with and perhaps reducible to location-denoting nouns that show possessive agreement with their NP argument. I have not worried about this distinction too much, because even though examples of this type might technically be agreement on N rather than on P, they probably still count as downward agreement of a head with its complement, and hence provide evidence of a “no” setting for parameter (1).

196


that the range of variation in how Ps agree in languages of the world is in keeping with what one would expect given the parameters in (1) and (2). 5.7

Agreement on v

I have saved almost until last the investigation of “object” agreement – the agreement that one often finds on v. This is probably the second most common kind of agreement after T agreement, being found in roughly 50% of the languages in my sample. It is thus much more prominent and grammatically significant than agreement on C or P. My parameters do indeed make clear predictions about how object agreement should be different in NC languages as compared to IE languages. The reason that I have put off discussion until now is an empirical one: object agreement happens to be rare or nonexistent in the IE languages, and there are some complications concerning it in the NC languages. Nevertheless, I can work through the logic of the parameters as they apply in this domain, and point to some possible instances in the NC and IE languages. Then when we turn to other language families we will find clearer examples that illustrate the expected range of possibilities. 5.7.1 v agreement in NC and IE languages In Bantu languages, the implications of (1) for object agreement are exactly parallel to its implications for agreement on Ps: there should be no object agreement on v unless the agreed-with NP moves to Spec, vP or higher. The object could in principle stay in Spec, vP, or it could move on to a peripheral position, as in a dislocation structure. For some Bantu languages, such as Chichewa, this looks promising. There is indeed no object prefix if the direct object is internal to the verb phrase, but there is an object prefix if the object is dislocated to a right or left-peripheral position, as discussed in detail by Bresnan and Mchombo (1987) (data from Mchombo 2004:80–1).29 (65)

a. Alenje [a-ku-ph´ık-´ıl-á anyan´ı z´ı-túmbûwa]. 2.hunters 2S-pres-cook-appl-fv 2.baboons 8-pancakes ‘The hunters are cooking the baboons pancakes.’

29 Note that “object markers” in Chichewa and Kinande cannot appear on verbs in passive or reciprocal voice (Mchombo 2004:91,107), which I tentatively take to be evidence that they are manifestations of agreement on active v, not cliticized pronouns comparable to those in Romance languages and Greek (see section 3.3.3, especially note 24). Another relevant difference is that in complex tenses object markers always stay on the main verb in the Bantu languages, whereas in IE languages object clitics climb onto the finite auxiliary, as shown by the contrast between Kinande and French in (i). Thus, object markers stay close to the v projection in Bantu, as expected if they are agreement on v, in contrast to object clitics in IE languages, which are attracted to the T node.

Agreement on v

197

b. Alenje [a-ku-wá-ph´ık-´ıl-á - z´ı-túmbûwa] anyan´ı. 2.hunters 2S-pres-2O-cook-appl-fv 8-pancakes 2.baboons ‘The hunters are cooking the baboons pancakes.’

Essentially the same is true in Kinande, except that left dislocation of the agreedwith object is fairly rare and marked, and right dislocation is forbidden. One usually finds object agreement only with a null object in Kinande, but (66b) is possible. (66)

a. N-a-(*ri)-gul-a eritunda. (Baker 2003c) 1sS-t-5O-buy-fv fruit.5 ‘I bought a fruit.’ b. Eritunda, n-a-*(ri)-gul-a. fruit.5 1sS-t-5O-buy-fv ‘The fruit, I bought it.’

In contrast, an IE language could in principle have object agreement with an object that remains inside VP, as long as v values the case of the agreedwith NP, presumably as accusative. This prediction is largely moot, since most IE languages do not have agreement on v at all. It is possible, however, that what were historically clitics have been reanalyzed as true object agreement in some varieties. Ormazabal and Romero (2006a) claim that this is true of le´ısta dialects of Spanish, in which the formerly dative clitics le and les are used (also) with animate/human objects in accusative case. Ormazabal and Romero argue that le and les (in contrast to lo/los/la/las) have the status of object agreement markers. Part of their evidence is that these morphemes can appear with even an indefinite or quantified NP in the normal direct object position, without it undergoing movement or dislocation. (67) is an example, where both the word order and the indefinite interpretation of a unos jóvenes ‘some young people’ suggest that it is still in VP. (67)

(i)

unos jóvenes al pueblo. (p. 18) (Yo) les-llevé a I 3pO-carry acc a.pl youngsters to.the town ‘I gave some young people a ride to the town.’ a. Tu-a-by-a tu-ka-ri-ly-a. (Not: *tu-a-ri-by-a tu-ka-ly-a.) 1pS-t-be-fv 1pS-ptpl-5O-eat-fv 1pS-t-5O-be-fv 1pS-ptpl-eat-fv ‘We were eating it.’ b. Je l’ai mangé. (not: *J’ai le mangé) I it-have eaten ‘I ate it.’

198


If Ormazabal and Romero’s interpretation of examples like these is correct, then it confirms that v can in principle look downward into VP for something to agree with in IE languages, as expected.30 Zulu reveals another important factor in evaluating predictions in this domain. Like Chichewa, Zulu has optional object agreement on the verb: (68)

a. Ngi-bona umu-ntu. (Doke 1963:299) 1sS-see cl1.sg-person ‘I see a person.’ b. Ngi-ya-m-bona umu-ntu. 1sS-disj-1.sg.O-see cl1.sg-person ‘I see the person.’

But unlike Chichewa, the agreed-with object is not dislocated to the periphery of the clause. Thus in Zulu it is possible to have V-NP1-NP2 order where the verb agrees with NP1 but not NP2, as in (69b) (contrast this with (65b) from Chichewa). (69)

a. Ngi-leth-el-a umfundisi incwadi. (Doke 1963:299) 1sS-bring-appl-fv teacher.1.sg book.5.sg ‘I am bringing the teacher a book.’ b. Ngi-ya-m-leth-el-a umfundisi incwadi. 1sS-disj-1.sg.O-bring-appl-fv teacher.1.sg book.5.sg ‘I am bringing the teacher – the one who told me to do so – a book.’

At first, this seems problematic. It is, however, perfectly possible to maintain that the (first) object in (68b) and (69b) has moved to Spec, vP. This is defensible as long as we believe that the verb in Zulu has moved past v into the T node (or perhaps to some aspect or mood head above v but below T, as in Julien 2002 and Carstens 2005:227). Moving the object to Spec, vP would put the object before the verb, but subsequent movement of the verb to T (or Aspect/Mood) restores the V-NP order: (70)

[TP pro T+v+bring [vP teacher [VP book]]]

30 Kayne (1989) discusses a rather different kind of object agreement in Romance languages – agreement on past participles – which seems to be possible only if the object moves to a position higher than the participle. I have no explanation for why this kind of agreement apparently needs to be upward, despite the “no” setting for parameter (1) in IE languages. I only suggest that this phenomenon needs to be studied in the light of the broader properties of participle agreement, including those mentioned in section 5.9.

Agreement on v

199

Indeed there is good independent reason to think that verb-raising happens in Zulu and other Bantu languages. First, the verb is morphologically united with tense-aspect-mood morphology into a single morphological word; head movement is one of the primary mechanisms by which such morphologically complex words can be formed (Baker 2002). Second, the finite verb comes before indefinite subjects that stay in vP in Zulu and other Bantu languages, resulting in V-S-X order (see (13a) above). It is expected that the same verbraising that takes the verb to the left of unraised subjects in Spec, vP will also take the verb to the left of shifted objects, which are also in a specifier of vP. This is at least consistent with the idea that Bantu languages have V-to-T movement. There is some semantic evidence that agreed-with objects have undergone movement in Zulu, even though that movement is concealed by other, independently motivated movements. Doke observes that whether the object is agreed with or not corresponds to something like definiteness: agreed-with objects are interpreted as definite, whereas unagreed-with objects are interpreted as indefinite, as indicated in the glosses of (68)–(69). We know that in languages like Dutch and German leftward object shift produces strong, definite readings of the shifted object (from Broekhuis 1992:81): (71)

a. dat Jan waarschijnlijk morgen iemand zal bezoeken that Jan probably tomorrow someone will visit ‘that Jan probably will visit someone (or other) tomorrow’ b. dat Jan iemand waarschijnlijk morgen zal bezoeken that Jan someone probably tomorrow will visit ‘that Jan probably will visit someone (a particular person) tomorrow’

A standard account of this sort of phenomenon since Diesing 1992 is to say that the verb phrase is the domain of existential closure. NPs that remain in the verb phrase are within the domain of existential closure, and get weak/indefinite/nonspecific readings, whereas NPs that escape the verb phrase get strong generic, specific, or (in articleless languages) definite readings. If something like this picture is right, then it is attractive to analyze the difference between (68a) and (68b) in Zulu in the same way. The NP stays in VP and gets a weak indefinite reading in (68a), but leaves the VP and gets a strong or definite reading in (68b). The different structures do not result in different surface word orders, because the verb moves to T (and there are no adverbs that left-adjoin to VP in most Bantu languages). But the difference is revealed indirectly by agreement on v, given the parameter in (1). When the NP moves to Spec, vP (or higher) it can agree with v, whereas when it stays inside VP (the domain of existential closure) it does not c-command v and cannot agree with it:

200


(72) a.

b.

TP

TP

DPi pro

T´ T

DPi vP

ti

pro v´ ∃

v see [*CL1,sg]

T´ T

vP v´

ti VP

NP

V

NP

person [CL1,sg]

*Agree by (1)

person [CL1,sg]

v´ v see

Agree is OK

∃ V

VP NP

Thus, there are at least two ways to observe whether or not an NP has moved higher than v: one is word order, and the other is whether it must have a strong/generic/definite reading. Given these background assumptions, the impossibility of a verb showing object agreement with a weak/indefinite NP in Zulu is a result of the restriction on agreement in Bantu languages stated in (1). The minimal contrast between the Zulu example in (68a) and the Spanish example in (67) thus testifies again to the parametric difference in the syntax of agreement between the two families. Diercks (2006) shows that this analysis of Zulu can also be applied to the very similar “optional” object markers found in Swahili.31 5.7.2 v agreement in other languages Let us then briefly survey the behavior of object agreement on transitive verbs in other languages. There are two things to look for. The crucial question with 31 There is one situation in which the Bantu v might agree downward with something in its c-command domain. In section 3.3.3, I suggested that the Bantu language Shambala has twoand-a-half agreement, in which v agrees with both the goal object in its Spec and the theme object still in VP. Assuming for the sake of argument that Shambala has the same setting for parameter (1) as most other Bantu languages, this stands out as an anomaly. What is special about this case is that a single functional head (v) agrees twice, once in the proper way (probing upward) and once not. This suggests that it is a “minimal compliance” effect in the sense of Richards 1998: if the parametric condition in (1) has been satisfied once for a given head, that head becomes free of the condition for further applications of Agree. (An alternative would be to say that the object markers in Shambala are clitics, not true instances of agreement. In that case Shambala is not relevant to the issues discussed in section 3.3.3.)

Agreement on v

201

respect to parameter (1) is whether object agreement requires the agreed-with NP to move out of VP or not, as evidenced either by word order or by the strong/definite reading of the object NP. The crucial question with respect to parameter (2) is whether object agreement requires that the agreed-with NP bear a particular case, such as accusative, or if agreement with oblique-case objects is also tolerated. In fact, the full space of possibilities is realized for v agreement, just as it was for T agreement and C agreement. Unfortunately, even languages that clearly have object agreement do not always provide clear evidence as to whether object agreement is contingent on movement. The reason is because object agreement with third person objects is often phonologically null, whereas agreed-with first and second person pronouns are usually pro-dropped except in special discourse conditions (topic and focus) which can themselves affect word order. When this is true, one cannot expect to observe that object agreement depends on word order as in (65) from Chichewa or on definiteness as in (68) from Zulu, because the form with agreement and the form without agreement would be homophonous. Languages of this indeterminate kind include quite a few languages of the Americas (Mapudungun, Maricopa, Choctaw, Acoma, Halkomelem, Wich´ı, Pirahã, Quechua, Guaran´ı, Ika) and some languages of Australia (e.g., Gooniyandi). But agreement with third person objects is not always phonologically null, and when it is not we find both languages in which object agreement is contingent on word order and/or definiteness and languages in which it is not. A good example of the former is Nahuatl. (73) compares an example in which the direct object is indefinite and adjacent to the verb with an example in which the direct object is definite and not adjacent to the verb. Both word order and semantic interpretation suggest that the direct object is inside VP in (73a) but outside it in (73b); nevertheless, the verb shows explicit object agreement with its direct object in both examples (Launey 1981:38). (73)

a. Ø-quim-itta c¯oc¯ohua in pilli. 3sS-3pO-see snakes det child ‘The child saw (some) snakes.’ b. Ø -qu-itta in cihu¯atl in calli. 3sS-3sO-see det woman det house ‘The woman saw the house.’

(73a) implies that downward agreement is permitted in Nahuatl. Other languages in which object agreement is possible with indefinite objects that stay close to the verb include Basque, Chukchi, Abkhaz, Georgian, Burushaski, Tauya, Amele, Alamblak, Yimas, Mangarayi, Wari, Ojibwa, Maung, Tiwi, Dani,

202


Warlpiri, Lavukaleve, Mohawk, Wichita, Hixkaryana, Lakota, Kiowa, Makah, and Pirahã. This seems to be the larger class. There are also a reasonable number of languages in which agreement does vary with the word order and/or interpretation of the object, however. One is Canela-Krahô. (74a) shows that when the object is adjacent to the verb – its normal position – the verb does not agree with it. The verb does, however, agree with the object when it is displaced to the left, either in a focus-like construction in which the object lands above the subject and tense marker ((74b)), or in a scrambling-like construction in which the object lands at the left edge of the verb phrase, below the subject and the tense maker ((74c)). (74)

a. H˜umre te po curan. (Popjes and Popjes 1986:139) man past deer kill ‘The man killed a deer.’ b. Po wa i-te ih-curan. (p. 136) deer I 1sS-past 3O-kill ‘It was a deer that I killed.’ c. Wa ka capi p˜ı na cu-m˜e. (p. 140) I fut Capri wood about 3O-order ‘I ordered Capri about the wood (to cut it).’

Another language that shows a Bantu-like relationship between word order and object agreement is the Athapaskan language Slave (Rice 1989:1197): (75)

a. Li [ʔ ehkee ka-yi hshu] dog boy adv-3S.bit ‘The dog bit the boy.’ b. ʔ ehkeei li [ei ka-ye-yi hshu] dog boy adv-3O-3S.bit ‘The boy, a dog bit (him).’

Yet another is Fijian (see Massam 2001 for a detailed analysis of the difference between these sorts of sentences in the related language Niuean): (76)

a. [E’au.i vola mai] a cauravou. (Dixon 1988:49) deliver letter to.here the youth ‘The youth is delivering letters.’ b. [E’au-ta - - mai] a-i-vola yai a cauravou. deliver-tr.3O to.here the-letter this the youth ‘The youth is delivering the letter.’

As a final example, I cite Apurinã. Recall from section 5.2.2 that this language has a very unusual word order, alternating between OSV and OVS. I analyzed

Agreement on v

203

this as a head-final language in which specifiers appear on the right – the mirror image of English and the Bantu languages. So movement upward in the structure is movement to the right in Apurinã. We find the same thing with object movement: low objects that appear before the verb are not agreed with, whereas high objects that appear after the verb are agreed with, as shown in (77) (Facundes 2000:547). (77)

a. Uwa ata maporoka. 3.m.sg 1.pl root.up ‘We pulled it up.’ b. Owa maporoka-ru uwa. 3.f.sg root.up-3O.m 3.m.sg ‘She pulled it up.’

So the kind of relationship between word order and object agreement that is found in the Bantu languages is also attested in other parts of the world. Moreover – anticipating section 5.11 below – it is very suggestive that the languages mentioned here are also languages that were seen to be like Bantu languages rather than like IE languages in other domains: Canela-Krahô and Apurinã are also like Bantu in showing a dependency between word order and subject agreement (section 5.2.2); Slave and Fijian are also like Bantu in showing agreement on Ps only when the object of P is outside the PP (section 5.6.2). This encourages the view that (1) is a single unified parameter, governing all the functional categories of the language.32 The crucial issue for the Case-Dependency of Agreement Parameter is whether or not v can agree with objects that bear oblique case as well as with objects that bear the normal structural case. This too will not be observable in all languages, but only in those that overtly mark a distinction between (say) accusative case and dative or instrumental case. Quite a few agreementrich languages do not have this sort of morphological case system, including most languages of Africa and the Americas. But among the subset that 32 A different sort of language in which v can only agree upward is Tariana, as analyzed in section 3.2.2.2. In Tariana, v agrees with the thematic subject, rather than with a moved object. As far as I know, this is a rare pattern. Perhaps it only arises in languages in which v has no EPP feature (so agreeing upward with a shifted object is not an option) and T is not a probe for agreement (so it does not compete with v for the privilege of agreeing with the thematic subject). Some ergative languages may also have vs that only agree upward with the thematic subject, such as Abaza (section 3.2.2.1). The easiest way to explain this would be to say that agreement in these languages is case-dependent: v agrees with NP only if v values NP’s case as ergative. If there are any languages that have this sort of ergative agreement but show other evidence of agreement not being case-dependent in the sense of (2), they might be problematic for my proposals.

204


do show case distinctions, we find both kinds of languages. For example, Georgian is a language in which object agreement is not dependent on case. Some verbs in Georgian take nominative objects in the aorist tense, whereas others take objects in the dative case. Some verbs, such as ‘hit,’ even alternate between the two case frames, depending on the dialect and register (Hewitt 1995:550): (78)

a. Mc’vrtnel-ma lom-s s-cem-a (formal/conservative) trainer-erg lion-dat 3sO-hit-3sS ‘The trainer hit the lion.’ b. Mc’vrtnel-ma lom-i s-cem-a (colloquial/innovative) trainer-erg lion-nom 3sO-hit-3sS ‘The trainer hit the lion.’

This difference in the case-marking of the object has no effect on verbal agreement: the same third person singular agreement with the internal argument ‘lion’ is present on the verb in both (78a) and (78b). Similarly, Georgian verbs show the same kind of agreement with dative objects of ditransitive verbs as they do with nominative or dative/accusative objects of monotransitive verbs. Burushaski ((79)) and Warlpiri ((80)) also fit this profile:33 (79)

gu-yε ts-m. (Agr with absolutive object) a. (U:n ) you-abs 2sO-see-1sS (Lorimer 1935) ‘I saw you.’ b. U:n ə r hik trn gu-ˇciˇc-m. (Agr with dative object) you-dat one half 2sS-give-1sS (Lorimer 1935) ‘I shall give a half to you.’

(80)

a. Ngajulu-rlu ka-rna-ngku nyuntu nya-nyi. (Simpson 1991:158) I-erg pres-1sS-2sO you see-npst ‘I see you.’ b. Yurrkunyu-rlu-lpa-ngku yu-ngkarla mangarri manu police-erg-impf-2sO give-irr food and nalija rdaku-ngka-ku-ju. (Simpson 1991:206) tea jail-loc-dat-eu ‘The police must give you food and tea in jail.’

33 Interestingly, dative objects can sometimes be replaced with allative-marked NPs in Warlpiri; these apparently never trigger agreement on the Aux node. I assume that allative case markers are really members of category P; as such the phrase they head is a phase and hence impenetrable to agreement from the outside. There is a difference between agreement with an absolutive object and agreement with a dative object only in the third person singular in Warlpiri. I put this aside as a secondary effect, perhaps a matter of how agreement is realized at PF.

Agreement on v

205

Note that these are the very same languages in which subject agreement is not sensitive to whether the subject has ergative case or nominative case (section 5.2.2). Thus, the languages that do not obey (2) for agreement on T also do not obey it for agreement on v. Other languages of this type are Basque, Chukchi, Imbabura Quechua, and perhaps Ika.34 In contrast to these languages, there are also languages in which the verb agrees only with an object that has a particular case. Greenlandic Eskimo is a clear example, where only absolutive case objects trigger agreement on the verb, never internal arguments that bear dative or instrumental case. This is seen in the minimal pair in (81): in (81a) the theme argument is in absolutive case and triggers plural agreement on the verb, whereas in (81b) the goal argument is in absolutive case and triggers singular agreement on a closely related verb (Fortescue 1984:88–9). (81)

a. Aningaasa-t Niisi-mut tunniup-pai money-abs.pl Nisi-dat give-3sS/3pO ‘He gave money to Nisi.’ b. Niisi aningaasa-nik tuni-vaa Nisi.abs.sg money-instr.pl give-3sS/3sO ‘He gave Nisi money.’

Apurinã provides another example. In this language, most ditransitive verbs take an oblique goal argument, and the verb cannot agree with that argument ((82a)). The only ditransitive verb that can take two non-oblique arguments is ‘give’ ((82b)); this is also the only verb that can show object agreement with the notional goal ((82c)) (Facundes 2000:287–90). (82)

a. Mipake owara-pira-ta-ru/*i u-sãkire pita-monhi. Mipake teach-language-3O.m/*2O 3.m-language you-goal ‘Mipake taught his language to you.’ b. Nhi-yowata-ne pita nota suka. 1sP-knife-poss you I give ‘I gave you my knife.’

34 My typology is more restrictive on this point than Bobaljik’s (to appear). Bobaljik distinguishes between unmarked case (nominative, absolutive), dependent case (accusative, ergative), and oblique case (all others). In his theory, a language can be specified as allowing agreement with dependent cases but not oblique cases, so he expects to find languages in which ergative subjects trigger agreement but dative NPs do not. I distinguish only between languages in which agreement on a particular head requires the trigger to have a particular case and languages that do not. I agree that there are a few languages that seem to match Bobaljik’s prediction, but I would analyze them as languages that have goal PPs, not dative case NPs (see, for example, note 13 on Nepali) – a distinction that the theory needs anyway (see note 31 on dative versus allative in Warlpiri).

206

Parameters of agreement c. Pu-suka-no nota. 2sS-give-1sO me ‘Give (it) away to me.’

The possibility of agreement with the goal in (88c) but not (88a) testifies to the case-dependence of agreement in Apurinã. (Recall that agreement is sensitive to word order in Apurinã because of parameter (1); that is why there is no agreement with either object in (82b).)35 Overall, then, we find the range of object agreement systems that we expect to find, given the existence of the two independent parameters for agreement in (1) and (2). 5.8

Agreement on the linker head

The last functional head whose agreement properties I consider is a rather unusual one that is unique to Kinande among the languages in my 108-language sample. Baker and Collins (2006) describe a particle that appears between the two complements of a triadic verb in Kinande; some examples are given in (83). (83)

okugulu k’ omo-kihuna. a. Mo-n-a-hir-ire aff-1sS-t-put-ext leg.15 lk.15 loc.18-hole.7 ‘I put the leg in the hole.’ b. Kambale a-seng-er-a omwami y’ ehilanga. Kambale 1S/t-pack-appl-fv chief.1 lk.1 peanuts.19 ‘Kambale packed peanuts for the chief.’

Collins and I analyze this as a head that we call linker (Lk), which selects a VP complement and is itself the head of the complement of v. One NP from within VP then moves to Spec, LkP, where it gets accusative case from v. If necessary, the other complement gets case from Lk itself. Interestingly, either NP can move to Spec, LkP. Thus, the sentences in (84) are also possible, in more or less free variation with the corresponding sentences in (83). (84)

omo-kihuna m’ okugulu. a. Mo-n-a-hir-ire aff-1sS-t-put-ext loc.18-hole.7 lk.18 leg.15 ‘I put the leg in the hole.’

35 There is a serious practical problem in distinguishing NPs marked with dative (or oblique) case from PPs. I suggested that agreement with the goal argument in (82a) is bad because Arupinã has parameter (2) set as “yes.” An alternative interpretation is that monhi is a P and hence blocks agreement with the goal by the phase condition. This indeterminacy of analysis is a general problem, and clearer and more reliable ways to distinguish case and adpositions are needed. (In Greenlandic it is relatively clear that dative is a case marking in the sense familiar from IE languages, because modifiers agree with the head in case, including dative.)

Agreement in auxiliary constructions

207

b. Kambale a-seng-er-a ehilanga hy’ omwami. Kambale 1S/t-pack-appl-fv peanuts.19 lk.19 chief.1 ‘Kambale packed peanuts for the chief.’

Consider now the agreement properties of this particle. It always agrees with the NP to its left, and never with the NP to its right. In structural terms, it agrees with its specifier, which asymmetrically c-commands it, never with something inside its VP complement. This is in accordance with (1). Furthermore, case assignment does not play a crucial role in this agreement: if Baker and Collins are correct, the first NP in (83) and (84) gets its case from v, not from Lk; nevertheless, Lk agrees with it. This shows once again that (2) does not apply in Kinande. Since IE languages have the opposite values for the agreement parameters in (1) and (2) from Kinande, I expect that IE languages could not have verb-phraseinternal particles with the same agreement properties as Kinande’s linker. That is consistent with the facts: to the extent that IE languages have VP-internal particles at all (e.g., English I gave Chris back the book) they do not bear agreement. An IE particle could conceivably agree with an NP that it assigned case to, given its “yes” value for (2). But if particles never assign case in IE languages – and never undergo case concord – then they can never have agreement. The value of this functional head for broad typological investigation is essentially nil. The only languages that are known to have overt linkers outside the NC languages are certain Khoisan languages studied by Collins and his students. But the linkers in those languages do not bear agreement. The value of mentioning this unusual element in Kinande comes from its contribution to my argument that the agreement parameters apply to languages as wholes, not just to particular heads in those languages. The linker particle in Kinande has recognizably the same agreement behavior as many other, more familiar heads in Kinande, including Tense, Complementizer, Determiner, Preposition, and v. The inventory of functional heads might vary to some extent from one language to another, but the parameters in (1) and (2) apply automatically to whatever agreement-bearing functional heads a language happens to have, I claim.

5.9


5.9.1 Auxiliary constructions in Bantu vs. IE languages There is one more rather prominent difference in agreement between Bantu languages and IE languages that can be studied in terms of my parameters. This

208


difference is seen in complex tenses that are made up of an auxiliary plus a main verb. IE languages consistently have full verbal agreement on the auxiliary verb but not on the main verb:36 (85)

a. I am crying b. You are crying. c. He is crying(*s)

(86)

el libro. (*leyendas) a. Las muchachas están leyendo the girls.f.pl are reading(m.sg) the book reading.f.pl b. Las muchachas han le´ıdo el libro. (*le´ıdas) the girls.f.pl have read(m.sg) the book read.f.pl

In contrast, Bantu languages have full person-number-gender agreement on both verbs in a complex tense. This construction has been studied with particular care from a generative perspective by Kinyalolo (1991) for Kilega and Swahili; it also exists in Kinande and Zulu. (See also Carstens 2001 for important implications of this construction for aspects of Chomsky’s Minimalist theory of agreement.) (87)

(Mimi) ni-li-kuwa ni-ngali ni-ki-fanya kazi. (Swahili) (1.sg) 1sS-past-be 1sS-still 1sS-perf-do 9.work ‘I was still working.’ (Carstens 2001:150)

(88)

a. Abakali ba-lwe ba-ka-ly-a amatunda. (Kinande) women.2 2S-leave 2S-ptpl-eat-fv fruits.6 ‘The women were eating fruits.’ b. Tú-lwé tú-ká-ly-a. 1pS-leave 1pS-ptpl-eat-fv ‘We were eating.’

This sort of double agreement is also found in Lokaa and Ibibio, agreement-rich non-Bantu languages of the Niger-Congo family. The question then is why are two full agreements with the same NP argument tolerated in NC but not in IE languages. My answer centers on the fact that IE languages are subject to (2), whereas Bantu languages are not. Suppose that the T-like participial head associated with the lower verb in a complex verbal construction agrees with a particular 36 Past participles in IE languages sometimes agree with moved direct objects in number and gender (although not in person) under complex conditions that vary from language to language (see Kayne 1989 and many others). I have nothing to say about this complex topic here. The question at hand is whether a language allows two verbs in the clause to agree with the same noun phrase, and the possibility of object agreement on participles is not directly relevant to this.


209

NP X. In an IE language, this will only be possible if the participial head values the case feature of X as (say) nominative. X then raises into the domain of the higher T associated with the auxiliary verb. This T cannot value the case feature of X, because X’s case was already valued by the lower T. Therefore, the upper T cannot agree with X. In short, the sort of double agreement with the same NP that is found in NC languages is ruled out in most IE languages because both agreeing Ts must value the case feature of NP and case valuation can only be done once for each NP.37 The NC languages, in contrast, are not subject to the Case-Dependency of Agreement Parameter in (2). Therefore, there is nothing to rule out two distinct T nodes from agreeing with the same NP in these languages. All that would be required would be for the NP to asymmetrically c-command both Ts. This is accomplished by NP movement. That explains why examples like (87) and (88) are possible in languages of this family. This reasoning extends to raising-to-subject constructions involving verbs like seem. In IE languages, these have agreement only on the matrix verb; the lower verb must be an uniflected infinitive:38 (89)

a. The children seem – to be tired. b. *The children seem – are tired.

In Kinande, it is possible to have full agreement on both the matrix verb and the embedded verb, as shown in (90b). (90)

a. A-li-nga aba-kali mo-ba-gend-ire. 1S-be-if 2-woman aff-2S-go-ext ‘It seems that the women have left.’ b. Aba-kali ba-li-nga mo-ba-gend-ire. 2-women 2S-be-if aff-2S-go-ext ‘The women seem to have left.’

37 Double agreement in these auxiliary-plus-verb constructions can be contrasted with having agreement on both C and T, which is possible in some West Germanic languages (see (35)). The difference, in my view, is that C and T work together to assign a single instance of nominative case (Watanabe 1996), whereas the two T-like nodes in a complex tense structure are competing to assign nominative case twice. 38 European Portuguese is unusual among the IE languages in that it has infinitives that can show agreement. Interestingly, the complement of a raising verb in Portuguese can be an agreeing infinitive if and only if the subject does not raise into the matrix clause, where it agrees with the matrix T (Schütze 1997:127). This proves that it is the agreement/no-agreement distinction that is crucial in IE languages, not the finite/nonfinite distinction per se, just as my theory would expect.

210


In (90b), the lower verb can agree with the subject ‘woman’ without necessarily valuing its case as nominative. So can the higher verb, and double agreement is therefore allowed in raising structures in Kinande. While the parameter in (2) is the primary factor at work in this domain, one can observe some influence of the Direction of Agreement Parameter as well. The matrix verb in Kinande can agree with the thematic subject of the lower clause only if that subject appears in the matrix clause. When the thematic subject stays in the lower clause, the ‘seem’-class predicate cannot agree with it, and must be a default third person singular form. Thus, (91) is ungrammatical, in contrast to (90a). (91)

*Ba-li-nga aba-kali mo-ba-gend-ire. 2S-be-if 2-woman aff-2S-go-ext ‘It seems that the women have left.’

This follows from the fact that T cannot search downward to find an agreement trigger in Bantu, the way it does in IE structures like (92). (92)

There seem to be some children in the basement.

Returning to complex tense constructions, it must be acknowledged that these sometimes do show a limited form of double agreement even in IE languages. But this never happens when the lower verb is a fully verbal, finite form. It only happens when the lower verb is an adjective-like participle, which agrees with the subject in number and gender but not in person. (93) gives examples from Serbian, where the phenomenon is quite general. (Many western European languages have this sort of double agreement only in passive or unaccusative clauses, for reasons that I do not investigate.) (93)

a. Já sam cˇ`ıta-o/ cˇ`ıta-la. (Magner 1991:268–9) I am read-m.sg read-f.sg ‘I was reading; I read.’ b. On/ona je cˇ`ıta-o/ cˇ`ıta-la. he/she was read-m.sg read-f.sg ‘He/she was reading; he/she read.’ c. Mi smo cˇ`ıta-li/ cˇ`ıta-le we are.1pS read-m.pl read-f.pl ‘We were reading; we read.’

Why is this possible? The answer is that participial heads do not value the case of the NP that they agree with. Thus they do not compete with the finite T associated with the auxiliary verb in this respect, so nothing prevents that


211

T from both assigning case to and agreeing with the NP. Why then can the participle agree with NP without valuing NP’s case, given (2)? Because, like adjectives and determiners, participles have an unvalued case feature of their own, which the NP can value. In other words, they can agree with the NP in case. In many situations the case on the participle will be unmarked nominative, but it can sometimes bear other cases; for example, accusative participles show up in ECM constructions in Icelandic (Thráinsson 1979:362–4): (94)

a. Mar´ıa var tekin af lögreglunni. Maria was taken.f.sg.nom by police.the ‘Maria was apprehended by the police.’ ´ tel b. Eg Mar´ıu hafa verið tekna (*tekin) af lögreglunni. I believe Mary.acc to.have been taken.f.sg.acc by police ‘I believe Maria to have been apprehended by the police.’

There are thus at least two types of T when it comes to case properties: finite Ts that assign (nominative) case to NPs, and participial Ts that can receive case from NPs. Bantu languages can have complex tenses with multiple Ts of the first kind; IE languages can only have complex tenses in which there is one T of the first kind and all the other Ts are of the second kind. Of course, lexical factors also play into whether the potential for multiple agreement in complex tenses is realized or not. The “participial” T on the main verb in a Bantu language could, for example, happen not to be a functional head that undergoes agreement. This happens sometimes in Kinande: some auxiliaries select an infinitival form of the main verb, with prefix eri, rather than the participial form shown in (88), with prefix ka. Verbs with eri do not agree, so there is only IE-like single agreement on the auxiliary in these particular constructions (e.g., tu-lwé tu-ka-lya ‘we-are we-eating’ vs. tu-lwe b’eri-rya ‘we-are to-eat’, meaning ‘we ought to eat’). The Bantu language Luvale seems to have only complex tenses based on the infinitival form of the verb. As a result, it never has double agreement: (95)

Tu-li na-(*tu)-ku-va-ngula. (Horton 1949) 1pS-be with-(1pS)-inf-2O-teach ‘We are teaching them.’

The implications we derive in this domain thus go only one way. It follows from a positive setting of parameter (2) that a language will not have full double agreement in auxiliary-plus-main verb constructions, but it does not follow from a negative setting of parameter (2) that a language must have double

212


agreement in auxiliary-plus-main verb constructions. That depends also on the lexical properties of the morphemes involved in the construction. 5.9.2 Auxiliary constructions in other languages Checking whether other languages have multiple agreement in auxiliary constructions is relatively straightforward. Some languages do not appear to have any relevant auxiliary constructions; either these languages simply lack nonthematic verbs that select VP complements, or the main verb incorporates into the auxiliary verb to form a single complex head, which cannot bear more than one set of agreements. Such languages do not have evidence of this sort for how parameter (2) is set. Examples include Mohawk, Abaza, Mapudungun, Mayali, Gooniyandi, Yimas, Nez Perce, Mixtec, Karok, Wichita, Pirahã, and Wari. Of the languages that do have some sort of auxiliary constructions, a healthy minority (about 23) are like Bantu languages in having full person-numbergender agreement on the main verb as well as the auxiliary in at least some cases. Examples of this type include Berber, Arabic, Georgian, Burushaski, Arapesh, Fijian, Maricopa, Tukang Besi, Halkomelem, Ojibwa, Slave, Otomi, Imbabura Quechua, Guaran´ı, Canela-Krahô, Jarawara, Jakaltek, and Zoque. For example, Cherchi (1999:25–6) lists some contexts in which an auxiliary verb is used in Georgian, and his examples show double agreement with a first person subject: (96)

a. v-my.er-i-v-ar. 1sS-sing-ptpl-1sS-be ‘I am singing.’ b. c.a-v-sul-v-ar. prev-1sS-go-1sS-be ‘I have gone.’

Thus, agreement with the first person argument appearing on the main verb does not disqualify the auxiliary verb from agreeing in person with the very same argument, as expected if Georgian is not subject to (2). Similarly, (97) shows that both the auxiliary and the main verb bear full agreement in Burushaski (Lorimer 1935). (97)

u:n ε xidmt ε cˇ -a ba. (p. 317) a. Jε I.nom your service do-1sS be.1sS ‘(for these many years) I have been at your service.’ b. ε cˇ -u bo; ε cˇ -i bi; ε cˇ -a ba:n (p. 274) do-FsS be.FsS do-xS be.xS do-1S be.1pS ‘she is doing it; X is doing it; we are doing it’


213

Burushaski and Georgian are also languages in which T can agree with ergative subjects as well as nominative ones, and v can agree with dative objects as well as with nominative ones. So there is converging evidence that these languages are not subject to the parameter in (2).39 There are also many languages (about 44) in which the auxiliary verb bears agreement but the main verb does not; these include Bagmiri, Lango, Hausa, Hebrew, Finnish, Basque, Turkish, Chukchi, Kannada, Asmat, Tauya, Amele, Alamblak, Kewa, Mangarrayi, Maung, Tiwi, Dani, Lavukaleve, Chamorro, Greenlandic, Acoma, Yagua, Warao, Barasano, Choctaw, Hixkaryana, and Wich´ı. Crucially we expect those languages in which agreement is tied to case to be in this group, and that is true. For example, Greenlandic verbs do not agree with oblique-case arguments, but only with arguments in structural case (absolutive and ergative), as shown in (81). This goes along with the fact that only the auxiliary verb bears agreement in complex tenses in Greenlandic (Fortescue 1984:48): (98)

Niqi niri-niq ajur-pa-a-t. meat eat-noml neg.hab-ind-3sO-3pS ‘They don’t eat meat.’

Apurinã was another language in which agreement seems to be case dependent in other respects (see (82)), and it too has agreement only on the auxiliary when one is present: (99)

komeru. (Facundes 2000:295) Pitxeka i-txa grow 3S.m-aux manioc.m.sg ‘The manioc grew.’

Some readers may have noticed that nothing in my account requires that the sole agreement in a language subject to (2) be on the auxiliary rather than on the main verb. My line of reasoning implies that there can only be one agreement per argument in such a language, but it says nothing about where that agreement must be. In fact, having agreement on the main verb but not the auxiliary does not seem to be common, but is attested in, for example, the Mayan language Tzotzil: (100)

a. Ba j-ta-tikotik j7ilol. (Aissen 1987:17) went 1E-find-1.pl.ex shaman ‘We went to find the shaman.’

39 Warlpiri was another language of this type, but it does not have multiple agreement in auxiliary constructions. I assume that this is a result of lexical factors, not the parameter settings, as in Luvale.

214

Parameters of agreement b. Laj j-maj-ot. end 1E-strike-2sA ‘I’m done hitting you.’

The impossibility of agreement appearing on the higher verb as well is consistent with other evidence that Tzotzil is a case-dependent language; its verbs never agree with oblique complements marked with ta, for example: (101)

7i-tal ta Saklum. (Aissen 1987:11) compl-come to Saklum ‘They came to Saklum.’ (Not: *7i-s-tal ta Saklum ‘compl-3E-come-3A to Saklum’)

I thus leave open the possibility that agreement can be on the main verb rather than on the auxiliary verb – although one would like to know why this is not very common (see section 5.11.3.2 for another example of this kind from Huallaga Quechua). Finally, I point out that there could be constructions in which the same argument triggers agreement on more than one head other than the auxiliary constructions focused on here. The existence of any such construction will tend to show that (2) is not obeyed in the language, given that an NP can only be casevalued once. Another situation where this arises is the long distance agreement constructions in Passamaquoddy and other Algonquian languages, discussed briefly in section 3.3.4. In these constructions the same argument triggers subject or object agreement on the lower verb and object agreement on the higher verb. Neither the lower T nor the matrix v undergoes case concord, and both of them agree, so these Algonquian languages must not be subject to (2). Similar multiple agreement in LDA constructions is found in Arabic, Canela-Krahô, and Quechua. However, this type of agreement is too rare to be especially useful for typological investigation; I therefore omit a full discussion here. 5.10

A third value for the Direction of Agreement Parameter

There is one other case of multiple agreement with a single argument that is worthy of special discussion, however, because it points to a third possible setting for the Direction of Agreement Parameter.40 This is the interesting fact 40 The analysis of Burushaski presented in this section, and its implications for the statement of the Direction of Agreement Parameter, was a late-breaking discovery made possible by my becoming aware of sources on this language other than Lorimer 1935. It was added to the book while it was already in production, so its implications could not be fully integrated into the text as a whole. I thank Beatriz Fernández for calling the additional sources to my attention, and for relevant discussion.


215

that unaccusative verbs show both subject and object agreement with their sole argument in Burushaski, as seen in (102). (102)

Das´ın há-e le mó-yan-umo. (Willson 1996:3, 19, 21) girl(abs) house-obl in 3sO.f-sleep-3sS.f/past ‘The girl slept in the house.’ (similarly, ‘die,’ ‘wake up,’ ‘rot,’ ‘be lost’ . . .)

At a minimum, examples like (102) show us that the non-theta-marking v present in unaccusative structures can be a probe for agreement in Burushaski, just as the theta-marking v found in transitive constructions is. Burushaski is different in this respect from many languages with object agreement, in which only the active transitive v is an agreement-bearer. But given this, the possibility of (102) is not so surprising, since we already know that the Case Dependency of Agreement Parameter is set “no” in Burushaski (see (21), (79), and (97)). Thus v need not assign objective case to the theme argument in order to agree with it in Burushaski, and v agreeing with this NP does not prevent T from also agreeing with the same NP, the way it would in an Indo-European language in which parameter (2) is set “yes.” It is notable, however, that unergative verbs in Burushaski do not show object agreement as well as subject agreement with their sole argument, as seen in (103). (103)

Das´ın há-e le huruT-umo. (Willson 1996:3, 19, 21) girl(abs) house-obl in sit-3sS.f/past ‘The girl sat in the house.’ (similarly, ‘dance,’ ‘walk,’ ‘cry,’ ‘come out’ . . .)

The question is why. The v used in unergative constructions is no different from the v used in transitive constructions in most conceptions, and we know that the latter does bear agreement – so why can’t the former? An interesting possibility is that v in unergative constructions cannot agree with the subject because the syntactic configuration is wrong for agreement to take place in this language. That would be the case if, in fact, functional heads in Burushaski need to probe downward to find a goal to agree with, rather than upward. Suppose then that we revise the Direction of Agreement Parameter stated in (1) so that it includes a third possible setting, as expressed in (104). (104)

The Direction of Agreement Parameter (i) F agrees with DP/NP only if DP/NP asymmetrically c-commands F, or (ii) F agrees with DP/NP only if F c-commands DP/NP, or (iii) F agrees with DP/NP only if F c-commands DP/NP or vice versa.

(104i) is the now-familiar Bantu setting of this parameter and (104iii) is the IE setting; (104ii) is the third logical possibility, which I propose for Burushaski.

216


This version of the parameter is more symmetrical than the original one: it says that a language can limit itself to downward agreement as well as to upward agreement. It is, therefore, to be preferred on conceptual grounds. The parameter setting in (104ii) accounts for the difference between (102) and (103) as follows. The sole NP argument of an unaccusative verb originates inside the VP complement of v; hence v c-commands NP (prior to any movement) and can agree with it in accordance with (104ii). In contrast, the sole overt argument of an unergative verb originates in Spec, vP, a position that v does not c-command. Therefore, v cannot agree in this structure, given (104ii). The two structures are compared in (105). (105)

a. [TP -- [vP [VP girl sleep ] v+AGR ] Past+AGR ]

b. [TP -- [vP girl [VP sit ] v+*AGR ] T+AGR ]

Object agreement in Burushaski has a second somewhat unusual property, which can also be attributed to this parameter setting. Willson (1996) shows that verbs like ‘tell’ and ‘give’ that select arguments in dative case show object agreement with their dative argument ((106a)). In contrast, verbs that appear with a freely added benefactive expression in dative case do not show object agreement with the benefactee ((106b)). (106)

mu-ú-imi. (Willson 1996:3) a. Hilés-e das´ın-mo-r toofá boy-erg girl-obl-dat gift.x(abs) 3sO.f-give.x-3sS.m/past ‘The boy gave the girl a gift.’ b. Jé-e gó-e-r cái tayáar e´ -t-am (Willson 1996:36) I-erg you-obl-dat tea.y(abs) prepare 3sO.y-do-1sS/past ‘I prepared tea for you.’

Most other languages treat examples like (106) the same: either they show object agreement with neither oblique phrase (if agreement is case-dependent in the language, or if the oblique phrase counts as a PP), or they show agreement with both dative expressions (e.g., Basque). Although the Case Dependency of Agreement Parameter cannot draw this distinction, the parameter setting in (104ii) arguably can. Verbs like ‘give’ select goal expressions as a lexical property; therefore, the goal phrase is generated inside the VP complement of v along with the theme argument. In contrast, verbs like ‘prepare’ do not inherently select a benefactive expression; this phrase is arguably theta-marked instead by an applicative head (Marantz 1993), which is phonologically null in


217

Burushaski but overt in many other languages. Suppose then that we add the assumption that the applicative head takes vP as its complement rather than the other way around in Burushaski. Then the structures of (106a,b) are roughly (107a,b). (107)

a. [TP -- [ vP boy [VP girl-DAT gift give ] v+AGR ] Past+AGR ]

b. [TP -- [ApplP you-DAT [ vP I [ VP tea prepare ] do+AGR ] Appl ] Past+AGR ]

The v c-commands the dative phrase in (107a) but not in (107b); therefore it can agree with the dative phrase in (107a) but not in (107b), given (104ii). The same parameter that explains why there is object agreement with “low” subjects but not higher ones ((102) vs. (103)) can also explain why there is object agreement with dative phrases selected by V but not with unselected dative phrases. There is some reason to say that the parameter setting in (104ii) applies to other functional heads in Burushaski as well. Some postpositions and quantifiers (‘both,’ ‘all’) agree with their NP complements in Burushaski; this confirms that downward agreement is possible in this language, although not that it is required. A hint that T also must agree downward comes from transitive sentences with a “scrambled” OSV order, such as (108); I tentatively analyze these as being the result of moving the object to Spec, TP rather than the thematic subject (see section 9.5.2 above). (108)

a. Ne hir cél-e Ø-cú-m-i (Grune 1998:16) the man(abs) water-erg 3sO.M-carry.away-past.II-3sS.n ‘The man was carried away by the water.’

b. [TP man [vP water-ERG [VP carry ] v+AGR ] Past+AGR]

Notice that the agreement suffix associated with finite T still agrees with the thematic subject in Spec, vP, not with the moved object in Spec, TP (see also (23)). Burushaski is like the IE languages rather than like Kinande in this respect. The question is why. For an IE language, agreement between T and the fronted object is blocked by the Case Dependency of Agreement Parameter: the fronted object has accusative case, not nominative case assigned by T; therefore T cannot agree with it. But this reasoning does not extend to Burushaski, because the Case Dependency of Agreement Parameter is set “no”; T clearly can agree with NPs that bear absolutive case as well as those that bear ergative case

218


(see (102) and (103)). Fortunately the parameter setting in (104ii) can explain this pattern. T cannot agree with the thematic object in the post-movement structure shown in (108b), because T does not c-command the object in Spec, TP. T does c-command the thematic object in its base position inside VP prior to movement. However, T also c-commands the thematic subject in Spec, vP, and the subject therefore blocks T from agreeing with the object in its pre-movement position, by the intervention condition on agreement. The “downward only” setting of the Direction of Agreement Parameter thus explains why T always agrees with the subject in Burushaski. If this analysis is correct, it should also have implications for agreement on predicate adjectives in Burushaski, since the subject of predication is never in the c-command domain of FA , the functional head associated with adjectives (see chapter 2 and section 5.3 above). And there is something interesting to explain in this area. Whereas Lorimer (1935:111) says only that adjectives “sometimes appear in plural forms when the noun to which they refer is in the plural,” Tiffou and Pesot (1989:53–4) observe a distinction between attributive adjectives and predicate adjectives. They imply that agreement in number is required when the adjective is an attributive modifier, but say that agreement is optional when the adjective is predicative: (109)

a. iskó e´ ser-um-iŋ (*éser-um) γatéha-ŋ three sharp-adj-pl sharp-adj sword-pl.y ‘three sharp swords’ b. We guˇs-iŋ a kanˇce´ ni/ kanˇce´ ni-tiŋ ban. these women-f.pl flighty flighty-f.pl be.f.pl ‘These women are flighty.’

I interpret this as follows. When an AP is an attributive modifier of NP, the projection of FA merges directly with NP and hence FA can agree with NP in accordance with (104ii).41 When AP is the complement of Pred, FA does not merge with NP, and the subject NP is never in its c-command domain; hence FA cannot agree with NP in these circumstances, in accordance with (104ii). (104ii) thus induces a difference in agreement between attributive structures and predicative structures in Burushaski that is not seen in Bantu languages or IE languages.

41 The technical definition of c-command would need to be adjusted somewhat so that FA counts as c-commanding the NP (or as being c-commanded by it) in the relevant adjunction structure. While I cannot enter into the technicalities here, the crucial structural difference between the two structures involving APs should be clear.

Many little parameters or two big parameters?

219

To complete this account, I need to say why agreement is possible as an option in (109b), whereas agreement is either required or forbidden in most other contexts. I claim that the reason is because (109b) is structurally ambiguous: it can be parsed as a simple predicate adjective structure, or it can be parsed as a predicate nominal structure in which the adjective is an attributive modifier of a null noun. The latter structure is possible in Burushaski by all accounts; Lorimer (1935:107) observes that when the subject is singular, the predicate adjective is often followed by the indefinite determiner n, and says that such sentences are properly translated as in (110). (110)

Huq mat-um-n bi (cf. mt-um huk) dog black-adj-one be.3.sg.n black-adj dog ‘The dog is a black one.’

The version of (109b) with agreement is, I claim, analogous to (110): the null N head is plural in agreement with its antecedent (not part of Agree: see section 4.2), and FA agrees with the N it modifies in number, just as it does in (109a). In contrast, when the complement of the copula is a simple AP, agreement never takes place. If this is right, it provides converging evidence that functional heads only probe downward for something to agree with in Burushaski – in contrast to Kinande, in which they only probe upward. The properties of Burushaski thus lead us to a more symmetrical and attractive conception of the Direction of Agreement Parameter. We no longer need to ask what is special about upward agreement, such that some languages would restrict themselves to that kind of agreement. It turns out that there is nothing particularly special about upward agreement; languages can also restrict themselves to downward agreement. Unfortunately, I do not have the kind of data that would distinguish the (104ii) setting of the parameter from the (104iii) setting for languages other than Burushaski and IE languages. Therefore, I recombine these two logically distinct cases back into one for purposes of the broad typological survey that the next section engages in. 5.11


I have presented a detailed comparison between agreement in NC languages and agreement in IE languages, showing that there are consistent differences that show up in a wide range of syntactic environments. This suggests that the differences are created by two overarching parameters; any other view seems to be missing a generalization. I have also shown that a similar range of phenomena can be seen in various other languages, filling out the range of possibilities and

220


proving that the issues are not unique to these two families. I also noted in passing that in several instances the parameter values in these other languages seem to be consistent across various functional heads. The skeptic could still say that it is a coincidence that all of the functional categories in NC languages work in a consistent way, and that all of the functional categories in IE languages work consistently in the opposite way. These are possible languages, of course, but perhaps languages that have some functional heads that behave one way with respect to agreement and other functional heads that behave another way are equally possible. The same skeptic might be suspicious that I have cherry-picked those languages that look best for my view. And the skeptic would be right. To really nail down the question, we have to find a way to look more broadly at what kind of languages are possible and common versus what kinds (if any) are impossible or surprisingly rare. In other words, we need to do some statistics. 5.11.1 A 108-language survey of agreement behavior One way to pose the question is to ask how many kinds of languages there are with respect to the syntax of agreement. Are there roughly the four kinds of languages predicted by the two logically independent, binary-valued parameters in (1) and (2)? Or are there the hundreds of different kinds that could in principle result from applying (1) and (2) to each functional head in a language individually? I investigated this by considering the 100 languages in the World Altas of Language Structures core language sample, with a few substitutions, plus a few additional languages that I had good information on and found interesting. More specifically, I replaced Ambrat with Tauya, Rama with Ika, and Kutenai with Nez Perce, because I could not find sources on those languages. Also in a handful of cases I switched a language in the WALS sample for another in the same family that I know better: Mohawk for Oneida, Choctaw for Koasati, Ojibwa for Cree, Tzotzil for Jakaltek. Finally, the eight languages that I added were: Halkomelem, Mayali, Warlpiri, Yimas, Nahuatl, Kinande, Tariana, and Jarawara. The result is a relatively large sample, genetically and areally diverse, and not personally constructed by me to prove a point. For each language, I looked for evidence relevant to the following eight specific issues: (111)

a. Is subject agreement dependent on the position of the subject? b. Is subject agreement sensitive to the case of the subject? c. Is object agreement dependent on the position or definiteness of the object?


221

Table 1 Agree must be up ((1) = Yes)

Agree can be up or down ((1) = No)

Agreement dependent on case ((2) = Yes)

Turkish, Lango, Greenlandic, Apurinã, Chamorro, Mapudungun (n = 6)

IE languages (7), Hausa, Finnish, Abkhaz-Abaza, Kannada, Asmat, Amele, Alamblak, Maung, Mangarrayi, Tiwi, Lavukaleve, Daga, Yimas, Lakhota, Tzotzil, Warao, Barasano, Yagua, Wich´ı, Choctaw, Hixkaryana, Hebrew, Wari’, Chukchi, Makah (n = 32)

Agreement independent of case ((2) = No)

Zulu, Swahili, Kinande, Berber*, Arapesh, Tariana, Fijian, Tukang Besi, Slave, Canela-Krahô, Jarawara (n = 11)

Georgian, Arabic, Persian, Warlpiri, Dani, Kewa, Burushaski, Mayali, Halkomelem, Tauya, Ojibwa, Nez Perce, Karok, Otomi, Zoque, Ika, Basque, I. Quechua, Guaran´ı (n = 19)

r Downward agreement possible, case dependence indeterminate: Khoekhoe, r r r r r

Kiowa, Nahuatl, Pirahã, Gooniyandi Upward agreement required, case dependence indeterminate: Luvale, Mohawk? Not case dependent, direction indeterminate: Maricopa Case dependent, direction indeterminate: Bagirmi, Acoma Both parameters indeterminate: Wichita No Agreement: Sango, Yoruba, Supyire, Grebo, Krongo, Koyra Chiini, Harar Oromo, Khalkha, Japanese, Korean, Lezgian, Mandarin, Burmese, Meithei, Hmong, Thai, Vietnamese, Paiwan, Indonesian, Malagasy, Rapanui, Tagalog, Imonda, Kayardild, Martuthunira, Ngiyambaa, Yaqui, Mixtec, Sanuma (n = 29)

d. e. f. g.

Is object agreement sensitive to the case of the object? Is there agreement on complementizers, and if so what kind? Is there agreement on adpositions with an unmoved complement? Is there agreement between a determiner/quantifier and the associated NP, and if so, does it depend on word order? h. Is there double agreement with the subject in auxiliary-verb constructions?

In general, the answers to (111a,c,e–g) are relevant to the setting of parameter (1), and the answers to (111b,d,h) are relevant to the setting of parameter (2). The details of what I found are summarized in the table in the appendix. It turns out that most of the languages can be categorized into one of five types, as shown in table 1.

222


Not all parameter settings turn out to be equally common in this survey. Languages in which agreement is dependent on case are about 33% more common than languages in which agreement is not dependent on case (40 to 31). More strikingly, languages that allow downward agreement between a probe and its goal outnumber languages that require upward agreement by an almost three-toone ratio (53 to 19). Languages in which agreement is subject to both restrictions seem to be particularly rare, with only a handful of plausible examples attested. It would be nice to understand these asymmetries in the distribution of language types. But for now I take them to be relatively insignificant to the point at hand. There is no strong expectation that both values of a binary parameter are equally distributed in the languages that are currently spoken in the world. The world is too small a place, and the way that languages diffuse through it and interact with one another is too nonrandom to expect complete statistical independence. What I take to be significant about table 1 is the fact that each of the four agreement types is found in a variety of language families and in different areas of the world. Even the rarest kind, the “Yes-Yes” languages, are attested on four continents and in six distinct language families. Thus, we are not dealing with a unique areal phenomenon, like the use of clicks in the languages of Southern Africa, or the existence of object-initial word order in Amazonian languages. Rather, the different agreement types are robustly attested in languages of the world in a way that is comparable to the famous Greenbergian word order types of SOV, SVO, and VSO (Greenberg 1963, Dryer 1992). Even more important for the theory of agreement – and the theory of parameters – is the question of how consistent are the parameter settings across functional categories. Suppose there was a language that only had agreement on T. One could use the properties of that agreement to classify the language in terms of table 1, following the model of section 5.2. But such a language would not provide evidence that all functional categories are governed by the same parameter settings. One might also wonder about the (fairly few) languages that I found to be indeterminate with respect to one parameter or the other. Were the results indeterminate because I found no convincing evidence for how the parameter in question is set, or because I found conflicting evidence regarding how the parameter is set? If it was the latter, then a “microparametric” theory of agreement that respected the Borer-Chomsky Conjecture might turn out to be better than the “macroparametric” version in terms of grammatical parameters that I have presented. What we need to know in this regard is how many languages have consistent behavior across more than one functional category, versus how many languages have inconsistent behavior. Consider first the Case Dependency of Agreement Parameter. In practice, there are three principal opportunities to observe the value of this parameter:


223

whether T agrees with oblique subjects or not, whether v agrees with oblique objects or not, and whether the language allows more than one head to agree with the same argument. What then is the degree of convergence between these three data points? For T agreeing with an oblique subject and v agreeing with an oblique object, there is not a lot of relevant data. To evaluate this, we need languages that have overt subject agreement, overt object agreement, ergative or dative subjects, and a dative/accusative distinction for objects. Not that many languages have all four properties. I found approximately sixteen, which pattern as follows: (112)

a. Consistently case-sensitive: 7 (Greenlandic?, Choctaw, Tzotzil?, Yimas?, Daga, Wari’, Mangarrayi, also Peruvian Quechua – see section 5.11.3.2) b. Consistently insensitive to case: 7 (Georgian, Burushaski, Warlpiri, Ika, Basque, Canela-Krahô, Imbabura Quechua – see section 5.11.3.2) c. T case-sensitive, v not: 1? (Nez Perce – see section 5.11.2.2) d. v case-sensitive, T not: 1 (Gooniyandi)

There are more consistent languages than inconsistent ones, but there are a couple of candidates for the latter, and the numbers are too small to draw any firm conclusions. Better results come from collapsing the two possible instances of agreement with oblique NPs into a single data point and correlating it with instances of multiple agreement with the same NP. The implicational universal predicted by my parameters is that languages which rule out agreement with oblique arguments should also not permit multiple agreement with the same argument (see section 5.9). This is nicely confirmed by the data: (113)

a. No agreement with obliques; no multiple agreement: 20 languages (Hindi, Apurinã, Daga, Choctaw, Hixkaryana . . .) b. Agreement with obliques; multiple agreement OK: 11 languages (Georgian, Burushaski, Ika, I. Quechua, Maricopa, Jarawara . . .) c. Agreement with obliques; no multiple agreement: 6 languages (Warlpiri, Basque, Chukchi, Tauya, Dani, Kewa) d. No agreement with obliques; multiple agreement OK: 0 languages

I take this to be a very positive result for the parametric theory. Note also that the “mixed” type in (113c) is somewhat less common than either of the consistent types in (113a) and (113b). Such examples can arise when the parameter values allow multiple agreement, but a suitable situation for multiple agreement to appear never arises because of lexical factors. Apparently such lexical conspiracies to hide the parameter value are not impossible, but are not especially common.

224


Next consider the Direction of Agreement Parameter. Here the results are even more impressive. There are five opportunities to observe this: agreement on T, on v, on D, on C, and on P. Most languages do not have all five kinds of agreement, but many have at least two, and quite a few have three or more. Call a language consistent up if at least two functional heads only agree with an NP that asymmetrically c-commands them, and all other functional heads either confirm this value, or are indeterminate, or are not applicable because they do not bear agreement. Conversely, call a language consistent down if at least two functional heads can agree with an NP that they c-command, and all other functional heads either confirm this value, or are indeterminate, or are not applicable because they do not bear agreement. Call all other combinations inconsistent. Obviously there is a large variety of imaginable languages that would count as inconsistent by this criterion. But there are not many actual languages that are inconsistent. On the contrary, there are almost none:42 (114)

a. Consistent up languages: 16 b. Consistent down languages: 43 c. Inconsistent languages: maybe 2 (Berber, Nez Perce)

I take this to be strong support for the idea that the parametric values for the agreement properties of functional heads are not independent of each other. On the contrary, languages tend to be remarkably uniform in these respects. This supports the classical GB notion of a parameter as variation in the formulation of a principle of grammar, rather than the Borer-Chomsky Conjecture that all parameters reduce to variation in the feature-specifications of individual functional items. 5.11.2 Two apparent exceptions Having already cherry-picked some of the more interesting positive examples for my grammatical parameters in previous sections, let me now cherry-pick two of the more interesting problematic examples for a bit more discussion. 42 There are two languages, Abkhaz-Abaza and Tzotzil, that have ergative agreement systems which can be analyzed as v agreeing upward with the agent in Spec, vP while T agrees downward with the object (see section 3.2.2.1). I assume that v is forced to agree upward in these languages not by parameter (1) but by parameter (2): v agrees only with an NP that it gives ergative case. Ergative case is not marked overtly on NPs in these languages, but following O’Herin (2002) I assume it is present in the syntactic representation and guides agreement. If this analysis proves untenable, these two languages might be switched from the consistent-down category to the inconsistent category.


225

Showing why they seem problematic hopefully illustrates what my thesis is, for better or for worse. These languages are, in my current understanding, as bad as it gets for my hypothesis, and I believe that they are not so bad. There are other things going on in the languages which, while far from being fully understood, might very well affect how the parameters apply. I am not so bold as to say that these languages support my parameters. I am open to the idea that they are neutral or even genuine counterexamples to it – either because lexical parameters happen to conceal the grammatical parameter settings, or for some other reason. But the macroparametric view has at least the virtue of pressing one to dig deeper into the syntax of these languages, and when one does one discovers some interesting things. I consider the two possibly inconsistent languages mentioned in (114), Berber and Nez Perce. 5.11.2.1 Berber Berber is an Afroasiatic language, related to Hebrew, Arabic, and Hausa. It is like the Niger-Congo languages and Arabic (but unlike Hebrew or Hausa) in allowing multiple agreement in complex tense constructions, as shown in (115). (115)

la teddu-n . . . (Ouali and Pires to appear) La-n past-be.perf-3pS pres leave-impf-3pS ‘They were leaving/had been leaving . . .’

This suggests that agreement in Berber (like Arabic) is not case-dependent in the sense of parameter (2). However, Berber does not have an overt dative case or other oblique cases, so we cannot confirm this parameter setting by checking whether agreement is possible with oblique noun phrases or not. Berber is thus consistent with respect to this parameter, although not in a particularly interesting way. The difficult issue concerns the Direction of Agreement Parameter in (1). Berber is a VSO language, in which the finite verb routinely agrees with the postverbal subject: (116)

urgaz lfilm. (Sadiqi 1997) a. I-zra 3sS.m-see man.cs film ‘The man saw a film.’ b. N-zra-t nuKni. 1pS-see-him we ‘We saw him.’

226


This strongly suggests that Berber allows downward agreement from T to the subject in Spec, vP (or some other specifier position lower than Spec, TP). In contrast, D-like heads in Berber only agree with NPs to their left, not with NPs to their right. For example, maTa ‘which’ precedes the associated NP and is invariant, whereas the ordinal numeral ‘second’ follows the NP and agrees with it: (117)

a. maTa urgaz (Sadiqi 1997: 128) which man.m.sg.cs ‘which man?’ b. argaz wiSin; tamtut tisnat man.m.sg second.m.sg woman.f.sg second.f.sg ‘the second man; the second woman’

Even more striking, the word kuLu ‘all’ can come either before the associated nominal or after it. When it comes before the nominal, it is invariant, but when it follows the nominal it agrees with that nominal: (118)

a. kuLu irgzn all men.m.pl ‘all (the) men’

(Sadiqi 1997:128–9)

b. irgzn kuLu-tn; tajtˇsin kuLu-tnt men.m.pl all-m.pl women.f.pl all-f.pl ‘all the men; all the women’

This is strongly reminiscent of the behavior of D-like heads in Kinande and Swahili, and suggests that D-like heads in Berber can only agree with something that c-commands them. We have, then, a possible instance of inconsistent parameter settings in Berber: parameter (1) is set “no” for finite T and “yes” for D. (Other functional heads in Berber give no additional data on this issue, since they are not agreement bearers.) Can anything be made of this relatively rare anomaly, beyond identifying it is as an apparent counterexample? One dull but reasonable possibility is that Berber does in general allow downward agreement, as IE languages do. That covers (116) in a straightforward way. One can then say that it is simply a coincidence (synchronically speaking) that the Ds that precede the associated NPs happen not to be agreement-bearers and (some of) the Ds that follow NP are. The word kuLu ‘all’ could have two distinct lexical entries, one that agrees and one that does not. Such an analysis would be descriptively adequate. It may fail to capture a generalization about the relationship between word order and agreement within DP, but since D is a closed class with a relatively small


227

number of members the loss would be a fairly small one.43 (Note that there are also Ds that follow NP and do not manifest agreement with it, such as the demonstratives ad ‘this’ and ann ‘that’ (afrux-ad ‘this boy’ vs. ifruxn-ad ‘these boys’ (Ouhalla 2005a: 628)). Thus lexical properties that determine whether a D agrees or not must come into the story on all accounts.) A riskier but potentially more rewarding possibility is that the Ds show that the true setting of the Direction of Agreement Parameter in Berber is “yes,” and something special is going on in VSO examples like (116). What could it be? Conceivably the subject NP moves from Spec, vP to Spec, TP, thereby triggering upward agreement on T, just as in a Bantu language. The difference could be that (exceptionally) the lower copy of the subject NP is spelled out in Berber and the higher copy is deleted, whereas it is the higher copy that is spelled out in Kinande and most other languages. This derivation is sketched in (119), where strikethrough indicates the copy of a moved item that is deleted at PF. (119)

[TP mani T{AGRi }+seek [vP mani v [VP seek film]]]

Bobaljik (2002) defends the idea that sometimes it is the lower copy of a movement chain that is spelled out at PF rather than the higher one, and explores some of the ramifications of this idea (see his article for other references). This analysis would be a kind of revival in current terms of an earlier proposal concerning VSO in Berber by H.-S. Choe (1986), who claimed that the subject moves downward from the normal subject position so as to appear after the finite verb in Berber. One should probably not consider adopting this second, riskier analysis unless one has at least a glimmer of an answer to why the subject chain should be spelled out differently in Berber than in most other languages. Fortunately, Choe had an idea about this that still seems relevant. She claimed that principles of case assignment determine where the subject NP appears in Berber. More specifically, she claimed that case is only assigned by the agreement-bearing verb to its right in Berber. And indeed there are some features of subject casemarking in Berber that do seem typologically unusual and could plausibly 43 Probably I would take this route for related Hebrew. For Hebrew, the evidence seems quite strong that T can agree downward with the subject, even if something else has moved to Spec, TP, given detailed studies like Borer 1995 and Shlonsky 1997. There are a couple of individual lexical items in Hebrew that seem to agree only with a DP on their left (’eyn ‘not’ and kol ‘all’; Shlonsky 1997) but there are no large classes of items that systematically behave this way. I thus assume that this agreement behavior reflects the lexical properties of those particular items, not the general parameter settings of Hebrew.

228


be related to the language’s anomalous agreement and word order properties. Subject case in Berber is definitely not the familiar nominative case known from Indo-European languages. Rather, the postverbal subject appears in the so-called construct state (Guerssel 1983, Sadiqi 1997). This form is also found on the objects of prepositions and on NPs within NP, so it corresponds not only to the IE nominative but also to the IE dative and genitive. In each of these domains, the NP in construct state immediately follows its case-assigner (V, P, or N). The phonological realization of construct state is also somewhat unusual: whereas nominative case is usually unmarked, the construct state is a marked form, realized as a phonological change at the beginning of the NP that involves deleting the first vowel of the word along with sometimes inserting a glide (see Guerssel 1983 for details). Perhaps then construct case is really a form of phonological weakening that results from the case-marked NP forming a prosodic unit with the word that immediately precedes it. If so, it is not implausible that the PF rules that spell out case in this unusual way might take precedence over the normal rules of chain realization, causing the lower copy to be spelled out rather than the higher one, consistent with Bobaljik’s (2002) vision. Another hint that this might be so involves the placement of overt subjects in complex tenses. This seems not to be entirely fixed: the subject can appear immediately to the right of the auxiliary verb or immediately to the right of the main verb, as shown in (120). (120)

a. Ila Ali da thiddu. (Ouali and Pires to appear (14)) past.be.3sS.m Ali fut go.aor.3sS.m ‘Ali was going to leave.’ b. Da illi-n la teddu-n lwashun. fut be.aor-3pS pres go-impf.3pS children ‘The children will be leaving.’ (Ouali and Pires to appear (19))

Still, the freedom of placement of the subject is not unrestricted; it seems never to appear between the tense markers da and la and the main verb, for example. This limited freedom is captured by the descriptive statement in (121). (121)

The case-marked subject must be spelled out right-adjacent to an inflected verb (with which it forms a phonological unit?).

But the statement in (121) also explains why the copy of the subject in Spec, TP cannot be spelled out in (119), whereas the copy in Spec, vP can be. To the extent then that the PF spell-out rule in (121) is already motivated by (120), we


229

have independent reason to think that PF spell-out also governs the appearance of VSO rather than SVO word order in Berber. Finally, this theory makes the prediction that objects and PPs should not be able to move to Spec, TP to derive X-V-S orders in which the verb agrees with the subject in Berber. Such sentences are sometimes possible in languages that really do allow downward agreement, including Icelandic, Yiddish, Finnish, and Hebrew. The reason why this should not be possible in Berber is simply because the Spec, TP position is occupied in the syntax by the highest copy of the agreed-with subject; otherwise the subject would not be agreed with, given a “yes” setting of parameter (1). If the subject is in Spec, TP in Berber, then it follows that that position is not an open landing site for movement of another phrase. This prediction seems correct: phrases other than the subject can appear on the left edge of the clause, but only as a result of clefting or left dislocation, not as a result of simple movement to Spec, TP. Thus, the fronted object in (122) requires there to be an object clitic on the verb, whereas in (123) the fronted object lands in the specifier of a C/Focus head rather than in Spec, TP. (122)

Tabrat i-ara-*(T) hmed. (Sadiqi 1997:148) letter.f.sg 3sS.m-wrote-it.f Ahmed. ‘The letter, Ahmed wrote it.’

(123)

Argaz a t-zra. (Sadiqi 1997: 217) man c 3sS.f-see ‘It’s a man that she saw.’

There are of course many other reasons why a particular language might forbid the movement of a direct object to Spec, TP. But there are languages that do allow such movement, and it is consistent with the proposed analysis that Berber is not one of them. The important point here is probably not that I can think of ways to save my theory; that might always be the case. The most important point about problematic-looking languages like Berber is how rare they are, as documented in section 5.11.1. But one does also learn more when one looks into the details about why a given language might seem anomalous. In this particular instance, one learns something about the special way that “case” is realized in Berber, and how that might interact with issues of agreement and word order. 5.11.2.2 Nez Perce Nez Perce, a Sahaptian language spoken in Idaho, has an interesting and unusual case and agreement system discussed originally by Noel Rude (1982, 1986a, 1986b, 1988) and in generative terms based on his work by Ellen Woolford

230


(1997) (see also Carnie and Cash Cash 2003). The puzzle for my parameters stands out clearly in Woolford’s discussion, so I begin by following her analysis. (Throughout the discussion, I concentrate on the person-marking affixes of Nez Perce. The language also has affixes that mark plural number for subjects and definite objects, as seen for example in (125b), but I leave these aside because they appear to raise no new issues.) Consider first subject agreement, which is presumably on T (note that it is present only in finite clauses, not in nonfinite ones (Rude 1988)). In Nez Perce, this is primarily realized by the alternation between Ø for first or second person versus hi for third person. (124)

a. Ø-páay-na ‘´ıin. (Rude 1986a:126) 1S-arrive-asp I ‘I arrived.’ b. Hi-páay-na háama. (Rude 1986a:126) 3S-arrive-asp man ‘The man arrived.’

Downward agreement on T is presumably allowed because T can agree with a postverbal subject, as it does in Spanish but not Zulu. More generally, word order is extremely free in Nez Perce, and has no effect on agreement (Rude 1986a:125). Thus, there is no evidence that the position of an NP is crucial to whether it is agreed with or not, making Nez Perce quite different from Kinande, Apurinã, and other languages in which (1) is set as “yes.” The Nez Perce verb also shows hi agreement with its third person subject regardless of whether that subject is in nominative case or ergative case, as shown in (125) (Rude 1988:Ex. (31)) (Woolford 1997:Ex. (36)) (as long as it is not blocked by the more specific morpheme pee; see (127b)). (125)

hi-‘w´ı-ye wewúkiye. (Rude 1988:552) a. Háama man.nom 3S-shoot-asp elk ‘The man shot an elk.’ b. Háama-nm hi-néec-‘wi-ye wewúkiye-ne. (Rude 1986a:127) man-erg 3S-pl.O-shoot-asp elk-obj ‘The man shot the elk (pl).’

Therefore, subject agreement is independent of case marking in Nez Perce, as argued by Woolford (1997:213). In my terms, the Case Dependency of Agreement Parameter in Nez Perce is also set as “no.” So far then, Nez Perce seems to be a No-No language, in the same category as Georgian and


231

Burushaski. Moreover, these are the same parameter values that we observed for agreement on C-like heads in Nez Perce; see the discussion of (43) in section 5.4.2. (P and D do not manifest any agreement in ϕ-features with their associated NPs, so they give no information about the setting of these parameters.) The problems arise when one considers agreement with objects. Object agreement is shown most clearly by the alternation between Ø (for 1/2 objects) and ‘e (for third person objects) (Rude 1986a): (126)

‘ime-né. (Rude 1986a:126) a. ‘´ıin Ø-‘ew´ı-ye I (1S/2O)-shoot-asp you-obj ‘I shot you.’ b. ‘´ıin ‘e-‘w´ı-ye wewúkiye-né. (Rude 1986a:126) I (1S)/3O-shoot-asp elk-obj ‘I shot the elk.’

When the subject is third person and the object is first or second person, the prefix is hi as expected ((127a)). A slight (but significant) wrinkle is that when the subject and object are both third person, one does not get the sequence of morphemes hi+‘e, but rather the portmanteau morpheme pée ((127b)). (127)

a. Háama-nm hi-‘w´ı-ye ‘iin-e. (Rude 1986a:126) man-erg 3S/(1O)-shoot-asp I-obj ‘The man shot me.’ b. Háama-nm pée-‘wi-ye wewúkiye-ne. (Rude 1986a:126) man-erg 3S/3O-shoot-asp elk-obj ‘The man shot the elk.’

Now comes the complication: the morphemes ‘e and pée, which explicitly show third person object agreement, are in a sense optional. They appear if and only if the object is definite/topical in the discourse, and if it is marked by the objective case marker ne as in (126b) and (127b). If the object is indefinite/nontopical and not marked by ne, the agreement prefix is Ø or hi, the same prefixes used on intransitive verbs as in (124): (128)

a. Kawá taxc qáamsit Ø-wiyáamk-o’ (Rude 1986a:23) then soon qáamsit 1S-peel-asp ‘Then soon I will peel the qáamsit fruit . . .’ b. Háama hi-‘w´ı-ye wewúkiye. man.nom 3S-shoot-asp elk ‘The man shot an elk.’

(= (125a))

So unlike the subject prefix hi, which is used regardless of whether the subject bears ergative case or unmarked nominative, the object prefix ‘e is only used

232


when the object bears objective case, not when it is unmarked accusative (in Woolford’s terminology). It thus seems that agreement on v is case-dependent in a way that agreement on T is not. Furthermore, the fact that verbs agree with the object only when the object is somehow definite/topical in Nez Perce recalls the behavior of “optional” object agreement in Zulu, analyzed in section 5.7.1. I analyzed the Zulu facts as indicating that v could only agree upward, with an object that had moved to Spec, vP, out of the domain of existential closure. By parity of reasoning, it would be reasonable to view object agreement in Nez Perce in the same light, and indeed Carnie and Cash Cash (2003) propose essentially that analysis. If so, then agreement on v must probe upward, whereas agreement on T can probe downward. It seems, then, that v in Nez Perce has quite different agreement properties from T and C in the same language, contrary to what one would expect if the agreement parameters are grammatical in nature. But once again, if one uses this cognitive dissonance as a spur to look more deeply into the language, one finds other interesting properties that might be relevant. Notice that I moved very quickly from “agreement with the object” to “agreement on the v head” in the previous paragraph, without citing any evidence for this equivalence. The v head is the usual locus for agreement with an object, but that is not necessarily universal. In fact, there is some interesting evidence that object agreement, like subject agreement, is a property of T in Nez Perce. If so, the picture changes dramatically. Some of the evidence that object agreement is on T rather than v is relatively superficial. First, there are never two distinct person agreement morphemes on the Nez Perce verb; rather there is one morpheme that expresses both (see especially pée in (127b)). This systematic fusion could be a sign that there is only one head (T) that agrees with both arguments. Second, I mentioned that subject agreement does not appear on nonfinite verbs in Nez Perce, as one would expect. But neither do object agreement morphemes like ‘e (Noel Rude, personal communication). Object agreement is just as dependent on finiteness as subject agreement is, suggesting that both are housed in T. In this respect, Nez Perce is different from Bantu languages, Mapudungun, and Quechua, in which object agreement is found in nonfinite clauses (evidence that object agreement is agreement on v, not T in those languages). Third, reflexive morphology in Nez Perce replaces both subject agreement and object agreement (Aoki 1973), rather than bleeding ordinary object agreement and leaving subject agreement unaffected, as in many other languages. The most sophisticated evidence that object agreement is on T in Nez Perce comes from clauses that have a possessed subject. A verb that has an intransitive third person subject that is not possessed bears the prefix hi, as in (124b). But a


233

verb that has an intransitive subject that contains a third person possessor may bear the ‘e prefix, and if the subject contains a first or second person possessor, the verb may have the Ø prefix: (129)

a. (‘´ıinim) pike Ø-w´ıine. my.gen mother 3+1-weep.asp ‘My mother wept.’

(Rude 1986b:124–5)

b. Kaa cáalii-nm pike ‘e-wéeyikse. (Rude 1986b:113) and Charlie-gen mother 3+3-cross.asp ‘And Charlie’s mother went across.’

Woolford (1997) puts this type of agreement aside; Rude (1986b) discusses it but does not fully integrate it with other kinds of agreement in Nez Perce. This is understandable: it is an unusual sort of agreement, and not easy to integrate into the system. Nevertheless, the ‘e and Ø morphemes in (129) are presumably the same morphemes that are used as object agreement morphemes in (126), since they are both homophonous with them and sensitive to the same person features (‘e signals that there is a third person NP distinct from the subject in both structures; Ø signals the presence of a first or second person NP in both). It is thus desirable to collapse the two into a unified analysis. What then is the nature of the ‘e prefix in (129b)? If it were a realization of agreement on v, then it could not manifest agreement with the possessor of the thematic subject, because neither one would c-command the other at any point in the derivation. The only way that the c-command condition on agreement is satisfied is if ‘e is a spell-out of agreement on T, because T c-commands the possessor when the possessed subject is still in Spec, vP. In other words, the structure of (129b) must be something like (130). TP

(130)

T

vP Agree(1)

Agree(2)

NP

NP

v´ N

Charlie mother (GEN)

v

VP V go.across

The same is true for the null allomorph of agreement found with a nonthird person possessor in (129a). Now assuming that the occurrences of ‘e and Ø as object agreement in (126) are instances of the same morphemes, then these too

234


must be realizations of agreement on T. More specifically, I propose that finite T probes its domain twice in Nez Perce. The first time, it finds the subject and agrees with it; the second time it finds either the direct object or the possessor of the subject. The result is an ordered pair of features on T. T is then spelled out at PF according to a set of disjunctive realization rules something like the following:44 (131)

a. b. c. d. e.

T <3, 3> → peeT → ‘eT → ØT <3 . . .> → hiT <1/2 . . .> → Ø-

What are the implications of this analysis for the agreement parameters? One clear implication is that there is no longer any question of agreement on v having different syntactic behavior than agreement on T, simply because we now hold that there is no agreement on v. The question now is whether the two agreement relationships that T enters into are consistent with respect to the parameters in (1) and (2). The answer seems to be yes. The object/possessor agreement realized as ‘e or Ø may be downward agreement, as shown in (130). That is consistent with the parameter setting that we observed for simple subject agreement on T and for agreement on C. Also the object/possessor agreement realized as ‘e or Ø can be agreement with an NP marked with the objective case ne, as in (126), or it can be agreement with an NP marked with the genitive/ergative case nm, as in (129). Thus, the second T agreement is not case-dependent, consistent with the parameter setting that we saw for simple subject agreement (hi) and for agreement on C. Once we realize that agreement with objects and agreement with possessors is (at least partly) the same phenomenon, it becomes clear that this sort of agreement can be downward and is not case-sensitive. There is thus no inconsistency in the parameter settings in Nez Perce after all. One remaining question is why indefinite objects which remain in VP and do not bear the objective case ne cannot trigger object agreement on the verb. 44 This is not a complete solution to the mysteries of Nez Perce agreement. Unlike (131b-c), the rule in (131a) must be limited to transitive clauses: it spells out Ts where the object as well as the subject is third person, but not Ts in which the possessor as well as the subject is third person. Second, these rules do not account for the hi prefix found with a third person subject and a first or second person object in (127a). Perhaps this needs to be a second hi morpheme, ordered before (131c). The underlying question here is whether it is better to treat hi or Ø as the default morpheme in this system, since neither is used in a natural class of environments.


235

I do not have a definitive answer to this, but a promising suggestion is that one might make use of Chomsky’s (2000, 2001) idea that transitive vPs are phases. As a phase, the vP should be opaque to agreement relationships. In the normal case, when object agreement is agreement on v, the Agree relationship between v and the object is internal to the vP phase, so there is no problem. But when object agreement is on T, as in Nez Perce, it is in a different phase from the base position of the object. The phase condition on agreement then implies that T can agree with the object (downward) if and only if the object raises to the edge of the vP phase, taking it outside the domain of existential closure, as shown in (132). (132)

TP T Agree(1) vP NP

vP

man NP v´ (ERG) elk v ∃ VP Agree(2) OK (OBJ) V NP *Agree(2) by phase condition

shoot

object shift

Therefore object agreement is possible only with strong/definite objects in Nez Perce, not because agreement must be upward (as in Zulu), but rather because the agreeing head is in a different phase from the agreed-with NP. This makes the strong prediction that (barring parameterization in whether vP is a phase or not) all languages in which T agrees with the object must be languages in which only definite objects are agreed with, regardless of other parameter settings. I do not know if this is true, but Inuit and Chukchi are two other languages that seem to support this. They have also been analyzed as having Ts that agree with both the subject and the object (Bok-Bennema 1991 for Inuit, Bobaljik and Branigan 2006 for Chukchi). Moreover the agreed-with object in absolutive case clearly moves higher than propositional operators on the verb in Inuit (Bittner 1994) and Bobajlik and Branigan observe that this seems to be true in Chukchi as well, although the matter has not been studied as carefully. So the overall approach to Nez Perce seems promising, and this language should not cause us to doubt too much the existence of grammatical parameters for agreement.

236


5.11.3 Two minimal comparisons One of the strongest tests that a macroparametric proposal can be put to is the test of closely related languages. Suppose that, through a process of historical change, a language develops a setting for a grammatical parameter that is different from that of the protolanguage, which is passed down to most of the daughter languages. Such changes might not be very frequent, but they must be possible given that even languages we think of as unrelated presumably came from a single language at a more remote time (e.g., the languages of South America). The strong prediction is that the effects of this parametric change should tend to show up in different areas of the innovative language’s syntax. In contrast, a purely microparametric analysis, in which grammatical parameters are avoided in favor of lexical parameters, does not have this expectation. On the microparametric view, the properties of one functional category could very well change without having any effect on the behavior of another functional category. I have not looked systematically for evidence that bears on this question, for both practical reasons and theoretical reasons. Practically speaking, the choice to work with the 100 languages from the WALS core language sample amounts to working with a set of languages that are more distantly related. There are not many minimal pairs of closely related languages in that sample, by virtue of its basic design. From a theoretical perspective, I would expect lexical factors to be very prominent in cases where parametric change has happened, perhaps hiding the grammatical parameter settings to an unusual degree (otherwise parametric change might not be possible at all). Nevertheless, I have stumbled across a few cases that are relevant to this issue, more or less by accident. One is Makhuwa, a Bantu language that seems to differ from its kin in allowing downward agreement, thereby having a different setting of parameter (1). Another is Imbabura (Ecuadorian) Quechua as compared to Huallaga (Peruvian) Quechua, the two languages apparently differing in the setting of parameter (2).45 I bring this study to a close with a short presentation of these two case studies, claiming that they illustrate in microcosm what the 108language survey illustrates on a larger scale – the fact that the two parameters apply to a range of different functional heads. 45 Another case of parametric variation within a single family, not chosen for illustration here, is Nepali and Maithili having a different setting for the Case-Dependency of Agreement parameter from Hindi and other IE languages; see note 4 for an outline of the data relevant to this. There seems to be a fair amount of variation in the Afroasiatic languages: Hausa and Hebrew differ from Arabic in the setting of parameter (2), and Berber differs from Arabic in the setting of (1) if the analysis proposed at the end of section 5.11.2.1 is correct.


237

5.11.3.1 Makhuwa versus Kinande/Chichewa Makhuwa is spoken in Mozambique, on the eastern extreme of the Niger-Congo area; I thank Jenneke van der Wal for calling it to my attention and sharing data with me. Van der Wal (2006) observes that Makhuwa is different from many other Bantu languages in that the verb does agree in noun class with the subject in thetic clauses with verb-subject word order, as shown in (133). (133)

mwanámwáne. a. O-hoó-khwá 1S-perf.disj-die 1a.child ‘There has died a child.’ b. A-náa´ -ttónyá maátsi. 6S-pres.disj-drip 6.water ‘There is water leaking out.’

The finite verb also agrees with the postverbal subject in clauses that have a PP-V-S order, in contrast to Kinande and Chichewa where agreement is with the fronted PP (van der Wal, personal communication): (134)

a-kwaatú. Watarátú a-r´ı 16.roof 2S-be 2-cats ‘On the roof are cats.’

Van der Wal (2006) treats the difference between Makhuwa and other Bantu languages like Sesotho in more or less microparametric terms. She claims that other Bantu languages are like French in having an expletive (il) that has ϕ-features, which the verb must agree with. In contrast, Makhuwa is like Spanish and Italian in not having an expletive with ϕ-features, so the verb is left free to agree with the postverbal subject in sentences like (133). This may indeed be a factor in the overall analysis. But her analysis also assumes that finite T can agree with a subject that is still inside VP in (133) and (134), a kind of downward agreement that we have seen to be banned in other Bantu languages. And if there is a macroparametric difference between Makhuwa and other Bantu languages in this respect, then we might expect to see the effects of this on other aspects of agreement in the language. In this light, it is striking that object agreement in Makhuwa also behaves very differently from object agreement in languages like Kinande and Chichewa (van der Wal, personal communication). Kinande and Chichewa have a complete set of object markers, and these appear if and only if there is no overt object or the object is dislocated to the edge of the clause. In contrast, Makhuwa has lost its object markers for all but class 1 and class 2 objects. Moreover, the object markers for these two classes act as obligatory object agreement: they

238


are required whenever the object of the verb is class 1 or 2, even when that object occurs in immediate postverbal position and has an indefinite reading: (135)

namarokoló. a. Ki-n-´ın-thúmá 1sS-pres-1O-buy 1a.rabbit ‘I am buying a rabbit.’ b. O-hi-n-thel-é nth´ıyáná oowoótha 2sS-neg-1O-marry-opt 1.woman 1.lying ‘Don’t marry a lying woman.’

There is no reason from word order or definiteness to think that the object has moved out of the verb phrase in (135b); nevertheless it triggers class 1 agreement on the transitive verb (technically, on v). This shows that v also agrees downward in Makhuwa. Thus, both subject agreement and object agreement work in a systematically different way in Makhuwa than in many of its Bantu relatives. I take this as support for the macroparametric view that parameter (1) governs the complete set of functional heads in the language. The other functional heads in Makhuwa are less striking, but are consistent with (1) being set as “no.” There is no agreement on C or P, hence no evidence from those domains that agreeing heads must probe upward. The other relevant head is D. Recall that most D-like heads in Kinande and Swahili follow the associated NP and agree with it, but the one or two that precede NP cannot agree with it. Given that Makhuwa allows heads to agree downward, it might allow D-NP order with agreement on the D, as IE languages do. In fact, judging by Woodward 1926, most D-like heads follow the NP in Makhuwa, so are not relevant to the prediction. Woodward does, however, give some examples of demonstrative doubling, where one token of the demonstrative comes before the NP and the other one comes after it, as shown in (136). (136)

a. yo-la m-tu’ o-la (Woodward 1926:285) 1-this 1-man 1-this ‘this very man’ b. I-la inupa i-la y-ulupale. (Woodward 1926:300) 3-this 3.house 3-this 3-large ‘This house is large.’

Observe that the preceding demonstrative agrees with the NP in number and noun class as much as the following demonstrative does. This is possible additional support for the claim that Makhuwa allows downward agreement – although one would like to know more about the structure of nominals in the language to be sure that we know how to interpret the data in (136).


239

Makhuwa does not seem to differ from its Bantu relatives with regard to its setting for parameter (2). In particular, it is like the other Bantu languages in permitting complex tenses in which the auxiliary verb and the main verb both show full person-number-gender agreement with the same argument (Jenneke van der Wal, personal communication): (137)

a. Ki-háana ki-khumáka 1sS-have 1sS-go.out.dur ‘I have to go out.’

The fact that a Bantu language can show a consistent change with respect to the direction of agreement without agreement developing case sensitivity confirms that (1) and (2) are two independent parameters. One cluster of properties changes together when the parameter setting changes, but the other cluster of properties does not. 5.11.3.2 Ecuadorian and Peruvian Quechua My second case study involves a change in the value of the Case Dependency of Agreement Parameter. Ecuadorian Quechua – specifically the Imbabura Quechua described by Cole (1985) and others – seems to differ from other Quechua varieties in having a negative value for this parameter. (I do not know for sure which is the innovative value of the parameter, but tentatively assume that it is the Imbabura value, since it seems to have a narrower distribution.) One sign of the parametric difference comes from auxiliary constructions. While agreement only on the auxiliary is the norm in both types of Quechua, Cole points out that some Imbabura Quechua dialects have agreement on both the main verb and the auxiliary in past conditional tenses: (138)

(Cole 1985:155–6) a. Shamu-y-man ka-rka-ni. come-1sS-cond be-past-1sS ‘I would have come.’ b. Shamu-ngui-man ka-rka-ngui. come-2sS-cond be-past-2sS ‘You would have come.’

Cole himself points out that other Quechuan languages have only default third person agreement on the auxiliary in examples comparable to (138). That is true in Huallaga Quechua, for example, as shown by (139), which is otherwise perfectly cognate to (138b).

240 (139)

Parameters of agreement Sha-mu-nki-man ka-ra-n. (Weber 1989:104) come-afar-2sS-cond be-past-3S ‘You should have come.’

(138) shows that parameter (2) is set “no” in Imbabura Quechua, whereas the systematic absence of such double agreement suggests that it is set “yes” in Peruvian Quechua. Given this difference, the macroparametric view predicts that Imbabura Quechua should be more amenable to having the verb agree with oblique NPs than Peruvian Quechua. This is true. Imbabura has verbs that take objects marked with dative case man. Moreover, the same first singular object agreement appears on the verb regardless of whether the object bears accusative case or dative case, as shown in (140). (140)

(Jake 1985:30) a. Maria-ca (ñuca-ta) maca-wa-rca-mi. Maria-top I-acc hit-1sO-3.past-val ‘Maria hit me.’ b. Maria-ca (ñuca-man) cara-wa-rca-mi. Maria-top I-dat serve-1sO-3.past-val ‘Maria served me.’

Indeed, Jake (1985:30) shows that some Imbabura verbs can even agree with objects in genitive case (benefactees) and instrumental case (comitatives). Thus, agreement on v is clearly not sensitive to the accusative/oblique distinction in Imbabura Quechua, as expected if parameter (2) is set “no”. In contrast, there does seem to be a close relationship between case and object agreement in Huallaga Quechua: an argument triggers object agreement only if it bears accusative case. It is true that the goal argument of a ditransitive verb like ‘give’ can trigger object agreement in Huallaga, as in Imbabura (see (141a)), but that is only because goal arguments bear the same accusative case marking ta as theme arguments in Huallaga (unlike Imbabura, see (141b)). (141)

(Weber 1989:98) a. Pay qam-ta qu-ma-ra-n he you-obj give-1sO-past-3S ‘He gave you to me.’ b. Hwan-ta papa-ta qo-ykusha. (Weber 1989:23) Juan-obj potato-obj give-past.3S ‘He gave the potato to Juan.’

Only NPs marked with ta trigger object agreement in Huallaga Quechua, a sign that its value for parameter (2) is “yes.” The two Quechuas thus differ not only


241

in the pattern of agreement in auxiliary constructions, but also in the ability to agree with oblique NPs, as expected. Finally, we want to check if the same contrast between the varieties is found with respect to oblique subjects. I argue that the answer is yes. On the basis of its “no” setting for parameter (2), one might expect Imbabura Quechua to allow agreement with oblique subjects, whereas Peruvian Quechua would not. In fact, there is an important difference between the languages when it comes to oblique subjects, discussed especially by Hermon (1985, 2001). Imbabura Quechua has experiencer constructions in which the experiencer argument bears overt accusative case and there is no nominative subject (Cole 1985:107–8): (142)

a. Juzi-ta puñu-naya-n. José-acc sleep-desid-3S ‘José wants to sleep; José is sleepy.’ b. Juzi-ta rupa-n. José-acc be.hot-3S ‘José is hot.’

These accusative case experiencers behave like subjects in certain respects, and not like the direct objects that they superficially resemble. For example, the experiencer of a desiderative verb like (142a) can be controlled when the verb itself is an infinitive selected by a suitable main clause verb (Hermon 2001:161): (143)

˜ Nuka-ka [- - puñu-naya-y-ta] kati-ju-rka-ni. I-top sleep-desid-inf-acc continue-prog-past-1sS ‘I continued to desire to sleep.’

Similarly, the experiencer argument of such a verb can be treated as the subject of the clause for the purposes of switch reference marking (Hermon 2001:162): (144)

[-- mika-naya-shpa] aycha-ta randi-rka-ni. eat-desid-ss meat-acc buy-past-1sS ‘When I desired to eat, I bought meat.’

The inflection shpa on the dependent verb is normally used only when the subject of that verb is coreferential with the subject of the matrix verb. Its use in (144) shows that the experiencer argument of a desiderative verb counts as a subject for these purposes, despite being case-marked like an object. So there are accusative-marked subjects in Imbabura Quechua, as well as nominativemarked ones.46 46 Hermon (1985, 2001) shows that there are some differences between the accusative subjects of desiderative verbs like (142a) and those of simple psych verbs like (142b). Descriptively

242


In contrast, there seem not to be accusatively marked subjects in Peruvian Quechua. Constructions that are equivalent to those in (142) seem to be absent in Huallaga Quechua, for example. There is no mention of such a use for accusative case in Weber (1989), and the cognate of the desiderative morpheme in (142a) is not used in the same way in this variety (Weber 1989:170). Hermon (1985, 2001) shows that examples that are superficially similar to (142) are found in Huanca Quechua, a Central Peruvian variety: (145)

a. Yaqa-kta tushu-na-ykaa-ma-n. (Hermon 2001:171) me-acc dance-desid-prog-1sO-3S ‘I desire to dance.’ b. Yaqa-kta alalaa-ma-n. me-acc cold-1sO-3S ‘I am cold.’

But it is striking that accusative case experiencers in Huanca Quechua do not behave syntactically like subjects the way their equivalents in Imbabura Quechua do. (146a) shows that the experiencer of a desiderative verb cannot be controlled in Huanca; (146b) shows that it cannot license the use of same-subject marking (Hermon 2001:171). (146)

a. *Yaqa muna-a-chu [-- mika-na-y-ta]. I want-pres.1-neg eat-desid-inf-acc ‘I want to desire to eat.’ b. *[-- kasara-na-la-l] alli wamla-kta ashi-lla-a. marry-desid-stat-ss good girl-acc look.for-affect-pres.1sS ‘When I desire to marry, I will look for a good girl.’

Indeed, Hermon says that accusative case experiencers have virtually no subject properties in Huanca Quechua, and behave almost identically to the normal objects of transitive verbs.47 Thus, Imbabura Quechua has accusative case subjects and Peruvian Quechua does not. I analyze these differences as being rooted in the setting of parameter (2). Since (2) is set “yes” in Peruvian Quechua, T cannot agree with the experiencer, which is assigned an inherent accusative case. Therefore, the experiencer does speaking, the experiencers of desiderative verbs are fully subject-like in their syntactic behavior, whereas those of simple psych verbs show a more mixed behavior. I offer no analysis of this interesting contrast. 47 The one difference between accusative experiencers and normal direct objects in Huanca Quechua that Hermon (1985, 2001) discovered is that accusative experiencers cannot be extracted from embedded clauses in a manner reminiscent of the ECP of Chomsky (1981). I have nothing to say about this fact.


243

not have the “subject” properties that go along with being in an agreement relationship with T. In contrast, (2) is set “no” in Imbabura Quechua, so T can agree with the accusative-marked experiencer in this variety. Therefore, the experiencer is similar to an ordinary nominative subject in those respects that relate to it having an Agree relationship with T. For example, nonfinite T in Quechua makes it possible for the argument it agrees with to be PRO; that extends to experiencers of desiderative verbs in Imbabura but not in Huanca. Similarly, the dependent T in an adverbial clause is spelled out as a same-subject marker if the NP it agrees with is referentially dependent on the NP that the T in the main clause agrees with. That includes the experiencer of a desiderative verb in Imbabura but not in Peruvian varieties. This is not a full theory of control or switch reference in Quechuan languages, of course, and I cannot provide one here. But on the (reasonable, I think) assumption that the Agree relationship plays a role in the theory of control and switch reference in these languages, then the differences in the syntax of accusative experiencers are reflections of the difference in how parameter (2) is set in two branches of this family. There is a complication to this story: T does not in fact appear to agree with the accusative-case experiencer in Imbabura Quechua any more than it does in Huanca Quechua. When the experiencer is first person in a sentence like (147), the finite verb does not agree with it in person, but rather bears the third person suffix n. (147)

n˜ uka-ta puñu-naya-n. (*puñu-naya-ni). (Cole 1985:111) I-acc sleep-desid-3S sleep-desid-1sS ‘I want to sleep; I am sleepy.’

Cole, Jake, and Hermon all very reasonably interpret this as default agreement, the third person ending being filled in when there is no argument to agree with. But in light of the overarching pattern of facts reviewed in this section, I interpret it as partial agreement, comparable to the agreement that a predicate adjective would have with its subject. The details could go something like this. Since experiencer subjects have a different θ -role than agentive subjects, they are generated in a different structural position – say Spec, ApplP rather than Spec, vP.48 In that position, they receive inherent accusative case from the applicative head. This inherent case freezes them in place, so that (unlike NPs generated in Spec, vP) they are unable to move to Spec, TP. The principles of agreement still allow T to agree with the experiencer in Imbabura, because the 48 This specific proposal is based on the similarities between experiencers and goals, not only crosslinguistically, but also in Imbabura (see especially Jake 1985 for relevant data).

244


universal conditions on Agree are satisfied, agreement is not case-dependent in Imbabura, and all Quechuan languages permit downward argument. (The primary evidence for this is that all Quechuan languages have reasonably free word order, and alternations between (say) SOV and OSV order have no affect on subject or object agreement.) This agreement relation between T and the experiencer contributes to an understanding of the control and switch reference properties reviewed above. But the principles of agreement do not allow T to agree with the experiencer in person in this configuration; rather, agreement can only be in number and gender, by the SCOPA. Now it so happens that gender is never marked in Quechua, and there is no distinction between singular and plural in the third person (Cole 1985:143, Hermon 1985:22). Therefore the partial agreement forced by the SCOPA happens to be indistinguishable on the surface from default agreement, simply because there happens to be only one nonfirst/nonsecond inflectional ending in the language. Nevertheless, the existence of a genuine agreement relationship can be detected indirectly by its role in control and switch reference. Positing it also allows us to identify a unified source for the various observable differences in the relationship between agreement and case that distinguish Imbabura Quechua from its Peruvian cousins – a testimony to the reality of (2) as a unified grammatical parameter.49 5.12

General conclusion

The first four chapters of this book developed a universal theory of agreement that applies equally well to all syntactic categories – to adjectives, nouns, adpositions, determiners, and complementizers as well as to verbs. This quest bore theoretical fruit in two specific ways: it led to the result that agreement can be upward as well as downward, and to the result that person agreement needs a more strictly local syntactic relationship than other kinds of agreement do. In this chapter, we have seen how having a broader theory of agreement can bear fruit along another dimension. Once we have all agreement-bearing categories fully in view, two typological results can also become clear. On the one hand, the agreement-bearing functional heads behave rather differently from one language to another with respect to the precise conditions under which they agree. On the other hand, all of the agreement-bearing heads in a given language tend to work the same way with respect to these conditions. This balance between 49 My analysis crucially predicts that if a dialect of Imbabura Quechua comes to light that distinguishes third plural agreement from third singular agreement, one would find the plural form in a sentence like ‘We are sleepy,’ not the default singular form. But no such dialect has been documented to my knowledge.

General conclusion

245

a limited degree of inter-language diversity and high degree of intra-language consistency is exactly what one expects given Chomsky’s classic Principles and Parameters vision of the nature of the human language faculty. But the pattern can only be seen clearly if one has in view both a broad range of natural languages and a theory that applies to a wide range of syntactic categories. This work, then, has culminated in an experiment in combining a formal-generative depth of analysis, concerned with capturing significant generalizations across different grammatical constructions, with a functionalist-typological methodology of putting claims to the test against a broad sample of languages. I like to think that the experiment has been a successful one, giving a big picture that is more robust and convincing than could be constructed simply by pursuing either style of inquiry by itself. But I admit that you are likely to be a better judge of that than I am.

Appendix: Table of languages and their agreement properties

The following table displays the properties of agreement that are most relevant for assessing how the Direction of Agreement Parameter and the Case Dependence of Agreement Parameter are set, for each of the 108 languages in my sample. The languages are listed in the same order as they are in the description of the World Atlas of Language Structures core sample, starting in Africa and moving northward and eastward in roughly the presumed ways that language-speakers spread across the globe by migration, ending in South America. The first two columns identify the primary word order and case-marking properties of full noun phrases in the language; in almost every instance, these values are taken from WALS maps 81–3 and 98. This information is included as background and plays no direct role in the analysis. The last eight columns record the answers that I found to the following eight questions concerning the syntax of agreement: a. Is subject agreement dependent on the subject being in a particular syntactic position? b. Is subject agreement sensitive to the case of the subject? For example, does the verb agree only with nominative subjects, or can it show the same sort of agreement with ergative or dative subjects as well? c. Is object agreement dependent on the position or definiteness of the object (two kinds of evidence that the object has moved to a higher position)? d. Is object agreement sensitive to the case of the object? For example, does the verb agree only with accusative/absolutive/unmarked objects, or can it show the same sort of agreement with dative or oblique complements? e. Is there agreement on complementizers, and if so what kind? f. Is there agreement on adpositions with an unmoved complement? g. Is there agreement between a determiner/quantifier and the associated NP, and if so does it depend on the word order of the NP with respect to the determiner? h. Is there full person-number-gender agreement with the subject on both the auxiliary verb and the main verb in complex tense constructions, or is there person agreement on one verb only? In general, the answers to (a, c, e–g) are relevant to the setting of Direction of Agreement Parameter, and the answers to (b, d, h) are relevant to the setting of the Case Dependence of Agreement Parameter. Note that if WALS gives no indication that a language has agreement with subjects, objects, or objects of adpositions, and does not indicate that

246

Table of languages and their agreement properties

247

it has grammatical gender (so adjectival agreement is unlikely), then I did not look up the language, and scored all categories as “none.” Such languages are only included to give an overall sense of how common each type of agreement is in the languages of the world. See chapter 5 for extensive further discussion of these properties, how they relate to the proposed parameters, and illustrative examples. (See also below for explanation of some of the less obvious notations used in the table.)

SVO SVO SVO

SVO SOV SVO VSO SVO SVO SVO VSO SVO SOV VSOSVO SVO SOV SVO

Swahili Zulu Kinande

Yoruba Supyire Grebo Krongo Bagirmi Lango Koyra Chiini Berber Hausa Harar Oromo Arabic

Hebrew Basque English

SOV SVO SVO

Khoekhoe Sango Luvale

Language

Word Order

neutral neutral neutral neutral neutral neutral neutral? accusative neutral accusative accusativeneutral accusative ergative neutral

neutral neutral neutral

accusative neutral neutral

Casemarking pattern

No No No

None None None None Can’t tell Can’t tell None No* Can’t tell None No

Yes Yes Yes

No None Yes

AgrS needs order?

Yes No N/A

None None None None N/A N/A None N/A N/A None N/A

N/A N/A N/A

(Yes) None N/A

AgrS depends on case

None None None None (clitic) (clitic) None (clitic) None None No/ clitic (clitic) No None

Yes Yes Yes

(clitic) None Yes

AgrO only if mov’t

N/A No None

None None None None N/A N/A None N/A N/A None N/A

N/A N/A N/A

N/A None N/A

AgrO depends on case

None Up None

None None None None None None None None? None None Down

None None Up

None None None

Agr on C

(clitic) None None

None? None Only if moved (clitic) (clitic) Only if moved None None None None (clitic) (clitic) None None None None Yes/ clitic

Agr on P with object

? Yes (Yes)

None None None None None Yes if N-D None Only if N-D Yes None Yes

Only if N-D Yes, N-D Only if N-D

Yes None Yes+ N-D

Agr on Det

Single Single Single

None None None None Single Single None Double Single None Double

Double Double Double

N/A None Single

Agr in Aux+Verb

accusative accusative split ergative accusative neutral accusative accusative accusative accusative accusative ergative accusative accusative neutral

ergative split ergative ergative accusative neutral accusative accusative neutral

Mixed SVO SOV

SOV SVO SVO SVO SVO SOV SOV Free SOV SOV SOV

SOV SOV

SOV SOV SVO SOV SOV SVO

German Greek Hindi

Persian French Spanish Russian Finnish Khalkha Turkish Chukchi Japanese Korean Abkhaz/ Abaza Lezgian Georgian

Burushaski Kannada Mandarin Burmese Meithei Hmong

No No None None None None

None No

No No No No No None Yes No(2) None None No

No No No

No Yes None None None None

None No

N/A? N/A? Yes Yes Yes None Yes No(2) None None (Yes)

Yes Yes Yes

No None None None None None

None No

(clitic) (clitic) No/Clitic None None None None (none) None None No

None (clitic) None

No None None None None None

None No

None None N/A None None None None (none) None None (Yes)

None None None

No C None None None None None

None None

None None None None None None None None None None None

Down None None

Some None None None None None

None Some?

(clitic) None None None Yes None None None None None Yes

None None None

Yes None None None None None

None Yes

None Yes Yes Yes Yes None None No D? None None None

Yes Yes Yes

(cont.)

Double Single None None None None

None Double

Double Single Single Single Single None Single Single None None No aux

Single Single Single

SVO SVO V-first V-first

VSO VOS VSO V-first VOS SVO SOV V-last SVO SOV free SOV SOV SOV SOV SOV SOV V-last

Rapanui Malagasy Chamorro Tagalog Tukang Besi Indonesian Tauya Imonda Arapesh Alamblak Yimas Asmat Dani Kewa Amele Daga Lavukaleve Gooniyandi

Word Order

Thai Vietnamese Paiwan Fijian

Language

Cont.

neutral accusative neutral neutral ergative neutral neutral active neutral neutral neutral neutral ergative ergative neutral neutral neutral ergative

neutral neutral neutral neutral

Casemarking pattern

None None Yes None Yes None No None Yes No No No No No No No Can’t tell No

None None None Yes

AgrS needs order?

None None N/A None No None No None N/A (Yes) (Yes) (Yes) No No (Yes) Yes N/A No

None None None N/A


None None None None Yes None No None Yes No No No No None No ? No No

None None None Yes


None None None None N/A None N/A None N/A N/A Yes N/A N/A None N/A Yes Yes Yes

None None None N/A


None None (Up) None None None None None Up None No C None None None None Down No C? No C

None None None None

Agr on C

None None None Only if moved None None None None None None Yes None None None None None Some None None Some Yes None


None None None None None None None None Yes if N-D? Yes Some None None None Yes None Yes None

None None None None

Agr on Det

None None Single None Double None Single None Double Single No aux Single Single Single Single Single Single No aux

None None None Double

Agr in Aux+Verb


Free SOV SOV Free

SOV VO

Yaqui Nahuatl

accusative neutral

neutral accusative

Mohawk ChoctawChicasaw Wichita Lakhota Kiowa Acoma

neutral

V-last Free Free SVO SOV SOV

SVOfree Free SOV

ergative ergative accusative neutral ergative neutral

SVO OV SVO

Maung Mangarrayi Martuthunira Ngiyambaa Warlpiri Kayardild Tiwi Greenlandic Slave

Ojibwa

(weak ergative) neutral accusative accusative

Free

Mayali

None No

No No Can’t tell Can’t tell

Yes No

No

(clitic) No None No Yes(2) Yes?

No No None

No

None N/A

N/A (Yes) N/A N/A

N/A Yes

N/A

N/A No None Yes Yes N/A

(Yes) Yes None

No

Yes Can’t tell No No No Can’t tell None No

No

(clitic) No None No (none) Yes

No No None

No

None N/A

N/A N/A N/A N/A

NoLDA N/A Yes

N/A No None N/A (none) N/A

Yes Yes None

N/A

None None

None None None None

None None

None

No C No C None None No C None

None None None

None

None Yes

No P Yes None None

No P Yes

No P None None None None Only if moved No P

None None None

None

None Yes

None Yes Yes None

None None

Yes

No D No D None Yes Yes if N-D Yes if N-D

Yes Yes None

Yes

(cont.)

None No aux

No aux Single No aux Single

No aux Single

Double

None Single None Single Single Double

Single Single None

No aux

VSO Free

VSO

SOV

Free VSO VOS VOS VOS SOV

OV OSV

SOV VSO SOV

varies

Halkomelem

Maricopa

Karok Mixtec Otomi Zoque Tzotzil Ika

Barasano Warao

Sanuma Yagua I. Quechua

Apurinã

Word Order

Makah Nez Perce

Language

Cont.

neutral

ergative neutral accusative

accusative neutral

neutral neutral neutral ergative neutral ergative

accusative

neutral

neutral tripartite

Casemarking pattern

Yes

None No No

No No

No (clitic) No No No No

Can’t tell

No

No No

AgrS needs order?

N/A

None N/A No

Yes Yes

No None N/A No (Yes) No

No

N/A

N/A No


None (clitic) Can’t tell Yes

Can’t tell Can’t tell No None No No No Can’t tell None None

No Yes*


Yes

None N/A No

None None

N/A None No? N/A Yes? No

NoLDA N/A

Yes Yes*


None

None None No C

None None

None None None None None No C

None

None

None Down

Agr on C

None

None Yes or clitic None Yes None

None (clitic) None None Yes None

None

None

No P None


Yes in D-N

None Yes None

Yes? Yes

None None Yes None Yes None

None (yes)

Yes

None None

Agr on Det

None Single Some double Single

Single Single

No aux No aux Double Double Single Double

Double

Single No aux/ double Double

Agr in Aux+Verb


VOS SOV SVO VOfree SOV SOV

accusative neutral

None* Yes

No Can’t tell No Yes

Yes

No

No?

None No

Yes N/A (Yes) Yes

No

N/A

N/A

Yes* None

No No No Yes?

Can’t tell Yes

No

N/A None

NoLDA Yes N/A N/A Yes

N/A

Yes

None None

Up? None None No C

None

None

No C

None None

Yes Yes Some None

(clitic)

Yes or clitic (clitic)

Only if N-D Some if N-D

Yes None None? None

None

Yes

None

Double Double

No aux No aux Single No aux

Double

Double

Single

Clarificatory notes: r “None” means the relevant kind of agreement is not present in the language. r N/A means that the agreement is present, but cannot be evaluated for the parameter for some other reason (e.g. no oblique case marking). r “Can’t tell” means that the agreement is present, but I could not find evidence as to whether it is sensitive to the position of the agreed-with NP (e.g. subject always in Spec, TP; subject or object agreement is null). r (Yes) means there is no overt case, but case sensitivity could insure that T agrees only with thematic subject despite the subject not being in a fixed position. r N/Y(erg) means that there is agreement with a weak/limited/optional ergative as well as with an unmarked NP. r (clitic) means that a bound pronominal form can attach to the relevant head. This was counted as equivalent to “none,” but conceivably could be reanalyzed as being agreement given more data. r (2) indicates a kind of ergative language analyzed in the literature as having two instances of agreement on T and none on v. Alternative analyses of this are conceivable.

Tariana Jarawara

neutral

SOV

CanelaKrahô Wari’ Pirahã Wich´ı Mapudungun

accusative

SVO

Guaran´ı

neutral

OVS

Hixkaryana

References

Abney, Steven. 1987. The English noun phrase in its sentential aspect. MIT, Cambridge, Mass.: Doctoral dissertation. Adesola, Oluseye. 2005. Pronouns and null operators: A-bar dependencies and relations in Yoruba. Rutgers University: Ph.D. dissertation. Adger, David, and Daniel Harbour. 2005. Syntax and syncretisms of the Person Case Constraint. In Minimalist Approaches to Clause Structure (MITWPL #50), ed. Ken Hiraiwa and Joey Sabbagh, 1–36. Cambridge, Mass.: MIT Linguistics Department. Aikhenvald, Alexandra. 2003. A grammar of Tariana. Cambridge: Cambridge University Press. Aissen, Judith. 1987. Tzotzil clause structure. Dordrecht: Reidel. Allen, Barbara, Donald Frantz, Donna Gardiner, and David Perlmutter. 1990. Verb agreement, possessor ascension, and multistratal representation in Southern Tiwa. In Studies in Relational Grammar 3, ed. Paul Postal and Brian Joseph, 321–84. Chicago: University of Chicago Press. Anagnostopoulou, Elena. 2003. The syntax of ditranstives: evidence from clitics. Berlin: Mouton de Gruyter. Anand, Pranav, and Andrew Nevins. 2004. Shifty operators in changing contexts. Paper presented at SALT 14, Northwestern University. Aoki, Haruo. 1973. Nez Perce grammar. Berkeley: University of California Press. Aoun, Joseph, and Lina Choueiri. 2000. Epithets. Natural Language and Linguistic Theory 18:1–39. Ashton, E. O. 1949. Swahili grammar. New York: Longmans. Baker, Mark. 1985. The Mirror Principle and morphosyntactic explanation. Linguistic Inquiry 16:373–415. Baker, Mark. 1988. Incorporation: a theory of grammatical function changing. Chicago: University of Chicago Press. Baker, Mark. 1991. Clitic climbing and the boundedness of head movement. In Clitics in the languages of Europe, ed. Hank van Riemsdijk, 369–73. Berlin: Mouton de Gruyter. Baker, Mark. 1996. The polysynthesis parameter. New York: Oxford University Press. Baker, Mark. 1997. Thematic roles and syntactic structure. In Elements of grammar, ed. Liliane Haegeman, 73–137. Dordrecht: Kluwer. Baker, Mark. 2002. Building and merging, not checking: the nonexistence of (Aux) SVO languages. Linguistic Inquiry 33:321–8.

254

References

255

Baker, Mark. 2003a. Lexical categories: verbs, nouns, and adjectives. Cambridge: Cambridge University Press. Baker, Mark. 2003b. On the loci of agreement: inversion constructions in Mapudungun. In Proceedings of NELS 33, ed. Makoto Kadowaki and Snigeto Kawahare, GLSA, University of Massachusetts-Amherst. Baker, Mark. 2003c. Agreement, dislocation, and partial configurationality. In Formal approaches to function in grammar, ed. Andrew Carnie, Heidi Harley, and Mary Ann Willie, 107–34. Amsterdam: John Benjamins. Baker, Mark. 2005. On verb-initial and verb-final orders in Lokaa. Journal of African Languages and Linguistics 26:125–64. Baker, Mark, and Chris Collins. 2006. Linkers and the internal structure of vP. Natural Language and Linguistic Theory 24:307–54. Baker, Mark, and Natalia Kariaeva. 2006. On the relationship of case valuation and phi-feature agreement. Ms., Rutgers University, New Brunswick, N.J. Barss, Andrew, and Howard Lasnik. 1986. A note on anaphora and double objects. Linguistic Inquiry 17:347–54. Bejar, Susana. 2003. Phi-syntax: a theory of agreement. University of Toronto: Ph.D. dissertation. Belletti, Adriana, and Luigi Rizzi. 1981. The syntax of ne: some theoretical implications. The Linguistic Review 1:117–54. Benmamoun, Elabbas. 2000. The feature structure of functional categories: a comparative study of Arabic dialects. New York: Oxford University Press. Bhatt, Rajesh. 2005. Long distance agreement in Hindi-Urdu. Natural Language and Linguistic Theory 23:757–807. Bickel, Balthasar, and Yogendra Yadava. 2000. A fresh look at grammatical relations in Indo-Aryan. Lingua 110:343–73. Bittner, Maria. 1994. Case, scope, and binding. Dordrecht: Kluwer. Bittner, Maria, and Kenneth Hale. 1996. Ergativity: toward a theory of a heterogeneous class. Linguistic Inquiry 27:531–604. Bobaljik, Jonathan. 1993. Nominally absolutive is not absolutely nominative. In Proceedings of the Eleventh West Coast Conference on Formal Linguistics, ed. Jonathan Mead, 44–60. Stanford: CSLI. Bobaljik, Jonathan. 2002. A-Chains at the PF-interface: copies and “covert” movement. Natural Language and Linguistic Theory 20:197–267. Bobaljik, Jonathan. to appear. Where’s Phi? Agreement as a post-syntactic operation. In Phi theory: Phi features across interfaces and modules, ed. David Adger, Daniel Harbour, and Susanna Béjar. Oxford: Oxford University Press. Bobaljik, Jonathan, and Phil Branigan. 2006. Eccentric agreement and multiple case checking. In Ergativity: Emerging Issues, eds. Alana Johns, Diane Massam and Juvenal Ndayiragije, 47–77. Dordrecht: Springer. Bobaljik, Jonathan, and Susi Wurmbrand. 2005. The domain of agreement. Natural Language and Linguistic Theory 23:809–65. Boeckx, Cedric. 2000. Quirky agreement. Studia Linguistica 54:354–80. Boeckx, Cedric. 2004. Long-distance agreement in Hindi: some theoretical implications. Studia Linguistica 58:23–36.

256

References

Boeckx, Cedric, and Fumikazu Niinuma. 2004. Conditions on agreement in Japanese. Natural Language and Linguistic Theory 22:453–80. Bok-Bennema, Reineke. 1991. Case and agreement in Inuit. Berlin: Foris. Bonet, Eulàlia. 1991. Morphology after syntax: pronominal clitics in Romance languages. MIT: Ph.D. dissertation. Borer, Hagit. 1984. Parametric syntax: case studies in Semitic and Romance languages. Dordrecht: Foris. Borer, Hagit. 1991. The causative-inchoative alternation: a case study in parallel morphology. The Linguistic Review 8:119–58. Borer, Hagit. 1995. The ups and downs of Hebrew verb movement. Natural Language and Linguistic Theory 13:527–606. Branigan, Phil, and Marguerite MacKenzie. 2002. Altruism, A-bar movement, and object agreement in Innu-aimun. Linguistic Inquiry 33:385–408. Bresnan, Joan. 1994. Locative inversion and the architecture of Universal Grammar. Language 70:72–131. Bresnan, Joan, and Joni Kanerva. 1989. Locative inversion in Chichewa: a case study of factorization in grammar. Linguistic Inquiry 20:1–50. Bresnan, Joan, and Sam Mchombo. 1987. Topic, pronoun, and agreement in Chichewa. Language 63:741–82. Broekhuis, Hans. 1992. Chain-government: issues in Dutch syntax. University of Amsterdam: Ph.D. dissertation. Bruening, Benjamin. 2001. Syntax at the edge: cross-clausal phenomena and the syntax of Passamaquoddy. MIT: Ph.D. dissertation. Burzio, Luigi. 1986. Italian syntax: a government-binding approach. Dordrecht: Reidel. Campana, Mark. 1992. A movement theory of ergativity. McGill University: Ph.D. dissertation. Carnie, Andrew, and Phil Cash Cash. 2003. Nez Perce case: a note on Woolford 1997. Ms., University of Arizona. Carstens, Vicki. 2001. Multiple agreement and case deletion: against phi-(in)completeness. Syntax 4:147–63. Carstens, Vicki. 2003. Rethinking complementizer agreement: agree with a case-checked goal. Linguistic Inquiry 34:393–412. Carstens, Vicki. 2005. Agree and EPP in Bantu. Natural Language and Linguistic Theory 23:219–79. Cherchi, Marcello. 1999. Georgian. Munich: LINCOM Europa. Choe, Hyon-Sook. 1986. An SVO analysis of VSO languages and parameterization: a study of Berber. Ms., MIT. Chomsky, Noam. 1973. Conditions on transformations: a festschrift for Morris Halle. New York: Holt, Rinehart and Winston. Chomsky, Noam. 1981. Lectures on government and binding. Dordrecht: Foris. Chomsky, Noam. 1986. Barriers. Cambridge, Mass.: MIT Press. Chomsky, Noam. 1991. Some notes on economy of derivation and representation: principles and parameters in comparative grammar. Cambridge, Mass.: MIT Press.

References

257

Chomsky, Noam. 1993. A minimalist program for linguistic theory. In The view from Building 20, ed. Kenneth Hale and Samuel Jay Keyser, 1–52. Cambridge, Mass.: MIT Press. Chomsky, Noam. 1995. The minimalist program. Cambridge, Mass.: MIT Press. Chomsky, Noam. 2000. Minimalist inquiries: the framework. In Step by Step, ed. Roger Martin, David Michaels, and Juan Uriagereka, 89–155. Cambridge, Mass.: MIT Press. Chomsky, Noam. 2001. Derivation by phase. In Ken Hale: a life in language, ed. Michael Kenstowicz, 1–52. Cambridge, Mass.: MIT Press. Cinque, Guglielmo. 1990. Types of A-bar dependencies. Cambridge, Mass.: MIT Press. Cole, Peter. 1985. Imbabura Quechua. London: Croom Helm. Collins, Chris. 2004. The agreement parameter. In Triggers, ed. Anne Breitbarth and Hank van Riemsdijk, 115–36. Berlin: Mouton de Gruyter. Cowell, Mark. 1964. A reference grammar of Syrian Arabic. Washington, D.C.: Georgetown University Press. Craig, Colette Grinevald. 1977. The structure of Jacaltec. Austin: University of Texas Press. Croft, William. 1991. Syntactic categories and grammatical relations. Chicago: University of Chicago Press. Davies, William. 1986. Choctaw verb agreement and universal grammar. Dordrecht: Reidel. den Dikken, Marcel, and Pornsiri Singhapreecha. 2004. Complex noun phrases and linkers. Syntax 7:1–54. Diercks, Michael. 2006. Object agreement in Swahili and the interpretation of objects. Ms. Georgetown University, Washington, D.C. Diesing, Molly. 1990. Verb movement and the subject position in Yiddish. Natural Language and Linguistic Theory 8:41–80. Diesing, Molly. 1992. Indefinites. Cambridge, Mass.: MIT Press. Dixon, R. M. W. 1982. Where have all the adjectives gone? Berlin: de Gruyter. Dixon, R. M. W. 1988. A grammar of Boumaa Fijian. Chicago: University of Chicago Press. Dixon, R. M. W. 2004. The Jarawara language of Southern Amazonia. Oxford: Oxford University Press. Doke, Clement. 1955. Zulu syntax and idiom. Cape Town: Longmans Southern Africa. Doke, Clement. 1963. Textbook of Zulu grammar. London: Longmans. Dryer, Matthew. 1992. The Greenbergian word order correlations. Language 68:81– 138. Duranti, Alessandro. 1979. Object clitic pronouns in Bantu and the topicality hierarchy. Studies in African Linguistics 10:31–45. Etxepare, Ricardo. 2006. Number long distance agreement in (substandard) Basque. Ms., CNRS-IKER. Evans, Nicholas. 2003. Bininj Gun-wok: a pan-dialectal grammar of Mayali, Kunwinjku and Kune. Canberra: Pacific Linguistics. Facundes, Sidney da Silva. 2000. The language of the Apurinã people of Brazil (Maipure/Arawak). State University of New York: Ph.D. dissertation.

258

References

Foley, William. 1991. The Yimas language of New Guinea. Stanford, CA: Stanford University Press. Fortescue, Michael. 1984. West Greenlandic. London: Croom Helm. Fortune, Reo. 1977. Arapesh. New York: AMS Press. Franks, Steven. 1995. Parameters of Slavic morphosyntax. New York: Oxford University Press. Frantz, Donald. 1993. Empty arguments in Southern Tiwa. Ms., University of Lethbridge. Fukui, Naoki, and Margaret Speas. 1986. Specifiers and projections. MIT Working Papers in Linguistics 8:128–72. Geach, Peter. 1962. Reference and generality. Ithaca, N. Y.: Cornell University Press. Giorgi, Alessandra, and Giuseppe Longobardi. 1991. The syntax of noun phrases: configuration, parameters and empty categories. Cambridge: Cambridge University Press. Givón, Talmy. 1976. Topic, pronoun, and grammatical agreement. In Subject and topic, ed. Charles Li, 149–89. New York: Academic Press. Greenberg, Joseph. 1963. Universals of language. Cambridge, Mass.: MIT Press. Grimshaw, Jane. 1990. Argument structure. Cambridge, Mass.: MIT Press. Grimshaw, Jane. 1991. Extended projection. Ms., Brandeis University. Grune, Dick. 1998. Burushaski: an extraordinary language in the Karakoram mountains. Ms., Vrije Universiteit. Guerssel, Mohamed. 1983. Some notes on the structure of Berber. Ms., MIT. Gupta, Anil. 1980. The logic of common nouns. New Haven, Conn.: Yale University Press. Halle, Morris, and Alec Marantz. 1993. Distributed morphology and the pieces of inflection. In The view from building 20, ed. Kenneth Hale and S. J. Keyser, 111–76. Cambridge, Mass.: MIT Press. Harley, Heidi, and Elizabeth Ritter. 2002. A feature-geometric analysis of person and number. Language 78:482–526. Harlow, Stephen. 1981. Government and relativization in Celtic. In Binding and filtering, ed. Frank Heny, 213–54. Cambridge, Mass.: MIT Press. Haspelmath, Martin. 2004. Explaining the ditransitive person-role constraint: a usagebased approach. Ms., Max-Planck-Institut für evolutionäre Anthropologie. Haspelmath, Martin, Matthew Dryer, David Gil, and Bernard Comrie, eds. 2005. The world atlas of language structures. Oxford: Oxford University Press. Heath, Jeffrey. 1984. Functional grammar of Nunggubuyu. Canberra: Australian Institute of Aboriginal Studies. Hermon, Gabriella. 1985. Syntactic modularity. Dordrecht: Foris. Hermon, Gabriella. 2001. Non-canonically marked A/S in Imbabura Quechua. In Noncanonical marking of subjects and objects, ed. Alexandra Aikhenvald, R. M. W. Dixon, and Masayuki Onishi, 149–76. Amsterdam: John Benjamins. Hewitt, B. G. 1995. Georgian: a structural reference grammar. Amsterdam: John Benjamins. Horton, A. E. 1949. A grammar of Luvale. Johannesburg: Witwatersrand University Press.

References

259

Hyman, Larry, and Alessandro Duranti. 1982. On the object relation in Bantu. In Syntax and Semantics 15, 217–39. New York: Academic Press. Iwara, Alexander. 1982. Phonology and Grammar of Lokaa, University of London: Masters thesis. Jackendoff, Ray. 1977. X-bar syntax. Cambridge, Mass.: MIT Press. Jake, Janice. 1985. Grammatical relations in Imbabura Quechua. New York: Garland. Julien, Marit. 2002. Syntactic heads and word formation. New York: Oxford University Press. Kaplan, David. 1989. Demonstratives. In Themes from Kaplan, ed. J. Almog, H. Wettstein, and J. Perry, 563–81. New York: Oxford University Press. Kayne, Richard. 1984. Connectedness and binary branching. Dordrecht: Foris. Kayne, Richard. 1989. Facets of Romance past participle agreement. In Dialect variation and the theory of grammar, ed. Paula Benincà, 85–105. Dordrecht: Foris. Kayne, Richard. 2000. Parameters and universals. New York: Oxford University Press. Kayne, Richard. 2005. Some notes on comparative syntax, with special reference to English and French. In The Oxford handbook of comparative syntax, ed. Guglielmo Cinque and Richard Kayne, 3–69. New York: Oxford University Press. Kimball, J., and Judith Aissen. 1971. I think, you think, he think. Linguistic Inquiry 2:242–6. Kimenyi, Alexandre. 1980. A relational grammar of Kinyarwanda. Berkeley, Calif.: University of California Press. Kinyalolo, Kasangati. 1991. Syntactic dependencies and the SPEC-head agreement hypothesis in KiLega. UCLA: Ph.D. dissertation. Koopman, Hilda. 1984. The syntax of verbs. Dordrecht: Foris. Koopman, Hilda. 2006. Agreement configurations: in defense of “Spec head”. In Agreement systems, ed. Cedric Boeckx, 159–200. Amsterdam: John Benjamins. Koopman, Hilda, and Dominique Sportiche. 1989. Pronouns, logical variables, and logophoricity in Abe. Linguistic Inquiry 20:555–89. Koopman, Hilda, and Dominique Sportiche. 1991. The position of subjects. Lingua 85:211–58. Kornfilt, Jaklin. 1997. Turkish. New York: Routledge. Kratzer, Angelika. 1998. More structural analogies between pronouns and tenses. In Proceedings of SALT VIII, ed. Devon Strolovitch and Aaron Lawson, 92–110. Ithaca, N.Y.: Cornell Linguistics Club. Larson, Richard. 1988. On the double object construction. Linguistic Inquiry 19:335–92. Launey, Michel. 1981. Introduction a` la langue et a` la littérature aztèques. vol. 1. Paris: L’Harmattan. Levin, Juliette, and Diane Massam. 1984. Surface ergativity: case/theta relations reexamined. In Proceedings of the Fifteenth Annual Meeting of the North-East Linguistics Society, 286–301. Amherst, Mass.: GLSA. Li, Yafei. 1990. Xy-binding and verb incorporation. Linguistic Inquiry 21:399–426. Lorimer, D. L. R. 1935. The Burushaski language 1: introduction and grammar. vol. 1. Cambridge, Mass.: Harvard University Press. Machobane, Malillo. 1993. The Sesotho locative constructions. Ms., National University of Lesotho.

260

References

Magner, Thomas. 1991. Introduction to the Croatian and Serbian language. University Park, Penn.: Pennsylvania State University Press. Marantz, Alec. 1991. Case and licensing. Paper presented at The 8th Eastern States Conference on Linguistics, University of Maryland, Baltimore. Marantz, Alec. 1993. Implications of asymmetries in double object constructions. In Aspects of Bantu Grammar 1, ed. Sam Mchombo, 113–50. Stanford, Calif.: CSLI. Marantz, Alec. 2000. Reconstructing the lexical domain with a single generative engine. Ms., MIT. Markman, Vita. 2005. The syntax of case and agreement: its relationship to morphology and argument structure. Rutgers University: Ph.D. dissertation. Massam, Diane. 2001. Pseudo noun incorporation in Niuean. Natural Language and Linguistic Theory 19:153–97. McCloskey, James. 1996. On the scope of verb movement in Irish. Natural Language and Linguistic Theory 14:47–104. Mchombo, Sam. 2004. The syntax of Chichewa. Cambridge: Cambridge University Press. Mohanan, Tara. 1995. Woodhood and lexicality: noun incorporation in Hindi. Natural Language and Linguistic Theory 13:75–134. Müller, Gereon, and Wolfgang Sternefeld. 1993. Improper movement and unambiguous binding. Linguistic Inquiry 24:461–507. Munro, Pamela, and Lynn Gordon. 1982. Syntactic relations in Western Muskogean. Language 58:81–115. Murane, Elizabeth. 1974. Daga grammar: from morpheme to discourse. Norman, Okla.: Summer Institute of Linguistics. Murasugi, K. 1992. NP-movement and the ergative parameter. MIT: Ph.D. dissertation. Ndayiragije, Juvénal. 1999. Checking economy. Linguistic Inquiry 30:399–444. Nichols, Johanna. 1986. Head-marking and dependent-marking grammar. Language 62:56–119. Noonan, Michael. 1992. A grammar of Lango. Berlin: Mouton de Gruyter. O’Herin, Brian. 2002. Case and agreement in Abaza. Dallas, Tex.: Summer Institute of Linguistics. Ormazabal, Javier, and Juan Romero. 2006. Object clitics and agreement. Ms., University of the Basque Country and University of Alcalá. Ormazabal, Javier, and Juan Romero. 2007. The object agreement constraint. Natural Language and Linguistic Theory 25:315–47. Ortiz de Urbina, Jon. 1989. Parameters in the grammar of Basque: a GB approach to Basque syntax. Dordrecht: Foris. Ouali, Hamid, and Acrisio Pires. to appear. Complex tenses, agreement, and whextraction. In Proceedings of the Berkeley Linguistics Society 31. Berkeley, Calif.: Berkeley Linguistics Society. Ouhalla, Jamal. 2005a. Clitic placement, grammaticalization, and reanalysis in Berber. In The Oxford handbook of comparative syntax, ed. Guglielmo Cinque and Richard Kayne, 607–38. New York: Oxford University Press. Ouhalla, Jamal. 2005b. Agreement features, agreement and antiagreement. Natural Language and Linguistic Theory 23:655–86.

References

261

Pesetsky, David, and Esther Torrego. 2001. T-to-C movement: causes and consequences. In Ken Hale: a life in language, ed. Michael Kenstowicz, 355–426. Cambridge, Mass.: MIT Press. Pica, Pierre. 1991. On the interaction between antecedent-government and binding: the case of long-distance reflexivization. In Long-distance anaphora, ed. Jan Koster and Eric Reuland, 119–35. Cambridge: Cambridge University Press. Polinsky, Maria, and Eric Potsdam. 2001. Long distance agreement and topic in Tsez. Natural Language and Linguistic Theory 19:583–646. Pollard, Carl, and Ivan Sag. 1994. Head-driven phrase structure grammar. Chicago: University of Chicago Press. Popjes, Jack, and Jo Popjes. 1986. Canela-Kraho. In Handbook of Amazonian languages, ed. Desmond Derbyshire and Geoffery Pullum, 128–99. Berlin: Mouton de Gruyter. Reinhart, Tanya. 1983. Anaphora and semantic interpretation. Chicago: University of Chicago Press. Reinhart, Tanya, and Eric Reuland. 1991. Anaphors and logophors: an argument structure perspective. In Long-distance anaphora, ed. Jan Koster and Eric Reuland. Cambridge: Cambridge University Press. Rezac, Milan. 2003. The fine structure of cyclic Agree. Syntax 6:156–82. Rice, Keren. 1989. A grammar of Slave. Berlin: Mouton de Gruyter. Richards, Norvin. 1998. The principle of minimal compliance. Linguistic Inquiry 29:599–629. Riemsdijk, Hank van. 1978. A case study in syntactic markedness. Dordrecht: Foris. Ritter, Elizabeth. 1991. Two functional categories in noun phrases: evidence from Modern Hebrew. In Syntax and Semantics 26, ed. Susan Rothstein, 37–62. San Diego, Calif.: Academic Press. Rizzi, Luigi. 1990. Relativized minimality. Cambridge, Mass.: MIT Press. Roberts, John. 1987. Amele. London: Croom Helm. Robins, R. H. 1989. A short history of linguistics. London: Longman. Rögnvaldsson, E., and H. Thráinsson. 1990. On Icelandic word order once more. In Modern Icelandic syntax, ed. Joan Maling and Annie Zaenen, 3–40. San Diego: Academic Press. Rood, David. 1976. Wichita grammar. New York: Garland. Rood, David, and Allan R. Taylor. 1996. Sketch of Lakhota, a Siouan language. Handbook of North American languages 17 (Languages), 440–82. Rude, Noel. 1982. Promotion and topicality of Nez Perce objects. In Proceedings of the Eighth Annual Meeting of the Berkeley Linguistics Society, ed. Monica Macaulay, 463–83. Berkeley: University of California. Rude, Noel. 1986a. Topicality, transitivity, and the direct object in Nez Perce. International Journal of American Linguistics 52:124–53. Rude, Noel. 1986b. Discourse-pragmatic context for genitive promotion in Nez Perce. Studies in Language 10:109–36. Rude, Noel. 1988. Ergative, passive, and antipassive in Nez Perce. In Passive and Voice, ed. M. Shibatani, 547–60. Amsterdam: John Benjamins. Sadiqi, Fatima. 1997. Grammaire du berbère. Paris: Editions L’Harmattan. Safir, Ken. 2004. The syntax of anaphora. New York: Oxford University Press.

262

References

Safir, Ken. 2005. Person, context, and perspective. Rivista di Linguistica 16:107–53. Salaburu Etxeberria, Pello. 1981. Ikaslearen Esku-gramatika (Basque grammar). Bilbao: Mensajero. Saltarelli, Mario. 1988. Basque. London: Croom Helm. Schlenker, Philippe. 2003. A plea for monsters. Linguistics and Philosophy 26:29– 120. Schlenker, Philippe. 2005. Person and binding. Rivista di Linguistica 16:155–218. Schütze, Carson. 1997. Infl in child and adult language: agreement, case and licensing. MIT: Ph.D. dissertation. Shlonsky, Ur. 1997. Clause structure and word order in Hebrew and Arabic. New York: Oxford University Press. Sigurð sson, Halldór. 2000. The locus of case and agreement. Working Papers in Scandinavian Syntax 67:103–51. Sigurð sson, Halldór. 2002. To be an oblique subject: Russian vs. Icelandic. Natural Language and Linguistic Theory 20:691–724. Sigurð sson, Halldór. 2005. The syntax of person, tense, and speech features. Rivista di Linguistica 16:219–51. Simpson, Jane. 1991. Warlpiri morpho-syntax: a lexicalist approach. Dordrecht: Kluwer. Smeets, Ineke. 1989. A Mapuche grammar. University of Leiden: Ph.D. dissertation. Sridhar, S. N. 1979. Dative subjects and the notion of subject. Lingua 49:99–125. Sridhar, S. N. 1990. Kannada. London: Routledge. Stassen, Leon. 1997. Intransitive predication. Oxford: Oxford University Press. Stechow, Armin von. 2003. Feature deletion under binding. In Proceedings of the North East Linguistic Society 33, ed. Makoto Kadowaki and Shigeto Kawahara, 379–403. Amherst, Mass.: GSLA. Stowell, Timothy. 1981. Origins of phrase structure. MIT: Ph.D. dissertation. Taraldsen, Knut. 1995. On agreement and nominative objects in Icelandic. In Studies in comparative Germanic syntax, ed. Hubert Haider, S. Olsen, and Sten Vikner, 307–27. Dordrecht: Kluwer. Thráinsson, Höskuldur. 1979. On complementation in Icelandic. New York: Garland. Travis, Lisa. 1984. Parameters and effects of word order variation. MIT: Ph.D. dissertation. Ura, Hiroyuki. 1996. Multiple feature-checking: a theory of grammatical function splitting. MIT: Ph.D. dissertation. Van Geenhoven, Veerle. 1998. Semantic incorporation and indefinite descriptions: semantic and syntactic aspects of noun incorporation in West Greenlandic. Stanford, Calif.: CSLI Publications. Vitale, Anthony. 1981. Swahili syntax. Dordrecht: Foris. Wal, Jenneke van der. 2006. How object-like is the postverbal subject? Paper presented at The Bantu-Romance Connection, Leeds. Watanabe, Akira. 1996. Case absorption and WH-agreement. Dordrecht: Kluwer. Webelhuth, Gert. 1992. Principles and parameters of syntactic saturation: Oxford studies in comparative syntax. New York: Oxford University Press. Weber, David. 1989. A grammar of Huallaga (Huanuco) Quechua. Berkeley: University of California Press.

References

263

Wechsler, Stephen, and Larisa Zlatić. 2003. The many faces of agreement. Stanford, Calif.: CLSI. Williams, Edwin. 1989. The anaphoric nature of θ -roles. Linguistic Inquiry 20:425–56. Willson, Stephen. 1996. Verb agreement and case marking in Burushaski. Work Papers of the Summer Institute of Linguistics North Dakota 40:1–71. Woodward, Herbert. 1926. An outline of Makua grammar. Bantu Studies 2:269–325. Woolford, Ellen. 1997. Four-way case systems: ergative, nominative, objective, and accusative. Natural Language and Linguistic Theory 15:181–227. Woolford, Ellen. 1999. More on the anaphor agreement effect. Linguistic Inquiry 30:257–88. Wurmbrand, Susanne. 2003. Infinitives: restructuring and clause structure. Berlin: Mouton de Gruyter. Zaenen, Annie, Joan Maling, and Höskuldur Thráinsson. 1985. Case and grammatical functions: the Icelandic passive. Natural Language and Linguistic Theory 3:441– 83.

Index

Abaza-Abkhaz agreement on P in 113, 194 agreement parameters in 166, 201 ergative agreement in 81–2, 83, 203, 224 active agreement systems 54 activity blocking agreement with oblique subjects 89 as condition on agreement 40, 43–4, 71, 76, 79, 80–2 subsumed by Case Dependency of Agreement Parameter 155, 173–4 Adesola, Oluseye 125, 129, 135–7 Adger, David 102 adjectives agreeing but not in person 2, 14–26, 27, 38–9, 85–6, 109, 144, 147–8 attributive versus predicative 17, 60–4, 218–19 compared to determiners 184–6 compared to verbs 6, 65–7, 107–8 definition of 28, 33–4 ergative and raising type 67–74, 89, 145, 171 functional heads associated with 35–6, 38 not case assigners 174 partial agreement with anaphors 150–1 person agreement on 56–8, 86 and principles of agreement 44–5, 48, 56, 171–5, 177 structure of, as predicates 36–8 transformed into verb 25, 26, 27–34, 36 adjunction and agreement 147–8, 185–6, 193 and c-command 51, 144, 218 as intermediate landing site 193 adpositions, see prepositions, prepositional phrases

264

agreement active patterns in 54 with anaphors 150–1, 152 in case 175–8 crosslinguistic variation in 7, 153–5 default 72, 159 in definiteness 17 on different lexical categories 1–3, 13–27, 48–56 ergative-absolutive patterns in 79, 80–1, 82, 83, 101 independent of movement 160 languages without 39, 153–4 long distance 103–7, 151, 214 morphological aspects of 7–9, 154–5 multiple, with single NP 223 not used to define lexical categories 5 partial 104, 109–10, 243–4 as property of functional heads 34–9 between operators and variables 121–4 semantic aspects of 10–11, 108 theoretical principles of 6, 40, 45, 46–7, 48, 65–6, 149–50 Aikhenvald, Alexandra 16, 23, 83, 84 Aissen, Judith 75–7, 82, 86 Allen, Barbara 96 Amele 165–6, 213 Amharic 125, 129, 134, 138 Anagnostopoulou, Elena 88–9, 94, 95, 151 Anand, Pranav 125, 129, 133–4 anaphora locality conditions on 139, 148 only nouns involved in 31, 32 special agreement with anaphors 150–2 as test for subjecthood 91 Aoun, Joseph 126 applicatives 43, 94, 102, 216–17, 243 Apurinã 167–8, 190, 202–3, 205–6, 213

Index Arabic agreement in definiteness in 17 gender agreement in 8 noun versus adjective agreement in 15, 18–19, 50 parameter settings in 165, 195, 212, 214, 236 verb versus adjective agreement in 22–3 Arapesh 179, 182, 190, 212 Ashton, E.O. 38 attributive modifiers, see modifiers auxiliary verbs and multiple subject agreement 174, 208–14, 225, 239–40 and person agreement 57–8 and position of the subject 228 role in structure 45, 77–8 Baker, Mark on copula types 62 on head movement 140, 199 Mirror principle 60 on Mohawk 54 particular analyses of agreement 154, 172, 206–7 on prepositions 112, 113 principles of category theory 20, 26, 27–34, 38, 45, 172–3 on trace licensing 69 Basque auxiliaries in 77–8 ergative agreement in 77–80, 85 long distance agreement in 104–5 parameter settings in 166, 183, 201, 205, 213 Bejar, Susana 77, 79, 100 Belletti, Adriana 67 Benmamoun, Elabbas 59 Berber 212, 225, 236 Bhatt, Rajesh 40–1, 46, 91, 92 Bickel, Balthasar 91, 156, 170 Bittner, Maria 81, 101, 235 Bobaljik, Jonathan on agreement at PF 9, 41, 87, 155, 205 on agreement domains 40, 42, 107, 160 on copy deletion 227–8 on ergativity 80, 83, 101, 169, 235 Boeckx, Cedric 40, 41, 42, 88–9, 94 Bok-Bennama, Reineke 83, 101, 235 Bonet, Eulalia 94, 151

265

Borer, Hagit 35, 156, 227 Borer-Chomsky Conjecture 156, 219–24 Branigan, Phil 83, 101, 107, 235 Bresnan, Joan on locative inversion 158, 159, 160 on object markers 124, 196 Broekhuis, Hans 42, 76 Bruening, Benjamin 77, 87, 105–6, 107, 151 Burushaski agreement only downward in 215–19 agreement on postpositions in 195 object agreement in 201, 204 parameter settings in 166, 169–70, 212–13 Burzio, Luigi 67, 69 Campana, Mark 81 Canela-Krahô 166–7, 194, 202, 203, 212, 214 Carnie, Andrew 232 Carstens, Vicki 160, 172, 174, 181, 208 case agreement in, see concord, in case assignment of 162, 174, 180 compared to adpositions 204, 206 determining position of moved NP 227–8 ergative, effect on agreement 81, 169–70 marking definite objects 231–2 and operator–variable agreement 122 quirky, on subjects 87–93 relationship to agreement 7, 40, 92, 155, 173–8 shared between noun and determiner 191 types of 205 uniqueness of assignment 162–3, 209 see also Case Dependency of Agreement Parameter, Case theory Case Dependency of Agreement Parameter applied to auxiliary constructions 208–10, 211, 212–14 applied to complementizer 180, 182, 183–4 applied to determiner 187–8, 191 applied to linker 207 applied to preposition 191, 193 applied to tense 159–60, 161–2, 163, 169–71, 230 applied to v 201, 203–6, 215, 232 consistency of 156, 222–3, 224, 225, 234, 239–44 distribution of settings of 221–2, 236 evidence relevant to 246–7 statement of 155, 173–8

266

Index

Case theory 72–3, 79, 176 Cash Cash, Phil 232 causatives, agreement in 96 c-command and agreement in person 51–2 as condition on agreement 7, 40–1, 45–6, 63–4, 148, 233 as condition on movement 53 definition of 40, 51, 79, 115, 144, 147, 218 in definition of intervention 47, 128, 144 downward, for agreement 67–70, 89–90, 200, 215, 224 not required for operator–variable agreement 122 symmetrical 187 upward, for agreement 42, 45–7, 74–85, 155, 224 see also Direction of Agreement Parameter Chicasaw 92–3, 109 Chichewa agreement with anaphors in 150–1 gender in 33 no agreement on Ps in 191 object agreement in 42, 99, 196–7 subject agreement in 158–9 Chinese 39, 100 Choctaw 94 Choe, Hyon-Sook 227 Chomsky, Noam bare phrase structure 52, 79, 114 on c-command 40, 46–7, 51 on ergativity 80 on parameters 156 on phases 43, 192, 235 theory of Agree 6, 12, 40, 43, 44, 45, 52, 65–6, 75, 160, 172, 174–5 theory of unvalued features 4 Choueiri, Lina 126 Chukchi 83, 140–1, 201, 205, 213, 235 Cinque, Guglielmo 67–9, 72–3 classifiers, as gender agreement 16, 19–20, 23–4 clauses as domain for phrasal movement 140–1 structure of 54 see also complementizer phrases clitics anaphoric 139 distinguished from agreement 98, 194, 196–7, 200 functional heads as 189

Cole, Peter 239 Collins, Christopher 164, 172, 206–7 complementizer phrases (CPs) agreement into 43, 44, 71, 77, 103–7 hosting person operators 125–6, 129–31, 138, 146 complementizers agreement on 23, 54, 117–20, 145–7, 181, 184, 209 distinguished from verbs 179 licensing subject traces 182 complements agreement with 52, 113–14, 187–8, 191–2 not allowed with modifier 62 relationship to specifier 47 concord in case 174–5, 211 reduced to agreement 7, 65–6, 108, 114, 173, 177 control of logophoric operators 146 of person operators 126, 129–32, 133–4 copula agreement on 72, 89 needed with nonverbal predicates 27, 28–9, 37 types of 62 criterion of identity 31–3 Croft, William 25, 26, 27 cyclicity 47 Daga 183 Davies, William 93, 94 definiteness and agreement 169 agreement in 17 of object, as sign of movement 199–200, 201, 232, 234–5 degree, agreement on 17, 117 deletion of higher copies 81, 227–9 den Dikken, Marcel 20 derivation, morphological 25, 26, 27–34, 36 determiners agreeing with complement 38, 114–16, 184–91, 217, 226–7, 238 agreeing with possessor 48, 60, 115–16, 151, 173 as locus of phi-features 63–4 Diercks, Michael 200 Diesing, Molly 160, 199

Index Direction of Agreement Parameter applied to complementizer 180–1, 183–4 applied to determiner 186–9, 190 applied to linker 207 applied to preposition 191, 192, 195 applied to tense 159–60, 161, 165–9, 210, 230 applied to v 196, 198–9, 201–3, 232 consistency of 198, 200, 224, 225–7, 234, 238 distribution of settings of 221–2, 236 evidence relevant to 246–7 statement of 155, 173, 214–19 Distributed Morphology 9, 17, 38, 154 ditransitive constructions, see double object construction Dixon, R. M. W. 26, 167 Doke, Clement 199 domination 40, 51 double object constructions agreement in 42–3, 94–8, 100, 151, 200, 204–6, 240 contrasted with applicatives 216–17 Duranti, Alessandro 97, 98 Dutch 42, 64, 199 Edo 135–6, 137–8 English agreement in expletive constructions in 43, 69–70 agreement on demonstratives in 187 agreement on predicate nominals in 10–11, 14 agreement on verbs in 75–7, 86–7, 109, 159, 162, 163, 208–14 properties of lexical categories in 27–8 semantic agreement in 10 EPP features not on adjectival heads 53, 74 related to agreement, in Bantu 172 on Tense, causing movement 52, 55, 100, 160, 166 two on same head 100, 101 on v 95, 203 ergativity in agreement patterns 79, 80–1, 82, 83, 101, 203, 224 effects of ergative case marking 169–70, 203 Etxepare, Ricardo 104–5

267

expletive constructions agreement in 43, 69, 89, 91, 159, 163–4, 172, 177 phi-features of expletive 237 extended projection 34, 50, 52, 53, 74, 86 features different systems of 17, 108 expressed multiple times 99–100 morphological effects on realization of 8–9 role in defining local domains 42, 141–3 stipulated on individual lexical items 4, 13, 44, 50, 174 see also gender, number, person, phi-features Fijian 195, 202, 203, 212 Foley, William 14 Franks, Steven 90 Frantz, Donald 96 French 139, 196 Fukui, Naoki 156 functional categories associated with nouns and adjectives 38, 74 as bearers of agreement 4, 20–1, 34–9, 53–4, 55 bearing person agreement 57–60, 112–21, 143–5 and case assignment 174 having consistent properties in a language 220, 222–4, 236 iteration of 49–50, 53–5, 91 made into variables by agreement 123–4 Person Licensing Condition applied to 138–9, 143–5 probehood of 100, 102, 153–4 Gardiner, Donna 96 Geach, Peter 31, 32 gender and agreement on verbs 8–9 different systems of 17 expressed on some nouns only 39 inherent only on nouns 33 long distance agreement in 103–4 mismatches of 66 and semantic agreement 10 Georgian 170, 177–8, 195, 201, 204, 212 German 64, 90–1, 199 Germanic languages, complementizer agreement in 120, 146, 179–80, 209

268

Index

Givón, Talmy 124 goals experiencer subjects as 243 as NP or PP 205 object agreement with 94–9, 205–6, 216–17, 240 position of 95 subject agreement with 101 see also double object constructions Gordon, Lynn 92–3 Greek 90–1, 98, 187 Greenlandic, see Inuit Grimshaw, Jane 50, 74 Gujarati 91–2 Gupta 31, 32 Hale, Kenneth 81 Halle, Morris 9, 17 Harbour, Daniel 102 Harley, Heidi 17, 33 Haspelmath, Martin 98 Hausa 236 Haya 98 Hebrew 35–6, 213, 227, 236 Heim, Irene 123, 125 Hermon, Gabriella 241–2 Hindi incorporated objects in 169 long distance agreement in 40–1, 46 oblique subjects in 159, 169–70 Icelandic agreement on adjectives in 173–4, 176–7 agreement on participles in 174, 211 anaphora in 139 copular clauses in 89 defective intervention in 41–2, 48 object fronting in 160 quirky case subjects in 87–90, 109, 145, 159, 176–7 raising adjectives in 70, 71–2 imperatives 127 impoverishment, rules of 9 incorporation and agreement 168 locality of 140–1 as test for functional categories 35–6 into verbs only 29

ineffability 109 infinitives, agreement on 46, 209, 211, 232 Innu-Aimun 107 intervention apparent violations of 81, 115 definition of 47, 144 and operator–variable agreement 122 restricting agreement 40, 41–2, 43, 47–8, 59, 71–2, 76, 161, 218 source of adjective versus noun difference 48–51 Inuit agreement on determiners in 151, 190 agreement with objects in 235 case in 206 ergativity in 83, 101 parameter settings in 195, 205, 213 Italian agreement on adjectives in 47–8, 67–9, 70–1, 72–3, 145 impersonal sentences in 91 oblique subjects in 159 unaccusativity in 29–30 Itelmen 107 Iwara, Alex 103, 104, 164 Jacaltec 112–13, 165, 183, 194, 212 Jackendoff, Ray 33, 112 Jake, Janice 240, 243 Japanese 39 Jarawara 167, 190, 212 Kanerva, Joni 158, 160 Kannada 61–2, 90, 166, 213 Kaplan, David 124 Kariaeva, Natalia 177 Kayne, Richard 30, 52, 125 on agreement 75–6, 77, 86, 162, 198, 208 on crosslinguistic variation 156 Kilega agreement on complementizers in 181 agreement on prepositions in 193 subject agreement in 159, 160, 191, 208 Kimball, J. 75–7, 86 Kinande adjectival agreement in 24, 171 agreement on C in 120, 146, 147, 179, 180–2 linker head in 206–7 object agreement in 196, 197

Index subject agreement in 158–9, 160, 163–4, 208, 209–10 upward agreement in 75, 187, 191–3 Kinyalolo, Kasangati 52, 172, 191, 192, 193, 208 Kinyarwanda 160 Kirundi 160, 164 Koopman, Hilda 52, 113, 125, 129, 135, 166 Kornfilt, Jaqueline 58 Kratzer, Angelika 123 labels, of categories 52, 55–6, 79 Laka, Itziar 45 Lakhota 100 Lango 58 Launey, Michel 58, 102 Levin, Juliette 80 lexical categories definitions of 5, 26, 27–34 differing agreement properties of 1–3, 4, 12, 13–27 not true bearers of agreement 34–6 Li, Yafei 35 linkers in attributive constructions 64 as locus of agreement in NPs 20–1, 119 in verb phrases 206–7 Li’s Generalization 35–6 locality conditions 139–43 see also intervention, Person Licensing Condition locative inversion 158–9 Lokaa agreement on complementizer in 118–19, 120, 148 agreement on tense in 164, 208 long distance agreement in 103–4, 109 Luvale 164, 190, 211 MacKenzie, Marguerite 107 Madariaga, Nerea 117 Maithili 91, 156 Makhuwa 237–9 Mapudungun 115–16, 154 Marantz, Alec 9, 17, 38, 42, 162 Markman, Vita 89, 177 Massam, Diane 80, 202 matching, as condition on agreement 44

269

Mayali adjective versus verb agreement in 24–5 gender in 36, 39, 99 noun versus adjective agreement in 15–16, 19 Mchombo, Sam 124, 196 Merge 47, 52, 112–14, 143 Minimal Compliance 200 Minimalism 46, 114, 156 see also Chomsky, Noam Mirror Principle 60, 82 modifiers adjectives as 33–4 agreement properties of 17–21, 24, 50–1, 60–4, 147–8, 218 as appositional structures 17 nouns as 21 Mohawk active agreement in 54 agreement on nouns in 116 agreement parameters in 99, 201 incorporation in 29, 140 morphology, influence on agreement 7–9, 17, 87, 154–5, 204, 234 movement conditions on 53, 139–42 creating agreement configurations 161, 163, 188, 192, 196, 198–9, 227 of goal arguments 95 interactions with agreement 74–5, 81, 165–6, 198, 206 not dependent on Agree 160, 167 out of PP 193 as precondition for agreement 77, 81, 103, 106–7, 235 types of 161 Müller, Gereon 141 Munroe, Pamela 92–3 Murasugi, Kumiko 81 Nahuatl agreement in ditransitive constructions in 94–6, 98, 99, 102, 109 agreement on nouns in 58–60, 116 agreement on prepositions in 113, 195 parameter settings in 165, 201 Ndayiragije, Juvénal 162, 164 negation 27 Nepali 91, 156, 169–70 Nevins, Andrew 125, 129, 133–4

270

Index

Nez Perce 101, 184, 229–34, 235 Niinuma, Fumikazu 42 Noonan, Michael 58 Noun Licensing Condition 20, 28, 34 nouns absence of agreement on 2–3, 17–21, 48–51 agreeing with possessor 116 definition of 28, 30–2 functional heads associated with 36, 38, 39, 188–9 as having intrinsic phi-features 31–3, 126, 127 locative, compared to adpositions 195 as modifiers 21 person agreement on 56–60 see also predicate nominals number agreement in 86–7, 105, 109 different systems of 17 impoverishment of 9 location of, in NP 38, 62–3 mismatches of 66 not affected by ergative displacement 80 only nouns specified for 32 and semantic agreement 10–11 objects agreement with 3, 22, 35, 53–5, 101, 156, 173, 196–204, 206, 223, 237–8, 240 agreement with, versus clitics 98, 196–7 compared to oblique subjects 242 complementizer agreeing with 184 definiteness of, affecting agreement 199–200, 231–2, 234–5 moved to Spec, TP 160–3, 164, 229 participles agreeing with 198, 208 position of 62 tense agreeing with 76, 87, 91, 224, 232–5 see also double object constructions O’Herin, Brian 81–2, 224 operators agreeing with variables 121–4 agreement with 118–19, 180, 181, 182–3 first and second person 120–1, 125–8, 131, 136, 138, 142–3, 180 licensing person agreement on heads 145–7 logophoric 125, 135–7, 146, 179 see also wh-phrases Ormazabal, Javier 95, 197–8

Ottosson, Kjartan 70 Ouhalla, Jamal 5 parameters concerning agreement 7, 44 independence of 168–9, 239 location of phi-features as 64 nature of 155–7, 219–24, 236, 245 in probing up or down 85 in whether agreement happens 100, 154 see also Case Dependence of Agreement Parameter, Direction of Agreement Parameter participles, agreement on 174–5, 198, 208, 209, 210–11 particles 207 Passamaquoddy 77, 87, 105–7, 151, 214 passive voice and agreement 84–5, 101–2, 107 distinguishing clitics and agreement 98, 196 pseudopassives 192 Perlmutter, David 96 person distribution of agreement in 2, 3, 21, 85–91, 103–7, 108–9 in ergative agreement systems 101 on functional heads 112–21, 185 impoverishment of 9 inherent only on nouns 32, 126–7 marked on verbs but not adjectives 21–7, 51–6, 73–4, 89, 174 more restricted than number and gender 7, 149–50 on nouns and adjectives 56–60, 86 participles agreeing in 208 as result of operator–variable agreement 123–4, 148 special locality conditions on 92, 124–38 underspecification of 77 see also Person Licensing Condition, Structural Condition on Person Agreement Person Case Constraint (PCC) 94–8, 151 Person Licensing Condition (PLC) motivation for 126–8, 129, 132–4, 137–8 SCOPA derived from 138–9, 142–5, 147–9 phases clauses as 75 not restricting operator–variable agreement 121

Index prepositional phrases as 44, 71, 192–3, 204, 206 as restricting agreement 40, 43–4, 47, 48, 71, 75, 77 vP as 235 phi-features +anaphor as 150 completeness of, and case assignment 174 definiteness not included among 17 in different kinds of agreement 123 of expletive subjects 237 intrinsic only on nouns 31–3 left unspecified 76, 79, 80 of locative phrases 160–1 of oblique subjects 88–9 position of in structure 37, 63–4, 148 on predicate nominals 49–50 related to EPP, in Bantu 172 spelled out on functional heads 36 see also gender, number, person Pica, Pierre 139 Polinsky, Maria 103 Pollard, Carl 31, 108 Portuguese 209 possessors adjectival 185 agreement on determiners with 60, 115–16 agreement on tense with 232–4 postpositions, see prepositions Potsdam, Eric 103 Pred blocking person agreement on adjectives 144, 218 in nonverbal predication 29–30, 37, 45 possibility of agreement on 45, 47, 59–60 predicate, nonverbal agreement on 45, 60–4, 188–9, 219 diagnostics for 26 in instrumental case 177 predicate nominals absence of agreement on 2–3, 14–17, 49–50 person agreement on 56, 59 semantic agreement on 10–11, 61–2 structure of 28–9, 36–8 tense marked on 30 prepositional phrases as islands for agreement 43, 44, 70–1, 73 movement out of 193 oblique subjects as 89, 90–1 as phases 71

271

prepositions agreement on 112–13, 114, 143, 151, 173, 191–4, 196, 217 compared to case markers 204, 205, 206 compared to locative nouns 195 as functional categories 33 probes functional heads stipulated as being 39, 100, 102, 153, 234 not defined by unvalued features 44 searching upward or downward 47 pro-drop 110, 127 pronouns agreeing with their antecedent 121–3 agreement with 24, 89, 114–15, 118, 138–9, 194 compared to agreement 7, 149–50 licensing person features of 124–7, 128, 131, 138, 142–3 logophoric 125, 135–8 modification of 147–8 as source of agreement in predication structures 61–2 properties 25–6 quantifiers, see determiners Quechua differences among dialects of 239–44 parameter settings in 166, 195, 201, 205, 212, 214 raising 156, 209–10 Reference-Predication Constraint 28, 33 referential indices 30–1, 127–8, 148 Reimsdijk, Henk van 193 Reinhart, Tanya 40, 152 relative clauses 118–19, 147, 148, 183 restructuring 40, 46, 140–1 Reuland, Eric 152 Rezac, Milan 77–9, 80, 81, 100 Rice, Keren 129–33 Richards, Norvin 200 Ritter, Elizabeth 17, 33, 38 Rizzi, Luigi 67, 141, 182 Rögnvaldsson, E. 160 Romero, Juan 95, 197–8 Rood, David 100 Rude, Noel 229, 230 Russian 90–1, 117, 159, 177

272

Index

Safir, Kenneth 121, 125, 135, 138 Sag, Ivan 31, 108 Saltarelli, Mario 63 Schlenker, Philippe 123, 125, 126, 129, 132 Schütze, Carson 41, 87, 209 scrambling 165, 202, 217 semantics, influence on agreement of 10–11, 108, 121 Serbian 46, 66, 108, 210 Sesotho 159 Shambala 97, 98, 99, 109, 200 Shlonsky, Ur 227 Sigurð sson, Halldór 48, 70, 71 on person operators 125 on quirky subjects 88–9, 90–1, 176 Singhapreecha, Pornsiri 20 Slave parameter settings in 151, 189–90, 195, 202, 203, 212 shifted pronouns in 126, 129–33, 136, 137, 142 small clauses 48 Spanish agreement on functional heads in 38, 114, 117, 187 clitics/object agreement in 98, 197–8 linking particle in 20 location of phi-features in 64 noun versus adjective agreement in 15, 17–18 verbal agreement in 8, 39, 126–7, 208 verb versus adjective agreement in 22, 171 Speas, Margaret 156 specifiers adjectives not having 68, 74 of CP, agreement with 75–7 as defining property of verbs 28–30, 37, 51, 52 as distinguished position for agreement 52, 79, 113–14, 172–3 multiple 54, 144 properties of 47, 190 on right edge of phrase 168, 203 Sportiche, Dominique 52, 125, 129, 135 Sridhar, S. N. 61 Stassen, Leon 3, 25–7 Stechow, Armin von 123, 125 Sternefeld, Wolfgang 141 Stewart, O. T. 135–7

Stjepanovic, Sandra 46 Stowell, Timothy 180 Structural Condition on Person Agreement and agreement on adjectives 73, 85 and agreement on nouns 59–60 and agreement on verbs 89, 93, 94–5, 103, 244 and anaphoric features 150–2 applied to functional heads 112–21 compared to other conditions 92 definition of 52–6, 65 derivation of 111, 138–9, 143–6, 147–9 subjects agreement with 158–71, 173, 230, 237 case marking of 162 complementizer agreeing with 180–1, 183, 184 with oblique case 87–91, 94, 145, 159, 160, 169–70, 176–7, 223, 241–4 postverbal 225–6, 227–9 v agreeing with 77–85, 215–16 see also ergativity Swahili agreement on determiners in 114, 187 agreement on lexical categories in 1–3, 12, 14–15, 18, 22, 36, 38–9 copulas in 2 gender agreement in 8, 10 linking particle in 20–1 object agreement in 54–5, 150, 200 subject agreement in 159, 160, 164, 208 Taraldsen, Knut 87–93 Tariana nonagreeing verbs in 23 noun versus adjective agreement in 15, 16–17, 19–21 parameter settings in 189 v agreeing with subject in 83–5, 203 verb versus adjective agreement in 23–4, 55, 147 Taylor, Allan 100 tense agreeing downward 217–18, 227 agreeing upward 74–7, 86–7, 162 agreeing with object 76, 81–2, 87–93, 94, 224 agreeing with possessor 232–4 agreeing with postverbal subjects 81, 89, 159, 226, 237

Index agreement parameters applied to 156, 158–70, 171, 208–11, 223, 229, 230, 242–4 and assignment of case 162, 174 differences in, affecting agreement 80 double agreement on 83, 100–1, 102, 232–5 and locality of agreement 42, 81–2 as locus of subject agreement 34–5, 52, 54, 57–8, 59 nonfinite 75, 87 not agreeing with subject 84 only verbs inflect for 27, 30 person agreement on 143 variable probehood of 39, 100–2, 167 verb moving to 198–9 thematic (theta) roles 28, 31 Theta Criterion 34 Thráinsson, Höskuldur 160 Tiwa, Southern agreement in ditransitive constructions in 96–7, 101–2, 109, 151 verb incorporation in 35 topics 90–1, 93 Travis, Lisa 140 Tsez 103 Turkish agreement on all categories in 56–8 parameter settings in 168–9, 183, 213 Tzotzil ergative agreement in 82, 83 parameter settings in 213–14, 224 unaccusativity of adjectives 67–8 special agreement on unaccusative verbs 83–4, 101, 215–16 of verbs versus adjectives and nouns 29–30, 45 unergative verbs 215–16 universals of agreement 3, 26, 27–34 Ura, Hiroyuki 163 v agreeing upward with subject 77–85, 203, 224 agreement parameters applied to 196–206, 215–17, 232, 238, 240 in double object constructions 94–100, 102, 200

273

as locus of object agreement 35, 42–3, 53–4, 196–7, 223 Van der Wal, Jenneke 237–9 variables agreeing with their operators 121–4 created by agreement 124 verbs agreeing upward 74–85 without agreement 23 and agreement in person 2, 85–107, 109 agreement on, compared to adjectives 21–6, 51–5, 65–7, 107–8, 173 definition of 28–30 derived from adjectives 36 gender agreement on 8–9 impoverishment of features on 8–9 moving to Tense 198–9 structure of clause headed by 34–5, 36–8 subject agreement on 157–71 Warlpiri 170–1, 201, 204, 213 Watanabe, Akira 180 Webelhuth, Gert 156 Wechsler, Stephen 31, 66, 108, 114 Welsh 194 wh-phrases agreement on complementizer with 181–2 agreement on verb with 75–7, 86–7, 109, 162 Wichita 29 Williams, Edwin 31 Woolford, Ellen 150, 229–30 word order, influence on agreement 165–9, 186, 189–90, 192, 202–3 Wurmbrand, Susanne 40, 107, 160 Yadava, Yogendra 91, 156, 170 Yiddish 160 Yimas 14, 195, 201 Yoruba 136–7 Zanuttini, Rafaella 127 Zazaki 133–4, 136 Zlatić, Larisa 31, 66, 108, 114 Zulu agreement on modifiers in 147, 190 agreement on quantifiers in 114–15, 185–6 object agreement in 198–200 subject agreement in 159, 163, 208

The Syntax of Agreement and Concord - Mark C. Baker

Recommend Documents