Aspects of the Syntax of Agreement

Aspects of the Syntax of Agreement

Routledge Leading Linguists EDITED BY CARLOS P. OTERO, University of California, Los Angeles, USA

1. Essays on Syntax and Semantics James Higginbotham

10. Towards an Elegant Syntax Michael Brody

2. Partitions and Atoms of Clause Structure Subjects, Agreement, Case and Clitics Dominique Sportiche

11. Logical Form and Linguistic Theory Robert May

3. The Syntax of Specifiers and Heads Collected Essays of Hilda J. Koopman Hilda J. Koopman 4. Configurations of Sentential Complementation Perspectives from Romance Languages Johan Rooryck 5. Essays in Syntactic Theory Samuel David Epstein 6. On Syntax and Semantics Richard K. Larson 7. Comparative Syntax and Language Acquisition Luigi Rizzi 8. Minimalist Investigations in Linguistic Theory Howard Lasnik 9. Derivations Exploring the Dynamics of Syntax Juan Uriagereka

12. Generative Grammar Theory and its History Robert Freidin 13. Theoretical Comparative Syntax Studies in Macroparameters Naoki Fukui 14. A Unification of Morphology and Syntax Investigations into Romance and Albanian Dialects M. Rita Manzini and Leonardo M. Savoia 15. Aspects of the Syntax of Agreement Cedric Boeckx

Aspects of the Syntax of Agreement

Cedric Boeckx

New York

London

First published 2008 by Routledge 270 Madison Ave, New York, NY 10016 Simultaneously published in the UK by Routledge 2 Park Square, Milton Park, Abingdon, Oxon OX14 4RN This edition published in the Taylor & Francis e-Library, 2007. “To purchase your own copy of this or any of Taylor & Francis or Routledge’s collection of thousands of eBooks please go to www.eBookstore.tandf.co.uk.”

Routledge is an imprint of the Taylor & Francis Group, an informa business © 2008 Taylor & Francis All rights reserved. No part of this book may be reprinted or reproduced or utilised in any form or by any electronic, mechanical, or other means, now known or hereafter invented, including photocopying and recording, or in any information storage or retrieval system, without permission in writing from the publishers. Trademark Notice: Product or corporate names may be trademarks or registered trademarks, and are used only for identification and explanation without intent to infringe. Library of Congress Cataloging in Publication Data Boeckx, Cedric. Aspects of the syntax of agreement / by Cedric Boeckx. p. cm. — (Routledge leading linguists) Includes bibliographical references and index. ISBN 978-0-415-96254-4 1. Grammar, Comparative and general—Agreement. I. Title. P299.A35B64 2007 415—dc22 ISBN 0-203-93033-9 Master e-book ISBN

ISBN10: 0-415-96254-4 (hbk) ISBN10: 0-203-93033-9 (ebk) ISBN13: 978-0-415-96254-4 (hbk) ISBN13: 978-0-203-93033-5 (ebk)

2007046221

For Youngmi, With fondest love

Contents

Preface Acknowledgments Introduction

ix xi 1

PART I—AGREEMENT RESTRICTIONS

1

Quirky Agreement

21

2

Intricacies of Agreement in Icelandic

42

PART II—AGREEMENT AND AGREE

3

Long-Distance Agreement in Hindi: Theoretical Implications

65

4

Conditions on Agreement in Japanese

77

5

Honorification as Agreement

95

6

Multiplicity: Symmetries and Asymmetries

104

PART III—AGREEMENT AND LOCALITY

7

8

Agree Versus Attract: A Relativized Minimality Solution to a Proper Binding Problem

113

Conflicting C-Command Requirements

130

viii 9

Contents Raising Across Experiencers Cross-Linguistically

10 The Fine Structure of Intervention in Syntax

150 162

PART IV—AGREEMENT AND INTERPRETATION

11 Scope Reconstruction and A-Movement

185

Notes Bibliography Index

215 241 259

Preface

The essays collected in this volume represent work of the past 10 years on issues that I started investigating when I entered graduate school. Chapters 2 and 9 appear here for the first time. The other essays are reprinted in their original form, with the following exceptions: Typographical errors have been corrected; occasionally, a clearer example has been substituted for the original; minor changes have been made to bring the chapters more in line with each other in terms of style and exposition. Also all bibliographical references have been updated, cross-references to chapters have been added, and a unified bibliography provided. Finally, I have attached a brief introductory note to each chapter putting the material in context. Some basic themes and examples appear in several chapters. I’ve let this stand, so as to provide the readers with more or less self-contained essays. I would like to take this opportunity to thank, first and foremost, my syntax teachers at the University of Connecticut, Željko Bošković and Howard Lasnik, for their excellent teaching, admirable mentoring, constant interest, and never-failing encouragement. The atmosphere they managed to create was ideally suited to the pursuit of theoretical linguistics, and remains a model that I seek to emulate. Looking back, it is hard for me to imagine an environment that would be more conducive of research and free exploration of the Chomskyan program for the study of the mind/brain. In addition to my teachers, I owe a special debt to my fellow graduate students, especially Koji Sugisaki, Nobu Miyoshi, Adolfo Ausín, and Debbie Chen Pichler, for help in countless ways. From the very beginning I have benefited from the support and insights of Noam Chomsky, Norbert Hornstein, and Juan Uriagereka. I am most grateful to them. Noam deserves special thanks for reading a poorly written first draft of what is here Chapter 1 and convincing me to do something with it. Kleanthes K. Grohmann has read everything I have written, and made the pieces here, and many others, much better than they would otherwise have been. I feel very fortunate indeed to have a friend like him. I would also like to thank my coauthor Fumi Niinuma for working with me on Japanese honorification, Jean-Roger Vergnaud for pointing to the relevance of case in linguistic theory and for much-cherished discussions

x

Preface

whenever I was around the University of Southern California (USC), Kjartan Ottosson for much-needed help with Icelandic, and the following list of people for valuable comments and discussions, then and since, on the nature of agreement: Halldór Ármann Sigurðsson, Christer Platzack, Anders Holmberg, Esther Torrego, Sam Epstein, Daniel Seely, Paul Pietroski, Adam Szczegielniak, Jairo Nunes, James Yoon, Robert Freidin, Abbas Benmamoun, Rajesh Bhatt, Marcel Den Dikken, Andrea Moro, Massimo PiattelliPalmarini, Heidi Harley, Richard Kayne, Juan Romero, Artemis Alexiadou, Elena Anagnostopoulou, Naoki Fukui, Milan Rezac, Ken Hiraiwa, Shigeru Miyagawa, Alec Marantz, and Mark Baker. Thanks also to the anonymous reviewers for the journals in which some of the chapters appeared, and the audiences to whom some of this material was presented. I am most grateful to Hirohisa Kiguchi, Cilene Rodrigues, Usama Soltan, Pritha Chandra, Tomo Fujii, Brent Henderson, Angel Gallego, and JungMin Jo, for asking me to serve on their thesis committees, and for engaging with the material collected here in their own works. My own students at Harvard deserve special credit for participating in seminars where this material was discussed, and making me feel that the present volume would be worthwhile. Thanks in particular to Bridget Samuels, Dennis Ott, Beste Kamali, Clemens Mayr, Suleyman Ulutas, Hiroki Narita, Balkiz Ozturk, Conor Quinn, Ju-Eun Lee, Taka Kato, Hiro Kasai, Masa Kuno, Soo-Yeon Jeong, Inna Livitz, Clay Kaminsky, and Jeremy Hartman. Three people deserve special mention, and thanks. Carlos Otero invited me to contribute to this extraordinary Routledge series. His interest in my work is an honor. Terje Lohndal volunteered to help me prepare the manuscript for publication, and went well beyond what this work deserves to see the project come to fruition. Last but not least, thanks fall short of expressing my gratitude and admiration for my wife. Youngmi not only contributed to this work by coauthoring Chapter 10, making it the best piece of the whole collection; she also provided the kind of love and support that added significance to the whole project, and convinced me that the road taken was the right one. I dedicate this volume to her.

Acknowledgments

Chapter 1 first appeared in the journal Studia Linguistica 2000, 54: 451–480. It is reprinted here with the kind permission of Blackwell Publishing. Chapter 3 first appeared in the journal Studia Linguistica 2004, 58: 23–36. It is reprinted here with the kind permission of Blackwell Publishing. Chapter 4 (written with F. Niinuma) first appeared in the journal Natural Language and Linguistic Theory 2004, 22: 453–480. It is reprinted here with the kind permission of Springer. Chapter 5 first appeared in the journal Natural Language and Linguistic Theory 2006, 24: 385–398. It is reprinted here with the kind permission of Springer. Chapter 6 first appeared in Cedric Boeckx and Kleanthes K. Grohmann (eds.), Multiple Wh-fronting, 2003, John Benjamins, pp. 13–22. It is reprinted here with the kind permission of John Benjamins. Chapter 7 first appeared in Artemis Alexiadou (ed.), Formal Approaches to Universals, 2002, John Benjamins, pp. 41–64. It is reprinted here with the kind permission of John Benjamins. Chapter 8 first appeared in the journal Studia Linguistica 1999, 53: 227–250. It is reprinted here with the kind permission of Blackwell Publishing. Chapter 10 (written with Y. Jeong) first appeared in Chungja Kwon and Wonbin Lee (eds.), Issues in Current Linguistic Theory: A Festschrift for Hong Bae Lee, 2004, Kyungchin, pp. 83–116. It is reproduced here with the kind permission of Kyungchin. Chapter 11 Chapter 5 first appeared in the journal Natural Language and Linguistic Theory 2001, 19: 503–548. It is reprinted here with the kind permission of Springer.

Introduction

This collection of essays deals with various aspects of the syntax of agreement. Taken together, the chapters that follow this introduction seek to identify properties of agreement systems in natural language that once properly analyzed would illuminate our understanding of the language faculty, and ultimately the human mind. I do not think that I have identified enough such properties of agreement systems for me to be able to present a full-fledged theory of agreement in this volume, but I hope that the analyses provided here point to a fruitful direction of research, one that will eventually lead to a characterization of agreement and its role in grammar that is adequate at both the descriptive and explanatory levels. This introduction attempts to situate the chapters that follow in a broader context of linguistic inquiry, and highlights the theoretical relevance of the aspects of agreement that I have chosen to include in this volume.

1. THE EMERGENCE OF AGREEMENT AS A THEORETICALLY SIGNIFICANT VARIABLE Although the agreement relation between a Noun Phrase (typically, the ‘subject’) and a Verb was already captured in Chomsky’s (1957) Syntactic Structures (see Transformation 15 on p. 112), it was not until the Principlesand-Parameters era that the outlines of a generative theory of agreement emerged. There are several reasons why the significance of agreement only became clear at a relatively late stage in the development of linguistic theory. First, until the Principles-and-Parameters era there was no formally coherent notion of Inflection; hence there couldn’t be any formally coherent notion of that part of the inflectional paradigm of a language we call agreement. Pollock’s (1989) justly influential ‘split-Infl’ hypothesis filled this important lacuna. Second, it was only after Jean-Roger Vergnaud insightfully pointed out in a famous letter to Noam Chomsky and Howard Lasnik in 1977 that purely formal features well known to traditional grammarians like case (and agreement) may lead to considerable simplification of formal statements

2

Aspects of the Syntax of Agreement

(such as filters) in the theory that linguists came to realize that a detailed investigation of the properties of such formal features may shed light on the nature of the language faculty. Third, the very recent formulation of a minimalist program for linguistic theory (Chomsky 1993), which seeks to go beyond explanatory adequacy and answer why the language faculty has the properties that previous research uncovered, forces linguists to provide a rationale for apparent ‘aberrations’ (“imperfections”) such as case and agreement (properties that at first sight fall short of ‘virtual conceptual necessity’). Consider case. Vergnaud gave us (to my mind, compelling) reasons to believe that case is more than mere morphological clothing. He was indeed the first to note that licensing of case features on noun phrases may to a very large extent explain the syntactic behavior and distribution of these noun phrases. But if case is at the heart of the syntax of noun phrases, why is it that it fails to receive an interpretation? The same could be said about agreement proper: Agreement is clearly present on verbal forms in the syntax, but it is never interpreted there. It is as if formal features like case and agreement appear in the syntax only to be wiped out before syntactic elements are interpreted. Perhaps uninterpreted features like case and agreement are like the irritants that give rise to pearls. Pearls, one must remember, begin as irritants; sand, a pebble, or pesky parasitic organisms get inside the oyster’s shell. To reduce irritation, the oyster coats the intruder with layers of a solid, slick material called nacre, ultimately giving rise to the thing of beauty we are all familiar with. Irritants, then, are the engine, the driving force of pearl formation. It may not be too far-fetched to think of case and agreement as parasitic, ‘misplaced’ features that drive syntactic computations. This was in fact the intuition behind Chomsky’s (1986a: 137, 201) introduction of the principle of Last Resort into syntactic theory—a principle from which linguistic minimalism was born. Syntactic processes switch into higher gear to make sure that formal features are licensed. Once all the formal features of an element have been appropriately licensed, the element becomes syntactically inert. Syntax, for that element, ends, and the interfaces take over. The picture just sketched reflects the results achieved after decades of intensive investigation into the nature of formal features. The investigation focused on three core issues—the three major parts of this volume: (I) the range of formal features that count as ‘irritants’ (all instances of case? all instances of agreement?); (II) the syntactic mechanisms needed to license formal features; and (III) the conditions imposed on such mechanisms to limit their power and range, in accordance with the search for a restrictive theory of syntax. (The fourth part of this volume focuses on the interpretive consequences of formal feature licensing.) Following Vergnaud’s insight, the study of formal features began with case, more specifically, those instances of case that are devoid of any clear interpretive correlate, dubbed structural cases in Chomsky (1986a: 202–3). However, it was soon recognized that case and agreement are intimately related—perhaps not always, perhaps not in all languages, but the relation was felt to

Introduction

3

be sufficiently strong as to extend the mechanisms that had been used for case licensing to the domain of agreement proper (licensing of “phi-features” such as person, number, and gender/class). As a matter of fact, in recent years, inquiry into the nature of formal features begins with phi-features, perhaps because, unlike case features, which never seem to be interpreted on any syntactic elements, phi-features are interpreted on nominals, and as such do not consistently beg questions about why such features emerged in the first place. Be that as it may, when I retrace the history of how formal features have been accommodated in modern generative grammar in the pages that follow, I will not always indicate whether a particular mechanism or configuration was first introduced to handle case or agreement. For much of what follows in this introduction, case and agreement can be treated as two sides of the same coin.

2. THE ROAD TO AGREE The development of case/agreement theory nicely illustrates the interplay between conceptual arguments and empirical considerations. All else equal, one would favor theories that provide a unique licensing mechanism for all instances of case/agreement. This drive toward symmetry was present from the very beginning of investigations into case and agreement. It was at the heart of the development of the theory of government. Unfortunately, as we will see, the unification achieved by government was not particularly satisfactory. From the very beginning of the Principles-and-Parameters era, four manifestations of structural case played a major role in the formulation of case/agreement theory: nominative case assignment to subjects, accusative assignment to objects, accusative case assignment to derived objects in Exceptional-Case Marking (ECM) contexts, and nominative case assignment to associate nominals in existential constructions. Although linguists pushed the idea that all these instances of case were assigned uniformly under government, it was clear to everyone that government was a cover term for a variety of configurations that didn’t have much in common. Thus, nominative Case to subjects was assigned under m-command by IP, as in (1). (1)

InflP ru NPi Infl’ ru Infl VP 6 . . . ti . . . [IP Johni [ I0 [VP ti kick the ball]]]

4

Aspects of the Syntax of Agreement

Accusative Case, by contrast, was assigned under c-command by V (2). (2)

VP | Vb ei V NP [IP Johni [ I0 [VP ti [V’ kick [NP the ball]]]]]

So both specifiers and complements were relevant ‘slots’ for case assignment. In addition, the definition of government had to be modified to encompass two more configurations, one for accusative case assignment in ECM contexts (targeting the specifier of the complement) (3), and a mechanism of case-transmission from specifier to the complement domain for nominative case assignment (in the so-called existential construction) (4). (3)

VP | Vb ei V InflP ei NPi Inflb ei Infl VP 6 . . . ti . . . [IP John [ I0 [VP [V’ believe [IP [NP the ball]i [I’ to [VP ti be blue]]]]]]]]

(4)

InflP ei Therei Inflb ei Infl VP | Vb ei V NPi [IP there [I’ is [VP a man in the room]]]

The prospects for a truly uniform theory of case configurations changed when the clausal skeleton was enriched, under the impetus of Pollock’s

Introduction

5

(1989) seminal study. By dissociating the Tense component and the Agreement component of IP, and treating each as a distinct projection (TP and AGRP), Pollock made it possible to begin asking precise questions about agreement (many of which are still with us today). In particular, it led Chomsky to make a key observation. Chomsky (1991: 146) noted that although Pollock originally split up Infl into Tense and AGR with TP dominating AGRP, one may have expected AGRP to dominate TP to capture subject–verb agreement in finite clauses, the logic here being that if T is [+finite], then AGR is available for subject–verb agreement. The only way to express this dependency of AGR on T[finite] in a framework that assumes that syntactic structures are built bottom up and in a cyclic fashion is to have AGR project once T has projected. Chomsky furthermore observed, following Belletti (1990), that in languages where T and AGR are distinct morphemes, agreement is expressed further away from the stem than tense is, as in the following French example: (5)

tu parle-ra-s you talk-future-2sg

By the logic of Baker’s (1985) Mirror Principle (a generalization that expresses that the order of morphemes mirrors the order of functional projection), (5) should correspond to the structure in (6). (6)

[AGRP [TP [VP V] T] AGR]

However, as Chomsky (1991: 147) noted, Pollock’s treatment of the wellknown verb-positioning contrast between French and English (7)–(8), as well as Chomsky’s own treatment in (1991), demand that TP dominate AGRP. (7)

a. J’embrasse souvent Marie. b. *Je souvent embrasse Marie.

(8)

a. *I kiss often Mary. b. I often kiss Mary.

Chomsky’s solution to this paradoxical situation was to assume the existence of two AGR projections, one dominating TP (and responsible for subject–verb agreement) and the other dominated by TP. Chomsky suggested that the lower AGR be the locus of object agreement. From this point on, the higher instance of AGRP came to known as AGRsP and the lower instance of AGRP as AGRoP. The structure in (9) became the basic clause structure for many researchers. (9)

[AGRsP AGRs [TP T [AGRoP AGRo [VP V]]]]

6

Aspects of the Syntax of Agreement

Chomsky noted that the identification of the lower AGRP as AGRoP fitted nicely with Kayne’s (1989a) study of agreement between (displaced) objects and active past participles in French, illustrated in (10). (10) Quelle fille Jean a(-t-il) vu-e? which girl Jean has-he seen-agr.fem ‘Which girl did Jean see?’ Sentences like (10) clearly show the existence of an agreement relation established lower than the position of the finite auxiliary and distinct from subject–verb agreement, precisely what the structure in (9) leads us to expect. Chomsky furthermore suggested that structural case is correlated with agreement (see already Chomsky 1981: 52), and reflects a structural relation between the relevant NP and the appropriate AGR element. Accordingly, AGRsP became associated with nominative case assignment, and AGRoP with accusative case assignment. Here the phenomenon of Object Shift in Scandinavian languages provided yet another rather strong piece of evidence for the structure in (9). As originally discussed in Holmberg (1986), objects in Icelandic may shift (move outside the VP, whose edge can be identified with negation), but only if they are Case-marked NPs, not PPs. (11) Nemandinn las bókinna ekki. student.the read book.the not ‘The student didn’t read the book.’ (12) *Jón talaði [við Maríu] ekki. Jon spoke with Maria not ‘John didn’t speak with Maria.’ This shifting process was plausibly reinterpreted as movement driven by structural-case reasons, and AGRoP appeared to provide just the position needed to host such shifted objects. Pretty soon many phenomena were reinterpreted as evidence for the need to establish a spec-head relation for the licensing of case/agreement. This theoretical shift is best illustrated by means of French data about past participle agreement like (13), and analyzed in a seminal paper by Kayne (1989a). (13) a.

Jean a vu-*e la fille. Jean has seen-agr.fem the girl ‘Jean saw the girl’ b. Jean l’a vu-e. Jean her-has seen-agr.fem ‘Jean saw her.’ c. Quelle fille Jean a(-t-il) vu-e? which girl Jean has-he seen-agr.fem

Introduction

7

‘Which girl did Jean see?’ d. Cette fille a été vu-e. this girl has been seen-agr.fem ‘This girl was seen.’ As can be seen in the examples just given, past participle agreement is only possible with displaced objects (cliticized, wh-moved, and passivized in the examples at hand). Quite naturally, linguists took agreement in this case, and, by hypothesis, in all cases, to require raising to some specifier position—the spec-head relation. This line of thought has been very productive (see Chung 1998, Koopman 1992, 1995, 2001, and Sportiche 1998, to cite but a few examples that make abundant use of Spec-Head agreement relations), and was at the core of the early minimalist conception of Case/agreement (Chomsky 1993). Finally, a primitive relation like spec-head emerged as the likely candidate to cover all instances of case/agreement licensing. Rather strong evidence for this came from a reexamination of ECM. Lasnik and Saito (1991) revived original observations by Postal (1974) that strongly indicate that accusative case-marked nominals in ECM have raised into the matrix clause. (14) The DA proved [two meni to have been at the scene of the crime] during each otheri’s trials. (*?The DA proved [that two meni had been at the scene of the crime] during each otheri’s trials) (*The DA proved [there to have been two meni at the scene of the crime] during each otheri’s trials) (15) The DA proved [no suspecti to have been at the scene of the crime] during hisi trial. (*?The DA proved [that no suspecti had been at the scene of the crime] during hisi trial) (*The DA proved [there to have been no suspecti at the scene of the crime] during hisi trial) (16) The DA proved [no onei to have been at the scene of the crime] during anyi of the trials. (*?The DA proved [that no onei had been at the scene of the crime] during anyi of the trials) (*The DA proved [there to have been no onei at the scene of the crime] during anyi of the trials) (17) *Joan believes himi to be a genius even more fervently than Bobi does. (Joan believes that hei is a genius even more fervently than Bobi does)

8

Aspects of the Syntax of Agreement (18) John believes Bob incorrectly to be a genius. (19) John made Bill out to be a fool.

The movement was plausibly interpreted as movement to SpecAGRoP (with subsequent movement of the main verb to some high projection to yield the observed VO word order). The drive toward a generalized Spec-head AGRbased case/agreement theory was such that even when raising didn’t seem to take place overtly, it was hypothesized to happen in covert syntax (extending a logic familiar from work on wh-in-situ). Such covert raising fits like a glove with the most popular analysis of existential constructions at the time (Chomsky 1986a, 1991), according to which the associate nominal in sentences like there were two men in the room either adjoins or literally replaces the expletive in covert syntax, thereby establishing the now-required spec-head configuration for nominative case assignment. Assuming literal expletive replacement, at LF, a sentence like there is a man in the garden looks like (20). (20) [A man]i is [ti in the garden]: LF-expletive replacement The Expletive Replacement Hypothesis straightforwardly accounts for the somewhat unusual agreement configuration that obtains in existential constructions. Descriptively, the finite verb in existential constructions agrees with the associate NP to its right, not with the element in SpecTP, which appears to be the more common agreement configuration in English (and many other languages; so-called ‘Spec-Head agreement’). Contrast (21a) and (21b). (21) a. There are/*is three men in the car. b. They are/*is one and the same element. The common agreement configuration obtains in existential constructions, albeit at LF. This adds a certain twist to case/agreement. Because we observe the agreement despite the fact that it is established covertly, agreement (and, by symmetry, case) was taken to be something that requires licensing/checking, not some ‘empty’ featural slot that requires filling (assignment). There were rather strong arguments in favor of the expletive-replacement hypothesis. Aside from the agreement issue, the analysis was also able to explain why expletives must have associate NPs (if there is no associate, the expletive can’t be replaced, and the sentence will be LF-deviant (cf. *there is in the garden), if we assume, plausibly enough, that meaningless elements like expletive there cause a sentence to crash at the interfaces). More generally, this approach provides an explanation for why expletive– associate pairings pattern with chains of A-movement, as in the contrast in (22):

Introduction

9

(22) a.

*A man seems [t has been arrested]. (cf. ‘A man seems to have been arrested.’) b. *There seems [a man has been arrested]. (cf. ‘There seems to have been a man arrested.’)

Any version of the Last Resort Condition can account for the ungrammaticality of (22a), as a man has its Case/agreement requirements met in the lower clause; by essentially the same reasoning, this type of explanation should also account for the parallel (22b) involving an expletive–associate pair, entailing the, this time inappropriate, displacement of the associate. But, despite its obvious virtues, the expletive replacement analysis was criticized as soon as it was proposed. As Lori Davis first observed in Chomsky’s 1985 class at MIT, with many researchers following her (see Lasnik 1999a for references), the expletive-replacement/covert spec-head analysis gets the scope facts wrong. Typically, indefinites in subject positions are scopally ambiguous (see (23a)). The covert rasing analysis predicts that such ambiguity should exist in existential constructions as well, contrary to fact. The associate in (23b) only has the narrow scope reading. (23) a.

Someone from New York is likely to be at the party. (someone >> likely / likely >> someone) b. There is likely to be someone from New York at the party. (likely >> someone / * someone >> likely)

Likewise, contrary to what the covert raising analysis would lead us to expect, associates in existential constructions are unable to establish a binding relation with elements that they would c-command after covert raising, as shown in (24b). (24) a. A mani seems to himselfi to be doing something wrong. b. *There seems to himselfi to be a mani doing something wrong. Data like (23b–24b) revealed the first crack in the generalized spec-head edifice being erected—a crack that Chomsky (1995) took to be a fatal flaw. First, Chomsky questioned the legitimacy of functional heads like AGR, which consist exclusively of features that are never interpreted. By eliminating AGRPs, Chomsky removed the hosts of the movement that made the generalized spec-head agreement theory tenable. Second, Chomsky questioned the motivation of actual displacement for purposes of feature checking. He did so in two steps. The first step was taken in Chomsky (1995). There Chomsky noted that if movement is driven to check features that would otherwise be illegitimate at the interfaces, it is natural to expect that ‘the operation Move [. . .] seeks to raise just F[eature]’ (Chomsky 1995: 262). We therefore expect under

10

Aspects of the Syntax of Agreement

minimalist assumptions that, if possible, the computational component can raise just what is needed (features to carry out the checking operation), leaving behind any extra lexical material. Relying on the feature-movement hypothesis, Chomsky proposes that in existential constructions only formal (K) features of the associate NP move (head-adjoin) to Infl, leaving all phonological and semantic features behind. Raising of K-features immediately accounts for the fact that finite agreement in existential constructions is controlled by the feature specification of the associate. As Lasnik (1999a) showed, the feature-movement account provides a straightforward explanation for the narrow scope of the associate NP in (6b), assuming that the establishment of scopal relations requires a full phrasal category, and not just a raised feature. Although the feature-movement analysis captured the core properties of existential constructions, it was silent regarding those instances of featuremovement accompanied by displacement of the full category containing the relevant features (those cases that lent strong empirical support for the spechead theory). Chomsky (1995) hypothesized that phrasal displacement was forced by morpho-phonological considerations that were independent of case and agreement. Chomsky took his second step toward the complete divorce of movement and case/agreement-licensing in Chomsky (2000). There Chomsky questioned the legitimacy of feature-movement (specifically, the fact that feature-movement took the form of head-adjunction, an operation that posed problems for the definition of cyclicity), and proposed the operation Agree. Agree was defined as a process of long-distance feature checking (or valuation) with no displacement. The configuration in which it obtains is very reminiscent of the notion of long-distance government under c-command (see Raposo and Uriagereka 1990). In an Agree analysis, a Probe (a functional head) searches inside its c-command domain for a Goal (the agreeing element) with a matching feature. Once the Goal is found, it checks the features of the Probe, triggering agreement. Any displacement of the agreeing category was assumed to be motivated by an independent Fill-Spec/ EPP-requirement. Like the generalized spec-head analysis, the Agree-based theory takes all instances of case-agreement to be established in the same manner, but this time movement does not figure as part of the theoretical symmetry established. The Agree-analysis expresses Chomsky’s intuition that existential constructions display the mechanism of case/agreement-licensing in all its purity. It would be unfair of me to fail to mention that there are many alternative analyses of existential constructions, many of them dispensing with the idea that the associate NP is directly responsible for agreement on the finite verb, and/or the idea that the associate NP receives case from finite Infl (see Moro 1997, Belletti 1988, Lasnik 1992, Hornstein and Witkos 2003, among many others). If correct, these analyses may enable us to maintain a generalized spec-head analysis for case/agreement-licensing. The data discussed in many

Introduction

11

chapters of this volume suggest that this would be the wrong move, and that Agree is a better way to unify all instances of case/agreement (for converging evidence, based on data not discussed here, see Soltan 2007). Before turning to a brief discussion of the kind of data required to make an argument in favor of Agree, let me sum up the preceding discussion by highlighting the desire for symmetry in the formulation of grammatical principles. Empirical evidence suggests two ways to proceed: Either all instances of case/agreement pattern like nominative case for canonical subjects, i.e., licensing under Spec-head, or else, all instances of case/ agreement pattern like accusative case for canonical objects, i.e., licensing under Head-complement (/Agree). The situation exemplifies the spirit of minimalist research: At the heart on our inquiry is a conceptual desire for good design (symmetry). This is what defines the research program. Empirically (/technically), the program can be articulated in two distinct ways: Spec-head or Head-complement (Agree). The challenge was (and still is) to determine which way will lead most naturally to a comprehensive treatment of case/agreement, one meeting the minimalist desideratum of symmetry.

3. AGREEMENT IN THE PRESENT VOLUME AND BEYOND To establish the superiority of Agree (i.e., separate case/agreement from the possibility of movement), it was necessary to go past the many factors that are involved in existential constructions and look at a variety of cross-linguistic phenomena that reveal (i) instances of movement unaccompanied by case-agreement licensing (quirky subjects; Chapters 1 and 2), (ii) instances of case-agreement for which there is no plausible movement source (Hindi long-distance agreement; Chapter 3), and (iii) instances of restrictions on case/agreement that follow straightforwardly if case/agreement can be established in the absence of displacement, but that would be expected to be voided if movement could reorder elements before case/ agreement is licensed (bleeding effects) (Japanese honorification; Chapters 4 and 5; Multiple Agree relations; Chapter 6). Agree can also be pressed into service to account for various asymmetries in locality effects that would be difficult to explain if case/agreement relations always required movement. Chapters 7–10 examine such asymmetries. It is hardly necessary to point out that all of the analyses developed in the following chapters remain controversial, and subject to reinterpretation (something I indulge in myself, as I reinterpret parts of Chapter 1 in Chapter 2, for example). But I suspect that the chapters that constitute the bulk of this collection will remain theoretically relevant even if they turn out to be wrong on matters of detail, because they show how an Agree-based analysis of case and agreement may be empirically substantiated. Specifically, they show that in addition to existential constructions, other kinds of data,

12

Aspects of the Syntax of Agreement

such as long-distance agreement or agreement asymmetries, may provide evidence for a theory that divorce case/agreement from movement. They also show the kind of data that will have to be looked at if one attempts to refute an Agree-based theory (for valuable attempts, see Chandra 2007, and Hornstein 2007; but see Boeckx 2007a,b). Having said this, let me stress that even if the arguments in favor of Agree provided here turn out to hold, some rather pressing questions will have to be addressed in future work. First, if case and agreement are truly divorced from movement, what accounts for EPP-effects pertaining to A-chains (raising, passive, etc.)? Ever since it was proposed in the early days of the Principles-and-Parameters approach (see Chomsky 1981, 1982), the so-called Extended Projection Principle (EPP) has stood out as an important generalization in need of a deep explanation. Implicating case and agreement in the formation of A-chains, as in the generalized spec-head theory, gave us the hope of a substantive hook on which to hang our understanding of that formal requirement. Put differently, the generalized spec-head approach gave us the hope that the EPP would become a problem. But once case and agreement are divorced from movement, as in an Agree analysis, it becomes harder to even begin to unravel the EPP-issue, keeping the latter firmly in the category of mysteries. Chomsky (2004: 114) is right to point out that “(. . .) raising of B (. . .) is always restricted to some category of constituents (. . .), hence some feature of B (or complex of features).” What the range of possible features identifying elements satisfying the EPP is still remains an open question. Chapter 6 of the present work suggests that a decomposition of phi-features and a careful examination of their (distinct) interface properties may shed light in this important issue, but much more remains to be done before even considering this avenue of research worth pursuing (for an attempt, see Boeckx 2007a). Another issue that the present volume fails to resolve is the nature of the relation between case and agreement. That a relation exists between the two can hardly be doubted, but which form this relation takes is unclear. What exactly does it mean to say that case and agreement are two sides of the same coin? Is case always the reflex of an agreement relation, as Chomsky has suggested in recent work (see Chomsky 2000; see also Chapter 3)? Or should case be regarded as the nominal counterpart of uninterpreted formal features (agreement) found on verbal functional heads, as suggested by Pesetsky and Torrego 2001 (see also Chapter 6)? Regrettably, the data discussed in the following pages are not conclusive enough to enable me to take a confident stance on this issue, and I must leave this question for future work (for relevant considerations, see Boeckx 2007a). The two big issues just raised do not exhaust the range of questions that the present work is silent on. Take, for example, the range of variation in the morphological expression of agreement, which is quite extensive: Apart from well-known cases of suffixation and prefixation, one also finds cases of fusion, circumfixation, infixation, portmanteau morphemes, fission, etc.

Introduction

13

Interestingly, Cinque (1999: 127) notes on the basis of an extensive survey of the world’s languages that [by the logic of the mirror principle], “the order [of functional heads] appears crosslinguistically invariant. The limited cases of apparent variation all seem to involve agreement and negation.” But such cases are not limited at all. Concluding her survey of over 500 genetically unrelated languages, Julien (2000: 359) observes that “there is one inflectional category which does not so easily fit into the . . . rigid framework that syntactic analyses [assuming a rigid mirror principle] provide. This category is agreement.” Why this is the case is far from clear. Likewise, several hypotheses were formulated in the GB era that tied (morphological) richness of agreement to phenomena such as pro-drop (see Rizzi 1982, Jaeggli and Safir 1989) or polysynthesis (Jelinek 1984, Baker 1996). But to this day no satisfactory characterization of “rich agreement” has been formulated, despite repeated attempts. And yet few would deny that agreement is deeply implicated in these phenomena. Another area that merits attention is the extent to which AGRP, if it exists, should be split in smaller projection reflection Person, Number, and Gender agreement, respectively. One of the phenomena discussed in this context is the nature of nominative NPs in Icelandic, discussed in Chapters 1 and 2. Although the language does not impose any K-feature restriction on nominative subjects, it prohibits non-3rd person nominative objects. Sigurðsson (1996) and Taraldsen (1995) (see also Sigurðsson and Holmberg in press for additional evidence) took the contrast to indicate that agreement with nominative objects in Icelandic was restricted to 3rd person, which, following Benvéniste’s well-known suggestion, they took to mean absence of Person. Hence agreement is restricted to Number. This led them to claim that AGRoP licensing nominative objects is a NumberP, whereas nominative subjects can relate to a PersonP. Similarly, the fact that past participle agreement is often restricted to Number and Gender (no Person) suggests that finer-grained distinctions are needed in the characterization of AGRP. A comprehensive theory of agreement should also touch on the relation between agreement and clitics, the general weakening/impoverishment of agreement with post-verbal subjects, the phenomenon of anti-agreement in the context of wh-extraction, the nature of wh-agreement (agreement established by moving wh-phrases), the near-total absence of overt agreement in some languages, the general lack of agreement relations with covertly moved NPs (such as covertly raised quantifiers), differences between agreement and concord, agreement within nominals, adjectival agreement, and so on. All of this will have to await future research. In concluding this introduction, I would like to stress my belief that Vergnaud’s insight that purely formal features like case and agreement play a key role in linguistic theory is likely to have many more lasting repercussions than what we have been able to gather from the syntax of agreement and the mechanisms underlying it. Chapter 11 of the present collection argues

14

Aspects of the Syntax of Agreement

that a detailed investigation of case-licensing may not only account for the distribution of nominals, but may also determine their interpretive properties. Specifically, case-licensing may help us characterize the range of reconstruction effects found in the domain of A-chains. (Boeckx and Hornstein in press suggest that case/agreement-licensing considerations may even bear on reconstruction effects in A-bar dependencies.) In independent work (see Boeckx 2003b, to appear; see also Boeckx and Hornstein 2006, and Rodrigues 2004), I have argued that ‘strong’ agreement (taken to mean agreement in all phi-features) has a major effect on island-formation. In the course of arguing for a specific (movement) analysis of structures involving resumptive pronouns, I was struck by the fact that in many cases, the resumptive pronoun and the wh-phrase antecedent were not in an agreement relation. The examples in (25)–(28) illustrate the lack of agreement (i.e., lack of phi-feature identity) one finds in language after language in the context of resumption.

Anti-Person Agreement (25) A Alec, tusa a bhfuil an Béarla aige. Hey Alec you aN is the English at-him ‘Hey Alec you that know(s) English.’

(Irish)

Anti-Number Agreement (26) Na daoine a chuirfeadh isteach ar an phost sin. the men C put-cond-3sg in for the job that ‘The men that would apply for that job.’

(Irish)

Anti-Gender Agreement (27) Dè a’mhàileid a chuir thu am peann ann? (Sc. Gaelic) Which the.bag.fem C put you the pen in.3.masc ‘Which bag did you put the pen in?’

Anti-Case Agreement (28) a.

Bha thu a’geàrradh na craoibhe. be.pst you cutting the three.gen ‘You were cutting the tree.’ b. Dè a’chraobh a bha thu a’geàrradh? which tree.nom C be.pst you cutting ‘Which tree were you cutting?’

(Sc. Gaelic)

Based on such facts, Boeckx (2003b) hypothesized that lack of agreement is a pre-condition on successful extraction (lack of opacity/islandhood).

Introduction

15

There is independent evidence that movement is closely related to what is known in the literature as ‘anti-agreement’ effects. Consider the following examples from Northern Italian dialects. (29) a.

La Maria l’ è venuta. the Maria she is come ‘Maria came.’ b. Gli è venuto la Maria. It is come the Maria ‘Maria came.’ c. Quante ragazze gli è venuto con te? how.many girls it is come with you ‘How many girls came with you?’ d. *Quante ragazze le sono venute con te? how.many girls they are come with you e. Chi hai detto che __ e partito? who has said that is left ‘Who did he say that left?’

(Italian)

(29a) shows that preverbal subjects relate to a K-feature matching clitic (3rd fem. sg). By contrast, post-verbal subjects don’t (29b), the clitic bearing default morphology. (29c) and (29d) show that subject extraction requires the use of a non-agreeing clitic. Based on (29c), we can conclude (as did Rizzi 1982) that in standard Italian examples like (29e), subject extraction takes place from a post-verbal position related to a silent non-agreeing clitic (pro), which obviates the [that-trace] effect. In a similar vein, the fact that objects are islands for standard wh-extraction in languages with object agreement, like Basque (30), suggests that agreement tends to turn an otherwise transparent element (object) into an island. (30) *Nori buruzko sortu zitusten aurreko asteko istiluek who about-of create scandals last week scandals zurrumurruak? rumors ‘Who have last week’s scandals caused [rumors about]?’ (Basque) Why should agreement have this effect? Boeckx (2003b) argues that chains can contain at most one ‘strong’ positions, where ‘strong position’ can be equated with [+wh]-checking position or with ‘strong agreement’ (see Boeckx 2003b for a more precise characterization; for a similar intuition, see Richards 1997, 2001, Rizzi 2006, Rizzi and Shlonsky 2007). As a result of this ban on “chains that are too strong,” elements that normally agree (“A”-type agreement, or what Chomsky 2001

16

Aspects of the Syntax of Agreement

calls ‘complete K-feature agreement’) must fail to agree (i.e., ‘anti-agree’) in order for them to successfully enter into a checking relation with an A-bar target (Wh-/A-bar-feature checking). The intuition behind Boeckx’s analysis is Last Resort. Once a chain contains a strong position (of any kind), it cannot contain another, equally strong position (of any kind). I have shown in Boeckx (2003b) that this view helps us understand resumptive structures, and the lack of island effects in such contexts. (Boeckx claims that the role of a resumptive pronoun is to take care of “A-style” agreement, freeing the wh-phrase from establishing a successful relationship with an A-bar target.) I believe the approach captures the subject-part of the CED, and it may extend to other instances of ‘opaque’ domains (for additional evidence, see Rackowski and Richards 2005, and especially Gallego 2007 and Henderson 2006; for a somewhat different take on the relevance of agreement in the domain of extraction, see Mayr 2006). If true, this line of research will show agreement playing a key role in a module of grammar where its role has so far been neglected. Yet another module of the grammar where agreement may play a bigger role than previously assumed is phrase structure, and the characterization of the most fundamental process in syntax, Merge. Since Chomsky (1993) it has become standard to regard movement as a special instance of merge (specifically, re-merge of an item already part of the domain dominated by the element targeted by ‘movement; what Chomsky 2004 calls ‘Internal Merge’). Although, as noted earlier, the relationship between feature-checking and movement is far from clear, many linguists would bet that some feature (Agree-relation) must underlie movement (if only because some such Agree relation appears necessary to identify the right kind of moving elements). As Boeckx (2003a, to appear) notes, if move is just a special instance of merge (‘internal merge’), then symmetry considerations would lead us to claim that all instances of merge, including standard (‘external’) merge, are licensed by some featural (Agree) relation. Put differently, no application of merge is free (see also Collins 1997, Chomsky 2000, Hornstein 2001, Pesetsky and Torrego 2007, Gallego 2007, Moro 2007, Mayr 2007). Boeckx (2003a, to appear) contends that understanding (external) merge as consisting of an Agree-relation (equivalently, Probe-Goal relation) may help us make sense of the asymmetry at the heart of merge referred to as labeling or projection. Exploring this idea here would take me too far afield, but I want to mention that, if correct, the study of Agree may well shed light on the nature of endocentricity and the specific kind of hierarchy used in language—properties that may help us characterize the defining features of the faculty of language in the narrow sense (see Hauser, Chomsky, and Fitch 2002; Boeckx, to appear). I could summarize all the remarks made in this introduction by saying that I regard the study of agreement understood as the investigation of the

Introduction

17

properties of formal features and of the mechanisms required to license them at the interfaces as a truly unique window onto the architecture of the language faculty. The present study focuses on only a few aspects of the syntax of agreement. I hope that the reader of this collection will get a sense of the importance of detailed, cross-linguistic syntactic investigations to address questions pertaining to agreement. Above all, I hope that the reader will recognize the relevance of those aspects of the syntax of agreement discussed here to some future formulation of a theory of agreement, and of syntactic relations.

I

Agreement Restrictions

1

Quirky Agreement

Verb agreement with a nominative element is severely constrained in Icelandic when a Quirky subject is present: only partial (number, not person) agreement obtains. Here I propose an account of this restriction by taking it to be an instance of Bonet’s (1994) Person-Case-Constraint, which blocks ‘object’ person agreement when agreement with a dative element takes place. For the Person-Case-Constraint account to hold, it is necessary to claim that agreement obtains with Quirky subjects, although it fails to show up morphologically on the verb for non-syntactic reasons. It is also crucial to assume that case and agreement can be licensed ‘at a distance’ (i.e., without movement), subject to minimality effects.

1.1. INTRODUCTION The phenomenon of Quirky Case in Icelandic has received a great deal of attention since Zaenen, Maling and Thráinsson’s (1985) detailed study of it. Regardless of whether Case is assigned (GB theory) or checked (Minimalist Program), Quirky Case sits square on conceptions of Case Theory in the Principles-and-Parameters framework. A quick perusal of the literature reveals various stages in the investigation of Quirky Case. There has first been the question of how Quirky Case (usually, dative) is assigned, and why Quirky Case-marked elements surface in subject position despite their having no nominative Case features usually associated with ‘subjecthood.’1 Next, there has been the fact that Quirky elements, which behave like subjects, do not induce agreement on the verb, an otherwise prototypical feature of subjects. More recently, the fact that elements bearing nominative Case behave like objects2 when a Quirky element is present, and yet trigger agreement on the verb, has been discussed. Finally, the fact that this agreement is partial and optional, and even barred in some cases, has received some attention. I will not dwell upon why the phenomena associated with Quirky Case are important for the Principles-and-Parameters approach to language. Nor will I attempt to review the vast amount of analyses put forward to deal

22

Aspects of the Syntax of Agreement

with the facts just mentioned. Rather, I will focus on one particular quirk: the agreement pattern, especially that found with raising predicates (seemconstructions) in Icelandic. This chapter is organized as follows: section 1.2 provides the necessary background for the analysis to come; section 1.3 develops a new view on how agreement functions in Icelandic; section 1.4 applies the proposal to raising predicates; section 1.5 extends it to other languages, and tries to uncover the parameter underlying the availability of Quirky Case, and in so doing gain some insight into the nature of inherent Case. Section 1.6 is a summary.

1.2. SOME QUIRKY FACTS This section provides some necessary background on Icelandic Quirky subjects. I give a simple overview of the constructions in question, highlighting the properties of Quirky elements that make them “subjects.” Quirky subjects behave like ordinary nominative subjects (and not like ‘topicalized’ objects) with respect to numerous syntactic phenomena, including such familiar subjecthood tests as subject-oriented reflexivization, subject–verb inversion, ECM, raising, subject control, and conjunction reduction (see Zaenen, Maling, and Thráinsson 1985, henceforth ZMT, for fuller discussion).3 Compare the behavior of ‘structural’ and ‘Quirky’ subjects in (1) and (2), respectively. (Examples are from Sigurðsson 1992: 5f.) (1)

a.

Húni sá myndina sínai. Reflexivization she.nom saw picture self’s4 ‘She saw her own picture.’ b. Hefur hún séð myndina? “Inversion” has she.nom seen picture ‘Has she seen the picture?’ c. Ég tel [hana hafa séð myndina]. ECM I believe her.acc have seen picture ‘I believe her to have seen the picture.’ d. Húni virðist [ ti hafa séð myndina]. Raising she.nom seems have seen picture ‘She seems to have seen the picture.’ e. Hún vonast til [að PRO sjá myndina]. Control she hopes for to PRO.nom see picture5 ‘She hopes to see the picture.’ f. Hún horfði og (hún) sá myndina. Conj. reduc. she.nom looked and (she.nom) saw picture ‘She looked and saw the picture.’

Quirky Agreement (2)

23

a.

Hennii leiðist bókin síni. her.dat bores book self’s ‘She finds her own book boring.’ b. Hefur henni leiðst bókin? has her.dat bored book ‘Has she found the book boring?’ c. Ég tel [henni hafa leiðst bókin]. I believe her.dat have bored book ‘I believe she found the book boring.’ d. Hennii virðist [ ti hafa leiðst bókin]. her.dat seems have bored book ‘She seems to have found the book boring.’ e. Hún vonast til [að PRO leiðast ekki bókin]. she hopes for to PRO.dat bore not book ‘She hopes not to find the book boring.’ f. Hún var syfjuð og (henni) leiddist bókin. she was sleepy and (her.dat) bore the book ‘She was sleepy and found the book boring.’

With this much background, we can turn to subject–verb-agreement patterns in ‘Quirky’ constructions. Agreement in Icelandic finite clauses involving nominative subjects proceeds as in many other languages: agreement obtains between the verb and the subject, usually in the highest spec (ignoring topicalization, post-copular subjects, etc.). (3)

Við kusum stelpuna. we.nom.1pl elected.1pl girl.acc ‘We elected the girl.’

(4)

Þær virðast tr hafa verið tb kosnar t. they.nom.3pl seem.3pl have been elected.nom.3pl ‘They seem to have been elected.’ Sigurðsson (1992: 2, (1), (4))

However, once we turn to Quirky subjects, facts get murky. If there is only one argument—the Quirky Case-marked subject—agreement does not obtain between the verb and the subject. The verb takes what has come to be called “default agreement” (a form homophonous with the third person singular neuter). (5)

Stelpunum var hjálpað. the girls.dat.pl.fem was.3sg helped.neuter.sg ‘The girls were helped.’

24

Aspects of the Syntax of Agreement (6)

Þeim virðist tr hafa verið tb hjálpað t. them.dat.3pl seems.3sg have been helped.neuter.sg ‘They seem to have been helped.’ Sigurðsson (1992: 3,(5b),(8))

If there is another argument—an “object”—agreement depends on the Case of the latter; if the Case of that argument is dative, genitive, or accusative, no agreement obtains ((7)–(9)).6 If the Case is nominative, the verb agrees with the nominative NP, not with the Quirky subject (10). (7)

Mig iðrar þess. me.acc repents.default this.gen ‘I repent this.’

Yip et al. (1987: 230, (9e))

(8)

Henni var skilað peningunum. her.dat was.default returned money.dat ‘She was given the money back.’

(9)

Mig vantar peninga. me.acc lacks.default money.acc ‘They seem to have been helped.’ ZMT (1985: 454, (29a); 459, (42b))

(10) Henni voru gefnar bækurnar. her.dat were.3pl given books.nom.pl ‘She was given the books.’ (Sigurðsson (1992: 5, (5f)) However, this is not the whole story. As the careful studies of Sigurðsson (1996) and Taraldsen (1995) have revealed, agreement with nominative objects is only partial. If the object is first or second person, only number, not person, agreement obtains (11); if it is third person, ‘full’ (i.e. person and number) agreement obtains (12).7 (11) ?*Henni leiddust við. her.dat.3sg bored.3pl we.nom.1pl ‘She was bored with us.’

Taraldsen (1995: 309, (7))

(12) Henni leiddust/?*leiddist þeir. her.dat.3sg bored.3pl/3sg they.nom.3pl ‘She was bored with them.’ Taraldsen (1995: 307, (1)) Those facts are puzzling, and have given rise to an explosion of possible analyses, which I won’t attempt to review here. Suffice it to say that two main approaches can be envisaged: One is to rely on Agreement projections (for example, on split AGRPs like AGRnumber and AGRperson) (see

Quirky Agreement

25

Taraldsen 1995 for an approach along these lines), and the other is to adopt Chomsky’s (1995) Attract-F hypothesis and stipulate that only certain features can raise to check agreement without triggering pied-piping.

1.3. THE LIMITS OF AGREEMENT IN ICELANDIC In this chapter, I will focus on agreement patterns with raising predicates like seem. I believe that these constructions can give us a particularly useful and revealing angle to tackle the problem of Icelandic finite agreement, for they involve a Quirky Case-marked experiencer (the analog of the to-phrase in it/John seems to me . . .)8 and an embedded clause containing a subject, and possibly an object. The embedded subject can bear either structural or Quirky Case. Thus, such constructions offer a full range of possibilities, and can display matrix versus embedded agreement asymmetries, depending on whether the embedded subject (features) raise(s), and on which Case it (they) ‘check(s).’ Put differently, raising constructions add the further complexity of having two Quirky arguments, which I will try to show is just the level of complexity needed to understand what determines agreement in simple root contexts.

1.3.1. Basic Facts Consider the agreement patterns in the following examples, exhibiting the full range of argument structures. r
One Quirky element, agreement on matrix verb with embedded nominative element: (13) Mér ?virðist/virðast hestarnir hafa verið gefnir me.dat seem.3sg/pl. horses.nom.pl have been given konunginum. king.dat ‘It seems to me that horses have been given to the king.’ (14) Honum eru taldir hafa verið gefnir him.dat are.pl thought have been given.nom.masc.pl peningarnir. money.nom.masc.pl ‘The money is thought to have been given to him.’ (15) Jóni ?*virðist/virðast vera *talið/taldir John.dat seem.sg/pl be believed.neuter.sg/nom.masc.pl líka hestarnir. like horses.nom.masc.pl ‘John seems to be believed to like horses.’

26

Aspects of the Syntax of Agreement (16) Mér *virðist/virðast þeir vera skemmtilegir. me.dat seem.3sg/3pl they.nom be interesting ‘It seems to me that they are interesting.’ Schütze (1997: sec.4.1, passim) r
Absence of agreement on the matrix verb with embedded nominative element, presence of an embedded Quirky subject: (17) Mér fannst/*fundust henni leiðast þeir. me.dat seemed.3sg/3pl her.dat bore they.nom ‘I thought she was bored with them.’ (18) Mér hefur/*hafa alltaf virst honum líka bækur. me.dat has/have often seemed him.dat like books.nom.pl ‘It has often seemed to me that he likes books.’ (19) Mér virðist/?*virðast Jóni líka hestarnir. me.dat seem.sg/pl John.dat like horses.nom.pl ‘It seems to me that John like horses.’ (20) Mér virðist/*?virðast stráknum líka þessir bílar. me.dat seem.3sg/3pl boy.dat like these cars.nom.pl ‘It seems to me that the boy likes these cars.’ (21) Mér hefur/*hafa alltaf virst honum hafa verið seldar/*selt me.dat has/have often seemed him.dat have been sold.pl/sg þessar bækur á alltof hár verði. these books.nom.pl at far.too high price ‘It has often seemed to me that he has been sold these books at far too high a price.’ Schütze (1997: sec.4.1, passim) r
Default agreement: (22) Mér þótti María vera gáfuð. me.dat seem.3sg Mary.nom be gifted.nom.fem.sg ‘I thought Mary was smart.’ Schütze (1997: sec.4.1, assim) r
Failure of all combinations: (23) Henni *leiddumst/?*leiddust/?*leiddist við. her.dat bored.1pl/3pl/default we.nom ‘She was bored with us.’ Sigurðsson (1996: 28, (56))

Quirky Agreement

27

At first sight, these patterns seem to resist any unified account, even if one restricts attention to a small class of predicates. Sigurðsson (1996: 29) has noted that agreement with raising predicates (his ‘Dat and Nom infinitive’ verbs, or ‘D/NcIs’) is “one notch weaker” than in the monoclausal Dat-nom constructions, where (at least) number agreement is the preferred strategy (‘default’ agreement is marginal, and only found with a few verbs, see Sigurðsson 1996: 16ff.). With raising predicates, default agreement seems to be the preferred option. Schütze (1997) has noted another generalization. There seems to be a general ‘adjacency’ requirement governing agreement in raising contexts: a verb can only agree with a nominative ‘object’ if the latter immediately follows the matrix verb. As Schütze remarks (p. 109), this seems to be a minimality effect: the closest lower DP is the only one that can be attracted into the matrix clause; if that closer element is Quirky Case-marked, the matrix verb takes default agreement. In other words, the relevant K-features of the embedded nominative object cannot be attracted across an intervening Quirky argument (which can only be the one contained in the embedded clause, the matrix experiencer (the “to-phrase”) always raises, and so never blocks feature-movement,9 see later discussion10). Though fairly straightforward, this minimality effect turns out to be difficult to implement within the minimalist program.11 It is indeed assumed (Chomsky 1995) that potential attractees compete not because of their categorial features, but because of some subcategorial feature they share (this is the basic insight of Relativized Minimality rephrased in Attract-F terms). As such, we would not expect a Quirky DP to intervene in the examples under discussion since there is good reason to believe that its K-features are inert for agreement. Given that, it seems that the “minimality effect” will have to be recaptured in somewhat different terms. At the moment, I am aware of no such analysis in the literature, which means that we will have to take a fresh start.12

1.3.2. The Proposal I will first lay out some basic properties of raising constructions in Icelandic.13 As in English, the experiencer is optional in Icelandic raising predicates. (24) Þeim hafði virst Ólafur vera gáfaður. them.dat has seemed Olaf.nom be intelligent ‘They regard Olaf as intelligent.’ (25) Ólafur hafði virst vera gáfaður. Olaf.nom has seemed be intelligent ‘Olaf seemed to be intelligent.’ (modeled on Sigurðsson (1996: 29, (57)))

28

Aspects of the Syntax of Agreement

However, unlike in English, Icelandic experiencers block raising. (26) *Ólafur hafði virst þeim vera gáfaður. Olaf.nom has seemed them.dat be intelligent ‘They regard Olaf as intelligent.’ This blocking effect is not peculiar to raising predicates: Quirky elements always block raising of nominative ‘objects’ (or ‘sobjects’ to use Sigurðsson’s 1996 term) to the highest (‘subject’) position, irrespective of the agreement pattern.14 (27) Strákarnir hafa leiðst henni. boys.nom has bored.3pl her.dat ‘She found the boys boring.’ (28) *Strákarnir hafa henni leiðst. (modeled on Taraldsen (1995: 307)) Note further that even if the embedded subject is Quirky, the experiencer blocks movement. (29) Mér virðist Jóni líka hestarnir. me.dat seems.default John.dat like horses.nom ‘It seems to me that John likes horses.’ (30) *Jóni virðist mér líka hestarnir. In short, raising of a full category is allowed only when the experiencer is absent.

1.3.3. Insular Scandinavian Is Not a Secluded Island The state of affairs reported in 1.3.2 is by no means restricted to Icelandic. In an interesting study of experiencers in Spanish, Torrego (1996, 1998) shows that many Romance languages (in fact, all except French and Italian) prohibit raising if there is an overt experiencer.15 (Illustration from Spanish.) (31) Este taxistai parece [ti estar cansado]. this taxi-driver seems be tired ‘This taxi-driver seems to be tired.’ (32) *Este taxistai nos parece [ti estar cansado]. this taxi-driver to.us seems be tired ‘This taxi-driver seems to us to be tired.’ Torrego (1996: 102, (2))

Quirky Agreement

29

This parallels the Icelandic facts noted earlier (26). It thus appears that English, French, and Italian are the exceptions rather than the rule.16 (33) This driver seems to me to be tired. (34) Ce conducteur me semble être fatigué. this driver to.me seems be tired ‘This driver seems to me to be tired.’ (35) Questo autista mi sembra essere stanco. this driver to.me seems be tired ‘This driver seems to me to be tired.’ The Romance data highlight the fact that Icelandic may not be the marked/‘Quirky’ case, but rather the specific instantiation of universal processes. Further facts from Romance reinforce this position. Schütze, for instance, notes (1997: 109 fn.17) that the status of (36) becomes “mysterious” once the auxiliary is dropped. Neither the agreeing nor the nonagreeing form seems good (37). (36) Mér hefur/??hafa alltaf virst honum hafa verið me.dat has/have often seemed him.dat have been seldar/*selt þessar bækur á alltof hár verði. sold.3pl/3sg these books.nom.pl at far.too high a.price ‘It has often seemed to me that he was sold these books at far too high a price.’ (37) ???Mér virðast/virðist honum hafa verið seldar/selt þessar me.dat seemedsg/pl him.dat have been sold.3pl/3sg these bækur á alltof hár verði. books.nom.pl at far.too high a.price ‘It has often seemed to me that he was sold these books at far too high a price.’ This parallels the French data given in (38)–(39): (38) Marie a semblé à Jean être fatiguée. Marie has seemed to Jean be tired ‘Marie seemed to Jean to be tired.’ (39) ??Marie semble à Jean être fatiguée. Marie seems to Jean be tired ‘Marie seems to Jean to be tired.’

30

Aspects of the Syntax of Agreement

For as yet unclear reasons, the French sentence is odd when the auxiliary is dropped (it is good if the experiencer is cliticized: Marie lui semble être fatiguée ‘Marie to-him seems be tired,’ which is impossible in Icelandic). I will postpone an explanation for the blocking effect of experiencers until section 1.5, for I believe that a proper understanding of this phenomenon requires some non-trivial departure from previous assumptions. Indeed, even if these facts suggest that Icelandic may not be that ‘quirky’ after all (at least, not so ‘isolated’ as is often claimed), we still have to explain why agreement, when it obtains, is only partial. Besides, we have to address the more general question of why Quirky subjects, which behave like true subjects in many respects (cf. section 1.2), do not trigger agreement. The puzzle may be derived from George/Kornfilt’s (1981) theory (or, rather, Chomsky’s 2000 interpretation of it) that agreement and Case are but one phenomenon (one ‘feature’; see also Raposo and Uriagereka 1996, and Schütze 1997), and that agreement on T is only possible with nominative.17 But this leaves us with the question of why agreement is not ‘total’ when it obtains with some nominative elements (‘sobjects’). I will show that cross-linguistic considerations again reveal that this is by no means a language-specific ‘inconsistency,’ but rather may be part of a universal constraint.

1.3.4. Explaining the Agreement Constraints: Agreement Does Obtain with Quirky Subjects 1.3.4.1. The Person-Case Constraint The universal constraint I have in mind is the one discussed in Bonet (1994) under the name “Person-Case Constraint” (PCC). The latter basically states that human language disallows the presence of first or second person agreement with a direct object when there is also dative agreement. (40) The Person-Case Constraint (PCC)If Dative (agreement) n Accusative (agreement) = 3rd Bonet (1994: 36) Here I illustrate the PCC with French data, but Bonet shows that it is valid in many, typologically unrelated languages, which leads her to claim that the constraint is universal. (41) *Jean me lui a recommandé. Jean me him has recommended ‘Jean recommended him to me.’ (compare: Jean le lui a recommandé. Jean it him has recommended ‘Jean recommended it to him.’)

Quirky Agreement

31

Bonet was not the first one to observe this constraint (see, e.g., Kayne 1975 and Perlmutter 1971), but the constraint is still poorly understood. Here, I will follow Bonet in giving it a morphophonological treatment, assuming a Distributed Morphology framework (Halle and Marantz 1993), according to which (some/most) morphological operations take place after syntax, in Morphology, part of the PF component.18 This will allow the constraint to be “inoperative” in (narrow) syntax; its effects will be manifest only in overt morphology. Put differently, nothing syntactically bars a configuration flouting the PCC (i.e., a configuration in which both the dative and the accusative elements fully agree with the verb); if syntax yields such an output, Morphology will use repair strategies (on which see Bonet 1994) that will mask the PCC-violation, preventing the derivation from crashing at PF. For our present purposes, it may mean that some agreement process takes place in syntax but fails to map onto Morphology, and so doesn’t surface at PF because it would flout the PCC. I will argue that this is indeed the case in Icelandic (and elsewhere, see section 1.5); that is, I will put forward the view that agreement between the Quirky element and the verb actually obtains (an instance of AGRschecking)19—thus making Quirky subjects “full-fledged subjects”—but that this agreement cannot be morphologically realized for independent Economy reasons. By contrast, “AGRo-checking” can surface on the verb iff a nominative element is involved, because there is an intricate link between agreement on T and nominative (cf. George-Kornfilt/Chomsky’s view). The morphological shape of this agreement relation is constrained by the PCC; let us now see how and why the latter holds in Icelandic. 1.3.4.2. Partial Agreement We have already noted that agreement with nominative objects is only partial; the verb takes third person agreement, only number agreement varies (being available in some, well-defined cases). This is similar to the PCCresult, which states that if agreement obtains with a dative element (in the case of Icelandic, a Quirky element),20 accusative agreement (here referred to as AGRo) must be third person. True, there is no ‘surface’ manifestation of dative agreement on the verb, but I would like to claim that there is on the dative element itself. Delsing (1996) has expressed the view that Case has lost most if not all of its function in modern Icelandic, and can now be seen as a ‘decorative’ device. This corroborates Sigurðsson’s (1996: 28) intuition that nominative is becoming Quirky like all other Cases in Icelandic. The decoration Delsing mentions might be reinterpreted as ‘sufficient’ for agreement purposes. That is, Economy considerations might dictate the non-overt manifestation of agreement on the verb, because the information is already visible on the noun (it seems a general fact about human languages that the checked feature is realized on only one member of the checking relation), a sort of

32

Aspects of the Syntax of Agreement

“Generalized Doubly Filled Comp Filter.”21 More subtly, dative agreement is also manifested in the obligatory third person morphology on the verb. Our claim is then that agreement takes place with Quirky subjects, but the result is almost ‘null’ for the verb’s morphological makeup; it only forces third person agreement, leaving morphological room for object number agreement with objects (if any). If no syntactic clash (Case-mismatch) arises, number agreement obtains with the nominative object (George/Kornfilt theory). 1.3.4.3. Beyond the PCC: Point of View Bonet develops the PCC in Optimality terms, and goes on to show what its effects are, how they can be circumvented, etc., but does not actually try to explain why there should be such a thing like the PCC in UG. Here, I can only be speculative. A large survey on dative Case assignment/checking (such as van Belle & van Langendock 1996) reveals that the interpretive ‘role’ associated with the dative (understood here as Quirky, cf. fn. 17) is typical of [+human/+animate], which is best characterized in terms of [+person], rather than gender or number. [+Person] is intrinsic to humans, which might lead to a ‘correspondence rule’ of the type: dative › person. As Sigurðsson (1996) and many others have claimed, [+person] = 1 or 2 person, third person is the absence of the [+person] feature; so dative does not block the latter, it only blocks 1st or 2nd person, i.e., [+person] features, which might be said to encode Point-of-View ([+/– speaker], [+/– addressee]). Point-of-view has recently received some attention in the generative literature, and is now assumed to be represented structurally via some projection (or several such projections, see Rizzi 1997), be it FP (Uriagereka 1995a, 1995b), PolP (Culicover 1992), or 4P (Laka 1990), dominating IP. Uriagereka (1995a, 1995b), especially, has tied clitic distribution, Point-ofView, and IP. Further, Chomsky (2000) argues that T’s EPP-feature should be kept distinct from its Case/K-feature, and, in the case of what he calls defective T (where the EPP is checked by an element distinct from the one that checks Case/Agreement), the EPP might be viewed as an impoverished [Person]feature (which, he argues, expletives can check.). I suggest we relate this EPP/[Person]-feature to the dative/Quirky feature, both of which encode Point-of-View. This will favor raising of the Dative/Quirky element over the nominative ‘(s)object’ for obvious feature-matching reasons. 1.3.4.4. Explaining Icelandic Agreement Given the PCC, we can now explain why person agreement is unavailable for nominative objects. The formulation is misleading; it should be something like: AGRs features are checked by Quirky subjects (whenever Quirky Case is involved), AGRo features by objects. Some universal morphological process says: Whenever AGR features are checked by nominative, they can

Quirky Agreement

33

be phonologically realized (George/Kornfilt’s theory). This coupled with the PCC says that AGRo features checked by a nominative element can surface but, now that some dative element is involved, person feature realization is impossible (restricted to third person). Only number features can surface, which is what we get in Icelandic. It is interesting to note that no matter which agreement pattern is chosen, sentences involving a Quirky 3rd person element and a 1st or 2nd person nominative ‘object’ are hopelessly bad. (42) Henni *leidumst/?*leiddust/?*leiddist við. her.dat bored.1pl/3pl/default us.nom ‘She was bored with us.’ Sigurðsson (1996: 28) (43) *Henni voruð sýndir/sýndar þið. her.dat were shown.masc/fem you.nom ‘You were shown to her.’ (44) *Henni vorum sýndir/sýndar við. her.dat were shown.masc/fem us.nom ‘We were shown to her.’

Sigurðsson (1996: 32)

I take this as a confirmation of the view just expressed: such sentences yield a “Point-of-View” clash. The 1st and 2nd person features (borne by a nominative element) express point-of-view, which is also associated with dative (‘Quirky’) elements; if a dative nominal prevents raising of a 1st or 2nd person nominative element to the projection encoding Point-of-View, the sentence crashes. In those cases, the language will resort to another configuration allowing the 1st or 2nd person bearing element to express point-of-view and agreement syncretically (45)–(46) (compare (43)–(44)). (45) Þið voruð sýndir/sýndar henni. you.nom were shown.masc/fem her.dat ‘You were shown to her.’ (46) Við vorum sýndir/sýndar henni.

Sigurðsson (1996: 32)

1.3.4.5. A Binding Extension Arguably, the analysis allows us to explain why there is no nominative reflexive in Icelandic (Lyle 1997: 322).22 (47) *Hennii finnst REFLi veik. her finds refl sick ‘Shei considers herselfi sick.’

34

Aspects of the Syntax of Agreement (48) Icelandic third person reflexives nom – acc sig dat sér gen sín

True, this might have to do with independent properties of psych-predicates and the binding configurations they allow,23 but, I would like to claim, it is also the result of the PCC. Reflexives of the sig-type have no inherent number features; we saw that nominative checking is linked with overt realization of K-features (inasmuch as they are allowed, given the PCC). The nominative reflexive has to check its features (cf. Lebeaux 1983 and much subsequent work for the claim that reflexives raise at LF). Since it only has features that are blocked by the PCC (person features), the Morphological Component will not know how to realize number features (the number features are realized on the Quirky element, not on the verb, see earlier discussion), and the sentence will crash at PF. Given that, language resorts to repair strategies and uses anaphoric pronouns instead of anaphoric reflexives.24 (49) Henni finnst hún veik. her.dat finds she.nom sick ‘She finds herself sick.’ It thus seems that our theory on Quirky Case can account for why agreement is only partial with nominative ‘sobjects,’25 and for why agreement seems to be absent with otherwise ‘well-behaved’ subjects: Quirky elements. The basic idea is that syntactic agreement with Quirky subjects is morphologically manifested by the anti-person-agreement with nominative objects.

1.4. AGREEMENT IN RAISING PREDICATES Having grasped how agreement works in Icelandic, we can now reconsider the data in (13)–(23). Two generalizations have to be accounted for: (a) agreement in raising contexts is “one notch weaker” than in monoclausal environments (Sigurðsson 1996: 29), and (b) the minimality effect noted in Schütze (1997) that agreement between the matrix verb and the embedded nominative element is blocked if there is an embedded Quirky subject intervening. It will be remembered that the minimality effect could not be captured in an Attract-F framework given that the K-features of Quirky elements were considered inert for agreement, and hence should not prevent attraction. But now that we have seen that the K-features of Quirky subjects are active in the syntax, the minimality effect follows at once. The K-features

Quirky Agreement

35

that the attractor (the matrix verb) will encounter are those of the embedded Quirky subject. Since attraction is “myopic” (Chomsky 1995, Collins 1997), it cannot go beyond and see the features of the nominative element. Given that, agreement will surface on the matrix verb as default (third person singular). (50) Mér fannst/*fundust henni leiðast þeir. me.dat seemed.3sg/3pl her.dat be.bored they.nom.pl ‘It seemed to me that she was bored with them.’ (51) Mér hefur/*hafa alltaf virst honum líka bækur. me.dat has/have often seemed him.dat like books.nom.pl ‘It often seemed to me that he liked books.’ This maximally simple view also accounts for why agreement is “one notch weaker” in raising predicate contexts. In monoclausal environments, there is no intervening Quirky element, and attraction can proceed unhindered. (52) Henni leiddust strákarnir. her.dat bored.3pl boys.nom.pl ‘She was bored with the boys.’ Interestingly, agreement in the embedded clause of raising constructions takes place as in monoclausal contexts (as shown on the participle), there being no intervening Quirky element. (53) Mér virðist/*virðast Jóni vera taldir líka me.dat seems.3sg/3pl John.dat be believed.nom.masc.pl like hestarnir. horses.nom.masc.pl ‘I perceive John to be believed to like horses.’ 1.5. AGREEMENT CONSTRAINTS ELSEWHERE

1.5.1. PCC Effects in English: Partial Agreement in Raising Contexts26 Howard Lasnik (p.c.) has pointed out to me that the present analysis might account for the following English agreement facts, which, to my knowledge, have gone unnoticed in the literature (for further discussion, see Boeckx 1999a, 1999b). (54) John seems to Mary to be the best. (55) John and Bill/The men seem to Mary to be the best.27

36

Aspects of the Syntax of Agreement (56) There seems to Mary to be a man in the room. (57) *?There seem to Mary to be men in the room. (58) There seems to Mary to be men in the room.

Whatever the [person] content of the nominative element, matrix verb agreement obtains when the element raises overtly ((54)–(55)). However, (assuming a covert raising analysis of the “associate,” a fact to which I return later), agreement is blocked when the nominative element raises covertly (57) (I will argue that the third person singular form in (56) should be conceived of as default), which suggests that dative agreement takes place in English, activating the PCC.28 The data in (54)–(58) can be accounted for as follows. In (54)–(55), the nominative element raises overtly, it is the true (syntactic) subject of the sentence (it can, for instance, bind the experiencer: Johni seems to hisi boss to be lazy), and it thus checks the agreement (AGRs) features on T, allowing them to be realized morphologically (George/Kornfilt’s view). When the nominative element is the associate of an expletive, the PCC is operative (cf. (57)), but why should this be so? More precisely, given a checking theory where inflection is present from the start, why should there be an asymmetry between overt and covert operations? For the PCC to be operative, we have to say that dative agreement takes place, but that should block full agreement in all cases (the agreement features are already encoded on the verb). However, this is not quite correct. The PCC says that: (59) If dative agreement n accusative (‘AGRo’) = third person When the nominative element moves overtly, it is a true syntactic subject, and the PCC has nothing to say about nominative “subjects.” When the nominative element moves covertly, it is an associate, not a true subject; rather, the expletive is, as can be seen from various tests, such as tags, where only syntactic subjects are repeated.29 (60) There was a man in the room, wasn’t there/*a man/*he? (61) John was in the room, wasn’t he? As is well known, expletives have no (full) agreement features, so they do not play a role in agreement patterns (one might say they are neutral as to Point-of-View); the ‘associate’ status of the nominative element, however, does. Lasnik (1992, 1995b) has argued that the associate of the expletive checks a special Case (partitive Case, see Belletti 1988 for the origin of this

Quirky Agreement

37

idea; see also Rooryck 1997a for additional arguments in favor of Lasnik’s approach), in a manner similar to objects, that is, in a specific functional projection parallel to AGRo. Capitalizing on this, I propose that only in those cases, agreement between the matrix verb and the nominative element is an instance of AGRo, and can (actually, is) disrupted by dative agreement, in accordance with the PCC.30 We must now explain how dative agreement (agreement with the experiencer) takes place in English. Here I will build upon work by Kitahara (1997) (see also Ferguson 1994 for a similar line of reasoning regarding the invisibility of some PP nodes). Kitahara (1997: 63) notes the following. In (62), (62) Theyi seem to him [ ti to like John]. John must be interpreted as disjoint in reference from him. Chomsky (1995: 304) takes such disjoint reference effects as evidence that him c-commands into the embedded clause. Given that, Condition C of the Binding theory is violated if him takes John as its antecedent. But if that is so, him must also c-command they prior to raising, and should therefore block raising (the features of him are closer for attraction), yielding something like: (63) *Himi seems to ti [they to like John]. To prevent this, Kitahara suggests that him does not c-command into the embedded clause prior to raising. This is made possible via the following Larsonian shell (Kitahara 1997: 64). (64)

TP ru T vmax ru v Vmax ei PP Vb ru ru to him seem TP 6 . . . they . . .

Assuming the standard definition of c-command (Reinhart 1976) (roughly, A c-commands B if the first branching node that dominates A dominates B), the PP-node dominates him but not the embedded clause, hence him does not c-command into the latter; raising is thus made licit. As regards the disjoint reference effect, Kitahara assumes that Binding applies at LF31; by then, him must c-command John. To achieve this,

38

Aspects of the Syntax of Agreement

Kitahara suggests that him raises and adjoins to P to check its Case; it is only after this movement that him can c-command into the embedded clause, which forces the disjoint reference effect (to avoid a Condition C violation). As for raising, since him checks its features covertly, it does not bleed any overt operations. I agree in essence with Kitahara’s account, but will suggest a revision that will turn out to have wide-ranging empirical consequences.32 Fujita (1996: 157) proposes that prepositional objects check their Case in a SpecHead relation (AGRprep for him, or a specifier position adjoined to PP, as in Kitahara 1997, following Chomsky 1995), and that this might explain the well-known reanalysis facts of English. (65) Johni was talked about ti. (66) Whoi did you talk to ti. Fujita argues that in the case of reanalysis, P and V form a complex predicate and that AGRprep = AGRo; put differently, the prepositional object checks its Case with the [V-P] head, not with P alone. This is what I would like to suggest for ‘seem-to’ constructions: to incorporates into (‘reanalyzes with’) V (seem), equating AGRprep and AGRo; this mechanism does not affect Kitahara’s account: reanalysis takes place at LF, the c-command relations he discusses are maintained; but the revision allows us to claim that seem checks the Case of the experiencer, in other words, agrees with a ‘dative’ element, which activates the PCC, as desired. Not only does this simple mechanism account for intricate data from English, it also allows us to explain the blocking effect of experiencers for overt raising in languages like Icelandic or Spanish (cf. (26), (32)) if we assume that reanalysis takes place in the overt component in those languages, which makes the experiencer closer for attraction, barring overt raising of the embedded nominative. In English, reanalysis takes place covertly, so raising is allowed.33 One question that arises given this line of reasoning is why there is no blocking effect with clitics in French (cf. the discussion that follows example (39)). Torrego (1996, 1998) has shown that those languages where overt raising is impossible when an experiencer is present obligatorily double the clitic with the equivalent of a to-phrase, Italian and French being the exceptions to the rule. A plausible explanation is that clitics are base-generated as heads on, say, T, and therefore do not block overt raising from below (see Boeckx 1998a for extensive discussion).

1.5.2. The Nature of Quirky Case The overt/covert reanalysis asymmetry (which falls within the range of generally assumed parameters) might potentially account for why certain

Quirky Agreement

39

languages have Quirky experiencers, and others don’t. The question is very broad, and cannot be fully explored here, but I will suggest a possible line of reasoning, focusing on German and Icelandic, which differ minimally on the surface, though only one language (Icelandic) has Quirky Case, as originally observed by Zaenen, Maling, and Thráinsson (1985). (67) Ihm wurde geholfen. him.dat was.3sg helped ‘He was helped.’ (68) *Ihm hofft, PRO geholfen zu werden. him.dat hopes PRO.dat helped to become ‘He hopes to be helped.’ (69) *Er kam und (ihm) wurde geholfen. he came and (him.dat) became helped ‘He came and (he) was helped.’ These are but a few subjecthood tests (control, conjunction reduction) that German dative elements fail to pass (compare Icelandic ‘Quirky’ subjects in section 2). Given that, ZMT put forward the now widely accepted view that fronted dative elements in German are topicalized constituents, not subjects. What I would like to suggest is that the Icelandic/German contrast results from an overt/covert asymmetry. The reanalysis process (which might underlie every instance of dative agreement) takes place covertly in German, but overtly in Icelandic. One piece of evidence in favor of this hypothesis comes from raising predicates: German experiencers do not block raising, which suggests that reanalysis does not take place prior to Spell-Out in the language. (70) Peteri scheint mir, ti zufrieden zu sein. Peter.nom seems me.dat happy to be ‘Peter seems to me to be happy.’ Sentences like (71) are well-formed in German: (71) Mir scheint Peteri, ti zufrieden zu sein. me.dat seems Peter.nom happy to be ‘Peter seems to me to be happy.’ But this is no evidence for ‘experiencer-blocking.’ The Verb-second nature of German obscures cases of Topicalization; but once we are aware of that fact (it is difficult to see how German would allow both (70) and (71) otherwise, except if one allows for optional reanalysis, which should be discarded on minimalist grounds), (71) becomes identical to the Italian facts mentioned in fn. 31 (repeated here under (72)).

40

Aspects of the Syntax of Agreement (72) A Piero Gianni non sembra fare il suo devere. to Piero Gianni not seems do the his duty ‘Gianni doesn’t seem to Piero to do his duty.’

The covert reanalysis option in German immediately explains why dative elements fail all subjecthood tests like control, binding, etc. These phenomena are associated with full category movement, not with feature movement (cf. Lasnik 1995c,d), hence their incompatibility with covert processes.

1.5.3. The Nature of Inherent Case The present study has argued that agreement takes place with dative elements in the syntax, but that this process cannot be morphologically realized, roughly because the information is already contained on the nominal item. As a result, agreement surfaces as default (third person singular). This casts doubt on the well-established inherent/structural Case distinction. The latter was empirically motivated on agreement facts from passives in German like (73)–(74).34 (73) Die Kuchen wurden/*wurde gegessen. The cakes.nom.pl became.3pl/sg eaten ‘The cakes were eaten.’ (74) Den Kindern wurde/*wurden geholfen. The children.dat.pl became.3sg/pl helped ‘The children were helped.’ The PCC does not force us to view inherent Case as special, or ‘inert’ for agreement; rather, it leads us to argue in favor of viewing inherent Case as equally ‘active’ as structural Case in the syntax. In other words, our approach unifies structural and inherent Cases—a welcome result from a minimalist point of view (see Collins and Thráinsson 1996 and Lasnik 1995b for further arguments that the inherent/structural distinction is spurious and should be done away with). Obviously, the consequences of such a unification remain to be investigated, but it seems to me that it might overcome many problems which the inherent/structural distinction had encountered over the years.

1.6. CONCLUSION I started by investigating some of the properties of raising predicates in Icelandic, concentrating on agreement patterns between the matrix finite verb and the nominative element in the embedded clause.

Quirky Agreement

41

By resorting to the Person-Case constraint studied in Bonet (1994), according to which object agreement is limited to third person if dative agreement takes place, I have been able to explain why only partial (i.e., number) agreement takes place with nominative objects, given the fact that agreement with Quirky subjects take place. I was able to maintain (indeed, strengthen) the George and Kornfilt theory that agreement and Case are but one feature. At no point have I felt the need to resort to Agreement projections—a desirable result (cf. Chomsky 1995). The claim that agreement takes place with experiencer (dative/Quirky) elements has proven empirically correct, not only for Icelandic, but also for English, Spanish, and other languages. The present theory has been shown to have wide-ranging empirical and conceptual consequences for (among other things) the nature of the associate of expletives (which I have argued should be regarded as an object of some sort), for Relativized Minimality effects in an Attract-F framework, for the availability of ‘Quirky’ elements, and for the inherent/ structural Case-theoretic distinction. In view of its empirical coverage and conceptual simplicity, the path taken here seems very promising.

2

Intricacies of Agreement in Icelandic

This chapter revisits the agreement restrictions with nominative objects in Icelandic discussed in Chapter 1. I point out several problems for my original analysis and for studies that built on it, and propose a novel account, based in part on novel data. In particular, I reconsider the parallelism between the Icelandic facts and standard Person-Case-Constraint effects, and suggest the two should not be equated, despite initial similarity. I also reject the claim that agreement with the nominative objects in Icelandic is established long-distance, pointing out that my original argument for an Agree relation in this case was based on the false assumption that nominative case has a unique source in Icelandic.

2.1. BASIC FACTS ABOUT AGREEMENT AND NOMINATIVE OBJECTS IN ICELANDIC The syntax of agreement involving nominative objects in Icelandic has been the subject of intensive research within the minimalist program in recent years (see Sigurðsson 1992, 1996, 2000, 2002, 2003, 2004a, 2004b, Taraldsen 1995, Schütze 1997, Boeckx 1998a, Chapter 1, Chomsky 2000, Hiraiwa 2001, Holmberg and Hróarsdóttir 2001, Haeberli 2002, Anagnostopoulou 2003, Bejar and Rezac 2003, Ormazabal and Romero 2007, among others). It is the aim of this chapter to offer a theoretically satisfactory account of the intricacies of finite verb agreement (and absence thereof) in the context of nominative objects. Let us begin with a couple of generalizations about Icelandic syntax that will be central to the present investigation. First, Nominative objects are found only in the context of Quirky subject constructions. Contrast (1) and (2). (1)

Henni voru gefnar bækurnar. she.dat were.pl given.pl books.nom.pl ‘She was given the books.’

Intricacies of Agreement in Icelandic (2)

43

Við kusum *hún/hana. we.nom elected she.nom/her.acc ‘We elected her.’

Second, Quirky subjects, unlike nominative subjects, do not trigger morphological agreement on the finite verb. This is illustrated in (3). (3)

Stelpunum var hjálpað. the.girls.dat.pl.fem was.3sg helped.sg ‘The girls were helped.’

Third, agreement between the finite verb and the nominative object is limited to number agreement. In particular, person agreement is excluded. Compare (1) and (4). (I return below to why (4) remains unacceptable when other finite verb forms such as leiddust ‘bored.3.pl.’ or leiddist ‘bored.3.sg.’ are used.) (4)

*Henni leiddumst við. her.dat bored.1pl us.nom ‘They were bored with us.’

Fourth, finite verb agreement with the nominative object is excluded if a Quirky element is within the c-command domain of the verb at Spell-Out (‘surface structure’), as exemplified in (5a). (5)

a.

Mér fannst/*fundust [henni leiðast þeir]. me.dat seemed.3sg/3pl her.dat bore they.nom ‘I thought she was bored with them.’ b. Mér virðast þeir vera skemmtilegir. me.dat seem.3pl they.nom be interesting ‘It seems to me that they are interesting.’

The four generalizations just shown constitute the core of Sigurðsson’s (1996) study, which provided the impetus for all subsequent studies on the topic. As we will see later, they require empirical refinement, but they will nevertheless stand us in good stead throughout. In section 2.2, I briefly go over one of the first attempts to provide a principled account of the generalizations highlighted in this section (Chapter 1), and point out several problems for that approach and subsequent studies that built upon it. Section 2.3 provides an alternative analysis, based in part on novel data. Section 2.4 introduces some further necessary empirical refinements and show how they can be accommodated within the present analysis. Section 2.5 is a summary.

44

Aspects of the Syntax of Agreement

2.2. QUIRKY AGREEMENT In Chapter 1 I argued that there is more agreement in Icelandic than meets the eye. In particular, I claimed that Quirky elements are full-fledged subjects, agreeing with the finite verb in narrow syntax. However, that agreement does not surface morphologically due to a condition akin to the DoublyFilled Comp Filter, given in (6). (6)

A K/Case-checking relation between B and C can receive a morphological exponent on either B or C, but not on both.

The claim that Quirky subjects trigger syntactic agreement immediately accounts for facts like (7a) (brought to my attention by Norbert Hornstein, p.c.). (7)

a.

*Í dag hefur Jóni virst [t var hjálpað]. today has John.dat seemed was helped ‘Today, John seems like he was helped.’ b. *Today John seemed [t was helped].

(7a) is parallel to the English sentence in (7b), except that the English subject bears nominative and not Quirky Case. It has been argued, quite convincingly in my opinion, that (7b) is out because ‘agreement’ (in particular full agreement {person, number, gender} with a “Case-assigning head”) freezes the relevant NP (see Chomsky 2000, Raposo and Uriagereka 1996, Lasnik 1995b, etc.; see Boeckx 2003a for review and additional references). Given that the same freezing effect obtains in (7a), it is natural to say that Quirky subjects enter into an agreement relation with the finite verb. (Note, incidentally, that sentences like (7a) argue against the idea that Quirky subjects move for ‘pure’ EPP reasons. Pure EPP positions, such as intermediate landing sites in long-distance dependencies, have no such freezing effect.) The idea that Quirky subjects trigger syntactic agreement also provides an account of the otherwise puzzling fact that agreement between the finite verb and nominative objects is restricted to number (recall (4)). Chapter 1 noted that the constraint was very reminiscent of what Bonet (1994) called the Person-Case Constraint (PCC) (a similarity later explored by Anagnostopoulou 2003, Bejar and Rezac 2003, and Ormazabal and Romero 2007). The PCC is a generalization over cases like the French example in (8) (see already Perlmutter 1971, Hale 1973, and Kayne 1975), or the Basque examples in (9). (8)

Jean *me/le lui donna. Jean me/it.acc him.dat gave ‘Jean gave me/it to him.’

Intricacies of Agreement in Icelandic (9)

45

a.

Azpisapoek etsaiari misilak saldu d-Ø-izki-o-te. traitors.erg enemy.dat missiles.abs sell abs3-dat3-erg3 ‘The traitors sold the missiles to the enemy.’ b. *Azpisapoek etsaiari ni saldu na-i-o-te. traitors.erg enemy.dat me.abs sell abs1-dat3-erg3 ‘The traitors sold me to the enemy.’

Bonet’s formulation of the PCC is given in (10). (10) Person-Case Constraint In the presence of dative agreement on a verbal form/dative clitic, accusative agreement with that verb/accusative clitic is confined to 3rd person As Chapter 1 argues, a slight modification of (10) such as (11) accommodates the restriction imposed on nominative object agreement straightforwardly. (11) Person-Case Constraint (revised) In the presence of dative (/Quirky) agreement on a verb, the direct object can only agree with that verb in number, not person What is crucial at this point is that for (11) to apply to Icelandic, Quirky subjects must agree. Once Quirky subjects are said to agree, the blocking effect in (5) falls straightforwardly under Minimality: (12) T [Dat V Nom] z----=--------m The impossibility of combining both a nominative subject and a nominative object in one clause (recall (1)) follows under Chapter 1’s analysis if nominative Case is uniquely licensed by R (To), and Icelandic To, unlike, say, Japanese To, cannot assign nominative Case more than once. The Chapter 1 analysis captures the core facts discussed in (10), but admittedly raises many questions. In particular, the status of the DoublyFilled-Comp-Filter-like principle in (6) is unclear and, as first pointed out to me by Peter Svenonius (p.c.), appears to be violated by sentences like (13), where both members of the agreement relation morphologically encode the relation. (13) The girls have breakfast Second, the nature of the PCC was left vague in Chapter 1. I tentatively followed Bonet (1994) in taking it to be a morphological constraint, but arguments for a syntactic treatment are offered in Anagnostopoulou (2003), Bejar and Rezac (2003). Simplifying somewhat, both works argue that the

46

Aspects of the Syntax of Agreement

PCC results from the fact that the Quirky NP and the Nominative NP check their features against one and the same head (To). The result is a split checking procedure, whereby the Quirky NP checks the person feature of T, and the Nominative NP checks the number feature of T. Why the opposite checking split (Quirky-Number, Nominative-Person) does not obtain is unclear. This is particularly puzzling given the fact that Quirky elements do show evidence of morphological agreement with Number, but never with Person, as in the following cases involving a floated quantifier (14a) or a secondary predicate (14b). (See Sigurðsson 2004a and Boeckx and Hornstein 2003 for discussion.) (14) a.

Strákunum leiddist öllum/*allir the.boys.dat bored.3sg all.dat.pl.masc/nom.pl.masc í skóla. in school ‘The boys were all bored in school.’ b. Strákarnir sýndu kennaranum óvirðingu the.boys.nom showed teacher.dat.sg.masc disrespect drukknum. drunk.dat.sg.masc ‘The boys showed the teacher disrespect (when he was) drunk.’

As such, syntactic accounts of the PCC remain stipulative. In addition, neither work offers an account of why agreement with a Quirky element never surfaces morphologically. Apart from the issues around (6) and around the PCC, Chapter 1’s account does not address the question of how nominative Case is licensed in the absence of agreement in (5a), as Hiraiwa (2001) observes. In addition, it fails to explain why the Quirky element does not block agreement in simple contexts like (3). In other words, why doesn’t a trace of a Quirky element block agreement? (15) Quirky-NPi T0 [ti V Nominative-NP] z-------_m Agree: √ The problem is particularly acute if agreement is established cyclically, as structure is being built (Chomsky 2000, Boeckx 2003a, Chapter 4, 10). At the point when the attractor To is introduced into the derivation, the Quirky element qua NP, not yet qua trace/silent copy, should intervene. Adopting the gist of Chapter 1’s analysis, Chomsky (2000, 2001) addresses the intervention problem and the Case-in-the-absence-of-agreement problem. With respect to the latter, Chomsky argues that some instances of nominative Case in Icelandic are inherent (see also Sigurðsson 1996). However,

Intricacies of Agreement in Icelandic

47

treating nominative Case as (sometimes) inherent would (if no extra assumption is made) predict that both nominative subjects and objects can occur within one clause (by taking one instance of nominative to be inherent, the other being structural), contrary to fact (recall (2)). With respect to the intervention problem, Chomsky (2001) relaxes cyclicity somewhat by letting Closest (Attract) be computed phase-by-phase (specifically, locality with respect to elements of phase P is computed at the next higher phase P’). The details of a phase-based analysis will not be discussed, since I want to adopt the strictest version of cyclicity—the null hypothesis. But I want to note that Chomsky’s analysis of intervention in (5) has another feature that is undesirable theoretically, and should therefore be dispensed with if possible. According to Chomsky, the intervention effect we see in (5) is of a specific type. It is an instance of defective intervention. Intervention is defective if the intervener is a ‘passive’ blocker as opposed to an active element. When it comes to Icelandic, the passive character of the intervening Quirky element is due to the fact that Quirky elements do not trigger agreement: Quirky elements cannot agree, but they block agreement. This is unlike other intervention effects, where interveners in principle can agree, check a [wh-] feature, etc. Chapter 10 examines the structure of defective intervention in detail and concludes that the notion of defectiveness is too powerful to capture intervention effects adequately. I endorse this conclusion here, and set it as a requirement for the analysis to be developed in the next sections.

2.3. QUIRKY AGREEMENT REVISITED In this section I offer an alternative analysis of the four generalizations listed in section 2.1. I reject the claim that Quirky subjects enter into an Agree relation. This forces me to reconsider the parallelism I drew between the Icelandic facts and the PCC. I provide novel data showing that the two should not be equated, despite initial similarity. I also reject the claim that nominative objects agree with finite To. Instead I treat nominative objects as objects, not as subjects of sorts, that is, as elements licensed by vo. The number restriction on agreement with nominative objects is shown to follow from restrictions on VP-level agreement (i.e., object agreement) in general.

2.3.1. Nominative Objects The first issue to clarify is the nature of nominative Case licensing. Since there is every reason to associate nominative Case on nominative subjects and finite To in Icelandic, the null hypothesis is to claim that the source of nominative Case is always finite To. However, the Icelandic example in (5a) shows that we cannot associate nominative Case assignment on objects with finite To, since finite To is not accessible to the nominative NP in the presence

48

Aspects of the Syntax of Agreement

of an intervening (overt) Quirky NP. Claiming that nominative Case may come from non-finite To won’t help, as there is no R-feature checking taking place with non-finite T-heads. Taking non-finite To to be the source of nominative Case on objects would go against the idea that has emerged in recent years (see Chomsky 2000 and Boeckx 2003a for references) that Case is a reflex of agreement. The only remaining possibility is vo. That vo may enter into K-feature checking is now standardly assumed for object agreement languages (see Chapter 4 for a clear example). It is also reasonable to assume that vo is the locus of past participle agreement in Icelandic. Consider (16). (16) Við virðumst hafa verið kosnar. we seemed.1pl have been elected.nom.pl ‘We seemed to have been elected.’ Interestingly, agreement on the participle in Icelandic is limited to number, gender, and Case. Crucially, participles show no sign of person agreement. This is highly relevant for us, as person agreement is impossible with nominative objects. That would follow from the licensing role of vo for nominative Case on objects. It would have nothing to do with whether Quirky subjects agree or not. (This is not to say that it has nothing to do with the presence of Quirky subjects; see next subsection.) Adopting the idea that vo is the licenser of nominative objects (see Alexiadou 2003a for a related proposal) leaves us with two sources for nominative Case in Icelandic. This potentially introduces some unwanted optionality into the grammar, in addition to opening up the possibility for both To and vo to license nominative Case on distinct arguments, giving rise to (2). A way of constraining the choice of nominative Case licenser must be found, a task I turn to in the next subsection.

2.3.2. Nominative Objects Only in the Presence of Quirky Subjects As pointed out earlier, taking vo to license nominative objects has the effect of dissociating the agreement possibilities of nominative objects from the agreement possibilities of Quirky subjects. No appeal to the PCC is needed. This is welcome in light of the problems raised for a PCC account in section 2.2, but we now have to ensure that nominative objects are only licensed in the context of Quirky subjects (recall (1)). At a general level, we are here facing a situation in which the availability of a given structural Case (Nominative) is tied to the presence of thematic information (thematic/Quirky Case). The situation is strongly reminiscent of Burzio’s Generalization (Burzio 1986). The latter expresses the idea that (structural) accusative Case is available only in the presence of an element bearing the external theta-role assignment. Chomsky (1995) captures the correlation by the position that both external theta-role assignment and

Intricacies of Agreement in Icelandic

49

(structural) accusative Case are properties of one and the same head: vo. Recast in Chomsky’s terms, Burzio’s expresses the idea that theta-role assignment by vo determines the latter’s Case-licensing property. In a similar vein, I would like to propose that a verbal head vo is endowed with the option of nominative Case licensing only if it assigns a theta-role realized as Quirky Case to an NP in its specifier. Several studies have now appeared (Svenonius 2002, Eythórsson 2000, Jónsson 2001) that indicate that the thematic nature of elements bearing Quirky Case is not as random as one might have thought. Most Quirky NPs are experiencers, Goals, or beneficiaries. For the sake of concreteness, I will assume that Quirky-Case-marked elements are introduced as specifiers of vQP. Within the clausal hierarchy/ functional sequence, vQP sits between VP (which introduces the most internal argument) and vP (which introduces the external argument). (Defined this way, vQP is just another name for applicative phrases like Pylkkänen’s 2002 ApplHighP). The reason for choosing a dedicated projection for Quirky elements distinct from the v-head that introduces agents has to do with the fact, known since Maling, Yip, and Jackendoff (1987), that Quirky elements resist the [agent] theta-role. I take this to mean that Quirky elements are excluded from SpecvP. But since I want to establish a parallel between Quirky-Case/R-role and (structural) Nominative Case on the one hand, and External R-role and (structural) Accusative Case on the other, I want to use a v-like projection to align my proposal with Chomsky’s (1995) take on Burzio’s Generalization. The parallelism I am proposing is expressed structurally in (17). (17) a.

vP b. vQP 2 2 NP vb NP vQb 2 2 v VP vQ VP | 5 | 5 {ExtR} NP {QuirkyR} NP {AccCase} {NomCase}

The proposal just made is enough to exclude the possibility of vo assigning nominative Case to the object NP and To assigning nominative Case to the subject NP, since subject NPs are incompatible with vQP. If no experiencer theta-role is assigned (i.e., if there is no vQ projection), then (finite) To must have the relevant property to license nominative Case. Crucially, this requirement can be stated very deterministically (leaving no room for crashing derivations, in accordance with the model of Crash-proof syntax developed by Frampton and Gutmann 2000) in the following way: (finite) T cannot have the relevant property to license nominative Case when it merges with a v that has that capacity too. Conversely, (finite) T must have the relevant property if it merges with a v that did not assign

50

Aspects of the Syntax of Agreement

an experiencer theta-role. The issue, then, reduces to one of selection, as stated in (18).1, 2 (18) a. *TFinite+Nom k vQ b. *TFinite-Nom k v(agent) (18) captures the facts in (1) and (2). In other words, it establishes a relation between Quirky NP and nominative object NP. What it does not do is capture the PCC effects. The next subsection presents novel data showing that this is a virtue of the analysis.

2.3.3. PCC Effects I already noted earlier that PCC effects can be captured without appealing to any version of the PCC if we combine the result of section 2.3.2: that nominative Case on objects is licensed only if the licensing head assigns Quirky Case/R-role, and the (independently needed) assumption that agreement in the “v”-domain in Icelandic is limited to Number, to the exclusion of Person. Setting aside the few cases that exist like Basque where objects trigger Person agreement on the finite verb like subjects do, object agreement appears limited to Number cross-linguistically (see Alexiadou and Anagnostopoulou 2002). Furthermore, instances of agreement inside VP such as part participle agreement are again limited to Number (and Gender); crucially, they do not involve Person. It is not at all clear why that restriction holds, but we can adopt this restriction as an independent fact of language for present purposes, and apply it to Icelandic.3 Interestingly, there is empirical evidence for rejecting the PCC as the cause of agreement limitations in Icelandic. As I observed in section 2.1, a reflex of the PCC, or agreement limitations with nominative objects in Icelandic, is the fact that those pronominal forms that are specified for [person] (i.e., 1st and 2nd person pronouns) are excluded from the object position. Consider (4), repeated here under (19). (19) *Henni leiddumst við. her.dat bored.1pl us.nom ‘They were bored with us.’ It is not difficult to capture (19). All we have to say is that (nominative) pronouns must trigger agreement (see Chapter 1, and Den Dikken 2001). Nominative [+Person] pronouns will therefore be excluded from object positions since their K-feature specifications conflict with the prohibition on [person] agreement for nominative objects. Interestingly, as Sigurðsson (1996) originally observed, [+person] nominative pronouns are acceptable in those contexts where agreement is not obligatory. Consider (20) first.

Intricacies of Agreement in Icelandic

51

(20) Mér *virðist/virðast þeir vera skemmtilegir. me.dat seem.3sg/3pl they.nom be interesting ‘It seems to me that they are interesting.’ For reasons that I will not discuss until section 2.3.6, Icelandic does not force agreement with the nominative element if the latter is the subject of a non-finite clause. Agreement is only forced if the nominative element and the agreeing head are clause-mates, as (21) indicates. (21) Henni leiddust/??*leiddist strákarnir. her.dat bored.3pl/3.sg the.boys.nom ‘She found the boys boring.’ Now consider (22). (22) Þeim hefur/*höfum/*hafa alltaf fundist [við vinna vel]. them.dat have.3sg/1pl/3pl always found we.nom work well ‘They have always thought that we worked well.’ Here we see a [+person] nominative pronoun that is licensed, but only in the absence of agreement. As Norbert Hornstein observes (p.c.), based on (22), we expect [+person] nominative pronouns to be acceptable in a context like (5a) (repeated here as (23)), where the presence of a Quirky element inside the embedded clause blocks agreement between the matrix verb and the embedded nominative object. (23) Mér fannst/*fundust [henni leiðast þeir]. me.dat seemed.3sg/3pl her.dat bore they.nom ‘I thought she was bored with them’ However, Icelandic speakers reject [+person] nominative pronouns in this context. (Data from Kjartan Ottósson, p.c.). (24) *Jóni virtist [Bjarna hafa líkað ég/við/þið]. John.dat seemed Bjarni.dat have liked I.nom/we.nom/you.nom ‘It seems to John that Bjarni likes me/us/you.’ Because the analysis I am proposing does not make nominative Case dependent on To, it can capture (24) straightforwardly. Nominative Case in (24) is licensed by vQ (forcing [-person]) inside the embedded clause. A PCC-based account cannot predict (24), because agreement and nominative licensing is tied to the putative source of quirky agreement, finite T. Indeed, the data from Basque in (25) (from Ormazabal 2000: 246) contrast

52

Aspects of the Syntax of Agreement

minimally with the Icelandic facts in (24), and show that the PCC and the restrictions on nominative object agreement cannot be equated. (25) a.

*Zuk ni etsaiari saldu naiozu. you.erg me.abs enemy.dat sell 1abs.3dat.2erg ‘You sold me to the enemy.’ b. Gaizki iruditzen zait zuk ni etsaiari saltzea. wrong seem aux you-erg me.abs enemy.dat sell. nomin. ‘Your selling me to the enemy seems wrong to me.’

As Ormazabal observes, the Basque facts in (25) indicate that the PCC is really about agreement in finite contexts. In (25b), no PCC effect arises because both the dative and the absolutive direct object are in a non-finite context. As a result, the direct object need not be [3rd person]. In Icelandic, however, it is not enough for the nominative object and the dative to be in a non-finite context to render non-3rd person nominative pronouns acceptable. I take the contrast between (24) and (25) to indicate that divorcing the Icelandic facts under consideration from the PCC is on the right track.4

2.3.4. Raising Quirky Subjects Since there is no longer any need to appeal to the PCC, the main motivation for taking Quirky subjects to trigger agreement disappears. But if Quirky subjects don’t agree, how do they manage to function as full-fledged subjects? The question is particularly salient when the mechanism of chain formation put forth in Chomsky (2000, 2001, 2004) is taken into account. According to Chomsky (2000: 122), elements may relate (featurally) in the following three ways. (26) a. Features of B and C Match b. (Properties of) Features of B and/or C trigger Agree (Valuation) c. (Properties of) Features of B and/or C trigger Move Chomsky (2001: 5) characterizes Match as a relation holding of two items sharing a feature. Agree is an agreement/valuation relation taking place at a distance between two elements (which Chomsky 2000 calls Probe and Goal). Move is the result of the satisfaction of the EPP-property of a probe that demands a filled specifier position. Crucially, the EPP-requirement is more specific than “a probe P requires a SPEC.” As Chomsky (2004: 114) emphasizes, “it seems that raising of B from XP is always restricted to some category of constituents of XP, hence some feature F of B (or complex of features).” In other words, it is not the case that anything can satisfy a

Intricacies of Agreement in Icelandic

53

probe’s EPP property. Rather, at the very least, something featurally related to the head of which the EPP holds can satisfy it. Thus, insertion of an adverb in SpecIP won’t lift the EPP-requirement: *[IP Today [I’ is [VP John coming]]]). This summary of Chomsky’s assumptions concerning chain formation entails that an element may not raise solely to ‘check the EPP feature’ (see Boeckx 2003a for extensive discussion). According to Chomsky, prior to EPP satisfaction, Match and Agree must take place. But, if this is true, this would force upon us the idea that Quirky subjects agree, since they move. To reconcile Chomsky’s position on chain formation and the guidelines I am trying to adhere to here, I depart from Chomsky in the following way (see also Boeckx 2001, 2003a, 2003b, Chapter 6 and references therein). From Chomsky I adopt the idea that Agree cannot take place in the absence of Match. This seems uncontroversial: For two elements to agree, they must have some feature in common. They must match. But I depart from Chomsky in not taking Agree to be a prerequisite for Move. In particular, I propose that Move may take place solely under Match. Boeckx (2001, 2003a) shows that such a departure from Chomsky’s position provides a basis for understanding the distribution of inflected (agreeing) and non-inflected (non-agreeing) forms in various domains where displacement is arguably involved. The same is true, I contend, in Icelandic Quirky subject constructions: Quirky elements move under pure Match, hence the absence of agreement on the verb. It now remains to determine the identity of the matching feature providing the basis for Move. The simplest answer is Case. It is remarkable that most non-nominative subjects are datives, a Case value intimately linked to some notion of location (goal, path, etc.). Taking location and time to be two reflections of a single feature (featural non-distinctness), we may say that dative elements (and by extension, all other quirky elements, since they all bear a locative-like theta-role) match the T feature of INFL. So far we have accounted for three of the four issues we were interested in: why Quirky subjects do not trigger agreement (answer: Move under Pure Match), why nominative objects are confined to quirky subject constructions (answer: Nominative is licensed by vQ), and why nominative objects, unlike nominative subjects, are incapable of triggering person agreement on the verb (answer: v-agreement is restricted to Number). The last issue to tackle is the intervention effect in (5a).

2.3.5. Intervention Effects To capture the intervention effects found in Icelandic it would be enough to adopt Chomsky’s notion of Defective Intervention, according to which an element C (c-commanding H and c-commanded by B) blocks the establishment of an Agree relation between two other elements B and H even if C itself could not agree with B. Since under the present analysis Quirky elements

54

Aspects of the Syntax of Agreement

match vQ, they could block the relevant relation between vQ and the nominative object in a situation like (27). (27) . . . vQ[Nom] . . . [QuirkyNP . . . V . . .NomNP] z-----_--=-----------m However, Chapter 10 argues that Defective Intervention is inadequate in capturing intervention effects in natural language. Although it appears to capture the Icelandic facts, it is too powerful in other domains. If intervention takes place, it must be ‘direct’ (based feature identity, with all members of the relevant configuration being active). The Icelandic case at hand argues for Defective Intervention only if Intervention is defined at the level of actual valuation (in which case Quirky elements are inactive), but at the level of Match, the very first suboperation in the establishment of a Probe– Goal relation, Quirky elements are as active as any other. So all we need to enforce is that locality (Minimality) be computed right at the level of Match, the very first step in an Agree relation so as to avoid recourse to Defective Intervention. This makes sense if locality (Minimality) is part of the operation Agree. Since the latter applies as soon as possible, it is plausible to claim that the former also applies as soon as possible. In this case, it means that Minimality is computed at the very first level of Agree: Match. An independent argument in favor of this position comes from facts discussed by Heck and Müller (2000). The core cases appear in (28). (Similar facts have been reported for Lebanese Arabic in Aoun and Li 2003.) (28) a.

Wen hat Fritz [nachdem er was gemacht hat] getroffen. who has Fritz after he what made has met ‘Who did Fritz meet after he did what.’ b. *Wen hat Fritz [nachdem er was gemacht hat] gesagt [dass who has Fritz after he what made has said that Maria liebt]. Maria loves ‘Who did Fritz say, after he did what, that Maria loves.’ c. *Was hat Fritz [nachdem er gemacht hat] gesagt [dass Maria what has Fritz after he made has said that Maria wen liebt]. who loves ‘Which thing is such that Fritz said, after he did it, that Maria loves who.’

The facts above constitute an interesting case of apparent defective intervention. Consider (28b). Descriptively speaking, an embedded wh-phrase cannot move to matrix C in the presence of a matrix adjunct clause containing a wh-element. Note that the wh-element contained inside the adjunct cannot move to SpecCP (an adjunct condition effect), as shown in (28c).

Intricacies of Agreement in Icelandic

55

Intervention is therefore defective. (28a) is accounted for by assuming that the wh-element that reaches SpecCP has first scrambled past the adjunct clause. Crucial for our purposes is that (Boeckx 2001, 2003a) has provided extensive arguments that elements inside adjuncts (and other strong islands) are inaccessible to Agree (which derives the adjunct condition). Importantly, such elements are accessible to Match (as Boeckx documents extensively on the basis of resumption patterns). That means that the blocking element in (28b) could not block Agree. The only level at which it can be active is that of Match. If this is correct, the German data provide an argument in favor of keeping the Match and Valuation steps within Agree distinct. Let me now turn to a pressing issue in the realm of intervention. Recall that Chapter 1 (or any analysis that makes nominative Case on objects dependent on finite To) has no account of why the Quirky element does not block agreement between the verb and a clause-mate nominative object (15). The present analysis accounts for this fact. Since it is vQ and not T that licenses nominative Case, we do not expect any blocking effect by the Quirky subject in monoclausal contexts, as the latter is not within the c-command domain of vQ to start with, as shown in (29). (29) [vQP Quirky-NP vQ [VP V Nom-NP]] z------m A similar analysis covers cases like (30), where the nominative object is prevented from agreeing with the matrix predicate due to the presence of a Quirky element, but is nevertheless able to trigger (number) agreement on the participle in its own clause. (30) Mér hefur/*hafa alltaf virst honum hafa verið seldar/*selt me.dat has/have often seemed him.dat have been sold.pl/sg þessar bækur. these books.nom ‘It often seemed to me that he had been sold these books.’ The example in (30) is crucial, as it addresses one problem raised by Ken Hiraiwa for Chapter 1’s account: How can nominative Case assignment obtain if agreement (the other side of the same Case-AGR relation) is blocked by an intervening Quirky element? (30) shows that Case and agreement need not be dissociated, as both take place inside the embedded clause.

2.3.6. Clause-Mate Effects and Another Source for Nominative Case However, there are several differences between monoclausal and biclausal contexts in Icelandic when it comes to agreement with nominative objects, some of which have already been illustrated earlier. I analyze them now.

56

Aspects of the Syntax of Agreement

First, although agreement is virtually obligatory in monoclausal contexts (31), it appears to be optional in biclausal environments (32). (See Sigurðsson 1996 for extensive discussion.) (31) a.

Henni leiddust strákarnir. her.dat bored.3pl the.boys.nom ‘She found the boys boring.’ b. ??*Henni leiddist strákarnir. her.dat bored.3sg the.boys.nom ‘She found the boys boring.’

(32) Mér virðist/virðast þeir vera skemmtilegir. me.dat seem.3sg/3pl they.nom be interesting ‘It seems to me that they are interesting.’ Another difference is that non-3rd-person pronouns are excluded from monoclausal contexts (33), but are licit in biclausal contexts (34). This difference is related to the first difference, as non-3rd person pronouns are licit in biclausal contexts only if agreement does not take place. Since agreement is forced in monoclausal contexts, (33) is out. (33) *Henni leiddumst við. her.dat bored.1pl us.nom ‘They were bored with us.’ (34) a.

Þeim hefur/*?höfum alltaf fundist [við vinna vel]. them.dat have3sg/1pl always found we.nom work well ‘They have always thought that we worked well.’ b. Þeim hefur/*?höfum alltaf fundist [við them.dat have.3sg/1pl always found we.nom duglegar]. industrious ‘They have always found us industrious.’

There are two questions to address here: (i) Why is agreement obligatory in monoclausal contexts, and (ii) how can nominative Case be licensed in the absence of agreement in biclausal contexts? Several asymmetries between monoclausal and biclausal contexts shed light on these issues. First note that the complement of raising verbs is clausal (IP), but it is nominal (NP/DP) in the monoclausal environments we are interested in. Second, although the Quirky element in monoclausal contexts is obligatory, the experiencer in raising constructions is optional.5 I would like to tie such asymmetries to the differences noted above and formulate the following hypothesis:

Intricacies of Agreement in Icelandic

57

(35) Since Quirky R is optionally assigned, Nominative Case (and concomitant agreement) on vQ is also optional present. The hypothesis in (35) is in the spirit of the Burzio’s-Generalization-style account I offered for the availability of nominative Case in section 2.3.2. It relates thematic and Case properties, and establishes a parallelism between them. Note, though, that (35) does not say that if the experiencer (Quirky R) is present, nominative Case must be. The converse of (35) forces agreement in monoclausal contexts. Because agreement is forced, and it is limited to Number, pronouns are incapable of functioning as nominative complements. I would like to relate this fact to the fact that pronominal NP- complements in Icelandic obligatorily undergo Object Shift. By that I mean that they obligatorily satisfy some EPP-requirement. If the EPP-requirement is formulated in terms of Person-feature checking (see Boeckx 2003a for justification), and agreement is obligatory (i.e., Move cannot take place under Pure Match), the only 3rd-person pronouns will be able to function as nominative NP-complements. No such restriction applies in biclausal contexts, as agreement (and Object Shift) is not obligatory. The hypothesis formulated in (35) entails that some instances of nominative Case, in particular those that do not trigger agreement, are the result of a default Case assignment, which I assume takes place in the PF component.6 (For independent evidence in favor of default nominative Case assignment in non-finite contexts, see Boeckx and Hornstein 2003; see already Hornstein 1990. For general discussion, see Schütze 2001.) In this I follow Chomsky’s (2000: 123) suggestion that the Case requirement of the moving element need not be checked. Evidence for this claim comes from gapping, for which I will adopt Johnson’s (1994) analysis. Johnson suggests that a sentence like (36) is derived by Across-the-Board movement of the verb, with the second conjunct lacking a T-layer, as represented in (37). (36) John reads Plato, and Mary Aristotle. (37) [IP Johni Infl [XP [X’ readsj [VP ti [v’ tj Plato]], and [VP Mary [V’ tj Aristotle]]]]] As noted by Kayne (2000b: 165f.), in the absence of a T-layer in its domain, the subject in the second conjunct is expected to lack Case. The grammaticality of the sentence in (36) supports Chomsky’s (2000) conjecture that what is primary is not structural Case itself, but the K-features of the head (T/v) the argument interacts with. Put differently, failure to check Case on Mary does not lead to a crash (the features of Infl are checked against those of John). What matters is what Howard Lasnik dubbed the Inverse Case Filter: the requirement that the {(Case/)K-} features of the probe be checked. (For extensive discussion of the inverse Case Filter, see Bošković 1997c, Martin 1999, and Boeckx 2003a.)

58

Aspects of the Syntax of Agreement

2.3.7. Summary Summing up so far, we have been able to account for the four generalizations discussed in section 2.1 and avoid the pitfalls uncovered in section 2.2. My starting point was the rejection of the idea that nominative objects agree with finite T. Instead I have treated nominative objects as objects, not as subjects of sorts, that is, as elements licensed by vo. The number restriction on agreement with nominative objects follows from restrictions on VP-level agreement (i.e., object agreement) in general, with no need to appeal to the PCC. Indeed, I have provided new data showing that the agreement restrictions on Icelandic and the PCC should not be equated. With the role of the PCC gone, I have abandoned the idea that Quirky subject agree with T. A simple matching procedure is enough to capture their movement behavior, as well as the intervention effects they trigger. In the next section I examine in detail several situations of (non-)intervention for agreement with nominative objects in Icelandic discussed by Holmberg and Hróarsdóttir (2001).

2.4. A-BAR MOVED QUIRKY ELEMENTS AND AGREEMENT Until recently, the generalization regarding intervention effects like (5a) in Icelandic was that finite verb agreement with a nominative object is excluded if a Quirky element is within the c-command domain of the verb at Spell-Out (‘surface structure’). This is true if a Quirky element is part of the embedded clause and c-commands the nominative object, as in (5a). It is also true if the matrix Quirky element fails to raise, as in existential constructions like (38) (compare with (39)).7 (38) Það virðist/*virðast einhverjum manni hestarnir vera seinir. there seem.3sg/3pl some man.dat horses.nom be slow ‘Tt seems to some man that the horses are slow.’ (39) Mér virðast hestarnir vera seinir. me.dat seem.3pl horses.nom be slow ‘It seems to me that the horses are slow.’ However, Holmberg and Hróarsdóttir (2001) observe that this generalization is not true in situations where the matrix Quirky element undergoes A-bar movement. In such cases, agreement is blocked, as if the Quirky element were in the c-command domain of the Probe at surface structure. We thus face an A-/A-bar asymmetry. If Quirky elements A-raise, they do not block. If they A-bar raise, they do.8 The examples in (40)–(42) illustrate the blocking effect of an A-bar raised Quirky element with wh-movement (40), topicalization (41), and relativization9 (42).

Intricacies of Agreement in Icelandic

59

(40) Hverjum veist þú að virðist/*virðast hestarnir vera seinir. who.dat know you that seem.3sg/3pl horses.nom be slow ‘To whom do you know that the horses seem to be slow.’ (41) Þessum stúdentum veit ég að virðist/*virðast hestarnir this student.dat know I that seem.3sg/3pl horses.nom vera seinir. be slow ‘To this student I know that the horses seem to be slow.’ (42) Þetta er stúdentinn sem virðist/*virðast hestarnir vera seinir. these is student.nom that seem.3sg/3pl horses.nom be slow ‘this is the student to whom the horses seem to be slow.’ More surprisingly still, although A-bar raised elements block agreement with a downstairs nominative element, they render raising of that NP possible, as shown in (43). (43) Hverjum hafa/*hefur hestarnir virst vera seinir. who.dat have.3pl/3sg horses seemed be slow ‘To whom have the horses seemed to be slow.’ This is puzzling, as raising across an experiencer in Icelandic, unlike in English, is disallowed, as (44) indicates. (44) Hestarnir virðist/*virðast mér vera seinir. horses.nom seem.3sg/3pl me be slow ‘The horses seem to me to be slow.’ Raising of the nominative element in (44) is limited to A-bar raising, like topicalization, as indicated by the absence of agreement on the finite verb. This contrasts sharply with (43) where agreement between the raised nominative NP is obligatory, not just optional. As they stand, the facts just reviewed, especially the contrast between (40) and (43), are at variance with Chomsky’s (2000) claim adopted here that agreement underlies movement, as opposed to the previously popular idea that movement rendered agreement possible (so-called Spec-Head agreement). Before developing an account, let me review Holmberg and Hróarsdóttir’s (2001) analysis. The salient properties of their analysis are: (i) A-traces are invisible for purposes of Agree; (ii) A-bar traces are visible for the purposes of Agree; and (iii) Move may take place in the absence of agreement, and in fact, Move may feed agreement. Setting aside the question of where nominative Case comes from, which Holmberg and Hróarsdóttir do not discuss in detail, and the idea that differences types of traces/copies act differently for purposes of intervention (a possibility that was feasible in GB, when A- and

60

Aspects of the Syntax of Agreement

A-bar traces had distinct properties, but no longer is under a strict Minimalist Program),10 the most problematic aspect of Holmberg and Hróarsdóttir’s analysis is the idea that Move may take place in the absence of agreement, for pure EPP reasons. They claim that such movement is attested in Icelandic in the guise of Stylistic Fronting, which is analyzed by Holmberg (2000) as an instance of movement for pure EPP purposes. Ignoring the conceptual issues that arise from Holmberg’s analysis of Stylistic Fronting (see Boeckx 2003a for extensive arguments against pure-EPP-triggered movement), we can point to empirical differences between Stylistic Fronting and the type of raising that takes place in (43). First, as Jónsson (1991) observes, Stylistic Fronting only apply to heads. Holmberg (2000) provides some instances of Stylistic Fronting affecting non-heads such as (45), but all such examples are degraded for all my informants, which contrasts sharply with raising in (43), which is virtually perfect to native speakers’ ears. (45) ??*Þeir sem þessa erfiðu ákvörðun verða að taka. they that this difficult decision must.pl to take ‘those who must take this difficult decision, . . .’ Second, stylistic fronting never feeds agreement, even in the putative examples involving NPs reported by Holmberg (2000). Witness (46). This again contrasts with (43). (46) Sá sem þessar ákvarðanir ??*verður/**verðs að taka. he that these decisions must.sg/pl to take ‘those who must take this difficult decision, . . .’ Third, stylistic fronting is clause-bounded, as shown in (47). (Both examples involve crossing a non-finite clause boundary.) (47) a.

*Þeir sem vera telja Maríu hæfa era mjögfáir. they that be consider Mary competent are few ‘Those who consider Mary to be competent are few.’ b. ??Sá sem elda virđist matinn er . . . he that cook seems the.food is . . . ‘The one who seems to cook the food is . . .’

This, once more, contrasts with (43), where raising crosses a non-finite clause boundary. The only similarity between raising in (43) and stylistic fronting is that both appear to be licit if the subject is A-bar moved. I think it is fair to say that this similarity is outweighed by the differences I just provided. I take them to cast serious doubt on Holmberg and Hróarsdóttir’s analysis. The alternative account I would like to develop is twofold. My claim is that two possible derivations underlie cases like (40)–(42), one blocking agreement (and movement), and akin to the surface form obtained in

Intricacies of Agreement in Icelandic

61

(38), and the other allowing agreement and movement. Call the former the ‘blocking’ scenario and the latter the ‘movement’ scenario. The blocking scenario rests on the assumption that (40)–(42) all contain a covert expletive það (which is independently attested in Icelandic grammar). The presence of það indicates that finite T is not the kind of T that hosts Quirky element (call that T TMatch, cf. (18)). It is a T that has agreement potential. Call it TK. By (18), the presence of TK implies that vQ cannot assign nominative and agree with it. The only functional head that can host agreement is T, but the nominative element is inaccessible to it due to the presence of a Quirky element, which either stays there (38) or later (i.e., once the option of agreement has been computed) undergoes A-bar movement. So agreement is blocked under this very cyclic derivation. Everything hinges on not endowing vQ with nominative Case and agreement possibilities. If that choice is made, agreement with the nominative element downstairs may obtain, but vQ cannot combine with TK hosting the (covert) expletive. From this scenario, the status of (38) and (40)–(42) follows. Since there is no expletive in (39), vQ may be selected with an agreement option, and no blocking effect takes place in this case. The A-/A-bar distinction that was central to Holmberg and Hróarsdóttir’s account turns out to be tangential. We now have to account for (43). Here comes the ‘moving’ scenario. As stated earlier, the ‘blocking’ scenario depended on the presence of an expletive. If no expletive is selected, then nothing forces us to select TK. If TK is not present, vQ may be selected with agreement and Case possibilities. Agreement will not be blocked by the Quirky element since the latter is outside the c-command domain of the Probe. Now, suppose that as a result of this agreement possibility, (Nominative) Object Shift may take place. In other words, vQ may have an EPP-property but crucially only when it has K/Casefeatures. If Object Shift takes place, targeting the outer SpecvQP, then, by minimality, the Quirky element can vacate vQP by A-bar movement only, as schematized in (48). A-movement to SpecTP is unavailable, as it would have to cross an A-specifier (occupied by the Shifted object).11 (48) [__[C [__[T [vP Nom-NPi [vQP QuirkyNP [ vQ [VP V [TP ti [ T . . . ]]]]]]]]]] # z-----=--------m z---------------_--m As Marcel Den Dikken points out (p.c), this analysis predicts that if the nominative element is a wh-phrase, even A-bar movement will be ruled out. The pair in (49)–(50) (data from Kjartan Ottósson) shows that the prediction is borne out. (49) ?*Hverjum hafa hvada hestar/hverjir virst vera seinir. who.dat have which horses/who seemed be slow ‘To whom did which horses/who seemed to be slow.’

62

Aspects of the Syntax of Agreement (50) a.

Hverjum hefur virst hvada hestar/?hverjir vera seinir. who.dat has seemed which horses/who be slow b. *Hverjum hafa virst hvada hestar/hverjir vera seinir. who.dat have.pl seemed which horses/who be slow

As (50) indicates, the string under discussion is acceptable as long as the nominative wh-phrase stays within the embedded clause (which blocks agreement, see (50b)), which strongly suggests that we are dealing with an intervention effect in (49). Another confirmation of the present analysis comes from Kjartan Ottósson’s observation (p.c.) that the raising possibility in (43) is limited to 3rdperson elements (a fact not noted by Holmberg and Hróarsdóttir 2001). Non-3rd-person nominative pronouns are excluded from such contexts, as (51) shows. (51) *Hverjum hefur/höfum við virst vera gáfaðir. who.dat has.3sg/3pl we seemed be intelligent ‘To whom have we seemed to be intelligent.’ Recall that non-3rd-person pronouns are precisely those that cannot trigger agreement. (51) clearly indicates that overt movement is dependent upon the possibility of agreement (as opposed to feeding agreement), as Chomsky’s (2000) Agree theory predicts. In sum, the two scenarios developed here account for the intricate facts studied by Holmberg and Hróarsdóttir, and additional data discussed here for the first time. The present analysis avoids the conceptually problematic apparatus that they required to account for the facts. As a matter of fact, the present account strengthens Chomsky’s claim about the interaction of Agree and Move.

2.5. CONCLUSION In this chapter I have proposed an analysis of the syntax of agreement in Icelandic that is able to capture that behavior of Quirky subjects and nominative objects in a way that appears to be theoretically more constrained than in current alternatives in that it does not need to rely on Filters, Defective Intervention, or Morphological rules obscuring syntactic operations. By refining the licensing mechanisms of nominative Case assignment and letting Move take place under pure Match, the major intricacies of Icelandic ‘Quirky Agreement’ fall out immediately.

II

Agreement and Agree

3

Long-Distance Agreement in Hindi Theoretical Implications

Here I offer a detailed analysis of long-distance object agreement in Hindi. After reviewing the basic facts, and establishing similarities between the Hindi facts and facts that motivated an Agree-analysis in existential constructions of the English-type, I show that an Agree-based account of Hindi long-distance object agreement is superior to various alternatives: SpecHead agreement, feature movement, etc. I also suggest that the phenomenon of long-distance agreement in Hindi is related to restructuring, a specific analysis of which enables me to treat case and agreement licensing as working in tandem.

3.1. THE SPEC-HEAD RELATION Kayne (1989a) provided a powerful argument for capturing instances of agreement as resulting from the establishment of a Spec-Head relation. Kayne built his argument around the following paradigm from French. (1)

a.

Jean a vu-*e la fille. Jean has seen-agr.fem the girl ‘Jean saw the girl.’ b. Jean l’a vu-e. Jean her.has seen-agr.fem ‘Jean saw her.’ c. Quelle fille Jean a(-t-il) vu-e. which girl Jean has-he seen-agr.fem ‘Which girl did Jean see?’ d. Cette fille a été vu-e. this girl has been seen-agr.fem ‘This girl was seen.’

As can be seen here, past participle agreement is only possible with displaced objects (cliticized, wh-moved, and passivized in the examples at hand). Quite naturally, Kayne took agreement in this case (and, by hypothesis, in

66

Aspects of the Syntax of Agreement

all cases) to require raising to some specifier position—the Spec-Head relation. This line of thought has been very productive, and was at the core of the early minimalist conception of Case/agreement (Chomsky 1993). However, recently, several arguments were presented in favor of another agreement/Case configuration, which Chomsky (2000) dubbed Agreement at a distance, or Agree, for short. The evidence for Agree came from existential constructions and instances of long-distance agreement in Icelandic with nominative objects, illustrated in (2) and (3), respectively. In both cases there is compelling evidence (not summarized here; see Boeckx 2003a) showing that the agreeing NP has not raised to the locus of agreement at any point, making it difficult for a SpecHead analysis to be entertained.1 (2)

There *seems/seem to be two men in the boat.

(3)

Mér *virðist/virðast þeir vera skemmtilegir. me.dat seem.3sg/3pl they.nom be interesting ‘It seems to me that they are interesting.’

Based on such facts, Chomsky proposed a simplification of feature checking (including Case/agreement) by eliminating actual feature displacement (posited in Chomsky 1995 and originally used in the context of (2)–(3); see Chomsky 1995, and Chapter 1) in favor of a checking procedure taking place at a distance. Under Agree, a functional head, called Probe, checks its uninterpretable feature against a matching element, called Goal, located in the Probe’s complement domain, as schematized in (4). (4)

TP ru Tb ru To TP 6 . . . NP . . .

vP ru vb ru vo VP 6 . . . NP . . .

All instances of Case/agreement-checking are hypothesized to fall under Agree. In a sense, the nature of Case/agreement in existential constructions has become the model to which all other instances of Case/agreement must reduce. For Chomsky, any instance of categorical displacement is said to be an instance of EPP-satisfaction. Strictly speaking, EPP-satisfaction is a process distinct from Case/agreement checking, and I will say nothing more about it here. Crucially, for Chomsky, the Spec-Head relation is dismissed as a significant syntactic relation. Chomsky (2004: 109, 114) notes that a SpecHead relation is beyond the set of relations available within a reasonable definition of minimal search space (essentially, the complement domain).

Long-Distance Agreement in Hindi

67

Furthermore, as Boeckx (2003a) argues, a Spec-Head relation is also inconsistent with a label-less theory of the type outlined in Collins (2002), since Spec-Head is just another term for Case Assignment under m-command, command by a maximal projection, which does not exist in the absence of labels. It is therefore important to revisit the evidence adduced in favor of the Spec-Head relation for Case/agreement, and see to what extent it can be captured under the more restrictive Agree-theory. Clearly, the task is vast, as can be gathered from the discussion in Chung (1998), Koopman (1992, 1995, 2001), and Sportiche (1998), to cite but a few works that rely abundantly on the Spec-Head relation to capture a rich array of facts. This chapter has the more modest aim of discussing an agreement phenomenon in Hindi, LongDistance (Object) Agreement (henceforth, LDA). LDA shares most of the characteristics that have been associated with past participle agreement in French in Kayne (1989a). However, unlike French, Hindi provides a rather strong case for Agree and against Spec-Head, or so I will show in what follows. If I am correct, the properties of French past participle agreement ought to be dissociated from the Spec-Head relation, and an alternative analysis should be developed. I develop such an analysis toward the end of the chapter.

3.2. PROPERTIES OF HINDI LONG-DISTANCE AGREEMENT LDA refers to the ability of a verb, taking an infinitival complement, to agree with an object that is the internal argument of the infinitive, as illustrated in (5). (Throughout, the reader should focus on feminine nouns, which trigger a type of agreement that is morphologically distinct from default agreement.) (5)

Vivek-ne [kitaab parh-nii] chaah-ii. Vivek-erg book.f read-inf.f want-pfv.f ‘Vivek wants to read the book.’

LDA has several properties that will be the focus of this chapter. First, note that the matrix verb chaah-ii does not agree with the matrix subject. This conforms to the general rule of agreement in Hindi according to which a subject triggers agreement on a verb if and only if the subject is not overtly Case-marked (see, for instance, (6)). If the subject is overtly Case-marked, the object may trigger agreement on the verb (7). If both arguments are overtly Case-marked, the verb bears default inflection (8).2 (6)

Rahul kitaab parh-taa thaa. Rahul.masc book.f read-hab.msc be.pst.msc ‘Rahul used to read (a/the) book.’

68

Aspects of the Syntax of Agreement (7)

Rahul-ne kitaab parh-ii thii. Rahul.erg book.f read-hab.pfv.f be.pst.msc ‘Rahul had read the book.’

(8)

Rahul-ne kitaab-ko parh-aa thaa. Rahul-erg book-acc read-pfv.msc be.pst.msc ‘Rahul had read the book.’

A second property of LDA is that it is only possible with arguments of non-finite complements. Agreement across a finite clause boundary is impossible (9). (9)

Firoz-ne soch-aa/*-ii ki [Mona ghazal gaa-tii Firoz-erg think-pfv.3msg/3fsg that Mona ghazal.f sing-hab.f hai]. be.prs ‘Firoz thought that Mona sings ghazals.’

Third, LDA is not possible if the infinitival clause has an overt subject (contrast (10) and (11)). (10) *Firoz-ne [Shabnam-kaa rotii khaa-nii] chaah-ii. Firoz-erg Shabnam-gen bread.f eat-inf want-pfv.3fsg ‘Firoz wanted Shabnam to eat bread.’ (11) Firoz-ne [roti khaa-nii] chaah-ii. Firoz-erg bread.f eat-inf want-pfv.3fsg ‘Firoz wanted to eat bread.’ Fourth, LDA appears to be optional. Thus, (12) is as acceptable as (11). (12) Firoz-ne [rotii khaa-naa] chaah-aa. Firoz-erg bread.f eat-inf.msc want-pfv.msc ‘Firoz wanted to eat bread.’ Fifth, LDA only goes from bottom to top. That is to say, we do not find any instance of LDA where an argument of a matrix predicate triggers agreement on some embedded predicate. (13) *Mona peR-ko dekh-nii chah-tii thii. Mona tree.masc.acc see-inf.f.sg want-hab.f be-pst.f ‘Mona wanted to see the cat.’ Finally, for a vast majority of speakers,3 agreement on the infinitival predicate is ‘parasitic’ on agreement with the embedding predicate. That is, agree-

Long-Distance Agreement in Hindi

69

ment with the infinitival verb fails if LDA fails (14b), and agreement with the infinitival verb must obtain if LDA obtains (14c). (14) a.

Shahrukh-ne tehnii kaat-nii chaah-ii. Shahrukh-erg branch.f cut-inf.f want-pfv.f b. *Shahrukh-ne tehnii kaat-nii chaah-aa. Shahrukh-erg branch.f cut-inf.f want-pfv.msc c. *Shahrukh-ne tehnii kaat-naa chaah-ii. Shahrukh-erg branch.f cut-inf.msc want-pfv.f ‘Shahruck wanted to cut the branch.’

There have been few analyses of LDA in Hindi. Here I will focus on Mahajan’s (1989) and Bhatt’s (2001) analyses, as the other analysis of LDA I am familiar with, Butt (1995), is meant to account for dialects in which (10b) is grammatical (see fn. 2). Mahajan’s (1989) account of LDA ties the phenomenon to his claim that infinitivals optionally assign Case to their objects. If they do, no LDA takes place. If they don’t, objects must agree with the embedding predicate, triggering LDA. As Bhatt (2001) observes, Mahajan’s proposal immediately suffers from the fact that it predicts (incorrectly) that the subject of an infinitival clause may be Case-licensed by the embedding predicate if the infinitive fails to assign Case to it. However, (15) shows that the prediction is not borne out. The subject of the infinitival clause must be licensed by the infinitive. (15) *Ram-ne Mohan jaa-naa chaah-aa. Ram-erg Mohan go-inf want-pfv ‘Ram wanted Mohan to go.’ Because of this problem, and because of the more general issue of optional Case assignment in minimalism, I will not pursue Mahajan’s intuition further. Bhatt (2001) treats LDA as an instance of agreement at a distance, following Chomsky’s (2000) Agree-based Case/agreement licensing mechanism. Crucially, he dissociates agreement from Case. According to him, Case is assigned to the object by the infinitival predicate. Agreement is a ‘post-cyclic’ phenomenon that happens after the whole embedding clause has been built. The analysis I will propose in the next section shares with Bhatt’s the idea that Agree underlies LDA. In fact, this is the central point of the analysis, as it will enable me to provide a strong argument against the Spec-Head relation. Unlike Bhatt, I claim that Case and Agreement need not be dissociated, consistent with recent proposals that treat Case and agreement as having a common nature (see Boeckx 2003a for discussion and references).

70

Aspects of the Syntax of Agreement

3.3. AN AGREE-BASED ANALYSIS OF HINDI LONG-DISTANCE AGREEMENT From the perspective of current syntactic theorizing, the ‘parasitic’ agreement relation between the embedded object and the embedded infinitival is certainly the most interesting. It raises issues similar to those found when trying to come to grips with successive cyclicity: Should the intermediate steps be taken independently of the trigger for the final step, as an emphasis on local economy and minimization of look-ahead would have it, or should it be part of chain formation upon the introduction of the trigger for the final step of movement? The interesting twist coming from an investigation of the Hindi pattern at hand is that movement is, as I will show, irrelevant: All the conditions for LDA are computed at the level of Agree. Already at some abstract level, one can show that several characteristics of LDA reviewed in the previous section fall out immediately from the nature of Agree. For purposes of illustration, I will use examples from existential construction and Icelandic nominative object agreement, as these two are representative cases of Agree. First, Agree cannot reach across a finite CP boundary. (16) *There seem that three men are in the room. (17) Mér virðist/*virðast að þeir lesi bókina. me.dat seem.3sg/3pl that they.nom read book.the.acc ‘It seems to me that they read the book.’ Second, Agree is subject to Relativized Minimality. In particular, Agree is subject to defective intervention, whereby an element in between the Probe and the Goal blocks Agree even if that intervener itself cannot enter into agreement (see Chomsky 2000, Chapter 10). Instances of defective intervention in existential constructions and in Icelandic are given in (18)–(19). (For a detailed analysis of these, see Chapter 10.) (18) There seems/*?seem to a woman to be three men in the room. (cf. There *?seems/seem to be three men in the room.) (19) Mér fannst/*fundust henni leiðast þeir. me.dat seemed.3sg/3pl her.dat bore they.nom ‘I thought she was bored with them.’ (cf. Mér *virðist/virðast þeir vera skemmtilegir. me.dat seem.3sg/3pl they.nom be interesting ‘It seems to me that they are interesting.’) LDA is also subject to defective intervention, as shown in (10), repeated here as in (20): The presence of a genitive subject in the infinitival clause blocks LDA.

Long-Distance Agreement in Hindi

71

Crucially, the intervention is defective, as the embedded subject of an infinitival clause cannot trigger agreement on the matrix predicate (recall (15)). (20) *Firoz-ne [Shabnam-kaa rotii khaa-nii] chaah-ii. Firoz-erg Shabnam-gen bread.f eat-inf want-pfv.3fsg ‘Firoz wanted Shabnam to eat bread.’ (cf. Firoz-ne [roti khaa-nii] chaah-ii. Firoz-erg bread.f eat-inf want-pfv.3fsg ‘Firoz wanted to eat bread.’) Third, Agree is insensitive to movement, as it applies prior to displacement (EPP-satisfaction; Chomsky 2000). Thus, Agree obtains in the context of relativization (21), topicalization (22), or scrambling in the case of LDA (23). (21) The three men that there seem to be in the garden . . . (22) Þeiri *virðist/virðast mér ti vera skemmtilegir. they.nom seem.3sg/3pl me.dat be interesting ‘It seems to me that they are interesting.’ (23) Kitaabi Vivek-ne [ti parh-nii] chaah-ii. book.f Vivek-erg read-inf.f want-pfv.f ‘Vivek wants to read the book.’ On the basis of such similarities, it seems clear that a unified analysis, in terms of Agree, is desirable. Before spelling out the details of an Agree-analysis, I want to show that no other mechanism currently available in the theory will be able to explain LDA. I can think of three alternative possibilities: a feature-movement analysis, an overt Spec-Head agreement analysis, and a covert phrasal movement analysis. The three of them are inadequate. Let us start with the feature-movement analysis. On the surface it looks like LDA can be handled nicely under Move-F. The difference between Agree and Move-F is very subtle. Chomsky (2000) offers only a conceptual argument for abandoning Move-F in favor of Agree: Move-F require featurechains, which, like head-chains more generally, are at odds with optimization considerations.4 Move-F could explain the clause-boundedness of LDA (assuming, with Chomsky 1995, that feature-chains are like head-chains, hence clause-bounded), and defective intervention effects (see Chapter 1 for a Move-F analysis of Icelandic object agreement, recast in Agree-terms in Chomsky 2000), but it would be more difficult to apply it to data showing that agreement is independent of movement. Here is why. As originally suggested by Chomsky (1995) and exploited in Ochi (1999a) and Lasnik (2003), when feature-movement takes place overtly, the element out of which the feature has moved is ‘PF-defective’ and must raise

72

Aspects of the Syntax of Agreement

to a position close enough to the feature to be PF-legible (assume with Ochi and Lasnik that ‘close enough’ here means the specifier position to which the feature has adjoined). If we assume the SOV order in Hindi to be derived (from an SVO base, as in Kayne 1994), then, the derivation of a sentence like (5) (Vivek-ne [kitaab parh-nii] chaah-ii) is likely to proceed as in (24). (24) a. Vivek-ne chaah-ii [kitaab parh-nii] n Move-F b. Vivek-ne [Fi + chaah-ii] [[kitaab]i parh-nii] n raising of parh-nii c. Vivek-ne parh-niij [Fi + chaah-ii] [[kitaab]i tj]

n raising of kitaab or of [[kitaab]i tj] n d. Vivek-ne [[kitaab]i]k parh-niij [Fi + chaah-ii] [tk tj] or db. Vivek-ne [[kitaab]i tj]k parh-niij [Fi + chaah-ii] tk As can be seen from either (24d) or (24db), kitaab will not be adjacent to the raised feature triggering agreement. The derivation should therefore crash at PF. A PF-crash could be avoided if feature movement were to take place in a separate covert component (where PF-legibility, hence adjacency is irrelevant), as it did in existential constructions (Chomsky 1995). But covert raising is not an option here as LDA can take place with scrambled objects (see (23)). Scrambling would take the object beyond the c-command/attraction domain of the agreement matrix predicate in overt syntax, leaving no possibility for covert attraction. This could be prevented if scrambling is assumed to reconstruct and covert raising is allowed to take place after reconstruction. But if such an option is chosen, it leaves the defective intervention/minimality effect in (20) unexplained, as lowering essentially voids minimality effects (see Bošković and Takahashi 1998). A similar problem for Move-F arises if we assume that Hindi has an SOV base. Here too, it is easy to see that the object will not be adjacent to the agreeing predicate after feature-movement (the embedded participle will again intervene between them). And covert feature-movement is excluded for the reason mentioned in the previous paragraph. Thus it appears that the adjacency condition on overt Move-F is lethal to the feature-movement analysis of LDA.5 An overt Spec-Head relation fares no better, as evidenced from sentences like (25). (25) Vivek-ne [jaldise kitaab andhereme parh-nii] chaah-ii. Vivek-erg quickly book.f in.the.dark read-inf.f want-pfv.f ‘Vivek wanted to read the book quickly in the dark.’ As (25) shows, LDA obtains when the agreeing NP is sandwiched between adverbs belonging to the embedded clause. It is very hard to see how the

Long-Distance Agreement in Hindi

73

relevant NP would be picked out to sit in the relevant specifier of the relevant head to the exclusion of the adverbs sandwiching it. I don’t see any non-stipulative way of doing that. As for a covert Spec-Head relation, it suffers from the same problem I raised for the covert featural movement analysis: Any covert operation would fall short of capturing the fact that the relevant relation can obtain prior to an operation that has an overt reflex. Agree faces none of these problems. In the next section, I provide a precise characterization of how LDA obtains within an Agree-based theory.

3.4. AGREE AND RESTRUCTURING Let me start by drawing a parallelism between Hindi LDA and the wellknown and better studied phenomenon of clitic climbing in Romance. The parallelism will provide a basis for understanding the mechanics of LDA. Clitic climbing refers to a phenomenon where an argumental clitic from the embedded clause attaches to the embedding predicate, as shown in (26) from Italian. (26) Maria li ha voluti prendere. Maria them has wanted take ‘Maria wanted to take them.’ As has been extensively documented since at least Rizzi (1976), clitic climbing cannot take place in the presence of an overt subject in the infinitival clause (27), and cannot take place out of a finite clause (28), properties that it shares with Hindi LDA. (27) *Maria li ha voluti Gianni prendere. Maria them has wanted Gianni take ‘Maria wanted Gianni to take them.’ (28) *Gianni li credo che Maria ha voluti prendere. Gianni them thinks that Maria has wanted take ‘Gianni thinks that Maria wanted to take them.’ Rizzi (1976) and many subsequent studies have argued that clitic climbing is contingent upon restructuring, an operation that combines the embedding predicate and the infinitive into a complex predicate of sorts. Rizzi took restructuring to be an instance of reanalysis. Recently, Wurmbrand (2001a) has provided compelling evidence in favor of viewing infinitival complements undergoing restructuring as bare VPs. By bare VPs, she means that ‘restructured’ infinitival complements lack a v-layer, and thus the external argument that v is assumed to introduce,

74

Aspects of the Syntax of Agreement

along with objective Case resources (see Chomsky 1995 for the origin of the claim that v has both thematic and Case functions). This bareness implies that internal arguments of the infinitive will have to be Case-licensed in the matrix clause, hence will have to ‘climb.’ Just like LDA in Hindi, clitic climbing/restructuring is optional. Wurmbrand captures the optionality in the latter case by assuming that the embedding predicate may either select a bare VP (in which case, restructuring results (obligatorily)), or a larger phrase (IP or CP), which will lack all the characteristics of ‘clause union’ as it now contains enough functional projections to check the Case of the internal argument and introduce (and license) the external argument (vP and IP). I would like to extend Wurmbrand’s selection-based account to Hindi. In particular, I claim that when LDA takes place, the infinitival complement is a bare VP. The internal argument is Case-licensed (via Agree) by the matrix predicate, which, by the same token, agrees long distance with it. In other words, there is no Case-agreement dissociation (unlike in Bhatt’s 2001 analysis mentioned earlier). In LDA cases, the infinitival clause fails to license the Case on its internal argument because it lacks the functional layer (vP, to be precise). If the matrix predicate selects a phrase larger than VP, say IP or CP, the presence of a (possibly null) external argument triggers a defective intervention effect, which blocks LDA. (29) [v [V [VP V Obj ]]] z___________m Agree (30) [v [V [CP/IP C/I (. . .) [Subj v [V Obj]]]]] z______________________=-________m

LDA

no LDA

The proposed extension of Wurmbrand’s analysis of restructuring captures the apparent optionality of LDA in such a way that is consistent with the Minimalist intuition that there is no optionality when it comes to checking features. Optionality resides in which functional heads are selected from the lexicon (see also Lasnik 1999c). The Agree-based account also captures the directionality of LDA (see also Bhatt 2001): Agree is a ‘downward’ operation: It operates strictly within the c-command/internal domain of the attractor (Probe, in Chomsky’s 2000 terminology). As for the parasitic nature of agreement on the infinitive, I would like to argue that it is best captured as an instance of multiple Agree, of the type explored by Hiraiwa (2001, 2005), Nomura (2002a), Chapter 6, and Chomsky (2004). Instances of multiple agreement in a way mirror processes like vowel harmony or tone spreading in phonology, where all potential elements between the upper bound (one may say ‘Probe’) and the lower bound (the ‘Goal’) are valued due to spreading. In the particular case of LDA, the Probe (matrix V) looks down into its c-command domain, and finds a bare VP containing an infinitival form a Goal NP, and valued both

Long-Distance Agreement in Hindi

75

simultaneously. I speculate here that the infinitival verb in Hindi manifests agreement overtly because it is gerundive in character (thus more nominal than a straight infinitive, on a par with participles, which also manifest agreement).6 At any rate, a multiple Agree analysis explains why, if no LDA takes place, no agreement with the infinitive will take place, as the latter does not take place independently of the former.7 The application of multiple Agree in LDA is schematized in (31). (Linear order is irrelevant.) (31) [v [V [VP V Obj ]]] z________z___m Multiple Agree

LDA

All in all, it appears that an Agree-based analysis can capture the relevant properties of LDA in an explanatory fashion. The final issue I would like to tackle is French past participle agreement. In particular, I would like to argue that agreement on the past participle comes about as the result of the presence of an additional functional head associated with clitic objects (see Sportiche 1992 and much subsequent work; interestingly, Taraldsen 2002 also posits a clitic projection to capture participle agreement facts in Germanic and Romance). This probe, not active if the object is not a clitic, would agree with both the participle and the object clitic, as schematized in (32). (32) [ClP Cl [. . . [VP Part. Objclitic]]] z___________z____m Multiple Agree

LDA

The fact that object clitics subsequently undergo displacement is independent of agreement. Past participle agreement also obtains in two contexts in French: when the object is A-bar moved (wh-movement; clitic left dislocation), and when the object is passivized. I think that the two cases should not be treated on a par. For situations involving A-bar movement, I assume, as I did in Boeckx (2001, 2003b), that there is a phonologically null (resumptive) clitic associated with the wh-phrase (in the case of clitic left dislocation, the clitic is overt), as in (33). (33) [ClP Cl [. . . [VP Part. [Objclitic Wh-phrase]]]] z____________z____m Multiple Agree

LDA

Assuming the presence of a null resumptive clitic in (33) immediately accounts for the semantic effect of specificity noted by Obenauer (1994), Déprez (1998), Richards (2001), and Rizzi (2001b), and allows agreement to be computed locally (without look-ahead up to the C-domain). For instances of agreement under passivization, I assume that the participle is in fact an adjective (adjectives independently agree with the NP they modify in French). That is, we are dealing with a different situation

76

Aspects of the Syntax of Agreement

altogether. The fact that ‘past participle’ agreement under passivization is obligatory in French, as opposed to being optional in other contexts, in fact suggests that a non-uniform analysis is called for. My analysis of past participle agreement in French, if correct, suggests that what Kayne’s study and much work in its wake uncovered is not the need of a Spec-Head configuration for agreement but (in some cases, not in all) the presence of a functional layer. Once this functional layer is identified, an Agree analysis is at the very least equally valid.

3.5. CONCLUSION In this chapter I have offered an analysis of LDA in Hindi that crucially relies on the operation Agree introduced in Chomsky (2000). I have shown how an Agree-based account captures the core facts pertaining to LDA, and is superior to feature movement or Spec-Head agreement alternatives. From this I conclude that Agree should be generalized, and the temptation to use Spec-Head relations resisted, even in face of apparent empirical evidence to the contrary. This is welcome, as Agree can be shown to yield a more restrictive theory of syntactic relations, as argued in Chomsky (2000, 2004). Aside from this theoretical point, I have also argued that LDA is a phenomenon akin to clitic climbing, and extended Wurmbrand’s (2001) analysis of restructuring in terms of bare VP complement selection to Hindi. This extension has allowed me to view Case and agreement as two sides of the same coin, which concurs with much recent work in syntactic theorizing (see Boeckx 2003a for review).

4

Conditions on Agreement in Japanese

This chapter integrates Japanese object honorification within a larger crosslinguistic context, and provides a principled explanation for an otherwise puzzling property: the fact that direct object honorification is blocked in the presence of a dative argument. Following a well-established tradition in the generative literature, honorification is here treated as a case of agreement, but, unlike previous approaches, which rely on Spec-Head configurations, Agree is shown to be able to capture (object) agreement in a more straightforward manner. The blocking effect of dative elements is a reflex of a more general locality constraint, ‘defective intervention’ (a special case of minimality), and is related to Person-Case Constraint effects.

4.1. OBJECT HONORIFICATION IN JAPANESE This chapter investigates the nature of the object honorification in Japanese and its implications for principles of Universal Grammar. Honorification in Japanese is largely determined by two factors. One is sociolinguistic (the element associated with honorification must be socially superior to and respected by the speaker). The other is syntactic. We will be focusing here exclusively on defining the syntactic conditions for honorification.1 We will not concentrate on the fairly well-studied case of subject honorification, in which an honorific marker on the verb is associated with the subject noun phrase, as in (1): (1)

Tanaka sensei-ga hon-o o-yomi-ni-nat-ta. Prof. Tanaka-nom book-acc read-sh-past ‘Prof. Tanaka read the book.’

Since Shibatani (1976), subject honorification has been treated as an instance of (abstract) subject verb agreement. Rather compelling evidence in favor of this position is to be found in Toribio (1990), Ura (2000), and Hasegawa (2002).

78

Aspects of the Syntax of Agreement

Example (2) illustrates the phenomenon of object honorification, where the verb bears an honorific marker associated with the object noun phrase, not the subject noun phrase: (2)

Taro-ga Tanaka sensei-o o-tasuke-si-ta/tasuke-ta. Taro-nom Prof.Tanaka-acc help-oh-past/help-past ‘Taro helped Prof. Tanaka.’

As shown in (2), the morphological shapes of subject and object honorifics are different. The main difference pertains to the suffix associated with the verb. In the case of subject honorification, ni nar is used, whereas su or itas function as object honorifics. The morphology of honorification in Japanese is summarized in (3) (from Harada 1976: 504).2 (4) illustrates both types of honorifics: (3)

Morphology a. HP+INF ni nar(Subject Honorification) b. HP+INF su- / itas(Object Honorification) HP = the ‘honorific prefix’, i.e. o-/go(common to both forms of honorification) INF = the infinitive form

(4)

a. hanasi-ta b. o-hanasi ni nat-ta c. o-hanasi si-ta

‘talk-past’ ‘talk-sh-past’ ‘talk-oh-past’

In the first modern study of object honorification, Harada (1976: 530) proposes a rule called Object Honorific Marking: (5)

Object Honorific Marking Mark the predicate as [Object Honorification] when an SSS (a person who is socially superior to the speaker) is included in (a) the indirect object, if the predicate is ditransitive, or (b) the direct object, if the predicate is transitive.

The relevant examples appear in (6)–(7). In (6), the verb is transitive, and it agrees with the direct object in honorification. In (7), we have a ditransitive predicate, and the verb agrees in honorification with the indirect object: (6)

Taro-ga Tanaka sensei-o o-tasuke-si-ta. Taro-nom Prof.Tanaka-acc help-oh-past ‘Taro helped Prof. Tanaka.’

(7)

Hanako-ga Tanaka Sensei-ni Mary-o go-syookai-si-ta. Hanako-nom Prof. Tanaka-dat Mary-acc introduce-oh-past ‘Hanako introduced Mary to Prof. Tanaka.’

Conditions on Agreement in Japanese

79

(8) illustrates the core phenomenon of the present analysis. The predicate is ditransitive, as in (7), but this time the NP capable of triggering honorification (Tanaka sensei-o) functions as the direct object. In such a case, object honorification is impossible. If an object honorific marker surfaces on the verb, we obtain the odd interpretation that Taro respects Mary. In other words, the honorific marker in a ditransitive predicate can only associate with the indirect object, not the direct object: (8)

*Hanako-ga Mary-ni Tanaka Sensei-o go-syookai-si-ta. Hanako-nom Mary-dat Prof. Tanaka-acc introduce-oh-past ‘Hanako introduced Prof. Tanaka to Mary.’

Note, incidentally, that changing the word order does not affect the object honorification relation, as shown in (9). The object is unable to control object honorification in the presence of an indirect object3: (9)

a.

Hanako-ga Mary-o Tanaka Sensei-ni go-syookai-si-ta. Hanako-nom Mary-acc Prof. Tanaka-dat introduce-oh-past ‘Hanako introduced Mary to Prof. Tanaka.’ b. *Hanako-ga Tanaka Sensei-o Mary-ni go-syookai-si-ta. Hanako-nom Prof. Tanaka-acc Mary-dat introduce-oh-past ‘Hanako introduced Prof. Tanaka to Mary.’

Our goal in this study is to formulate an empirically correct condition on object honorification that captures the restriction exemplified by (8) in an explanatory fashion. As we describe in section 4.2, Harada’s formulation extends beyond simple cases of indirect objects. After providing the necessary empirical refinements to Harada’s generalization, we offer an analysis of object honorification based on Chomsky’s (2000) Agree operation. We then show that Chomsky’s (2000, 2001) Defective Intervention Effect on Agree accounts straightforwardly for the conditions on object honorification (section 4.3). Section 4.4 embeds Japanese object honorification within a larger context of agreement (or lack thereof), which includes constraints on clitic clustering in Romance and nominative object agreement in the context of Quirky subjects in Icelandic. Section 4.5 is a summary. Appendix 1 discusses the issue of Case assignment in double object constructions. Appendix 2 provides two potential arguments in favor of an Agree-based account of object honorification over a Spec-Head treatment.

4.2. ADDITIONAL OBJECT HONORIFICATION CONFIGURATIONS In this section, we illustrate the range of objects capable of triggering object honorification in Japanese.

80

Aspects of the Syntax of Agreement

4.2.1. Object Honorification with Ni-NP-Objects Harada’s rule in (5) captures the basic facts about Japanese ni-phrases (indirect objects). As shown in (10)–(14), the verbs kasu ‘lend,’ ageru ‘give,’ tsukuru ‘make,’ osieru ‘teach,’ and kakeru ‘call’ take indirect ni-objects, which are responsible for Object honorification. (That this is the case is confirmed by the ungrammaticality of the sentences that follow here if ‘Professor Tanaka’ is replaced by ‘Mary,’ which is incapable of triggering honorification.): (10) Taro-ga Tanaka sensei-ni/*Mary-ni Yamada sensei-no Taro-nom Prof. Tanaka-dat/Mary-dat Prof. Yamada-gen hon-o o-kasi-si-ta. book-acc lend-oh-past ‘Taro lent Prof. Yamada’s book to Prof. Tanaka/to Mary.’ (11) Taro-ga Tanaka sensei-ni/*Mary-ni tanjoobi purezento-o Taro-nom Prof. Tanaka-dat/Mary-dat birthday present-acc sasiage-ta. give-oh-past ‘Taro gave the birthday present to Prof. Tanaka/to Mary.’ (12) Taro-ga Tanaka sensei-ni/*Mary-ni keeki-o o-tsukuri-si-ta. Taro-nom Prof. Tanaka-dat/Mary-dat cake-acc make-oh-past ‘Taro made the cake to Prof. Tanaka/to Mary.’ (13) Taro-ga Tanaka sensei-ni/*Mary-ni insutooru-no sikata-o Taro-nom Prof. Tanaka-dat/Mary-dat install-gen way-acc o-osie-si-ta. teach-oh-past ‘Taro taught Prof. Tanaka/Mary how to install.’ (14) Taro-ga Tanaka sensei-ni/*Mary-ni denwa-o o-kake-si-ta. Taro-nom Prof. Tanaka-dat/Mary-dat phone-acc call-oh-past ‘Taro made a phone call to Prof. Tanaka/to Mary.’

4.2.2. Object Honorification with kara-NP-Objects However, the ni-phrase is not the only element which plays a role in the Object honorification in Japanese. As shown in (15)–(17), a ‘from NP’ can also control object honorification: (15) Taro-ga Tanaka sensei-kara/*Mary-kara hon-o Taro-nom Prof.Tanaka-from/Mary-from book-acc o-kari-si-ta/haishaku-si-ta. borrow-oh-past/borrow-oh-past ‘Taro borrowed the book from Prof. Tanaka/from Mary.’

Conditions on Agreement in Japanese

81

(16) Taro-ga Tanaka sensei-kara/*Mary-kara hana-o Taro-nom Prof. Tanaka-from/Mary-from flower-acc itadai-ta. receive-oh-past ‘Taro received the flower from Prof. Tanaka/from Mary.’ (17) Taro-ga Tanaka sensei-kara/*Mary-kara hanasi-o Taro-top Prof. Tanaka-from/Mary-from story-acc ukagat-ta. hear-oh-past ‘Taro heard the story from Prof. Tanaka/from Mary.’

4.2.3. Object Honorification with To-NP-Objects Also, a to-phrase can agree with the verb in honorification. The verbs hanasi ‘talk’ and sagasi ‘look for’ in (18)–(19) take to-phrases and they can agree neither with the ‘about’ NP nor with the accusative NP: (18) Taro-ga kinoo Tanaka sensei-to/*Mary-to Taro-nom yesterday Prof.Tanaka-with/Mary-with Yamada sensei-nitsuite o-hanasi-si-ta. Prof. Yamada-about talk-oh-past ‘Taro talked yesterday to Prof. Tanaka/Mary about Prof. Yamada.’ (19) Taro-ga Tanaka sensei-to/*Mary-to Taro-nom Prof. Tanaka-with/Mary-with Yamada sensei-no o-taku-o o-sagasi-si-ta. Prof. Yamada-gen house-acc look for-oh-past ‘Taro looked for Prof. Yamada’s house with Prof. Tanaka/Mary.’

4.2.4. Object Honorification with Benefactives A reviewer points out that null (dative) benefactives can also trigger object honorification (a fact already discussed by Harada 1976: 527). Witness the cases in (20)–(24): (20) Watasi-ga obentoo-o otabe-si-masyoo. I-nom lunch-acc eat-oh-pres ‘I’ll eat (your) lunch for you.’ (21) Watasi-ga kawarini Eberesuto-ni o-nobori-si-masu. I-nom instead Everest-dat climb-oh-pres ‘I’ll climb Mt. Everest instead (of you).’

82

Aspects of the Syntax of Agreement (22) Watasi-ga ryoori-o o-tori-si-masu. I-nom dish-acc take-oh-pres ‘I’ll take the dishes for you.’ (23) Watasi-ga sore-o o-kaki-si-masu. I-nom that-acc write-oh-pres ‘I’ll write it (for you).’ (24) Watasi-ga denwabangoo-o o-sirabe-si-masu. I-nom phone number-acc check-oh-pres ‘I’ll check the phone number (for you).’

As the reviewer notes, all the cases just listed are acceptable if they receive a benefactive interpretation. We take those benefactives to be represented syntactically as instances of pro (possibly as instances of null applicatives). To summarize the facts so far, Harada’s Object Honorific Marking rule extends beyond simple ni-marked indirect objects, and encompasses other instances of (argumental) ‘datives.’ In all cases, the presence of a ‘dative’ controls object honorification (i.e., the direct object is incapable of associating with the honorific marker on the verb). To accommodate all cases presented in this section, we will refer to ‘indirect objects’ as ‘datives.’ Following Shibatani, Toribio, and others, we take honorification to be an instance of (abstract) agreement on a par with the more common K-feature agreement found in many languages of the world. We further assume that all datives are capable of triggering agreement because their K-features are accessible in Japanese (for a similar conclusion, based on the behavior of dative subjects, see Ura 1999, 2000). That is, the postposition/Case-marker associated with the NP does not prevent the latter from agreeing with v. The simplest hypothesis, which we will adopt, is to treat such postpositions/Case-markers as adjoined to DP (for arguments in favor of this position, see Niinuma 2000 and Takano 1997, among others; for a related position, see Kayne 2000b, 2004). Thus, we suggest that all datives are NPs, visible to the verb.4 Having said that, we revise Harada’s generalization as in (25)5: (25) Object Honorification Agreement generalization Mark the predicate as [Object Honorification] when an SSS (a person who is socially superior |to speaker) is (a) an (argumental) Dative (b) the accusative object, if the predicate does not take a dative argument 4.3. CAPTURING HONORIFICATION VIA AGREE6 Chomsky (2000, 2001, 2004) proposes as an alternative to the Spec-Head relation the relation Agree to capture properties of agreement. The latter

Conditions on Agreement in Japanese

83

amounts to a process of feature checking (in his terms, valuation) at a distance. Chomsky (2000) proposes that Agree take place under Match, but not every matching pair induces Agree (p. 122). In particular, Chomsky provides one argument in favor of distinguishing Match from Agree. The argument rests on the existence of what he calls “defective intervention effects” (Chomsky 2000: 123). Defective intervention arises when an element A matches the featural requirements of a probe P, but fails to agree with it (for reasons we will not go into here; in the case Chomsky discusses, A bears inherent Case, which renders its K-features inert). Crucially, in such cases, no more deeply embedded element B that matches the featural requirements of P, and is able to agree with it in other circumstances, is accessible for checking, due to the presence of A. A clear case of intervention arises in Icelandic Quirky subject constructions (the data are taken from Chapter 1, where the agreement facts are discussed at length). As is well known, Quirky subjects fail to trigger agreement on the finite verb (26), despite the fact that they behave for all other purposes as bona fide subjects (again we assume that this follows from the fact that Quirky elements bear inherent Case, and that inherent Case-marked elements have inert K-features): (26) Stelpunum var hjálpað. the.girls.dat.pl.fem was.3sg helped.neuter.sg. ‘The girls were helped.’ However, their presence blocks the establishment of an agreement relation between the verb and a nominative element (27), which is otherwise possible (28): (27) Mér fannst/*fundust henni leiðast þeir. me.dat seemed.3sg/3pl her.dat bore they.nom ‘I thought she was bored with them.’ (28) Mér *virðist/virðast þeir vera skemmtilegir. me.dat seem.3sg/3pl they.nom be interesting ‘It seems to me that they are interesting.’ In terms of Agree, (27) and (28) can be schematized as in (29) and (30), respectively: (29) (Dative-NP) (Infl) Verb [. . . Nominative NP . . .] | Agree | (30) (Dative NP) (Infl) Verb [ . . . Dative NP . . . [. . .Nominative NP. . .]] | X |

84

Aspects of the Syntax of Agreement

If Valuation were the only significant relation, the intervention effect in (30) would be unexpected, since the quirky element cannot participate in Valuation. However, if Match exists independently of Valuation, the blocking effect in (30) falls into place. Being a closer matching element, the quirky NP renders the nominative NP inaccessible to the finite verb. We will now argue that the failure of object honorification with a direct object in the presence of an indirect object is another case of ‘defective intervention.’7 That intervention is indeed defective is shown by the fact that the indirect object itself cannot trigger honorific agreement (it fails to refer to a socially superior person), but nevertheless prevents the direct object (which has the relevant features) from agreeing with v. (Our account, like Chomsky’s, raises the question of how Case is checked/assigned to the direct object, if it is not accessible to v. We address this issue in Appendix 1; see also Chapter 6.) Thus, schematically, the impossibility of direct object honorification in (8) can be represented as in (31) (parallel to (30)): (31) v [Dative-NP [ V Accusative-NP]] z____________=_______________m The first, correct, prediction of our analysis is that in the absence of a dative element, object honorification with the direct object can take place unhindered. As we saw earlier, this is the case. In monotransitive contexts, v can probe for the direct object, and agree with it: (32) a.

Taro-ga Tanaka sensei-o o-tsure-si-ta. Taro-nom Prof. Tanaka-acc take-oh-past ‘Taro took Prof. Tanaka.’ b. [vb v [ DO V]] z___m Agree

Another prediction of our proposal is that if the object bearing the relevant feature for honorification is embedded within a larger noun phrase, it should be inaccessible to Agree. As shown in (33)–(34), this prediction is borne out. In (33), Tanaka sensei ‘Professor Tanaka’ is embedded inside a noun phrase. By Relativized Minimality, the closer element for an Agree relation with v is ‘Mary.’8, 9 Here too ‘Mary’ does not possess the relevant feature for honorification, but nonetheless blocks the search. A similar situation obtains in (34), where o-tonari-san-o ‘neighbor’ intervenes10 (taken from Harada 1976) (unless o-tonari-san-o is taken away, in which case Tanaka sensei-no becomes accessible to v, as in (35)): (33) *Watasi-ga Mary-no Tanaka sensei-nituite-no kizi-o I-nom Mary-gen Prof. Tanaka-about-gen article-acc haidoku-si-ta. read-oh-past ‘I read Mary’s article about Prof. Tanaka.’

Conditions on Agreement in Japanese

85

(34) *Watasi-ga Tanaka sensei-no otaku-no o-tonari-san-o I-nom Prof. Tanaka-gen house-gen neighbor-acc o-tazune si-ta. visit-oh-past ‘I visited the house next door to Prof. Tanaka’s.’ (35) Watasi-ga Tanaka sensei-no otaku-o o-tazune si-ta. I-nom Prof. Tanaka-gen house-acc visit-oh-past ‘I visited Prof. Tanaka’s house.’ Several consequences of our analysis are worth pointing out at this point. First, in order to capture the relevant defective intervention effects, it is crucial for us that the dative element c-command the accusative element. If the reverse were possible, the accusative element would be closer to v, and there would be no defective intervention. That, in turn, would entail that the base order in Japanese is rigid, as originally argued on independent grounds by Hoji (1985). In other words, Miyagawa’s (1997) proposal that both and word orders are basic in Japanese cannot be correct. If it were, (direct) object honorification in the presence of an indirect (dative) object would be predicted to be possible, as schematized in (36) (the structure that Miyagawa would assign to (9b))11: (36) [vb v [DO [IO V]]] z___m Agree As a matter of fact, our analysis provides a powerful empirical argument against any base-generation analysis of scrambling. As we saw in (9) earlier, word order alternations do not affect object honorification, which follows if the latter is determined via Agree. Take, for concreteness, Bošković and Takahashi’s (1998) analysis of scrambling, which to date is the most detailed base-generation analysis of the phenomenon. According to them, ‘scrambled’ phrases are in fact base-generated in their surface positions, and undergo covert (LF) lowering to check their theta-roles, thereby satisfying Full Interpretation (FI). Under Bošković and Takahashi’s analysis, a scrambling sentence like (37) is analyzed as (38): (37) Sono hon-o John-ga [Mary-ga katta to] omotteiru. that book-acc John-nom Mary-nom bought that thinks ‘John thinks that Mary bought that book.’ (38) Numeration n (a) n (c) (LF); n (b) (PF) a. [sono hon-o [John-ga [[Mary-ga [katta ]] to] omotteiru]] b. sono hon-o John-ga Mary-ga katta to ometteiru c. t [John-ga [[Mary-ga [[sono hon-o] katta]] to] omotteiru] |____LF-lowering__________m

86

Aspects of the Syntax of Agreement

Note that if Agree is taken as a derivational process (i.e., it applies as soon as the probe is introduced), as it is in Chomsky (2000, 2001, 2004), Bošković and Takahashi would predict that object honorification is not possible if the object is scrambled, contrary to fact (see (9)). This is so because for them the object is not in the c-command domain of the agreeing verb at the point when Agree applies. A probe-Goal relation could therefore not be established. (As a reviewer notes, this is especially true in the case of long-distance scrambling, for which Bošković and Takahashi merge the scrambled object in the matrix clause, where there cannot be any c-command relation between the object and the embedded v upon First-Merge.) Bošković and Takahashi also incorrectly predict no intervention effect by datives in (9b). To see this, it is worth discussing how they account for the absence of superiority effects under scrambling. As has often been noted, scrambling of both objects in a ditransitive structure may result in either order. Consider (39): (39) a.

sono hon-o John-ni Mary-ga watasita. that book-acc John-dat Mary-nom handed ‘Mary handed the book to John.’ b. John-ni sono hon-o Mary-ga watasita.

Bošković and Takahashi account for the absence of superiority by capitalizing on the fact that scrambling is lowering. Since movement is not to a c-commanding position, they predict no Relativized Minimality (Superiority) effects, since the latter arise only when ‘Closeness,’ which is defined in terms of c-command, is violated. Such an account works nicely for (39), but it clearly fails to predict the distribution of the object honorific marker in Japanese, as it treats both objects of a ditransitive predicate as equally close to v. Based on the object honorification facts discussed here, we conclude that scrambling must be a movement-based account. Our analysis also provides evidence for Nemoto’s (1993) and Kitahara’s (1999) claims that the position occupied by the object under “short scrambling” is not a VP-adjoined position (contra Ura 1996, 2000, Takano 1998). It must be higher than v. This is so because if the VP-adjunction analysis were correct, the structure of (9b) would be (40). (40) [vP [v’ v [VP Obj1 [VP Dat . . . t1 . . .]]] z_____m Agree At the point of adjunction to VP, the direct object is closer to v than the indirect object, and we would thus expect v to be able to agree with the direct object, contrary to fact. Under Nemoto’s and Kitahara’s analyses, short scrambling applies after the merger of v, and therefore fails to affect the relevant Agree relation (41)12:

Conditions on Agreement in Japanese

87

(41) [vP Obj1 v [VP Dat t1 V]]] Before closing this section, we would like to point out that taking v to be the relevant probe (locus of object honorification) captures the distribution of the honorific marker more straightforwardly than an analysis that assumes a (full-blown) ‘Split VP’ structure. Building upon Johnson (1991), Koizumi (1993, 1995) (see also Lasnik 1995a) argues that the locus of object agreement is below v, as in (42): (42) [vP vo [AGRoP AGRoo [VP Vo [ OBJ]]]] Bures (1992) already argued that in double object constructions, the relevant structure must be as in (43) (for additional arguments and related proposals, see Collins and Thráinsson 1996, Koizumi 1995, Kayne 2001b): (43) [AGRioP AGRioo [vP vo [AGRdoP AGRdoo [VP IO [Vo [ DO]]]]] Note that such a structure would fail to capture the object honorification facts in Japanese. Here the direct object moves to an AGRdo, and hence would be able to trigger agreement.13 An alternative structure for ditransitives, explored in Lasnik (1995a) and subsequent work, is given in (44): (44) [vP vo [AGRioP AGRioo [VP IO [Vo [AGRoP AGRoo [VP Vo [ DO]]]]] Here, each argument is dominated by its Case shell, and no notion of equidistance is needed to account for Case-marking. But note that here too, the direct object raises to some AGR projection; hence, if no additional assumption is made, we expect direct object honorification agreement to be possible in ditransitives. In sum, taking v as a probe for object agreement seems to yield the most straightforward account of object agreement in Japanese.

4.4. FROM HONORIFICATION IN JAPANESE TO AGREEMENT CROSS-LINGUISTICALLY The picture that emerges from our account of object honorification in Japanese as subject to defective intervention effects is that (honorific) agreement in Japanese is no different from (K-) agreement cross-linguistically. As already pointed out in section 4.3, defective intervention accounts for agreement constraints in Icelandic (see Chapters 1 and 2 for fuller discussion and references). In the presence of a Quirky “dative” in the c-command domain of the probe, a nominative object fails to trigger agreement on the finite verb.

88

Aspects of the Syntax of Agreement

Chapter 1 argues that the Icelandic facts mirror constraints on clitic clustering in Romance, in particular, the so-called me-lui constraint (Perlmutter 1971, Kayne 1975, Bonet 1994). This constraint basically says that the presence of the dative clitic prevents the presence of a 1st/2nd person accusative clitic. If an accusative clitic is used, it must be 3rd person.14 Taking 3rd person to be the default, Boeckx argues that the presence of a dative clitic prevents the presence of an agreeing accusative clitic—another case of defective intervention effect on agreement (we assume that the dative clitic c-commands the accusative clitic upon first Merge (before cliticization), as in Japanese; hence the intervention effect): (45) Me-lui Constraint If AGRdative n AGRobj = 3rd person (default) a. *Jean me lui a recommandé. Jean me.acc him.dat has recommended ‘Jean recommended him to me.’ b. Jean le lui a recommandé. Jean it.acc him.dat has recommended ‘Jean recommended it to him.’ We can now liken the Japanese facts to those found in Romance and Icelandic.15 If so, our analysis begs the question of whether functional heads in Japanese are as impoverished as has sometimes been claimed (see, especially, Fukui 1986, 1995). Given that we find constraints on agreement in Japanese similar to those found cross-linguistically (i.e., in languages where agreement is more pervasive, morphologically speaking), it would be undesirable to say that agreement is absent or fundamentally different in Japanese. There is, however, a sense in which agreement in Japanese is ‘different.’ As Shigeru Miyagawa has observed (in a comment following our presentation of this material at FAJL 3), agreement in Japanese seems to be restricted to honorification (or ‘person’ features). One does not find agreement in number (or gender). It is interesting to note in this context that Chomsky 2000 treats expletive elements as defective K-feature bearers, where defectiveness is taken to be [+person]. Following Chomsky, we may say that agreement is ‘defective’ if restricted to [person] features, thereby making agreement in Japanese comply with Fukui’s claim that Infl is defective in the language. What is important for us is that even defective agreement be subject to the same universal locality constraint on Agree.

4.5. CONCLUSION To conclude, we have examined the properties of object honorification in Japanese. Refining original observations of Harada’s (1976), we have argued for the following generalization:

Conditions on Agreement in Japanese

89

(46) Mark the predicate as [Object Honorification] when an SSS (a person who is socially superior to speaker) is (a) a Dative argument. (b) the direct object, if the predicate does not take a dative argument. Chomsky’s (2000) Agree relation captures the whole array of facts once it is assumed to be subject to defective intervention effects, which are attested independently. Our analysis enables us to bring the Japanese agreement facts fully in line with agreement constraints cross-linguistically, and thus provides another piece of evidence that, abstractly, languages are near-invariant.

APPENDIX 1: CASE-ASSIGNMENT IN DITRANSITIVES In this appendix, we will return to the question of Case assignment left open in section 4.3. Recall that for us the direct object is inaccessible to v if a dative element is present. This begs the question of how the direct object checks/is assigned Case. We would like to offer three possible solutions. The simplest answer is to treat Case as a default, possibly via the presence of a Case-particle, as argued in Takano (1997). Alternatively, we may follow Chomsky’s (2000: 123) suggestion that the Case requirement of the moving element need not be checked. Evidence for this claim comes from gapping, for which we will adopt Johnson’s (1994) analysis. Johnson suggests that a sentence like (47) is derived by Across-the-Board movement of the verb, with the second conjunct lacking a T-layer, as represented in (48)16: (47) John reads Plato, and Mary Aristotle. (48) [IP Johni Infl [XP [Xb readsj [VP ti [vb tj Plato]], and [VP Mary [Vb tj Aristotle]]]]] As noted by Kayne (2000b: 165f.), in the absence of a T-layer in its domain, the subject in the second conjunct is expected to lack Case. The grammaticality of the sentence in (47) supports Chomsky’s (2000) conjecture that what is primary is not structural Case itself, but the K-features of the head (T/v) the argument interacts with. Put differently, failure to check Case on Mary does not lead to a crash (the features of Infl are checked against those of John). What matters is what Howard Lasnik dubbed the Inverse Case Filter: the requirement that the {(Case/) K-} features of the probe be checked. (For extensive discussion of the Inverse Case Filter, see Bošković 1997c, Martin 1999, and Boeckx 2000a.) A third option would be to treat Case and K-features separately. Accordingly, we may claim that Case is assigned ‘simultaneously’ to all relevant elements within a phase, as in Hiraiwa (2001). An alternative would be to say

90

Aspects of the Syntax of Agreement

that all elements within a given domain (say, the vP-phase) are co-valued, as in Lopez (2002) (for a related proposal, see Frampton and Gutmann’s 2000 idea of Agreement as Feature Sharing). Equivalently, Case may be treated as an [Attract-All] feature in the sense of Bošković (1999) (see also Niinuma 1999), which, for reasons that we will not review here, is not subject to Attract Closest. By contrast, the K-features of v have an [Attract-One] property, subject to Attract Closest, and hence the (defective) intervention effect discussed in this chapter. For a detailed exploration of this idea, see Chapter 6. Choosing among the three alternatives sketched here would take us too far afield. We leave this issue for future research.

APPENDIX 2: POTENTIAL ARGUMENTS AGAINST A SPECHEAD AGREEMENT TREATMENT OF HONORIFICATION17 In this chapter we proposed an Agree-based analysis of object agreement in Japanese. We think that Agree captures the relevant intervention effect of datives very naturally. Since Chomsky (2004) recently argued that SpecHead agreement is an epiphenomenon, and suggested that Agree underlies all instances of agreement, it would be interesting to find arguments against Spec-Head agreement in the realm of object honorification. In this appendix we offer two potential arguments to that effect. We stress that the arguments are contingent upon specific analyses of independent phenomena such as the structure of head-internal relative clauses and secondary predication. This prevents us from saying that the arguments are fully conclusive. In addition, we think that in the absence of a full-fledged theory of what counts as a possible functional projection, there is no way to exclude the possibility of Spec-Head agreement, as it will always be possible to claim the existence of an abstract XP to implement Spec-Head agreement mechanically (though, we feel, at the cost of explanatory adequacy). A Spec-Head agreement analysis of object honorification would take the following form: Object honorification is the result of the relevant NP raising to SpecvP, as schematized in (49) (linear order is immaterial, here and throughout): (49)

vP ty NP1 vb ty VP v ty t1 Vb ty NP2 V

Conditions on Agreement in Japanese

91

Assuming that all ‘datives’ are first-merged higher than accusative NPs (as we argued earlier), the intervention effect caused by datives would reduce to a standard Attract Closest account. Two pieces of evidence against such an analysis of object honorification based on Spec-Head agreement can be made. The first piece of evidence comes from head internal relative clauses.18 Consider (50): (50) John-wa [Tanaka sensei1-ga kuukoo-ni o-tuki-ni natta no]-o pro1 John-top Prof. Tanaka-nom airport-at arrive-sh-past-no-acc o-mukae-si-ta. greet-oh-past ‘John greeted Prof. Tanaka, who arrived at the airport.’ Following Murasugi (1995) and Hoshi (1996), we assume that in the case of head-internal relative clauses, a pro functions as the head of the relative clause.19 Hoshi’s structure, which we adopt, is given in (51) (notice that the hierarchical relation between the head-internal relatives and pro is important; if the order were reversed, a violation of Condition C of the Binding Theory would result): (51)

vP ty vb ty XP vb 6 ty . . NP1 . . VP v ty pro1 V

Assuming the correctness of Hoshi’s analysis, let us now reconsider (50). The sentence is acceptable with object honorification. Under Hoshi’s analysis, the object pro agrees with the verb (and the latter, by transitivity, with ‘Prof. Tanaka’). If the Spec-Head configuration were responsible for object honorification, the structure needed for (50) would be as in (52). However, such a structure would rule out the sentence as a Condition C violation:20 (52)

vP ty pro1 vb ty XP vb 6 ty . . NP1 . . VP v ty t1 V

92

Aspects of the Syntax of Agreement

The second piece of evidence against treating object honorification as an instance of Spec-Head agreement comes from secondary predicates in Japanese. Building upon Koizumi (1994), Yatsushiro (1999) assumes the following structural representation for the subject depictive phrase and the object depictive phrase (see also Pylkkänen 2002): (53) Depictive Predicates in Japanese TP ty Tb ty vP T ty NP vb ty SDP vb ty VP v ty NP Vb ty ODP V Under her analysis, the subject depictive phrase is adjoined to vb. Evidence for this position comes from the following VP preposing data21: (54) VP Preposing a. Taro-ga hadaka-de katuo-o tabeta. Taro-nom naked bonito-acc ate ‘Taro ate the bonito naked.’ b. [VP katuo-o tabe-sae]1 Taro-ga hadaka-de t1 sita. bonito-acc eat-even Taro-nom naked did ‘Even eat the bonito, Taro did naked.’ c. [vP hadaka-de katuo-o tabe-sae]1 Taro-ga t1 sita. naked bonito-acc eat-even Taro-nom did ‘Even eat the bonito naked, Taro did.’ Yatsushiro (1999) argues that VP preposing in Japanese may target either VP or vP. In (54b), the subject depictive phrase hadaka-de is outside of VP, and thus fails to be part of the preposed VP. (54c) is an instance of vP preposing. Koizumi (1994) further argues that the subject depictive phrase is structurally higher than the position of the indirect object. Evidence for this assumption comes from variable binding (Koizumi 1994: 41). In (55a), the variable ‘soitu’ cannot be bound by the NP in indirect object position. However, if the indirect object moves to the higher position due to the scrambling, then the variable can be c-commanded, and thus be bound:

Conditions on Agreement in Japanese

93

(55) Variable Binding a. ?*Hanako-wa [soitu1-no mottomo sukina kakkoo]-de Hanako-top it-gen most favorite fashion [Yamada-to yuu otoko to Tanaka-to yuu otoko]1-ni Yamada called man and Tanaka called man-dat tanzyoo purezento-o watasita. birthday present-acc gave ‘Hanako gave a birthday present to [a man called Yamada and a man called Tanaka]1 in his1 most favorite dress.’ b. [Yamada-to yuu otoko to Tanaka-to yuu otoko]1-ni2 Yamada called man and Tanaka called man-dat Hanako-wa [soitu1-no mottomo sukina kakkoo]-de t2 Hanako-top it-gen most favorite fashion tanzyoo purezento-o watasita birthday present-acc gave The VP structure we arrive at is given in (56), where the subject depictive phrase (SDP) is generated higher than the indirect object: (56) [vb SDP [VP NP-ni [Vb NP-o . . . ]]] Combining Yatsushiro’s and Koizumi’s results, we now have a strong prediction concerning object honorification. If the Spec-Head agreement is responsible for object honorification in Japanese, then the dative NP must move to Spec vP at some point in the derivation so as to be able to trigger agreement. It should therefore be able to bind a variable inside the depictive phrase. However, as shown in (57), this is not the case: (57) a.

?*Taro-wa [so1-no kata-no mottomo sukina kakkoo]-de Taro-top it-gen person-gen most favoriate fashion [Yamada sensei to Tanaka sensei]1-ni tanzyoo purezento-o Prof. Yamada and Prof. Tanaka-dat birthday present-acc o-watasi-sita. give-oh-past ‘Taro gave a birthday present to [Prof. Yamada and Prof. Tanaka]1 in his1 most favorite dress.’ b. Taro-wa [Yamada sensei to Tanaka sensei]1-ni Taro-top Prof. Yamada and Prof. Tanaka-dat [so1-no kata-no mottomo sukina kakkoo]-de it-gen person-gen most favoriate fashion tanzyoo purezento-o o-watasi-sita. birthday present-acc give-oh-past ‘Taro gave a birthday present to [Prof. Yamada and Prof. Tanaka]1 in his1 most favorite dress.’

94

Aspects of the Syntax of Agreement

As shown in (57a), even though object honorification obtains, the bound variable reading is not available. This reading is possible when the dative NP moves to the higher position via scrambling, as illustrated in (57b). If the arguments made on the basis of the head-internal relative and the depictive phrase data are tenable, we can conclude that the most straightforward Spec-Head agreement analysis is incapable of capturing the facts about object honorification in Japanese. In both cases, we have shown that the object triggering object honorification fails to move to SpecvP. Our conclusion would then corroborate Chomsky’s (2004) statement that the SpecHead relation is irrelevant to matters of agreement, contrary to much work in the late 1980s and in the 1990s (see Belletti 2000 for a comprehensive overview).

5

Honorification as Agreement

This chapter addresses Bobaljik and Yatsushiro’s (2006) criticisms leveled at the analysis proposed in Chapter 4, specifically, the claim that object honorification in Japanese can (and should) be assimilated to more familiar instances of (verb-object) agreement. It is shown that Bobaljik and Yatsushiro’s criticism is too weak to abandon the kind of analysis proposed in Chapter 4. I conclude that object honorification is indeed agreement, and that the most straightforward account of restrictions on object honorification should be couched in terms of Agree.

5.1. INTRODUCTORY REMARKS Bobaljik and Yatsushiro (2006), hereafter B&Y, take issue with Chapter 4’s claim that object honorification in Japanese can (and should) be assimilated to more familiar instances of (verb-object) agreement, operating under Chomsky’s (2000) conception of checking/valuation (Agree-ata-distance). In this note I wish to examine the problems B&Y raise for Chapter 4’s analysis, and the logic behind them, and show that upon close scrutiny B&Y offer no reason to abandon Chapter 4’s analysis. As a starting point, it is, I think, useful to briefly go over what Chapter 4 proposed, and what it was trying to achieve, since this is not entirely clear from B&Y’s presentation. Chapter 4 can be summarized as follows. (I) Chapter 4 treats honorification in Japanese, in particular, object honorification, on a par with familiar object–verb agreement. (II) Chapter 4 focuses on the fact that the presence of a “dative” (phrases suffixed by -ni, -kara, -to, null benefactives) precludes direct object honorification marking on the verb (which is otherwise possible in transitive contexts). (III) Chapter 4 capitalizes on the fact that word order variation (scrambling) has no ‘repair’ effect on the constraint in (II) to argue that honorification marking (i.e., agreement) takes place prior to movement

96

Aspects of the Syntax of Agreement

operations involving objects. Specifically, Chapter 4 argues that honorification takes place under Agree (feature checking at a distance), and not under Spec-Head agreement, corroborating Chomsky’s (2000) claim for more familiar instances of agreement. Chapter 4 follows Chomsky in taking Agree to be constrained by some version of (relativized) minimality, and claims that (II) is an intervention effect. (IV) Chapter 4 liken the intervention effect responsible for (II) to other agreement-blocking effects on direct objects/accusatives caused by the presence of indirect objects/datives such as the well-known Person-Case/me-lui constraint. It furthermore argues that the intervention effect responsible for (II) is an instance of what Chomsky (2000) calls “defective intervention”—a situation where the intervener blocks agreement (checking/valuation, more generally) despite the fact that the intervener itself cannot trigger agreement (checking/valuation). I want to make two remarks on (I)–(IV) before turning to B&Y’s critique. First, I want to make clear what is new and what is not about Chapter 4’s analysis (see Chapter 4). The claim that honorification is in part a matter of syntax is not new. The central thesis of Harada (1976/2000), the groundbreaking work on honorification in a generative context, was that honorification is “conditioned by grammatical factors” (p. 266). The view that honorification is a reflex of agreement is also not new (see Shibatani 1978 and Toribio 1990, among others). The empirical observation in (II) also goes back to Harada (1976/2000: 301), although it is left there as “an ad hoc condition on object honorific marking.” Toribio (1990:541) suggests briefly (actually, stipulates) that the presence of the indirect object causes a bounding problem for movement of the direct object to the specifier position where honorific agreement takes place.1 What is new about Chapter 4 is the claim that honorification does not require movement (III), and that the blocking effect of the dative (II) can be seen as a reflex of minimality in its minimalist instantiation. Second, as B&Y acknowledge at various points, “the shape of [Chapter 4’s] analysis is straightforward.” I take this to be a very desirable virtue. On the one hand, the simplicity of Chapter 4’s analysis makes it easy to see where it can go wrong. On the other hand, the simplicity of the analysis comes from Chapter 4’s reliance on concepts such as Agree and Minimality that current minimalist research takes to be part of good design in language. The simplicity of the analysis alone suggests that something like Chapter 4’s analysis is to be abandoned only under the pressure of strong empirical arguments to the contrary. This could be case if it were shown that (object) honorification is a non-syntactic phenomenon. It is hard to see what nonsyntactic condition would capture (II), especially in light of the fact that datives block what clearly is syntactic object agreement in a variety of circumstances (see IV). That said, of course, Chapter 4’s analysis is not immune

Honorification as Agreement

97

to criticism, but I want to stress that the kind of analysis it offers is, at the moment and to the best of my knowledge, the only one that captures the relevant facts. Consider, in this context, B&Y’s claim that “an assimilation of honorification to object agreement along the lines [Chapter 4] propose[s] is a much more ambitious task than [the chapter] suggests.” I find this assertion misleading. In the absence of a full-fledged alternative conception of honorification, rejecting the kind of analysis Chapter 4 entertains is a very difficult task, much more difficult than B&Y’s statement suggests. The independently attested blocking effect of datives on object agreement seems to me to suggest that Chapter 4’s analysis is not very “ambitious”; rather, it appears to be the null hypothesis.2 In sum, pending an alternative conception of honorification, criticism is likely to be directed at levels of technical detail. This is exactly what B&Y’s article is about. As I will show, the kind of arguments raised by B&Y are too weak to reject Chapter 4’s analysis, although some of their arguments suggest that refinements are needed, and that improvements can be made—a welcome result.

5.2. ALLEGED PROBLEMS B&Y cast doubt on Chapter 4’s analysis by questioning (i) the appropriateness of morphological glosses, (ii) the c-command relation holding between “datives” and direct objects, and (iii) the degree to which honorification is similar to agreement. I tackle each issue in turn. I should point out to the reader that due to space limits I will not be able to reproduce the type of evidence Chapter 4 and B&Y offer in support of their generalizations. Interested readers are referred to the original articles.

5.2.1. Glossing over Glosses Chapter 4’s analysis of (II) in terms of (III) can be schematized as follows [IO = indirect object; DO = direct object]: (1)

[vP v[+hon-K] [VP IO V DO]] z__________=_______m

(linear order irrelevant)

B&Y object to (1) because they claim that nothing in the data indicates that v is the honorific-K probe triggering agreement. As Harada (1976/2000) already noted, the only genuinely honorific morpheme on the verb is the prefix (g)o-, a prefix that attaches to a nominal(ized)/gerund/infinitive form of the verb, which then combines with the light verb -suru (followed by tense and complementizer markers that will play no role here), as shown in (2).

98

Aspects of the Syntax of Agreement (2)

Taroo-ga Tanaka sensee-o o-tasuke-si-ta. Taro-nom Tanako Prof.-acc hon-help-suru-past. ‘Taro helped Prof. Tanaka.’

B&Y are correct in claiming that (g)o- cannot be equated with v or any projection c-commanding IO and DO. But I think that B&Y are wrong to conclude from this that honorification cannot be taken to be an Agree relation between v (the head standardly taken to probe for object K-features) and IO/DO. At worst (for Chapter 4), what the gloss in (2) suggests is that there is a syntax–morphology mismatch. After all, the existence of suppletive forms discussed in Chapter 4 suggests that the syntax of honorification is not always morphologically transparent. Although such mismatches are never nice things to appeal to (I, like so many others, would like to hold on to Baker’s 1985 Mirror Principle as long as possible), it is a fact that they do exist, to be captured, perhaps, as readjustment rules; see Halle and Marantz (1993). But before we are forced to explore the mismatch hypothesis, I would like to suggest a more appealing way to reconcile (1) and (2). As Harada (1976/2000) originally noted, the prefix (g)o- always co-occurs in the verbal domain with light verbs: -suru (for object honorification) or -ni nar- (for subject honorification). Suppose these light verbs are ‘flavors’ of v (in the sense of Folli and Harley 2005) that match (g)o-V in a way similar to usual v-V combinations. (To avoid complications pertaining to the morphological status of the nominal(ized)/gerund/infinitive form of the verb the honorific prefix attaches to, I will label/treat V as a root (√), as is done in Distributed Morphology; see, e.g., Marantz [2000]. Nothing crucial hinges on this assumption, as far as I can see.) If this Root-v is tenable, it is not implausible to assume some featural transmission (including, crucially, honorific K) between the two heads (see, e.g., Collins 2002 on selectional requirements as Agree between heads).3 The resulting structure is given in (3).4 (3)

[vP vsuru [“–P” IO [[hon-–] DO]]]

As should be obvious, (3) is a minimal departure from (1), but it is enough to accommodate the morphological concerns raised in B&Y not taken into consideration by Chapter 4. It leaves Chapter 4’s syntactic analysis unaffected.

5.2.2. (Absence of) C-Command between “Datives” and Direct Objects B&Y point out that the “datives” interveners over the range listed by Chapter 4 do not all obviously c-command DO. For example, kara-NP ‘datives’ block object honorification but fail to license otagai ‘each other’ inside DO, as shown in (4).

Honorification as Agreement (4)

99

*Taro-ga [Yamamoto-sensee to Tanaka-sensee]i Taro-nom Yamamoto Prof and Tanaka-Prof -kara otagaii-no hon]-o kari-ta. -from each-other-gen book-acc borrow-past. ‘Taro borrowed each other’s books from Prof Yamada and Prof Tanaka.

As B&Y acknowledge in a footnote (note 17), speakers’ judgments vary concerning (4). Indeed, quite a few speakers judge (4) with the intended binding relation as “ok.” B&Y note that they have also encountered “variation in testing Chapter 4’s examples.” They do not provide any example numbers, so it is not clear to me which examples they have in mind. However, B&Y point out in the same footnote that “what is important for [their] point is not the absolute judgments of course, but rather the failure of any systematic correlation between honorification possibilities and other c-command diagnostics.” It is not clear that there is no systematic correlation between the effect of the presence of kara-datives on object honorification and the status of sentences like (4). Of the 20 native speakers of Japanese on the Harvard campus I interviewed (January 2005), all of them regard honorification with the direct object in sentences like (5) as unacceptable. (5)

Taro-ga Tanaka sensee-kara hon-o o-kari si-ta. Taro-nom Tanaka Prof-from book-acc hon-borrow suru-past. ‘Taro borrowed the book from Prof Tanaka.’

However, more than half of them (12 speakers) judge (4) as “ok” with the intended binding relation. This would suggest, following B&Y’s use of otagai data, that c-command obtains between IO and DO. So, for more than half of my informants, there may be a correlation between binding data and honorification possibilities, going in the direction of Chapter 4’s analysis. More robust evidence (in the sense of being less subject to variation in my testing sample) for the c-command relation between kara-datives and direct objects comes from variable binding cases like (6), constructed by Taka Kato (p.c.), based on works by Hoji (2003) and Ueyama (1998). (All speakers I interviewed accept (6)–(8) with the relevant binding relation.) (6)

Monkasyoo-ga [50 izyoo-no daigaku]-kara The Ministry of Education-nom 50 or more-gen university-from [so-ko-no tosyokan-nikansuru hookokusyo]-o uketotta. that-place-gen library-about report-acc received. ‘The Ministry of Education received from fifty or more universities a report on their libraries.’

Similar examples obtain for ni- (7) and to- (8) phrases, which also block direct object honorification (see Chapter 4).

100

Aspects of the Syntax of Agreement (7)

Monkasyoo-ga [50 izyoo-no daigaku]-ni The Ministry of Education-nom 50 or more-gen university-dat [so-ko-no tosyokan-o kaitikusuru tame-no sikin]-o ataeta. that-place-gen library-acc rebuild for-gen fund-acc gave. ‘The Ministry of Education gave fifty or more universities a fund to rebuild their libraries.’

(8)

Monkasyoo-ga [50 izyoo-no daigaku]-to The Ministry of Education-nom 50 or more-gen university-with [so-ko-no tosyokan-no kaitiku]-nikansite kyoogisita. that-place-gen library-gen rebuilding-about consulted. ‘The Ministry of Education consulted with 50 or more universities about the rebuilding of their libraries.’

What the preceding data show is that testing the c-command relation between IO and DO involving licensing of otagai does not give rise to very robust results, while testing the same relation by using variable binding data does.5 To conclude this section on c-command, I would like to point out that even if the empirical considerations I have just raised proved incorrect, it would still not be enough to reject Chapter 4’s analysis. C-command paradoxes (conflicting c-command data) of the type B&Y claim arise are not restricted to the domain of honorification in Japanese.6 To cite but two relatively familiar examples, Reinhart (1983), Kayne (1994), and others have proposed various syntactic solutions to the contrast in (9)–(10). (9)

Every girli’s father loves heri.

(10) *Every girli’s father believes herselfi to be a genius. Similarly, Kitahara (1997), Chapter 8 and others have discussed the c-command paradox that arises in raising constructions like (11)–(12): If c-command obtains, as variable binding indicates, why is anaphor binding blocked, or why doesn’t movement violate minimality? (11) Pictures of himselfi seem to every boyi to be attractive. (12) *Pictures of Mary seem to every boyi to put himselfi in trouble. In light of such examples, I conclude that at worst (for Chapter 4) B&Y’s examples force one to reconsider the nature and compatibility of various c-command tests, and of the role of c-command in the definition and scope of syntactic intervention (see Aoun and Li 2003: Chapter 2 and Müller 2004 for recent discussion and potentially problematic cases for the c-command

Honorification as Agreement

101

condition on intervention). At best, for Chapter 4, B&Y’s examples just point to the quirky nature of otagai for some speakers.

5.2.3. On Honorification Being Like Agreement B&Y’s last criticism concerns the defective intervention hypothesis advanced in Chapter 4. Their criticism is twofold: On the one hand, they claim that datives are not defective interveners in Japanese. On the other hand, they claim that the intervention effect we find in Japanese is unlike other kinds of intervention effects in the realm of agreement such as the Person-Case Constraint discussed by Chapter 4. Chapter 4 claimed that intervention in (1) is a case of defective intervention, a descriptive term meant to capture the fact that the intervening dative need not be [+honorific] to block direct object honorification. B&Y take issue with the notion of defective intervener, pointing out that the honorific marker is not ruled out in intervention cases; rather, the dative controls honorification, leading to unacceptability in case the dative is not of the appropriate social status in the universe of discourse. As B&Y point out, “the dative in Japanese is a non-defective intervener.” The point is well taken, and in fact was recognized independently in Niinuma (2003) and Chapter 10, both of which recognize that the defective character of datives in Chapter 4’s system is an artifact of the featural makeup assumed there. It is also acknowledged in Chapter 4 in fn. 7 (not mentioned in B&Y). The contents of Niinuma’s and Chapter 10’s studies concerning honorification were not reworked into the published version of Chapter 4 in part due to the fact that whether intervention is defective or not is a descriptive issue, and does not affect the main point of the chapter. Whether intervention is defective or not doesn’t make honorification more or less like agreement.7 What matters is the presence of intervention effects, which follows if honorification is a matter of Agree(ment). B&Y further criticize Chapter 4’s honorification-as-agreement hypothesis by claiming that Chapter 4’s analogy between (1) and Person-Case Constraint (PCC) effects in Romance and other languages is incorrect. According to B&Y, the constraint on honorification in (II) is unlike PCC effects because PCC effects refer to situations where the “dative restricts the possible choices for a lower element.” B&Y further note that “[Chapter 4] present[s] no evidence that the direct object is in any way restricted in the presence of a dative.” B&Y’s criticism stems from the fact that by PCC they mean something different from what I had in mind. In Chapter 4 (and elsewhere; see later description), PCC effects are cases where the presence of a dative in the relevant configuration (similar to the one in (1)) blocks person agreement with the direct object. Specifically, if clitics are taken to be agreement markers (a common, though not uncontroversial view; see Sportiche 1996), PCC effects amount to the absence of clitic/agreement [+Person] direct object. If realized at all, the direct object

102

Aspects of the Syntax of Agreement

must be a strong pronoun or a full noun phrase, not an agreement marker/ clitic. Put differently, PCC effects amount to the following: no [+Person] agreeing/clitic DO in the presence of IO. This is exactly what we find in Japanese: Taking honorification to be [+Person] agreement, as argued in Chapter 4, the presence of a dative precludes the presence of a [+Person] agreeing direct object. The similarity seems to me pretty clear. Following Anagnostopoulou (2003) and Jeong (2004) and references therein, we take PCC effects to be the result of syntactic intervention (like (1)). If the PCC is taken to be the reflex of syntactic intervention, the conception of the PCC I just gave is the only possible one. In current terms, Agree constrains the content of K-features on the Probe (v, T, etc.), not on the Goal (argument NPs), since Agree is a valuation operation, and Goals come out of the lexicon with fullyvalued K-sets. So, whatever the PCC does, if it is a constraint on valuation, it can only affect the shape of the verb, and it cannot be a “constraint on possible choices of a direct object” (B&Y), although it may, of course, have consequences for what kinds of direct objects (clitics, e.g.) can be licensed. B&Y admit that honorification can be seen as agreement if we restrict ourselves to languages that conform to the following generalization: (13) Morphological non-subject agreement is controlled by: a. The indirect object if there is one. b. The direct object otherwise. This is exactly what Chapter 4 claims. However, B&Y claim that the pattern in (13) is not universal, and cite Moravcsik (1978) to support their view. Roughly put, Moravcsik claims that some languages conform to (13), while other languages allow agreement with both dative and accusative NPs, and in yet other languages, “depending on some conditions,” the verb agrees with either accusative or dative NPs. Moravcsik notes that no language displays the reverse of (13) (morphological non-subject agreement is controlled by the direct object if there is one, the indirect object otherwise). Moravcsik’s passage in fact lends support to Chapter 4. If we assume, along the lines of a principle like UTAH (Baker 1988), that in all languages the indirect object (dative, in the unmarked case) NP8 is always higher than the direct object (accusative, in the unmarked case) NP (see Kayne 1984, Larson 1998, and, more recently, Pylkkänen 2002, Jeong 2006, 2007a, Lee 2004), it will be the case that intervention of the type found in (1) will result. It follows that languages will conform to (13), but not the reverse of (13). To the extent that both direct and indirect objects agree, one will find PCC effects that will limit the range of agreement markers associated with the direct object. (As B&Y note, “when the pattern in [(13)] does arise, it is often with concomitant restrictions on the person, number, and other properties of the DO.”) To the extent that one finds languages where DO proceeds

Honorification as Agreement

103

unhindered by the presence of IO, one must assume that IO is inert for case/ agreement in these situations (see McGinnis 2005, Jeong 2006, 2007a). In sum, I conclude that the constraint on honorification in (II) really is like the type of constraint one finds in agreement systems. Note, crucially, that I focused on syntactic constraints on agreement; I said virtually nothing about morphology (see Chatper 1 for one argument that agreement in syntax and agreement in morphology do not always go hand in hand). B&Y are right in saying that there is a “relatively wide range of variation attested in (morphological) agreement systems.” But this is not the premise that syntacticians focusing on agreement (as in Chapter 4) work on. Rather, they assume that the computational system of narrow syntax is uniform across languages, and that there is only one agreement system, one that predicts the pervasive presence of intervention effects like (1) once combined with UTAH (i.e., the claim that datives are higher than direct objects).

5.3. CONCLUDING REMARKS In this brief note I have reviewed B&Y’s arguments against Chapter 4’s analysis of object honorification in Japanese. B&Y do not offer an alternative analysis of any of the data discussed in Chapter 4, and confine themselves to pointing out technical problems with the specific proposal Chapter 4 made. I have shown that none of the problems force us to abandon Chapter 4’s hypothesis. This is not to suggest that Chapter 4’s analysis is correct in all its aspects, or complete. There are a few issues left open in Chapter 4 that deserve careful investigation. Most prominent among these, in my opinion, is how Chapter 4’s hypothesis relates to the typology of ditransitive structures in terms of high/low applicatives (Pylkkänen 2002, McGinnis 2003, Miyagawa and Tsujioka 2004).9 In addition, Chapter 4 calls for a unified analysis of object honorification, subject honorification, and other manifestations of honorification in Japanese and other languages. As usual, there is no knowing in advance whether any of these extensions will prove problematic. All I have shown here is that B&Y offer no reason to stop entertaining the kind of analysis Chapter 4 explored.

6

Multiplicity Symmetries and Asymmetries

Here I show that an investigation of the minutiae of Agree can lead to a better understanding of symmetries and asymmetries in instances of multiple agree (in both the A- and A-bar domain). Crucial to this account is the decomposition of Agree into suboperations (Match and Value), the pervasive role of minimality as a locality condition, and the idea that the computational system is strongly derivational in character. Of particular relevance here is the claim that case and agreement features must be regarded as features with a parallel function. Specifically, agreement features are analyzed as interpretable on the Goal (NP) but uninterpretable on the Probe (T/v). The reverse situation obtains for Case: It is an interpretable feature on the Probe (corresponding to the Tense/Aspect feature), but uninterpretable on the Goal (NP).

6.1. INTRODUCTION This chapter seeks to refine the formal aspect of multiple wh-fronting by refining the concept of Multiple Agree, recently advocated by Hiraiwa (2001, 2005), and Chomsky (2004). I will mostly be dealing with the pattern of multiple wh-fronting found in Bulgarian. The language may be characterized abstractly as requiring movement of all wh-phrases (which I will call the symmetric part of the process), while at the same time favoring movement of the first wh-phrase, by imposing a superiority condition on it (call this the asymmetric part of the process). Examples in (1) illustrate both aspects of the process. (1)

a.

Koj kogo kakvo e pital? who whom what is asked ‘Who asked whom what?’ b. Koj kakvo kogo e pital? c. *Kogo kakvo koj e pital? d. *Kakvo kogo koj e pital? e. *Kakvo koj kogo e pital?

Multiplicity: Symmetries and Asymmetries f. g. h. i. etc.

105

*Kogo koj kakvo e pital? *Koj e pital kogo kakvo? *Kogo e pital koj kakvo? *Kakvo e pital koj kogo?

There is reason to believe that the Bulgarian pattern is more basic than the one found in Serbo-Croatian (among other languages). As Bošković has shown in a series of work (Bošković 1997b, 1998b, 1999, 2002b), SerboCroatian appears to lack the asymmetric aspect of multiple wh-fronting. Witness the absence of superiority in (2).1 (2)

a.

Ko šta gdje kupuje? who what where buys ‘Who buys what where?’ b. Ko gdje šta kupuje? c. Šta ko gdje kupuje? d. Šta gdje ko kupuje? e. Gdje ko šta kupuje? f. Gdje šta ko kupuje?

However, Bošković has conclusively argued that the lack of superiority is only true in matrix questions in which there is no overt evidence for a complementizer. Elsewhere (e.g., embedded questions, long-distance questions, correlatives, matrix questions with overt complementizers), Serbo-Croatian is like Bulgarian, as (3) shows, using long-distance questions. (3)

a.

Ko si koga turdio da je istukao? who are whom claimed that is beaten ‘Who did you claim beat whom?’ b. *Koga si ko turdio da je istukao?

To situate the investigation within current syntactic theorizing, I will take as my starting point Bošković’s (1999) characterization of Bulgarian. Bošković argues that two features are checked against C0 in the process of multiple wh-fronting.2 Asymmetry arises from the checking of a [wh]feature, which can only be checked once (in Bošković’s terms, it has an [Attract-1] property). Superiority follows, as checking of the [wh]-feature must be by the closest matching element (wh-phrase). Symmetry emerges from the checking of a [Focus]-feature,3 which, unlike the [wh]-feature, can be checked multiply (it has an [Attract-all] property). Because all features must be checked in the most economical way (a mainstay of the minimalist program), the order of elements checking the [Focus] feature does not matter (i.e., no superiority effect arises), as the same number of nodes will ultimately be crossed (notice the global aspect of economy calculation here).

106

Aspects of the Syntax of Agreement

In what follows I will preserve the gist of Bošković’s analysis, in particular the idea that two features ([wh] and [focus]) are involved in the process of multiple wh-fronting. But I want to embed the checking of those features within a different framework, so as to eliminate what I take to be suboptimal aspects of Bošković’s analysis such as ‘Attract-1/Attract-All’ and global economy. This will require a detour into the nature of multiple checking in the A-domain, which is better understood, as it has been subject to extensive investigation in recent years.

6.2. MULTIPLE AGREE IN THE A-DOMAIN Consider the case of a ditransitive predicate, which for the purpose of discussion takes two structurally Case-marked NP-objects. For illustration, I will use Japanese examples. (For equivalent constructions in a wide-range of languages, see Boeckx 2003a.) Japanese double objects have an interesting property that has been documented in much detail in Chapter 4 and 10. The property pertains to object honorification, which I take to be an instance of agreement (see the references cited earlier for extensive discussion of this issue). In the first modern study of object honorification, Harada (1976: 530) proposes the following rule called Object Honorific Marking: (3)

Object Honorific Marking Mark the predicate as [Object Honorification] when an SSS (a person who is socially superior to the speaker) is included in (a) the indirect object, if the predicate is ditransitive, or (b) the direct object, if the predicate is transitive.

The relevant examples appear in (4)–(5). In (4), the verb is transitive, and it agrees with the direct object in honorification. In (5), we have a ditransitive predicate, and the verb agrees in honorification with the indirect object. (4)

Taro-ga Tanaka sensei-o o-tasuke-si-ta. Taro-nom Prof.Tanaka-acc help-oh-past ‘Taro helped Prof. Tanaka.’

(5)

Hanako-ga Tanaka Sensei-ni Mary-o go-syookai-si-ta. Hanako-nom Prof. Tanaka-dat Mary-acc introduce-oh-past ‘Hanako introduced Mary to Prof. Tanaka.’

However, Chapter 4 observes that (3) has to be refined in light of cases like (6). (6)

*Hanako-ga Mary-ni Tanaka Sensei-o go-syookai-si-ta. Hanako-nom Mary-dat Prof. Tanaka-acc introduce-oh-past ‘Hanako introduced Prof. Tanaka to Mary.’

Multiplicity: Symmetries and Asymmetries

107

Here, the predicate is ditransitive, as in (5), but this time the NP bearing the relevant feature to trigger honorification functions as the direct object. In such a case, object honorification (i.e., agreement between the verb and the direct object) is impossible. If an object honorific marker surfaces on the verb, we obtain the odd interpretation that Taro respects Mary. What this amounts to is that the honorific marker in a ditransitive predicate can only associate with the indirect object, not the direct object. (7) summarizes the distribution of the object honorific marker in Japanese in light of (6). (7)

Object Honorification Agreement generalization Mark the predicate as [Object Honorification] when an SSS (a person who is socially superior to speaker) is (a) a Dative argument (b) the accusative object, if the predicate does not take a dative argument

Chapters 4 and 10 argue that (6) constitutes a case of intervention (minimality). Since switching the surface order of the direct object and of the indirect object does not affect honorification marking (see (8)), Chapters 4 and 10 claim that it must be the case that honorific agreement takes place prior to word order permutation. (8)

*Hanako-ga Tanaka Sensei-o Mary-ni go-syookai-si-ta. Hanako-nom Prof. Tanaka-acc Mary-dat introduce-oh-past ‘Hanako introduced Prof. Tanaka to Mary.’

I take this to mean that agreement takes place under Chomsky’s derivational version of Agree, which applies as soon as the Probe (in our case, v) is introduced into the derivation. To capture the relevant intervention effect, I take the dative element to c-command the accusative element upon First Merge. If the reverse were a possible base order, the accusative element would have a chance of being closer to the functional head triggering agreement (say, v), and there would be no defective intervention. The relevant structure is provided in (9). (9)

[vP v [VP IO [Vb DO V]]] z____=______m Agree

The authors of the work just discussed leave open the question of Case assignment in ditransitives. What their analyses show is that the direct object is inaccessible to v if a dative element is present as far as K-feature checking is concerned. But the direct object gets accusative (by assumption, structural) Case,4 which must come from v (Chomsky 1995).5 The paradox, then, is that the direct object is both accessible and inaccessible to v

108

Aspects of the Syntax of Agreement

for checking purposes. More precisely, it is accessible for Case, but not for K-feature checking. The problem is far from trivial, if we take seriously Chomsky’s (2000) claim that Case and K-features are two sides of the same coin. But the same state of affairs is arrived at on independent grounds by Nomura (2002a) and Hiraiwa (2005). Analyzing constructions from a wide range of languages, they show that Case is often assigned to several elements by one and the same Probe (a process they formalize in terms of Multiple Agree, on which see next section), but K-feature checking is always asymmetric (it always takes place with the closest element). The situation is very reminiscent of the Bulgarian pattern of multiple whfronting. Assuming that overt displacement is a matter of EPP-satisfaction (Chomsky 2001, 2004), one can characterize Bulgarian Co as a probe that Agrees multiply for [Focus] and just once for [wh]. At least two questions arise at this stage. First, how can an element both asymmetrically and symmetrically probe within the same domain? Second, why is it [Focus]/[Case] that can be probed multiply, and [wh]/[K] once, and not the other way round? The question is especially pressing, since [Case]/ [K] and [focus]/[wh] are taken to be reflexes of the same feature. This chapter aims to answer those questions.

6.3. TOWARD A SOLUTION Let me start by saying what I take Chomsky’s statement to mean in saying that [Case] and [K] (and similarly for [focus]/[wh]) are two sides of the same coin. First, I do not assume that they are the same feature, but instead entertain the possibility that they are features that have the same function. In particular, I take K-features to be interpretable on the Goal (NP) but uninterpretable on the Probe (T/v). The reverse situation obtains for Case: It is an interpretable feature on the Probe (corresponding to the Tense/Aspect feature), but uninterpretable on the Goal (NP).6 I assume that the presence of an uninterpretable feature on an element is forced to ensure ‘syntactibility’ (which expresses the same intuition as Chomsky’s 2000 notion of ‘visibility’): to ensure that the element can be combined (i.e., can participate in concatenation). I further assume, following Chomsky (2000, 2001a, 2004), that the Agree relation is a composite operation consisting of a Matching procedure and a Valuation (agreement proper) procedure.7 Match is, by definition, symmetric: B and C match. By contrast, and quite crucially, Value is asymmetric: B values C. At first, it appears that the preceding paragraph offers a solution to our symmetry/asymmetry issue. All we have to do is pair the symmetric feature checking ([Case]/[focus]) with Match and the asymmetric feature checking ([K]/[wh]) with Value. Although I will ultimately adopt this conclusion, it

Multiplicity: Symmetries and Asymmetries

109

cannot be as straightforward, for plainly [Case]/[focus] is valued on the Goal. Why isn’t valuation asymmetric in this case? I’d like to propose that the answer to this question lies in the fact that in the case of symmetric feature checking, valuation proceeds from the Probe down to the Goal. I propose that the valuation (asymmetry) here piggybacks on (asymmetric) c-command (understood in terms of derivational Merge, in the spirit of Epstein 1999). By contrast, asymmetric valuation for [K]/[wh] cannot piggyback on asymmetric c-command. Instead it capitalizes on another aspect of derivational order: closest/most recently merged element. The minimality effect follows. Summing up this section, asymmetric/symmetric checking is not due to an intrinsic difference between [Case]/[focus] and [K]/[wh], but instead is a result of where they are placed.8, 9 The interpretable feature located on the Probe can agree multiply, as the asymmetry is located elsewhere (the Probe c-commands the Goal). By contrast, the uninterpretable feature on the Probe can only be valued once, as the asymmetry arises via Closest (matching element). There is one more issue we have to touch on. As stated earlier, I follow Chomsky in taking Value to depend on Match. As Youngmi Jeong (p.c.) observes, two situations could be envisaged here: The Probe matches (and enters into valuation with) the first Goal, and then matches the second Goal (without valuation from the Goal), as shown in (10). (10) Probe . . . Goal1 . . . # 1(Match) _" " 2(Value) _# # 3(Match)

Goal2 "

# _#

Or else, the Probe matches both Goals as part of the same (simultaneous) operation. Then, valuation takes place (from the closest Goal), as schematized in (11). (11) Probe . . . Goal1 . . . # 1(Match) _" " 2(Value) _#

Goal2 _"

(11) appears to be the only tenable option. If Valuation took place right after the first Matching element is found, the Probe would cease to be a Probe (its uninterpretable feature would be valued, hence invisible/inactive), and no subsequent matching/detection would be necessary (hence, by economy, should be ruled out). Returning now to the process of multiple wh-movement proper, we have to add that, from the preceding discussion, multiplicity as a result of Match cannot be parametrized (contra Fukui 1999): All languages match multiply (down to the next Probe).10 What is parametrized is whether languages

110

Aspects of the Syntax of Agreement

associate that matching with overt displacement.11 When they do, the EPP requirement is tied to the matching interpretable feature on the Probe. This converges with Boeckx’s (2001, 2003a, 2003b) claim that Move may take place under Match, independently of Agree. In fact, Serbo-Croatian matrix questions with no superiority effect (see (2)) illustrate this perfectly: No valuation from the Goal takes place, but Match and Move happen.

6.4. CONCLUSION In this study I have argued that an investigation of the minutiae of Agree can lead to a deeper understanding of symmetries and asymmetries in instances of multiple agree such as multiple wh-fronting, without abandoning important guiding ideas of current syntactic theorizing such as local economy, intervention effects, parallelism among features, etc. Crucial was the distinction between Match and Agree, and the idea that the computational system is derivational (which allows for simultaneous operations, operations taking precedence over others, etc.).

III

Agreement and Locality

7

Agree Versus Attract A Relativized Minimality Solution to a Proper Binding Problem

I examine an extraction paradigm first discussed by Kroch and Joshi (1985), which Lasnik (2002) takes to provide an argument for feature movement. I show that Lasnik’s solution is problematic on both conceptual and empirical grounds. I offer an alternative approach that in contrast to previous solutions does not rely on the Proper Binding Condition (whose status in minimalism is far from clear), but instead deeply implicates Relativized Minimality. Although I take issue with Lasnik’s analysis, I agree with him that the Kroch–Joshi paradigm bears on the nature of feature checking, and agreement. In particular, I show that an Agree-based account handles the facts straightforwardly.

7.1. INTRODUCTION Considerable insight has been gained in recent years into the nature of word order by focusing on possibilities of remnant movement. Abstractly, remnant movement takes the form in (1). An element B is moved out of an element C, which is subsequently moved to a position higher than (and featurally distinct from) B’s derived position. (1)

[[C . . . tB . . .] . . . [ . . . B . . . [ . . . tC . . .]]]

The process is illustrated in (2) (from German). (2)

[Y ti gelesen]j hat [X das Buch]i keiner tj read has the book no one ‘No one has read the book.’

Because of its two-step character, remnant movement has come to be regarded as a powerful argument for a derivational approach to syntactic computation (see Müller 1998; but see Brody and Szabolcsi 2003). One of the most fascinating aspects of remnant movement is the problem it poses

114

Aspects of the Syntax of Agreement

for the Prober Binding Condition (PBC), which demands the traces be bound (hence c-commanded) by their antecedents. (3)

Proper Binding Condition Traces must be bound

(Fiengo 1977)

The influence of remnant movement analyses in recent years has grown steady. For instance, projects are being developed to show that most (perhaps all) instances of head-movement can, and should, be reanalyzed in terms of remnant movement (see Koopman and Szabolcsi 2000 and Mahajan 2000, among others). Kayne (1998, 2001) has even suggested that traditional instances of covert processes such as scope reversal via QR or reconstruction be captured by making massive use of remnant movement, with no need for a distinct LF component. Remnant movement raises many intriguing properties that I will not discuss here (see Müller 1998 for what is to date the most detailed study of remnant movement).1 Rather, I will focus on a fairly narrow issue that arises in the remnant movement approach, and suggest that once properly analyzed the problem is only apparent. In addition, our investigation will enable us to draw some important conclusions about the structure of the grammar—in particular, about the status of a distinct LF component.

7.2. HOW LIKELY Consider the contrast in (4)–(5), originally reported in Kroch and Joshi (1985). (4)

a. John is likely to win. b. There is likely to be a riot.

(5)

a. How likely to win is John. b. *How likely to be a riot is there.

(5a) appears to be a paradigm case of remnant movement. In a first step, John is raised from its VP-internal position to its surface position (SpecIP). Next, the non-finite predicate how likely . . ., containing the trace of John, is raised to SpecCP. This is schematized in (6). (6)

(=5a) [CP [how likely [ti to win]]j [Cb isk [IP Johni tk [VP tj]]]]

Surprisingly, an example like (4b), which is minimally different from (4a), fails to yield a grammatical output if the how likely . . . chunk is raised, as in (5b) (compare (6) and (7)).

Agree Versus Attract (7)

115

(=5b) [CP [how likely [ti to be a riot]]j [Cb isk [IP therei tk [VP tj]]]]

The crucial difference between (4a) and (4b) is that John raises to matrix SpecIP in (4a), but not in (4b). Instead in (4b) an expletive is inserted in the embedded SpecIP (due to the preference of Merge-over-Move, see Chomsky 1995: 348; 2000: 111), and undergoes movement to SpecIP. (I return to the status of expletive raising later. Note that if expletives are taken to undergo predicate movement, as in Moro 1997 and related work, (5b) is equally puzzling.) The contrast in (5) is puzzling in more than one respect. Müller (1998: 7 fn. 10) mentions the paradigm in (5) and suggests we treat (5b) as an unexplained exception. It is indeed difficult to see what grammatical property would exclude (5b) while ruling in (5a). Further, it has been suggested in Vukić (1998) and Bošković (2002a) that expletives are merged in their surface position without undergoing any movement. (Both Vukić and Bošković treat expletives as grammatical formatives that are merged as late as possible. See Bošković for evidence that there is no EPP-checking in the infinitival complement of raising predicates.) If correct, the ‘late-insertion’ view of expletives replaces (7) with (8). (8)

[CP [how likely [to be a riot]]j [Cb isk [IP there tk [VP tj]]]]

That (8) yields an ungrammatical output, and (6) doesn’t, is clearly unexpected. If anything, one would expect the reverse pattern of grammaticality, as (6) appears to violate the PBC, whereas (8) does not (the raised predicate does not contain any trace at all). It is this puzzle that I will concentrate on in this chapter. On grounds that I have discussed elsewhere (see Boeckx 2001), I will assume that the derivation of (5b) given in (8) is the correct one (i.e., expletives are merged in their surface positions). The thesis I will entertain here is that the contrast in (5) has nothing to do with remnant movement or the Proper Binding Condition. Whatever their ultimate status, these will remain unaffected by (5), for reasons to be developed shortly. I will also argue, contra Lasnik (2002), that the contrast in (5) fails to provide an argument for the existence of feature movement (Chomsky 1995) and for a distinct LF component.

7.3. MORE FACTS Before offering a solution to the contrast in (5), let me expand the data base by showing that the contrast in (4)–(5) is not limited to existential constructions. An effect similar to (5b) is found with idiom chunks (9b).2 (9)

a. Advantage is likely to be taken of John. b. *How likely to be taken of John is advantage?

116

Aspects of the Syntax of Agreement

Descriptively speaking, raising of portion of an idiom is unproblematic (9a), unless the movement is followed by predicate raising (9b). An additional puzzle of the how likely paradigm is that the contrasts observed so far disappear if only portion of the predicate is raised, as in (10). (10) a. How likely is John to win? b. How likely is there to be a riot? c. How likely is advantage to be taken of John? Relying on Kroch and Joshi’s original observations, Lasnik and Saito (1992: 141) take the how likely paradigm to require an explanation in terms of the PBC. According to them, (5b) and (9b) are out because the trace (of there and of advantage, respectively) fails to be properly bound after predicate raising. (Note that when only part of the predicate raises, as in (10), all traces are bound, as the portion of the predicate containing the trace remains in situ.) According to Lasnik and Saito, what saves (5a) is the existence of an alternative derivation that does not contain an unbound trace. Lasnik and Saito appeal to a long-standing view that modal predicates such as likely are ambiguous between raising and control predicates. Thus, (4a) may be represented as (11) or (12). (11) John is likely [t to win]. (12) John is likely [PRO to win]. If the derivation in (11) is chosen, (5b) will violate the PBC, as shown in (13). (13) [CP [how likely [*ti to win]]j [Cb isk [IP Johni tk [VP tj]]]] If, however, the control derivation is chosen, no trace will be contained in the raised predicate, and the sentence will be grammatical. (14) [CP [how likely [PRO to win]]j [Cb isk [IP John tk [VP tj]]]] Lasnik and Saito note that since expletives and idiom chunks cannot control PRO, a derivation like (14) is unavailable to them. (15) a. *there tried [PRO to be a riot] b. *advantage wants [PRO to be taken of John] (16) a. *[how likely PRO to be a riot] is there b. *[how likely PRO to be taken of John] is advantage By capitalizing on the PBC and on a raising/control ambiguity for predicates like likely, Lasnik and Saito account for the whole how likely paradigm.

Agree Versus Attract

117

There are, however, several problems with Lasnik and Saito’s solution. The first I want to mention is the raising/control distinction of modal predicates. As already said, it is standard to capture the root versus epistemic modal distinction via a control versus raising analysis. The main motivation for such an analysis is that root modals (but not epistemic modals) appears to assign a theta-role to the subject. Assuming some version of the thetacriterion, root modals then have to project their own subjects. Aside from the status of the theta-criterion in minimalist models of grammar, I note that the claim that root modals assign a theta-role to the subject has not gone unchallenged (see already Newmeyer 1975). Further, if one follows Kratzer (1991) in taking the root/epistemic distinction not to be the result of a lexical ambiguity, but of different modal bases that are provided by different conversational backgrounds, different syntactic (/thematic) structures are a priori not necessary to capture the distinction. In addition, Wurmbrand (1998) (see also Bobaljik and Wurmbrand 2000) and Barbiers (1995) provide detailed arguments based on unrelated phenomena (restructuring and ellipsis, respectively) against a control/raising analysis of the root/epistemic modal distinction.3 In addition, if control reduces to raising, as Hornstein (1999) has argued, Lasnik and Saito’s PBC account of the contrasts above cannot be correct. Perhaps the strongest argument against a PBC account of the how likely paradigm comes from the ungrammaticality of (17) (pointed out to me by Koji Sugisaki, p.c.; see also Nomura 2001 and Abels 2003).4 (17) *Who said that there was how likely to be a riot? A similar fact obtains with idiom chunks. (18) *Who said that advantage was how likely to be taken of John? Note, in contrast, the grammaticality of (19).5 (19) Who said that John was how likely to win? The three sentences just given mirror the contrast between (5a) and (5b)– (9b). The importance of the present cases is that they do not involve the remnant movement part (predicate raising). Hence they cannot be ruled out via the PBC.

7.4. A MOVE-F ACCOUNT Recently, Lasnik (2002) has revisited the how likely paradigm, and argued that it provides an argument for Chomsky’s (1995) treatment of covert movement as feature movement. To understand the argument, it is useful

118

Aspects of the Syntax of Agreement

to retrace certain developments in the minimalist program concerning the nature of covert movement and the timing of operations.

7.4.1. Theories of Movement Although the first Minimalist paper (Chomsky 1993) had taken over the socalled Y-model (Chomsky and Lasnik 1977) to some extent (compare (20) and its minimalist variant (21)), subsequent research within the Minimalist Program (henceforth MP) has led to interesting modifications of the central architecture of the language faculty. (20) D(eep)-Structure | S(urface)-Structure 3 LF PF

(21) Initial Array | Spell-Out 3 LF PF

Chomsky (1995) still adopts the ‘temporal’ asymmetry between overt and covert movement (i.e., pre- vs. post-Spell-Out operations), much like in Chomsky (1993), but suggests that we view ‘covert operations’ as consisting not of movement of categories that happen to receive no pronunciation (as was the case in 1993, and in work within the GB-framework), but rather of movement of formal features. For Chomsky (1995), “the operation Move (. . .) seeks to raise just F[eature]” (p. 262). Chomsky’s reasoning is that movement is triggered to check features. We therefore expect under Minimalist assumptions that if the computational component can raise just what is needed (features to carry out the checking operation), it will do so. The question now arises as to why sometimes whole categories, and not just formal features, move. Chomsky’s answer (in 1995) is that overt movement is to be decomposed in the following way. Applied to the feature F, the operation Move creates at least one, perhaps two “derivative chains” alongside the chain CHF = (F, tF) constructed by the operation itself. One is CHFF = (FF[F], tFF[F]), consisting of the set of formal features FF[F] and its trace; the other is CHCAT = (B, t), B a category carried along by generalized pied-piping. (Chomsky 1995: 265) Chomsky assumes that any overt operation is the result of moving features first (so far, overt and covert movements are indistinguishable, except perhaps in terms of timing, a point I will discuss extensively later), and then an operation of pied-piping which carries along the ‘remnants’ of the item from which the features have been moved. Chomsky assumes that

Agree Versus Attract

119

For the most part—perhaps completely—it is properties of the phonological component that require pied-piping. Isolated features and other scattered parts of words may not be subject to its rules, in which case the derivation is canceled; or the derivation might proceed to PF with elements that are ‘unpronounceable,’ violating F[ull] I[nterpretation]. (Chomsky 1995: 266) As emphasized in the first quote, within a Move-F framework, overt and covert operations are indistinguishable up to a certain point (formation of a derivative chain/pied-piping). This leads Chomsky to claim that “such considerations could permit raising without pied-piping even overtly, depending on morphological structure” (p. 266). What Chomsky means by ‘overtly’ here is ‘the overt (pre-Spell-Out) component’ in (21). Before Move-F, it was assumed without discussion that covert operations took place after Spell-Out. Given Move-F, it is now possible to view covert movement as the ‘first’ part of overt movement, that is, as an operation that does not require any distinct component. This is indeed the conclusion that Chomsky (2000, 2001, 2004) embraces: “There is a single cycle; all operations are cyclic. Within narrow syntax, operations that have or lack phonetic effects are interspersed. There is no distinct LF component within narrow syntax” (Chomsky 2000: 131). Other proponents of the so-called single output syntax (ignoring important differences among them) are Bobaljik (1995, 2002) Groat and O’Neil (1996), Pesetsky (2000), and, from a representational perspective, Brody (1995, 1997). (For a very different implementation, see Kayne 1998.) Proponents of the SOS make use of an insight expressed in Chomsky (1993), reviving the analysis of movement in Chomsky (1955), that traces are copies of the moved elements. The important idea is that chains (the objects of syntactic computation) consist of sequences of copies of a given element, but that at the interfaces (LF and PF) only one position in a given chain (link) is (typically) privileged or ‘interpreted.’ (At least in the general case, see Nunes 1999, 2004, and Bošković 2000b, 2001b, for valuable discussion and necessary refinements). Departing from Chomsky (1993, 1995), SOS defenders argue that not only may LF privilege either the higher or the lower copy (see Chomsky’s 1993 discussion of A-bar movement reconstruction in such terms), but that PF also may choose which copy to privilege (i.e., pronounce). On this view, ‘overt’ and ‘covert’ movements are distinguished not by temporal ordering in the derivation, but rather by the choice of which copy to pronounce and which copy to delete.6 Or, put differently, pronunciation of the highest copy corresponds to ‘overt’ movement, and pronunciation of the lowest copy corresponds to ‘covert’ movement. The major point of departure for Chomsky from other Single-Output syntax models like Groat and O’Neil (1996) and Bobaljik (1995, 2002) is in its acceptance of a covert operation that is distinct from ‘covert’ (i.e., unpronounced) phrasal movement (either feature movement or Agree). (Kayne

120

Aspects of the Syntax of Agreement

1998: 172 remains somewhat agnostic; Brody 1997 and Pesetsky 2000 assume some version of Agree.) The need for such an operation is best illustrated on the basis of existential sentences.7 Originally, Chomsky (1986) proposed that the associate-indefinite NP in such cases replaces the expletive there in the covert component, as illustrated in (22). (As far as I can see, unless some extra stipulation is built in, such as Bobaljik’s 2002 Minimize PF-LF mismatch, Groat and O’Neil 1996 and Bobaljik 1995, 2002 make the same prediction as Chomsky 1986.) (22) a. There is a man in the garden b. A man is [t in the garden]

S(urface)-Structure LF-expletive replacement

This analysis was criticized as soon as it was proposed (apparently, first, by Lori Davis; Howard Lasnik p.c.): The expletive replacement analysis gets the scope facts wrong. As is well known, indefinites in subject positions are scopally ambiguous (see (23)). (22b) predicts that such ambiguity exists in existential constructions. But this is not the case. The associate in (24) only has a low reading. (23) Someone from New York is likely to win the lottery. (someone >> likely/likely >> someone) (24) There is likely to be someone here. (*someone >> likely/likely >> someone) Chomsky (1991) puts forward a new analysis of existential constructions. He suggests that at LF the associate does not literally replace the expletive but adjoins to it, as shown in (25). (25) [a man [there]] is [t in the garden] There are many problems with this analysis, and I won’t review them here. They are thoroughly discussed in Lasnik (1992). Chomsky (1995) proposes a much more satisfactory account. Chomsky’s reasoning is that movement is triggered to check features. We therefore expect under Minimalist assumptions that demand minimization wherever possible that if the computational component can raise just what is needed (features to carry out the checking operation), it will do so (recall “the operation Move (. . .) seeks to raise just F[eature]” (Chomsky 1995: 262)). Thus, Chomsky argues for the existence of feature movement (Move-F). Relying on the Move-F hypothesis, Chomsky proposes that in existential constructions only formal (K-) features of the associate NP move (headadjoin) to Inflo, leaving all phonological and semantic features behind. Raising of K-features immediately accounts for the fact that finite agreement

Agree Versus Attract

121

in existential constructions is controlled by the feature specification of the associate, as illustrated in (26). (I here set aside semi-formulaic examples like there’s two men in the garden.) (26) a. There is/*are a man in the garden. b. There *is/are two men in the garden As Lasnik has extensively discussed (see the essays in Lasnik 1999a), the Move-F account provides a straightforward explanation for the narrow scope of the associate NP in these constructions if we assume, quite plausibly, that the establishment of scopal relations is more than a matter of formal features, and requires phrasal displacement (see Pesetsky 2000: 2–5 for some discussion).8 On largely conceptual grounds, Chomsky (2000: 123) dispenses with feature movement altogether and captures its effects via the operation Agree. The latter amounts to a process of feature checking (in his terms, valuation) at a distance. The summary of minimalist views on covert movement given here, especially the covert process that accounts for agreement in existential sentences will now enable us to examine Lasnik’s (2002) argument in favor of feature movement based on the how likely paradigm discussed in previous sections.

7.4.2. The Proper Binding Residue Lasnik (2002) uses the preceding paradigm in support of the Move-F hypothesis. His argument runs as follows. Barss (1986) rules out (5b) by capitalizing on Chomsky’s (1986) expletive-replacement analysis of existential constructions. Barss notes that if a riot must replace there at LF, the movement will be illicit because it is sideward. Lasnik agrees that the expletive replacement account cannot be correct, but he claims that Barss’s analysis can be maintained under the feature movement analysis, crucially not under an Agree analysis. Assume that the K-features of a riot are attracted in (5b); the element becomes PF-deficient (see Chomsky’s 1995 quote given earlier; see also Ochi 1999a, 1999b, Lasnik 1999b, and Uriagereka 1999a). How likely fronting removes the category from the c-command domain of the moved features, the necessary repair strategy cannot be carried out, and the derivation crashes due to the presence of scattered features at the interfaces. Lasnik claims that an Agree account cannot capture the fact that in (5b), as in the absence of (feature) movement, there is no feature scattering to start with, hence no requirement for the associate NP to remain within the c-command domain of the expletive. Lasnik’s analysis is appealing for the level of subtlety it reaches. In contrast to many other studies (see, e.g., Wurmbrand 2001b), it is not concerned with whether or not some non-phrasal

122

Aspects of the Syntax of Agreement

‘covert’ process exists, but with the more difficult question of which form that process takes. However, Lasnik’s account faces many problems. First, by adopting a feature movement analysis, Lasnik inherits the conceptual difficulties that led Chomsky to reject feature chains in favor of Agree. In particular, it is not clear what feature scattering (crucial for Lasnik) means under the copy theory of movement. Movement of the feature will leave a copy behind, rendering the need for repair obscure. Further, if feature movement chains reduce to head-chains (Chomsky 1995, Bošković 1998), they inherit the problems associated with the latter (see Chomsky 2000, 2001; Brody 2000; Boeckx and Stjepanović 2001, and Mahajan 2000). Second, in order to account for why feature movement is not accompanied by repair in standard existential sentences, Lasnik has to assume that that instance of feature movement takes place after Spell-Out, in a separate LF component (where feature scattering does not cause any crash). He is thus forced to a return to the Y-model. Third, Lasnik has to postulate that feature movement out of the copy left by remnant movement is impossible (contra Bošković 1997 and Nishioka 1997). If it were, the remnant movement case (5b) would be virtually identical to (4b), as illustrated in (27). (27) [how likely to be a riot] [is [there [] | F-movement | Fourth, Lasnik’s solution says nothing about the badness of (17), as it crucially relies on remnant movement to exclude (5b). Fifth, it is not clear how the feature movement account of (5b) extends to (9b) (let alone (18)). Unlike the expletive–associate relation, the raising of an idiom chunk to the idiom remnant has never been treated in terms of feature movement, as far as I know. On the basis of the problems it faces, I think it is fair to say that Lasnik’s analysis is inadequate.

7.5. RELATIVIZED MINIMALITY We have seen that neither a raising versus control/PBC account nor a feature movement account adequately captures the how likely paradigm. In this section I propose a novel way of looking at the facts that not only accounts for the whole range of data, but also allows us to preserve the arguably more elegant single output model of syntax, and does not jeopardize any conclusions about remnant movement reached by previous studies. The format of the solution I would like to argue for is well known. It essentially amounts to a Relativized Minimality violation (Rizzi 1990). An element B enters into a relation with an element C if there is no H that meets

Agree Versus Attract

123

the requirement(s) of B (i.e., that matches B), and H either c-commands C. The illicit situation is schematized in (28). (28) [B . . . [ . . . H . . .[ . . . C . . .]]] (H c-commands C) In the following I will adopt Starke’s (2001) conception of Relativized Minimality, as it leads to what I think is a clearer solution. The portion of Starke’s view on chains that will be relevant for us is roughly as follows. (29) a. B . . . B . . . B b. BC . . . B . . . BC In (29a) we have three elements of the same type (B). Attempting to relate the first and the third (in linear order) leads to a violation of Relativized Minimality. The situation in (29b) is more complex. The intervening element (second element in linear order) is of type (B). The first and the third elements that the grammar is trying to relate both contain an B feature. In addition, they contain a C feature which is missing from the intervener. Starke’s point is that if the first and the third element are B-related, the situation that obtains is equivalent to that in (29a), and is thus ruled out by Relativized Minimality. If, however, the first and the third elements are C-related, no intervention effect emerges, as the potential intervener is not “of the same type.” A concrete case of (29a) is a superiority condition of the type we found in (30)–(31). (30) *whati did who buy ti (31) [C [ who T [buy what]]] B B B +wh +wh +wh Starke claims that the situation in (29b) corresponds to a weak island. As is well known from the work of Cinque (1990), Rizzi (1990), and Szabolcsi and Zwarts (1993), weak islands such as wh-islands are ‘selective’ islands, in the sense that certain elements associated with well-defined readings can extract, while others, which lack these readings, can’t.9 Classic cases are the theta-/non-theta related contrast (32), and the D-linked/non-D-linked contrast (33).10 (Judgments are contrastive, rather than absolute.) (32) a.

*How many pounds do you wonder whether he weighed t? ANSWER 1 [non-referential/non-theta reading]: he weighed 100 lbs (he was skinny) b. How many pounds do you wonder whether he weighed t? ANSWER 2 [referential/theta reading]: he weighs 100 lbs (by lifting the package)

124

Aspects of the Syntax of Agreement (33) a. *What do you wonder whether Mary read? b. Which book do you wonder whether Mary read?

Starke’s view is that the good instances of extraction (32b, 33b) correspond to the situation in (29b) when the moving element C-relates to its final landing site. The bad instances of extraction correspond to an B-relation. In what follows, I argue that the B/C-relation in (29b) plays a role in the how likely paradigm. The reader should understand that I will not try to rule out the bad cases by appealing to some inadequacy of the remnant movement step fronting how likely . . . . The ungrammaticality of (17)–(18) suffices, in my view, to show that the badness of (5b)–(9b) is independent of remnant movement. In other words, they should be ruled out prior to the application of remnant movement. Put differently, in order to derive the contrast between (4b)/(9a) and (5b)/(9b), we must find a difference between the following stages of the derivation: (34) a. is [likely to be a riot] b. is [likely to be advantage taken of John] (35) a. is [how likely to be a riot] b. is [how likely to be taken advantage of John] The difference cannot affect the good cases (4a)/(5a): (36) is [likely John to win] (37) is [how likely John to win] With Chomsky (2000) I assume that in existential constructions, Inflo and the associate stand in a checking relationship by Agree, and that feature matching takes place as the derivation unfolds, not in a distinct LF-component. It will therefore be crucial to examine the various cases step by step. The good cases (4a, 4b, 9a) are derived straightforwardly, as in (38)–(39). (38) [T [likely [. . . NP . . .]]] | | (39) a. [T [likely [John to win]]] Agree (T,John) + Move (John) b. [T [likely [to be a riot]] Agree (T,[a riot]) + Merge there m_: there-insertion c. [T [likely [to be taken advantage of John]]] Agree (T,advantage) + Move (advantage) The proposal I would like to make to rule out the cases in (5b, 9b) is that the presence of how in how likely . . . blocks the Agree relation between T and some NPs.

Agree Versus Attract

125

It has often been argued that that wh-words are decomposable into a whpart and an indefinite part (an idea going back to Chomsky 1964 and Katz and Postal 1964). Suppose that how in how likely actually consists of a whpart and an indefinite part, roughly as ‘wh-indefinite (degree).’11 I would like to argue that the indefinite part of how creates an intervention/Relativized Minimality effect,12 as schematized in (40). (40) [T [how likely [. . . NP . . .]]] [+NP] [+WH,+NP] [+NP] z_____________=_______m It may be objected that the indefinite part of the wh-phrase does not c-command the associate, hence should not be a blocker for Agree (recall (28)). However, there are various ways around this well-known ‘almost c-command’ problem. For concreteness, I will assume that c-command out of the specifier of an XP is possible (Kayne 1994). (Note that some features of how must ‘head’ the whole phrase to trigger pied-piping under wh-movement.) Although (40) rules out the crucial Agree relation in (39b,c), it appears to do so in (39a) as well, predicting (5a) to have the same status as (5b)–(9b), contrary to fact. However, here Starke’s characterization of weak islands (29b) comes handy. Recall that blocking is obviated if there is an ‘alternative’ Agree-relation involving a feature that is absent from the blocker. I will argue that the [+NP] corresponds to B in (29b). The idea now being that there is another feature that is present and can partake in Agree in (39a), but not in (39b,c). The feature that I will make use of is [+D]. It is often assumed that definite noun phrases are DPs, while indefinites are NPs (see, e.g., Chomsky 1995: 342, 350). Further, indefinites are ambiguous between a DP reading (e.g., specific indefinites) and an NP reading. The ambiguity may account for the two readings in (41). (41) Someone is likely to win the lottery. (someone >> likely/likely >> someone) As is well known, the ambiguity in (41) is missing in existential sentences. (42) There is likely to be a lottery winner. (*a winner >> likely/likely >> a winner) Let us take the absence of the wide scope reading in (42) to mean that the indefinite NP in existential constructions is a pure NP (it lacks the DP reading). (This restriction may underlie the well-known definiteness effect.) Now let us go back to the examples in (39) and the intervention effect in (40). The presence of the indefinite feature on how blocks the Agree relation that relates T and the noun phrase in the infinitive complement. This is the B-relation in (29b). I propose that Agree can succeed if a D-feature is

126

Aspects of the Syntax of Agreement

involved. This would correspond to the C-relation in (29b). The D-feature is absent from the intervener how. Having established that the indefinite noun phrase in existential construction is a pure NP, the C-relation (Agree [+D]) cannot be established in (39b). Likewise, the saving C-relation cannot be established in (39c). As is well known, portions of idioms are non-referential noun phrases, and thus quite plausibly lack the DP-reading. Summarizing the discussion so far, (39b,c) are excluded due to intervention as in (43). (43) [T [how likely [. . . NP . . .]]] [+NP] [+WH,+NP] [+NP] | * | B B
B (39a) is rescue by the presence of a D-feature on the noun phrase being attracted. (44) [T [how likely [. . . DP . . .]]] [+DP] [+WH,+NP] [+DP] | | C B C An interesting, and correct, prediction of the present analysis is that in case an indefinite is used, and attracted in the how likely frame, it cannot have an indefinite, non-referential reading (that reading would match the indefinite part of how, and the element could not be attracted); it must have a specific, (quasi-)referential reading. As (45) shows, the prediction is borne out. (45) [how likely to win the lottery] is someone from New York (someone >> likely/*likely >> someone) So far our proposal is able to capture the basic how likely paradigm without any appeal to remnant movement, distinct LF-component, or move-F, which I take to be desirable. What remains to be explained is the improvement in (10b,c) (repeated). (46) a. How likely is there to be a riot? b. How likely is advantage to be taken of John? In an earlier version of the present work (Boeckx 1999a), I claimed that extraposition took place in (46) (more precisely, I adopted Larson’s 1988 treatment of extraposition as resulting from Light-Predicate Raising; see already Fiengo 1977; see also Kayne 1994, and, for a precise formulation of the Light-Predicate Raising rule, Runner 1995). The derivation I assumed is given in (47).

Agree Versus Attract (47) a.

[how [likely [to be a riot]]]

b. [to be a riot]i [how [likely [ti]]] c. is [to be a riot]i [how [likely [ti]]] d. [how [likely [ti]]]i is there [to be a riot]i tj

127

n predicate-raising/ ‘extraposition’ n Agree between Infl0 and a riot Remnant movement

The crucial step was (47b), which brings the NP past the intervener, allowing the establishment of the B-relation (Agree [+NP]), thus avoiding the fate of (39b,c). However, as pointed out in Nomura (2001) there are at least two problems with the extraposition analysis (problem #1 was also brought to my attention by Klaus Abels p.c.). First, what prevents extraposition from taking place in (17), repeated here as (48)? (48) *Who said that there was how likely to be a riot? Indeed, pending a more precise formulation of extraposition in (47), nothing seems to block a derivation like (49). (Irrelevant stages omitted.) (49) a.

[how [likely [to be a riot]]]

b. [to be a riot]i [how [likely [ti]]] c.

n predicate raising/ ‘extraposition’ n remnant movement Agree between Inflo and a riot n

is [[how [likely [ti]]]j[to be a riot]i [tj]] d. there is [[how [likely [ti]]]j [to be a riot]i [tj]] e. who said that there is [[how [likely [ti]]]j[to be a riot]i [tj]]

The problematic step here is (49c), where an Agree/B-relation can be established. By remnant-moving how likely . . . in (49b), the (almost) c-command relation between how and the lower NP (a riot) is broken, preventing intervention from taking place. (49) thus predicts (17) to be grammatical (a similar problem arises for (18), which I won’t illustrate here). A second problem for the extraposition approach is that the extraposition step in (47a) patterns unlike familiar instances of extraposition (see Nomura 2001). (The contrasts are subtle, but nonetheless significant.) As shown in (51), wh-movement out of an extraposed infinitival is degraded. (50) Whati did Bill ask Mary [to fix ti] yesterday? (51) ??Whati did Bill ask Mary yesterday [to fix ti]?

128

Aspects of the Syntax of Agreement

If wh-movement out of an extraposed clause crosses a wh-island, the result is ungrammatical (more so than if extraposition does not take place, (53)). (52) *Whati did you wonder [how nicely Bill asked Mary yesterday [to fix ti]]? (53) ?*Whati did you wonder [how nicely Bill asked Mary [to fix ti] yesterday]? Crucially, the deviance in (52) is not replicated to the same degree in parallel cases involving how likely. (54) ??Whati do you wonder [how likely Bill is [to fix ti]]? (55) ???Whati do you wonder [how likely Bill is [to ask Mary to fix ti]]? As I see no straightforward solution to the problems raised for the extraposition account by Nomura (2001), I reject the solution to (10b,c) offered in Boeckx (1999a), and turn to an alternative. The proposal I would like to make is based on an intuition going back to Rosenbaum (1967: 108 fn.1).13 The idea is that what is traditionally referred to as raising adjectives (likely, certain) are in fact “peculiar adverbs” (Rosenbaum’s term). Treating raising adjectives as “adjuncts” opens up a new possibility of dealing with (10b,c). It has become popular since Lebeaux (1988) to view adjuncts as being inserted acyclically (in contrast to complements, which conform to a strict view of the cycle, along the lines of Chomsky’s 1993 Extension Condition). (See, e.g., Chomsky 1993 and much subsequent work.) Suppose then that there are two ways of inserting how likely: cyclically, in which case it takes the to-infinitive as its complement, or ‘acyclically;’ separately from its alleged complement (to-infinitive). If the latter option is chosen, a sentence like (56) is derived as in (57). (56) How likely is John to win? (57) a. is [John to win] b. [Johni [is [ti to win]]]

n attraction of John n acyclic insertion of how likely14 c. [Johni [is [how likely] [ti to win]]] n wh-movement d. [[how likely]j isk [Johni [tk [tj] [ti to win]]]

Acyclic insertion essentially allows us treat how likely as a constituent, independent from the to-infinitive. This option therefore does not affect cases where wh-movement of how likely pied-pipes the infinitive. This is an important point because if acyclic insertion could be involved in ‘pied-

Agree Versus Attract

129

piping cases,’ nothing would prevent a derivation like (58), which is clearly unwanted, as it would incorrectly rule in cases like (5b).15 (58) a.

is [to be a riot]

n Agree (T,riot) (+ there insertion) b. [there [is [to be a riot]]] n acyclic insertion of how likely c. [there [is [how likely] [to be a riot]]] n wh-movement (+ pied-piping) d. [[how likely [to be a riot]]j isk [there [tk [tj]]]]

Acyclic insertion of how likely does not face the second problem raised by Nomura for the extraposition account. In the absence of extraposition, we do not expect (52) and (54) to pattern the same way. As for the first problem (how likely in situ, as in (17)), it also does not arise under acyclic insertion. If how likely is an adjunct, (17) reduces to (59). That is, whatever excludes in-situ wh-adjuncts in multiple questions will exclude the acyclic insertion option for a case like (17). The only option available will be the cyclic insertion, which gives rise to intervention, as demonstrated earlier. (59) *Who left why? The acyclic insertion analysis of how likely thus seems superior to the extraposition analysis in accounting for the grammaticality of (10b,c), without running afoul of (17)–(18).

7.6. CONCLUSION To conclude, I have examined a paradigm first discussed by Kroch and Joshi (1985), which Lasnik (2002) took as an argument for feature-movement. I have shown that Lasnik’s solution is problematic on several grounds. In particular, it fails to provide a solution for part of the paradigm (the idiom case) and moreover leads to an organization of the grammar that contains a separate LF component and feature chains. I have offered an alternative approach to the how likely paradigm that deeply implicates Relativized Minimality. Not only does the present analysis capture the full paradigm straightforwardly, it also need not assume the existence of feature-movement chains or of a distinct LF-component.16 In so doing, the present analysis lends credence to the conceptually more elegant mechanism of Agree and the One-cycle model of syntax. Finally, the account reconciles the how likely paradigm with independent conclusions about remnant movement, which previous analyses (Müller 1998) had failed to do.

8

Conflicting C-Command Requirements

Here I concentrate on the locality problem that arises in the context of raising verbs when the embedded subject moves over the experiencer. I review the major approaches put forth in the minimalist program, and highlight their shortcomings. I propose a solution that crucially relies on the timing of operations, and thus on a strictly derivational architecture of the grammar. This chapter highlights the interpretive limits of pure Agree-relations and the role of case in the characterization of interveners, and argues that effects reminiscent of the Person-Case-Constraint may obtain in the domain of experiencer constructions in English.

8.1. INTRODUCTION The chapter examines the structural status of the experiencer (to-phrase) in an English sentence like (1).1 (1)

Johni seems to Bill [ti to be a genius]

One problem raised by such experiencer phrases was recently brought to light in Chomsky 1995 (p. 305ff.) (The problem was already noted in the GB-era by, for instance, Lasnik and Saito 1992: 139). To allow raising of John, given Shortest Move/Closest Attract (2)–(3), which in essence captures Relativized Minimality effects, the experiencer cannot c-command (into) the embedded clause. (2)

(Shortest) Move B can raise to target K if there is no legitimate operation Move C targeting K, where C is closer to K (Chomsky (1995: 296 (82)))

(3)

(Closest) Attract K attracts F if F is the closest feature that can enter into a checking relation with a sublabel of K (Chomsky (1995: 297 (84)))

Conflicting C-Command Requirements



131

r

C is closer to [K] than B if C c-commands B and is not in the minimal domain of [(B,t)] (Chomsky (1995: 299 (87))) r

B

.BY B) is the smallest maximal projection including B b. The domain E([(B,t)]) of [(B,t)] is the set of categories included in Max(B) that are distinct from and do not contain B or t c. The minimal domain Min(E([(B,t)])) of [(B,t)] is the smallest subset K of E([(B,t)]) such that for any H Ž E([(B,t)]), some C Ž K reflexively dominates H (Chomsky (1995: 299 (86))

If the experiencer c-commands the embedded subject, it will be attracted, yielding the ungrammatical: (4)

*Billi seems to ti [John to be a genius].

At the same time, there is evidence that the experiencer c-commands (into) the embedded clause. In (5), him and John must be disjoint in reference (Condition C effect). (5)

Theyi seem to himj [ti to like Johnk/*j].

Assuming that there is an irreducibly geometric aspect (c-command) to binding (see Lasnik 1989 for compelling evidence), the disjoint reference effect in (5) can only be captured if him c-commands John. It thus seems that we have conflicting evidence as to the status of the experiencer. This is what is meant by ‘conflicting c-command requirements’ (a term borrowed from Kitahara 1997). Various solutions have been offered to solve the ‘experiencer paradox.’ They will be reviewed in the next section. Roughly, Kitahara (1997), Epstein et al. (1998), Ferguson (1994, 1996), and Boeckx (1998a) all try to find a way of having the experiencer c-command into the embedded clause after raising of the embedded subject. Without going into any detail at this stage, one can immediately note two major problems with such an approach. First, the very term ‘after’ entails the existence of two cycles, or countercyclic operations, which goes against much current research in the minimalist program (see the Single Cycle Hypothesis (Bobaljik 1995, Groat and O’Neil 1996, Pesetsky 1998, 2000, Chomsky 2000), and strictly derivational approaches (Epstein 1999, Epstein et al. 1998, Uriagereka 1998, 1999b)). Second, it is not sufficient to let the experiencer c-command into the embedded clause after subject raising, for there is evidence that the experiencer c-commands into the raising subject: (6)

[Pictures of himself]i seem to John [ti to be ugly].

132

Aspects of the Syntax of Agreement

If we assume, following Chomsky (1995), and the compelling arguments in Lasnik (1999c), that there is no reconstruction with A-movement, then binding of himself in (6) can only take place during the course of the derivation, prior to subject raising.2 One might try to get away with the (still poorly understood, and rarely defined) notion of logophoricity, and characterize himself as a logophor, but there is some evidence that at least some anaphors embedded in (picture)-NPs are to be analyzed in syntactic terms.3 Torrego (1984) convincingly argues that subject–auxiliary inversion in embedded clauses provides evidence for the successive cyclic nature of long-distance wh-movement. (7)

a. Que fotosi cree Juan tbi que admiraj Pedro ti tj. b. Que fotosi cree Juan que Pedro admira ti.

Torrego takes the absence of inversion in (7b) to indicate that the wh-phrase has skipped that Comp-position. Crucially, if we embed an anaphor within the moving wh-phrase, and replicate the pattern in (7), binding of the anaphor by some element c-commanding the embedded Comp is substantially worse if inversion does not take place (Esther Torrego p. c.). (8)

a. [Que fotos di si mismoj/k]i cree Juanj tbi que admiram Pedrok ti tm. b. [Que fotos di si mismo*j/k]i cree Juanj que Pedrok admira ti.

This fact receives a straightforward explanation under Torrego’s (1984) hypothesis, and at the same time shows that binding of anaphors embedded in picture-NPs must be characterized syntactically (at least in part). Moreover, were one to dismiss the Spanish data as inconclusive, the following datum, where a negative polarity item is used instead of an anaphor to remove any notion of logophoricity, clearly shows that the experiencer licenses the various variables we have looked at (reflexive, NPI). (9)

Pictures of any artists appear to no critics [t be revealing].

(9) provides a powerful argument against approaches following in essence Kitahara (1997) that establish the c-command domain of the experiencer after raising of the embedded subject. Chomsky (2000) offers a different solution to the ‘experiencer paradox.’ Chomsky argues that the experiencer is inherently Case-marked, and as such ‘inactive’ (inert, in the sense of McGinnis 1998) for Attraction. Put differently, even though the experiencer c-commands (into) the embedded clause from the very start (upon merger), it does not block raising of the embedded subject because it has no structural Case-features to match those of the attractor (matrix T). While it avoids all the problems raised against the Kitahara-style accounts, Chomsky’s solution forces us to view the experiencer as inherently Case-marked, and inherent Case as ‘inactive’

Conflicting C-Command Requirements

133

or ‘invisible’ for attraction. While both these ideas have been entertained in various places in the literature (see McGinnis 1998, and references therein), they are certainly not forced upon us by ‘virtual conceptual necessity,’ and, not surprisingly, both ideas have been dismissed in the literature. Boeckx (1998a) examines agreement data that show that the experiencer constrains the agreement possibilities on matrix T. When agreement can be shown not to be default (3sg) as in (13), it is clear that the experiencer blocks the agreement relation between matrix T and the embedded subject. (10) John seems to Mary to be the best. (11) John and Bill/The men seem to Mary to be the best. (12) There seems to Mary to be a man in the room. (13) *?There seem to Mary to be men in the room. (14) There seems to Mary to be men in the room. Facts like these, which will be examined in great detail in section 8.3, militate against the inertness of the experiencer. In addition, the assumption that inherent Case in general is inert has been questioned, with some empirical success, in (among others) Collins and Thráinsson (1996), Stjepanović (1997), and Lasnik (1995b), who argue in favor of a Case-checking mechanism for inherent Case identical to that of structural Case. The purpose of this chapter is to offer a solution to the experiencer paradox that is free of the criticisms just listed. The solution will consist in claiming that even though the experiencer c-commands (into) the embedded clause right from the start (licensing the anaphor/NPI prior to subject raising, and triggering the Condition C effect), it fails to be attracted by matrix T, allowing raising of a ‘less close’ element. The failure to attract the experiencer is not due to the inherently Case-marked status of the latter, but to the need for reanalysis. Keeping the account at the intuitive level for the moment, I will claim that in order for the experiencer to raise (be attracted), the preposition to must reanalyze with seem. This necessary step makes the feature communication between matrix T and the embedded subject much more ‘direct’ (in a sense to be made precise later in this chapter). As a result, the embedded subject raises. Given the crucial role played by the algorithmic timing of operations, the present account offers an argument in favor of a fully derivational approach to syntax. The chapter is organized as follows. In section 8.2, I review Kitaharastyle solutions to the experiencer paradox, and spell out the problems such accounts face. In section 8.3, I turn to Chomsky’s (2000) solution, and show its inadequacy based on the agreement facts in (10)–(14). Section 8.4 contains my solution. Section 8.5 is a summary.

134

Aspects of the Syntax of Agreement

8.2. KITAHARA (1997) Kitahara (1997) addresses the experiencer paradox, illustrated in (5) (repeated here as (15)), and suggests the following derivational account, which is similar to the one independently proposed by Ferguson (1994, 1996). (15) Theyi seem to himj [ti to like Johnk/*j]. To ensure c-command by him and account for the Condition C effect, and at the same time prevent something like himi seems to ti [they to like John] (expected under Closest Attract (3), if him c-commands into the embedded clause), Kitahara suggests that him does not c-command into the embedded clause prior to raising. This is made possible via the following Larsonian shell (Kitahara 1997: 64). (16)

TP ru T vmax ru v Vmax ei PP Vb ru ru to him seem TP 6 . . . they . . .

Assuming the standard definition of c-command (Reinhart 1976) (17), the PP-node dominates him but not the embedded clause, hence him does not c-command into the latter; raising is thus made licit. (17) For A, B nodes in a tree, A c-commands B iff every branching node dominating A dominates B and neither A nor B dominates the other To capture the disjoint reference effect, Kitahara assumes that by LF him must c-command John (assuming, following Chomsky 1993 et seq. that binding is an interpretive procedure established by inspecting LF). To achieve this, Kitahara suggests that him covertly raises and adjoins to P to check its Case4; it is only after this movement that him can c-command into the embedded clause, which forces the disjoint reference effect. Since the experiencer checks its features covertly, it does not bleed any overt operations. Note that Kitahara’s account crucially relies on the existence of a distinct covert component. Recent work (Bobaljik 1995, and Groat and O’Neil

Conflicting C-Command Requirements

135

1996, Pesetsky 1998, 2000, Chomsky 2000, among others) favors a Single Cycle model, where overt and covert operations are interleaved. In such a model, it is unclear (at best) whether raising of the experiencer would not block raising of the embedded subject. This is especially true in frameworks like Bobaljik (1995), and Groat and O’Neil (1996), where covert operations amount to movement of whole categories except for phonological features. Chomsky’s (2000) model might provide a way out in that it distinguishes between feature movement (or long-distance agreement) and category movement (see also Pesetsky 2000). One could say that raising of the experiencer consists of mere feature movement, and as such does not block raising of a category, allowing ‘overt’ raising of the embedded subject. Kitahara’s account might then be viewed as an argument in favor of distinguishing feature and category movements, or the traditional Y-model, with a distinct covert component, if one does not want to explore the way out just sketched in a Single Output syntax. There is, however, a major problem with Kitahara’s account, both the original, and the revised single-cycle one. As pointed out to me by Howard Lasnik (p.c.), the covert/feature movement account proposed by Kitahara is at odds with the otherwise powerful argument put forth in Lasnik (1995c,d) that covert/feature-movement does not affect interpretive relations. Lasnik’s argument is based on the following data (based on Den Dikken 1995b; see Lasnik and Saito 1991, and Bošković 1997c for similar examples in ECM-contexts). (18) a. Some applicants seem to each other to be eligible for the job. b. No applicants seem to any of the deans to be eligible for the job. (19) a.

*There seem to each other to be some applicants eligible for the job. b. *There seem to any of the deans to be no applicants eligible for the job.

Assuming covert/feature raising of the associate indefinite NP in (19), we expect these sentences to be good, other things being equal: At LF (the level where binding is standardly assumed to be determined), the sentences in (19) are similar to those in (18). Crucially, things are not equal. In (19), movement of the ‘logical subject’ is overt (category movement). In (18), movement consists of mere displacement of formal features (Chomsky 1995), which, as Lasnik plausibly concludes, is not enough to affect semantic relations: Something more than formal features is needed. I will refer to the ‘vacuous’ character of feature-movement for interpretive relations like binding/NPI licensing as ‘Lasnik’s Generalization,’ stressing that Lasnik provided an explanation for it.5

136

Aspects of the Syntax of Agreement

Returning to Kitahara’s account, we see that his covert/feature movement must affect binding relations: It is crucial to capture the Condition C effect. If we maintain the view that covert movement is mere feature displacement, Kitahara’s solution to the experiencer paradox fails. If we pursue the idea that this particular instance of covert movement is movement of a whole category (granting some independent mechanism to determine when movement of a whole category is needed), the account still fails in a Single Output Syntax model, for then we lose any distinction between experiencer raising and embedded subject raising: Both raise categories, and blocking would be expected, contrary to fact. Based on this, it seems that Kitahara’s account must be abandoned. Epstein et al. (1998: Chapter 2) offer a different execution of Kitahara’s original account. Couched within a strictly derivational account, the Epstein et al. argument is that all interpretive procedures (binding in the case at hand) should be done ‘on-line,’ i.e., as the derivation proceeds.6 To understand this claim, a note on strictly derivational frameworks is in order. The nature of derivation has been a much discussed issue in the minimalist framework (see Brody 1995, 1997, Chomsky 1995, 2000, Epstein 1999). Here I concentrate on Epstein (1999), Epstein et al. (1998), and Uriagereka (1998, 1999b). Epstein (1999) argues that syntactic relations can only be characterized in a non-stipulative way by appealing to the derivational/ dynamic character of Merge/Move (Generalized and singulary transformations, respectively). Within this model, the transformational component is implicated in all aspects of structure-building. There is no (static) bifurcation as there was from the Standard theory through GB-theory, between the creation of D-Structure (the ‘base’) and the transformational derivation of S-structure. As Epstein argues, syntactic relations (up to and including the ‘is-a’ relation of Lasnik & Kupin 1977) are better conceived of not as properties defined over representations generated by syntactic rules, but as properties inherent to and established by the application of the rules Merge and Move themselves. The set of syntactic relations holding of a linguistic expression follows directly, and, by hypothesis, entirely from the rules that generate the expression and their order of application, i.e. from the derivation. The result is a fully derivational approach to syntax. Given the absence of (static) ‘bifurcation’ in this model, semantic (conceptual-intensional) and phonological (articulatory-perceptual) interpretations are ‘invasive’ to the syntax, being isomorphic to the syntactic derivation and interpreting it as it proceeds, rather than interpreting an output phrase marker representation (a ‘level’) generated by the derivation. There is no phrase marker that serves as the sole object of semantic interpretation (no LF ‘level’), nor is there a unique phrase marker that is spelled out to the phonological component (see especially Uriagereka’s 1998, 1999b Multiple Spell-Out hypothesis). The T-model (Chomsky & Lasnik 1977) collapses. For every syntactic operation there is at least one corresponding interpretive procedure in the semantic and/or phonological components.

Conflicting C-Command Requirements

137

As Epstein et al. (1998) note (p. xvi), this theory of syntax might comport well with a compositional approach to semantic interpretation. Instead of cyclic syntactic construction of an entire phrase marker followed by a cyclic semantic interpretation which retraces the steps of the syntax,7 the syntax serves as a set of instructions to the semantic component to build cyclically composed structures. For binding relations, for instance, there is no longer any need for the semantic interpretive component to examine the LF phrase marker and determine whether a c-command relation holds; rather, binding takes place when the configuration is created, upon merger of the relevant elements, no matter whether the binding configuration is destroyed afterwards. We see how this framework potentially makes room for what appear to be “conflicting c-command requirements” when c-command is viewed representationally, as we did when we laid out the problem in section 1. Derivationally, it is possible for an element B to enter different (‘conflicting’) configurations giving rise to binding effects. This is what Epstein et al. capitalize on. Ignoring technical details (for which I refer the reader to Epstein et al. (p. 71ff.)), one can summarize the Epstein et al. solution as follows. Epstein et al. assume the essence of Kitahara (1997): The experiencer does not c-command (into) the embedded clause prior to raising of the embedded subject. C-command is established after raising. Epstein et al. departs from Kitahara in the way c-command is established in the case of the experiencer. For them, raising of the experiencer is not at issue. Rather, the preposition, which prevents the experiencer from c-commanding (into) the embedded clause, is ‘eliminated’ by some (stipulated) process (20): (20) [to XP] n [XP] Once (20) has applied, the experiencer (XP) is ‘remerged’ into the tree, and c-commands (into) the embedded clause, forcing disjointness in reference with material in the embedded clause. There are various problems with that account. First, the stipulative character of (20) is blatant, and is the kind of technical ‘trick’ the Minimalist enterprise precisely tries to avoid. Second, and this problem is shared with Kitahara’s account, the‘binding-after-raising’ account has virtually no way to capture the binding by the experiencer of an anaphor contained in the raising embedded subject, as in (21) (or, alternatively, the NPI-licensing in section 8.1). (21) [Pictures of himself]i seem to John [ti to be ugly]. I already made it clear in section 8.1 that such binding must be determined syntactically. For that, the experiencer must bind into the embedded subject. Either it does so prior to subject raising, in which case we have to find

138

Aspects of the Syntax of Agreement

another account for why raising is not blocked given Closest Attract, or it binds into the copy of the subject. The latter solution is clearly at odds with the derivational approach to interpretive procedure Epstein et al. push forward (see, especially, their Chapter 6). I therefore take this possibility to be unavailable. But then, we are left with no explanation for the full range of “experiencer effects.” Moreover, upon closer scrutiny, the Epstein et al. account appears to be inconsistent with the strictly derivational approach. The very fact of establishing c-command for the experiencer after raising of the subject by remerging the experiencer without the preposition is clearly counter-cyclic. There might be technical ways out, but not only would these ways be tricks, they would also undermine the original appeal of the whole derivational framework and its alleged ‘minimalist’ superiority. The same criticism is valid for Boeckx’s (1998a) account. Boeckx agrees in essence with Kitahara’s account, but suggests a revision to avoid violating ‘Lasnik’s Generalization.’ The revision consists in not raising the experiencer at any point, but in reanalyzing the preposition and the verb after raising of the subject, thereby extending the binding domain of the experiencer. The result is very similar to the application of (20) in Epstein et al.; indeed, Boeckx suggests we conceive of reanalysis in this case as “covert incorporation/deletion of the P(P) node.” Conceptually, it is more appealing in that reanalysis is independently motivated in that it seems to capture cases like8: (22) Johni was talked to ti. (23) Whoi did you talk to ti? Still, reanalysis, like the process in (20), and Kitahara’s original experiencer raising solution are really at odds with the strictly derivational/cyclic models the authors assume and are trying to support. I therefore take all of the preceding analyses to be unavailable. This forces us to look for an alternative account.

8.3. INHERENT CASE AND INERTNESS An alternative account to the experiencer paradox is put forward in Chomsky (2000). Chomsky argues that the experiencer c-commands (into) the embedded clause, thereby straightforwardly accounting for the various binding effects the experiencer induces.9 The reason the experiencer does not block raising of the embedded subject despite its c-commanding it (its being ‘closer’ for attraction) is, Chomsky stipulates, due to the fact that the experiencer is inherently Case-marked. For Chomsky (see also McGinnis 1998), inherent Case renders the NP inert/invisible for attraction. As a result, the closest match for matrix T (the attractor) is the (structurally Case-

Conflicting C-Command Requirements

139

marked) embedded subject. As already pointed out in section 8.1, treating inherent Case this way is really not forced by ‘virtual conceptual necessity.’ If anything, as argued on empirical grounds in Lasnik (1995b), Stjepanović (1997), and Collins and Thráinsson (1996), the null hypothesis is to assume that inherent Case is licensed by checking, in a Spec-Head relation, much like structural Case. Moreover, irrespective of how we treat inherent Case in general, there seems to be some evidence that the experiencer in the constructions under discussion is not inert/inactive. Consider the following agreement patterns (whose relevance was first pointed out to me by Howard Lasnik (p.c.)): (24) John seems to Mary to be the best. (25) John and Bill/The men seem to Mary to be the best.10 (26) There seems to Mary to be a man in the room. (27) *?There seem to Mary to be men in the room. (28) There seem to be men in the room. The agreement pattern in (24)–(25) is much as expected. It is a case of natural agreement (subject-verb agreement), typical of English. (26) does not tell us anything. It might be that the matrix verb agrees with the expletive, or with the experiencer, or with the embedded subject, which following established (though somewhat misleading)11 terminology I will be referring to as the ‘associate.’ By contrast, (27) is something of a puzzle. Assuming verb agreement to take place with the associate in expletive constructions (contra Sobin 1997; see Chomsky 1995: 384 fn. 43 for arguments against the latter),12 agreement appears to be blocked when an experiencer intervenes (contrast (27) with (28)). In case an experiencer intervenes, the verb seems to take on a default agreement (3sg): (29) There seems to Mary to be men in the room. (27) clearly shows that the experiencer is active, and interferes in the feature communication between matrix T and the associate. The experiencer does not trigger agreement on the matrix verb, as shown in (30). (30) *There seem to Mary and Sue to be a man in the room. (31) There seems to Mary and Sue to be a man in the room. (Due to English poor agreement morphology, (31) does not tell us anything; it might involve default agreement. See Appendix for discussion.)

140

Aspects of the Syntax of Agreement

But even though direct agreement with the experiencer is not taking place, I would nonetheless like to interpret the preceding facts as showing that there is some (‘indirect’) agreement between matrix T and the experiencer. More precisely, I would like to regard (27) as a case of a more general phenomenon, known as the me-lui constraint (Perlmutter 1971, Kayne 1975), or Person-Case Constraint (Bonet 1994), which roughly says that: (32) Person-Case Constraint (PCC) If Dative (agreement) n Accusative (agreement) = 3rd (Bonet (1994:36)) In plain English: (33) If a verb agrees with a dative element and an accusative element, the morphology of the agreement relation with the accusative element is necessarily 3rd person When Perlmutter, Kayne, Bonet, and many others put forward the PCC as an agreement constraint, they were concerned with accounting for the restriction in Romance languages that there cannot be a 1st or 2nd person accusative clitic if there is a dative clitic (though Bonet claims the restriction to be universal). This is illustrated in (34) (from Catalan). (34) *Me li ha recomanat la senyora Boffil. Me.acc.1sg him.dat.3sg has recommended the Mrs. Boffil ‘Mrs. Boffil has recommended me to him.’ I have argued extensively elsewhere (Boeckx 1998a, Chapter 1) that the PCC is indeed universal, as Bonet speculated, and should not be restricted to clitic-cluster, or other surface dative–accusative cases, but rather is to be extended to abstract ‘dative’–‘accusative’ agreement relations. In particular, I have shown that the PCC yields an elegant account of the murky agreement facts in Icelandic sentences involving ‘Quirky’ datives and nominative ‘objects.’ Space limitations prevent me from considering the whole range of complexities discussed in Boeckx (1998a, Chapter 1), but a few examples are in order. Icelandic Quirky elements never agree with the verb (which takes 3rd person singular default agreement morphology) (35), even though they otherwise behave as full-fledged subjects (see Zaenen, Maling, & Thráinsson 1985, and much subsequent work on that point). When a nominative element cooccurs with a Quirky element, the verb agrees with the nominative element, but only in number, not in person (36), whereas full (number and person) agreement takes place between the verb and the nominative element if there is no Quirky element (37).

Conflicting C-Command Requirements

141

(35) Stelpunum var hjálpað. the.girls.dat.3pl was.3sg helped.3sg ‘The girls were helped.’ (36) ?*Henni leiddust við. her.dat.3sg bored.3pl us.nom.1pl ‘She was bored with us.’ (37) Við kusum stelpuna. We.nom.1pl elected.1pl the.girl.acc.3sg ‘We elected the girl.’ To account for these facts, I have argued that agreement between the Quirky element and the verb actually obtains—thereby making Quirky subjects ‘full-fledged subjects’—but that this agreement cannot be morphologically realized on the verb for independent Economy reasons (agreement is already morphologically visible as rich Case on the noun; agreement morphology on the verb would violate some ‘Generalized Doubly-Filled C(ategory) Filter’; see Boeckx 1998a, Chapter 1, for detail). By contrast, nominative agreement always surfaces on the verb (because of some intricate link between agreement on T and nominative (see George/Kornfilt’s 1981 view, as revised by Chomsky 2000; see also Raposo and Uriagereka 1996, and Schütze 1997)). The morphological shape of this nominative– verb agreement relation is constrained by the PCC. If there is no dative/ Quirky element,13 agreement can proceed unhindered (37); if a dative/ Quirky is present, it is the subject (for independent reasons; see, among others, Chapter 1, Boeckx 1998a, and Harley 1995b); the nominative element is an object. This situation is identical to the Catalan example given earlier (34): We have a dative subject and an object, both agreeing with the verb. We have noted that in those cases in Icelandic, agreement with nominative objects is only partial; the verb takes third person agreement, and only number agreement varies (being available in some well-defined cases). This is similar to the PCC result, which states that if agreement obtains with a dative element (in the case of Icelandic, a Quirky element), accusative agreement (here ‘nominative object agreement’) must be third person. That agreement obtains with the Quirky element is detectable, in some subtle form, by the obligatorily third person morphology on the verb (if there were no agreement, there would be no detectable PCC effects, and the restriction on nominative object agreement would be mysterious). Put differently, agreement takes place with Quirky subjects, but the result is almost ‘null’ for the verb’s morphological makeup—it only forces third person agreement, leaving morphological room for object number agreement with (nominative) objects (if any). I won’t try to deduce the PCC from something deeper here (see Boeckx 1998a, and Ormazabal & Romero 1998); I’ll just focus on its effects.

142

Aspects of the Syntax of Agreement

The digression on Icelandic was necessary because I believe that the same PCC effects are at work in English, and can provide a principled account for the puzzling grammaticality of (27). Consider again the full range of agreement possibilities (sentences repeated). (38) John seems to Mary to be the best. (39) John and Bill/The men seem to Mary to be the best. (40) There seems to Mary to be a man in the room. (41) *?There seem to Mary to be men in the room. (42) There seem to be men in the room. The data can be accounted for as follows. In (40)–(41), the nominative element raises overtly, it is the true (syntactic) subject of the sentence, and it thus checks the agreement (AGRs) features on T, allowing them to be realized morphologically (George/Kornfilt’s view). When the nominative element is the associate of an expletive, we seem to detect PCC effects (see (41)): Some agreement relation with a nominative element is blocked by the presence of a dative element, but why should this be so? More precisely, given a checking theory where inflection is present from the start, why should there be an asymmetry between overt and covert operations (39) versus (41)? If the PCC were operative (i.e., if dative agreement (agreement with the experiencer) took place), agreement with the nominative element should be blocked in all cases. This is indeed what first comes to mind, but this is not quite right. The PCC says that: (43) If dative agreement n accusative (‘AGRo’) = third person When the nominative element moves overtly, it is a true syntactic subject, and the PCC has nothing to say about nominative “subject” agreement. When the nominative element moves covertly, it is an associate, not a true syntactic subject; rather, the expletive is, as can be seen from various tests, such as tags, where only syntactic subjects are repeated. (44) There was a man in the room, wasn’t there/*a man/*he? (45) John was in the room, wasn’t he? The ‘associate’ status of the nominative element is crucial. Lasnik (1992, 1995b) has argued that the associate of the expletive checks a special Case (partitive Case, see Belletti 1988 for the origin of this idea; see Rooryck

Conflicting C-Command Requirements

143

1997 for additional arguments in favor of Lasnik’s approach), in a manner similar to objects, that is, in a specific functional projection, which Lasnik equates with AGRo. Capitalizing on this, I propose that only in those cases where the (misleadingly called) “nominative” element is an associate, agreement between the matrix verb and the nominative element is an instance of ‘AGRo-’ or ‘object-agreement,’ and can be (actually, is) disrupted by dative (experiencer) agreement, in accordance with the PCC. In case the nominative element is an associate (an ‘object’ of sorts), and there is a dative element (an experiencer), agreement between matrix T and the associate is blocked (limited to 3rd) (in the Appendix, I return to why a singular form surfaces).14 Note that this principled explanation for the puzzling ungrammaticality of (41) is only available if the experiencer agrees with matrix T. But that means that the experiencer is not inert, inactive, or invisible for attraction, as Chomsky (2000) claims. Indeed, based in part on Boeckx (1998a) (cited as Boeckx 1997), Chomsky (2000) agrees that the agreement constraints in Icelandic are best explained if Quirky elements bear a structural Case feature. But if they do in Icelandic, and if we find the same agreement constraints in English, we are virtually forced to say that English experiencers bear ‘structural’ Case. If so, Chomsky’s account of the experiencer paradox is no longer tenable. Being structurally Case-marked, i.e., eligible for attraction, the experiencer should block subject raising in (46), contrary to fact: (46) a. *Billi seems to ti [John to be a genius]. b. Johni seems to Bill [ti to be a genius]. This means that we are again left with no explanation for the behavior of experiencers. In the next section, I will propose an account that does not suffer from the shortcomings of both Kitahara- and Chomsky-style accounts.

8.4. THE PROPOSAL I believe that Chomsky is on the right track in saying that the experiencer c-commands (into) the embedded clause right from the start (upon merger). This seems to me to be the only way to maintain a fully derivational account of syntax and interpretive procedures in its full force. What must be done then is find a way of letting the embedded subject raise despite its being not closer to matrix T, or, alternatively, find how to deduce the invisibility of the PP node. To achieve the desired result, I would like to take up the idea from Boeckx (1998a) (but not the technical execution, for reasons given earlier) that what prevents the experiencer from being attracted is the presence of P, or, more precisely, of un-reanalyzed P.

144

Aspects of the Syntax of Agreement

Cross-linguistically, the ‘experiencer paradox’ seems not to arise. It is a robust fact across languages that the experiencer c-commands (into) the embedded clause, licenses anaphors, and triggers Condition C effects. Furthermore, the experiencer blocks raising of the embedded subject, the expected state of affairs if the experiencer c-commands the embedded clause. This is made clear in the following Icelandic and Spanish data. That it is the experiencer that blocks raising of the subject is made clear by (47c,f), where the experiencer is not lexically realized. Similar facts hold in full force in many languages (see Torrego 1996 and McGinnis 1998 for cross-linguistic variation). (47) a.

Þeim hafði virst Ólafur vera gáfaður. them.dat has seemed Olaf.nom be intelligent ‘Olaf seemed to them to be intelligent.’ b. *Ólafur hafði virst þeim vera gáfaður. c. Ólafur hafði virst vera gáfaður. d. Nos parece [este taxista estar cansado]. us seems this taxi-driver be tired ‘This taxi-driver seems to us to be tired.’ e. *Este taxistai nos parece [ti estar cansado]. f. Este taxistai parece [ti estar cansado].

Icelandic

Spanish

English appears to be a special case in allowing raising of the subject across the experiencer. I take this fact to be what is to be explained. That is, I believe that the best way to capture the whole range of data reviewed so far is to assume that the experiencer c-commands (into) the embedded clause right upon merger. As such, there is nothing more to be said for binding of anaphors, or Condition C effects, assuming these to be determined derivationally (see Boeckx 1999e for a derivational theory of binding relations). The proposal I would like to make to capture the failure of matrix T to attract the experiencer is that in those (few) languages (including English) where the subject of the embedded clause is allowed to raise across the experiencer, the latter is dominated by a (possibly phonetically null) preposition. The preposition is overt in the case of seem, covert in the case of strike (see note 1). This preposition is what prevents ‘direct’ (feature) communication (/relation) between matrix T, the attractor, and the experiencer NP/DP (but—crucially—it does not prevent the experiencer from c-commanding (into) the embedded clause). Communication being mediated, it is intuitively obvious that if the attractor can communicate more directly with an ‘equidistant’ element, the latter will be attracted. I take this to be the case in the construction we have examined. That is, matrix T would attract the experiencer were the preposition not present. In the presence of a preposition, the first NP/DP matrix T encounters in its search toward some feature that might eliminate

Conflicting C-Command Requirements

145

its inadequacy is the matrix subject. To put it in Chomsky’s (2000) terms, the preposition consitutes a ‘deficient barrier’ (not itself attracting, but preventing attraction from above) (see also Bošković 1998a for examples of ‘deficient barriers’ (although Bošković does not use the term)). There are two crucial assumptions I have to make here. One is that the embedded subject and the experiencer are in some sense ‘equidistant,’ that is equally close to the target. One might say that they are in the same domain (Spec and Complement of Vseem/strike, respectively). As such, they are both eligible for attraction. This result can be achieved given the following definition of Equidistance, and the representation in (49). (48) Equidistance B and C are ‘equidistant’ for the purpose of attraction, if they are within the same minimal domain [on which see (49)] (49)

VP ru Exper. Vb ru seem IP (={IP,{NPsubj,IP}})

I added the full label of IP in (49) because, as in Uriagereka (1999b), labels can be ‘reified’ and act as ‘stands-in’ for the elements they immediately dominate (NPsubject in our case). In other words, NPsubject becomes part of the minimal domain of V by virtue of being a sister of IP, which is part of the minimal domain of V.15 (Alternatively, one might claim that all elements within a ‘phase’ (in the sense of Chomsky 2000) are equidistant. With IP not being a phase (Chomsky 2000), the experiencer and the embedded subject in (49) are “equidistant.” I leave the precise formulation of how equidistance is achieved in the case at hand for future research.) As for (48), something like ‘equidistance’ has been assumed since the early days of the minimalist program, and survives in Chomsky (2000). I therefore take it to be well motivated empirically. The second assumption I have to make is that P has not (yet) reanalyzed with V by the time Attraction by matrix T takes place. Otherwise, raising of the experiencer would be allowed in much the same way as it is in (50). (50) Theyi were talked about ti. As Baltin and Postal (1996) point out, reanalysis is an operation that has never been precise, but it seems so adequate in accounting for (50) and similar sentences that I will adopt it, without trying to make its nature much more precise, although the proposal I am about to make might shed some light on it.

146

Aspects of the Syntax of Agreement

What I would like to suggest is that reanalysis is an operation triggered ‘from above.’ We have no way of determining whether reanalysis has applied in a sentence like: (51) John talked about Peter. It might have, but it is not visible. By contrast when there is some feature from ‘above’ to check by category movement, as in (50), reanalysis seems to be ‘forced.’ This is essentially the conclusion reached by Kayne (1993), who is forced to say that A-movement triggers incorporation (/reanalysis). In his account of the have/be alternation (sketched in (52a-b)), Kayne suggests that (52a) the specifier of a D/P is an A-bar position because of the nature of D/P, but that by incorporating into BE (52b), which has an A-specifier, D/P becomes the sort of element which has an A-specifier, and thereby allows A-movement through its specifier. (52) a. BE [D/PP B D/Po [. . .]] b. D/Poi-BE [D/PP B ti [. . .]] Based on this, I would like to argue that reanalysis is a Last Resort operation that applies if the complement of the preposition is the only potential checker of some ‘above’ feature (i.e., of a feature from a category c-commanding the V–P complex.) By reanalyzing, V and P free up the way for movement. For the sake of concreteness, I assume that reanalysis amounts to rendering P invisible, making the object of the preposition a direct object. Once reanalysis has applied the object stands in an immediate feature communication with the attractor: They can ‘agree,’ some feature of the object can be ‘attracted,’ triggering category raising as some sort of repair (on which see Chomsky 1995; Ochi 1998, 1999b, and Lasnik 1999b). Note that I am assuming, following Chomsky (1995 et seq.), that there is a syntax of features. The members of a checking relation must first ‘agree’ for category movement to proceed (see Pesetsky 2000). This assumption is embodied in Chomsky’s (1995) view on overt movement: (53) Applied to the feature F, the operation Move creates at least one, perhaps two “derivative chains” alongside the chain CHF = (F, tF) constructed by the operation itself. One is CHFF = (FF[F], tFF[F]), consisting of the set of formal features FF[F] and its trace; the other is CHCAT = (B, t), B a category carried along by generalized pied-piping (Chomsky 1995: 265) The adequacy of (53) has been demonstrated in Agbayani (1998a, 1998b), Ochi (1998, 1999b), Lasnik (1999b), and Pesetsky (2000) (among others).

Conflicting C-Command Requirements

147

Applying it to the present cases yields desired results, as I will now show. The above view on reanalysis makes it clear that the operation manipulates categories, not single features. One might read (53) as saying that the syntax of features takes precedence over the syntax of categories (category-chains are ‘remnant-chains,’ feature-chains are formed first). If so, we predict that if an operation solely involving features can be established within the same domain, it will be favored. This is all that is needed to account for why raising of the embedded subject past the experiencer is allowed in those languages where the experiencer is dominated by a preposition, i.e., in those language where P must first reanalyze with V to enable its complement to communicate ‘featurally’ with some above attractor. If reanalysis doesn’t apply, the experiencer loses ground over the embedded subject that is within the same domain of attraction, and moreover is not prevented by any category from communicating/agreeing with matrix T. Once the communication with the embedded subject is established, it cannot be disrupted, irrespective of whether reanalysis has taken place. Following Chomsky’s (1995) definition of strong feature (roughly, a feature triggering overt movement) (54), one could say that the operation that must apply immediately upon insertion of the strong feature is feature agreement/ attraction. The very next step is pied-piping of the category from which/of which the feature has been attracted/has agreed.16 (54) Strong Feature (‘Virus Theory’) A strong feature must be checked as soon as it enters the derivation In the case of experiencer constructions, reanalysis doesn’t apply, which prevents the experiencer from agreeing ‘first’ with matrix T. The first element to agree with matrix T is the (equidistant) embedded subject. Once Agree has taken place, the very next operation will consist in raising the category that constitutes the embedded subject (pied-piping involving look-ahead, ‘the very next step’ requirement reduces computational complexity). This raises the question of whether reanalysis takes place after raising. Although there might be ways of defining the cycle in such a way as to allow such a ‘countercyclic’ operation taking place within the same minimal domain (see Richards 1997, and Chomsky’s (1997 Fall class lectures, MIT) Featural Cyclicity), I believe that any ‘after’ operation is at odds with a strictly derivational framework like the one assumed here. I therefore assume that reanalysis does not take place. But then, how do we account for the agreement facts that prove crucial in rejecting Chomsky’s (2000) analysis? The answer is straightforward in a framework like Chomsky (1995, 2000), where ‘covert’ feature-raising of objects is taken to target T too (see Chomsky (1995: 361, his (195)). The feature will have to be checked in the complex [v-V-P-T] (assuming all heads to conglomerate at LF; see Chomsky 2000 for a hint at such an idea). This is schematized in (55) (irrelevant

148

Aspects of the Syntax of Agreement

details omitted). We again avoid acyclic operations, without any empirical loss.17 (55)

T ru FF[experiencer] T ru FF[subject] T ru [v v-V-P] T

In languages where experiencers are not dominated by a preposition (Spanish, Icelandic, etc., see earlier discussion), reanalysis does take place (there being no preposition to reanalyze), and strict ‘Attract Closest’ is observed, and raising of a lower element (the embedded subject) is blocked. We seem to have solved the experiencer paradox. It is worth noting that the present account crucially relies on temporal asymmetries (in terms of algorithmic time); it clearly distinguishes between two ‘times:’ one corresponding to feature syntax, the other to category syntax. The two are interleaved (in accordance with recent Single Output models; see references cited earlier), but they can be teased apart at some level of abstraction. As such, the present study might be viewed as an argument in favor of the derivational approach to syntax, supporting Chomsky’s (2000) intuition that the derivationalism–representationalism is more than merely terminological (see Boeckx 2000a for another argument that the derivationalism–representationalism issue is real, and that empirical evidence can be adduced in support of the derivational approach). As pointed out to me by Howard Lasnik (p.c.), the present account, which distinguishes between feature- and category-syntax, might provide an interesting alternative to Pesetsky’s (1995) ‘bi-dimensional syntax’ (cascade and layered syntax) put forth to account for the peculiar properties of prepositional phrases. However, given the intricacies of the data discussed by Pesetsky, this promising line of research will be left for future research.

8.5. CONCLUSION To sum up, this chapter has focused on the paradoxical status of the experiencer in raising constructions. After reviewing major proposals made in the minimalist program to explain why an experiencer does not block raising of the embedded subject despite c-commanding it, I have suggested an analysis that makes crucial use of Chomsky’s (1995, 2000) suggestion that the core computational component operates on features, not categories, and on the idea that syntax is strongly derivational. To the extent that it is correct, the

Conflicting C-Command Requirements

149

present analysis provides evidence of the strongest kind for both a syntax of features, and a strongly derivational view of syntax.

APPENDIX I here discuss some further facts related to ‘experiencer agreement’ in English. As noted in the text, the PCC is operative in a case like (A1) (A1) There seems to a man to be two women in the room. As pointed out to me by Mona Anderson (p.c.), the PCC effect seems to disappear when the experiencer is a pronoun. Consider (A2). (A2) There seem to him to be two women in the room. To account for this contrast, I would like to claim that English ‘suppresses’ all agreement morphology (not just person agreement morphology) in PCC environment. English is forced to do that to make PCC effects visible at all, due to its poor agreement morphology. This does not seem to capture (A2), though. With respect to (A2), I would like to argue that here English really is Icelandic. Pronouns are the only elements that bear visible Case in English. Much like Dative/Quirky elements, English pronouns “wear their Cases on their sleeves.” Due to the generalized Doubly Filled C[ategory] Filter alluded to in the text (agreement surfaces either on the verb, or as rich Case morphology on the noun), the pronoun encodes the ‘dative’ agreement relation in (A2), and leaves (morphological) room for number agreement to be realized with the embedded ‘subject.’ In case the experiencer has no special Case morphology (A1), the result of the PCC has to be encoded on the verb, which I take to entail total lack of agreement with the downstairs subject, due to poor verbal inflection. Note that, if correct, the present account entails that the -s/† morphology on T reflects Number, not Person agreement (see already Kayne 1989b, 1995). At the same time, if all agreement is suppressed in (A1), -s is also a default agreement marker. The present account thus entails that -s is ambiguous between default and number agreement morphology, a not unusual state of affairs in the languages of the world.

9

Raising Across Experiencers Cross-Linguistically

Building on the previous chapter, I show here how the seemingly different possibilities of raising across an experiencer in Italian and English can receive a uniform treatment. My proposal amounts to saying that ‘overt’ (Agree + Remerge) and ‘covert’ (pure Agree) operations have the same locality, but because they involve distinct sets of features, they are not blocked by the same class of intervening elements. The present analysis supports the elimination of the overt/covert component distinction (the temporal asymmetry between overt syntax and covert syntax), but crucially assumes that not all ‘covert’ operations can be analyzed in terms of lower copy pronunciation. At least some instances of long-distance feature checking must not involve any movement of any sort. The chapter suggests that Person-features may be crucially implicated in the nature of EPP effects.

9.1. INTRODUCTION Consider the following asymmetry between English and Italian.1 (1)

a. Johni seems to Mary [ti to be nice]. b. Johni seems to her [ti to be nice].

(2)

a.

*Giannii sembra a Maria [ti essere stanco]. Gianni seems to Maria to-be ill ‘Gianni seems to Maria to be ill’ b. Giannii gli sembra [ti essere stanco]. Gianni to-her seems to-be ill ‘Gianni seems to her to be ill’

As these examples show, the possibility of raising an embedded subject across an experiencer appears to be parameterized. Although English readily allows such raising (1a,b), Italian allows it only to a limited extent (in a sense that will be made precise in the next section). In the present contribution I show

Raising Across Experiencers Cross-Linguistically

151

that the difference reduces to an independently attested parameter: pro-drop and the status of ‘overt’ subjects.

9.2. ITALIAN Let us start by examining the Italian facts in more detail. If an experiencer is a full noun phrase (3), raising is barred. If the experiencer is a clitic (4a), raising is licit. Likewise, raising is allowed if the experiencer is fronted either by dislocation (4b) or wh-movement (4c).2 (3)

*Giannii sembra a Maria [ti essere stanco]. Gianni seems to Maria be ill ‘Gianni seems to Maria to be ill.’

(4)

a.

Giannii gli sembra [ti essere stanco]. Gianni her seems be ill ‘Gianni seems to her to be ill.’ b. A Maria, Giannii gli sembra [ti essere stanco]. to Maria Gianni her seems be ill ‘To Maria, Gianni seems to be ill.’ c. A chi sembra Giannii [ti essere stanco]? to whom seems Gianni be ill ‘To whom does Gianni seem to be ill?’

9.3. ENGLISH As the translations of the Italian examples attest, English does not seem to impose any morphological or positional restriction on the experiencer and readily allows raising across any such element. However, there is somewhat more subtle evidence showing that English does make distinctions among experiencers, and interestingly, it appears to impose the same restrictions as Italian. Consider the paradigm in (5)–(10), first reported in Chapter 8 on the basis of observations by Howard Lasnik (see also Den Dikken 2001).3 (5)

John seems to Mary to be the best.

(6)

John and Bill/The men seem to Mary to be the best.

(7)

There seems to Mary to be a man in the room.

(8)

*?There seem to Mary to be men in the room.

152

Aspects of the Syntax of Agreement (9)

There seems to Mary to be men in the room.

(10) *?There seems to be men in the room. When the possibility of default (3sg) agreement can be excluded, as in (8), it can be seen that the experiencer blocks the agreement relation between matrix T and the embedded subject (contrast with (10)). The experiencer itself is unable to trigger agreement, as shown in (11). (11) There *seem/seems to two women to be several men in the room. We thus face a case of defective intervention, in the sense of Chomsky (2000): a situation in which the element blocking an agreement relation is itself unable to trigger agreement. (For extensive discussion of the nature of defective intervention, see Chapter 10.) Taking agreement on the finite verb in existential constructions to be established by some ‘covert’ raising process (Chomsky’s 1995 feature movement analysis or his 2000 Agree analysis), we can claim on the basis of (8) that English experiencers block covert ‘raising’ of the embedded subject, as evidenced by the impossibility of verbal agreement. Interestingly, as pointed out by Mona Anderson (p.c.), the intervention effect noted in (8) disappears when the experiencer is pronominal. (12) There seem/*?seems to her to be two men in the room. Likewise, when the experiencer is topicalized, as in (13a), or wh-fronted, as in (13b), the intervention effect is absent (Howard Lasnik p.c.). (13) a. To Mary, there seem /*?seems to be men in the room. b. John wonders to whom there seem/*?seems to be men in the room. Significantly, pronouns and fronted experiencers are precisely those that do not block raising in Italian, as we saw in section 9.2. Thus, the data reviewed here strongly suggest that experiencer intervention for subject raising can be more pervasive than a basic minimal pair like (1) and (2) would indicate. The generalization seems to be that what Italian experiencers do ‘overtly,’ English experiencers do ‘covertly.’ In other words, while some Italian experiencers block covert subject raising, the corresponding English experiencers block covert raising. To the best of my knowledge, no current account of the contrast in (1)–(2) predicts such a deep parallelism between the two languages (see Chapter 8 for discussion of recent accounts of (1)–(2) and their limitations).

Raising Across Experiencers Cross-Linguistically

153

9.4. TOWARD AN ACCOUNT I will now show that the overt/covert distinction is misleading once the status of subjects in the two languages is taken into account. It has often been pointed out in the literature that preverbal subjects in Italian, and in pro-drop languages in general, have A-bar properties (see Cardinaletti 1997 for review and references). A particularly salient illustration of this property is the fact that preverbal subjects in pro-drop languages block overt wh-movement, as shown in the following Italian examples. (14) a.

*Chi Gianni ha visto? who Gianni has seen ‘Who did Gianni see?’ b. Chi ha visto Gianni?

The contrast follows if the preverbal subject Gianni has A-bar properties. (14a) is a straightforward case of intervention (Relativized Minimality). Largely because it renders the parallelism I am trying to establish more transparent, I will assume, as was standard in the GB-literature, that the subject position in Italian is occupied by pro, a base-generated “expletive-like” element in SpecIP (for recent reappraisal of this position, see Goodall 2001).4 If the treatment of pro as an expletive-like element is correct, it allows us to reduce the relation between a post-verbal subject (like Gianni in (14b) and INFL in Italian) to a feature-movement/Agree process, akin to the relationship between Infl and the indefinite NP in English existential constructions. The EPP (i.e., filled spec IP requirement) is satisfied by an ‘expletive’ in both cases. This conception of pro illuminates the similarity between English and Italian at the heart of this study. The generalization that emerges is that if the subject requirement (filled SpecIP) is met by independent means (presence of an expletive: there in English; pro in Italian), a full NP experiencer blocks any featural relation between matrix Infl and the embedded subject. By contrast, experiencer clitics and traces created by wh-movement/topicalization don’t. Granting the generalization arrived at, we now have to explain (i) why some experiencers block ‘covert,’ but not ‘overt’ raising, and (ii) why some experiencers do not have any blocking effect at all. Let us deal with issue (ii) first.

9.4.1. Raised Experiencers The generalization we arrived at can be refined for Italian as follows: A raised experiencer has no blocking effect whatsoever: If the experiencer is cliticized or A-bar moved, it is transparent. Such a refinement extends to English if we grant the possibility of pronoun raising in sentences like (1b). That (1b) may be analogized to (2b) is supported by the following contrast between simple pronouns and coordinated/emphatic ones.

154

Aspects of the Syntax of Agreement (15) a. There seems/*?seem to him and her to be men in the room. b. There seems/*?seem to HER to be men in the room.

Lack of stress and failure to be coordinated are defining properties of clitics (or reduced pronouns) (see Kayne 1975, and Cardinaletti and Starke 1999, among many others). Furthermore, Bošković (2001a, 2004) provides compelling empirical evidence for his claim that quantifiers like all are floated (à la Sportiche 1988), but never in theta-positions. If this is correct, (16) also suggests that raising of the experiencer has taken place. (16) There seem/?*seem to them all to be men in the room. On the basis of (15)–(16), I conclude that the pronoun in (1b) is a weak nominal form that has undergone movement. Since pronoun movement is not at all unattested in the grammar of English (it arguably underlies the contrast between John picked him up/*up him and John picked Bill up/up Bill), it is plausible to assume some (pronominal) Object Shift in the cases at hand (see Koopman 1999 for evidence for pronominal object shift in English). The question that arises at this point is why raised experiencers are transparent. Put in a larger context, the observation is actually not new. Uriagereka (1988) already noted that ‘traces’ are invisible for attraction. Chomsky (2001) develops a mechanism of evaluation for locality that achieves the same result: Traces (/unpronounced copies) do not have any blocking effect. This generalization alone would have the desired effect for our purposes. But before addressing the next question on our agenda I would like to suggest a reason for why raising renders an element invisible. It is an old observation (see already Chomsky 1986) that an element having checked Case (i.e., having raised to a K-complete functional projection for EPP purposes) ceases to be available for movement, as illustrated in (17). (17) *Johni seems [ti is nice]. In Chomsky’s (2000: 123) terms: It is structural Case that enables the closest goal G to select P(G) to satisfy the EPP by Merge. Thus, if structural Case has already been checked (deleted), the phrase P(G) is “frozen in place,” unable to move further to satisfy the EPP in a higher position. In contrast to the freezing effect caused by overt movement in (17), we see in examples like (18) that the K-features of an NP may be involved in multiple Agree relations.

Raising Across Experiencers Cross-Linguistically

155

(18) a.

Það voru skrifaðar þrjár bækur. (Icelandic) there were written.pl three books ‘There were three books written.’ b. ?There look as though there are two men in the garden.

In (18a), the NP þrjár bækur triggers agreement on both the past participle and the finite verb. In (18b) (modeled on an example in Chomsky 2000: 149 fn. 93(ii) and Groat 1999), the associate indefinite NP can Agree with both finite Infls. There thus seems to be a contrast between Move and Agree: multiple Move (triggered by features of the same type) is disallowed, but multiple Agree isn’t. The asymmetry has been recognized in several works. For instance, upon examining a wide range of apparently unrelated phenomena, Richards (1997) formulates a condition on what constitutes a legitimate chain at the interface. His first principle is given in (19). (19) PF must receive unambiguous instructions about which part of a chain to pronounce Richards argues that (19) must be understood in conjunction with the principle in (20). (20) A strong feature (EPP associated with a full K-feature set) instructs PF to pronounce the copy in a chain with which it is in a feature checking relation. (20) may then be rephrased as: an element may only check one EPP feature (of the same type in order to allow for overt Case/EPP checking by a moving wh-phrase, for reasons discussed in Boeckx 2001). The invisibility of raised elements may thus follow from Full Interpretation (Chain Legibility at the interfaces). Once raised, elements are frozen in place. The sentences in (21) furthermore suggest that a raised element is not only frozen for further movement, but also for any Agree relation: (for extensive discussion of (21b), and the way it contrasts with (18b), see Fujii 2002.) (21) a. *There seem [several men are her]. b. *There look as though two men are in the room. In sum, raised elements are invisible for both Move and Agree. The failure of moved experiencers to intervene may thus be schematized as follows (I leave aside the exact identity of the landing site of experiencer movement): (22) T . . . [XP EXPi . . . [VP ti . . . [TP . . . SUBJ . . .]]] However, there is one more issue to address. (22) straightforwardly corresponds to the cases where the experiencer has cliticized, but not obviously

156

Aspects of the Syntax of Agreement

to those cases where it is A-bar moved. Such cases may correspond to the derivation in (23), where the experiencer checks its Case in situ (23a), and stays put until C is introduced (23b). (23) a. T . . . [VP wh-EXP . . . [TP . . . SUBJ . . .]]] b. [CP wh-EXPi [C [T . . . [VP ti . . . [TP . . . SUBJ . . .]]] Taking agreement relations to be established derivationally (see Chapter 10 for extensive evidence for this position), we incorrectly predict that the whexperiencer will intervene at the crucial T-stage (23a), since at that stage T would not know that the experiencer will be raised to C later on. Crucially, from a derivational viewpoint, for the experiencer to be invisible, it must undergo some raising prior to the T-stage, as in the clitic case. To achieve this, I claim that A-bar elements (wh-phrases, topics, etc.) are either merged directly in SpecCP and are associated with a clitic-like pronominal element (null in English, overt in Italian clitic left dislocation), which gets the relevant theta-role and undergoes movement like other pronouns, or else such experiencers are decomposable into a quantifier part and a pronominal part (e.g., who = wh + one). I assume that the pronominal (sub)element behaves like a clitic. In other words, it undergoes overt movement like other clitics, in this case pied-piping the whole experiencer phrase, and thereby rendering the latter invisible for attraction by T.

9.4.2. The Overt/Covert Asymmetry Having accounted for why raised experiencers are transparent for purposes of attraction by T, I now turn to issue (i) raised at the end of section 9.3: why experiencers that fail to raise block Agree (Italian, English existential sentences), but not Move. This situation is puzzling if Move is a composite operation consisting of Agree + Re-merge (Chomsky 2000, 2001). Whether Agree is coupled with Re-merge should not matter; the experiencer should have the same effect. It should block in all situations, contrary to fact. Here I argue that the situation is indeed paradoxical as long as we do not specify the features that are involved in Agree. More precisely, we have so far proceeded on the assumption that the same K-features are involved in Agree at a distance and Agree coupled with EPP-satisfaction. But this assumption should be questioned. For one thing, as Chomsky (2004) and Boeckx (2003a) (among others) have emphasized, the EPP is not a feature in the technical sense; it is a requirement associated with the checking of a feature. Since the EPP on T is not optional, it may be desirable to say that the EPP-requirement correlates with the presence of a feature that is absent from the set of features triggering long-distance agreement. Put another way, Agree at a distance and Agree underlying Move (in fact, forcing Move if Boeckx 2003a is right) range over different features.

Raising Across Experiencers Cross-Linguistically

157

Here I will follow Boeckx (2003a) in taking the distinguishing feature to be [+person] (nothing substantial bears on this choice here, though). More precisely, suppose that Agree at a distance only involves number (and possibly gender), not person features, while Agree underlying/forcing Move involves a full set of K-features: person and number (and gender). If this is tenable, we are now in a position to account for the fact that experiencers in situ block Agree but not Move. All we need to say is that experiencers are not eligible for person-feature checking. They thus fail to match the probing feature of T, letting the embedded subject become the closest Match when person-feature checking is involved. As for why experiencers cannot check the person feature, it would be odd to say that it is because experiencers lack the person feature entirely (experiencers may be 1st person, 2nd person, etc: to me/you, . . .). Instead, I propose that the fact that they are Case-marked by P render their personfeature transparent. (Cross-linguistically, it is extremely rare to find elements Case-marked by preposition triggering agreement, but to the extent that we regard ‘concord’ situations as instances of agreement, we see that agreement is limited to [number], and never involves [person]. For fuller discussion, see Boeckx 2003a.) Having said this, we are now on familiar ground. The general format of the blocking effect of the experiencer is well known. It essentially amounts to a Relativized Minimality violation (Rizzi 1990). An element B enters into a relation with an element C if there is no H that meets the requirement(s) of B (i.e., that matches B), and H either c-commands C. The illicit situation is schematized in (24). (24) [B . . . [ . . . H . . .[ . . . C . . .]]] (H c-commands C and B c-commands H) In the following I will adopt Starke’s (2001) conception of Relativized Minimality, as it leads to what I think is a clearer solution. The portion of Starke’s view on chains that will be relevant for us is roughly as follows. (25) a. B . . . B . . . B b. BC . . . B . . . BC In (25a) we have three elements of the same type (B). Attempting to relate the first and the third (in linear order) leads to a violation of Relativized Minimality. The situation in (25b) is more complex. The intervening element (second element in linear order) is of type B. The first and the third elements that the grammar is trying to relate both contain an B feature. In addition, they contain a C feature, which is missing from the intervener. Starke’s point is that if the first and the third element are B-related, the situation that obtains is equivalent to that in (25a), and is thus ruled out by Relativized Minimality. If, however, the first and the third elements are C-related, no

158

Aspects of the Syntax of Agreement

intervention effect emerges, as the potential intervener is not “of the same type.” A concrete case of (25a) is a superiority condition of the type we found in (26)–(27). (26) *Whati did who buy ti (27) [C [who T [buy what]]] B B B +wh +wh +wh Starke claims that the situation in (25b) corresponds to a weak island. As is well known from the work of Cinque (1990), Rizzi (1990), and Szabolcsi and Zwarts (1993), weak islands such as wh-islands are ‘selective’ islands, in the sense that certain elements associated with well-defined readings can extract, while others, which lack these readings, can’t. Classic cases are the theta-/non-theta related contrast (28), and the D-linked/non-D-linked contrast (28). (Judgments are contrastive, rather than absolute.) (28) a.

*How many pounds do you wonder whether he weighed t? ANSWER 1 [non-referential/non-theta reading]: he weighed 100 lbs (he was skinny) b. How many pounds do you wonder whether he weighed t? ANSWER 2 [referential/theta reading]: he weighs 100 lbs (by lifting the package)

(29) a. *What do you wonder whether Mary read? b. Which book do you wonder whether Mary read? Starke’s view is that the good instances of extraction (28b, 29b) correspond to the situation in (25b) when the moving element C-relates to its final landing site. The bad instances of extraction correspond to an B-relation. My claim is similar: the blocking effect of the experiencer corresponds to the situation in (25a). The EPP-requirement (overt raising) introduces a [person] feature that corresponds to the C-feature in (25b), and that voids the otherwise expected blocking effect.

9.5. SUMMARY In sum, externalized experiencers are invisible to both Infl-Agree and InflMove as they have already checked an EPP requirement. In situ experiencers block Infl-Agree (since they haven’t checked any EPP requirement, they are still visible). However, being Case-marked by a preposition, they cannot enter into an Agree relation involving a [person] feature, thus failing to be

Raising Across Experiencers Cross-Linguistically

159

a potential Match for Infl-Agree when the latter is associated with an EPP requirement (i.e., with [person] checking). Overt raising will therefore be possible across them. Covert raising, by contrast, will be blocked, as by hypothesis Agree in such cases does not involve [person] feature checking. The preceding paragraphs show how the seemingly different possibilities of raising across an experiencer in Italian and English receive a uniform treatment. My proposal amounts to saying that ‘overt’ (Agree + EPP/‘Move’) and ‘covert’ (Agree) operations have the same locality, but because they involve distinct sets of features, they are not blocked by the same elements. In effect, the present analysis eliminates the overt/covert component distinction (the temporal asymmetry between overt syntax and covert syntax), but is not compatible with any Single Output Syntax model. In particular, it appears to require the existence of a non-phrasal covert operation like Agree. It is thus compatible with the claims made in Chomsky (2000, 2001), Pesetsky (2000), and Brody (1997) (among others), but not those made in Bobaljik (2002) and Groat and O’Neil (1996), who assimilate all covert operations to phrasal movement.

ADDENDUM To close this study, I would like to go back to the generelizations arrived at for English and Italian. The pattern we have uncovered is strongly reminiscent of a generalization arrived at by Alexiadou and Anagnostopoulou (2001). They note that in various contexts such as quotative inversion in English and stylistic inversion in French, there exists a ‘transitivity’ restriction that disallows the presence of both a subject and an object DPs inside the verbal complex, as in (30)–(31). (The crucial examples are (30c) and (31d).) (30) a. “It’s raining,” said the weatherman. b. “It’s raining,” said the weatherman to Jane. c. *“It’s raining,” told the weatherman Jane. (31) a.

Je me demande où Marie mangera demain. I wonder where Marie will.eat tomorrow ‘I wonder where Marie will eat tomorrow.’ b. Je me demande où mangera Marie demain. c. Je me demande où Marie mangera sa pomme. I wonder where Marie will.eat her apple ‘I wonder where Marie will eat her apple.’ d. *Je me demande où mangera Marie sa pomme. e. Je me demande quand Marie parlera à Pierre I wonder when Marie will.speak to Pierre ‘I wonder when Marie will speak to Pierre.’ f. Je me demande quand parlera Marie à Pierre.

160

Aspects of the Syntax of Agreement

Since nothing goes wrong if the subject remains in situ in the presence of a PP inside VP, Alexiadou and Anagnostopoulou logically conclude that the ungrammaticality of (30c)–(31d) is due to a Case problem. More precisely, they argue that the configurations in the two cases would demand covert Case checking by a head/Case-assigner that has raised, and ceases to c-command after adjunction to another head. In order for their account to work, Alexiadou and Anagnostopoulou make the following assumptions. In all the cases of subject in situ they examine, the EPP is ‘independently’ checked (by an expletive, verb movement, etc.) and the verb has raised overtly to T. Verb raising leads to adjunction of the (canonical) accusative Case checker (say, V or v) to the (canonical) nominative Case checker (T). In other words, verb raising results in the formation of a composite head bearing two Case features. Crucially, Case-checking of in situ arguments (in early minimalist terms, checking of weak Case features) takes place in a distinct LF component (much as in the traditional Y-model), after all overt operations have taken place. In other words, Case checking of the DPs in VP in (30)–(31) takes place after the creation of a composite [V-T] head. Alexiadou and Anagnostopoulou claim that creation of a complex head prior to multiple Case checking leads to a c-command problem: the accusative Case-checker (V) no longer c-commands the object after adjunction to T. In the presence of an independent Case feature on T, the Case feature on V cannot percolate. Crucially, if one of the DP arguments is externalized (i.e., raises in overt syntax), covert multiple Case checking is avoided (Case percolation may take place). Hence, they provide a strong argument for viewing Case as one trigger of overt movement. The similarities between Alexiadou and Anagnostopoulou’s generalization and the raising sentences discussed here should be clear enough at this point. In both sets of cases, externalization of one element rescues derivations where two arguments including a subject in situ would be contained within VP. However, two important differences emerge from the preceding discussion. First, as should be clear from (30b)–(31f), Alexiadou and Anagnostopoulou’s generalization is confined to instances where there are two DPs without Case. It says nothing about {DP,PP} sequences. Now in both Italian and English raising cases (2a, 5 through 10), the experiencer is a PP (clitic cases aside in Italian). Second, overt verb raising is crucial to Alexiadou and Anagnostopoulou’s account. The English raising cases examined here do not obviously fit the pattern, as it is standardly assumed that main verbs in English do not raise (to Infl). As for the first difference, although we are dealing with a PP experiencer, it is undeniable that the latter has NP properties. There is indeed ample evidence that the evidence that the experiencer NP c-commands into the embedded clause, as shown by the Condition C effect it induces in (32). (For additional evidence, see Chapter 8 and Castillo, Drury, and Grohmann 1999a, 1999b.)

Raising Across Experiencers Cross-Linguistically

161

(32) Theyi seem to himj [ti to like Johnk/*j]. To account for the c-command relation here one may assume that the preposition to does not head the PP, but is adjoined to the NP, functioning as some sort of Case-marker (for alternatives, see Kitahara 1997, Epstein et al. 1998, and Chapter 8). Still, the experiencer NP would be Case-marked within VP, and it is therefore not clear why the derivation would be identical to Alexiadou and Anagnostopoulou’s cases.5 As for verb raising, a lot may depend on the exact position of main verbs in English, an area I do not wish to enter here. It is clear that there are some asymmetries between the cases I seek to relate, but the similarities seem to be more than mere accidents. The solution I offer for (1)–(2) differs in several respects from Alexiadou and Anagnostopoulou’s line of research. First, it is crucial for them that at least some instances of externalization not be triggered by the need to check the EPP. They thus draw a sharp distinction between Case and EPP. Second, they crucially rely on the existence of two separate cycles. Third, they treat headmovement as a syntactic phenomenon. These three assumptions are at odds with the strictly cyclic model developed by Chomsky (2000, 2001), which I adopt here. In particular, Chomsky is quite explicit about the fact that the EPP requirement is more specific than “a probe P requires a SPEC.” If this view of the EPP is correct, Alexiadou and Anagnostpoulou’s formal instantiation of the generalization cannot be maintained. Also, my analysis does not appeal to any role of head-movement, which is desirable in light of the many problematic aspects of the latter (see Brody 2000, Chomsky 2001, Boeckx and Stjepanović 2001, Mahajan 2000, among others). And, finally, it does not require the existence of two separate cycles, which is desirable conceptually. Whether the solution proposed here extends to all the cases Alexiadou and Anagnostopoulou discuss is a task beyond the scope of this chapter. It is, however, noteworthy that Alexiadou and Anagnostpouolou reject the possibility that their generalization reduces to a locality condition (say, Closest Attract) because such a reduction would require covert movement to be more local than overt movement. My analysis does not say that covert movement (/Agree) is more local; it merely says that distinct feature bundles are involved.

10 The Fine Structure of Intervention in Syntax

This chapter revisits the interplay between locality and feature checking by taking a look at putative examples of defective intervention effects, i.e., instances of intervention by an element whose formal features have all been checked. Contrary to Chomsky (2000) and the evidence adduced in favor of defective intervention in Chapter 4, it is argued that only interveners that fully match the Probe and Goal featurally should block Agree. Apparent cases of Defective Intervention have been reanalyzed in a way consistent with this position. This is achieved by careful examination of the content of features and feature hierarchies. If correct, the analysis entertained here provides support for distinguishing between two suboperations (Match and Value) in the formulation of Agree.

10.1. INTRODUCTION This chapter is concerned with the nature of locality in syntax. Since Ross (1967), a major concern in generative grammar has been to adequately constrain the set of possible syntactic relations in natural languages. The locality principle that has gained importance in recent years is Rizzi’s (1990) Relativized Minimality principle, which appears in its schematic form in (1). (1)

B>C>H z_______m

Relativized Minimality says that in a situation like (1), C blocks the establishment of a syntactic relation between B and H if C c-commands H (and is c-commanded by B) and is of the same type as B and H. Whereas the c-command requirement does not appear to be problematic, the characterization of what determines whether two elements are of the same type is still a matter of debate. Originally, Rizzi (1990) took the relevant types to be head, A-, and A-bar. Accordingly, if C is a head, it blocks the formation of a head chain, and

The Fine Structure of Intervention in Syntax

163

similarly for elements targeting A- and A-bar positions. Classic examples illustrating the three cases are given in (2). (2)

a. *Beeni John has ti killed? b. *Johni seems that it was told ti that it was raining c. *Whoi does John wonder whether Bill saw ti

With the advent of the minimalist program the three-way distinction that Rizzi relied on is no longer available. Instead more fine-grained features must be sought. But as soon as finer featural distinctions are made, one runs the risk of failing to properly constrain syntactic relations. As is well known, the class of interveners (C) is often very general, and not point-by-point identical to B/H. Consider (3). (3)

*Whoi did you say that [to Sue]j Bill introduced ti tj

Here a topic (to Sue) blocks the creation of a [+wh]-chain. Both topics and wh-phrases count as A-bar elements, but featurally, they are distinct ([+topic] vs. [+wh/+focus]). A detailed featural characterization would incorrectly rule (3) in if it required featural identity as the cause of intervention. On the basis of facts like (3), Rizzi (2001a) argues that intervention should be defined in terms of feature classes (see also Starke 2001). In his own words, What seems to be needed is an intermediate position between the geometric approach and the one based on identity of features with the respective heads. Features determining chain formation seem to cluster into natural classes, such that Minimality effects are determined within classes, but not across them. (Rizzi 2001a: 101) The syntactician’s task then is to define the relevant classes of features. The main goal of this chapter is shed light on this very issue. Along with refining the nature of syntactic interveners, we want to address the question whether Relativized Minimality should be defined representationally or derivationally—and, if the latter, whether it should be defined as an independent principle, or as part of an independently motivated operation. The first aspect of the question has a long tradition, starting with Rizzi’s (1986) chain formation algorithm, and has been hotly debated in recent years as part of the developments of strictly derivational versus representational models of grammar (see Brody 1995; Epstein, Groat, Kawashima, and Kitahara 1998, among many others). The second aspect of the question has only begun to be investigated. Its starting point is Chomsky’s 1995 suggestion that the Minimal Link Condition (i.e., Relativized Minimality) be part of the definition of movement: “We (. . .) add to the definition of Move

164

Aspects of the Syntax of Agreement

[/Attract] the condition (. . .) expressing the M[inimal] L[ink] C[ondition]” (Chomsky 1995: 296). Rizzi (2001a: 101) argues against Chomsky’s proposal on the basis of examples like (3). Chomsky’s claim indeed appears to be too selective, as Move/Attract is defined in very precise featural terms. Once Minimality is made part of Move, the class of interveners is defined too narrowly (e.g., [+topic] elements are not expected to interfere in the establishment of a [+wh/focus] dependency). As a result, Chomsky (2000: 123) refined his view by introducing the notion of defective intervention. The latter is best illustrated by means of the following paradigm from Icelandic (the data are taken from Chapter 1, where the agreement facts are discussed at length). As is well-known, Quirky subjects fail to trigger agreement on the finite verb (4), despite the fact that they behave for all other purposes as bona fide subjects (see Zaenen, Maling, and Thráinsson 1985). (4)

Stelpunum var hjálpað. the girls.dat.pl.fem was.3sg helped.sg. ‘The girls were helped.’

However, the presence of a Quirky element inside the internal domain of the agreeing verbal element at the point of Spell-Out (‘surface structure’) blocks the establishment of an agreement relation between the verb and a nominative element (5), which is otherwise possible (6). (5)

Mér fannst/*fundust henni leiðast þeir. me.dat seemed.3sg/3pl her.dat bore they.nom ‘I thought she was bored with them.’

(6)

Mér *virðist/virðast þeir vera skemmtilegir. me.dat seem.3sg/3pl they.nom be interesting ‘It seems to me that they are interesting.’

For Chomsky, the Quirky element henni in (5) is a defective intervener. It blocks an agreement relation even though it itself lacks the relevant property to trigger agreement in (the Quirky element’s K-features are said to be ‘inactive’). Understood this way, defective intervention may be represented as in (7). (7)

B # Probe

B # inactive F

B # Goal

However, the representation in (7) raises a host of questions. Note that defective intervention is defined in terms of (non-) activity of a feature,

The Fine Structure of Intervention in Syntax

165

which may reasonably be characterized as a feature value (+/–). This goes against the grain of Chomsky’s claim that featural relation like identity is defined in terms of feature, not feature value (Chomsky 2000: 124). Chomsky’s position is clear: Whether an element is 1st or 2nd person has little incidence on syntactic intervention (3rd person may be different from 1st and 2nd person, as in obviation phenomena, but these may be explained away by treating 3rd person as the absence of a person feature, following Benvéniste’s 1966 well-known suggestion). The problem gets worse for a representation like (7) if we follow Uriagereka’s (2000: 2) proposal that “(. . .) de-activation of [a] feature [be understood] as feature deletion.” If this suggestion is adopted, (7) is to be replaced by (8). (8)

B

ø

B

Clearly, no one expects intervention in a situation like (8), as the intervener lacks the relevant feature. However, prior to discarding the notion of defective intervention, one may want to address the following potential objection to our reasoning. Chomsky ties the inactivity of the quirky element’s K-features in the Icelandic example motivating defective intervention to the lack of structural Case. So inactivity of a feature F in this case is linked to the absence of a feature F’. If so, by taking B to correspond to K-features and C to the structural Case feature, one may posit a configuration like (9) for the Icelandic example (5). (We enclose the [-C] in brackets, as its presence depends on one’s position regarding Uriagereka’s proposal stated earlier.) (9)

BC

B(-C)

BC

But for (9) to be a representation of intervention one must prevent the establishment of a C-relation by requiring that B and C be treated as an inseparable bundle for syntactic purposes. However, there is compelling evidence against such a requirement. For instance, accusative Case can be checked without triggering agreement on the verb. Likewise, number agreement can take place in the absence of person agreement (e.g., past participle agreement in Romance), etc. So it is not at all clear how (9) constitutes an improvement over (7). In this study we show that a schema like (9) obtains in many cases, but crucially without giving rise to intervention at all (see also Starke 2001). The only valid representation of intervention appears in (10) (what we dub ‘Direct Intervention’ for exposition purposes), where three relevant elements are specified (positively) for a given feature F. (10) B

B

B

After establishing the validity of (10), we will turn to its nature: Should (10) be understood representationally, or derivationally? Which feature (sets)

166

Aspects of the Syntax of Agreement

should it operate on? Should (10) be part of ‘Attract’? And, if so, which suboperation within ‘Displacement’ (‘Attract’) should it be tied to? Our study is structured as follows. We begin by providing several cases that appear to support the concept of Defective Intervention. Then, we turn to cases that show that defective intervention makes incorrect predications.1 After demonstrating the adequacy of Direct Intervention for the latter, we reexamine the arguments put forth in favor of defective intervention, and provide an alternative analysis for the alleged cases of defective intervention. Throughout, we assume that features are hierarchically organized into types, that is, labeled sets. We see no other way to characterize featural identity, which is the basis for what we call Direct Intervention. Our chapter focuses on cross-linguistic data involving A-chains and A-bar chains. We set head-chains aside, as some doubts exist as to their existence (see Boeckx and Stjepanović 2001, Chomsky 2001, and references therein). The chapter contains four more sections. Section 10.2 provides additional evidence in favor of Defective Intervention. Section 10.3 provides evidence against the latter. Section 10.4 reconciles the evidence by reanalyzing the evidence discussed in section 10.2 in terms of Direct Intervention. Section 10.5 is the conclusion.

10.2. ALLEGED ARGUMENTS FOR DEFECTIVE INTERVENTION In this section we offer three cases that appear to provide rather strong evidence for the concept of defective intervention. We start with two cases that haven’t figured in the generative literature: Japanese object honorification (section 10.2.1), Spanish clitic doubling (section 2.2), the well-known superraising facts (section 10.2.3), and some wh-extraction facts involving (selected) adjuncts in Italian (section 10.2.4).

10.2.1. Japanese Object (Honorific) Agreement The process of object honorification in Japanese (11) provides a very clear case of what it would mean for an intervener to be defective. (11) Taro-ga Tanaka sensei-o o-tasuke-si-ta/tasuke-ta. Taro-nom Prof.Tanaka-acc help-oh-past/help-past ‘Taro helped Prof. Tanaka.’ Since Shibatani (1977), subject honorification has been treated as an instance of (abstract) subject–verb agreement.2 Rather compelling evidence in favor of this position is to be found in Toribio (1990), Ura (2000), and Hasegawa (2002). By parity of reasoning we will take object honorification to be an instance of object–verb agreement.

The Fine Structure of Intervention in Syntax

167

In the first modern study of object honorification, Harada (1976: 530) proposes the following rule called Object Honorific Marking: (12) Object Honorific Marking Mark the predicate as [Object Honorification] when an SSS (a person who is socially superior to the speaker) is included in (a) the indirect object, if the predicate is ditransitive, or (b) the direct object, if the predicate is transitive. The relevant examples appear in (13)–(14). In (13), the verb is transitive, and it agrees with the direct object in honorification. In (14), we have a ditransitive predicate, and the verb agrees in honorification with the indirect object. (13) Taro-ga Tanaka sensei-o o-tasuke-si-ta. Taro-nom Prof.Tanaka-acc help-oh-past ‘Taro helped Prof. Tanaka.’ (14) Hanako-ga Tanaka Sensei-ni Mary-o go-syookai-si-ta. Hanako-nom Prof. Tanaka-dat Mary-acc introduce-oh-past ‘Hanako introduced Mary to Prof. Tanaka.’ However, Chapter 4 observes that (12) has to be refined in light of cases like (15). The predicate is ditransitive, as in (14), but this time the NP bearing the relevant feature to trigger honorification functions as the direct object. In such a case, object honorification (i.e., agreement between the verb and the direct object) is impossible. If an object honorific marker surfaces on the verb, we obtain the odd interpretation that Taro respects Mary. What this amounts to is that the honorific marker in a ditransitive predicate can only associate with the indirect object, not the direct object. (15) *Hanako-ga Mary-ni Tanaka Sensei-o go-syookai-si-ta. Hanako-nom Mary-dat Prof. Tanaka-acc introduce-oh-past ‘Hanako introduced Prof. Tanaka to Mary.’ (16) summarizes the distribution of the object honorific marker in Japanese in light of (15). (16) Object Honorification Agreement generalization Mark the predicate as [Object Honorification] when an SSS (a person who is socially superior to speaker) is (a) a Dative argument (b) the direct object, if the predicate does not take a dative argument

168

Aspects of the Syntax of Agreement

Chapter 4 argues that (15) constitutes a case of defective intervention. That intervention is indeed defective comes from the fact that the indirect object itself cannot trigger honorific agreement (it fails to refer to a socially superior person, hence lacks the relevant feature), but nevertheless prevents the direct object (which has the relevant feature) from agreeing with the verb. Since switching the surface order of the direct object and of the indirect object does not affect honorification marking (see (17)), Chapter 4 claims that it must be the case that honorific agreement takes place prior to word order permutation. (17) *Hanako-ga Tanaka Sensei-o Mary-ni go-syookai-si-ta. Hanako-nom Prof. Tanaka-acc Mary-dat introduce-oh-past ‘Hanako introduced Prof. Tanaka to Mary.’ I take this to mean that agreement takes place under Chomsky’s derivational version of Agree, which applies as soon as the Probe (in our case, v) is introduced into the derivation. To capture the relevant defective intervention effect, I take the dative element to c-command the accusative element. If the reverse were a possible base order, the accusative element would have a chance of being closer to the functional head triggering agreement (say, v), and there would be no defective intervention. The relevant structure is provided in (18). (18) [vP v [VP IO [Vb DO V]]] | X | Agree Analyzed in this light, Japanese object honorification is virtually identical to the Icelandic case that Chomsky took to motivate Defective Intervention: An element C blocks the establishment of an Agree relation between B and H, even though C lacks the crucial property to enter into an Agree relation with B, as schematized in (19). (19) B v | +hon

C IO | –hon

H DO | +hon

n

Defective Intervention

10.2.2. Spanish Clitic Doubling In Spanish, dative clitic doubling (20) is quite commonly allowed (sometimes even necessary; see Torrego 1998, among others). In the vast majority of cases, the clitic matches its double in terms of (K-) features (gender, number, person).

The Fine Structure of Intervention in Syntax

169

(20) a.

Le dije a Pedro que se fuese. to.him.dat.sg said.I to Peter.dat.sg that se leave.he ‘I told Peter to leave.’ b. Les dije a Pedro y Juan que se fuese. to.them.dat.pl said.I to [Peter and Juan.dat]pl that se leave.they ‘I told Peter to leave.’

However, little discussed3 but quite robust feature mismatches are found with dative clitic doubling. (21) provides an illustration of the phenomenon we will focus on here. (In evaluating the examples, native speakers will want to set aside prescriptive pressure that demands the use of les in (21).) (21) Le/*Les envié las notas a mis alumnus. to.them.dat.sg/*pl sent.I the grades to my students.dat.pl ‘I sent my students their grades.’ In (21) we expect the plural form of the clitic (les), but the singular form (le), arguably the default in Spanish, surfaces. Because third person singular is the default, we cannot tell if we are facing a case of agreement mismatch in (22), where the double is third person singular, but, on grounds of uniformity and simplicity, we will hypothesize that mismatch obtains here too, and that the clitic form used here is a default. (22) Le pedí los apuntes a Miguel. to.him.dat.sg asked.I the notes to Michael.dat.sg ‘I asked Michael for his notes.’ The relevant factor distinguishing (20) from (21)–(22) seems to be the presence of an accusative element in the latter. At first, it looks like defective intervention could capture the relevant facts by taking the direct object to intervene, and block the agreement between the clitic and its double. Crucially, the intervention appears to be defective as accusative elements in Spanish cannot double clitics (unless they are preceded by ‘a,’ for which they must meet several interpretive conditions; we set such cases aside here). As in Icelandic, it is not clear how to implement featurally the cause of the defectiveness of the relevant element, in this case, the impossibility of clitic doubling by ‘bare’ (i.e., non-a marked) accusative objects. Because we will ultimately propose an analysis that eschews the issue altogether, we will for convenience only resort to a [-clitic] feature to characterize the defectiveness of accusative elements, but the reader should feel free to substitute any more substantive feature for the one we adopt. By means of the [clitic] feature the intervention effect in (21)–(22) can be represented as in (23) (CL = Clitic; DO = Direct Object; IO = Indirect Object).

170

Aspects of the Syntax of Agreement (23) B CL | +cl

C DO | –cl

H IO | +cl

n

Defective Intervention

The situation arrived at for Spanish is virtually identical to those in Icelandic and, especially, in Japanese since we are dealing with ditransitive contexts. The only difference between Japanese and Spanish appears to be that in Japanese dative elements block, while in Spanish accusative elements do. We contend that the difference is superficial. We assume that in both languages indirect objects (‘datives’) start off higher than direct objects (‘accusatives’). In Japanese the probe (v) is introduced right on top of the thematic layer assigning R-roles to both direct and indirect objects. By contrast, in Spanish we assume that the direct object undergoes movement across the indirect object prior to the introduction of the probe, so that at the point of application of the Minimality configuration (1) the direct object intervenes between the probe and the goal (indirect object). The exact nature of the movement involved is not clear. Movement of the direct object across the indirect object has been appealed to in a number of studies, in its most detailed form in Vukić (1998), but the justification offered for it is ‘Case/Agreement,’ factors that we believe, following Chomsky (2000) and subsequent work, do not require overt movement. We tentatively assume that the direct object undergoes scrambling over the indirect object (an operation immune to Minimality, for reasons we will not go into here) and adjoins to VP, as depicted in (24).4 (For an alternative analysis, see Bowers 2002.) (24) [VP DOi [VP IO [Vb V ti]]] As indicated in (23) we take the probe reaching for the indirect object in the cases at hand to be the clitic itself. Following Sportiche (1996) and, especially Bleam (1999) (for reasons that are laid out in Jeong 2002), we assume that dative clitics are base-generated as heads of functional projections. We take the relevant functional projection to be v. Clitic doubling is an instance of agreement (implemented as an Agree relation) between the clitic (probe) and its double (goal). Combining this idea with the structure in (23) yields a configuration like (25) that is the relevant input for Minimality. (25) [vb clitic-vo [VP DOi [VP IO [Vb V ti]]]]

10.2.3. Superraising Superraising sentences like (26) have been part of the classic Minimality cases. (26) *Johni seems [it was told ti [that it’s raining]]

The Fine Structure of Intervention in Syntax

171

Upon closer examination they appear to constitute a genuine case of defective intervention, if we assume (as is fairly standard) that overt A-movement is Case-driven. In (26) the element that is standardly taken to be the intervener (the embedded subject it) has checked its Case, yet it blocks Casedriven movement of John, as schematized in (27). (27) [IP John (. . .) [IP it (. . .) [VP t]]] z__________=___________m Deletion of the Case-feature of it yields a representation like (28). (28) B John | +Case

C it | –Case

H t | +Case

n

Defective Intervention

10.2.4. Selected Wh-Adjunct Extraction Rizzi (2001) discusses the following case of wh-adjunct extraction in Italian. (29) *Come hai detto chi si è comportato? how have.you said who si is behaved ‘How did you say who behaved?’ Example (29) is slightly different from standard extraction facts involving adjuncts, as this time the adjunct is selected by the downstairs verb. The adjunct is thus ‘argument-like.’ Yet the deviance of extraction across a wh-island is comparable to the one found with non-selected adjuncts like (30a). In other words, it is stronger than extraction of arguments like (30b). (30) a.

*Come non sai chi abbia letto questi articoli? how not know who has read those articles ‘How do you wonder who read those articles?’ b. Che articoli non sai chi abbia letto? which articles not know.you who has read ‘Which articles do you wonder who read?’

Although (29) may be taken to constitute a standard case of intervention defined over a wh-feature, as in (31), we need to take into account the fact that the improvement noticed in (30b) is often accounted for by appealing to the presence of a theta-role that is missing on the intervener (see Rizzi 1990). Accordingly, (29) could be represented as in (32).

172

Aspects of the Syntax of Agreement (31) come . . . . chi . . . . t [+wh] [+wh] [+wh] z____________=_____________m (32) come . . . . chi . . . . t [+wh] [+wh] [+wh] [+mnr] [+mnr] z____________=_____________m

The representation in (32) is a case of defective intervention: unlike the probe and the goal, the intervener lacks a [+manner] feature, and in this sense is defective. In sum, we have seen four cases that support the idea that grammatical theory must contain a notion of defective intervention. However, the last case discussed highlights an important question. We represented (29) as in (32) by analogy with Rizzi’s account of the improvement in (30b), schematized in (33). (33) chi . . . . si . . . . t [+wh] [+wh] [+wh] [+R] [+R] z____________=_____________m Interestingly, the instance of defective intervention in (33) gives rise to a mildly deviant (for some speakers, even non-deviant) output, whereas the one in (32) yields sharp deviance. In the next section we document four cases where defective intervention obtains and yet the result is grammatical. Our task in the remaining sections will be to relieve the tension that arises from the conflicting results obtained in this study. 10.3. NO DEFECTIVE INTERVENTION

10.3.1. Raising Across an Experiencer Although English is known to allow overt raising of an embedded subject over an experiencer as in (36), Boeckx (2002b, and Chapters 1and 8) shows that it disallows ‘covert raising’ (Agree) (at least when the experiencer is a full NP; see the references above for details). (34) Johni seems to Mary [ti to be nice]. Thus, in existential constructions like (35), it can be shown that the NP-associate is not accessible to matrix T, due to the presence of an experiencer. (35) a.

*?There seem to Mary to be men in the room.

The Fine Structure of Intervention in Syntax

173

b. There seems to Mary to be men in the room. c. *?There seems to be men in the room. d. There seem to be men in the room. The Move/Agree asymmetry found in raising over an experiencer constructions is puzzling if we take seriously Chomsky’s (2000) suggestion that Move incorporates Agree as a suboperation (Chomsky takes Move to be the sum of Match, Agree, and Remerge/EPP-satisfaction). If Agree is part of Move, and Agree is blocked in (35), it should block Move in (34). Boeckx (2002b) claims that the situation just described is indeed paradoxical as long as we do not specify the features that are involved in Agree. More precisely, we have so far proceeded on the assumption that the same K-features are involved in Agree at a distance and Agree coupled with the EPP. But this assumption should be questioned. For one thing, the EPP is not a feature in the technical sense; it is a requirement associated with the checking of a feature (see Boeckx 2002c for extensive discussion). Since the EPP on T is not optional, it may be desirable to say that the EPP requirement correlates with the presence of a feature that is absent from the set of features triggering long-distance agreement. Boeckx claims that the distinguishing feature is [+person]. (Chomsky 2000 claims that expletives are endowed with a (minimal; i.e., 3sg) person feature. Since the function of an expletive is to satisfy the EPP, Boeckx suggest we collapse [EPP] and [person] features (at least for the T-domain); more precisely, let us say that checking of a [person] feature requires movement to Spec—in other words, [person]-checking amounts to the EPP effect. See Boeckx 2002b, 2002c for a way of deducing this.) Accordingly, Agree at a distance only involves number (and possibly gender), not person features. We are now in a position to account for the fact that experiencers in situ block Agree but not Move, without giving up the idea that Agree is a suboperation of Move. The situation found with raising over an experiencer in English can be summarized as in (36) (recall the discussion in section 10.1). (36) a. B . . . B . . . B b. BC . . . B . . . BC Here B would correspond to [Number] and C to [Person]. Since the intervening element is eligible for B-checking only, it fails to intervene when the relevant checking operation involves a [person] feature. A strict view on feature type (call it Direct Intervention) makes the right predictions in this case. But Defective Intervention clearly doesn’t, as it would predict blocking by the experiencer in the case of overt raising. The intervener is defective, as it lacks the relevant [person] feature,5 but that would be enough to block raising/Agree as part of Move. Note that if intervention were defined in a general fashion, say over K-features, then the experiencer would block (as it contains some (active)

174

Aspects of the Syntax of Agreement

K-feature, viz. [Number]). So intervention must be defined very precisely in terms of feature hierarchies to capture the full paradigm.

10.3.2. Multiple Expletives (Chomsky 2000, Groat 1999) A second case of apparent defective intervention yielding a grammatical result we would like to discuss is exemplified in (37). (37) There look as though there are two men in the room. The example, first discussed in Groat (1999) and Chomsky (2000), provides a case where the associate of the expletive in the embedded clause is able to govern agreement in the matrix clause. By hypothesis, let us take the type of agreement triggered by the associate NP in existential constructions to be limited to [Number] (on English finite verb agreement as [+Number], see Kayne 1989b).6 Once this assumption is made, (37) can be represented as in (38). (38) a.

[IP there Infl (. . .) [IP there (. . .) [XP two men]]] z______________________________m b. Infl there two men +Numb +Pers +Numb(;+Pers)

The question that arises at this point is why the expletive in the downstairs clause does not block the Agree relation between the associate and the upstairs expletive. If defective intervention were a valid theoretical construct, intervention would be predicted in this case: Even though the expletive is contains a [person] feature only (Chomsky 2000), it would be able to (defectively) interfere with the establishment of a [Number]-agreement relation, as schematized in (39). (39) B Infl | +Numb

C there | –Numb

H NP | +Numb

That no intervention is found in (37) argues against Defective Intervention.

10.3.3. French Combien Extraction A third argument against Defective Intervention comes from the well-known asymmetry in amount wh-phrase extraction in French. As discussed extensively in Obenauer (1984) and Rizzi (1990), French allows wh-extraction of the whole combien phrase across a quantificational adverb like beaucoup

The Fine Structure of Intervention in Syntax

175

(‘a lot’) (40), but prohibits the extraction of the combien portion in such a context (41). ((42) illustrates the fact that extraction of the combien portion is independently attested.) (40) Combien de livres a-t-il beaucoup lu how-many of books has-he a lot read ‘how many books did he read a lot?’ (41) *Combien a-t-il beaucoup lu de livres (42) Combien a-t-il lu de livres (41) is the easiest case to represent. It may be reasonable to posit a schema like (43), where the target of movement and the adverb both possess a quantificational feature of sorts (say, Q). In such a case it is not surprising to see an intervention effect. (43) C | +Q

ADV | +Q

combien ([de NP]) | +Q

The example in (40) requires an additional piece of information to be generated (under any theory, as far as we can tell). As has been recently observed by Obenauer (1994) and Rizzi (2001b), movement of the whole combien de NP phrase yields a specific, more-D-linked-like reading that is absent from a bare combien extraction. Rizzi goes even as far as saying that the whole combien de NP phrase raises not to FocusP (the normal landing site for wh-movement), but to a higher TopicP (see Grohmann 1998, 2000 on wh-phrases as topics). We believe that Rizzi’s intuition is correct, but instead of appealing to a Topic feature (which would be odd for a wh-feature), we would like to make use of K-features to characterize presuppositional readings of wh-phrases (the idea being that K-features further encode the wh-phrase into context). Thus we obtain a representation like (44). (44) C | +Q+K

ADV | +Q

combien de NP | +Q+K

Assuming Direct Intervention, (44) correctly rules (40) in (adverbs lacking K-features generally). By contrast, Defective Intervention, as shown in (45), incorrectly predicts the sentence to be ungrammatical (the presence of a Q-feature on the adverb is enough to trigger intervention).

176

Aspects of the Syntax of Agreement (45) why not:
B C | +Q +K

C ADV | +Q -K

H combien de NP | +Q +K
n Defective Intervention

The combien extraction facts in French thus offer an additional argument against Defective Intervention.7

10.3.4. Successful wh-Island Extractions8 The fourth case against Defective Intervention we would like to offer has already been hinted at in section 10.2.4. Defective Intervention predicts that any wh-extraction across a wh-phrase will be deviant. However, as has been pointed out by Rizzi (1990), Cinque (1990), and many others since then, D-linked argumental wh-phrases extract fairly easily out of wh-islands. (The slightly marginal status of the sentence may be a parsing factor.) Consider (46). (46) Which of the girls did you ask whether Bill kissed t? Rizzi argues that the presence of thematic information on wh-phrases renders extraction easier. Cinque claims that D-linking is the relevant factor. By taking presupposition to be encoded in terms of K-features (which are absent from adjuncts), as we did earlier, we can couple both proposals into one and characterize (46) as in (47). (47) C | +Q+K

whether | +Q

which NP | +Q+K

Direct Intervention, unlike Defective Intervention, predicts (47) to yield a grammatical result. As a matter of fact, Direct Intervention makes the—as far as we can tell, correct—prediction (pointed out to us by Alan Munn) that extraction of a presupposed wh-phrase across another wh-phrase of the same type is worse than (46). (48) bears this out. (48) *?Which girl do you wonder which boy kissed t? The contrast is reinforced by forcing a non-D-linked reading on the intervener, such as by using an aggressively non-D-linked wh-phrase of the whthe-hell type, as in (49). (49) Which girl do you wonder who the hell kissed t?

The Fine Structure of Intervention in Syntax

177

The minimal pair in (48)–(49) provides rather strong support to a DirectIntervention-based definition of locality.

10.4. REVISITING THE EVIDENCE Let us take stock. So far we have obtained conflicting results. All the cases in section 10.2 correspond schematically to (50), or, if features are privative (monovalent), to (51). (50) +B+C

+B-C

+B+C

(51) +B+C


+B

+B+C

In all cases, such configurations yield deviance. Surprisingly, the very same configurations were shown to yield grammatical results in section 10.3. For reasons outlined already in section 10.1 ‘Defective Intervention’ does not appear to be an optimal concept, relying as it does on a value/feature of a feature. Therefore we would like to explore the possibility of dispensing with it by revisiting the evidence gathered in section 10.2. Specifically, we would like to argue that in each case a different feature from the one used above can be appealed to, which will have the effect of triggering a Direct Intervention. Consider the Japanese object honorification case first. From the perspective of the [honorific] feature, a sentence like (15), repeated here as (52), indeed offers a case of defective intervention. (52) *Hanako-ga Mary-ni Tanaka Sensei-o go-syookai-si-ta. Hanako-nom Mary-dat Prof. Tanaka-acc introduce-oh-past ‘Hanako introduced Prof. Tanaka to Mary.’ But the argument for defective intervention disappears once we claim that honorification is a specific value of a more general [person] feature (think of the many languages like French who use a special form of the 2nd person to mark honorification). Recall now that values do not matter in syntax; only features do. Once [person] is taken into account, the dative element in (52) is a ‘direct’ intervener: it may not have the appropriate value for honorification, but it certainly does have a [person] feature, which causes intervention. So, as far as Japanese is concerned, the perspective of a mother feature cancels the argument for defective intervention. A somewhat different case obtains in Spanish clitic doubling constructions. Recall that despite the fact that it cannot be doubled, an accusative element blocks dative clitic doubling, as repeated in (53). (53) Le/*Les envié las notas a mis alumnus. to.them.dat.sg/*pl sent.I the grades to my students.dat.pl ‘I sent my students their grades.’

178

Aspects of the Syntax of Agreement

We claimed earlier that if the possibility of doubling is tied to the presence of a [clitic] feature, the intervening accusative element intervenes in a defective way. However, here too, a different perspective is available that need not force us into Defective Intervention. Let us keep to the idea mentioned earlier that dative clitics are base-generated as functional heads (following Sportiche 1996), and relate to their doubles via Agree (see Jeong 2002 for an independent argument in favor of the position we are about to adopt). By contrast, accusative clitics are merged with their doubles (following Uriagereka 1995a), and move away from them (i.e., no Agree relation is involved here). In other words, the possibility of doubling is not due to the presence of a feature like [clitic]. If dative clitics relate to their doubles via Agree, we expect the latter to be subject to locality conditions. Assume further that the Agree relation is one of [number].9 Then, we expect [+Number] elements to block the relevant Agree relation. We contend that this is the case in (53). Certainly the accusative element has a [Number] feature. This feature is not enough for it to act as a double, but it is enough for purposes of intervention. Note here that intervention is Direct, as Agree is defined over a feature that is possessed by the intervener. The third case of Defective Intervention we discussed earlier is superraising. From the point of view of Case the intervener in (54) is indeed defective. (54) *John seems that it was told t that it was raining. But this is not the only way of analyzing (54). We would like to offer two alternatives that dispense with the need for Defective Intervention. The first is to reject that idea that Case is involved, and appeal to K-features. As far as these are concerned, it is no less defective than John. The second alternative we would like to mention preserves the idea that Case is at play, but instead of focusing on the Case (or absence thereof) of the expletive, we would like to point out that both the expletive and the element we are trying to raise (John in (54)) are embedded inside an argumental that-clause, which also requires Case (see Bošković 1995 for extensive arguments). If this is correct, it is plausible to claim that the whole argumental clause prevents John from raising (via some version of the A-over-A, possibly reducible to Relativized Minimality, as argued in Fukui 1997), as represented in (55). (55) __ seems [CP (it) . . . John . . .] | | | +Case +Case +Case
B B B n Direct Intervention The final case of Defective Intervention we discussed in section 10.2 concerns the extraction of (selected) adjuncts across wh-islands in Italian. We repeat the crucial sentence in (56) and its representation in (57).

The Fine Structure of Intervention in Syntax

179

(56) *Come hai detto chi si è comportato? how have.you said who si is behaved ‘How did you say who behaved?’ (57) come [+wh] [+mnr] |

chi [+wh]

t [+wh] [+mnr]

x

|

In this section we claim that (57) is not the correct representation for (56). Note that Agree in (57) is defined over the already formed chain (come, t). But this is illegitimate if we follow Chomsky (2000, 2001) in taking Agree to take place as soon as possible, establishing a relation between a functional head (Probe) and an element in its domain (Goal). (56) should therefore be represented as (58). (58) C [+wh]

chi [+wh]

come [+wh] [+mnr]

Unlike in (57), when the point of view of the target is taken, as in (58), the crucial feature that made the intervener defective (i.e, [+manner]) is missing. It makes no sense to put a manner feature on C.10 Only a [+wh] feature is justified. (56) thus turns out to be a case of Direct Intervention. Note, crucially, that (58) is the representation obtained prior to chain-formation. As we saw, computing locality/Minimality after chain formation leads to the wrong result. In other words, (58) provides an argument for a derivational approach to syntax. So far the four cases presented in section 10.2 in favor of Defective Intervention have been reanalyzed with no loss in terms of Direct Intervention. To do so, we have resorted to two strategies. In some cases (Japanese honorification), the point of view of a mother node inside the feature hierarchy must be taken to achieve Direct Intervention. In other cases (wh-extraction), the point of view of a sister node salvages instances of Agree that would otherwise give rise to Direct Intervention. This is schematized in (59)–(60). (59) +B | +C

+B | –C

+B | +C

(60) BC

B

BC

Two more cases need to be discussed here: the Icelandic facts that served as the original evidence for Defective Intervention, and the Topic island exemplified in (3). They too must be made compatible with Direct Intervention.

180

Aspects of the Syntax of Agreement

Let us start with the Topic island. On the face of it, the example does not lend itself to Direct Intervention, as Topicalized and Focused elements do not obviously share any feature. But if we regard them both as quantificational elements (forming operator-variable chains), possessing a [+Q], feature, then (3) may be represented as in (61). (61) C [+Q]

Topic-NP [+Q] [+K]

wh-NP [+Q]

From this perspective, [wh] and [topic] would be values of a [Q]-feature. The feature structure of A-bar elements have been less studied than that of A-elements, and offering a full-blown justification for (61) goes well beyond the present chapter. What we just did is show a way of making (3) compatible with Direct Intervention. In light of the theoretical and empirical worries raised here against Defective Intervention, the hypothesized structure in (61) certainly gains in validity. The final case we discuss is Icelandic. Icelandic cases like (5) argue for Defective Intervention only if Intervention is defined at the level of actual valuation (in which case Quirky elements are inactive), but at the level of Match, the very first suboperation in the establishment of a Probe–Goal relation (Chomsky 2000, Boeckx 2001), Quirky elements are as active as any other, and certainly possess the relevant K-features. So all we need to enforce is the computation of locality (Minimality) right at the level of Match, the very first step in an Agree relation so as to avoid recourse to Defective Intervention (see also Boeckx 2002a). This makes sense if locality (Minimality) is part of the independent operation Agree. Since the latter applies as soon as possible, it is plausible to claim that the former also applies as soon as possible. In this case, it means that Minimality is computed at the very first level of Agree: Match. An independent argument in favor of this position comes from facts discussed by Heck and Müller (2000). The core cases appear in (62a–c). (62) a.

Wen hat Fritz [nachdem er was gemacht hat] getroffen? who has Fritz after he what made has met ‘Who did Fritz meet after he did what?’ b. *Wen hat Fritz [nachdem er was gemacht hat] gesagt who has Fritz after he what made has said [dass Maria liebt]? that Maria loves ‘Who did Fritz say, after he did what, that Maria loves?’ c. *Was hat Fritz [nachdem er gemacht hat] gesagt what has Fritz after he made has said [dass Maria wen liebt]? that Maria who loves ‘*What did Fritz say, after he did, that Maria loves who?’

The Fine Structure of Intervention in Syntax

181

The preceding facts constitute an interesting case of apparent defective intervention. Consider (62b). Descriptively speaking, an embedded wh-phrase cannot move to matrix C in the presence of a matrix adjunct clause containing a wh-element. Note that the wh-element contained inside the adjunct cannot move to SpecCP (an adjunct condition effect), as shown in (62c). Intervention is therefore defective. (62a) is accounted for by assuming that the wh-element that reaches SpecCP has first scrambled past the adjunct clause (a generalization of Bošković’s [1997a] idea). Crucial for our purposes is that Boeckx (2001) has provided extensive arguments that elements inside adjuncts (and other strong islands)11 are inaccessible to Agree (which derives the adjunct condition). Importantly, such elements are accessible to Match (as Boeckx documents extensively on the basis of resumption patterns). That means that the blocking element in (62b) could not block Agree. The only level at which it can be active is that of Match.12 If this is correct, we just provided an argument in favor of keeping the Match and Valuation steps within Agree distinct, a position defended at length and on independent grounds in Boeckx (2001) (see also Nomura 2002b). It may be said that defining Minimality at the level of Match, as opposed to that of actual valuation, makes locality more representational, since Match is a more ‘passive’ step than Valuation. Though correct, this conclusion does not force upon us the idea that syntax is representational. (As a matter of fact, we argued earlier concerning the representation in (58) that syntax must be derivational.) It simply means that syntax makes use of the representations that are generated in the course of the derivation as inputs for further operations (see Lasnik 2001a for a related point).

10.5. CONCLUSIONS This chapter has offered a detailed investigation of locality conditions in syntax. Our central question has been whether intervention effects should hold when the putative intervener is ‘defective.’ Contrary to Chomsky (2000), and despite the apparent evidence we adduced in favor of his position in section 10.2, we have concluded that only interveners that fully match the Probe and Goal featurally should block Agree. Apparent cases of Defective Intervention have been reanalyzed in a way consistent with our position. In order to do so, we were led to carefully examine the content of features and feature hierarchies. What we saw is that values (like [2ndperson]/ honorific]) do not matter. But, equally importantly, intervention should not be defined too generally, say over K-features. For then, [Number] elements would block relations over [person] features, incorrectly (see section 10.3.1, e.g.). So subdistinctions matter. A natural characterization we would like to propose is that intervention is defined over mothers of feature values, not above it. Needless to say, many more cases must be examined before this conclusion can be endorsed.

182

Aspects of the Syntax of Agreement

Our results have several implications. First, as just said, they are compatible with the idea that syntax does not take feature values into account. They also open up the possibility—preferable on grounds of Economy of representation—of dispensing with negatively specified features: Nothing is lost if syntactic features are monovalent/privative. Our account also eliminates the need to appeal to the notion of ‘inactive’ feature, and as such may provide a potential argument against a phase-based theory of syntax in which (in) activity of features is central. Finally, it offers an argument for keeping the two suboperations Match and Agree/Valuation within the establishment of a Probe–Goal relation distinct, and for viewing the computation of locality derivationally.

IV

Agreement and Interpretation

11 Scope Reconstruction and A-Movement

This chapter reexamines recent arguments against A-movement reconstruction in general, and argues that reconstruction effects with A-chains exist, although they are more limited than previously thought. I argue that the reason why A-movement reconstruction is limited is due to the fact that A-moved elements only become ‘interpretable’ once their case features have been checked. This checking typically happens in a position above all potential reconstruction sites. I show that the Case-checking condition on scope taking can be obviated in the case of indefinites by means of covert insertion of an expletive, and go on to provide arguments that lowered readings of indefinites involve literal lowering. This movement is shown to be subject to Relativized Minimality, thereby further supporting for the claim that Quantifier Movement is a syntactic operation.

11.1. INTRODUCTION The goal of this study is to test the validity of the claim made by Chomsky (1993, 1995) that A-movement does not give rise to reconstruction effects. Recent work by Lasnik suggests that, although far from being the standard assumption, Chomsky’s claim is largely correct. In section 11.2, I lay out the basics of Chomsky’s (and Lasnik’s) claim. In section 11.3, I reexamine Lasnik’s examples, and contrast them with other cases that point to the need for reconstruction. The picture that emerges is that non-indefinites exhibit what I call ‘partial reconstruction’ (connectivity effects); indefinites, by contrast, do reconstruct (‘total reconstruction’). In section 11.4, I provide an explanation for this split in terms of Case. Roughly, A-moved elements can be interpreted only in the position where their Case—a feature rendering elements uninterpretable—is deleted. Portions of complex NPs headed by non-indefinites are allowed to reconstruct to the extent that they are not tied to Case assignment in their surface positions. Indefinites reconstruct because the computational system can resort to expletive insertion, which allows them to check their Case lower than the position in which they appear. Section 11.5 tackles the question of whether in the case of total reconstruction

186

Aspects of the Syntax of Agreement

we are dealing with copy deletion or actual lowering. I provide evidence from intervention effects that seems to show that literal lowering is to be favored. Section 11.6 revisits the ‘partial negation’ sentences brought to bear on A-movement reconstruction by Chomsky (1993), and shows how the lowering analysis of section 11.5 accounts for the full range of data. Section 11.7 summarizes the main results of this study, and suggests that Lasnik’s conclusion that A-movement does not leave a copy can be maintained.

11.2. IS THERE A-MOVEMENT RECONSTRUCTION? In this section, I introduce the core data that led Chomsky and Lasnik to the claim that A-movement does not reconstruct. I also review the major analyses proposed to account for this state of affairs. Chomsky’s revamping of his (1955) copy theory of movement in the Minimalist Program (Chomsky 1993, 1995) offers an elegant account of interpreting syntactic objects in a position they do not occupy at the surface (Spell-Out) without any form of ‘backward’ movement. Chomsky (1993) gives convincing evidence in favor of this view on reconstruction for A-bar movement,1 with complementary deletion in the Operator-variable sequence (chain). As for A-movement, Chomsky notes the absence of reconstruction effects in general, and suggests that the copies left by A-movement, unlike those of A-bar movement, are ignored by interpretive mechanisms. Chomsky gives four arguments in favor of his claim. But he offers only one really strong argument in favor of not leaving copies with A-movement, and that is the one which will concern us here. He revives the old observation that every cannot take scope under negation in (1b) (Chomsky offers no explanation for why it can in (1a), the so-called partial negation reading): (1)

a. (It seems that) everyone isn’t there yet. b. Everyone seems not to be there yet.

( >> ¬; ¬>> ) ( >> ¬; *¬>> )

Lasnik (1999c) notes that the same absence of reconstruction effects is manifested in the case of object-raising. He motivates his claim on the basis of pseudogapping examples, which he takes to involve overt object raising (Lasnik 1995a and subsequent work): (2)

Mary proved every Mersenne number not to be prime, and John will every Fibonacci numberi [prove ti not to be prime]. ( >> ¬/*¬ >> )

Lasnik observes that absence of scope-reconstruction in the case of A-movement is also clear in subject-to-object/ECM contexts, which Lasnik and Saito (1991), based in part on previous work by Postal (1974), have shown to manifest overt raising to object position:

Scope Reconstruction and A-Movement (3)

187

John proved every defendanti [ti not to be guilty] during hisi trial. ( >> ¬; *¬ >> )

However, Lasnik (1999c) observes that the conclusion regarding ECM-contexts may be too hasty. There is indeed no reconstruction/scope ambiguity effect in the contexts which Lasnik and Saito discuss (anaphor binding, negative-polarity-item licensing, etc.; see Lasnik 1999c for data and discussion), but this is not the case in all ECM contexts. Thus, most speakers detect some ambiguity in sentences like (4): (4)

a. I believe everyone not to have arrived yet. ( >> ¬/¬ >> ) b. John proved every Mersenne number not to be prime. ( >> ¬/¬ >> )

The facts in (4) constitute an interesting challenge: they either question the absence of reconstruction with A-movement (leaving (1b)–(3) unexplained), or indicate that raising is not obligatory in ECM cases. Lasnik (1999c, 2001b) notes that there is some evidence in favor of the latter conclusion. After all, what Postal and Lasnik and Saito show is that raising is available, not that it is obligatory. Indeed, Lasnik observes that in cases where the surface word order leaves no doubt as to whether object raising has taken place, scope is unambiguous. This is the case in (2)–(3), and it is also the case in a construction first discussed in Kayne (1985) involving the raising verb make . . . out (see Den Dikken 1995, Safir n.d., and Lasnik 1999c for recent discussion of the construction): (5)

John made Billi out [ti to be a fool].

Make . . . out provides an excellent test for (unambiguous) raising, if we assume with Johnson (1991) and Lasnik (1999c, 1999d) that the particle out occupies a fixed position. As with all particle verbs (see (7)), overt raising of the pronouns is forced with make . . . out (6): (6)

a. Mary made himi out [ti to be a fool]. b. *Mary made out [him to be a fool].

(7)

a. Mary picked himi up ti. b. *Mary picked up him.

Also, as with other particle verbs, raising of full DPs/NPs is optional with make . . . out (though some speakers have a marked preference for overt raising, as indicated by the % sign): (8)

a. Mary made Billi out [ti to be a fool]. b. %Mary made out Bill [to be a fool].

188

Aspects of the Syntax of Agreement (9)

a. Mary picked Johni up ti. b. Mary picked up John.

As the following examples indicate, make . . . out is a perfect test for overt object raising. If the NP/DP appears to the right of the particle, as in (11), it means that it has not raised, and thus does not show the behavior discussed in Postal (1974) and Lasnik and Saito (1991): (10) a.

The DA made the defendantsi out [ti to be guilty] during each other’s trials. b. The DA made no suspecti out [ti to have been in New York] during his trial.

(11) a.

*?The DA made out [the defendants to be guilty] during each other’s trials. b. *?The DA made out [no suspect to have been in New York] during his trial.

This strongly suggests that Johnson (1991) and others were right in taking the position of the particle to be fixed, signaling overt raising. The crucial test for reconstruction consists in examining scope relations in the make . . . out construction, to see whether scope is ambiguous when the embedded subject appears to the left of the particle, indicating raising. If it is, we have a rather strong argument for A-movement reconstruction. If it isn’t, we have a strong argument against A-movement reconstruction, and for optionality of object raising: (12) The mathematician made every even number out not to be the sum of two prime. ( >> ¬/*¬ >> ) (13) The mathematician made out every even number not to be prime. (? >> ¬/¬ >>) As the data in (12)–(13) show, when raising has taken place, scope is unambiguous, which indicates that the word order optionality manifested in the make . . . out constructions reflects a more general optionality, namely, the optionality of subject-to-object raising of full DPs. Note that this is not the best possible state of affairs for the strongest minimalist thesis, whose core notion of ‘Last Resort’ bans any kind of (pure) optionality. However, since the issue of optionality is tangential to the argument made in this chapter, I will refrain from discussing it here. The facts just reviewed led to the claim that A-movement does not reconstruct. Chomsky and Lasnik’s facts indeed seem to point at the lack of reconstruction with A-movement, but, though suggestive, the evidence is not overwhelming, and massive evidence against scope reconstruction

Scope Reconstruction and A-Movement

189

(‘Quantifier Lowering’) can be found in the literature. The most famous argument for A-movement reconstruction is given in May (1977), based on (14): (14) Some politician is likely to address John’s constituency. (some >> likely/likely >> some) May characterizes the ambiguity in (14) as follows: [(14)] may be taken as asserting either (i) that there is a politician, e.g., Rockefeller, who is likely to address John’s constituency, or (ii) that it is likely that there is some politician (or other) who will address John’s constituency. (p. 189) I will return to May’s characterization of the ambiguity of (14) below. For now, I will adopt it at face value. Importantly, Chomsky (1993) does not deny the existence of lowered readings like the one in (14) (see also Chomsky 2001: 36 fn. 12). He argues that in such cases, the lowered reading of the A-moved element could result from adjunction of the matrix quantifier to the lower IP (ccommanding the trace of raising and yielding a well-formed structure if the trace of quantifier lowering is deleted, along the lines of May’s original proposal). But reconstruction in the A-chain does not take place, so it appears. (p. 327) Hornstein (1995) claims that the theory would be more elegant if any kind of reconstruction were handled by ‘activation’ of a copy, as in the case of A-bar movement. Lasnik agrees with Hornstein, but unlike him, does not argue against Chomsky’s claim that A-movement does not reconstruct. Apart from the object raising evidence already described, Lasnik notes the following example: (15) Every coin is 3% likely to land heads. ( >> likely; *likely >> ) Lasnik notes that the situation in (15) strongly biases the sentence toward the lowered reading (likely >> every), but that reading still is not possible. That is (for Lasnik), (15) cannot be accurately paraphrased as (16): (16) It is 3% likely that every coin will land heads. Facts like these, and others to be discussed shortly, lead Lasnik to claim that even the process of literal lowering that Chomsky allows to capture the lowered reading in (14) and similar sentences is not available. For Lasnik, the apparent lowered reading of the indefinite NP stems from another

190

Aspects of the Syntax of Agreement

source, viz. the vagueness (specific/non-specific ambiguity) of indefinites. In the same breath, Lasnik proposes that the absence of reconstruction with A-movement is explained by the hypothesis that A-movement does not leave a copy.2 In this, Lasnik differs also from Fox (1999), who notes the absence of A-movement reconstruction, and proposes (p. 192), much in the spirit of work in the EST framework, that A-movement (optionally) leaves a simple trace, not a copy. Fox’s solution suffers from the same defect as Chomsky’s, in that it weakens the view that there is no process of reconstruction, as argued in Chomsky (1995). Moreover, it is not explanatory, because it does not tell us why certain elements (A-bar elements) can leave a full-fledged copy, while others (A-elements) leave a simple trace. Things could have just as well been the reverse. Finally, Fox’s proposal violates the Inclusiveness Condition of Chomsky (1993, 1995), which bars the introduction of elements extraneous to the numeration in the course of the syntactic computation. While copies satisfy Inclusiveness, traces clearly don’t. This, coupled with the hybridity Fox’s system introduces into CHL, seems to me to be sufficient to disregard his alternative. Lasnik’s solution is less stipulatory than it appears at first sight. It is motivated by Chomsky’s (1993) claim that reconstruction is a property of operator-variable (i.e., A-bar) chains. Lasnik notes that A-bar movement typically creates an operator-variable relation, so at least two elements (the moved element and a copy) are necessary. For A-movement, the copying operation appears to be a theoretical excrescence inherited from GB-theory and, more particularly, from the Projection Principle.3 There are not two distinct interpretive roles for an A-moved element and its copy to fulfill, so copying in this case does not follow from ‘virtual conceptual necessity.’ Chomsky, however, parried that argument as follows: “In the phonological component, traces [copies] delete. We have found no reason to extend the convention to the NnM computation, and indeed cannot; were we to do so, R-positions would be invisible at LF” (p. 301). Chomsky here (and elsewhere) adopts the configurational approach to theta-role assignment (see Hale and Keyser 1993 et seq.). There is, however, an alternative conception of thematic relations, viz. one that views thetaroles as ‘features’ that need to be checked in the course of the derivation. By contrast, Lasnik, following work by Bošković and Takahashi (1998), Lasnik (1995a), and Hornstein (1999), among others, takes R-roles to be features, which renders A-traces deletable. I will not question Lasnik’s view on R-roles, but rather will examine his argument based on failure of paraphrasability to motivate the absence of reconstruction with A-movement. I will show that Lasnik’s cases are either inconclusive or incorrect, and that some instances of A-movement exhibits reconstruction effects. To sum up, we have seen how Chomsky and Lasnik interpreted data from both subject and object raising as evidence for the absence of reconstruction with A-movement. I have reviewed Fox’s and Lasnik’s explanations for this state of affairs, and provided arguments that Lasnik’s explanation

Scope Reconstruction and A-Movement

191

is superior. The issue now is: Are there really no reconstruction effects with A-movement?

11.3. ON THE VALIDITY OF QUANTIFIER LOWERING AS A MEANS OF RECONSTRUCTING A-MOVEMENT In this section I examine Lasnik’s argument against standard tests for A-movement reconstruction based on paraphrasability (11.3.1.) and control versus raising (11.3.2.). Subsection 11.3.3 introduces additional data relevant to the issue at hand.

11.3.1. Paraphrasability The first test I discuss is paraphrasability: Can sentences involving A-movement be paraphrased in such a way that the A-moved element appears in its thematic position? Despite Lasnik’s claim to the contrary, I conclude that paraphrasability is not entirely relevant as a test for A-movement reconstruction. Recall that May’s (1977) argument for A-movement reconstruction, i.e., Quantifier Lowering, rested on the paraphrasability of (14) as ‘it is likely that some politician (or other) will address John’s constituency’. Lasnik notes that many examples involving indefinites are not paraphrasable in that way. Some of Lasnik’s examples are reproduced under (17)–(20)4: (17) No large Mersenne number was proven to be prime. (18) #It was proven that no large Mersenne number is prime. (19) No one is certain to solve the problem. (20) #It is certain that no one will solve the problem. Lasnik’s examples indeed show that where we expect to find reconstruction, we do not. However, a few notes are in order here. Crucially, the examples contain no one, or some variant thereof. No one is a weak NP (e.g., it appears in existential sentences), and we will see that weak NPs typically exhibit reconstruction effects. However, as Kim (2000) observes, no one is also unlike other weak NPs in that it contains a negation operator. Put differently, no one is a complex expression, consisting of an operator (no) and a variable (one). It is possible that in (17)–(19), only one reconstructs, with no remaining in its surface position. That would mask any reconstruction effect, and definitely prevent paraphrases like those in (18)–(20). I will come back to how this fact might be instantiated formally, but at this point I want to show that although the absence of reconstruction is uncontroversial in

192

Aspects of the Syntax of Agreement

(17)–(20), ‘external’ factors might influence this fact. Likewise, Lasnik’s crucial example (15) involves the predicate 3% likely. This seems quite close to the typical raising verb likely, but it is again unclear whether they are exactly the same. In fact, it seems to me that adding a degree modifier changes the syntax of the construction in a way that makes it look more like control than raising, with subsequent loss of reconstruction possibilities. Some independent facts seem to point to that conclusion. Consider the following paradigm from Lasnik and Saito (1992): (21) a. John is likely to win. b. There is likely to be a riot. c. Advantage is likely to be taken of John. (22) a. How likely to win is John? b. *How likely to be a riot is there? c. *How likely to be taken of John is advantage? Following Kroch and Joshi (1985), Lasnik and Saito (1992) conclude that likely is a control predicate in (22). As we know, expletives and portions of idioms cannot control, hence the ungrammaticality of (22b,c). Interestingly, scope with how likely is fixed in a way that is not when likely is used: (23) [How likely to win the lottery] is someone from New York? (someone >> likely/*likely >> someone) Although I defer a possible explanation for the forced wide scope in Lasnik’s examples, I want to stress at this point that it is quite likely that we are not dealing with genuine raising predicates, and that this might be the factor eliminating reconstruction possibilities. When we turn to typical raising predicates, lowered readings emerge. Put differently, Lasnik’s crucial examples contain factors that once understood and controlled for might show that A-movement reconstructs. Consider now (24). (24) Someone from New York is likely to win the lottery. Lasnik’s point that we might be dealing here with the ‘vague’ nature of indefinites is well taken. Semanticists readily agree that the reliance on the de dicto/de re type paraphrase, as May does, is not the strongest kind of argument for reconstruction. More interesting is the fact that in (24), there is a truth-conditional difference between the two readings, which can be strengthened by simply taking the strongest modal, as in (25) (a suggestion I owe to Bill Ladusaw p.c.): (25) Someone from New York is guaranteed to win the New York lottery.

Scope Reconstruction and A-Movement

193

(26) It is guaranteed that someone from New York will win the lottery. Assuming, counterfactually, that the New York lottery is open only to New York residents and that the rules are that numbers are drawn until there is a winner, then it will be true that ‘it is guaranteed that someone from New York will win the lottery, but false that ‘there is someone from New-York who is guaranteed to win the lottery.’ These are the kind of cases that seem to lend themselves to the conclusion that one needs to allow interpretation of the subject in the lower clause. In fact, for this one case, the lower construal is the more salient reading. Here we are dealing not just with lexical ambiguity, since there is an intervening operator that makes a truth-conditional difference. Similar cases are readily available: (27) No two people are likely to choose the same password. (it is likely that no two people will . . .) (28) Someone seems to be absent from class. (it seems that someone is absent from class) (28) is a clear case of ambiguity for most semanticists, though it is admittedly hard to say how it is truth-conditional, verging as it does on the de dicto/de re ambiguity. It can be strengthened with a negative: (29) No one seems to be absent from class today. (it seems that no one is absent from class today) (29) is most plausibly construed as saying that it seems that no one is absent from class, and is not a strong quantificational claim over people. The latter reading can be made to surface by putting a polarity item in the upper clause—a trapping effect of the type discussed extensively by Fox (1999) and Romero (1998) (see also Lebeaux 1988 through 1998 for original insights) (I return to trapping effects later): (30) No one seems to anyone to be absent from class today. It seems to me that cases like these militate in favor of A-movement reconstruction.5 What Lasnik seems to have shown is that we do not understand all the factors influencing the availability of scope interpretations. The observed possibilities seem to be even more restricted than we thought. I have shown that refining some of the standard examples help bring out the reconstructed reading. In the next subsection, I turn to another argument of Lasnik’s, which capitalizes on the alleged absence of lowered readings with control. Here too I show that the argument does not quite hold.

194

Aspects of the Syntax of Agreement

11.3.2. Control Versus Raising This section addresses the claim made by Lasnik that contrasts between control versus raising, especially those based on paraphrases, are irrelevant to the issue of A-movement reconstruction. Again, I show that this claim is incorrect. Lasnik (1998a, 1998b, 1999c) replicates the well-known paradigm from May (1985), intended to show the contrast between raising (31) and control (32): (31) A unicorn is likely to be apprehended. (it is likely that a unicorn will be apprehended) (32) A unicorn is eager to arrive. (#/*it is eager that a unicorn will arrive) May claims that raising gives rise to reconstruction possibilities, while control doesn’t. The raising versus control issue is particularly relevant in the present context, given the claim put forth by Hornstein (1999) that control is A-movement. Although Lasnik does not deny the contrast between (31) and (32), he denies its relevance for A-movement reconstruction. He notes that the failure of paraphrase in (32) may not be due to the absence of a lowered reading, but simply be a violation of the theta-criterion. Lasnik’s point is well taken, but it seems to me that Lasnik’s claim is equivalent to the claim that (32) does not have a lowered reading because it is a control structure. As already said, paraphrasability, or failure thereof, is not the best kind of evidence for reconstruction. And here too, we can find a better test.6 Thus, the pair in (33) are contradictory, while the pair in (34) are not: (33) a.

??Even though there are no unicorns, Mary found them in her garden. b. ??Unicorns are fictional entities, but out there a unicorn is eager to come in.

(34) a.

Even though there are no unicorns, yet a unicorn seems to be approaching. b. Unicorns are fictional entities, yet a unicorn seems to be approaching.

One can also note, as does Chierchia (p.c.), that coordination of extensional verbs forces a specific reading, while coordination with an intensional verb does not: (35) A unicorn walked in and coughed.

Scope Reconstruction and A-Movement

195

(36) A pair of Calvin Klein jeans seemed to be much wanted by John and were finally bought by Bill. This contrast shows that there is a genuine difference between raising and control regarding reconstruction. Here too, raising seems to point to the need for A-movement reconstruction. In fact, there is one ‘control’ context where the paraphrase does not violate the theta-criterion. And crucially, the lowered reading is absent: (37) Many people are tough to please. (y it is tough to please many people) Regardless of how we treat tough constructions (null operator relation, movement from theta-position to theta-position, etc.), we are dealing here with a predication relation that is quite similar to control. And, as pointed out in Epstein (1989), there is no scope reconstruction in tough constructions. I find this fact significant. In the next section, I turn to further cases of reconstruction with A-movement.

11.3.3. Reconstructing Quantifiers As the attentive reader will have already noted, there does not seem to be any case where the ambiguity is clearer or the tendency to interpret the raised NP below is stronger than with indefinites. This fact is clearly in need of an explanation, and it will be my goal in the next sections to provide one. However, I would like to point out that one can find examples of what look like reconstruction effects with strong quantifiers having undergone A-movement. Consider the following (Mats Rooth p.c.; attributing the example to Dorit Abusch): (38) Everybody who shows up is likely to be a psycholinguist. (39) The one who will show up is likely to be a psycholinguist. (38) shows that a present tense in the raised subject can have a future interpretation, as if compositionally it were in the scope of the future-oriented raising predicate. As a matter of fact, this is precisely what Abusch has argued. She claims that the complement has a futurity operator that interacts with the matrix tense. On her analysis, the subject would have to have scope in the complement at LF to get the intended reading.7 One should also mention the fact that reconstruction does not seem limited to genuine indefinite descriptions, as the following examples involving negative and non-monotonic quantifiers make clear: (40) Nobody is believed to be in the reactor room.

196

Aspects of the Syntax of Agreement (41) Exactly one person is likely to get an offer.

To recap, I have shown on the basis of carefully chosen examples that there is a need for A-movement reconstruction. I have not shown that lowered readings have to be analyzed as activation of a copy. In other words, Chomsky’s (1993) claim that lowered readings are the result of literal lowering thus still holds. However, I have at least shown that one cannot deny the presence of lowered reading as Lasnik does. More than the mere vagueness of indefinites is at stake.8 Having shown the need for A-movement reconstruction, one still has to account for the numerous cases where reconstruction is absent. In particular, one has to account for the contrast we started with (repeated here): (42) Everyone hasn’t arrived yet. (43) Everyone seems not to have arrived yet.

( >> ¬; ¬>> ) ( >> ¬; *¬>> )

One also has to account for the weak versus strong quantifier reconstruction asymmetry, as well as for cases where reconstruction with an indefinite does not obtain. This section has shown that although the vast majority of cases of reconstruction with A-movement involve indefinites, reconstruction with A-raised strong quantifiers sometimes does obtain.

11.4. CASE AND SCOPE Chomsky claims that “reconstruction in the A-chain does not take place” (1995: 327), but does not offer any explanation for why that is the case. We saw that Fox’s hypothesis that A-movement (optionally) leaves a simple trace, not a copy, is undesirable. Lasnik’s claim that A-movement does not leave a trace or copy might still be adequate, provided all cases of lowered readings are the result of literal lowering as proposed by May and Chomsky), but even if that turns out to be true, one still has to explain why only indefinites lower. For now, let us assume that A-movement leaves a copy, and we will return later to the issue of whether the account can be reconciled with Lasnik’s claim. The goal of this section is to offer a plausible reason for why in many cases A-movement fails to reconstruct. The starting point for my proposal is an observation that Fox (1999: 193, fn. 57) attributes to Irene Heim and David Pesetsky, namely, that “the necessary stipulation about A-movement could be derived from an assumption that has an air of an explanation to it, namely, the assumption that copies must receive Case.” That A-traces and A-bar traces differ in terms of Case is familiar from the GB-literature, but it is hard to see why Case would be the relevant factor in allowing

Scope Reconstruction and A-Movement

197

reconstruction. The oddity, however, disappears when we take Chomsky’s (1995) claim seriously that Case is an uninterpretable feature. My proposal is that Case checking renders the element visible for interpretation, where ‘visible’ is used in the sense of Aoun’s (1979) and Chomsky’s (1986b) visibility condition (see also Longobardi’s 1994 N-to-D raising). The original visibility condition says that Case-checking makes an argument visible to the theta-criterion. I generalize the visibility condition and claim that Casechecking marks an element as interpretable, not just for thematic purposes, but also for notions like scope. The idea behind my reasoning is that as long as an NP has an unchecked Case feature, its feature set is uninterpretable. Once Case is checked, the elements are freed for interpretation.9 The Case checking condition on scope taking is meant to account for why A-moved elements take scope in their surface, Case-checking position. Unfortunately, the claim that Case position correlates with scope taking cannot be maintained in its simplest form. The reason is that Chomsky (2000), building upon Chomsky (1995), argues that Case/K-features do not necessitate category raising to the surface position, but can be checked via longdistance matching (Agree; a reworking of feature movement). If so, we can no longer maintain that scope is determined where Case is checked, with Case now being eliminable in situ.10 Further, as pointed out by a reviewer, the earliest minimalist assumption is that checking of a feature takes place in all the members of the chain, not just in the copy that happens to be in the checking configuration. If so, all the copies should be visible for interpretation after checking, which would amount to saying that reconstruction is possible, and no explanation would be available for why reconstruction fails in so many cases. Clearly, saying that there is a preference for interpreting the highest copy at LF (by analogy with the preference for pronouncing the highest copy of a chain at the PF side) is not explanatory. One would still have to explain why in the case of A-bar chains, interpretation is more liberal than pronunciation. Fortunately, I believe that the problem for the Case account can be solved. Chomsky (1995) distinguishes between checking and erasure: A feature can be checked, and yet remain accessible for further computational purposes. Once it is erased, it is no longer available to the syntax. Although Chomsky (2000) tries to eliminate the checking/deletion–erasure distinction, the latter is maintained precisely in the case of Case. In Chomsky (2000), Case is said to make an element visible for Attraction. In the case of successive cyclic movement, Case has to remain available. I take that to mean that Case is expunged after the element has reached its final landing site (at least in the case of A-movement). The account works technically, but relies on notions such as deletion and erasure that might not follow from virtual conceptual necessity. It too much resembles the coding tricks that Minimalism precisely seeks to eliminate. However, here too, a slight reworking appears to be possible. Pesetsky and Torrego (2001) capitalize on the body of work initiated by George and Kornfilt (1981), who first showed that there is an

198

Aspects of the Syntax of Agreement

intimate relationship between nominative Case and finiteness. George and Kornfilt’s paper was also the point of departure for Chomsky’s claim that Case and K-features go hand in hand. It seems overwhelmingly true crosslinguistically that nominative is the Case the element associated with Tense receives when endowed with K-features. Pesetsky and Torrego give an interesting twist to George and Kornfilt’s insight. Whereas Chomsky lumps Case and K-features together (while nonetheless distinguishing between them), Pesetsky and Torrego argue that Case is really an uninterpretable Tensefeature. By uninterpretable, they actually mean something different from Chomsky (1995). For them, nominative Case is an inherently interpretable Tense-feature that happens to be on the ‘wrong’ element (a noun, and not a verb), which makes the feature ‘uninterpretable’. What Pesetsky and Torrego mean is that uninterpretable features are actually misplaced features. The shift from uninterpretable to misplaced might appear innocuous, but it actually is not. In particular, for our purposes, misplaced features are those that are interpreted (rather than merely ‘interpretable’) in the position where they are put back into place. The features in question are interpretable per se, but they are interpreted on just one occurrence, the highest one in the case of A-movement. Crucially, misplacement and replacement are quite different from checking. The latter affects the whole chain, whereas replacement targets one particular position. This is precisely what we need, since we want just one position (the highest occurrence) to be actually interpreted.11 There is an interesting parallel to draw here between ‘Scope freezing’ (i.e., no A-reconstruction) and ‘Case-freezing’, the generalization that once Case is checked, the element stays put (Lasnik 1995b, Chomsky 1995, Chomsky 2000). In more concrete terms, there seems to be an interesting parallelism between (44) and (45) (cf. (46)): (44) Every coin is 3% likely <every coin> to <every coin> land heads.  Case n interpretation (45) John is likely <John> to <John> be intelligent.  Case: final landing site n PF-interpretation (46) *John is likely <John> is intelligent.  Case n frozen

Hyperraising

The present proposal that ties scope interpretation and Case-checking (feature replacement, i.e., interpretation of features) allows A-movement to be much like A-bar movement in terms of the copy theory. A-movement can leave a full copy and Case will prevent the interface from using members of the chain other than the head, since only the highest occurrence is the one

Scope Reconstruction and A-Movement

199

where all features appear to be in place. Besides allowing us to maintain the copy theory in its simplest form (i.e., that all movement leaves a full copy), this theory makes an interesting prediction: Although Case forces the head of the chain to be interpreted in the case of A-movement,12 it does not say anything about elements that are pied-piped under A-movement. Take the case of a relative clause complement or an of phrase. It is standardly assumed that such elements check their Cases NP-internally, not after A-movement is completed. This means that such elements are not frozen in the final position occupied by the A-moved element; rather, they are accessible for interpretation upon merger. They are therefore expected to give rise to reconstruction or connectivity effects. Examples like (47) and (49) bear out that prediction: (47) Everybody who shows up is likely to be a psycholinguist. (48) Everybody who shows up is likely <everybody who shows up> to be <everybody who shows up> a psycholinguist. (49) Pictures of himself frighten John. (50) Pictures of himself frighten John . Such facts seem to lend rather strong support to the present analysis. Note that reconstruction here is much as in the case of A-bar movement a function of the copy theory, not of literal lowering. To sum up this section, I have argued that A-movement does not generally give rise to reconstruction effects because the A-moved element has to wait until all its features have been ‘replaced,’ which happens in the highest occurrence. Only there does the whole feature set of the NP become visible for interpretation. The result of that is that A-moving elements are both scope and Case-frozen.13 Support for the present analysis comes from partial reconstruction facts. These are cases where portions of the A-moved elements reconstruct whose features have been checked before being pied-piped. Having accounted for why A-movement does not reconstruct, we now have to face cases where A-movement does reconstruct, that is, instances of A-moved indefinites.

11.5. CAPTURING RADICAL RECONSTRUCTION This section concentrates on the fact that in the vast majority of cases only weak NPs reconstruct under A-movement. I show that two recent accounts for this are inadequate, both conceptually and empirically. I then turn to my own proposal, which relies on the possibility of covert insertion of an expletive in just those cases where A-movement gives rise to reconstruction.

200

Aspects of the Syntax of Agreement

As we saw earlier, indefinites seem to stand on their own when it comes to A-movement reconstruction. In contrast to other quantifiers, indefinites appear to give rise to what one might call radical reconstruction, that is reconstruction of the head of the A-chain. Consider the following example again: (51) Someone from New York is likely to win the lottery. It appears that one reading of (51) has someone (from New York) under likely: (52) e is likely <someone from New York> to win the lottery. There are two questions to address in this case. One is how come indefinites do not exhibit freezing effects. The other is how to account for the lowered readings of indefinites. Should it be done via copy-deletion, as in the case of reconstruction with A-bar movement and partial A-movement reconstruction, or via literal lowering as in Chomsky (1995) and, ultimately, May (1977). In some sense, reconstruction with indefinites is different from other cases of reconstruction in that other cases of reconstruction, be they instances of A-bar movement or of A-movement (the partial reconstruction cases in (47)–(49)), the operation always leaves something in the head position of the chain that they form. Thus, A-bar reconstruction typically leaves the operator in SpecCP, while A-movement reconstruction is partial as well, as we saw in the previous section. The fact that reconstruction is total in the case of indefinites poses a problem for a hypothesis that has received considerable support from the literature. That is Heim and Kratzer’s (1998) suggestion to treat movement as creating a derived predicate, with a lambda-abstract being formed. For reasons of semantic simplicity, Heim and Kratzer follow Cooper (1979) and suggest that movement creates a predicate, which then applies to the moved constituent. On this view, movement that targets YP, at least XP movement, creates two new syntactic constituents above YP. The lower one of these (YPb) corresponds to a functional abstract over the variable left in the position that XP originated from. The higher one (YPr) is interpreted as the result of the function YPb applying to the meaning of the moved constituent XP. This is summarized and schematized in (53) and (54), respectively: (53) Movement: When XP moves from a position inside YP to the sister position of YP, the following operations take place: a. XP is replaced with a variable x b. A new node YPb is formed with the two daughter nodes YP and MP c. A new node YPr is formed with the two daughter nodes YP and XP

Scope Reconstruction and A-Movement (54)

201

YPr ru XP1 YPb ru M1 YP 6 . . . t1 . . .

This view of movement has received considerable support from Sauerland (1998a, 1998b), and Nissenbaum (1998). I hasten to add that one need not reject the Principle of Inclusiveness (see earlier discussion) if one adopts (53). That is, one need not take (53) as applying in the syntax per se; it might be the way the interface translates syntactic movement, which is the view I will adopt here.14 Still, one should not allow for semantically driven operations like Quantifier Movement that would ultimately run afoul of (53). The problem with radical reconstruction is that it would leave the lambdaabstract unbound, violating the Proper Binding Condition, i.e., the requirement first outlined in Fiengo (1977) that traces must be bound.

11.5.1. A PF Analysis and Its Problems The first account of radical reconstruction I discuss is the one proposed by Sauerland and Elbourne, which capitalizes on the possibility of movement in the PF component to give the illusion of reconstruction. The Proper Binding problem was independently noted by Sauerland (1999a) and Elbourne and Sauerland (1999).15 Their solution consists in having EPP-driven movement apply in the PF component. Under that view, there is no need for reconstruction, whether it be copy-activation or literal lowering. For all syntactic and semantic purposes, the moving element is treated as if in situ, as is represented in (55): (55) [Someone from New York] is likely <someone from New York> m______PF-movement_____________|______________ to <someone from New York> win the game. _____________| Ÿ Spell-Out: LF-visibility Obviously, Sauerland and Elbourne and Sauerland must allow for the option of A-movement in the syntax being able to characterize the high reading of indefinites in raising constructions. This very option poses a problem in the case of Trapping-sentences like (56): (56) Someone seems to himself tb to be t a genius. m___________________|______|

202

Aspects of the Syntax of Agreement

The issue here is how to decide which operation has to apply in which component. That is, how do we avoid looking-ahead? How does the computational system choose between performing a given movement in the syntax (with ensuing semantic consequences) and performing it in PF? In the case of Trapping effects, movement must apply in the syntax, since if it applies in PF, the variable himself will not be bound. Syntactic movement is always an option in Sauerland and Elbourne and Sauerland’s work, but one might want to reduce the option to just one, in light of computational concerns familiar in the minimalist program (see Collins 1997, Chomsky 2000 for discussion). (Similar considerations hold for examples (67)–(68) later, which show that the ability of A-movement to reconstruct is sensitive to intervening quantifiers. I do not see any conceptually appealing way of making PF movements sensitive to quantifier crossing.) The second major problem I can see with a solution like the PF-movement approach is that it flatly violates the otherwise robust generalization that movement in PF, if it exists at all,16 is extremely local, limited to adjacent positions (see Halpern’s 1995 Prosodic Inversion operation, arguably the best motivated PF-driven movement). This is clearly not the case in the examples Sauerland and Elbourne and Sauerland discuss. As far as I can see, their solution is yet another case where movement is shifted to the PF branch because it poses a problem for syntax, not because there are compelling PF reasons for locating it there.17 Given the problems just listed (see also notes 15 and 17), I am inclined to look for an alternative solution to the radical reconstruction problem.

11.5.2. A Semantic Analysis of A-Movement Reconstruction and Its Problems One solution for the reconstruction facts under A-movement that comes to mind has been pursued in unpublished joint work by Sigrid Beck and Kyle Johnson. Beck and Johnson propose that indefinites are ambiguous between an argument reading and a predicate reading. We know independently that predicates obligatorily reconstruct (see Huang 1993, for instance). Beck and Johnson’s solution amounts to saying that when indefinites reconstruct, they are predicates. Their solution might be related to our proposal in the previous sections. Since predicates do not have (structural) Case, they are not frozen, and therefore give rise to reconstruction effects. Though appealing at first sight, the solution suffers from a major problem: It is not realistic from a syntactic point of view to say that some NPs violate the Case-Filter, although they might surface as Case-marked morphologically in languages where Case is overt. There does seem to be Case variation among indefinites across languages (see Mahajan 1990, Enç 1991, De Hoop 1992), but the variation concerns their de dicto/de re distinction, or their specificity distinction, which we saw should not be equated with the pure scope ambiguity we are focusing on here.18 In addition, it is unclear to me how forced reconstruction, as expected

Scope Reconstruction and A-Movement

203

under the predicate-based solution, would account for intermediate reconstruction facts like (57), a variant of which was first discussed in Zubizarreta (1982), and studied recently by Higginbotham (1995) and Lasnik (1998b) (their example have a universal instead of an indefinite subject.) (57) A man is likely not to win the lottery. In (57) the indefinite a man can scope under likely, but cannot reconstruct all the way to its original position. Rather, it obligatorily takes scope over negation and other operators that might appear in the infinitival clause. The situation is different from Huang’s (1993) core data, of which (58) is an example, where predicates reconstruct all the way regardless. (58) *How proud of herself does Mary think that John is. In light of this absence of parallelism between indefinites and predicates, I do not adopt Beck and Johnson’s solution, and turn to my own proposal.

11.5.3. Lowering Reconsidered In this section, I will argue that ‘lowerable’ Quantifiers, that is, those subject to radical reconstruction, are precisely those that can appear in theresentences. (59) and (60) shows this correlation. (59) a. b. c. d.

Someone from New York is likely to win the lottery. Nobody is believed to be in the reactor room. Exactly one person is likely to get an offer. #Every coin is likely to land heads.

(60) a. b. c. d.

There is someone in the garden. There is nobody in the garden. There is exactly one person in the garden. *There is everybody in the garden.

I contend that the way to capture this generalization is to allow for a null counterpart of there to be inserted post-Spell-Out, in the covert component. A late insertion theory has already been put to good use in Bošković (1998a, 2000a) and Bošković and Lasnik (1999) for various constructions. Bošković argues that the framework of Chomsky (1995) allows covert insertion of phonetically null elements. He claims that such an option is realized in the case of French interrogative strategies. Thus, French allows matrix Co to be inserted post Spell-Out, triggering LF wh-movement. Crucially, Bošković and Lasnik go one step further: They note that Chomsky’s (1995) definition of strength (61) has the effect of disallowing acyclic checking of heads with strong features:

204

Aspects of the Syntax of Agreement (61) Suppose that the derivation D has formed 4 containing B with a strong feature F. Then, D is canceled if B is in a category not headed by B.

Bošković and Lasnik suggest (61) as an alternative to the Extension Condition to derive cyclicity effects. What this means is that elements with weak features will be allowed to be inserted acyclically. I will now argue in favor of acyclic insertion in the case of there, whose covert counterpart I will represent as thereLF. Suppose we allow thereLF to be inserted on top of a raised indefinite or other lowerable elements, like negative and non-monotonic quantifiers. Insertion is allowed on grounds made clear in Bošković and Lasnik (1999), and Chomsky (2000), since it takes place at the IP (non-phase) level. Let us furthermore assume that thereLF does not have any strong feature.19 The function of thereLF is to turn the overtly raised quantifier into an associate, and ‘push it down’ the tree, as it were, for purposes of interpretation. Put differently, insertion of thereLF is what allows lowerable quantifiers to undergo radical reconstruction. As is well known from the vast literature on existential sentences, associate NPs obligatorily take narrow scope,20 and resort to a somewhat special mechanism for Casechecking.21 This very peculiar property of associates is, I claim, what allows them not to be Case-frozen. Insertion of thereLF also has the advantage of solving the Proper Binding Condition violation of radical reconstruction discussed earlier.22 ThereLF will act as a binder for the lambda-abstract by virtue of sharing features with the moved quantifier.23 As we can see, insertion of thereLF allows us to explain the peculiar behavior of some quantifiers. To sum up, I have shown that neither a PF-movement analysis nor a purely interpretive solution can account for radical reconstruction. I have argued that the possibility of inserting an expletive in the covert component achieves the desired results. 11.6. COPY AND DELETION OR LOWERING? The remaining issue I would like to touch on is whether radical reconstruction amounts to activation of a lower copy, or literal lowering. That is, which of (62a) or (62b) is the correct representation of the lowered reading of someone from New York?: (62) a.

[There [someone from NY]] is likely <[someone from NY]> to win the lottery. b. [There [someone from NY]] is likely ___ to win the lottery. |______________________m

We will see that an answer to that question will provide a possible explanation for partial negation readings, as well as for contexts where even indefinites do not reconstruct.

Scope Reconstruction and A-Movement

205

I would like to pursue the idea that insertion of thereLF makes English really look like Icelandic at LF. As is well known, Icelandic allows multiple subjects in overt syntax, as illustrated in the so-called Transitive Expletive Construction (TEC), which contains both a expletive and a contentful subject many students: (63) Það hafa sennilega margir stúdentar lesið bókina. there have probably many students read the.book ‘Many students have probably read the book.’ (Bobaljik and Jonas 1996: 212 (20)) Chomsky (1995: 343, 2000) observes that, contrary to what had been thought, similar multiple subject constructions might be available in English. Some examples are given in (64): (64) a. There entered the room a man from England. b. There hit the stands a new journal. c. There visited us last night a large group of people who traveled all the way from India. Chomsky notes that such constructions require ‘heavy,’ arguably extraposed subjects (‘associates’), and speculates that the difference between English and Icelandic might just be a PF phenomenon, forcing themes to appear at the edge in English, but not in Icelandic (due to V2 effects in the latter language). Chomsky further argues that the correct representation of TECs is one involving multiple specifiers of TP, separated by the verb at PF: (65) a.

Tmax ru Expletive Tb ru indefinite Tb n ru v-V Tb ru To VP max b. T ru Expletive Tb ru indefinite Tb ru v-V Tb ru To VP

206

Aspects of the Syntax of Agreement

If what I said about (62) is correct, thereLF creates a configuration identical to (65a). I would push the Quantifier Lowering–TEC parallel further and say that much like overt multiple subject constructions in English force extraposition of the subject at PF, thereLF forces lowering (intraposition) of the relevant quantifier at LF. That is, I am claiming that the lowered reading of quantifiers comes about by movement, in this case, literal reconstruction. Interestingly, there seems to be some evidence favoring literal lowering (reconstruction) over activation of a lower copy. The evidence comes from intervention effects.24 Beck (1996) and, following her, Pesetsky (2000) take the island-like effect in (66) to be indicative of movement of the wh-phrase wo at LF. (66) a.

*Wen hat niemand wo gesehen? who has nobody where seen ‘Where did nobody see who?’ b. *Wen hat niemand wo gesehen? z__________=______m

German

What I would like to argue is that similar intervention effects obtain under A-movement reconstruction. At first sight, one might take Trapping effects to be such a case of intervention, favoring movement over the copy theory. However, Trapping effects have been successfully analyzed under the copy-deletion mechanism (see Fox 1999, Romero 1998). Furthermore, it is unclear why an anaphor would block movement, where by blocking I mean some effect akin to Relativized Minimality.25 More revealing is the contrast in (67)–(68): (67) A red car seems to me to be parked at the corner. (= it seems to me that there is a red car . . .) (68) A red car seems to every driver to be parked at the corner. (*it seems to every driver that there is a red car . . .) (67) is a standard example of raising across over an experiencer, where nothing unexpected happens. By contrast, in (68), the presence of a quantifier in the experiencer position seems to force the high reading of the raised indefinite subject. I would like to claim that the quantifier blocks lowering of the subject. Under a copy-deletion view, it is unclear why the presence of a quantifier would block deletion of the higher copy of the subject. In a derivational framework like the one assumed here, the copy-deletion mechanism is blind to intervention effects (things are slightly different in a framework like Brody 1995, and especially Brody 1999). By contrast, it is natural to assume that the experiencer blocks lowering when it is of the

Scope Reconstruction and A-Movement

207

same class (quantifier). The same effect appears to be found with negation as an intervener: (69) A man is likely not to win the lottery. ( >> likely >> ¬/likely >>  >> ¬/*likely >> ¬ >> ) The blocking effect of negation will become significant in the so-called partial negation readings of the type illustrated in (70): (70) Someone hasn’t arrived yet.

( >> ¬; *¬ >> )

The present analysis thus offers evidence for Chomsky’s (1995) position that the lowered reading of (1a) does not result from copy-activation, but from literal lowering (see also Hornstein 1998, who withdraws from his 1995 conclusion and views QL as literal lowering). The intervention effects just seen might also add to the evidence adduced by Bruening (2001) and Sauerland (1999b) in support of the claim that quantifier movement is subject to Relativized Minimality,26 thereby strengthening May’s original conclusion that scope facts result from syntactic movement. To subsume the intervention effects under QL, one will have to modify the standard definition of Relativized Minimality to encompass cases of lowering. This is easily done. All that needs to be said is that two positions B and C can relate to one another (in the sense ‘raising’ or ‘lowering’) if one (say B) c-commands the other (C) and there is no position H of the same type as (B, C), and H also c-commands C. This is a typical Relativized Minimality formulation, in effect close to Chomsky’s (2000) definition of Agree, which is also a ‘downward’ relationship.27 I will conclude this section by showing how the intervention effects just discussed shed light on the status of experiencer phrases. In particular, we will see how the preceding facts favor Kitahara’s (1997) analysis of raising over an experiencer. The central problem raised by experiencers in raising constructions was recently brought to light in Chomsky (1995: 305f.).28 To allow raising of John, given Relativized Minimality (or any of its minimalist versions), the experiencer cannot c-command (into) the embedded clause. If the experiencer c-commands the embedded subject, it will be attracted, yielding the ungrammatical (71): (71) *Billi seems to ti [John to be a genius]. At the same time, there is evidence that the experiencer c-commands (into) the embedded clause. In (72), him and John must be disjoint in reference, a typical Condition C effect. (72) Theyi seem to himj [ti to like Johnk/*j].

208

Aspects of the Syntax of Agreement

Assuming that there is an irreducibly geometric aspect aspect to binding (see Lasnik (1989) for compelling evidence), the disjoint reference effect in (72) can be captured only if him c-commands John. At this point, we are facing a paradox: The experiencer cannot c-command into the embedded clause so as not to interfere with subject raising, and yet it must c-command into the embedded clause to account for the Condition C effect. Kitahara (1997) addresses the experiencer-paradox, and suggests the following derivational account. To ensure c-command by him and account for the Condition C effect in (72), and at the same time prevent something like himi seems to ti [they to like John] (expected under Closest Attract, if him c-commands into the embedded clause), Kitahara suggests that him does not c-command into the embedded clause prior to raising. He achieves the right configuration by adopting the Larsonian shell in (73) (Kitahara 1997: 64). (73)

TP ru T vmax ru v Vmax ei PP Vb ru ru to him seem TP 6 . . . they . . .

Assuming the standard definition of c-command, the PP node dominates him but not the embedded clause, and hence him does not c-command into the latter. Raising is therefore possible. To capture the disjoint reference effect, Kitahara assumes that by LF him must c-command John. In particular, he suggests that him covertly raises and adjoins to P to check its Case; it is only after this movement that him can c-command into the embedded clause, which forces the disjoint reference effect. Since the experiencer checks its features covertly, it does not bleed any overt operations. Although there are problems with Kitahara’s solution (see Chapter 8 for discussion, and possible solutions), the intervention effect of the experiencer in (62) points to the correctness of such an account: the experiencer does not block overt raising of the subject, but blocks its covert lowering. Another conclusion that comes out of the present analysis is that the representational copy-theory mechanism to capture reconstruction effects is too limited. The thereLF insertion analysis forces us to take the view that there is a distinct LF-component29 (or that QL is a post-cyclic operation in the sense of Chomsky 2000),30 and that syntax is even more derivational than previously thought.

Scope Reconstruction and A-Movement

209

11.7. PARTIAL NEGATION AND FURTHER RECONSTRUCTION FACTS RECONSIDERED Having accounted for basic reconstruction effects with A-movement, let us now go back to the data that motivated Chomsky’s and Lasnik’s conclusion that there are no reconstruction effects with A-movement. I will show that none of the crucial examples are unexpected under the present analysis, as a closer look at the syntax reveals that QL is blocked in all the relevant cases. Let us start with the partial negation facts in (1), repeated here for ease of exposition: (74) a. (It seems that) everyone isn’t there yet. b. Everyone seems not to be there yet.

( >> ¬; ¬>> ) ( >> ¬; *¬>> )

Let us focus on how universal quantifiers can scope under negation when they are ‘clause-mates’ (by which I mean that both relate to the same T-head31), as in (75). (75) a. Everyone isn’t there yet. b. [TP Everyone [T isn’t [there yet]]]. As mentioned above, and in Lasnik (1998a,b, 1999c), the reading where negation scopes over the subject is restricted (for most speakers) to some, mainly universal, quantifiers. This restriction makes it implausible to try to argue for a reconstruction effect in this case (say, reconstruction to the VPinternal subject position along the lines of Hornstein 1995). If reconstruction were involved, one would expect it not to be sensitive to the universal nature of the reconstructed element. If anything, one would expect indefinites to at least pattern the same way as universals, since they are prone to ‘lower’ readings, as we saw earlier. If reconstruction is not involved, how do we account for the scope reversal in (75)? Before extending the syntactic mechanism developed in the previous sections, I will refute the idea that scope reversal in (75) is due to intonation.32 The importance of intonation for scope reversal in a context similar to (75) was already pointed out in Jackendoff (1972), who notes that negation can scope over a subject under a fall–rise contour: (76) All the men didn’t go. (contour: ___fˉˉˉˉh: neg > all) Scope reversal under fall–rise intonation has been studied extensively for German by Krifka (1998). Quite crucially, Krifka notes (p. 82) that “scope inversion in English is more restricted and of a different nature than scope inversion in German.” Krifka’s point is that if intonation were at stake, one would expect many more cases of scope reversal under specific intonation

210

Aspects of the Syntax of Agreement

in English, which is not the case, as scope reversal of the type illustrated in (74)–(75) seems limited to negation. My proposal is that scope reversal of the sort found in (74)–(75) is due to covert raising of negation. Clearly, such a proposal has to address many questions. It has to restrict negation raising to those cases where scope reversal is found (roughly, cases with clause-mate universal subject quantifiers). It has to ensure that negation cannot raise to a higher clause (which would incorrectly allow (73b)), and that it cannot raise across an indefinite, which would, again incorrectly, predict scope reversal in two men haven’t arrived yet. More generally, it has to reconcile the idea of neg-raising with the consensus since Ladusaw (1980) that negation raising should be disallowed, as it would give rise to readings that do not exist. In short, neg-raising has to be extremely restricted. I believe that it is possible to allow for neg-raising in only a limited number of cases in a non-stipulative way. Let us see how. First, I adopt the idea that negation associates with a projection immediately dominating TP, say 4P (see Zanuttini 1997). This is the position I take negation to raise to when it scopes over the universal subject in (74). To block long-distance raising of negation to matrix 4, and thereby correctly widen scope of negation in, say, every boy seems not to have arrived yet, I appeal to ‘superraising.’ As is well known, movement from a Case-checking position to another Case-checking position is disallowed: (77) a.

John seems [tb is t happy]. z___=______g____m b. 4 Everyone seems [4 tb T not to be t there yet]. z_______=____________g_______m

I want to say that movement from 4 to 4 in (77b) is blocked for the same reason as Case-to-Case movement is. The question now becomes why negation can raise over a subject quantifier (an apparent case of superraising) but cannot reach matrix 4 (hyperraising). Although the nature of superraising and hyperraising still lacks a comprehensive discussion (a topic too large for me to go into here, but see Bruening 2001 and Boeckx 2000d for possible avenues of research), I want to suggest the beginning of an explanation for the licit character of neg-raising over subjects, and for the badness of neg-raising over 4. Suppose negation is indeed raising. Assume further that it undergoes head-movement. In partial negation cases (75), movement is allowed to cross the subject quantifier, which, being an XP, does not block head-movement (Rizzi 1990). By contrast, in (77b), a quantifier head (seem) intervenes, which I take to be what blocks raising of negation. Note that if correct, the analysis of (77b) provides another piece of evidence that quantifier-raising is subject to Relativized Minimality.33 I believe that an explanation along similar lines is also available for other cases of unwanted neg-raising. Consider (78)–(79) (typical examples that led Ladusaw 1980 and others to argue against neg-raising):

Scope Reconstruction and A-Movement

211

(78) John {often/usually} doesn’t smoke.

(often/usually > not)

(79) John doesn’t {often/usually} smoke.

(not > often/usually)

These sentences are unambiguous, exhibiting surface scope only. If negation is allowed to raise across the subject to yield the partial negation reading in (75), how do we exclude raising of negation scoping over the adverbs in (78)? My proposal is that neg-raising in (78) is blocked for the same reason as in (77b). In both cases, negation and its targeted landing site are separated by a quantifier head (in this case an adverb).34 Hence, raising is blocked.35 So far, we have been able to restrict neg-raising only to those cases where partial negation is found. One case still has to be examined: the lack of partial negation with indefinite subjects.36 Upon closer scrutiny, one observes that partial negation is available with those quantifiers that do not lower: (80) a. b. c. d.

All the men did not come. Every man did not come. Some boys did not come. Many boys did not come.

(*neg > some) (*neg > many)

There is a nice explanation to be given in this case for the absence of negraising. In (80a,b), neg-raising is the only possible source for the narrow scope of the subject quantifier, quantifier lowering being unavailable. By contrast, lowering is available in (80c,d), and, I argue, is precisely what I take to block neg-raising. Chomsky (1995, 2000) argues that Merge is preferred over Move because Move is a more complex operation (which incorporates Merge). Assume this is correct. Hence the same economy considerations block neg-raising. Recall that in order for the relevant subject quantifiers to ‘lower,’ thereLF is inserted in LF. This is the instance of Merge that blocks Move—in this case, movement of negation. Crucially, we take the view argued most forcefully in Collins (1997) that economy is computed locally. At the point of deciding whether to merge thereLF or move negation, the computational system favors thereLF-insertion. The system does not know that by so doing it will block negation taking scope over the subject quantifier, for once thereLF is inserted, a chain can no longer be established between there and the position where the subject quantifier started. As we saw in section 11.7, negation acts as an intervener for lowering. In a nutshell, thereLF-insertion blocks neg-raising (the preference of Merge over Move), and negation blocks subsequent lowering (Relativized Minimality). The two conditions block any reading where negation takes scope over lowerable subject quantifiers. The analysis I just gave provides rather strong evidence for local economy. Had the computational system have access to global representations, it would have known that favoring Merge (insertion of thereLF) would ultimately block the lower reading for the subject quantifier, and it would have favored raising of negation to reverse scope. Such

212

Aspects of the Syntax of Agreement

knowledge is unavailable under local economy, and we see that this is the result we want.37 If tenable, the present analysis accounts not only for why only universal quantifiers are affected by neg-raising, but also for why the reading is lost across clauses, across adverbs, and across ‘lowerable’ subject quantifiers—in short, in all the cases where partial negation readings are not found.38 To close this section, I will return to some of the cases discussed in Lasnik (1998a, 1998b, 2001b) that show that reconstruction is blocked even in those cases where it is expected to take place. Consider (15), (17), and (19) again, all repeated here under (81): (81) a. Every coin is 3% likely to land heads. b. No one is certain to solve the problem. c. No large Mersenne number was proven to be prime. Let us first concentrate on (81b,c). As pointed out earlier, Kim (2000) has noted that no one and like elements have a dual character, consisting of an operator and a variable. Suppose we say that like other weak NPs, no one can lower, but, in order to provide negation with a standard operator-variable structure, only the variable part (one) lowers. As a result, it is impossible for no one to occur below the relevant predicates (certain, proven). Reconstruction seems to be absent (there is no scope reversal), although it does take place, but only partially: a portion of the subject lowers. As for (81a), we saw that universal quantifiers do not lower generally, but for many speakers lowering is blocked even if a lowerable quantifier is used: (82) 3 coins are 30% likely to land heads.

(?* likely > 3)

What Lasnik failed to consider is that likely and n% likely might differ syntactically. This is what I will capitalize on, arguing that 3% acts as an intervener blocking lowering. In fact, we see that the presence of a degree modifier generally blocks lowered readings. Consider the following contrast: (83) a.

Someone from New York is likely to win the lottery. (likely > some) b. [How likely to win the lottery]i is someone from New York ti. (someone >> likely/*likely > someone)

Drawing on Lasnik and Saito’s (1992) conclusion, who in turn build upon work by Kroch and Joshi, Martin (1992) claims that in (83b) we are dealing with a control structure, which we know independently disallows lowering. Like Bobaljik and Wurmbrand (2000), Sauerland (1999a) rejects this account on the grounds that there are good reasons to believe that the control/raising ambiguity is inadequate to capture the contrast under discussion: likely is always a raising predicate. Assume that this is correct.

Scope Reconstruction and A-Movement

213

There is still another explanation for (84b), and, by extension (83a). It is not implausible to assume that how is some sort of degree quantifier that behaves like quantificational experiencers or negation discussed in section 6, and block quantifier lowering upon thereLF insertion. If this is correct, it is predicted that the lower reading for the subject quantifier is absent in (83a). The account just given predicts that it will be impossible for there to be inserted overtly in the context of n% likely, as the presence of a degree head would block the Agree relationship between there and the associate NP. The prediction is borne out, as (84) shows: (84) *There is 30% likely to be a man in the garden. In sum, Lasnik’s crucial examples in (81a) might indeed show the absence of reconstruction, but for a reason that he had not apparently suspected: that n% likely is not really the same as likely. The merit of Lasnik’s study is to have shown that syntax places far greater constraints on interpretation than May and much work following him had suspected. Such a conclusion I think the vast majority of syntacticians would be willing to endorse.

11.8. CONCLUSIONS AND POSSIBLE EXTENSIONS To conclude, I have examined in some detail the claim made in Chomsky (1995) and strengthened in Lasnik (1998a, 1998b, 1999c) that A-movement does not reconstruct. I have shown that at least Lasnik’s position cannot be maintained, and that in various cases A-movement does reconstruct. I have argued that A-movement leaves a full copy (contra Fox 1999), but that this copy is not what gives rise to lowered readings of quantifiers. Rather, the latter are the result of literal lowering, a possibility rendered possible by covert merger of there. I have argued that failure to reconstruct in the case of A-movement is the result of the relevant elements bearing a Case feature, making them uninterpretable in sites other than the one where this Case feature is expunged. This is what I called the ‘Scope-freezing effect’ of Case-checking. I have shown that in the case of indefinites, this effect can be obviated by inserting an expletive in the covert component, which allows a lower copy of the quantifier to be activated/interpreted (an phenomenon akin to the cases investigated by Hornstein and Uriagereka under the label ‘reprojection’).39 Such a reconstruction operation was shown to be subject to Relativized Minimality effects, which militates for a syntactic treatment of scope assignment (i.e., Quantifier Movement). Indeed, it militates for a derivational account of syntax at large. I showed how the account successfully captures the partial negation facts that motivated Chomsky’s claim, as well as the crucial examples provided by Lasnik. The account provided for those cases reinforces the view that quantifier raising is subject to relativized

214

Aspects of the Syntax of Agreement

minimality, and that a local computation of economy is superior to its global alternative. In concluding, I would like to point out that the present picture is not altogether incompatible with the idea that A-movement does not leave a trace or copy (Lasnik 1998a, 1998b, 1999c). After all, the best cases of A-movement reconstruction have been argued to result from literal lowering. The cases of partial reconstruction (with non-lowerable quantifiers) might be captured via semantic reconstruction (chain-binding). Indeed, Akmajian (1972) had already claimed that not all connectivity effects are the result of movement. Subsequent work by Barss (1986), Cresti (1995), Ceccheto and Chierchia (1999), Lechner (1997), and Sharvit (1997), among others, have strengthened that conclusion. If semantic reconstruction is possible, why wouldn’t it be available in the cases of partial A-movement reconstruction? The appeal to semantic reconstruction is virtually forced upon us if work arguing against successive cyclic A-movement is on the right track (see Boeckx 2000b, Castillo, Drury, and Grohmann 1999a, Epstein and Seely 2006). If the intermediate copies used for ‘reconstruction’ are not there (because such positions were never targeted by movement), then clearly semantic reconstruction must be available. This raises an interesting question regarding the tension between syntax and semantics. Fox (1999, 2000) and Romero (1998) have shown the superiority of syntactic reconstruction (copy-interpretation) for A-bar movement. If semantic reconstruction is available for A-movement, why not extend it to all cases of reconstruction? Here, I would like to follow Cecchetto and Chierchia (1999), and argue that semantic reconstruction is available only as a Last Resort, when syntactic reconstruction is unavailable, that is, when there is no movement, hence no copy to make use of. Since A-bar movement leaves copies, syntactic reconstruction is available, and takes precedence over semantic reconstruction. When no copies are available, semantic reconstruction is at work. Further, if Ausín (2000) is right that idiom interpretation does not necessitate LF-adjacency (contra Chomsky 1995), then there is really no compelling argument for A-traces.40 What emerges is a more diversified view of the syntax–semantics interface, where syntactic reconstruction, reprojection, and semantic reconstruction conspire to yield the full range of observed ‘reconstruction’ effects.41

Notes

PART I—AGREEMENT RESTRICTIONS CHAPTER 1 1. See Harley (1995b: Chap. 1) for a useful and critical discussion of phenomena associated with subjecthood. 2. The term ‘object behavior’ draws on the results of such tests as those applied by Harley (1995a, 1995b) to determine whether the nominative element hestar in (i) is Quirky or not, and if not, where it checks its features. (i)

Konungi hafa verið gefnir hestar. king.dat have been given horses.nom ‘A king has been given horses.’

The conclusion Harley arrives at is that nominative Case is checked in AGRo, the prototypical position of ‘objects.’ As Harley shows (see also Jónsson 1996), there are good reasons to believe the nominative element is an object of some sort: Unlike ‘subjects,’ it does not check the EPP (the Quirky element does), it fails to pass the subjecthood tests devised by Zaenen, Maling, and Thráinsson 1985 (see our section 1.2), it undergoes Object Shift, and it behaves like objects with respect to Negative Polarity items, ECM constructions, passivization, wh-extraction, and control (see Harley 1995a, 1995b for data and discussion). 3. It is generally agreed that, though poorly understood, those phenomena are to be associated with subjects. Some of them, like ECM, seem robust (see Bošković 1997c and Martin 1996, among many others, for relevant discussion). 4. Compare: (i)

Henni fannst kennarinn sinn/*hennar leiðinlegur. her.dat found teacher.nom her[+refl]/[-refl] boring ‘She found her teacher boring.’ Yip, Maling and Jackendoff (1987: 224,(5a))

5. See Sigurðsson (1991) for convincing arguments that PRO bears Case in Icelandic (here, nominative Case). 6. These examples do not reflect all the possible Case combinations in Icelandic. See Yip, Maling, and Jackendoff (1987) for such a survey. Note also that the specific Case of the Quirky subject does not matter. For all ends and purposes, we can equate ‘Quirky’ with ‘non-nominative.’ See later discussion.

216

Notes

7. Here I limit attention to agreement with the finite verb, not with past participles, for instance. Also, note that the term “full” agreement may not be adequate, for it might be that only number agreement is involved in Quirky Case constructions, and that the third person is copied from default agreement. See Taraldsen (1995: 312) for such a view. 8. As we will see later, the to-phrase is optional in Icelandic (as in English); we will, however, focus on sentences containing it so as to make the Case-array as complete as possible within one sentence. 9. I continue to use the term ‘movement,’ though Chomsky (2000) has argued that no actual displacement, but mere agreement, is what is involved in the syntax of features. ‘Feature-Communication’ (a term due to David Pesetsky) might be the most appropriate term. See Boeckx (1999a) for extensive discussion, and ways to tease apart Agree from Attract. 10. On the invisibility of A-traces, see Chomsky (2000); for a possible explanation, see Lasnik (1999c). 11. There is a way of capturing the minimality effects in an Attract-F framework, as argued by Ochi (1999b) and Boeckx (1999b), viz. by capitalizing on the Two-Chain hypothesis (Chomsky 1995: 265), according to which overt movement consists in the creation of two chains, the Feature chain and the Pied-piping chain, with only the latter subject to all the locality conditions on movement. However, this possibility is of no avail in the case under discussion since all that moves are features, hence only Feature chains come into play. Chomsky (2000), based in part on the present study, distinguishes ‘feature’ from ‘feature-value,’ and claims that only the former is relevant to capture Relativized Minimality effects. The present proposal, however, makes it possible to just talk about feature-value, given that, unlike Chomsky, I eventually argue against a structural/inherent Case distinction. 12. One possibility, suggested to me by Željko Bošković (p.c.), as an extension of Bošković (1998a), is to say that in the following abstract configuration, the embedded Infl blocks Attraction from the matrix clause. (i)

Experiencer SEEM Quirky to Inf. Nominative z_____________+____________m

However, it is unclear to me why the embedded Infl would block attraction given that it has a different feature specification from the matrix verb, except if we focus on features, not feature-values (see previous note), and Boeckx (1999b) for discussion within a larger context. 13. In addition to virðast ‘seem,’ sýnast ‘appear,’ finnast ‘find, consider,’ þykja ‘find, consider,’ and heyrast ‘sound (as if)’ are common raising predicates (Sigurðsson 1996: 29). To this list, I add “ECM” verbs like telja ’believe,’ which involve raising into the matrix clause (see Thráinsson 1979 and much subsequent work), though here the ‘experiencer’ problem is absent, but the ‘Quirkyinterceptor’ remains valid. 14. I use complex tenses in the examples because in the simplex tense, (ii), corresponding to (28), would be grammatical with henni as a topic. (i)

Henni leiddust strákarnir. her.dat bored.3pl boys.nom ‘She found the boys boring.’

(ii)

Strákarnir leiddust henni. (modeled on Taraldsen 1995: 307)

15. On Spanish experiencers, and possible connections with Icelandic Quirky subjects, see also Fernandez-Soriano (1999). For some discussion of Germanic, see section 1.5.2.

Notes

217

16. I will gloss over the differences among the three languages. See Boeckx (1998a, 1999b) for discussion. 17. Pesetsky and Torrego (2000) suggest that one might view Nominative as uninterpretable T-feature on D/NPs. 18. Ormazabal (2000) and Ormazabal and Romero (1998) argue that the PCC is best analyzed as a syntactic constraint. To the extent that the present analysis is correct, it argues against such a view. It is difficult to determine the different empirical predictions the two views make, for the very place of morphology is not made clear in Ormazabal & Romero’s work. Conceptually, it is, I think, desirable to restrict the scope of the PCC to the PF component; that way, we avoid look-head (why would syntax care about the output of morphology?). 19. I use ‘AGRs’ and ‘AGRo’ as mnemonics for subject and object agreement, respectively. As a matter of fact, the present study could be viewed as an argument of doing away with AGR projections, given that it captures agreement facts in a language (Icelandic), where it has been argued AGRs and T should be split (see Bobaljik and Jonas 1996, Bobaljik and Thráinsson 1998). 20. Here I use ‘dative’ as an umbrella term for ‘Quirky.’ As can be seen in such detailed studies as Yip, Maling, and Jackendoff (1987), dative is indeed the prototypical instantiation of Quirky Case. See Smith (1994, 1996) for an interesting conjecture about this “Dative Sickness” phenomenon. 21. This may be the equivalent of ‘Avoid structure,’ ‘avoid morphology’ (. . . up to crash) (i.e., avoid superfluous symbols, Minimize B (as much as possible)), familiar from the literature. 22. I concur with Lyle that this should receive a deeper explanation than an appeal to some morphological gap (Maling 1984). 23. See Tagalog, though. (i)

Gusto ng kanyang sarilii si Amadoi likes A his self T Amado ‘Himselfi likes Amadoi.’

Harley (1995b: 205 (39))

24. This is just tentative; the issue of binding in Icelandic is too complex for me to tackle here. 25. Partial agreement in Icelandic brings to mind the agreement patterns in Romance se-constructions discussed in Raposo and Uriagereka (1996). The basic pattern is as follows. (i)

Ontem compraram-se demasiadas salsichas no talho Sanzot. yesterday bought.3pl-se too-many sausages at butcher S. ‘Yesterday too many sausages were bought at the butchershop.’

(ii)

Essas salsichas compraram-se ontem no talho Sanzot. those sausages bought.3pl-se yesterday at butcher S.

(iii) Compra-se sempre demasiadas salsichas no talho Sanzot. buy.3sg-se always too-many sausages at butcher S. ‘Too many sausages are always bought at the butchershop S.’ Raposo and Uriagereka (1996: 750 (1)–(2)) In (iii), agreement looks like default. Interestingly, Raposo and Uriagereka adhere to the George/Kornfilt hypothesis, and show that due to its morphological defectiveness, se bears a special Case (Null Case) and cannot check person agreement, as a result, these are free to be realized with nominative Case, which is borne by the object, which raises across se (as in (ii)), when the F feature is strong (F = point-of-view, cf. Uriagereka 1995a, 1995b). Thus, [+person] forces projection of a category on top of IP, another instance of our Person/Point-of-View criterion.

218

26.

27. 28. 29.

30.

Notes It seems that, like Icelandic, Romance objects can check nominative; the only difference between Icelandic and Romance is that person features are not frozen by se (which is not dative, and thus fails to activate the PCC); as a result they will be checked high in the tree (for point-of-view reasons), and the object will have to raise. Objects in Icelandic do not raise because ‘high’ positions related to [+person] features are frozen. Romance is thus the mirror image of Icelandic, if the present analysis is correct. Note also that we achieve a parallelism between Null Case and Quirky Case; both cases are special with respect to agreement; interestingly, both are checked by PRO (Martin 1996) in non-finite clauses, where agreement is not involved. Again, a case of perfection in language it seems to me. Other cases of partial agreement come to mind, such as that found in classical Arabic (see Aoun, Benmamoun, and Sportiche 1994 for detailed discussion), where the subject and the verb agree in ‘full’ if the subject has raised; if the subject appears on the right, person agreement is not morphologically realized, again as expected under our approach (note, incidentally, that Benmamoun 1997 has argued that partial agreement in classical Arabic was the result of a PF process). The ‘partial agreement’ data to be discussed are different from the facts studied in Kayne (1995). Kayne assumes that (certain varieties of) English has (have) only number agreement. The present analysis crucially relies on the availability of both number and person agreement features. I use the men as an alternative to avoid problems inherent to coordinated structures. Howard Lasnik (p.c.) notes that the intervention effect is stronger if the experiencer is indefinite. See Boeckx (1999a) for an explanation of this fact in terms of Chomsky’s (2000) defective intervention constraint. As for the status of the expletive, I assume with Chomsky (1995) that it is merged directly in IP, and not raised from a VP-internal small clause (as in the predicate-raising hypothesis, see Moro 1997 and much related work). See Boeckx (1999a) for arguments against the predicate-raising approach. Mona Anderson (p.c.) points out that the pattern noted in (57) is not found when the experiencer is a pronoun. (i)

*There seems to her to be men in the room.

(ii)

There seem to her to be men in the room.

This fact does not cast doubt on the analysis developed here. As is well known, pronouns are special. In French, for instance, a full NP/DP experiencer seems to block overt raising of the subject (at least in the present tense, for reasons that are not clear to me), whereas raising is fine when a pronoun is used. (iii) ??Jean semble á Marie être heureux. Jean seems to Marie to-be happy (iv) Jean lui semble être heureux. Jean to-her seems to-be happy A similar fact is found in Italian, where a full experiencer NP/DP must topicalize to allow raising, while a clitic pronoun is irrelevant for raising purposes. (v)

A Piero, Gianni non sembra fare il suo devere. to Piero Gianni not seems do the his duty ‘Gianni does not seem to Piero to do his duty.’

(vi) *Gianni non sembra a Piero fare il suo devere

Notes

219

(vii) Gianni non gli sembra fare il suo devere Gianni not to-him seems do the his duty Torrego (1996: 106–107) Still, the special behavior of pronouns, though suggestive, does not explain why we do not detect PCC effect in that case in English. Here is what I would like to suggest (see Chapter 8: Appendix for extensive discussion). Pronouns, unlike full NPs/DPs in English are morphologically marked for Case; they may then be said to realize agreement on themselves, much like Quirky elements in Icelandic, and thus leave morphological room for associates to realize their agreement checking relations for number (person being excluded by the PCC), but note that English only has number, not person morphology, hence the apparent ‘full’ agreement with the associate in (ii). Put differently, the morphology of pronouns may be said to tease apart number and person agreement, freezing only person, not number, which can then be realized in accordance with the associate. That pronouns behave differently from DPs/NPs with respect to agreement and other phenomena is well known. Thus, in classical Arabic, agreement is partial with full NPs/DPs located on the right of the verb, while it is total with pronouns in the same position. (viii) Naama/*Naamuu l-!awlaad-u. Slept.3.masc.pl/sg the-children.nom ‘The children slept.’ (ix) !al-!awlaad-u naamuu/*naama (x)

Naamuu/*naama hum. slept they ‘They slept.’

(xi) Hum naamu/*naama. Aoun, Benmamoun and Sportiche (1994: 197 (5); 203 (21)) 31. As far as I can see, nothing in Kitahara’s analysis necessitates that assumption; Binding Theory could be said to apply everywhere, the relevant relation being formed at LF in the example under discussion. 32. See Chapter 8 for a full-fledged account of ‘conflicting c-command requirements’ of this kind. The account in Chapter 8 is slightly different from the one presented here, which will do for present purposes given the intricacies one would have to go into. 33. See Boeckx (1998b) for extension of that approach to other constructions and other languages. 34. Apart from agreement facts, Chomsky (1986a) also mentions the fact that inherent Case is intricately linked to theta-selection, thus differing from structural Case, which entertains a much freer relation with theta-role assignment. I don’t find the argument compelling; English has theta-selection processes similar to German, yet is often said to lack inherent Case (thus allowing prepositional passives, preposition-stranding, etc.; see Hornstein & Weinberg 1981 for a classic argument along these lines).

CHAPTER 2 1. The proposal has to be amended slightly if there are verbs that take experiencers bearing Nominative Case. Some cases have been reported in the literature, but they are very few, so we can assume that they can be learned on a case

220

Notes by case basis. It is interesting to note that children tend to generalize Quirky (Dative) Case on experiencers and that parents themselves are unsure about which Case to use (see Svenonius 2002). A similar amendment (or rule-breakdown) holds for the handful of transitive verbs with Quirky subjects that take non-nominative objects. Here too there is evidence that the pattern is marked, and progressively being replaced by the more robust [Quirky Subject-Nominative Object] frame. I should note as well that, not surprisingly, passivization interacts with (18). In active sentences, assignment of Quirky Case to the Goal does not lead to availability of nominative Case for the Theme (see (i)), but it does in passives (ii). (i) (ii)

Jón hefur selt honum bækur. Honum hefur verið seldar/*selt bækur. him.dat has.3sg been sold.3pl/3sg books.nom ‘He was sold the books.’

(ii) can be accommodated straightforwardly if passivization means absence of v(agent+acc). For (i) to be captured, we must incorporate (iii) into (18). (iii) *v(agent+acc) <-> vQ (iii) is plausible since Quirky Case is incompatible with Agentivity. 2. (18) and the amendment in the previous note are clearly not very explanatory, but they shouldn’t be, since we are dealing with an idiosyncratic, languageparticular issue: specific Case realizations. How that parametric choice interacts with other properties of the system is what should be explained by UG principles. 3. Since number agreement with the nominative objects shows up on the finite verb/auxiliary, I assume, with Den Dikken (1994) and Bošković (2001b), that auxiliaries start off under v and raise to Infl, much like main verbs. Thanks to Norbert Hornstein for useful discussion of this issue. 4. Anagnostopoulou (2003: Chap. 5) discusses some apparent differences between the Icelandic facts and classic PCC facts, but manages to explain them away. I don’t think the contrast between (24) and (25) can be so dismissed. 5. For compelling arguments from Romance that experiencers in monoclausal contexts and biclausal contexts are different, see Torrego (2002). 6. It is important to note that default nominative Case assignment need not be assumed in cases like (5a), since I argue that in such cases nominative Case is assigned by the downstairs vQ. 7. Holmberg and Hróarsdóttir (2001) observe that all the instances of intervention by Quirky elements in (38), (40)–(42) improve somewhat (though not up to full acceptability) if the Quirky element is plural, as in (i). (i)

?(?)Það virðist/*virðast mörgum stúdentum hestarnir vera there seem.3sg/3pl many students.dat horses.nom be seinir. slow ‘it seems to many students that the horses are slow.’

I have no account of this fact. 8. Marcel Den Dikken observes (p.c.) that the A-/A-bar asymmetry strengthens Zaenen et al.’s (1985) conclusion that Quirky subjects undergo A-movement of the standard variety in normal contexts, and are not topics. If they were topics, one would expect them to block agreement in cases like (39). 9. It is worth recalling that the overt Case realization of the head of the relative clause in Icelandic is not determined internal to the relative clause, but after

Notes

221

relative clause formation. Hence, the nominative Case on stúdentinn ‘student’ in (42), even though the raising verb inside the relative clause takes a Quirky (dative) experiencers. 10. In addition, as Holmberg and Hróarsdóttir point out, raising to SpecIP (A-movement) prior to raising to SpecCP in examples like (40) must be prevented, in order to predict a difference between (39) and (40). How to execute this in a cyclic fashion is far from trivial. 11. Skipping TP in (48) has implications for the EPP which I can’t discuss here, but see Boeckx (2003a). (See also McCloskey (2000: 78).)

PART II—AGREEMENT AND AGREE CHAPTER 3 1. I should note that it is always possible to describe facts like those under discussion in terms of massive remnant movement, but such an option clearly lacks restrictiveness, as it could capture virtually any fact. 2. Ideally, one would like to develop a theory of agreement that captures all the facts in (6)–(8) (e.g., the fact that ergative subjects don’t trigger agreement) in terms of primitives of the theory. But this is a task for future research. Here I focus on LDA only. For general considerations on agreement in Hindi, see Nevins (2003). 3. For some speakers (see, e.g., Butt 1995), agreement with the infinitival clause may take place independently of LDA (i.e., for them (10b) is grammatical). I will set the dialects that accept (10b) aside, coming back to them toward the end of the chapter. 4. For an empirical argument in favor of Agree over Move-F, see Chapter 1. 5. Of course, one could deny the adjacency requirement on Move-F, but if ones does so, Agree and Move-F become notational variants, with only conceptual arguments distinguishing the two (in favor of Agree). 6. I do not wish to imply that all instances of past participle/infinitival agreement come about via Multiple Agree. At the moment, it seems to me that participles/infinitival that are ‘nominal’ such as in Hindi, where infinitives are gerunds morphologically, and in Icelandic, where participles bear nominal Case morphology, participate in Multiple Agree, by virtue of being a close (categorical) match to the NP Goal. In languages like, say, Italian, where participles do not display overt ‘nominal’ characteristics such as Case, a different, more direct relation is probably at play (participles may be an independent Probe). The difference between the two types of languages would account for why Icelandic rules out past participle agreement with a quirky dative (the participle gets default morphology), as it does for any instance of finite verb agreement (i), while in Italian, quirky clitic si can trigger agreement on the participle while it is unable to agree with INFL (ii) (see D’Alessandro 2002). (i)

Stelpunum var hjálpað. the.girls.dat.pl.fem was.3sg helped.neuter.sg ‘The girls were helped.’

(ii)

Si è arrivati presto si is.3sg arrived.masc.pl early ‘People arrived early.’

222

Notes

7. For speakers of Kashmiri and Hindi-Urdu who allow agreement in the embedded clause in the absence of LDA (see Butt 1995, Subbarao and Munshi 2000, Bhatt 2001; see fn. 3), I have to assume that they allow v to be [+K] if selected by T[-finite], as schematized in (i); an option that most speakers of Hindi reject. Clearly, this is an instance of micro-variation (in the realm of categorical selection). (i)

[v [V [TP T (. . .) [Subj v [V Obj]]]]] |__|__| (Multiple) Agree

Kashimiri

CHAPTER 4 1. One reviewer objects to our treatment of (object) honorification as agreement on grounds that unlike agreement in, say, English, which is obligatory, honorific agreement is optional and conditioned by extra-grammatical factors. We do not deny the importance of pragmatics in the study of honorification, but it seems to us that the nature of the constraint we discuss in this chapter does not obviously lend itself to a pragmatic explanation. The fact that we are able to provide a reasonable syntactic account of an important set of sentences, in our view, argues for a syntactic characterization of honorification. (See also Hasegawa’s 2002 comprehensive overview of honorification phenomena in Japanese, where extensive arguments for a syntactic account of them are provided.) The reviewer notes that a syntactic account is inappropriate, since failing to trigger honorification leads to ‘discourse inappropriateness’ as opposed to ungrammaticality. We disagree. All we need to say to account for the difference the reviewer noted is that the honorification feature is optional. But once present agreement is obligatory, and is governed by the same mechanism of feature-checking. In other words, optionality obscures genuine ungrammaticality. Norbert Hornstein (p.c.) reminded us that the fact that object honorification with a reflexive is impossible (see (i)) provides a strong argument for treating (object) honorification as (object) agreement: (i)

*Tanaka sensei-ga zibun-o o-tasuke-si-ta. Prof. Tanaka-nom self-acc help-oh-past ‘Prof. Tanaka helped himself.’

As Woolford (1999) extensively showed, anaphors typically resist agreement. We therefore expect anaphors to resist honorification. Note, incidentally, that other grammatical phenomena like focus- and topic-assignment have received comprehensive treatment in terms of syntactic mechanisms like Agree/feature-checking, even though they too depend on extra-linguistic factors. Also, in the past purely syntactic accounts of ‘optional’ agreement (sometimes associated with different linguistic registers) such as past participle agreement (see especially Kayne 1989a) have led to a deepening of syntactic theory, which is also our goal in this chapter. 2. Some verbs use a suppletive form instead of the affixes in (3), as illustrated in (i): (i)

Suppletive Forms a. go-ran ni nar-u b. mesiagar-u c. haiken su-ru d. itadak-u

‘see-(sh)-pres’ ‘eat-(sh)-pres’ ‘see-(oh)-pres’ ‘eat-(oh)-pres’

Simple Form mi-ru ‘see-pres’ tabe-ru ‘eat-pres’ mi-ru ‘see-pres’ tabe-ru ‘eat-pres’

Notes

223

The verbs in examples (11), (15), and (17) are instances of this morphological process. 3. A reviewer points out several apparent counterexamples to this generalization: (i)

Watasi-wa kaizyoo-ni Tanaka sensei-o o-ture-si-ta. I-top place-dat Prof. Tanaka-acc take-oh-past ‘I took Prof. Tanaka to the place.’

(ii)

Watasi-wa heya-ni Tanaka sensei-o go-annai-si-ta. I-nom room-dat Prof. Tanaka-acc usher-oh-past ‘I ushered Prof. Tanaka to the room.’

In both (i) and (ii) object honorification with the direct object appears to be possible despite the presence of a dative element. However, note that the dative elements at stake are locatives, arguably generated lower than direct objects (see Larson 1988), or adjoined to the thematic domain as a whole—at any rate, plausibly not in a position that would trigger an intervention effect of the kind discussed in the text. The example in (iii), also provided by the reviewer, is perhaps the only one that constitutes a genuine counterexample to our generalization: (iii) Watasi-ga tyoosyuu-ni sensei-o o-hikiawase-si-masu. I-nom audience-dat teacher-acc o-introduce-oh-polite ‘I’ll introduce the teacher to the audience.’ However, the example may be made compatible with our approach if we treat tyoosyuu-ni as a locative of sorts. If that is correct, the sentence should be glossed as ‘I’ll take the teacher to the audience.’ 4. Shigeru Miyagawa points out (p.c.) that honorification agreement with the direct object is unavailable even if the indirect object is passivized (as in the following niyotte passive example). (But note that there is a potential interfering factor: The passive morpheme -rare also functions as a subject honorification marker in Japanese. A morphological conflict may prevent the realization of both passive and honorific markers. We abstract away from this in this note.) (i)

*Mary-ga Hanako-ni-yotte Tanaka sensei-o Mary-nom Hanako-dat-by Prof. Tanaka-acc go-syookai-sare-ta. introduce-oh-pass-past ‘Mary was introduced to Prof. Tanaka by Hanako.’

This fact can be captured under our analysis if we treat the passive morpheme as a ‘clitic’ doubled by the passivized argument (see Baker, Johnson, and Roberts 1989). The passive morpheme would absorb the Case/K-features of v, thus blocking agreement with the direct object. 5. As pointed out to us by Takae Tsujioka (p.c.), object honorification in monotransitive environments seems absent with nominative objects: (i)

Boku-ga Tanaka sensei-*ga/o o-tsure-si-tai. I-nom Prof. Tanaka-nom/acc take-oh-want-pres ‘I want to take Prof. Tanaka.’

This fact may follow from the analysis developed below if T, and not little v, acts as a probe for nominative objects, as argued in Niinuma (1999). (Under Niinuma’s analysis, T attracts both the subject and the nominative object, forming mirrored specifiers, as in Richards 1997.) For a detailed treatment of honorification in the realm of nominative objects, see Boeckx and Jeong (in preparation).

224

Notes

6. Thanks to Youngmi Jeong for extensive discussion on the nature of Agree as it relates to the domain of honorification. 7. Chapter 10 discusses the possibility of dispensing with the notion of ‘defectiveness’ in the realm of intervention. We will not be concerned with this theoretical issue here, and refer the interested reader to Chapter 10. 8. A reviewer observes that in his judgment, the sentence is still odd if ‘Mary-no’ is absent, as in (i). (i)

9.

10.

11.

12.

13.

14. 15.

16.

*Watasi-ga Tanaka sensei nituite-no kizi-o haidokusita. I-nom Tanaka Prof. about-gen article-acc read-oh-past ‘I read an article about Prof. Tanaka.’

The reviewer suggests (and we follow him in this respect) that the sentence may be excluded under our analysis by positing a null pro in SpecDP. We assume that nominals like kizi-o ‘article,’ or otaku-o ‘house’ are not eligible candidates for honorific agreement. Unlike ‘datives,’ which are animate, and thus match the K-features of v, inanimate nouns such as kizi-o or otaku-o don’t even match the features of v. Hence, they are not involved in computing locality conditions. See Chapter 10 for discussion of this point. For intervention to obtain in this configuration, we assume, following Fukui (1997), that the A-over-A configuration can be reduced to a Minimality configuration. Such a reduction is possible if we assume that a Probe first sees a label/head such as the NP o-tonari-san before the specifier (Tanaka sensei-no). For related discussion, see Collins (2002). Norvin Richards points out (p.c.) that a base-generation analysis à la Miyagawa could capture the honorification facts if the direct object were basegenerated higher than v (in which case, it could not act as a goal, hence trigger agreement). However, in the absence of a principled characterization of possible base-generation sites, this analysis seems ad hoc. Further, it does not appear to exclude the possibility of base-generating the indirect object higher than v, in which case, it would predict honorification agreement to be possible with the direct object, contrary to fact. Note, incidentally, that the object undergoing short scrambling may move through a VP-adjoined position on its way to SpecvP, if successive cyclic movement is not a series of independent steps (as in Chomsky 2000, 2001, 2004), but rather part of the operation Form Chain, which applies only after the introduction of the highest probe (v, in our case), as in Takahashi (1994), Collins (1994), Boeckx (2001), and Bošković (2002a). Most accounts adopting (43) rely on the notion of equidistance (Chomsky 1993), which, in a way not dissimilar to the base-generation analyses discussed earlier, allows for the direct object and the indirect object to be ‘equally close’—precisely what we must avoid. This constraint holds irrespective of the order of clitics within the cluster, which varies across Romance. For a recent overview, see Manzini and Savoia (2001). Richard Kayne (p.c.) reminds us that the me-lui constraint is less sharp across Romance with 2nd-person clitics. Consider the following example involving a 2nd-person honorific clitic: *?Jean vous lui a recommandé ‘J. you(polite) him recommended’/’J. recommended him to you. To our ears, the me-lui constraint is still quite sharp in such cases. Should there be a difference between Romance and Japanese, we would like to claim that it is due to the ‘defective’ character of agreement in Japanese, which does not leave much room to maneuver in non-agreement situations (see the later discussion in the text). We simplify the labels in (48) somewhat, and depart from Johnson’s claim that the verb ATB-moves to To, which we regard as problematic in the case of English main verb.

Notes

225

17. Many thanks to Nobuhiro Miyoshi for helping us construct the arguments that appear in this appendix. 18. We concede that the argument is very theory-internal as it depends on the correctness of Murasugi’s and Hoshi’s analyses of head-internal relations. We refer the reader to these works for arguments in favor of the position adopted in the text (see also fn. 19). 19. A strong piece of evidence in favor of Hoshi’s and Murasugi’s view comes from Kuroda’s (1975/76) observation that head-internal relatives can have multiple internal heads, as shown in (i): (i)

[zyunsa1-ga doroboo2-o kawa-no-hoo-e policeman-nom thief-acc river’s direction-toward oitumete-itta no]-ga [pro1,2] ikioi amatte futaritomo tracked down-nom power exceed both-two kawa-no-naka-e tobikonda. river-into jumped ‘A policeman was tracking down a thief toward the river, who both, losing control, jumped into the river.’

In this example, both the subject ‘policeman’ and the object ‘thief’ of the relative clause are heads, that is, they together function as the subject of the matrix verb, ‘jumped into’ (as attested by the use of ‘both’). Hoshi’s pro-analysis of head internal relatives captures Kuroda’s observation by letting pro, in the complement position of V, take both the subject and the object of the relative clause as referents. 20. Recall that in section 4.3, we rejected the possibility that the agreeing object raises to some specifier position lower than vP, as in Koizumi’s (1993, 1995) analysis. Note that such an analysis would not predict any Condition C effect in (50). 21. A reviewer objects to Yatsushiro’s analysis by observing that she missed the generalization that Japanese does not allow a remnant VP to be preposed, as evidenced by the following cases, provided by the reviewer: (i)

a. *[ti tabe-sae] Taro-ga katuo-oi sita. eat-even Taro-nom bonito-acc did ‘Taro even ate the bonito.’ b. *[ti huri-sae] ame-gai sita. fall-even rain-nom did ‘The rain even fell.’ c. *[ti hiraki-sae] doa-gai sita. open-even door-nom did ‘The door even opened.’

To our knowledge, the generalization the reviewer appeals to has never been studied in any depth in the Japanese literature, and it is not clear what it would follow from, were it correct (given that, as the reviewer notes, many languages allow remnant VP-movement). At this point, it clearly requires further investigation.

CHAPTER 5 1. For a detailed discussion of the limitations of Toribio’s system, see Niinuma (2003). 2. B&Y note that “various omissions in [Chapter 4] have led [the authors] to posit a greater degree of similarity than the data actually show.” As I argue later, the data actually do suggest the type of similarity Chapter 4 posited.

226

Notes

3. Andrew Nevins (p.c.) points out that this transmission comes close to treating the (g)o- + suru combination as a circumfix. 4. Alternative structures are equally plausible, and depend on the structure of double objects as (low) applicatives (Pylkkänen 2002, Jeong 2006, 2007a, among others) and light verb constructions in Japanese. (i)

[vP v [“–P” hon-– [ApplP IO Applº DO]]]

(ii)

[vP v [“–P” IO [Predº-max [hon-–] Predº] DO]]

(iii) [vP v [PredP [Predº-max [hon-–] Predº] [ApplP IO Applº DO]]]

5.

6. 7.

8.

9.

Some of these heads may be phonetically null, an idea (“an avenue of retreat,” as B&Y characterize it) B&Y claim I cannot appeal to, since Chapter 4 argued against the Spec-Head relation in part because in order to maintain Spec-Head relations, ad hoc projections are often posited. Unfortunately, B&Y fail to note that my remark on Spec-Head doesn’t carry over to the case of v, which is part of the ‘core functional projection’ (Chomsky 1995), and whose presence is widely justified on independent grounds. B&Y acknowledge my skepticism about the use of otagai as a c-command test in their foonote 17, and note that “[the authors] expect that [their] point will be replicable with any familiar c-command diagnostic that can be applied to the examples.” Examples like (6)–(8) cast doubt on this expectation. My skepticism about the use of otagai is grounded in Hoji’s (1998, 2003: sec. 4.2) arguments against taking otagai to be the Japanese equivalent of the local anaphor each other. I will not go over Hoji’s arguments here. Suffice it to say that given otagai’s unclear status, arguments based on it are not foolproof. As a matter of fact, the English translations of sentences (6) and (8) offer another example of binding from within PPs that may pose another c-command paradox (see Pesetsky 1995 for discussion). Note that Chapter 4’s analysis also stands even if it turns out that, say, Icelandic nominative agreement data discussed in Chapter 4 do not constitute cases of defective intervention. Icelandic was used for illustrative purposes only. It is the cross-linguistically widely attested phenomenon of dative intervention behind the specific Icelandic data that matters. I set aside adjunct datives and cases where the grammatical function ‘indirect object’ and ‘direct object’ do not map onto the morphological cases ‘dative’ and ‘accusative,’ respectively. I set these aside to make B&Y’s quote from Moravcsik (1978) most relevant to the debate about object honorification in Japanese. See Jeong (2006, 2007a) and Lee (2004) for specific proposals for how to relate the two.

CHAPTER 6 1. A third pattern exists, exemplified by Basque, where superiority regulates not only the first instance of wh-movement, but all of them, as shown in (i). For a treatment of that aspect of Basque multiple wh-questions, see Jeong (2007b). (i)

a. Nork nori zer eman dio? who.erg to whom.dat what.abs give aux ‘Who gave what to whom?’ b. *Nork zer nori eman dio? who what to whom give aux c. *Nori nork zer eman dio? to whom who what give aux d. *Nori zer nork eman dio? to whom what who give aux

Notes

227

2. Bošković (2002b) has argued that scrambling may be another source of whfronting in some cases. See also Grohmann (2000, 2003a). 3. Bošković here relies on Stjepanović’s (1998, 1999) finding that (non-wh) (contrastively) focused phrases in Serbo-Croatian must move overtly (just like wh-phrases) to positions that seem to be identical to the landing sites of wh-phrases. Elsewhere, Rizzi (1997) has argued that the Focus Phrase is the landing site of wh-movement. Numerous subsequent studies confirmed the isomorphism between focus-movement and wh-movement (with some exceptions, such as Lee 1999, to which I come back briefly later). 4. The possibility of honorification in simple transitive contexts such as (i) suggests that accusative Case is structural in Japanese (inherently Case-marked elements rarely show overt signs of agreement). (i)

Taro-ga Tanaka sensei-o o-tasuke-si-ta/tasuke-ta. Taro-nom Prof.Tanaka-acc help-oh-past/help-past ‘Taro helped Prof. Tanaka.’

5. I am assuming that v also assigns Case to the indirect (dative) object. Which Case value surfaces on which NP is arguably a case of morphology (late insertion; see Harley 1995, developing ideas of Alec Marantz). 6. My solution is in some sense related to Pesetsky and Torrego’s (2001, 2004) claim that Case is an uninterpretable T-feature. For me, Case is an uninterpretable feature corresponding to (but not necessarily identical to) T/Asp. I thus depart from Chomsky’s (2000) claim that feature matching is an instance of feature identity. I would like to claim that matching is in terms of featural compatibility, restricting featural identity to instances of adjunction. See Boeckx (to appear) for full discussion. 7. Note, incidentally, that in Chomsky’s (2001, 2004) framework, Valuation cannot be a morphological operation, as valuation inactivates lexical items for the purposes of narrow syntax. 8. Abstractly, this is reminiscent of Bošković’s (1998b) use of both (Probeoriented) Attract and (Goal-oriented) Greed in an earlier version of his (1999) Attract-1/Attract all account sketched in section 6.1. (Interestingly, Bošković 1997c also argues that Case-checking, as opposed to K-feature checking, necessitates a notion like Greed). The present solution captures the same intuition, but without the computational complexity/globality associated with Greed. 9. The differences between wh-fronting and focus-movement noted by Lee (1999) may follow from this extrinsic difference. 10. This is reminiscent of Manzini and Roussou’s (2000) Scopal Agree, which lets Agree take place down to the next Attractor. Boeckx (to appear) explores the possibility of recasting control (as raising, see Hornstein 1999) in terms of multiple theta matching. Other areas of multiplicity, such as parasitic gap licensing, may also be fruitfully recast in those terms. 11. If they associate the EPP with Valuation, then only one overt movement process will take place.

PART III—AGREEMENT AND LOCALITY CHAPTER 7 1. Saito (2003) argues that remnant movement is possible if and only if the remnant contains no trace (in his terms, if the remnant is a complete constituent). Besides the unclear status of what it means to contain no ‘trace’ under the

228

Notes

copy theory of movement (Chomsky 1995), Saito’s account is designed to rule in only those cases of remnant movement out of which A-movement has taken place (following Lasnik 1999c, Saito assumes that A-movement leaves no trace). Müller contains many examples which appear to falsify Saito’s proposal, as they arguably involve A-bar movement. 2. Kayne (2001) notes that the predicate part of small clauses pattern like idioms in this respect. (i)

a. The winneri is likely to be [Small Clause John t]. b. *[How likely to be John ti]j is the winneri tj.

I will not discuss such examples here, as the simple cases of raised predicates like (ia) pose non-trivial questions (why is there no Relativized Minimality effect when an NP, the winner, crosses another one, John?), to which I cannot provide an answer here. If the exact nature of the first step of movement out of the small clause to SpecIP is ignored, the solution I offer in the text for (5b)– (9b) extends to (ib). 3. The raising-control ambiguity appears to be supported by Martin’s (1992, 1996) argument, based on observations in Barss (1986), that a sentence like (i) allows for only one reading of the indefinite, in contrast with regular raising cases (see (ii)), but alongside with control predicates (see May 1985) (iii). (i)

[How likely to win the lottery] is someone from NY ? (someone >> likely/*likely >> someone)

(ii)

Someone from NY is likely to win the lottery. (someone >> likely/likely >> someone)

(iii) A unicorn is eager to be apprehended. (a unicorn >> eager/*eager >> a unicorn) Setting aside the issue of whether control predicates do indeed block reconstruction (see Hornstein 2001: 139–140), I note that Sauerland (1999) provides an analysis of the contrast between (i) and (ii) that is independent from raising versus control. The explanation I provide later for the how likely paradigm also captures the asymmetry. 4. Contrast: (i)

Who said that it was how likely that there will be a riot?

5. Some speakers find (19) marginal (see note 14 below for a possible explanation for this). Crucially, they still perceive a contrast between (17)/(18) and (19). (Thanks to Masashi Nomura for discussion of the relevant examples.) 6. For a different tack on the overt/covert asymmetry, see Nissenbaum (2000) and Chomsky (2004), where covert movement is taken to be phrasal movement taking place after (cyclic) spell-out. I will not discuss this option here, as many details still remain to be worked out. 7. The LF-intervention effects on A-bar movement discussed in Pesetsky (2000) seem to provide another argument in favor of non-phrasal ‘covert’ operations. 8. Lasnik also points out that the Move-F analysis captures the paradigm discussed in Den Dikken (1995) (see also Lasnik and Saito 1991 for similar examples in ECM contexts), which is problematic under expletive-replacement. (i)

a. Some applicants seem to each other to be eligible for the job. b. No applicants seem to any of the deans to be eligible for the job.

(ii)

a. *There seem to each other to be some applicants eligible for the job.

Notes

229

b. *There seem to any of the deans to be no applicants eligible for the job. As the data in (ii) show, the associate is incapable of licensing an NPI/anaphor located in the matrix clause, which is unexpected under the expletive replacement analysis since according to the latter (i) and (ii) share the same LFs. Lasnik takes the ungrammaticality of the sentences in (ii) to mean that such licensing mechanisms require more than formal features. But see Branigan (1999), Yatsushiro (1999), and Watanabe (2000b) for some arguments that binding (but not scope) can be established via feature movement. 9. Strictly speaking, this is not Cinque’s (1990) view. According to him, all islands are absolute. Apparent (good) extractions out of islands are cases of base-generation and (non-overt) resumption. I ignore this detail here (for arguments that resumption demands a movement-approach, see Boeckx 2001). 10. I will not be concerned here with defining the relevant factors easing extraction out of weak islands. For valuable discussion, see Frampton (1999), Szabolcsi and Den Dikken (1999), and Starke (2001). 11. Masashi Nomura (p.c., attributing the original observation to Howard Lasnik) points out that if the proposal I make is correct, we expect no substantial difference between (i) and (ii). My proposal indeed seems to predict a Relativized Minimality effect triggered by somewhat in (ii). (i)

(Who said that) there is how likely to be a riot?

(ii)

There is somewhat likely to be a riot.

The fact of the matter is that for some speakers (ii) is much better than (i). To account for the contrast, I am forced to say that for those speakers, there must be a structural difference between (i) and (ii). In other words, for them, the (offending) indefinite part of the degree modifier is (structurally) more prominent in (i) than it is in (ii) (i.e., it c-commands a riot). Intuitively, this seems correct. In (i), how must project in some way or other in order for the complex how likely to act as a wh-phrase. Exactly how to formulate this prominence distinction is a task I leave for future research. 12. Properly speaking, the intervention effect would be of the type that Chomsky (2000: 123) characterizes as “defective intervention effect.” Defective intervention arises when an element B matches the featural requirements of a probe P, but fails to agree with it. (In other words, H blocks the raising of C to B even though H itself cannot raise to B.) 13. Howard Lasnik (p.c.) informs me that Noam Chomsky has made a proposal similar to Rosenbaum’s at various times. Neither Lasnik nor I have been able to locate the proposal in Chomsky’s writings. It may have been made in class lectures. 14. Alternatively, one could follow Rizzi (1990), Uriagereka (1988), Law (1991, 1993), Boeckx (2001), and Starke (2001), among others, and take wh-adjuncts to be directly inserted in COMP. As far as I can see, this possibility yields identical results to the text discussion (i.e., it is unavailable in the ‘pied-piping’ cases and in the in situ examples, for reasons I discuss in the text immediately following). Klaus Abels points out that many speakers of English find long-distance questions with how likely like (i) deviant. (i)

*?How likely did John say that there was to be a riot?

This suggests that how likely patterns like how come, which is known to lack long-distance construal.

230

Notes (ii)

*How come did John say that Mary left t?

The facts in (i) and (ii) may demand an analysis in terms of base-generation in [+wh] SpecCP for how come and how likely. Such an analysis may account for the fact that some speakers find (19) deviant. 15. Unless Relativized Minimality is viewed as a condition on representation (see Rizzi 1986). I reject this option as I follow Chomsky (2000) in taking Agree to be derivationally established. 16. Lasnik’s (2002) second argument in favor feature movement (based on the interaction of head-movement and ellipsis) is equally problematic, as discussed in Boeckx and Stjepanović (2001).

CHAPTER 8 1. The formulation given here is somewhat misleading because the same problem arises in sentences where the experiencer (Bill) is not headed by a preposition, as in (i). Following standard assumptions, I will assume that the object of strike is headed by an empty preposition, which reduces (i) to (1). (i)

Johni strikes Bill [ti as being a genius].

2. The conclusion is forced in strictly derivational models (Epstein et al. 1998, e.g.), where traces of any kind do not play any role; all ‘interpretive mechanisms’ (binding included) taking place ‘on-line.’ 3. This test was suggested to me by Howard Lasnik. 4. Ferguson (1994) assumes that the target of movement is some AGRprepP. Kitahara assumes adjunction to P, following Chomsky’s (1995) dismissal of AGRPs. 5. The conclusion is also reached in Pesetsky (2000), who sees the need to distinguish between two sets of ‘covert’ operations (operations without consequences for PF): feature-movement/agreement (Chomsky 1998), and category-movement. 6. For important precursors to the cyclic interpretive approach, see Jackendoff (1972), and Lasnik (1972, 1976). For arguments in favor of a cyclic application of phonological processes, see Bresnan (1971). 7. In a way much like the cycle did, when coupled with the Projection Principle (see Epstein et al. 1998: introduction for discussion). 8. The reanalysis account for (22)–(23) goes back to at least Chomsky (1955), and has remained virtually unchallenged ever since. (Baltin & Postal 1996 criticize ‘reanalysis hypotheses,’ but do not offer an alternative.) 9. Assuming binding to be determined as the derivation proceeds. This is not Chomsky’s assumption (Chomsky assumes binding to be determined strictly at LF), but it is virtually forced upon us if we assume with Chomsky that A-movement does not give rise to reconstruction effects. I see no other way to reconcile the licensing of an anaphor/NPI embedded in the embedded subject with the absence of reconstruction with A-movement. 10. I use the men as an alternative to avoid problems inherent to coordinated structures. 11. If we follow Chomsky (1995) in viewing the ‘associate’ to relate to To, not to the expletive. 12. Setting aside the question of how the agreement mechanism should be captured: feature-movement (Chomsky 1995; Lasnik 1995c, 1995d), featureagreement (Chomsky 1998; Boeckx 1999b, 1999d), or some (rather vague, and unminimalist) co-indexing/feature-transfer (Moro 1997; Hoekstra & Mulder 1990).

Notes

231

13. I use the term “dative” element so as to make the relation with the PCC more transparent. Dative is the prototypical instantiation of Quirky Case. 14. Note that the present account does not—correctly—predict (i) to be good. (i)

*There seem to two men to be a woman in the garden.

The PCC does not say that agreement with the dative surfaces, it merely says that person agreement with the accusative is blocked. I take -s to be a default form. See Appendix for further elaboration. Thanks to an anonymous reviewer for raising the point made in this note, and for encouraging me to address further data points in the appendix. 15. Such transitivity is intended to capture the relation ‘dependent of complement’ extensively discussed in Pesetsky (1995), for instance. 16. I leave open the ultimate reason(s) of pied-piping. 17. At this point, one might wonder whether we don’t incorrectly rule in sentences like (i) by saying that binding takes place derivationally. Boeckx (1999e) shows that such sentences are excluded on independent grounds, which I cannot go into here (roughly, himself is viewed as a (part of a) predicate, and is said not to have the matching feature necessary to check the EPP). (i)

Himself seems to John to be a genius.

CHAPTER 9 1. Languages like Spanish (i) and Icelandic (ii) disallow any kind of subject raising from an infinitival clause across an experiencer (i.e., even if the latter is pronominal). ((iib) is good only if Ólafur is topicalized. See Ura 2001 for discussion.) (i)

a. *Este taxistai lo parece a Juan [ti estar cansado]. this taxi-driver to.him seems to Juan be tired ‘This taxi-driver seems to Juan to be tired.’ b. *Este taxistai nos parece [ti estar cansado]. this taxi-driver to.us seems be tired ‘This taxi-driver seems to us to be tired.’

(ii)

a. Þeim/Stelpunum hafði virst Ólafur vera gáfaður. them.dat/the girls.datt has seemed Olaf.nom be intelligent ‘They regard Olaf as intelligent.’ b. *Ólafur hafði virst þeim/stelpunum vera gáfaður.

However, there are independent reasons for the absence of raising across an experiencer in both languages which for this reason make them less interesting than Italian. Thus, for Spanish, Ausín and Depiante (2001) (see also Anagnostopoulou 2003), building upon original observations by Torrego (1996), provide compelling arguments that raising verbs like parecer cease to act like raising verb when they combine with an experiencer (for instance, Ausín and Depiante show that depending on whether an experiencer is used parecer takes different morphological paradigms). As for Icelandic, it is well known that dative experiencers are ‘quirky,’ and behave like subjects in virtually all respects. Hence the fact that they block raising of a lower NP should not come as a surprise. The examples in (ii) are straightforwardly captured under any version of Relativized Minimality. For lack of space, I will not document other patterns of raising (see Chapter 9 for a fairly comprehensive survey; see also Anagnostopoulou (2003) for

232

Notes

extensive discussion of Greek, which in many respects constitutes a mix of Italian and Spanish). 2. It is not the only place where pronouns and A-bar raised elements behave alike, and in contrast to in situ full NP objects. One thinks here of the past participle agreement patterns in French discussed by Kayne (1989a) (i) and the ‘exceptional’ ECM cases analyzed by Kayne (1984) (ii). (i)

a. Jean a vu-*e la fille. Jean has seen-agr.fem the girl ‘Jean saw the girl.’ b. Jean l’a vu-e. Jean her.has seen-agr.fem ‘Jean saw her.’ c. Quelle fille Jean a(-t-il) vu-e. which girl Jean has-he seen-agr.fem ‘Which girl did Jean see?’

(ii)

a. *Jean croit Pierre être intelligent. Jean believes Pierre be intelligent ‘Jean believe Pierre to be intelligent.’ b. Jean le croit être intelligent. Jean him believes be intelligent ‘Jean believe him to be intelligent.’ c. Qui Jean croit(-il) être intelligent? who Jean believes(-he) be intelligent ‘Who does Jean believe is intelligent?’

Whether all such cases should receive a uniform treatment is a question I leave for future research. 3. The contrasts reported here are subtle, as many speakers allow singular agreement in there-constructions (see Schütze 1999 for extensive discussion), but they are nevertheless quite robust across speakers. 4. As far as I can see, no substantial result of this study is affected if we analyze pro is a feature moving from the subject position in the lexical domain, adjoining to Infl, as does Uriagereka (1999a), or if the existence of pro is denied, and the EPP is checked by a head (the raising verb) in pro-drop languages, as in Alexiadou and Anagnostopoulou (1998). 5. Although Alexiadou and Anagnostopoulou (2001: 204 fn. 11) observe that the ban on multiple of-phrases in nominals (*the destruction of the enemy of the city) may fall under their generalization if of is treated as a Casemarker.

CHAPTER 10 1. Most of the cases we discuss differ markedly from those discussed in Atkinson (2001), who also argues against Defective Intervention. Atkinson concentrates on alleged cases where the intervener is an element has checked the relevant feature (thereby becoming inactive, in Chomsky’s [2000] sense), but nevertheless manages to block certain relations. A typical example would be superraising: (i)

*Johni seems [it was told ti [that it’s raining]].

Here the intervener has checked its Case feature, and yet blocks movement of John (which by assumption is Case-driven). Atkinson shows that none of the cases fitting that description are conclusive. Be that as it may, Atkinson’s

Notes

233

cases are very theory-internal as they rely on notions like deletion, phase, and feature status (feature deletion), which all arguably weakens the idea that syntactic computations are strictly cyclic. For this reason we think that the conclusions that can be drawn from them are less general than the cases we examine. We therefore set them aside, noting that Atkinson’s arguments, if correct, support our main conclusion about Defective Intervention. 2. The treatment of honorification as an agreement process is controversial. The issue is discussed in Chapter 4, which we rely on here. 3. The phenomenon has been reported in a few traditional grammars of Spanish (e.g., Kany 1976; see Martinez-Miera 2002 for references). In the generative literature, we find some scattered examples of agreement mismatch in Romance languages, and a few mentions of the phenomenon, among them Kayne (2000b: Chap. 8), Boeckx (2001: Chap. 2), and Ledgeway (2000). 4. Like Vukić (1999) we hypothesize that this movement takes place only if the IO is a Prepositional Phrase. This accounts for the surface orders in (i) for English. (i)

a. John gave Mary a book. b. John gave a book to Mary.

For Spanish it enables us to explain why there is no feature mismatch between the dative clitic and its double when the latter is non-overt (i.e., when it is a pro/NP). In the absence of movement of the direct object, no intervention will take place, and the probe and goal will match in K-features: (i)

Les/*le envié pro las notas. to.them.dat.pl sent.I pro.dat.pl the grades ‘I sent them the grades.’

Note also that the NP/PP distinction just mentioned renders the scrambling operation referred to in the text unavailable in Japanese, where dative arguments are NPs (with a Case-marker), as argued in Chapter 4. 5. Perhaps more precisely, the [person] feature is not active on experiencers. See Boeckx (2002b) for discussion. 6. The intuition behind the proposal is, on the one hand, that the nature of the associate is always 3rd person (see *there is me/you in the room), which since Benvéniste (1966) has been taken to signify the absence of person; and, on the other, that the expletive is specified for [person] features only (Chomsky 2000), leaving room for the associate to agree in [Number]. 7. For a similar case, based on contrasts like (i)–(ii), corresponding to (40)–(41), see Chapter 7. (Thanks to Norbert Hornstein for reminding us of these cases in the present context.) (i)

a. John is 5% likely to win the race. b. How likely to win the race is John?

(ii)

a. *There is 5% likely to be a man in the room. b. *How likely to be a man in the room is there?

8. For a similar treatment of wh-island extractions, see Starke (2001). 9. All the cases we deal with here are all about 3rd-person NPs. Interestingly, if the dative is 1st/2nd person, the clitic must match the dative double in person and number. So, examples like (ii) obtain, as opposed to cases like (i). (i)

*Juan me dio flores a nosotros. Juan to.me.dat.sg gave flowers to us.dat.pl ‘Juan gave us flowers.’

234

Notes (ii)

Juan me/te/nos dio flores a mi/ti/nosotros. Juan to.me.sg/you.sg/us.pl.dat gave flowers to me.sg/you.sg/us. pl.dat ‘Juan gave me/you/us flowers.’

This pattern is reminiscent of the Experiencer case (section 3.1): Agreement across [Num] is only possible if [Per] agreement between a Probe and a Goal takes place. Assuming that 3rd person is ‘absence of person’ following an idea with a long tradition in grammar, we can say that [Person] is missing from direct objects. The feature schema we will get, thus, is as follows: (iii) a. B [Num] b. BC [Num][Per]

B [Num] B [Num]

B [Num] (n the cases we had so far) BC [Num][Per]

If Defective Intervention in the sense of Chomsky is valid, the DO should block agreement in [Num] between the dative clitic and its double even though it itself doesn’t bear [Per] feature, which is not the case, as seen in (ii). (Note that cases where DO has [Per] feature (non-3rd person) cannot be tested, as the DO would then be a clitic, and should be 3rd person, by the Person-Case/ me-lui constraint, that forces 3rd person features on accusative clitics in the presence of a dative argumental clitic; see Perlmutter 1971, Kayne 1975, Bonet 1994, etc.). 10. Alan Munn (p.c.) points out that by the same reasoning it may be wrong to put a K-feature on C in the case of D-linked wh-phrases, as we did in section 10.3.3. We disagree. Encoding presupposition in C (in the guise of K-features) has some plausibility, but [manner]-features do not seem to belong to the C-domain in any way. 11. Heck and Müller provides instances parallel to those in (62) involving an intervening wh-element contained inside a relative clause or a (complex) NP, which, like adjuncts, constitute strong islands. 12. One question remains regarding the intervention effect in (62): How can the wh-element contained in the adjunct block wh-movement in the absence of c-command? To get around this problem, it is tempting to assume that the wh-phrase raises covertly to the edge of the adjunct, as argued for by Nissenbaum (2000) in the case of parasitic gaps. For evidence that wh-elements can undergo covert movement in German/Dutch, see Den Dikken (1999), building upon Ausín (1998). Note, incidentally, that given the idea that a Probe-Goal relation is established as soon as possible (Chomsky 2000, 2001), the facts in (62) indicate that a forced late insertion of adjuncts is not a viable theoretical option.

PART IV—AGREEMENT AND INTERPRETATION CHAPTER 11 1. Throughout this chapter, I will use the terms ‘A-movement’ and ‘A-bar movement’ in a somewhat pre-theoretical sense (roughly, contrasting wh-movement and A-raising), thus avoiding the well-known intricacies in properly defining the class of A- and A-bar positions. 2. Epstein and Seely (2006) reach a similar conclusion on independent grounds. Manzini and Roussou (2000) explain the lack of A-movement reconstruction

Notes

235

by denying the existence of A-movement: If there is no movement, there is no copy, hence no reconstruction. See Boeckx (2000a) for arguments against Manzini and Roussou. 3. To the best of my knowledge, the argument that A-movement copies (‘traces’ at the time) are needed only to satisfy the Projection Principle was first developed in Hoekstra (1987). 4. Sentence (19) was first noted, to my knowledge, by Partee (1971), who discussed its relevance for surface interpretation. 5. Another possibility would be to say that interpretation is cyclic (Higginbotham 1995), and so reconstruction not required. Martin and Uriagereka (1998) contain a discussion of quantifier-interactions in those terms (see also Tsai 1994, 1998). However, it is hard to see how such a cyclic interpretation, stemming from Multiple Spell-Out (Uriagereka 1999b, Chomsky 2000, 2001), can be reconciled with the standard notion of compositionality. (Jackendoff 1972 claims that interpretation is cyclic, but his approach does not rely on compositionality). The following passage from Heim and Kratzer (1998: 46–47) illustrates the potential conflict between compositionality and strongly derivational syntax. The kind of semantic theory we are developing here is compatible with a wide range of approaches to syntax. It may be worth pointing out, however, that it is nevertheless fundamentally incompatible not only with many conceivable proposals regarding the structures of the input to semantic interpretation, but even with some that have actually been made. For instance, Jackendoff and other representatives of the Extended Standard Theory argued that the meaning depended in both Deep Structure and Surface Structure. Roughly, Deep Structure was to determine predicate-argument relations, and Surface Structure scope and binding relations. In our terms, this would mean that the domain of interpretation function [[ ]] should consist of (or include) something like pairs of phrase structure trees. What this would amount to in concrete detail is far from clear, and our current conception of semantic composition would have to be substantially altered and/or enriched to make it work. However, Heim and Kratzer’s quote should not be construed as entailing that ‘phases’ in Chomsky’s (2000) sense have no consequence for interpretation. See Boeckx (2000d) for arguments that they do. 6. Thanks to Gennaro Chierchia (p.c.) for suggesting these tests to me. 7. For more evidence that sequence of tense phenomena are ‘syntactic,’ which would require lowering of the subject in (38), see Higginbotham (1998). 8. Further examples illustrating the same point are produced here. Take (i)–(ii), where pure scope relations, and not specificity/vagueness, play a role: (i)

Some student seems to me to have read every paper.

(ii)

Some student seems to herself/her TA to have read every paper.

Here, the issue is whether or not the choice of student can vary with the choice of paper (i.e., scope, as it is generally understood). The reported facts (from David Lebeaux) are that the choice of student can vary with the choice of paper in (i) (it seems that for each paper there is a student who has read it), but not in (ii). The effect in (ii) is ‘trapping’ in the high position, as extensively studied by Fox (1999) and Romero (1998). Both Fox and Romero give evidence that we are not dealing here with connectivity (also known as chainbinding, or semantic reconstruction). In the case of trapping, a function of the element like binding forces it to stay high.

236

Notes As pointed out to me by Jonathan Bobaljik (p.c.) (see also Bobaljik and Wurmbrand 2000), the excluded reading in (iii) is easy to construct, for example, with the paraphrase, which, Bobaljik notes, shows again that paraphrasability is not a good test for scope, as argued by Lasnik: (iii) For every paper, there is some student who seems to herself to have read it. However, what Bobaljik and Wurmbrand failed to note is that the reading in (iii) is not the reading excluded in (ii) that we are after. (iii) allows for the two quantifiers to take scope over seem, whereas what we want in (ii) is have both quantifiers under seem. The point is worth making, as it shows that Lasnik’s criticism against A-movement reconstruction is not so easy to dismiss. Nonetheless, other cases do show the need for reconstruction. Among such cases, one can also mention the well-known Belletti and Rizzi (1988) facts, and the each-paradigm, originally discussed in Burzio (1981), who took (v) as evidence for ‘layered’ traces (i.e., copies), and A-movement reconstruction (thanks to Richard Kayne p.c., for reminding me of the each-paradigm in the present context). Typical examples appear in (iv)–(v): (iv) Pictures of himself seem to John to be on sale. (v)

One translator each is likely to be assigned to the athletes.

With regard to psych-verbs, Lasnik (1998a) gives the following: (vi) Each other’s supporters frightened the candidates. (vii) Each other’s supporters seem to the candidates to be unscrupulous. which he contrasts with: (viii) *Each other’s supporters attacked the candidates. (ix) *Each other’s supporters asked the candidates to be more honest. Lasnik reproduces the standard judgments for (vi)–(ix), but confesses that he is no longer confident that (vi)–(vii) are as good as they are always claimed to be, or that (viii)–(ix) are as bad. He notes that if they do not really contrast, then, needless to say, there is no clear argument for A-movement reconstruction in this case. My informants find a contrast between (vi)–(vii) and (viii)– (ix), and hence I am inclined to believe that the argument for A-reconstruction holds in this case as well (for discussion of Belletti and Rizzi facts, see Pesetsky 1995). 9. A similar proposal has been made independently by Kitahara (1998, 1999) to account for a different set of cases. 10. Note, however, that there is a technical problem in Chomsky’s system that makes in situ elimination unavailable. For Chomsky (2000), an NP can undergo movement if and only if it is active. To be active, NP must have a Case feature; hence Case deletion must follow NP-movement. But the K-features of T can agree with the K-features of N only if T c-commands N. Given that T does not c-command its specifier, K-agreement must precede NP-movement. But if K-agreement automatically executes Case deletion, then NP becomes inactive before undergoing movement. To solve the problem, one could assumes (as Kitahara 1999 proposes) that Agree and re-Merge constitute a single application of Move. Uninterpretable features would delete upon completion of a single application checking all formal features. It is unclear to me whether this truly allows for in situ checking of Case. 11. There is an interesting parallel to be drawn between the present proposal tying Case and scope and Sportiche’s (1997) solution to the absence of A-movement

Notes

12.

13.

14.

15.

237

reconstruction, in which portions of nominals (determiner and quantifier head) are generated outside the thematic domain. Sportiche’s solution allows A-movement to reconstruct. Since the determiner or quantifier crucial to yielding the various readings never started in the lower position, it will not appear in lowest copy. Reconstruction appears to be semantically vacuous. Though interesting, Sportiche’s conclusion necessitates non-trivial departures from basic assumptions about phrase structure. It requires massive sideward movement. It also loses the generalization due to Longobardi (1994) that raising of N to D turns the noun into an argument, which seems to require the presence of D in the thematic domain—precisely what Sportiche does not want. I leave a detailed comparison of his approach with mine for future research. On the view that control is A-movement (Hornstein 1999), I assume that reconstruction of strong quantifiers in this case fails due to the absence of Case checking inside the embedded clause. Double theta-role assignment might also add to the opacity of control structures. For weak quantifiers, the impossibility of there-insertion in control contexts readily accounts for the absence of reconstruction. Sandra Stjepanović (p.c.) raises an interesting question in the present context. She notes that the present theory predicts that inherently Case-marked NPs will exhibit reconstruction effects on the (standard) assumption that inherent Case, unlike structural Case, is interpretable. The prediction appears to be difficult to test. On the one hand, many studies treat inherent Case and structural Case alike for syntactic purposes; i.e., as requiring checking (see Lasnik 1995b, Collins and Thráinsson 1996, Stjepanović 1997). On the other, inherently Case-marked NPs are often treated as immobile (Chomsky 1993, Ura 1996). The of-phrase in the Belletti-Rizzi cases might be relevant, if of is a spell-out of inherent Case, as is defended in Chomsky (1986a). A reviewer also raises an interesting issue. Based on the possibility of reconstruction of genitive phrases (pictures of each other vs. each other’s pictures), it might be possible to pinpoint the exact location of Case on the noun. I leave a full examination of reconstruction with prenominal modifier for another occasion (see Safir 1999 for valuable discussion), as the issue of where Case is located in the NP is tangential to the main point of this chapter. The assumption that Heim and Kratzer’s mechanism does not apply in the syntax is also made by Nissenbaum (2000a), who rejects his earlier (1998) assumption that lambda-abstraction applies at the point in the derivation where movement occurs. Nissenbaum (2000a) assumes instead that lambdaabstraction is simply an interpretive reflex of a configuration involving a chain, and that is implemented as a type-shifting operation that raises the semantic type of its sister from to <e,U>. Bhatt (1999: 103ff.) also makes the observation that radical reconstruction poses a Proper Binding Condition problem once we assume Heim and Kratzer’s mechanism. His point is made in the context of reconstruction in relative clauses. Note that in this context, one cannot adopt a head-raising analysis, the PF-solution Elbourne and Sauerland propose fails since the reconstructing element undergoes further (syntactic) movement after raising out of the relative clause. Given Elbourne and Sauerland’s account, one would have to assume that all further movement is in PF, or else that the element can be reintroduced in the syntax after PF-raising, both of which seem to me implausible. Movement of the relative head has to be syntactic, as the head later participates in logico-syntactic processes like theta-discharge. Bhatt suggests a modification of Heim and Kratzer (1998) to accommodate radical reconstruction that relies on a type adjustment rule type-lifting the predicate into a proposition with a free variable.

238

Notes

16. See Bošković (2001b) for compelling arguments that it does not; see also Stjepanović (1999) and Kayne (2000a). 17. For further empirical problems for Elbourne and Sauerland’s approach, see Den Dikken (1999), who shows the inadequacy of Elbourne (1999), which is central to Elbourne and Sauerland. 18. This point is missed, I believe, in Brown (2000), where an argument is made for A-chains on the basis of Case assignment and interpretation of Case chains in Russian. The readings to be teased apart there revolve around the concept of specificity, for which an derivational account along the lines of Tsai (1998) seems to me to be adequate. 19. If one assumes that expletives have Case (Lasnik 1995b), I take it that thereLF does not have (strong) Case, which may be what allows its insertion in LF, strong Case being forced to be inserted ‘early’ to avoid a crash. 20. On the parallelism between there-sentences and quantifier lowering, see Aoun (1982). 21. The precise mechanism that associate NPs make use of to check their Cases is immaterial for present purposes. Long-distance agreement (Chomsky 2000), partitive Case (Belletti 1988; Lasnik 1995b), and Case-transfer with the expletive (Chomsky 1986a) are viable options. 22. It is interesting to note that Fiengo (1977) argues for expletive insertion to avoid a Proper Binding violation. 23. This solution is reminiscent of Williams’s (1984) view on there as a scope marker. See also Hornstein (1991). 24. The parallel between Beck-effects and unavailable readings due to constraints on quantifier movement has independently been made by Kyle Johnson (1998; class lectures Spring 1999). There is a further interesting parallel to be made between the kind of LF-insertion I am proposing here to capture reconstruction effects and the ‘reprojection’ analysis proposed by Hornstein and Uriagereka (1999) to capture Beck-effects (on ‘reprojection,’ see also Epstein et al. 1998: 70). 25. The presence of an anaphor could block lowering if one assumes an analysis based on anaphor-raising, which would effectively block lowering of the antecedent if the rationale behind anaphor movement is to reach their antecedents. 26. Note, incidentally, that the intervention effect of the experiencer argues in favor of LF-raising of the latter, as in Kitahara (1997), since the experiencer does not block overt raising. 27. It is legitimate to question the need of actual lowering at this point. Suppose we let there enter the derivation at LF (thereLF). Following Chomsky (2000, 2001), we take the expletive to be a probe, trying to agree with a goal (the associate). One could then take this ‘Agree’ relationship to amount to the ‘reactivation’ of an otherwise ‘dead’ copy, namely, the copy left by overt movement of the indefinite. Reactivation is just another term for ‘reconstruction’ (interpretation of a lower copy), and would be worth relating to van Riemsdijk’s (1989) notion of regeneration. However, we will see later that issues of pied-piping arise for what I have so far referred to as LF-lowering, which in turn suggests that actual lowering is more adequate a reactivation approach. 28. The problem was already noted in the GB-era by, for instance, Lasnik and Saito (1992: 139). 29. For arguments in favor of a distinct LF-component, see Richards (1997) and Nissenbaum (1998, 2000a), as well as the work by Bošković and Lasnik mentioned earlier. 30. The need for a distinct LF-component is to be contrasted with the recent attempts to make QR a cyclic, ‘overt’ operation like Kayne (1998), Fox and Nissenbaum (1999), and Pesetsky (2000), among others.

Notes

239

31. Assume for now that there is no AGR projection, and that the subject sits in SpecTP (Chomsky 1995, 2000). 32. Also, I do not think that the English sentence (74a) is derived in the way Frey (1989) and Lechner (1998) have analyzed the German sentence (i): (i)

Ein Elektriker ist nicht gekommen. a electrician is not come

( >> ¬/¬ >> )

They account for the scope ambiguity by appealing to a structural ambiguity: (ii)

a. [Ein Elektriker]i ist [[DP/NP nicht ti] gekommen. b. [Ein Elektriker]i ist [NEG nicht [VP ti gekommen]].

If constituent negation is chosen (and stranded), the ‘¬ >> ’ reading results. If we opt for sentential negation, the other reading comes about. This is clearly not the case for English (74b), where the negation is contracted (contraction is a test for sentential negation), and yet narrow scope for the universal quantifier is available. 33. Of course, one would like to know why standard scope ambiguity sentences (a girl kissed every boy) are direct violations of Relativized Minimality. For a possible explanation reconciling the text’s claim and standard cases of scope ambiguity, see Boeckx (2000d) and Watanabe (2000a). 34. If one believes, with Cinque (1999), that adverbs are XPs occupying Spec positions, neg-raising will be blocked by whatever head hosts the adverb. 35. A reviewer observes that the present account seems to predict that neither (i) nor (ii) will be ambiguous, due to the presence of adverbs: (i)

Everyone wasn’t here (by six).

(ii)

Everyone usually isn’t here (by six).

Since most English speakers I asked did detect an ambiguity in (i)–(ii), the present account appears to be inadequate. However, it is not clear to me that my analysis predicts sentences like (i)–(ii) not to be ambiguous. The adverb in (i) is a locative adverb which is typically taken to occur rather low in the structure (see Cinque 1999). As such, the present account predicts negation to be insensitive to its presence, and free to raise. As for (ii), if it is true that I predict the absence of scope reversal between usually and negation, I leave room for usually to pied-pipe negation when it crosses everyone, thereby allowing negation to take scope over the quantified subject. 36. Baltin (1999) notes that partial negation is unavailable in (ib) (contrast with (ia)): (i)

a. I would prefer for all the men not to leave. b. I would prefer for all the men to not leave.

(neg > all) (*neg > all)

The contrast is easily explained if we take not in (ib) to be a case of constituent negation, as argued persuasively in Wurmbrand (1998). 37. The correlation between ‘lowerable’ quantifiers and partial negation is almost perfect. Baltin (1999) has noted that more than x can be outscoped by negation, as in (i): (i)

More than two people didn’t leave.

(neg > more than)

More than x can occur in an existential context (ii), and, as expected, lowers in the relevant contexts (iii): (ii)

There were more than 10 students at the party.

(iii) More then 10 people from New York are likely to win the lottery.

240

38.

39.

40.

41.

Notes We face a puzzle since here too we expect neg-raising to be blocked and negation to block lowering. Yet negation can take scope over more than x in (i). The explanation for why neg-raising is blocked in (87c,d) is, I think, too deep. It is hard to see how it could be parametrized in such a way as to allow negraising in (i). Although I cannot provide a detailed account of (i) at this point, it seems to me that the presence of a negative element or feature (which we know from the licensing of NPIs in comparative contexts must be present) inside more than could lead to attraction (absorption) of n’t, leading to scope reversal. (On scope reverting absorption of this kind, see Watanabe 2000a.) Baltin (1999) has questioned, correctly, as I now believe, the particular instantiation of the neg-raising analysis given in Boeckx (2000c). The account there assumed that neg-raising took place for pragmatic reasons, which were not present when indefinite subjects were used. This was clearly ad hoc. Baltin’s alternative is that neg-raising is an effect of Spec-head agreement. In addition to being forced to rely on some notion of m-command to characterize scope in this case, a somewhat unminimalist assumption in the light of Epstein (1999), Baltin’s account cannot make the relevant distinctions between the types of quantifiers that may scope under negation (all subject quantifiers are presumably in the same Spec position). Note that Baltin’s argument that the availability of partial negation correlates with the position of neg-licensing cross-linguistically (on which see Laka 1990, Zanuttini 1997) can be directly imported into the present account, which also relates negation and its licensor (4P). Bobaljik and Wurmbrand (2000) also arrive at the conclusion that only indefinites reconstruct, although they do not provide an explanation for this fact. Their conclusion is based on a detailed investigation of the interaction between subject quantifiers and modals, which I have refrained from discussing here, as it is unclear to me whether modal auxiliaries are like other heads in being interpreted in situ. We know from work on the semantics of modality that modals are associated with empty operators, which might be an interfering factor at the time of computing scope relations, and the relevance of modals for quantifier lowering. It is conceivable that the modal operator is higher than the subject to start with, and allows the modal to take scope without the need for the subject to lower. It is significant in this context to note that modalraising is usually said to be the only clear case where head-movement appears not to be semantically vacuous. There would still be the issue of what it means for A-movement not to leave a trace or copy. As noted at the end of Lasnik (1999a), the absence of copy not only seems to force upon us the view that theta-roles are features (contra Chomsky 1995, 2000), but also begs the question of how copies, being terms, can be erased, thereby going against bare phrase structure assumptions. Addressing the issue is beyond the scope of this chapter. Safir (1999) reaches the same conclusion in his detailed discussion of A-bar movement reconstruction. Interestingly, instead of copy-deletion, Safir appeals to a process of vehicle change (treating an R-expression as a pronoun), which could be seen as yet another instance of reprojection.

Bibliography

Abels, Klaus. 2003. Successive Cyclicity, Anti-locality and Adposition Stranding. Doctoral dissertation, University of Connecticut, Storrs. Agbayani, Brian. 1998a. Feature attraction and category movement. Doctoral dissertation, University of California, Irvine. Agbayani, Brian. 1998b. Generalized pied-piping and island effects. In Proceedings of NELS 28, 1–14. GLSA, University of Massachusetts, Amherst. Akmajian, Adrian. 1972. Getting tough. Linguistic Inquiry 3:373–377. Alexiadou, Artemis. 2002. On nominative Case features and split agreement. Ms., University of Stuttgart. Alexiadou, Artemis, and Elena Anagnostopoulou. 1998. Parametrizing AGR. Natural Language and Linguistic Theory 16:491–539. Alexiadou, Artemis, and Elena Anagnostopoulou. 2001. The subject-in-situ generalization and the role of case in driving computations. Linguistic Inquiry 32:193–231. Alexiadou, Artemis, and Elena Anagnostopoulou. 2002. Person and animacy splits: From hierarchies to features. Paper presented at GLOW 25. Anagnostopoulou, Elena. 2003. The Syntax of Ditransitives. Berlin: Mouton de Gruyter. Aoun, Joseph. 1979. Case and theta-theory. Ms., MIT, Cambridge, Mass. Aoun, Joseph. 1982. On the logical nature of the binding principles: Quantifier lowering, double raising of there and the notion empty element. In Proceedings of NELS 12, 16–35. GLSA, University of Massachusetts, Amherst. Aoun, Joseph, Elabbas Benmamoun, and Dominique Sportiche. 1994. Agreement, word order, and conjunction in some varieties of Arabic. Linguistic Inquiry 25:195–220. Aoun, Joseph, and Y.-H. Audrey Li. 2003. Essays on the Representational and Derivational Nature of Grammar: The Diversity of wh-Constructions. Cambridge, Mass.: MIT Press. Atkinson, Martin. 2001. Defective intervention effects, Die! In Essex Research Reports in Linguistics 31, 1–29. Ausín, Adolfo. 1998. Chinese-type questions in English. In Proceedings of WCCFL 17, 30–43. Stanford, Calif.: CSLI. Ausín, Adolfo. 2000. Where does idiom interpretation apply? In Proceedings of NELS 30, 43–54. GLSA, University of Massachusetts, Amherst. Ausín, Adolfo, and Marcela Depiante. 2001. On the syntax of parecer (‘to seem‘) with and without an experiencer. In Hispanic Linguistics at the Turn of the Millennium, ed. by Héctor Campos, Elena Herburger, Alfonso Morales-Front, and Thomas J. Walsh, 155–170. Somerville, Mass.: Cascadilla Press.

242

Bibliography

Baker, Mark. 1985. The Mirror Principle and morphosyntactic explanation. Linguistic Inquiry 16:373–416. Baker, Mark. 1988. Incorporation. Chicago: Chicago University Press. Baker, Mark. 1996. The Polysynthesis Parameter. Oxford: Oxford University Press. Baker, Mark, Kyle Johnson, and Ian Roberts. 1989. Passive arguments raised. Linguistic Inquiry 20:214–251. Baltin, Mark. 1999. A case on non-inverse inverse scope. Ms., New York University. Baltin, Mark, and Paul Postal. 1996. More on reanalysis hypotheses. Linguistic Inquiry 27:127–145. Barbiers, Sjef. 1995. The Syntax of Interpretation. The Hague: HAG. Barss, Andrew. 1986. Chains and anaphoric dependence. Doctoral dissertation, MIT, Cambridge, Mass. Beck, Sigrid. 1996. Quantified structures as barriers for LF-movement. Natural Language Semantics 4:1–46. Bejar, Susana, and Milan Rezac. 2003. Person licensing and the derivation of PCC effects. In Proceedings of LSRL 32, 49–62. Amsterdam: John Benjamins. Belle, William van, and Willy van Langendock. 1996. The Dative (Vol. 1). Amsterdam: John Benjamins. Belletti, Adriana. 1988. The case of unaccusatives. Linguistic Inquiry 19:1–34. Belletti, Adriana. 1990. Generalized Verb Movement. Turin: Rosenberg and Sellier. Belletti, Adriana. 2000. Agreement projections. In The Handbook of Contemporary Syntactic Theory, ed. by Mark Baltin and Chris Collins, 483–510. Oxford: Blackwell. Belletti, Adriana, and Luigi Rizzi. 1988. Psych verbs and theta-theory. Natural Language and Linguistic Theory 6:291–352. Benmamoun, Elabbas. 1997. Agreement in Arabic at the PF-interface. Proceedings of WCCFL 15, 33–48. Stanford, Calif.: CSLI. Benvéniste, Émile. 1966. Relationships of person in the verb. In Problèmes de Linguistique générale, ed. by Émilie Benvéniste, 165–204. Paris: Gallimard. Bhatt, Rajesh. 1999. Covert modality in non-finite contexts. Doctoral dissertation, University of Pennsylvania. Bhatt, Rajesh. 2001. Word order, long distance agreement and incorporation. Paper presented at SALA 01, Konstanz, Germany. Bleam, Tonia. 1999. Leísta Spanish and the syntax of clitic doubling. Doctoral dissertation, University of Delaware. Bobaljik, Jonathan. 1995. Morphosyntax: The syntax of verbal inflection. Doctoral dissertation, MIT, Cambridge, Mass. Bobaljik, Jonathan. 2002. A-Chains at the PF-interface: Copies and “covert” movement. Natural Language and Linguistic Theory 20:197–267. Bobaljik, Jonathan, and Phil Branigan. 2006. Eccentric agreement. In Ergativity, ed. by Diane Massam, Alana Johns and Juvenal Ndayiragije, 47–77. Dordrecht: Kluwer. Bobaljik, Jonathan, and Dianne Jonas. 1996. Subject positions and the role of TP. Linguistic Inquiry 27:195–236. Bobaljik, Jonathan, and Höskuldur Thráinsson. 1998. Two heads aren’t always better than one. Syntax 1:37–71. Bobaljik, Jonathan, and Susi Wurmbrand. 2000. Modals, raising, and reconstruction. Ms., McGill University. Bobaljik, Jonathan, and Kazuko Yatsushiro. 2006. Problems with Honorification-asagreement in Japanese: A reply to Boeckx and Niinuma. Natural Language and Linguistic Theory 24: 355–384. Boeckx, Cedric. 1997. On the contention surrounding the content of AGR(P). Ms., University of Connecticut, Storrs.

Bibliography

243

Boeckx, Cedric. 1998a. Quirky agreement in Icelandic and elsewhere. Working Papers in Scandinavian Syntax 61:1–35. Boeckx, Cedric. 1998b. Raising in Romance. Ms., University of Connecticut, Storrs. Boeckx, Cedric. 1999a. Agree or Attract? Ms., University of Connecticut, Storrs. Boeckx, Cedric. 1999b. Expletive split: Existentials and presentationals. In Proceedings of NELS 29, 57–69. GLSA, University of Massachusetts, Amherst. Boeckx, Cedric. 1999c. Extractions. Ms., University of Connecticut, Storrs. Boeckx, Cedric. 1999d. How did it get there? Ms., University of Connecticut, Storrs. Boeckx, Cedric. 1999e. Remarks on obviation. Ms., University of Connecticut, Storrs. Boeckx, Cedric. 1999f. Remnants of movement. Ms., University of Connecticut, Storrs. Boeckx, Cedric. 1999g. Restrictions on movement in a minimalist program. Ms., University of Connecticut, Storrs. Boeckx, Cedric. 2000a. A note on contraction Linguistic Inquiry 31:357–366. Boeckx, Cedric. 2000b. EPP eliminated. Ms., University of Connecticut, Storrs. Boeckx, Cedric. 2000c. Interpreting A-chains at the interface. In Proceedings of NELS 30, 69–81. GLSA, University of Massachusetts, Amherst. Boeckx, Cedric. 2000d. Interpretation by phase. Ms., University of Connecticut, Storrs. Boeckx, Cedric. 2001. Mechanisms of chain formation. Doctoral dissertation, University of Connecticut, Storrs. Boeckx, Cedric. 2002a. Case and Agreement systems. Presented at the Mayfest 2002. Boeckx, Cedric. 2002b. Quirky Agreement revisited. Ms., University of Illinois, Urbana-Champaign. Boeckx, Cedric. 2003a. Case matters and minimalist concerns. Harvard Working Papers in Linguistics 8, 159–197. Department of Linguistics, Harvard University. Boeckx, Cedric. 2003b. Islands and Chains: Resumption as Stranding. Amsterdam: John Benjamins. Boeckx, Cedric. 2007a. Case and Agreement—The syntax of A-dependencies. Ms., Harvard University. Boeckx, Cedric. 2007b. Understanding Minimalist Syntax: Lessons From Locality in Long Distance Dependencies. Oxford: Blackwell. Boeckx, Cedric. To appear. Bare Syntax. Oxford: Oxford University Press. Boeckx, Cedric, and Norbert Hornstein. 2006. Raising and control. Syntax 9:118–130. Boeckx, Cedric, and Norbert Hornstein. In press. Superiority, reconstruction, and islands. In Foundational Issues in Linguistic Theory, ed. by Robert Freidin, Carlos P. Otero, and Maria-Luisa Zubizaretta. Cambridge, Mass.: MIT Press. Boeckx, Cedric, and Youngmi Jeong. In preparation. Nominative objects in Japanese: Principles and parameters. Ms., University of Maryland, College Park. Boeckx, Cedric, and Sandra Stjepanović. 2001. Head-ing toward PF. Linguistic Inquiry 32:345–355. Bonet, Eulalia. 1994. The Person-Case constraint: A morphological approach. In MIT Working Papers in Linguistics 22, 33–52. Department of Linguistics and Philosophy, MIT, Cambridge, Mass. Bošković, Željko. 1995. Case properties of clauses and the Greed principle. Studia Linguistica 49:32–53. Bošković, Željko. 1997a. On certain violations of the superiority condition, AGRo, and economy of derivation. Journal of Linguistics 33:227–254.

244

Bibliography

Bošković, Željko. 1997b. Superiority effects with multiple wh-fronting in SerboCroatian. Lingua 102:1–20. Bošković, Željko. 1997c. The Syntax of Non-Finite Complementation: An Economy Approach. Cambridge, Mass.: MIT Press. Bošković, Željko. 1998a. LF-Movement and the Minimalist Program. In Proceedings of NELS 28, 43–57. Amherst: University of Massachusetts, GLSA. Bošković, Željko. 1998b. Multiple wh-fronting and economy of derivation. In Proceedings of WCCFL 16, 49–63. Stanford, Calif.: CSLI. Bošković, Željko. 1999. On multiple feature checking: Multiple wh-fronting and multiple head movement. In Working Minimalism, ed. by Samuel D. Epstein and Norbert Hornstein, 159–187. Cambridge, Mass.: MIT Press. Bošković, Željko. 2000a. Sometimes in situ, Sometimes in SpecCP. In Step by Step: Essays on Minimalist Syntax in Honor of Howard Lasnik, ed. by Roger Martin, David Michaels, and J. Uriagereka, 53–87. Cambridge, Mass.: MIT Press. Boskovic, Zeljko. 2000b. What is special about multiple wh-fronting? In Proceedings of NELS 30, 83–107. GLSA, University of Massachusetts, Amherst. Bošković, Željko. 2001a. Floating quantifiers and R-role assignment. In Proceedings of NELS 31, 59–78. GLSA, University of Massachusetts, Amherst. Bošković, Željko. 2001b. On the Nature of the Syntax–Phonology Interface. London: Elsevier. Bošković, Željko. 2002a. A-movement and the EPP. Syntax 5:167–218. Bošković, Željko. 2002b. On multiple wh-fronting. Linguistic Inquiry 33:351–383. Bošković, Željko. 2004. Be careful where you float your quantifiers. Natural Language and Linguistic Theory 22:681–742. Bošković, Željko, and Howard Lasnik. 1999. How strict is the cycle? Linguistic Inquiry 30:689–697. Bošković, Željko, and Daiko Takahashi. 1998. Scrambling and Last Resort. Linguistic Inquiry 29:343–366. Bowers, John. 2002. Transitivity. Linguistic Inquiry 32:183–224. Branigan, Phil. 2000. Binding effects with covert movement. Linguistic Inquiry 31:553–557. Bresnan, Joan. 1971. Sentence stress and syntactic transformations. Language 47:257–281. Brody, Michael. 1995. Lexico-Logical Form. Cambridge, Mass.: MIT Press. Brody, Michael. 1997. Perfect chains. In Elements of Grammar, ed. by Liliane Haegeman, 139–167. Dordrecht: Kluwer. Brody, Michael. 1999. Relating syntactic elements: Remarks on Norbert Hornstein’s “Movement and Chains.” Syntax 2:210–226. Brody, Michael. 2000. Mirror theory. Linguistic Inquiry 31:29–56. Brody, Michael, and Anna Szabolcsi. 2003. Overt scope in Hungarian. Syntax 6:19–51. Brown, Sue. 2000. A-Chain reconstruction: Evidence from Russian. Paper presented at FASL 9, Indiana. Bruening, Benjamin. 2001. QR obeys superiority: ACD and frozen scope. Linguistic Inquiry 32:233–273. Bures, Anton. 1992. Recycling expletives (and other) sentences. Ms., MIT, Cambridge, Mass. Burzio, Luigi. 1981. Intransitive verbs and Italian auxiliaries. Doctoral dissertation, MIT, Cambridge, Mass. Burzio, Luigi. 1986. Italian syntax. Dordrecht: Reidel. Butt, Miriam. 1995. The Structure of Complex Predicates in Urdu. Stanford, Calif.: CSLI. Cardinaletti, Anna. 1997. Subjects and clause structure. In The New Comparative Syntax, ed. by Liliane Haegeman, 33–63. London: Longman.

Bibliography

245

Cardinaletti, Anna, and Michael Starke. 1999. The typology of structural deficiency: A case study of three classes of pronouns. In Clitics in the Languages of Europe, ed. by Henk van Riemsdijk, 185–234. Berlin: Mouton de Gruyter. Castillo, Juan-Carlos, John Drury, and Kleanthes Grohmann. 1999a. Merge over Move and the Extended Projection Principle. In University of Maryland Working Papers in Linguistics 8, 66–103. University of Maryland, College Park. Castillo, Juan-Carlos, John Drury, and Kleanthes Grohmann. 1999b. No more EPP. In Proceedings of WCCFL 19, 153–166. Somerville, Mass.: Cascadilla Press. Cecchetto, Carlo, and Gennaro Chierchia. 1995. Reconstruction in dislocation and the syntax/semantics interface. In Proceedings of WCCFL 17, 132–146. Stanford, Calif: CSLI. Chandra, Pritha. 2007. (Dis)agree: Movement and agreement reconsidered. Doctoral dissertation, University of Maryland, College Park. Chomsky, Noam. 1955. The logical structure of linguistic theory. Ms., MIT/Harvard University, Cambridge, Mass. [Published in part 1975, Plenum, New York.] Chomsky, Noam. 1957. Syntactic Structures. The Hague: Mouton. Chomsky, Noam. 1964. Current Issues in Linguistic Theory. The Hague: Mouton. Chomsky, Noam. 1981. Lectures on Government and Binding. Dordrecht: Foris. Chomsky, Noam. 1982. Some Concepts and Consequences of the Theory of Government and Binding. Cambridge, Mass.: MIT Press. Chomsky, Noam. 1986a. Knowledge of Language. Its Nature, Origin, and Use. New York: Praeger. Chomsky, Noam. 1986b. Barriers. Cambridge, Mass.: MIT Press. Chomsky, Noam. 1991. Some notes on economy of derivation and representation. In Principles and Parameters in Comparative Grammar, ed. by Robert Freidin, 417–454. Cambridge, Mass.: MIT Press. Chomsky, Noam. 1993. A minimalist program for linguistic theory. In The View from Building 20: Essays in linguistics in honor of Sylvain Bromberger, ed. by Ken Hale and Samuel Jay Keyser, 1–52. Cambridge, Mass.: MIT Press. Chomsky, Noam. 1995. The Minimalist Program. Cambridge, Mass.: MIT Press. Chomsky, Noam. 2000. Minimalist inquiries: The framework. In Step by Step: Essays on Minimalist Syntax in Honor of Howard Lasnik, ed. by Roger Martin, David Michaels, and Juan Uriagereka, 89–151. Cambridge, Mass.: MIT Press. Chomsky, Noam. 2001. Derivation by phase. In Ken Hale: A Life in Language, ed. by Michael Kenstowicz, 1–52. Cambridge, Mass.: MIT Press. Chomsky, Noam. 2004. Beyond Explanatory Adequacy. In Structures and Beyond— The Cartography of Syntactic Structure (Vol. 3), ed. by Adriana Belletti, 104–131. Oxford: Oxford University Press. Chomsky, Noam, and Howard Lasnik. 1977. Filters and control. Linguistic Inquiry 7:425–504. Chomsky, Noam, and Howard Lasnik. 1993. The theory of principles and parameters. In Syntax: An International Handbook of Contemporary Research, ed. by Joachim Jacobs, Arnim von Stechow, Wolfgang Sternefeld, and Theo Vennemann, 506–569. Berlin: Mouton de Gruyter. Chung, Sandra. 1998. The Design of Agreement. Chicago: Chicago University Press. Cinque, Guglielmo. 1990. Types of A-Bar Dependencies. Cambridge, Mass.: MIT Press. Cinque, Guglielmo. 1999. Adverbs and Functional Heads: A Cross-Linguistic Perspective. Oxford: Oxford University Press. Collins, Chris. 1994. Economy of derivation and the generalized proper binding condition. Linguistic Inquiry 25:45–61. Collins, Chris. 1997. Local Economy. Cambridge, Mass.: MIT Press. Collins, Chris. 2002. Eliminating labels. In Explanation and Derivation in the Minimalist Program, ed. by Samuel D. Epstein and T. Daniel Seely, 42–64. Oxford: Blackwell.

246

Bibliography

Collins, Chris, and Höskuldur Thráinsson. 1996. VP-Internal Structure and Object Shift in Icelandic. Linguistic Inquiry 27:391–449. Cooper, Robin. 1979. The interpretation of pronouns. In Syntax and Semantics 10, ed. by Frank Heny and Helmut S. Schnelle, 61–92. New York: Academic press. Cresti, Diana. 1995. Extraction and reconstruction. Natural Language Semantics 3:79–122. Culicover, Peter. 1992. Polarity, inversion, and focus in English. Proceedings of ESCOL’91: Proceedings of the Eighth Eastern States Conference on Linguistics, 46–68. Department of Linguistics, Ohio State University, Columbus. D’Alessandro, Roberta. 2002. Agreement in Italian impersonal constructions: A phase-based account. Ms., University of Stuttgart. De Hoop, Helen. 1992. Case configuration and noun phrase interpretation. Doctoral dissertation, Groningen University. Delsing, Lars-Olof. 1996. The reduction of a Case system—Old Swedish. Talk delivered to the Linguistic Institute, University of Tromsø. Dikken, Marcel den. 1994. Auxiliaries and participles. In Proceedings of NELS 24, 65–79. GLSA, University of Massachusetts, Amherst. Dikken, Marcel den. 1995a. Particles. Oxford: Oxford University Press. Dikken, Marcel den. 1995b. Binding, expletives, and levels. Linguistic Inquiry 26:347–354. Dikken, Marcel den. 1999. Pluringulars. Paper presented at ESCOL ’99, University of Connecticut. Dikken, Marcel den. 2000. The syntax of features. Journal of Psycholinguistic Research 29:5–23. Dikken, Marcel den. 2001. Pluringulars, pronouns, and Quirky agreement. The Linguistic Review 18, 19–41. Déprez, Viviane. 1998. Semantic effects of agreement: The case of French past participle agreement. Probus 10:1–65. Elbourne, Paul. 1999. Some correlations between semantic plurality and quantifier scope. In Proceedings of NELS 29, 81–92. GLSA, University of Massachusetts, Amherst. Elbourne, Paul, and Uli Sauerland. 1999. Total reconstruction, PF-movement, and derivational order. Ms., MIT, Cambridge, Mass., and Tübingen University. Enç, Mürvet. 1991. The semantics of specificity. Linguistic Inquiry 22:1–26. Epstein, Samuel D. 1989. Quantification in null operator constructions. Linguistic Inquiry 20:647–658. Epstein, Samuel D. 1999. Un-principled syntax: The derivation of syntactic relations. In Working Minimalism, ed. by Samuel D. Epstein and Norbert Hornstein, 317–345. Cambridge, Mass.: MIT Press. Epstein, Samuel D., Erich Groat, Ruriko Kawashima, and Hisatsugu Kitahara. 1998. The Derivation of Syntactic Relations. Oxford: Oxford University Press. Epstein, Samuel D., and T. Daniel Seely. 2006. Derivations in Minimalism. Cambridge: Cambridge University Press. Eythórsson, Thórhallur. 2000. Dative vs. nominative: Changes in quirky subjects in Icelandic. In Leeds Working Papers in Linguistics 8, 1–28. Department of Linguistics and Phonetics, Leeds University. Ferguson, K. Scott 1994. Deriving the invisibility of PP-nodes for Command from AGR-P+P Case checking. In Harvard Working Papers in Linguistics 4, 30–36. Department of Linguistics, Harvard University. Ferguson, K. Scott 1996. A feature-relativized shortest move requirement. Doctoral dissertation, Harvard University. Fernández-Soriano, Olga. 1999. Two types of impersonal sentences in Spanish: Locative and dative subjects. Syntax 2:101–140. Fiengo, Robert. 1977. On Trace Theory. Linguistic Inquiry 8:35–61.

Bibliography

247

Folli, Rafaella, and Heidi Harley. 2005. Flavors of v: Consuming results in Italian and English. In Aspectual Inquires, ed. by Paula Kempchinsky and Roumyana Slabakova, 95–120. Dordrecht: Springer. Fox, Danny. 1995. Economy and scope. Natural Language Semantics 3:283–341. Fox, Danny. 1998. Economy and semantic interpretation. Doctoral dissertation, MIT, Cambridge, MIT. Fox, Danny. 1999. Reconstruction, binding theory, and the interpretation of chains. Linguistic Inquiry 30:157–196. Fox, Danny. 2000. Economy and Semantic Interpretation. Cambridge, Mass.: MIT Press. Fox, Danny, and Jon Nissenbaum. 1999. Scope and extraposition: A case for overt QR. In Proceedings of WCCFL 18, 132–144. Somerville, Mass.: Cascadilla Press. Frampton, John. 1999. The fine structure of wh-movement and the proper formulation of the ECP. The Linguistic Review 16:43–61. Frampton, John, and Sam Gutmann. 2000. Agreement as feature sharing. Ms., Northeastern University. Frampton, John, and Sam Gutmann. 2002. Crash-proof syntax. In Derivation and Explanation in the Minimalist Program, ed. by Samuel D. Epstein and T. Daniel Seely, 90–105. Oxford: Blackwell. Fujii, Tomohiro. 2002. Long-distance A-movement. Ms., University of Maryland, College Park. Fujita, Koji. 1996. Double objects, causatives, and derivational economy. Linguistic Inquiry 27:146–173. Fukui, Naoki. 1986. A theory of category projection and its application. Doctoral dissertation, MIT, Cambridge, Mass. Fukui, Naoki. 1995. The Principles-and-Parameters approach: A comparative syntax of English and Japanese. In Approaches to Language Typology, ed. by Masayoshi Shibatani and Theodora Bynon, 327–372. Oxford: Oxford University Press. Fukui, Naoki. 1997. Attract and the A-over-A Principle. In UCI Working Papers in Linguistics 3, 51–67. Department of Linguistics, University of California, Irvine. Fukui, Naoki. 1999. The multiplicity parameter. Glot International 4:26–27. Gallego, Angel. 2007. Phase theory and parametric variation. Doctoral dissertation, Universitat Autònoma de Barcelona. George, Leland, and Jaklin Kornfilt. 1981. Finiteness and boundedness in Turkish. In Binding and Filtering, ed. by Frank Heny, 105–127. Cambridge, Mass.: MIT Press. Goddall, Grant. 2001. Subjects and the EPP in Spanish. In Objects and Other Subjects, ed. by William Davies and Stanley Dubinsky, 193–223. Dordrecht: Kluwer. Groat, Erich. 1999. Raising the case of expletives. In Working Minimalism, ed. by Samuel D. Epstein and Norbert Hornstein, 27–43. Cambridge, Mass.: MIT Press. Groat, Erich, and John O’Neil. 1996. Spell-Out at the LF interface. In Minimal Ideas, ed. by Werner Abraham, Samuel D. Epstein, Höskuldur Thráinsson and C. JanWouter Zwart, 113–139. Amsterdam: John Benjamins. Grohmann, Kleanthes K. 1998. Syntactic inquiries into discourse restrictions on multiple interrogatives. In Groninger Arbeiten zur germanistischen Linguistik 4, 21–60. Grohmann, Kleanthes K. 2000. Prolific peripheries: A radical view from the left. Doctoral dissertation, University of Maryland, College Park. Grohmann, Kleanthes K. 2003a. German is a multiple wh-fronting language! In Multiple Wh-Fronting, ed. by Cedric Boeckx and Kleanthes K. Grohmann, 99–130. Amsterdam: John Benjamins. Grohmann, Kleanthes K. 2003b. Prolific domains. Amsterdam: John Benjamins.

248

Bibliography

Haeberli, Eric. 2002. Features, Categories, and the Syntax of A-Positions. Dordrecht: Kluwer. Hagstrom, Paul. 1998. Decomposing questions. Doctoral dissertation, MIT, Cambridge, Mass. Hale, Ken. 1973. Person marking in Walbiri. In A Festshrift for Morris Halle, ed. by Stephen Anderson and Paul Kiparsky, 308–344. New York: Holt, Rinehart, and Winston. Hale, Ken, and Samuel Jay Keyser. 1993. On the argument structure and lexical expression of syntactic relation. In The View from Building 20: Essays in Linguistics in Honor of Sylvain Bromberger, ed. by Ken Hale and Samuel Jay Keyser, 53–109. Cambridge, Mass.: MIT Press. Halle, Morris, and Alec Marantz. 1993. Distributed morphology and the pieces of inflection. In The View from Building 20: Essays in Linguistics in Honor of Sylvain Bromberger, ed. by Ken Hale and Samuel Jay Keyser, 111–176. Cambridge, Mass.: MIT Press. Halpern, Aaron. 1995. On the Placement and Morphology of Clitics. Stanford, Calif.: CSLI. Harada, Shin-Ichi. 1976. Honorifics. In Syntax and Semantics 5, ed. by Masayoshi Shibatani, 499–561. New York: Academic Press. [Republished 2000 in: Syntax and Meaning: S. I. Harada Collected Works in Linguistics, ed. by Naoki Fukui, 265–338. Tokyo: Taishukan. Page references to this edition.] Harley, Heidi. 1995a. Abstracting away from abstract Case. Proceedings of NELS 25, 207–221. GLSA, University of Massachusetts, Amherst. Harley, Heidi. 1995b. Subject, events, and licensing. Doctoral dissertation, MIT, Cambridge, Mass. Hasegawa, Nobuko. 2002. Honorifics. Ms., Kanda University. Hauser, Mark D., Noam Chomsky, and W. Tecumseh Fitch. 2002. The faculty of language: What is it, who has it, and how did it evolve? Science 298:1569–1579. Heck, Fabian, and Gereon Müller. 2000. Successive cyclicity, long-distance superiority, and local optimization. In Proceedings of WCCFL 19, 218–231. Somerville, Mass.: Cascadilla Press. Heim, Irene, and Angelika Kratzer. 1998. Semantics in Generative Grammar. Oxford: Blackwell. Henderson, Brent. 2006. The syntax and typology of Bantu relative clauses. Doctoral dissertation, University of Illinois, Urbana-Champaign. Higginbotham, James. 1995. Semantic computation. Ms., Oxford University, Oxford. Higginbotham, James. 1998. Why is sequence of tense obligatory. Ms., Oxford University, Oxford. Hiraiwa, Ken. 2001. Multiple agree and the defective intervention constraint in Japanese. In MIT Working Papers in Linguistics 40: Proceedings of HUMIT 2000, 67–80. Department of Linguistics and Philosophy, MIT, Cambridge, Mass. Hiraiwa, Ken. 2005. Dimensions of symmetry in syntax: Agreement and clausal architecture. Doctoral dissertation, MIT, Cambridge, Mass. Hoekstra, Teun. 1987. SOV en extrapositie. Tabu 17:133–142. Hoekstra, Teun, and Réne Mulder. 1990. Unergative as copular verbs: Locational and existential predication. The Linguistic Review 7:1–79. Hoji, Hajime. 1985. Logical form constraints and configurational structures in Japanese. Doctoral dissertation, University of Washington. Hoji, Hajime. 1998. Otagai. Ms., University of Southern California. Hoji, Hajime. 2003. Falsifiability and repeatability in generative grammar: A case study of anaphora and scope dependency in Japanese. Lingua 113:377–446. Holmberg, Anders. 1986. Word order and syntactic features in Scandinavian languages. Doctoral dissertation, University of Stockholm.

Bibliography

249

Holmberg, Anders. 2000. Scandinavian stylistic fronting: How any category can become an expletive. Linguistic Inquiry 31:445–483. Holmberg, Anders, and Thorbjörg Hróarsdóttir. 2001. Agreement and movement in Icelandic raising constructions. Working Papers in Scandinavian Syntax 69:147–168. [Revised publication in Lingua 113:997–1019.] Hornstein, Norbert. 1990. Verb raising in Icelandic infinitives. In Proceedings of NELS 20, 215–229. GLSA, University of Massachusetts, Amherst. Hornstein, Norbert. 1991. Expletives: A comparative study of English and Icelandic. Working Papers in Scandinavian Syntax 47:1–88. Hornstein, Norbert. 1995. Logical Form: From GB to Minimalism. Oxford: Blackwell. Hornstein, Norbert. 1998. Movement and chains. Syntax 1: 99–127. Hornstein, Norbert. 1999. Movement and control. Linguistic Inquiry 30:69–96. Hornstein, Norbert. 2001a. Move! A Minimalist Theory of Construal. Oxford: Blackwell. Hornstein, Norbert. 2001b. On A-chains: A reply to Brody. Syntax 3:129–143. Hornstein, Norbert. 2007. A theory of syntax. Ms., University of Maryland, College Park. [To appear with Cambridge University Press.] Hornstein, Norbert, and Juan Uriagereka. 1999. Labels and projections: A note on the syntax of quantifiers. In University of Maryland Working Papers in Linguistics 8, 249–270. University of Maryland, College Park. [Also published as Hornstein, Norbert, and Juan Uriagereka. 2002. Reprojections. In Derivation and Explanation in the Minimalist Program, ed. by Samuel David Epstein and T. Daniel Seely, 106–132. Malden, Mass.: Blackwell.] Hornstein, Norbert, and Amy Weinberg. 1981. Case theory and preposition stranding. Linguistic Inquiry 12:55–92. Hornstein, Norbert, and Jacek Witkos. 2003. Yet another approach to existential constructions. In Grammar in Focus: Festschrift for Christer Platzack 18 November 2003, ed. by Lars-Olof Delsing, Cecilia Falk, Gunlög Josefsson, and Halldór Á. Sigurðsson, 167–184. Lund University: Department of Scandinavian Languages. Hoshi, Koji. 1996. Multiple case-checking in the head-internal relativization in Japanese. In MIT Working Papers in Linguistics 29, 21–40. Department of Linguistics and Philosophy, MIT, Cambridge, Mass. Huang, James. 1993. Reconstruction and the structure of VP: Some theoretical consequences. Linguistic Inquiry 24:103–138. Jackendoff, Ray. 1972. Semantic Interpretation in Generative Grammar. Cambridge, Mass.: MIT Press. Jaeggli, Osvaldo, and Ken Safir. 1989. The Null Subject Parameter. Dordrecht: Foris. Jelinek, Eloise. 1984. Empty categories, case, and configurationality. Natural Language and Linguistic Theory 2:39–76. Jeong, Youngmi. 2002. Clitic doubling strategies. Ms., University of Maryland, College Park. Jeong, Youngmi. 2004. Asymmetries in multiple case assignment. In Proceedings of WCCFL 23, 413–421. Somerville, Mass.: Cascadilla Press. Jeong, Youngmi. 2006. The landscape of applicatives. Doctoral dissertation, University of Maryland, College Park. Jeong, Youngmi. 2007a. Applicatives. Amsterdam: John Benjamins. Jeong, Youngmi. 2007b. Multiple Wh-fronting in Basque. In University of Maryland Working Papers in Linguistics, 98–142. University of Maryland, College Park. Johnson, Kyle. 1991. Object positions. Natural Language and Linguistic Theory 9:577–636. Johnson, Kyle. 1994. Bridging the gap. Ms., University of Massachusetts, Amherst.

250

Bibliography

Johnson, Kyle. 1998. Ways of hiding scope. Paper presented at MIT, Cambridge, Mass. Jónsson, Jóhannes Gísli. 1991. Stylistic fronting in Icelandic. Working Papers in Scandinavian Syntax 48:1–43. Jónsson, Jóhannes Gísli. 1996. Clausal architecture and case in Icelandic. Doctoral dissertation, University of Massachusetts, Amherst. Jónsson, Jóhannes Gísli. 2001. Not so quirky: On subject case in Icelandic. Ms., University of Iceland. Julien, Marit. 2000. Syntactic heads and word formation: A study of verbal inflection. Doctoral dissertation, University of Tromsø. [Published 2002 as Syntactic Heads and Word Formation. Oxford: Oxford University Press.] Kany, Charles. 1976. Sintaxis Hispanoamericana. Madrid: Gredos. Katz, Jerrold, and Paul Postal. 1964. An Integrated Theory of Linguistic Descriptions. Cambridge, Mass.: MIT Press. Kayne, Richard S. 1975. French Syntax: The Transformational Cycle. Cambridge, Mass.: MIT Press. Kayne, Richard S. 1984. Connectedness and Binary Branching. Dordrecht: Foris. Kayne, Richard S. 1985. Principles of particle constructions. In Grammatical Representation, ed. by Jaqueline Guéron, Jean-Yves Pollock, and Hans-Georg Obenuer, 101–140. Dordrecht: Foris. Kayne, Richard S. 1989a. Facets of Romance past participle agreement. In Dialect Variation and the Theory of Grammar, ed. by Paola Beninca, 85–103. Dordrecht: Foris. [Reprinted in Kayne 2000b: 25–39.] Kayne, Richard S. 1989b. Notes on English agreement. CIEFL Bulletin 1:41–67. [Reprinted in Kayne 2000b: 187–205.] Kayne, Richard S. 1993. Toward a modular theory of auxiliary selection. Studia Linguistica 57:3–31. [Reprinted in Kayne 2000b: 107–130.] Kayne, Richard S. 1994. The Antisymmetry of Syntax. Cambridge, Mass.: MIT Press. Kayne, Richard S. 1995. Agreement and verb morphology in three varieties of English. In Studies in Comparative Germanic Syntax, ed. by Hubert Haider, Susan Olsen, and Sten Vikner, 159–165. Dordrecht: Kluwer. [Reprinted in Kayne 2000b: 206–211.] Kayne, Richard S. 1998. Overt vs. covert movement. Syntax 1:128–191. [Reprinted in Kayne 2000: 223–281.] Kayne, Richard S. 2000a. On the left edge in UG: A reply to McCloskey. Syntax 3:44–51. Kayne, Richard S. 2000b. Parameters and Universals. Oxford: Oxford University Press. Kayne, Richard S. 2001. Raising, reconstruction, and remnant movement. Talk given at the Asymmetry conference, May, UQAM. Kayne, Richard S. 2004. Prepositions as probes. In Structures and Beyond: The Cartography of Syntactic Structures (Vol. 3), ed. by Adriana Belletti, 192–212. Oxford: Oxford University Press. [Reprinted in: Kayne, Richard S. 2005. Movement and Silence, 84–104. New York: Oxford University Press.] Kim, Kwang-Sup. 2000. (Anti-)connectivity. Ms., Hanshin, Seoul. Kitahara, Hisatsugu. 1997. Elementary Operations and Optimal Derivations. Cambridge, Mass.: MIT Press. Kitahara, Hisatsugu. 1999. A derivational interpretation of scrambling sites. Paper presented at the 2nd Asian GLOW Conference. Kitahara, Hisatsugu. 2002. A derivational analysis of scrambling sites. In Derivation and Explanation in the Minimalist Program, ed. by Samuel D. Epstein and T. Daniel Seely, 167–183. Oxford: Blackwell.

Bibliography

251

Koizumi, Masatoshi. 1993. Object agreement phrases and the split VP hypothesis. In MIT Working Papers in Linguistics 18, 99–148. Department of Linguistics and Philosophy, MIT, Cambridge, Mass. Koizumi, Masatoshi. 1994. Secondary predicates. Journal of East Asian Linguistics 3:25–79. Koizumi, Masatoshi. 1995. Phrase Structure in Minimalist Syntax. Doctoral dissertation, MIT, Cambridge, Mass. Koopman, Hilda. 1992. On the absence of case chains in Bambara. Natural Language and Linguistic Theory 10:555–594. Koopman, Hilda. 1995. The spec-head relation. Ms., UCLA. Koopman, Hilda. 1999. The internal and external distribution of pronominal DPs. In Beyond Principles and Parameters, ed. by Kyle Johnson and Ian Roberts, 91–132. Dordrecht: Kluwer. Koopman, Hilda. 2001. DP-structure in Masaai. Ms., UCLA. Koopman, Hilda, and Anna Szabolcsi. 2000. Verbal Complexes. Cambridge, Mass.: MIT Press. Kuroda, Sige-Yuki. 1992. Japanese Syntax and Semantics. Dordrecht: Kluwer. Kratzer, Angelika. 1991. Modality. In Semantics. An International Handbook of Contemporary Research, ed. by Arnim von Stechow and Dieter Wunderlich, 639–650. Berlin: Mouton de Gruyter. Krifka, Manfred. 1998. Scope inversion under the rise–fall contour in German. Linguistic Inquiry 29:75–112. Kroch, Anthony, and Aravind Joshi. 1985. The linguistic relevance of tree-adjoining grammar. Ms., University of Pennsylvania. Ladusaw, William. 1980. Polarity Sensitivity as Inherent Scope Relations. New York: Garland. Laka, Itziar. 1990. Negation in syntax. Doctoral dissertation, MIT, Cambridge, Mass. Larson, Richard. 1988. On the double object construction. Linguistic Inquiry 19:335–391. Lasnik, Howard. 1972. Analyses of negation in English. Doctoral dissertation, MIT, Cambridge, Mass. Lasnik, Howard. 1976. Remarks on coreference. Linguistic Analysis 2:1–22. Lasnik, Howard. 1989. Essays on Anaphora. Dordrecht: Kluwer. Lasnik, Howard. 1992. Case and expletives: Notes toward a parametric account. Linguistic Inquiry 23:381–406. Lasnik, Howard. 1995a. A note on pseudogapping. In MIT Working Papers in Linguistics 27: Papers in Minimalist Syntax, 143–163. Department of Linguistics and Philosophy, MIT, Cambridge, Mass. [Reprinted in: Lasnik 1999a, 151–174.] Lasnik, Howard. 1995b. Case and expletives revisited: On Greed and other human failings. Linguistic Inquiry 26:615–633. [Reprinted in Lasnik 1999a, 74–96.] Lasnik, Howard. 1995c. Last Resort. In Minimalism and Linguistic Theory, ed. by Shosuke Haraguchi and Michio Funaki, 1–32. Tokyo: Hituzi Syobo. [Reprinted in Lasnik 1999a, 120–150.] Lasnik, Howard. 1995d. Last Resort and Attract F. In Papers From the Sixth Annual Meeting of the Formal Linguistics Society of Midamerica, 62–81. Indiana University, Bloomington, Indiana University Linguistics Club. Lasnik, Howard. 1998a. Some reconstruction riddles. In Penn Working Papers in Linguistics 5, 83–98. Lasnik, Howard. 1998b. On a scope reconstruction paradox. Ms., University of Connecticut, Storrs. [Reprinted in Lasnik 2003, 118–124.] Lasnik, Howard. 1999a. Minimalist Analysis. Oxford: Blackwell. Lasnik, Howard. 1999b. On feature strength: Three minimalist approaches to overt movement. Linguistic Inquiry 30:197–217.

252

Bibliography

Lasnik, Howard. 1999c. Chains of arguments. In Working Minimalism, ed. by Samuel D. Epstein and Norbert Hornstein, 189–215. Cambridge, Mass.: MIT Press. Lasnik, Howard. 1999d. Some alternative principles of particle constructions. Paper presented at UQAM. Lasnik, Howard. 2001a. Derivation and representation in modern transformational syntax. In The Handbook of Contemporary Syntactic Theory, ed. by Mark Baltin and Chris Collins, 62–88. Malden: Blackwell. Lasnik, Howard. 2001b. Subjects, objects, and the EPP. In Objects and Other Subjects, ed. by William D. Davies and Stanley Dubinsky, 103–121. Dordrecht: Kluwer. Lasnik, Howard. 2002. Feature movement or agreement at a distance? In Dimensions of Movement, ed. by Artemis Alexiadou, Elena Anagnostopoulou, Sjef Barbiers and Hans-Martin Gärtner, 189–208. Amsterdam: John Benjamins. Lasnik, Howard. 2003. Minimalist Investigations in Linguistic Theory. London: Routledge. Lasnik, Howard, and Joseph Kupin. 1977. A restrictive theory of transformational grammar. Theoretical Linguistics 4:173–195. Lasnik, Howard, and Mamoru Saito. 1991. On the subject of infinitives. In Papers from the 27th Regional Meeting of CLS, 324–343. University of Chicago. Lasnik, Howard, and Mamoru Saito. 1992. Move B. Cambridge, Mass.: MIT Press. Law, Paul S. 1991. Effects of head movement on theories of subjacency and proper government. Doctoral dissertation, MIT, Cambridge, Mass. Law, Paul S. 1993. On the base position of wh-adjuncts and extraction. Paper presented at the 67th Annual Meeting of the Linguistic Society of America, Los Angeles. Lebeaux, David. 1983. A distributional difference between reciprocals and reflexives. Linguistic Inquiry 14:723–730. Lebeaux, David. 1988. Language acquisition and the form of the grammar. Doctoral dissertation, University of Massachusetts, Amherst. Lebeaux, David. 1998. Where does the binding theory apply (II)? Ms., NEC Research Institute. Lechner, Winfried. 1998. Two kinds of reconstruction. Studia Linguistica 52:276–310. Ledgeway, Adam. 2000. A Comparative Syntax of the Dialects of Southern Italy: A Minimalist Approach. Oxford: Blackwell. Lee, Felicia Ann. 1999. Antisymmetry and the syntax of San Luca Quiavini Zapotec. Doctoral dissertation, UCLA. Lee, Ju-Eun. 2004. Ditransitive structures and (anti-)locality. Doctoral dissertation, Harvard University. Longobardi, Guiseppe. 1994. Reference and proper names: A theory of N-raising in syntax and logical form. Linguistic Inquiry 25:609–665. Lopez, Luis. 2002. On agreement: Locality and feature-valuation. In Formal Approaches to Universals, ed. by Artemis Alexiadou, 165–209. Amsterdam: John Benjamins. Lyle, James. 1997. Oblique subjects and nominative anaphors. In Proceedings of WCCFL XV, 321–336. Stanford, Calif.: CSLI. Mahajan, Anoop. 1989. Agreement and agreement phrases. In MIT Working Papers in Linguistics 10: Functional Heads and Clause Structure, 217–252. Department of Linguistics and Philosophy, MIT, Cambridge, Mass. Mahajan, Anoop. 2000. Eliminating head-movement. Ms., University of California, Los Angeles. Maling, Joan. 1984. None-clause-boundedness reflexives in Modern Icelandic. Linguistics and Philosophy 7:211–241. Manzini, Rita, and Anna Roussou. 2000. A Minimalist Theory of A-movement and control. Lingua 110:409–447.

Bibliography

253

Manzini, M. Rita, and Leonardo M. Savoia. 2001. Clitics: Coocurrence and mutual exclusion patterns. Ms., University of Florence. Marantz, Alec. 2000. Words. Ms., MIT, Cambridge, Mass. Martin, Roger. 1992. On the distribution and content of PRO. Ms., University of Connecticut, Storrs. Martin, Roger. 1996. A minimalist theory of PRO and control. Doctoral dissertation, University of Connecticut, Storrs. Martin, Roger. 1999. Case, the Extended Projection Principle, and minimalism. In Working Minimalism, ed. by Samuel D. Epstein and Norbert Hornstein, 1–25. Cambridge, Mass.: MIT Press. Martin, Roger, and Juan Uriagereka. 1998. Collapsed waves in natural languages. Ms., Tsukuba University and University of Maryland, College Park. Martinez-Miera, Carmen. 2002. Defective intervention effects in Spanish. Ms., University of Illinois, Urbana-Champaign. May, Robert. 1977. The grammar of quantification. Doctoral dissertation, MIT, Cambridge, Mass. May, Robert. 1985. Logical Form. Cambridge, Mass.: MIT Press. Mayr, Clemens. 2006. Internal Merge triggers pied-piping: On remerge and left branches. Ms., Harvard University. Mayr, Clemens. 2007. Stylistic inversion and VP externalization. Ms., Harvard University. McCloskey, Jim. 2000. Quantifier float and wh-movement in an Irish English. Linguistic Inquiry 31:57–84. McGinnis, Martha. 1998. Locality in A-movement. Doctoral dissertation, MIT, Cambridge, Mass. McGinnis, Martha. 2003. Variation in the syntax of applicatives. Linguistic Variation Yearbook 1:105–146. McGinnis, Martha. 2005. UTAH at Merge: Evidence from multiple applicatives. In MIT Working Papers in Linguistics 49: Perspectives on Phases, 182–200. Department of Linguistics and Philosophy, MIT, Cambridge, Mass. Miyagawa, Shigeru. 1997. Against optional scrambling. Linguistic Inquiry 28:1–25. Miyagawa, Shigeru, and Takae Tsujioka. 2004. Argument structure and ditransitive verbs in Japanese. Journal of East Asian Linguistics 13:1–38. Moravcsik, Edith. 1978. Agreement. In Universals of Human Language, ed. by Joseph Greenberg, 331–374. Stanford, Calif.: Stanford University Press. Moro, Andrea. 1997. The Raising of Predicates. Cambridge: Cambridge University Press. Moro, Andrea. 2000. Dynamic Antisymmetry. Cambridge, Mass.: MIT Press. Moro, Andrea. 2007. Some notes on unstable structures. Ms., Harvard University. Murasugi, Keiko. 1995. Head-internal relative clauses and the Pro-drop parameter. Ms., Kinjo Gakuin University. Müller, Gereon. 1998. Incomplete Category Fronting. Dordrecht: Kluwer. Müller, Gereon. 2004. Phase impenetrability and wh-intervention. In Minimality Effects in Syntax, ed. by Arthur Stepanov, Gisbert Fanselow, and Ralf Vogel, 286–325. Berlin: Mouton de Gruyter. Nemoto, Naoko. 1993. Chains and case positions: A study from scrambling in Japanese. Doctoral dissertation, University of Connecticut. Nevins, Andrew. 2003. Ergativity. Ms., MIT, Cambridge, Mass. Newmeyer, Frederick. 1975. English Aspectual Verbs. The Hague: Mouton. Niinuma, Fumikazu. 1999. On nominative objects and overt A-movement in Japanese. In ESCOL ’99: Proceedings of the Eighth Eastern States Conference on Linguistics, 149–160. Cornell University. Niinuma, Fumikazu. 2000. On the role of postposition in Japanese. Ms., University of Connecticut, Storrs.

254

Bibliography

Niinuma, Fumikazu. 2003. The syntax of honorification. Doctoral dissertation, University of Connecticut, Storrs. Nishioka, Nobuaki. 1997. On trace movement. English Linguistics 14:182–202. Nissenbaum, Jon. 1998. Movement and derived predicates: Evidence from parasitic gaps. In MIT Working Papers in Linguistics 25: The Interpretive Tract, 247–295. Department of Linguistics and Philosophy, MIT, Cambridge, Mass. Nissenbaum, Jon. 2000a. Covert movement and parasitic gaps. In Proceedings of NELS 30, 541–556. GLSA, University of Massachusetts, Amherst. Nissenbaum, Jon. 2000b. Investigations of covert phrase movement. Doctoral dissertation, MIT, Cambridge, Mass. Nomura, Masashi. 2001. Extraposition or scattered deletion. Ms., University of Connecticut, Storrs. Nomura, Masashi. 2002a. Multiple case marking via multiple match. Ms., University of Connecticut, Storrs. Nomura, Masashi. 2002b. Case and agreement in Icelandic and Japanese: Multiple Case-marking and multiple phi-feature agreement. Ms., University of Connecticut. Nunes, Jairo. 1999. Linearization of chains and the phonetic realizations of chain links. In Working Minimalism, ed. by Samuel D. Epstein and Norbert Hornstein, 217–249. Cambridge, Mass.: MIT Press. Nunes, Jairo. 2004. Linearization of Chains and Sideward Movement. Cambridge, Mass.: MIT Press. Obenauer, Hans-Georg. 1984. On the identification of empty categories. The Linguistic Review 4:153–202. Obenauer, Hans-Georg. 1994. Aspects de la syntaxe A-barre. Doctoral dissertation, Université de Paris VIII. Ochi, Masao. 1998. Move or Attract? In Proceedings of WCCFL 16, 319–333. Stanford, Calif.: CSLI. Ochi, Masao. 1999a. Constraints on feature checking. Doctoral dissertation, University of Connecticut, Storrs. Ochi, Masao. 1999b. Some consequences of Attract F. Lingua 109: 81–107. Ormazabal, Javier. 2000. A conspiracy theory of case and agreement. In Step by Step: Essays in Minimalist Syntax in Honor of Howard Lasnik, ed. by Roger Martin, David Michaels, and Juan Uriagereka, 235–260. Cambridge, Mass.: MIT Press. Ormazabal, Javier, and Juan Romero. 1998. The me-lui constraint. Ms., University of the Basque Country and MIT, Cambridge, Mass. Ormazabal, Javier, and Juan Romero. 2007. The object agreement constraint. Natural Language and Linguistic Theory 25:315–347. Partee, Barbara. 1971. On the requirement that transformations preserve meaning. In Studies in Linguistic Semantics, ed. by Charles Fillmore and D. Terence Langendoen, 1–22. New York: Holt, Rinehart, and Winston. Perlmutter, David. 1971. Deep and Surface Constraints in Generative Grammar. New York: Holt, Rinehart, and Winston. Pesetsky, David. 1995. Zero Syntax. Cambrigde, Mass.: MIT Press. Pesetsky, David. 1998. Some optimality principles of sentence pronounciation. In Is the Best Good Enough?, ed. by Pilar Barbosa, Danny Fox, Paul Hagstrom, Martha McGinnis, and David Pesetsky, 337–383. Cambridge, Mass.: MIT Press. Pesetsky, David. 2000. Phrasal Movement and Its Kin. Cambridge, Mass.: MIT Press. Pesetsky, David, and Esther Torrego. 2001. T-to-C movement: Causes and consequences. In Ken Hale: A Life in Language, ed. by Michael Kenstowicz, 355–426. Cambridge, Mass.: MIT Press. Pesetsky, David, and Esther Torrego. 2004. Tense, case, and the nature of syntactic categories. In The Syntax of Time, ed. by Jacqueline Gúeron and Jacqueline Lecarme, 495–537. Cambridge, Mass.: MIT Press.

Bibliography

255

Pesetsky, David, and Esther Torrego. 2007. Probes, goals and syntactic categories. Ms., MIT, Cambridge, Mass., and University of Massachusetts, Amherst. Pica, Pierre, and William Snyder. 1994. Weak crossover, scope, and agreement. In Proceedings of WCCFL 13, 334–349. Stanford, Calif.: CSLI. Pollock, Jean-Yves. 1989. Verb movement, universal grammar, and the structure of IP. Linguistic Inquiry 20:365–425. Postal, Paul. 1974. On Raising. Cambridge, Mass.: MIT Press. Pylkkänen, Lina. 2002. Introducing arguments. Doctoral dissertation, MIT, Cambridge, Mass. Rackowski, Andrea, and Norvin Richards. 2005. Phase edge and extraction: A Tagalog case study. Linguistic Inquiry 36:565–599. Raposo, Eduardo, and Juan Uriagereka. 1990. Long distance case assignment. Linguistic Inquiry 21:505.537. Raposo, Eduardo, and Juan Uriagereka. 1996. Indefinite ‘se’. Natural Language and Linguistic Theory 14:749–810. Reinhart, Tanya. 1976. The syntactic domain of anaphora. Doctoral dissertation, MIT, Cambridge, Mass. Reinhart, Tanya. 1983. Anaphora and Semantic Interpretation. Chicago: Chicago University Press. Richards, Norvin. 1997. What moves where when in which language. Doctoral dissertation, MIT, Cambridge, Mass. Richards, Norvin. 2001. Movement in Language. Oxford: Oxford University Press. Riemsdijk, Henk van. 1989. Movement and regeneration. In Dialect Variation and the Theory of Grammar, ed. by Paola Benincà, 105–136. Dordrecht: Foris. Rizzi, Luigi. 1976. Ristrutturazione. Rivista di Grammatica Generativa 1:1–54. Rizzi, Luigi. 1982. Issues in Italian Syntax. Dordrecht: Foris. Rizzi, Luigi. 1986. On chain formation. In Syntax and Semantics 19: The Syntax of Pronominal Clitics, ed. by Hagit Borer, 65–95. Orlando: Academic Press. Rizzi, Luigi. 1990. Relativized Minimality. Cambridge, Mass.: MIT Press. Rizzi, Luigi. 1997. The fine structure of the left periphery. In Elements of Grammar, ed. by Liliane Haegeman, 281–337. Dordrecht: Kluwer. Rizzi, Luigi. 2001a. Relativized minimality effects. In The Handbook of Contemporary Syntactic Theory, ed. by Mark Baltin and Chris Collins, 89–110. Malden: Blackwell. Rizzi, Luigi. 2001b. Reconstruction, weak island sensitivity, and agreement. In Semantic Interfaces, ed. by Carlo Cecchetto, Gennaro Chierchia, and Maria Teresa Guasti, 145–176. Stanford, Calif.: CSLI. Rizzi, Luigi. 2006. On the form of chains: Criterial positions and ECP effects. In Wh Movement: Moving On, ed. by Lisa Lai-Shen Cheng and Norbert Corver, 97–133. Cambridge, Mass.: MIT Press. Rizzi, Luigi, and Ur Shlonsky. 2007. Strategies of subject extraction. In Interfaces + Recursion = Langauge? Chomsky’s Minimalism and the View from SyntaxSemantics, ed. by Hans M. Gärtner and Uli Sauerland, 115–160. Berlin: Mouton de Gruyter. Rodrigues, Cilene. 2004. Impoverished morphology and A-movement out of case domains. Doctoral dissertation, University of Maryland, College Park. Romero, Maribel. 1998. The correlation between scope reconstruction and connectivity. In Proceedings of WCCFL 16, 351–366. Stanford, Calif.: CSLI. Rooryck, Johan. 1997. Passive as partitive quantification. In Projections and Interface Conditions, ed. by Anna Maria Di Sciullo, 201–237. Oxford: Oxford University Press. Rosenbaum, Peter Steven. 1967. The Grammar of English Predicate Complement Constructions. Cambridge, Mass.: MIT Press.

256

Bibliography

Ross, John Robert. 1967. Constraints on variables in syntax. Doctoral dissertation, MIT, Cambridge, Mass. Runner, Jeffrey. 1995. Noun phrase licensing and interpretation. Doctoral dissertation, University of Massachusetts, Amherst. Safir, Ken. n.d. Semantic selection and complementation. Ms., Rutgers University. Safir, Ken. 1999. Vehicle change and reconstruction in A-bar chains. Linguistic Inquiry 30:587–620. Saito, Mamoru. 2003. A derivational approach to the interpretation of scrambling chains. Lingua 113:481–518. Sauerland, Uli. 1998. The meaning of chains. Doctoral dissertation, MIT, Cambridge, Mass. Sauerland, Uli. 1999a. Scope reconstruction without reconstruction. In Proceedings of WCCFL 17, 582–596. Stanford, Calif.: CLSI. Sauerland, Uli. 1999b. Relativized minimality effects with quantifier raising. Ms., Tübingen University. Schütze, Carson. 1997. INFL in child and adult language: agreement, Case and licensing. Doctoral dissertation, MIT, Cambridge, Mass. Schütze, Carson. 1999. English expletive constructions are not infected. Linguistic Inquiry 30:467–484. Schütze, Carson. 2001. On the nature of default case. Syntax 4:205–238. Sharvit, Yael. 1997. The syntax and semantics of functional relative clauses. Doctoral dissertation, Rutgers University. Shibatani, Masayoshi. 1977. Grammatical relations and surface cases. Language 53:780–809. Sigurðsson, Halldór Ármann. 1991. Icelandic Case-marked PRO and the licensing of lexical arguments. Natural Language and Linguistic Theory 9:327–364. Sigurðsson, Halldór Ármann. 1992. The case of Quirky subjects. Working Papers in Scandinavian Syntax 49:1–26. Sigurðsson, Halldór Ármann. 1996. Icelandic finite verb agreement. Working Papers in Scandinavian Syntax 57:1–46. Sigurðsson, Halldór Ármann. 2000. The locus of Case and agreement. Working Papers in Scandinavian Syntax 65:65–108. Sigurðsson, Halldór Ármann. 2002. To be an oblique subject: Russian vs. Icelandic. Natural Language and Linguistic Theory 20:691–724. Sigurðsson, Halldór Ármann. 2003. Case: abstract vs. morphological. In New Perspectives on Case Theory, ed. Ellen Brandner and Heike Zinsmeister, 223–268. Stanford, Calif.: CSLI. Sigurðsson, Halldór Ármann. 2004a. Agree and agreement: Evidence from Germanic. In Focus on Germanic Typology, ed. by Werner Abraham, 61–103. Berlin: Akademie Verlag. Sigurðsson, Halldór Ármann. 2004b. Icelandic non-nominative subjects: Facts and implications. In Non-Nominative Subjects (Vol. 2), ed. by Peri Bhaskararao and Karumuri Venkata Subbarao, 137–159. Amsterdam: John Benjamins. Sigurðsson, Halldór Ármann, and Anders Holmberg. In press. Icelandic dative intervention: Person and number as separate probes. In Agreement Restrictions, ed. by Roberta D’Alessandro, Susan Fischer and Gunnar Hrafn Hrafnbjargarson. Berlin: Mouton de Gruyter. Smith, Henry. 1994. “Dative sickness” in Germanic. Natural Language and Linguistic Theory 12:675–736. Smith, Henry. 1996. Restrictiveness in Case Theory. Cambridge: Cambridge University Press. Sobin, Nicholas. 1977. Agreement, default rules, and grammatical viruses. Linguistic Inquiry 28:318–343.

Bibliography

257

Soltan, Usama. 2007. On formal feature licensing in minimalism. Doctoral dissertation, University of Maryland, College Park. Sportiche, Dominique. 1988. A theory of floating quantifiers and its corollaries for constituent structure. Linguistic Inquiry 19:425–449. Sportiche, Dominique. 1990. Movement, agreement, and case. Ms., UCLA. [Published in Sportiche 1998, 88–243.] Sportiche, Dominique. 1996. Clitic constructions. In Phrase Structure and the Lexicon, ed. by Johan Rooryck and Laurie Zaring, 213–276. Dordrecht: Kluwer. [Reprinted in Sportiche 1998, 244–307.] Sportiche, Dominique. 1998. Atoms and Particles of Phrase Structure. London: Routledge Starke, Michael. 2001. Move dissolves into Merge: A theory of locality. Doctoral dissertation, University of Geneva. Stjepanović, Sandra. 1997. Is inherent case structural? Proceedings of FASL. The Indiana Meeting 1996, 295–311. Ann Arbor, Mich.: Michigan Slavic Publications. Stjepanović, Sandra. 1998. Short-distance movement of wh-phrases in Serbo-Croatian matrix clauses. Paper presented at the Comparative Slavic Morphosyntax workshop, Indiana University, Spencer. Stjepanović, Sandra. 1999. What do second position cliticization, scrambling, and multiple wh-fronting have in common? Doctoral dissertation, University of Connecticut, Storrs. Subbarao, Karumuri Venkata, and Sadaf Munshi. 2000. Agreement in Kashmiri and Minimalist inquiries: The framework. Ms., University of Delhi. Svenonius, Peter. 2002. Icelandic Case and the structure of events. Journal of Comparative Germanic Linguistics 5:97–225. Szabolcsi, Anna, and Marcel Den Dikken. 1999. Islands. Ms., New York University and CUNY. Szabolcsi, Anna, and Frans Zwarts. 1993. Weak islands and algebraic semantics for scope taking. Natural Language Semantics 1:235–284. Takahashi, Daiko. 1994. Minimality of movement. Doctoral dissertation, University of Connecticut, Storrs. Takano, Yuji. 1997. Linear order and parametric variation in syntax. Ms., Keio University. Takano, Yuji. 1998. Object shift and scrambling. Natural Language and Linguistic Theory 16:817–889. Taraldsen, Knut Tarald. 1995. On agreement and nominative objects in Icelandic. In Studies in Comparative Germanic Syntax, ed. by Hubert Haider, Susan Olsen, and Sten Vikner, 307–327. Dordrecht: Kluwer. Taraldsen, Knut Tarald. 2002. Case and agreement in Germanic and Romance. Ms., University of Tromsø. Thráinsson, Höskuldur. 1979. On Complementation in Icelandic. New York: Garland. Toribio, Almeida. 1990. Specifier-head agreement in Japanese. In Proceedings of WCCFL 9, 535–548. Stanford, Calif.: CSLI. Torrego, Esther. 1984. On inversion in Spanish and some of its effects. Linguistic Inquiry 15:103–129. Torrego, Esther. 1996. Experiencers and raising verbs. In Current Issues in Comparative grammar, ed. by Robert Freidin, 101–120. Dordrecht: Kluwer. Torrego, Esther. 1998. The Dependencies of Objects. Cambridge, Mass.: MIT Press. Torrego, Esther. 2002. Arguments for a derivational approach to syntactic relations based on clitics. In Derivation and Explanation in the Minimalist Program, ed. by Samuel D. Epstein and T. Daniel Seely, 249–268. Oxford: Blackwell. Tsai, W. T. Dylan. 1994. On economizing the theory of A-bar dependencies. Doctoral dissertation, MIT, Cambridge, Mass.

258

Bibliography

Tsai, W. T. Dylan. 1998. Subject specificity, copy theory, and the extended mapping hypothesis. In MIT Working Papers in Linguistics 25: The Interpretive Tract, 1–28. Department of Linguistics and Philosophy, MIT, Cambridge, Mass. Ueyama, Ayumi. 1998. Two types of dependency. Doctoral dissertation, University of Southern California. Ura, Hiroyuki. 1996. Multiple feature checking: A theory of grammatical function splitting. Doctoral dissertation, MIT, Cambridge, Mass. Ura, Hiroyuki. 1999. Dative subjects in Japanese and Korean. Journal of East Asian Linguistics 7:223–254. Ura, Hiroyuki. 2000. Checking Theory and Grammatical Functions in Generative Grammar. Oxford: Oxford University Press. Ura, Hiroyuki. 2001. Raising in Japanese and its theoretical implications in checking theory. Ms., Osaka University. Uriagereka, Juan. 1988. On government. Doctoral dissertation, University of Connecticut, Storrs. Uriagereka, Juan. 1995a. Aspects of the syntax of clitic placement in Western Romance. Linguistic Inquiry 26:79–123. Uriagereka, Juan. 1995b. An F-position in Western Romance. In Discourse-Configurational Languages, ed. by Katalin É. Kiss, 153–175. Oxford: Oxford University Press. Uriagereka, Juan. 1998. Rhyme and Reason. Cambridge, Mass.: MIT Press. Uriagereka, Juan. 1999a. Minimal restrictions on Basque movements. Natural Language and Linguistic Theory 17:403–444. Uriagereka, Juan. 1999b. Multiple Spell-Out. In Working Minimalism, ed. by Samuel D. Epstein and Norbert Hornstein, 251–282. Cambridge, Mass.: MIT Press. Uriagereka, Juan. 2000. T. Ms., University of Maryland, College Park. Ueyama, Ayumi. 1998. Two types of dependency. Doctoral dissertation, University of Southern California. Vukić, Sasa. 1999. Attract F and the minimal link condition. Linguistic Analysis 28:185–226. Watanabe, Akira. 2000a. Absorption: Interpretability and feature strength. Ms., Tokyo University. Watanabe, Akira 2000b. Feature copying and binding: evidence from complementizer agreement and switch reference. Syntax 3:159–181. Williams, Edwin. 1984. There-insertion. Linguistic Inquiry 15:131–152. Woolford, Ellen. 1999. More on the anaphor-agreement effect. Linguistic Inquiry 30:257–287. Wurmbrand, Susi. 1998. Infinitives. Doctoral dissertation, MIT, Cambridge, Mass. Wurmbrand, Susi. 2001a. Infinitives: Restructuring and Clause Structure. Berlin: Mouton de Gruyter. Wurmbrand, Susi. 2001b. Move or Agree? Presented at WCCFL 20, February, University of Southern California. Yatsushiro, Kazuko. 1999. Case licensing and VP structure. Doctoral dissertation, University of Connecticut, Storrs. Yip, Moira, Joan Maling, and Ray Jackendoff. 1987. Case in tiers. Language 63:217–250. Zaenen, Annie, Joan Maling, and Höskuldur Thráinsson. 1985. Case and grammatical functions: The Icelandic passive. Natural Language and Linguistic Theory 3:441–483. Zanuttini, Raffaela. 1997. Negation and Clausal Structure. Oxford: Oxford University Press. Zubizarreta, Maria-Luisa. 1982. On the relationship of the lexicon to syntax. Doctoral dissertation, MIT, Cambridge, Mass.

Index

A adjacency 27, 72, 214, 221 adjunct condition 54, 55, 181 Agree 10–12, 16, 42, 47, 52–55, 59, 62, 65–67, 69–71, 73–77, 79, 82–90, 95–96, 98, 101–102, 105, 107–108, 110, 113, 119–122, 124–127, 129–130, 150, 152–159, 161–162, 168, 170, 172–174, 178–182, 197, 207, 213, 216n, 221–222n, 227n, 230n, 236n, 238n Match 52–57, 62, 83–84, 104, 108–110, 157, 159, 162, 173, 180–182 Multiple 11, 74–75, 104, 108, 110, 154–155, 221–222n Valuation 10, 52, 54–55, 83–84, 95–96, 102, 108–110, 121, 180–182, 227n agreement dative 30–32, 36–37, 39, 41, 45, 142, 149 long distance (LDA) 11–12, 65–76, 135, 156, 173, 221–222n, 238n participle 6–7, 13, 48, 50, 65, 67, 75–76, 165, 221–222n, 232n rich 13 Alexiadou, A. on transitivity restriction 159 Anagnostopoulou, E. on Person-Case Constraint 44–45 on transitivity restriction 159 Arabic 54, 218–219n Atkinson, M. on Defective Intervention 232–233n Attract 90, 105–106, 164, 166, 216n, 227n

Closest 47, 90–91, 130, 134, 138, 148, 161, 208 F 25, 27, 34, 41, 216n attraction 34–35, 37–38, 72, 128, 132–133, 138, 143, 145, 147, 154, 156, 197, 216n, 240n Ausín, A. on raising verbs 231n B Baker, M. C. on Mirror Principle 5, 98 on Uniformity of Theta Assignment Hypothesis 102 Baltin, M. on neg-raising 240n on reanalysis 145, 230n Barbiers, S. on root/epistemic distinction 117 Basque 15, 44–45, 50–52, 226n Bejar, S. on Person-Case Constraint 44–45 Benveniste, E. on 3rd person 13, 165, 233n Bhatt, R. on Long Distance Agreement 69, 74, 222n on radical reconstruction 237n Bleam, T. on dative clitics 170 Bobaljik, J. D. on object honorification 95–103 Boeckx, C. on adjunct condition 55, 181 on Agree 12, 16, 53, 55, 156, 173, 216n on clitics 38, 88 on derivations vs. representations 148

260

Index

on Defective Intervention 54, 71, 101, 107, 152, 162 on EPP 12, 53, 57, 60, 155–156, 173, 221n on extraposition 126, 128 on freezing 44 on honorification 223n on islands 14–16, 55, 181 on Move 16, 53, 110, 156 on neg-raising 240n on Person-Case Constraint 141 on raising 38, 138, 210, 218n on Spec-Head 66–67 on unifying structural and inherent Case Bonet, E. on Person-Case Constraint 21, 30–32, 41, 44–45, 88, 140, 234n Bošković, Ž. on Attract 90, 105–106 on expletives 115 on inverse Case Filter 57, 89 on late insertion 203–204 on multiple wh-fronting 105–106 on scrambling 72, 85–86, 181, 227 Burzio’s Generalization 48–49 C c-command 4, 9–10, 37–38, 43, 53, 55, 58, 61, 72, 74, 85–86, 88, 93, 97–100, 107, 109, 114, 121, 123, 125, 127, 131–134, 137–138, 143–144, 146, 157, 160–162, 168, 207–208, 219n, 226n, 229n, 234n, 236n Case default 57 freezing 198 inherent 22, 40–41, 83, 132–133, 138–139, 216n, 219n, 237n multiple 160 null 217–218n structural 2–3, 6, 40–41, 48, 57, 89, 107, 132–133, 139, 143, 154, 165, 202, 219n, 237n Catalan 140–141 chain feature 71, 122, 129, 147, 216n head 71, 122, 162, 166 checking Case 44, 133, 160, 185, 197–198, 210, 213, 227n, 237n

feature 9–10, 15–16, 48, 57, 66, 83, 96, 107–109, 113, 121, 151, 155, 157, 159, 162, 222n, 227n Chomsky, N. on agreement and Case 6, 12, 30–31, 66, 89, 108, 141, 198 on AgrP 5–6, 9, 239n on Agree 10, 52–53, 62, 66, 70–71, 76, 79, 82–83, 86, 89–90, 95–96, 107–108, 122, 168, 173, 179–180, 207, 230n on binding 37, 134, 230n on Burzio’s Generalization 48–49 on Case 2, 30–31, 46, 57, 132, 154, 165, 197, 219n, 237n on chain 52–53, 71, 122, 216n, 224n on Closest Attract 130 on covert movement 8, 117–119, 147, 159, 228n on defective intervention 46–47, 53, 70, 79, 89, 96, 152, 164–165, 168, 180, 218n, 229n, 234n on EPP 12, 32, 52–53, 66, 108, 156, 161 on experiencer 130, 132, 138, 143, 207 on expletives 8–10, 88, 120–121, 124, 139, 173–174, 205, 233n, 238n on freezing 44, 154 on Icelandic 46–47, 143, 165 on Last Resort 2 on lowering 189–190, 196, 200, 207 on Move 16, 59, 71, 115, 118, 120, 146, 156, 163, 173, 211 on Multiple Agree 74, 104 on multiple expletives 174 on phase 47, 145 on raising 52–53 on reconstruction 185–186, 188–190, 196, 209, 213 on Shortest Move 130 on Spec-Head 7, 66, 82, 90, 94 on timing 118 Cinque, G. on functional categories 13 on islands 123, 158, 176, 229n clitic 7, 13, 15, 30, 32, 38, 45, 65, 75, 79, 88, 101–102, 140, 151, 153–156, 160, 178, 218n, 221n, 223–224n climbing 73–74, 76

Index doubling 166, 168–170, 177, 233–234n combien-extraction 174–176 compositionality 235n concord see Case matching Condition C effect 131, 133–134, 136, 144, 160, 207–208, 225n connectivity 185, 199, 214, 235 control 22, 39–40, 116–117, 122, 192–195, 212, 215n, 218n as movement 117, 194, 227–228n, 237n copy deletion 186, 200, 206, 240n theory 122, 186, 198–199, 206, 208, 228n covert component 72, 120, 134–135, 150, 159, 203–204, 213 movement 117–119, 121, 136, 161, 228n, 234n Culicover, P. on Point-of-View 32 cycle 119, 128–129, 131, 135, 147, 161, 230n cyclic interpretation 235n D Defective Intervention 47, 53–54, 62, 70–72, 74, 78–79, 83–85, 87–90, 96, 101, 152, 162, 164–166, 168–181, 226n, 229n, 232–234n; see also intervention effects Depiante, M. on raising verbs 231n derivational syntax 86, 104, 107, 109–110, 113, 130–131, 133–134, 136–138, 143–144, 147–149, 156, 163, 165, 168, 179, 181–182, 206n, 208n, 213n, 230–231n, 235n derived predicate 200 Direct intervention 165–166, 173, 175–180 dislocation 75, 151, 56 Distributed Morphology 31, 98 Doubly-Filled Comp-Filter 45 E each-paradigm 236n economy 31, 105–106, 109–110, 141, 182 local 70, 110, 211–212, 214

261

English 5, 8, 27–29, 35–38, 41, 44, 59, 65, 130, 139, 142–144, 149–154, 156, 159–161, 172–174, 205–206, 209–210, 216n, 218–219n, 222n, 224n, 229n, 233n, 239n Epstein, S. D. on compositionality 137 on derivations 109, 131, 137, 163, 230n on experiencers 131, 136–137, 161 on interpretive procedures 138 equidistance 87, 145, 224n erasure 197 Erlbourne, P. on radical reconstruction 201–202, 237n Exceptional Case Marking (ECM) 3–4, 7, 22, 135, 186–187, 215–216n, 228n, 232n existential constructions 3, 8–11, 58, 65–66, 70, 72, 115, 120–121, 124–125, 152–153, 172, 174 experiencer 25, 27–28, 30, 36–39, 41, 49–50, 56–57, 59, 130–161, 172–173, 206–208, 213, 216n, 218–221n, 230–231n, 233–234n, 238n agreement see dative agreement expletive 8–9, 32, 36, 41, 61, 88, 115–116, 120–122, 139, 142, 153, 160, 173, 178, 185, 192, 199, 204, 213, 218n, 228–230n, 233n, 238n insertion 185, 238n covert 61 multiple 174 Extended Projection Principle (EPP) 10, 12, 32, 44, 52–53, 57, 60–61, 66, 71, 108, 110, 115, 150, 153–156, 158–161, 173, 201, 215n, 221n, 227n, 230–232n Extended Standard Theory (EST) 235n Extension Condition 128, 204 extraposition 126–129, 206 F feature interpretable 104, 108–110, movement 10, 27, 40, 65, 71–72, 76, 113, 115, 117, 119–122, 129, 135–136, 152–153, 197, 229–230n strong 147, 155, 203–204

262

Index

uninterpretable 66, 108–109, 197–198, 227n, 236n Fiengo, R. on Proper Binding Condition 114, 201, 238n French 5–6, 28–30, 38, 44, 65, 67, 75–76, 159, 174–177, 203, 218n, 232n Fujita, K. on prepositional objects 38 Full Interpretation 85, 155 G George, L. on agreement and nominative Case 30–33, 36, 41, 141–142, 197–198, 217n German 39–40, 55, 75, 113, 206, 209, 219n, 234n, 239n government 3–4, 10 Government and Binding (GB) 13, 21, 59, 118, 130, 136, 153, 190, 196, 238n Groat, E. on compositionality 137 on derivations 109, 131, 137, 163, 230n on experiencers 131, 136–137, 161 on interpretive procedures 138 on multiple expletives 174 H Hale, K. on Person-Case Constraint 44 on theta-roles 190 Halle, M. on Distributed Morphology 31, 98 Hasegawa, N. on honorification 222n Harada, S.-I. on object honorification 78–82, 88, 96–98, 106, 167 Harley, H. on nominative Case 215n on Case assignment 227n head-movement 114, 161, 210, 230n, 240n Heim, I. on movement 196, 200, 235n Hindi 11, 65, 67, 69–70, 72–76, 221–222n Hindi-Urdu see Hindi Hiraiwa, K. on Multiple Agree 74, 89, 104, 108

Holmberg, A. on Quirky elements 58–62, 220–221n on stylistic fronting 60 honorification 11 object 77–103, 106–107, 166–168, 177, 222–223n, 226n Person feature 177 Subject 77–78, 98, 103, 166, 223n Hornstein, N. on control 117, 194, 227–228n, 237n how likely paradigm 116–117, 121–122, 124, 126, 129, 228n Hróarsdóttir, Th. on Quirky elements 58–62, 220–221n I Icelandic 6, 13, 21–34, 38–62, 66, 70–71, 79, 83, 88, 140–144, 148–149, 155, 164–165, 168–170, 179–180, 205, 215–221n, 226n, 231n Inclusiveness Condition 190, 201 interpretive procedures 136, 143 intervention effect 47, 53–54, 58, 62, 71, 74, 79, 83–86, 90–91, 96, 101, 103, 107, 110, 123, 125, 152, 158, 169, 175, 181, 186, 206–208, 218n, 223n, 228–229n, 234n, 238n; see also Defective Intervention Inverse Case Filter 57, 90 islands 15, 176, 178 strong 55, 181, 234n weak 123, 125, 158, 229n Italian 15, 28–29, 38–39, 73, 150–153, 156, 159–160, 166, 171, 178, 218n, 221n, 231–232n J Jackendoff, R. on Quirky Case 49, 217n Japanese 11, 45, 77–82, 84–103, 106–107, 166–168, 170, 177, 179, 222–227n, 233n Jeong, Y. on dative clitics 170, 178 on Person-Case Constraint 102 Johnson, K. on object agreement 87 Joshi, A. on how likely paradigm 114, 116, 129, 192, 212

Index Julien, M. on agreement 13 K Kashimiri 222n Kayne, R. S. on participle agreement 6, 65, 67, 76, 222n, 232n on Person-Case Constraint 31, 44, 88, 140, 224n, 234n on raised predicates 126, 187, 228n Kawashima, R. on compositionality 137 on derivations 109, 131, 137, 163, 230n on experiencers 131, 136–137, 161 on interpretive procedures 138 Keyser, S. L. on theta-roles 190 Kitahara, H. on compositionality 137 on derivations 109, 131, 137, 163, 230n on experiencers 37–38, 100, 131–138, 143, 161, 207–208, 219n, 238n on interpretive procedures 138 Koizumi, M. on object agreement 87, 225n Kornfilt, J. on agreement and nominative Case 30–33, 36, 41, 141–142, 197–198, 217n Kratzer, A., on movement 200, 235n on root/epistemic distinction 117 Kroch, A. on how likely paradigm 114, 116, 129, 192, 212 L Laka, I. on Point-of-View 32 Larson, R. on Light-Predicate Raising 126 Lasnik, H. on binding 131, 208 on ditransitives 87 on ECM constructions 7 on feature movement 10, 40, 71, 115, 121–122, 129, 135, 230n on how likely paradigm 114–117, 121–122, 129

263

on Inverse Case Filter 57, 89 on late insertion 203–204 on Move-F 121, 228–229n on object agreement 87, 146 on partitive Case 36, 142, 238n on reconstruction 132, 185–196, 203, 209, 212–214, 236n Last Resort 2, 9, 16, 146, 188, 214 late insertion 115, 203, 227n, 234n light verb 97–98, 226n Logical Form (LF) 8, 34, 37–38, 85–86, 114–115, 118–122, 124, 126, 129, 134–137, 147, 160, 190, 195, 197, 201, 203, 205–206, 208, 211, 214, 219n, 228–230n, 238n look-ahead 70, 75, 147 lowering 72, 85–86, 185–186, 189, 191, 196, 199–201, 203–204, 206–208, 211–214, 235n, 238n, 240n M Mahajan, A. on Long Distance Agreement 69 Maling, J. on Quirky Case 21–22, 39, 49, 140, 164, 215n, 217n Marantz, A. on Distributed Morphology 31, 98 on Case assignment 227n Martin, R. on raising-control ambiguity 228n May, R. on Quantifier Lowering 189, 191 me-lui constraint see Person-Case Constraint Merge 16, 49, 86–88, 91, 107, 109, 115, 124, 132, 136–137, 143–144, 154, 156, 178, 199, 211, 213, 218n remerge 16, 137, 150, 156, 173, 236n Merge-over-Move 115 Minimalist Program 2, 21, 27, 42, 60, 105, 118, 130–131, 145, 148, 163, 186, 202 modal predicates 116–117 Moro, A. on small clauses 218n Move-F 71–72, 119–121, 126, 221n, 228n Müller, G. on remnant movement 114, 129

264

Index

N negation 6, 13, 186, 191, 203, 207, 210–211, 213, 239–240n partial 186, 204, 209, 212–213, 239n Niinuma, F. on honorification 101, 223n Nomura, M. on extraposition 127–129 on Multiple Agree 74, 108 null resumptive clitic 75 Nunes, J. on pronunciation of copies 119 O Object Shift 6, 57, 61, 154, 215n of-phrase 232n, 237n optionality 48, 74, 188, 222n Ormazabal, J. on Person-Case Constraint 44, 51–52, 141, 217n P parameter 22, 38, 150 pro-drop 13, 151, 153, 232n Perlmutter, D. on Person-Case Constraint 31, 44, 88, 140, 234n Person-Case Constraint (PCC) 21, 30–38, 40–41, 44–48, 50–52, 58, 88, 101–102, 140–143, 149, 217–220n, 231n, 234n Pesetsky, D. on Case 12, 197–198, 217n, 227n on misplaced features 198 on uninterpretable features 12, 197–198, 217n, 227n phase 47, 90, 145, 182, 204, 233n, 235n phi(K)-features 3, 10, 12–15, 27, 32, 34, 57, 82–83, 88–90, 98, 102, 107–109, 120–121, 154–157, 164–165, 168, 173–176, 178, 180–181, 197–198, 222–224n, 227n, 233–234n, 236n Phonetic Form (PF) 31, 34, 57, 71–72, 85, 118–121, 155, 197–198, 201–202, 204–206, 21–218n, 230n, 237n pied-piping 25, 118–119, 125, 129, 146–147, 156, 216n, 229n, 231n, 238n Point-of-View 32–33, 36, 217–218

Pollock, J.-Y. on Split-Infl hypothesis 1, 4–5 Postal, P. on rasing 7, 186–188 on reanalysis 145, 230n Principles-and Parameters 1, 3, 12, 21 Principle of Inclusiveness, see Inclusiveness Condition pro 15, 82, 91, 153, 225n, 232–233n PRO see control Probe/Goal Projection Principle 190, 230n, 235n Proper Binding Condition 113–115, 201, 204, 237n pseudogapping 186 Q Quantifier floating 46, 154 lowering 189, 191, 206, 211, 213, 238n, 240n Quirky Case 21–23, 25, 27, 29, 32, 34, 38–39, 44, 48–50, 216–218n, 220n, 231n Quirky subject 11, 21–24, 26, 30–32, 34–35, 39, 41–45, 47–48, 52–53, 55, 58, 62, 79, 83, 141, 164, 215–216n, 220n R raising 7–10, 12, 36–37, 60, 62, 66, 72, 90, 119–120, 122, 145–146, 158, 170–171, 178, 194–195, 197–198, 228–229n, 234n, 237n negation 210–212, 239–240n objects 28, 186–190 across an experiencer 28, 32–33, 38, 59, 100, 134, 150–154, 156, 159, 172–173, 206–207, 231n, 238n particle verbs 187–188 predicates 22, 25, 27, 34–35, 39–40, 115–117, 126–127, 192, 195, 212, 216n, 218n quirky subject 52 subject 131–133, 135–138, 143–144, 147–148, 208, 218n, 231n verb 56, 130, 160–161, 192, 221n, 232n Raposo, E. on partial agreement 217n

Index reanalysis 38–40, 73, 133, 138, 145–148, 230n reconstruction 14, 72, 114, 119, 132, 185–214, 228n, 230n, 234–240n indefinites 200 intermediate 203 partial 199 radical 199–204, 237n Relativized Minimality 27, 41, 70, 84, 86, 96, 113, 122–123, 125, 129–130, 153, 157, 162–163, 178, 185, 206–207, 211, 213, 216n, 228–231n, 239n remnant movement 113–115, 117, 122, 124, 126–127, 129, 221n, 227–228n representational syntax 119, 136–137, 148, 163, 165, 181, 208, 211, 230n Rezac, M. on Person-Case Constraint 44–45 Richards, N. on chains 155 on strong features 15 Rizzi, L. on clitic climbing 73 on combien-extraction 174–175 on islands 15, 123, 158 on pro 13, 15 on Relativized Minimality 122, 157, 162–164, 210, 230n on wh-extraction 171, 176, 229n Romance 28–29, 73, 75, 79, 88, 101, 140, 165, 217–218n, 220n, 224n, 233n Romero, J. on Person-Case Constraint 44, 141, 217n Rosenbaum, P. S. on raising adjectives 128 S Saito, M. on ECM constructions 7 on how likely paradigm 116–117, 192, 212 on remnant movement 227–228n Sauerland, U. on radical reconstruction 201–202, 237n Schütze, C. on default case 57

265

on raising predicates 27 scope 100, 120, 185–188, 192–193, 195, 197, 199, 202–203, 207, 213, 217n, 229n, 235–236n, 238n freezing 198, 213 narrow 9–10, 121, 204, 211, 239n reversal 114, 209–210, 212, 239–240n wide 125, 192 scrambling 71–72, 85–87, 93–95, 170, 224n, 227n, 233n secondary predicate 46, 92 seem see raising predicate selection 50, 74, 76, 98, 219n, 222n Serbo-Croatian 105, 110, 227n Shibatani, M. on honorification 77, 82, 96, 166 Shortest Move 130 Sigurðsson, H. Á. on agreement 13, 24, 27–28, 31–32, 50 on Quirky Subjects 22 on PRO 215n on raising predicates 27, 34, 216n Single Output syntax 119, 135–136, 159 subjects 28, 30, 34 Spanish 28, 38, 41, 132, 144, 148, 166, 168–170, 177, 216n, 231–233n Specifier-Head (Spec-Head) agreement 6–12, 38, 59, 65–67, 69, 71–73, 76–77, 79, 82, 90–94, 96, 139, 226n, 240, Spell-Out 39, 43, 58, 118–119, 122, 136, 164, 186, 201, 203, 228n, 237n Multiple 136, 235n Sportiche, D. on dative clitics 170, 178 Starke, M. on Relativized Minimality 123–124, 157, 163, 165 on weak islands 125, 158, 229n Stylistic Fronting 60 subject 1, 3, 5–6, 9, 11, 13, 15–16, 21, 67–71, 73, 89, 92–93, 117, 120, 130, 150–153, 159–160, 171–172, 186, 188, 193, 195, 203, 207, 209–213, 215n, 217–218n, 221n, 225n, 230–232n, 235n, 239–240n;

266

Index

see also Quirky subjects; subject honorification; subject raising; Transitive Expletive Construction successive cyclicity 70 superraising 166, 170–171, 178, 210, 232n T Takahashi, D. on scrambling 72, 85–86, 181, 227 Taraldsen, K. T. on agreement 13, 24 on Quirky Subjects 216n theta role 48–50, 53, 85, 117, 156, 170–171, 219n, 237n configurational 190 criterion 117, 194–195, 197 as features 190, 240n Thráinsson, H. on Quirky Case 21–22, 39, 140, 164, 215n Toribio, A. on honorification 77, 82, 96, 166 Torrego, E. on clitic doubling 168–169 on Case 12, 197–198, 217n, 227n on misplaced features 198 on uninterpretable features 12, 197–198, 217n, 227n tough construction 195 trace see copy Transitive Expletive Construction (TEC) 205–206 trapping effect 193, 202, 206

U Ura, H. on honorification 77, 166 Uriagereka, J. on Defective Intervention 165 on Multiple Spell-Out 136, 235n on partial agreement 217n on Point-of-View 32, 217n V Visibility Condition 197 Vukić, S. on expletives 115 W weak NP 191, 199, 212 wh-movement 58, 75, 109, 125, 127–129, 132, 151, 153, 175, 203, 226–227n, 234n Wurmbrand, S. on infinitivial complements 73–74, 76 on root/epistemic distinction 117 Y Y-model 118, 122, 135, 160 Yatsushiro, K. on object honorification 95–103 Yip, M. on Quirky Case 49, 217n Z Zaenen, A. on Quirky Case 21–22, 39, 140, 164, 215n