Negative and Positive Polarity: A Binding Approach (Cambridge Studies in Linguistics)

In Negative and Positive Polarity, Ljiljana Progovac presents crosslinguistic data on negative polarity, reflexive bind...

Author: Ljiliana Progovac

93 downloads 717 Views 3MB Size Report

This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!

Report copyright / DMCA form

DOWNLOAD PDF

In Negative and Positive Polarity, Ljiljana Progovac presents crosslinguistic data on negative polarity, reflexive binding and the subjunctive mood, and proposes a unified analysis for various languages, including English and Serbian/Croatian. She argues that Negative Polarity Items (NPIs), such as 'anyone' and 'ever', are anaphoric in nature and must be bound in their governing category, while Positive Polarity Items (PPIs), such as 'someone' and 'already', are subject to Principle B of the Binding theory. She also suggests that possible binders (and SUBJECTS) for polarity items are negation or else a polarity operator in the Comp of questions, conditionals, and other clauses with an unfixed truth-value. Her analysis not only captures many similarities between polarity and anaphora, but also accounts for a number of hitherto unexplained facts about polarity items, for example, their scope restrictions, Comp involvement in NPI licensing, cross-linguistic variation, and their behavior in subjunctive clauses.

CAMBRIDGE STUDIES IN LINGUISTICS General Editors: J. BRESNAN, B. COMRIE, W. DRESSLER, R. H U D D L E S T O N , R. L A S S , D . L I G H T F O O T , J. L Y O N S , P H . M A T T H E W S , R. P O S N E R , S. R O M A I N E , N . V . S M I T H , N. V I N C E N T

Negative and Positive Polarity

In this series 35 MARTIN ATKINSON Explanations in the study of child language development 36 SUZANNE FLEISCHMAN The future in thought and language 37 JENNY CHESHIRE Variation in an English dialect 38 WILLIAM A. FOLEY and ROBERT VAN VALIN JR Functional syntax and Universal

Grammar 39 MICHAEL A. COVINGTON Syntactic theory in the High Middle Ages 40 KENNETH J. SAFIR Syntactic chains

41 J. MILLER Semantics and syntax 42 H. c. BUNT Mass terms and model-theoretic semantics 43 HEINZ J.GIEGERICH Metrical phonology and phonological structure 44 JOHN HAIMAN Natural syntax

45 BARBARA M.HORVATH Variation in Australian English: the sociolects of Sydney 46 GRANT GOODALL Parallel structures in syntax: coordination, causatives, and restructuring 47 JOHN M. ANDERSON and COLIN J. EWEN Principles of dependency phonology

48 BARBARAA.FOX Discourse structure and anaphora 49 LAUREL J.BRINTON The development of English aspectual systems 50 DONNA JO NAPOLI Predication theory

51 NOEL BURTON-ROBERTS The limits to debate: a revised theory of semantic proposition 52 MICHAEL s. ROCHEMONT and PETER w. CULICOVER: English focus

constructions and the theory of grammar 53 PHILIP CARR: Linguistic realities: an autonomist metatheory for the generative enterprise 54 EVESWEETSER: From etymology to pragmatics: metaphorical and cultural aspects of semantic structure 55 REGINABLASS: Relevance relations in discourse: a study with special reference to Sissala 56 ANDREW CHESTERMAN: On definiteness: a study with special reference to English and Finnish 57 ALESSANDRA GIORGI and GIUSEPPE LONGOBARDK The syntax of noun

phrases: configuration, parameters and empty categories 58 MONIKCHARETTE: Conditions on phonological government 59 M. H. KLAIMAN: Grammatical voice 60 SARAH M.B. FAGAN: The syntax and semantics of middle constructions: a study with special reference to German 61 ANJUMP. SALEEMI: Universal grammar and language learnability 62 STEPHEN R. ANDERSON: A-Morphous Morphology

63 LESLEY STIRLING: Switch reference and discourse representation 64 HENKJ.VERKUYL: A theory of aspectuality: the interaction between temporal and atemporal structure 65 EVEVCLARK: The lexicon in acquisition 66 ANTHONY R WARNER: English auxiliaries: structure and history 67 p. H. MATTHEWS: Grammatical theory in the United States from Bloomfield to Chomsky 68 LJILJANAPROGOVAC: Negative and positive polarity: a binding approach

Supplementary

volumes

ANNICKDEHOUWER: The acquisition of two languages from birth: a case study LILIANEHAEGEMAN: Theory and description in generative syntax. A case study in West Flemish.

Earlier issues not listed are also available

NEGATIVE AND POSITIVE POLARITY A binding approach LJILJANA PROGOVAC English Department, Wayne State University

CAMBRIDGE UNIVERSITY PRESS

CAMBRIDGE UNIVERSITY PRESS Cambridge, New York, Melbourne, Madrid, Cape Town, Singapore, Sao Paulo Cambridge University Press The Edinburgh Building, Cambridge CB2 2RU, UK Published in the United States of America by Cambridge University Press, New York www.cambridge.org Information on this title: www.cambridge.org/9780521444804 © Cambridge University Press 1994 This publication is in copyright. Subject to statutory exception and to the provisions of relevant collective licensing agreements, no reproduction of any part may take place without the written permission of Cambridge University Press. First published 1994 Hardback version transferred to digital printing 2005 Digitally printed first paperback version 2005 A catalogue recordfor this publication is available from the British Library Library of Congress Cataloguing in Publication data Progovac, Ljiljana. Negative and positive polarity : a binding approach / Ljiljana Progovac. p. cm. — (Cambridge studies in linguistics; 68) Includes bibliographical references and index. ISBN 0 521 44480 2 (hardback) 1. Grammar, Comparative and general-Syntax. 2. Anaphora (Linguistics) 3. Government-binding theory (Linguistics) I. Title. II. Series. P291.P776 1994 415-dc20 93-12938 CIP ISBN-13 978-0-521-44480-4 hardback ISBN-10 0-521-44480-2 hardback ISBN-13 978-0-521-02379-5 paperback ISBN-10 0-521-02379-3 paperback

To Stefan and Ana

Contents

Preface

xi

0.1. 0.2 0.3

Introduction Polarity Binding theory Overview

1 3 10 20

1 1.1 1.2

Negative contexts: Serbian/Croatian Issues in Serbian/Croatian syntax Negative polarity in Serbian/Croatian

23 23 33

2 2.1 2.2 2.3

Negative contexts: English Positive Polarity Items (PPIs) Negative Polarity Items (NPIs) and LF-raising Independent evidence for LF-raising

54 54 54 57

3 3.1 3.2

62 62

3.3 3.4 3.5

Non-negative polarity contexts Introductory words Non-negative licensing patterns with superordinate negation Operator in Comp Non-negative licensing from Comp Independent evidence for polarity operator

64 66 70 75

4 4.1 4.2

Language variation Raising options Binding options

79 79 90

5 5.1

Rhetorical questions H^-questions

97 97

X

Contents

5.2 5.3

Why and How Yes/No questions

101 102

6 6.1 6.2 6.3 6.4 6.5

Subjunctive: domain extensions

104 104 108 116 119 120

7 7.1 7.2 7.3 7.4 8 8.1 8.2 8.3 8.4

Introductory words Sub-categorized clauses Adverbial clauses Relative clauses Summary and discussion Free-choice items

Introductory words Free-choice Bilo in Serbian/Croatian An analysis Free-choice Any

122 122 123 127 130

Ladusaw Linebarger Polarity/operator interactions Conclusion

132 132 136 139 141

Notes References Index

143 157 165

Ladusaw and Linebarger

Preface

This book offers cross-linguistic data on negative polarity, reflexive binding and the subjunctive mood and proposes a unified analysis for various languages. It can be used as a textbook for intermediary or advanced courses in syntax and semantics, especially courses dealing with negative polarity, binding, or syntax/semantics interfaces in general. The discussion presupposes acquaintance with only an elementary syntax text, such as Haegeman's or Radford's introduction to the syntactic theory. The monograph is based on my (1988) dissertation "A Binding Approach to Polarity Sensitivity", although it has been considerably modified and updated. There are three major additions to the dissertation: (i) a novel treatment of the subjunctive mood, as involving deletion of functional categories Infl and Comp in LF (chapter 6); (ii) the discussion of the Relativized SUBJECT approach to long-distance anaphora (section 0.2); and (iii) an overview of the basic issues in Serbian/Croatian1 syntax (section 1.1.). Moreover, the monograph provides a discussion of some recent developments in the treatment of negative polarity, such as Zanuttini (1991) and Laka (1990), especially concerning the analysis of negative concord. I am indebted to many people for their valuable help with the dissertation. The deepest thanks go to my advisor, Joseph Aoun, who at the same time taught, directed, and inspired, and continues to do so. For their precious comments, I owe a great deal to Marc Authier, Miirvet Enq, Irene Heim, Larry Hyman, Osvaldo Jaeggli, Audrey Yen-Hui Li, and Dominique Sportiche. For his brilliant comments and the time he devoted to the present version of the monograph, I am deeply grateful to Richard Kayne. I am also very grateful to the two CUP reviewers and the editor, Catherine Max, who are directly responsible for many substantive improvements in the monograph. Thanks also to Phil Connell, Steve Franks, James Huang, Alice ter Meulen, Johan Rooryck, Jindfich Toman, Lisa Travis,

xii

Preface

Natsuko Tsujimura, and Raffaella Zanuttini for comments on different aspects of the monograph. Some ideas in the monograph have been (or will be) published in the papers listed in the references. Many thanks to the reviewers and editors for numerous insightful comments and suggestions, especially to Chisato Kitagawa. Parts of the monograph have been presented at different conferences or colloquia. Thanks to the audiences for helpful questions and discussions, especially to Ellen Barton, Susan Bennett, Helen Goodluck, Sabine Iatridou, Kyle Johnson, Bill Ladusaw, Barbara Partee, Pierre Pica, Martha Ratliff, Marissa Rivero, Thomas Stroik, Arhonto Terzi, and Olga Tomic. Many thanks to Clark Iverson for a most careful proof-reading. Warmest thanks to my husband Dusan and my children Ana and Stefan for patience and love. Notes For the reasons why I use the term Serbian/Croatian for my native language, which has traditionally been referred to as Serbo-Croatian, see note 1 of the Introduction.

Introduction

This monograph advocates a relativized approach to the bindingtheoretic notions of binder and SUBJECT, making it possible to subsume polarity items under the Binding theory. The Binding theory was primarily designed to cover locality conditions between anaphors and pronominals and their antecedents. I will defend the view that binding/ obviation principles only happened to be formulated for reflexives/ pronouns first, but that there is no reason to think that they are restricted to them. In fact, in a maximally restrictive model, the same locality conditions should hold for all dependent and anti-dependent phenomena. Since reflexives and negative polarity items are both dependent, they both must be in the scope of (or bound to) their licenser or antecedent, and since pronouns and positive polarity items are both anti-dependent, they both must be interpreted outside of the scope of a local antecedent. These scope properties of polarity items must be stipulated in other frameworks, but follow directly from the binding analysis. It is a longnoted puzzle that negative polarity items always receive narrow-scope interpretation with respect to negation, and that positive polarity items must receive a wide-scope interpretation with respect to local negation, but either narrow or wide-scope interpretation with respect to superordinate negation. This sets polarity items apart from other quantifiers, which enter freely into scope ambiguities with negation, exhibiting both wide- and narrow-scope readings (cf many and every in (1)). (1)

Many/every arrow(s) didn't hit the target. (a) For every/many X, it is true that X did not hit the target. (b) It is not true for every/many X that X hit the target.

Negative Polarity Items = NPIs (e.g. any) are licensed in negative, but not in the corresponding affirmative sentences: (2)

John did not insult anyone.

2 (3) (4)

Introduction I do not say that John insulted anyone. *John insulted anyone.

They are also licensed in questions, antecedents of conditionals, complements of adversative predicates, etc: (5) (6) (7)

Did Mary insult anyone! If Mary insulted anyone, she should apologize. I doubt that Mary insulted anyone.

NPIs will be differentiated from the homophonous free-choice items (FCIs), which are typically licensed by modals: (8)

Anyone can climb that tree.

Positive Polarity Items = PPIs (e.g. some), on the other hand, cannot appear in the scope of a clausemate negation, but are licit elsewhere, e.g. with superordinate negation and in affirmative contexts: (9) (10) (11)

#Mary did not insult someone. I don't say that Mary insulted someone. Mary insulted someone.

(The symbol '#' indicates that someone must be interpreted as taking wide scope with respect to negation.) It will be argued that the best way to capture the distribution and interpretation of negative and positive polarity items is within the Binding Theory (for semantically based approaches to polarity, e.g., Ladusaw's (1980) and Linebarger's (1981), see chapter 8). More precisely, I will argue that negative polarity items obey Principle A, whereas positive polarity items obey Principle B: (12) (13)

NPIs are subject to Principle A of the Binding Theory. PPIs are subject to Principle B of the Binding Theory.

A potential binder for polarity items is an A'- polarity operator: either negation or an empty polarity operator (Op) generated in Comp (see Aoun (1985, 1986) for Generalized Binding, and also section 0.2.3.). The arguments presented in this monograph are based on data from English and Serbian/Croatian1, a language which exhibits two different types of NPIs and therefore provides good insight into the nature of NPIlicensing. References are also made to other languages, such as Catalan,

Polarity 3 Chinese, French, Icelandic, Italian, Korean, Russian, and Turkish, to further support the analysis. The first part of the introduction (0.1.) is an overview of the analysis of polarity items within the Binding Theory. The second part (0.2.) offers a brief discussion of the Binding Theory adopted in the monograph. Section 0.3. offers a chapter-by-chapter overview of the book.

0.1. Polarity 0.1.1. Parallelism between anaphora and polarity Just as reflexives are local (see section 0.2.1. for a local account of longdistance reflexives), so are NPIs in many languages. Example (14) shows that a reflexive must take a clausemate antecedent, whereas examples (15) to (23) illustrate that NPIs must be licensed by a clausemate negation across languages: English: (14) Janej believes [that Maryj respects herself*^.] Serbian/Croatian: (15) Marija *(ne) voli ni(t)ko-ga.2 Mary not loves noone-ACC 'Mary does not love anyone.' (16) *Ne tvrdi-m da Marija voli ni(t)ko-ga. not claim-1SG that Mary loves noone-ACC 'I do not claim that Mary loves anyone.' (17) *Da li Marija voli ni(t)ko-gal that Q Mary loves noone-ACC 'Does Mary love anyone?' Russian: (18) Maria *(ne) vidit nikogo. Mary not sees noone 'Mary does not see anyone.' (19) * Peter ne skazal, [cto Maria vidit nikogo]. Peter not says that Mary sees noone 'Peter did not say that Mary can see anyone.'

4 (20)

Introduction *Ja somnevayus\ [cto Maria vidit nikogo]. I doubt that Mary sees noone 'I doubt that Mary can see anyone.'

Chinese (Audrey Y.-H. Li, p. a): (21) Ta conglai meiyou kanjian Mali, he ever did-not see Mary 'He has never seen Mary.' (22) *Zhangasan meiyou shuo ta conglai kanjian quo Mali. John did-not say he ever see Asp Mary 'John did not say that he had ever seen Mary.' (23) *Ta conglai kanjian Mali ma? he ever see Mary Q 'Has he ever seen Mary?' (See also Milner (1979) for personne in French, Sedivy (1990) for Czech, Hasegawa (1987) for Japanese, Linebarger (1981) for strict NPIs in English, etc.) On the other hand, just as pronouns must be obviative from a local antecedent, as in (24), so must PPIs take wide scope with respect to local negation (as in (25)): (24) (25)

Maryi saw her*^. #Mary did not insult someone. There is a person X such that Mary didn't insult X. (wide-scope reading) *Mary insulted no one. (narrow-scope reading - unavailable)

Likewise, just as a pronoun can be either bound to, or obviative from, a non-local antecedent, so can a PPI take either wide or narrow scope with respect to a non-local negation: (26) (27)

Peter, doesn't say that John insulted ^ Peter doesn't say that John insulted someone. There is a person X such that Peter doesn't say that Mary insulted X. (wide) Peter doesn't say that Mary insulted a person/any person. (narrow)

The following crucial question arises: why should obviation correspond to wide-scope reading (Richard Kayne, personal communication)? Generally speaking, wide scope corresponds to "outside of the influence

Polarity 5 of," whereas narrow scope with respect to something corresponds to "within the influence of something." For example: (28)

John did not shoot many arrows.

On the wide-scope reading, many escapes the influence of negation, and has its dictionary meaning: a great number. On the narrow-scope reading, on the other hand, it amalgamates with negation (in the sense of Cinque 1990, pp. 80-81), falls within the influence of negation, to result in something corresponding to few. Likewise, someone (meaning one person) when in the scope of a universal quantifier can refer to different individuals: (29)

Everybody loves someone.

Only on the wide-scope reading is its dictionary meaning (one person) unaffected. If one can indeed claim that wide-scope corresponds to "outside the influence of," then it must be that obviation, rather than binding, should be subsumed under this notion. While a bound expression draws its reference and all features from its antecedent, an obviative expression avoids such an influence on its reference completely. Furthermore, polarity items show the same behavior as reflexives and pronouns in Exceptional Case Marking (ECM) environments: (30) (31)

John considers [IP himself (to be) intelligent]. Johni considers [IP him*^ (to be) intelligent].

Serbian/Croatian: (32) Milan ne smatra [IP ni(t)ko-ga vredn-im Milan not regards no-one-ACC worthy-INSTR tolik-e paznj-e]. such-GEN attention-GEN 'Milan does not regard anyone worthy of so much attention.' (33) #John doesn't consider [IP someone worthy of so much attention]. There is someone (a person) who John does not consider worthy of so much attention, (wide) *John considers no one worthy of so much attention, (narrow) Obviously, in the ECM subject position all the four items extend their domain. Next, the domain for both anaphora and polarity extends in subjunctive clauses (for an analysis of subjunctive, see chapter 6):

6

Introduction-

Reflexives: Icelandic: (Hyams and Sigurjonsdottir (1990)) (34) Jonj segir ad Peturj raki sig ^ John says that Peter shaves-SUBJ self (35) Joni veit ad Peturj rakar sig*^. John knows that Peter shaves-IND self Italian (Giorgi (1984)): (36) Giannii suppone che tu sia innamorato della propria 'Giannii supposes that you are-SUBJ in love with self s^ moglie. wife.' (37) *Giannii mi ha detto che tu sei innamorato della propriax moglie. 'Giannii told me that you are-IND in love with self Si wife.' See also Pica (1985) for French data. Negative polarity: French (Kayne (1981)): (38) ?Je n'ai exige qu'ils arretent personne. I neg-have required that they arrest nobody Italian (Rizzi (1982)): (39) Non pretendo che tu arresti nessuno. neg require-I that you arrest no-one The consistent parallelism between reflexives and NPIs, on the one hand, and between pronouns and PPIs, on the other, seems to call for a unified analysis. 0.1.2. Polarity: a binding analysis A binding analysis of polarity seems highly desirable since it offers a costfree way to capture the near-complementary distribution between NPIs and PPIs, also attested for reflexives and pronouns. Let us propose that negative and positive polarity items are subject to the same locality conditions that have been embodied in the Binding Theory: (40)

NPIs are subject to Principle A: they must be bound to negation (or other truth-functional operator) in their governing category (see Milner (1979), Aoun (1985, 1986), Progovac (1988, 1991, 1992b))

Polarity 1 (41)

PPIs are subject to Principle B: they must not be bound to (fall within the scope of) negation (or a truth-functional operator) in their governing category (see Progovac 1988, 1991, 1992b).

This immediately explains the basic polarity facts established in examples (2) to (11) above: (42)

NPIs (as anaphoric) need a licenser. The licenser for NPIs has to be local. PPIs (as pronominal) need no licenser. PPIs cannot be bound to a local truth-conditional operator.

One potential obstacle for treating NPIs on a par with reflexives comes from the fact that the domain for the two rarely coincides in any given language. Thus English has local reflexives, but long-distance NPIs. On the other hand, Russian has long-distance reflexives, but local NPIs. However, there is no reason to expect the same domain for reflexives and NPIs, on the assumption that governing categories and binders are relativized. While potential antecedents for reflexives in English are NPs in specifier positions (see section 0.2.1.), potential antecedents for NPIs are functional categories: negation in Infl or a truth-conditional operator in Comp. In fact, there are many languages in which even different types of reflexives have different binding domains, as illustrated below for ziji and ta ziji in Chinese: (43)

Zhangsani renwe [Lisij zhidao [Wangwuk xihuan ziji^^ / thinks knows likes self ta zijuirj/k\] he-self 'Zhangsan thinks that Lisi knows that Wangwu likes himself.'

Here, the simple reflexive ziji shows no domain effect, whereas the complex reflexive ta ziji is strictly clause-bound, just like English himself. We will see in section 0.2.1. that this difference need not be stipulated, but follows from the different morphological properties of these reflexives and, thus, from a different choice of binder. Besides capturing the basic (anti)licensing of polarity items by negation, this approach also extends successively to non-negative polarity contexts, such as conditionals, questions, etc: (44) (45)

Did Mary insult any one \ someone] If Mary insults anyone]someone, she will regret it.

8

Introduction

In English and Serbian/Croatian, licensing in non-negative contexts patterns with licensing by superordinate negation, rather than by clausemate negation, in that both NPIs and PPIs are allowed. I argue that a null polarity operator appears in the Comp of all non-negative polarity clauses (see Progovac (1988, 1992b) and Laka (1990)). Since the first potential antecedent for polarity items is negation in Infl (or NegP), the governing category for these items cannot extend further than IP. This explains why NPIs and PPIs must be in complementary distribution only within IP. An Operator in Comp does not anti-license PPIs since it is outside of the binding domain. Only those NPIs which can raise in LF higher than IP can extend their domain and be bound to the operator. Informally, the Op is assumed to represent a switch with a + /- value, reflecting the fact that all the clauses headed by Op have an indeterminate truth value (e.g., questions, conditionals, etc.). The negative value of Op is now responsible for NPI licensing, reducing it to negation. Three pieces of evidence are presented for the Comp involvement in NPI licensing, dealing with (i) complements to adversative predicates, (ii) complements to universal quantifiers and (iii) comparatives. In each case it is only in clausal (as opposed to phrasal) complements that NPIs are licensed. The following contrast illustrates the point for adversative predicates: (46) (47)

I doubt [CP Op that John understood anything]. *I doubt anything.

The NPI is licensed in (46) because the complement of doubt is sentential, and therefore hosts a Comp position. On the other hand, in (47) there is no Comp position for Op selection. In addition to strict locality of negative polarity in numerous languages, contrasts like that between (46) and (47) argue strongly against treating polarity in purely semantic terms (for additional arguments, see chapter 8). Although I argue that LF movement of NPIs does not in itself constitute an explanation for their scope properties, I do not exclude the possibility that they raise in LF, the way other quantifiers do. In fact, there is evidence for movement of English any, as discussed in chapter 2 (see Progovac (1988, 1992b) and nessuno (see Longobardi (forthcoming)). This explains why English any can be bound outside its local clause, in fact, in all polarity contexts: (48) (49)

John did not insult anyone. I don't say that John insulted anyone.

Polarity 9 (50) (51)

Did Mary insult anyone! I doubt that Mary insulted anyone.

The main empirical advantage of the binding approach to polarity sensitivity is in its ability to predict variation in NPI-licensing across languages (see chapter 4). A wide range of attested variation can be captured by the interplay of two separate factors: different raising options and being subject to different binding principles. Some NPIs do not raise at all, and therefore remain strictly local, e.g., ni(t)ko-type NPIs illustrated above. The inability to raise will be reduced to their morphological complexity. Some NPIs raise either by IP-adjunction or by movement through Comp, e.g., English any, and can consequently be bound either by truth-conditional operators in Comp, or by superordinate negation. Some NPIs can move only by IP-adjunction, e.g., certain NPIs in Romance and Chinese, and can thus be bound by operators in Comp, but not by superordinate negation. Those NPIs which move only through Comp can be bound by superordinate negation, but not by operators in Comp, e.g., NPIs in Turkish. Along the binding dimension, all NPIs must be bound by negation or a truth-conditional operator. Most NPIs are subject to Principle A, e.g., NPIs in English, Italian, Chinese, etc. Some NPIs, however, can obey Principle B, e.g., Serbian/Croatian I-NPIs, resulting in the requirement that they be bound only outside their local clause, either by superordinate negation, or by operators, but not by clausemate negation. PPIs, on the other hand, can be subject either to Principle B only, as in English and Serbian/Croatian, or to Principle C (which subsumes Principle B), as in Russian. In Russian, PPIs must be disjoint from (i.e., fall outside the scope of) not only local negation, but any polarity licenser in the sentence, including superordinate negation. What distinguishes the NPIs that move from those that do not? I take the ni(t)ko type NPIs in Russian and Serbian/Croatian to be local due to their morphological complexity. The item ni(t)ko consists of three morphemes: n-i-(t)ko (neg-any-who), and thus contrasts with English any: (52)

DP OP

DP D

any

NP one

10 Introduction (53)

DP NegP

DP QP DP D

n

i

NP (t)ko

If only QPs, i.e., any part could raise, then we have an explanation for the strict locality for ni(t)ko\ the raising of / in ni(t)ko would result in a minimality violation, where the trace of / would be minimally governed by morphological negation (for details, see section 2.2). 0.2. Binding theory

The purpose of this section is to discuss the essentials of the binding theory, on which I base my analysis of polarity. I assume the original formulation of Principles A and B but argue for a relativization of the notions 'binder' and 'SUBJECT.' 0.2.1. Principle A: relativized SUBJECT

What is appealing about the original formulation of the binding principle A, given in (54), is its simple intuition that the first potential antecedent (i.e., SUBJECT) closes off the binding domain for an anaphor (Chomsky (1981)): (54a) Principle A: An anaphor must be bound in its governing category. (54b) The governing category for an anaphor is the smallest maximal projection containing the anaphor, the governor for the anaphor, and a SUBJECT accessible to the anaphor. SUBJECTS for reflexives are [NP,IP], [NP,NP], or AGR. The core-binding facts illustrated below thus follow from the formulation in (54), where binding across either [NP,IP] or [NP,NP] subject is prohibited: (55) (56) (57)

Maryi believes [that Janej respects herself*^]. Maryi believes [Janej to respect herself*^]. Maryi admires [Jane'Sj stories about herself*^].

Binding theory

11

While it is intuitively appealing and theoretically justified in many modules to assume that a potential antecedent closes off the locality domain, there are two major problems with the choice of SUBJECTS (i.e., potential antecedents) in (54). First of all, accruing cross-linguistic evidence has cast doubt on the validity of the rigid notion of SUBJECT, given that certain anaphors seem to be insensitive to certain instances of SUBJECT (the so-called long-distance (LD) reflexives), as illustrated in the following examples from Korean (Yang (1983)) and Russian (Rappaport (1986)): (58)

(59)

Johni-in [Billj-i [Maryka [Tomriy caki-^^x taehan John-TOP Bill-NOM Mary-NOM Tom's self toward thaeto]-lil silhh9-n-ta-ko] saengkakha-n-ta-ko] attitude-ACC hate-ASP-DEC-COMP think-ASP-DEC-COMP mit-nm-ta. believe-ASP-DEC 'John believes that Bill thinks that Mary hates Tom's attitude toward self.' Professori poprosil assistentaj [PROj citat' svoji/j doklad] Professor asked assistant read-INF self s report 'The professori asked the assistant^ to read self Si/j report

In (58) the monomorphemic reflexive caki (self) can be bound across two instances of [NP,IP] and one instance of [NP,NP]. In (59), svoj (self s) can be bound across PRO, again an instance of [NP,IP]. The same facts obtain for monomorphemic reflexives across languages (see Yang (1983)), pointing to the conclusion that [NP,IP] and [NP,NP] cannot be SUBJECTS for monomorphemic reflexives. On the other hand, AGR seems to be a SUBJECT for such reflexives since binding across finite clauses is only possible in languages without overt agreement, e.g., in Korean (see (58)), but not in Russian (see (64) below). The second problem with the choice of SUBJECTS in (54) is the fact that the set does not constitute a natural class and is essentially arbitrary. Notice that the set represents a mixture of functional heads (AGR) and lexical specifiers (NP and IP subjects). This is especially problematic in the light of the recent discoveries that X-bar compatibility plays a crucial role in other modules of grammar (for movement, see Emonds (1976), Baltin (1982), Chomsky (1986a)); for government, see Rizzi (1990a)). To solve the problems mentioned above, I will adopt the Relativized SUBJECT formulation of Principle A, as proposed in Progovac (1992),

12 Introduction Progovac and Connell (1991) and Progovac and Franks (1991) although my analysis of polarity is not necessarily incompatible with other formulations. On this approach, a reflexive selects only an X-bar compatible category for its SUBJECT, thus monomorphemic (X°) reflexives take X° antecedents (i.e., SUBJECTS), whereas XP (complex) reflexives take XP specifier SUBJECTS: (60)

Relativized Principle A: A reflexive R must be bound in the domain D containing R and an X-bar compatible SUBJECT; If R is an X° (morphologically simple) reflexive, then its SUBJECTS are X° categories only, i.e., AGR (as the only salient (ccommanding) head with pronominal features). If R is an x m a x (morphologically complex) reflexive, its SUBJECTS are Xmax specifiers with pronominal features, thus [NP,IP] and [NP,NP].

Both Yang (1983) and Nakamura (1987) notice that AGR is the only possible SUBJECT for simple reflexives, and they parametrize the domain accordingly. In this analysis the fact that AGR is the only SUBJECT is derived. Relativization of SUBJECT solves both of the problems pointed out above. First, the choice of SUBJECT is no longer arbitrary, given that in other modules as well, e.g., movement and government, specifiers are relevant for maximal projections, and heads for other heads. Second, we can now explain why simple X° reflexives can be bound across specifiers, i.e., long-distance: they do not recognize any maximal categories as their SUBJECTS.

Before I return to simple reflexives, let us see how we can safely exclude AGR as a SUBJECT for complex reflexives. This will have a desired consequence of eliminating the stipulative accessibility requirement from Principle A. The core case where AGR seems required to close off the domain for a complex anaphor is given below (see Chomsky (1981)): (61)

*John believes [that himself AGR will win].

There is growing consensus in the literature that ungrammaticahty of (61) is due to a module other than the binding theory (see Lebeaux (1983) for the ECP, and Rizzi (1990b) for local AGR-anaphor chains). I will explore the possibility that a reflexive is banned from the subject position (in languages with overt agreement) because, being dependent on reference

Binding theory

13

and pronominal features, a reflexive does not have an index (see Manzini (1992: 142)), and thus cannot pass one onto the AGR node. By convention, subject and AGR are coindexed (see Chomsky (1981)). In other words, if Principle A applies only in LF, as assumed by most, a subject reflexive will acquire an index from its antecedent only in LF, and will not be able to support an overt, S-structure agreement. This explains why subject reflexives are in general permitted in languages without overt agreement, such as Chinese, Japanese, Korean, etc. In these languages AGR features, being non-overt, need not receive an index until LF. The i-within-i fact in (62) below follows from this approach without a need for the stipulative accessibility condition of the original formulation: (62)

Johni believes [that [pictures of himself^ AGR will be on sale].

Since AGR does not act as a SUBJECT for complex reflexives, the domain extends up to the matrix clause. Exactly as predicted, such an extension will not be possible for simple reflexives, since they recognize AGR as their SUBJECT (see Yang (1983), Pica (1987)): Russian: (63) *Vanjai znaet [cto [stat'ja o sebet] AGR -NM knows that article-NM about self pojavilas' v gazete] appeared in newspaper This difference in behavior between simple and complex reflexives cannot follow from the LF-movement approaches, according to which simple reflexives are farther reaching. We would also not expect accessibility to be parametrized for different types of reflexives. This is because it is a highly abstract principle, the positive evidence for which is not readily available in the input. Going back to the formulation of Principle A in (60), it predicts exactly three logical possibilities for the length of the domain for (pure) anaphors: 1. local XP whose specifier has pronominal features (i.e. NP or IP) for complex (bimorphemic) reflexives; 2. the first clause showing AGR for monomorphemic reflexives in languages with overt agreement; 3. no domain for monomorphemic reflexives in languages with no overt agreement

14

Introduction

This is exactly what is attested. Type 3 domain is illustrated for languages with no overt AGR, such as Korean (see (58)), Japanese, and Chinese (see Huang (1982), Aoun (1986) and Fukui (1986) for claims that there is no (referential) AGR in these languages). Type 2 domain is attested in Russian. The contrast between (59) above and (64) below shows that the domain for the simple reflexive can stretch only across infinitival boundaries in Russian, which language shows overt subject/verb agreement: (64)

Vanjai znaet [cto Volodjaj ljubit svoju*^ zenu]. Vanja knows that Volodja loves self s wife

This difference in domain-length between Russian-type and Korean-type languages cannot be captured by the LF-movement analysis without a stipulation. On the other hand, it is exactly what a relativized SUBJECT analysis predicts. Type 1 domain is attested for complex reflexives across languages, including English (see examples (55) to (57) above). Complex reflexives such as herself seem to consist of a specifier her and a head self, as explicitly suggested in Helke (1971) and Katada (1991). This analysis is supported by the possibility to insert adjectives between the two, for example, her usual self her normal self etc. That the domain type is not a property of the language, but rather a property of the reflexive, is best illustrated in the following minimal contrast from Chinese: (65)

Zhangsanj renwei [Lisij zhidao [Wangwu^ xihuan zijiXj^ / thinks knows likes self ta z//7*i/*j/k]] he-self 'Zhangsan thinks that Lisi knows that Wangwu likes himself.'

Here, the simple reflexive ziji shows no domain effect, whereas the complex reflexive ta ziji is strictly clause-bound, just like English himself. What is the exact mechanism for the domain extension in languages with no overt AGR? A first guess would be that the lack of morphological AGR entails the lack of the AGR node altogether. This would trivially predict that the domain for simple reflexives will never be closed, since their only SUBJECT will be consistently missing. However, this would lead to a paradox. Since we started with the assumption that SUBJECT is a potential binder, we would end up with the conclusion that a

Binding theory

15

simple reflexive will have no binder in languages with no overt agreement. Besides, subject-orientation and blocking effects (to be discussed below) would remain unexplained. For these reasons I will assume with Borer (1989) that the absence of morphological AGR in Chinese-type languages and infinitivals does not entail the absence of syntactic AGR (see also Huang and Tang (1989), Cole, Hermon and Sung (1990)). Rather, null AGR can be dependent or anaphoric in some relevant sense. Suppose that null AGR may or may not form a chain with a higher ccommanding AGR (cf., controlled vs. arbitrary PRO), as suggested to me by James Huang (personal communication). On the other hand, overt AGR, possibly subject to Principle C or B, must not depend on the higher AGR for its content. An X° reflexive will always be bound to the local AGR, obeying relativized Binding Principle A, given in (60). Consider the following example: (66)

Zhangsani AGR2i shuo [Lisij AGRlj chang piping zijiv say often criticize self

AGR-1 is the SUBJECT and binder for the X° reflexive ziji. If the local AGR-1 is coindexed with the next higher AGR-2, by transitivity, the reflexive will also be bound to AGR-2, and can be coindexed with either of the subjects. This would be parallel to the binding of anaphors by other anaphors (James Huang, personal communication): (67)

John believed [himself\ to have persuaded himself2 [to criticize himself^.

Himself?, is ultimately bound to John, which looks like long-distance binding, but this is only achieved via a set of local anaphoric dependencies. Notice that / in (66) is a feature index (in the sense of Huang and Tang (1989), see also Manzini (1992) for "address" indices), which does not involve actual co-reference, so that feature-indexing Zhangsan and Lisi with / does not imply that they are co-referential. By convention, AGR is also feature-coindexed with the subject (see Chomsky (1981)). This then explains why, in case AGRs are coindexed, subjects must be feature-compatible, but no more than that, i.e., we have an immediate explanation for the so-called blocking effects in Chinese (Huang and Tang (1989)). Long-distance binding of ziji is suspended if the intervening subject shows different person features:

16 Introduction (68) (69)

Zhangsani AGR2i shuo [WOJ AGRlj chang piping ziji*^]] Zhangsan say I often criticize self Zhangsa^ AGR3 shuo [WOJ AGR2 zhidao [Lisik AGRl chang Zhangsan say I know Lisi often piping

ziji*irj/k]].

criticize self Here AGRl cannot be coindexed with AGR2 since they do not share features, and therefore the reflexive can only be bound to the local AGRl, and by transitivity, to the local subject. Another salient property of monomorphemic reflexives is Subject orientation, i.e., the fact that they can be bound to subjects, but not to objects, as illustrated below for Chinese ziji and Russian sebja: (70) (71)

Zhangsani songgei Lisij zijii^ de shu. Zhangsan gave Lisi self 's book Milicioneri rassprasival arestovannogOj o sebe{/*y policeman questioned suspect about self

Subject orientation follows without stipulation from this system. Since a simple reflexive must be bound to AGR by X-bar compatibility, it ends up being bound to subjects, since generally only subjects are coindexed with AGR (see Chomsky (1981)). I consider Relativized SUBJECT approach to long-distance anaphora superior to the LF-movement analyses (see Pica (1987), Battistella (1987), Huang and Tang (1989), Cole, Hermon, and Sung (1990), Katada (1991)) because of the following questions that remain unanswered in the latter approaches: If reflexives/pronouns are in principle allowed to move, why is it that X° reflexives can move successive-cyclically by adjoining to Infl's, whereas movement of XP reflexives must be strictly local? Notice that head-movement is generally more restricted than XP movement. Why aren't movement constraints obeyed in the construal of LD reflexives, as observed in Huang and Tang (1989))? The following example involves a reflexive which can be bound across a relative clause island (and across a specific NP): (72)

Zhangsa^ bu xihuan [neixie piping ziji-^ de renj] not like those criticize self REL person 'Zhangsan does not like those people who criticize self.'

Binding theory

17

The same type of LF head movement in Chinese, illustrated in A-not-A questions, does obey the relative clause island (see Huang (1982)): (73)

*Ni zui xihuan ta mai-bu-mai de shu? you most like he buy-not-buy REL book

Also, why are binding domains for X° reflexives wider in languages without overt agreement (e.g., Chinese) than in languages with overt agreement? Furthermore, even though a reflexive may move outside of the c-command domain of an object, its trace will still be in the scope of the object. Why isn't it then possible for the object to bind the reflexive by binding its trace? With XP reflexives, neither S-structure nor LFmovement bleeds binding possibilities: (74) (75)

Which picture of himself^ does [IP Johni think [CP t [IP Peterj has seen t]]]? Whoi said that Billj wants Peterk to buy [which picture of himselfim]7

In sum, I will advocate an analysis of reflexives which both preserves the original and simple formulation of the Binding Principle A, and derives the differences between complex and simple reflexives from their morphological properties without stipulating LF movement. 0.2.2. Principle B

The main insight of Principle B is that items subject to it must be disjoint in a certain domain, as in the standard formulation in Chomsky (1981): (76)

A pronoun must be free in its governing category.

While it may be controversial how to define the domain, I believe that the case for the necessity of Principle B is strong. Contrary to some recent attempts to abolish Principle B (Burzio (1990, 1991, forthcoming)), I will assume that this part of Principle B must remain a primitive, at least at this stage of our knowledge. To the extent to which the binding analysis of positive polarity items proves adequate, we will have indirect evidence for the necessity of Principle B. Strong support for Principle B also comes from mixed categories, the so-called anaphoric pronominals (such as Norwegian seg). I will also show that such mixed categories exist in the polarity domain as well: Serbian/Croatian I-NPIs are completely parallel to anaphoric pronominals in that they must be free in a local domain, but

18 Introduction bound in a wider domain. It is hard to see how one can account for the distribution of anaphoric pronominals without recourse to Principle B. Given the original formulation in (76), the governing category for a pronoun coincides with that of an anaphor, and contains the pronoun, the governor for the pronoun, and a SUBJECT accessible to the pronoun. This characterization of the governing category has been revised in different ways, for example, to exclude the accessibility requirement (Huang (1983)), or to exclude SUBJECT (Chomsky (1986b)). The exclusion of the accessibility requirement is a desirable outcome, and as shown in section 0.2.1, it can also be achieved for reflexives. But I will assume that we still have to operate with the notions of SUBJECT (especially when it comes to anti-subject oriented pronouns) and governor (especially when it comes to domain extension with subject pronouns in subjunctive clauses, which, on the assumption that governor plays a role, reduces to the ECM cases, as discussed in chapter 6.). Some LF-movement analyses claim to be doing away with SUBJECTS by raising pronouns either up to Infl, or up to the specifier of a governor (see Hestvik (1990, 1991)). Notice, however, that their claim is just a terminological variant: stipulating the landing site for movement is equivalent to stipulating the SUBJECT. In addition, the stipulation that pronouns move to the Spec of their governor obviously reestablishes the relevance of the governor. Thus all the ingredients of the binding principles remain, although in disguise. 0.2.3. Generalized binding Allowing AGR to be a binder commits one to the view that A'-binding is possible. This is in the spirit of Aoun's (1985a, 1986) extension of binding to include both A and A'-binding. Notice that the N value of an NP should be an A'-, rather than an A-element, on the simplest possible assumption that argument positions can only be the ones in which argument NPs are generated. If N, under which a simple reflexive is generated, is a non-argument, it is no surprise that it gets bound to a nonargument, Agreement. Aoun's generalized binding analysis is primarily designed to subsume the ECP effects under the binding theory. For this purpose, wh-traces are treated as (non-overt) A'-anaphors, which have to be A'-bound in their governing category. The following are the generalized binding principles (from Aoun (1985)):

Overview 19 (77) (78) (79) (80)

(81)

Principle A: An anaphor must be X-bound in its governing category. Principle B: A pronominal must be X-free in its governing category. Principle C: A name must be A-free. Y is a governing category for X if and only if Y is the minimal maximal category containing X, a governor of X, and a SUBJECT accessible to X. X is accessible to Y if and only if Y is in the c-command domain of X and coindexing of (X,Y) would not violate any grammatical principle.

Apart from i-within-i condition, which Chomsky (1981) takes as relevant for accessibility, Aoun also makes use of Principle C. Wh-traces are subject both to Principle C, as is commonly held, and to the generalized version of Principle A. Subject extraction will always be local since AGR will act as a SUBJECT for the trace. Object trace, on the contrary, will have no accessible SUBJECT, and therefore no domain, because neither [NP,IP] nor AGR are accessible to an object trace. This is because coindexing of an object trace to either AGR or subject will result in a Principle C violation, where the wh-trace gets A-bound to the subject. Thus, the following subject/object asymmetry in wh-movement is derived from the Binding Theory: (82) (83)

*Who did John say that t adores Mary? Who did John say that Mary adores t?

In this monograph, I will only make use of the basic insight of Generalized Binding - that A'-binding is possible. This move is essential for treating polarity within the Binding Theory, given that negation and truth-conditional operators act as binders for negative and positive polarity. Generalized binding is an instance of the relativization of the notion 'binder,' along the A/A'-dimension. I rely on such relativization in two ways: first, I assume that NPIs require an A'-binder and SUBJECT: second, I relativize the notion of binder and SUBJECT for reflexives along the X-bar compatibility dimension (see section 0.2.1). Relativization renders it possible to keep binding principles maximally simple, while making binders compatible to the bound elements.

20

Introduction

0.3. Overview

The monograph is organized in the following way: In Chapter 1 I discuss Serbian/Croatian polarity items in negative contexts. There are two types of NPIs: NI-NPIs and I-NPIs. NI-NPIs (beginning with the negative prefix ni) can appear only with clausemate negation. Therefore, NI-NPIs are parallel to English reflexives in that they have to be bound in their local clause: (84)

Johnj believes [that Peterj admires himself*^.]

I-NPIs (beginning with the prefix /) are licit in all polarity contexts except clausemate negation. I-NPIs will be argued to parallel anaphoric pronominals in that they obey Principle B, in addition to being subject to the general binding requirement in the sentence. As for PPIs, they cannot appear in the scope of a clausemate negation, although they can be bound by superordinate negation. In this respect they parallel pronouns, which have to be free from a clausemate antecedent: (85)

Johnj believes [that Peterj admires himi/*j.]

In Chapter 2, I extend the binding analysis to English negative contexts. Any is also subject to Principle A of the Binding Theory. The reason why any in English is licensed by both clausemate and superordinate negation follows from the assumption that any can raise at LF, the way quantifiers have been argued to raise (see May (1977, 1985)). Independent motivation is offered for the LF raising of any. they obey typical movement constraints such as the ECP, Subjacency Condition, and the Specificity Condition. In Chapter 3, I establish that, in both English and Serbian/Croatian, licensing in non-negative contexts (i.e., questions, conditionals, etc.) patterns with licensing by superordinate negation, rather than by clausemate negation. I propose that a null polarity operator appears in the Comp of all these environments, and that it falls outside of the governing category for polarity items since the first potential antecedent for polarity items is Infl. The existence of an operator in this position is motivated on the basis of two pieces of independent evidence: inversion and epistemic modals (see McDowell (1987)). This approach enables us to capture the following surprising contrast: (86) (87)

I forgot [CP Op that anyone dropped by]. *I forgot anything.

Overview

21

The NPI is licensed in (86) because the complement of forget is sentential, and therefore hosts a Comp position. On the other hand, in (87) there is no Comp position for Op selection. In Chapter 4, I consider NPI licensing in different languages and propose two factors whose interplay can account for wide cross-linguistic variation: (i) being subject to different binding principles and (ii) raising options. Some NPIs are subject to Principle A only (e.g., Serbian/ Croatian NI-NPIs and English NPIs) while others are subject to both Principle B and a general binding requirement in the sentence (Serbian/ Croatian I-NPIs). In addition, some NPIs do not raise at LF at all (Serbian/Croatian NI-NPIs), whereas others do. While English NPIs can raise by either IP-adjunction or through the Spec of CP, some NPIs are argued to raise only through the Spec of CP (Turkish NPIs) or only via IP-adjunction (Catalan and Chinese NPIs). This accounts for the fact that English NPIs are licit in all the polarity contexts, whereas Chinese NPIs appear in all but superordinate negation contexts, and Turkish NPIs are licensed only in negative contexts, either by superordinate or clausemate negation. In Chapter 5, it is argued that the rhetorical force in wh-questions containing NPIs: (88)

When did Mary ever insult anyone*?

can be derived by Spec/Head AGR in CP (see Chomsky (1986a) and Rizzi (1990a)). Since a wh-word in the Spec of CP triggers wh-AGR in the head of CP, Op cannot appear in this position. In order for the question to be salvaged the wh-word must lose its wh-force by suppressing wh-AGR. The only other option for the wh-word to remain licit in Comp would be for it to become a negative expression. This is only possible by Spec/Head AGR if Op switches to its negative value, thus giving rise to a negative rhetorical interpretation. Chapter 6 attempts a unified explanation for domain extensions in subjunctive clauses with negative polarity, reflexives, and pronouns. It is proposed there that subjunctive clauses, having anaphoric (or dependent) tense, are characterized by LF deletion of functional categories (Infl and Comp) in LF. If both potential antecedents for negative polarity are absent in LF, the domain for NPIs extends up to the first indicative Infl. The domain for simple reflexives likewise extends across subjunctive clauses because their only SUBJECT, AGR, will be missing in LF. As for

22 Introduction pronouns, they will extend their domain only when in the subject position of a subjunctive clause. This is because in this position pronouns have no governor in LF, where Infl and Comp are deleted, and their next available governor will be the matrix verb. This way, pronoun obviation in subjunctive clauses is completely parallel to the ECM obviation at Sstructure. Chapter 7 discusses Free-Choice Items = FCIs {any in English and bilo in Serbian/Croatian) which are generally assumed to be licensed by modals and to receive wide-scope universal interpretation (e.g., Ladusaw (1980), Linebarger (1981), Carlson (1981)): (89)

Anyone can beat a child.

On the basis of Serbian/Croatian, which differentiates lexically between FCIs and NPIs {bilo (t)ko vs. i(t)ko), it will be shown that the distribution of FCIs is much wider than assumed for any in English. In particular, bilo appears in all the polarity contexts, in which it receives the same interpretation as an NPI in its place would. More precisely, bilo receives wide-scope interpretation only if the trigger is local (i.e., clausemate negation or modal). If the trigger is distant (e.g., superordinate negation or the polarity operator in Comp), bilo receives narrow-scope existential interpretation. I argue that these facts follow from the assumption that FCIs are subject to Principle B: they can be bound by a distant, but not by a local, binder. Finally, in Chapter 8 I discuss two leading approaches to polarity sensitivity: Ladusaw's and Linebarger's. Ladusaw (1980) proposes that an NPI will be triggered if and only if in the scope of a downwardentailing element, whereas a PPI will be anti-triggered by a clausemate negation. Linebarger (1981) argues that NPIs are licensed either if in the immediate scope of negation at LF, or if they appear in sentences with a negative implicature. The binding approach to locality conditions in NPI licensing has the following general advantages: it offers an account of cross-linguistic variation; it relies on the already established principles of the binding theory, requiring no extra mechanisms; and it can capture naturally our intuitive feeling that there is a symmetric relation between NPIs and PPIs. In addition, it offers a simple explanation for the otherwise puzzling fact that non-negative polarity licensing must proceed through Comp, as pointed out and illustrated in section 0.1.2.

1 Negative contexts: Serbian/ Croatian

Serbian/Croatian offers good insight into the nature of polarity sensitivity because it has two types of Negative Polarity Items (NPIs). A need for establishing locality restrictions on polarity sensitivity emerges clearly in Serbian/Croatian. I will argue that the best way to capture these locality conditions is to invoke the principles of the Binding Theory. This analysis will be extended to English in Chapter 2, and to other languages in Chapter 4. The main appeal of the binding approach is in its ability to capture locality conditions imposed on polarity items by invoking already established and otherwise needed principles. Section 1.1. is a brief introduction to the main issues in Serbian/ Croatian syntax, while Section 1.2. covers the basic facts about negative polarity in Serbian/Croatian.

1.1. Issues in Serbian/Croatian syntax

This section provides some background information on the syntax of Serbian/Croatian, especially in cases where it departs from English, and where the departure is theoretically interesting or relevant for the discussion in subsequent chapters.

1.1.1. Word order: full NPs versus clitics

The unmarked word order in Serbian/Croatian is essentially the same as in English, SVO: (1)

Stefan crta An-u. Stefan draws Ana-ACC 'Stefan is drawing Ana.' 23

24 Negative contexts: Serbian I Croatian (2)

Ana daje cokolad-u Stefan-u. Ana gives chocolate-ACC Stefan-DAT 'Ana is giving chocolate to Stefan.'

For emphasis, however, constituents can be freely scrambled, and appear in any order: (3) (4) (5) (6) (7)

Crta Stefan Anu. Crta Anu Stefan. Anu crta Stefan. Anu Stefan crta. Stefan Anu crta.

(8) (9) (10) (11) (12) (13) (14) (15) (16) (17) (18) (19) (20) (21) (22) (23) (24) (25) (26) (27) (28) (29) (30)

Ana daje Stefanu cokoladu. Ana Stefanu daje cokoladu. Ana Stefanu cokoladu daje. Ana cokoladu daje Stefanu. Ana cokoladu Stefanu daje. Daje Ana cokoladu Stefanu. Daje Ana Stefanu cokoladu. Daje cokoladu Ana Stefanu. Daje cokoladu Stefanu Ana. Daje Stefanu Ana cokoladu. Daje Stefanu cokoladu Ana. Stefanu Ana daje cokoladu. Stefanu Ana cokoladu daje. Stefanu daje Ana cokoladu. Stefanu daje cokoladu Ana. Stefanu cokoladu daje Ana. Stefanu cokoladu Ana daje. Cokoladu daje Ana Stefanu. Cokoladu daje Stefanu Ana. Cokoladu Stefanu Ana daje. Cokoladu Stefanu daje Ana. Cokoladu Ana daje Stefanu. Cokoladu Ana Stefanu daje.

Word order freedom correlates with rich inflectional morphology in Serbian/Croatian. As evident in examples (1) and (2), accusative and

Issues in Serbian/'Croatian syntax

25

dative NPs are marked with distinct Case endings, so that a fixed word order is not necessary to determine the grammatical function of an NP. Clitics are the only NPs that show fixed word order (see Browne (1974) for a detailed description of the clitic system in Serbian/Croatian). Although direct (DO) and indirect objects (IO) can freely permute if full NPs, as evident in the above examples, the DO must follow the IO if they are cliticized: (31) (32) (33) (34)

Ana joj ga daje. Ana her it gives. *Ana ga joj daje. Ana raw ga daje. Ana him it-MASC gives *Ana ga mu daje.

Traditionally, clitics in Serbian/Croatian have been claimed to have two possible landing sites: either the second constituent position, or the second word position; i.e., clitics appear either after the first constituent, or after the first word (Browne (1974), Comrie (1981)): (35)

(36)

Anina drugarica raw nudi cokoladu. Ana's girlfriend him offers chocolate 'Ana's friend is offering him chocolate.' Anina raw drugarica nudi cokoladu. Ana's him girlfriend offers chocolate

I will assume that clitics in Serbian/Croatian adjoin to the head of CP, Comp, given that they always immediately follow a complementizer in subordinate clauses, and that they can stack up: (37)

Stefan tvrdi [da raw ga je Petar poklonio]. Stefan claims that him it is Peter given 'Stefan claims that Peter has given it to him as a present.'

In a matrix clause, when there is no complementizer, clitics in Comp cannot remain clause-initial, since they are phonologically dependent, and need a host. One possibility for clitics to satisfy this requirement is to lower down and attach to the right hand side boundary of the first constituent or word. I will adopt another alternative, suggested to me by Marissa Rivero (personal communication), that clitics stay in Comp (or wherever their ultimate landing site turns out to be), and that support for clitics is achieved by moving constituents to the Spec of CP. Thus in (35),

26 Negative contexts: Serbian I Croatian Anina drugarica is in the Spec of CP, and the clitic is adjoined to Comp. The phenomenon of clitic placement is thus reminiscent of verb-second in Germanic, where a verb in Comp is always preceded by a constituent in the Spec of CP. A wh-phrase in the Spec of CP can support clitics, which is also reminiscent of verb-second: (38)

(T)Koga je Milan zbunio? whom is Milan confused 'Who did Milan confuse?'

It remains to account for (36). It has often been claimed that wordsecond clitic placement in Serbian/Croatian violates the most basic principle of Universal Grammar-constituent dependency, see e.g. Comrie (1981). This is because in examples like (36) the clitic is placed after the first word, which does not look like a constituent. This is only at first sight, though. Both possessive and other adjectives behave as constituents in Serbian/Croatian in that they can be displaced from the head noun, and wh-questioned: (39)

(40)

Anina dolazi sestra. Ana's comes sister 'Ana's sister is coming.' Cija dolazi sestra? whose comes sister 'Whose sister is coming?'

Thus, it is plausible to assume that in (36) "Anina" is a constituent and that it moves to the Spec of CP to support the clitic in Comp. Thus the difference between (35) and (36) reduces to the difference between preposing a full NP versus preposing the possessive specifier of the NP. This way we can avoid the conclusion that clitics in Serbian/Croatian either follow the first constituent or the first word. More evidence for the claim that clitic placement is not blind to the structure comes from the fact that clitics cannot be placed after any old word in a sentence. For example, they cannot be hosted by a preposition: (41)

Na sto ga ostavi. on table it leave 'Leave it on the table.'

Issues in Serbian!Croatian syntax (42)

27

*Na ga sto ostavi. on it table leave

This would follow from the independent fact that prepositions can neither be displaced from the complement NPs, nor questioned. This way, they cannot move to the Spec of CP to support the clitic in Comp. Notice that it would not help to argue that prepositions cannot host clitics because they do not receive stress. It is true that in normal speech prepositions are not stressed, but neither is the complementizer da, which can otherwise support clitics (see (37) above). Moreover, prepositions can be stressed for emphasis or contrast, but this does not influence their ability to host clitics: (43)

*Prema je Milanu Marija koracala, a ne od njega. toward is Milan-DAT Mary walked and not from he-GEN 'Mary walked toward Milan, not away from him.'

Therefore, it seems safe to conclude that clitics in main Serbian/ Croatian clauses are supported by preposing a constituent to the Spec of CP, provided that the constituent can move in principle. No such movement is necessary in subordinate clauses because the subordinate complementizer can act as a host for the clitic (see (37)). This main/ subordinate clause asymmetry is again reminiscent of verb-second in Germanic. Notice that a wh-phrase in the specifier of a subordinate CP can host clitics: (44)

Milan se pita (t)koga ce Marija da poslusa. Milan self asks whom will Mary that listens-to 'Milan wonders who Mary will listen to.'

In sum, the basic (underlying) word order in Serbian/ Croatian is SYO, but constituents can freely invert, except for clitics, which observe a rigid order. 1.1.2. Wh-movement Wh-phrases in Serbian/Croatian are fronted in syntax: (45) (46)

(T)Ko t voli Mariju? who loves Mary (T)Koga Marija voli t? whom Mary loves

28 Negative contexts: Serbian I Croatian Long-distance Wh-movement in Serbian/Croatian shows a subtle Subject/Object asymmetry, which has often escaped notice (see Browne (1986)). Although both (47) and (48) sound acceptable, it is only because subjectless "mislis da" is inserted, almost as a kind of parenthetical: (47)

(48)

?(T)Ko misli-s [da t voli Mariju]? Who think-2SG that loves Mary *'Who do you think that loves Mary?' (T)Koga mislis [da Marija voli t]? Whom think-2SG that Mary loves 'Who do you think that Mary loves?'

As soon as the matrix clause is expanded, as below, the asymmetry sharpens, for which fact I have no explanation: Questions: (49)

(50)

?*(T)Ko Milan misli [da t voli Mariju]? who Milan thinks that loves Mary *4Who does Milan think that loves Mary?' (T)Koga Milan misli [da Marija voli t]? Whom Milan thinks that Mary loves 'Who does Milan think that Mary loves?'

Relative clauses: (51) (52)

?*Covek koji Milan misli [da e Voli Mariju]... man who Milan thinks that loves Mary Covek (t)koga Milan misli [da Marija voli]... man whom Milan thinks that Mary loves

The language also employs a non-movement, resumptive pronoun strategy, which obliterates the subject/object asymmetry: Questions: (53) (54)

Za (t) koga for whom Za (t)koga for whom

Milan Milan Milan Milan

misli [da e voli Mariju]? thinks that loves Mary misli [da ga Marija voli]? thinks that him Mary loves

Issues in Serbian I Croatian syntax 29 Relative clauses: (55) (56)

Covek man Covek man

za (t)koga for whom za (t)koga for whom

Milan Milan Milan Milan

misli [da e voli Mariju] ... thinks that loves Mary misli [da ga Marija voli]... thinks that him Mary loves

Here, the wh-word is case marked by the initial preposition za suggesting that movement is not involved in such constructions (see Goodluck (1992)). If it were, the wh-chain would receive case twice, in violation of the Case Filter. There is a strong reason to believe that the resumptive pronoun (or pro) is linked (or bound) to the wh-phrase in Comp. "Za (t)koga" strategy is never available for local extraction, presumably as a result of Principle B violation: (57)

(58)

*Za (t)kogaj proj voli Mariju? for whom loves Mary 'Who loves Mary?' *Za (t)kogai gaA Marija voli? for whom him Mary loves 'Who does Mary love?'

One may be tempted to say here that "za (t)koga" strategy is only used as a last resort, for example, when long-distance extraction results in a violation. This would explain why it is not available for local extraction. But this is clearly not the case since the resumptive strategy is also used for long- distance extraction of objects, which yields a perfect result with movement (see (54)). A (resumptive) pronoun is obligatory with object questioning (see (54)), in case the object is animate. It is prohibited, however, in the subject position: (59)

*Za (t)koga Milan misli [da on voli Mariju]? for whorrii Milan thinks that hei loves Mary

The empty category in the subject position is most probably pro, as argued for in Browne (1986). One may wonder why it is not possible to have an overt pronoun in the case of questioning the subject, given that pro-drop is generally optional. Actually, pro-drop in Serbian/Croatian becomes obligatory if a pronoun is bound to an A'-operator, such as a quantifier or a wh-word (see section 1.1.3). The fact that "za (t)koga"

30 Negative contexts: Serbian/'Croatian patterns with these cases suggests that it occupies an A'-position, most probably Spec of CP position. Since it is not derived by movement, I conclude that it is generated in this position. A non-movement analysis is strengthened by the fact that islands are not obeyed in case of "za (t)koga" constructions, as opposed to regular extraction (see Goodluck (1992) for the parallel contrasts with relative clauses). Complex-NP Constraint: (60)

(61)

IZa (t)kogat je Milan cuo [NP pricu [CP da gai Marija voli]]? for whom is Milan heard story that him Mary loves 'About whom did Milan hear the story that Mary loves him?' *(T)Kogai je Milan cuo [NP pricu [CP da Marja voli tj]? whom is Milan heard story that Mary loves 'About whom did Milan hear the story that Mary loves him?'

Wh-Islands: (62)

(63)

Za (t)kogat se Milan pita [CP zasto gaA Marija voli]? for whom self Milan asks why him Mary loves 'About whom does Milan wonder why Mary loves him?' *(T)Kogat se Milan pita [CP zasto Marija voli tj? whom self Milan asks why Mary loves ?*'Whom does Milan wonder why Mary loves?'

Multiple wh-questions are formed in Serbian/Croatian by moving all whwords in syntax. This seems to relate to the possible absence of LF movement in the language (see section 1.1.4). 1.1.3. Pro-drop As a pro-drop language, Serbian/Croatian optionally allows null subjects: (64)

(Stefan) svira violin-u. Stefan plays violin-ACC

Pro-drop becomes obligatory if a pronoun is bound to an A'- operator, such as a quantifier or a wh-word (see Xu (1984), Montalbetti (1984) and Aoun (1985b) for comparable facts in Chinese and Spanish):

Issues in Serbian I Croatian syntax 31 (65)

(66)

(67)

Svakoi misli [da je (*onO pametan]. everyone thinks that is he smart 'Everyonei thinks that hei is smart.' Ne(t)koi misli [da je (*oni) pametan]. someone thinks that is he smart 'Someonej thinks that hei is smart.' (T)KOi misli [da je (*onO pametan]? who thinks that is he smart j thinks that hej is smart?'

1.1.4. No LF movement in Serbian/Croatian? As far as I can tell, there is no evidence for LF movement in Serbian/ Croatian. Multiple wh-questions are formed by multiple syntactic extraction, and quantifier ambiguities are resolved at S-structure by fronting the wide-scope quantifier, possibly by scrambling. 1.1.4.1. Quantifier scope Consider the following examples, each involving an existential and a universal quantifier: (68)

(69)

Ne(t)ko voli svako-ga. someone loves everyone-ACC 'There is a person X such that X loves everybody.' Svako voli ne(t)ko-ga. everybody loves someone-ACC 'For every X, there is a Y, such that X loves Y.'

To my judgment, and to the judgments of my informants, only the widescope reading of the existential quantifier is possible in (68), and only the wide-scope reading of the universal quantifier in (69). However, if objects are fronted the judgments are reversed: (70)

(71)

Svako-ga ne(t)ko voli. everyone-ACC someone loves 'For every X, there is a Y, such that Y loves X.' ??Ne(t)ko-ga svako voli. someone-ACC everyone loves 'There is an X such that everyone loves X.'

32 Negative contexts: Serbian I Croatian In (70), the universal quantifier takes wide scope, whereas in (71), to the extent that it is acceptable at all, the fronted existential quantifier takes wide scope. It seems, therefore, that scope possibilities are read off the surface structure since the preceding quantifier, whether base-generated or preposed, must take scope over the following one. On a very speculative note, this may follow from the assumption that Serbian/Croatian does not exploit LF movement. In order to take wider scope, quantifiers in object positions must prepose. This speculative conclusion seems strengthened by the fact that multiple questions are also formed by overt extraction, as discussed in the following section. 1.1.4.2. Multiple wh-extr action: Multiple questions in Serbian/Croatian are formed by overtly fronting all the wh-phrases: (72)

(73)

(74)

(T)Ko (t)ko-ga slusa? who who-m listens-to 'Who obeys whom?' Gde (t)ko spava? where who sleeps 'Who sleeps where?' Sta (t)ko (t)ko-me nudi? what who who-DAT offers 'Who offers what to whom?'

Rudin (1988a, 1988b) argues that the first wh-word is always in the Spec of CP, whereas the others IP adjoin. Thus the (72) would be represented as (75) below: (75)

CP

(t)kOj

Jc^

c

n>

(t)kogaj

IP tj

1'

i

yp V slusa

NP

Negative polarity in Serbian I Croatian 33 If wh-words are left in situ, they are interpreted as d-linked, in Pesetsky's (1987) terms: (76) (T)Ko vodi (t)ko-ga za ruku? who takes who-m by hand One would utter (76) in case the wh-phrases are already linked in discourse (d-linked). For example, there is a group of children established in the discourse, and we only need to determine who is going to hold whose hand. Pesetsky argues that such d-linked wh-phrases need not raise in LF in languages where LF movement otherwise takes place. If this is true, it would also explain why such d-linked wh-phrases can stay in situ in Serbian/Croatian. On the basis of the similar behavior of Polish wh-phrases, Pesetsky (1987) concludes that there is no LF wh-movement in Polish (see also Wachowicz (1974)). Thus, we are left with the following situation in Serbian/Croatian: whenever movement is necessary for scope purposes, it takes place at SS. Quantifiers in object positions must prepose at SS to take wider scope with respect to subject quantifiers. Wh-phrases in multiple questions all move to the front of the sentence to take wh-scope. This leaves us with the possibility that LF movement, as a UG option, is not exploited in Serbian/Croatian. This may correlate with the fact that overt scrambling can obviate the necessity for LF movement by placing arguments in the desired scope positions already at S-structure. 1.2. Negative polarity in Serbian/Croatian 1.2.1. Placement of negation In English, modals and Tense necessarily precede negation, suggesting that negation is below Infl: (77) I cannot do that. (78) Peter doesn't understand that. One possibility is to assume with Pollock (1989) that negation is in the separate NegP projection: (79)

I Tense/Agr

34

Negative contexts: Serbian I Croatian

If finite forms are under I, this would account for the ordering possibilities of negation in English with respect to modals and Tense constituents, which negation necessarily follows (but see Kayne (1989) for the discussion of contracted modal forms). In (79) I depart from Pollock's suggestion that Infl splits into Tense and Agreement projections, but rather continue to assume that these are features on Infl (see also Iatridou (1990) for arguments against splitting Infl). The move to grant a separate projection to negation, but not to Agreement or Tense is not as arbitrary as it may seem. Negative particles are words, rather than morphemes, and show more distributional freedom. More importantly, Tense and Agreement are mutually exclusive in English (as well as with modal verbs), suggesting that they might all occupy one and the same position. This is not the case with negation, however, which can freely co-occur with any of these categories. In addition, my analysis of subjunctive follows more coherently if Agreement and Tense are treated as features on Infl (see Chapter 6.). Alternately, one might view the negative particle not as an adverb, generated somewhere within VP, for example VP-adjoined, as suggested in Hasegawa (1987) and Pollock (1987). This view would easily capture the fact that not can be generated as a constituent adjunct in the positions where one would not want to postulate a NegP, e.g: (80) I would not like to leave, not at all. (81) Not everybody will attend the lecture. I will not try to resolve these two options, but will tentatively adopt the latter view for two reasons: first, it can capture the difference in negation placement in English and Serbian/Croatian; and, second, it might help explain why NPIs are long-distance bound in English without invoking movement in LF (see section 2.2. for discussion). Negation in Serbian/Croatian is realized as a negative particle ne, which cliticizes to the left of the first finite verb form, whether an auxiliary or main verb:1 (82)

(83)

Milan nece pobeci. Milan not-will run-away 'Milan will not run away.' *Milan ce ne pobeci.

Negative polarity in Serbian/'Croatian 35 (84)

(85)

Milan ne poznaje Mariju. Milan not knows Mary 'Milan does not know Mary.' *Milan poznaje ne Mariju.

This suggests that, at least at S-structure, negation in Serbian/ Croatian is either in Infl, or above Infl. In the former case, we could assume that it originates in a NegP below Infl, and then moves with the verb to Infl, as suggested to me by a reviewer. In the latter case, a NegP would be generated above Infl. Different placement of negation in English and Serbian/Croatian seems to correlate with the possibility to license a clausemate subject NPI. Compare the following contrast between the two languages: (86) (87)

* Anyone does not know Mary. Niko ne poznaje Marij-u. no-one not knows Mary-ACC 'No one knows Mary.'

Although in English the subject NPI is not licensed by clausemate negation, in Serbian/Croatian it is: there is no subject/object asymmetry with respect to NI-NPI licensing (see also section 3.2.3). Adopting the definition of c-command proposed in Aoun and Sportiche (1983) in terms of the first maximal projection (m-command), we predict that the subject position will be c-commanded by negation in Serbian/Croatian, but not in English.2 Apart from Serbian/Croatian, there are other languages in which clausemate negation licenses subject NPIs, for example, Turkish (Enq (1988)), Catalan (Casanova-Seuma (1988)), and Japanese (Hasegawa (1987)). Turkish: (88) Hiq kimse Ali-yi gor-me-di. any person Ali-ACC see-not-PAST 'No one saw Ali.' Catalan: (89) Ningu no ha vingut. nobody not has come 'Nobody has come.'

36

Negative contexts: Serbian I Croatian

Japanese: (90) Daremo ko-nakkatta anyone came-neg(Past) 'No one came.1 On the other hand, there are also other languages like English whose clausemate negation does not license NPIs, e.g., Chinese (Li (1985)) and KiNande (Progovac (in press)): Chinese: (91) #Shei meiyou kanjian Zhangasan?3 who did-not see John 'Nobody saw John.' KiNande: (92) *Mukali si anzire Yohani. any-woman not likes John 'No woman likes John.' In the light of this cross-linguistic variation, it appears necessary to distinguish between different placement options for negation. 1.2.2. Negative concord

Another difference between Serbian/Croatian and English should be noted. In Serbian/Croatian, every instance of sentential negation has to be realized through the negative particle ne. In other words, Serbian/ Croatian does not have the option, available in English, to express sentential negation by negating a constituent only: (93) (94)

Mary will see nobody. *Marija ce videti niko-ga. Mary will see no-one-ACC

Standard English contrasts with Serbian/Croatian not only in that it can express sentential negation by negating a constituent, but also in that it does not tolerate negative concord (multiple negation): (95)

*John did not see nobody.

Serbian/Croatian also differs from Romance languages, which seem to fall midway between English and Serbian/Croatian. While in Romance languages negative concord (multiple negation) is necessary only with

Negative polarity in Serbian I Croatian

37

post-verbal negative polarity items, in Serbian/Croatian it must affect both pre-verbal and post-verbal NPIs. Examples (96) and (97) from Zanuttini (1991) show the subject/object asymmetry with Italian negative concord, while (98) and (99) illustrate the absence of such asymmetry in Serbian/Croatian: (96)

(97)

(98)

(99)

Mario *(non) ha visto nessuno. Mario neg has seen nobody 'Mario did not see anybody.' Nessuno (*non) ha visto Mario, nobody neg has seen Mario 'Nobody saw Mario.' Mario *(ne) vidi ni(t)ko-ga. Mario neg sees nobody-ACC 'Mario cannot see anyone.' Ni(t)ko *(ne) vidi Mari-ja. Nobody neg sees Mario-ACC 'Nobody can see Mario.'

Zanuttini (1991) concludes that nessuno is not a real NPI, since it can remain unlicensed in the pre-verbal position. Negation is required with post-verbal NPIs since sentential negation must be expressed from a position not lower than Infl. Even if these two conclusions are relevant for Italian, they cannot be correct for Serbian/Croatian, since every instance of an NPI must be licensed by the negative particle. This is also true of Russian, Czech, and other languages (see section 0.1.1.). I will return to the Italian data in Chapter 6 and argue that the conclusions are not necessary for Italian either. While I agree with Zanuttini that negated NPIs are different from the non-negated ones, I disagree with her that this difference is to be captured by saying that the former are not polarity items. Here I agree with Laka (1990) that n-words across languages (see nessuno in Italian and ninguno in Spanish) can all be analyzed as negative polarity items. I will argue that their partial independence follows from the fact that they host a negative morpheme, capable of triggering negative Spec/Head agreement if in a specifier position. The triggered negation can then license the polarity item (see section 6.2.1. for details). For obvious reasons, NPIs which show no morphological negation (e.g., alcuno in Italian) do not have this option. Likewise, object NPIs, even if negated, cannot trigger negative AGR, assuming that they are in a complement, rather than a specifier

38

Negative contexts: Serbian/Croatian

position. The difference between Serbian/Croatian and Italian might then follow from a stricter requirement in the former language that NPIs be licensed by overt negation. Zanuttini's (1991) basic claim is that negated items in Romance are negative quantifiers which need to be interpreted in the NegP projection. For this reason, they have to raise to the Spec of a NegP. No matter how many such quantifiers raise to NegP, and whether an overt negative marker is present or not, the meaning of the sentence will be that of a single negation, due to Spec/Head agreement in NegP. The well-known subject/object asymmetry in Romance, illustrated above for Italian, is captured by stipulating that NegP selects TP, and that a negative quantifier cannot cross a TP unless it is L-marked by negation, and thus debarrierized. This makes the negative particle obligatory only in case of negative quantifiers being extracted from the object position (see the contrast between (96) and (97)). This explanation for negative concord raises the following questions. First of all, it allows for more than one negative quantifier to appear in the Spec position of NegP, involving a multiply-filled position. Zanuttini assumes that negative quantifier movement is similar to the LF movement of wh-in-situ, which, presumably, also multiply-fills the Spec of CP position. However, recent arguments point to an alternative analysis of wh-in-situ (see Aoun and Li (1991), Watanabe (1991), Chomsky (1992)), where it is argued that wh-in-situ does not move by itself, but that there is an overt movement of a null operator to the head of Comp, which is then responsible for certain island effects. Chomsky (1992) concedes that the interpretation of wh-in-situ is done, not through movement in LF, but rather through absorption associating two whphrases to form a generalized quantifier (see Higginbotham and May (1981)). If such an analysis is preferable for wh-in-situ, it is even more so for negative quantifiers which have traditionally been treated as NPIs. In their in-situ position they will be bound (or absorbed) with the negative particle to produce only one instance of negation, i.e., the reading of negative concord. Indeed, a more constrained grammar would be the one which does not allow for doubly-filled positions at any level. Laka's (1990) analysis raises the same problem of doubly-filling the position. She argues that NPIs must move to the Spec of Sigma-projection in LF in order to be interpreted. Notice that negative words move to Spec of NegP only in LF. This means that there is nothing in the head of NegP at S-structure, and the

Negative polarity in Serbiani'Croatian

39

NegP projection is not there at all. In that case it is not clear what marks the landing site for the movement of n-words in LF. Notice that even if the LF wh-movement analogue were to be invoked here, the parallel breaks down: wh-movement requires an S-structure scope marker or place-holder. Furthermore, saying that nessuno is a negative quantifier commits one to the view that it will express negation no matter what the clause-type, if in a pre-Infl position. However, as discussed by Zanuttini, nessuno in questions and conditionals does not have such force, but is interpreted as any in English: (100) Ha telefonato nessuno! has called nobody 'Has anybody called?' In other words, if nessuno is a negative quantifier, it remains unclear how to differentiate it from English no one, which always carries negation with it: (101) Did nobody call? It is also not clear how an object n-word can cross a TP boundary in questions and conditionals, in order to raise to Spec of CP, without a negation to L-mark it. Recall that the claim that a TP boundary cannot be crossed unless it is L-marked by negation was the main argument for obligatory negation with object n-words (see the famous contrast between (96) and (97)). Zanuttini's strongest argument for claiming that n-words are not NPIs comes from examples of obviously unlicensed negative words in Italian: (102) Ha detto cio con nessuna malizia. he said so with no malice However, the argument weakens in the light of the fact that even obvious NPIs, such as English ever, can remain unlicensed when in non-argument positions: (103) the ever-smiling countenance; ever-lasting love; ever-changing seasons (104) They lived happily ever after.

40

Negative contexts: Serbian I Croatian

Although the phenomenon seems more restricted with ever, the fact that it is possible at all casts doubt on the necessity to abandon a polarity analysis of negative concord. 1.2.3. Distribution of polarity items This section describes the distribution of Serbian/Croatian NPIs and PPIs in negative contexts. There are two types of NPIs: NI-NPIs (beginning with the prefix ni) and I-NPIs (beginning with the prefix /). In the scope of negation, NI-NPIs and I-NPIs are in complementary distribution: while the former co-occur only with clausemate negation, the latter are licensed only by superordinate negation. 4 PPIs, on the other hand, are disallowed in the scope of a clausemate negation, but are licit elsewhere. 1.2.3.1. NI-NPIs (Negative Polarity Items) I will call NI-NPIs those NPIs which begin with the negative prefix ni, e.g.: ni(t)ko (no one), nista (nothing), nikud (nowhere), nikad (never), etc. Morphologically, these items are complex, consisting of three morphemes: (105)

n-i-(t)ko neg-any-who (106) n - i - s t a neg-any-what

They will be glossed as nobody, nothing, etc., only because they show morphological negation. However, their distribution is by no means identical to that of English negated NPIs, as will become evident immediately. The distribution of NI-NPIs is straightforward: they can only occur in a clause which contains an overt negative particle ne no matter whether they are in the object, subject, or adjunct position: (107) Milan *(ne) vidi nista. Milan not sees nothing 'Milan cannot see anything.' (108) Ni(t)ko *(ne) vidi Milan-a. nobody not sees Milan-ACC 'Nobody can see Milan.'

Negative polarity in Serbian I Croatian 41 (109) Milan ni-kada *(ne) vozi. Milan no-when neg drives 'Milan never drives.' (110) Milan to ni-kako *(ne) odobrava. Milan that no-how not approves 'Milan does not approve of that at all.' A negative particle in a superordinate clause does not license a NI-NPI (but see section 1.2.6. for subjunctive-like constructions and infinitivals): (111) *Milan Milan (112) *Milan Milan (113) *Milan Milan (114) *Milan Milan

ne not ne not ne not ne not

tvrdi [da claims that tvrdi da claims that tvrdi da claims that tvrdi da claims that

Marija poznaje ni(t)ko-ga]. Mary knows no-one-ACC ni(t)ko vidi Marij-u. no-one sees Mary-ACC Marija nikada vozi. Mary never drives Marija to ni-kako odobrava. Mary that no-how approves

If both clauses contain the negative particle ne, the sentence is interpreted as having two logical negations: the number of negative particles equals the number of logical negations. This number is unaffected by the number of NI-NPIs which appear in the sentence: (115) Milan ne tvrdi [da Marija ne poznaje ni(t)ko-ga\. Milan not claims that Mary not knows no-one-ACC 'Milan does not claim that Mary knows no one.' (116) Milan ne tvrdi da ni(t)kone vidi Marij-u. Milan not claims that no-one not sees Mary-ACC 'Milan does not claim that no one can see Mary.' (117) Milan ne tvrdi da Marija nikada ne vozi. Milan not claims that Mary never not drives 'Milan does not claim that Mary never drives.' (118) Milan ne tvrdi da Marija to ni-kako ne odobrava. Milan not claims that Mary that no-how not approves 'Milan does not claim that Mary does not approve of that at all.' 1.23.2. I-NPIs (Negative Polarity Items) I will call I-NPIs those NPIs which begin with the prefix /, e.g.: i(t)ko (anyone), ista (anything), ikud (anywhere), ikad (ever). Morphologically, these items consist of two morphemes:

42 Negative contexts: Serbian!Croatian (119) i-(t)ko any-who (120) i- sta any-what They will be glossed as any although their distribution differs considerably from that of English awy-phrases. I-NPIs are licensed in the embedded clause of a negated matrix, appearing in object, subject, and adjunct positions: (121) Milan ne tvrdi [da Marija poznaje i(t)ko-ga\ Milan not claims that Mary knows anyone-ACC 'Milan does not claim that Mary knows anyone.' (122) Milan ne tvrdi da i(t)ko voli Marij-u. Milan not claims that anyone loves Mary-ACC 'Milan does not claim that anyone loves Mary.' (123) Milan ne tvrdi da Marija ikada vozi. Milan not claims that Mary ever drives 'Milan does not claim that Mary ever drives.' (124) Milan ne tvrdi da Marija to i-kako odobrava. Milan not claims that Mary that any-how approves 'Milan does not claim that Mary approves of that in any way.' However, I-NPIs are disallowed with a clausemate negation: (125) *Marija ne poznaje i(t)ko-ga. Mary not knows anyone-ACC (126) *I(t)ko ne vidi Milan-a. anybody not sees Milan-ACC (127) * Milan ikada ne vozi. Milan ever not drives (128) *Milan to i-kako ne odobrava. Milan that any-how not approves I-NPIs are equally unacceptable if there is no negation anywhere in the sentence: (129) *Milan Milan (130) *Milan Milan

tvrdi [da claims that tvrdi da claims that

Marija poznaje i(t)ko-gd\. Mary knows anyone-ACC i(t)ko voli Marij-u. anyone loves Mary-ACC

Negative polarity in Serbian I Croatian 43 (131) *Milan Milan (132) *Milan Milan

tvrdi da Marija ikada vozi. claims that Mary ever drives to i-kako odobrava. that any-how approves

In sum, I-NPIs are licensed only by superordinate negation, and never by clausemate negation. In this respect, they are in complementary distribution with NI-NPIs when in the scope of negation. NI-NPIs and I-NPIs share one important characteristic of negative polarity items: neither is licensed in positive statements. It should be noted at this point that there are other contexts which allow I-NPIs, although they do not seem to be negative, for example, yes/ no questions, conditionals, adversative predicates, etc. The discussion of these environments will be postponed until Chapter 3 because it depends on the analysis of licensing by negation. For the moment, suffice it to say that only I-NPIs can occur in these contexts; NI-NPIs are restricted to contexts with a clausemate negative particle: (133) Da li Milan voli i(t)ko-ga / * ni(t)ko-gal5 that Q Milan loves anyone-ACC no-one-ACC 'Does Milan love anyone?' 1.2.3.3. Positive Polarity Items (PPIs) Positive Polarity Items (PPIs) are those lexical items which are in some way 'anti-triggered' (i.e., disallowed) by negation. Ladusaw (1980) notices that PPIs in English, such as some, already, etc., cannot occur in the scope of a clausemate negation: (134) #John did not see someone. The symbol '#' is used here to indicate that a PPI cannot be interpreted in the scope of negation. In particular, (134) cannot have the following reading: (135) It is not the case that John saw a (any) person. It can only have an interpretation where someone takes wide scope with respect to negation: (136) There is a person X such that John did not see X. The facts in Serbian/Croatian are the same: PPIs are not allowed in the scope of clausemate negation:6

44 Negative contexts: Serbian I Croatian (137) #Milan nije uvredio ne(t)ko-ga. Milan not-is insulted someone-ACC 'Milan has not insulted someone.' (138) #Ne(t)ko nije dosao. someone not-is come 'Someone has not come.' (139) #Milan ne dolazi nekada. Milan not comes sometimes 'Milan does not come sometimes.' (140) #Milan to ne odobrava ne-kako. Milan that not approves some-how 'Milan does not approve of that somehow.' On the other hand, PPIs can fall within the scope of superordinate negation: (141) Marija ne smatra da je Milan uvredio ne(t)koga. Mary not thinks that has Milan insulted someone 'Mary does not think that Milan insulted someone.' (142) Marija ne smatra da ne(t)ko dolazi. Mary not thinks that someone comes 'Mary does not think that someone is coming.' (143) Marija ne smatra da Milan dolazi nekada. Mary not thinks that Milan comes sometimes 'Mary does not think that Milan sometimes comes.' (144) Marija ne smatra da Milan to ne-kako odobrava. Mary not thinks that Milan that some-how approves 'Mary does not think that Milan approves of that somehow.' If there is no negation in the sentence, PPIs are licit: (145) Milan je uvredio ne(t)ko-ga. Milan is insulted someone-ACC 'Milan has insulted someone.' (146) Ne(t)ko je dosao. someone is come 'Someone has come.' (147) Milan dolazi nekada. Milan comes sometimes 'Milan sometimes comes.'

Negative polarity in Serbian I Croatian 45 (148) Milan to razume ne-kako . Milan that understands some-how 'Milan understands that somehow.' 1.2.4. A binding analysis

As established above, NI-NPIs are licensed only by clausemate negation, whereas I-NPIs are licensed only by superordinate negation. On the other hand, PPIs are anti-triggered by clausemate negation, while licit elsewhere. Therefore, NI-NPIs and PPIs are in a perfect complementary distribution. It is these facts that I will try to account for by invoking binding principles. A binding approach to negative polarity has already been suggested in the literature, although only for negative contexts (see e.g., Aoun (1985; 1986) and Milner (1979)). Since both NI-NPIs and I-NPIs are dependent on the presence of negation, let us assume that they are both, in a sense, anaphoric. Since PPIs cannot appear in the scope of a local negation, let us assume that they are pronominal. In order to explain the difference between the two NPI-types, let us propose that NI-NPIs are anaphors, subject to Principle A of the binding theory, whereas I-NPIs are bound pronouns, subject to Principle B, but which are also required to be bound in the sentence. Following the parallel further, PPIs will be subject only to Principle B. Since the binder for NPIs is negation situated in an A'-position, e.g. Infl, I will adopt Aoun's (1985; 1986) insight that both A- and A'-binding are possible7. The distribution of Serbian/Croatian NPIs would now follow from the following principles: (149) NI-NPIs (ni(t)ko, nista, etc.) are anaphoric, subject to Principle A: they have to be A'-bound by negation in their governing category. (150) PPIs (ne(t)ko, nesto, etc.) are pronominals, subject to Principle B: they have to be A'-free in their governing category. (151) I-NPIs (i(t)ko, ista, etc.) are anaphoric pronominals, subject to Principle B: they need to be free in the local clause, but bound in the sentence.8 It is important to note in the very beginning that the notion of binding referred to here is purely syntactic, intended to capture locality conditions between an NPI and its licenser, negation. Since NPIs are non-referential (see Givon (1969)), co-indexing which results from binding of NPIs by

46

Negative contexts: Serbian I Croatian

negation cannot be interpreted as co-reference. However, it may be that non-referential expressions require a truth-conditional operator to negate their referentiality. If so, the most restrictive hypothesis would be that the same locality conditions that hold between anaphors and their antecedents would also hold between an NPI and its licenser. Principles (149) to (151) account immediately for the following facts: (152) 1. NPIs need a licenser. 2. The licenser for NPIs has to be local (non-local NPIs will be shown in Chapter 2 to involve LF raising) 3. PPIs need no licenser. 4. PPIs cannot be bound to a local licenser. Let us now review Serbian/Croatian data in the light of the proposed analysis. The following example is grammatical because the NI-NPI is bound by negation in the local clause: (153) Milan ne postuje ni(t)ko-ga. Milan not respects no-one-ACC 'Milan does not respect anyone.' Negation in the superordinate clause, however, is outside of the governing category for the NI-NPI, and therefore cannot license it: (154) *Ne tvrdi-m

da ni(t)ko postuje Milan-a.

not claim-1SG that no-one respects Milan-ACC With no antecedent anywhere, NI-NPIs are equally unacceptable: (155) *Ni(t)ko postuje Marij-u. no-one respects Mary-ACC The following sentence is ungrammatical because the anaphoric requirement for the I-NPI is not met. There is no appropriate antecedent for it: (156) *I(t)ko postuje Milan-a. anyone respects Milan-ACC The following example, on the other hand, violates the pronominal requirement on I-NPIs. The I-NPI is A'-bound by negation in the local clause, where it should be A'-free.

Negative polarity in Serbian I Croatian 47 (157) *Milan ne postuje i(t)ko-ga. Milan not respects anyone-ACC In order for an I-NPI to satisfy both of its requirements, it must be free from negation in the local domain, but bound by one in a superordinate clause: (158) Ne tvrdi-m da i(t)ko postuje Marij-u. not claim-1SG that anyone respects Mary-ACC 'I don't claim that anyone respects Mary.' It is possible for I-NPIs to be bound across more than one clause boundary, as long as they remain free in the local clause: (159) Ne kaze-m [da Milantvrdi [da i(t)ko i(t)ko-ga postuje]] not say-lSG that Milan claims that anyone anyone-ACC respects. 'I don't say that Milan claims that anyone respects anyone.' 1.2.5. Parallelism with reflexives and pronouns

Let us for a moment concentrate on NI-NPIs and PPIs only. The parallel with A-binding is striking. Morphologically complex reflexives (see English him + self) require a local antecedent, just like NI-NPIs: (160) *John thinks [that Mary likes himself]. Morphologically simple reflexives, on the other hand, can be bound longdistance (see section 0.2.1. for discussion). The local NPI n-i-(t)ko (negany-who) is morphologically more complex than the long-distance NPI i(t)ko (any-who) or any-one. I return to the account of this contrast below. Furthermore, pronouns, which are subject to Principle B, do not require an antecedent. If they have one, it cannot be local: (161) Mary saw him. (162) *Mary thinks that Peteri saw himx. This is exactly parallel to PPIs. The parallel also extends to Exceptional Case Marking (ECM) (see Chomsky (1981)). Consider the following pair: (163) John considers [IP himself {to be) intelligent]. (164) *Johni considers [IP himx (to be) intelligent].

48 Negative contexts: Serbian I Croatian In the ECM environment, a reflexive in the subject position can take an extra-clausal antecedent. A pronoun in the same position must be disjoint from the matrix subject. Now consider the following examples with polarity items: (165) Milan ne smatra [IP ni(t)ko-ga inteligentn-im]. Milan not regards no-one-ACC intelligent-INSTR 'Milan does not regard anyone intelligent.' (166) #John doesn't consider [IP someone (to be) intelligent]. (167) #Milan ne smatra [IP ne(t)koga inteligentnim]. Milan not regards someone intelligent A local NPI (ni(t)ko) is licensed by extraclausal negation in the ECM environment (see (165)), just like the reflexive in (163) is. Moreover, a PPI (someone) cannot fall within the scope of the superordinate negation in (166) and (167), just like the pronoun in (164) cannot take the matrix subject as its antecedent. Notice that in order to achieve this result, we need to recast the notions 'narrow scope' and 'wide scope' into binding terminology (see also discussion in section 0.1. ). I take a PPI to have a narrow-scope reading with respect to negation if it is bound by negation. On the other hand, a PPI is interpreted as taking wide scope with respect to negation if it is obviative, that is, if it is not bound by negation. In the following example: (168) John saw him. him has to be obviative, that is, not bound by John. We can just as well say that in this case him is not in the scope of John to make it parallel with PPIs. However, this is not just a terminological variant. While a scope approach has to stipulate that PPIs cannot appear in the scope of a clausemate negation, this fact follows automatically from the binding terminology. In fact, if NPIs and PPIs are quantifiers not subject to the binding principles proposed above, it is hard to explain why they do not enter freely into scope relations with negation, the way other quantifiers do. For example, the universal quantifier every in (169) can take either wide or narrow scope with respect to negation, giving rise to two different readings, (170) and (171): (169) Everybody didn't come. (170) For every X, it is true that X did not come; or

Negative polarity in Serbian I Croatian 49 (171) It is not true, for every X, that X came. Crucially, such scope possibilities are not granted to otherwise quantificational NPIs and PPIs. Their scope with respect to negation is fixed. A PPI must take wide scope with respect to a clausemate negation, while a NPI must always take narrow scope with respect to negation. This fact, too, follows from the binding approach. If NPIs are bound by negation, they must be in its scope. If PPIs are subject to Principle B, they may not be bound by clausemate negation, i.e., they may not fall within its scope. So far we have seen that NI-NPIs correspond to local complex reflexives of the type himself m English. English any, on the other hand, corresponds to morphologically simple long-distance reflexives of the type ziji (self) in Chinese (see section 0.2.1. for details). One might wonder, however, if there is any A-binding equivalent of the anaphoric pronominal I-NPI. The distribution of the Greek reflexive o idhios (oneself) is exactly parallel to that of I-NPIs (see Iatridou (1986)). O idhios has to be A-free in its governing category, but A-bound in the sentence: (172) O Yanis pistevi oti o idhios tha kerdhisi. John believes Comp himself will win 'Johni believes that hej will win.' (173) *O Yanis aghapa ton idhio. John loves him-self 'John loves himself.' In addition, the Norwegian reflexive seg shows a similar distribution: it can be bound to a matrix subject across an infinitival clause, but not to a local subject (see Yang (1983) and originally Halvorsen (1982)): (174) Knuti ba Ola, [PROj korrigere segi/*j] asked correct self 'Knut asked Ola to correct self.' (175) *Olai korrigerer segx. 'Ola corrected self.' Iatridou (1986) introduces another binding principle to capture the distribution of the Greek anaphor: "What seems to be missing is the option 'bound in the whole sentence but free in the governing category' which we may provisionally call 'binding condition D'" (p. 769).

50

Negative contexts: Serbian I Croatian

Before concluding this section, it is advisable to draw a parallel between PPIs, as pronominals, and I-NPIs, as anaphoric pronominals. As their names suggest, they should show both similarities and differences, which is indeed the case. Neither I-NPIs nor PPIs can be bound by clausemate negation. On the other hand, both PPIs and I-NPIs are compatible with superordinate negation and with non-negative polarity environments. This similarity can be ascribed to their common pronominal property. However, PPIs differ from I-NPIs in that they do not require negation, or any other polarity context, in order to be felicitous. This difference follows from the assumption that I-NPIs are anaphoric, whereas PPIs are not. 1.2.6. Subjunctive-like constructions

As pointed out in section 1.2.3, there is a class of verbs in Serbian/ Croatian (zeleti 'wish', moci 'be able to', traziti 'ask for', etc.) with which the complementary distribution between NI-NPIs and I-NPIs breaks down. The embedded complement of these verbs can host either NI-NPIs or I-NPIs: (176) Ne zeli-m da vidi-m ni(t)ko-ga / i(t)ko-ga not wish-lSG that see-lSG no-one-ACC anyone-ACC 'I do not wish to see anyone.' In the subject position, however, only I-NPIs are allowed: (177) Ne zeli-m da !*ni(t)ko / i(t)ko poseti Marij-u. not want-lSG that no-one / anyone visits Mary-ACC 'I don't want that anyone visits Mary.' This subject/object asymmetry is reminiscent of the asymmetry in Romance subjunctives (see section 1.2.2). Just as in Romance, the governing category for the embedded NPIs with these verbs seems to extend one clause up. Curiously, the set of Serbian/Croatian verbs discussed in this section coincides with those verbs that take Subjunctive complements in other languages, e.g., in Romance and Icelandic. Subjunctives have been argued to have anaphoric tense, by e.g., Anderson (1982), Pica (1984), Everaert (1984), Jakubowicz (1985) and Johnson (1985). It seems that the absence of Tense (assuming that Present is a default Tense) in the complement clause allows for a domain extension. Serbian/Croatian

Negative polarity in Serbian]Croatian

51

provides supporting evidence for this claim. The verbs that allow domain extensions do not show any tense alternations in the complement clause, i.e., present tense must be used; contrast (176-177) with (178,179): Past/Perfect: (178) *Ne zelim da sam video Mariju. not wish-lSG that am seen Mary 'I don't wish to have seen Mary.' Future: (179) *Ne zelim da cu videti Mariju. not wish-lSG that FUT see Mary 'I don't wish to see Mary.' Notice that if the conditional modal bih is used in the matrix clause, both present and past tense become possible: Present: (180) Ne bih zeleo da vidim Mariju. not COND-1SG wished that see Mary 'I wouldn't like to see Mary.' Past/Perfect: (181) Ne bih zeleo da sam video Mariju. not COND-1SG wished that am seen Mary 'I wouldn't like to have seen Mary.' However, NI-NPIs are now illicit in the complement: (182) Ne bih zeleo da sam video *ni(t)koga / i(t)koga. not COND-1SG wished that am seen no-one anyone 'I wouldn't like to have seen anyone.' Suppose now that if Tense is absent, the functional sentential projections (Comp and Infl) can become invisible at LF: see Lasnik and Saito (1984) for deletion of recoverable elements at LF (see Chapter 6 for a detailed analysis of subjunctive).9 Given this much, we can explain why the domain for polarity items extends one clause up. If both Infl and Comp are absent, the first potential antecedent for an NPI or a PPI is negation in the matrix Infl. Thus the matrix IP counts as the governing category (see section 3.3. for a definition of governing category based on the first potential antecedent). Thus, PPIs also extend their domain in

52

Negative contexts: Serbian I Croatian

subjunctive-like constructions. In the following example, the PPI must take wide scope with respect to negation: (183) #Ne zeli-m da vidi-m ne(t)ko-ga. Not want-lSG that see-lSG someone-ACC 'I don't want to see someone.' It is worth noting that the domain with the listed verbs extends not only with NPIs but with other clause-bound processes as well. For example, preposing out of the embedded clause is possible only with these verbs: (184) To ne zeli-m da izgubi-m. this not wish-lSG that lose-lSG This, I don't want to lose.' (185) l*To ne kaze-m da sam izgubio. this not say-lSG that am lost ??This, I don't say that I have lost.' Clitics originating in the embedded clause can attach to the matrix subject only with these verbs: (186) ?Milangfli zeli [da vidi tj. Milan him wants that sees 'Milan wants to see him.' (187) *Milangfli kaze [da vidi tj. Milan him says that sees 'Milan says that he can see him.' While long-distance extraction over negation is generally unacceptable even for arguments in Serbian/Croatian (see Rizzi (1990a) for adjuncts), it is grammatical with these verbs: (188) Stai ne zelis [da mi kazes tj? what not want that me tell 'What don't you want to tell me?' (189) ?*Stai ne kazes [da volis tj]? what not say that like 'What don't you want to say that you like?' At the moment, the facts can only be taken as suggestive of some kind of 'clause union effect,' which notion will be made more precise in Chapter 6. Unfortunately, it is not possible to check whether the binding domain

Negative polarity in Serbian/'Croatian

53

for reflexives in Serbian/Croatian also extends, since the embedded subject must be a pro (or PRO), coindexed with the matrix subject (for a detailed discussion of these phenomena, see Progovac (1993)).

2 Negative contexts: English

In Chapter 1 polarity sensitivity in Serbian/Croatian negative contexts was captured within the Binding theory by the following claim: NPIs must be bound, whereas PPIs are subject to Principle B of the Binding theory. Negation serves as a possible binder for polarity items. The following question immediately arises: can this binding account of polarity sensitivity carry over to other languages as well, or is it confined to Serbian/Croatian alone? I will argue in this chapter that the analysis extends to English as well. Some other languages will be considered in Chapter 4. 2.1. Positive Polarity Items (PPIs) The proposal for Serbian/Croatian PPIs carries over directly to English PPIs, which show the same distribution. English PPIs, such as some, already, etc., are not allowed in the scope of a clausemate negation, but are acceptable in the scope of a superordinate negation: (1)

#John has not already arrived.

(2)

Mary did not say that John has already arrived.

Likewise, PPIs are acceptable in positive contexts: (3) John has already arrived. All one needs to say about PPIs in English is that they are subject to Principle B, and therefore have to be free from negation in their governing category. 2.2. Negative Polarity Items (NPIs) and LF-raising The story is not as simple with NPIs, however. At first sight, English NPIs appear incompatible with a binding account since no locality 54

NPIs and LF-raising 55 requirements seem to be imposed: NPIs are licensed in the scope of any polarity operator, negation or other:1 Clausemate Negation (4) Mary did not see anyone. Superordinate Negation (5) Mary did not say that she had seen anyone. Yes/No Questions (6) Did Mary see anyone! Conditionals (7) If Mary saw anyone, she will let us know. Adversative Predicates (8) I doubt that Mary saw anyone. NPIs in English are also licensed by negation which is more than one clause away: (9)

Mary did not say [that Peter claimed [that she had seen anyone]].

In this respect, any resembles the so-called long-distance reflexives that have been attested in AGR-less languages such as Chinese, Japanese and Korean. Such reflexives can, in principle, be bound across any number of clauses. As pointed out in section 0.2.1, morphologically simple, X° reflexives seem to take only AGR (an X° category) as their SUBJECT (see Progovac (1992c), Progovac and Connell (1991), Progovac and Franks (1991)). Since Chinese-type languages lack overt (morphological) AGR, there is nothing to delimit the domain. It may be possible to explain the lack of locality conditions with any in a similar fashion. Suppose that cross-linguistically the only SUBJECT for NPIs is negation in a head position of a functional category, IP or NegP. In languages in which negation is clearly in a head position, whether under I or Neg, the domain will be local (e.g., Serbian/Croatian, Italian, etc.). However, if not is an adverb adjoining to a maximal projection (see section 1.2.1. for discussion of Hasegawa (1987) and Pollock (1987)), there will be no SUBJECT for any in English (see also Rizzi's (1990a) arguments for treating not as a specifier). I leave this possibility open for future research. For the moment, however, I will assume that the long-distance effect with any is achieved via successive-cyclic movement of any (see section 4.1.2. for details). Unlike reflexives, which do not obey movement constraints (see Huang and Tang (1989)), any does, as will be shown below.

56

Negative contexts: English

The following universal and parameters will be taken as responsible for variation in NPI licensing: (10) (11)

(12)

Universal: All NPIs must be bound and are subject to Binding Principles. Parameter 1: Some NPIs are subject to Principle A (e.g., Serbian/ Croatian NI-NPIs and English NPIs), whereas others are subject to Principle B (e.g., Serbian/Croatian I-NPIs). Parameter 2: Some NPIs raise at LF (e.g., English NPIs ), whereas others do not (e.g., Serbian/Croatian NI-NPIs).

In Chapter 4 these parameters will be refined, and some new languages added. In the rest of this section, I will argue that Parameter 2 can be derived for English and Serbian/Croatian NPIs from their morphological properties. Recall (from section 1.2.3) that NI-NPIs are morphologically more complex than English any. (13) (14)

any-one n- i- ko neg-any-who

Suppose that the following is the internal structure of anyone? (15)

DP QP

DP D

NP

any one Being quantificational in nature, any QPs can raise at LF, and can thus satisfy their anaphoric requirement extraclausally (for the landing sites, see section 4.1.2.). On the other hand, NI-NPIs are assigned the following structure: (16)

DP NegP

DP QP

DP D

NP (t)ko

Independent evidence for LF-raising

57

The anaphoric part ni cannot raise because it is not a constituent at all. To raise / alone would not suffice to satisfy the anaphoric property of the whole word. Besides, the trace of / would violate the ECP (Empty Category Principle) due to the minimality condition (see Chomsky (1986a), Aoun and Li (1989), Rizzi (1990a)). The trace of QP would be bound by the minimal binder, the negative particle n, and not by the QP itself, resulting in vacuous quantification.3 2.3. Independent evidence for LF-raising

Let us consider the following example which involves long-distance NPIlicensing in English, i.e., licensing by a superordinate negation: (17)

Mary does not claim [that John misplaced anything].

Since in (17) the binder for the A'-anaphor anything is not within its governing category at S-Structure, suppose that the NPI raises at LF, thus extending its domain. It has been established that overt movement is subject to some constraints, for example: Island Constraints (see Ross (1967)) and the Empty Category Principle (Chomsky (1981)). Island constraints prohibit movement from islands (e.g., Complex NPs, Coordinate Structures, WhClauses, etc.): (18)

*WhOi did Mary hear the story that ti killed Peter?

In addition, the ECP prohibits movement from the positions which are not properly governed. However, all these restrictions on movement (island constraints and the ECP) seem far less obvious at LF (see Huang (1982)).4 The that-tmce effect, for example, which has been subsumed under the ECP, is also not always observed at LF. Nevertheless, I would like to argue that long-distance licensing of any does obey movement constraints, in particular the Specificity Constraint, the Island conditions, the ECP, and a ban on extraction from topicalized positions. 2.3.1. Specificity The specificity of an NP can create an island for extraction (see Fiengo and Higginbotham (1981)):

58 Negative contexts: English (19) (20)

Who did you see pictures of? *Who did you see the picture of?

The definiteness of the determiner and the singular number of N' contribute to the specificity of an NP. Fiengo and Higginbotham show that the Specificity Constraint applies at LF because of its interaction with QR, an LF rule. They cite the following examples to illustrate that a quantifier cannot raise out of a specific NP: (21) (22)

People from every walk of life like jazz. The people from every walk of life like jazz.

While (21) has a wide-scope quantifier reading, (22) does not, and is therefore odd. Since Specificity does apply at LF, we should expect it to affect the LF extraction of NPIs. And, indeed, this seems to be the case. May (1985) offers the following contrast: (23) (24)

John never reads books which have any pages missing. *John never read the book that has any pages missing.

While the specific NP in (24) blocks a relation between negation and the NPI, the non-specific NP in (23) allows the link to be established. 2.3.2. Island conditions

Any seems to obey the Coordinate Structure and the Adjunct Constraint, while the Complex NP Constraint is not obeyed consistently. Consider the following examples illustrating any within different islands: Coordinate Structure Constraint: (25) ?*I am not asking you to prepare this and bring anything. Adjunct island: (26) *I did not make a pie after I received anyone. Complex NP Constraint (CNPC) (see Ross (1967) and Baker (1970b)): (27) ?*We weren't aware of the fact [that anyone had left]. There are examples which are acceptable although the Relative Clause island should be violated (Ladusaw (1980)), for which fact I have no explanation:

Independent evidence for LF-raising 59 (28)

I never met [a man [who anybody tried to kill]].

I return to island effects in section 6.2.1, where I discuss Italian nessuno.5 233. The ECP The ECP (Empty Category Principle) prohibits movement from the positions which are not properly governed. Objects are lexically governed by the verb and need no antecedent government, whereas subjects (see also adjuncts of the how/why type) are not lexically governed, and therefore require an antecedent (Chomsky (1981)). This results in a subject/object asymmetry with respect to long-distance extraction, as in: (29) (30)

*WhOi did Mary say that ti hurt Bill? Whoi did Mary say that Bill hurt ti?

Antecedent government is only blocked if there is an overt complementizer in the Comp of the embedded clause (for an analysis of how it is blocked, see the Comp-Indexing algorithm proposed in Aoun, Hornstein, and Sportiche (1981)). Therefore the counterpart of (29) without the complementizer will be grammatical: (31)

Whoi did Mary say ti hurt Bill?

At first sight, there are no ECP effects in the long-distance binding of any in English. The following examples show no subject/object or adjunct asymmetry:6 English: (32) John does not believe [that anyone is coming to the party]. (33) John does not believe [that Mary is seeing anyone]. (34) John does not believe [that Mary ever snores]. However, contrary to appearances, NPIs seem to obey the ECP in that they show a non-bridge verb effect. The following bridge/non-bridge verb contrast was illustrated by Martin (1987):7 (35)

John did not say that anyone else is dangerous.

(36)

*John didn't whisper that anyone else is dangerous.

This contrast weakens with object and adjunct extraction: (37)

Mary did not say that John killed anyone.

60 Negative contexts: English (38) (39)

?Mary did not whisper that John killed anyone. ?John did not whisper [that Mary ever came].

Notice that adjuncts also pattern with objects with wh-extraction in that they do not show that-trace effects (see Lasnik and Saito (1984), Rizzi (1990a)): (40) (41) (42)

*Which student do you think [that t could solve the problem]? Which problem do you think [that Bill could solve t]? How do you think [that Bill could solve the problem t]?

It remains to be explained why bridge verbs behave differently from non-bridge verbs. As noticed in Erteschik (1973), extraction from the subject position is never acceptable with non-bridge verbs, while extraction from object position is possible (Stowell (1981)): (43) (44)

*Who did Bill mutter/whisper was playing too much poker? Who did Bill mutter/whisper that he does not like?

The question to address at this point is the following: what makes the LF movement out of non-bridge verb complements parallel to overt whmovement, and different from the LF wh-movement out of bridge verb complements? The answer seems to be related to another difference between bridge and non-bridge verbs. With non-bridge verbs the complementizer is always obligatory (see Erteschik (1973)): (45)

*Bill muttered/whispered Denny was playing too much poker.

If we assume, with Stowell (1981), that non-bridge verbs do not properly govern the head of the clause they embed, i.e., Comp, the complementizer cannot be deleted at S-Structure or at LF. Assuming, furthermore, with Lasnik and Saito (1984) that that, having no syntactic content, can freely delete at LF with a bridge verb which can lexically govern its trace, we can explain why that-trace effects do not obtain with LF extraction from complements to bridge verbs. On the other hand, with non-bridge verbs, which do not allow that deletion, the effects are observed.8 Following Rizzi (1990a), we can say that objects and VP adjuncts are head-governed by the verb, whereas subjects are not. Since subjects need to be head-governed within the immediate projection of the head, they can only be governed by Comp. The reason why subject extraction improves in the absence of the complementizer in Comp follows from Rizzi's assumption that AGR in the head of CP, triggered by the wh-

Independent evidence for LF-raising

61

element or its trace in the Spec of CP, can head govern the subject position. Given that non-bridge verbs cannot delete their complementizer for independent reasons, CP agreement will be blocked by the presence of that, which is inert for government. 2.3.4. Topicalized NPIs It has long been observed that once a phrase is in an A'-position at SStructure, it remains unavailable for LF movement. Lasnik and Uriagereka (1988) give the following contrast: (46) (47)

Who thinks that I like who! *Who thinks that who, I like t?

The topicalized NP is not eligible for movement to the Spec of CP, thus producing ungrammaticality in (47). Exactly the same contrast, also noted by Lasnik and Uriagereka (1988), is observed with NPIs:9 (48) (49)

I don't think that Mary solved any problems. *I don't think that any problems, Mary solved.

While in (48) movement from an argument position is permitted, (49) illustrates that movement from an IP-adjoined position is prohibited. This is a strong argument for the movement analysis: if NPIs do not move, why should (49) be worse than (48), given that the NPI in (49) is even closer to its licenser than in (48)? In sum, there is evidence for NPI-Raising, although its application at LF makes this evidence less straightforward. In any event, it seems that NPIs offer at least as much evidence for a raising analysis as some other LF phenomena, e.g., wh-in-situ in Chinese (see Huang (1982)).

3 Non-negative polarity contexts

3.1. Introductory words

In Chapters 1 and 2 a binding account of polarity sensitivity was proposed for negative contexts. It was argued that locality restrictions on polarity items can best be captured by claiming that NPIs are subject to Principle A, whereas PPIs are subject to Principle B of the Binding Theory. How does this account carry over to non-negative polarity contexts, in which there is no visible element from which locality conditions can be established? Although licensing by negation is generally assumed to be the core case of NPI licensing (see Baker (1970b), Klima (1964), and Linebarger (1981); but see Ladusaw (1980) for a different view), there are many other contexts which allow NPIs: yes/no questions, conditionals, adversative predicates1 (such as deny, doubt, prevent), universal quantifiers, etc. (see Linebarger (1981) for a detailed list): (1) (2) (3) (4)

Did Mary insult anyone! If Mary insulted any of these people, she should apologize. I doubt that Mary insulted anyone. Every man who insulted anyone should apologize.

I will argue that all the non-negative polarity contexts have an empty polarity operator (Op) in Comp which is responsible for NPI licensing. There are a few possible ways to predict the licensing of Op (i.e., NPIs) in these contexts. One way is to assume with Ladusaw (1980, 1982, 1983) that NPIs are licensed by DE (Downward Entailing) operators. On Ladusaw's DE approach, a polarity item is licensed if and only if in the scope of a DE operator, e.g., negation, conditional, adversative predicate, etc. (for details, see section 8.1.). In the scope of a DE operator the direction of entailment is from supersets (a vegetable below) to subsets (kale): 62

Introductory words 63 (5) (6)

Mary did not eat a vegetable. Mary did not eat kale.

To reconcile Ladusaw's semantic licensing conditions with the locality requirements, one can suggest that a DE clause licenses a polarity operator in its Comp, which, in turn, licenses NPIs (see Progovac (1992, a and b)). This would still leave yes/no questions unexplained (see section 8.1. for more details). As Ladusaw (1980) himself points out, yes/no questions are not DE, yet they license NPIs: (7) (8) (9)

Did Mary eat a green vegetable? Did Mary eat brussels sprouts? Did Mary eat anything!

yes yes

Ladusaw assumes that this is so because yes/no questions are compatible with both positive and negative answers, so that basically anything is possible in questions. However, this cannot be the whole truth, as evidenced in the following contrast (see also footnote 5, Chapter 1, for a surprising fact in Serbian/Croatian): (10) (11)

*Mary ate anything! Did Mary eat anything!

Athough both examples involve questions with an unfixed truth value, only the latter licenses NPIs. Notice that the following intonation question without an NPI is fully acceptable: (12)

Mary ate her lunch?

This fact suggests that a formal licenser for NPIs is indeed necessary even in questions. Alternatively, one can assume that Op is licensed either by unfixed truth conditions in unselected contexts (questions, conditionals, etc.), or by selection by certain verbs and prepositions (e.g., doubt, forget, without). One striking similarity is shared by all the non-negative polarity clauses: none of them has its truth value fixed positively. For example, yes/no questions allow for either a negative or an affirmative answer: (13)

Op Did Mary visit you yesterday?

Antecedents of conditionals are not committed to the positive outcome: (14)

Op If Mary comes, we'll let you know.

64 Non-negative polarity contexts One can assume that Op is either negative (Progovac (1988), Laka (1990)) or that it represents a switch with a +/- choice (see section 3.3. for more details concerning the former possibility). Either way we would explain why the truth value of sentences headed by Op is not fixed positively. I will not attempt to decide between different options here. Instead, I will concentrate on the locality conditions that hold between NPIs and their licenser, Op, whatever turns out to be the ultimate licenser for Op itself. 3.2. Non-negative licensing patterns with superordinate negation

In this section I will show that licensing in non-negative polarity contexts patterns with licensing by negation from a superordinate clause. There are three pieces of evidence that lead to this conclusion: choice of NPIs in Serbian/Croatian, distribution of PPIs in both English and Serbian/ Croatian, and subject/object asymmetries in English. 3.2.1. Choice of NPIs in Serbian/Croatian In non-negative polarity contexts, only I-NPIs are licensed, and never NI-NPIs. Yes/no questions: (15) Da li Milan voli i(t)ko-ga / *ni(t)ko-gal that Q Milan loves anyone-ACC / no-one-ACC 'Does Milan love anyone?' Conditionals: (16) Ako Milan povredi i(t)ko-ga / *ni(t)ko-ga, bi-ce kaznjen. if Milan hurts anyone-ACC/ no-one-ACC be-FUT punished 4 If Milan hurts anyone, he will be punished.' Adversative predicates: (17) Sumnja-m da Milan voli i(t)ko-ga / *ni(t)ko-ga. doubt-1SG that Milan loves anyone-ACC / no-one-ACC 4 I doubt that Milan loves anyone.' Universal Quantifiers: (18) Svako (t)ko povredi i(t)ko-ga / *ni(t)ko-ga, mora everyone who injures anyone-ACC / no-one-ACC must

Non-negative licensing patterns 65 biti kaznjen. be punished 'Everyone who injures anyone must be punished.' The same is true of licensing by superordinate negation (see section 1.2.): (19)

Marija ne tvrdi da Milan voli i(t)koga / *ni(t)koga. Mary not claims that Milan loves anyone-ACC / no-one-ACC 'Mary does not claim that Milan loves anyone.'

3.2.2. PPIs In both English and Serbian/Croatian, PPIs are licit in non-negative polarity environments, confirming our generalization that non-negative licensing patterns with licensing by superordinate negation. Recall that PPIs are also compatible with superordinate negation (section 1.2.3.3). Here are the relevant examples from English and Serbian/Croatian: Yes/No Questions: (20) Has John insulted someone! (21) Da li je Jovan ne(t)koga uvredio? that Q is John someone insulted Conditionals: (22) If John has insulted someone, he should apologize. (23) Ako je Jovan ne(t)koga uvredio, treba da se izvini. if is John someone insulted ought that self apologize Adversatives: (24) I doubt that John has insulted someone. (25) Sumnja-m da je Jovan ne(t)koga uvredio. doubt-1SG that is John someone insulted Compare occurrence of PPIs with superordinate negation: (26) (27)

I do not claim that John has insulted someone. Ne tvrdi-m da je Jovan ne(t)koga uvredio. not claim-1SG that is John someone insulted

3.2.3. Subject /object asymmetry in English English provides yet another piece of evidence for the claim that licensing by Operator patterns with licensing by superordinate negation. Although

66

Non-negative polarity contexts

local negation fails to license a subject NPI, both negation in the superordinate clause and non-negative polarity contexts render the subject NPI acceptable: (28) (29) (30)

*Anyone didn't invite John. I don't think that anyone invited John. Did anyone invite John?

3.3. Operator in Comp It remains to be explained why in both English and Serbian/Croatian non-negative NPI licensing patterns with licensing from a superordinate clause. To this end, let us assume that all non-negative polarity environments discussed so far involve a null polarity operator (Op) which is capable of licensing NPIs. I will take Op to be located in the head of Comp, in keeping with the recent proposals that negation is a head element (see Pollock (1989), Laka (1990), etc.).2'3 (31)

CP ^Op

IP

I will further assume that Op is a clitic which must attach to lexical material in Comp, either a complementizer or an inverted verb. This would provide motivation for inversion in polarity contexts (see section 3.5.1). In fact, Serbian/Croatian seems to have an overt counterpart of Op, which is easily demonstrated to be a Comp clitic: the particle li. It appears with yes/no questions and conditionals, cliticizing either to a complementizer or to an inverted verb: (32)

(33) (34) (35)

Da // Marija dolazi? that Mary comes 'Is Mary coming?' Dolazi // Marija? comes Mary Ako (//) Marija dodje, proslavi-ce-mo. If Mary comes celebrate-will-lPL Dodje li Marija, proslavi-ce-mo. comes Mary celebrate-will-1 PL

Operator in Comp 67 It can never precede Comp elements: (36) (37) (38) (39)

*Li da Marija dolazi? *Li dolazi Marija? *Li ako dodje Marija, proslavicemo. *Li dodje Marija, proslavicemo.

Nor can it attach to a non-Comp element: (40)

*Marija // dolazi?

As shown in Laka (1990), Basque also has an overt reflex of the polarity operator. In Basque, verbs which license NPIs must appear with the negative complementizer enik, if there is an NPI in their complement: (41)

Amaiak [inork gorrotoa dio-m'A;] ukatu du Amaia anyone hatred has-that denied has 'Amaia denied that anybody hated her.'

Below are the proposed representations for different non-negative polarity contexts: Yes/no Questions: (42) [Cp [c Has Op [rP anyone come?]]] Conditionals: (43) [Cp [c if Op [ip anyone comes]]], let me know. Adversative Predicates: (44) I doubt [cp [c that Op [IP anyone has come.]]] Universal Quantifiers: (45) [NP Every man [Cp who [c has Op [iP read anything by Chomsky]]]] will attend the lecture. Op may be licensed in (42) and (43) by virtue of unfixed truth values. More precisely, one might say that in unselected contexts, an Op is licensed in a clause whose truth value is not set positively. This would allow for an Op in counterfactual conditionals: (46)

[Cp Had Op [IP anyone misbehaved], we would have left.]

In embedded contexts, however, Op must be selected by the matrix predicate, as in (44), or by a quantifier, as in (45). One fact seems to speak in favor of the selection analysis. As Marc Authier has pointed out

68 Non-negative polarity contexts (personal communication), some adversative predicates (e.g., forget) optionally select either a wh-word or Op: (47)

Mary forgot [where Peter put her books].

(48)

Mary forgot [that anybody visited her on Monday].

However, if a wh-Comp is selected, an NPI is no longer licit:4 (49) *Mary forgot [where anybody put his books]. If it is true that there is a polarity operator in the Comp of yes/no questions, if-clauses, and adversative complements, why is it the case that this operator licenses only I-NPIs, and not NI-NPIs? In order to be consistent with our analysis that I-NPIs are anaphoric pronominals, which have to be free in their governing category, we would have to assume that this Op somehow falls outside of the governing category for polarity items. This would also capture the fact that NI-NPIs are not licensed by this Op, and that PPIs are not incompatible with it. But how is it possible for Op to fall outside of the governing category for polarity items? According to the Generalized Binding framework, whtraces are A'-anaphors that have to be bound in their governing category from Comp (see section 0.2.3). Obviously, Comp has to be within their governing category, not outside it. There is one obvious difference between wh-traces and polarity items: the first potential antecedent (FPA) for the former is Comp, whereas the FPA for the latter is Infl or head of NegP. Assuming a relativized governing category (see Huang (1982), Aoun (1985) and Chomsky (1986b) and section 0.2.1. for Relativized SUBJECT), the governing category for the wh-trace will be the minimal maximal projection containing the potential antecedent and the anaphor, i.e., CP. For NPIs, on the other hand, the first potential antecedent will be negation within IP, Infl or NegP, thus rendering IP the governing category for NPIs. This way we obtain the desired result: the null operator in Comp will be outside of the governing category for polarity items. That this is not just a convenient manoeuvre is confirmed by the following contrast from Serbian/Croatian (see also footnote 5, Chapter 1): (50)

Zar *i(t)koj ni(t)ko od vas ne zna kako se to radi? Q anyone no one of you not knows how self that does Ts it possible that none of you knows how to do it?'

Operator in Comp 69 (51)

Da ne zna i(t)kol*ni(t)ko od vas kako se to radi? that not knows anyone no one of you how self that works 'Do any of you know how to do it, by any chance?'

Both (50) and (51) are negative yes/no questions, but they differ both in interpretation and NPI licensing. While in (50) the predicate know is negated, it is not in (51). While (50) licenses NI-NPIs, (51) licenses INPIs. Where do these differences come from? Both zar and da are complementizers, and it is highly unlikely that they have any effect on these differences. The only other structural difference between (50) and (51) is in the position of negation: ne is in its usual (Infl) position in (50), following the subject, but is in a pre-subject position in (51). Actually, both the negative particle and the verb precede the subject in (51), and I will assume that they are adjoined to Comp. If so, the negative particle will only be able to license I-NPIs in (51) since it falls outside of the local IP, just established as the governing category for NI-NPIs. On the other hand, since negation in (50) is IP internal, it will license NI-NPIs. The significance of the contrast above is twofold. First of all, it shows that the determining factor in NPI licensing is the position of the licenser, rather than the type of licenser, making it possible to unify operator licensing with negation licensing. Furthermore, it shows that polarity operators may well be negative, and that a unification of the class of polarity licensers can be pushed even further: it could reduce to negation only (see Progovac (1988) for arguments in this direction, and also Laka (1990) for adversative predicates). In other words, it may be possible to argue that negation can either be in Infl (or in some other position within IP, see section 1.2.1), in which case the clause receives negative interpretation, or negation can be in Comp, in which case the truth value of the clause is only indeterminate.5 This is exactly what the contrast between (50) and (51) shows. I will continue to take a weaker stand in this monograph, i.e., that the operator involves a + /- switch, but will leave the stronger (and much more interesting) position for future research. Only now can we explain the mysterious facts about non-negative licensing: why it tolerates both PPIs and NPIs, and why it licenses only INPIs, but not NI-NPIs. Since the null polarity operator falls outside of the governing category for polarity items, it patterns with superordinate, rather than clausemate, negation.

70 Non-negative polarity contexts Notice that this binding approach to polarity sensitivity allows us to capture the fact that NPIs and PPIs are not always in complementary distribution in a principled way. This is a considerable improvement over Ladusaw's analysis, which stipulates that PPIs are disallowed only by clausemate negation, and not by the whole set of downward entailing expressions which license NPIs. Given this analysis, which recognizes locality requirements, it can be maintained that all the NPI licensers are also PPI anti-triggers. If both NPIs and PPIs are allowed in certain contexts, it is because the licenser is outside of the governing category for PPIs, and because NPIs are allowed to raise at LF, thus extending their governing category. In the next section I argue that non-negative licensing indeed must involve Comp. 3.4. Non-negative licensing from Comp In the previous section I proposed that there is an Op in the Comp of non-negative polarity contexts in order to maintain the binding analysis for NPIs and PPIs. In this section I will provide evidence that the licenser for polarity items in non-negative contexts must indeed reside in Comp. This is unexpected under Ladusaw's (1980, 1982, 1983) view that NPIs are licensed directly in the scope of a downward entailing (DE) operator. However, it is consistent with the Operator-in-Comp-proposal, which makes the following strong prediction: licensing of NPIs in non-negative contexts is a property of clauses, rather than phrases, since only clauses have Comp positions. Evidence from licensing by adversative predicates, universal quantifiers, comparatives, the determiner only and ECM verbs gives support for this claim. 3.4.1. Adversative predicates Ladusaw's approach rightly predicts that the adversative predicate forget will be an NPI licenser, as in (54), since it is a DE operator (see (52) and (53)), however, it fails to capture the ungrammaticality of (55): (52) (53) (54) (55)

John forgot that Mary ate a vegetable. John forgot that Mary ate kale. I forgot that anyone dropped by yesterday. *I forgot anything.

Non-negative licensing from Comp

71

The contrast between (54) and (55) follows directly from the assumption that forget selects an Op in Comp, and that this Op does the licensing: no clausal projection intervenes between forget and the NPI in (55), leaving no possibility to generate Op. For more examples with adversative predicates and negative NPs, see Progovac (1988) and Laka (1990). 3.4.2. Comparatives In Serbian/Croatian, there are two types of comparatives: clausal and prepositional: (56)

(57)

Marija je visa nego (sto je) Jovan. Mary is taller than what is John 'Mary is taller than (is) John.' Marija je visa od Jovan-a. Mary is taller from John-GEN

Only clausal comparatives license NPIs (see also Hoeksema (1983) for comparable facts from Dutch): (58)

(59)

Marija je visa nego i(t)ko u razredu. Mary is taller than anyone in class 'Mary is taller than anyone in the class.' *Marija je visa od i(t)ko-ga u razredu. Mary is taller from anyone-GEN in class

Only free-choice items are licit in prepositional comparatives: (60)

Marija je visa od bilo (t)ko-ga u razredu. Mary is taller from FC-anyone-GEN in class

3.4.3. Universal quantifiers Universal quantifiers exhibit a similar clausal/phrasal asymmetry: (61) (62)

Every man [CP who owns any guns] must report to the police station. #Every man [PP with any guns] must report to the police station.

Example (62) may be grammatical only on the free-choice reading of any (see Chapter 7 for definition and discussion). In Serbian/Croatian, which

72

Non-negative polarity contexts

has separate lexical items for NPIs and free-choice items (FCIs), only the latter are possible in the PP: (63)

(64)

*Svaki covek [PP sa ikakvim oruzjem] mora se javiti u every man with any-NPI guns must self report in policiju. police Svaki covek [PP sa bilo-kakvim oruzjem] mora se javiti u every man with any-FCI guns must self report in policiju. police

Example (63) contrasts with the grammatical: (65)

Svaki covek [CP koji poseduje ikakvo oruzje] mora se every man who owns any-NPI guns must self javiti u policiju. report in police

Also, substituting any for an NPI which cannot have a free-choice reading in English (e.g., ever, a single) is only possible with clausal complements, on the relevant reading (see Laka (1990) for the test): (66) (67)

Every man [CP who owns a single gun] must report to the police station. #Every man [PP with a single gun] must report to the police station.

While in (66) the item has an existential reading equivalent to "at least one gun", no such reading is available in (67). In (67) a single gun implies "exactly one gun", which is not a polarity reading. (Irrelevantly, this reading is also available in (66)). The reason why NPIs are licensed in relative clauses, but not in PPs, follows from the fact that only the former involve a Comp position. Licensing by negative prepositions poses a potential problem for this analysis: (68)

Mary left [without any warning].

I will assume that negative prepositions select a clause with a Comp projection. As opposed to positive PPs, they can easily be extended into an obvious clausal structure (see Stowell (1981) and Milsark (1988) for the assumption that gerunds can have Comp):

Non-negative licensing from Comp 73 (69) (70)

Mary left without saying a word. *Mary left with saying a word.

Some support for the assumption that a Comp position is involved in such expansions comes from the corresponding Serbian/Croatian examples with the complementizer da: (71)

(72)

Marija je otisla bez [CP da je rekla ijednu rec]. Mary is left without that is said any word 'Mary left without saying any word.' *Marija je otisla sa da je rekla jednu rec. Mary is left with that is said one word

One may wonder why negative prepositions should differ from their positive counterparts in this respect. If Rizzi's (1990a) assumption is correct that negation has to be in a scopal position, the facts above would follow. A negative preposition requires an Infl/Comp projection in order to realize its negative value. 3.4.4. Only The behavior of the determiner only seems to further support the initial claim that non-negative polarity licensing must be from a Comp position. Only behaves somewhat curiously with respect to NPI licensing. Although judgements tend to be uncertain, the following descriptive pattern emerges: only can license NPIs only from subject or Comp position:6 (73) (74) (75) (76) (77)

[IP Only Mary showed any respect for the visitors.] [CP Only to his girlfriend [C did [IP John give any flowers.]]] [CP Only last year [C did [IP John get any grey hairs.]]] ?*John gave only his girlfriend any flowers. ?*John told only Mary about any books.

Notice that this restriction does not apply to overt negation (see Barss and Lasnik (1986)): (78)

John gave no girlfriend any flowers.

Here I will question the assumption that only is in the subject position in (73), at least not in LF. I propose that only can become a licenser only if

74 Non-negative polarity contexts raised to Comp, at S-Structure or at LF. Notice that 6>«/y-subjects can never be preposed to Comp at S-structure: (79)

*Only Mary did see John.

This bears striking resemblance to w/2-movement. Subject wh-words always remain in situ at S-structure, judging by the absence of inversion:7 (80)

*Who did t see John?

The above contrasts minimally with object extraction: (81)

Who did Mary see t?

According to Chomsky (1986a), vacuous movement, i.e., movement to an adjacent position, need not apply until LF. If true, this would also generalize to only movement: the only phrase would be in Comp at LF.8 As a result of Spec-Head Agreement in Comp (see Chomsky (1986a), Schneider-Zioga (1987), Rizzi (1990a), Rooryck (1991)), the covertly negative only will license Op in the head of CP, which, in turn, will license NPIs.9 Notice that vacuous movement is granted only to adjacent positions, correctly predicting that the following extraction of the ECM subject (brought to my attention by a reviewer) will trigger inversion in the matrix clause: (82)

Only Bush would John believe [t able to govern].

The grammaticality of (83) below will follow from the assumption that CP can be iterated (see section 3.5.1. for details): (83)

John claimed that only Mary showed any respect for the visitors.

Only Mary would be in the specifier of the lower CP, licensing an Operator in the head of the same projection. 3.4.5. ECM verbs While opinions are divided on whether infinitivals and gerunds have Comp or not, it seems to be generally assumed that ECM complements cannot host a Comp position (see Chomsky (1981)). This is because an intervening Comp would prevent exceptional case-assignment to the embedded subject:

Independent evidence for polarity operator 75 (84)

John believes [IP Peter to be a fool].

If we are correct in claiming that adversative predicates select an operator in Comp, we predict that there will be no adversative predicates that can assign exceptional case. This seems to be confirmed in the following minimal contrasts involving adversative predicates and their nonadversative counterparts: (85) (86)

?*John doubts Peter/anyone to be a fool. John believes Peter to be a fool.

(87) (88)

?*John denied Mary/anyone to be the best candidate. John affirmed Mary to be the best candidate.

The proposal that adversative predicates select an operator in the embedded Comp can explain the otherwise curious asymmetries above. Examples (85) through (88) also show that a negative Operator/feature must be realized in the Comp position, rather than in some position lower than IP, contrary to the conclusion reached in Frank and Kroch (1992). If a negative feature could be realized on the head of the complement, in this case on Infl (see Laka (1990)), there would be no reason for the existence of the adversative/non-adversative contrasts exemplified above. 3.5. Independent evidence for polarity operator

There are two pieces of independent evidence for the existence of an Operator in Comp of non-negative polarity environments. The evidence comes from subject-verb inversion and epistemic modals. 3.5.1. Inversion

Subject-Auxiliary Inversion (SAI) in English has the effect of switching the order of the subject and the first auxiliary. Typically, it applies in whquestions and with negative preposing: (89) (90)

Why did he complain about his salary? Never has he complained about his salary.

It is generally held that inverted verbs occupy the head position of CP, C, and that they are therefore mutually exclusive with overt complementizers across different languages (see Koster (1975), den Besten (1978), Koopman (1983) and Radford (1988) for additional references). I am

76 Non-negative polarity contexts also assuming with Emonds (1976), Chomsky (1986a), Radford (1988), Rizzi (1990a), Weerman (1989) and others, that Negative Preposing has Spec of CP as its landing site. For arguments dealing with the incompatibility of wh-question formation and negative preposing, see Emonds and Radford. The sentence (91) below, brought up by a reviewer, raises a problem for these assumptions: (91)

John swore that never again would he dance the polka.

One would have to assume here that the embedded CP node is iterated, following Authier (1992) for English, De Haan and Weerman (1986) for Frisian, Platzack (1986a and b) for Swedish, Danish, Norwegian, and Icelandic, Adams (1987) for Old French (see also Laka (1990) for a SigmaP below Comp and Culicover (1991) for Polarity Phrase below Comp). This move is necessary in order to provide enough space for both the complementizer that (in the first C), and the fronted negative phrase and would (in the Comp of the second CP). In (91), obviously, the complementizer may co-occur with the inverted modal, suggesting that there are two C's in the sentence. Suppose that SAI is triggered whenever there is a clitic-like element in the Head of CP. The verb moves there in order to support the clitic. The clitic can either be the Op, or AGR with an element in the Spec. For example, movement of a wh-element to the Spec of CP will trigger whAGR in the head of CP, and the verb will have to move in order to support the AGR clitic (see (89)).10 Likewise, a moved negative phrase would trigger negative AGR in Comp, which also requires a verb to support it (see (90)). In the same vein, an Op in Comp will require inversion in order to be supported by the verb: (92) (93)

Op Did he complain about his salary? Op Had he complained about his salary, he would have received a raise.

Notice that a Polarity Operator in the Comp of yes/no questions has an important advantage over the old Q operator (see Katz and Postal (1964) and Baker (1970a)). One can get an interrogative interpretation even without inversion, given the right intonation: (94)

He complained about his salary?

Independent evidence for polarity operator 11 Also, as noted in Grimshaw (1977) and Weerman (1989), a Q operator is not essential for deriving an interrogative reading. On the other hand, inversion is a necessary prerequisite for licensing of NPIs: (95) (96)

?*He complained about anything! Did he complain about anything]

In other words, we now have independent motivation for an operator in Comp: licensing of NPIs. 3.5.2. Epistemic modals According to McDowell (1987), epistemic modals raise to Comp at LF thus taking wide scope over the whole proposition. Thus, epistemic modals will be in Comp in LF whether or not they are inverted at Sstructure. The LF representation of (97) will be (98): (97) (98)

Someone must have killed Yuri. [Cp musti [IP someone ti have killed Yuri.]]

She observes that epistemic interpretations are incompatible with whquestions:11 (99)

?*Who must have killed Yuri?

The raising of modals is not permitted in wh-questions because their Comp position is already occupied by a contentful element, wh-AGR. To illustrate that the ungrammaticality of (99) is not due to a pragmatic incompatibility of asking for information and asserting a high probability, McDowell gives the following acceptable discourse: (100) Someone must have killed Yuri, because it clearly wasn't suicide, but who? An even more convincing argument that the whole phenomenon cannot be pragmatic comes from negative preposing, as evident from the following contrast: (101) *Never must John have loved Mary. (102) John must never have loved Mary. According to McDowell, yes/no questions, if-clauses, and adversative predicates are all incompatible with epistemic modals:12

78 Non-negative polarity contexts (103) ?*Must John know the answer? (104) *If John must know the answer, let's ask him. (105) *I doubt that John must know the answer. Our analysis, in conjunction with McDowell's, offers an explanation for the ungrammaticality of the above sentences: if all these contexts involve a null operator situated in Comp, then the epistemic modal cannot raise there. Although Op may share the Comp position with the complementizer that or with an inverted auxiliary, it cannot do so with a raised epistemic modal (nor with wh-agreement, see Chapter 5). As pointed out by a reviewer, this difference would follow from the fact that epistemic modals (and wh-agreement) need scope over the clause, thus resulting in a scope clash with the operator, which also takes scope over the clause, determining its truth value. Neither complementizers nor inverted auxiliaries need to be interpreted in Comp, and can thus co-occur with Op. In sum, I have argued for the existence of a null operator in Comp of non-negative polarity contexts on the basis of two kinds of independent evidence: the fact that they trigger inversion and that they are incompatible with epistemic modals.

4 Language variation

The main empirical advantage of the binding approach to polarity sensitivity is in its ability to predict variation in NPI-licensing across languages. In this Chapter I will argue that a wide range of attested variation can be captured by the interplay of two separate factors: different raising options (section 4.1.) and being subject to different binding principles (section 4.2). Along the raising dimension, some NPIs do not raise at LF at all, e.g., NI-NPIs in Serbian/Croatian and strict NPIs in English (section 4.1.1). Some NPIs raise either by IP-adjunction or by movement through Comp, e.g., English NPIs (section 4.1.2). Some NPIs can move only by IP-adjunction, e.g., certain NPIs in Romance and Chinese (section 4.1.3). Finally, some NPIs can only move through Comp, e.g., Turkish NPIs (section 4.1.4). Parametrization of landing sites also proves useful in other instances of A-bar movement, such as topicalization, wh-movement, and scrambling (see Miiller and Sternefeld (forthcoming a, forthcoming b)). Along the binding dimension, most NPIs are subject only to Principle A of the Binding Theory (section 4.2.1). Some NPIs, however, are subject to Principle B, e.g., Serbian/Croatian I-NPIs (section 4.2.2). PPIs, on the other hand, can be subject either to Principle B only, as in English and Serbian/Croatian, or to Principle C (which subsumes Principle B), as in Russian (section 4.2.3).

4.1. Raising options 4.1.1. No raising

As a first approximation, let us assume that the raising factor is responsible for the following difference among NPIs with respect to their ability to be bound long distance: 79

80 (1)

Language variation -Some NPIs can raise at LF (e.g., English NPIs); -Other NPIs cannot raise at LF (e.g., Serbian/Croatian NI-NPIs).

Those NPIs that cannot raise at LF are predicted to occur only with local negation, and never in non-negative polarity contexts or with superordinate negation. This is because the governing category for NPIs is IP, given that their first potential antecedent is negation in Infl or NegP. Since both the polarity operator in Comp and superordinate negation fall outside of this domain, NPIs cannot be bound to them in their governing category. Recall (from section 1.2.3.1) that NI-NPIs in Serbian/Croatian are licensed only by clausemate negation: Clausemate negation: (2) Marija ne voli ni(t)ko-ga. Mary not loves no-one-ACC 'Mary does not love anyone.' Superordinate negation: (3) *Ne veruje-m da Marija voli ni(t)ko-ga. not claim-1SG that Mary loves no-one-ACC T do not claim that Mary loves anyone.' Yes/no questions: (4) *Da li Marija voli ni(t)ko-gal that Q Mary loves no-one-ACC 'Does Mary love anyone?' In sections 1.2.4. and 2.2. I offered a possible explanation for the fact that NI-NPIs cannot raise. While English any is a QP constituent which can freely raise at LF, ni is not a constituent at all: (5)

DP DP D

NP

any (6)

one DP

NegP

DP OP

DP D

n

i

NP (t)ko

Raising options

81

There are also other NPIs which show the same distribution as NI-NPIs but which do not involve morphological negation: the Chinese NPI conglai and English strict NPIs. 1 Conglai ( = ever) is licensed only by clausemate negation, and never in any other polarity environment (Audrey Li, personal communication). 2 Clausemate negation: (7) Ta conglai meiyou kanjian Mali, he ever did-not see Mary 'He has never seen Mary.' Yes/no questions: (8) *Ta conglai kanjian Mali ma? he ever see Mary Q 'Has he ever seen Mary?' (9) *Ta conglai kan-mei-kanjian Mali? he ever se-not-see Mary 'Has he ever seen Mary?' Conditionals: (10) *Ruguo ta conglai kanjian Mali qing gaosu wo. if he ever see Mary then tell me 'If he has ever seen Mary, tell me.' Adversative predicates: (11) *Ta jujue conglai gen Mali shuo-hua. he refuse ever with Mary talk-language 'He refused to ever talk to Mary.' Superordinate negation: (12) *Zhangasan meiyou shuo ta conglai kanjian quo Mali. John did-not say he ever see ASP Mary 'John did not say that he had ever seen Mary.' English also has NPIs which behave like NI-NPIs. The so-called strict NPIs (e.g., until) are only licensed by clausemate negation (see Linebarger (1981)): (13) (14) (15) (16)

John did not arrive until seven o'clock. *I am not saying that John arrived until seven o'clock. *Did John arrive until seven o'clock? *If John arrived until seven o'clock, he was in time.

82 Language variation (17)

*Peter denied that John arrived until seven o'clock.

It may well be that conglai and until cannot raise at LF because they lack quantificational force.3 This suggestion will also be brought up with respect to idiomatic NPIs of the kind budge an inch, to be discussed in section 5.3. 4.1.2. Two raising options In Chapter 2 it was proposed that English regular NPIs raise at LF in order to be bound locally. As opposed to non-raising NPIs, English any is licensed in all the polarity environments:4 (18) (19) (20) (21) (22)

John didn't hurt anyone. Mary does not say that John hurt anyone. Did John hurt anyonel If John hurt anyone, he must apologize. I doubt that John hurt anyone.

It is only now that we have established all the potential antecedents for NPIs that we can determine the exact mechanism of NPI-Raising. This is because the governing category for NPIs must be defined on the basis of the first potential antecedent (see section 3.3): (23)

The governing category for NPIs is the first maximal projection containing the NPI and its first potential antecedent.

There are two cases of long-distance licensing which must involve NPIraising: licensing by superordinate negation and licensing by the polarity operator: (24)

Mary does not claim that John hurt anyone.

(25)

Did John hurt anyonel

I will argue that at LF (24) and (25) are represented as follows: (26) (27)

Mary did not claim [Cp anyone^ [cf that [IP John hurt tj]] [cp Op has [IP anyone^ [ip John hurt tj]]

In (26), the NPI raises to the Spec of CP, whereas in (27), it IP-adjoins. The governing category for the NPI in (26) will be the matrix IP since the first potential antecedent is negation within IP. In (27), on the other

Raising options 83 hand, the governing category for the NPI will be the embedded CP since the first potential antecedent will be the Op in the head of CP. In other words the rule of any raising has two options: to move through the Spec of CP, or to adjoin to IP (for IP-adjunction of QPs, see May (1977)). This conclusion follows from both theory-internal and independent reasons. Let us first consider the theory-internal argument. The following structure highlights all the potential antecedents for the underlined NPI: (28)

[CP Op [IP [r NEG [VP [CP Op [IP [r NEG [VP any]]]]]]]]

There are four potential antecedents for any: clausemate negation, Op in the embedded Comp, superordinate negation, and Op in the matrix Comp. If any only had the option to IP-adjoin, it would never be eligible for long-distance binding from the matrix clause. This is because there would always be a potential antecedent in the embedded Comp so that the embedded CP would count as the opaque domain for the IP-adjoined NPI. The only way for an NPI to be licensed by superordinate negation would be for it to move to the embedded Spec of CP. That way, its governing category would extend to the matrix IP, as desired. On the other hand, for an NPI to be licensed by an Op in Comp, it must not move to the Spec of CP. If it did, by the minimality condition (see Chomsky (1986a), Aoun and Li (1989) and Rizzi (1990a)) the trace of the NPI would be bound by Op in Comp, which is a closer A' governor, and the raised NPI would violate the condition on vacuous quantification: (29)

[Cp anyone^ [c Op has [IP John hurt tj]]

On the other hand, if anyone moves only as far as IP, it will minimally bind its trace, and will in turn be bound to Op. There is an important difference between base-generated Op and the raised NPI. While Op is an unselective binder, which can bind more than one NPI, or indeed none, as shown in (30) and (31) below, a raised NPI is subject to the one-to-one mapping requirement between traces and raised elements. In other words, the Op is not subject to the Vacuous Quantification condition, just like an NP which does not bind a reflexive is not (see (32) and (33)): (30) (31) (32) (33)

Op Did Mary give anything to anyone] Op Did Mary see Peter? Peter talked to himself about himself. Peter talked to John.

84

Language variation

Here are some independent reasons for allowing NPIs to move through the Spec of CP. First of all, this is the only way to explain the difference between NPI-Raising and Quantifier-Raising (QR), which only has the option of adjoining: while QR seems clause-bound, NPI-Raising in English is not. As noticed by Longobardi (forthcoming), adjunction to IP allows a quantifier to take scope over the constituent, but it cannot be successive cyclic. The following sentence with two quantificational NPs in different clauses is not ambiguous, indicating that QR cannot apply over a (finite) clause boundary: (34)

Someone said [that everybody was on time].

The sentence does not have a reading where everybody takes scope over someone.

The following examples illustrate that an NPI can be licensed by negation that is more than one clause away, suggesting that movement proceeds through the Spec of CP: (35)

Mary did not say that Peter claimed that John hurt anyone.

In sum, both raising options seem to be granted to the quantiflcational NPIs in English: IP-adjunction and movement through the Spec of CP. Assuming with Muller and Sternefeld (forthcoming a, forthcoming b) that moving from an adjunction site to Spec of CP, or vice versa, would constitute improper movement, I would say that an NPI moves either by adjoining to IP, or to Spec of CP, but once moved to one landing site type, it cannot continue to move to another. In the next section I will argue that both options need not be exploited by NPIs in different languages. Some NPIs can raise by IP-adjoining only, whereas others can raise by moving through the Spec of CP only. This way we will be able to account for attested variation in NPI licensing. 4.1.3. IP-adjunction: Romance and Chinese

On the basis of English and Serbian/Croatian data, we have established that licensing in non-negative contexts patterns with licensing from a superordinate clause (see section 3.2). This fact was captured by postulating a polarity operator in Comp in non-negative polarity contexts (section 3.3). Assuming that the first potential antecedent creates an opaque domain, the polarity operator falls outside of the governing category for polarity items. Our analysis so far predicts that

Raising options

85

whenever a certain NPI is licensed by superordinate negation, it will also be licensed in non-negative contexts. And, conversely, if a certain NPI is not licensed by superordinate negation, it will not be licensed in nonnegative polarity contexts either. This difference among NPIs would be ascribed to raising possibilities: those NPIs that are licensed long distance can raise at LF, and those that are not licensed long distance cannot. However, looking at additional languages falsifies this claim in both directions. First of all, there are NPIs that are licensed in non-negative contexts and by clausemate negation, but not by superordinate negation, e.g., Italian nessuno, Catalan ningu, and Chinese renhe (Li (1985)). With these NPIs, non-negative licensing patterns with clausemate negation, rather than with superordinate negation. On the other hand, there are NPIs that are licensed only by negation, clausemate or superordinate, but not in non-negative polarity environments, e.g., Turkish NPIs (Enq (1988)). Here, clausemate and superordinate negation pattern together, and differ from non-negative polarity contexts. I will argue below that this variation can be captured in the following way: while nessuno, ningu, and renhe can only IP-adjoin, Turkish NPIs can only move through the Spec of CP. I leave it for future research to provide independent motivation for this difference. At the moment, suffice it to say that landing sites can be parametrized for other A-bar movement landing sites, such as topicalization, scrambling, wh-movement, etc. (Miiller and Sternefeld (forthcoming a, forthcoming b)). Italian n-words (e.g., nessuno, niente) are licensed by clausemate negation, and in non-negative polarity contexts, but not by superordinate negation (examples from Zanuttini (1991)):5

Clausemate negation: (36) Non ho visto nessuno. not has seen no-one 'He hasn't seen anyone.

Yes/no questions: (37) Ha telefonato nessunol has called no-one 'Has anyone called?'

86

Language variation

Adversative Predicates: (38) Dubito che venga nessuno. doubt-1SG that comes no-one 'I doubt that anyone will come.' Comparatives: nessuno di voi. (39) Maria lavora di piu di quanto lavori Maria works of more of how much works-SUBJ no-one of you 'Maria works more than anyone of you does.' Superordinate negation: (40) ?*Non ho deciso [che faro niente]. not have decided that do nothing T haven't decided that I'll do anything.' The same is true of n-words in other Romance languages, e.g., Catalan (see Casanova-Seuma (1988)): Clausemate negation: (41) No vull res not want-lSG anything 'I do not want anything.' Yes/No Questions: (42) Ha vingut ningut has come nobody 'Has anybody come?' Conditionals: (43) Si ve ningu avisam. if comes nobody let-me-know 'If anybody comes, let me know.' Superordinate Negation: (44) *No crec que ve ningu. not believe-I that comes-IND nobody 'I do not believe that anyone is coming.' The NPI renhe in Chinese shares the distribution of nessuno and ningu. It is licensed by clausemate negation and in non-negative polarity contexts, but not by superordinate negation:

Raising options 87 Clausemate Negation: (45) Ta bu xihuan renhe dongxi. he not like any thing 'He does not like anything.' Yes/No Questions: (46) ?Ta xi-bu-xihuan renhe dongxi? he li-not-like any thing 'Does he like anything?' (47) Ta xihuan renhe dongxi ma? you like any thing Q 'Do you like anything?' Conditionals: (48) Ruguo ta xihuan renhe dongxi qing gaosu wo. if he like any thing then tell me 'If he likes anything, tell me.' Adversative Predicates: (49) Ta jujue gen renhe ren shuo-hua. he refuse with any one talk-language 'He refused to talk to anyone.' Superordinate Negation: (50) *Wo meiyou gaosu guo ta ni zuo renhe shiqing. I did-not tell ASP he you do any matter 'I did not tell him that you did anything.' Suppose that these NPIs can only IP-adjoin, hence cannot move through the Spec of CP. Since the first potential antecedent for the IP-adjoined NPI will be the polarity operator in the head of CP, the embedded CP will count as its opaque domain. The following structure presents potential antecedents for the IP-adjoined NPI ningu: (51)

[IP [r NEG [VP [Cp Op [IP ningu [IP [r Infl [VP]]]]]]]]

This way, the NPI can never be licensed by superordinate negation because it will fall outside of its governing category. In other words, such NPIs will be licensed by clausemate negation and in non-negative contexts, but not by superordinate negation.

88

Language variation

Since IP-adjunction does not seem to be successive-cyclic (cf. the locality of Topicalization and Quantifier Raising (see also section 4.1.2.)), we predict that NPIs will not be licensed by superordinate questions: (52)

?*Ni gao-mei-gaosu guo Mali Zhangsan xihuan renhe dongxi? you te-not-tell ASP Mali Zhangsan like any thing 'Did you tell Mali that Zhangsan likes anything?'

So far, we have considered the following raising options: (53)

Raising Parameters: 1. No raising (e.g., NI-NPIs, English strict NPIs, and Chinese conglai). 2. Raising by either IP-adjunction or through the Spec of CP (English regular NPIs). 3. Raising by IP-adjunction only (e.g., Italian nessuno, Catalan NPIs and Chinese renhe).

In the next section, I consider the remaining logical option: 4. Raising through the Spec of CP only (Turkish). 4.1.4. Movement through Comp: Turkish In this section I consider the last logical possibility for NPI raising: movement through the Spec of CP only. In order to see what distribution this option would predict, let us look at the following LF representation: (54)

[CP [,P [r NEG [VP [CP NPI, [ c Op [IP [ r Infl [VP V tj]]]]]]]

If the NPI; can raise as far as the embedded Spec of CP, it automatically becomes eligible for binding by matrix negation. Since the first potential antecedent for an NPI in the Spec of CP will be negation in Infl, the matrix IP will count as the opaque domain for the NPI. NPIs will be eligible for licensing by clausemate negation even if they do not raise at LF, or their trace will be bound. What about licensing in non-negative contexts? If the NPI in (54) stays in situ at LF, its first potential antecedent will be the embedded Infl, and its governing category - the embedded IP. Therefore, it could not be bound by an operator in Comp since it falls outside of its governing category. On the other hand, if the NPI moved to the Spec of CP it would induce a minimality violation if there is an Op in Comp (see section

Raising options

89

4.1.2). The minimal binder for the NPI trace would be the Op in Comp, so that the raised NPI would have nothing to quantify over, resulting in vacuous quantification. In other words, if an NPI has only the option of moving through the Spec of CP, we predict that it will have the following distribution: it will be licensed by clausemate and superordinate negation, but not in non-negative polarity contexts. Exactly the predicted distribution is shown by NPIs in Turkish (Enq (1988)).6'7 Clausemate negation: (55) Ali hif kimse-yi gor-me-di. Ali any person-ACC see-not-PAST 'Ali did not see anyone.' Superordinate negation: (56) Ali hiqrhir -§ey-in bozul-dug-u-nu Ali any-a thing-GEN breakdown-GERUND-3sg.-ACC soyle-me-di. say-not-PAST 'Ali didn't say that anything broke down.' NPIs in Turkish are illicit in non-negative polarity contexts:8 Conditionals: (57) *(Eger) Ali hiq kimseyi getirirse, televizyon seyred-eceg-iz. if Ali any person brings TV watch-FUT-we 'If Ali brings anyone over, we will watch TV.' Adversatives: (58) *Zafer Ali'nin hiq kimse-yle evlen-me-si-ne Zafer Ali any person-with marry-GERUND-3SG-DAT surprise 'Zafir was surprised that Ali married anyone.' Universal Quantifiers: (59) *Dibilim hakkinda hiq-bir §ey okumu§ herkes gel-ecek linguistics about any-a thing read everyone come-FUT 'Everyone who read anything about linguistics will come.' This section completes the raising paradigm: (60)

Raising parameters: 1. No raising; 2. Raising either by IP-adjunction or through the Spec of CP;

90 Language variation 3. Raising by IP-adjunction only; 4. Raising through the Spec of CP only. However, this does not exhaust the variation possibilities. The following section explores the binding dimension. As pointed out by a reviewer, it would be desirable to derive, rather than stipulate, the parameters in (60). At this point (60) is a little more than a restatement of facts, hopefully capturing the right generalization. If negative polarity items are a homogeneous class, why should they select different landing sites? First of all, they might not be a homogeneous class at all. Some NPIs are negative (e.g., ni(t)ko in Serbian/Croatian and nessuno in Italian). Other NPIs are homophonous with wh-words, e.g., Chinese sheme (see section 5.1). In fact, Serbian/ Croatian n-i-(t)ko (neg-any-who) consists of three morphemes, including both negation and a wh-word. Perhaps the clue can be sought in their morphological properties. Second, even for elements that are considered a homogeneous class, landing sites may differ cross-linguistically. Rudin (1988a, 1988b) argues convincingly that overt multiple wh-movement in some Slavic languages, including Serbian/Croatian, involves movement to Spec of CP for the first wh-phrase, but IP-adjunction for all the consequent wh-phrases. One piece of evidence for this conclusion is the fact that clitics, which attach to Comp, must immediately follow the first wh-phrase. Likewise, although Topicalization in English seems to involve IP adjunction (see e.g., Lasnik and Uriagereka (1988)), it is generally held that it has Spec of CP as its landing site in Verb-Second languages, e.g., German and Dutch, in which languages Topicalization necessarily triggers verb movement to Comp. Also, Miiller and Sternefeld (forthcoming a, forthcoming b) argue that cross-linguistic variation in scrambling can be captured by parametrizing the choice of landing sites. While German selects VP and IP as possible adjunction sites for scrambling, Russian, in addition, also allows scrambling by adjunction to CP and NP. 4.2. Binding options

The crux of the binding approach to polarity sensitivity are the following claims: (61) (62)

All NPIs must be bound. All PPIs are subject to Principle B of the binding theory.

Binding options

91

However, nothing in our analysis prevents NPIs and PPIs from being subject to other universal principles of grammar. In particular, in this section I will defend the following generalizations: (63)

(64)

NPI Binding Parameter: Most NPIs are subject to Principle A (e.g. Serbian/Croatian NINPIs, English NPIs, Turkish NPIs, Catalan NPIs, Chinese NPIs); Some NPIs are subject to Principle B (e.g. Serbian/Croatian INPIs).9 PPI Binding Parameter: Some PPIs are subject to Principle B only (PPIs in English and Serbian/Croatian); Some PPIs are also subject to Principle C (PPIs in Russian).

4.2.1. NPIs: Principle A This section reviews only those NPIs which are subject to Principle A and which cannot raise at LF (the distribution and analysis of the raising NPIs was given in section 4.1). These include: Serbian/Croatian and Russian NI-NPIs, Chinese conglai, and strict NPIs in English, as discussed in section 4.1.1.10 Recall from section 1.2.3.1. that NI-NPIs in Serbian/Croatian are licensed only by clausemate negation: Clausemate negation: (65) Marija ne voli ni(t)ko-ga. Mary not loves no-one-ACC 'Mary does not love anyone.' Superordinate negation: (66) *Ne veruje-m da Marija voli ni(t)ko-ga. not believe-1SG that Mary loves no-one-ACC 'I do not believe that Mary loves anyone.' Yes/no questions: (67) *Da li Marija voli ni(t)ko-gat that Q Mary loves no-one-ACC 'Does Mary love anyone?' The same holds of Russian NPIs (see also section 0.1.1):

92

Language variation

Clausemate negation: (68) Maria ne vidit niko-go. Mary not sees no-one-ACC 4 Mary does not see anyone.'

Superordinate negation: (69) *Peter ne skazal, [cto Maria vidit niko-go]. Peter not says that Mary sees no-one-ACC

Adversative predicates: (70) *Ja somnevayus', [cto Maria vidit niko-go]. I doubt that Mary sees no-one-ACC This distribution is captured automatically if one assumes that NI-NPIs are subject to Principle A. Since their first potential antecedent is negation in Infl (or NegP), their governing category is the first maximal projection containing Infl and the NPI, that is, IP. This explains why they can only be licensed by negation within IP. Both superordinate negation and the polarity operator in Comp fall outside that domain, and therefore fail to bind NI-NPIs in their governing category. For Serbian/Croatian and Russian, one might argue that there is an alternative explanation, in particular, that NI-NPIs are the result of 'negative concord' which forces NPIs in the scope of a clausemate negation to receive the feature [ +Negative].11 This would render licensing in Serbian/Croatian parallel to licensing in English: NPIs (INPIs) would be allowed whenever in a polarity environment, except that in Serbian/Croatian, unlike in English, there is a negative concord rule which transforms I-NPIs in the scope of clausemate negation into NINPIs (but see footnote 5, Chapter 1 for a problem for such a view). More conclusive evidence that NPIs are indeed subject to Principle A of the Binding Theory would come from those NPIs which do not have morphological negation but show the same distribution as NI-NPIs. The Chinese NPI conglai ( = ever), for example, is such an NPI. It is licensed only by clausemate negation, and never in any other polarity environment. Since it involves no morphological negation, a negative concord analysis cannot be responsible for its distribution. I repeat the relevant facts from section 4.1.1:

Binding options

93

Clausemate negation: (71) Ta conglai meiyou kanjian Mali, he ever did-not see Mary 'He has never seen Mary.' Yes/no questions: (72) *Ta conglai kanjian Mali ma? he ever see Mary Q 'Has he ever seen Mary?' (73) *Ta conglai kan-mei-kanjian Mali? he ever se-not-see Mary 'Has he ever seen Mary?' Conditionals: (74) *Ruguo ta conglai kanjian Mali qing gaosu wo. if he ever see Mary then tell me 'If he has ever seen Mary, tell me.' Adversative predicates: (75) *Ta jujue conglai gen Mali shuo-hua. he refuse ever with Mary talk-language 'He refused to ever talk to Mary.' Superordinate negation: (76) *Zhangasan meiyou shuo ta conglai kanjian quo Mali. John did-not say he ever see Asp Mary 'John did not say that he had ever seen Mary.' The distribution of conglai is exactly the same as that of NI-NPIs, although it does not involve morphological negation. Conversely, there are also NPIs (e.g., in Romance) which are morphologically negative but nonetheless licensed in non-negative polarity contexts with the meaning of any (see also section 4.1.3): Italian: (77) Ha telefonato nessunol has called no-one 'Has anyone called?'

94

Language variation

Catalan: (78) Ha vingut ningul has come nobody 'Has anyone come?' (79) Si ve ningu avisam. if comes nobody let-me-know 'If anyone comes, let me know.' As established in section 4.1.1, strict NPIs in English (e.g. until) are also only licensed by clausemate negation (see Linebarger (1981)): (80) (81) (82) (83) (84)

John did not arrive until seven o'clock. *I am not saying that John arrived until seven o'clock. *Did John arrive until seven o'clock? *If John arrived until seven o'clock, we'll consider his case. *Peter denied that John arrived until seven o'clock.

In sum, NI-NPIs, conglai, and English strict NPIs behave like English reflexives and reciprocals in that they have to be bound in their governing category. The simplest way to account for their distribution would be to assume that they are subject to Principle A of the Binding Theory, and that they cannot raise in LF. 4.2.2. NPIs: Principle B It was proposed in section 1.2.4. that I-NPIs are subject to two binding requirements: Principle B and a general requirement to be bound in the sentence. This entails that I-NPIs are never licensed by clausemate negation, but are licit in other polarity environments (see also section 1.2.3.2): Clausemate negation: (85) *Marija ne voli i(t)ko-ga. Mary not loves anyone-ACC Superordinate negation: (86) Ne tvrdi-m da Marija voli i(t)ko-ga. not claim-1SG that Mary loves anyone-ACC 'I do not claim that Mary loves anyone.'

Binding options

95

Yes/no questions: (87) Da li Marija voli i(t)ko-gat that Q Mary loves anyone-ACC 'Does Mary love anyone?' Conditionals: (88) Ako Marija prepozna i(t)ko-ga, bi-ce dobro. if Mary recognizes anyone-ACC be-FUT good 'If Mary recognizes anyone, everything will be O.K.' Adversative predicates: (89) Sumnja-m da Marija voli i(t)ho-ga. doubt-1SG that Mary loves anyone-ACC 'I doubt that Mary loves anyone.' I-NPIs are licit in (86) through (89) because superordinate negation and the polarity operator in Comp fall outside of the governing category for NPIs (IP), so that I-NPIs remain free in their domain. At the same time, I-NPIs are bound by superordinate negation or by the polarity operator in the sentence. Example (85) is ungrammatical because the I-NPI is bound by local negation in its governing category. 4.2.3. PPIs: Principle B or C?

It seems that PPIs too can be subject to cross-linguistic variation. Given the binding approach to polarity, one additional option for PPIs is predicted. By definition, PPIs (i.e., POSITIVE polarity items) are antitriggered by negation. One possibility is for them to be anti-triggered by a clausemate negation only, in which case they obey Principle B, as is true of English and Serbian/Croatian PPIs. Another possibility not mentioned so far is for PPIs to be subject to Principle C as well. Such PPIs would be disallowed by all the polarity licensers: clausemate negation, superordinate negation and the polarity operator. As shown in Brown (1989), Russian pronoun kto-to shows exactly these properties.12 It is parallel to English someone in non-polarity contexts being ambiguous between a specific and a non-specific reading (all the examples are from Brown (1989)): (90)

Kto-to prisel segodnja utrom. someone came today in-the-morning

96

Language variation

'Someone (a certain person)/ Someone (anyone) came this morning.' In all the polarity contexts, however, kto-to can only receive a specific interpretation: Clausemate negation: (91) #Ja kogo-to ne videla. I someone not saw 'I didn't see someone (a certain person).' Superordinate negation: (92) #Maria ne govorit, cto ona cto-to znaet o Mary not say that she something knows about svojem prijatele. her (own) friend 'Mary doesn't say that she knows (that certain) something about her friend.' Yes/no questions: (93) #Ty kogo-to videl? you someone saw 'Did you see (that certain) someone?' Conditionals: (94) #Esli by ja cto-to znal o Dime, ja by kazala tebe. if COND I something knew about Dima I COND tell you 'If I knew (that certain) something about Dima, I would tell you.' Adversative Predicates: (95) #Ja somnevajus', cto ona cto-to znaet ob Ivane. I doubt that she something knows about Ivan 'I doubt that she knows (that certain) something about Ivan.' So, the only requirement imposed on PPIs is for them to be obviative from clausemate negation. The rest is a matter of choice: PPIs can be subject to either Principle B or Principle C (which subsumes Principle B). Therefore, the PPI universal has to take the following disjunctive form: (96) PPI Universal: All PPIs are subject to Principle B of the Binding theory. (97) PPI Binding Parameter: Some PPIs are, in addition, subject to Principle C (which subsumes Principle B).

5 Rhetorical questions

5.1. Wh-questions This chapter will be concerned with rhetorical questions of the following type: (1) (2)

When did Mary insult anyone? Who did Mary ever kiss on the first date?

Here, the rhetorical force is induced by the presence of two items: a wh-word and an NPI. Rhetorical force is only optionally available in wh-questions with no NPIs (3 and 4), or in yes/no questions with NPIs (5 and 6): (3) (4) (5) (6)

When did Mary insult Peter? Who did Mary kiss on the first date? Did Mary insult anyone! Did Mary kiss anyone on the first date?

However, the rhetorical reading is obligatory, rather than optional, in wh-questions with NPIs, suggesting that there is something about the combination of the two that forces rhetorical readings (see Borkin (1971) and Lawler (1971)). Examples (3) and (4) permit a range of answers: yesterday, last week, etc. (for (3)) and Peter, John, etc. (for (4)). However, as soon as NPIs are inserted, as in (1) and (2), the wh-word ranges over an empty set, as paraphrased in (7) and (8): (7) (8)

Never did Mary insult anyone. No one did Mary ever kiss on the first date.

The same rhetorical readings are enforced in Serbian/Croatian whquestions containing NPIs, suggesting that this is not a language-specific phenomenon: 97

98 (9)

(10)

Rhetorical questions Kada je Marija i(t)koga uvredila? When is Mary anyone insulted 'When did Mary insult anyone?' (T)Ko-ga je Marija ikada poljubila na prvom sastanku? Who-ACC is Mary ever kissed on first date 'Who did Mary ever kiss on the first date?'

Here I will not be concerned with the possibility of rhetorical readings in general, which may be a pragmatic factor, but will rather try to explain the necessity of such readings in questions containing both NPIs and whwords. I will argue that obligatory rhetorical force in wh-questions with Negative Polarity Items (NPIs) is a direct consequence of Spec/Head AGR in Comp. This simple result can be achieved only if we accept the necessity of a polarity Operator in Comp to license NPIs, as established in section 3.3. Descriptively speaking, two things happen if an NPI is inserted in a wh-question (see (7) and (8)): - the wh-phrase loses its wh-force; - it is interpreted as a negative expression, more precisely, a negated NPI {never= neg + ever; no one = neg + anyone).

Why should the presence of an NPI induce such changes? I assume that a wh-phrase in the Spec of CP triggers wh-AGR in the head of CP (for Spec/Head AGR in Comp see Chomsky (1986a), Schneider-Zioga (1987), Rizzi (1990a), Rooryck (1991), etc.). Thus (1) is represented as follows: (11)

[Cp when [a AGRwh did [IP Mary insult anyone]]]

On the other hand, an NPI is licensed by a polarity Operator (Op) in Comp (see 3.3). If Op is there, as evidenced by the presence of inversion in a question, an NPI is licensed; however, if there is no Op, NPIs are illicit: (12) (13)

Op Did Mary invite anyone! *Mary invited anyone! el

In order to derive rhetorical force in wh-questions it is enough to assume that wh-AGR and Op in Comp are incompatible, as in (14) below, both requiring a separate interpretation in the Comp position (see also section 3.5). This is not to say that wh-questions cannot license NPIs, since they

Wh-questions 99 obviously can, as in (1) and (2). It is only to say that this incompatibility will lead to a special reading, actually the one where wh-agreement is suppressed. (14)

*[CP when [C AGRwh Op did [IP Mary insult anyone]]]

That this incompatibility is real can be demonstrated with verbs like forget which select either AGRwh or Op:1 (15) (16)

I forgot who dropped by. I forgot that anyone dropped by.

Crucially, however, if wh-AGR is selected, NPIs become illicit (Marc Authier, personal communication): (17)

*I forgot when anyone dropped by.

In order to salvage (14), two things can happen. One possibility is to suppress the Op. But in that case the sentence would be ungrammatical since there is no licenser for the NPI. The only remaining option is to suppress wh-AGR in Comp, resulting in the loss of the wh-force. This process would have to take place after wh-movement and inversion. But if the wh-word is not a wh-word any longer, what else can it be? It is well known that Spec of CP in English can also host negative expressions (see section 3.5.1): (18)

[CP Never [C has [IP Mary insulted anyone.]]]

So, the only other option for the wh-word is to be interpreted as a negative NPI. It is true cross-linguistically that wh-words can serve as NPIs, e.g., in Chinese and Serbian/Croatian: (19)

(20)

Ni xiang chi sheme ma? you like eat what Q 'Would you like to eat anything?' Da li je Milan (\-)sta doneo? that Q has Milan any-what brought 'Has Milan brought anything?'

Since only negated NPIs are tolerated in the Spec of CP (cf. (21)), Op in Comp must set its switch to the negative value, and due to Spec/Head AGR, the wh-word gets interpreted as a negated NPI, by a process of absorption, or amalgamation, in Cinque's (1990) terms (cf. (22) and (23)).

100 Rhetorical questions I am assuming that Op involves a + / - switch (see also discussion in section 3.3): (21) (22) (23)

*Ever hasn't Mary insulted Peter. [CP when [C Op Neg did [IP Mary insult anyone]]]? -* [CP neg-ever [C OpNeg did [IP Mary insult anyone]]]

This process is reminiscent of the amalgamation of the quantifier many with negation to produce few, as in the following example from Cinque (1990) (see also Klima (1964)): (24)

Franco non ha visto mold film di Forman. 'Franco has not seen many films by Forman.'

The following non-rhetorical sentence, brought up by a reviewer, is potentially a problem for this analysis of rhetorical force in wh-questions: (25)

When(ever) Mary insults anyone all hell breaks loose.

I will assume that Op is generated in the Comp of (25), but that inversion cannot (and need not) apply due to the presence of a complementizer in C, possibly ever, or its non-overt counterpart, parallel to embedded whquestions (see Koopman (1983)): (26)

I wondered when Mary insulted Peter.

The difference between free relatives and genuine embedded whquestions is that the latter have wh-force, whereas the former do not. Whenever is roughly equivalent to if in (25), and does not question a specific time reference, such as yesterday, or today. In (26) we can still provide an answer that refers to a specific point in time, such as yesterday, today, etc. In other words, (25) involves no scope competition between wh-AGR and Op in free-relatives, since free-relatives do not have wh-force at all, and therefore no wh-AGR in Comp is necessary. This explains why NPIs do not induce rhetorical readings in freerelatives. On the other hand, embedded wh-questions become rhetorical if an NPI is inserted due to the incompatibility of wh-force (or whagreement) and Op: (27)

I wondered when Mary insulted anyone.

Why and How 101 5.2. Why and How As noticed by Borkin (1971) and Lawler (1971), occurrences of NPIs in why and how questions do not give rise to rhetorical interpretations. These look like genuine questions (examples from Borkin): (28) (29)

Why did anybody go home so soon? How did anybody break open the safe?

Unlike questions in (1) and (2), they presuppose that the action has taken place so that Lawler (1971) calls why and how 'factives'. But notice that the same presupposition facts obtain if we analyse these wh-words as negated NPIs: (30)

Somebody went home for no (obvious) reason.

(31)

Somebody broke open the safe in no (obvious) way.

This way, the problem reduces to the familiar contrast between: (32) Mary uttered a curse at no conference. (33) Mary uttered a curse for no reason. While (33) presupposes that Mary uttered a curse, (32) asserts the opposite. Nonetheless, both sentences contain negated expressions. Why there is such a difference between why /how on the one hand, and other wh-words on the other, is a separate question. Notice that negated NPIs of the why/how type neither trigger inversion when appearing sentence-initially, as noted in Lakoff (1971) and Lawler (1971): (34)

For no reason, Mary left home.

(35)

*For no reason did Mary leave home.

They also do not license NPIs: (36) *For no reason, Mary left any pets at home. Negated temporal NPIs both trigger inversion and license NPIs in the clause: (37)

At no time did John keep any pets at home.

Notice that it would not do here to say that why and how differ because they are non-arguments, since temporal expressions above are likewise not arguments. A more promising direction would be to differentiate them along the referential/non-referential dimension, as in e.g., Aoun

102 Rhetorical questions (1985, 1986) and Rizzi (1990a). Aoun argued that why and how cannot be long-distance extracted because they are non-referential (i.e., there are no indexicals corresponding to them, cf. then and there, corresponding to when and where), and are therefore not subject to Principle C. Still, it is not clear why this property should affect applicability of inversion and NPI licensing. I leave it for future research to account for this intriguing fact. 5.3. Yes/No questions

Since there arises no AGR incompatibility in yes/no questions, we do not expect obligatory rhetorical readings in them. This prediction is only partly correct. Regular NPIs (e.g., anyone) do not force rhetorical readings: (38)

Did Mary help anyone!

The question is compatible with two answers: yes and no. However, idiomatic NPIs (e.g., lift a finger, budge an inch, etc.) do induce rhetorical readings even in yes/no questions (see Sedivy (1990) for this and other differences between regular and idiomatic NPIs): (39)

Did Mary lift a finger/budge an inch to help her boyfriend?

Why is this so? Suppose that idiomatic NPIs are listed in the lexicon as negative, e.g.: neg-budge an inch, neg-lift a finger. They occur most naturally with clausemate negation, and are never licensed by superordinate negation: (40) (41)

Mary didn't lift a finger to help her boyfriend. *John didn't say that Mary lifted a finger to help her boyfriend.

Superordinate negation, however, can license non-idiomatic NPIs (see Chapter 2): (42)

John didn't say that Mary helped anyone.

If true that negation is a part of the lexical item for idiomatic NPIs, then the value of polarity operator must be set negatively in yes/no questions in order to satisfy the lexical properties of the idiom. This results in a negative rhetorical reading.

Yes/No questions 103 It is generally true that idiomatic expressions cannot cross clause boundaries. The idiom "the cat is out of the bag" cannot be split into two clauses (see Radford (1988))2: (43)

*The cat is trying [CP PRO to be out of the bag.]

If this analysis is correct, then idiomatic NPIs are not independent lexical items, like other NPIs, which enter into locality relations with negation and Operator. Rather, they are idioms whose integral part is negation. This explains why idiomatic NPIs behave in the same way crosslinguistically (see Sedivy (1990)), although the distribution of regular NPIs may vary considerably. In sum, the binding analysis of NPIs, coupled with the recent proposal that there is Spec/Head AGR in Comp, can provide a natural account of rhetorical force in questions containing NPIs.

6 Subjunctive: domain extensions

6. 1. Introductory words It has long been noticed that Subjunctive clauses extend the domain for negative polarity items, for example in Italian and French. The domain with subjunctives also extends for other clause-bound processes, such as, reflexivization and pronominalization. Explanations for the Subjunctive puzzle so far have not been able to unify the three processes, and in particular they have failed to differentiate between morphologically simple reflexives, which extend their domain in subjunctives, and morphologically complex reflexives, which do not. This chapter offers a unified syntactic account for all the three processes. According to the most influential syntactic analysis, Tense in Subjunctive clauses is anaphoric (see Anderson (1982), Pica (1984), Everaert (1984), Jakubowicz (1984), Johnson (1985), Picallo (1984), etc.). Coindexing of the subjunctive tense with the superordinate indicative tense is argued to automatically extend the domain up to the superordinate clause. I will assume that they are right as far as the first claim goes: (1)

Subjunctive morphology is an overt way to mark absence of independent tense.1

I will further claim that the absence of independent Tense is only a necessary (but not a sufficient) condition for domain extension. In particular, absence of Tense makes it possible for Infl and/or Comp projections to delete in LF. This leads to the following proposal: (2)

104

Subjunctive Functional Deletion: Subjunctive Infl/Comp projections delete in LF up to recoverability.

Introductory words 105 Given that Comp and Infl are sentential heads, their deletion at the level of logical interpretation gives rise to what can descriptively be termed a "clause union" effect. This is not to say, however, that the subjunctive clause and its matrix literally become one in any absolute sense. Although the two share one Infl/Comp complex in LF, there are still two subjects and two verbs in them. As will be argued in this chapter, the domain extends only for those syntactic processes whose domain is defined on the basis of Infl or Comp, when Infl or Comp are either missing governors or missing SUBJECTS. For example, Infl deletion is sufficient to extend the domain for simple reflexives, which pick AGR (in Infl) as their only SUBJECT (see section 0.2.1). On the other hand, Infl deletion will not affect complex reflexives whose set of SUBJECTS includes [NP,IP] and [NP,NP]. As pointed out above, subjunctive clauses do have an [NP,IP], and the domain for the complex reflexive closes off with the subjunctive clause. On the other hand, since subjunctive clauses lack Infl projection at the relevant level, the domain for AGR sensitive reflexives extends. Thus the descriptive term "clause-union" should be interpreted only in a relative sense, since obviously no absolute clause union takes place. AGR and Case properties of Infl are satisfied either at S-structure or PF, thus rendering Infl deletable in LF. If, for reasons independent of Tense, Infl or Comp cannot delete, the domain will not extend. For example, Russian, Polish, and Rumanian subjunctives contain a modal in the Infl position, and the domain in these languages does not extend. Likewise, the domain for pronouns and negative polarity does not extend in relative subjunctive clauses because their Comp contains a relative operator and is thus unrecoverable. Only recoverable material is subject to LF-deletion (see Lasnik and Saito (1984) and Chomsky (1989)). I am assuming that Comp carries a truth value index only in case the clause it heads has a truth value. Thus the domain for pronouns and negative polarity extends only with volitional subjunctives (selected by verbs like want), which have been assumed to carry no truth value. This way, the semantic categories of Tense and Truth influence the binding domain only indirectly: the absence of independent Tense or Truth specification renders the sentential functional projections unnecessary for interpretation, and they can delete in LF. In the rest of the introduction I briefly outline the domain extension mechanism for each of the local dependencies. As argued in the previous chapters, negative polarity items (NPIs) are A'-anaphors which have to be bound by either negation in Infl (split or not) or the polarity operator

106 Subjunctive: domain extensions in Comp (in questions, conditionals, etc.). If both projections are missing in LF (in subjunctive clauses), the next potential antecedent will be Infl in the superordinate clause, thus allowing licensing by superordinate negation. Both Comp and Infl transparency are required for the domain extension. On the assumption that non-volitional subjunctives (e.g., complements of factive verbs) have a truth-value index in their head, Comp, their Comp cannot delete in LF, and the domain must remain local. Assuming that a negated NPI (e.g., Italian nessuno) in the subject position triggers negative AGR in Infl, an instance of Spec/Head AGR, we can explain the well-known subject/object asymmetry in some Romance languages, e.g., French and Italian. Subject NPIs do not extend their domain in subjunctive clauses because Infl that they mark negative cannot delete in LF, negation constituting unrecoverable material. This same assumption would also take care of the subject/ object asymmetry in simple sentences: while subject NPIs can appear without a negative licenser, object NPIs cannot. This is because the triggered negative AGR in Infl can act as a licenser for the NPI. Subject Pronoun Obviation effects in volitional subjunctives also follow naturally from this approach. If Infl/Comp projections are invisible in LF with volitional subjunctives, the first available governor for the subject pronoun will be the immediately dominating verb. The first maximal projection including this governor and a SUBJECT for the pronoun [NP, IP] will be the next higher IP. Thus, a subjunctive subject is disjoint in reference only from an immediately dominating subject, even if this subject itself is a part of a subjunctive clause embedded in another matrix. This is in contrast with X° reflexives which extend their domain across any number of subjunctive clauses. Since the only SUBJECT for an X° reflexive (AGR) is missing in subjunctive clauses, the reflexive has to look up for an indicative AGR. A next higher NP, on the other hand, is a possible SUBJECT for a pronoun, and thus delimits its domain. Another difference between X° reflexives and pronouns concerns the object position. While reflexives extend their domain even if in the object position, pronouns do not. This again follows from the Relativized SUBJECT approach, coupled with Comp/Infl deletion in LF. A local IP is the governing category for object pronouns since it contains both their governor, the verb, and a SUBJECT, [NP,IP]. X° reflexives, on the other hand, lack a SUBJECT (i.e., AGR) in this domain, although they are governed by the Verb. That the domain for a pronoun is sensitive to the

Introductory words

107

presence of a governor is indicated by the following contrast (see Chomsky (1981)): (3) (4)

Peteri believes [that hei is the best]. *Peter[ wants [himi to be the best].

While in (3) the local Tense governs the pronoun, in (4) the governor is the matrix verb. Thus the domain extends in (4), but not in (3). Given this analysis, pronoun obviation in Subjunctive clauses is made parallel to the Exceptional Case Marking (ECM) cases in (4), but applies in LF, rather than at S-structure. As pointed out above, only morphologically simple (X°) reflexives extend the domain with subjunctives (see Yang (1983) and Pica (1987)). It is argued in section 0.2.1 that morphologically simple (X°) reflexives recognize X° SUBJECTS only. The only salient (c-commanding) X° (head) category that can act as a SUBJECT for reflexives, by virtue of having relevant person and number features, is AGR. Since the AGR node (assuming that it is a part of Infl) becomes invisible in LF with subjunctives, AGR having hopped onto the verb at S-Structure, the next higher AGR will define the domain for the reflexive. In contrast, morphologically complex (XP) reflexives select XP specifiers for their SUBJECTS, e.g., [NP,IP] and [NP,NP]. The Indicative/Subjunctive distinction will not affect XP reflexives because the local subject, [NP,IP], will always close off their binding domain, and no long-distance effect is expected. This approach to the domain extensions in Subjunctive clauses seems to have the following general advantages. First and most important, it provides a unified account for the behavior of reflexives, pronouns, and negative polarity in subjunctive clauses. Second, it derives syntactically the correlation between binding, on the one hand, and Tense and Truth, on the other. This explanation makes it possible to keep the Binding Theory as it is, while still making it sensitive to the issues of tense and truth. Third, the analysis makes only natural and independently motivated assumptions that functional categories can delete in LF unless unrecoverable (see Lasnik and Saito (1984)). Fourth, we can now explain why only X° reflexives extend their domain in Subjunctive clauses, as opposed to XP anaphors which remain local. I explore domain extensions in (i) sub-categorized subjunctive clauses in section 2, (ii) adverbial subjunctive clauses in section 3, and (iii) relative subjunctive clauses in section 4. It is necessary to discuss these types of

108 Subjunctive: domain extensions clauses separately because of the following differences between them: sub-categorized clauses are both c-commanded by the matrix elements, and contain deletable Comp material; adverbial clauses, on the other hand, are neither c-commanded by the immediately dominating subject, nor do they have a recoverable Comp; relative clauses are c-commanded by the immediately dominating clause, but do not have recoverable Comp due to the presence of a relative Operator. 6. 2. Sub-categorized clauses

6.2.1. Negative polarity If the basic analysis of NPIs is correct so far, i.e., if NPIs must be bound in the domain of either negation in Infl or Operator in Comp, then we have an explanation for the subjunctive facts. Since both Infl and Comp are deleted in LF with volitional subjunctives, the domain extends to the first indicative clause. Negative Polarity Items nessuno and personne in Italian and French show two types of subject/object asymmetries that require explanation: (i) local and (ii) long-distance (LD) asymmetry (see also section 1.2.6.). The former concerns the fact that the object NPI has to be licensed by negation in the local clause, whereas the subject NPI need not be so licensed (in fact, nessuno must not appear with local negation when in subject position), as shown below: (5)

(6)

Mario *(non) ha visto nessuno. Mario neg has seen no-one 'Mario has not seen anyone.' Nessuno (*non) ha visto Mario, no-one neg has seen Mario 'No one has seen Mario.'

Following Zanuttini (1989), Ladusaw (1991) argues that the head of a negative chain must be either Infl or Subject. On this view, one must give up the idea that nessuno is a negative polarity item. One also has to accept a disjunctive notion of head of chain. But suppose instead that nessuno is an NPI, and that it does require negation to license it. The object facts follow without further stipulation. Suppose now that the following is true (for Spec/Head Agreement in IP and CP, see Chomsky (1986a) and Rizzi (1990a)):

Sub categorized clauses 109 (7)

A negated NPI when in the specifier of a functional projection (IP or CP)3 triggers negative morphological AGR in the head of the same projection.

Suppose, next, that negative AGR in Infl suffices to license the subject NPI. The immediate independent advantage of this proposal is in that we have a unified notion of licenser, or head of chain in Ladusaw's sense: negation in Infl. This also offers an explanation for the local subject/ object asymmetry, cf. (5) to (6). The object NPI, on the assumption that it is not in a specifier position, does not have the option of triggering AGR, and must be licensed by a negative particle. The LD asymmetry concerns the contrasts (8) through (11) below, which show that NPIs in subjunctive clauses can be licensed by superordinate negation only if they are in the object position (with indicative clauses, neither of the positions can be licensed long-distance), see Kayne (1981) for French, and Rizzi (1982) for Italian:4 Italian: (8) Non pretendo [che tu arresti nessuno]. neg require-I that you arrest-SUBJ no-one. 'I don't require that you arrest anybody.' (9) *Non pretendo [che nessuno ti arresti]. neg require-I that no-one you arrest-SUBJ. 'I don't require that anybody be arrested.' French: (10) ?Je n'ai exige qu'ils arretent personne. I neg-have required that-they arrest nobody (11) *Je n'ai exige que personne soit arrete. I neg-have required that nobody be arrested The most influential explanation for the contrasts assumes that nessuno and personne raise in LF, and that their traces obey the Empty Category Principle (ECP), see e.g., Kayne (1981) and Rizzi (1982). This account still leaves unexplained the indicative/subjunctive contrast, in other words, it does not explain what blocks LF movement in indicative clauses (but see Zanuttini (1991) for an attempt to explain the contrast).5 On this approach, the subjunctive/indicative contrast follows from the assumption that indicative Infl cannot delete, due to its unrecoverable tense and truth specifications. The domain can extend only in those subjunctive

110 Subjunctive: domain extensions clauses whose Infl contains no unrecoverable material, as in (8) and (10). The presence of an NPI in the subject position of a subjunctive clause (see (9) and (11)) precludes the possibility for Infl-deletion since the negative AGR constitutes unrecoverable material. Thus the domain cannot extend.6 (Irrelevantly, (9) and (11) can have the reading where the embedded clause is negated, i.e., when NPIs are licensed by the triggered AGR in the local Infl.) This analysis makes the following prediction: the domain for a reflexive will not extend in a subjunctive clause with a nessuno subject. The prediction is confirmed in the following contrast, due to Raffaella Zanuttini (personal communication). In (12), the reflexive can take Gianni as its antecedent (although on an unpreferred reading), whereas in (13) it cannot: (12) (13)

Giannii vuole che Mariaj parli di sel^y Gianni wants that Maria speak of self. Giannii non vuole che nessunoj parli di se*-^. Gianni doesn't want that no one speak of self.

Se cannot refer to Gianni in (13) because nessuno triggers negative AGR in Infl, rendering the content of Infl unrecoverable. Since Infl cannot delete in LF, the domain for the reflexive remains the local clause, since it contains a SUBJECT, i.e. AGR.

6.2.2. Pronoun disjointness

As noted by many, a subject pronoun in a subjunctive clause must be disjoint in reference from the subject of its matrix clause in certain Romance languages (see e.g. Anderson (1982), Picallo (1984), Salamanca (1981), Pica (1984), Everaert (1984), Jakubowicz ((1984), Johnson (1985)). The following indicative/subjunctive contrasts obtain in French and Italian. (Unless otherwise indicated, examples in this section are from Johnson (1985)): French: (14) *Jeani veut qu'i/i aille-SUBJ a Paris. John wants that-he go to Paris (15) Jeanj croit qu7/i est-IND a Paris. John thinks that-he is in Paris

Sub categorized clauses

111

Italian: (16) *Giovannii vuole che prox telefone-SUBJ John wants that (he) telephone (17) Giovannii sa che prox ha telefonato-IND. John thinks that (he) has telephoned Furthermore, the disjointness effect holds only between the subjunctive subject and the subject of the immediately dominating clause, whether it be in the indicative or subjunctive mood: (18)

Jeani veut qu'il*i desire-SUBJ qu% aime-SUBJ Marie. John wants that-he desires that-he loves Mary.

The governing category for an element X (ranging over anaphors and pronominals) is defined in Chomsky (1981) as that domain which contains X, a governor for X, and a SUBJECT accessible to X. If a governor for a subject pronoun is missing in the local clause, its domain extends automatically: (19)

*Peterj wants [him^ to be the best].

Since there is no Tense or AGR in infinitivals to govern the subject, the verb want assumes this role. Thus the domain extends to the matrix IP. If a governor is present in the local clause, the domain does not extend: (20)

Peteri believes [that he, is the best].

A situation parallel to ECM (Exceptional Case Marking) arises in subjunctive clauses. With deleted Infl/Comp projections there is no local governor available for the subject pronouns in (14) or (16), rendering the matrix verb the closest governor in LF, where I assume the binding principles apply. Thus the domain for subject pronouns extends to the matrix IP, to include the governor, explaining why subject pronouns must be disjoint from matrix subjects. Since Infl/Comp projections cannot delete in indicative clauses, no domain extension is evidenced in (15) and (17). The domain for pronouns will not extend across more than one subjunctive clause, as shown in (18), because the first higher IP contains all the relevant elements: the pronoun, the governor, and a SUBJECT. In contrast, the domain will extend for X° reflexives across any number of subjunctive clauses because subjunctive clauses contain no SUBJECT for these reflexives in LF, i.e., no c-commanding AGR (see section 6.2.3.). As

112 Subjunctive: domain extensions predicted by our analysis, double embedding of pronouns within infinitivals will likewise show only the immediately dominating subject effect: (21)

John; wants [him*; to expect [himx to come]].

Furthermore, the domain in subjunctives extends only for pronouns in the subject position. As shown in the following example, an object need not be disjoint in reference from the matrix subject: French: (22) Jeanj veut que Marie /i'aime-SUBJ. John wants that Mary him-love This is as predicted since objects are always governed by the local verb, and their domain will not be affected by the absence vs. presence of the Infl/Comp complex. The local IP is the governing category for the object pronoun since it contains both the governor for the pronoun (verb) and a SUBJECT, local [NP,IP]. Our analysis also predicts that postverbal subject pronouns in Romance, being governed by the verb, do not show obviation effects. This seems to be true, given the possibility (although unpreferred) for the following construal (see Johnson (1985)): (23)

Mariai spera che vinca-SUBJ lei\. Mary hopes that wins she

As pointed out by Johnson (1985), however, the obviation effect is somewhat weakened with overt pronouns, and this prediction is only possible to check with overt pronouns. Johnson's (1985) i-within-i explanation (which relies on Infl coindexing) cannot carry over to nominal complements, which also show obviation effects with subjunctives. Johnson attributes this observation to Jim Harris: (24) (25)

*E1 miedo de Juanj que pro, llegue tarde. (Spanish) *La peur de Jeanj qu7/i arrive en retard. (French) the fear of John that (he) arrive-SUBJ late

Although there can be no Tense coindexing here, since there is no Tense in the NP, the obviation effect holds. Given our analysis, the subjunctive Infl/Comp projections delete by virtue of hosting no referential Tense, regardless of whether this Tense is coindexed to a higher Tense or not.

Sub categorized clauses

113

6.2.3. Long-distance reflexives It has long been noticed that subjunctive clauses extend the domain for X° reflexives up to the first indicative clause, as reported for Icelandic (Thrainsson: (1976, 1990), Maling (1984), and Sigurdsson (1990)), Italian (Napoli (1979) and Giorgi (1984)), and French (Pica (1984)). The following examples show the Subjunctive/Indicative contrast in these languages: Icelandic: (Hyams and Sigurjonsdottir (1991)) (26) Joni segir ad Peturj raki sig^ says that shaves-SUBJ self (27) Joni veit ad Peturj rakar sig*i/y knows that shaves-INDicative self Italian (from Giorgi (1984)): (28) Giannii suppone che tu sia innamorato della propriax 'Gianni^ supposes that you are-SUBJ in love with self Si moglie. wife.' (29) *Giannii mi ha detto che tu sei innamorato della propriax moglie. 'Giannii told me that you are-IND in love with self Si wife.' French: (from Pica (1985)): (30) Oni ne souhaite jamais que les gens ne regardent que soix. one not wishes never that people not look-SUBJ at oneself. 'One never wishes that people look only at oneself (31) *Oni ne dit jamais que les gens disent du mal de soix. one not says never that people slander-IND oneself 'One never says that people slander oneself The domain does not extend with XP (morphologically complex) reflexives: Icelandic (Hyams and Sigurjonsdottir (1991)): (32) Jonj segir ad Peturj raki sjalfan-sig*^ says that shaves-SUBJ him-self Italian (Giorgi (1984)): (33) Quel dittatorei pensava che i libri di storia avrebbero parlato a lungo di sex / *se stessox e delle sue gesta.

114 Subjunctive: domain extensions That dictator thought that the books of history would talk for a long time about self and his deeds.' I argued in section 0.2.1 that long-distance properties of morphologically simple (X°) reflexives can be derived by relativizing the notion of SUBJECT in the following way: (34) (35)

A SUBJECT for an X° reflexive can only be a c-commanding X° category (head) carrying person/number features, i.e. AGR. A SUBJECT for an XP reflexive can only be a c-commanding XP specifier carrying person/number features, therefore [NP,NP] and [NP,IP].

The choice of SUBJECT is relativized according to the X-bar status of the reflexive (see the Structure Preserving Principle for movement in Baltin (1982), Chomsky (1986), and Rizzi's (1990a) relativized minimality). The parallel with movement landing sites is striking: only specifiers seem to be potential SUBJECTS for XP reflexives, and only heads for X° reflexives. Objects, even though in some cases they can act as binders, can never define the domain. The Relativized SUBJECT approach has an immediate explanation for the contrast between simple and complex reflexives in subjunctive clauses. XP reflexives do not extend their domain in subjunctives (see examples (32) and (33)) since the subjunctive subject always counts as a SUBJECT for a reflexive, thus delimiting its domain to the local IP. This subject, on the other hand, is not a SUBJECT for an X° reflexive, and will therefore not define its domain. It remains to explain why AGR, which is obviously present in subjunctives,7 does not count as a SUBJECT.

Suppose that AGR and Tense are features on the Infl node (following Iatridou (1990), and contra Pollock (1989) who argues that they head separate projections), and that these affixes hop onto the verb at Sstructure, as suggested in Chomsky (1981). The traces of Tense and AGR remain in Infl, Tense being relevant for temporal interpretation of the sentence in LF. However, if Tense is dependent on the matrix tense, or on the matrix verb, as is the case with subjunctives, then its specification in Infl becomes redundant. In fact, as pointed out in footnote 7, Tense or AGR in English subjunctive clauses need not be represented even at Sstructure. I propose that precisely in such cases, i.e., when Tense has no independent reference, it is possible for Infl to delete in LF, having fulfilled its function at S-structure: Case assignment to the subject and

Sub categorized clauses

115

affix hopping. If Infl is deleted in LF, the AGR feature (or its trace) cannot serve as a SUBJECT for an X° reflexive, and the domain extends to the first visible (i.e. Indicative) Infl. The AGR suffix on the verb cannot count as a SUBJECT since it does not c-command the reflexive. 6.2.4. Subjunctive without extension Curiously, subjunctive clauses in Russian, Polish, and Rumanian do not show domain extensions. What subjunctive clauses in these languages have in common, as opposed to the languages discussed so far, is the presence of a subjunctive particle in Infl. Given this approach, we can readily explain the exceptional behavior: the presence of a subjunctive particle in Infl makes it unrecoverable, and Infl cannot delete in LF. In Russian (36) from Comrie (1980), the NPI must remain local due to the presence of the particle by in Infl:8 (36)

*On ne hocet, cio-by ja obrascalsja ni k komu. he not want that-should I turn-to no-one 'He doesn't want that I turn to anyone.'

Likewise, the domain for the otherwise long-distance X° reflexive sebja does not extend across a subjunctive clause: (37)

Vanjai xocet, cio-by vsej ljubili sebja*^ wants that-should everybody loves self 'Vanja wants everybody to love himself

This is in contrast with Russian infinitivals, which allow for the extension, as pointed out in section 0.2.1. (the following Russian examples are slightly modified from Comrie (1980); translations are Comrie's, glosses mine): (38)

Petjai velel mnej kupit' sebe-^ cvety. Peter told me buy self flowers 'Peter told me to buy himself/myself some flowers.'

However, even infinitivals become opaque if the particle by appears in them: (39)

*Predsedatel'i ne tak umen, cio-by PRO arb sebjax uvazat. chairman not so clever that-should self respect 'The chairmani is not clever enough for one to admire sehV

116

Subjunctive: domain extensions

This fact suggests that domain extensions in infinitivals can be seen in the same light: their domain extends only to the extent to which their functional projections can become invisible in LF. This has an important parallel in S'-deletion at S-structure, discussed in section 6.2.2. In Polish there is likewise no domain extension with pronoun binding, negative polarity, or reflexives (Chominsky (1992)). Polish subjunctive clauses also host the modal by: Pronoun Binding: (40) *Jani chce, zeby prox pojechat do Paryzu John wants that he goes to Paris. Negative Polarity: (41) *Nie zadalem, zzby oni nikogo aresztowali. I have not required that they nobody arrest. Reflexives: (42) Janj woli, zebys (ty)j byl zakochany w swojej*^ zonie. John prefers that you be in love with self s wife. As predicted by this approach, Rumanian subjunctives, which have a particle sa in Infl, do not show obviation effects (from Kempchinsky (1986)):9 (43)

Ana vrea sa mearga. Anaj wants (she;) SUBJ go.

The evidence from Romanian, Polish, and Russian, although not conclusive support for the analysis of Subjunctive proposed in this chapter, at least suggests that the proposal is on the right track. 6.3. Adverbial clauses As shown below, the domain for reflexives in Icelandic does not extend across counterfactual conditionals and adverbial clauses introduced by nema 'unless' and 6o-ad 'although' (examples from Thrainsson (1990); see also Sigurdsson (1990)): (44) (45)

*J6nj vaeri gladur ef Maria kyssti sig\. 'John, would be glad if Mary kissed-SUBJ himsehV *J6ni kemur ekki nema Maria kyssi sigx. 'John, come not unless Mary kisses-SUBJ himself|.'

Adverbial clauses 117 (46)

*J6rii er her Oo ad Maria kyssi sigx. 'John is here although Mary kisses-SUBJ himsehy The same is true of Italian (Raffaella Zanuttini, personal communication): (47) *Maria non e' venuta sebbene Peter abbia ucciso nessuno. 'Mary did not come although Peter killed-SUBJ anyone.' I will assume that this follows from the lack of c-command between the matrix Infl and the adverbial clause. That this assumption is on the right track is supported by the fact that further embedding under a verb taking subjunctive makes LD binding from the matrix clause possible (examples from Thrainsson (1990) and Sigurdsson (1990) respectively): (48) Joni segir ad Haraldurj komi fyrst Maria 'Johnj said that Haraldj comes-SUBJ since Mary elski

sigij*y

loves-SUBJ selfi/*j.' (49) Olafuri segir ad Mariaj se her enn 06 ad eg Olafj says that Maryj is-SUBJ here still although I skammi sig[. scold-SUBJ selfi/*j. 'Olaf says that Mary is still here although I scold him/*her.' As is obvious from the above examples, the immediately dominating clause remains inert for binding even under this embedding. The same fact holds in Italian (Giorgi (1984)): (50) Mariai sperava che Osvaldoj ritornasse in patria prima che il fisco sequestrasse il proprio^ patrimonio. 'Maria hoped that Osvaldo would return to his country before the public-treasury sequestred self s estate.' Suppose that adverbials are IP adjoined to the right, as shown below: (51)

Mary

meet

Elaine

likes

self

118 Subjunctive: domain extensions Suppose further that binding cannot cross a maximal projection, not even in case of inclusion (see Chomsky (1986a)). In that case Mary cannot ccommand the reflexive. However, the subject of a higher matrix will uncontroversially c-command the reflexive:

Peter meet Elaine likes self If there is indeed no c-command between the subjunctive and the immediately higher subject, we also predict the absence of Principle C effect: (53)

?*Hei will meet Elaine if John* likes her.

Although far from perfect, the example sounds slightly better than: (54)

*Hei said that Johni likes Elaine.

The acceptability of (53) seems on a par with: (55)

?*Hei is here. Johnj likes Elaine.

The oddity of (53) and (55) may be ascribed to a pragmatic precedence condition. Principle C should not apply across a clause boundary.10 In sum, binding of reflexives across adverbial clauses can be explained on the grounds of lack of c-command. Since reflexives are only sensitive to Infl, the unrecoverable content of Comp does not play a role. However, given that negative polarity and pronoun disjointness are sensitive to both Infl and Comp content, we expect no long-distance effects with either of the two processes across an adverbial subjunctive, even if it is embedded under two clauses. This prediction is confirmed by the following polarity example from Italian, provided by Raffaella Zanuttini:

Relative clauses 119 (56)

*Pietro non ha vuole [che Maria non venga-SUBJ [sebbene abbiaSUBJ visto nessuno]]. ""Peter didn't want that Mary not come although she saw anyone.'

Neither of the superordinate negations can license the NPI in (56) above since the local Comp cannot become transparent due to the presence of the unrecoverable complementizer. For independent reasons, however, it is not possible to check if subject pronouns show an obligatory obviation effect. This is because pro cannot normally skip a possible antecedent: (57)

Pietroj dice [che Giannij verra [sebbene pro*^ abbia visto Peter says that John will come although pro saw Maria. Mary.

6.4 Relative clauses What does our approach predict of relative clauses? Since they must be ccommanded by the matrix subject, long-distance binding of the reflexive will be possible. On the other hand, the domain will not extend for pronouns and negative polarity since the relative Comp contains unrecoverable material, the relative operator (see Chomsky (1986a)). As predicted, reflexive binding across Subjunctive relative clauses is possible. Icelandic (Thrainsson (1990)): (58) Joni segir ad 0etta se stiilkan sem elski sigx. Johni says that this is-SUBJ the-girl that loves-SUBJ selfi. Italian (Giorgi (1984)): (59) Un soldato valorosoi uccide sempre coloro che minacciano in armi il propriox paese. 'A brave soldier always kills those who threaten self s country.' On the other hand, negative polarity items in relative clauses cannot be licensed by superordinate negation: French (Marc Authier, personal communication): (60) *Marie n'a jamais vu un homme qui ait tue personne. Mary neg-has ever seen a man who has killed anyone 'Mary has never seen a man who has killed anyone.'

120 Subjunctive: domain extensions Italian (Raffaella Zanuttini, personal communication) (61) ?*lo non ho mai avuto un amico che avesse fatto del male a nessuno. 'I have never had a friend who hurt anyone.' Likewise, subject pronouns in relative subjunctive clauses do not show an obligatory obviation effect: (62)

Maria pensa alFuomo che avrebbe dovuto incontrare Mary; thinks of-the-man that proi should have met ieri. yesterday.

Thus, the domain in relative subjunctive clauses extends only for reflexives, which are sensitive to Infl properties only, AGR being their only SUBJECT. The domain does not extend for negative polarity and pronoun obviation because their governing category is sensitive to the properties of both Infl and Comp, and the latter cannot delete in LF since it contains a relative operator. 6.5. Summary and discussion In conclusion, the solution to the Subjunctive puzzle proposed in this chapter has the following general advantages: it relies on the natural assumption that recoverable functional projections can delete in LF; it provides a unified account for all the domain extension phenomena; and it derives the difference between languages that extend the domain and those that do not. Even more importantly, we are now able to capture the interaction between Binding, on the one hand, and Tense and Truth, on the other, without altering or complicating the Binding Theory. The Binding Theory can still operate with structural concepts only, while the issues of Tense and Truth will determine the status of functional projections in LF. As pointed out in the introductory section, no absolute clause-union takes place in subjunctive clauses. Instead, the domain extends only for those processes whose governing category is defined on the basis of Infl and/or Comp, the projections missing in LF. Since SUBJECTS for complex (X-self) reflexives are base-generated specifiers ([NP, IP] or [NP, NP], and since their governor is the verb (recall from section 0.2.1. that reflexives do not generally appear in subject positions), the delimitation of their

Summary and discussion 121 governing category will be unaffected by the presence vs. absence of the Infl/Comp complex. On the other hand, SUBJECTS for simple (se/self) reflexives (AGR) and for negative polarity (negation or operator) are either in Infl or in Comp, and the deletion of the latter two will extend their domain. Likewise, governors for pronouns are either in Infl (Tense and/or AGR) or prepositional complementizers in Comp, again predicting a domain extension in case of Infl/Comp deletion. The characterization of Subjunctive proposed in this Chapter implies that subjunctive clauses are similar to infinitivals, which also lack unrecoverable Infl and Comp. However, there is one important difference between the two: infinitivals show little (if any) evidence of Infl even at Sstructure, whereas subjunctives have a fully-fledged Infl at S-structure. In particular, infinitivals in general show no Tense or AGR at S-structure, thus they do not have to wait until LF for their domain to extend. In fact, as pointed out by a reviewer, the island effects induced by Tense are observable even on S-structure movement, as the following contrast illustrates (see Chomsky (1986a)): (63) (64)

Which car do you wonder how to repair? *Which car do you wonder how John repaired?

Subjunctives, on the other hand, have Tense and AGR at S-structure, and can only extend their domain in LF. This predicts that only LF phenomena will show long-distance dependencies in subjunctive clauses.

7 Free-choice items (FCIs)

7.1. Introductory words

So far I have considered closely only polarity any in English, with occasional mention of free-choice (FC) any (see section 3.4.2.). Polarity and ¥C-any are homophonous in English, which makes it hard to delimit their distribution. I include this chapter on FCIs for two reasons. First of all, Serbian/Croatian has a separate lexical item for ¥C-any so that its distribution can be easily determined, with no possibility for confusion. Second, the binding analysis of polarity sensitivity proves useful in accounting for the interpretation of FCIs. In the literature it is usually assumed that free-choice any is a universal quantifier with wide scope over its trigger, i.e., the element which licenses it, usually taken to be a modal (see, for example, Lasnik (1975), Hintikka (1977), Klima (1964), Ladusaw (1980) or Linebarger (1981)). Some of these approaches take this to distinguish between free-choice any and polarity any, which is argued to be an existential quantifier taking narrow scope with respect to its trigger (see, e.g. Carlson (1980), Klima (1964), Ladusaw (1980) and Linebarger (1981)). So, the sentence in (1) can be semantically represented as in (2): (1) (2)

Anyone can beat a child. Ax POSSIBLE (x beat a child) For every x, it is possible for x to beat a child.

Here, free-choice any takes scope over the modal. On the other hand, polarity any always takes narrow scope with respect to its licenser, for example, negation: (3)

Mary did not insult anyone.

This sentence is represented as follows: 122

Free-choice 'Bilo' in Serbian/Croatian

(4)

123

NOT Ex, Mary insulted x There is no x such that Mary insulted x.

What I will try to question in this chapter is the claim that free-choice any is always a universal quantifier taking wide scope with respect to its licenser, and with it the whole attempt to distinguish between the two any's in terms of their quantificational force. In contrast, I will propose a binding account for this distinction. As will be established in section 7.2., Serbian/Croatian bilo appears in all the contexts where free-choice any does, receiving the same interpretation as free-choice any does. In addition, bilo is also licensed by superordinate negation and by the polarity operator (see Chapter 3), the contexts which are usually not associated with free-choice any in English. Notably, bilo in these contexts receives an existential interpretation, that is, the same interpretation as an NPI in its place would. So, if free-choice any does appear in these contexts, it can easily be mistaken for the polarity any. And I will assume that this is indeed the case (see section 1.4.). While with extraclausal licensers (such as superordinate negation and the polarity operator) bilo can receive the same existential interpretation as a corresponding NPI, this is not possible in clausemate contexts (such as clausemate negation and modals). I will assume that those readings which allow for the substitution of an NPI in polarity contexts without a change in meaning are bound readings since they show the same semantic correlation between negation and the FCI which has already been established for NPIs. The other possible readings will be referred to as free readings. The fact that bilo can be bound by extraclausal antecedents, but not by local antecedents, will lead to the proposal that it is subject to Principle B of the Binding Theory (section 7.3.). Potential binders for bilo are modality operators: negation, modals, and polarity operator.

7.2. Free-choice 4Bilo' in Serbian/Croatian

In Serbian/Croatian, the lexical item corresponding to free-choice any is bilo. Bilo can occur with modals, with negation, and with the polarity operator.1 With clausemate negation and modals, bilo receives a widescope interpretation. With superordinate negation and the polarity operator, bilo receives a narrow-scope existential interpretation.

124 Free-choice items 7.2.1. With modals In modal contexts, the distribution and interpretation of bilo mirror those of English free-choice any: (5)

(6)

Milan ce uraditi bilo sta za novae. Milan will do be-it what for money 'Milan will do anything for money.' Bilo (t)ko ce to uraditi za novae, be-it who will that do for money 'Anyone will do that for money.'

Bilo in modal contexts is interpreted the same way free-choice items in English are: (7) (8)

Given S, the set of actions, if X belongs to S, then Milan will do X for money. Given S, the set of persons made relevant by the pragmatic context, if X belongs to S, then X will do that for money.

In addition, bilo is not licensed in the corresponding sentences without modals (excluding generic tense which also seems to be a licenser): (9)

(10)

*Milan je uradio bilo sta za novae. Milan is done be-it what for money *'Milan did anything for money.' *Bilo (t)ko je to uradio za novae. be-it who is that done for money *4Anyone did that for money.'

7.2.2. With negation With clausemate negation, bilo necessarily receives a contrastive reading: (11)

Goran nije uvredio bilo (t)ko-ga, (vec Predsednik-a). Goran not-is insulted be-it who-m but President-ACC 'Goran did not insult just anyone, (but the President).'

With superordinate negation, the preferred reading is the existential, narrow-scope reading:2

Free-choice 'Bilo' in Serbian I Croatian 125 (12)

Goran ne tvrdi da je bilo (t)ko dosao. Goran not claims that is be-it who come 'Goran does not claim that anyone has come.'

This reading corresponds to an NPI-reading. In fact, bilo can be substituted for an I-NPI with no change in meaning: (13)

Goran ne tvrdi da je i(t)ko dosao. Goran not claims that is anyone come 'Goran does not claim that anyone has come.'

Sentences with bilo show the same negative implicature as those with i(t)ko\ (14)

Milan nije ukrao knjigu zato-sto ga je \(t)ko\ bilo (t)ko Milan not-is stolen book because him is anyone/be-it who nagovorio. persuaded. 'Milan has not stolen the book because anyone persuaded him to.'

Both cases give rise to the following implicature: (15) No one persuaded Milan to steal the book. On the other hand, the substitution of bilo for an NPI in contexts with clausemate negation leads to a totally different reading: (16)

Goran nije uvredio ni(t)koga. (*vec Predsednik-a). Goran not-is insulted no-one but President-ACC 'Goran did not insult anyone. (*but the President).'

The contrastive reading of (11) is lost in (16). 7.2.3. With the polarity operator The polarity operator in Comp can also have bilo in its scope (see section 3.3.): Yes/No Questions: (17) Da li je bilo (t)ko dosao? that Q is be-it who come 'Did anyone come?'

126

Free-choice items

Conditionals: (18) Ako je Mira uvredila bilo (t)koga, pokaja-ce se. if is Mira insulted be-it whom regret-FUT self 'If Mira has insulted anyone, she will regret it.' Adversative predicates: (19) Sumnja-m da je Mira uvredila bilo (t)ko-ga. doubt-1SG that is Mira uvredila be-it who-m 'I doubt that Mira has insulted anyone.' Just as with superordinate negation, the preferred reading in all these contexts is the one usually associated with NPIs: the existential narrowscope reading.3 Substituting the polarity item i(t)ko (an I-NPI, see section 1.2.2.) for bilo produces no difference in meaning. Yes/No Questions: (20) Da li je i(t)ko dosao? that Q is anyone come 'Has anyone come?' Conditionals: (21) Ako je Mira uvredila i(t)ko-ga, pokaja-ce se if is Mira insulted anyone-ACC regret-FUT self 'If Mira has insulted anyone, she will regret it.' Adversatives: (22) Sumnjam da je Mira uvredila i(t)ko-ga. doubt-1SG that is Mira insulted anyone-ACC 'I doubt that Mira has insulted anyone.' 7.2.4. Discussion

If bilo were a universal quantifier, necessarily taking wide scope over its licenser (e.g., negation and the polarity operator), we would expect the following readings to be ascribed to the sentences (12), (17), (18), (19): Superordinate negation: (23) For every person X, it is not the case that Goran claims that X has come. Yes/no Questions: (24) For every person X, is it true of X that X has come?

An analysis

127

Conditionals: (25) For every person X, if Mira has insulted X, she will regret it. Adversatives: (26) For every person X, I doubt that Mira has insulted X. While it might be possible to argue that the ascribed readings are correct for superordinate negation and adversative predicates, it is clear that those given for yes/no questions and conditionals are not the correct readings. Instead, the readings ascribed to (17) and (18) are the following: Yes/No Questions: (27) Was there at least one person such that that person came? Conditionals: (28) If there is at least one person, such that Mira insulted that person, she will regret it. The above established distribution of bilo indicates that it cannot be treated as a universal quantifier necessarily taking wide scope with respect to its licenser. I will propose in section 7.4. that free-choice any is licensed in the same contexts and receives the same interpretation in these contexts as bilo. Due to the lexical homophony of free-choice any and polarity any in English, and to their synonymy in these contexts, we are not able to detect its occurrence. Notice that it is exactly those contexts in which confusion is possible (polarity contexts) that have not been established as free-choice any contexts in English. 7.3. An analysis

As established above, bilo can receive either a universal or existential interpretation. It is interpreted universally when appearing with clausemate licensers: modals or negation. It is interpreted existentially with superordinate negation and a polarity operator in Comp. In other words, its existential/universal interpretation does not seem to be an intrinsic property, but rather to depend on the distance of the modality operator. Universal readings are forced only when an operator is in the closest possible Apposition, i.e., negation and modals in Infl. However, if the operator is further away (superordinate negation or Comp), then bilo receives an existential interpretation.

128

Free-choice items

Suppose that bilo is bound by an operator in Comp, or by superordinate negation, because in these contexts bilo receives the same interpretation as, and is interchangeable with, the corresponding NPI. The question is: why cannot bilo be bound by a clausemate modal or negation, the way polarity items can? Recall that in clausemate negation contexts bilo cannot receive an NPI interpretation, nor is it substitutable for an NPI (section 7.2.2.). Since the contrastive reading bilo receives in this context is clearly not a bound reading, I will take it to be a free reading. Let us propose now that bilo obeys the following condition: (29)

Bilo is subject to Principle B of the Binding Theory. (Potential binders for bilo are modality operators: modals, negation, and the polarity operator in Comp)4

If the first potential antecedent creates the opaque domain in the case of bilo and polarity items (see section 3.3.), then negation and modals in Infl would be the first potential antecedents for bilo. Its governing category in that case would be the first maximal projection containing bilo and its first potential antecedent, i.e., IP. This way, operators in Comp and superordinate negation will fall outside of this domain, thus being able to bind the pronominal bilo. In other words, as a pronominal, bilo has to be free only within its governing category, which results in wide-scope, universal reading. Given this analysis, the wide- vs. narrow-scope interpretation is a semantic reflection of a syntactic distinction: free vs. bound status. I will assume with Ladusaw (1980) that FCIs need no trigger. This is consistent with the proposal here that they are pronominal. As shown in Davison (1979), there are more contexts which license FCIs than those that do not. Since it would be hard to establish a natural class for all the licensers, it seems more promising to exclude NPIs from certain contexts by a semantic filter. Notice that the bound distribution of bilo parallels that of I-NPIs and PPIs. I-NPIs have been argued to be anaphoric pronominals, subject to Principle B and to a general binding requirement (see section 1.2.4.). PPIs have been argued to be subject to Principle B of the Binding Theory. Both I-NPIs and PPIs can appear in all polarity contexts with the exception of clausemate negation. This has been argued to be due to their pronominal requirement: they need to be free from a potential binder in their governing category.

An analysis

129

The proposal in (29) would explain a striking parallelism between bilo, on the one hand, and I-NPIs and PPIs, on the other hand. Neither of them can be bound by clausemate negation: (30)

(31)

(32)

*Mira nije uvredila i(t)ko-ga. Mira not-is insulted anyone-ACC 'Mira did not insult anyone.' #Mira nije uvredila ne(t)ko-ga. Mira not-is insulted someone-ACC #'Mira did not insult someone.' #Mira nije uvredila bilo (t)koga vec svoj-u majk-u. Mira not-is insulted be-it whom but her-ACC mother-ACC #'Mira has not insulted anyone but her mother.'

Just like I-NPIs and PPIs, bilo can be bound by superordinate negation: (33)

(34)

(35)

Goran ne tvrdi da je Mira i(t)ko-ga uvredila. Goran not claims that is Mira anyone-ACC insulted 'Goran does not claim that Mira has insulted anyone.' Goran ne tvrdi da je Mira ne(t)ko-ga uvredila. Goran not claims that is Mira someone-ACC insulted 'Goran does not claim that Mira has insulted someone.' Goran ne tvrdi da je Mira bilo (t)ko-ga uvredila. Goran not claims that is Mira be-it who-m insulted 'Goran does not claim that Mira has insulted anyone.'

In fact, all three examples above are synonymous, on the preferred reading of each. The same facts obtain in all the non-negative contexts, as illustrated in (36) - (41) for yes/no questions and adversative predicates: Yes/No questions: (36) Da li je Mira i(t)ko-ga uvredila? that Q is Mira anyone-ACC insulted 'Has Mira insulted anyone?' (37) Da li je Mira ne(t)ko-ga uvredila? that Q is Mira someone-ACC insulted 'Has Mira insulted someone?' (38) Da li je Mira bilo (t)koga uvredila? that Q is Mira be-it whom insulted 'Has Mira insulted anyone?'

130

Free-choice items

Adversative predicates: (39) Sumnja-m da je Mira i(t)ko-ga uvredila. doubt- 1SG that is Mira anyone-ACC insulted 'I doubt that Mira has insulted anyone.' (40) Sumnja-m da je Mira ne(t)ko-ga uvredila. doubt-1SG that is Mira someone-ACC insulted 'I doubt that Mira has insulted someone.' (41) Sumnja-m da je Mira bilo (t)ko-ga uvredila. doubt-1SG that is Mira be-it who-m insulted 'I doubt that Mira has insulted anyone.' Again, each set of examples is synonymous. The proposal that bilo is subject to Principle B carries over to modal contexts, on the assumption that the wide-scope reading of bilo is its free reading. This analysis explains why bilo must always take wide scope with respect to clausemate modals. 7.4. Free-choice 'Any'

Finally, I would like to suggest that the analysis of bilo can shed new light on ¥C-any in English. As established above, bilo can receive either a universal or existential interpretation. Notice that a similar phenomenon has been observed for FC-any, hence the current debate whether it is universal or existential. Carlson (1981) claims that both FC-any and polarity any are present in some contexts (such as comparatives and adversative predicates), with no ambiguity between the universal and existential reading. Instead, existential readings are assigned to both any's (examples from Carlson):5 (42)

Bob ran faster than anyone did.

From these data Carlson concludes the following: "f-c any is in fact a universal, and [...] it must, somehow, appear equivalent to an existential under certain circumstances" (p. 20). Strikingly, Carlson (1981) wonders why, according to his tests (e.g., modification by nearly or almost), FCany does not appear with negation, questions, and certain conditionals, given that they do not commit one to the existence of objects (for details, see Carlson): (43)

*Did almost anyone just walk into the room?

Free-choice 'Any'

131

Interestingly, bilo in Serbian/Croatian appears in exactly these contexts. In fact, FCIs appear in such contexts in other languages as well, e.g., French (Scullen (1991)) and Portuguese (Gabriel (1991)). The nullhypothesis would be that this is also true for ¥C-any in English. If it, too, could appear in these contexts and if it received the same existential interpretation as bilo does, it would be impossible to distinguish it from polarity any. Moreover, all the usual tests for ¥C-any would fail here (e.g., modification by nearly or almost) because what they detect is the universal interpretation, rather than ¥C-any, given that they are also valid for other universal quantifiers, e.g., every. (44)

Almost everyone passed that exam.

So, if ¥C-any is existential in some contexts, it cannot be detected by the above mentioned tests in these contexts. In sum, it has been proposed in this chapter that free-choice bilo in Serbian/Croatian is a pronominal which can receive either a bound (narrow-scope existential) reading or a free (wide scope universal) reading, depending on the distance of the modal operator. On the other hand, polarity items, as anaphors (see section 1.2.4.), have only one option, and that is to be bound by their licenser thus receiving a narrowscope interpretation. If the same is true for ¥C-any and polarity any in English, it would suggest, as also assumed by Carlson, that there can be no unified account of polarity any and ¥C-any in English, neither in existential nor universal terms.

8 Ladusaw and Linebarger

This chapter will briefly summarize the two leading approaches to negative polarity: Ladusaw's (section 8.1.) and Linebarger's (section 8.2.). It will be pointed out that their analyses need to incorporate locality conditions in order to accommodate the data presented in this work. Section 8.3. provides an alternative account for certain operator/polarity interactions captured by Linebarger's analysis.

8.1. Ladusaw

Ladusaw (1980; 1982; 1983) attempts to completely reduce the phenomenon of negative polarity to semantics. He argues that licensing of polarity items can be determined on the basis of semantic criteria alone: what predicts triggerhood is a semantic property of being downward entailing. For Ladusaw (1980, 131), licensers cannot be reduced to negation, negations are only one "very obvious subclass of downward entailing expressions." By downward entailment Ladusaw means the following (1980, 149): (la)

(lb)

132

For any determiner D, if "D men walk" entails "D fathers walk" then D is downward entailing, (e.g., No/every man walks. -• No/ every father walks.) If "D fathers walk" entails "D men walk", then D is upward entailing, (e.g. Some fathers walk. -• Some men walk.) For any quantifier NP P, if "P walks" entails "P walks slowly", P is downward entailing, (e.g., No man walks -> No man walks slowly.) If "P walks slowly" entails "P walks" then P is upward entailing, (e.g., Every man walks slowly -• Every man walks. Some men walk slowly -> Some men walk).

Ladusaw 133 In other words, "an expression is affective [ = downward entailing] if it licenses inferences in its scope from supersets to subsets" (Ladusaw: 1982). Take, for example, the following two propositions, P and Q: (2) (3)

John ate brussels sprouts. (P) John ate a green vegetable. (Q)

The direction of entailment is from the subset {brussels sprouts) to the superset {green vegetables), that is, P entails Q, but not vice versa. This is an instance of upward entailment. Consider, on the other hand, the negated versions of (2) and (3): (4) (5)

John didn't eat a green vegetable (NOT Q) John didn't eat brussels sprouts (NOT P)

Here, the entailment relations are reversed: NOT P does not entail NOT Q, but rather NOT Q entails NOT P. Since the entailment is from the superset {green vegetable) to the subset {brussels sprouts), this is an instance of downward entailment. Given the above definition of downward entailment, Ladusaw (1980, 112) proposes the following principles which should account for the distribution of both NPIs and APIs (Affirmative Polarity Items, his term for PPIs): (6a)

(6b)

A NPI must appear in the scope of a trigger [a downward entailing element]. If its trigger is in the same clause as the NPI, the trigger must precede the NPI. An API may not appear within the scope of a clause-mate negation.

Since it is only downward entailing environments that license NPIs, Ladusaw predicts, correctly, that an NPI will be licensed in (7) but not in (8): (7) (8)

John did not eat anything. *John ate anything.

This analysis extends successfully to many other NPI environments, for example, //-clauses: (9) (10)

If you have a pet, you will not be allowed in. If you have a cat, you will not be allowed in.

134

Ladusaw and Linebarger

The antecedent of the conditional is downward entailing since (9) entails (10), i.e., the entailment proceeds from the superset (pet) to the subset (cat). And, as expected, NPIs are allowed in //-clauses: (11)

If you have any pets, you will not be allowed in.

Ladusaw's analysis also extends to contexts with universal quantifiers: (12) (13)

Everyone who has a pet will get in free. Everyone who has a cat will get in free.

Again, the entailment proceeds from (12) to (13), that is, from the superset (pet) to the subset (cat). Being a downward entailing context, it is predicted to license NPIs, and indeed, it does: (14)

Everyone who has any pets will get in free.

However, without recognizing locality restrictions, Ladusaw's theory faces the problems brought up below. First of all, it offers no obvious way to accommodate the parametric variation described in Chapter 4. This is so because it operates with the semantic notion of downward entailment which is constant across languages, predicting that NPIs in all languages will be licensed in the same contexts as in English. Although all the established downward entailing contexts for English will also be downward entailing in other languages, they will not all license NPIs in other languages. In addition, Ladusaw's analysis cannot explain the fact that nonnegative licensing must involve Comp (see section 3.4.), and thus cannot account for the following contrast: (15) (16)

I doubt that anyone is here. #1 doubt anyone.

Furthermore, Ladusaw's attempt to avoid the reduction of licensing to negation is not consistent. In his analysis, PPIs are disallowed only by overt negations (cf. his principle (6b)), and not by the whole set of downward entailing expressions which function as licensers for NPIs. In my analysis, which recognizes locality requirements, one can maintain that what allows NPIs also disallows PPIs (see section 3.3). In addition, yes/no questions, as noticed by Ladusaw himself, are not downward entailing, yet they license NPIs: (17)

Do you have a pet? (Yes) -/>

Ladusaw 135 (18) (19)

Do you have a cat? (Yes) Do you have any pets?

According to Kartunnen (1977), for a question A to entail a question B, every true answer to A has to entail B. It is evident that a positive answer to (17) above does not entail a positive answer to (18). Ladusaw (1980, 186) speculates about questions in the following way: (20)

Questions may be formed from any sentence regardless of the properties that its meaning may have, and hence both NPIs and APIs may appear in them independent of the presence or absence of triggers and anti-triggers.

However, this is not completely true (see also section 3.1.). Consider the following contrast: (21) (22)

Has anyone come? *Anyone has come?

Although (22) is a question, just like (21), it does not tolerate NPIs. Some formal licenser seems to be necessary, even in questions. One possible way to integrate questions into Ladusaw's framework would be to assume the following: (23)

An NPI is prohibited in an Upward Entailing clause (see Progovac (1992 b)).

Since questions are not upward entailing (see below), they would be NPI licensers: (24) (25)

Do you have a cat? (No) -/> Do you have a pet? (No)

However, as pointed out by Bill Ladusaw (personal communication) this characterization would wrongly predict licensing by exactly, which is not upward entailing: (26) (27) (28)

Exactly 5 fathers came. -/> Exactly 5 men came. *Exactly 5 men saw anyone.

136

Ladusaw and Linebarger

8.2. Linebarger

As opposed to Ladusaw, Linebarger (1981; 1987) argues that NPI triggerhood can be reduced to negation. For her (1987, 326), "the distribution of negative polarity items in English reflects an interplay between syntax and pragmatics." Linebarger's analysis mirrors that of Baker (1970b) who takes NPI licensing to be twofold: in the core case, an NPI is licensed by c-commanding negation in its sentence; in all other cases, an NPI is licensed by an implicature which must have the form of a core case licenser. She departs from Baker's theory in that she requires NPIs to be in the immediate scope of negation, and in that she determines scope relations at LF, and not at S-Structure. The core case of licensing (Part A), i.e., licensing by overt negation, is accounted for by her Immediate Scope Constraint (ISC), which is a syntactic condition (Linebarger (1987, 338)): (29)

A negative polarity item is acceptable in a sentence S if in the LF of S the sub-formula representing the NPI is in the immediate scope of the negation operator. An element is in the immediate scope of NOT only if (i) it occurs in a proposition that is the entire scope of NOT, and (ii) within this proposition there are no logical elements intervening between it and NOT. Logical elements are defined in Linebarger (1981, 30) as elements capable of entering into scope ambiguities.

Linebarger (1987, 336) uses this Constraint to explain the fact that "the same negation cannot license the NPI budge an inch and at the same time take the because-clause in its scope," as in (30): (30)

*He didn't budge an inch because he was pushed, (but because he fell).

She represents (30) as (31), where S-l is the adjunct clause and S-2 the matrix clause: (31)

NOT CAUSE (S-l, S-2) It is not true that S-l causes S-2.

Linebarger argues that it is the intervention of the predicate CAUSE between NOT and S-2 that makes the NPI budge an inch in S-2 impossible. In her terms, the NPI is not in the immediate scope of the

Linebarger

137

negative operator. Obviously, her constraint has to be determined at LF where the scope relations between logical operators are established. The other possible representation for (30) would be: (32)

CAUSE (S-l, NOT S-2) It is because he was pushed that he didn't budge an inch.

In this representation, the NPI is licensed in the matrix clause because it is in the immediate scope of negation, but it cannot be licensed in the adjunct clause because it is not in the scope of negation at all. Notice that NPIs are allowed in the because-cX&wse on the wide-scope reading of negation although the NPI is not in the immediate scope of negation (see the representation in (31)): (33)

He didn't move because anyone pushed him.

As will be shown below, Linebarger (1987) invokes Part B (the entailment part) to account for this sentence. Another instance where an intervening logical element blocks the NPI licensing, and which can therefore be captured by the ISC, concerns sentences with universal quantifiers (Linebarger (1987)): (34) (35)

*John didn't give a red cent to every charity. NOT Ax (charity, x) (John gave a red cent to x)

The sentence is ungrammatical because the universal quantifier intervenes between NOT and the NPI. Part B of Linebarger's theory is meant to cover those cases of NPI licensing which fail to satisfy Part A, the paradigm case. These include mostly non-negative polarity contexts (but also examples like (33) above). She proposes, following in essence Baker (1970b), that such sentences are acceptable because they allude to the paradigm case in some way. (36)

(37)

Expectation of negative implicatum is itself a conventional implicature. A negative polarity item contributes to a sentence S expressing a proposition P the conventional implicature that the following two conditions are satisfied. Availability of negative implicatum. There is some proposition NI (which may be identical to P) which is implicated or entailed by S and which is part of what the speaker is attempting to convey in uttering S. In the LF of some sentence S' expressing NI, the lexical representation of the NPI occurs in the immediate scope of

138

(38)

Ladusaw and Linebarger

negation. In the event that S is distinct from S', we may say that in uttering S the speaker is making an allusion to S'. NI strengthens P. The truth of NI, in the context of the utterance, virtually guarantees the truth of P. (Linebarger (1987, 346))

The following example from Linebarger (1981) is taken to illustrate this point: (39) (40)

I was surprised that she contributed a red cent} IMPLICATURE: I had expected her not to contribute a red cent.

By virtue of implicating (40), (39) is able to license an NPI. Let us now return to the example (33), repeated here for convenience: (41)

He didn't move because anyone pushed him.

Recall that the sentence is not accounted for by Part A since the NPI is not in the immediate scope of negation. However, now it can be captured by Part B by virtue of implying: (42)

No one pushed him.

Notice the contradiction in the following example: (43)

*He didn't move because anyone pushed him, although someone did push him.

The same basic objections brought up with regard to Ladusaw's analysis hold also of Linebarger's: inability to predict cross-linguistic variation and inability to account for the role of Comp in non-negative licensing. Furthermore, Linebarger's assumptions that NPI licensing takes place at LF, and that negation is an operator which is raised at LF, predict no subject/object asymmetry with respect to NPI licensing, since both subject and object will be in the scope of negation at LF. However, subject NPIs in English are never licensed by clausemate negation (see section 1.2.1. for discussion): (44)

* Anyone didn't come to the party.

Also, Linebarger's analysis provides no explanation for the fact that in the following example an NPI in the matrix cannot be licensed derivatively, as observed by Enq (1988): (45)

*He didn't budge an inch because anyone pushed him.

Polarity I operator interactions

139

In other words, it is not clear why (45) cannot implicate (46): (46)

He didn't budge an inch because no one pushed him.

However, Linebarger's Immediate Scope Constraint provides an account for two types of interaction concerning NPIs: interaction with universal quantifiers and with the operator because. In the next section, I will try to provide an alternative explanation for the two types of interaction which will be consistent with the binding analysis.

8.3. Polarity/operator interactions

Let us consider negated adjuncts first. Negation in the following sentence cannot at the same time license an NPI in the matrix clause and take the adjunct clause in its scope: (47)

*Mary did not budge an inch because she was pushed, but because

According to Linebarger's (1987) representation of this sentence, neither the matrix nor the adjunct clause is presupposed: what is negated is the causal link between the two clauses: (48)

NOT CAUSE (S-l, S-2)

However, it seems to me that the matrix clause is presupposed, given the possibility of the following paraphrase (for more discussion, see Progovac (1988)): (49)

Mary moved, not because she was pushed, but because . . .

On this reading, negation is not a part of the matrix clause at all which explains why NPIs are not licensed in this clause. Suppose that (49) is the D-structure representation of (47). Suppose, next, that the surface structure in (47) is derived by a special rule of Neg-Raising which moves negation from the embedded pre-IP position to the usual matrix position of negation (see Horn (1978)). This way the ambiguity arises between the matrix and the lower negation reading. Now, suppose that at LF the transposed negation is reconstructed to its D-structure position. This will explain why on this reading NPIs are allowed only in the adjunct clause, and not in the matrix.

140

Ladusaw and Linebarger

As noticed by Irene Heim (personal communication), this type of negation movement from an embedded pre-constituent position to the matrix negation position need not satisfy the ECP because reconstruction may take place before the ECP. In fact, this might be the reason for enforcing reconstruction in the first place. Suppose that the following is a partial representation for (47): (50)

i

C

The adjunct CP, not being L-marked, would be a barrier for the antecedent government of tj by negation (see Chomsky (1986a)). The only way to salvage the structure would be to reconstruct negation to its original position before the application of the ECP. In fact, this account seems superior to Linebarger's in two respects. First, it extends to cases which do not involve logical operators at all, but rather adverbial phrases, which do not raise at LF: (51)

*John didn't budge an inch in the afternoon, but in the evening.

In this respect Linebarger's analysis proves too weak. Second, only some clause operators give rise to the ambiguity discussed above: (52)

Mary did not come although she had promised to.

The above example is not ambiguous, lacking the following reading: (53)

*Mary came, not although she had promised to, but although . . .

Linebarger's analysis proves too strong in this case. The only possible explanation would be to stipulate that although is not an operator which raises at LF, while because is. It is hard to see what this distinction would follow from. Example (53) poses no problem for our analysis as long as we are aware of the Surface Structure difference between because and although, which is illustrated by the following contrast:

Conclusion 141 (54)

Mary came, not because she was ill, but because . . .

(55)

*Mary came, not although she was ill, but although . . .

Let us consider polarity/quantifier interactions next: (56) #John didn't give any money to every charity. The following reading, where negation takes scope over the QP every, is ungrammatical: (57)

*NOT QP NPI

However, the LF ordering where QP takes wider scope with respect to negation is acceptable: (58)

QP NOT NPI

Given QR and our binding analysis of NPIs, assume that the two LF representations take the following form: (59) (60)

*NOTi QPj NPIi tj QPj NOTi NPIi tj

There are two A'-chains: one connecting the NPI and its binder, negation, and the other, connecting the raised QP to its variable. The difference between the two representations lies in the relation between the two dependencies: while on the offending reading (59) the two A'-chains intersect, on the acceptable reading (60) the two paths are nested. Thus (59), but not (60), violates Pesetsky's (1982) Crossing Constraint. 8.4. Conclusion

This monograph looks into syntactic locality conditions on polarity sensitivity and proposes that NPIs are anaphors that need to be A'-bound in their governing category, while PPIs are pronominal-like in that they obey Principle B of the Binding Theory. Potential binders for polarity sensitive items are either negation or a polarity operator in Comp in the case of non-negative polarity licensing. This gives an immediate explanation for the otherwise curious fact that Comp must be involved in non-negative polarity licensing. The binding approach makes it possible to explain the distribution and interpretation of NPIs and PPIs cross-linguistically, as well as to capture the intuitive feeling that NPIs and PPIs are the poles of one and the same

142

Ladusaw and Linebarger

phenomenon (cf. the traditional term "polarity" items), which is not possible to achieve on purely semantic grounds. According to Ladusaw's analysis, for example, NPIs are licensed by the whole set of downward entailing operators, whereas PPIs are anti-licensed only by clausemate negation. Not only is negation just one member of the set of downward entailing operators, but it is not even the case that all instances of negation have an anti-licensing effect on PPIs. Only clausemate negation does. Subjecting NPIs and PPIs to the Binding Principles makes it possible to unify the class of NPI licensers and PPI anti-licensers. All the operators that license NPIs at the same time anti-license PPIs. Since PPIs are pronominal, the anti-licensing effect will be felt only within their governing category, i.e., IP. Both superordinate negation and operators in Comp fall outside of the local IP, which explains why they do not antilicense PPIs. In addition, a binding approach captures directly otherwise puzzling scope restrictions imposed on polarity items. NPIs always take narrow scope with respect to negation because they must be bound by negation. On the other hand, PPIs must take wide scope with respect to clausemate negation, and are free to take either wide or narrow scope with respect to a distant negation, because they are pronominal. A pronominal must not fall within the scope of (or must not be bound to) a local binder. These scope possibilities must be stipulated in an approach that does not appeal to the binding principles. Not only does the binding approach to polarity sensitivity capture numerous parallels between anaphora and polarity items, but it also contributes to the restrictiveness of the model of grammar by reducing two different phenomena to the same UG principles.

Notes

Introduction 1 I will refer to my native language as Serbian/Croatian, rather than SerboCroatian, as was traditionally the practice. Since the gradual break up of the Socialist Federal Republic of Yugoslavia during the past couple of years, Croatia has referred to the language as Croatian, while the remaining part of Yugoslavia, consisting of Serbia and Montenegro, has referred to the language as either Serbian or Serbo-Croatian. In order to be sensitive to the new terminological developments in both states, I will call the language Serbian/ Croatian. This monograph is about both Serbian and Croatian. All the judgments relevant for the discussion in this book work the same way for both variants. There are two differences that will surface in the monograph. First of all, the lexical item for who is ko in Serbian, but tko in Croatian. Throughout the examples I will use (t)ko to cover both variants. Since (t)ko is an integral part of polarity items, it will appear quite frequently in the monograph. Another difference concerns the use of infinitival versus finite subordinate complements. While Croatian prefers infinitival complements, the preference in Serbian is for finite complements introduced with the complementizer da. Since this is mainly a preference, the difference will be pointed out only in case it is relevant for the comparison between the two variants. I am deeply grateful to Zvjezdana Vrzic for her patient judgements on Croatian. 2 See note 1 for the distinction between niko and nitko. 1 Negative contexts: Serbian/Croatian 1 The underlying word order in Serbian/Croatian is the same as in English: Subject-Verb-Object (SVO). However, Serbian/Croatian is a so-called free word order language in that it allows almost any permutation of constituents (see section 1.1.1.). 2 At first sight, as pointed out by Audrey Li (personal communication), it may seem that there should be a difference between English and Serbian/Croatian with respect to PPIs in the subject position because negation does not ccommand the subject position in English, whereas it does in Serbian/Croatian. However, given our analysis of PPIs (see section 1.2.4), no difference is 143

144

Notes to pages 36-43

expected. Both in English and Serbian/Croatian, subject PPIs will be forced to take wide scope with respect to negation, although for different reasons. In Serbian/Croatian, the wide-scope reading of subject PPIs is enforced by their pronominal character: they have to be obviative from a potential binder in their domain, i.e., from clausemate negation. In English, on the other hand, the wide-scope reading of a PPI in subject position is a direct consequence of the failure of negation to c-command it. 3 The symbol '#' is used here to indicate that the sentence is ungrammatical only on the intended reading, i.e., when the wh-word is interpreted as an NPI. The sentence is perfect on the w/z-reading: (i) Who didn't see John? 4 There is a group of verbs in Serbian/Croatian, taking subjunctive-like complements, with which the complementary distribution between NI-NPIs and I-NPIs breaks down. These will be discussed in section 1.2.6. 5 The only way for questions to contain NI-NPIs is for them to contain the negative particle ne: (i) Zar ni(t)ko ne zna? (really) no-one not knows Ts it possible that no one knows?' However, this is not the whole story. There is another way to form negative yes/ no questions in Serbian/Croatian: (ii) Da ne zna Marija nesto o Jovanu? that not knows Mary something about John 'Does Mary know something about John, by any chance?' As opposed to (i), in (ii) negation and the verb precede the subject, and I will assume that they adjoin to Comp. Although both (i) and (ii) are negative questions, involving an overt negative particle ne, they show an important contrast with respect to negative polarity. While (i) type questions license NINPIs, as illustrated in (i), (ii) type questions only license I-NPIs: (ii) Da ne zna Marija ?ista/ *nista o Jovanu? that not knows Mary anything/nothing about John 'Does Mary know anything about John by any chance?' This follows directly from our assumption that NI-NPIs can only be licensed by negation within the local IP, but not by superordinate negation, or Operator in Comp (see especially sections 1.2.4. and 3.3.). I return to the relevance of this contrast in section 3.3. 6 Morphologically speaking, PPIs in Serbian/Croatian consist of the "nonspecific" morpheme ne (which, rather surprisingly, happens to be homophonous with the negative particle) and the Wz-word, e.g.: (i) ne-(t)ko NE-who

Notes to pages 45-56

145

7 Aoun's Generalized Binding framework extends the Binding Theory of Chomsky (1981), which is concerned with elements in A-position, to elements in A'-position. In this framework, wh-traces are treated as A'-anaphors which have to be A-free but A'-bound in their governing category (see section 0.2.3. for details). 8 This formulation is due to a reviewer. Originally I assumed that I-NPIs are subject to two binding principles simultaneously: Principle A at LF and Principle B at S-structure. The idea that an element can have different properties at different levels is not new. Finer (1984) proposes that the Dogrib word ye obeys Principle A at S-Structure but falls under Principle B at LF. As noticed by Saxon (1984), in order to be licit ye must have a local antecedent like an anaphor, but must be disjoint from it: (i) John j>e-hk'e ha. John him-3.shoot FUT 'John; is going to shoot himj/*i.' (ii) * ?ekaani je-enda thus he-3. survive 'He lives this way.' Given the original formulation, I-NPIs would have to raise out of their clause in LF in order to be bound locally by superordinate negation. The issue deserves further investigation. I adopt a non-raising analysis here primarily because Serbian/Croatian in general does not allow LF-movement, as established in section 1.1.4, but see section 2.3. for possible indications of movement. 9 I assume that the same is true of infinitivals, if they have Infl/Comp projections at all. NI-NPIs are licensed by matrix negation in embedded infinitivals: (i) Milan ne zeli ici ni-kuda. Milan not wishes go-INF no-where 'Milan does not want to go anywhere.' (ii) Milan ne zeli videti ni(t)koga. Milan not wishes see-INF no-one. 'Milan does not want to see anyone.' 2 Negative contexts: English 1 Notice that in English there are also so-called 'strict' NPIs whose distribution differs from that of any. Strict NPIs, like until, are only licit with clausemate negation, as pointed out in Linebarger (1981) (see also section 4.1.1). 2 As pointed out by a reviewer, the structure with any in D would be simpler: (i)

DP D

NP

I

I

i/any (t)ko/one

146

Notes to pages 57-9

However, in section 4.1.1 argue that NPI raising involves either IP adjunction or movement to Spec of CP in case of long-distance licensing. On the assumption that only maximal projections move to other maximal projections (see Baltin (1982) and Chomsky (1986a)), the quantifier cannot originate in the head of DP. However, if NPI movement can be characterized as adjunction to Infl or Comp, then the structure in (i) will indeed turn out preferable. At the moment, however, I do not see how adjunction to Infl and/or Comp would predict variation described in section 4.1. 3 As pointed out in section 1.2.6, negated NPIs in Romance and Serbian/ Croatian do show subject/object asymmetries, which are usually ascribed to the ECP (Kayne (1981), Rizzi (1982)). Since I am assuming that negated NPIs do not raise, given their strict locality in indicative clauses, I cannot appeal to the ECP explanation. Instead, I argue that negated NPIs, if in a specifier position, trigger negative AGR in the head position. That accounts for the local subject/ object asymmetries, unexplainable in terms of LF raising. Triggered negative AGR in Infl precludes the possiblity for transparency of functional categories in subjunctive clauses, which I take to be responsible for domain extensions (see Chapter 6 for details). 4 Later in his work, however, Huang argued that more movement conditions are observed at LF than he originally acknowledged (see e.g. Fiengo et al. (1988) and Huang and Tang (1989)). 5 I-NPIs in Serbian/Croatian also show island effects: Specificity: (i) ?Jovan nikada ne cita knjige koji-ma nedostaje i-jedna stranica. John never not reads books which-DAT misses any-one page (ii) *Jovan nikada ne cita ovu knjigu koj-oj nedostaje i-jedna John never not reads this book which-DAT misses any-one stranica page Complex NP Constraint: (iii) ?*Ne veruje-m u [mogucnost [da ce nam dozvoliti da not believe-1SG in possibility that will us allow that \(t)ko-ga poseti-mo]]. anyone-ACC visit-1 PL *I don't believe the possibility that we will be allowed to visit anyone.' (iv) *Ni-sam trazio [devojku [koja je i(t)ko-ga opljackala]]. Neg-am looked-for girl who is anyone-ACC robbed 4 ?I did not look for a girl who robbed anyone.' Coordinate Structure Constraint: (v) ?*Ne trazi-m da [[to pripremi-s] i [ista donese-s]]. Not ask-lSG that this prepare-2SG and anything bring-2SG '?*! am not asking you to prepare this and bring anything.'

Notes to pages 59-61

147

Adjunct Island: ugostila]. (vi) *Ni-sam napravila pit-u [posto sam i(t)ko-ga Neg-am made pie-ACC after am anyone-ACC received '*I did not make a pie after I received anyone.' The fact that long-distance construal of I-NPIs obeys islands does not necessarily run contrary to the assumption in section 1.2.4. that I-NPIs are bound anaphors. Our analysis only claims that they need not raise in order to satisfy their binding requirement, but they may raise for other reasons. I-NPIs are quantifiers, with a similar morphological make-up as English anyone (see section 2.2.), and are in principle eligible for raising (see also section 6.2.1, in particlular footnote 5, dealing with the raising of nessuno. However, such conclusion would be inconsistent with the suggestion in 1.1.4. that there is no LF movement in Serbian/Croatian. I leave the resolution of this question for future research. 6 As pointed out by the reviewer, the fact that non-argument phrases like budge an inch cannot be licensed by long-distance negation may follow from the ECP: (i) John did not budge an inch to help, (ii) *Mary did not say that John budged an inch to help. Given that they are not head-governed, they obey the fto-trace effect. However, since other adjuncts can be bound by negation long-distance (see (34) in the text), I conclude that (ii) receives a different explanation. In particular, I argue in section 5.3. that idiomatic NPIs are actually listed in the lexicon with the negative particle: (iii) neg-budge-an-inch This would predict that they require clausemate negation. Moreover, given this treatment it is possible to explain rhetorical effects in yes/no questions (see 5.3. for details). 7 The same subject/object asymmetry with non-bridge verbs is attested in Serbian/Croatian (see also footnote 3): (i) *Jovan nije sapnuo [da je i(t)ko opasan]. John neg-is whispered that is anyone dangerous poljubila]. (ii) ?Jovan nije sapnuo [da je Marija i(t)koga John neg-is whispered that is Mary anyone-ACC kissed 8 Assuming the Comp Indexing convention, we would have to say that by deleting that we also delete its index (as proposed in Aoun (1986)), thus allowing the trace in Comp to bind its variable. 9 I am grateful to Marc Authier (personal communication) for bringing these facts to my attention. This assumption, however, says nothing about the marginal status (for some speakers) of wh-extraction from the same position: (i) ??Who thinks that who left?

148

Notes to pages 62-74

It should be noted, however, that the absence of the complementizer does not make the sentence any better, suggesting that the ungrammaticality of (i) is not due to the ECP: (ii) ??Who thinks who left? 3 Non-negative polarity contexts 1 The group of adversative predicates includes verbs and adjectives such as doubt, deny, refuse, stupid, surprise, etc., which are usually felt to be 'inherently negative'. 2 However, the proposal in Progovac (1988) that Op is in the Spec of CP would also lead to the correct results on the assumption that an Op in the Spec of CP triggers AGR in Comp (for Spec/Head AGR in Comp, see Chomsky (1986a), Schneider-Zioga (1987), Rizzi (1990a)). That way, adversative predicates would select for a negative Agr in Comp, and indirectly for an Op in the Spec position. This is exactly parallel to wh-selection, which is satisfied by a wh-phrase in Spec by virtue of Spec-Head AGR. 3 For the problem raised by sentences of the type in (i), brought up by a reviewer, see section 3.5.1. (i) John swore that never again would he dance the polka. 4 In sentences like (i) below, the NPI will be licensed by whether, which can be treated as an overt counterpart of Operator in English. (i) Mary forgot whether anybody visited her. 5 The Serbian/Croatian fact in (51) is hardly unprecedented. As I point out in section 4.1.4, footnote 8, yes/no questions are formed in Turkish by a negative particle in Comp. These are interpreted as regular, rather than negated questions. Also, as Irene Heim pointed out to me, Latin verbs of fearing select a negative particle in the embedded Comp, but the embedded clause is not interpreted as negative (from Allen and Greenough (1983, 365)): (i) Time-6 ne Verres fecerit . . . fear-lSG NEG Verres has-done 'I fear that Verres has done . . .' The authors actually advise to translate ne as 'that'. 6 For arguments that negative preposing has Spec of CP as its landing site, see section 3.5.1. below. 7 Alternatively, one might assume with Koopman (1983) that wh-words move to Comp even at S-structure, but that the movement of the verb to Comp is blocked by the ECP, resulting in a /to-trace configuration. 8 Notice that the sentence given below is not necessarily a problem for this conclusion, although Comp is not available for the raising of only: (i) For whom did only Mary show any respect?

Notes to pages 74-81

149

Given that NPIs are licensed in wh-questions even without only (see section 5.1), only need not be a licenser here: (ii) For whom did Mary show any respect? 9 Examples of the following type may be excluded either by a precedence requirement, or by a strict c-command condition: (i) *Only anybody likes broccoli. However, none of these requirements is imposed on licensing of the subject NP by overt negation in Infl (see section 1.1.). In any event, such examples are at least as problematic for an approach in which only is a direct licenser. 10 Notice that inversion is not obligatorily triggered in Serbian/Croatian whquestions: (i) Sta Marija kupuje? what Mary buys 'What is Mary buying?' (ii) Sta kupuje Marija? This may be because wh-phrases in Serbian/Croatian also have the option to IP-adjoin, as argued in Rudin (1988a and 1988b). 11 A reviewer points out that an echo question tolerates epistemic interpretation of modals: (i) Yuri must have killed whom? This fact probably suggests that echo wh-questions do not undergo movement in LF, following Kayne (1981), Lasnik and Saito (1984), Van Riemsdijk and Williams (1986), but contra May (1985) and Sirbu-Dumitrescu (1987). This is a plausible conclusion given that wh-words in echo-questions neither take scope over the clause, nor do they necessarily appear in extractable positions: (ii) John saw Peter and whom? (iii) John saw Peter in the white what? 12 Some speakers do not find these examples as bad as indicated. McDowell (1987) points out that the sentences are good on echo readings, and it may be that such readings are available to those speakers (see also footnote 11). 4 Language variation 1 According to Hasegawa (1987), Japanese NPIs show the same distribution as Chinese ones, i.e., they are only licensed by clausemate negation. This is also true of Russian NPIs (see section 0.1.1). 2 Notice that it sometimes appears in non-polarity contexts for emphasis, as pointed out by Audrey Li (personal communication):

150 (i)

Notes to pages 82-9 ?Ta conglai jin-shi hao haizi. he ever EMPHASIS-be good boy 'He is always a good boy.'

A similar use of ever is attested in English (see section 1.2.2. for more examples of unlicensed NPIs, and discussion): (ii) They lived happily ever after 3 The reviewer points out that all these local NPIs in Chinese and English are non-arguments. Thus, their lack of quantificational force can follow from their inability to raise in LF. However, there are NPIs in English and Serbian/ Croatian which are adjuncts, but can be licensed long-distance (see also sections 1.2.3.2. and 2.3.3): (i) Peter did not say that Mary insulted Jane in any way. (ii) Peter did not say that Mary ever insulted Jane, (iii) Petar nije rekao da je Marija i-malo uvredila Jan-u. Peter neg-is said that is Mary any-little insulted Jane-ACC (iv) Petar nije rekao da je Marija ikada uvredila Jan-u. Peter neg-is said that is Mary ever insulted Jane-ACC Unfortunately, a non-circular definition of quantifier force is lacking here, as rightly pointed out by a reviewer. 4 Serbian/Croatian I-NPIs differ from English NPIs in that they are subject to Principle B, as well as to a general binding requirement (section 1.2.4). This prohibits the occurrence of I-NPIs with clausemate negation. 5 Superordinate negation in Romance can license an NPI in the embedded clause only if it is in the subjunctive mood (see section 6.2.1. for discussion and references): Italian: (i) Non pretendo [che tu arresti nessund]. not require-1SG that you arrest-SUBJ no-one 'I don't require that you arrest anyone'. 6 According to Enq (1988), the distribution of partitive NPIs is different. Since partitive NPIs in general seem to behave exceptionally (see Pesetsky (1987) and Enq (1991)), I will concentrate on non-partitive NPIs only. 7 Marc Authier (personal communication) informs me that the same seems to hold for the French NPI de. It is perfectly acceptable if licensed by clausemate negation: (i)

Marie n'a pas lu de livre Mary NE-has not read any book 'Mary didn't read any book.'

It is also acceptable, although marginally, if licensed by superordinate negation:

Notes to page 89

151

(ii) ??Je n'ai pas dit qu'il a lu de livre. I NEG-have not said that has read any book 'I didn't say that he had read any book' However, de is not acceptable in non-negative polarity contexts: Yes/No Questions: (iii) *Est-ce-que Marie a lu de livre? Q Mary has read any book 'Did Mary read any book?' Conditionals: (iv) *Si Marie lit de livre, je serai content. if Mary reads any book I will-be happy 'If Mary reads any book, I will be happy.' Adversatives (v) *Je doute que Marie ait lu de livre. I doubt that Mary have-SUBJ read any book I doubt that Mary read any book 8 There is one potential counterexample concerning Turkish NPI licensing: yes/ no questions in Turkish license NPIs: (i) Ali hie, kimse-yi gdrdu mii? AH any one-ACC saw Q 'Did Ali see anyone?' It seems, however, that yes/no questions in Turkish actually involve overt negation and that they can therefore be subsumed under negative licensing. Two facts point to this conclusion. First of all, Turkish embedded yes/no questions are indisputably formed by overt negation (EnQ (1988)): (ii) Ali [Fatma-nin gel-ip gel-me-di-gini] biliyor. Ali Fatma-NOM come-and-come-not-PAST knows 'Ali knows whether or not Fatma came.' Second, the question particle in Turkish has the same stem as negation: m: (iii) Ali hie kimse-yi gdr-me-di. Ali any person-ACC see-not-PAST 'Ali did not see anyone.' (iv) Ali Fatma-yi gor-dii mii? Ali Fatma-ACC see-PAST? 'Did Ali see Fatma?' A similar phenomenon is also observed in Latin yes/no questions and Chinese A-not-A questions. A yes/no question in Latin is formed by "adding the enclitic -ne to the empathic word" (see Allen and Greenough (1983, 205), as in (v). The authors note that the particle ne often has a negative force when added to a verb, as illustrated in (vi), see section 3.3 for more Latin facts):

152

Notes to pages 91-5

(v) T-u-ne id veritus es? YOU-Q that feared 'Did you fear that?' (vi) Recte-ne interpreter sententiam tuam? rightly-Q interpret-I meaning your 'Do I not rightly interpret your meaning?' Chinese A-not-A questions likewise involve a positive/negative disjunction: (vii) Mali zai jia bu zai jia? Mary at home not at home 'Is Mary at home?' 9 In fact, it has been argued in Aoun (1985a; 1986) and Aoun and Hornstein (1985) that Italian nessuno and French personne are subject to principles A and C. If true, this would exhaust all the logical possibilities for the binding dimension. Both personne (see Kayne (1981)) and nessuno (see Rizzi (1982)) exhibit a subject/object asymmetry when licensed by superordinate negation (see sections 1.2.2 and 6.2.1): (i) Non pretend-o che tu arresti nessuno neg require-1 SG that you arrest nobody (ii) *Non pretendo che nessuno ti arresti neg require-1SG that nobody you arrests (iii) ?Je n'ai exige qu'ils arretent personne. I NEG-have required that-they arrest nobody (iv) *Je n'ai exige que personne soit arrete. I NEG-have required that nobody be arrested Aoun (1985; 1986) and Aoun and Hornstein (1985) relate this fact to the subject/object asymmetry in wh-extraction, which is captured in Generalized Binding by subjecting wh-traces to principles A and C (see section 0.2.3. for generalized binding). Notice, also, that in Milner's (1979) dialect only clausemate negation can license personne. He points out that the distribution of personne in his dialect is similar to that of bound anaphors. The basic problem with this approach is that both nessuno and personne can be licensed without negation if in the subject position. For further discussion, see chapter 6, section 6.2.1.. 10 See also footnote 9 for French personne in Milner's (1979) dialect. 11 Of course, in that case one may wonder why only clausemate negation has such an effect on NPIs. Even on this assumption locality restrictions would be necessary. 12 As pointed out by a reviewer, Russian PPIs might be subject to a long-distance version of Principle B, as attested for example with Icelandic long-distance pronouns, forcing complementarity between pronouns and long-distance reflexives (see e.g. Johnson (1985)). However, there are two basic differences between Russian PPIs, on the one hand, and Icelandic long-distance pronouns, on the other, which seem to set them apart. While Icelandic

Notes to pages 99-109

153

pronouns show long-distance obviation only in subjunctive clauses, which are characterized by domain extension effects in general (see Chapter 6, section 6.2.2), Russian PPIs must be obviative in indicative contexts as well. Also, while the long-distance pronoun obviation in Icelandic has the effect of forcing complementary distribution between long-distance reflexives and pronouns, no such effect is in force in Russian. The domain for kto-to is wide despite the fact that the domain for the NPI niko remains local (see section 4.2.1.). 5 Rhetorical questions 1 The existence of Op in (16) is confirmed by the occurrence of the NPI anyone. Not all verbs select an Op (see section 3.4.1): (i) *I think that anyone dropped by. 2 . . . unless a part of it moves in raising structures: (i) [The cat]i seems [IP U to be out of the bag]. In this case, as pointed out by Radford, the idiom was clausemate at Dstructure, and the trace left mediates a clausemate relationship at other levels of representation. 6 Subjunctive: domain extensions 1 Of course, this does not mean that it is necessary to mark the absence of independent tense in this, or any other way. As pointed out in Sigurdsson (1990), in Faroese, a language related to Icelandic, the domain extends for reflexives with the same verbs with which it extends in Icelandic, although there is no subjunctive mood in Faroese. Similarly, in Serbian/Croatian the domain for NPIs extends with volitional verbs in spite of the lack of morphological subjunctive marking, as pointed out in section 1.2.6. 2 However, in the absence of a formal characterization of truth value indexes, this suggestion can only remain a promissory line for future investigation, as pointed out by a reviewer. 3 Like Subject NPIs, preposed NPIs do not require negation to license them, although they can co-occur with it (see Zanuttini (1991)): (i)

Proprio niente, (non) ho detto. absolutely nothing, (I) neg have said If preposing is to the Spec of CP, then by Spec/Head AGR in CP, Comp will be marked negative, and it will be capable of licensing the NPI. Since negative AGR is in Comp, it is not incompatible with overt Infl negation, explaining the difference in behavior between preposed elements and subjects, contrast (i) and (6) in the text. Thus the local subject/object asymmetry is more correctly a specifier/complement asymmetry. I continue to use the former term because of its familiarity. 4 For comparable facts in Spanish, see Kempchinsky (1986).

154

Notes to pages 109-114

5 Longobardi (forthcoming) discusses many instances of islands which nessuno seems to obey, e.g., Complex NP Constraint, Adjunct Constraint, Coordinate Structure Constraint, etc. Islandhood, of course, is a characteristic of movement. However, given the proposal that Infl/Comp projections delete in LF, many island violations can be derived. Adjunct condition, for example, follows from the fact that adjunct complementizers are unrecoverable in LF, blocking the domain extension. Likewise, relative clauses have a relative operator in Comp which is not subject to LF deletion. Longobardi in fact noticed that nessuno obeys only clausal islands, as opposed to phrasal islands, which asymmetry is predicted on the approach advocated here. Only clauses which have unrecoverable material in Comp or Infl will block extraction out of them. 6 As pointed out by a reviewer, the negative particle can appear in the subjunctive clause if there is no subject NPI. However, the presence of negation does not seem to have an effect on pronoun obviation. According to Raffaella Zanuttini, pro in (i) still has to be obviative from Mary, suggesting that clause union has taken place: (i) *Mariai sostiene che prox non abbia visto Pietro. Mary claims that not had-SUBJ seen Peter. 'Mary claims that she did not see Peter.' One can assume here that non does not occupy the Infl position, but heads (or specifies) its own projection (see Pollock (1989)). This way, deletion of Infl will be unaffected by the presence of non. On the other hand, the negative feature triggered by Spec/Head agreement would have to be in Infl, and this would explain why only triggered negation blocks domain extension. I continue to assume with Iatridou (1990) that both AGR and Tense are features in Infl. It is at least true of English that Infl elements (modals, AGR and Tense) are mutually exclusive, although each of them can occur with negation. Notice that nessuno in the subject position does not co-occur with overt negation, and is thus not possible to tell whether the placement of non in NegP would have any bearing on the c-command requirement in this case (see section 6.2.1. for more details). 7 . . . . with a notable exception of English whose subjunctive shows neither tense nor agreement: (i) John demands that Peter go. (ii) *John demands that Peter went. In addition, subjunctives in English do not tolerate do-support: (iii) *John demands that Peter doesn't/don't go. (iv) *John demands that Peter does/do go. (v) John demands that Peter not go. In short, there is no evidence of Infl even at S-Structure. It may be that in English the verb moves to Infl at S-Structure, instead of affixes hopping onto the verb, as I am assuming for Romance and Icelandic. I leave it for future

Notes to pages 115-30

155

research, however, to decide on the ultimate analysis of the distinction between English and Romance subjunctives. 8 For arguments that by is a modal particle cliticizing onto Comp, see Brecht (1974). 9 Notice that pronouns are necessarily obviative in Russian subjunctives (Michael Yadroff, personal communication): (i) Vanjai xocet, cto-by on} ljubil Natasu. Vanja wants that-SUBJ he loves Natasa. But this is also true of indicatives. If co-reference is intended, pro drop must be used. I have no explanation for the fact that pro drop is not an option in Russian subjunctives (Michael Yadroff, personal communication). 10 Notice, incidentally, that in our account it is the feature AGR on Infl that binds a long-distance reflexive, rather than the subject itself. Even if the subject turns out to c-command the reflexive, a case can be made that Infl does not, being lower in the tree. 7 Free-choice items 1 It also appears in all the other ¥C-any contexts, e.g. imperatives, generic tense, relative clauses (for a detailed list, see Davison (1979)). 2 Another possible reading for bilo is similar to the contrastive reading obtained with clausemate negation, roughly paraphrasable as follows: (i) 'Goran does not claim that just anyone has come. (What he claims instead is that the President has come.)' 3 In addition, bilo in these contexts can also have a contrastive reading also available with clausemate and superordinate negation (see section 7.2.2.). 4 Saying that bilo in certain contexts has a bound reading does not necessarily commit one to the view that bilo is an anaphor, since pronominals can also have bound readings, provided the binder is outside their opaque domain: (i) Maryi did not realize that Peter wanted to leave heri. 5 Notice also that it has already been observed that some occurrences of any in English, usually taken to be instances of polarity any, on closer inspection turn out to be instances of free-choice any. For example, Hoeksema (1983) argues, on the basis of Dutch data, that the occurrences of any in NP-comparatives are instances of free-choice any, rather than polarity any, as illustrated by the following example: (i) This girl is taller than any boy. See section 3.4.3. for similar facts in Serbian/Croatian.

156

Notes to page 138

8 Ladusaw and Linebarger

1 A reviewer points out that (39) is at best marginal for him/her. It may be that a red cent is an idiomatic expression for the reviewer, requiring a clausemate negation as its integral part (see section 5.3. for discussion). This would explain why grammaticality improves in the reviewer's example (i), involving a counterfactual conditional: (i)

I would be surprised if she contributed a red cent.

The implicature in (i) is not only that "I expect her not to" but also that "she will not contribute a red cent."

References

Abney, S. P. (1987) The English Noun Phrase in Its Sentential Aspect. Ph.D. Dissertation, MIT, Cambridge, Massachusetts. Adams, M. (1987) Old French, Null Subjects and Verb Second Phenomena. Ph.D. Dissertation, UCLA, Los Angeles. Allen, J. H. and J. B. Greenough (1983) New Latin Grammar. College Classical Series, New Rochelle, New York. Anderson, S. R. (1982) Types of Dependency in Anaphors: Icelandic (and Other) Reflexives. Journal of Linguistic Research 2.2., 1-22. Aoun, J. (1985a) A Grammar of Anaphora, The MIT Press, Cambridge, Massachusetts. (1985b) Bound Pronouns in Chinese, Proceedings of NELS 16. (1986) Generalized Binding, Foris, Dordrecht. Aoun, J. and N. Hornstein (1985) Quantifier Types, Linguistic Inquiry 16, 623637. Aoun, J., N. Hornstein, and D. Sportiche (1981) Some Aspects of Wide Scope Quantification. Journal of Linguistic Research 1, 69-95. Aoun, J. and A. Li (1989) Scope and constituency. Linguistic Inquiry 22.1, 1-25. Aoun, J. and A. Li (1991) Wh-Elements in-situ: Syntax or LF? ms. University of Southern California, Los Angeles. Aoun, J. and D. Sportiche (1983) On the Formal Theory of Government, The Linguistic Review 2, 211-236. Authier, J. -M. (1992) Iterated CPs and Embedded Topicalization, Linguistic Inquiry 23.2, 329-336. Baker, C. L. (1970a) Notes on the Description of English Questions: The Role of an Abstract Question Morpheme. Foundations of Language 6, 197-219. (1970b) Problems of Polarity in Counter-factuals. In J. M. Sadock and A. L. Vanek, (eds.), Studies Presented to Robert B. Lees by his Students, 1-15, Linguistic Research, Carbondale, USA / Edmonton, Canada. Baltin, M. R. (1982) A Landing Site Theory of Movement Rules, Linguistic Inquiry 13.1., 1-38. Barss, A. and H. Lasnik (1986) A Note on Anaphora and Double Objects. Linguistic Inquiry 17, 347-354. Battistella, E. (1987) Chinese Reflexivization. Paper presented at the 2nd Harbin Conference on Generative Grammar, Heilongjiang University, Harbin, People's Republic of China. 157

158

References

Besten, H. den (1978) On the Interaction of Root Transformations and Lexical Deletive Rules. Paper presented at the 1978 GLOW Conference, Amsterdam. Blight, R. C. (1992) Long Distance Binding and Subordinate Clauses in Latin. Manuscript, Wayne State University. Borer, H. (1989) Anaphoric AGR. In O. Jaeggli and K. Safir, (eds), The Null Subject Parameter, Reidel, Dordrecht. Borkin, A. (1971) Polarity Items in Questions. Papers from the 7th Regional Meeting of the Chicago Linguistic Society, 53-62, Chicago. Brecht, R. D. (1974) Deixis in embedded structures. Foundations of Language 11, 489-518. Brown, S. E. (1989) Russian Polarity Items. Manuscript, Indiana University, Bloomington. Browne, W. (1974) On the Problem of Enclitic Placement in Serbo-Croatian. In R. D. Brecht and C. V. Chvany, (eds.), Slavic Transformational Syntax, 3652, Michigan Slavic Material 10, Ann Arbor. (1986) Relative Clauses in Serbo-Croatian in Comparison with English. Ph.D. Dissertation, Institute of Linguistics, University of Zagreb, Yugoslavia. Burzio, L. (1990) On the Non-Existence of Disjoint Reference Principles. Rivista Di Grammatica Generativa 14, 1-27, Unipress, Padova. (1991) The Morphological Basis of Anaphora. Journal of Linguistics 27, 81-105. (forthcoming) The Role of the Antecedent in Anaphoric Relations. In R. Freidin, (ed.), Current Issues in Comparative Grammar, Kluwer, Dordrecht. Carlson, G. N. (1980) Polarity any is Existential. Linguistic Inquiry 11, 799-804. (1981) Distribution of Free-Choice any. Papers from the Seventeenth Regional Meeting of the Chicago Linguistic Society, 8-23. Casanova-Seuma, L. (1988) Binding and Negative Polarity Items in Spanish and Catalan. Manuscript, University of Southern California, Los Angeles. Chominsky, L. (1992) A Crosslinguistic Analysis of the Subjunctive: English and Polish. Manuscript, Wayne State University. Chomsky, N. (1981) Lectures on Government and Binding, Foris, Dordrecht. (1986a) Barriers, the MIT Press, Cambridge, Massachusetts. (1986b) Knowledge of Language: Its Nature, Origin, and Use. Praeger, New York. (1989) Some Notes on Economy of Derivation and Representation. MIT Working Papers 10, 43-74. (1992) A Minimalist Program for Linguistic Theory. MIT Occasional Papers in Linguistics 1, 1-67. Cinque, G. (1990) Types of A'-Dependencies, the MIT Press, Cambridge, Massachusetts. Cole, P., G. Hermon, and L.-M. Sung (1990) Principles and Parameters of LongDistance Reflexives. Linguistic Inquiry 21.1, 1-22. Comrie, B. (1980) Clause Structure and Movement Constraints in Russian. Morphosyntax in Slavic, 98-113, Slavica, Columbus, Ohio. (1981) Language Universals and Linguistic Typology: Syntax and Morphology. Chicago University Press, Chicago.

References

159

Culicover, P. (1991) Topicalization, Inversion, and Complementizers in English. Manuscript, The Ohio State University, Columbus. Davison, A. (1979) Any as Universal or Existential. J. Van der Auwera, (ed.), The Semantics of Determiners, 11-40, Croom Helm, London. DeHaan, G. and F. Weerman (1986) Finiteness and Verb-Fronting in Frisian. In H. Haider and M. Prinzhorn, (eds.), Verb-Second Phenomena in Germanic Languages, Foris, Dordrecht. Emonds, J. E. (1976) A Transformational Approach to English Syntax. Academic Press, New York. Enc, M. (1988) University of Southern California Class Lectures, Spring 1988, Los Angeles. (1991) The Semantics of Specificity. Linguistic Inquiry 22.1., 1-25. Erteschik, N. (1973) On the Nature of Island Constraints, Ph.D. Dissertation, MIT, Cambridge, Massachusetts. Everaert, M. (1984) Icelandic Long Reflexivization and Tense-Connectedness. Working Papers in Scandinavian Syntax 12, 1-35, University of Trondheim. Fauconnier, G. (1975) Polarity and the Scale Principle. Papers from the Eleventh Regional Meeting of the Chicago Linguistic Society, 188-199. Fiengo R. and J. Higginbotham (1981) Opacity in NP. Linguistic Analysis 7, 395421. Fiengo R., et al (1988) Wh-in-situ and Bounding in LF. Paper presented at the Seventh West Coast Conference on Formal Linguistics, University of California, Irvine. Finer, D. (1984) On the Relation Between Binding and Coreference. Manuscript, MIT, Cambridge, Massachusetts. (1985) The Syntax of Switch Reference. Linguistic Inquiry 16, 35-55. Frank, R. and A. Kroch (1992) Nominal Structures and Structural Recursion. Manuscript, University of Pennsylvania, Philadelphia. Fukui, N. (1986) A Theory of Category Projection and Its Applictions, Ph. D. Dissertation, MIT, Cambridge, Massachusetts. Gabriel, L. (1991) Free-Choice Items: Existential? Manuscript, Indiana University, Bloomington. Giorgi, A. (1984) Toward a Theory of Long-Distance Anaphors: A GB Approach. The Linguistic Review 3, 307-361. Givon, T. (1969) Studies in ChiBemba and Bantu Grammar, Studies in African Linguistics 3, The University of California, Los Angeles. Goodluck, H. (1992). Notes on Relative Clauses in Serbo-Croatian. Manuscript, University of Ottawa. Grimshaw (1977) English Wh-Constructions and the Theory of Grammar. Ph.D. Dissertation, University of Massachusetts, Amherst. Haegeman, L. 1991. Introduction to Government and Binding Theory, Blackwell, Oxford. Haider, H. (1986) V-Second in German. In H. Haider and M. Prinzhorn, (eds.), Verb-Second Phenomena in Germanic Languages, 49-75, Foris, Dordrecht. Halvorsen, P. K. (1982) Norwegian Reflexives and Lexical Functional Grammar. Manuscript, MIT, Cambridge, Massachusetts.

160

References

Hasegawa, N. (1987) INFL Movement and the Scope of Negation. Manuscript, University of Massachusetts, Amherst. Helke, M. (1971) The Grammar of English Reflexives, Garland, New York. Hestvik, A. (1990) LF-Movement of Pronouns and the Computation of Binding Domains. Ph.D. Dissertation, Brandeis University, Waltham, Massachusetts. (1991) Subjectless Binding Domains. Natural Language and Linguistic Theory 9, 455-496. Higginbotham, J. and R. May (1981) Questions, Quantifiers, and Crossing. Linguistic Review 1, 41-79. Hintikka, J. (1977) Quantifiers in Natural Languages: Some Logical Problems II. Linguistics and Philosophy 1, 135-172. Hoeksema, J. (1983) Negative Polarity and the Comparative. Natural Language and Linguistic Theory 1, 403-434. Horn, L. R. (1978) Remarks on Neg-Raising. In P. Cole, (ed.), Syntax and Semantics 9, 129-220, Academic Press, New York. Huang, C. - T. J. (1982) Logical Relations in Chinese and the Theory of Grammar, Ph.D. Dissertation, MIT, Cambridge, Massachusetts. (1983) A Note on the Binding Theory. Linguistic Inquiry 14.3, 554-561 Huang, C.-T. J. and C.-C. J. Tang (1989) On the Local Nature of the LongDistance Reflexives in Chinese. Proceedings of NELS 19, 191-206, GLSA, University of Massachusetts, Amherst. Hyams, N and S. Sigurjonsdottir (1990): The development of Long Distance Anaphora: A Cross-Linguistic Comparison with Special Reference to Icelandic. Language Acquisition 1, 57-93. Iatridou, S. (1986) An Anaphor not Bound in its Governing Category. Linguistic Inquiry 17, 766-772. (1990) About Agr(P). Linguistic Inquiry 21.4, 551-577. Jakubowicz, C. (1985) Do Binding Principles Apply to INFL. Proceedings of NELS 15, 188-206, GLSA, University of Massachusetts, Amherst. Johnson, K. (1985) Some Notes on Subjunctive Clauses and Binding in Icelandic. MIT Working Papers in Linguistics 6, 102-134. Karttunen, L. (1977) Syntax and Semantics of Questions. Linguistics and Philosophy 1, 3-44. Katada, F. (1991) The LF Representation of Anaphors. Linguistic Inquiry 22.2, 287-313. Katz, J. J. and P. M. Postal (1964) An Integrated Theory of Linguistic Descriptions, the MIT Press, Cambridge, Massachusetts. Kayne, R. S. (1981) Two Notes on the NIC. In A. Belletti, L. Brandi, and L. Rizzi, (eds.), Theory of Markedness in Generative Grammar: Proceedings of the 1979 GLOW Conference Scuola Normale Superiore di Pisa, 317-346. (1989) Notes on English Agreement. CIEFL Bulletin 1, 41-67. Kempchinsky, P. M. (1986) Romance Subjunctive Clauses and Logical Form. Ph.D. Dissertation, University of California, Los Angeles. Klima, E. (1964) Negation in English. In J. Fodor and J. Katz, (eds.), The Structure of Language, 246-323, Prentice-Hall, New Jersey.

References

161

(1965) Studies in Diachronic Transformational Syntax. Ph.D. Dissertation, Harvard University. Koopman, H. (1983) ECP Effects in Main Clauses. Linguistic Inquiry 14, 346-350. Koster, J. (1975) Dutch as an SOV Language. Linguistic Analysis 1, 111-136. Kroch, A. S. (1975) The Semantics of Scope in English. Indiana University Linguistics Club, Bloomington. Ladusaw, W. A. (1980) Polarity Sensitivity as Inherent Scope Relations. Garland, New York. (1982) On the Notion Affective in the Analysis of Negative-Polarity Items. Journal of Linguistic Research 2, 1-16. (1983) Logical Form and Conditions on Grammaticality. Linguistics and Philosophy 6, 373-392. (1991) Interpreting Negative Concord Structures. Paper presented at the 1991 LSA Annual Meeting in Chicago. Laka, I. (1990) Negation in Syntax: On the Nature of Functional Categories and Projections, Ph.D. Dissertation, MIT, Cambridge, Massachusetts. Lakoff, G. (1971) Adverbs and Modal Operators. Unpublished mimeo, University of Michigan Phonetics Lab. Larson, K. R. and W. A. Ladusaw (1986) Notes on Negation and Polarity. Manuscript, MIT, Cambridge, Massachusetts. Lasnik, H. (1975), On the Semantics of Negation. Hockney et al., (eds.), Contemporary Research in Philosophical Logic and Linguistic Semantics, 279311, Reidel, Dordrecht. Lasnik, H., and M. Saito (1984) On the Nature of Proper Government. Linguistic Inquiry 15, 235-289. Lasnik, H., and J. Uriagereka (1988) A Course in GB Syntax, MIT, Massachusetts. Lawler, J. M. (1971) "Any Questions"? Papers from the 7th Regional Meeting of the Chicago Linguistic Society, 163-173. Lebeaux, D. (1983) A Distributional Difference between Reciprocals and Reflexives. Linguistic Inquiry 4, 723-730. Li, Y.-H. A. (1985) Abstract Case in Chinese. Ph.D. Dissertation. University of Southern California, Los Angeles. Linebarger, M. C. (1981) The Grammar of Negative Polarity. Indiana University Linguistics Club, Bloomington. (1987) Negative Polarity and Grammatical Representation. Linguistics and Philosophy 10, 325-387. Longobardi, G. (forthcoming) In Defense of the Correspondence Hypothesis. In J. Huang and R. May, (eds.), Logical Structure and Linguistic Structure, Reidel, Dordrecht. Maling, J. (1984) Non-Clause-Bounded Reflexives in Modern Icelandic. Linguistics and Philosophy 7, 211-241. Manzini, M. R. (1992) Locality: A Theory and Some of Its Empirical Consequences, MIT, Cambridge, Massachusetts. Martin, J. (1987) Some Implications of Treating Polarity Phenomena Within Binding Theory. Paper presented at the Winter LSA Meeting, San Francisco.

162

References

May, R. (1977) The Grammar of Quantification. Ph.D. Dissertation. MIT, Cambridge, Massachusetts. (1985) Logical Form, MIT, Camrbidge, Massachusetts. McDowell, J. P. (1987) Assertion and Modality. Ph.D. Dissertation. University of Southern California, Los Angeles. Milner, J.-C. (1979) Le systeme de la negation en francais et l'opacite du sujet. Langue Franfaise 44, 110-163. Milsark, G. L. (1988) Singl-ing. Linguistic Inquiry 19.4, 611-634. Montalbetti, M. (1984) After Binding. Ph.D. Dissertation, MIT, Cambridge, Massachusetts. Muller, G. and W. Sternefeld (forthcoming a) Improper Movement and Unambiguous Binding. To appear in Linguistic Inquiry 24.3. (forthcoming b) Scrambling as A-bar Movement. To appear in N. Corver and H. van Riemsdijk, (eds.), Studies on Scrambling, Walter de Gruyter, Berlin, London, New York. Nakamura, M. (1987) Parametrized Extension of Binding Theory. MIT Working Papers in Linguistics 9, 193-223.

Napoli, D. J. (1979) Reflexivization Across Clause Boundaries in Italian. Journal of Linguistics 15, 1-28. Pesetsky, D. (1982) Paths and Categories, Ph.D. Dissertation. MIT, Cambridge, Massachusetts. (1987) Wh-in-Situ: Movement and Unselective Binding. E. J. Reuland and A. G. B. ter Meulen, (eds.), The Representation of Indefiniteness, 98-129. MIT, Cambridge, Massachusetts. Pica, P. (1984) On the Distinction between Argumental and Non-argumental Anaphors. W. de Geest and Y. Putseys, (eds.), Sentential Complementation, 185-193. Foris, Dordrecht. (1985) "Subject, Tense and Truth: Towards a Modular Approach to Binding. In J. Gueron, H.-G. Obenauer and J.-Y. Pollock, (eds.), Grammatical Representation, 259-291, Foris, Dordrecht. (1987) On the Nature of the Reflexivization Cycle. Proceedings of NELS 17, 483-499, GLSA, University of Massachusetts, Amherst. Picallo, M.C. (1984) The Infl Node and the Null Subject Parameter. Linguistic Inquiry 15, 75-102.

Platzack, C. (1986a) The Position of the Finite Verb in Swedish. In Haider, H. and M. Prinzhorn, (eds.), Verb-Second Phenomena in Germanic Languages, 27-47. Foris, Dordrecht. (1986b) Comp, Infl, and Germanic Word Order. In L. Hellan and K. Koch Christensen, (eds.), Topics in Scandinavian Syntax, Reidel, Dordrecht. Pollock, J.-Y. (1987) Sur la Syntaxe Comparee de la Negation de Phrase en Francais et en Anglais: Deplacement du Verbe et Grammaire Universelle. Manuscript, Universite de Paris 12. (1989) Verb Movement, UG and the Structure of IP. Linguistic Inquiry 20, 365424. Progovac, L. (1988) A Binding Approach to Polarity Sensitivity, Ph.D. Dissertation, University of Southern California, Los Angeles.

References

163

(1990a) Free-Choice Bilo in Serbo-Croatian: Existential or Universal. Linguistic Inquiry 21.1, 130-135. (1990b) Inversion as a Purely Structural Phenomenon. Mid-America Linguistics Conference Papers, 277-293, University of Kansas, Lawrence. (1991) Polarity in Serbo-Croatian: Anaphoric NPIs and Pronominal PPIs. Linguistic Inquiry 22.3, 567-572. (1992a) Non-Negative Polarity Licensing Must Involve Comp. Linguistic Inquiry 23.2, 341-347. (1992b) Negative Polarity: a Semantico-Syntactic Approach. Lingua 86, 271299. (1992c) Relativized SUBJECT: Long-Distance Reflexives Without Movement. Linguistic Inquiry 23.4, 671-680. (In press) Non-Augmented NPs in KiNande. In Sam Mchombo, (ed.), Theoretical Aspects of Bantu Grammar, CSLI, Stanford. (In preparation) Binding Domains and Functional Categories in Slavic. In Jindrich Toman, (ed.), Functional Categories in Slavic Syntax. (1993) Locality and Subjunctive-like Complements in Serbo-Croatian. Journal of Slavic Linguistics 1, 116-144, Indiana University, Bloomington. Progovac, L. and P. Connell (1991) Long-Distance Reflexives, Agr-suBJEcrs, and Acquisition. Paper presented at the Second Annual Meeting of the Formal Linguistics Society of MidAmerica. Ann Arbor, Michigan. Progovac, L. and S. Franks (1991) Relativized SUBJECT for Reflexives. 22nd Meeting of the North East Linguistics Society (NELS), University of Delaware, Newark. Quine, W. (1960) Word and Object, MIT, Cambridge, Massachusetts. Radford, A. (1988) Transformational Grammar, Cambridge: Cambridge University Press. Rappaport, G. C. (1986) On Anaphor Binding in Russian. In Natural Language and Linguistic Theory 4, 97-120. Rizzi, L. (1982) Issues in Italian Syntax, Foris, Dordrecht. (1990a) Relativized Minimality, MIT, Cambridge, Massachusetts. (1990b) On the Anaphor-Agreement Effect. Rivista di Linguistica 2,1, 27-42. Rooryck, J. (1991) Negative Islands Without the ECP. Paper presented at the Parasession on Negation, 27th Meeting of the Chicago Linguistic Society [To appear as "Negative and Factive Islands Revisited," Journal of Linguistic Research 28.] Ross, J. R. (1967) Constraints on Variables in Syntax. Ph.D. Dissertation, MIT, Cambridge, Massachusetts. Rudin, C. (1988a) On Multiple Questions and Multiple Wh-Fronting. Natural Language and Linguistic Theory, 6.4., 445-501. (1988b) Multiple Questions in South Slavic, West Slavic, and Romanian. Slavic and East European Journal 32, 1-24. Salamanca, D. (1981) Remarks on Spanish Subjunctives. Manuscript. MIT, Cambridge, Massachusetts. Saxon, L. (1984) Disjoint Anaphora and the Binding Theory. West Coast Conference on Formal Linguistics 3, 242-251.

164

References

Schneider-Zioga, P. (1987) Object/Subject Asymmetries in KiNande, a Bantu Language. Manuscript. University of Southern California, Los Angeles. Scullen, M. E. (1991) Free-Choice Items in English, Serbo-Croatian, and French: A Unified Analysis. Manuscript. Indiana University, Bloomington. Sedivy, J. (1990) Against a Unified Analysis of Negative Polarity Licensing. Cahiers Linguistiques D'Ottawa 18, 95-105. University of Ottawa. Sigurdsson, H. A. (1990) Long-Distance Reflexives and Moods in Icelandic. Syntax and Semantics 24, ed. by J. Maling and A. Zaenen. Sirbu-Dumitrescu, D. (1987) The Syntax of Echo Questions and Other Repetitive Structures (With Special Regard to Spanish and Romanian). A dissertation proposal, University of Southern California, Los Angeles. Stowell, T. A. (1981) Origins of Phrase Structure. Ph.D. Dissertation. MIT, Cambridge, Massachusetts. Thrainsson, H. (1976) Reflexives and Subjunctives in Icelandic. Papers from NELS 6, 225-239. (1990) A Semantic Reflexive in Icelandic. J. Maling and A. Zaenen, (eds.), Syntax and Semantics 24\ Modern Icelandic Syntax, 289-307. Academic Press, San Diego. Travis, L. (1984) Parameters and Effects of Word Order Variation. Ph.D. Dissertation. MIT, Cambridge, Massachusetts. Van Riemsdijk, H. and E. Williams (1986) Introduction to the Theory of Grammar. MIT, Cambridge, Massachusetts. Vendler, Z. (1967) Linguistics in Philosophy. Cornell University Press, Ithaca. Wachowicz, K. (1974) Against the Universality of a Single WH-Question Movement. Foundations of Language 11, 155-166. Watanabe, A. (1991) Wh-in-situ, Subjacency, and Chain Formation. Manuscript. MIT, Cambridge, Massachusetts. Weerman, F. (1989) The V2 Conspiracy: A Synchronic and Diachronic Analysis of Verbal Positions in Germanic Languages. Foris, Dordrecht. Yang, D-W. (1983) The Extended Binding Theory of Anaphors. Language Research 19, 169-192.

Xu, L. J. (1984) Empty Categories in Chinese. Manuscript. Fudan University, Shangai. Zanuttini, R. (1988) Two Strategies for Negation: Evidence from Romance. Proceedings of the Fifth ESCOL. (1989) The Structure of Negative Clauses in Romance. Manuscript, University of Pennsylvania. (1991) Syntactic Properties of Sentential Negation. A Comparative Study of Romance Languages. Ph.D. Dissertation, University of Pennsylvania, Philadelphia.

Index

A'-anaphor 18, 57, 68, 105, 145 Adams 76 adjuncts 34, 40, 42, 52, 58, 59, 60, 101, 102, 107, 108, 116, 117, 118, 136, 137, 139, 140, 147, 150, 154 adversative predicates 2, 8, 43, 55, 62, 64, 65, 67, 68, 70, 71, 75, 77, 81, 86, 87, 89, 92, 93, 95, 96, 126, 127, 129, 130, 148, 151 Allen 148, 151 anaphoric pronominals 17, 18, 20, 45, 49, 50, 68, 128, 147 Anderson 50, 104, 110 Aoun 2, 6, 14, 18, 19, 30, 35, 38, 45, 57, 59, 68, 83, 101, 102, 145, 147, 152 Authier 67, 76, 99, 119, 147, 150 Baker 58, 62, 76, 136, 137 Baltin 11, 114, 146 Barss 73 Basque 67 Battistella 16 Besten 75 binding principles 1, 2, 6, 7, 9, 10, 11, 12, 13, 15, 17, 18, 19, 20, 21, 22, 23, 29, 45, 47, 48, 49, 54, 56, 62, 79, 91, 92, 94, 95, 96, 102, 111, 118, 123, 128, 130, 141, 142, 145, 150, 152 binding theory 1, 2, 3, 6, 10, 12, 18, 19, 20, 22, 23, 45, 54, 62, 79, 90, 92, 94, 96, 107, 120, 123, 128, 141, 145 blocking effects 15 Borer 15 Borkin 97, 101 Brecht 155 bridge verbs; 59, 60 non-bridge verbs 59, 60, 61, 147 Brown 95 Browne 25,28,29 Burzio 17 Carlson 22, 122, 130, 131 Casanova-Seuma 35, 86

case (see also ECM) 5, 18, 25, 29, 47, 74, 105, 107, 111, 114 Catalan 2, 21, 35, 85, 86, 88, 91, 94 Chinese 3, 4, 7, 9, 13, 14, 15, 16, 17, 21, 30, 36, 49, 55, 61, 79, 81, 84, 85, 86, 88, 90, 91, 92, 99, 149, 150, 151 Chominsky 116 Chomsky 10, 11, 12, 13, 15, 16, 17, 18, 19, 21, 38, 47, 57, 59, 68, 74, 76, 83, 98, 105, 107, 108, 111, 114, 118, 119, 121, 140, 145, 146, 148 Cinque 5, 99, 100 clitics 23, 25, 26, 27, 52, 66, 76, 90, 151, 155 Cole 15, 16 comparatives 8, 70, 71, 86, 130, 155 Comrie 25, 26, 115 conditionals 2, 7, 8, 9, 20, 39, 43, 51, 55, 62, 63, 64, 65, 66, 67, 77, 81, 86, 87, 89, 93, 95, 96, 106, 116, 126, 127, 130, 133, 134, 156 Connell 12, 55 coordination 58, 146, 154 Croatian 143 Culicover 76 Czech 4, 37 d-linking 33 Danish 76 Davison 128, 155 De Haan 76 Dogrib 145 downward entailment (see also 'upward entailment') 22, 62, 63, 70, 132, 133, 134, 142 Dutch 71, 90, 155 ECM (Exceptional Case Marking) 5, 18, 22,47, 48, 74, 75, 107, 111 ECP (Empty Category Principle) 12, 18, 20, 57, 59, 60, 109, 140, 146, 147, 148 Emonds 11, 76 EnQ 35, 85, 89, 138, 150, 151

165

166

Index

epistemic modals 20, 75, 77, 78, 149 Erteschik 60 Everaert 50, 104, 110

Jakubowicz 50, 104, 110 Japanese 4, 13, 14, 35, 36, 55, 149 Johnson 50, 104, 110, 112, 152

Faroese 153 Fiengo 57, 58, 146 Finer 145 Frank 75 Franks 12, 55 free-choice items 2, 22, 71, 72, 122, 123, 124, 127, 130, 131, 155 French 3, 4, 6, 76, 104, 106, 109, 110, 112, 113, 119, 131, 150, 152 Frisian 76 Fukui 14

Kartunnen 135 Katada 14, 16 Katz 76 Kayne 4, 6, 34, 109, 146, 149, 152 Kempchinsky 116, 153 KiNande 36 Klima 62, 100, 122 Koopman 75, 100, 148 Korean 3 , 1 1 , 1 3 , 1 4 , 5 5 Koster 75 Kroch 75

Gabriel 131 generalized binding 2, 18, 19, i 8, 145, 152 German 26, 27, 90 gerunds 72, 74, 89 Giorgi 6, 113, 117, 119 Givon 45 Goodluck 29, 30 Greek 49 Greenough 148, 151 Grimshaw 77

Ladusaw 2, 22, 43, 58, 62, 63, 70, 108, 109, 122, 128, 132, 133, 134, 135, 136, 138, 142 Laka 8, 37, 38, 64, 66, 67, 69, 71, 72, 75, 76 Lakoff 101 Lasnik 51, 60, 61, 73, 90, 105, 107, 122, 146, 149 Latin 148, 151 Lawler 97, 101 Lebeaux 12 LF (Logical Form) 8, 13, 14, 16, 17, 18, 20, 21, 22, 30, 31, 32, 33, 34, 38, 39, 46, 51, 54, 56, 57, 58, 60, 61, 70, 73, 74, 77, 79, 80, 82, 85, 88, 91, 94, 104, 105, 106, 107, 108, 109, 110, 111, 114, 115, 116, 120, 121, 136, 137, 138, 139, 140, 141, 145, 146, 147, 149, 150, 154 Li 4, 36, 38, 57, 81, 83, 85, 143, 149 Linebarger 2, 4, 22, 62, 81, 94, 122, 132, 137, 138, 139, 140, 145 Longobardi 8, 84, 154

Halvorsen 49 Hasegawa 4, 34, 35, 55, 149 Heim 140, 148 Helke 14 Hermon 15, 16 Hestvik 18 Higginbotham 38, 57, 58 Hintikka 122 Hoeksema 71, 155 Horn 139 Hornstein 59, 152 Huang 14, 15, 16, 17, 18, 55, 57, 61, 68, 146 Hyams 6, 113 i-within-i 13, 19, 112 Iatridou 34, 49, 114, 154 Icelandic 3. 6, 50, 76, 113, 116, 119, 152, 153, 154 idiomatic NPIs 82, 102, 103, 147, 156 implicature 22, 125, 136, 137, 138, 139, 156 infinitivals 14, 15,41,49, 111, 112, 115, 116, 121, 143, 145 inversion 20, 27,'66, 74, 75, 76, 77, 78, 98, 99, 100, 101, 102, 149 Italian 3, 6, 9, 37, 38, 39, 55, 59, 85, 88, 90, 93, 104, 106, 109, 110, 111, 113, 117, 118, 119, 120, 150, 152

Miiller 79, 84, 85, 90 Maling 113 Manzini 13, 15 Martin 59 May 20, 38, 58, 83, 149 McDowell 20, 77, 78, 149 Milner 4, 6, 45, 152 Milsark 72 minimality 10, 57, 83, 88 relativized minimality 114 modals (see also 'epistemic modals') 2, 20, 22,33,34, 51,76, 77, 105, 116, 122, 123, 124, 127, 128, 130, 131, 149, 154, 155 Montalbetti 30 Nakamura 12 Napoli 113

Index Neg-Raising 139 negative concord 36, 37, 38, 40, 92 Norwegian 17, 49, 76 only

70, 73, 74, 148, 149

Pesetsky 33, 141, 150 Pica 6, 13, 16, 50, 104, 107, 110, 113 Picallo 104, 110 Platzack 76 Polish 33,105,115,116 Pollock 33, 34, 55, 66, 114, 154 Portuguese 131 Postal 76 pro 29, 30, 53, 119, 154, 155 QR 58, 84, 88, 141 quantifiers 1, 5, 8, 20, 29, 30, 31, 32, 33, 38, 39, 48, 58, 62, 64, 67, 70, 71, 82, 83, 84, 88, 89, 100, 122, 123, 126, 127, 131, 132, 134, 137, 139, 141, 146, 147, 150 questions 2, 4, 7, 8, 16, 17, 20, 21, 27, 28, 29, 30, 31, 32, 33, 38, 39, 43, 55, 62, 63, 64, 65, 66, 67, 68, 69, 73, 75, 76, 77, 80, 81, 85, 86, 87, 88, 91, 93, 95, 96, 97, 98, 100, 101, 102, 103, 106, 125, 126, 127, 129, 130, 134, 135, 144, 147, 148, 149, 151, 152 echo questions 149 rhetorical questions 21, 97, 98, 100, 101, 102, 103, 147 Radford 75, 76, 103, 153 Rappaport 11 recoverability 51, 104, 105, 106, 107, 108, 109, 110, 115, 118, 119, 120, 121, 154 Reinhart 146 relative clauses 16, 17, 28, 29, 30, 58, 72, 100, 105, 107, 108, 119, 154, 155 resumptive pronouns 28, 29 Rivero 25 Rizzi 6, 11, 12, 21, 52, 55, 57, 60, 73, 74, 76, 83, 98, 102, 108, 109, 114, 146, 148, 152 Rooryck 74, 98 Ross 57, 58 Rudin 32, 90, 149 Rumanian 105, 115, 116 Russian 3, 7, 9, 11, 13, 14, 16, 37, 79, 90, 91, 92, 95, 105, 115, 116, 149, 152, 153, 155 Saito 51, 60, 105, 107, 149 Salamanca 110 Saxon 145 Schneider-Zioga 74, 98, 148

167

scope 1, 2, 4, 5, 7, 8, 9, 17, 20, 22, 31, 32, 33, 39, 40, 43, 44, 45, 48, 49, 52, 54, 55, 58, 62, 70, 77, 78, 84, 92, 100, 122, 123, 124, 125, 126, 127, 128, 130, 131, 133, 136, 137, 138, 139, 141, 142, 144, 149 scrambling 31, 33, 79, 85, 90 Scullen 131 Sedivy 4, 102, 103 Serbian 143 Serbian/Croatian 2, 3, 5, 8, 9, 17, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 40, 43, 45, 46, 50, 52, 53, 54, 55, 56, 63, 64, 65, 66, 68, 71, 73, 79, 80, 84, 90, 91, 92, 95, 97, 99, 122, 123, 131, 143, 144, 145, 146, 147, 148, 149, 150, 153, 155 Serbo-Croatian 143 Sigurdsson 113, 116, 117, 153 Sigurjonsdottir 6, 113 Sirbu-Dumitrescu 149 Spanish 30, 37, 112, 153 Spec/Head Agr 21, 37, 38, 98, 99, 103, 106, 108, 148, 153, 154 specificity 16, 20, 57, 58, 146 Sportiche 35, 59 Sternefeld 79, 84, 85, 90 Stowell 60, 72 Structure Preserving Principle 114 subjacency 16, 17, 20, 28, 29, 30, 38, 57, 58, 59, 121, 146, 147, 154 SUBJECT

1, 10, 11, 12, 13, 14, 15, 18, 19, 21,

105, 106, 107, 110, 111, 112, 114, 115, 120, 121 accessibility 10, 12, 13, 18, 19, 111 relativized SUBJECT

1, 10, 11, 14, 16, 106,

114 subject orientation 15, 16, 18 subject/object asymmetries 19, 28, 35, 37, 38, 50, 59, 64, 65, 106, 108, 109, 138, 146, 147, 152, 153 Sung 15, 16 Swedish 76 Tang 15, 16, 55, 146 tense 21, 33, 34, 50, 51, 104, 105, 107, 109, 111, 112, 114, 120, 121, 124, 153, 154, 155 Thrainsson 113, 116, 119 topicalization 57, 61, 79, 85, 88, 90 truth 6, 7, 8, 9, 19, 46, 63, 67, 105, 106, 107, 109, 120, 138, 153 Turkish 3, 9, 21, 35, 79, 85, 88, 89, 91, 148, 151 upward entailment (see also 'downward entailment') 132, 133, 135

168

Index

Uriagereka

61, 90

vacuous movement 74 vacuous quantification 57, 83, 89 Van Riemsdijk 149 verb-second 26, 27, 90 Wachowicz 33 Watanabe 38 Weerman 76, 77

Williams Xu

149

30

Yadroll 155 Yang 11, 12, 13,49, 107 Zanuttini 37, 38, 39, 85, 108, 109, 110, 117, 118, 120, 153, 154