Derivations and Evaluations: Object Shift in the Germanic Languages 9783110207200, 9783110198645

Citation preview

Derivations and Evaluations

≥

Studies in Generative Grammar 97

Editors

Henk van Riemsdijk Jan Koster Harry van der Hulst

Mouton de Gruyter Berlin · New York

Derivations and Evaluations Object Shift in the Germanic Languages

by

Hans Broekhuis

Mouton de Gruyter Berlin · New York

Mouton de Gruyter (formerly Mouton, The Hague) is a Division of Walter de Gruyter GmbH & Co. KG, Berlin.

The series Studies in Generative Grammar was formerly published by Foris Publications Holland.

앝 Printed on acid-free paper which falls within the guidelines 앪 of the ANSI to ensure permanence and durability.

Library of Congress Cataloging-in-Publication Data Broekhuis, Hans. Derivations and evaluations : object shift in the Germanic languages / by Hans Broekhuis. p. cm. ⫺ (Studies in generative grammar ; 97) Includes bibliographical references and index. ISBN 978-3-11-019864-5 (cloth : alk. paper) 1. Germanic languages ⫺ Syntax. 2. Germanic languages ⫺ Grammar, Generative. 3. Germanic languages ⫺ Direct object. 4. Germanic languages ⫺ Indirect object. 5. Minimalist theory (Linguistics) 6. Optimality theory (Linguistics) I. Title. PD369.G76 2008 4301.045⫺dc22 2007051988

Bibliographic information published by the Deutsche Nationalbibliothek The Deutsche Nationalbibliothek lists this publication in the Deutsche Nationalbibliografie; detailed bibliographic data are available in the Internet at http://dnb.d-nb.de.

ISBN 978-3-11-019864-5 ISSN 0167-4331 쑔 Copyright 2008 by Walter de Gruyter GmbH & Co. KG, D-10785 Berlin. All rights reserved, including those of translation into foreign languages. No part of this book may be reproduced in any form or by any means, electronic or mechanical, including photocopy, recording, or any information storage and retrieval system, without permission in writing from the publisher. Cover design: Christopher Schneider, Berlin. Printed in Germany.

Preface and acknowledgements

The title of this book, Derivations and Evalutions, aims at expressing the central idea underlying this study, namely that developing an explanatorily and descriptively adequate theory of syntax requires that restrictions be formulated on both the syntactic derivations and the resulting syntactic representations. This will be obtained by combining certain aspects of the minimalist program (MP) and certain aspects of optimality theory (OT); more specifically, I will assume that representations created by some version of the computational system of human language CHL from MP are evaluated in an optimality-theoretic fashion. The main reason for seriously investigating this idea and for being optimistic about its explanatory and descriptive adequacy lies in the insight that whereas MP has been especially successful in formulating a restrictive theory of core grammar, OT has been very successful in describing the more peripheral, languagespecific properties of languages as well as the variations between languages. Although the idea of combining these aspects of MP and OT in the socalled derivation-and-evaluation (D&E) model seems highly controversial, Chapter 1 will extensively argue that it is actually not: MP, OT, and D&E are all specific instantiations of the more general model in Figure 1. Input

Generator

Output representations

Evaluator

Optimal output

Figure 1: The architecture of grammar

The subtitle Object shift in the Germanic languages refers to the main object of this study, and contends that, contrary to what is normally assumed, object shift is not a phenomenon restricted to the Scandinavian languages, but should be identified with one of the two types of scrambling that can be found in Dutch and German. Chapter 2 through Chapter 4 will provide a detailed analysis of the relevant data, which is based on (i) the minimalist assumption that movement is subject to the Last Resort Condition, and (ii) a limited set of violable constraints that can be traced back to the conditions and principles that were or still are widely accepted in the principles-and-parameters framework; see the appendix at the end of the book for a review of the constraints used in this study. I will set out to argue –following earlier proposals– that object shift is A-movement, and

vi Preface and acknowledgements

that we must actually distinguish between two types of object shift. The first type will be called short object shift and is triggered by the ij-features on the verbal root V: it will be argued that we find this type of movement in all Germanic languages (including English). The second type, which is the type normally discussed in the literature on object shift, will be called regular object shift, and is triggered by the case features on the light verb v. Short object shift and regular object shift will be the main topics of Chapter 2 and Chapter 3, respectively. The latter chapter will analyze the restrictions that are normally attributed to object shift, such as Holmberg’s Generalization, which prohibits regular object shift across the main verb, the indirect object, or a verbal particle. The analysis given in Chapter 3 will raise a number of new questions concerning the interaction of object shift with verb movement, A-movement of the subject, and A′-movement. These will be discussed in Chapter 4. Chapter 5, finally, aims at describing various types of predicate movement by appealing to the same set of constraints that were used in the analysis of object shift, thus trying to maximally exploit these constraints in the grammar. The research reported in this book started in 1997 with my collaboration with Joost Dekkers, who was working at that time on a Ph.D. project on OT-syntax at the French department of the Faculty of Arts of the University of Amsterdam. Since there was no one on the regular staff who was really comfortable or even acquainted with this kind of work, and since I was working as a postdoc researcher at that time, I was asked to co-supervise Joost’s project. This was probably more advantageous to me than to Joost since this arrangement resulted in Joost giving me a crash course in OTsyntax. In this period Joost and I also developed what I referred to as the D&E model above. This model grew out of a number of heated debates, which focused on my concerns about the lack of an explicit theory about the generator GEN and the unrestricted nature of the universal constraint set CON, and Joost’s enthusiasm about the new and exciting avenues of research opened by the new OT framework. Our collaboration and discussions ultimately led to a joint paper on relative constructions (Broekhuis & Dekkers 2000), which was perhaps not very original in that it simply applied Pesetsky’s (1997;1998) OT-analysis of relative clauses to a set of new data from Dutch (dialects), but which was highly important to us because it enabled us to clearly formulate certain programmatic statements about the relation between OT and the minimalist framework. Notwithstanding the fact that we both committed ourselves to these programmatic statements in later work, it was clear that the routes we took from there diverged: Joost Dekkers’ impressive, award-winning Ph.D. thesis from

Preface and acknowledgements vii

1999, in which he rephrased in OT terms some of the filters from the preGB era and the GB principles and conditions that grew out of them, is definitely more OT-oriented than my paper on object shift (Broekhuis 2000), which ultimately led to the present book and which I believe fits more in the minimalist tradition. I still regret that Joost’s line of research ended when he decided in 1999 to leave the academic world and make his career elsewhere. That there is a gap of ten years between the writing my original paper on object shift in 1997-98 and the publication of this book has mainly to do with the fact that I was involved in another, more descriptively oriented, project from 1998 to 2003. It was only in 2003, when I received a so-called VIDI-grant from the Netherlands Organisation of Scientific Research (NWO) that I was able to take up the D&E project again. Without this grant, for which I am greatly indebted, I would never have been able to start writing this book. Much of what is included in this book has been presented orally at various places. I had the opportunity to teach this material at the LOT Winter School at Leiden University in 1998; at the Zentrum für Allgemeine Sprachwissenschaft, Typologie und Universalienforschung (ZAS) in Berlin and the University of Potsdam in 1999; at the University of Venice in 2001; and at the LOT Winter School at the Radboud University Nijmegen in 2007. In addition, I presented parts of this work at a Workshop on Scrambling held in Leiden in 2003, the Comparative Germanic Syntax Workshop 19 held in New York in 2004, the NWO/OTKA meeting held in Leiden in 2004, the workshop Descriptive and Explanatory Adequacy in Linguistics organized by Ralf Vogel and myself in 2005, the Comparative Germanic Syntax Workshop 22 held in Stuttgart in 2007, several Linguistics in the Netherlands meetings organized by the Linguistic Society of the Netherlands (AVT), and at many more informal meetings at all linguistic departments in the Netherlands (Amsterdam, Groningen, Leiden, Nijmegen, and Utrecht) and abroad (Eberhard Karls Universität Tübingen, Graduate Center of CUNY, Newcastle University, University of Connecticut, University of Massachusetts–Amherst, and Harvard University). It goes without saying that the input I received from the audiences of these meetings has been of considerable help in advancing the project. Parts of the present study have been published elsewhere. It also goes without saying that the comments that I got from the editors and the mostly anonymous reviewers have been of invaluable help and often have given rise to considerable changes in the details of the proposal, even if that did not immediately show in the published versions of the papers.

viii Preface and acknowledgements

– Chapter 1: an earlier and shorter version (without section 1.3.3) was published as Broekhuis (2006a); section 1.3.3 summarizes some results of Broekhuis & Klooster (2001; to appear). – Chapter 2: a shorter version of section 2.3 was published as Broekhuis (2006b). – Chapter 3: a shorter version of 3.1.2.4 was published as Broekhuis (2007b); section 3.2 contains material published earlier as part of Broekhuis (2000) and Broekhuis (1997b;1999), but is considerably simplified by incorporation of two order preservation constraints. – Chapter 4: section 4.2 contains material that was published as part of Broekhuis (2007a). – Chapter 5: sections 5.1 and 5.2 develop some ideas first presented in Broekhuis (2005); parts of sections 5.2-5.4 are contained in Broekhuis & HegedĦs (to appear); section 5.5 builds on some ideas first presented in Broekhuis & Migdalski (2003). An important turning point in the conduct of the research presented here was initiated by a long and intensive discussion with Susi Wurmbrand during the preparation of Broekhuis (2007a) in 2004/5. This ultimately resulted in my giving up all hope that Holmberg’s Generalization could be derived by some version of the Minimal Link Condition, and convinced me that appealing to the above-mentioned order preservation constraints (originally proposed by Müller 2000;2001, and based on work by Williams 2003) was the right way to go. Without me taking this step, this book might never have been completed. Special mention is due to Cedric Boeckx, Tonya Kim Dewey, Marcel den Dikken, Liliane Haegeman, and Ellen Woolford, who all read an earlier version of this book and contributed to the final version by providing helpful comments that enabled me to make substantial improvements in the final manuscript. Not all their comments have been incorporated in the present text, and no doubt I will be sorry for that at some point in the future. Anders Holmberg, and Halldór Sigurðsson occasionally helped me with the Scandinavian data. This book would certainly not have been what it is now without the assistance of Carole Boster (UConn), who meticulously copy-edited the prefinal manuscript. Her contribution to this book went much further than simply correcting my English; she also discussed several problematic data with me and discovered a number of serious errors that would have made the discussion unintelligible at places. I cannot thank her enough, and I am sure that all remaining errors in the final version of the book are due to my last minute revisions.

Preface and acknowledgements ix

This work is also much influenced by the members of the Comparison of Grammatical Models group in Tilburg. First and foremost I must mention Henk van Riemsdijk and Norbert Corver, who have always had a keen interest in the current work and have played a crucial role in obtaining the VIDI-grant that enabled me to further the D&E project, and also commented on substantial parts of this book. Other members of the group also contributed to my study by discussing the underlying assumptions with me, commenting on earlier versions of the material included, providing me with relevant information and data, and by simply being good colleagues and friends: Elena Anagnostopoulou, Jutta Hartmann, Veronika HegedĦs, Ben Hermans, Riny Huybregts, Evelien Keizer, Krzysztof Migdalski, Nataša Miliüeviü, and Craig Thiersch. Other people from the University of Tilburg that should be mentioned for supporting me in several other ways are Bertjan Busser and Leen Jacobs. Unfortunately the Comparison of Grammatical Models group has ceased to exist as of 2005 due to a cost-cutting effort by the Faculty of Arts. I much admire all members of the group for keeping up such good spirits during the time period when the group faded out of existence; for this reason I dedicate this book to them. I also want to thank those members of the faculty of arts that morally supported our group in that difficult period. Finally, I want to thank Norbert Corver, Theo van de Hoek, Jelle Kaldewaij, Ans van Kemenade, Jos Schaeken, Arie Verhagen, Henk van Riemsdijk and Fred Weerman for helping me to find a new position in a more fitting environment, as well as my new colleagues at the Leiden University Center for Linguistics for welcoming me there and making me feel comfortable again. December 25, 2007

Table of contents

Preface and acknowledgements ................................................................. v Chapter 1 Derivations (MP) and Evaluations (OT) ................................................... 1 1.1. Where MP and OT are similar: The architecture of syntax.......... 2 1.1.1. Principles & Parameters Theory........................................................4 1.1.2. Optimality Theory ...........................................................................16 1.1.3. Conclusion.......................................................................................23 1.2. Where MP and OT differ: derivations and evaluations .............. 23 1.2.1. Universal properties of language (the generator) ............................24 1.2.2. Variation (the evaluator)..................................................................27 1.2.3. Conclusion.......................................................................................30 1.3. The derivation-and-evaluation framework ................................. 31 1.3.1. The generator...................................................................................32 1.3.2. The evaluator ...................................................................................34 1.3.2.1. CHL constraints.................................................................................35 1.3.2.2. The interface (PF and LF) constraints .............................................41 1.3.3. Input and output of CHL ...................................................................48 1.3.3.1. Is the input a numeration? ...............................................................50 1.3.3.2. Is the output semantically restricted? ..............................................52 1.3.3.3. Conclusion.......................................................................................56 1.4. Summary .................................................................................... 56 Chapter 2 Short object shift ....................................................................................... 61 2.1. Object shift/scrambling is A-movement..................................... 61 2.1.1. Arguments supporting the hypothesis .............................................62 2.1.2. Arguments against the hypothesis ...................................................64 2.2. Two types of object shift ............................................................ 71 2.3. The universal base hypothesis: VO or OV? ............................... 73 2.3.1. The Linear Correspondence Axiom and the Branching Constraint.73 2.3.2. The trigger problem.........................................................................78 2.3.3. Complex verb constructions ............................................................80 2.3.3.1. Barbiers (2005)................................................................................81

xii

Table of contents

2.3.3.2. 2.3.3.2.1. 2.3.3.2.2. 2.3.3.3. 2.3.4. 2.3.5. 2.4. 2.4.1. 2.4.1.1. 2.4.1.2. 2.4.1.3. 2.4.2. 2.4.2.1. 2.4.2.2. 2.4.2.3. 2.4.2.4. 2.4.2.5. 2.5.

Haider (2003;2005) .........................................................................83 Haider’s proposal........................................................................ 83 Comparison to the VO-approach................................................ 87 Conclusion.......................................................................................90 Evidence in favor of “short” object shift in English........................90 Conclusion.......................................................................................96 Short object shift......................................................................... 96 Simple tense constructions ..............................................................97 The Germanic VO-languages: {EPP(ij), EPP(v)} >> *MOVE ............97 The Germanic OV-languages: EPP(ij) >> *MOVE>> EPP(v) ..........100 A puzzle: the placement of VP-adjuncts .......................................101 Perfect tense constructions ............................................................104 The Germanic VO-languages: the constraint EPP(Asp).................104 The Germanic OV-languages ........................................................105 The Germanic VO-languages again: the constraint H-COMPL .......108 The ranking of H-COMPL in the Germanic OV-languages .............109 Unifying EPP(v) and EPP(asp): *STRAY FEATURE ..........................111 Conclusion................................................................................ 117

Chapter 3 Regular object shift ................................................................................. 119 3.1. Holmberg’s Generalization ...................................................... 121 3.1.1. Deriving HG from case theory ......................................................122 3.1.2. Deriving Holmberg’s Generalization from Locality theory ..........123 3.1.2.1. The Minimal Link Condition.........................................................124 3.1.2.2. The MLC and object shift in embedded clauses in Dutch/German 129 3.1.2.3. The MLC does not block object shift in perfect tense constructions..132 3.1.2.4. Why the MLC cannot derive Holmberg’s Generalization.............135 3.1.2.4.1. Defective intervention effect .................................................... 136 3.1.2.4.2. Agreement in Icelandic quirky subject constructions............... 138 3.1.2.4.3. Intervening datives in Dutch: the proper definition of closeness 141 3.1.2.4.4. Timing of feature elimination................................................... 144 3.1.2.4.5. More Icelandic agreement facts................................................ 146 3.1.2.4.6. The person constraint on the nominative argument.................. 149 3.1.2.4.7. Defective intervention does not exist ....................................... 152 3.1.2.5. Conclusion.....................................................................................152 3.1.3. Deriving HG from factors external to CHL ....................................153 3.1.4. Conclusion.....................................................................................157 3.2. The restrictions on regular object shift ..................................... 157 3.2.1. Object shift of DPs and pronouns..................................................159 3.2.1.1. Type 1: languages without object shift..........................................160 3.2.1.2. Type 2: languages with full object shift ........................................161

Table of contents xiii 3.2.1.3. 3.2.1.4. 3.2.1.5. 3.2.2. 3.2.2.1. 3.2.2.2. 3.2.2.3. 3.2.2.4. 3.2.3. 3.2.3.1. 3.2.3.2. 3.2.3.3. 3.2.3.4. 3.2.4. 3.3.

Type 3: languages with pronoun shift only ...................................162 A mixed type: optional pronoun shift............................................163 Conclusion.....................................................................................164 Object shift across the indirect object............................................165 Type 2: languages with full object shift ........................................166 Type 3: languages with pronoun shift only ...................................177 The mixed type: languages with optional pronoun shift................181 Concluding remarks.......................................................................183 Object shift across the main verb...................................................183 Type 1: languages without regular object shift..............................184 Type 2: languages with full object shift ........................................185 Type 3: languages with pronoun shift only ...................................188 The mixed type: optional object shift ............................................189 Object shift across a verbal particle...............................................190 Summary .................................................................................. 199

Chapter 4 Object shift and other movement types................................................. 205 4.1. Object shift and verb movement............................................... 205 4.1.1. The Danish/Swedish verb movement puzzle.................................206 4.1.2. V-to-I in the Germanic languages..................................................210 4.1.3. Solving the Danish/Swedish verb movement puzzle.....................213 4.2. Object shift and subject shift (A-movement) ........................... 216 4.2.1. Subject shift and the constraint EPP(case)......................................218 4.2.2. Subject shift and the constraint EPP(ij) ..........................................224 4.2.3. Subject shift and the constraint H-COMPL ......................................229 4.2.4. Subject shift and the constraint RELMIN ........................................234 4.2.5. Conclusion.....................................................................................237 4.3. Object shift and A′-movement.................................................. 238 4.3.1. Wh-movement (“shortest steps”) ...................................................239 4.3.2. Quantifier movement and Neg-movement ....................................246 4.3.3. Topicalization ................................................................................254 4.4. Conclusion................................................................................ 260 Chapter 5 Predicate movement................................................................................ 263 5.1. The underlying assumptions..................................................... 264 5.2. Locative inversion in English................................................... 268 5.2.1. Hoekstra & Mulder’s analysis from an MP perspective................268 5.2.2. Locative inversion in the D&E framework ...................................271 5.2.3. Fine-tuning the analysis.................................................................274 5.2.3.1. The Minimal Link Condition.........................................................275

xiv Table of contents 5.2.3.2. 5.2.3.3. 5.2.3.4. 5.2.3.5. 5.2.3.6. 5.2.4. 5.2.5. 5.2.5.1. 5.2.5.2. 5.2.5.3. 5.3. 5.3.1. 5.3.2. 5.3.2.1. 5.3.2.1.1. 5.3.2.1.2. 5.3.2.1.3. 5.3.2.1.4. 5.3.2.1.5. 5.3.2.1.6. 5.3.2.2. 5.3.2.2.1. 5.3.2.2.2. 5.4. 5.4.1. 5.4.2. 5.4.3. 5.4.4. 5.5. 5.5.1. 5.5.2. 5.5.3. 5.6.

Resultative constructions...............................................................278 Copular inversion constructions ....................................................281 Copular inversion and agreement ..................................................284 Predicate inversion and A′-movement of the postverbal subject...286 Conclusion.....................................................................................287 Extending the analysis: VP-movement..........................................288 Conclusion: three brief remarks ....................................................291 Present progressive tense (be V-ing) .............................................292 Locative inversion and topicalization............................................295 Locative inversion and focus.........................................................299 Predicate movement in Dutch .................................................. 300 Locative inversion in Dutch?.........................................................301 Predicate movement in Dutch........................................................304 The placement of small clause predicates .....................................305 The trigger ................................................................................ 305 The constraint A-OVER-A......................................................... 307 Why locative inversion is blocked in Dutch............................. 313 Comparison of the placement of DPs and predicative phrases. 314 A final question ........................................................................ 317 Conclusion................................................................................ 318 Extending the analysis: VP-movement..........................................319 VP-movement in Dutch perfect tense constructions ................ 319 The analysis of the perfect tense construction in the VO-languages 324 Verbal modifiers in Hungarian................................................. 325 Core data and previous analysis ....................................................326 What do the verbal modifiers have in common? ...........................330 The distribution of verb modifiers and focused phrases................333 Conclusion.....................................................................................335 L-participle fronting in Bulgarian............................................. 336 Some properties of l-participles.....................................................337 Long l-participle fronting ..............................................................337 Some speculations on short l-participle fronting ...........................344 Conclusion................................................................................ 347

Chapter 6 Summary and conclusion........................................................................ 349 Appendix: Constraints used in this study ................................................................ 353 Bibliography ............................................................................................ 361 Subject index............................................................................................ 377

Chapter 1 Derivations (MP) and Evaluations (OT)

This chapter describes and discusses the theoretical framework underlying this study: the derivation-and-evaluation model. The central idea of this framework is that developing an explanatorily and descriptively adequate theory of syntax requires that restrictions be formulated both on the syntactic derivations and the resulting syntactic representations. This is obtained by assuming that the framework combines certain aspects of the minimalist program (MP) and of optimality theory (OT). More specifically, I will assume that representations created by some version of the computational system of human language CHL from MP are evaluated in an optimality-theoretic fashion, as indicated in Figure 1. Input

CHL

Output representations

OTEvaluator

Optimal output

Figure 1: The derivation-and-evaluation model

One reason for seriously investigating the properties of the model in Figure 1 and for being optimistic about its explanatory and descriptive adequacy lies in the insight that whereas MP has been especially successful in formulating a restrictive theory of core grammar, that is, the universal properties of grammar as encoded in CHL, OT has been very successful in describing the more peripheral, language-specific properties of languages and the variation between languages. Before I continue I want to make a clarification on the use of the notions of “core” and “periphery” in this chapter. I will use these notions in the sense of Chomsky & Lasnik (1977), without the implication that only the former is part of UG. On the contrary: I will adopt the OT-claim that the constraints that enter the evaluation are part of a universal constraint set CON, and that the only thing that must be acquired by the speaker is the ranking of these constraints. This also implies that the evaluator is part of the “core of linguistic investigation” and that the “true periphery” therefore lies outside the model in Figure 1 and consists of everything that must be learned on an item-by-item or construction-by-construction basis. This will be made more explicit in Figure 13 on page 59.

2 Chapter 1: Derivations (MP) and Evaluations (OT)

The model in Figure 1 goes against the often tacitly adopted but apparently generally accepted view that MP and OT are incompatible, and thus competing, frameworks. In earlier work (Broekhuis & Dekkers 2000; Broekhuis 2000) I have argued, however, that MP and OT are actually complementary frameworks, which can therefore be advantageously combined into a single overarching theory of grammar: MP is mainly a derivational theory that aims at accounting for the universal properties of language, whereas OT is rather a representational theory that focuses on the language-specific properties of language. This chapter will take the earlier claim even one step further, and argue that, despite all the differences between them, MP and OT basically assume the same kind of architecture of grammar, which comes very close to the one in Figure 1. The widely held, and in my view erroneous, belief that MP and OT are incompatible theories of grammar seems mainly due to the fact that the proponents of the two frameworks more or less exclusively focus on only one of the two components of the model in Figure 1: most work in MP focuses on properties of CHL, whereas most work in OT focuses on properties of the OT-evaluator. This chapter is organized as follows. Section 1.1 will substantiate the claim that MP and OT adopt essentially the same architecture of the grammar. Actually it will be shown that this is true for most grammars that have been developed during the principles-and-parameters (P&P) period of generative grammar, which I take to start with Chomsky’s Conditions on transformations (1973). Whereas section 1.1 mainly highlights the similarities between MP and OT, section 1.2 discusses some differences between the two research programs, and argues that these do not inherently follow from the two systems themselves. The discussion in 1.1 and 1.2 will lead to the conclusion that it is readily possible to combine MP and OT into a single overarching model of grammar, and that this gives rise to the derivation-and-evaluation model in Figure 1. Section 1.3 provides a more specific formulation of this model, and briefly illustrates some of its properties. The discussion and claims in this chapter are restricted to syntax, but it goes without saying that I believe that the proposal as worked out in section 1.3 should be extended to other parts of grammar like phonology (see LaCharité & Paradis 2000 for relevant discussion of the role of rules/the generator in OT-phonology; also some more recent work by John McCarthy is similar in spirit). 1.1. Where MP and OT are similar: The architecture of syntax This section will argue that most grammars that have been developed during the P&P period of generative grammar assume the architecture in

Where MP and OT are similar: The architecture of syntax 3

Figure 2, where the Generator and the Evaluator can be held responsible for respectively the universal and language-specific properties of languages. The essential property of this model is that the generator defines a set S of potentially well-formed expressions that can be generated on the basis of a given input, and that the evaluator selects the expressions from S that are actually grammatical in a given language L. Input

Generator

Output representations

Evaluator

Optimal output

Figure 2: The architecture of grammar

The general idea has been very clearly formulated by Chomsky & Lasnik in Filters and Control (1977), where they argue that “to attain explanatory adequacy it is in general necessary to restrict the class of possible grammars, whereas the pursuit of descriptive adequacy often seems to require elaborating the mechanisms available and thus extending the class of possible grammars” (p.44-5). In order to solve this tension they propose that “there is a theory of core grammar with highly restricted options, limited expressive power, and a few parameters” next to a more peripheral system of “added properties of grammar”, which “we may think of as the syntactic analogue of irregular verbs” (p.47). Core grammar consists of the phrase structure and transformational component (the generator in Figure 2), whereas the more peripheral system consists of language-specific surface filters (the evaluator). Chomsky & Lasnik’s main claim is that the introduction of these filters contributes to the simplification of the transformational rules by bearing “the burden of accounting for constraints which, in the earlier and far richer theory, were expressed in statements of ordering and obligatoriness, as well as all contextual dependencies that cannot be formulated in the narrower framework of core grammar”. Although ideas about which aspects of grammar should be considered part of core grammar and which should be considered part of the periphery have changed over the years (and no doubt will change in years to come), the gist of the proposal has survived in the more recent minimalist incarnations of the theory, where core syntax can be more or less equated with CHL, and the periphery with the interface (or bare output) conditions. The task of reducing core grammar as much as possible has been very successful: the reduction of CHL to its absolute minimum (internal and external merge) contributes much to the explanatory adequateness of the theory. But, as expected, the contribution of core grammar to descriptive

4 Chapter 1: Derivations (MP) and Evaluations (OT)

adequacy has diminished accordingly, so that in this respect we have to rely more and more on the interface conditions. Below, I will give a necessarily sketchy overview of the ways in which the global architecture in Figure 2 has been given shape in the various proposals that have been put forth over the last thirty years. I will start in section 1.1.1 with discussing some subsequent proposals within the P&P framework, and show that although the proposed grammars from the earlier period diverge in several respects from the overall structure in Figure 2, the more recent minimalist proposals more and more converge with it. After this, I will give a brief discussion of OT in section 1.1.2, and show that OT fits neatly to the global architecture in Figure 2, which is clear from the fact that some version of it can be found in virtually all introductory texts on OT. 1.1.1.

Principles & Parameters Theory

Since Chomsky & Lasnik (1977), the global organization of the different P&P models has had more or less the shape given in Figure 2 above, although in the earlier proposals this is masked by the fact that instead of a fully linear model, a so-called T- or inverse Y-model was assumed, according to which the derivation of the LF- and the PF-representation diverge after a certain point (s-structure or Spell-Out). The classical GBmodel (Chomsky 1981), for example, looks more or less as indicated in Figure 3, where certain movements can apply between s-structure and LF, which are therefore not reflected in the PF-side of the grammar (so-called covert movement). Lexicon D-structure Move α

S-structure Move α

Phonetic Form (PF)

Logical Form (LF)

Figure 3: The GB model

This property of the early P&P models disappears in the later versions of MP with the introduction of mechanisms like feature movement, spell out

Where MP and OT are similar: The architecture of syntax 5

of copies and Agree, which void the need for covert movement. As a result, these later versions fully accord with the linear model in Figure 2. The answer to the question what is part of the generator (core grammar) and what is part of the evaluator has of course changed over the years. The that-trace filter, for example, was originally proposed as a languagespecific filter for English, but the Empty Category Principle, which ultimately grew out of it, was rather assumed to be part of core grammar. Furthermore, it is not always easy to determine which ingredients were considered part of generator and which of the evaluator since these were normally not discussed in these terms. It is clear, however, that at least the phrase structure and transformational component have consistently been considered part of the generator in all proposals so far. In what follows I will compare the various stages of the P&P framework with the global architecture in Figure 2. First consider the model of grammar adopted by Chomsky & Lasnik in Filters and Control, which is given in Figure 4 below. Input

Core grammar

LF-component

PF-component

Filters

Optimal PF-output

LF-output

Figure 4: The Filters and Control model (Chomsky & Lasnik 1977)

The input of the system is a set of lexical items. The generator contains a phrase structure and a transformational component. The phrase structure component consists of phrase structure rules constrained by X-bar-theory, which combine the lexical elements from the input into a d-structure representation. The transformational rules are constrained by a set of general conditions and modify the d-structure representation into an s-structure representation. The s-structure representation is subsequently fed to the LFand the PF-component of the grammar, where it undergoes further computation. The LF-wing of the grammar contains rules that assign a semantic interpretation to the s-structure representation, for example, rules of construal (binding and control) and quantifier interpretation. The PFwing of the grammar contains rules that assign a phonetic interpretation to the s-structure representation. Among these phonological rules we find deletion and stylistic rules. The language-specific filters, finally, evaluate the resulting PF-representations: only those representations that pass these filters are acceptable in the language under discussion.

6 Chapter 1: Derivations (MP) and Evaluations (OT)

The introduction of a filter component was motivated by the fact that this made a more restrictive formulation of core grammar possible by eliminating ordering statements and language-specific properties from the transformational component of the core grammar. By way of demonstration let us consider the derivation of the relative clauses in (1). (1)

a. the man who I know b. the man that I know c. the man I know d. *the man who that I know

The relative pronoun who is generated in the regular object position, so that the d-structure of the examples in (1) is as given in (2a). Chomsky & Lasnik further propose that universal grammar (UG) contains a universal principle “Move wh-phrase” that requires that relative pronouns (and other wh-phrases) be placed to the left of the complementizer, as in the s-structure representation in (2b). (2)

a. b.

the man [that I know who] the man [[COMP who that] I know twho]

(d-structure) (s-structure)

The examples in (1) can now be derived by assuming a deletion rule that freely deletes the relative pronoun who or the complementizer that. The resulting PF-representations are given in (3). Chomsky & Lasnik further assume the language-specific filter in (4) that prohibits the simultaneous realization of the relative pronoun and the complementizer. This excludes representation (3d). (3)

a. the man [[COMP who that] I know twho] b. the man [[COMP who that] I know twho] c. the man [[COMP who that] I know twho] d. *the man [[COMP who that] I know twho]

(4)

Doubly Filled COMP Filter: *[COMP wh-phrase complementizer]

Although the deletion rule is freely applicable in principle, the resulting representation is subject to a recoverability principle, which requires that deleted elements be locally recoverable. This is needed to block deletion of the wh-phrase in representations like (5): the recoverability principle in tandem with the Doubly Filled COMP Filter in (4) ensures that the examples in (5b-d) are excluded.

Where MP and OT are similar: The architecture of syntax 7 (5)

a. I wonder [who that you met twho] b. *I wonder [who that you met twho] c. *I wonder [who that you met twho] d. *I wonder [who that you met twho]

By the same means, deletion of the preposed PP in relative clauses like (6) is blocked. Deletion of about which violates the recoverability principle because the preposition about cannot be recovered locally. (6)

a. the book [about which that he spoke tabout which] b. *the book [about which that he spoke tabout which] c. *the book [about which that he spoke tabout which] d. *the book [about which that he spoke tabout which]

The virtue of Chomsky & Lasnik’s proposal of the data above is that by accounting for the language-particular properties of the English constructions by means of the filter in (4), we can keep the transformational rule that derives s-structure (2b) maximally simple. This, in turn, makes it possible to attribute this rule to UG. Example (7) gives Chomsky & Lasnik’s first tentative formulation of this rule. (7)

The rule “Move wh-phrase” places the wh-phrase in the COMP position, to the left of the complementizer.

In the Government-and-Binding (Chomsky 1981) and Barriers (Chomsky 1986) period, the model of grammar remains essentially the same. The attempts to further reduce the transformational component of the core grammar led to the formulation of the general rule Move Į. As far as the filter component was concerned, it turned out that some of the filters proposed in Chomsky & Lasnik (1977) had a wider application and could be reformulated as more general principles. For example, the thattrace filter, which prohibits a trace immediately to the right of the English complementizer that, was reformulated as/reduced to the Empty Category Principle (ECP), which requires that a trace be properly governed. This change is depicted in Figure 5. Input

Core grammar

PF-component

LF-component

Principles

Filters

LF-output

Figure 5: The GB/Barriers model (Chomsky 1981;1986)

Optimal PF-output

8 Chapter 1: Derivations (MP) and Evaluations (OT)

Although the ECP was claimed to be universal, that is, to be part of UG, its function is more or less the same as that of the that-trace filter: it excludes structures that have been created by core grammar. Therefore the formulation of the ECP is not a reason to look with a skeptical eye on the notion on the notion of filter: it should rather give us hope that we can also obtain a certain degree of explanatory adequacy in the domain of filters. In the Minimalist Program, as developed by Chomsky since the mid 1980s, core grammar seems to have been reduced to its absolute minimum. The computational system of human language CHL, as it is now called, consists of essentially one merge operation in two guises. External merge, which I will refer to as Select, combines two independent syntactic objects into a larger syntactic unit, whereas internal merge, which I will refer to as Move, takes some element from an existing syntactic object, and merges it to the root of this object, thus deriving the effect of movement. (8)

• The operations of CHL (Merge): a. Select (external merge): take two independent syntactic objects and merge them. b. Move (internal merge): take an element from a syntactic object and merge it to the root.

Merge is subject to a number of general conditions. For example, it never involves more than two elements at the same time, which results in binary branching phrase structures. Move obeys certain locality restriction and is further subject to the Last Resort Condition, which requires that movement be triggered by some uninterpretable/unvalued formal feature. As in Chomsky & Lasnik (1977), descriptive adequacy lies mainly outside the core system: for example, Chomsky (1995a: section 4.7.3) suggests (rightly or wrongly) that “rearrangement” phenomena like extraposition, right-node raising, VP-adjunction, and scrambling are essentially the result of stylistic rules of the phonological component. Although the notion of filter is not used, MP also heavily relies on the filter component. It seems that this filter component has taken various guises in the various stages in the development of the program. The organization of grammar in Chomsky (1995a: ch.3) is more or less as indicated in Figure 6.

Where MP and OT are similar: The architecture of syntax 9

Input

Generator CHL

PF-component

Filters

LF-component

Economy conditions

LF-output (satisfying FI)

Optimal PF-output (satisfying FI)

Figure 6: The early MP model (Chomsky 1995a: ch.3)

Many of the filters as discussed in Chomsky & Lasnik (1977) have not found an alternative account in MP, but the fact that they are not discussed is, of course, no guarantee that they are not needed: this motivates the postulation of a set of PF-filters in Figure 6. Furthermore, Chomsky (1995a) explicitly assumes that CHL generates a set of converging (=potentially well-formed) derivations satisfying Full Interpretation, the reference set. It is further assumed that the optimal output is the representation that satisfies a number of global economy conditions best: derivations with a smaller number of derivational steps are preferred (Fewest Steps), as are derivations with shorter movement chains (Shortest Steps). The language L thus generates three relevant sets of derivations: the set D of derivations, a subset DC of convergent derivations of D, and a subset DA of admissible derivations of D. [Full Interpretation] determines DC, and the economy conditions select DA. [...] DA is a subset of DC. (Chomsky 1995a:220)

It is not so clear to what extent the global economy conditions still play a role in the current formulation of MP. It seems that very soon they lost their independent status by being successfully incorporated into the definition of the movement operation. Fewest Steps was replaced by the Last Resort Condition (Chomsky 1995a:280) and Shortest Steps was, at least partly, replaced by the Phase Impenetrability Condition in Chomsky (2001); cf. section 4.3.1 for discussion. As a result, DC and DA can be considered identical and we are left with only two sets of derivations: the set of derivations D and the set of converging derivations DC. Another important innovation in Chomsky (1995a: ch.4, p.221) is the introduction of the bare output conditions, which are later normally referred to as the interface conditions. According to Chomsky, these interface conditions are “imposed from the outside” by the performance systems that make use of the representations created by CHL, and which include (perhaps at most) the articulatory-perceptual and the conceptual-intentional system. Chomsky claims that the interface conditions may be involved in the displacement property of language, and we will see shortly that in later

10 Chapter 1: Derivations (MP) and Evaluations (OT)

work, he formulates one of these conditions in the format of a filter on the output of CHL (Chomsky 2001); cf. the discussion of (13/24) below. So let us provisionally assume that the interface conditions can be formulated as filters on the output of the PF- and the LF-component, as in Figure 7. Input

Generator CHL

PF-component

LF-component

LF-Filters

PF-Filters

Optimal PF-output (satisfying FI)

Optimal LF-output (satisfying FI)

Figure 7: The later MP model (Chomsky 1995a: ch.4)

The assumption that the interface conditions can be formulated as filters on the output of the PF- and the LF-component does of course not necessarily imply that all filters on (or, stated more neutrally, the evaluation of) the output of CHL can be fully traced back to the properties of the articulatoryperceptual or the conceptual-intentional system (cf. Chomsky 1995a:222-3). But one of the main goals of my investigation is to show that it is at least possible for the range of data discussed in this study. As was noted at the beginning of this section, a conspicuous property of the P&P models discussed above is that they differ from the linear model in Figure 2 in that the derivation of the PF- and LF-representations diverge at a certain point in the derivation in order to account for the fact that there can be certain mismatches between linear order and semantic interpretation. Very early in the development of MP, proposals were put forth to eliminate this property from the grammar. Groat & O’Neil (1996), for example, noted that the copy theory of movement made it possible to account for the discrepancies in PF- and LF-representations by assuming that phonology could either spell out the lower or the higher copy in a movement chain (cf. also Bobaljik 2002). Chomsky (1995a: ch.4) noted that economy considerations can account for these discrepancies by assuming that it is more economical to move a syntactic category without its phonological features, pied piping of the phonological features being possible only when there are independent reasons to do so. Finally, the introduction of Agree (feature checking at a distance) in the Minimalist Inquiry framework (Chomsky 2000 and subsequent work) made overt movement totally superfluous from the point of view of core grammar. As a result of this we can assume that the derivation of the LF- and PF-representations proceed in fully parallel fashion. The model of grammar assumed in this framework is therefore as indicated in Figure 8.

Where MP and OT are similar: The architecture of syntax 11

Generator CHL

Input

Output PF/LFrepresentations (satisfying FI)

PF/LFFilters

Optimal output

Figure 8: The Minimalist Inquiry model (Chomsky 2000 and later)

Since Agree makes movement superfluous as far as core grammar is concerned, movement must be forced by external factors, more specifically by the interface conditions imposed on the output representations of CHL. Actually, the intuition underlying this proposal is much older than the Minimalist Inquiry framework. For example, it has been argued that the motivation for wh-movement is that a wh-phrase can only be interpreted if it heads an operator-variable chain; cf., e.g., Chomsky (1991:440) and Rizzi (1996). Consequently, Full Interpretation forces wh-movement. This, in turn, implies that in theories without “covert” movement like the Minimalist Inquiry framework, wh-in-situ languages like Japanese or Chinese also have “overt” wh-movement. This is precisely what has been claimed by Watanebe (1991), who argued on empirical grounds that whmovement in these languages involves empty operator movement. Chomsky (2001) aims at showing that certain types of A-movement are also externally motivated. We will look at this in some detail in what follows. According to MP, movement of a syntactic object S is subject to the Last Resort Condition: it must be triggered by some unchecked or unvalued formal feature of a higher functional head H that can be checked or valued by a corresponding feature of S. In the earliest proposal it was assumed that these features of H come in two forms: weak and strong features. A strong feature on H must be checked before the projection of H is merged with some higher head; if checking does not take place, the derivation is canceled. A weak feature on H, on the other hand, cannot be checked before Spell-Out as a result of the economy condition Procrastinate. This proposal led to a very rigid system in which the question whether a certain movement does or does not apply overtly is mechanically determined by the feature constellation of the functional head H. However, it is clear that movement may be sensitive to other factors as well. Consider the case of object shift in the Icelandic examples in (9). (9)

a. b.

Jón keypti ekki bókina. Jón bought not the book Jón keypti bókinai ekki tbókina

bókina ⊂ focus bókina ⊂ presupposition

The examples in (9) demonstrate that it is possible in Icelandic to move the direct object to the left, across the negative adverb ekki. This movement is,

12 Chapter 1: Derivations (MP) and Evaluations (OT)

however, not obligatory and depends on the information structure of the clause: object shift applies only when the object is part of the presupposition (“old” information) of the clause; it is excluded when the object is part of the focus (“new” information) of the clause. Let us provisionally assume that object shift is triggered by the case feature on the light verb v*, as proposed by, e.g., Vikner (1994); cf. also Chomsky (2001), which will be reviewed below. If this case feature were strong, we would wrongly expect this movement to be obligatory; if it were weak, we would wrongly predict it to be impossible. In order to account for the apparent optionality of object shift, we must therefore introduce additional means. One possibility would be to make the strength of the case feature sensitive to the information structure of the clause: only when the object is part of the presupposition of the clause does v* have a strong case feature. Apart from being ad hoc, this option is not descriptively adequate since object shift is never possible in complex tense constructions like (10): object shift is excluded irrespective of the information structure of the clause, and (10a) is therefore ambiguous. (10)

Jón hefur ekki keypt bókina. bókina ⊂ focus or presupposition Jón has not bought the book b. *Jón hefur bókina ekki keypt tbokina a.

Another possibility is to follow Holmberg (1999) in claiming that object shift is actually not part of core grammar. He proposes that object shift is a phonological operation that is driven by the interpretation of the object: in the terminology used above, object shift is only possible if the object is part of the presupposition of the clause. This is stated in (11a), which paraphrases Chomsky’s (2001: (54a)) summary of Holmberg’s claim. Holmberg (1999:15) accounts for the ungrammaticality of (10b) by postulating the additional restriction on the application of object shift in (11b): object shift is blocked in (10b) because it would move the object across the main verb. (11)

a.

b.

Object shift is a phonological movement that satisfies condition (11b) and is driven by the semantic interpretation INT of the shifted object: (i) INT: object is part of the presupposition of the clause. (ii) INT′: object is part of the focus of the clause. Holmberg’s Generalization (1999 version): Object shift cannot apply across a phonologically visible category asymmetrically c-commanding the object position except adjuncts.

Where MP and OT are similar: The architecture of syntax 13

Chomsky (2001:32) argues that Holmberg’s proposal is problematic because “displacement rules interspersed in the phonological component should have little semantic effect” (p.15), and he therefore develops a proposal according to which object shift takes place in core syntax. The relevant configuration is given in (12), where Objectș is the thematic position of the object, and Objects is a specifier position of v* created by object shift (note that Chomsky assumes a multiple specifier approach). (12)

... [Į Objects [Subject v* [V ... Objectș ]]]

Note that (12) is an intermediate stage in the derivation: at some later stage in the derivation the subject is moved into SpecTP; in simple tense constructions the v*+V complex is moved to T. Given this, Chomsky (2001: (61)) tries to account for the properties of Icelandic object shift in (11) by adopting the assumptions in (13), where INT and INT′ are defined as in (11a). (13)

a. b. c.

v* is assigned an EPP-feature only if that has an effect on outcome. The EPP position of v* is assigned INT. At the phonological border of v*P, XP is assigned INT′.

The EPP-feature mentioned in (13a) has the same function as the strong features in the earlier proposals in that it forces movement of some element into a specifier position of the head that it is assigned to. The statement in (13a) must be considered an invariant principle of grammar, which expresses that v* is only assigned an EPP-feature if the resulting movement has some effect on the output representation. According to Chomsky this is the case when the movement affects the interpretation of the clause, or when it makes A′-movement possible (by placing the object at the phonological edge of the v*P-phase). We will see shortly that this leads to a less rigid system in the sense that movement can be made sensitive to factors other than the feature constellation of the attracting head. Chomsky (2001) claims that (13b) is also an invariant principle: in the terminology employed earlier, this claim expresses that an object occupying the position Objects in (12) must be construed as being part of the presupposition of the clause. It is important to note that (13b) is only concerned with shifted objects, and leaves open the option that non-shifted objects are ambiguously interpreted as being part of either the focus or the presupposition of the clause. This is needed in order to allow the non-shifted objects in Icelandic examples like (10a) to be interpreted as part of the presupposition of the clause, and, of course, also correctly predicts that the

14 Chapter 1: Derivations (MP) and Evaluations (OT)

objects in languages like English, which do not have object shift of the Icelandic sort, can be part of either the focus or the presupposition of the clause. Given that (13b) does not restrict the interpretation of non-shifted objects, we need something in addition to account for the fact that object shift is obligatory in examples like (9b). This is where (13c) comes in. Let us first consider the notion of phonological border, which is defined as in (14).1 (14)

XP is at the phonological border of v*P, iff: a. XP is a v*P-internal position; and b. XP is not c-commanded by v*P-internal phonological material.

The main difference between the examples in (9) and (10) is that in the former the main verb has moved out of v*P into T, whereas in (10) it has not and thus occupies a v*P-internal position. Example (10a) is therefore correctly predicted to be ambiguous: since the v*+V complex is v*Pinternal and c-commands the object, clause (13c) does not apply and the object can be interpreted either as part of the focus of the clause (INT′) or as part of the presupposition of the clause (INT). Example (10b) is consequently blocked by (13a) because object shift has no effect on the outcome as the object can also be assigned the interpretation INT in its base position in (10a). Therefore, in constructions like (10), the EPP-feature can only be assigned to v* if it is needed to enable A′-movement. In (9), on the other hand, there is no v*P-internal phonological material that c-commands 1 Since Chomsky assumes that XP is in an outer specifier of v*, clause (14a) should actually say that XP is part of the internal domain of v*, that is (dominated by) the complement of v*. I will not digress on this since I will adopt a slightly different view on movement that does not involve multiple specifiers. Furthermore, in order to make this proposal more widely applicable, it might be necessary to add to the leftto clause (14b).ҏ Suppose that the type of Dutch scrambling that derives (ib) from (ia) is also A-movement (cf. section 2.1). When one assumes that the OVorder is base-generated, the addition to the left is needed since then the participle c-commands the object but nevertheless does not block scrambling; when one assumes an underlying VO-basis, the OV-order is the result of leftward movement of the object out of the c-command domain of the participle, and the addition is not needed. It seems that Chomsky has to opt for the first since he claims there to be an OV/VO-parameter distinguishing Icelandic from Dutch and German.

(i)

a. b.

Jan heeft waarschijnlijk het boek opgehaald. het boek ⊂ focus Jan has probably the book prt.-fetched Jan heeft het boeki waarschijnlijk ti opgehaald. het boek ⊂ presupposition

Where MP and OT are similar: The architecture of syntax 15

the ș-position of the object. Consequently, if the object occupies this position, (13c) states that it must be assigned INT′. Movement of the object into the Objects position in (12) therefore has an effect on the outcome because it enables the assignment INT, so that (13a) allows assignment of an EPP-feature to v*. It is important to note that statement (13c) clearly functions as a filter in the sense of Chomsky & Lasnik (1977). First, it is clear that it cannot be considered a condition on the derivation; when we apply it to the intermediate stage in (12), the desired distinction between (9) and (10) cannot be made locally (in the sense of Collins 1997, that is, in the course of the derivation), because the verb and the subject are moved out of the v*P only at a later stage in the derivation. Chomsky (2001) therefore assumes that it applies at the higher phase level (CP). Second, (13c) is a language-specific statement; Icelandic (as well as the continental Germanic languages) is subject to it, and therefore object shift is forced in examples like (9b); the Romance languages, on the other hand, are not subject to it, so that (13a) blocks object shift in comparable Romance examples. Thus, statement (13c) has two characteristic properties of the PF-filters proposed in Chomsky & Lasnik (1977). It differs from these filters, however, in that it is sensitive both to phonological and to semantic information. But this is, of course, to be expected if filters in one way or another reflect the fact that the output of CHL is fed both to the articulatory-perceptual and to the conceptualintentional system. Let me conclude with pointing out a potential problem with Chomsky’s account of Holmberg’s Generalization. His proposal crucially presupposes that reordering of phonological material is not enough to satisfy (13a). If reordering would suffice, we could no longer block object shift in (10b) given that it inverts the order of the main verb and the object, so that it has an effect on the outcome at PF. However, restricting the notion of “effect on the outcome” to LF is surprising given that in Chomsky’s (1995a:294) earlier work this notion explicitly referred to properties of PF; cf. Sabel (2005) for an extensive discussion and interesting application of this idea. See in this connection also Chomsky’s (1986: section 9) Vacuous Movement Hypothesis and Williams’ (2003:162) suggestion that linearization and interpretation play a similar role in making movement “visible”. This subsection has shown that all grammars proposed during the P&P era have the global architecture of grammar indicated in Figure 2, although this was obscured in the early period by the fact that it was assumed that the derivations of the PF- and LF-representation diverge at some point. It has been shown that by rejecting this assumption Chomsky’s recent Minimalist

16 Chapter 1: Derivations (MP) and Evaluations (OT)

Inquiry framework fully conforms to the architecture in Figure 2 in that the grammar now consists of a generative component that creates PF/LFrepresentations that are subsequently evaluated by a filter component. The filters place both semantic and phonological constraints on the output of CHL, which reflects the fact that the PF/LF-representation(s) that pass these filters are subsequently fed to the articulatory-perceptual and the conceptualintentional system where they undergo further computation in order to receive a phonetic and a semantic interpretation. 1.1.2.

Optimality Theory

Optimality theory (OT) fits nicely to the global architecture of grammar in Figure 2, which is clear from the fact that it can actually be found in virtually all introductory texts on OT. Nevertheless, it is certainly not easy to describe the substantive contents of each of the components mentioned in the model. The input, for example, depends on the part of grammar we are talking about. For phonology, for example, it is generally assumed that the input consists of underlying phonological representations, which are of course not suitable for syntax. But even if we restrict our attention to syntax, it is clear that there is hardly any consensus on the question of what the nature of the input is: in some proposals it is assumed that the input consists of a set of lexical elements comparable to the numeration in MP, in other proposals the input is a structured meaning, and sometime it is even assumed that the input consists of prefabricated syntactic representations (thus leaving open the question how these are created). Something similar holds for the generator. McCarthy & Prince (1993) assume that the generator consists of linguistic operations subject to “very general considerations of structural well-formedness”. As a rule we only find scattered remarks on the nature of these operations and the restrictions they are subject to: Grimshaw (1997), for example, claims that the generator builds structures in accordance with some version of X-bar-theory. We can therefore conclude that the generator is still largely unanalyzed in OT, certainly where syntax is concerned. Nevertheless, it is crucial that the generator is an overgenerating system. It creates a candidate set from which the evaluator selects the optimal candidate(s). It is generally assumed that this candidate set is infinite and contains many candidates that will never surface because they are harmonically bound by some other candidate (where A is harmonically bound by B if A violates at least one constraint on top of the constraints violated by B). Note that OT differs in this respect from most P&P approaches, according to which the output of the computational system is finite and consists of potentially well-formed

Where MP and OT are similar: The architecture of syntax 17

representations (GB being a bit exceptional in this respect by categorically excluding some of these representations by means of, e.g., the ECP or the Case Filter). In OT, the focus of attention is on the evaluator. It consists of a set of constraints with the properties in (15a-c), which I will discuss more extensively below. (15)

The optimality-theoretic evaluator contains constraints that: a. are taken from a universal set of constraints CON; b. are violable; and c. have a language-specific ranking.

The constraints crucially differ from the language-specific filters assumed in the principles-and-parameters theories in that they are generally assumed to be universal, that is, to be part of UG. It is assumed that there is a universal set of constraints CON from which the constraints that are active in a given language are taken — in fact, normally it is assumed that all constraints from CON are active, but that the effects of some constraints are simply not observable; cf. (16a) below. The constraints can nevertheless be used to express language-specific properties due to the two other properties of the constraints: according to (15b) constraints need not be obeyed but can in principle be violated, although this normally comes with a certain penalty; according to (15c) languages may differ with respect to the ranking of these constraints, whereby violation of a higher ranked constraint is punished more severely than violation of a lower ranked constraint. The determination of the optimal candidate proceeds as in (16), which I adapted from Archangeli (1997). (16)

The evaluator finds the candidate that best satisfies the ranked constraints, such that: a. violation of a lower ranked constraint is tolerated in order to satisfy a higher ranked constraint, and b. ties by violation or by satisfaction of a higher ranked constraint are resolved by a lower ranked constraint.

The way the OT-evaluator works can readily be demonstrated by means of Pesetsky’s (1997;1998) analysis of relative clauses. This will also give me the opportunity to show how the OT-evaluator differs from the filters assumed in the P&P approaches. Consider again the representations of the relative clauses from examples (1) and (6), given here as (17) and (18). Recall that the judgments on these representations were accounted for in

18 Chapter 1: Derivations (MP) and Evaluations (OT)

Filters and Control by taking recourse to the Doubly Filled COMP Filter and the recoverability condition on deletion. (17)

a. the man [whoi that I know ti] b. the man [whoi that I know ti] c. the man [whoi that I know ti] d. *the man [whoi that I know ti]

(18)

a. the book [about whichi that he spoke ti] b. *the book [about whichi that he spoke ti] c. *the book [about whichi that he spoke ti] d. *the book [about whichi that he spoke ti]

When we contrast these examples with the French relative clauses in (19) and (20), we see that English and French differ in that the former allows a wider variety of constructions with a bare relative pronoun than the latter. However, when the relative pronoun is embedded in a PP (or an NP), the two languages behave the same. (19)

a. *l’homme [quii que je connais ti] b. l’homme [quii que je connais ti] c. *l’homme [quii que je connais ti] d. *l’homme [quii que je connais ti]

(20)

a. l’homme [avec quii que j’ai dansé ti] b. *l’homme [avec quii que j’ai dansé ti] c. *l’homme [avec quii que j’ai dansé ti] d. *l’homme [avec quii que j’ai dansé ti]

In order to account for the data in (17) to (20), Pesetsky proposed the constraints in (21), which I haved slightly simplified here for reasons of exposition. Constraint (21a) is simply the recoverability condition on deletion from Chomsky & Lasnik (1977), constraint (21b) is a constraint that expresses that embedded clauses tend to be introduced by a complementizer, and (21c) is a constraint that expresses that function words (like complementizers) tend to be left unpronounced. (21)

a. b. c.

RECOVERABILITY (REC):

a syntactic unit with semantic content must be pronounced unless it has a sufficiently local antecedent. LEFT EDGE (CP): the first leftmost pronounced word in an embedded CP must be the complementizer. TELEGRAPH (TEL): do not pronounce function words.

Where MP and OT are similar: The architecture of syntax 19

The ranking of these constraints will determine the optimal output. In order to see this, it is important to note that LE(CP) in (21b) and TEL in (21c) are in conflict with each other: the first wants complementizers to be pronounced, whereas the latter wants them to be deleted. Such conflicts make it possible to account for variation between languages: when we rank these constraints differently, we get languages with different properties. When we assume that LE(CP) outranks TEL, as in (22a), we get a language in which embedded declarative clauses must be introduced by a complementizer. When we assume the ranking in (22b), however, we get a language in which embedded declarative clauses are not introduced by a complementizer. When we assume that the two constraints are in a tie (ranked equally high), as in (22c), we get a language in which embedded declarative clauses are optionally introduced by a complementizer.2 (22)

a. b. c.

LE(CP) >> TEL: embedded declarative clauses are introduced by a complementizer. TEL >> LE(CP): embedded declarative clauses are not introduced by a complementizer. TEL LE(CP): embedded declarative clauses are optionally introduced by a complementizer.

The evaluation can be made visible by means of a tableau. Tableau 1 gives the evaluation of embedded declarative clauses with and without a pronounced complementizer in a language with the ranking in (22a). Tableau 1: No C-deletion in embedded declarative clauses LE(CP)

.... [ complementizer ....] .... [ complementizer ....]

)

TEL

* *!

The two asterisks indicate that the constraint in the header of their column is violated. The first candidate, with a pronounced complementizer, violates TEL but this is tolerated because it enables us to satisfy the higher ranked constraint LE(CP). The second candidate, with a deleted complementizer, violates LE(CP), and this is fatal, which is indicated by an exclamation mark, because the first candidate does not violate this constraint. The first candidate is therefore optimal, which is indicated by means of the pointed 2 The use of angle brackets “” indicates that there are two rankings active at the same time: A B means that both A >> B and B >> A are active, and the two rankings may each result in the selection of a different optimal candidate. We are dealing with grammar competition in the sense of Kroch (1989;1994).

20 Chapter 1: Derivations (MP) and Evaluations (OT)

finger: ). The shading of cells following a fatal constraint violation indicates that these cells do not play a role in the evaluation; this convention is mainly for convenience, because it makes it easier to read the tableaux. Now consider the evaluation of the same candidates in a language with the ranking in (22b), given in Tableau 2. Since TEL is now ranked higher than LE(CP), violation of the former is fatal, so that deletion of the complementizer becomes obligatory. Tableau 2: Obligatory C-deletion in embedded declarative clauses TEL

.... [ complementizer ....] .... [ complementizer ....]

LE(CP)

*! )

*

Tableau 3 gives the evaluation according to the ranking in (22c), where the two constraints are in a tie, which is indicated in the tableau by means of a dashed line. Under this ranking the two rankings in (22a&b) are simultaneously active. Therefore we have to read the tie in both directions: when we read the tie from left to right, the violation of LE(CP) is fatal (which is indicated by >), and the first candidate is optimal; when we read the tableau from right to left, the violation of TEL is fatal (which is indicated by

Let us now return to the difference between English and French with respect to the pronunciation of relative clauses. It is clear that English is a language of type (22c) given that the complementizer is normally optional in embedded declarative clauses. French, on the other hand, is a language of type (22a): the complementizer is obligatory in embedded declarative clauses. Pesetsky (1997) has shown that this also accounts for the differences between the English and French examples in (17) and (19), in which a bare relative pronoun is preposed. Assume that in both languages the constraint RECOVERABILITY outranks the constraints TEL and LE(CP); the rankings of the constraints in (21) are then as given in (23).

Where MP and OT are similar: The architecture of syntax 21 (23)

a. b.

French: REC >> LE(CP) >> TEL English: REC >> TEL LE(CP)

The evaluation of the French examples in (19) proceeds as in Tableau 4. Since the relative pronoun has a local antecedent it is recoverable after deletion, so that all candidates satisfy REC. The second candidate is the optimal candidate because it is the only one that does not violate LE(CP); the fact that this candidate violates the lower-ranked constraint TEL is tolerated since this in fact enables the satisfaction of the higher-ranked constraint LE(CP); the violations of TEL therefore do not play a role, which is again indicated by shading the cells of the non-optimal candidates. Tableau 4: Relative clauses with preposed relative pronoun

French l’homme [quii que je connais ti] l’homme [quii que je connais ti] l’homme [quii que je connais ti] l’homme [quii que je connais ti]

REC

LE(CP)

TEL

*! )

* *! *!

*

The evaluation of the English examples in (17), shown in Tableau 5, is slightly more complex than that of the French ones due to the fact that LE(CP) and TEL are in a tie: we are therefore dealing with two rankings at the same time: REC >> LE(CP) >> TEL and REC >> TEL >> LE(CP). The first ranking is actually the one we also find in French, and we have seen that this results in selection of the second candidate as optimal. Under the second ranking, violation of TEL is fatal, so that the first and third are selected as optimal. As a result, three out of the four candidates are grammatical in English. Tableau 5: Relative clauses with preposed relative pronoun

English the man [whoi that I know ti] the man [whoi that I know ti] the man [whoi that I know ti] the man [whoi that I know ti]

REC

) ) )

LE(CP)

TEL

*> *< *> *>

*
> TEL does not mean that the complementizer is always realized, but that this may depend on other factors; when the complementizer is preceded by some element that must be realized, TEL forces the complementizer to delete. Tableau 6: Relative clauses with preposed PP

French l’homme [avec quii que j’ai dansé ti] l’homme [avec quii que j’ai dansé ti] l’homme [avec quii que j’ai dansé ti] l’homme [avec quii que j’ai dansé ti]

REC

)

LE(CP)

TEL

* *! *!

* * *

*!

For the English examples in (18) we get the same result as in French: in Tableau 7 both the second and the third candidate are excluded by REC, and the fourth candidate is excluded because it is harmonically bound by the first candidate, that is, it has a fatal violation of TEL irrespective of whether we read the tie from left to right or from right to left (violations of tied constraints that are fatal on all rankings available in the language are simply indicated by means of an exclamation mark). Tableau 7: Relative clauses with preposed PP

English the book [about whichi that he spoke ti] ) the book [about whichi that he spoke ti] the book [about whichi that he spoke ti] the book [about whichi that he spoke ti]

REC

LE(CP)

TEL

* *! *!

* * *

*!

The discussion above has shown that OT fully adheres to the global architecture in Figure 2, although the focus of attention is on the evaluator. The OT view on the evaluator seems to be more optimistic in nature than that of the P&P approaches. The latter consider the evaluator as a more or less random collection of language-specific filters on the output of core grammar. Pesetsky’s work has shown, however, that at least some of the filters proposed by Chomsky & Lasnik (1977) can be decomposed into more atomic OT-constraints; see Dekkers (1999) for more examples. Furthermore, since the OT-constraints are claimed to be universal, they

Where MP and OT are similar: The architecture of syntax 23

make precise predictions about the range of language variation that is allowed: Pesetsky, for example, has shown that his proposal is able to account for the differences between English and French relative clause constructions, and Broekhuis & Dekkers (2000) and Dekkers (1999) have shown that his proposal can be readily extended to relative constructions in (dialects of) Dutch. 1.1.3.

Conclusion

This section has argued that the global architecture of grammar is as given in Figure 2 on page 3, and that the several proposals made within the principles-and-parameters approach do not differ in this respect from OTsyntax. The two frameworks are similar in assuming that we are dealing both with derivations and with evaluations: a generator creates a potentially multi-membered set of expressions S, and an evaluator determines which expressions from S are grammatical in a given language L. Although this section has mainly focused on the similarities in architecture between the P&P approaches and OT-syntax, it must be noted that there are other similarities between the two frameworks. For example, both MP and OTsyntax adopt some version of Frege’s principle of compositionality of meaning by claiming that meaningful elements must be interpreted: in MP it is assumed that interpretable semantic features cannot be deleted and must receive an interpretation (Full Interpretation); the fact that Pesetsky’s (1997;1998) constraint RECOVERABILITY is universally ranked high expresses more or less the same thing,3 as does Grimshaw’s (1997) claim that all candidates in a certain candidate set have the same meaning. I will not digress on this, however, and continue the discussion by focusing on some differences between the two frameworks. 1.2. Where MP and OT differ: derivations and evaluations The previous section has argued that MP and OT assume the same global architecture of grammar. However, there are also obvious differences. This subsection will briefly discuss these and argue that they do not have a principled linguistic motivation, but are the result of a more or less accidental difference in focus of attention between the two approaches: MP is mainly concerned with the universal, derivational aspects of grammar, 3 Given that there are no known cases in which RECOVERABILITY is violated, Broekhuis & Dekkers (2000:421) argued that it should actually not be considered a constraint but an inviolable condition on the operation Delete; see the introduction to section 1.3.3 for more discussion.

24 Chapter 1: Derivations (MP) and Evaluations (OT)

whereas OT-syntax rather focuses on more language-specific aspects of grammar, or, to put it differently, MP is basically a theory of CHL, the generator from the model in Figure 2, whereas OT is basically a theory of the evaluator. This difference between MP and OT is also reflected in the research strategies that the two approaches employ, which in a sense are each other’s opposite. Research in MP tends to attribute as many properties of languages to the generator CHL as possible; although we have seen in the discussion of Icelandic object shift in section 1.1.1 that MP does allow for filtering devices, researchers seem to take recourse to these as a last resort only, that is, when no derivational account is available. Research in OT, on the other hand, tends to attribute as many properties of languages to the evaluator; although it is generally acknowledged that the generator has certain universal properties, these are hardly ever invoked to account for the data. Given that MP is a theory of the generator and OT-syntax is a theory of the evaluator, it is not surprising that the empirical successes of the two approaches lie in different areas. MP is especially well equipped to account for the universal properties of languages, but there is no generally accepted view on how we should account for, or even approach, the many ways in which languages may differ from each other. OT, on the other hand, precisely provides such a general theory of language variation, but since there is no generally accepted theory of the generator, current OT-syntax fails to account for the “truly” universal properties of languages. These differences between MP and OT will be discussed more extensively below. 1.2.1.

Universal properties of language (the generator)

Both MP and OT-syntax hold the generator responsible for the invariant properties of language: the generator determines what representations are contained in the output, and hence can take part in the evaluation. The two frameworks differ, however, with respect to the extent that the generator is developed, or invoked in the analysis of the linguistic data. The investigation of the generator (CHL) is considered MP’s core business. It has resulted in a sophisticated, restrictive theory on the nature of the generator. It is assumed that CHL consists of a small set of operations that are subject to inviolable conditions that are relatively well understood. Perhaps CHL can be reduced to a single merge operation, which has two incarnations, Select (external merge) and Move (internal merge). As a result of this, the output of CHL is also highly restricted; although it can be a non-singleton set, the differences between the members of this set are very limited in nature, and perhaps only involve the number of movements that

Where MP and OT differ: derivations and evaluations 25

have occurred (cf. the discussion of Icelandic object shift in section 1.1.1). It seems that analyses that do not invoke filtering devices are more highly valued in MP than those that do. As a result, research tends to focus on those phenomena that can be successfully approached by means of a derivational account, with a concomitant reduction of the empirical scope of the theory; Chomsky (1995a: section 4.7.3), for example, suggests that “rearrangement” phenomena like extraposition, right-node raising, VPadjunction and scrambling are not part of core syntax. It is generally admitted in OT-syntax that the generator is the locus of the “truly” universal properties of language. For example, Grimshaw (1997) assumes that the structures formed by the generator conform to some version of X-bar-theory, Pesetsky (1998) and Anderson (2000) adopt some version of generative grammar as the generator, and Bresnan (2000) and Sells (2001) argue in favor of some version of Lexical Functional Grammar. The nature of the generator is, however, not a prominent subject of research, which is possibly also related to the fact that the current generation of OTsyntacticians has come from various theoretical frameworks with varying views on the nature of the generator. Furthermore, it is rather exceptional for an OT-researcher to account for some phenomenon by taking recourse to the generator; most research in OT-syntax rather focuses on the variation that can be found than on the universal properties of languages. Despite the differences in theoretical background (P&P, LFG, etc.), it seems that the view on the generator of many (if not most) OT-syntacticians crucially differs from that of the MP-researchers. This becomes especially apparent when we consider the differences in the view on the output of the generator. We have already seen that although MP allows for non-singleton reference sets, it is generally taken for granted that the reference set is very small and that differences between the members of this set are limited in type, perhaps confined to differences in movement. In OT, on the other hand, it is generally maintained that the output of the generator is in principle infinitely large, and that the members of the set may differ in a wide variety of ways. This seems to imply that the generator contains a larger set of operations in OT than is assumed in MP, and that these operations are probably confined in a less strict manner than the operations assumed in MP. As a result of this different view on the generator, MP and OT tend to provide entirely different explanations for similar phenomena, the former mainly appealing to properties of the generator and the latter to properties of the evaluator. This state of affairs seems to strengthen the widely accepted view that we are dealing with two competing and essentially incompatible frameworks. However, it can also be assessed differently, and

26 Chapter 1: Derivations (MP) and Evaluations (OT)

more positively. Since it is not a priori given whether a certain phenomenon belongs to core syntax or to the periphery, it is important to develop alternative analyses that can subsequently be compared and evaluated; the fact that in some domains competing MP- and OT-analyses are available therefore does not mean in itself that we are dealing with competing or conflicting theories. In fact, there are similar conflicts internally in MP. Take as an example verb second, which has long been considered a prototypical example of a phenomenon that is part of core syntax, and which has played an important role in the development of the theory of functional heads (especially the CP projection) and head movement (verb movement to C). Nevertheless, since CHL as developed in Chomsky (1995a: section 4.10) is no longer able to handle verb second in Icelandic transitive expletive constructions (p.354), Chomsky concluded that it should be considered part of the periphery, as the result of some not further explicated PF-rule (p.368). Taken to its extreme, this proposal may lead to the claim that verb second, like the other “rearrangement” phenomena mentioned earlier, is not part of core syntax at all, but essentially a PF-phenomenon; cf. Chomsky (2004) and especially Boeckx & Stjepanovic (2001), who explicitly argue that head-movement in general is a PF phenomenon. However, it is also conceivable that a slightly different view on CHL makes it possible to properly account for the problematic fact, thus maintaining the idea that verb second is part of core syntax (see fn.31 on p.128 for a concrete proposal). Note in this connection that claiming that head-movement is a phonological operation is actually incompatible with Chomsky’s (2001) analysis of Icelandic object shift, which I discussed in section 1.1.1. Chomsky convincingly argues that Holmberg’s (1999) proposal that object shift is a post-syntactic operation cannot be maintained since “displacement rules interspersed in the phonological component should have little semantic effect”, so that “surface semantic effects are restricted to narrow syntax” (2001:15). If this is indeed the case, we must conclude that verb second is likewise a syntactic operation; it is not likely that the application of a syntactic rule can be blocked by the failure of applying a phonological rule, given that the latter is not detectable in core syntax. Take Chomsky’s analysis of Icelandic object shift in (13) on p.13, and assume that verb second is a phonological rule. This implies that the main verb invariably occupies a VP-internal position in the syntax so that we cannot determine whether the object occupies the phonological border of the v*P before we have entered the phonological component, which, however, is too late for narrow syntax and LF. The claim that post-syntactic operations do not have semantic effects therefore makes it impossible to formulate the language-

Where MP and OT differ: derivations and evaluations 27

specific statement in (13c), according to which XP is assigned interpretation INT′ at the phonological border of v*P — such statements make sense only when verb second is also a syntactic operation.4 This section has argued that the fact that OT and MP provide competing analyses for the same phenomena does not show that MP and OT should be seen as competing or conflicting theories but should rather be seen as a normal reflex of the fact that it is not a priori given whether a certain phenomenon belongs to core syntax or to the periphery. The question which of the available analyses are most feasible is therefore essentially an empirical one. 1.2.2.

Variation (the evaluator)

One of the main concerns of both MP and OT is cross-linguistic variation. However, the way they approach this problem is entirely different - at least, at first sight. Let us start by discussing the way MP approaches the issue. Language variation is assumed to arise as a result of additional constraints on the application of the otherwise universal generator (CHL). The generator can basically perform two operations: Select (external merge) and Move (internal merge). Let us provisionally adopt the standard assumption in MP that Select is indispensable, given that it is needed in order to assemble lexical items into semantically interpretable structures, e.g., by saturating the thematic roles of a given lexical head. Despite the fact that movement may have certain semantic implications, it is not essential in the creation of semantically interpretable structures, so that we expect to find language variation in this domain.5 Note that since MP is mainly concerned with core syntax, it also mainly studies differences between languages that are somehow related to movement: variation found in other domains is attributed to other modules (like PF), and is generally not discussed any further. In early versions of MP, the locus of variation between languages is solely attributed to the lexicon. Differences in the displacement property of languages are due to differences in the “strength” property of the morphosyntactic features that trigger movement: strong features trigger overt 4

See in this connection also Lechner (2005), who likewise argues against the claim that head-movement is the result of a PF-operation by pointing out that headmovement may affect the interpretation. 5 Chomsky claims that due to this difference between Select and Move, the former operation is essentially “costless”, whereas the latter operation is “costly”, so that when the two operations compete, the former is preferred. I will argue in section 1.3.3 that this is actually not correct. Note in this connection that we have seen that, like Select, A′-movement operations that result in operator-variable chains may be indispensable in the creation of semantically interpretable structures.

28 Chapter 1: Derivations (MP) and Evaluations (OT)

movement, whereas the weak features allow covert movement (which is favored by Procrastinate). In the more recent Agree-based theories, which reject the idea of covert movement, the core idea is preserved by assuming that movement only takes place if a functional head F contains an EPPfeature, which requires that the specifier of F be present. Under this view, the task of the language learner is to determine whether the functional head F has a weak or strong feature, or, alternatively, whether it has an EPPfeature, and to store this information in the lexicon. The scope of OT goes much beyond the displacement property of languages: in principle, all (phonological, syntactic, semantic, pragmatic, etc.) properties can be fruitfully investigated, as long as one can plausibly postulate constraints bearing on the phenomenon in question. As we have already seen, variation between languages is attributed to the evaluator in Figure 2, more specifically to the differences in ranking of the otherwise universal constraints. Under this view, the task of the language learner is to determine the constraint ranking (and the lexicon) of the language. The discussion above seems to reveal another important difference between MP and OT: in the former cross-linguistic variation is solely due to differences in lexical specifications, whereas in the latter it is rather due to the ranking of the universal constraints. This is indeed the case when we compare early MP with OT-syntax, but it no longer holds when we compare the most recent Minimalist Inquiry framework and OT-syntax. The early MP thesis that the sole locus of cross-linguistic variation is the lexicon runs into severe problems when we consider variation within a single language, because it predicts that languages cannot have “optional” movement, by which I refer to movement operations that occur only under certain well-defined semantic or phonological conditions. One example of this type of movement is Icelandic object shift (already discussed in section 1.1.1), which can only apply when the object is part of the presupposition of the clause (cf. (9)), and when it does not cross the verb (cf. (10)) or other v*P-internal material. This kind of optionality cannot arise under the early MP thesis because the postulation of feature strength or an EPP-feature gives rise of to a very rigid system: when a feature is strong/an EPP-feature is present, movement must apply; when a feature is weak/an EPP-feature is not present, movement is blocked by Procrastinate. This problem has led to proposals according to which in some cases certain features are optionally strong or an EPP-feature is optionally present. In order to avoid circularity, the choice must be made sensitive to external factors like the semantic and phonological conditions imposed on the pertinent movement, and this is precisely what Chomsky (2001) did in his account of object shift in Icelandic in (13), repeated below as (24): as we

Where MP and OT differ: derivations and evaluations 29

have seen, the language-specific statement in (24c), in tandem with the universal principles in (24a&b), precisely derives the circumstances under which Icelandic object shift applies. (24)

a. b. c.

v* is assigned an EPP-feature only if that has an effect on outcome. The EPP position of v* is assigned INT. At the phonological border of v*P, XP is assigned INT′.

Chomsky (2001:36) presents clause (24c) as a parameter that distinguishes object shift from non-object shift languages. French, for example, has verb movement to I, but nevertheless object shift does not apply. This can be accounted for by assuming that (24c) does not hold for French. In this language, the interpretation INT can therefore also be assigned to the object when it is at the phonological border of v*P, and, as a result, movement of a presuppositional object into the EPP-position of v* is not needed, so that assignment of an EPP-feature to v* is consequently blocked by (24a). It seems, however, that (24c) is unlike the parameters of the earlier P&P framework in that it is not binary. This is clear from the fact that it is not the case that languages can be straightforwardly divided between object shift and non-object shift languages. This will become clear when we consider the Danish examples in (25) and (26), taken from Vikner (1994:502). The examples in (25) show that Danish, unlike Icelandic, does not have object shift of non-pronominal DPs, whereas the examples in (26) show that it does have object shift of weak pronouns. (25)

a.

Hvorfor læste studenterne ikke artiklen? why read the students not the article b. *Hvorfor læste studenterne artikleni ikke ti?

(26)

a.

Hvorfor læste studenterne deni ikke ti ? why read the students it not b. *Hvorfor læste studenterne ikke den?

This can be accounted for by assuming that clause (24c) must be further refined as in (24c′). This clause makes the following correct predictions: (i)non-pronominal DPs that are part of the presupposition of the clause (=INT) must undergo object shift in Icelandic, but not in Danish or the Romance languages; (ii) definite pronouns (which are assigned INT by definition) must undergo object shift in Icelandic and Danish; (iii) the Romance languages do not have any kind of object shift because the set of elements that is assigned INT′ in the relevant context is empty, which is indicated by “∅”.

30 Chapter 1: Derivations (MP) and Evaluations (OT) (24)

c′.

At the phonological border of v*P, XP is assigned INT′ (i) XP = DP (Icelandic) (ii) XP = weak definite pronoun (Danish) (iii) XP = ∅ (Romance)

For completeness’ sake, note that the fact that English does not have object shift does not follow from clause (24c′): since English does not have V-to-I movement, objects are never at the phonological border of v*P; (24c′) never applies and object shift is therefore blocked by (24a). What I want to stress here is that the adoption of language-specific statements like (24c) or (24c′) is a radical break with the early MP thesis that the sole locus of cross-linguistic variation is the lexicon. Since statements like these essentially function as language-specific filters on the output of CHL, they should be attributed to the evaluator in the model in Figure 2, and not to the lexicon. In fact, it seems that Chomsky’s proposal makes it possible to eliminate the EPP-features entirely: when we assume that movement is subject to the Last Resort Condition but applies optionally, we could simply replace clause (24a) by the statement that movement is possible only if it has an effect on the outcome. This would make it possible to attribute cross-linguistic language variation entirely to the evaluator, just as in OT. In (27), I attempt to rephrase Chomsky’s proposal such that reference to the notion of EPP-feature becomes superfluous. (27)

1.2.3.

a. b. c.

Move is possible only if it has an effect on outcome. The derived object position is assigned INT. At the phonological border of v*P, XP is assigned INT′ (i) XP = DP (Icelandic) (ii) XP = weak definite pronoun (Danish) (iii)XP = ∅ (Romance)

Conclusion

Since we have seen in section 1.1 that MP and OT adopt more or less the same global organization of grammar, we may conclude that the differences in the research strategies of MP and OT are somewhat accidental: as far as I can see, there are no theory-internal reasons for these frameworks to limit their investigation one-sidedly to, respectively, the generator or the evaluator. The fact that MP and OT occasionally provide alternative analyses for similar data as a result of these differences in research strategy does not follow from insurmountable theoretical differences between the two frameworks either, but simply reflects the fact that it is not a priori given whether a certain phenomenon belongs to core syntax or to the periphery.

Where MP and OT differ: derivations and evaluations 31

Early MP and OT-syntax do seem to adopt conflicting views on the nature of variation between languages: the former adopts the thesis that language variation can be reduced to differences in the feature specifications of the lexical elements (feature strength/EPP-features), whereas the latter assumes that language variation is due to the evaluator, that is, to differences in constraint rankings. In Chomsky’s current Minimalist Inquiry framework, however, the early MP thesis has been dropped: language variation is (also) attributed to parameters like (27c), which essentially function as language-specific filters on the output of CHL. Current MP and OT therefore both attribute language variation to the evaluator, and the main difference between MP and OT boils down to whether the evaluator takes recourse to output filters or to ranked constraints. In sum, we may conclude that MP and OT-syntax are actually much more alike than is generally assumed or one would be inclined to think at first sight. Given that the strengths and weaknesses of the two frameworks are somewhat complementary (MP being especially successful in accounting for the universal, derivational aspects of grammar, and OT-syntax being especially well equipped to account for variation), it is fully justified and useful to investigate whether the strengths of the two frameworks can somehow be combined. This will be the topic of the next section. 1.3. The derivation-and-evaluation framework This section aims at developing the derivation-and-evaluation (D&E) model in Figure 1, repeated below as Figure 9, in which the strengths of MP and OT are combined. The name of the model underlines the claim that the generator and the evaluator are equally important for providing descriptions and explanations of linguistic phenomena. The D&E model differs from the current versions of OT-syntax in that it adopts a version of CHL as its generator, and it differs from MP in claiming that the output of CHL is not evaluated by means of filters but in an optimality-theoretic fashion. Input

CHL

Output representations

OTEvaluator

Optimal output

Figure 9: The architecture of grammar

Adopting the model in Figure 9 makes it necessary to seriously investigate the interaction between the generator and the evaluator: after all, when both the generator and the evaluator are to be taken seriously, they are expected to interact in intricate ways so that properties ascribed to the former may have far-reaching consequences for the design of the latter, and vice versa.

32 Chapter 1: Derivations (MP) and Evaluations (OT)

Sections 1.3.1 and 1.3.2 will discuss the generator and evaluator, respectively, and compare the assumptions underlying the D&E framework with those normally adopted in MP and OT-syntax. 1.3.1.

The generator

D&E adopts the standard assumption from MP that the computational system CHL is universal and consists of operations that are conceptually necessary, such as the two incarnations of the merge operation, Select and Move, and possibly the operation Delete. The latter operation is needed to account for deletion of the phonological features of complementizers and relative pronouns (cf. the discussion of relative clauses above), although it is not a priori clear whether Delete should be considered an operation of CHL or of the phonological component; see section 1.3.3 for an argument in favor of the latter position. Furthermore, D&E adopts the claim that these operations are subject to inviolable conditions: any application of Move, for example, must satisfy the Last Resort Condition, according to which movement of a syntactic object S must be triggered by some unchecked or unvalued formal feature of a higher functional head H that can be valued by a feature corresponding to S, and Delete is subject to the recoverability condition (see fn.3 on p.23 and section 1.3.3 for discussion). The main difference between D&E and the “standard” versions of MP is that the former assumes that CHL is not parameterized: more specifically, it is assumed that there are no strength/EPP-features that may force or block the application of a certain operation, and neither can an operation be blocked by the availability of a more economical option.6 At any point P in the derivation, CHL may choose at random between applying or not applying the operation(s) that could in principle be performed (= would satisfy the Last Resort Condition) at P. Consequently, the number of candidates in the candidate set is at most 2n, where n is the number of operations that satisfy Last Resort; see Figure 10.

6 More specifically, section 1.3.3.2 will argue that there is no general preference for Select (external merge) over Move (internal merge), and that assuming this gives rise to false empirical predictions. The fact that Select seems to come for free in English is due to the fact that Move is generally not a favored option in this language, that is, that *MOVE is ranked fairly high and outranks *SELECT. We will see that in Dutch the ranking is arguably the inverse: *SELECT >> *MOVE.

The derivation-and-evaluation framework 33 Operation 1 yes

no

Operation 2 yes

no

yes

no

Operation 3 yes

no

yes

no yes

no

yes

no

Figure 10: The construction of the candidate set

CHL thus defines a candidate set that contains a limited number of candidates, and is defined by the optional application of the operations Select, Move, and Delete. Of course, the effects of the strength/EPP-features must be mimicked in some way, but we have seen in section 1.2.2 that the filters introduced in Chomsky (2001) in effect already determine whether certain movements may or may not apply; this makes the EPP-features superfluous: cf. the discussion above (27). We may therefore conclude that, as far as the generator is concerned, the D&E model in Figure 9 comes very close to the more current versions of MP. The D&E claim that the generator should be identified with the computational system CHL from MP breaks radically with the generally adopted OT-claim that the candidate set is infinite; the claim that the operations of the generator licensed by the Last Resort Condition can in principle be optionally applied results in candidate sets that are very small.7 By way of illustration, (28a&b) give the maximum size of the candidate sets for derivations with, respectively, 8 and 16 operations that satisfy the Last Resort Condition. Actually, it is even possible to reduce these numbers much further by adopting some version of phase theory. This is shown in (28a′&b′). (28)

7

The size of the candidate set: a. 8 operations: 28 = 256 a′. 8 operations in 2 phases of 4 operations each: 2 x 24 = 32 b. 16 operations: 216 = 65.536 b′. 16 operations in 4 phases of 4 operations each: 4 x 24 = 64

What is maintained, however, is that the candidate set can be assumed to be very similar for all languages: variation may arise but this is mainly the result of differences in the lexicon, such as the availability of certain lexical items, or the (non-)affixal status or the categorial nature of the lexical elements involved in the derivation.

34 Chapter 1: Derivations (MP) and Evaluations (OT)

I believe that this radical break with the OT-tradition is also advantageous from the OT point of view. First, of course, CHL can be invoked to provide a non-ad hoc account for the truly universal properties of languages, which the OT-evaluator by its very nature is not able to give. Secondly, since part of the descriptive burden is now placed on the generator, we may hope that this will enable us to considerably reduce the number of constraints in the universal constraint set CON. This, in turn, will result in a dramatic decrease in the number of constraint rankings, and, consequently, in the number of possible natural languages. Thirdly, the fact that CHL does not only limit the candidate set, but also the type of differences that can be found among the candidates in this set, which are defined by the application or nonapplication of the operations of CHL, suggests that it will be possible to also reduce the number of constraint types, and, consequently, also the ways in which natural languages can differ from each other. It goes without saying that all these consequences contribute to considerably enhancing the explanatory adequacy of OT-syntax. 1.3.2.

The evaluator

The previous subsection has briefly mentioned the D&E proposal that the EPP-features should be eliminated by attributing the intended effects of these features to the evaluator. From the point of view of MP, this step seems quite natural since I have already argued in section 1.2.2 that the filters introduced in Chomsky (2001) actually suffice to determine whether certain movements may or may not apply. A general problem with filters is, however, that they tend to take the form of ad hoc stipulations that simply reformulate descriptive generalizations or the description of certain states of affairs in a semi-formal language. Since it is not obvious that this will lead to any deeper insights, the D&E framework adopts the idea that filters should be subject to further investigation, and be derived from more primitive notions of the theory. It further assumes that this is precisely what OT does: work by Pesetsky (1997;1998) and Dekkers (1999) has already shown that at least some of the filters from Chomsky & Lasnik (1977) may receive a natural explanation in this way, and this section will show that the language-specific filter in (27c) can also be expressed by means of the interaction of a small set of more primitive constraints (cf. Costa 1998 and Broekhuis 2000). The previous section has also argued that by adopting CHL as the generator, the OT-evaluator can be considerably simplified: since the inviolable conditions on the operations of the generator carry part of the descriptive burden, we may expect a reduction of the number of constraints that are part of CON, and since the candidates in the candidate set only

The derivation-and-evaluation framework 35

differ from each other in a small number of well-defined ways (see Figure 10), we may also expect the number of constraint types to be rather small. In order to get some idea about the syntactic constraints and constraint types that we may expect, I will adopt as my point of departure the proposal that the OT-evaluator is a formalization of the interface conditions postulated in MP. If that is indeed so, we expect the syntactic constraints in CON to be somehow related to three components involved: the computational system CHL, which creates the relevant syntactic representations in the candidate set, and the two interpretive systems that interpret them, namely the articulatory-perceptual and the conceptual-intentional component. Let us therefore assume that the syntactic constraints in CON can be divided into the two basic types in (29). (29)

The syntactic constraints in CON are of two basic types: a. CHL constraints b. Interface (PF and LF) constraints

Before I discuss these constraint types, I want to point out that, in my view, it is not only desirable to restrict the number and kind, but also the possible format of the constraints. I will therefore adopt Eisner’s (1999) proposal that there are basically two formal types of constraints, which should be formulated as positive or negative generic statements. Eisner refers to these two types of statements as, respectively, the implication and the clash constraints: good examples of the former type are the EPP-constraints in (32) below, which require that all probes attract their goal, and a good example of the latter type is *Move, which bans movement of any kind. Furthermore, I will assume that the formulation of the constraints is simple in the sense that connectives like and, or, unless, etc. cannot be used. 1.3.2.1. CHL constraints The D&E framework assumes that the application of the operations of the generator is essentially free. Nevertheless, it is clear that most languages exhibit strict restrictions on the application of these operations. A good example of this is object shift: languages like Icelandic and Dutch/German have it, whereas the Romance languages do not. Given the claim that the generator is universal and cannot be parameterized, it must be the evaluator that penalizes the application of this movement. Therefore, we must postulate a set of clash constraints that favor non-application of the operations of CHL, which I will henceforth refer to as economy constraints. A first example of such an economy constraint is STAY, which I prefer to call *MOVE in order to highlight the fact that it is a clash constraint: *MOVE

36 Chapter 1: Derivations (MP) and Evaluations (OT)

forbids internal merge, and thus militates against superfluous movement steps in the derivation. Assuming this constraint seems uncontroversial: it is assumed in most work in OT-syntax, and it has its MP counterpart in the claim that movement is a costly operation. In the early MP period, this claim has played a crucial role in the formulation of principles like Procrastinate and Fewest Steps, and it has survived in the later period in the form of the proposal that movement is licit only when an EPP-feature is present. It has been proposed that the economy constraints on movement may take a more specific form. For example, Grimshaw (1997) proposes the constraint NO LEXICAL MOVEMENT (NOLEXM), which disfavors movement of the lexical (ș-role assigning) verbs. This constraint is a reformulation of Pollock’s (1989) ban of movement of lexical verbs to weak AGR-phrases: English has a weak AGR, and therefore movement of a lexical (but not an auxiliary or a modal) verb is blocked; French has a strong AGR, and consequently movement of a lexical (as well as an auxiliary or a modal) verb is possible. (30)

a. b.

John often Mary. Jean souvent Marie.

Note in passing that it is not so clear how this contrast between English and French is accounted for within current MP. Although the contrast was extensively discussed in Chomsky’s (1991) Some notes on economy of derivation and representation (included as chapter 2 in Chomsky 1995a), it has not received much attention in the MP literature since, which is perhaps due to the fact that some researchers claim that head movement is not part of core syntax, but of the periphery; cf. the discussion in section 1.2.1. Another potential economy constraint is NO VACUOUS MOVEMENT (NOVACM). This constraint, which states that movement must have an effect on the linear order of the output, will be used in section 2.4 to account for the fact that object shift normally crosses the adverbial phrases (if present).8 This constraint is simply an adapted version of Chomsky’s (1986: section 9) Vacuous Movement Hypothesis, which still features in current MP in the form of “effect on outcome” conditions like (24/27a); cf. discussion in section 1.1.1. It is interesting in the present context to note that Chomsky (1986:50) explicitly presents the Vacuous Movement 8

NOVACM may also play a role in the account of English do-support given that it makes the inversion of the subject and the inflectional node I visible, e.g., What [Idid] he tI say? vs. What [IØ] he tI said? or If this is on the right track, this will also allow us to account for do-support with, or Under no circumstances [Ido] we tI allow smoking in our limousines.

The derivation-and-evaluation framework 37

Hypothesis as a violable constraint that can be overruled by “more general considerations”. The logic that has led to the postulation of *MOVE leads to the expectation that there is also a constraint *SELECT that disfavors external merge, and thus militates against superfluous elements in the derivation. Postulating *SELECT is, however, controversial from the point of view of current MP, where it is generally claimed that external merge is costless. This claim is intimately related to the predominant view within MP that the input of the derivation is a numeration, which must be empty at the end of the derivation. More specifically, Chomsky (1995a:226) claims that “a derivation converges only if [Select/Merge] applies often enough to leave us with a single object, also exhausting the initial numeration”; insufficient application of Select thus implies that “no derivation is generated and questions of convergence and economy do not arise”. Section 1.3.3 below will argue, however, that the claim that the derivation takes a numeration as its input is empirically inadequate. Instead, I will adopt Grimshaw’s (1997) claim that the candidates in a given candidate set share the same meaning, where meaning is assumed to include notions from information structure, such as the distinction between the focus and the presupposition (new and old information) of the clause. If so, the numeration is not needed to define the possible output representations of the derivation, so that lexical elements can be assumed to be taken directly from the lexicon, which, in turn, implies that Chomsky’s argumentation in favor of the costless nature of Select cannot be maintained. In fact, section 1.3.3.2 will show that the claim that Select is costless wrongly predicts that the unvalued formal features on the functional heads would always be checked by merging some appropriate dummy element from the lexicon, if that is available. Above, we have seen that we must postulate a set of economy constraints that disfavor the application of the operations of CHL, and I have assumed that this set includes the constraints in (31), which block the application of (certain instances of) Select and Move. The motivation for the postulation of the constraints in (31) will become clear later in this study; note that, following the canonical literature, I will use NOLEXM and NOVACM instead of *LEXM and *VACM. (31)

a. b. c. d.

*SELECT: don’t apply external merge. *MOVE (STAY): don’t apply internal merge. NO LEXICAL MOVEMENT (NOLEXM): don’t apply internal merge to lexical (ș-role assigning) verbs. NO VACUOUS MOVEMENT (NOVACM): don’t apply internal merge when it is not visible in the output.

38 Chapter 1: Derivations (MP) and Evaluations (OT)

For completeness’ sake, it must be noted that we do not expect an economy constraint *AGREE that blocks the application of Agree, which has been argued to be a prerequisite for the application of Move (Chomsky 2000), and, one might argue, also for the application of Select.9 If Agree is blocked, that is, when an unvalued feature fails to agree with an appropriate goal, the derivation simply does not converge (=cannot be interpreted) and the resulting representation is consequently excluded from the candidate set. Given that the economy constraints in (31) block the application of the operations of the generator, we must also introduce means that allow or force the operations of CHL to apply. It doesn’t seem necessary to introduce special constraints in order to allow the application of Select (external merge): this operation may be largely semantically driven in the sense that it is forced by the conceptual-intentional component. For example, this component may require that all thematic roles are appropriately saturated (just like ș-theory did in the earlier P&P period), and therefore that the lexical heads are merged with their arguments. More generally, if Grimshaw (1997) is indeed correct in claiming that the candidates in a given candidate set share the same meaning, something similar will hold for the other elements in the clause. In effect, Grimshaw’s claim has more or less the same effect as Chomsky’s claim that external merge comes for free, although there are differences, to which I will return in section 1.3.3. Since we have seen that languages differ in their displacement properties, we cannot appeal to some general property of the conceptualintentional or the articulatory-perceptual component to force movement. Therefore, we have to postulate constraints that favor movement, so that the relative ranking of these constraints and the economy constraint *MOVE will determine whether a certain movement does or does not take place. Of course, we want to restrict the class of constraints that force movement as much as possible. In order to obtain this let us assume that all probes (=unvalued formal features) prefer movement of their goal (=the element containing a feature that may check/value the probe) into their local domain. In passing, note that the intended definition of local domain is identical to 9

Chomsky’s (2000:122) claim that Move is a complex operation consisting of the three more basic operations Associate, Agree and internal Merge can be readily extended to Select. The main difference between Move and Select is that in the former case the associate is internal to the syntactic object headed by the probe, whereas in the latter case it is external to it. From this difference and the requirement that the output of CHL be a single syntactic object, it also follows that Merge must immediately follow Association and Agree with an external object, whereas Merge need not follow Associate and Agree with an internal object. See section 1.3.3.2 for more discussion.

The derivation-and-evaluation framework 39

the one of checking domain in Chomsky (1995a: ch.3); I prefer using the notion of local domain, however, in order to avoid the connotation that movement into the local domain of a head H is required to value the unvalued features. The assumption that all probes prefer movement of their goal into their local domain means that, in a certain sense, we are generalizing the EPP to all unvalued features. The general form of the EPP constraints is therefore as given in (32), and states that movement of a goal into the local domain of its probe is required. Consequently, if the goal of probe F is an XP, this constraint forces it to move into a specifier of the head that has F as its sublabel, and, if the goal is a head, it will be adjoined to the head that has F as its sublabel. Potential specific instantiations of the “generalized” EPP constraint, which will play a key role in the analysis of object shift in this study, are given in (32a-c). The constraints EPP(case) and EPP(ij) require movement of a DP into the specifier of a head containing unvalued case or ij-features, and EPP(tense) requires head-movement of the finite verb to T. (32)

EPP(F): probe F attracts its goal. a. EPP(case): an unvalued case feature attracts its goal. b. EPP(ij): unvalued ij-features attract their goal. c. EPP(tense): an unvalued tense feature attracts its goal.

d.

etc.

It is obvious that the number of EPP constraints cannot be larger than the number of unvalued features that are postulated in the grammar. It is, however, less clear whether the two numbers are equal. Take wh-movement. Watanabe (1991) has argued on empirical grounds that wh-in situ languages like Japanese actually have overt wh-movement of an empty operator: among other things, this accounts for the fact that these languages also exhibit wh-island effects. Chomsky (1995a: ch.3) claimed on the basis of Watanabe’s findings that wh-features are universally strong. When we abandon covert movement in favor of Agree, Watanebe’s findings suggest that Agree does not suffice to license wh-constructions, but that movement must apply. So the question is: Why? Earlier proposals have maintained that a wh-phrase can be interpreted by the conceptual-intentional component only if it heads an operator-variable chain; cf., e.g., Chomsky (1991:440) and Rizzi (1996). If so, the obligatoriness of wh-movement follows immediately from semantic considerations, since any construction in which wh-movement does not apply will either crash as a violation of Full Interpretation or, at least, receive an anomalous interpretation. Consequently, the postulation of a constraint like EPP(wh) has no effect, so that we may safely assume that it does not exist. If movement of the goals of other [+affect] features like

40 Chapter 1: Derivations (MP) and Evaluations (OT)

[topic], [focus] or [neg] are likewise forced by semantic considerations, we may conclude that they do not fall under the generalized EPP constraint in (32) either. This would eliminate a large set of potential constraints from the grammar, and thus considerably reduce the set of possible grammars. Since establishing that there are no EPP constraints related to the [+affect] features constitutes a research program in its own right and since this issue does not play a prominent role in the present study, I will not pursue this any further. I will therefore leave it to future research, while noting that Watanabe (1991) in my view already conclusively established this for the wh-feature by showing that the wh-in situ languages also have overt wh-movement. A potential problem is, of course, that in multiple whquestions only a single wh-element needs to move, so that the in situ whphrases apparently violate the LF-requirement that they head an operatorvariable chain. It has been argued, however, that the in situ wh-phrases are different from moved wh-phrases in that they receive an interpretation by entering into a dependency relation with the moved wh-phrase (cf. Williams 2003: section 6.1 for discussion). The claim that there are no EPP constraints related to the [+affect] features seems also feasible for the [neg]feature, as Haegeman (1995) argues that in at least some languages failure to apply Neg-movement results in constituent negation; this seems confirmed by the findings on the Scandinavian languages in Svenonius (1996) and Christensen (2005). Focus-movement seems to constitute more of a challenge: although Focus-movement seems obligatory in languages like Hungarian (see section 5.4), there are also languages like English without Focus-movement, and languages like Dutch in which Focus-movement seems optional. It has been argued, however, that the two options are associated with two different readings: movement results in an exhaustive focus reading that expresses that the “background” of the proposition applies exclusively to the displaced focused element (see, e.g., Neeleman & Van de Koot, to appear, and Wagner 2007 for more discussion). Word order variation between languages is accounted for by assuming that the EPP constraints interact in an optimality-theoretic fashion with the economy constraints. Ranking (33a) expresses that probe F (normally) does not trigger movement due to the fact that the EPP constraint is outranked by the economy constraint *MOVE: this ranking will be called “weak”, since it is more or less equivalent to assuming that probe F is weak or has no EPPfeature associated with it. Ranking (33b), on the other hand, expresses that probe F (normally) does trigger movement due to the fact that the EPP constraint outranks the economy constraint *MOVE: this ranking will be called “strong”, since it is more or less equivalent to assuming that probe F is strong or has an EPP-feature associated with it.

The derivation-and-evaluation framework 41 (33)

a. b.

Weak ranking: *MOVE >> EPP(F) Strong ranking: EPP(F) >> *MOVE

The choice between the weak and the strong ranking of a certain EPP constraint constitutes one of the ways in which languages can be parameterized. In fact, (33) constitutes a clear example of what one may call a macroparameter. For example, if we continue to assume that object shift is triggered by the case features on v*, we may distinguish between languages like Icelandic, which has full object shift, and languages that have only partial object shift or no object shift at all, by the two rankings in (34). (34)

a. b.

*MOVE >> EPP(case): object shift is (normally) blocked. EPP(case) >> *MOVE: object shift (normally) applies.

Of course, we have seen that object shift is much more complicated than this, and this is where the interface (PF and LF) constraints come in. 1.3.2.2. The interface (PF and LF) constraints One of the disadvantages of early MP was that the postulation of feature strength or the association of an EPP-feature with certain formal features gave rise to a very rigid system: if a certain formal feature is assumed to be strong or to be associated with an EPP-feature, it is predicted that it invariably triggers movement; if a certain formal feature is assumed to be weak or not to be associated with an EPP-feature, Procrastinate predicts that the pertinent movement is invariably blocked. As we have seen above, Chomsky (2001) has tried to make the system more flexible by making the selection of the EPP-features dependent on semantic and phonological factors. The three statements in (24), repeated here as (35), ultimately have the effect that v* is only assigned an EPP-feature (i) when the object is assigned the interpretation INT (=when the object is part of the presupposition of the clause), and (ii) when the object is at the phonological border of v*P, that is, when object shift does not result in crossing v*Pinternal material. (35)

a. b. c.

v* is assigned an EPP-feature only if that has an effect on outcome. The EPP position of v* is assigned INT. At the phonological border of v*P, XP is assigned INT′.

The statement in (35c) is assumed to be a parameter: object shift languages have it, whereas non-object shift languages do not. Further, we have seen that the introduction of (35c) makes the postulation of EPP-features superfluous, so that we can simply replace (35a) by the assumption that

42 Chapter 1: Derivations (MP) and Evaluations (OT)

movement is optional in principle, as in (36a). Finally, we have seen that the parameter in (35c) does not suffice, since some languages like Danish have limited object shift with weak definite pronouns. This means that (35c) must be further refined as in (36c). (36)

a. b. c.

Movement is possible only if it has an effect on outcome. The derived object position is assigned INT. At the phonological border of v*P, XP is assigned INT′. (i) XP = DP (Icelandic) (ii) XP = weak definite pronoun (Danish) (iii)XP = ∅ (Romance)

Macro-parameters in the format of (33) introduce the same kind of flexibility as filters like (36c). Although movement is normally blocked under the weak ranking in (33a), movement can be forced provided that there is some higher ranked constraint A that favors this movement (cf. (37a)); in the terminology of Chomsky (1995a: ch.3), one might say that constraint A overrules “Procrastinate”. Similarly, although movement is normally forced under the strong ranking in (33b), it can be blocked if there is some higher ranked constraint B that disfavors it (cf. (37b)); in other words, constraint B overrules “Strength”. (37)

a. b.

A >> *MOVE >> EPP(F) Ÿ if A favors movement, “Procrastinate” is overruled. B >> EPP(F) >> *MOVE Ÿ if B disfavors movement, “Strength” is overruled.

The claim that I want to make here is that it is the function of the interface constraints to overrule macro-parameters of the type in (33). I will illustrate this more specifically for the macro-parameter in (34). We have seen that Danish has object shift of a more limited type: although non-pronominal DPs do not shift, definite pronouns do. This was illustrated by the examples in (25) and (26), repeated here as (38) and (39). (38)

a.

Hvorfor læste studenterne ikke artiklen? why read the students not the article b. *Hvorfor læste studenterne artikleni ikke ti?

(39)

a.

Hvorfor læste studenterne deni ikke ti ? why read the students it not b. *Hvorfor læste studenterne ikke den?

The derivation-and-evaluation framework 43

This can be accounted for by assuming that Danish has the weak ranking *MOVE >> EPP(case) in (34a), but that this weak ranking is overruled by a constraint that requires weak/definite pronouns to be vP-external. The claim that there is a restriction of this sort on the placement of pronouns is not new: Diesing (1997:380), for example, claims that definite pronouns are variables that due to their definiteness cannot remain within the nuclear scope of the clause (VP). Let us assume that something of the sort is indeed the case, and postulate the clash constraint D-PRONOUN in (40a), which requires that definite pronouns be vP-external. The fact that Danish has object shift with definite pronouns only can now be accounted for by assuming the ranking in (40b). This is shown by the evaluations of the examples in (38) and (39) in Tableaux 8 and 9.10 (40)

a. b.

D-PRONOUN:

A weak/definite pronoun must be vP-external: *[vP ... pron[+def] ...]. Danish: D-PRONOUN >> *MOVE >> EPP(case)

Tableau 8: Danish (no object shift of non-pronominal DPs) DPRONOUN

Hvorfor læste studenterne ikke artiklen? ) Hvorfor læste studenterne artikleni ikke ti

*MOVE

EPP

(case) * *!

Tableau 9: Danish (obligatory pronoun shift) DPRONOUN

Hvorfor læste studenterne ikke den Hvorfor læste studenterne deni ikke ti )

*MOVE

EPP

(case) *

*! *

The subranking D-PRONOUN >> *MOVE can again be seen as a macroparameter which subdivides the languages that do not have full object shift 10 One might of course also assume that Danish has a strong ranking of and try to formulate a constraint that blocks object shift of non-pronominal DPs; so far I have not been able to formulate such a constraint in a plausible way. Although the main text refers to Diesing’s semantic account of pronoun shift, and thus suggests that D-PRONOUN is an LF constraint on definite pronouns, it must be noted that there are also analyses of pronoun shift that focus on the phonological properties of these weak pronouns: cf. Vogel (to appear). Of course, this would not affect the main claim, i.e., that the weak/strong ranking can be overruled by the interface constraint. Since I have no conclusive arguments for deciding whether D-PRONOUN is indeed an LF- or rather a PF-constraint, I will leave both options open for the moment.

44 Chapter 1: Derivations (MP) and Evaluations (OT)

into languages that do and languages that do not allow pronoun shift; cf.Figure 11. This shows that the constraints we have introduced so far can successfully account for the division postulated by the clauses in (36ci-iii). Observe that the ranking of D-PRONOUN and *MOVE is immaterial for the full object shift languages, since movement of the pronoun is already forced by the strong ranking of EPP(case).

*MOVE >> EPP(case)

No full object shift

EPP(case) >> *MOVE Full object shift: Icelandic

D-PRONOUN >>*MOVE

*MOVE >>D-PRONOUN

Pronoun shift: Danish

No object shift: Romance

Figure 11: Macro-parameterization of languages w.r.t. object shift

The semantic conditions on the application of object shift in Icelandic can also be taken care of by means of an interface constraint. As we have seen in (9), repeated here as (41), object shift is normally obligatory in Icelandic, but blocked when the object is part of the focus (new information) of the clause. (41)

a. b.

Jón keypti ekki bókina. Jón bought not the book Jón keypti bókinai ekki ti

bókina ⊂ focus bókina ⊂ presupposition

When we adopt the constraint ALIGNFOCUS in (42a) from Costa (1998) and rank it above EPP(case), we will derive the desired result. The ranking in (42b) correctly predicts that all object DPs must undergo object shift, unless they are part of the focus of the clause: object shift of a nonpresuppositional object across some phonetically realized constituent is excluded. 11 The evaluations of the two examples in (41) are given in Tableaux 10 and 11, where ⊂ and ⊄ stand for, respectively, “is part of” and “is not part of” . 11 Note that the notion of prosodically unmarked focus refers to the new information of the clause and stands in opposition to the notion of presupposition, the old information of the clause. The notion of prosodically unmarked focus must certainly not be confused with the notion of exhaustive or contrastive focus, which is always marked by means of a marked accent and may trigger the previously mentioned focus movement into SpecFocusP.

The derivation-and-evaluation framework 45 (42)

a. b.

ALIGNFOCUS (AF):

the prosodically unmarked focus is the rightmost constituent in its clause. Icelandic: ALIGNFOCUS >> EPP(case) >> *MOVE

Tableau 10: Icelandic

object ⊄ focus Jón keypti ekki bókina Jón keypti bókinai ekki ti

AF

EPP(case)

*MOVE

*! )

*

Tableau 11: Icelandic

object ⊂ focus Jón keypti ekki bókina Jón keypti bókinai ekki ti

AF

)

EPP(case)

*MOVE

* *!

*

Note in passing that ALIGNFOCUS is a so-called gradient constraint. This means that it is interpreted such that each constituent following the focus of the clause results in a violation of this constraint; if the focus is followed by two constituents this gives rise to two violations of ALIGNFOCUS. This will become important later in this study. The discussion above has shown that we can appeal to the interface constraint D-PRONOUN in (43a) to account for the fact that some languages that do not have full object shift do exhibit pronoun shift. By introducing the LF constraint ALIGNFOCUS, on the other hand, we are able to account for the fact that object shift is sometimes blocked in languages that normally do have full object shift. Later in this study we will come across another LF constraint, viz. SCOPE in (43c). (43)

LF constraints: a. D-PRONOUN: a weak/definite pronoun must be vP-external: *[vP ... pron[+def] ...]. b. ALIGNFOCUS: the prosodically unmarked focus is the rightmost constituent in its clause. c. SCOPE: relative scope of quantifiers corresponds to the hierarchical order of their topmost A-positions.

By introducing the constraints in (43a&b), we have captured almost the same range of data as (36); the only thing that we have not accounted for yet is that object shift cannot cross v*P-internal material. For example, if the main verb is non-finite and thus appears v*P-internally, object shift is excluded. This was illustrated in (10), repeated here as (44).

46 Chapter 1: Derivations (MP) and Evaluations (OT) (44)

Jón hefur ekki keypt bókina. bókina ⊂ focus or presupposition Jón has not bought the book b. *Jón hefur bókina ekki keypt tbókina a.

In order to account for this, I will appeal to a number of PF constraints involving linearization (other PF constraint were already introduced in section 1.1.2, viz. LEFT EDGE (CP) and TELEGRAPH). Since these constraints effectively require that the underlying order of heads and arguments be maintained in the surface realization, I will refer to these as order preservation constraints. The constraints that are adopted in the study are given in (45). (45)

Order preservation (PF) constraints: a. RELATIVIZED MINIMALITY (RELMIN): if the foot of X-chain Į c-commands the foot of X-chain ȕ, the head of Į c-commands the head of ȕ (X-chain = A-, A′- or head chain). b. HEAD-COMPLEMENT ( H-COMPL): a head precedes all terminals originally dominated by its complement.

The constraint RELMIN in (45a) is of course a direct descendant of the most influential “order preservation” principle from the earlier P&P period, Rizzi’s (1990) Relativized Minimality, and is rephrased here as a constraint on the output of the generator; other proposals that have similar constraints/principles are, e.g., Williams (2003), Müller (2000;2001), Fox & Pesetsky (2005) and Koeneman (2006). The formulation of RELMIN in (45b) is modeled after the formulation of the constraint ORDER PRESERVATION in Engels & Vikner (2006). Although this will not be illustrated here, this constraint plays an important role in prohibiting object shift of a direct object across an indirect object (cf. Section 3.2.2 for extensive discussion).12 Since section 2.3 will argue in favor of Kayne’s (1994) conjecture that all languages have the underlying head-complement order, the constraint H-COMPL in (45b) can also be construed as an order preservation constraint. It disfavors object shift across the main verb because this results in a surface order that differs from the underlying order. Consequently, by assuming that H-COMPL outranks EPP(case) in Icelandic, object shift will be 12 The proposal here differs considerably from the one in Broekhuis (2000), where I still entertained the belief that the effects of RELMIN in (45a) could be derived by appealing to Chomsky’s (1995a: ch.3) notion of equidistance, which was originally proposed to capture the effects of (45b). As I will show in Chapter 3, there are both theoretical and empirical reasons to drop all hope that this is the case. See also Williams (2003:16-9) for useful discussion.

The derivation-and-evaluation framework 47

blocked in examples like (44); the evaluation of these examples is given in Tableau 12. Note that the relative ranking of H-COMPL and ALIGNFOCUS cannot be determined on the basis of the present set of data, since object shift in (44) will be blocked irrespective of whether the object belongs to the focus of the clause or not, that is, whether the star between parentheses is present or not. (46)

Icelandic: H-COMPL ALIGNFOCUS >> EPP(case) >> *MOVE

Tableau 12: Icelandic (no object shift in complex tense constructions)

Jón hefur ekki keypt bókina Jón hefur bókinai ekki keypt ti

HCOMPL

AF

*!

(*!)

EPP

*MOVE

(case) *

)

*

In order to avoid confusion, note that H-COMPL must not be confused with the alignment constraint HEAD-LEFT that can be found in much OT-work (e.g., Grimshaw 1997). This constraint also requires a head to precede its complement, but competes with its counterpart HEAD-RIGHT. Alignment constraints like these play a prominent role in OT-syntax, and have generated a lot of new insights; they have been employed by, e.g., Legendre (2000) in her account of the linearization of clitics in the Bulgarian clitic cluster, by Anderson (2000) to account for verb second and other second position phenomena, and by Sells (2001) for describing Swedish object shift. These alignment constraints do not figure in this study, and differ from H-COMPL in that they do not appeal to an underlying word order but directly express certain surface word order generalizations; I refer to section 3.1.3 for a brief discussion of Sells’ proposal. By introducing the interface constraints D-PRONOUN, ALIGNFOCUS, and H-COMPL, the present proposal accounts for the same range of facts as the set of claims in (36). There are, however, a number of reasons to prefer the present constraint approach to an approach that takes appeals to output filters. First, filters tend to take the form of ad hoc stipulations that simply reformulate descriptive generalizations or the description of certain states of affairs in a semi-formal language, and it is not obvious that this will lead to any deeper insights. The constraint approach, on the other hand, derives these generalizations from more primitive notions of the theory. Secondly, the constraint approach (but not the filter approach) makes very precise predictions about what types of natural languages are possible. The postulation of H-COMPL, for example, predicts that there are also

48 Chapter 1: Derivations (MP) and Evaluations (OT)

languages in which EPP(case) outranks H-COMPL, and which therefore allow object shift across the verb. This might be the case in one of the many OVlanguages in the world. Section 2.4 will argue that languages like Dutch and German do have some form of object shift across the main verb However, this movement differs from object shift in Icelandic in that it is not forced by the strong ranking EPP(case) >> *MOVE, but triggered by the ij-features on the verb; Dutch and German have the strong ranking EPP(ij) >> *MOVE. Thirdly, the constraint approach (but not the filter approach) provides us with a general format for approaching other word order phenomena, as well as the means to express directly that certain movements (or the lack thereof) are motivated by properties of the interpretive (articulatory-perceptual and conceptual-intentional) components of the grammar. Examples illustrating this will be given in Chapter 5 where I will discuss semantic and phonological restrictions on several types of predicate movement. 1.3.3.

Input and output of CHL

The two previous subsections have extensively discussed the properties of the generator and the evaluator. This subsection will briefly discuss some properties of the input and the output of the generator. The D&E claim that the generator is some version of CHL places severe restrictions on the input: the only elements that can be processed by this system are lexical items that can be merged into larger structures. Further, we have followed Chomsky’s (1995a:ch.3) suggestion that the representations in the candidate set consist of converging derivations only, that is, structures that satisfy Full Interpretation: all features in the output must receive an interpretation in the articulatory-perceptual or the conceptual-intentional system. Adopting Full Interpretation, of course, only makes sense when we assume some version of the Lexical Integrity Hypothesis (Di Sciullo & Williams 1987): for example, it seems necessary to assume that interpretable features can be neither deleted from nor added to a lexical item during the derivation. This seems to hold in full force for the semantic features, and it holds for the phonological features at least insofar as these features are not recoverable. We will see shortly that the fact that the restriction on the phonological features is less strict than that on the semantic features may be due to the fact that the phonological output interacts directly with the articulatory-perceptual system, where ease of production and perception play an important role. The adoption of both Full Interpretation and Lexical Integrity Hypothesis implies that there are no faithfulness constraints in syntax. So there are no syntactic counterparts of the phonological constraints MAX(X) and DEP(X), which disfavor deletion and insertion by requiring, respectively,

The derivation-and-evaluation framework 49

that (i) all elements in the input be also present in the output, and (ii) all elements in the output be also present in the input. In a sense, one could say MAX(X) and DEP(X) are satisfied in syntax by definition. This makes the D&E framework crucially different from OT-phonology, where the interaction of faithfulness constraints and the so-called markedness constraints, which place additional restrictions on the output of the generator, constitute the core of the system. The cause of this difference may lie in the fact that the phonological module directly interacts with the articulatory-perceptual system: the phonological markedness constraints normally favor outputs that are easier to produce, whereas the faithfulness constraints seem to counterbalance this by favoring outputs that facilitate the ease of perception. Since syntax is independent of the performance systems, markedness and faithfulness constraints are less likely to occur. For completeness’ sake, note that Full Interpretation and the Lexical Integrity Hypothesis can be seen as a reformulation of the classical idea of the autonomy of syntax, as well as of Frege’s principle of compositionality of meaning.13 Given the three assumptions above (Generator=CHL, Full Interpretation, and Lexical Integrity), the set of possible hypotheses concerning the nature of the input and the output is restricted even further, and seems to be exhausted by the following two options (or a combination thereof, which can be dismissed as least favored). (47)

a. b.

There are additional restrictions on the input, and the output (candidate set) is fully determined by the input. There are additional restrictions on the output (candidate set), and the generator has free access to the lexicon.

13 Of course, the conclusion that there are no faithfulness constraints in syntax follows from the fact that we are implementing the minimalist framework in D&E terms; it does not follow if we were to implement, e.g., Williams’ (2003) Representation Theory, where economy is expressed in terms of isomorphism between different levels of representation (like Theta, Case and Surface structure). This was already pointed out by Williams (2003:38) himself: “Since mismatches [between representation levels] are allowed […], the only question is whether there is a systematic way to resolve these conflicts. I will suggest that languages differ with respect to which representation relations are favored. This arrangement is somewhat like Optimality Theory, if we identify ‘shape-conserving representation relation’ with ‘faithfulness’.” Taking up this idea, Müller (2001) takes Williams’ shape-conserving principle to be a syntactic faithfulness constraint, whereas the constraints that favor movement and thus may cause disruption of the base order are considered markedness constraints. This view, though highly interesting, is not compatible with the proposal here, according to which the input is simply a set of lexical items.

50 Chapter 1: Derivations (MP) and Evaluations (OT)

The option in (47a) is adopted in MP, where it is assumed that the input is a numeration. The option in (47b) is adopted by Grimshaw (1997), who claims that the candidates in a candidate set have the same meaning, where meaning is assumed to include notions from information structure, such as the distinction between the focus and the presupposition (new and old information) of the clause. The difference between these two proposals is subtle but clear: hypothesis (47a) predicts that candidate A can never be blocked by candidate B when A and B contain different lexical items; hypothesis (47b) predicts that this is possible provided that A and B have the same meaning. In order to see which proposal is correct, let us consider the arguments that have been put forward in favor of each of them. 1.3.3.1. Is the input a numeration? Subject Raising constructions like (48) have been put forward as evidence in favor of the MP claim in (47a), and the intimately related assumption that Select (external merge) is inherently less “costly” than Move (internal merge); note that, for convenience, I will use here the terminology as introduced above, instead of the terminology that Chomsky (1995a: section 4.9) uses in his discussion of these facts. (48)

a. Therei is likely [IP ti to be someone here at 6:00] a′. *There is likely [IP someonei to be ti here at 6:00] b. Someonei is likely [IP ti to be ti here at 6:00]

Examples (48a) and (48a′) differ in that in the first case the expletive there is merged in the subject position of the infinitival clause in order to satisfy the EPP in the embedded clause, and subsequently moved into the subject position of the matrix clause in order to satisfy the EPP in the matrix clause (note that the notion of EPP is used here descriptively and refers to the fact that in English SpecIP must be filled). In (48a′), on the other hand, the subject someone has first moved into the embedded subject position in order to satisfy the EPP in the infinitival clause, and the expletive is subsequently merged in the subject position of the matrix clause in order to satisfy the EPP in the finite clause. This derivation is excluded despite the fact that movement of someone into the subject position of the infinitival clause is possible when no expletive is present. Of course, this cannot be observed directly from the phonetic output of (48b), but it is plausible given the fact that this movement must also apply in ECM-constructions, as is illustrated in (49b&b′).

The derivation-and-evaluation framework 51 (49)

a. John wants there to be someone here at 6:00. b. John wants someonei to be ti here at 6:00. b′. *John wants e to be someone here at 6:00.

Chomsky’s account of the data in (48a&a′) is very simple. At the stage in the derivation where the subject position of the infinitival clause must be filled, there are two options: SpecIP is filled either by Select, that is, external merge of the expletive there, or by movement of the NP someone. Chomsky claims that the choice between these two options must be made locally (in the sense of Collins 1997), so that the putative fact that Select is inherently less costly than Move forces the application of the former: this rules out the derivation that leads to (48a′). In (48b), the Select option is not available, since the numeration does not contain an expletive, and movement of the NP someone is the only option to satisfy the EPP. If the examples in (48a&b) indeed have the same meaning, the fact that they are both possible seem to favor the MP hypothesis in (47a) and the concomitant claim that Select is inherently less costly than Move over the alternative hypothesis in (47b). However, there is an alternative analysis available that does not appeal to the relative cost of Select and Merge, and that can actually be applied to a wider range of constructions. In this analysis, the expletive there is considered a small clause predicate that is moved into SpecIP by means of (obligatory) predicate inversion; cf. Moro 1997, Hoekstra & Mulder 1990, Den Dikken & Næss 1993, among others. According to the predicate inversion analysis, the examples in (48) are derived as follows. In (48a), the expletive there is moved by predicate inversion into the subject position of the embedded clause in order to satisfy the EPP in the embedded clause. After this, only there can be moved into the subject position of the matrix clause, because moving the subject someone into the matrix subject position results in a violation of the locality conditions on Move. The derivation of (48a) would therefore be as indicated in (50a). Example (48a′) cannot be derived, because after the subject someone has moved into the subject position of the embedded clause, movement of there into the subject position of the matrix clause would violate the locality conditions on Move (cf. Moro 1997:120). The derivation of (48b) does not involve predicate inversion, and is given as (50b). Note that the locative predicate here is assumed to be an adjunct in the expletive construction, and not the actual predicate of the small clause. (50)

a. Therei is likely [IP ti to be [SC someone ti] here at 6:00] a′. *Therei is likely [IP someonej to be [SC tj ti] here at 6:00] b. Someonei is likely [IP ti to be [SC ti here] at 6:00]

52 Chapter 1: Derivations (MP) and Evaluations (OT)

Independent evidence in favor of this predicate inversion analysis of the expletive construction can be found in (51). In (51a), the directional predicate down the hill has been moved into the subject position of the matrix clause via the subject position of the embedded clause, and the result is acceptable, which indicates that the locational predicate is able to satisfy the EPP, just like there (cf. Bresnan 1994 for similar data). In (51a′), on the other hand, movement of the predicate into the subject position is blocked by the locality conditions on Move. The derivation in (51b), of course, satisfies all the locality conditions on movement and is therefore licit. In other words, the analysis of the examples in (51) is essentially identical to the analysis of the expletive constructions in (50). (51)

a. Down the hilli seems [IP ti to roll [SC a baby carriage ti]] a′. *Down the hilli seems [IP a baby carriagej to roll [SC tj ti]] b. A baby carriagei seems [IP ti to roll [SC ti down the hill]]

If the predicate inversion analysis of the expletive constructions in (47) is tenable, this voids Chomsky’s evidence in favor of the two claims he has made: neither hypothesis (47a) nor the claim that Select is inherently less costly than Move receives independent empirical support. Actually there is evidence that both claims are wrong. This will be discussed in the next subsection. A more detailed analysis of predicate inversion (which adopts slightly different background assumptions) can be found in Chapter 5. 1.3.3.2. Is the output semantically restricted? Broekhuis & Klooster (2001;to appear) have argued that the notion of a numeration should be eliminated from the theory and be replaced by Grimshaw’s claim in (47b) that the candidates in the candidate set have the same meaning. Although he does not take the step to abandon the notion of a numeration (see below), essentially the same intuition is explored by Hornstein (2001: section 5.5) in order to account for do-support and the distribution of anaphors and bound pronouns. This is especially clear from his footnote 74 on p. 202. Broekhuis & Klooster argue further that Select is not inherently less costly than Move. They claim that these operations are disfavored by the economy constraints *SELECT and *MOVE, respectively, and that languages may differ in the ranking of these two constraints. This predicts that in some languages Move will be preferred over Select, and this subsection will show that this prediction is indeed borne out. But first allow me a brief digression. Although Hornstein (2001) follows the more traditional view that Move is more costly than Merge given that the

The derivation-and-evaluation framework 53

latter operation is a constituting part of the former (Move = Copy + Merge), he provides important conceptual reasons in favor of the proposal defended here that there is no inherent difference in the cost of these operations. He argues that “selecting atomic units from the lexicon for computational manipulation by the syntax involves the Copy operation” (p.211). This implies that the two operations Select and Move postulated in the present study both consist of the more primitive operations Copy and Merge — they differ only in that in the former case the merged element is copied from the lexicon, whereas in the latter case it is copied from the structure it is merged with. Interestingly, Hornstein ends his section 6.1 with a footnote in which he takes up this line of reasoning and concludes that “the main reason for postulating an array (=numeration) rests on empirical evidence [...] that economy plays a role in derivations” (p.223). I have already argued above that the evidence given in favor of the claim that Move is less costly than Select is rather feeble, and the goal of this section is to present new evidence that strongly suggests that Select is not inherently less costly than Move. Now, consider the examples in (52) and (53), which show that sentence negation can be expressed either by means of a negative NP, as in the Dutch example in (52), or by means of a NPI preceded by the negative adverb not, as in the English example in (53). (52)

a.

Jan is over niemand tevreden. Jan is about nobody satisfied b. *Jan is niet over ook maar iemand tevreden. Jan is not about anybody satisfied

(53)

a. *John is satisfied with nobody. b. John is not satisfied with anybody.

According to the MP thesis in (47a), according to which the generator takes a numeration as its input, the derivation of the grammatical Dutch example in (52a) has a numeration that contains at least a negative NP and the functional head Neg associated with it: cf. (54a). The derivation of the English sentence in (53b), on the other hand, has a numeration that contains at least a NPI, the negative adverb not and the functional head Neg associated with it; cf. (54b). (54)

a. b.

{..., Neg, DP[+neg], ...} {..., Neg, niet/not, NPI, ...}

The acceptability of both (52a) and (53b) shows that the abstract numerations in (54) can both give rise to an acceptable negative sentence. This is of course consistent with the idea that the unvalued feature of the

54 Chapter 1: Derivations (MP) and Evaluations (OT)

Neg-head can be valued by either an internal or an external syntactic object, that is, either by Move or by Select (cf. Chomsky’s analysis of the expletive construction discussed above). In the derivation that takes (54a) as its input, the unvalued feature of the Neg-head is valued by moving the negative DP into SpecNegP, as in (55a).14 In the derivation that takes (54b) as its input, the unvalued feature can be checked by external merge of the negative adverb in SpecNegP, as in (55b). (55)

a. b.

Jan is [NegP over niemandi [Neg′ Neg ... [AP tevreden ti ]]] John is [NegP not [Neg′ Neg ... [AP satisfied with anybody]]]

Let us now discuss what the examples in (52) and (53) tell us about the two hypotheses in (47). First note that since MP does not impose special requirements on the numeration, there is no obvious way of forcing Dutch to take the numeration in (54a) rather than (54b), or to force English to take (54b) rather than (54a): (54a) and (54b) should therefore be freely available in both languages. Secondly, given that the acceptability of (52a) and (53b) shows that both numerations in (54) give rise to a converging derivation, the ungrammatical examples in (52b) and (53a) cannot be ruled out by claiming that they are the result of a crashing derivation. Consequently, the obvious way to exclude these ungrammatical examples is to assume that they are somehow blocked by their grammatical counterparts in (52a) and (53b). However, this is predicted not to be possible by hypothesis (47a): since examples that are based on different numerations cannot be part of the same candidate set, the a- and b-examples in (52) and (53) do not compete, and we therefore wrongly predict that all examples in (52) and (53) are acceptable. According to the alternative hypothesis in (47b), the a- and b-examples in (52) and (53) do compete because they have the same meaning. The meaning of the a-examples can be expressed as in (56a). Hornstein (1984) has argued that a NPI like ook maar iemand or anybody can be represented as a universal operator with wide scope (i.e., with scope over the other operators in the clause). 15 This implies that the b-examples can be 14

Recall that section 1.3.2.1 suggested that movement of the goals of [+affect] features like [wh], [topic], [focus], or [neg] is assumed to be forced by semantic considerations, so that the negative DP cannot remain in situ. If the negative DP does not move, this gives rise to constituent negation (cf. Haegeman 1995 for West Flemish and Broekhuis & Klooster 2001 for Dutch). See also section 4.3.2. 15 Zanuttini (1991:116) has pointed out that universal quantifiers like everybody, all and everywhere can be modified by means of expressions like almost and just about, while existential quantifiers like some and any cannot. That would make ook maar iemand an existential quantifier. The actual choice is innocuous in the limited contexts discussed here, in view of the logical equivalence mentioned in the main text.

The derivation-and-evaluation framework 55

semantically represented as in (56b). Since the formulas in (56a) and (56b) are semantically equivalent (¬∃xF(x) ļ ∀x¬F(x)), the hypothesis in (47b) predicts that the a- and b-examples are part of the same candidate set, and hence compete with each other. (56)

a. b.

¬∃x (x:person) (Jan is satisfied about x) ∀x (x:person) ¬(Jan is satisfied about x)

In order to account for the examples in (52) and (53), we can take recourse to the two economy constraints in (31a&b), repeated here as (57), which block derivations with superfluous applications of Select and Move, respectively. (57)

a. b.

*SELECT: don’t apply external merge. *MOVE: don’t apply internal merge.

The contrast between the Dutch and English data can now be accounted for by assuming different rankings of these constraints. Given that in Dutch, sentence negation is expressed by means of moving a negative DP into SpecNegP, we must conclude that Dutch prefers movement over the application of Select: the Dutch ranking is therefore as given in (58a). English, on the other hand, prefers the use of a NPI and external merge of the negative adverb, so that we must assume that in this language the application of Select is preferred over Move; the English ranking is therefore as in (58b). The evaluations of the examples like (52) and (53) are therefore as given in Tableaux 13 and 14. Note that the violations of *SELECT and *MOVE that are shared by both candidates are not indicated. (58)

a. b.

Dutch: *SELECT >> *MOVE English: *MOVE >> *SELECT

Tableau 13: Negative sentences in Dutch

*SELECT ... [NegP niemandi [ Neg ... ti ... ]] ... [NegP niet [ Neg ... NPI ...]]

)

*MOVE *

*!

Tableau 14: Negative sentences in English

... [NegP nobodyi [ Neg ... ti ... ]] ... [NegP not [ Neg ... NPI ...]]

*MOVE *! )

*SELECT *

56 Chapter 1: Derivations (MP) and Evaluations (OT)

1.3.3.3. Conclusion This section has shown that hypothesis (47a), according to which the numeration determines which candidates are part of the candidate set, is not tenable in the light of examples like (52) and (53). These examples do support the alternative hypothesis in (47b), according to which the candidates in the candidate set have the same meaning. The discussion has also shown that the operations Select and Move are both costly, so that we have to postulate at least the economy constraints *SELECT and *MOVE. As a result, the differences between Dutch and English can be accounted for by assuming that the ranking of these constraints differs in the two languages. The number of facts discussed in this subsection is of course very small, and I refer to Broekhuis & Klooster (2001;to appear) for a more extensive discussion of the relevant Dutch and English data. Independent support for hypothesis (47b) can be found in Broekhuis & Dekkers (2000), but since this hypothesis will not play a prominent role in this study, I will not discuss this evidence here. 1.4. Summary This chapter has provided an updated version of the derivation-andevaluation (D&E) framework originally proposed in Broekhuis & Dekkers (2000) and Broekhuis (2000). The leading idea of the framework is that, in order to arrive at a descriptively and explanatorily adequate theory, restrictions must be placed both on the syntactic derivation and on the resulting syntactic representations. This has been given shape by assuming a framework in which aspects of the minimalist program (MP) and optimality theory (OT) are combined. More specifically, it was claimed that representations created by some version of the computational system of human language CHL from MP are evaluated in an optimality-theoretic fashion, as indicated in Figure 12. Input

CHL

Output representations

OTEvaluator

Optimal output

Figure 12: The derivation-and-evaluation model

In MP and OT-syntax the explanatory burden is normally placed on the generator and the evaluator, respectively. By placing the explanatory burden on both systems, these systems cannot be developed independently in the D&E framework: properties ascribed to the one may have farreaching effects on the format of the other. The following table summarizes

Summary 57

the central claims of D&E that we have discussed above, and compares these to those normally adopted in MP and OT-syntax. Derivation-and-evaluation model I. The generator is some version of CHL a all operations are subject to inviolable conditions b all operations are subject to the Last Resort Condition c the generator is autonomous and operations apply at random; there are no EPP-features. II. The evaluator consists of a ranked set of syntactic constraints a the syntactic constraints are taken from a universal set CON b the number of syntactic constraints in CON is small c the number of syntactic constraint types in CON is small III. The input and output a lexical items from the input are selected directly from the lexicon without the intervention of a numeration b all candidates in the candidate set share the same meaning

MP + + + —

OT — ? — +

— n.a. n.a. n.a.

+ + — —

—

?

+

+

It seems to me that MP and D&E have a descriptive apparatus of more or less the same size, and are facing a similar task in that they both have to identify the features that may trigger movement. The frameworks mainly differ in that they provide different answers to the question of what determines whether the movements that are allowed by the Last Resort Condition actually do take place in a given language L. In MP it is commonly assumed that movement is forced by the presence of an EPPfeature, and since certain movements, like Icelandic object shift, only apply under certain well-defined conditions, the question is raised what determines the distribution of these EPP-features. Chomsky (2001) claims that the distribution of the EPP-features is determined by certain “parameters” that take the form of language-specific output filters. In D&E the answer takes the form of an optimality-theoretic evaluation, as indicated in (33) and (37), repeated here as (59). (59)

a. b. c. d.

Weak ranking: *MOVE >> EPP(F) Strong ranking: EPP(F) >> *MOVE A >> *MOVE >> EPP(F) Ÿ if A favors movement, “Procrastinate” is overruled. B >> EPP(F) >> *MOVE Ÿ if B disfavors movement, “Strength” is overruled.

58 Chapter 1: Derivations (MP) and Evaluations (OT)

The constraint rankings in (59a&b) determine whether a certain formal feature F “normally” does or does not trigger movement of its goal, or, phrased in MP terms, does or does not have an EPP-feature. The constraints A and B in (59c&d) determine whether the “normal” case can be overridden, and thus correspond to the output filters postulated in Chomsky (2001). The advantage of adopting an OT-evaluation is, however, that this enables us to reduce the filters to more primitive notions of the theory. This may provide us with new and deeper insights into how Universal Grammar allows us to exploit, e.g., the displacement property of language for the more “functional” purpose of expressing the information structure of the clause. In a sense, OT also provides us with a restrictive theory on the format of “filters”, which enables us to make very precise predictions about the ways in which languages can differ (provided, of course, that we have established the syntactic constraints that are part of the universal constraint set CON). D&E differs from OT-syntax in that the former postulates the computation system CHL from MP as the generator. As a result of this, many imaginable derivations are blocked by the inviolable conditions on the operations of CHL, so that the number of candidates in the candidate set is very restricted, and the candidates in this set can differ in well-defined manners only. This has led to the conjecture that there are not only a limited number of syntactic constraints, but also a limited number of constraint types. In order to establish these types, I have assumed that the evaluator is actually a hypothesis about the interface conditions postulated in MP and, consequently, that the constraints fall into the two main classes in (60a-b). The CHL constraints can be further subdivided into two families of constraints, viz. the economy constraints that disfavor the operation of CHL to apply, and the EPP constraints that favor them to apply: the ranking of these constraints determines whether a certain operation normally does or does not take place. The interface (PF and LF) constraints seem to be more varied in nature, and it is still an open (empirical) question how many there actually are. The constraints that will be used in our discussion of object shift in Chapter 2 to Chapter 4 are given in (60). (60)

The syntactic constraints in CON are of three basic types: a. CHL constraints: (i) Economy constraints: *MOVE, NOLEXM, NOVACM (ii) EPP(F): probe F attracts its goal b. Interface constraints: (i) PF constraints: RELMIN, H-COMPL (ii) LF constraints: ALIGNFOCUS, D-PRONOUN

Summary 59

Note that I did not spell out the EPP(F) constraints, because we will consider several proposals: in the end it will turn out that we only need (i)EPP(gender), EPP(person), and EPP(case), which favor object and subject shift, and (ii) *STRAY FEATURE, which favors verb movement into the “verbal” functional heads in its extended projection and thus unifies conceivable constraints like EPP(v), EPP(Asp), EPP(tense). In a sense, the D&E framework directly descends from the model proposed in Chomsky & Lasnik’s (1977) Filters and Control in postulating two independent systems for generating and evaluating syntactic structures. Chomsky & Lasnik left open the option that the periphery (the evaluative component) uses “much richer resources, perhaps resources as rich as contemplated in the earlier theories of TG”, However, our hope should be that this will turn out not to be the case, and that the rules of the periphery will also be largely determined by our genetic endowment, that is, by the innate and thus universal constraint set CON. From the D&E perspective, Chomsky & Lasnik’s use of the notions “core” and “periphery” for the generator and the evaluator, respectively, is therefore misleading: the generator and the evaluator together constitute core syntax, the periphery instead rather lies outside these systems and should refer to everything that must be learned on an item-to-item or construction-to-construction basis. From now on, I will use the notions of core and periphery in this sense, while noting that it is of course not a priori clear what empirical phenomena are part of the core, and which belong to the periphery; much will depend on the descriptive success of the analyses developed within this framework. I hope, however, that the discussions in the following chapters will convince the reader that many of the phenomena that have been removed from the core in Chomsky’s recent work (e.g., head movement) can again be fruitfully investigated within the boundaries of core syntax as defined in Figure 13.

Periphery Input

CHL

output

OTEvaluator

Core syntax Figure 13: Core and periphery in syntax

Optimal output

60 Chapter 1: Derivations (MP) and Evaluations (OT)

To conclude, I want to note that, in some places, Chomsky & Lasnik (1977) seemed to have had a model in mind similar to that in Figure 13, given that they “think of theory of grammar T as consisting of two parts: a universal grammar UG that determines the class of potential grammars and the way they operate, and a system of evaluation that ranks potential grammars in terms of ‘optionality’ or ‘simplicity’” (Chomsky & Lasnik 1977:44). This seems a very apt description of the D&E framework.

Chapter 2 Short object shift

This chapter will develop two hypotheses concerning object shift and scrambling. The first hypothesis will be discussed in section 2.1 and contends that object shift and scrambling are of the same type: it is Amovement triggered by the unvalued nominal features on the verb. The second hypothesis will be discussed in section 2.2 and contends that there are two subtypes of object shift: the first type is A-movement triggered by the ij-features of the verbal root V, and the second type is A-movement triggered by the case features on the light verb v. Consequently, the object can in principle occupy the three A-positions indicated in (1): OBJ1 is the ș-position of the object, and the ij-features on V and the case feature on v may trigger movement of the object into position OBJ2 and OBJ3, respectively. (1)

[OBJ3 ... v [ OBJ2 ... V OBJ1]]

The structure in (1) adopts Kayne’s (1994) version of the universal base hypothesis according to which all languages have the same underlying specifier-head-complement order, which implies that all languages are SVO underlyingly. However, there are competing proposals that claim that assuming an underlying SOV order gives rise to a simpler grammar (e.g., Haider 2000 and Barbiers 2000). Section 2.3 will discuss these competing hypotheses, and argue that the SVO hypothesis is actually the correct one. Section 2.4 will conclude this chapter by providing a more detailed analysis of “short” object shift into position OBJ2. The movement of the object into position OBJ3 will be the main topic of Chapter 3. 2.1. Object shift/scrambling is A-movement The previous chapter has already introduced the main topic of this study, object shift. Object shift refers to the phenomenon that under certain circumstances DP-complements can be moved leftward across adverbial phrases (and, as we will see, sometimes also across other phonetically realized material). This phenomenon was illustrated by means of the Icelandic examples in (2), but similar examples can be given from languages like German and Dutch where this phenomenon is normally

62 Chapter 2: Short object shift

referred to as scrambling. Since the object movements are subject to similar semantic restrictions in the two languages, the null hypothesis is that object shift and scrambling are actually instantiations of the same movement type. This section will argue that object shift and scrambling are instances of A-movement, and will discuss the arguments that have been used to support and to contest this claim. (2)

a. b.

(3)

a. b.

2.1.1.

Jón keypti ekki bókina. Jón bought not the book Jón keypti bókinai ekki ti Jan koopt waarschijnlijk het boek. Jan buys probably the book Jan koopt het boeki waarschijnlijk ti

bókina ⊂ focus bókina ⊂ presupposition het boek ⊂ focus het boek ⊂ presupposition

Arguments supporting the hypothesis

Vikner (1994;2006) has argued that Scandinavian object shift is an instance of A-movement, that is, movement triggered by the case and/or the ij-features on the verb, and, as we have seen in Chapter 1, this claim was also adopted by Chomsky (2001) in his account of Icelandic object shift. The main argument in support of this claim is that it immediately accounts for the fact that PPs and DP-complements of PPs do not undergo this movement: PPs cannot be probed by the unvalued features on the verb because they do not have case or ij-features, and the DP-complements of PPs cannot be probed because they are no longer active at the relevant stage of the derivation due to the fact that their case feature is valued PPinternally. This is illustrated for Icelandic by the examples in (4), taken from Jónsson (1996:69). (4)

• No shift of PPs or complements of PP a. Jón las ekki (DO) Jón read this book not b. Ég talaði ekki (*PP-complement) I talked to Jón not c. Ég talaði ekki við (*DP-complement of P) I talked Jón not to

If Scandinavian object shift is indeed A-movement, we expect the same to be true for Dutch/German scrambling. There is, however, no consensus on this issue. The existing approaches to scrambling of objects can be

Object shift/scrambling is A-movement 63

divided into three different groups depending on whether it is considered to be A- or A′-movement, or to involve base-generation; a representative sample of these approaches can be found in Corver and Van Riemsdijk (1994). Webelhuth (1989;1992) has shown that Dutch/German object scrambling has properties of both A- and A′-movement, a fact that is often referred to as Webelhuth’s Paradox. For example, the fact that scrambling feeds binding is a typical A-movement property (cf. Vanden Wyngaerd 1988;1989b), whereas the fact that scrambling licenses parasitic gaps is generally considered an A′-movement property (cf. Bennis & Hoekstra 1984; see De Hoop & Kosmeijer 1991, and Broekhuis 1992, however, for some problems for this claim). The binding facts are illustrated in (5); note that example (5a) improves slightly if the adverbial phrase namens elkaar ‘on behalf of each other’ is assigned contrastive accent, which might be due to the fact that (5a) might then be derived from (5b) by means of reconstructable focus movement (which will be discussed shortly). The parasitic gap facts are illustrated in (6). (5)

a.

*Hij heeft namens elkaar de jongens he has on behalf of each other the boys b. Hij heeft de jongensi namens elkaar ti he has the boys on behalf of each other ‘He visited the boys on behalf of each other.’

(6)

a.

bezocht. visited bezocht. visited

*Hij heeft [zonder PRO pg te bekijken] het boek opgeborgen. he has without to look-at the book prt-filed b. Hij heeft het boeki [zonder PRO pg te bekijken] ti opgeborgen. he has the book without to look-at prt-filed ‘He filed the book without looking (at it).’

To my mind, the most plausible solution to Webelhuth’s paradox is to assume that the notion of scrambling is not a unitary phenomenon, but actually refers to (at least) two different phenomena (cf. Vanden Wyngaerd 1988;1989b; Déprez 1989; Mahajan 1990; Neeleman 1994b), which I will henceforth refer to as object shift and focus or neg-movement, respectively. The fact that the object in (7) is able to both bind the anaphor elkaar ‘each other’ and to license the parasitic gap can then be accounted for as follows: binding of the anaphor is made possible by object shift, that is, by the intermediate trace t′i, which occupies an A-position; the parasitic gap is licensed by focus movement, that is, by the DP preceding the adjunct clause, which presumably occupies the specifier position of a FocusP, that is, an A′-position.

64 Chapter 2: Short object shift (7)

Hij had de gasteni [zonder pg te bekijken] t′i aan elkaar ti voorgesteld. he had the guests without to look-at to each other introduced ‘He had introduced the guests to each other without looking (at them).’

An important difference between object shift and focus/neg-movement is that the former can be applied to nominal objects only, whereas the latter can be applied to a wider range of constituents; see the next section for some discussion. 2.1.2.

Arguments against the hypothesis

Holmberg & Platzack (1995: section 6.3.2) have argued against the claim that object shift is A-movement on the basis of the following two arguments: (i) object shift crosses the subject in its base position (under our assumptions: Specv*P), and thus violates Rizzi’s (1990) Relativized Minimality; (ii) object shift does not feed binding. A solution to the first problem was already provided by Chomsky (1995a: ch.3) by the introduction of the notion of equidistance, which amounts to saying that, under a number of strict conditions, violations of Relativized Minimality are allowed; I will return to this issue in section 3.1, and show there that Chomsky’s proposal can be readily revised such that object shift across the subject does not actually involve a violation of Relativized Minimality. The second argument was argued to be invalid by Jónsson (1996:70-73), who pointing out that reflexive binding in Icelandic is normally subject-oriented, hence objects do not count as suitable antecedents for a reflexive. According to Jónsson, reciprocals give rise to better results in similar contexts. Haider, Olsen & Vikner (1995:20-1) point out that the Danish examples in (8) also form a potential problem for the A-movement analysis of object shift. Example (8a) shows that the locational pro-form der ‘there’ undergoes some sort of object shift, despite the fact that it is not case-marked; example (8b) further shows that, just like pronoun shift, movement of der cannot cross the main verb. (8)

a.

b.

Peter sov alligevel ikke . Peter slept there after all not ‘Peter didn’t sleep there after all.’ Peter har alligevel ikke sovet . Peter has there after all not slept ‘Peter didn’t sleep there after all.’

I will postpone the discussion of the examples in (8) to Section 3.2.2.1 (p.176), where I will argue on the basis of similar examples in Dutch that

Object shift/scrambling is A-movement 65

the pro-form is moved into a position preceding the landing site of object shift. There are good empirical reasons to assume that this movement of the pro-form is the result of A′-movement (cf. Van Riemsdijk 1978: ch.5, Huybregts 1991, and Broekhuis 2002: section 1.5); this will be briefly discussed shortly on the basis of example (11) below. Although Vikner (1994;2006) extensively argues that Scandinavian object shift is A-movement, he also argues that Dutch/German scrambling differs in various respects from Scandinavian object shift, and should therefore be considered an instance of A′-movement. The differences that Vikner (2006:411) mentions are given in Table 1. Table 1: Alleged differences between object shift and Scrambling

I II III IV V VI VII

may cross the main verb/is independent of V-to-I movement may cross an indirect object may cross a preposition may cross a verbal particle may license parasitic gaps may move PP may target a position in between adverbials

Scandinavian object shift —

Dutch/German scrambling +

— — — — — —

+ + + + + +

The differences in I and II are real (although II does not hold for Dutch), and will be discussed extensively in sections 3.2.2 and 3.2.3, so let us immediately proceed to III. That Scandinavian object shift cannot cross a preposition was already illustrated in (4b) above. Vikner does not illustrate his claim that scrambling differs in this respect from object shift and can cross a preposition. However, this claim is clearly wrong. The examples in (9) show that, just as in Icelandic, it is impossible in Dutch to shift the complement of a PP leftward across the preposition. (9)

a.

b.

Hij praat niet graag over . he talks his wife not gladly about ‘He doesn’t like to talk about his wife.’ Hij praat niet graag over . he talks her not gladly about ‘He doesn’t like to talk about his wife.’

66 Chapter 2: Short object shift

Perhaps what Vikner has in mind in claiming that scrambling may cross the preposition is the phenomenon of R-extraction, which is illustrated in (10). (10)

a.

Hij praat niet graag He talks THERE not gladly ‘He doesn’t like to talk about that.’

over. about

The phenomenon of R-extraction certainly cannot be equated with scrambling and there are reasons to assume that the leftward movement of the R-word er targets a designated position for R-words (cf. Van Riemsdijk 1978). Furthermore, this position is probably an A′-position given that whmovement and topicalization of other R-words across this position are blocked when it is filled by a locative pro-form. This is shown in (11). (11)

a.

*Waari heeft hij erloc. vaak [PP ti over gepraat]. WHERE has he there often about talked Intended reading: ‘About what did he talk there often.’ b. *Hieri heeft hij erloc. vaak [PP ti over gepraat]. HERE has he there often about talked Intended reading: ‘About this he talked there often.’

Note in passing that the examples in (11) are grammatical when the clauseinitial R-word is construed as the locative pro-form, and er as part of the PP-complement, which follows given that these constructions do not involve crossing. The intricacies of the construction are extensively discussed in Huybregts (1991); unfortunately, this important paper is written in Dutch but an extensive review of the main findings can be found in Broekhuis (2002). Now that we have seen that claim III is false, let us continue with the claim in IV that object shift cannot cross verbal particles, whereas scrambling can. First consider the Scandinavian data. In languages where the particle must precede the object, object shift is excluded in verb-particle constructions: (12a) shows that the DP-objects must follow the particle in Swedish, and (12b) shows that object shift across the particle is excluded; for completeness′ sake note that the pronoun cannot be placed between negation and the particle either. (12)

a. b.

Peter har inte kastat bort . Peter has not thrown the-carpet away Peter kastade inte bort . Peter threw it not away

Object shift/scrambling is A-movement 67

However, in languages where the particle must (or may) follow the object, object shift is possible: (13a) shows that the DP-objects must precede the particle in Danish, and (13b) shows that, as usual, pronominal object shift is obligatory in verb-particle constructions. (13)

a. b.

Peter har ikke smidt ud . Peter has not thrown the-carpet away Peter smed ikke ud. Peter threw it not away

Vikner does not discuss the data that led him to the conclusion that scrambling across verbal particles is possible, and actually I do not understand what could motivate this claim: example (14a) shows that in Dutch and German, the object always precedes the particle, just as in Danish, so that Vikner’s reasoning above simply predicts that scrambling of the object is possible in verb-particle constructions. Example (14b) shows that this prediction is correct. Claim IV, therefore, does not support the hypothesis that object shift and scrambling are different phenomena. (14)

a. b. b′.

Peter heeft Peter has Peter heeft Peter has Peter heeft Peter has

weg gegooid . the carpet away thrown waarschijnlijk weggegooid. the carpet probably away-thrown waarschijnlijk weggegooid. it probably away-thrown

The fact that scrambling licenses parasitic gaps was extensively discussed in section 2.1.1, and has led to the conclusion that scrambling is actually not a unitary phenomenon, and refers to two different movement processes of, respectively, the A- and the A′-type: object shift and focus/negmovement. This also gives us a hold on the difference given in VI: leftward movement of PPs. That Scandinavian object shift of PPs is excluded was already illustrated in (4b). Scrambling of PPs is indeed possible in Dutch and German. This is illustrated in the examples in (15), taken from Neeleman (1994a). However, before we can conclude anything from these examples, it must first be shown that we are dealing with scrambling of the A-type. Below I will defend the claim that we are not, but that we are instead dealing with some sort of focus movement, that is, movement of the A′-type. (15)

a. b.

dat Jan nauwelijks op mijn opmerking reageerde. that Jan hardly on my remarks reacted dat Jan op mijn opmerking nauwelijks tPP reageerde.

68 Chapter 2: Short object shift

My claim that the movement in (15b) is some sort of focus movement is explicitly dismissed by Neeleman (1994a), who noticed that the PP in (15b) does not require contrastive focus accent. Since scrambling of PPs cannot be A-movement either, Neeleman (1994a) sets out to argue that scrambling is not movement at all: the two orders in (15) are simply base generated. Haeberli (2002) has tried to solve the problem by assuming that in (15b) we are dealing with A-movement of a more abstract type (according to his proposal: A-movement as the result of categorial feature checking). I will argue that both claims are wrong because the premise that (15b) involves a similar kind of movement as object shift is not empirically supported. The first reason to assume that the movement is (15b) is different from object shift has to do with the distribution of PPs that contain a definite pronoun. The primed examples in (16) show that scrambling is only possible when the pronoun has its non-reduced form; use of a weak pronoun is impossible in the scrambled order. This shows that the complement of the scrambled PP must be assigned stress, whereas object shift normally has the effect of destressing the moved element. This, in turn, suggests that Neeleman is actually wrong in assuming that the PP is not assigned focus accent in (15b). (16)

a. a′. b. b′.

dat Jan nauwelijks naar hem/’m luisterde. that Jan hardly to him listened dat Jan naar hem/*’m nauwelijks luisterde. dat Jan nauwelijks naar haar/’r keek. that Jan hardly at him looked dat Jan naar haar/*’r nauwelijks keek.

A second reason to assume that the movement in (15b) is of a different type than object shift is that scrambling of the PP is only possible with a restricted set of adverbial phrases. 16 When we replace the negative adverbial phrase nauwelijks in (15b) by the adverbial phrase gisteren ‘yesterday’, scrambling of the PP normally gives rise to a degraded result. This is illustrated in (17) with three different PP-complements.

16 It is not entirely clear at this moment which types of adverbial phrases do or do not allow this movement: my impression is that scrambling of PPs is mainly possible with negative adverbs like nauwelijks ‘hardly’, quantified adverbs like vaak ‘often’, and modal adverbs like waarschijnlijk ‘probably’. Since little is known about the conditions under which PP scrambling is (im)possible, it is clear that this issue requires more research in the future.

Object shift/scrambling is A-movement 69 (17)

a. a′. b. b′. c. c′.

Jan heeft nauwelijks/gisteren op mijn opmerkingen gereageerd. Jan has hardly/yesterday on my remarks reacted Jan heeft op mijn opmerkingen nauwelijks/*gisteren gereageerd. Jan heeft nauwelijks/gisteren naar Marie gekeken. Jan has hardly/yesterday at Marie looked Jan heeft naar Marie nauwelijks/*gisteren gekeken. Jan heeft gisteren op vader gewacht. Jan has yesterday for father waited *Jan heeft op vader gisteren gewacht.

Object shift, on the other hand, can readily cross adverbs like gisteren: Ik heb dat boeki gisteren ti gelezen ‘I have read that book yesterday’. Recall that we only consider the cases with a more or less neutral intonation pattern; adding exhaustive focus accent to the PP complement or emphatic accent to the adverb may improve the unacceptable cases in the primed examples. The examples in (18a&a′), finally, show that scrambling of a PPcomplement across an adverbially used PP is always blocked, whereas object shift across such adverbial PPs is possible. For completeness’ sake, note that the ungrammaticality of (18a) cannot be accounted for by assuming some constraint that blocks movement of a complement of a certain categorial type across an adverbial phrase of the same categorial type, since this would incorrectly exclude object shift in example (18b), in which we are dealing with the adverbially used DP deze middag ‘this afternoon’. (18)

a.

a′.

b.

dat Jan na de vergadering wachtte. that Jan for Marie after the meeting waited ‘that Jan waited for Marie after the meeting.’ dat Jan na de vergadering wegbracht. that Jan the book after the meeting away-brought ‘that Jan brought away the book after the meeting.’ dat Jan deze middag zal wegbrengen. that Jan that book this afternoon will away-bring ‘that Jan will bring away that book this afternoon.’

This discussion shows that PP scrambling exhibits behavior that is untypical of object shift. I therefore conclude that it is movement of the A′-type, viz. focus movement. Consequently, examples like (15b) cannot be used to argue against the claim that Dutch and German have object shift as well. This leaves us with Vikner’s claim in VII that object shift may never target a position between two adverbials, whereas scrambling can. This is illustrated by the Icelandic and Dutch examples in (19) and (20).

70 Chapter 2: Short object shift

The crucial difference is that whereas (19b) is ungrammatical, (20b) is acceptable. (19)

a.

Í gær las Pétur eflaust ekki bókina. Yesterday read Pétur doubtlessly not the book ‘Yesterday Peter undoubtedly did not read the book.’ b. *Í gær las Pétur eflaust bókina ekki tbókina. c. Í gær las Pétur bókina eflaust ekki tbókina.

(20)

a. b. c.

Gisteren had Peter ongetwijfeld niet het boek gelezen. Yesterday had Peter undoubtedly not the book read Gisteren had Peter ongetwijfeld het boek niet gelezen. Gisteren had Peter het boek ongetwijfeld niet gelezen.

The facts are, however, much more complex than Vikner suggests. For example, another crucial difference between the examples in (19) and (20) is that whereas the order in (19a) is an unmarked way of expressing sentential negation in Icelandic, (20a) cannot be used in this way in Dutch. In fact, nominal objects never follow the negative adverb niet ‘not’ when the latter expresses sentential negation: (20a) is only acceptable when we construe the sentence with constituent negation and assign the DP het boek contrastive accent; sentential negation can therefore only be expressed by means of (20b) or (20c). This shows that the contrast between (19) and (20) might be related to some difference concerning negation itself, so that these examples cannot be used to argue that object shift and Dutch/German scrambling are different types of movement. Furthermore, it is not entirely clear whether the restriction that object shift may never target a position between two adverbials holds in full force. Nilsen (2003:12/113) shows that in Norwegian double object constructions the objects can be placed in between any of the adverbs in (21), as long as the relative order of the arguments is maintained. This means that the indirect object can be shifted as in (21b-e), adapted from Nilsen (2007), and that in these constructions the direct object can be shifted to any of the positions indicated by a double cross “#”. A problem with taking these examples as an argument against Vikner’s claim, however, is that Norwegian is reported normally not to allow full object shift; cf. *Studenten leste bokeni ikke ti ‘the student didn’t read the book’(Thráinsson 2001; ex. (4c)). Since I did not investigate Norwegian in detail, I will refrain from commenting on the examples in (21) any further.

Object shift/scrambling is A-movement 71 (21)

a. a. b. c. d. e.

Derfor ga Jens ... Therefore gave Jens ... ... tydeligvis ikke lenger alltid Kari kyllingen kald. ... evidently not any-longer always Kari the-chicken cold ... tydeligvis ikke lenger Kari # alltid kyllingen kald. ... tydeligvis ikke Kari # lenger # alltid kyllingen kald. ... tydeligvis Kari # ikke # lenger # alltid kyllingen kald. ... Kari # tydeligvis # ikke # lenger # alltid kyllingen kald.

The discussion above has shown that of the seven arguments in favor of distinguishing object shift and scrambling mentioned in Table 1, only the first two can be upheld. More importantly, the three arguments that really militate against claiming that Dutch and German have object shift can be easily countered: Dutch and German do not allow scrambling of the complement of a preposition (contra claim III), and the leftward movement of the object that licenses parasitic gaps, as well as the leftward movement of PPs, is arguably not of the A- but of the A′-type. From this I conclude that object shift is not restricted to the Scandinavian languages, but also occurs in Dutch and German. 2.2. Two types of object shift Since Hale & Keyser (1993) and Chomsky (1995a: ch.3) it is generally assumed that verbs are not atomic units but consist of a verbal root V and a light verb v. The verbal root V can be held responsible for the thematic properties of the verb, whereas the light verb v can be held responsible for the “truly” verbal properties of the verb, such as the selection of an external argument and accusative case assignment. Given this bifurcation of the lexical properties of the verb, it seems justified to ask to which head the unvalued nominal features of verbs must be ascribed. Given that v is held responsible for accusative case assignment, it is clear that the unvalued case features must be attributed to it. Assume, however, that the ij-features are not part of v, but of the root V; cf. Broekhuis (2000) and also Chomsky (2005), who adopts this claim in the guise of feature inheritance — T and V are claimed to inherit the unvalued ij-features from the phase heads immediately dominating them, that is, C and v respectively. The claim that the case and ij-features are situated on separate heads makes it possible to adopt Kayne’s (1994) claim that all languages are SVO underlyingly. Assume that the verb phrase has the base structure in (22). (22)

[... v[case] [... V[ij] OBJ[ij]/[case]]]

72 Chapter 2: Short object shift

SOV-languages can be derived from (22) by requiring that in these languages the ij-features on V trigger obligatory object shift, as in (23b). I will henceforth refer to this type of object shift in (23) as short object shift. SVO-languages can be assumed either to lack short object shift, as in (22), or to undo the inversion that results from it by subsequent application of V-to-v, as in (23b). We will see later that short object shift is possible, and probably even obligatory, in all the Germanic languages, including English. This implies that the Germanic VO-languages also have obligatory V-to-v, whereas this movement need not apply in Dutch and German. This will be discussed more extensively in section 2.3.4. (23)

a. b.

[... v[case] [OBJi ... V tOBJ]] [...V+v[case] [OBJi ... tV tOBJ]]

(Germanic OV-languages) (Germanic VO-languages)

Since v still has unvalued case features that can be checked by the object, short object shift can be followed by a second instance of object shift. My claim is that this second instance is responsible for what I referred to earlier as Scandinavian object shift as well as Dutch/German scrambling (of the Amovement type). Since most researchers use the term “object shift” to refer to Scandinavian object shift, I will henceforth refer to the movements in (24a&b) as regular object shift. Recall from our discussion in Chapter 1 that, in the Scandinavian languages, regular object shift in (24b) is only possible if the V+v complex is subsequently moved to I, as in (24b′); we will discuss this movement in section 4.1. (24)

a. b. b′.

[OBJi ... v [tOBJ ... V tOBJ]] [OBJi ..V+v [tOBJ ... tV tOBJ]] [...V+v+I [OBJi ..tv [tOBJ ... tV tOBJ]]]

(Dutch/German scrambling) (Scandinavian object shift)

For completeness’ sake, note that using the notion of regular object shift for the object movement in (24) is preferable to using the notion of long object shift, because the latter is normally used to refer to cases in which weak object pronouns precede the subject in Swedish: cf., e.g., Josefsson (1992), Holmberg & Platzack (1995), and Sells (2001) for the relevant data, which will not be discussed in this study.17 17

Although I will not discuss Swedish long object shift, I do want to mention that I doubt whether it really involves movement of the object pronoun across the subject; the order of the arguments seems rather to reflect the underlying order, that is, the subject in these cases seems to be an internal argument of the verb. Long object shift in the sense of movement of weak object pronouns across the subject does occur in German, but we will see in section 4.2.4 that this can be analyzed as regular object shift across the subject.

The universal base hypothesis: VO or OV? 73

2.3. The universal base hypothesis: VO or OV? The universal base hypothesis postulates that all languages have the same underlying word order. From the perspective of language acquisition, assuming a universal base order is of course advantageous, since it reduces the task of the language learner. Since the language learner tacitly knows from which structure a certain surface string is derived, it becomes easier to establish what movements must be postulated in order to derive this string, and thus to establish the parameters of the target language. Since the universal base order is not a priori given, it is not surprising that there are competing hypotheses about it. Section 2.3.1 will discuss two well-developed proposals from the literature: Kayne’s (1994) Linear Correspondence Axiom predicts that phrases are uniformly spelled out in the specifier-headcomplement order, so that all languages should be SVO underlyingly; Haider’s (2000) Branching Constraint, on the other hand, predicts an underlying SOV order to be possible. This section will motivate why I will adopt Kayne’s proposal rather than Haider’s. I will start by briefly discussing the axioms that underlie the two proposals. After that the two proposals will be evaluated. 2.3.1.

The Linear Correspondence Axiom and the Branching Constraint

According to the Linear Correspondence Axiom (LCA), linear order is derived from the asymmetric c-command relation between the nonterminal nodes. More specifically, Kayne (1994: ch.4) argued that the asymmetric c-command relation is mapped to precedence rather than to subsequence: (25)

Let X, Y be nonterminals and x, y terminals such that X dominates x and Y dominates y. Then if X asymmetrically c-command Y, x precedes y.

The precise predictions that follow from (25) of course depend on other assumptions. In order to derive his claim that phrases are uniformly spelled out in the order specifier-head-complement, Kayne assumes that specifiers are adjoined to their phrase as in (26). (26)

HPsegment specifier

HPsegment H

complement

74 Chapter 2: Short object shift

The uniform specifier-head-complement order will now follow when we adopt the definition of c-command in (27): the specifier asymmetrically c-commands both H and the complement of H, and therefore the terminals dominated by the specifier precede the terminals dominated by H and the complement. Since H asymmetrically c-commands the nonterminal nodes dominated by its complement, the terminal immediately dominated by H precedes the terminals dominated by the complement. The clause in (27a) that X must exclude Y is crucial since otherwise the lower HP segment asymmetrically c-commands the nonterminals dominated by the specifier, so that the terminals dominated by the lower segment of XP should precede the terminal dominated by the specifier, which would then lead to a contradictory ordering. (27)

a. b.

X c-commands Y iff X and Y are categories and X excludes Y and every category that dominates X dominates Y. X excludes Y if no segment of X dominates Y.

Chomsky (1995b) derived more or less the same result by claiming that only heads and maximal projections are relevant in the computation of the linear order; intermediate projections only play a role during the derivation but are invisible in the phonological component. This means that only the [+min] and [+max] nodes in (28) enter the computation, and consequently we again derive the specifier-head-complement order. (28)

H[-min, +max] specifier

H[-min, -max] H[+min, -max]

complement

There is, however, an important difference between the two proposals concerning adjunction. The definition of c-command in (27) blocks adjunction to the HP in (26). Since in (29) neither the adjunct nor the specifier is dominated by HP, they are both immediately dominated by the first category above HP. Since the adjunct asymmetrically c-commands the nonterminal nodes dominated by the specifier, the terminals dominated by the adjunct should precede the terminals dominated by the specifier. Since the specifier asymmetrically c-commands the nonterminals dominated by the adjunct, the terminals dominated by the specifier should also precede

The universal base hypothesis: VO or OV? 75

the terminals dominated by the adjunct. This leads to a contradictory order, and hence the structure in (29) is blocked. (29)

HPsegment adjunct

HPsegment specifier

HPsegment H

complement

Note in passing, however, that Kayne seems to be working from the assumption that the LCA must be respected at all intermediate stages of the derivation. When we replace this assumption by the plausible claim that, since the LCA is essentially a PF-principle, it only has to be respected at spell-out, Kayne’s ban on adjunction no longer follows (cf., e.g., Chomsky 1995b, Broekhuis 1996, Nunes 1999;2004, and Moro 2000 for examples of studies that entertain this intuition). Under the new assumption adjunction to HP will be allowed when the specifier of HP is moved into some higher position in the course of the derivation. Chomsky’s structure in (28) also allows adjunction when the specifier is filled: in (30) the specifier is dominated by all segments of the maximal projection of H, and therefore does not c-command the adjunct. Consequently the adjunct asymmetrically c-commands the specifier, so that the terminals dominated by the former will precede the terminals dominated by the latter. (30)

H[-min, +max]/segment H[-min, +max]/segment

adjunct specifier

H[-min, -max] H[+min, -max]

complement

In this study I will adopt this option and assume for reasons that will become clear later that adjunction of, e.g., adverbial phrases is allowed in the normal case. More specifically, I will assume that sentence and VP-

76 Chapter 2: Short object shift

adverbs are adjoined respectively to vP and VP (or the functional projections immediately dominating them). This implies that I will not follow the cartographic approach, according to which adverbial phrases are generated in designated specifier positions of functional projections (Cinque 1999); functional projections are only adopted in this study when the head contributes to the meaning of the clause as, for example, CP (illocutionary force), TP (temporal information), AspP (aspectual information), NegP (sentential negation), etc.18 Kayne’s proposal in (25) predicts that all languages have an underlying SVO-order. This implies that the OV-languages are derived by leftward movement of various types of VP-internal constituents: DP-objects, PP-objects, predicative phrases, etc. Haider (1997a) and Barbiers (2000) have objected to this and pointed out that for many of these elements it is hard to find an obvious trigger that would motivate these movements. They propose an underlying SOV-order and claim that this gives rise to a simpler grammar, since the VO-languages can be derived by means of a single operation, namely V-movement across the VP-internal constituents. Later in this section I will challenge this claim, but before I do that let me first discuss Haider’s axiom that derives the underlying SOV order, the Branching Constraint (BC). The version of the BC that I will discuss here is taken from Haider (2000) and is given in (31); slightly different formulations can be found in Haider (1997a,b), and in section 2.3.3.2 I will discuss the slightly extended version of the BC found in Haider (2003). (31)

Branching Constraint: Projection-internal branching nodes on the (extended) projection line follow their sister node.

The BC conspicuously differs from the LCA in that it allows both the complement-head and the head-complement order when we are dealing with a lexical head L: both structures in (32) satisfy the BC, because there is no branching projection of L that is occupying a left branch. The complement of L, the branching node XP, may precede L since it is not a projection-internal branching node, but a complete extended projection. (32)

a. b.

VO-language: [L′ L XP] OV-language: [L′ XP L]

18 In section 4.3.2, I will even go further and argue that NegP and, in some cases, CP are not headed by a functional head, but are rather created by remerger of some lower head (V and I, respectively), and must therefore be analyzed as extended projections in the sense of Grimshaw (1997) and Nash & Rouveret (1997).

The universal base hypothesis: VO or OV? 77

Haider assumes that the choice between the two structures in (32) depends on a parameterized option on the directionality of licensing of arguments. (33)

Directionality parameter of licensing of arguments: a. VO-languages license their arguments from left to right Ÿ (32a) is selected b. OV-languages license their arguments from right to left Ÿ (32b) is selected

Note in passing that the adoption of (33) shows that Haider still assumes some version of the directionality parameter from the earlier P&P framework, that is, he does not assume a universal underlying OV-base, but merely postulates that this order can be base-generated. In this respect Haider’s proposal crucially differs from the one proposed by Barbiers (2000), who claims that nonverbal (but not verbal) complements are always generated in the position preceding the verb: unfortunately, Barbiers does not give the axiom that derives this. The BC forces the branching projection L′ in (32) to follow the specifier YP of L, as in (34). In the VO-languages this results in a conflict with the licensing condition on arguments, which requires that the specifier YP be licensed by L from the left. Haider claims that this conflict is solved by moving L into a position preceding YP, which gives rise to a Larsonian shell structure. In the OV-languages, placing the specifier YP in front of the branching projection L′ is in accordance with the licensing condition of arguments: the specifier YP is correctly licensed from the right by L. By assuming that the type of head movement that results in shell formation is subject to a Last Resort Condition in the sense that it must be forced by the licensing condition on arguments, it follows that shell structures do not arise in the OV-languages. (34)

a. b.

VO-languages: [LP YP [L′ L XP]] Ÿ [L [LP YP [tL XP]]] OV-languages: [LP YP [L′ XP L]]

Again Haider’s proposal differs somewhat from the one by Barbiers. Since Barbiers assumes that object-DPs are invariably generated to the left of the verb, he concludes that leftward verb movement is required to derive the VO-orders with nominal arguments. Since the BC forces specifiers to be on a left branch (cf. (34)), it also predicts that there is no rightward substitution movement: the structure in (35b) is excluded because the projection-internal branching node F′ precedes its specifier, the moved phrase WP: the only possibility is therefore as given in (35a).

78 Chapter 2: Short object shift (35)

a. [FP WPi [F′ F [LP YP [ti L]]]] b. *[FP [F′ F [LP YP [ti L]]] WPi]

If we further assume that the notion of projection-internal branching node is insensitive to the category/segment distinction, the BC also blocks rightadjunction to phrases, irrespective of whether it is the result of basegeneration or movement (see fn.22 on p.85 for some remarks on headadjunction): in the primed examples in (36) the lower XP-segment precedes the adjoined phrase, and therefore these structures are blocked; in the prime-less examples the lower segment follows the adjoined phrase, and the structures are admissible. (36)

a. [XP YP [XP …]] a′. *[XP [XP …] YP] b. [XP YPi [XP … tj ]] b′. *[XP [XP … tj ] YPi]

Finally, the BC in (31) blocks rightward placement of a functional head F if FP is an extended projection of the lexical head L: in (37) the branching projection LP is internal to the extended projection FP and it must therefore follow its sister F. (37)

a. [FP … F [LP YP [L′ XP L]]] b. *[FP … [LP YP [L′ XP L]] F]

In (38) I summarize the predictions concerning the underlying order of the verb and its complement made by the three proposal discussed in this section. The following section will discuss some ramifications of these proposals. (38)

a. b.

c.

2.3.2.

LCA (Kayne 1994): heads precede their complements Ÿ VO-order only. BC (Haider 2000): functional heads in an extended projection precede their complements; lexical heads either precede or follow their complements Ÿ both VO- and OV-orders are possible, depending on the licensing direction of the language. Barbiers (2000): OV-order only (if the object is nonverbal).

The trigger problem

According to Kayne’s (1994) LCA, the complement of a verb is basegenerated to the right of the verb. Consequently, when the complement surfaces to the left of that verb, as in the OV-languages, it must have been moved leftward across the verb. This raises the question of what triggers

The universal base hypothesis: VO or OV? 79

this movement in these languages. When the complement of the verb is a direct object, finding a trigger for the movement is not difficult: obvious candidates are the unvalued case or ij-features on the verb; cf. the discussion in section 2.2. In other cases, however, identifying a trigger is not that easy: predicative complements like groen ‘green’ in (39a), for example, are normally assumed to be generated as a complement of the verb, but it is not a priori clear what triggers the movement into the preverbal position in (39b). When we adopt the BC, the trigger problem does not arise, since we can simply assume the surface order in (39a) to be base-generated.19 (39)

a. b.

dat Jan het hek groen verft. that Jan the gate green paints dat Jan het hek groeni verft ti.

The derivation of the English order in (40a) is much the same in the three approaches. When we adopt the LCA, there are two possibilities: one option is to assume that the verb takes a small clause complement, so that the order in (40a) is base-generated as in (40b), and nothing more need be said. Alternatively, one may follow Hale & Keyser (1993) in assuming that the structure is as given in (40b′), where the predicate and its subject are generated as the complement and the specifier of V, respectively, and the surface order is derived by moving V to v. (40)

a. b. b′.

that John painted the gate green. that John painted [SC the gate green] that John paintedi [VP the gate [ti green]]

In the alternative approaches, there are also two options. Haider (1997a) assumes a structure comparable to (40b′) above, but with a different motivation: he claims that the verb and the predicate form a complex predicate, and thus “share” the argument the gate. Since the object must be licensed by the verb from the left, the latter must undergo head movement, as in (41b). Alternatively, one may assume that the verb takes a small clause complement and follow Barbiers’ (2000) claim that non-verbal complements are always base-generated in preverbal position: the only thing required then is verb movement across the small clause, as in (41b′). 19

Strictly speaking, the use of the notion base-generation is only appropiate when we discuss Haider’s and Barbiers’ proposals. Under our earlier assumptions that the LCA is a PF-principle and, thus, that linearization applies at PF, the notion base-generation should rather be replaced by something like spell out. For ease of presentation, however, I will maintain the improper use of the notion.

80 Chapter 2: Short object shift (41)

a. b. b′.

that John painted the gate green. that John paintedi [VP the gate [ti green]] that John [paintedi [SC the gate green] ti]

(Haider) (Barbiers)

Haider and Barbiers have claimed that their OV-approaches are superior to Kayne’s VO-approach, because they solve the trigger problem discussed above: both the VO- and the OV-order can be derived by appealing to leftward verb movement, which seems independently needed, so that taking recourse to leftward movement of arguments or predicative complements becomes superfluous. This argument against the VO-approach, of course, only holds water if verb movement of the type discussed above indeed suffices to derive all the established word orders. Section 2.3.3 will argue that this is not the case, and, consequently, that it remains an open question whether base-generated OV-orders should be allowed or not. Subsequently, section 2.3.4 will defend the OV-approach by showing that there is independent empirical motivation for adopting the object movement predicted by the OV-approach. Besides the trigger argument in favor of the OV-approach, which is the main topic of this section, there are at least two other arguments that have been given in favor of it. I will briefly address these here. The first is that, since no movement of the complements is needed to derive the OV-order, we immediately account for the fact that English and Dutch have basically the same surface order of arguments. This, of course, also follows from my proposal in section 2.2 that the leftward movements postulated for Dutch apply in English as well. The second argument is that the OV-approach immediately accounts for the fact that preverbal phrases are transparent for movement in the OV-languages; since the object is not moved into preverbal position, Freezing is not expected. In the VO-approach, we may account for the lack of Freezing by assuming that subextraction from XP is only possible if XP occupies the lowest position in which it could in principle appear: since short object shift into the position preceding the clause-final verbs is obligatory in Dutch, subextraction is possible from this position; regular object shift (scrambling), on the other hand, is optional in Dutch so that it does invoke a freezing effect. 2.3.3.

Complex verb constructions

This section will discuss complex verb constructions and argue that the theories of Haider and Barbiers are not able to derive the established orders between arguments/predicates and verbs in these constructions solely by appealing to the type of head movement discussed above. Since Haider

The universal base hypothesis: VO or OV? 81

(2003;2005) and Barbiers (2005) have divergent ideas on complex verb constructions, I will discuss these in two separate subsections. 2.3.3.1. Barbiers (2005) Barbiers (2000) claims that sentential/verbal complements (CPs, IPs and VPs) are base-generated on a right branch, that is, in the VO-order. Barbiers (2005) further shows that adopting this assumption makes it possible to derive all and only the attested verb orders in three-verb clusters by assuming that the modal and auxiliary verbs contain respectively unvalued mood and aspectual features that probe the more deeply embedded VP(s); cf. Broekhuis (1997a) and Haegeman (1998) for similar VP-movement proposals. Consider the verb orders in (42), in which V1 refers to the highest auxiliary or modal verb, and V3 to the main verb: the base order in (42a) surfaces when VP-movement does not apply; example (42b) can be derived by assuming that the unvalued feature on V2 attracts the projection of the main verb, VP3; example (42c) can be derived by assuming that VP3 remains in its base position, but that V1 attracts VP2; example (42d) is derived when both V2 and V1 trigger movement of their complement; (42e) can be derived by assuming that the projection of the main verb is moved successive-cyclically via the specifier position of VP2 to the specifier position of VP1; (42f), finally, is not attested and cannot be derived because movement of VP2 across V1 necessarily pied pipes VP3. This derives all and only the attested word orders. (42)

a. V1-V2-V3: [VP1 V1 [VP2 V2 [VP3 … V3]]] b. V1-V3-V2: [VP1 V1 [VP2 [VP3 … V3] V2 tVP3]] c. V2-V3-V1: [VP1 [VP2 V2 [VP3 … V3]] V1 tVP2] d. V3-V2-V1: [VP1 [VP2 [VP3 … V3] V2 tVP3] V1 tVP2] e. V3-V1-V2: [VP1 [VP3 … V3] V1 [VP2 t′VP3 V2 tVP3]] f. *V2-V1-V3

Note in passing that Broekhuis (1997a) has claimed that the V3-V1-V2 order in (42e) is only possible in perfect tense constructions. Barbiers shows in his study of Dutch dialects that these orders do occur in non-perfective constructions. This, however, is only possible as a secondary order, that is, next to one of the other orders in (42a-d): there is no dialect in which (42e) is the only possible order. Since Den Besten & Broekhuis (1992) already showed that some V3-V1-V2 orders involve nominalization of V3, I believe more research is needed to establish whether this order is a genuine one in non-perfective constructions.

82 Chapter 2: Short object shift

The crucial thing for the present discussion is that, according to Barbiers (2000), the nominal arguments of the main verb are basegenerated in the positions indicated by the dots. The surface realization of these arguments need not, however, coincide with these positions. This is illustrated in (43), adapted from Haegeman (1992), for the possible surface realizations of indirect and direct objects in West Flemish constructions with a V1-V2-V3 sequence. Under Barbiers’ assumptions, verb movement does not suffice to derive these orders; only the base order in (43a) will be available. The most plausible assumption is therefore that the examples in (43b-d) are derived by leftward object movement, which would imply that this movement is also needed when we postulate an underlying OV-order.20 (43)

a. b. c. d. e. f.

(NPsubj) V1 (NPsubj) V1 IO (NPsubj) IO V1 IO (NPsubj) V1 IO (NPsubj) IO V1 (NPsubj) IO DO V1

V2 IO V2 V2 DO V2 DO V2 V2

DO V3 DO V3 DO V3 V3 V3 V3

Note that I only brought up the examples in (43) to show that some form of object shift is also needed in the VO-approaches. It clearly goes far beyond the scope of this study to discuss in detail the variation found in West Flemish and to provide an answer to the question of which order arises under which conditions, not least because the information available from the literature is insufficient to enable a non-native speaker to accomplish this task without additional field work. However, Haegeman & Van Riemsdijk (1986) and Haegeman (1992: ch.3) provide a number of useful insights (e.g., concerning scope) and would be a good starting point for anyone who wishes to venture into this enterprise. If a resultative adjective like groen in (39) is analyzed as a predicative complement of the main verb, we have to draw the conclusion that Barbiers also needs some form of predicate movement. Consider the two-verb construction in (44). Under the OV-analysis, the order in (44a), which is a possible order in West Flemish, is base-generated. The order in (44b) is, however, also possible (and even obligatory in Dutch), so that under Barbiers’ assumptions we again need an additional mechanism that places 20

In view of the Barbiers’ (2005) analysis in (42), object movement is actually expected: since (42) presupposes that the leftward movement of the verbal complements is triggered by the aspectual/modal features on the auxiliary/modal verbs, we cannot rule out in a principled way that movement of DP-complements is likewise triggered by the ij- and/or case features on the verb(s), as is normally assumed by proponents of the VO-approach.

The universal base hypothesis: VO or OV? 83

the predicate in front of the finite verb. The most plausible assumption is that this involves leftward movement of the predicate. (44)

a.

b.

dat Jan het hek wil [groen verven] that Jan the gate wants green painted ‘that Jan wants to paint the gate green.’ dat Jan het hek groeni wil [ti verven]

We have seen above that Barbiers’ (2005) assumption that verbal complements (VP/IP) are base-generated to the right of the selecting auxiliary/modal verb inevitably leads to the conclusion that apart from verb movement we need an additional stipulation to allow arguments and predicative complements to precede the clause-final verbal sequence. This suggests that in Barbiers’ framework leftward movement of arguments and predicates is also needed. Since we will see in section 2.3.3.2.2 that Van Riemsdijk (1997) provides independent empirical evidence that shows that postulation of leftward movement of predicative phrases is also needed when we assume an OV-base, this voids Barbiers’ argument in favor of underlying OV-orders. 2.3.3.2. Haider (2003;2005) This subsection will discuss Haider’s (2003;2005) theory concerning the clause-final verbal sequence. Section 2.3.3.2.1 will start by showing that unlike Barbiers’ proposal, Haider’s does not require leftward movement of the complements of the main verb. Section 2.3.3.2.2 will show, however, that this comes at a certain cost in that it requires a large set of stipulations that are not needed in the LCA-based VO-approaches and that in some cases seem to lead to internal inconsistencies. I will conclude from this that the VO-approach is superior to Haider’s. 2.3.3.2.1. Haider’s proposal Haider’s (2003) proposal differs from Barbiers’ in that Haider does not assume VP/IP complements to be base-generated in the VO-order in Dutch and German. Rather, he assumes that these complements are subject to the same licensing requirements as DP-arguments and predicative complements, and he therefore predicts that they should precede their selecting verb in the OV-languages.21 The licensing condition on arguments therefore predicts that the base-order of the verbs is the inverse of what is assumed by Barbiers: V3-V2-V1, where V3 again refers to the main verb. Let us 21 Haider (1997b) adopts a special proviso for CP-complements and other ‘extraposed’ constituents, which are assumed to be base-generated to the left of the verb.

84 Chapter 2: Short object shift

assume for the moment that (45b) is indeed the underlying order, and consider the question of how the verb orders in (42a-e) can be derived. (45)

• Expected base orders in the OV-languages: a. *[VP1 … V1 [VP2 … V2 [VP3 … V3 …]]] (excluded by the licensing condition) b. [VP1 … [VP2 … [VP3 … V3] V2] V1] (predicted by the licensing condition)

A first option that comes directly to mind is rightward VP-movement, but this is excluded by the ban on rightward movement (cf. (35) and (36)). Moreover, this would also imply some form of leftward argument/predicate movement: since VP-movement pied pipes the VP-internal material, the word orders in (43b-f) can only be derived when rightward VP-movement is followed by leftward movement of the complement(s) of the main verb. Another option, which would be more in line with Haider’s BC, is leftward movement of the selecting verb in front of its VP/IP complement, as in (46). This at least derives the attested orders in (42a-d). The unattested order in (42f) cannot be derived, as desired. The only potential problem is that order (42e) cannot be derived either, but in my view this is not necessarily a disadvantage; cf. the discussion below (42). (46)

a. b. c. d.

[… [… [… V3] V2] V1] […V1 [… [… V3] V2] tV1] [… […V2 [… V3] tV2] V1] […V1 […V2 [… V3] tV2] tV1]

Although the derivations in (46) come very close to the mark, Haider cannot adopt this proposal since the postulated verb movements are not in accordance with the Last Resort Condition: the verb movements should establish the required licensing configuration on movement, but in (46) they actually destroy it. In addition, this proposal again implies some form of leftward argument/predicate movement, since if the selecting verb precedes its VP-complement, it also precedes the latter’s arguments, so that the word orders in (43b-f) can only be derived when leftward V-movement is followed by leftward movement of the complement(s) of the main verb. The ban on rightward VP-movement and “untriggered” leftward verb movement as well as the claim that leftward movement of complement(s) of the main verb does not occur force Haider to adopt an entirely different approach to complex verb constructions; note that the two options considered above are not discussed by Haider, so I cannot tell with certainty whether his actual approach, which will be discussed now, was

The universal base hypothesis: VO or OV? 85

motivated by the failure of these options. Haider’s proposal is based on the reformulation of the BC in (47). (47)

Branching Constraint (revised version): Projection-internal branching nodes on the (functionally or lexically extended) projection line follow their sister node.

The intended effect of this reformulation of the BC is that not only functional heads but also auxiliaries and modal verbs are part of the extended projection of a lexical verb. As far as linearization is concerned, (47) therefore predicts that, like functional heads, auxiliary and modal verbs precede their complement, as in (48a); cf. the discussion of (37). The order in (48b), which is preferred by the licensing conditions, is now excluded by the revised BC. (48)

a. [VP1 … V1 [VP2 … V2 [VP3 … V3 …]]] (allowed by the revised BC) (blocked by the revised BC) b. *[VP1 … [VP2 … [VP3 … V3] V2] V1]

For the VO-languages the reformulation of the BC has no consequences: both the BC in (47) and the licensing condition on arguments predict (48a) to be selected in these languages. For the OV-languages, however, the result is dramatic since it is now predicted that in these languages modal and auxiliary verbs cannot take a verbal projection as their complement: the structure in (48b) is ruled out by the BC, and the structure in (48a) violates the requirement that the verbal complement be licensed from the right by the modal/auxiliary verb (which, as is easy to see, cannot be repaired by means of leftward verb movement). Haider concludes from this that the verbs are inserted as a cluster in the OV-languages, and that the thematic properties of the main verb are simply inherited by the whole cluster. Within the cluster, the main verb precedes the higher ones in order to satisfy the licensing condition on arguments. Instead of (48a) or (48b), we therefore have the structure in (49), where the dots indicate the arguments of the main verb V3.22 22

The labeled bracketing given by Haider is [VP … [V3 [V2 V1]]]. So far I have not been able to make sense of this, especially since we will see that Haider assumes that V3 is able to right-adjoin to V2 (and V2+V3 to V1), which would amount to lowering under Haider’s bracketing. This problem is solved by assuming the labeled bracketing in (49). However, this structure raises the question whether the BC also applies to the nodes within the verb cluster; if so, (49) would be excluded because the branching node [V3 V2] precedes V1. I ignore questions like these in the discussion that follows, and assume for the sake of argument that the BC does not apply within the cluster.

86 Chapter 2: Short object shift (49)

[VP … [[V3 V2] V1]]

Let us now consider Haider’s (2003) account of the attested word order variation within verb clusters given in (42) above. Haider claims that this variation is the result of verb movement within the cluster. He distinguishes two types of verb movement: right-adjunction of VN+1 to VN and leftadjunction of a verb to the full cluster. Haider assumes that these types of verb movement are essentially similar to the verb movement type that we find in verb second constructions like (50b). Most noticeably, all these verb movements are assumed to obligatorily strand verbal particles like op in (50). (50)

a. b.

dat Jan dat boek opbergt. that Jan that book prt.-files Jan bergt dat boek op tbergt.

Assuming right-adjunction readily accounts for examples like (51a), in which the particle precedes the verbal sequence. This order can be derived by first adjoining the main verb bergen to the modal verb moeten, while stranding the particle op, followed by movement of the complex moeten+bergen to the modal verb zal. (51)

a.

b.

dat Jan dat boek op zal moeten bergen. that Jan that book prt. will must file ‘that Jan will have to file that book.’ dat Jan dat boek [[opbergen moeten] zal] Ÿ dat Jan dat boek [[op tbergen moeten+bergen] zal] Ÿ dat Jan dat boek [[op tbergen tmoeten+bergen] zal+moeten+bergen]

Left-adjunction to the verb cluster is necessary to account for the order in (52a), in which the particle remains adjacent to the main verb. Since rightadjunction of the verb obligatorily strands the particle, that is, since the particle cannot permeate the verbal cluster by pied piping, the only way to derive the order in (52a) is by leftward movement of the two modal verbs, as in (52b). (52)

a.

b.

dat Jan dat boek zal moeten op bergen. that Jan that book will must prt. file ‘that Jan will have to file that book.’ dat Jan dat boek [[opbergen moeten] zal] Ÿ dat Jan dat boek [zal [[opbergen moeten] tzal]] Ÿ dat Jan dat boek [zal [moeten [[opbergen tmoeten] tzal]]]

In order to derive the order in (53a), it must be assumed that the two types of verb movement may also apply simultaneously. This is shown in (53b).

The universal base hypothesis: VO or OV? 87 (53)

a.

b.

dat Jan dat boek zal op moeten bergen. that Jan that book will prt. must file ‘that Jan will have to file that book.’ dat Jan dat boek [[opbergen moeten] zal] Ÿ dat Jan dat boek [[op tbergen moeten+bergen] zal] Ÿ dat Jan dat boek [zal [[op tbergen moeten+bergen] tzal]]

Although Haider does not discuss this, it does not seem difficult to derive all the attested word orders in (42) by placing special restrictions on the application of the two movement types. Certain orders, like V1-V2-V3 discussed above, can even be derived in more than one way. It seems harder, however, to block the unattested order V2-V1-V3 in (42f). For example, in the derivation in (52b) I followed Haider’s implicit assumption that left-adjunction involves tucking in (cf. example (39) in Haider 2003), since if we did not assume this to be obligatory, the unattested order in (42f) would be derived. Furthermore, in order to block this order, we must also assume that left-adjunction of V2 across the structurally higher V1 is excluded. This, in turn, forces us to assume the countercyclic derivation in (52b)). In short, the unacceptability of (42f) does not follow in a natural way from the proposal and requires some additional tinkering. So far, Haider’s proposal does not account for the surface realization of the arguments in (43) either: it is predicted that the clause-final verbal sequence always follows the arguments of the main verb, because these precede the verb cluster: [VPIO[V′DO[V V-V]]]. In order to allow for the orders in (43), Haider proposes that the verbs cannot only be left-adjoined to the verbal cluster but also cluster-externally, that is, to one of the projections (V′ or VP) of the verbal clusters. 2.3.3.2.2. Comparison to the VO-approach Now that we have a more or less complete picture of Haider’s OVapproach, it is time to evaluate it, and see how it fares compared to the LCA-based VO-approach. Recall that the main problem under discussion is that the VO-approach forces us to assume leftward movement of elements for which no a priori trigger is available. How serious is this problem? Since leftward movement of DP-complements is normally assumed to be triggered by the ij- and/or case features on the verb, and since we can simply follow Barbiers (2005) in assuming that leftward VP-movement is triggered by the mood/aspectual features on the modal or auxiliary verb, the problem mainly involves leftward movement of predicative phrases into preverbal position, which is needed to derive examples like (44b), and leftward movement of verbal particles, which is needed to derive the

88 Chapter 2: Short object shift

a-examples in (51)-(53). Since it has been claimed that the verbal particles are also predicative complements (Den Dikken 1995a), these problems actually reduce to a single one. Interestingly, Van Riemsdijk (1997) has extensively argued that even when we adopt an underlying OV-order, there are reasons to postulate some form of leftward movement of predicative phrases. The argument is based on the relative position of predicative phrases and stranded prepositions of, e.g., adverbial phrases. Predicative phrases and stranded prepositions have in common that they normally must be adjacent to the clause-final verbal sequence. This is illustrated in (54a) for predicative phrases, and in (54b) for stranded prepositions. Interestingly, when the clause contains both a predicative phrase and a stranded preposition, the former may either follow or precede the latter. Since the stranded preposition mee in (54c) is the trace of a VP-adjunct, Van Riemsdijk claims that this is due to movement of the predicative phrase into some VP-external position; for arguments against some other conceivable derivations, like rightward movement of the stranded preposition, see Van Riemsdijk (1997: section 3). (54)

a.

b.

c.

dat Jan het hek morgen zal verven. that Jan the gate green tomorrow will paint ‘that Jan will paint the gate green tomorrow.’ dat Jan daar het hek morgen zal verven. that Jan there the gate with tomorrow will paint ‘that Jan will paint the gate with it tomorrow.’ dat Jan daar het hek morgen mee zal verven. that Jan there the gate tomorrow green with will paint ‘that Jan will paint the gate green with it tomorrow.’

This convincingly shows that even if we assume that Dutch is OV underlyingly, we have to postulate some form of leftward predicate movement. Van Riemsdijk claims that the predicate targets the specifier position of a PredP, and it is therefore a small step to evoke this position in order to derive the Pred-V order from an underlying V-Pred order, which leads us to an analysis of the sort proposed by Zwart (1993;1997) and Koster (1994). The only remaining problem would be that it is not immediately clear how the word order variation in (54c) can be accounted for if the predicative phrase is base generated in the postverbal position — at first sight, we seem to predict that movement of the predicate will always cross the stranded preposition. I will return to this issue in section 5.3.2.1.4. The conclusion we must draw from the discussion above is that the OV-approaches face the same trigger problem that arises in the VO-

The universal base hypothesis: VO or OV? 89

approach, provided at least that one assumes that the two orders in (54c) are not base-generated but transformationally derived from the same underlying source. We may therefore conclude again (cf. section 2.3.3.1) that the OV-approaches and the LCA-based VO-approach are in a tie in this respect. This suggests to me that the VO-approach is to be preferred to Haider’s OV-approach, given that the latter requires a large set of assumptions that are not needed in the former. The most important of these assumptions are given in (55). (55)

a. b. c.

Directionality parameter Base insertion of verb clusters Excorporation of verbs from the verb cluster

The directionality parameter is not needed within the LCA-based VOapproach. At first sight this seems balanced by the fact that the latter needs to postulate some trigger for the leftward movement of arguments and predicates to compensate, but since we have seen that in the end this is also needed in Haider’s approach, the directionality parameter must be seen as a true addition to the theoretical apparatus. Furthermore, assuming this directionality parameter has the result that linearity continues to play a role in the syntax, whereas the VO-approach can entertain a fully hierarchical view of syntax. Haider’s approach also differs fundamentally from the VOapproach in that it requires postulation of base-generated verb clusters, which in turn requires several additional mechanisms which were not extensively discussed here, such as “pooling” of the arguments of the verbs in the cluster (Haider 2003). Furthermore, the postulation of verb clusters requires assumption (55c), that verbs can excorporate from the verb cluster, in order to derive verb second constructions and the permutated verb sequences in (43a-e). In short, it seems that Haider’s OV-approach requires a considerably larger theoretical apparatus and thus defines a much larger class of possible grammars than the VO-approach based on Kayne’s LCA. Consequently, the latter is more highly valued as far as explanatory adequacy is concerned. An even more serious problem for Haider’s proposal on top of that is that it requires the postulation of the verb movements in (56), for which there is no (syntactic) trigger. Haider (2003:117-18) acknowledges this and motivates these verb movements instead by taking recourse to parsing considerations; the V3-V2-V1 order results in center embedding, which is known to create processing difficulties and application of the verb movements in (56) contribute to dissolving these. One possible objection to this is that it runs afoul of the postulate of the autonomy of syntax.

90 Chapter 2: Short object shift (56)

a. b.

Rightward adjunction of verbs to verbs Leftward adjunction of verbs to: (i) the verb cluster (ii) higher verbal projections

Recall from the discussion of (34) that Haider claims that verb movement in English is motivated by the fact that it establishes the required licensing relation between the verb and its object. In the derivation of the Dutch example (51a), however, the licensing relation is destroyed rather than created by rightward movement of the verbs. The same holds for the leftward verb movements involved in the derivation of (52a). Finally, the stipulation that leftward movement may target the cluster or any other higher verbal projection seems merely dictated by the data in (43), and does not follow from any independently motivated principle. 2.3.3.3. Conclusion This section has discussed the problem that, in order to derive the word order of OV-languages, assuming a universal underlying VO-order requires the postulation of leftward movement of predicative phrases and particles for which no obvious trigger is available. Barbiers (2000) and Haider (2000) have argued that this trigger problem does not arise when we assume that the universal base order is OV, and conclude from this that the OVapproach is superior to the VO-approach. This section has challenged this claim by showing that the postulation of leftward movement of predicative phrases and particles is also needed in the OV-approaches discussed. Furthermore, since Haider’s OV-approach requires a number of additional mechanisms in order to account for word order phenomena in complex verb constructions that are not needed in the VO-approach, I concluded on conceptual grounds that the latter approach is superior to the former. 2.3.4.

Evidence in favor of “short” object shift in English

The previous subsection has argued against Haider’s and Barbiers’ claim that assuming an underlying OV-order makes the assumption of leftward movement of arguments and predicates superfluous, and therefore dismissed this as an argument for their second claim that assuming an underlying OVorder is preferable to assuming an underlying VO-order. Actually, if anything can be concluded from the discussion above it is rather that the opposite may hold true. However, we have not seen so far any conclusive empirical evidence in favor of one of the two options. The goal of this section is to provide such evidence in favor of the VO-approach by showing that the criticized postulated leftward movement of the object does

The universal base hypothesis: VO or OV? 91

not only occur in the OV-languages but also in the VO-languages, as was already suggested by (22) and (23) in section 2.2, repeated here as (57). (57)

a. b. c.

[... v[case] [... V[ij] OBJ[ij]/[case]]] [... v[case] [OBJi ... V tOBJ]] [...V+v[case] [OBJi ... tV tOBJ]]

(underlying structure) (OV-languages) (VO-languages)

The analysis in (57) agrees with the OV-approaches discussed above that the difference between the OV- and the VO-languages is that only the latter has obligatory V-to-v, but it differs from them by claiming that the VOlanguages (may) also have short object shift. The evidence in favor of the second claim that will follow below is not new, but simply taken from the existing literature, and therefore the discussion can be brief. The idea that VO-languages like English may have some sort of object shift was first introduced by Johnson (1991) in order to account for –among other things– the fact that objects may and sometimes must be placed in front of certain particles; cf. also Koizumi (1993). (58)

a. b.

John looked up the information/*it. John looked the information/it up.

Conclusive evidence in favor of short object shift is provided in Lasnik (1999a: ch.2&8) based on binding and NPI licensing. First consider the unacceptable example in (59a), which involves a violation of binding condition A: the anaphor is embedded in an adjunct in the matrix clause and is therefore not c-commanded by the intended antecedent the defendants. However, when we replace the finite clause by an infinitival clause, as in (59b), the result improves considerably; according to Lasnik (59b) has more or less the same status as (59c). (59)

a. b. c.

*?

The DA proved [that the defendants were guilty] during each other’s trials. The DA proved [the defendants to be guilty] during each other’s trials. The DA accused the defendants during each other’s trials.

The acceptability of (59b) suggests that the surface position of the exceptionally case-marked subject is not internal to the infinitival clause, but that it has moved out of it into a position where it c-commands the anaphor. Note that, for this reason, the placement of the brackets in (59b) and some of the later examples is somewhat misleading. I use them only to indicate that the antecedent of the anaphor is base-generated as part of the infinitival clause, without implying that it is still part of this clause when

92 Chapter 2: Short object shift

the binding relations are established. It is important to note that the acceptability of (59b) is not directly related to the fact that the antecedent originates as the subject of the infinitival clause (e.g., by some sort of reconstruction process that destroys the structure of the infinitival clause); the examples in (60) show that the subject of the infinitival clause can bind the anaphor only when it is moved leftward, and consequently the structure of the infinitival clause must be intact. (60)

a.

The DA proved [two men to have been at the scene of the crime] during each other’s trials. b. *The DA proved [there to have been two men at the scene of the crime] during each other’s trials.

The binding argument can be replicated by the examples in (61) and (62). The first example shows that a bound variable reading of the possessive pronoun his is only possible when the subject of the infinitival clause has moved leftward, and (62) shows that licensing of the negative polarity item any also requires leftward movement of the subject. (61)

a.

The DA proved [no suspect to be at the scene of the crime] during his trial. b. *The DA proved [there to be no suspect at the scene of the crime] during his trial.

(62)

a.

The DA proved [no one to be at the scene of the crime] during any of the trials. b. *The DA proved [there to be no one at the scene of the crime] during any of the trials.

The binding facts in the infinitival constructions in (59) to (62) provide a first piece of conclusive evidence in favor of the claim that there is short object shift in English. In fact, under the standard assumptions that binding involves c-command and that adverbials are generated external to VP, the much simpler example The DA accused the defendants during each other’s trials in (59c) already shows the same thing: if short object shift does not apply, the object is VP-internal and therefore does not c-command the reciprocal embedded in the adverbial phrase, and binding would wrongly be predicted to be impossible; if we do assume short object shift, as in the schematic representation in (63), the c-command requirement is satisfied so binding is correctly predicted to be possible. (63)

[vP... v+V the defendants [VP during each other’s trials [VP tV tDO]]]

The universal base hypothesis: VO or OV? 93

A more controversial piece of evidence involves antecedent-contained deletion in example (64a). Assume that the relative clause contains an empty VP, as in (64b). The empty VP must be interpreted as a copy of its antecedent, the VP of the matrix clause, which would lead to the LFstructure in (64c), which again contains an empty VP; given that this empty VP must also be interpreted, this leads to infinite regress. (64)

a. b. c. d.

Dulles suspected everyone Angleton did. Dulles [VP suspected everyone Angleton did [VP e]] Dulles [VP suspected everyone Angleton did [VP suspect everyone Angleton did [VP e]]] etc.

May (1985) proposed to solve the problem of infinite regress by assuming that the object of the matrix VP is removed from the matrix VP by quantifier raising before the empty VP is interpreted. This leads to the derivation in (65): (65b) gives the structure after quantifier raising, and (65c) is the structure resulting from interpretation of the empty VP. (65)

a. b. c.

Dulles [VP suspected [DP everyone Angleton did [VP e]]] [DP everyone Angleton did [VP e]]i [Dulles [VP suspected ti]] [DP everyone Angleton did [VP suspect ti]]i [Dulles [VPsuspected ti]]

Although this solution is very elegant, there are reasons to assume that it cannot be maintained in the form above. First, quantifier raising, which in my view has always been a problematic operation, seems incompatible with the current theories that have given up covert movement. Secondly, Baltin (1987) has shown that there are several empirical problems with the quantifier raising approach to antecedent-contained deletion; cf. Hornstein (1995: ch.5) for a critical overview of the discussion. Fortunately, It has been shown very early in the development of the minimalist program that it is possible to replace the quantifier raising based approach by a theory that avails itself exclusively of operations that are independently motivated. If antecedent-contained deletion does not involve quantifier raising, we must identify an alternative operation that may move the object out of the VP. Hornstein (1995: ch.5) and Lasnik (1991: ch.3) have suggested that the case feature on the verb triggers covert movement of the object into SpecAgrOP before the empty VP is replaced by the matrix VP. Object shift thus leads to a structure like (66b), in which the VP again consists of the verb and a trace, so that it can replace the empty VP as in (66c) without any unwanted consequences.

94 Chapter 2: Short object shift (66)

a. b. c.

Dulles [AGRoP AGROP [VP suspected [DP everyone Angleton did [VP e]]]] Dulles [AGRoP [DP everyone Angleton did [VP e]]i AGROP [VP suspected ti]] Dulles [AGRoP [DP everyone Angleton did [VP suspect ti]]i AGROP [VP suspected ti]]

This proposal provides us with a good alternative for quantifier raising, but still takes recourse to covert movement: the movement cannot take place in overt syntax, since this wrongly predicts that the DP everyone Angleton did precedes suspected; cf. the structure in (66b). The proposal can, however, be readily reformulated such that covert movement is not needed by appealing to the hypothesis in (57) that English has the short object shift movement in (57b), which is triggered by the ij-features on the verbal root suspect. Although short object movement results in inversion of the verbal root suspect and the object, we can undo this inversion by subsequently moving the verbal root to v. This provides us with an explanation that is fully consistent with the facts and that does not require the postulation of covert movement operations.23 Of course, it goes without saying that the present analysis is much indebted to Koizumi’s (1993) proposal (which is also adopted in Lasnik’s later work; cf., e.g., Lasnik 1999a: ch.6;1999b). Let me conclude this discussion of antecedent-contained deletion with a final remark. May’s proposal and the alternatives sketched above differ in that the former assumes that the object is moved by means of A′-movement whereas the latter assume that it is moved by A-movement. Since A′- and A-movement differ in that only the former gives rise to reconstruction effects, we may safely assume that the semantic features of copies left by the former must be assigned an interpretation by the conceptual-intentional 23

This argument in favor of short object shift is controversial given that there is a set of problems that do not follow immediately from the present proposal. First, Lasnik (1999a: ch.3) has shown that antecedent-contained deletion is also possible when the DP containing the VP gap is embedded in a PP-complement: Dulles spoke to everyone Angleton did. Given that PPs do not undergo A-movement, this is not expected. It will follow, however, if we adopt Kayne’s (2000:part III) proposal that complement PPs are not base-generated as the complement of V, but created later in the derivation: the preposition is merged externally to VP and acts as a probe attracting the DP-complement of V (section 2.4.2.5 will provide some reasons to adopt this proposal). The most recalcitrant data are those discussed by Kennedy (1997) under the designation NP-contained antecedent-contained deletion; in this case the DP containing the VP gap is embedded in an object DP: Beck read a report on every suspect Kollberg did. Kennedy shows that run-of-the-mill application of the A-movement approach will give rise to the wrong interpretation. To my knowledge this problem has not been solved so far.

The universal base hypothesis: VO or OV? 95

system. Given this, there is independent reason to assume that May’s Quantifier Raising analysis cannot be correct. In order to see this, consider again the structure in (65c) and observe that it violates Chomsky’s (1981:212) i-within-i condition: *[Ȗ ... į ...] where Ȗ and į bear the same index. Stated in terms of current copy theory, the problem is that the preposed DP contains a semantically active copy of itself, so that the moved DP and its copy will be given different semantic interpretations by the conceptual-intentional system; this results in an incoherent interpretation. In short, we have replaced one problem by another. The new A-movement approach does not give rise to this problem: given that A-movement does not give rise to reconstruction effects, we may conclude that the semantic features of copies left by A-movement are not interpreted by the conceptual-intentional system, which will therefore only provide a semantic interpretation of the moved DP (the head of the A-chain). The result of this is that the incoherent interpretation will not arise. More evidence in favor of short object shift in English will be discussed later in this study. Section 2.4.1 will show that the placement of the VP-adjuncts also support short object shift in English: since the LCA predicts that VP-adjuncts precede the base position of the object, the fact that they normally occur after the object in the surface realization strongly suggests leftward movement of the object across the adjunct. Evidence of a slighty more ambiguous nature is provided by verb-particle constructions like (67). (67)

a. b.

John looked up . John looked up

If we follow Den Dikken (1995a) in assuming that the verbal particle up takes the DP the information/it as its internal argument, the order in which the DP precedes the particle must be derived by short object shift. Matters are somewhat complicated in this case, however, due to the fact that two orders are possible in (67a), which strongly suggests that short object shift of non-pronominal DPs is optional in this construction. I will provide an account for this in section 3.2.4.24 24

A final piece of evidence of a more ambiguous nature is provided by floating quantifiers. If floating quantifiers are licensed by A-movement operations like short object shift (section 3.1.2.3 will briefly discuss two proposals that derive this effect), we may expect that English direct objects can always be construed with a floating quantifier. The following example shows that this is indeed borne out for pronominal objects, but not for non-pronominal objects (Bobaljik 2003): I have read them/*the books all. At this moment I do not have an account for this contrast.

96 Chapter 2: Short object shift

2.3.5.

Conclusion

This section has investigated Haider’s and Barbiers’ claim that assuming an underlying OV-order is preferable to assuming an underlying VO-order, since in the latter case a set of phrasal movements must be assumed for which no a priori trigger is available. I have shown, however, that the same movements are also needed when we adopt Barbiers’ (2005) analysis of complex verb constructions. Haider’s (2003) analysis of complex verb constructions indeed does not need these movements, but at the same time requires the postulation of a set of verb movement operations that likewise lack a trigger. I therefore concluded that the OV-approaches do not really solve the trigger problem. Finally, I have argued that the object movement postulated by the VO-approach is empirically supported by (i) the fact that the exceptionally case-marked subject of an infinitival clause may enter into a binding relation with elements in the matrix clause, and (ii) the phenomenon of antecedent-contained deletion. From this I conclude that the VO-approach is superior to the OV-approaches. This of course does not imply that the trigger problem pointed out by Haider and Barbiers should not be taken seriously: Chapter 5 will take this issue up again. The present chapter will now continue with discussing short object shift in more detail. 2.4. Short object shift The previous section has argued in favor of Kayne’s (1994) version of the universal base hypothesis, according to which languages have a universal specifier-head-complement order, which implies that all languages have an underlying SVO order. It was further argued that the word order differences between the Germanic VO- and OV-languages are the result of a minor difference in the derivation. Both language types have the short object shift triggered by the ij-features on the verbal root V depicted in (68b), which inverts the order of the object and the verbal root V, but only in the VOlanguages is this inversion obligatorily undone by subsequent V-to-v movement as shown in (68c); the Germanic OV-languages lack this step in non-verb second contexts. (68)

a. b. c.

[VP V[uij] O[iij] ] [VP O[iij] (VP-adv) [V[uij] tO]] [vP S v+V [O[iij] (VP-adv) [ tV tO]]]

(base order) (ij-feature checking) (V-to-v)

Let us further provisionally assume that short object shift crosses the “low” VP-adverbs, such as manner and frequency adverbs, just as regular object shift crosses the “high” sentential adverbs (cf. section 2.4.1.3 for more

Short object shift 97

careful discussion). This accounts for the fact that, although Swedish does not have regular object shift of non-pronominal objects, these objects normally do precede the VP-adverbs. (69)

• Swedish a. Han he b. Han he

tvättar washes tvättar washes

gärna . (short object shift) the car gladly ju/inte . (*regular object shift) the car indeed/not

The fact that the direct object bilen ‘the car’ may also follow the VP-adverb gärne perhaps suggests that short object shift is optional, but this may be deceptive: Sells (2001:143) claims that these VP-adverbs can only appear in front of the direct object for “purposes of emphasis or due to other discourse related factors” (see Christensen, 2005:52, for a similar remark on Danish, and Koeneman, 2006:fn.7, for the claim that in Scandinavian VP-adverbs are put uniformly after the object). Consequently, the order ADV-DO may be the result of movement of the adverb into, e.g., a focus position. To indicate the marked nature of this order, I marked it in (69a) with a double cross. In the analysis below I will assume that short object shift applies obligatorily in all languages under discussion; more detailed study is needed, though, to fully corroborate this claim. 2.4.1.

Simple tense constructions

The derivation in (68) involves two movements: short object shift triggered by the ij-features on V, and V-to-v. This suggests that we have to postulate the two EPP constraints in (70a&b). As usual, their relative ranking with the economy constraint *MOVE in (70c) will determine whether the pertinent movement normally applies or not. (70)

Required constraints: a. EPP(ij): unvalued ij-features attract their goal. b. EPP(v): the verbal feature of v attracts its goal (the verbal root V). c. *MOVE: *t (don’t move).

2.4.1.1. The Germanic VO-languages: {EPP(ij), EPP(v)} >> *MOVE Let us start by discussing the derivation of the Germanic VO-languages. In principle, these can be derived in the three ways indicated in (71). The first option is to assume that the economy constraint *MOVE outranks the two EPP constraints; this predicts that short object shift and V-to-v are both blocked, and hence that both the verb and the object must follow the VP-

98 Chapter 2: Short object shift

adverbs. The second option is to assume that *MOVE is outranked by EPP(v) but not by EPP(ij); this predicts that the adverbs will intervene between the verb and its object. The third option is to assume that the two EPP constraints both outrank *MOVE; this predicts that both movements apply, and hence that both the verb and the object precede the VP-adverbs.25 (71)

• Germanic VO-languages (potential rankings) a. Option I: *MOVE >> {EPP(ij), EPP(v)} b. Option II: EPP(v) >> *MOVE >> EPP(ij) c. Option III: {EPP(ij), EPP(v)} >> *MOVE

The English examples in (72) show that the third option in (71c) is the correct one. For completeness’ sake, the evaluation of these examples is given in Tableau 1. Note that the data in (72) are not sufficient to determine the ranking of the constraints EPP(ij) and EPP(v). This is indicated in Tableau 1 by placing a dashed line between the two constraints. (72)

• Nominal complements a. *that John every day reads books. b. *that John reads every day tV books. c. *that John books every day reads tDO. d. that John reads books every day tV tDO.

Tableau 1: VO-languages (DP-complements)

[v adv [V O]] [v+V adv [tV O]] [v [O adv [V tO]]] [v+V [O adv [V tO]]]

EPP(ij)

EPP(v)

*> *!

*< *!

)

*MOVE * * **

Extraposition of PPs and/or sentential complements can now be reanalyzed as non-application of short object shift: since PPs and sentential complements do not have ij-features they are not attracted by V, and are hence predicted to remain in their base position following the VP-adverbs. 25 The use of curly brackets in (71) indicates that the ordering of the constraints is immaterial (as in the present case), or that there is insufficient evidence for the ranking of the constraints: {A, B} therefore means that either A >> B or B >> A is active, and that the two rankings select the same optimal candidate. The use of curly brackets is entirely different from that of the angle brackets which are used to indicate a constraint tie: A B means that both A >> B and B >> A are active, and that each of these rankings may result in the selection of a different optimal candidate (cf. fn.2 on p.19).

Short object shift 99

The evaluations of the examples in (73) and (74) are given in Tableau 2; the last two candidates are shaded to indicate that they cannot be generated by CHL because the movements they involve violate the Last Resort Condition, and are therefore not part of the candidate set. (73)

• PP complements a. *that John every day reads to his children. b. that John reads every day tV to his children. c. *that John to his children every day reads tPP. d. that John reads to his children every day tV tPP.

(74)

• Sentential complements a. *that John yesterday said that he will come. b. that John said yesterday tV that he will come. c. *that John that he will come yesterday said tV tCP. d. *that John said that he will come yesterday tV tCP.

Tableau 2: VO-languages (PPs and sentential complements) EPP(ij)

[v adv [V PP/S]] [v+V adv [tV PP/S]] ) [v [PP/S adv [V tPP/S]]] [v+V [PP/S adv [tV tPP/S]]]

EPP(v)

*MOVE

*! *!

* * **!

Observe that the fact that the PP in (73) may also precede the VP-adverbial every day is not yet accounted for. There are several potential solutions to this problem. For example, one might follow Haeberli (2002) in assuming that the categorial features of the PP can enter into a probe-goal relation (with V, in this case); I will not do this for reasons already discussed in section 2.1.2. There are also proposals according to which the word order variation is not due to movement of the PP, but leftward movement of the VP. For example, Barbiers (1995) suggests that complement PPs are generated VP-externally and that the postverbal placement of the PP is the result of VP-movement into the specifier of the PP. And Kayne (2000:part III) suggests that the preposition is not generated within the VP and that the PP is created in the course of the derivation; as in Barbiers’ work, the postverbal placement of the PP is the result of leftward VP-movement. Finally, one might want to claim that the two alternate orders depend on discourse-related factors (e.g., focus movement of the PP), and that we are therefore dealing with A′-movement of the PP. Since there is so much uncertainty concerning the proper analysis of (73d), I will put this issue

100 Chapter 2: Short object shift

aside for the moment, while concluding that at least the examples in (74) provide support for the proposed analysis. I will return to the problem concerning the PP in section 2.4.2.5, where I will argue that Kayne’s proposal is on the right track. 2.4.1.2. The Germanic OV-languages: EPP(ij) >> *MOVE>> EPP(v) Let us now turn to the Germanic OV-languages. Example (71) gives us five out of the six possible rankings of the constraints in (70). The ranking in (75) is the sixth possibility, and it is this ranking that gives rise to the properties that I have attributed to the OV-languages. (75)

Germanic OV-languages: EPP(ij) >> *MOVE >> EPP(v)

The subranking EPP(ij) >> *MOVE expresses that the ij-features of V trigger short object shift, whereas the subranking *MOVE >> EPP(v) expresses that there is (normally) no V-to-v. The evaluation in Tableau 3 shows that this ranking correctly predicts that the nominal complement in (76) is obligatorily placed in a position preceding the clause-final verb. (76)

a.

*dat Jan leest dat boek. that Jan reads that book b. dat Jan dat boek leest tDO.

Tableau 3: OV-languages (DP-complements) EPP(ij)

[v [V O]] [v+V [tV O]] [v [O [V tO]]] [v+V [O tV [tV tO]]]

*! *! )

*MOVE

EPP(v)

* * * **!

*

Extraposition of PPs and/or sentential complements can again be reanalyzed as non-application of short object shift: since PPs and sentential complements do not have ij-features, they are not attracted by V, and hence remain in their base position following the clause-final verb. This is shown in Tableau 4, in which the shading of the last two candidates again expresses that they cannot be generated by CHL and are therefore not part of the candidate set. The fact that the PP in (77a) may also precede the clausefinal verb is still not accounted for; cf. the discussion below Tableau 2.

Short object shift 101 (77)

a. b.

dat that dat that

Jan wacht . Jan for his father waits Jan zei . Jan that he ill was said

Tableau 4: OV-languages (PPs and sentential complements) EPP(ij)

[v [V PP/S]] ) [v+V [tV PP/S]] [v [PP/S [V tPP/S]]] [v+V [PP/S tV [tV tPP/S]]]

*MOVE

EPP(v)

* *! *! *!*

*

2.4.1.3. A puzzle: the placement of VP-adjuncts The analysis of the VO- and OV-languages given above seems relatively straightforward but there is a puzzle concerning the placement of the adverbial phrases that needs to be solved. Note that whereas short object shift must cross the VP-adverbs in VO-languages like English, this need not be the case in OV-languages like Dutch. This will become clear by comparing the English examples in (78), taken from Chomsky (1995a), to their Dutch counterparts in (79): in the English examples the shifted object must precede the adverbially used phrase every day, whereas in the Dutch examples it can either precede or follow elke dag. Note that I included the sentential adverb waarschijnlijk in the Dutch rendering of the English examples in order to exclude the possibility that (79b) is derived by means of regular object shift. (78)

a. *that John reads every day those books [VP tV tDO]. b. that John reads those books every day [VP tV tDO].

(79)

a. b.

dat Jan waarschijnlijk elke dag die boeken [VP leest tDO]. that Jan probably every day those books reads dat Jan waarschijnlijk die boeken elke dag [VP leest tDO].

I would like to suggest that the contrast between English and Dutch follows from the constraint NOVACM in (80), which penalizes string vacuous movement. This constraint can be considered a direct descendant of Chomsky’s (1986: section 9) Vacuous Movement Hypothesis, which, interestingly, was also considered to be violable (at least at LF). (80)

NOVACM:

Don’t apply internal merge when it is invisible in the output.

102 Chapter 2: Short object shift

Now, assume that the order of merging the adverbial phrase and other material adjoined to VP, on the one hand, and short object shift, on the other, is essentially free. If so, we can in principle derive the orders in (81) for VO-languages like English. Since in the surface order in (81b) there is no phonetically realized material between the object and its trace, the constraint NOVACM will block this order in favor of the order in (81a) where the adverb intervenes between the object and its trace. (81)

• VO-languages a. [V+v [O [adverb [tV tO]]]] b. [V+v [adverb [O [tV tO]]]]

(Merge adverb > Move O) (Move O > Merge adverb)

The structures that can be derived in OV-languages like Dutch are given in (82), which only differ from those in (81) in that V does not raise to v. As a result, NOVACM is satisfied in both representations in (82), due to the fact that verbal root V intervenes between the object and its trace. Consequently, both orders are expected to be possible. (82)

• OV-languages a. [v [O [adverb [V tO]]]] b. [v [adverb [O [V tO]]]]

(Merge adverb > Move O) (Move O > Merge adverb)

This, of course, does not mean that we expect that the selection of one of the two orders will always be entirely free; there may be other factors involved that favor one order over the other. For example, it is an old observation that the direct object must c-command, and therefore precede, a supplementive (=secondary predicate) in order to make secondary predication possible: while in (83a) the supplementive drunken ‘drunk’ can be predicated of either the subject or the object, in (83b) it can only be predicated of the subject. This shows that, when the object is the intended subject of the secondary predication, the c-command restriction on predication requires that the supplementive be merged before the application of short object shift. (83)

a.

b.

dat Marie de jongen dronken naar huis bracht. that Marie the boy drunk to home brought ‘that Marie brought the boy home, while she/he was drunk.’ dat Marie dronken de jongen naar huis bracht. that Marie drunk the boy to home brought ‘that Marie brought the boy home, while she/*he was drunk.’

The Dutch examples in (84) show that semantic factors may also be relevant: the choice between the two examples depends on the relative

Short object shift 103

scope of the object and the adverbial phrase. Example (84a) can only be interpreted with a wide scope reading of the adverbial phrase, so that Jan may read 14 different poems every week (or less if he rereads some of the poems); in example (84b), on the other hand, the object has wide scope, so that Jan is reading the same two poems over and over again. (84)

a. b.

dat Jan waarschijnlijk elke dag twee gedichten leest. (∀ > |2|) that Jan probably every day two poems reads dat Jan waarschijnlijk twee gedichten elke dag leest. (|2| > ∀)

We can account for these examples by introducing the LF constraint SCOPE in (85), which requires that relative scope corresponds to the asymmetrical c-command relations, hence linear order, of the topmost A-positions of the quantifiers (cf. Williams 2003 and the references cited there for slightly different proposals that are nonetheless comparable in spirit) or, to say it differently, that relative scope corresponds to linear order. We will see in section 4.2.1 that this constraint leaves open the possibility that, even in Dutch, scope and linear order diverge under certain circumstances. (85)

SCOPE:

relative scope of quantifiers corresponds to the hierarchical order of their topmost A-positions.

It is important to realize that assuming NOVACM in (80) does not entail that vacuous movement can never occur. Like all constraints, NOVACM can be violated when this is necessary to satisfy a higher ranked constraint. Consider, for example, the Dutch main clauses in (86). Since the verb has moved into second position in these examples, the object and its trace are no longer separated by any phonological material in (86a), and consequently we must conclude that this example violates NOVACM. This is allowed, however, when we assume that SCOPE outranks NOVACM, so that it is more important to reflect the relative scope of the adverb and the object in the linear order of the clause than to make short object shift visible. (86)

a. b.

Jan leest waarschijnlijk elke dag twee gedichten. Jan reads probably every day two poems Jan leest waarschijnlijk twee gedichten elke dag.

(∀ > |2|) (|2| > ∀)

The fact that the two orders may also occur in non-quantificational contexts suggests that NOVACM is ranked relatively low in Dutch, and that it is outranked by the constraints that enter into the determination of the relative order of the object and the VP-adverbs. The study of the constraints that determine the relative order of the object and the VP-adverbs in

104 Chapter 2: Short object shift

Dutch/German constitutes a research program in its own right, and may involve more pragmatic notions like “prominence in the focus field” (Broekhuis 2007a), which can perhaps be made more precise by appealing to accessibility and salience scales (Slioussar 2007). I will not digress on this subject any further but leave it to future research. 2.4.2.

Perfect tense constructions

This section briefly discusses short object shift in complex verb constructions like the perfect tense construction in (87). It is easy to show that short object shift does not apply only in clauses in the simple past or present, but also in constructions of this type. Consider the examples in (87). (87)

a. b.

The DA has proved [the defendants to be guilty] during each other’s trials. that John has read those books every day.

Example (87a) is the perfect tense counterpart of Lasnik’s example in (59b); since changing the past tense into the present perfect does not affect the acceptability of the example, we must conclude that in the perfect tense also, the subject of the infinitival clause undergoes short object shift into a higher position in the matrix clause from where it will c-command the anaphor. Example (87b) is the perfect tense counterpart of Chomsky’s example in (78b), and since the object must also precede the VP-adverbial in this example, we must again conclude that short object shift applies. Since relatively little is known about the structure of the perfect tense construction, the discussion below aims at establishing at least some of the basic properties of this construction in a step-by-step fashion. 2.4.2.1. The Germanic VO-languages: the constraint EPP(Asp) Since short object shift in (87) does not invert the order of the object and the participle, we must conclude that the original order is restored by movement of the participle into some higher functional projection comparable to v. Let us assume that this head is the aspectual head Asp. The derivation of the perfect tense construction is therefore globally as indicated in (88). (88)

a. b. c. d. e. f.

[Part O ] [O [Part tO ]] [Asp [O [Part tO ]]] [Part+Asp [O [tPart tO ]]] [aux [Part+Asp [O [tPart tO ]]]] [aux+v [taux [Part+Asp [O [tPart tO ]]]]]

Short object shift 105

The derivation starts out by merging the participle Part and the object, as in (88a). Since we know from the Romance languages that under certain conditions the participle and object may agree in ij-features, we can safely assume that the participle contains unvalued ij-features that probe the ij-features of the object.26 Given that we already know that EPP(ij) outranks *MOVE in the Germanic VO-languages, we predict short object shift to apply, as in (88b). In order to restore the VO order, we have to merge the aspectual head Asp, and move the participle into it, as in (88c-d). We may assume that this movement is triggered by some verbal feature on the functional head Asp, comparable to the verbal feature on the light verb v, and that the movement is forced by a strong ranking of the constraint EPP(Asp), which requires checking of the verbal features on Asp. After short object shift and V-to-Asp, the auxiliary is merged, as in (88e). We will see later that there are reasons to assume that the auxiliary is associated with a light verb v, as in (88f), and that the derivation continues from this point in the same way as in regular transitive constructions; I will not digress on this issue here, but return to it in section 3.1.2.2. 2.4.2.2. The Germanic OV-languages The derivation in (88) seems plausible for the VO-languages but we will see after the discussion of the derivation of the perfect tense construction in the Germanic OV-languages below that it is in need of further discussion. Consider the Dutch example in (89). (89)

Jan heeft waarschijnlijk dit boek gelezen. Jan has probably this book read ‘Jan has probably read this book.’

Since the object in this example occupies a position in between the sentential adverb waarschijnlijk ‘probably’ and the participle, we must conclude that in Dutch the object also undergoes short object shift. The fact that the participle follows the object suggests that it does not undergo head movement to Asp, and thus that the derivation is as indicated in (90), which 26

Recall that participle agreement in Italian and French is only possible when the theme argument precedes the main verb as the result of, e.g., NP-, clitic-, or whmovement; when the theme argument is in the regular postverbal object position participial agreement is excluded (see, e.g., Kayne 1985;1989). This could be accounted for by assuming that morphological object agreement is a reflex of the application of short object shift. From this assumption it would follow that short object shift normally does not apply in Italian and French and is only possible as an intermediate step of the movement operations mentioned earlier (cf. the more extensive discussion in section 4.3.1).

106 Chapter 2: Short object shift

differs from (88) only in that the participle and the auxiliary do not move into, respectively, Asp and v. (90)

a. b. c. d. e.

[Part O ] [O [Part tO ]] [Asp [O [Part tO ]]] [aux [Asp [O [Part tO ]]]] [v [aux [Asp [O [Part tO ]]]]]

The derivation in (90) is relatively straightforward but things are not as simple as they seem at first sight, and it can be readily shown that the derivation in (90) is in fact not correct. Ignore for the moment the questions of how and when the subject and the adverb waarschijnlijk are introduced in the structure. Assume further that example (89) is derived from (90d/e) by applying verb second to the auxiliary. If this really is the case, we expect that in the absence of verb second, the object would be sandwiched between the auxiliary and the participle. The examples in (91) show, however, that this expectation is not borne out, and that the object must precede the auxiliary verb.27 (91)

a.

dat Jan waarschijnlijk dit boek heeft gelezen. that Jan probably this book has read ‘that Jan has probably read this book.’ b. *dat Jan waarschijnlijk heeft dit boek gelezen.

This suggests that the derivational step in (90d) is not immediately followed by merging the light verb, but by an additional movement of the object into a position preceding the auxiliary. This means that the derivation is, rather, as indicated in (92). (92)

27

a. b. c. d. e. f.

[Part O ] [O [Part tO ]] [Asp [O [Part tO ]]] [aux [Asp [O [Part tO ]]]] [O [aux [Asp [t′O [Part tO ]]]]] [v [O [aux [Asp [t′O [Part tO ]]]]]]

It is hard to determine whether the order in (91b) is possible in any of the Germanic OV-languages because most of these languages require the participle to precede the auxiliary; the aux-participle order that can be found in Standard Dutch is the exception rather than the rule, and has long been promoted by prescriptive grammarians. Despite the fact that the order aux-participle is frequently used in colloquial speech, we should therefore wonder whether it should be considered part of core grammar; cf. section 5.3.2.2 for more discussion.

Short object shift 107

The Last Resort Condition on movement forces us to identify a trigger for the second movement of the object in (90e). What I want to propose is that this second movement is again triggered by unvalued ij-features, but this time not by those on the participle, but by those on the auxiliary verb. Perfective constructions thus involve agreement relations of multiple probes (the participle and the auxiliary) with a single goal (the object). That this type of multiple agreement exists and that auxiliaries can partake in it is clear from, e.g., the Italian passive construction in (93), where the DP Maria overtly agrees both with the passive participle and the passive auxiliary stata. (93)

Maria è stata accusata. Maria is beenfem,sg. beschuldigdfem,sg. ‘Maria has been accused.’

If the second leftward object movement in (90e) is indeed short object shift, that is, object shift triggered by the ij-features on the auxiliary, the ranking that we have established in section 2.4.1.2 for the OV-languages gives the correct result. I repeat this ranking in (94) in a slightly extended form in order to incorporate our findings concerning head movement of the participle into Asp. (94)

Germanic OV-languages: EPP(ij) >> *MOVE>> {EPP(v), EPP(asp)}

The strong ranking of EPP(ij) forces movement of the object into the local domains of the participle (step (92b)) and the auxiliary (step ((92e)), whereas the weak rankings of EPP(v) and EPP(asp) predict aux-to-v and Part-to-Asp not to take place. This gives us the derivation in (92). The evaluation is given in the Tableau 5. Note that I did not give the candidates with Part-to-Asp or aux-to-v. This is innocuous given that these head movements are blocked by the weak rankings *MOVE >> {EPP(v), EPP(asp)}: application of these head movements makes the resulting candidates worse by adding violations of the higher ranked economy constraint *MOVE. Tableau 5: OV-languages (complex tenses) EPP(ij)

... v [... aux [... Asp [... Part O]]] ... v [... aux [... Asp [O... Part tO]]] ... v [O ... aux [... Asp [tO... Part tO]]]

**! *! )

*MOVE * **

108 Chapter 2: Short object shift

2.4.2.3. The Germanic VO-languages again: the constraint H-COMPL Above we have seen that the derivation in (88), repeated here as (95), is a plausible approximation of the derivation of the perfect tense construction in the VO-languages. (95)

a. b. c. d. e. f.

[Part O ] [O [Part tO ]] [Asp [O [Part tO ]]] [Part+Asp [O [tPart tO ]]] [aux [Part+Asp [O [tPart tO ]]]] [aux+v [taux [Part+Asp [O [tPart tO ]]]]]

Now consider the constraint ranking in (96), the updated version of the ranking we had established for the VO-languages in section 2.4.1.1, in which it is assumed that EPP(asp) outranks *MOVE in order to account for the fact that the short object shift triggered by the participle in (95b) is followed by participle movement to Asp. (96)

Germanic VO-languages: {EPP(ij), EPP(v), EPP(asp)} >> *MOVE

The ranking in (96) indeed predicts the derivation to proceed as indicated in (95d-e). However, the strong ranking of EPP(ij) forces movement of the object not only into the local domain of the participle, but also into that of the auxiliary. This means that after merger of the auxiliary in (95e), which is repeated below, the derivation must continue with a second application of short object shift, as in (95f′). After this the light verb v is introduced and the strong ranking of EPP(v) forces movement of the auxiliary into v. Since aux-to-v crosses the shifted object it is wrongly predicted that the object will end up in between the auxiliary and the participle. (95)

e. f′. g. h.

[aux [Part+Asp [O [tPart tO ]]]] [O [aux [Part+Asp [t′O [tPart tO ]]]]] [v [O [aux [Part+Asp [t′O [tPart tO ]]]]]] [aux+v [O [taux [Part+Asp [t′O [tPart tO ]]]]]]

Of course, the fact that this problem arises, does not show that the approach that we are developing here is entirely on the wrong track, since we have to take into account the possibility that the strong ranking of EPP(ij) is overruled by some higher ranked constraint. In fact, the situation is slightly more complex since short object shift must be selectively blocked: the movement triggered by the participle should take place for the reasons discussed earlier, so that it is only the movement triggered by the auxiliary

Short object shift 109

verb that must be blocked. This is exactly what will happen when we assume that the order preservation constraint H-COMPL in (97a), which was already introduced and independently motivated in section 1.3.2.2, outranks the EPP constraint EPP(ij), as in (97b). (97)

a. b.

HEAD-COMPLEMENT ( H-COMPL): a head precedes all terminals originally dominated by its complement. Germanic VO-languages: H-COMPL >> {EPP(ij), EPP(v), EPP(asp)} >> *MOVE

The evaluation is given in Tableau 6. For simplicity, I have ignored the nine potential candidates that fail to apply V-to-v and/or Part-to-Asp. This is innocuous given that these head movements are forced by the strong rankings {EPP(v), EPP(asp)} >> *MOVE: non-application of these head movements makes the resulting candidates worse by adding violations of the higher ranked constraints EPP(v) and EPP(asp), while only reducing the number of violations of the lower ranked constraint *MOVE. Given this, the candidates in Tableau 6 involve the same number of violations induced by verb movement; for this reason the tableau only indicates the violations of *MOVE that result from short object shift (that is, we only count the object traces tO and not the verbal traces). Tableau 6: Perfect tense constructions in the VO languages HCOMPL

[aux+v [taux [Part+Asp [tPart O ]]]] [aux+v [taux [Part+Asp [O [tPart tO ]]]]] ) [aux+v [O [taux [Part+Asp [t′O [tPart tO ]]]]]]

*!

EPP

(ij) **! *

*MOVE (tO only) * **

2.4.2.4. The ranking of H-COMPL in the Germanic OV-languages This section concludes the discussion of short object shift in perfect tense constructions by investigation the ranking of H-COMPL in the Germanic OV-languages. I will show that the desired result follows when H-COMPL is outranked by the economy constraint *MOVE, as in the ranking in (98). (98)

Germanic OV-languages: EPP(ij) >> *MOVE >> {EPP(v), EPP(asp), H-COMPL}

The evaluation is as given in Tableau 7. First, note that the formulation of H-COMPL does not only disfavor object shift across the main verb but also object shift across the auxiliary: since the direct object is dominated by the

110 Chapter 2: Short object shift

complement of the auxiliary, AspP, it must follow the auxiliary. As a result, the third candidate in Tableau 7 does not have one but two violations of this constraint. Further note that, for simplicity, Tableau 7 does not give the nine potential candidates in which V-to-v and/or Part-to-Asp have applied. This is innocuous given that these head movements are blocked by the weak rankings *MOVE >> {EPP(v), EPP(asp)}: application of these head movements makes the resulting candidates worse by adding violations of the higher ranked constraint *MOVE, while only removing the violations of the lower ranked constraints EPP(v) and EPP(asp).

EPP(asp)

)

EPP(v)

*!* *!

H-OMPL

[v [aux [Asp [Part O ]]]] [v [aux [Asp [O [Part tO ]]]]] [v [O [aux [Asp [t′O [Part tO ]]]]]]

*MOVE (tO only)

EPP(ij)

Tableau 7: Perfect tense constructions in the OV languages

* **

* **

* * *

* * *

That *MOVE must outrank H-COMPL is clear from the fact that changing this ranking will exclude the third candidate; the ranking EPP(ij) >> H-COMPL >> *MOVE >> {EPP(v), EPP(asp)} will instead derive the aux-object-part order, as is shown in Tableau 8.

[v [O [aux [Asp [t′O [Part tO ]]]]]] [v [O [aux [Part+Asp [t′O [tPart tO ]]]]]] [aux+v [O [taux [Asp [t′O [Part tO ]]]]]] ) [aux+v [O [taux [Part+Asp [t′O [tPart tO ]]]]]]

**! **! * *

* * **!

* *

EPP(asp)

EPP(v)

*MOVE (tV only)

H-COMPL

EPP(ij)

Tableau 8: Perfect tense constructions in some hypothetical language

* *

Tableau 8 does not give the eight potential candidates in which short object shift fails to apply, but this is innocuous given that these movements are forced by the strong ranking of EPP(ij). Short object shift does, however, not give rise to the object-aux-part order, due to the fact that H-COMPL overrules the weak rankings of EPP(v) and EPP(asp); this requires that the

Short object shift 111

inversion that results from short object shift be made undone by V-to-v/Asp, if possible. As a result of this, short object shift will in a sense “push up” the auxiliary into v, which results in the aux-object-part order. Movement of the participle is, of course, blocked by the weak ranking of EPP(asp) given that V-to-Asp does not contribute to reducing the number of violations of H-COMPL. The fact that H-COMPL can evoke “push up” movement of the verb will be more extensively investigated in the following subsection. 2.4.2.5. Unifying EPP(v) and EPP(asp): *STRAY FEATURE Let me conclude the discussion of short object shift with two brief and more speculative remarks on the EPP constraints EPP(v) and EPP(asp). Although the postulation of these constraints is unproblematic given the general setup of the D&E framework, it seems justified to ask whether we really need them both. When we consider the rankings proposed above, it immediately becomes apparent that these two constraints are always tied: the rankings in (99) are not attested so far. (99)

a. b.

EPP(v) >> *MOVE >> EPP(asp) EPP(asp) >> *MOVE >> EPP(v)

Although it is not unthinkable that these are possible rankings, the fact that the two EPP constraints have the same effect of forcing movement of a verbal root into some higher verb-like head makes it plausible that we are actually dealing with a single constraint. Let us provisionally call this constraint EPP(v/asp), and assume that it requires that a verbal root amalgamate with the verbal features it is associated with. This means that we can simplify the rankings that we have established above as in (100). (100) a. b.

Germanic VO-languages: H-COMPL >> {EPP(ij), EPP(v/asp)} >> *MOVE Germanic OV-languages: EPP(ij) >> *MOVE>> {EPP(v/asp), H-COMPL}

This is a desirable simplification of the proposal since it enhances the explanatory adequacy of the theory by eliminating the rankings in (99) from universal grammar and thus delimiting the set of possible natural languages. I will therefore assume, until proven false, that EPP(v) and EPP(asp) are indeed instantiations of a single constraint EPP(v/asp). The discussion of short object shift above has shown that Scandinavian short object shift cannot cross the main verb. This claim is familiar from the literature on Scandinavian regular object shift, and is known as Holmberg’s Generalization (HG); cf. the earlier discussion in section 1.1.1. The leading

112 Chapter 2: Short object shift

idea in the literature so far has been that when movement of the main verb is blocked for some independent reason, object shift will be blocked as well. This idea has been implicitly adopted in the discussion above. However, this idea is by no means necessary: HG would in principle also be respected when object shift forces application of verb movement. So, in principle, there are the following two strategies to satisfy HG: (101) a. b.

Blocking strategy: failure to apply movement of the main verb results in failure of applying (short/regular) object shift. Push-up strategy: application of (short/regular) object shift forces movement of the main verb.

Now, consider again the ranking assumed for the VO-languages in (100a). It must be noted that this ranking can be subdivided into two subrankings that have more or less the same effect. The subranking EPP(v/asp) >> *MOVE straightforwardly forces movement of the verbal root into v/asp. However, the subranking H-COMPL >> EPP(ij) >> *MOVE also favors this head movement: it requires that the VO-order be restored after the obligatory application of short object shift. This is clear from the fact that the evaluations in Tableau 9 and Tableau 10 select the same candidate as the optimal one. The VO-languages of the hypothetical type B are languages that apply the push-up strategy: due to the weak ranking of EPP(v/asp), V-to-v normally does not apply, but short object shift will push the verbal stem V up into v in order to satisfy H-COMPL. Tableau 9: VO-languages, hypothetical type A (DP-complement) HCOMPL

[v adv [V O]] [v+V adv [tV O]] [v [O adv [V tO]]] [v+V [O adv [tV tO]]]

EPP

EPP

(ij) *> *!

(v/asp) *
> EPP(ij).28 Section 4.1 will argue that the two types of Germanic OV languages distinguished above indeed exist. In fact, it will be argued that EPP(v/asp) and EPP(tense) can be expressed by means of an even more general constraint *STRAY FEATURE in (102a), which requires that the verbal features in v/Asp and tense (and other potential functional projections in the extended projection of the verb) amalgamate with the verbal root. If this is tenable, type A would be a language like Icelandic, which has V-to-I movement in embedded clauses: the subranking *STRAY FEATURE >> *MOVE requires this movement in order to enable the tense features on I to amalgamate with the finite verb. Type B would be languages like Danish and Swedish that do not have V-to-I in this context due to the fact that the ranking MOVE >> *STRAY FEATURE disfavor this movement.29 (102) a. b.

*STRAY FEATURE: Amalgamate formal features of the functional heads with the root they are associated with. Germanic VO-languages: (i) Icelandic: H-COMPL >> EPP(ij) >> *STRAY FEATURE >> *MOVE (ii) Swedish/Danish: H-COMPL >> EPP(ij) >> *MOVE >> *STRAY FEATURE

Note in passing that I assume that EPP(ij) outranks *STRAY FEATURE in Icelandic, despite the fact that there is (at present) no evidence for the claim, in order to make the difference between the two rankings in (102b) as small as possible. It seems innocent to do this since these two constraints do not seem to conflict with each other. Although the unification of the constraint EPP(v/asp) and EPP(tense) is desirable for reasons of explanatory adequacy, there are serious problems with claiming that there are Germanic VO-languages of type B. The most important one is that the ranking in (102b,ii) predicts that in the type B 28

We will also select [v+V [O adv [tV tO]]] as the optimal candidate when we assume the ranking {EPP(ij), EPP(v/asp)} >> {*MOVE, H-COMPL}, but this of course wrongly predicts short object shift triggered by the ij-features on the auxiliary to be possible across the main verb, so that we derive the order aux-object-Perf. 29 The fact that regular object shift cannot apply in Danish and Swedish embedded clauses shows that it is not able to push-up the verb; it therefore differs in this respect from short object shift, which is able to push-up the verbal stem V to v/Asp in Danish and Swedish. I will return to this issue in section 4.1.

114 Chapter 2: Short object shift

languages movement of the verbal root is not necessary when there is no nominal object. This is shown in Tableaux 11 and 12, where the unhappy face indicates the wrong prediction; recall that, on the current assumptions, the third and fourth candidate violate the Last Resort Condition on movement and are therefore not even part of the candidate set, which is indicated by shadowing the full rows. Tableau 11: VO-languages, hypothetical type A (PP/S-complement) HCOMPL

[v adv [V PP/S]] [v+V adv [tV PP/S]] ) [v [PP/S adv [V tPP/S]]] [v+V [PP/S adv [tV tPP/S]]]

EPP

(ij)

*STRAY FEATURE

*MOVE

*! *!

*

* * **!

Tableau 12: VO-languages, hypothetical type B (PP/S-complement) HCOMPL

[v adv [V PP/S]] / [v+V adv [tV PP/S]] [v [PP/S adv [V tPP/S]]] [v+V [PP/S adv [tV tPP/S]]]

EPP

*MOVE

(ij)

*STRAY FEATURE *

*!

*! * *!*

*

There are several ways of circumventing the problem in Tableau 12. For example, the PP case can be solved when we assume, along the lines of Kayne (2000:part III), that complement PPs are not base-generated as the complement of V, but created later in the derivation: the preposition is merged externally to VP and acts as a probe attracting the DP-complement of V. Following Kayne, this movement should be followed by movement of the preposition to some higher functional head W. This part of the derivation is given in (103a-e). Since the order in (103e) violates H-COMPL, this constraint will force V-to-v, as in (103f). (103) a. b. c. d. e. f.

[VP V O] P [VP V O] [PP O P [VP V tO]] W [PP O P [VP V tO]] P+W [PP O tP [VP V tO]] V+v [WP P+W [PP O tP [VP tV tO]]]

Short object shift 115

This analysis has at least three advantages. First, we are now able to solve the problem noted in section 2.4.1 that complement “PPs” may precede the VP-adverbs in the Germanic VO-languages. Secondly, this analysis also accounts for the fact that complement “PPs” may precede the verbs in clause-final position in the Germanic OV-languages; since V-to-v does not apply in these languages the linear order in (103e) will surface. Thirdly, it solves the problem that we set out to solve, namely: Why is it that V-to-v also applies in Swedish and Danish when V takes a “PP”-complement? Note that the continuation in (103f) is not the one proposed by Kayne, who instead claims that the specifier position of WP is the landing site of the remnant VP, as in (103f′). Let us assume that this is also an option and that this gives rise to what is traditionally referred to as Extraposition, and thus derives the fact that in the Germanic OV-languages “PP”-complements can also follow the verb in clause-final position. (103) e. f′.

P+W [PP O tP [VP V tO]] [WP [VP V tO] P+W [PP O tP tVP ]]

An analysis similar to that for extraposition of “PPs” in (103f′) can be assumed to account for extraposition of sentential complements: extraposition of sentential complements should then be analyzed as intraposition, that is, leftward movement of the CP-complement into some higher specifier position, followed by leftward VP-movement (cf. Den Dikken 1995b; Kayne 2005: ch.11). This leftward VP-movement could then be seen as the result of H-COMPL, that is, the requirement that the verbal root V precedes its complement. We might even go further and assume that the complementizer that is like a preposition in that it attracts an IP complement of the verb into its specifier. This would give us a derivation that is formally identical to that of the PP cases. There are at least two additional arguments in favor of the derivation in (103). The first is more or less theory-internal: if PP-complements are just like DPs in being base-generated to the right of the verb, we would wrongly predict that extraction of so-called R-words from PP (Van Riemsdijk 1978) would strand the preposition in post-verbal position; cf. (104a). The fact that the stranded preposition must precede the verb follows straightforwardly, however, if the R-word daar must be moved via the specifier of WP, since this will block movement of VP into this position. (104) a.

b.

dat Jan daar niet wacht . that Jan there not for waits ‘that Jan does not wait for that.’ .... daar ... [WP t′daar op+W [PP tdaar top [VP wacht tdaar]]]

116 Chapter 2: Short object shift

Note that this proposal is essentially the same as the one given by Barbiers (1995), who assumes that the specifier of the complement PP, which is base-generated externally to VP, can be used either as an intermediate landing site of the R-word or as the landing site of VP. The second argument is that the derivation in (103) may contribute to solving a long-standing problem concerning remnant VP-topicalization (Broekhuis 2007a: fn.17; Haslinger 2007: section 4.5). Consider the examples in (105): (105a) shows that topicalized VPs may not contain a stranded preposition when the R-word daar is placed in the middle field of the clause; (105b) shows that VP-topicalization is possible when the stranded preposition is left behind in clause-final position (Den Besten & Webelhuth 1990). (105) a. *Op gewacht heeft Jan daar niet. b. Gewacht heeft Jan daar niet op. waited has Jan there not for ‘Jan has not waited for that.’

Under the traditional assumption that PP-complements are generated within VP, the expected judgments are rather the inverse from what is actually attested. If we are dealing with VP-topicalization in (105), we must then assume that the PP has been extracted from the VP before R-extraction takes place, as in (106a). However, since example (106b&b′) shows that leftward movement of the PP normally gives rise to a Freezing effect, we would expect (106a) to be impossible as well. (106) a. b.

[VP tPP gewacht] heeft Jan daar niet [PP tdaar op] Daar heeft Jan niet [VP [PP tdaar op] gewacht] there has Jan not for waited ‘Jan has not waited for that.’ b′. *Daar heeft Jan [PP tdaar op] niet [VP tPP gewacht]

However, the acceptability of example (105b) is expected when we adopt the derivation in (103): we first move the complement of the verb, daar, into the specifier of the PP, so that after movement of the preposition op to W and extraction of the R-word via the specifier of WP, the remnant VP can be moved in one fell swoop into SpecCP. (107) [VP Gewacht tdaar] heeft Jan daar niet [WP tdaar op+W [PP tdaar top tVP]]]

Of course the ungrammaticality of (105a) remains as mysterious as ever given that topicalization of WP is possible when the R-word daar remains

Short object shift 117

in the specifier of WP (Daarop gewacht heeft Jan niet), but at least we now have a straightforward account for the acceptability of (105b), which was lacking in the more traditional framework according to which the PP was base-generated within VP. The derivation suggested in (103), of course, raises a myriad number of questions that I am not able to address at this point. I therefore confine myself to pointing out that the suggested analysis is compatible with the proposal outlined in this section, and effectively solves the problem concerning Tableau 12, the problem raised earlier that complement PPs seemingly undergo short object shift, and at least one long-standing problem concerning VP-topicalization. Given this, I will take the optimistic view and assume that the three EPP constraints under discussion can indeed be reduced to one single constraint *STRAY FEATURE, and that there are indeed two types of Germanic VO-languages that are characterized by the constraint rankings in (108a,i) and (108a,ii), respectively. The ranking of the Germanic OV-languages is given in (108b). (108) a.

b.

Germanic VO-languages: (i) Icelandic: H-COMPL >> EPP(ij) >> *STRAY FEATURE >> *MOVE (ii) Swedish/Danish: H-COMPL >> EPP(ij) >> *MOVE >> *STRAY FEATURE Germanic OV-languages: EPP(ij) >> *MOVE >> {*STRAY FEATURE, H-COMPL}

2.5. Conclusion This chapter has argued that all Germanic languages have obligatory short object shift, that is, movement of the object triggered by the ij-features on the verbal root V. This is accounted for by assuming that these languages all have a strong ranking for EPP(ij): EPP(ij)>> *MOVE. The distinction between the VO- and OV-languages has been argued to follow from the fact that in the former the verbal root V obligatory moves into v/Asp, whereas in the latter it may stay in its base position. It was further suggested that two constraints are involved in forcing head movement in the VO-languages: H-COMPL and *STRAY FEATURE. In the VO-languages, (at least one of) these constraints outrank *MOVE and thus force V-to-v/Asp to undo the inversion that results from short object shift. In the OVlanguages these constraints are both outranked by *MOVE, and as a result V-to-v/Asp does not (normally) apply.

Chapter 3 Regular object shift

Section 2.2 has claimed that there are two types of object shift: short and regular object shift. Short object shift is triggered by the ij-features on the verbal root V and may cross the VP-adverbials, which are indicated in (1) by means of the lower-case abbreviation adv. The previous chapter has extensively discussed this type of object shift and argued that it obligatorily applies in all Germanic languages. Further, it has shown that the distinction between the Germanic OV- and VO-languages can be accounted for by assuming that in the latter the inversion of the verbal root V and its complement OBJ that results from short object shift must be made undone by subsequent V-to-v. (1)

• Short object shift a. [... v[case] [OBJ ...adv V tOBJ]] b. [...V+v[case] [OBJ ... adv tV tOBJ]]

(OV-languages) (VO-languages)

This chapter will discuss regular object shift, which is triggered by the case features on the light verb v and may cross the sentential adverbs ADV, as in (2a). As in the case of short object shift, the inversion that results from regular object shift must be made undone in the Germanic VO-languages by subsequent verb movement, more specifically V-to-I of the main verb; cf. (2b&b′). (2)

• Regular object shift a. [OBJi ... ADV v [t′OBJ ... V tOBJ]] b. [OBJi ... ADV V+v [t′OBJ ... tV tOBJ]] b′. [...V+v+I [OBJi ..tv [t′OBJ ... tV tOBJ]]]

(OV-languages) (VO-languages)

The requirement that a shifted object be preceded by the main verb in the Scandinavian languages is generally referred to as Holmberg’s Generalization (HG). Although there are several other restrictions on regular object shift, HG has certainly received most attention, and for this reason I will discuss in section 3.1 some proposals that try to derive this generalization. Section 3.1 in effect provides a brief overview of the most influential theories on regular object shift, and argues that a theory that attributes HG to factors external to CHL is to be preferred. In the D&E

120 Chapter 3: Regular object shift

framework adopted here this implies that we have to provide an optimalitytheoretic account of HG. Section 3.2 will continue with providing a full analysis of the restrictions on regular object shift, including HG. Holmberg (1986;1999) mentions the restrictions in Table 1. Table 1: Restrictions on the application of regular object shift

FinnishSwedish English Swedish Norwegian Danish Icelandic Dutch German

weak pronouns

DPs

across P

across IO

across V

—

—

n.a.

n.a.

n.a.

precedes non-verbal particles n.a.

— optional optional obligatory obligatory obligatory obligatory

— — — — optional optional optional

n.a. — — — — — —

n.a. — — — — — +

n.a. — — — — + +

n.a. — + + + + +

As I have already discussed in Chapter 1, the Scandinavian languages vary with respect to what kind of elements may undergo this shift: some of these languages have full regular object shift in the sense that they allow shift of both weak pronouns and non-pronominal DPs, some allow pronoun shift only, and some do not allow any form of regular object shift at all. This will be discussed in more detail in section 3.2.1. The fact that the complement of a preposition cannot undergo regular object shift was already discussed in section 2.1: since the case feature on the DP-complement of P is already valued by P, the complement is no longer active and therefore cannot be probed by the case feature on the light verb v, which triggers regular object shift.30 This restriction thus directly follows from the claim that regular object shift is triggered by the case feature on v, and it will therefore not be discussed any further in this chapter. A third restriction on Scandinavian regular object shift is that it cannot cross an indirect object. This also holds for Dutch but not for German. This will be discussed and analyzed in section 3.2.2. The fourth restriction on Scandinavian regular object shift is that it cannot cross the 30

This holds both in the traditional proposal, according which the PP is basegenerated within the VP, and under Kayne’s more recent proposal adopted in the previous chapter that the PP is formed later in the derivation by moving a DPcomplement into the local domain of a VP-external P.

Holmberg’s Generalization 121

main verb, which was already discussed above as HG. Let us provisionally assume, following Vikner (2006), that HG must be formulated as in (3). (3)

Holmberg’s Generalization (HG): regular object shift is only possible when the main verb moves into I.

From this formulation it follows, correctly, that regular object shift cannot apply in complex tense constructions or in embedded clauses without V-to-I in the Scandinavian languages. Dutch and German, on the other hand, do allow object shift in these constructions; for this reason I have placed a “+” in the column across V although regular object shift of course does not literally cross the main verb; only short object shift does. HG will be extensively discussed in section 3.1 and 3.2.3. Finally, Holmberg (1999) noticed that in Swedish a shifted object cannot precede a particle. This will be discussed in section 3.2.4. 3.1. Holmberg’s Generalization Most discussions on object shift in the literature (tacitly) consider HG as defined in (3) an inviolable condition on object shift. It is mainly for this reason that scrambling in Dutch and German is normally assumed not to involve the same operation as object shift: scrambling is possible in constructions in which the main verb fails to undergo V-to-I (see, e.g., Vikner 2006:396). This is illustrated in (4a) for an embedded clause and in (4b) for a perfect tense construction, both of which have the main verb in clause-final position. (4)

a.

b.

dat Jan < het boek> waarschijnlijk that Jan the book probably ‘that Jan is probably reading the book.’ Jan heeft < het boek> waarschijnlijk Jan has the book probably ‘Jan has probably read the book.’

leest. reads gelezen. read

Within the D&E framework the claim that HG is inviolable amounts to saying that it follows from some property of the generator (CHL), and it has indeed been claimed that this might be the case. In order to investigate this claim I will review some proposals in 3.1.1 and 3.1.2, and argue that these proposals cannot be maintained. Consequently, the fact that Dutch/German scrambling does not obey HG cannot be used as an argument for claiming that scrambling and object shift involve different types of movement (see section 2.1 for more discussion). Section 3.1.3 will continue with discuss-

122 Chapter 3: Regular object shift

ing an alternative set of theories that attribute HG to properties of the grammar external to CHL, and which all postulate some sort of restriction on linearization. 3.1.1.

Deriving HG from case theory

Holmberg (1986) gave the first comprehensive discussion of Scandinavian regular object shift within generative grammar. His account of HG was based on the following three presuppositions; cf. Holmberg (1986: section 6.4/7). (5)

a. b. c.

All and only heads of A-chains have case (except when they are headed by PRO). A verb in VP-internal position must assign case; a verbal trace assigns case optionally. Morphologically case-marked noun phrases may occur in a noncase marked position.

Assuming that regular object shift results in the creation of an A-chain, the clauses in (5a&b) predict that object shift across the main verb is excluded since this results in an A-chain in which the tail (and not the head) is marked for case. Apart from accounting for the optionality of object shift in Icelandic, clause (5c) also accounts for the fact that Scandinavian object shift of non-pronominal DPs is restricted to Icelandic: whereas Icelandic has morphological case-marking both on DPs and on pronouns, the mainland Scandinavian languages have morphological case-marking on pronouns only. Since clause (5c) goes explicitly against the Chomsky’s (1981) Case Filter, it does not come as a great surprise that Holmberg’s proposal has been reformulated such that it does comply with it. Vikner (1994:500) proposes that the verb need not assign case from its base position, but may also do that from its derived positions; consequently after V-to-I, the object can either be assigned accusative case in its base position by the trace of the verb, or in its shifted position by the verb in I. Since Vikner’s reformulation intends to eliminate clause (5c), we are no longer able to predict which elements may or may not undergo object shift, so we have to come up with a new alternative account for the pertinent facts. Vikner’s proposal crucially assumes that case must be assigned under government (to the right) and cannot be assigned under Spec-Head agreement: if the latter were possible, we would wrongly predict that the shifted object could be moved into SpecIP and thus precede the verb in I. Correspondingly, Vikner claims that the subject (in verb second languages at least) is assigned nominative case under government by C. Given that in

Holmberg’s Generalization 123

current minimalist theories government is no longer assumed to play a role, Vikner’s analysis cannot be maintained in these theories. Chomsky (1995a: ch.3) has argued, however, that HG can be derived by appealing to locality theory. This proposal will be discussed and simplified somewhat in the next section. 3.1.2.

Deriving Holmberg’s Generalization from Locality theory

Section 3.1.2.1 will review Chomsky’s (1995a: ch.3) proposal that HG is a theorem of the Minimal Link Condition (MLC) on movement: the MLC predicts that movement of the subject into SpecIP will be blocked by a shifted object unless the verb moves across the shifted object into I. If this is indeed the case, it is tempting to conclude that Vikner’s claim that Scandinavian regular object shift and Dutch/German scrambling are different types of operation is indeed correct. Recall that this claim was based on the fact that scrambling is possible in examples like (4), repeated here as (6), in which the main verb occupies a VP-internal position. (6)

a.

b.

dat Jan < het boek> waarschijnlijk that Jan the book probably ‘that Jan is probably reading the book.’ Jan heeft < het boek> waarschijnlijk Jan has the book probably ‘Jan has probably read the book.’

leest. reads gelezen. read

Section 3.1.2.2 will show, however, that it is readily possible to develop an analysis of (6a) that fully complies with the MLC. According to this analysis scrambling in (6a) only seemingly violates HG as formulated in (3), since the Germanic OV-languages do have V-to-I although this movement is not reflected in the visible word order. The fact that V-to-I is invisible can be attributed to the already independently established fact that in these languages V-to-v need not apply, so v-to-I can strand the verbal root V. Section 3.1.2.3 will propose an analysis of the perfect tense construction, from which it follows that scrambling in (6b) also satisfies the MLC. Since this implies that the MLC does not block object shift in the Scandinavian perfect tense constructions either, we need an additional constraint on object shift to derive HG. Given that this additional constraint can also account for the fact that Scandinavian object shift is blocked in embedded clauses without V-to-I, it actually makes the MLC entirely superfluous in accounting for HG. Section 3.1.2.4, finally, will provide the death blow to MLC-based accounts of HG by showing that there are empirical reasons to assume that

124 Chapter 3: Regular object shift

shifted objects are never able to block movement of the subject into SpecIP. I will therefore conclude that HG cannot be derived from the locality theory on movement, so we must develop another approach to deriving it. 3.1.2.1. The Minimal Link Condition Chomsky (1995a: ch.3) assumes that movement is constrained not only by the Last Resort Condition in (7a), but also by the Minimal Link Condition in (7b), and he further claims that the notion of closeness mentioned in (7b) should be defined in terms of c-command and minimal domains, as in (8). (7)

An element β can move into a landing site α in the local domain of a head H iff: (a) some unvalued formal feature on H takes β as a legitimate goal, and (b) β is the closest legitimate goal of H.

(8)

γ is closer to α than β in [α H [… γ … β]], iff: (a) γ c-commands β, and (b) α and γ are not in the same minimal domain.

For our present purposes, it suffices to say that movement of a Head Y into a higher head X, as in (9), creates a minimal domain for the chain (Y,t) that includes both SpecXP (= α) and SpecYP (= γ). Consequently, when both WP and (some element dominated by) ZP are potential goals of X, crossing of WP is allowed provided that the head movement takes place; when the head movement does not take place, WP will block the movement. In effect, this proposal relaxes Rizzi’s (1990) Relativized Minimality in configurations like (9), but only when the indicated head movement applies. (9)

XP

specifier

X′ X

YP WP

Y Y

′ ZP

Holmberg’s Generalization 125

Chomsky (1995a) has shown that the definition of closeness in (8) may account for HG. Assume for the moment that the (simplified) derivation of a transitive clause with regular object shift is as given in (10): the functional head I represents the combined heads AGRS and T, and V is simply an abbreviation of the amalgamation of the verbal root V and the light verb v. (10)

IP I′

SPEC1 I

AGROP AGRO ′

SPEC2 III

AGRO

VP Subj

IV I

V′ V

Obj

II

Chomsky further assumes that the nominal features on AGRO may take either the subject or the object as its goal. The clause in (8b) maintains that the object may move into SpecAGROP across the intervening subject in SpecVP if SpecAGROP and SpecVP are part of the same minimal domain. Since this is the case when V moves into AGRO (movement I), this predicts that object movement (movement II) is contingent on this head movement. The next step in the derivation is movement of the subject into SpecIP across the shifted object in SpecAGROP (movement IV). Since it is assumed that both the object and the subject are potential goals for the nominal features on I, this is only possible when SpecIP and SpecAGROP are part of the same minimal domain. Since this is the case when V+AGRO complex moves into I (movement III), this predicts that subject movement is also contingent on head movement. This means that HG actually consists of the two parts in (11).

126 Chapter 3: Regular object shift (11)

Holmberg’s Generalization (revised version): a. Object shift is contingent on movement of V-to-AGRO. b. In the object shift construction, movement of the subject into SpecIP is contingent on AGRO-to-I.

When object shift does not take place (overtly), the head movements in (11) are not necessary; movement of the subject into SpecIP does not cross any potential goal of I, so that domain extension is not required. Despite its apparent success, Chomsky has dropped the proposal above for conceptual reasons in later work. In the earlier proposal it was tacitly assumed that the AGR-heads are responsible for case-assignment. However, given that the verb itself determines whether accusative case is assigned or not, the question was raised whether these AGR-heads have any features at all. Chomsky (1995a: section 4.10) decided that they do not, and that they are therefore superfluous elements that should be eliminated from the theory. Chomsky did this by attributing case assignment to the light verb v, and by assuming that the unvalued case feature on v triggers movement of the object into an outer spec of vP, as in (12a). After object shift, the subject moves into SpecIP, as in (12b). (12)

a. b.

[vP object [v′ subject [v′ v [VP V tobject ]]]] [IP subject [I′ I [vP object [v′ tsubject [v′ v [VP V tobject ]]]]]]

Chomsky claims that in this theory, HG can no longer be derived. It seems likely, however, that this just results from the specific definition of closeness that he adopts. This definition is given in (13). According to (13b), movement of the object into the outer specifier of vP in (12) is admissible as both the inner and the outer specifier are part of the minimal domain of v (and also of the chain (V,t) when V-to-v applies). Further, movement of the subject across the object in the outer specifier of vP is allowed by (13c) as the subject and the shifted object in the outer specifier of vP are both in the minimal domain of v. (13)

γ is closer to α than β in [α H [… γ … β]], iff: (a) γ c-commands β; and (b) α and γ are not in the same minimal domain, or (c) γ and β are not in the same minimal domain.

Since neither V-to-v nor V-to-I movement is necessary to avoid a violation of the MLC on the definition of closeness in (13), HG can no longer be captured. It must be observed, however, that this is just due to the newly introduced clause in (13c): Kitahara (1997: section 2.3.3) has shown that

Holmberg’s Generalization 127

when we drop this clause and assume that the shifted object c-commands the subject, V-to-I is again required to license movement of the subject across the shifted object. Furthermore, there are various alternative ways of designing a theory that does not appeal to AGR-projections. For example, by adopting some version of Grimshaw’s (1991;1997) theory of extended projections we can do away with AGR, while maintaining that the derivation proceeds as indicated in (10), that is, without taking recourse to multiple specifiers. All we have to assume is that V has a case feature that must be checked by the object of the clause, and that in order to create the required checking configuration an additional projection must be created above VP; the checking relation is created by filling the empty head of the additional projection with V and moving the object into its specifier (see Ackema, Neeleman & Weerman 1992 and Nash & Rouveret 1997 for similar ideas). After V-to-I and movement of the subject into SpecIP, we end up with the representation in (14), which is virtually identical to the one in (10); the two structures differ only with respect to the labels of the nodes. For convenience, we may continue to think of eP as AGROP, but only in the sense of AGROP being an extended projection of V (in the same way as Grimshaw maintains the traditional labels in her work); we have to keep in mind, however, that after V has moved into the position e, eP will actually be a VP. (14)

IP I′

SPEC1 I

eP e′

SPEC2 III

e

VP Subj

IV I II

V′ V

Obj

128 Chapter 3: Regular object shift

An advantage of adopting this version of extended projections is that it makes a simpler and more natural formulation of closeness possible. 31 Since extended projections are introduced in order to create a checking configuration between a head H and some XP, the addition of these projections can be seen as part of the operation Move. For two reasons, an extended projection will never be generated for the specifier of a probing head H. First, since H does not c-command its specifier, it cannot probe it either. Secondly, given that checking of unvalued features may also result from Select (external merge), H and its specifier start out in a checking configuration, so that attraction of the specifier by H will be blocked by economy considerations (the constraint *MOVE), even when H c-commands its specifier at some later stage in the derivation. In short, extended projections only arise when H probes an XP in its c-command domain. Consequently, clause (8b) can be dropped, so that the definition of closeness becomes maximally simple: γ is closer to a head H than β is, iff γ c-commands β. This in fact makes the notion of closeness superfluous, so that we can simplify the definition of the MLC in (7b) as in (15b). (15)

MOVE: a head α attracts β iff: (a) β enters into a checking relation with a formal feature of α (or a formal feature of a head H adjoined to α), and (b) α cannot legitimately attract γ, where γ c-commands β.

The definition in (15), which reduces the MLC to “closest c-command”, is taken from Broekhuis (2000). Chomsky (2000:(40c)) arrives at the same 31

Another advantage is that the (Icelandic) transitive expletive construction (Jonas and Bobaljik 1993), which originally motivated the introduction of multiple specifiers, can be readily explained without taking recourse to the PF-operations invoked by Chomsky (1995a:354): in his analysis the expletive and the subject are, repectively, the outer and inner specifiers of I, so that the resulting word order expletive-subject-verb must be reordered at PF in order to derive the verb-second order expletive-verb-subject. The alternative account may take the following shape. First, the movement of the indefinite subject into SpecIP is triggered by the ijfeatures on I. However, the indefinite subject can value only a subset of these features, given that it is an NP and not a DP. Therefore, I must enter into an additional agreement relation with the expletive, which is a D by assumption. In order to create the required checking configuration, an additional projection must be created on top of IP: I moves into the empty head of this projection and the expletive is merged into its specifier. The expletive and the subject must be introduced in the indicated order if the expletive D has unvalued N-features that must be valued by the indefinite subject. This ensures that we derive the verbsecond structure in (i), without the aid of a phonological reordering rule. (i)

[IP EXPL [V-I] [IP subject tV-I [VP tsubject tV object]]]

Holmberg’s Generalization 129

formulation for the locality restrictions on Agree, but he maintains that clause (13c) continues to play a role (Chomsky 2000:(41)) in order to account for the fact that certain “intervention” effects are absent when γ and β are co-arguments. I will return to this in 3.1.2.4. 3.1.2.2. The MLC and object shift in embedded clauses in Dutch/German The discussion so far suggests that it is readily possible to formulate a plausible locality theory from which HG in (3) follows as a theorem. Given that Dutch/German scrambling apparently does not behave in accordance with this generalization, it is therefore tempting to accept Vikner’s claim that Scandinavian regular object shift and scrambling are different operations. This subsection will argue, however, that it is readily possible to develop an analysis of scrambling that fully complies with the MLC, and that the fact that scrambling violates HG as formulated in (3) is related to the already independently established fact that the Germanic VO- and OV-languages differ in that the former, but not the latter, have obligatory V-to-v; cf. sections 2.3 and 2.4. Consider again the problematic example in (6a), repeated here as (16), which shows that scrambling may occur when the finite main verb is in clause-final position: under the assumptions adopted so far the clause-final placement of the main verb means that the verb did not undergo V-to-I, and thus that scrambling violates HG in (3). (16)

dat Jan < het boek> waarschijnlijk leest. that Jan the book probably reads ‘that Jan is probably reading the book.’

However, scrambling in examples like (16) can readily be made compatible with the MLC when we follow Zwart (1997: ch.IV), who claims that, despite appearances, V-to-I does take place in Dutch embedded clauses. This movement is invisible, however, because it involves movement of the formal feature of the verb only, the phonological features being stranded for reasons of economy. Zwart’s argumentation that leads to this conclusion is given in (17). (17)

a. b. c. d.

Movement is triggered by the need for feature checking. Feature checking requires movement of formal features only. Economy considerations block movement of features that do not enter into a checking relation. Pied Piping of the phonological features is therefore blocked, if not needed for independent reasons.

130 Chapter 3: Regular object shift

From (17) it follows that I attracts the formal features of the verb only, and that the phonological features of the verb preferably remain in their base position. In order to account for those cases that do involve movement of the complete verb (i.e., of both the formal and phonological features of the verb), Zwart adopts the additional assumption in (18). (18)

Formal features must be phonologically supported at Spell-Out.

Now consider the main clause in (19a) and the embedded clause in (20a). At a certain stage in the derivation, the structure is roughly as in the b-examples (for the moment, we ignore issues that are related to the placement of the object and the subject). The unvalued tense feature on I attracts the interpretable tense feature of the verb, which will be abbreviated here as FF(V). In accordance with (17d), FF(V) moves to I in isolation, as in (19c) and (20c). The condition in (18) is, however, violated in the resulting structures. Consequently, the formal features FF(V) must look for an element that can phonologically support them. In the main clause in (19c), the only candidate is the verb itself, which must hence move and adjoin to FF(V), as in (19d). In the embedded clause in (20c), on the other hand, the formal features can move themselves and adjoin to the complementizer dat, as in (20d). (19)

a. b. c. d.

(20)

a. b. c. d.

Jan koopt het boek. Jan buys the book Jan I het boek koopt Jan FF(V)-I het boek koopt Jan koopt-FF(V)-I het boek tv dat Jan het boek koopt. that Jan the book buys dat Jan I het boek koopt dat Jan FF(V)-I het boek koopt FF(V)-I-dat Jan tFF(V) -I het boek koopt

Zwart claims that the derivation of embedded clauses in (20) is supported by the fact that in many Dutch and German dialects, the phenomenon of complementizer agreement occurs, which can be considered a morphological reflex of the movement of FF(V)-T to C. However, this claim was challenged by Van Koppen (2005: section 4.2.2), who has pointed out that complementizer and verbal agreement differ in ways that are not expected under Zwart’s proposal. This can be illustrated by example (21) from the Tegelen dialect. The crucial fact is that the complementizer may agree with

Holmberg’s Generalization 131

the first conjunct of the subject, which gives rise to the agreeing 2nd person singular form des, whereas the verb must agree with the full subject, which gives rise to the plural form treffe. If the formal features of the verb are really responsible for complementizer agreement, this difference between complementizer and verbal agreement is of course not expected. (21)

de-s [doow en ich] ôs treff-e. That2SG. [you2SG. and I1SG]1PL. each other1PL meetPL ‘that you and I will meet (each other).’

Another objection concerns the concept of feature movement itself. Assuming feature movement substantially enlarges the set of potential derivations, and thus raises a large set of new empirical questions (cf. also Chomsky 2000:119). For example, if the morphological reflex of movement is only due to condition (18), why does clitic movement in the Romance languages involve pied piping of the phonological features? Since the pronominal clitics are generally assumed to be adjoined to verbs, (18) should allow stranding of the phonological features. So the question is: Can we recast Zwart’s claim that V-to-I takes place in embedded clauses in the Germanic OV-languages without making use of feature movement? Broekhuis (2000) has argued that this is indeed possible. Recall from Chapter 2 that the Germanic VO-languages have obligatory V-to-v which undoes the inversion of the verbal root V and its complement OBJ that results from short object shift; cf. (22a). At a later stage in the derivation the functional head I is merged with the structure in (22a), as in (22b). Assume that the light verb v contains the tense feature that is probed by the unvalued tense feature on I. Now consider what happens when we provisionally assume that there is a constraint EPP(tense) that outranks *MOVE in Icelandic: since the strong ranking of EPP(tense) forces v-to-I, and since V has amalgamated with v earlier in the derivation, movement of v necessarily results in moving the complete V+v amalgam. (22)

a. b. c.

[...V+v [OBJ ... adv tV tOBJ]] [I [...V+v [OBJ ... adv tV tOBJ]]] [V+v+I [...tv [OBJ ... adv tV tOBJ]]]

(VO-languages)

The Germanic OV-languages differ from the VO-languages in that V-to-v need not apply, which results in the inverted OV-order in (23a). Now assume for the moment that, just as in Icelandic, EPP(tense) outranks *MOVE in Dutch and German. This again forces movement of v, but now the verbal root V need not be pied piped.

132 Chapter 3: Regular object shift (23)

a. b. c.

[... v [OBJ ...adv V tOBJ]] [I [... v [OBJ ...adv V tOBJ]]] [v+I [...tv [OBJ ...adv V tOBJ]]]

(OV-languages)

Of course, we now need some account for the fact that v-to-I movement pied pipes the verbal root in main, but not in embedded, clauses. We could do this by slightly adapting Zwart’s (1997) original proposal by assuming that v must be morphologically supported, and that this can be done by moving the v+I complex to C in embedded clauses, and by applying V-to-v prior to V-to-I in main clauses. However, if we take Van Koppen’s (2005) argument as valid evidence against this option, we must introduce some new means to account for this difference. We could do this by adopting a constraint LEXICALLY FILL TOP F, which requires that the highest functional projection in an extended projection be lexically filled: cf. Zwart (2001) for a comparable principle. This constraint would be satisfied by the complementizer dat ‘that’ in the embedded clause in (24a), but not in the main clause in (24b), so that V-to-v is forced to apply prior to V-to-v as a result; I refer to section 4.1 for more extensive discussion of this proposal. (24)

a. [CP dat [IP .. v+I [vP ... tv [VP ... V ...]]]] b. *[IP ... v+I [vP ... tv [VP ... V ...]]] b′. [IP ... V+v+I [vP ... tv [VP ... tV ...]]]

To avoid misunderstandings, note that the conclusion that Dutch and German have a strong ranking of EPP(tense) is dictated by the hypothesis that movement of the subject across the shifted object is otherwise excluded by the MLC. Since I will argue in the next two subsections that there are reasons not to adopt this hypothesis, nothing has been decided yet, and we will see in section 4.1 that there are actually reasons to assume that Dutch and German have a weak ranking of this constraint (or, rather, the more general constraint *STRAY FEATURE introduced earlier), just like, e.g., Danish. 3.1.2.3. The MLC does not block object shift in perfect tense constructions Now that we have seen that regular object shift in Dutch examples like (16) can be made compatible with the MLC, let us see whether the same holds for the perfect tense construction in (6b), repeated here as (25), which violates HG as formulated in (3) because scrambling applies while the main verb is still in its VP-internal position. (25)

Jan heeft < het boek> waarschijnlijk gelezen. Jan has the book probably read ‘Jan has probably read the book.’

Holmberg’s Generalization 133

In fact, there is not much known about the structure of perfect tense constructions. The derivation that I will propose here in a sense completes our earlier discussion of the perfect tense construction in section 2.4.2. Tha section concluded, on the basis of example (26a), that short object shift is not only triggered by the ij-features on the main verb, but also by the ij-features of the auxiliary; the strong ranking of EPP(ij) in Dutch then correctly predicts that the object must precede the auxiliary in clause-final position, and cannot be placed in between the auxiliary and the main verb. However, what I have not mentioned yet is that short object shift into the local domain of the auxiliary can be followed by regular object shift, as in (26b). (26)

a.

b.

dat Jan waarschijnlijk heeft gelezen. that Jan probably this book has read ‘that Jan has probably read this book.’ dat Jan dit boek waarschijnlijk heeft gelezen.

This suggests that the auxiliary not only has ij-features but can also assign accusative case, and hence must be associated with a light verb v. Given that it is normally assumed that light verbs with a case feature also have a specifier (= the external argument of the verb) the derivation of a perfective construction proceeds as indicated in (27). The steps in (27a-e) are those given in (92) in section 2.4.2, and involve the two applications of short object shift. The derivation is continued as in (27f) by merging the light verb v and the subject. From this point the derivation may continue as in a normal transitive construction: first the case feature on v may attract the direct object as in (27g); after this, I is merged and V-to-I allows the subject to move into SpecIP without violating the MLC; this results in the representation in (27h). (27)

a. b. c. d. e. f. g. h.

[Part O ] [O [Part tO ]] [Asp [O [Part tO ]]] [aux [Asp [O [Part tO ]]]] [O [aux [Asp [t′O [Part tO ]]]]] [S v [O [aux [Asp [t′O [Part tO ]]]]]] [O [S v [t′′O [aux [Asp [t′O [Part tO ]]]]]]] [S v+I [O [tS tv [t′′O [aux [Asp [t′O [Part tO ]]]]]]]]

The analysis in (27) has several properties to recommend itself (see Broekhuis & Van Dijk 1995 for more discussion). First, it is fully compatible with Hoekstra’s (1984) claim that past and passive participles should

134 Chapter 3: Regular object shift

receive a uniform analysis. It is also compatible with the proposal in Jaegli (1986) and Roberts (1987) that the passive participle morphology absorbs both the external argument and accusative case, and actually generalizes it to all participle forms. Furthermore it makes it possible to treat the perfect auxiliary HAVE as a regular transitive verb, just like the main verb have: its function is to reintroduce the agent and to assign accusative case to the internal argument of the main verb. The fact that the external argument is suppressed in the passive construction follows from the fact that the passive auxiliaries are unaccusative verbs and are therefore unable to reintroduce the agent and to assign accusative case to the object. Finally, it can be noted that the fact that in many languages participles of unaccusative verbs select the auxiliary BE is also compatible with the proposed analysis: since unaccusative verbs do not have an external argument and do not assign accusative case, selecting HAVE is not needed (although the question remains what the function of the participle morphology is in this case). The claim that the external argument is an argument of HAVE and not of the participle can also be supported empirically. Assume that floating quantifiers are base generated as part of the noun phrase they are associated with and that they are stranded by A-movement (Sportiche 1988). When we assume that the subject of a perfect tense construction is the external argument of the participle, we expect that a floating quantifier associated with the subject would follow the base position of the auxiliary verb HAVE. The analysis in (27), on the other hand, predicts this to be impossible. Now consider the Swedish examples in (28), taken from Holmberg (1999). Since Swedish has no V-to-I in embedded clauses, the two analyses make clear predictions about the placement of the floating quantifier alla ‘all’: the analysis that takes the subject to be an argument of the participle wrongly predicts (28a) to be acceptable and (28b) to be excluded; the analysis in (27), on the other hand, correctly predicts that the floating quantifier must precede the auxiliary HAVE. (28)

a.

*Jag undrar I wonder b. Jag undrar I wonder

varför why varför why

studenterna the-students studenterna the-students

inte har not have inte alla not all

alla all har have

åkt till Lund. gone to Lund åkt till Lund. gone to Lund

Note that in languages with V-to-I of (auxiliary) verbs in embedded contexts, like Icelandic, French, or English, the auxiliary may of course precede the floating quantifier. The English example in (29a) differs from the comparable Swedish example in (28b) in that the auxiliary has been moved across the floating quantifier and negation into I; cf. (29b).

Holmberg’s Generalization 135 (29)

a. b.

I wonder why the students have not all gone to Lund I wonder why the students I+have not all thave gone to Lund

Note further that Sportiche’s analysis of floating quantifiers is not essential for the argument to hold water: contrary to what Holmberg (1999:16) suggests, the same results follow under an adverbial analysis of floating quantifiers when we assume that the base position of the associate must be in the scope of –that is, be c-commanded by– the adverbial quantifier (see Doetjes 1997: ch.8). In fact, Doetjes’ proposal is probably even superior to the Stranding analysis as it does not invoke the unaccusative/passive problem discussed by Bobaljik (2003:117-8). Whereas the stranding approach wrongly predicts that the floating quantifier can be stranded in the base position of the internal argument and can hence be preceded by the main verb in these constructions, treating the floating quantifier as an adjunct correctly predicts that it precedes the main verb: He said that the students have arrived ; cf. Bobaljik (2003) for a more extensive discussion of the two approaches. From the discussion above I conclude that the derivation in (27) is well supported. It is important to stress that this implies that the impossibility of regular object shift in Scandinavian perfect tense constructions can no longer be derived from the MLC, so that at least some additional restriction is needed to block object shift in this construction. As we will see later, we can effectively obtain this by appealing to the linearization constraint HCOMPL, which was already introduced in section 2.4.2. However, since this constraint will also block object shift in embedded clauses without V-to-I, we must conclude that we actually do not need the MLC to account for the restrictions on Scandinavian regular object shift. Consequently, we must ask ourselves whether we need the MLC at all. This will be the topic of the next subsection. 3.1.2.4. Why the MLC cannot derive Holmberg’s Generalization Let us sum up the discussion so far. First, section 3.1.2.1 has shown that HG as reformulated in (11) can be derived as a theorem from the MLC. Section 3.1.2.2 has shown that it is possible to claim that in Dutch and German embedded clauses the light verb v moves into I (while stranding the verbal root in its base position), so that we can in principle maintain that scrambling is in fact the same as Scandinavian object shift. Thirdly, section 3.1.2.3 has argued that the MLC does not block object shift in perfect tense constructions and concluded from this that the Scandinavian ban on object shift in these constructions must follow from an additional constraint which blocks object shift across the main verb. Since this constraint also blocks

136 Chapter 3: Regular object shift

object shift in embedded clauses in which V-to-I does not apply, the MLC actually becomes superfluous in accounting for HG. The present section will argue that there are empirical reasons to assume that the MLC is not only superfluous, but is actually not able to derive HG. 3.1.2.4.1. Defective intervention effect The earlier discussion was based on the premise that the MLC can be fully reduced to “closest” c-command, that is, to clause (15b). Chomsky (2000: (40c)) arrives at the same formulation for the locality restrictions on Agree, but maintains his earlier clause (13c) in the form of (30). (30)

Terms of the same minimal domain are “equidistant” to probes.

We have seen earlier that (13c/30) makes it impossible to derive HG from the MLC: since the shifted object and the base position of the subject are arguably part of the same minimal domain, the former will never be able to block movement of the subject into SpecIP. It is therefore time to investigate the motivation for the claim in (30) in more detail. Chomsky (2000) claims that under certain circumstances (which will be made precise in 3.1.2.4.2) an Agree relation between a probe and a potential goal can be blocked by what he refers to as defective interveners, intervening potential goals that are inactive in the sense that all their features are valued. (31)

Defective Intervention (Chomsky 2000:123): (i) both γ and β match probe P in [… P [… γ ... β ...]]]; (ii) γ c-commands β; (iii) γ is inactive; and (iv) γ blocks the Agree relation between P and β.

The claim is based on Icelandic, where agreement between the finite verb (i.e., the inflectional node I) and the nominative argument is sometimes blocked by an intervening dative argument, which according to Chomsky must have its dative feature valued and therefore be inactive by the time that I is merged into the structure. This is illustrated by example (32), taken from Holmberg & Hróarsdóttir (2004); in this example the plural nominative phrase cannot trigger plural agreement on the finite verb, which must therefore appear with the default value singular.32 32 Holmberg & Sigurðsson (2006) show that there are actually three varieties of Icelandic: the variety described by Holmberg & Hróarsdóttir (which they call variety B) that exhibits dative intervention effects and which according to Ivona Kuþerová (p.c.) is spoken by the younger generation living in Reykjavik, a variety

Holmberg’s Generalization 137 (32)

það virðist einhverjum manni hestarnir vera seinir. there seemssg some mandat the horsesnom/pl be slow ‘The horses seem to some man to be slow.’

This section will argue that these facts indeed provide compelling evidence in favor of (30). However, I will argue that the dative phrase, which acts like a quirky subject, is actually still active at the time that I starts probing for a goal, hence, we are not dealing with defective intervention in the sense of (31).33 From this it will follow that in languages that do not have quirky subjects, the “defective” intervention effect does not arise. This is illustrated by the acceptability of Dutch examples like (33a), which is based on an actual sentence found on the internet and in which the finite verb and the nominative argument do agree in number despite the presence of the intervening dative phrase Jan/hem. That Agree is not blocked by the dative phrase in Dutch is also clear from the fact that the nominative phrase can be moved across the dative phrase into the regular subject position of the clause, as in (33b); the Last Resort Condition on movement implies that, in order to license this movement of the subject, I must enter into an Agree relation with the nominative phrase and, thus, that the dative phrase does not block this relation in Dutch. (33)

a.

b.

Daarom lijken Jan/hem de grafieken niet te kloppen. Therefore seempl Jan/himdat the chartsnom/pl not to be-correct ‘Therefore, the charts seem to be wrong to Jan/him.’ Daarom lijken de grafieken Jan/hem tde grafieken niet te kloppen.

The discussion in the next few subsections will ultimately lead to a slightly simplified version of the definition of closeness entertained by Chomsky (2000), and will provide a detailed analysis of the agreement facts in Icelandic and Dutch constructions with a dative intervener. A that is like Dutch in that it does not exhibit intervention effects at all, and a variety C that has no agreement at all. Since Chomsky’s proposal is based on the variety B, I will limit my attention to that variety as well. 33 Holmberg & Hróarsdóttir (2004) adopt a different definition of defective intervention (which they erroneously attribute to Chomsky): an intervener γ is defective if it cannot value the unvalued features of P but nevertheless blocks agreement between P and some other potential goal β. Under this definition the quirky subject is indeed a defective intervener since it does not trigger agreement on the finite verb. Neither Chomsky nor Holmberg & Hróarsdóttir discuss how nominative case is licensed on the DP hestarnir ‘the horses’. I refer to Sigurðsson (2003) and Woolford (to appear) for interesting proposals concerning case assignment in quirky subject constructions in Icelandic and Faroese; the two languages differ in that in the former the non-dative argument is assigned nominative case whereas in the latter it is assigned accusative case.

138 Chapter 3: Regular object shift

3.1.2.4.2. Agreement in Icelandic quirky subject constructions That something like (30) might indeed be correct is suggested by the agreement facts in Icelandic quirky subject constructions in (34) and (35) below, which have received a lot of attention in the more recent literature (Jónsson 1996, Sigurðsson 1996, Schütze 1997, Boeckx 2000, Chomsky 2000, Holmberg & Hróarsdóttir 2004, and many others). First consider the examples in (34), taken from Jónsson (1996:153). These examples show that the finite verb may agree with the nominative argument in quirky subject constructions when the dative and nominative are co-arguments, that is, arguments of the same verb. (34)

a.

b.

það líkuðu einhverjum þessir sokkar. there likedpl somebodydative these socksnom/pl ‘Somebody liked these socks.’ það voru einhverjum gefnir þessir sokkar. there werepl somebody given these socksnom/pl ‘Somebody was given the socks.’

The examples in (34) contrast sharply with those in (35), taken from Holmberg & Hróarsdóttir (2004:654), which show that agreement is blocked when the dative and the nominative are not co-arguments: in these quirky subject constructions, the dative argument is selected by the matrix verb, while the nominative is the subject of, respectively, a predicatively used adjectival phrase and an infinitival clause.34 (35)

a.

b.

það finnst/*finnast einhverjum stúdent tölvurnar ljótar. there findsg/findpl some studentdative the computersnom/pl ugly ‘Some student considers the computers ugly.’ það virðist/*virðast einhverjum manni hestarnir vera there seemsg/seempl some mandative the horsesnom/pl be seinir. slow ‘The horses seem to some man to be slow.’

The judgments on the examples in (35) are, of course, exactly what we expect when we assume that Agree is only possible under closest c-command: the dative argument asymmetrically c-commands the nominative 34

Holmberg & Hróarsdóttir (2004:654) note that the finite verb does appear in the plural form when both the dative and the nominative are plural. I will put this aside for the moment but return to it in section 3.1.2.4.5. Another fact that I will not discuss here is that first and second person subject pronouns are normally avoided in this construction; I return to this fact in section 3.1.2.4.6.

Holmberg’s Generalization 139

argument and is therefore a closer potential goal for the number feature on I. The examples in (34), on the other hand, show that the MLC cannot be fully reduced to closest c-command since this incorrectly predicts that agreement between the finite verb and the nominative argument would also be impossible in these examples. The facts in (34) therefore support some version of the claim in (30).35 Now consider the examples in (36), which are again taken from Holmberg & Hróarsdóttir (2004:653/5). These examples show that the intervention effect in (35) disappears when the dative argument is moved into clause-initial position (as was first noted by Schütze 1997). I will postpone the discussion of the fact that agreement is optional in (36a) to section 3.1.2.4.5. (36)

a.

b.

Mér finnst/finnast tölvurnar ljótar. medative findsg/findpl the computersnom/pl ugly ‘I consider the computers ugly.’ Mér virðist/virðast hestarnir vera seinir. medative seemsg/seempl the horsesnom/pl be slow ‘It seems to me that the horses are slow.’

The disappearance of the intervention effect can readily be accounted for under the assumptions in (37). The claim in (37a) that quirky subjects have an additional structural case feature, taken from from Jónsson (1996:146)/ Chomsky (2000:127), turns the dative argument into a potential goal for the nominal features on I, which can therefore move into SpecIP without violating the Last Resort Condition on movement. Since the movement in question is A-movement, the acceptability of (36) follows from assumption (37b), taken from Chomsky (2000:131). (37)

a. b.

Quirky Case is (ș-related) inherent Case with an additional structural Case feature. A-movement traces are “invisible” to the probe-associate relation.

35 It must be pointed out that the Icelandic data discussed here are not entirely clear: Holmberg & Sigurðsson (2006) claim that the speakers of the variety under discussion do not have agreement in mono-clausal contexts like (34) either, whereas others maintain that these speakers do have the judgments as presented in the main text; see Kuþerová (to appear) for a recent example. Given that Holmberg & Sigurðsson (2006) indicate that it is often not possible to draw clear distinctions between the three varieties, I will follow Kuþerová, who reports the judgments of a well-defined group of Icelandic speakers, viz., the younger generation living in Reykjavik. It must be noted, however, that if the lack of agreement in mono-clausal contexts reported by Holmberg & Sigurðsson (2006) is due to an intervention effect, this will cast a new light on the b-clause of the definitions of closeness that will be discussed in the main text.

140 Chapter 3: Regular object shift

Clause (37b) is also supported by example (38), again taken from Holmberg & Hróarsdóttir (2004), which shows that traces of A′-movement do not void the intervention effect; cf. section 3.1.2.4.5 for relevant discussion. (38)

Hvaða stúdenti veist þú að finnst/*finnast ti tölvurnar ljótar? the computers ugly which studentdative know you that findsg/findpl ‘Which student do you know considers the computers ugly?’

Note in passing that Holmberg & Hróarsdóttir also claim that wh-traces of dative arguments do not invoke an intervention effect for movement. This would of course be surprising if Agree is a prerequisite for movement. Their claim is weakened, however, by the fact that agreement is possible in the crucial example; cf. their fn.8 and the appendix to their article. We will return to the relation between intervention effects and A′-movement traces in section 4.3. The discussion above refutes our earlier assumption that the MLC can be fully reduced to closest c-command, and shows that we must complicate the definition of closeness by reintroducing the notion of minimal domain, as in (39b), and also by introducing further restrictions on the intervening element, as in (39c). Observe that we need the disjunction or, because (39b) accounts for the acceptability of agreement in (34), and (39c) for the acceptability of agreement in (36); agreement between the finite verb and the nominative argument is only blocked in quirky subject constructions when both clauses fail to apply.36 (39)

γ is closer to probe P than β in [… P [… γ … β]], iff: (a) γ c-commands β; and (b) γ and β are not in the same minimal domain, or (c) γ is not an A-movement trace.

36 Nunes (2004) provides four arguments in favor of the claim that A-movement traces do evoke intervention effects. The arguments in sections 2.3.3 and 2.3.4 of his book are crucially based on covert/feature movement and therefore do not carry over to the Agree-based framework adopted here. The argument in section 2.3.1, which is based on expletive it constructions, is furthermore voided when we adopt phase theory since the A-movement trace that allegedly invokes the intervention effect results from illicit A-movement across a CP-boundary (in one fell swoop from SpecTP of the embedded clause into SpecTP of the matrix clause). The final argument in 2.3.2 is based on subject raising constructions from French and English, and crucially rests on the assumption that (inherently case-marked) dative phrases do not evoke intervention effects. However, the comparable Icelandic subject raising constructions discussed in the main text unequivocally show that dative phrases (but not their A-movement traces) do in fact evoke intervention effects in this construction.

Holmberg’s Generalization 141

Of course, (39b) is identical to (13c) from Chomsky (1995a: ch.4), and we have seen earlier that it is this clause that makes it impossible to derive HG as defined in (3) from the MLC. Actually, the situation is even more hopeless than this for the locality approaches to HG, since there are reasons to assume that an “inactive” object can never act as a possible intervener for an Agree relation between I and some more deeply embedded nominative argument. I will show this in the remainder of this subsection. 3.1.2.4.3. Intervening datives in Dutch: the proper definition of closeness Chomsky (2000) has proposed that a goal must be active in the sense that it has at least one unvalued formal feature in order to act as a goal for some higher probe. Since a shifted object has its case feature valued by definition, it follows that it can no longer enter into an Agree relation with I. The null hypothesis therefore should be that it cannot block the Agree relation between the finite verb and the subject either. If so, clause (39c) should be replaced by clause (40c). The notion active in this clause is defined as in (41). (40)

γ is closer to probe P than β in [… P [… γ … β]], iff: (a) γ c-commands β, (b) γ and β are not in the same minimal domain, and (c) γ is active.

(41)

A goal γ is active, iff: (a) γ is the head of a (possibly singleton) A-chain, and (b) γ has an unvalued formal feature.

It is important to note that (40) makes use of conjunctions only, which should be considered a considerable improvement, given that Rizzi (1990) has argued that the disjunction or should be avoided in formalizations for conceptual reasons. Disjunctive statements are intrinsically unsatisfactory. Admitting a disjunctive formulation amounts to admitting that the nature of the generalization is not understood; if I write a principle saying that either property A or property B must be fulfilled, I am implicitly admitting that I do not understand the nature of the formal or functional equivalence holding between A and B. Of course a disjunctive formulation can turn out to be extremely productive, and even illuminating at certain stages of the comprehension of an issue [...] but the desideratum of avoiding disjunctions is an important one, even in the face of significant descriptive success. (Rizzi 1990:76-77)

142 Chapter 3: Regular object shift

Nevertheless, Chomsky (2000:123, 127-8) explicitly denies clause (40c) by referring to Icelandic examples like (35), where a quirky subject blocks agreement with the lower nominative argument. This blocking effect is, however, a typical property of the quirky subject construction, and does not occur in comparable examples in Dutch and German. First, consider the examples in (42). Since the dative argument Peter is preceded by adverbial material we may safely conclude that it is not in subject position, and hence occupies some position intervening between I and the nominative argument. The fact that the verb and the nominative phrase agree in person and number illustrates clause (40b): when the dative and the nominative argument are co-arguments, the former cannot block agreement with the latter. These examples are therefore completely parallel to the Icelandic ones in (34). (42)

a.

b.

Daarom bevallen waarschijnlijk Jan die sokken beter. therefore pleasepl probably Jandat those sockspl better ‘Therefore, those socks will probably please Jan more.’ Straks worden waarschijnlijk Jan die sokken aangeboden. later arepl probably Jandat those sockspl prt-offered ‘Those socks will probably be offered to Jan later.’

That agreement between I and the nominative argument is possible when the nominative argument and the dative phrase are co-arguments is also evident from the fact illustrated in (43), that the nominative argument can be moved across the dative argument into the regular subject position (SpecIP); see, e.g., Lenerz (1977), Koster (1978), Den Besten (1985), and Broekhuis (1992) for extensive discussions of the word order alternations in these examples. (43)

a.

b.

Daarom bevallen Jan beter. therefore please those socks Jandat better ‘Therefore, those socks please Jan more.’ Straks worden Jan aangeboden. prt-offered later are those socks Jandat ‘Those socks will be offered to Jan later.’

Now consider the examples in (44), in which the dative argument is selected by the raising verb lijken ‘to seem’, whereas the nominative argument is generated as the subject of a more deeply embedded predicate, the predicatively used adjectival phrase snel genoeg ‘fast enough’ and the infinitival clause te veel te drinken ‘to drink too much’, respectively. These examples show that the dative does not block the Agree relation that

Holmberg’s Generalization 143

establishes number agreement between the verb and the nominative, which is also clear from the fact that the nominative may be moved across the dative into the regular subject position of the clause. (44)

a.

a′. b.

b′.

Daarom leken niemand die computers snel genoeg. Therefore seemedpl nobodydat those computers fast enough ‘Therefore those computers seemed fast enough to nobody.’ Daarom leken die computers niemand snel genoeg. Soms lijken mij die jongens te veel te drinken. sometimes seempl medat those boys too much to drink ‘Sometimes those boys seem to me to drink too much.’ Soms lijken die jongens mij te veel te drinken.

The Dutch examples in (44) are similar to the Icelandic ones in (35) in that the dative argument and the nominative argument are not co-arguments. If Chomsky is correct in claiming that in such configurations the dative phrase blocks agreement between I and the nominative argument, we wrongly predict (i) that the finite verb does not agree with the nominative argument in person and number features, and (ii) that the movement of the nominative into SpecIP across the dative argument is blocked. We must therefore conclude that Icelandic and Dutch differ in that the dative phrases can only evoke intervention effects in the former language. The difference between the Icelandic examples in (35) and the Dutch examples in (44) can be readily accounted for by appealing to the c-clause of the definition of closeness in (40). First consider Dutch. By the time that I is merged, the unvalued features of the dative argument are already valued, and consequently this argument is not active when I probes for a goal. Since the dative argument is not active, the closest potential goal of I is the nominative argument, which accounts for the facts that the finite verb agrees with the nominative argument and that the latter can be moved into SpecIP. Icelandic crucially differs from Dutch in having a dative argument that functions as a quirky subject. According to (37a) quirky subjects have an additional structural case feature that is accessible to I, so we must conclude that the dative argument is still active by the time that I is merged. This makes the dative into a closer goal for I than the nominative argument, and it is therefore predicted that it will block agreement between I and the nominative phrase unless it is moved into SpecIP. The fact established above that inactive goals do not block the Agree relation between I and the nominative argument is the death blow to any attempt to derive HG from locality theory. Since objects have their case feature valued and are thus inactive by the time that I is introduced into the

144 Chapter 3: Regular object shift

structure, they cannot act as potential goals for I by definition, and consequently no blocking effect is expected. Hence, no V-to-I is needed in order to allow movement of the subject across the shifted object. This conclusively shows that we cannot account for HG by taking recourse to locality theory and must find new ways to account for this generalization. However, before we start looking into that, I want to complete the analysis of the (lack of) “dative” intervention effects by discussing three remaining problems that I have put aside so far. 3.1.2.4.4. Timing of feature elimination The discussion above presupposes a specific timing of inactivation/erasure of the valued uninterpretable features: since the in-situ quirky subject remains to block agreement between I and the lower nominative argument after its structural case feature is valued by I, we must assume that it is not inactivated before I has become inactivated as well. Similar timing issues arise in the Minimalist Inquiry framework where it is explicitly claimed that “[t]he probe-goal relation must be evaluated for the Minimal Link Condition at the strong-phase level [...]” (Chomsky 2001:27). However, Chomsky’s proposal that feature inactivation applies at the strong-phase level does not work for the Dutch constructions in (42) to (44) because the verbs involved in these constructions are unaccusative, and the light verb associated with unaccusative verbs is assumed not to induce a strong phase: consequently, the dative argument would still be active when I is merged to the structure and thus would block agreement between I and the nominative argument. An alternative possibility is to assume that features valued by a probe P are inactivated at the moment that P’s features are all valued and P is consequently inactivated itself. Following Chomsky’s (2000:132) suggestion in (45a), we may assume that inactivation of P must take place before the structure headed by P is merged with some higher head, which would also be in line with Chomsky’s (2001:15) interpretation of Pesetsky’s Earliness Principle in (45b). (45)

a. b.

Properties of the probe [...] must be exhausted before new elements of the lexical subarray are accessed to drive further operations. [P]erform computations as quickly as possible [...]. Thus, if local (P,G) match and are active, their uninterpretable features must be eliminated at once, as fully as possible.

If features valued by a probe P are indeed inactivated at the moment that P’s features are all valued, (45a) implies that the valued case features on the internal arguments are inactivated before vP is merged with I, so that the

Holmberg’s Generalization 145

dative arguments in (42) to (44) are indeed inactive by the time I starts probing for a goal. Chomsky (2001) weakens the claims in (45) by assuming that erasure of valued uninterpretable features takes place at the strong-phase level. This is related to his view that “case assignment [...] is a reflex of Agree holding of (probe, goal)”. The initial motivation for doing so lies in Chomsky’s (2001:17-8) discussion of Icelandic expletive passive participle constructions, which I review in a somewhat simplified form here. The core observation is that the passive participle agrees with the postverbal subject in case and agreement, which is illustrated for case in the following examples from Holmberg & Platzack (1995: section 1.32). (46)

a. b.

Það there Hún she

hafa verið [VP have been taldi hafa believed have

keyptir boughtnom verið [VP been

einhverjir bátar] some boatsnom keypta einhverja báta] bought some boatsacc

Under Chomsky’s assumption that case assignment is a reflex of agreement in ij-features, the case agreement on the participle implies that its unvalued ij-features cannot be erased at the VP level, that is, immediately after they are checked by the internal argument (the subject). The reason for this is that the participle and the internal argument cannot value each other case feature, and erasure of the ij-features of the participle at the VP level would make the participle invisible for the VP-external case assigner (the functional head I and the light verb v* associated with the ECM-verb in, respectively, (46a) and (46b)) with the result that the case feature on the participle “cannot be valued and the derivation crashes contrary to fact”. Consequently, Chomsky is forced to assume that erasure of the valued ij-features of the participle takes place no sooner than at the strong-phase level (v*P or CP). Since I did not adopt in this study the view that case assignment is a reflex of Agree in ij-features, this problem does not arise in my proposal, so I can maintain the claims in (45) in their strongest form, as is necessary to account for the Dutch data in (42) to (44): the unvalued features of the participle are erased as soon as they are valued by the internal argument, and before the projection of the participle is merged with a higher head. The fact that the proposed analysis of (42) to (44) forces us to make the derivation truly local in the sense of Collins (1997) by claiming that the operation of inactivating/erasing valued uninterpretable features is also subject to the syntactic cycle can be seen, I believe, as an important conceptual argument in support of my proposal.

146 Chapter 3: Regular object shift

3.1.2.4.5. More Icelandic agreement facts The discussion above has argued that agreement between the verb and the nominative is blocked in Icelandic examples like (35a), repeated here as (47), due to the fact that the intervening dative is still active by the time that I probes for a goal. (47)

það finnst/*finnast einhverjum stúdent tölvurnar the computersnom/pl there findsg/findpl some studentdat ‘Some student considers the computers ugly.’

ljótar. ugly

As was already indicated in footnote 34 on p.138, the discussion so far has ignored one important problem, namely that there are also examples in which the dative and the nominative are not co-arguments, but agreement can nevertheless occur across the intervening dative argument. An example of this is given in (48), again taken from Holmberg & Hróarsdóttir (2004), in which the agreement is optional. This problem will be tackled in this subsection. (48)

það finnst/finnast mörgum stúdentum tölvurnar the computersnom/pl there findsg/findpl many studentsdat ‘Many students consider the computers ugly.’

ljótar. ugly

Although Holmberg & Hróarsdóttir (2004) do not extensively discuss example (47), they suggest in their footnote 6 that agreement is only possible when the dative and the nominative argument have the same number marking. Given that the singular is the default marking of the verb, this singles out as special the case in which both the dative and the nominative are plural; only in this case can the finite verb be marked as plural. This accounts for the contrast between (47) and (48). A problem for Holmberg & Hróarsdóttir’s proposal, however, is that there are also cases that satisfy their criterion, but in which agreement is nevertheless impossible. This is illustrated by the examples in (49a) taken from Kuþerová (to appear). (49)

það finnst/*finnast fáum börnum tölvurnar few childrendat the computersnom/pl There findsg/findpl. ‘There are few children that find the computers ugly.’

ljótar ugly

The hypothesis put forth by Kuþerová is that the intervention effect is lifted when the dative phrase has undergone object shift. This immediately accounts for the following facts. First, the impossibility of agreement in (47) and (49) follows from the fact that object shift is not possible with noun phrases containing “weak” quantifiers like einhverjum ‘some’ and fáum

Holmberg’s Generalization 147

‘few’. Secondly, given that object shift is normally optional in the sense that it depends on the information structure of the clause, we derive that the plural marking on the verb in (48) is optional, and that it affects the interpretation of the clause. When the verb appears in the singular, the dative object mörgum stúdentum ‘many students’ must be in its non-shifted position and receive a non-specific interpretation as a result: it refers to a group of students that is not further specified. However, when the verb appears in the plural, the dative object must be in its shifted position and consequently it receives a specific interpretation: the dative refers to a group of students familiar from the discourse. Finally, it is predicted that the placement of the dative with respect to the adverbs will determine whether agreement will or will not arise. This is indeed the case: when the dative in (50) follows the adverb alltaf ‘always’ the finite verb must be singular. (50)

a. b.

það finnst There findsg það finnast There findpl

alltaf þremur börnum always three children (*alltaf) þremur börnum always three children

tölvurnar ljótar. the computersnom/pl ugly tölvurnar ljótar. the computersnom/pl ugly

Kuþerová’s proposal is descriptively successful but incompatible with the conclusions reached earlier in this paper. Since the dative object functions as a quirky subject, which can actually occur in the subject position, we must assume that even after object shift it still has an unvalued feature which can be probed by I. Consequently, we still expect it to behave as an intervener blocking agreement. The result can, however, be maintained in the present proposal by assuming that the movement of the quirky subject is not object shift but targets the subject position. A conclusive argument for assuming this is that the movement of the quirky subject does not obey Holmberg’s Generalization: example (51), cited from Kuþerová’s (to appear) article, shows that unlike object shift, the quirky subject can be moved across the main verb. (51)

það

voru konugi gefnar ambáttir í vinter. were kingdat. given slaves in winter ‘A king was given female slaves in winter.’ EXPL

Given the fact that Icelandic is a verb second language, the claim that the quirky subject targets the subject position amounts to saying that after this movement, the relevant structure is as indicated in (52). (52)

[CP það V+I+C [IP DPdat tV+I [ ... tdat … DPnom]]]

This structure makes clear that the head of the A-movement chain (DPdat, tdat) does not intervene between the I-position and the nominative argument,

148 Chapter 3: Regular object shift

and, consequently, the definitions in (40) and (41) correctly predict agreement to be possible. The fact that the dative appears to intervene between I and the nominative argument is due to the fact that the I+V amalgam moves into the C-position in order to satisfy the verb second requirement. For completeness’ sake, note in this connection that it is generally assumed that the expletive það does not occupy the subject position but is merged in SpecCP; like German es, but unlike English there, it functions as a kind of expletive topic (see, e.g., Vikner 1990, Holmberg & Platzack 1995, and Vangsnes 2002). This means that SpecIP is indeed available as a landing site for the quirky subject. Finally, consider again the examples in (36), repeated here as (53). Chomsky’s analysis, which was reviewed in section 3.1.2.4.2, presupposed that the dative pronoun is moved into the subject position, SpecIP. Given Zwart’s (1997) claim that subject-initial clauses in languages like Dutch and German can be IPs, this is indeed a plausible option for the quirky subjects in (53) as well. However, we clearly cannot rule out the possibility that the dative pronoun occupies the topic position SpecCP. If we assume that both options are available, the fact that agreement is optional in these examples will fall in place. When the dative pronoun mér occupies SpecIP, the trace tmér is an A-movement trace, which does not count as an active intervener according to the definitions in (40) and (41), and agreement between the verb and the nominative argument is predicted to be possible. However, when the dative pronoun is topicalized, the trace tmér is an A′movement trace, which does count as an active intervener, and, as a result, agreement between the verb and the nominative argument will be blocked in this case. 37 (53)

a.

b.

Mér finnst/finnast tmér tölvurnar ljótar. medat findsg/findpl the computersnom/pl ugly ‘I consider the computers ugly.’ vera seinir. Mér virðist/virðast tmér hestarnir medat seemsg/seempl the horsesnom/pl be slow ‘It seems to me that the horses are slow.’

37 A complication I did not discuss is that Taraldsen (1995:307&309) observes that some (but not all) speakers can also use the default, non-agreeing form in examples like (i), where the quirky subject and the nominative are co-arguments.

(i)

Henni leiddust/leiddist þeir. herdat bored3pl/bored3sg theynom ‘She was bored with them.’

Holmberg’s Generalization 149

This account of the (apparent) optionality of agreement in the examples in (53) presupposes that topicalization of the quirky subject may skip the subject position. This will follow from my proposal in section 4.3.1 that A′-movement must only be moved via accessible intervening A-positions when this is forced by phase theory. Kuþerová’s (to appear) account in terms of object shift, on the other hand, cannot be readily reconciled with the optionality of agreement in (53), given that object shift of pronouns is normally obligatory in Icelandic: Kuþerová’s proposal therefore wrongly predicts agreement to be obligatory. 3.1.2.4.6. The person constraint on the nominative argument So far I have also ignored the fact that quirky subject constructions exhibit a peculiar person restriction on the nominative argument: although, as we have seen, the verb may agree with the nominative in number, it does not agree with it in person. As a matter of fact the nominative can only be third person: first and second person nominative pronouns give rise to a degraded result (see, e.g., Sigurðsson 1996: section 2.5, Jónsson 1996:153, fn.36, and Sigurðsson & Holmberg 2006). This is illustrated for the first person plural and singular pronoun in the examples in (54), taken from Taraldsen (1995). (54)

a.

b.

Henni *leiðumst/?*leiddust/??leiddist herdat bored1pl/bored3pl/bored3sg ‘She was bored with us.’ ? Henni leiddist ég. herdat bored1sg me ‘She was bored with me.’

við. wenom

Note in passing that the judgments given on (54a) in Taraldson differ from those on similar examples in Sigurðsson (2003), which all get a star. Sigurðsson does not discuss examples like (54b), which, according to Taraldsen, “is significantly better than [(54a)] for some speakers (H.Thráinsson p.c.)”. What I want to focus on here is that Sigurðsson (2003) claims that the person constraint on the nominative is the result of an intervention effect invoked by the dative. His argument goes roughly as follows. Essentially following Taraldsen (1995), he adopts some version of the split-INFL hypothesis with separate Person, Number, and T heads (in that order). He proposes that the Person head enters into an Agree relation with, and triggers movement of, the dative argument; these are respectively the

150 Chapter 3: Regular object shift

operations I and II in (55). After this, the lower Number head may enter into an Agree relation with the nominative DP; this is operation III.38 (55)

PersP Pers′ Pers

NumP Num

TP T

vP

I II

v′

Dat v

III

VP V

Nom

Further, Sigurðsson assumes that the agreement between Person and the dative argument cannot be morphologically expressed on the verb due to the fact that the dative is already morphologically agreeing with the assigner of inherent case, which results in default third person agreement on the verb. Finally, Sigurðsson assumes that this default agreement on the verb, in turn, blocks person agreement with the nominative. In order for this argument to go through, we must assume that the dative acts as an intervener for the Agree relation between Person and the nominative argument, which would be unexpected given the definition of closeness in (40): first, the dative and nominative are co-arguments; 38 This raises the question, of course, why the Num head is not in an Agree relation with the dative argument as well, by which number agreement with the nominative argument should also be blocked. Taraldsen (1995) suggests that it is a general property of dative phrases that they do not enter into a number agreement relation, referring to Italian where dative clitics, as opposed to accusative clitics, do not trigger number agreement on the participle. However, since dative clitics in Italian and the quirky subject in Icelandic do not trigger morphological gender and person agreement either, we have to ask why we could not assume that dative arguments trigger default singular number agreement, just like they trigger default gender and person agreement. In my view, the question is therefore still open.

Holmberg’s Generalization 151

secondly, after movement of the dative into SpecPersP the intervening element is an A-movement trace and therefore inactive. This shows that Sigurðsson’s proposal is incompatible with our definition of closeness. Although the idea that the dative argument enters into an Agree relation with the unvalued person feature in the INFL-domain is fully compatible with our earlier claim that the quirky subject is active at the time that I enters the derivation, there are several problems with Sigurðsson’s account of the person restriction on the nominative. A first problem is that the proposed derivation is countercyclic since it is assumed that the dative moves into SpecPersP before the lower Number head can enter into an Agree relation with the nominative phrase, the dative being a potential intervener. This is, of course, an undesirable property. 39 Furthermore, it is actually not true that the dative must be A-moved in order for number agreement to be possible: we have seen in (34) that the quirky subject need not move into subject position to license number agreement when it is a coargument of the nominative. This considerably weakens the claim that quirky subjects may act as interveners in this context: if so, they are interveners for the person agreement relation only. A second problem for the proposal is that it is not cross-linguistically true that a goal cannot enter into a morphological agreement relation with multiple probes. This does occur in Italian passive, perfect unaccusative, and adjectival copular constructions in which the theme agrees both with the participle/adjective (in gender and number) and the auxiliary/copula (in person and number); cf. Burzio (1986), and example (3) in Chapter 5 for illustration. Sigurðsson’s claim that the default third person agreement on the verb is due to the fact that the quirky subject is already morphologically agreeing with the assigner of inherent case therefore seems rather stipulative (see Boeckx 2000 for an alternative proposal). Finally, and most importantly, Sigurðsson (2003) does not make fully explicit how the default person agreement on the verb forces the nominative argument to be third person. He claims that it follows “if, first, 3rd person is not “true” person [...] and, second, if the finite verb has to enter into a (3rd person) “defective agreement” relation with the subject and is thus blocked from agreeing in person with the nominative object” (p.260-1). Of course, it is perfectly true that under standard assumptions concerning feature valuing, the person feature of the verb cannot enter into an Agree relation with the nominative argument after it has been valued by the dative. However, it does not automatically follow from this that the non39

In Sigurðsson & Holmberg (2006), this problem is solved but replaced by an equally problematic proposal involving intrinsic rule ordering.

152 Chapter 3: Regular object shift

nominative must be third person: since the person feature on the nominative is inherently valued/interpretable, it need not enter into any Agree relation at all, so there is no reason to expect it to be restricted to the “spurious” third person. From this I conclude that Sigurðsson’s proposal actually does not have the intended result, so we have to find some alternative explanation for the person constraint.40 Consequently, this constraint does not provide evidence against the definition of closeness proposed in (40). 3.1.2.4.7. Defective intervention does not exist This subsection has discussed the complicated agreement patterns in Icelandic quirky subject constructions and concluded that the difference between Icelandic and Dutch is due to the fact that the quirky subject is not defective in the sense intended by Chomsky (2000); the quirky subject, but not the corresponding dative phrase in Dutch, still has an active feature by the time that I is merged. In the end, therefore, it turns out that defective intervention in the sense of (31), repeated here as (56), does not exist. (56)

Defective Intervention (Chomsky 2000:123): (i) both γ and β match probe P in [… P [… γ ... β ...]]]; (ii) γ c-commands β; (iii) γ is inactive; and (iv) γ blocks the Agree relation between P and β.

Further, the discussion has shown conclusively that we must incorporate statement (30), according to which terms of the same minimal domain are “equidistant” to probes, into the definition of closeness, which makes it impossible to derive HG from locality theory. 3.1.2.5. Conclusion This section has discussed several attempts to derive HG from the locality theory of movement. Although Chomsky’s (1995a: ch.3) original proposal based on the MLC discussed in 3.1.2.1 and 3.1.2.2 seemed very promising, section 3.1.2.3 has shown that the MLC is actually superfluous. Section 3.1.2.4, finally, showed that the MLC in (57a) is too strict, and that the definition of closeness must be relaxed by adding the clauses in (57b&c). 40

I have no new insights to offer here. See Taraldsen (1995) and Boeckx (2000) for alternative proposals. Boeckx’s proposal, which tries to relate the person restriction on the nominative argument in quirky subject constructions to the person case restriction that we find in the Romance languages, is especially interesting as it does not attribute this constraint to core syntax (which, as we have argued above, does not seem to block first and second person nominative arguments in quirky subject constructions) but to morphology (the PF component).

Holmberg’s Generalization 153 (57)

γ is closer to probe P than β in [… P [… γ … β]], iff: (a) γ c-commands β, (b) γ and β are not in the same minimal domain, and (c) γ is active.

First, Chomsky (2000) has argued that the agreement facts in Icelandic quirky subject constructions like (34/35) show that Agree of probe P and goal β can only be blocked by a goal γ when γ is not a co-argument of β; this motivates clause (57b). Secondly, I argued, contra Chomsky (2000), on the basis of Dutch nominative-dative constructions in (44) that inactive goals never block probe-goal relations; this motivates clause (57c). According to (57), a shifted object is not an intervener for the Agree relation between I and the subject for two reasons: (i) the shifted object is arguably in the same minimal domain as the base position of the subject, and (ii) it has its case feature valued and is thus inactive at the time that I is merged to the structure. Consequently, V-to-I is not needed to allow movement of the subject across a shifted object, and HG can therefore no longer be derived from the MLC. From this it follows that we have to find alternative ways to account for this generalization 3.1.3.

Deriving HG from factors external to CHL

If HG as formulated in (3) cannot be derived from the MLC, it seems plausible to assume that the restrictions on Scandinavian regular object shift are due to factors external to the computational system CHL. In recent years various proposals have been put forth to that effect, starting with Holmberg (1999). Holmberg argued that the formulation of HG in (3) is overly permissive since it only blocks regular object shift across a main verb, whereas in actual fact this movement is blocked by a wider range of elements, such as other arguments and verbal particles (see the discussion of Table 1 on p.120). Since object shift is not blocked by the traces of these elements, Holmberg argues that object shift is a PF- (or at least a postsyntactic) operation. Consider the contrast in (58). (58)

a.

b.

Jag har inte kysst . I have her not kissed ‘I haven’t kissed her.’ Kysst har jag henne inte. Kissed have I her not

Holmberg claims that analyzing object shift as a syntactic operation leads to a violation of the strict cyclicity condition. The argument goes roughly as

154 Chapter 3: Regular object shift

follows (cf. section 4.3.3 for a more careful discussion and a critical appraisal of the proposal). First, example (58a) shows that object shift cannot apply across the main verb when it is in situ. Holmberg concludes from this that (58b) can only be derived by first moving the participle into SpecCP, as in (59a), and subsequently applying object shift, as in (59b). The derivation in (59) is, however, countercyclic and thus violates Chomsky’s extension condition on movement. (59)

a. b.

[CP kysst [C′ har jag inte [VP tkysst henne]]] [CP kysst [C′ har henne inte [VP tkysst thenne]]]

When we assume that object shift is a post-syntactic operation, this problem is solved because the extension condition does not apply to post-syntactic operations. The contrast in (58) can now be accounted for by assuming that object shift is subject to a general condition that it cannot apply across phonologically visible VP-internal material (except adjuncts): the representation of (58a) prior to object shift is (60a), and hence object shift in (60a′) is blocked; the representation of (58b) prior to object shift is (60b), and hence object shift in (60b′) is allowed. (60)

a. jag har inte [VP kysst henne] a′. *jag har henne inte [VP kysst henne] b. kysst har jag inte [VP kysst henne] b′. kysst har jag henne inte [VP kysst henne]

As Holmberg himself has pointed out, regular object shift is subject not only to the condition that it cannot cross phonological VP-internal material, but also to the semantic condition that it applies to presuppositional material only. Chomsky correctly points out that phonological rules are not expected to have semantic effects (2001:15), and we have seen in section 1.1 that this led him to replace Holmberg’s proposal by the postulation of what is essentially an output filter. Section 1.3 has further argued that this output filter can be decomposed into a set of more primitive constraints. Sections 3.2.1-3.2.4 will provide an optimality-theoretical account of regular object shift in the Germanic languages based on these constraints. Since the core of the proposal was already discussed in Chapter 1, I will not elaborate on this further here. To conclude this subsection I want to briefly discuss two alternative proposals that resemble the proposals mentioned in this subsection in that they do not take recourse to the computational system. The first is by Bobaljik (1995;2002), who claims that HG is the result of a morpho-phono-

Holmberg’s Generalization 155

logical constraint on verb inflection that requires merger of the inflectional morphology in I and the finite verb under PF-adjacency. Consider the Swedish example in (61a). When object shift does not apply, the structure of the embedded clause is as given in (61b): the past morpheme -te is adjacent to the verbal root läs- and therefore merger of these elements is possible. When object shift does apply, the structure is as given in (61c): the shifted object den now intervenes between the past tense morpheme -te and the verbal root läs- and therefore merger of these elements is blocked. Note that, as in Holmberg (1999) and Chomsky (2001), adverbs are assumed not to count for PF-adjacency. (61)

a. b. c.

Det är troligt [att de läste ] it is probable that they it read att [IP de [I -te] [VP läs- den]]] att [IP de [I -te] den [VP läs- tden]]]

Bobaljik accounts for the fact that Dutch/German scrambling may apply in embedded clauses by assuming that Dutch and German are head-final languages, and consequently that I is placed on a right branch. As is shown in (62) object shift does not affect the adjacency of the verb and the present tense morpheme -t, and can therefore freely apply. (62)

a. b. c.

Hij zei [dat hij waarschijnlijk koopt] he said that he the book probably buys dat [IP hij [VP het boek koop-] [I -t]] dat [IP hij het boek [VP thet boek koop-] [I -t]]

A property of Bobaljik’s morpho-phonological account is that, like the locality approaches discussed earlier, it has nothing to say about the other restrictions pointed out by Holmberg, so that alternative accounts must be developed for those cases. A second drawback is that it forces us to drop the universal base hypothesis, since it crucially relies on the fact that the functional head I precedes the VP in the Scandinavian languages, whereas it follows the VP in Dutch and German. A third, and in my view conclusive, argument against Bobaljik’s proposal is more empirical in nature. Holmberg (1999:6-7) has argued that Bobaljik’s (1995) proposal incorrectly predicts that in main clauses, object shift is possible in perfect tense constructions. First, consider the Swedish examples in (63), which show that, whereas negation appears in front of the auxiliary in embedded clauses, it follows the auxiliary in main clauses. This can be accounted for by assuming (i) that negation is base-generated to the left of the auxiliary, and (ii) that the order is reversed in main clauses as the result of V-to-I.

156 Chapter 3: Regular object shift (63)

a. b.

att Per inte har kysst henne. that Per not has kissed her Per har inte taux kysst henne. Per has not kissed her

Now consider the ungrammatical example in (64a). The structure of this example should be approximately as given in (64b): the object pronoun is shifted in front of negation, and the auxiliary is moved to I. As a result of the second movement, I and the auxiliary are PF-adjacent: Bobaljik’s proposal therefore wrongly predicts (64a) to be acceptable. (64)

a.

*Per har henne inte taux kysst. Per has her not kissed b. [IP Per [I′ har + I] henne inte [auxP taux [PartP kysst thenne] ]]

Bobaljik (2002:fn.25) discusses this problem, and claims that example (64a) is ungrammatical because the pronoun has simply raised too high: he argues that the target position of object shift is the specifier of an AGRO projection associated with the main verb, which is therefore situated to the right of the auxiliary. This is, however, at odds with our earlier conclusion that the case feature on the light verb v associated with the auxiliary may trigger regular object shift in Dutch and German (cf. the discussion of (27)) as a result of which the shifted object is placed in a position to the left of (the base position of) the auxiliary. I therefore conclude that Holmberg’s objection still stands. Sells (2001) provides an OT-account of object shift in Swedish that makes extensive use of alignment constraints, which are very effective for expressing word order generalizations; cf. section 1.3.2.2 for some references. For example, the fact that the subject precedes the objects and the indirect object precedes the direct object in the Germanic languages can be readily expressed by ranking the three alignment constraints SU-L, IO-L and DO-L, which require that the respective arguments be aligned as closely to the left edge of the clause as possible, as in (65a). Now by assuming the alignment constraint F-PRED-L, which requires that the predicate be aligned as closely as possible to the left edge of the clause, the ranking in (65b) expresses that the objects (but not the subject) must follow the verb. Consequently, when the verb fails to undergo V-to-I, object shift will be excluded. (65)

a. b.

SU-L SU-L

>> IO-L >> DO-L >> F-PRED-L >> IO-L >> DO-L

Holmberg’s Generalization 157

The effects of the alignment constraint in (65) are comparable to those of the two order preservation constraints H-COMPL and RELMIN. However, since there are fewer conceivable order preservation constraints than alignment constraints, considerations of explanatory adequacy favor the former over the latter. 3.1.4.

Conclusion

This section has reviewed a number of earlier analyses of regular object shift. A main trend has been to assume that HG as defined in (3) can be derived from certain inviolable syntactic conditions, such as the case theory (Holmberg 1986 and Vikner 1994) or the locality theory on movement (e.g., Chomsky 1995a: ch.3). Since Dutch/German scrambling apparently does not obey HG, it is often concluded that Scandinavian object shift and scrambling are different operations. This section has argued, following Holmberg (1999) and Chomsky (2001), that the presupposition that HG is a theorem of some inviolable syntactic condition is wrong, and that we must rather appeal to factors external to CHL. As a result, it no longer follows that Scandinavian object shift and Dutch/German scrambling must be distinct operations. This does not come as a big surprise since section 2.1 already gave conclusive evidence in favor of the claim that object shift and scrambling (of the A-movement type) can be identified. 3.2. The restrictions on regular object shift The previous section has argued that the restrictions on Icelandic object shift (including HG) are not due to properties of the generator. Given the D&E framework outlined in section 1.3, this implies that the restrictions on Scandinavian object shift must be derived in an optimality-theoretic manner. The main ingredients of a theory that can derive the restrictions on regular object shift have already been introduced in section 1.3.2. For example, we have seen there that the three constraints in (66) enable us to distinguish three basic language types in Figure 1: (i) languages like Icelandic that have regular object shift with all types of noun phrases, (ii) languages like Danish that have regular object shift with pronouns only, and (iii) languages like French that have no regular object shift at all. (66)

• Constraints needed for deriving the basic typology w.r.t. object shift: a. EPP(case): an unvalued case-feature attracts its goal. b. *MOVE: *t (do not move). c. D-PRONOUN: a weak/definite pronoun must be VP-external.

158 Chapter 3: Regular object shift

*MOVE >> EPP(case)

EPP(case) >> *MOVE

No full object shift

Full object shift: Icelandic

D-PRONOUN >>*MOVE

*MOVE >>D-PRONOUN

Pronoun shift: Danish

No object shift: Romance

Figure 1: Macro-parameterization of languages w.r.t. object shift

Those languages that do have regular object shift may be subject to various additional restrictions on its application. Section 1.3.2.2 argued that this can be accounted for by adopting a small set of additional interface constraints that favor or disfavor object shift under certain circumstances. The constraints central to the analysis provided below are given in (67). (67)

• Constraints needed for fine-tuning the analysis: a. ALIGNFOCUS (AF): the prosodically unmarked focus is the rightmost constituent in its clause. b. RELATIVIZED MINIMALITY (RELMIN): if the foot of X-chain Į c-commands the foot of X-chain ȕ, the head of Į c-commands the head of ȕ (X-chain = A-, A′-, or head chain). c. HEAD-COMPLEMENT ( H-COMPL): a head precedes all terminals originally dominated by its complement.

The goal of this section is to derive the classification of the Germanic languages given in Table 2 from the constraints in (66) and (67). Table 2: Restrictions on object shift and scrambling

FinnishSwedish English Swedish Norwegian Danish Icelandic Dutch German

weak pronouns —

DPs

— optional optional obligatory obligatory obligatory obligatory

— — — — optional optional optional

—

across across IO V n.a. n.a. n.a. — — — — — +

n.a. — — — — + +

across nonverbal particles n.a. n.a. — + + + + +

The restrictions on regular object shift 159

3.2.1.

Object shift of DPs and pronouns

Section 1.3.2 has shown that languages fall into three main classes with respect to regular object shift: some languages allow it both with full noun phrases and pronouns, some allow it with pronouns only, and some do not allow it at all. It has also been shown that this basic classification can be derived from the constraints given in (66). Although these constraints allow the six rankings in (68), not every ranking gives rise to a typologically different language. (68)

a. b. c. d. e. f.

*MOVE >> EPP(case) >> D-PRONOUN *MOVE >> D-PRONOUN >> EPP(case) EPP(case) >> *MOVE >> D-PRONOUN EPP(case) >> D-PRONOUN >> *MOVE D-PRONOUN >> EPP(case) >> *MOVE D-PRONOUN >> *MOVE >> EPP(case)

The languages defined by (68a-b) are typologically identical with respect to regular object shift: the fact that the economy constraint *MOVE outranks both EPP(case) and D-PRONOUN expresses that regular object shift is categorically blocked. The languages in (68c-e) are also typologically identical: the fact that EPP(case) outranks *MOVE expresses that regular object shift is possible with all noun phrases. The ranking in (68f), finally, defines the third language type: regular object shift of weak pronouns is forced by the subranking D-PRONOUN >> *MOVE, despite the fact that the subranking *MOVE >> EPP(case) blocks regular object shift of full noun phrases. Table 3 summarizes the discussion above. Table 3: Regular object shift/scrambling in the Germanic languages (basic typology)

6

OBJECT SHIFT

LANGUAGES

1 *MOVE >> {EPP(case), D-PRONOUN}

2

—

2 EPP(case) >> *MOVE Ranking D-PRONOUN not relevant 3 D-PRONOUN >> *MOVE >> EPP(case)

3

+

1

pronouns only

FinnishSwedish Icelandic Dutch/German Danish

RANKING

The second column indicates the subranking that crucially gives rise to the language type in question. The third column gives the total number of rankings that have the crucial subranking. The fourth column gives the

160 Chapter 3: Regular object shift

defining property of each language type, and the final column provides typical examples of each type from the Germanic language family. The remainder of this subsection will discuss the several subtypes more extensively. 3.2.1.1. Type 1: languages without object shift In languages of type 1, the economy constraint *MOVE outranks the constraints EPP(case) and D-PRONOUN, which favor regular object shift. This results in languages that do not have object shift. Christensen (2005:153) mentions Finnish-Swedish and Falster Danish as examples of such languages, and illustrates this by citing a Finnish-Swedish example from Hugo Berghroth that shows that the definite pronoun det ‘it’ cannot be moved across the negative adverb inte ‘not’. (69)

Jag vet inte I know it not ‘I don’t know it myself.’

selv. myself

(Finnish-Swedish)

The evaluation of this example is given in Tableau 1. The representations of the two candidates in this tableau are slightly simplified for practical reasons. First, I have ignored the application of short object shift and the application of V-to-v, because we have seen in Chapter 2 that these operations must take place in all the Scandinavian languages; the verb phrase is therefore simply represented as [VP V O] — similarly, we will use [VPOV] for the OV-languages later in this chapter. Secondly, I have only included those candidates in which V-to-I takes place, postponing for the moment the question of what forces its application in main clauses; see section 4.1 for discussion. Thirdly, I have only indicated the traces left by object shift, as the candidates do not differ with respect to the number of traces left by verb movement. I will continue to make such simplifications in the tableaux that follow. Finally, the violations of D-PRONOUN are only present when the object is a weak definite pronoun, and are therefore given between parentheses. Tableau 1: Finnish-Swedish (DP and pronominal object)

[I+V Adv [tV O]] ) [I+V [O Adv [tV tO]]]

*MOVE (tO only)

D-PRONOUN

EPP(case)

*

*!

(*) (*)

Although the ranking between D-PRONOUN and EPP(case) cannot be determined on the basis of the available data, I will assume that in Finnish-

The restrictions on regular object shift 161

Swedish D-PRONOUN outranks EPP(case) for reasons that will become clear when we discuss the Swedish constraint ranking in (75c). Although English does not have regular object shift, we cannot conclude from this that English is a type 1 language. The reason for this is that V-to-I of English main verbs is not possible either, which is clear from the fact illustrated in (70a) that the main verb must follow sentential adverbs like probably. Consequently regular object shift in (70b) may be blocked for the independent reason that it either involves an invalid application of V-to-I or induces a violation of HG by crossing the main verb; cf. section 3.2.3. (70)

a. He probably it. b. *He it probably .

Given that Romance languages do have V-to-I, but do not have object shift, they can be assumed to be a type 1 language (unless one would like to maintain that clitic movement must be analyzed as some sort of object shift). 3.2.1.2. Type 2: languages with full object shift Languages of type 2 have the strong ranking EPP(case) >> *MOVE. As a result these languages have full regular object shift, that is, both with full noun phrases and with pronouns. Typical examples of languages with full regular object shift are Icelandic and Dutch, examples of which are given in (71) and (72), respectively. (71)

• Icelandic (e.g., Holmberg 1986, Jónsson 1996) a. Jón las ekki . Jón read it not b. Jón las ekki . Jón read this book not

(72)

• Dutch (e.g., Verhagen 1986, Broekhuis et al. 2003) a. Jan leest waarschijnlijk . Jan reads it probably b. Jan leest waarschijnlijk . Jan reads this book probably

These examples show that regular object shift of weak definite pronouns is obligatory. Regular object shift of non-pronominal DPs, on the other hand, is optional and depends on whether the object is part of the presupposition or the focus of the clause: object shift only applies in the former case. The evaluation of the Icelandic examples in (71b) is given in Tableaux 2 and 3. Tableau 2 provides the evaluation for those cases in which the

162 Chapter 3: Regular object shift

object is part of the presupposition of the clause, and shows that the strong ranking of EPP(case) forces application of regular object shift. Since weak definite pronouns are presuppositional by definition, the evaluation in Tableau 2 also implies that pronoun shift is obligatory in Icelandic: this shows that the ranking of D-PRONOUN is immaterial for languages with a strong ranking of EPP(case), at least insofar as simple data like (71) and (72) are concerned (section 3.2.2.1 will show on the basis of more intricate facts that D-PRONOUN actually outranks ALIGNFOCUS in Icelandic and Dutch; the discussion below will normally omit modifications of this kind). Tableau 2: Icelandic

DP ⊄ focus [I+V [Adv [tV O]]] [I+V [O Adv [tV tO]]]

AF

EPP(case)

*MOVE (tO only)

*! )

*

Tableau 3 provides the evaluation for those cases in which the object is part of the focus of the clause, and shows that the fact that ALIGNFOCUS outranks EPP(case) blocks regular object shift in this case. Tableau 3: Icelandic

DP ⊂ focus [I+V [Adv [tV O]]] [I+V [O Adv [tV tO]]]

AF

)

EPP(case)

*MOVE (tO only)

* *!

*

Putting aside questions concerning the application of verb movement, the evaluations of the Dutch examples in (72) is essentially identical to the ones given for Icelandic in Tableaux 2 and 3. 3.2.1.3. Type 3: languages with pronoun shift only Languages of type 3 have the ranking D-PRONOUN >> *MOVE >> EPP(case), and are characterized by the fact that they have regular object shift of pronouns only. A typical example of such a language is Danish: the examples in (73) illustrate that whereas pronoun shift is possible, regular object shift of non-pronominal DPs is excluded. (73)

• Danish (Vikner 1994:502) a. Hvorfor læste studenterne why read the students b. Hvorfor læste studenterne why read the students

ikke . it not ikke . the article not

The restrictions on regular object shift 163

Tableau 4 provides the evaluation of example (73b) and shows that the weak ranking of EPP(case) blocks the application of regular object shift of non-pronominal DPs. Tableau 4: Danish (DP-object) D-PRONOUN

*MOVE (tO only)

[I+V [Adv [tV O]]] ) [I+V [O Adv [tV tO]]]

EPP(case)

* *!

Tableau 5 gives the evaluation of example (73a): the fact that D-PRONOUN outranks the economy constraint *MOVE forces object shift of the weak pronoun in this example. Tableau 5: Danish (pronominal object) D-PRONOUN

[I+V [Adv [tV pron]]] [I+V [pron Adv [tV tO]]] )

*MOVE (tO only)

*!

EPP(case)

* *

3.2.1.4. A mixed type: optional pronoun shift In the previous subsections we have discussed the three basic types into which languages can be divided when it comes to regular object shift. There is, however, a fourth type, which is characterized by optional pronoun shift. According to Holmberg (1999:2), most varieties of Swedish as well as some varieties of Norwegian are of this type. In (74) some examples are given from Swedish. (74)

a. b.

Jag kysste inte . I kissed her not Jag kysste inte . I kissed Anna not

(Swedish)

Swedishs seem to be of a mixed type: it is like languages of type 3 in that it allows pronoun shift only, but resembles languages of type 1 in that it does not require it. A plausible way of accounting for this is to follow Kroch’s (1989;1994) grammar competition approach, and to assume that Swedish is undergoing a process of language change so that (most) speakers have two competing grammars available. In other words, Swedish speakers can freely apply either the Finnish-Swedish ranking in (75a) or the Danish ranking in (75b). We can readily express this by saying that the constraints *MOVE and D-PRONOUN are in a tie, as in (75c): cf. the discussion of the Doubly Filled COMP Filter in section 1.1.2.

164 Chapter 3: Regular object shift (75)

a. b. c.

Finnish-Swedish: *MOVE >> D-PRONOUN >> EPP(case) Danish: D-PRONOUN >> *MOVE >> EPP(case) Swedish: D-PRONOUN *MOVE >> EPP(case)

The evaluation of the Swedish examples in (74) can now proceed as indicated in the tableaux below. The evaluation of example (74b) is given in Tableau 6, and shows that just as in Finnish-Swedish and Danish, the weak ranking of EPP(case) blocks object shift of non-pronominal DPs. Tableau 6: Swedish (DP-object) D-PRONOUN

[I+V [Adv [tV O]]] ) [I+V [O Adv [tV tO]]]

*MOVE (tO only)

EPP(case)

* *!

The evaluation of example (74a) is given in Tableau 7, and shows that pronoun shift need not apply in Swedish. When the tie between D-PRONOUN and *MOVE is read from left to right, that is, when the “Danish ranking” in (75b) is used, the first candidate gives rise to a fatal violation of DPRONOUN, so that the second candidate with object shift is selected as the optimal one. However, when the tie is read from right to left, that is, when the “Finnish-Swedish ranking” in (75a) is used, the second candidate gives rise to a fatal violation of *MOVE, so that the first candidate without object shift is selected as the optimal one. Tableau 7: Swedish (pronominal object) D-PRONOUN

[I+V [Adv [tV pron]]] ) [I+V [pron Adv [tV tO]]] )

*MOVE (tO only)

*>

EPP(case)

* *
> ALIGNFOCUS >> EPP(case) >> *MOVE German: D-PRONOUN >> ALIGNFOCUS >> RELMIN >> EPP(case) >> *MOVE

The evaluations of the German examples in (88) are therefore as given in Tableaux 15-18. Tableau 15 shows that when both objects belong to the presupposition of the clause, the strong ranking of EPP(case) forces object shift of both objects; this selects example (88c) as the optimal candidate.

a. Subj I+V Adv IO DO tV b. Subj I+V IO Adv tIO DO tV c. Subj I+V IO DO Adv tIO tDO tV ) d. Subj I+V DO Adv IO tDO tV

*!* *! *!

*

*MOVE (tO only)

EPP(case)

RELMIN

AF

IO & DO ⊄ focus

DPRONOUN

Tableau 15: German

* ** *

174 Chapter 3: Regular object shift

Tableau 16 shows that when the objects are both part of the focus of the clause, ALIGNFOCUS blocks the application of object shift, as in (88a).

a. Subj I+V Adv IO DO tV ) b. Subj I+V IO Adv tIO DO tV c. Subj I+V IO DO Adv tIO tDO tV d. Subj I+V DO Adv IO tDO tV

* **! **!* **!

** * *

*

*MOVE (tO only)

EPP(case)

RELMIN

AF

IO & DO ⊂ focus

DPRONOUN

Tableau 16: German

* ** *

Tableau 17 shows that when only the direct object is part of the focus of the clause, the strong ranking of EPP(case) forces object shift of the indirect object, but object shift of the direct object will be blocked by ALIGNFOCUS. This selects (88b) as the optimal candidate.

a. Subj I+V Adv IO DO tV b. Subj I+V IO Adv tIO DO tV ) c. Subj I+V IO DO Adv tIO tDO tV d. Subj I+V DO Adv IO tDO tV

**! * *! *!*

*

*

*MOVE (tO only)

EPP(case)

RELMIN

AF

IO ⊄ focus; DO ⊂ focus

DPRONOUN

Tableau 17: German

* ** *

Thus far the evaluations of the German examples proceed in the same way as the Icelandic and Dutch ones. The languages diverge, however, in those cases where the indirect object is part of the focus and the direct object is part of the presupposition of the clause. The claim that ALIGNFOCUS outranks RELMIN in German now predicts that it is better to move the direct object across the indirect object than to apply the blocking strategy. Tableau 18 shows that as a result example (88d) is correctly predicted to be optimal.

The restrictions on regular object shift 175

a. Subj I+V Adv IO DO tV b. Subj I+V IO Adv tIO DO tV c. Subj I+V IO DO Adv tIO tDO tV d. Subj I+V DO Adv IO tDO tV )

*! *!* *!*

** * *

*

*MOVE (tO only)

EPP(case)

RELMIN

AF

IO ⊂ focus; DO ⊄ focus

DPRONOUN

Tableau 18: German

* ** *

For completeness’ sake, Tableaux 19 and 20 provide the evaluations for examples in which the direct object is not a lexical DP but a pronoun. It is correctly predicted that pronoun shift across the indirect object is possible.

*!

*

* ** *

*MOVE (tO only)

** *

*MOVE (tO only)

EPP(case)

RELMIN

*! *!

EPP(case)

a. Subj I+V Adv IO pron tV b. Subj I+V IO Adv tIO pron tV c. Subj I+V IO pron Adv tIO tDO tV ) d. Subj I+V pron Adv IO tDO tV

AF

IO ⊄ focus; DO = pronoun

DPRONOUN

Tableau 19: German

a. Subj I+V Adv IO pron tV b. Subj I+V IO Adv tIO pron tV c. Subj I+V IO pron Adv tIO tDO tV d. Subj I+V pron Adv IO tDO tV )

*! *!

RELMIN

AF

IO ⊂ focus; DO = pronoun

DPRONOUN

Tableau 20: German

* ** *!*

** * *

*

* ** *

176 Chapter 3: Regular object shift

This completes the discussion of object shift in double object constructions in languages with full object shift. For completeness’ sake it must be noted that in German, as well as in Dutch, direct object pronouns may also precede shifted indirect objects. This is illustrated for Dutch in (90): although the direct object pronoun may follow the full indirect object, it seems that there is some preference for placing them into some position preceding the shifted indirect object. (90)

a.

b. c.

dat Jan Marie waarschijnlijk tIO de boeken/*ze aanbood. that Jan Marie probably the books/them prt.-offered ‘that Jan offered Marie the books.’ dat Jan Marie de boeken/ze waarschijnlijk tIO tDO aanbood. dat Jan ze/*de boeken Marie t′DO waarschijnlijk tIO tDO aanbood.

This suggests that the regular object shift of object pronouns in (90b) can be followed by an additional movement that places the pronoun in an even more leftward position. The fact that this movement may cross the shifted indirect object in Dutch, as in (90), suggests that this movement is not of the A-movement type, because this would be blocked by RELMIN. That the movement in question is rather of the A′-movement type is further supported by the fact that the higher position is also accessible to locational pro-forms. Consider the examples in (91). (91)

a.

Jan heeft het boek waarschijnlijk tDO in de kast gelegd. Jan has the book probably in the closet put ‘Jan has probably put the book in the closet.’ b. Jan heeft het boek waarschijnlijk daar gelegd. b′. #Jan heeft het boek daar waarschijnlijk tdaar gelegd. b′′. Jan heeft daar het boek waarschijnlijk tdaar gelegd.

The predicative directional PP in example (91a) must be placed in the position immediately preceding the clause-final verb (unless it is wh-moved or topicalized). In (91b-b′′) the PP is replaced by the pro-form daar ‘there’. Example (91b) shows that this form can simply occupy the same position of the full PP in (91a). The unacceptability of (91b′) shows that the proform daar cannot undergo regular object shift, which does not come as a surprise given that it has no case feature; the double cross is used to indicate that example (91b′) is acceptable when daar is stressed — as before, I am only considering the neutral pronunciation pattern here with the moved elements being destressed. The acceptability of (91c), finally, shows that the locational pro-form daar can be moved into a position

The restrictions on regular object shift 177

preceding the landing site of the shifted direct object het boek; in section 2.1.2 I have already argued that the landing site of the pro-form daar is an A′-position. The fact that we have identified an additional landing site for pronouns and other pro-forms is important, because we have already seen that Haider, Olsen & Vikner (1995:20-1) have given the Danish examples in (92) as a problem for an A-movement account of regular object shift: (92a) shows that leftward movement of the pro-form der ‘there’ is possible despite the fact that it is not case-marked, and (92b) shows that, just like pronoun shift, movement of der cannot cross the verb. (92)

a.

b.

Peter sov alligevel ikke . Peter slept there after all not ‘Peter didn’t sleep there after all.’ Peter har alligevel ikke sovet . Peter has there after all not slept ‘Peter didn’t sleep there after all.’

(Danish)

If Danish is similar to German and Dutch in allowing A′-movement of pronouns and other pro-forms into a position to the left of the target position of regular object shift, this potential problem for the A-movement approach to regular object shift is partly solved. The question why movement of der, like object pronouns, is excluded across the main verb must still be answered, of course, but since I have no insights to offer here, I will leave this question to future research. 3.2.2.2. Type 3: languages with pronoun shift only This subsection will discuss regular object shift in double object constructions in languages that have pronoun shift only. Let us first consider double object constructions in which both the direct and the indirect object are pronouns, as in (93), taken from Vikner (1989;1990). The evaluation in Tableau 21 shows that the constraint ranking that was established earlier straightforwardly predicts that (93c) is the optimal candidate. (93)

a.

*Peter viste jo hende den. Peter showed indeed her it b. *Peter viste hende jo tIO den. c. Peter viste hende den jo tIO tDO. d. *Peter viste den jo hende tDO.

(Danish)

178 Chapter 3: Regular object shift

a. Subj I+V Adv tV pronIO pronDO b. Subj I+V pronIO Adv tV tIO pronDO c. Subj I+V pronIO pronDO Adv tV tIO tDO d. Subj I+V pronDO Adv tV pronIO tDO

*!* *! ) *!

* ** *

EPP(case)

DPRONOUN

IO & DO = pronoun

*MOVE (tO only)

Tableau 21: Danish

** * *

This ranking also straightforwardly predicts that when the indirect object is a pronoun and the direct object is a non-pronominal DP, as in (94), it is only the indirect object that undergoes object shift. This is shown in Tableau 22. (94)

a.

*Peter viste jo hende bogen. Peter showed indeed her the book b. Peter viste hende jo tIO bogen. c. *Peter viste hende bogenj jo tIO tDO. d. *Peter viste bogen jo hende tDO.

(Danish)

a. Subj I+V Adv tV pronIO DO b. Subj I+V pronIO Adv tV tIO DO c. Subj I+V pronIO DO Adv tV tIO tDO d. Subj I+V DO Adv tV pronIO tDO

*! ) *!

* **! *

EPP(case)

DPRONOUN

IO = pronoun; DO ≠ pronoun

*MOVE (tO only)

Tableau 22: Danish

** * *

The situation is getting more complex, however, when the indirect object is a non-pronominal DP and the direct object a pronoun. Consider the examples in (95), with the judgments assigned to them in Vikner (1989;1990). Vikner’s judgments suggest that there is some uncertainty concerning the grammatical status of the examples in (95a&c). Furthermore, there seems to be some disagreement on example (95a) as similar examples are judged

The restrictions on regular object shift 179

fully acceptable by Christensen (2005:155). Unfortunately, Christensen does not discuss examples like (95c), but since Vikner judges the two examples as equally acceptable, it seems safe to somewhat idealize the data and to assume that both are grammatical. (95)

a.

??

Peter viste jo Marie den. Peter showed indeed Marie it b. *Peter viste Marie jo tIO den. c. ??Peter viste Marie den jo tIO tDO d. *Peter viste den jo Marie tDO

(Danish)

The conclusion that (95c) is acceptable is quite remarkable, as this shows that Danish allows object shift of non-pronominal DPs in double object constructions, despite the fact that this is categorically blocked in monotransitive constructions. 43 We will see below, however, that this follows immediately when we assume that the constraint D-PRONOUN is outranked by RELMIN. That we need to introduce RELMIN is of course independently motivated by the fact that the ranking D-PRONOUN >> *MOVE >> EPP(case) wrongly predicts that example (95d) is acceptable.

a. Subj I+V Adv tV IO pron b. Subj I+V IO Adv tV tIO pron c. Subj I+V IO pron Adv tV tIO tDO d. Subj I+V pron Adv tV IO tDO

*! *! /

* **! *

EPP(case)

DPRONOUN

IO ≠pronoun; DO = pronoun

*MOVE (tO only)

Tableau 23: Danish

** * *

The evaluation in Tableau 24 shows that this problem is solved when we assume that RELMIN outranks *MOVE. The evaluation in Tableau 24 also makes clear that, just as in Icelandic, in Danish the constraint D-PRONOUN favors the push-up strategy.

43

Tarald Taraldsen (p.c.) has brought to my attention that in Norwegian and some varieties of Swedish object shift of non-pronominal indirect objects is possible without push up. This does, however, not hold for Danish (Holmberg & Platzack 1995:172).

180 Chapter 3: Regular object shift

a. Subj I+V Adv tV IO pron b. Subj I+V IO Adv tV tIO pron c. Subj I+V IO pron Adv tV tIO tDO d. Subj I+V pron Adv tV IO tDO

*! *! ) *!

* ** *

EPP(case)

RELMIN

DPRONOUN

IO ≠ pronoun; DO = pronoun

*MOVE (tO only)

Tableau 24: Danish

** * *

What we have not yet accounted for is the acceptability of example (95a) in Danish. Since this example involves the blocking strategy, and we have seen in the discussion of Icelandic and Dutch that ALIGNFOCUS favors this strategy, we must conclude that ALIGNFOCUS outranks D-PRONOUN. The evaluations in Tableaux 25 and 26 show that this will correctly predict the pattern in (95) provided that we also assume that ALIGNFOCUS in its turn is outranked by RELMIN. Tableaux 25 shows that the blocking strategy in (95a) will be preferred over the push-up strategy when the indirect object is part of the focus of the clause, Tableaux 26, on the other hand, shows that the push-up strategy in (95c) is preferred when the indirect object is part of the presupposition of the clause. observe that the subranking ALIGNFOCUS >> D-PRONOUN is also crucial, because when we reverse the order of these constraints, the push-up strategy will also arise when the indirect object is in focus; we have seen in Tableau 14 that this is what happens in Icelandic and Dutch.

a. Subj I+V Adv tV IO pron b. Subj I+V IO Adv tV tIO pron c. Subj I+V IO pron Adv tV tIO tDO d. Subj I+V pron Adv tV IO tDO

*! *! ) *!

* ** *

EPP(case)

*MOVE (tO only)

AF

RELMIN

IO ⊄ focus; DO = pronoun

DPRONOUN

Tableau 25: Danish

** * *

The restrictions on regular object shift 181

a. Subj I+V Adv tV IO pron b. Subj I+V IO Adv tV tIO pron c. Subj I+V IO pron Adv tV tIO tDO d. Subj I+V pron Adv tV IO tDO

)

* **! **! *!

* *

* ** *

EPP(case)

*MOVE (tO only)

AF

RELMIN

IO ⊂ focus; DO = pronoun

DPRONOUN

Tableau 26: Danish

** * *

3.2.2.3. The mixed type: languages with optional pronoun shift Since we have seen earlier that Danish and Swedish differ mainly in that in the latter language the constraints D-PRONOUN and *MOVE are in a tie, let us hypothesize that the Swedish ranking is as given in (96b). (96)

a. b.

Danish: RELMIN >> AF >> D-PRONOUN >> *MOVE >> EPP(case) Swedish: RELMIN >> AF >> D-PRONOUN *MOVE >> EPP(case)

As we have seen in section 3.2.1.4, the fact that D-PRONOUN and *MOVE are tied in Swedish predicts that pronoun shift is optional. Ranking (96b) now predicts that the same holds for the double object construction as long as this does not lead to a violation of RELMIN. This prediction is indeed borne out, as can be seen from the examples in (97), taken from Christensen (2005:159). (97)

a.

Jag gav inte henne den. I gave not her it ‘I didn’t give it to her.’ b. Jag gav henne inte thenne den. c. Jag gav henne den inte thenne tden. d. *Jag gav den inte henne tden.

(Swedish)

As is the case in Danish, complications arise when we have a pronominal direct object and a non-pronominal DP as indirect object. Consider the examples in (98), taken from Anagnostopoulou (2003:128) and Christensen (2005:162).

182 Chapter 3: Regular object shift (98)

a.

Jag visade inte Elsa den förran langt senare. I showed not Elsa it until much later b. *Jag visade Elsa inte tIO den förran langt senare. c. %Jag visade Elsa den inte tIO tDO förran langt senare. d. *Jag visade den inte Elsa tDO förran langt senare.

(Swedish)

Again the data are not crystal clear: According to Anagnostopoulou, example (98c) is judged fully acceptable by Anders Holmberg and Christer Platzack, while Christensen notes that Maia Andreasson assigns it two question marks. Again I will idealize the data and assume that both (98a) and (98c) are fully grammatical. The evaluations in Tableaux 27 and 28 show that this follows from the ranking in (96b). Tableau 27 shows that when the indirect object is part of the presupposition of the clause, example (98a) is selected as the optimal candidate when the tie d-pronoun *move is read from right to left, whereas (98c) is selected as the optimal candidate when it is read from left to right. Tableau 28 shows that when the indirect object is part of the focus of the clause the push-up strategy is blocked by ALIGNFOCUS, so that only example (98a) is selected as the optimal candidate.

a. Subj I+V Adv tV IO pron ) b. Subj I+V IO Adv tV tIO pron c. Subj I+V IO pron Adv tV tIO tDO ) d. Subj I+V pron Adv tV IO tDO

*> *> *!

*< *> EPP(ij) >> *STRAY FEATURE >> *MOVE (ii) Swedish/Danish: H-COMPL >> EPP(ij) >> *MOVE >> *STRAY FEATURE Germanic OV-languages: EPP(ij) >> *MOVE >> {*STRAY FEATURE, H-COMPL}

The question of whether regular object shift may or may not cross the main verb depends on the relative ranking of H-COMPL and the constraints that favor this movement, EPP(case) and D-PRONOUN. For simplicity, let us ignore questions concerning he relation between short object shift and verb movement and simply see what the typological predictions are for the ranking of the four constraints H-COMPL, EPP(case), D-PRONOUN and *MOVE. These constraints can be ranked in 24 different orders, but since some of these rankings give rise to identical language types, we only predict the language types given in Table 4, which shows that the three basic types we distinguished in section 3.2.1 can be divided into smaller subtypes. Four out of the six different subtypes can be found in my sample taken from the Germanic language family. The following subsections will discuss these subtypes in more detail. Table 4: Regular object shift/scrambling in the Germanic languages (extended typology) RANKING

1 2 A B C 3 A B

*MOVE >> {EPP(case), D-PRONOUN}; ranking H-COMPL not relevant EPP(case) >> {*MOVE, H-COMPL}; ranking D-PRONOUN not relevant H-COMPL >> EPP(case) >> *MOVE; H-COMPL >> D-PRONOUN D-PRONOUN >> H-COMPL >> EPP(case) >> *MOVE D-PRONOUN >> *MOVE >> EPP(case); D-PRONOUN >> H-COMPL H-COMPL >> D-PRONOUN >> *MOVE >> EPP(case)

# 24 8

—

V n.a.

8

+

+

3

+

—

1

+

3

pronouns only pronouns only

pronouns only +

1

OBJECT SHIFT

ACROSS

—

LANGUAGE

FinnishSwedish Dutch/ German Icelandic not in sample not in sample Swedish/ Danish

3.2.3.1. Type 1: languages without regular object shift The first language type is characterized by having no regular object shift at all, and arises when the economy constraint *MOVE outranks the constraints that favor this movement, EPP(case) and D-PRONOUN. Since object shift does not apply, the ranking of H-COMPL is of course immaterial. Consequently, 8

The restrictions on regular object shift 185

out of the 24 possible rankings give rise to this language type. An examples of this language type is Finnish-Swedish. Examples have already been given in section 3.2.1.1. 3.2.3.2. Type 2: languages with full object shift Table 4 distinguishes three subtypes of languages with full object shift: languages in which object shift may cross the main verb, languages in which the main verb blocks object shift, and languages in which the main verb block object shift of full noun phrases only. Languages of subtype 2A are characterized by having full regular object shift, which may also apply across the main verb. This type arises when EPP(case) outranks the constraints that disfavor object shift, *MOVE and HCOMPL. Since object shift is possible anyway, the ranking of D-PRONOUN is immaterial. This means that 8 out of the 24 possible rankings give rise to this language type. Examples of this language type are Dutch and German, although it must be noted that the crossing of the main verb in these languages is not the result of regular object shift but of short object shift, given that in these languages *MOVE and H-COMPL are also outranked by EPP(ij); cf. (101b). We illustrate this by means of the Dutch examples in (102). (102) a.

b.

Jan denkt dat Peter zeker het boek koopt. Jan thinks that Peter certainly the book reads ‘Jan thinks that Peter will certainly read the book.’ Jan denkt dat Peter het boek zeker koopt.

Tableaux 29 and 30 show that the order in (102a) arises when the object is part of the focus of the clause, whereas (102b) arises when the object is part of the presupposition of the clause. Tableau 29: Dutch embedded clauses

object ⊄ focus C [...Adv v ... V O] C [... Adv v ... O V tO] C [... O Adv v tO V tO] )

EPP

AF

(ij) *!

EPP

(case) * *!

*MOVE (tO only)

HCOMPL

* **

* *

*MOVE (tO only)

HCOMPL

* **

* *

Tableau 30: Dutch embedded clauses

object ⊂ focus C [...Adv v ... V O] C [... Adv v ... O V tO] ) C [... O Adv v tO V tO]

EPP

AF

(ij) *! * **!

EPP

(case) * *

186 Chapter 3: Regular object shift

The Dutch examples in (103) show that we find essentially the same facts in complex verb constructions. I provide the evaluations of these examples in Tableaux 31 and 32. Recall from section 2.4 that V-to-v/Asp does not apply in Dutch, and that both the verbal root V and the auxiliary have ij-features that may trigger short object shift, and from section 3.1.2.3 of this chapter that regular object shift is triggered by the case feature on the light verb v associated with the auxiliary. (103) a. b.

dat Jan waarschijnlijk het boek heeft gelezen. that Jan probably the book has read dat Jan het boek waarschijnlijk heeft gelezen.

* * *!

H-COMPL

**! *!

*MOVE (tO only)

...Adv v... aux ... Asp ... V O ...Adv v... aux... Asp ... O V tO ...Adv v... O aux... Asp ... tO V tO ... O Adv v... tO aux ... Asp ... tO V tO )

AF

EPP(ij)

object ⊄ focus

EPP(case)

Tableau 31: Dutch complex verb constructions

* ** ***

* ** **

* ** ***!

* * *

H-COMPL

*!* *!

*MOVE (tO only)

...Adv v... aux ... Asp ... V O ...Adv v... aux... Asp ... O V tO ...Adv v... O aux... Asp ... tO V tO ) ... O Adv v... tO aux ... Asp ... tO V tO

AF

EPP(ij)

object ⊂ focus

EPP(case)

Tableau 32: Dutch complex verb constructions

* ** ***

* ** **

The subtypes 2B and 2C differ from Type 2A in that the main verb may block regular object shift: type 2C allows object shift across the main verb but only when the object is a pronoun; type 2B does not allow any form of object shift across the main verb. This difference is again due to the relative ranking of H-COMPL and D-PRONOUN: When the former outranks the latter regular pronoun shift across the main verb is blocked; when the ranking is the inverse, pronoun shift may cross the verb.

The restrictions on regular object shift 187

To my knowledge, there are no examples of subtype 2C in the Germanic languages. The examples in (104) and (105) show that Icelandic belongs to subtype 2B: those in (104) show that whereas regular object shift is possible in simple tense constructions, it is blocked by the main verb in the perfect tense construction; the examples in (105), which are taken from Thráinsson (2001), show that this also holds for pronominal objects. (104) a.

Jón keypti ekki bókina. Jón bought not the book a′. Jón keypti bókina ekki tbókina. b. Jón hefur ekki keypt bókina. Jón has not bought the book b′. *Jón hefur bókinai ekki keypt tbókina.

(105) a.

Nemendurnir hafa ekki lesið hana. the students have not read it b. *Nemendurnir hafa hana ekki lesið thana.

Given the Icelandic subrankings established earlier, the ranking of the constraints involved in the evaluation of the perfect tense constructions in (104b&b′) and (105) must be as indicated in Tableau 33; because the table will not fit on the page if I put the relevant candidates in it, I give them separately as (106). Of course, the stars between parentheses only occur when the direct object is a pronoun, as in (105). (106) a. b. c. d.

I+v+aux ... ekki tv+aux ... taux ... Asp+V ... tV O I+v+aux ... ekki tv+aux ... taux ... Asp+V ... O tV tO I+v+aux ... ekki tv+aux ... O taux ... Asp+V ... tO tV tO I+v+aux ... O ekki tv+aux ... tO taux ... Asp+V ... tO tV tO

Tableau 33: Icelandic complex verb constructions HCOMPL

(106a) (106b) (106c) (106d)

EPP(ij)

**! *

) *! *!

DPRONOUN

(*) (*) (*)

EPP

(case) * * *

*MOVE (tO only) * ** ***

For completeness’ sake, recall that regular object shift is allowed in Icelandic embedded clauses with a simple tense due to the fact that V-to-I does apply in that context; as a result, regular object shift will not violate HCOMPL by definition. This is illustrated again in (107).

188 Chapter 3: Regular object shift (107) a.

b.

það var gott að hann keypti ekki bókina. it was good that he bought not the book ‘It was good that he didn’t buy the book.’ það var gott að hann keypti bókina ekki.

3.2.3.3. Type 3: languages with pronoun shift only Types 3A and 3B are characterized by having pronoun shift only. As we have seen earlier, this arises when we have the subranking D-PRONOUN >> *MOVE >> EPP(case). The two types differ in that in type 3A pronoun shift can apply across the main verb, whereas in type 3B it cannot. This difference is due to the relative ranking of D-PRONOUN and H-COMPL. When the former outranks the latter regular pronoun shift across the main verb is allowed; as far as I know there is no Germanic language of this type. On the inverse ranking regular pronoun shift across the main verb will be blocked, as in Danish. Since Danish does not have V-to-I in embedded clauses this can be readily illustrated by means of the examples in (108), which receive the evaluation in Tableau 34. (108) a.

at Peter ikke købte den. that Peter not bought it ‘that Peter didn’t buy it.’ b. *at Peter den ikke købte tden.

C [I ... ikke v+V ... tV pron] C [I ... ikke v+V ... pron tV tO] C [I ... pron ikke v+V ... tO tV tO]

*! ) *!

* *

* **

EPP(case)

*MOVE (tO only)

DPRONOUN

EPP(ij)

H-COMPL

Tableau 34: Danish pronoun shift in embedded clauses

* *

The examples in (109) show that regular object shift of pronouns is also blocked in Danish perfect tense constructions. The evaluation is given in Tableau 35; the relevant candidates are given as (110). (109) a.

Hvarfor har Peter aldrig læst den? why has Peter never read it ‘Why has Peter never read it?’ b. *Hvarfor har Peter den aldrig læst tden?

The restrictions on regular object shift 189 (110) a. b. c. d.

I+v+aux ... Adv tv+aux ... taux ... Asp+V ... tV pron I+v+aux ... Adv tv+aux... taux ... Asp+V ... pron tV tO I+v+aux ... Adv tv+aux ... pron taux ... Asp+V ... tO tV tO I+v+aux ... pron Adv tv+aux ... tO taux ... Asp+V ... tO tV tO

Tableau 35: Danish pronoun shift in complex verb constructions HCOMPL

(110a) (110b) ) (110c) (110d)

EPP

(ij)

**! * *! *!

DPRONOUN

* * *

*MOVE (tO only) * ** ***

EPP(case)

* * *

Before we continue with discussing Swedish, note that I have assumed EPP(ij) to be ranked above D-PRONOUN, despite the fact that we have seen no empirical evidence for that so far: the ranking H-COMPL >> {EPP(ij), D-PRONOUN} >> *MOVE >> EPP(case) gives the same results as the ranking in Tableaux 34 and 35. A first reason to opt for the latter ranking is that we will see in the next subsection that EPP(ij) outranks D-PRONOUN in Swedish: if we want to keep the difference between Swedish and Danish as minimal as possible, the Danish ranking in Tableau 34 is clearly the preferred one. A second and of course more conclusive reason is that the analysis of verb movement in Danish in section 4.1 will provide empirical evidence that the ranking in the tableaux is indeed correct. 3.2.3.4. The mixed type: optional object shift Let us conclude with Swedish, which was shown earlier to be of a mixed type due to the tie between D-PRONOUN and *MOVE. The examples in (111), which are taken from Holmberg (1999), show that Swedish behaves essentially like Danish in the respects relevant here. (111) a.

att jag inte kysste henne. that I not kissed her a′. *att jag henne inte kysste thenne. b. Jag har inte kysste henne. I have not kissed her b′. *Jag har henne inte kysste thenne.

The evaluation of the embedded clauses in (111a&a′) is given in Tableau 36, which shows that regular object shift is excluded. Note that in Swedish EPP(ij) must be assumed to outrank D-PRONOUN, since otherwise it follows

190 Chapter 3: Regular object shift

by transitivity that it is also in a tie with *MOVE, which would wrongly predict that short object shift is optional.

C [I ... inte v+V .. . tV pron] C [I ... inte v+V ... pron tV tO] C [I ... pron inte v+V ... tO tV tO]

*! )

* *

*!

EPP(case)

*MOVE (tO only)

DPRONOUN

EPP(ij)

H-COMPL

Tableau 36: Swedish pronoun shift in embedded clauses

* *

* **

The evaluation of the perfect tense constructions in (111b&b′) is given in Tableau 37, and shows that regular object shift of pronouns is blocked; the relevant candidates are given as (112). (112) a. b. c. d.

I+v+aux ... inte tv+aux ... taux ... Asp+V ... tV pron I+v+aux ... inte tv+aux ... taux ... Asp+V ... pron tV tO I+v+aux ... inte tv+aux ... pron taux ... Asp+V ... tO tV tO I+v+aux ... pron inte tv+aux ... tO taux ... Asp+V ... tO tV tO

Tableau 37: Swedish pronoun shift in complex verb constructions HCOMPL

(112a) (112b) (112c) (112d) 3.2.4.

) *! *!

EPP(ij)

DPRONOUN

**! *

* * *

*MOVE (tO only) * ** ***

EPP(case)

* * *

Object shift across a verbal particle

This subsection concludes the discussion of the restrictions on regular object shift with a discussion of Holmberg’s (1986;1999) observation that regular object shift of pronouns is impossible in Swedish when the clause contains a verbal particle, but not in the other Scandinavian languages. It has been repeatedly observed that this fact correlates with the fact that Swedish is also unique in that it does not allow non-pronominal objects to occur in between the verb and the verbal particle. I will start the discussion with illustrating the relevant facts by means of the examples in (113) to (115), taken from Vikner (2006:398-9).

The restrictions on regular object shift 191

Consider first the Danish examples in (113). Examples (113a&b) show that the non-pronominal object tæppet ‘the carpet’ must precede the verbal particle ud ‘away’, and examples (113b&c) show that non-pronominal and pronominal objects differ in the usual way: whereas the former cannot, the latter must undergo regular object shift. (113) a. b. c.

Peter har ikke smidt Peter has not thrown Peter smed Peter threw the-carpet Peter smed Peter threw it

ud . (Danish) the-carpet away ikke ud . not away ikke ud . not away

Now compare the Danish examples in (113a&c) with the Swedish examples in (114). Example (114a) shows that Swedish differs from Danish in that the non-pronominal DP mattan ‘the carpet’ must follow the verbal particle bort ‘away’. Example (114b) shows that the same holds for the pronominal object det ‘it’: it may occupy neither the position immediately preceding the verbal particle, nor undergo regular object shift into the position preceding negation. (114) a. b.

Peter har inte kastat bort . Peter has not thrown the-carpet away Peter kastade inte bort . Peter threw it not away

(Swedish)

Finally, consider the Icelandic examples in (115). Example (115a) shows that the non-pronominal object may either precede or follow the verbal particle. The examples in (115b) and (115c) shows that, as usual, nonpronominal DPs may and pronominal objects must undergo object shift. (115) a. b. c.

Pétur hefur ekki hent Pétur has not thrown Pétur henti Pétur threw the carpet Pétur henti Pétur threw it

út . (Icelandic) the carpet away ekki út . not away ekki út . not away

Now that we have reviewed the basic facts, let us see whether we are able to account for the differences between Danish, Swedish, and Icelandic. As we have seen in Chapter 2, verb-particle constructions have been used to argue that the VO-languages like English also have object movement (see, e.g., Johnson 1991): example (116b) is derived from (116a) by some sort of object shift.

192 Chapter 3: Regular object shift (116) a. b.

John looked up the information. John looked the information up ti.

Given our assumptions so far, the movement in (116b) must be considered short object shift. We have seen, however, that due to the strong ranking of EPP(ij), short object shift is normally obligatory in the Germanic VOlanguages. Consider again the rankings that we established in section 2.4.2.5 for the Scandinavian languages under discussion. Recall that the rankings in (117) predict that V-to-v/Asp applies obligatorily in these languages: this is forced by *STRAY FEATURE >> *MOVE in Icelandic, and by H-COMPL >> EPP(ij) >> *MOVE in Swedish and Danish., Since short object shift will never violate H-COMPL as a result, it follows that this shift is normally obligatory. (117) a.

Icelandic: H-COMPL

b.

>> EPP(ij) >> *STRAY FEATURE >> *MOVE Danish/Swedish: H-COMPL >> EPP(ij) >> *MOVE >> *STRAY FEATURE

If English is like these Scandinavian languages, this means that we must find some other means to block short object shift in English examples like (116a). It has been assumed in, e.g., Den Dikken (1995a: ch.2) that examples like (116a) are the result of reanalysis of the verb and the particle. Den Dikken claims that this reanalysis makes it possible to assign case to the object in its base position. When reanalysis does not take place, the object must move into the position right-adjacent to the verb in order to pick up case. Of course, this assumption cannot be adopted here since we have assumed that short object shift is triggered by the ij-features of the verb. However, let us maintain the basic idea that reanalysis of the verb and the particle results in the creation of a complex verb, and assume that, as a result of this, short object shift across the particle results in a violation of H-COMPL. (118) Short object shift across the particle of a reanalyzed V-prt complex invokes an additional violation of H-COMPL.

The constraint rankings in (117) in tandem with the assumption in (118) predict that short object shift across a reanalyzed particle is excluded. This implies that the data in (113) to (115) can be accounted for by assuming that the languages under discussion exhibit different behaviors as far as reanalysis is concerned. At this moment I do not have a definite answer to the question why these languages differ as indicated in (119), although the

The restrictions on regular object shift 193

discussion that will follow suggests that this may have to do with the availability of certain small clause-internal movements; cf. (132). (119) a. b. c.

Swedish: V-prt reanalysis is obligatory. Danish: V-prt reanalysis is blocked. Icelandic: V-prt reanalysis is optional.

If the assumptions in (119) are indeed correct, the differences between the perfect tense examples in (113a) to (115a), repeated here as (120), follow immediately. Since Swedish has obligatory V-prt reanalysis, short object shift in (120a) will give rise to a violation of H-COMPL, and is thus blocked. Since Danish does not allow V-prt reanalysis, short object shift in (120b) will not give rise to a violation of H-COMPL, so that the strong ranking of EPP(ij) will force it to apply; of course the ij-features on the auxiliary will not trigger short object shift since this gives rise to a violation of H-COMPL. The alternation in the Icelandic examples in (120c) follows from the optionality of reanalysis: when reanalysis applies, short object shift is blocked by H-COMPL; when reanalysis does not apply, short object shift is forced by the strong ranking of EPP(ij). (120) a. b. c.

Peter har inte kastat bort . Peter has not thrown the-carpet away Peter har ikke smidt ud . Peter has not thrown the-carpet away Pétur hefur ekki hent út . Pétur has not thrown the carpet away

(Swedish) (Danish) (Icelandic)

Now, let us consider the past tense examples. The difference between the Swedish and Danish examples in (113c) and (114b), repeated here as (121), again follows immediately. (121) a. b.

Peter kastade inte bort . Peter threw it not away Peter smed ikke ud . Peter threw it not away

(Swedish) (Danish)

Since Swedish has obligatory V-prt reanalysis, short object shift in (121a) will give rise to a violation of H-COMPL, and is thus blocked; this accounts for the fact that the object cannot appear in between the verb and the particle. Since we also have seen that D-PRONOUN is outranked by H-COMPL, regular object shift is blocked as well; this accounts for the fact that the object cannot be placed in front of negation. Since Danish does not allow V-prt reanalysis, short object shift in (121b) will not give rise to a violation

194 Chapter 3: Regular object shift

of H-COMPL, so that the strong ranking of EPP(ij) will force it to apply. However, the pronoun cannot remain in the position in between the verb and the particle since D-PRONOUN requires short object shift to be followed by regular object shift. This correctly predicts that the pronoun must be placed in the position preceding negation. Now consider the Icelandic examples in (115b&c), repeated here as (122). The placement of the nonpronominal DP mottunni ‘the carpet’ in example (122a) is as expected: as before, the fact that it may either follow or precede the particle follows from the optionality of V-prt reanalysis. Furthermore, when V-prt reanalysis does not apply, the strong ranking of EPP(case) will force the object to undergo regular object shift across negation unless the object is part of the focus of the clause. The placement of the pronominal object in example (122b), on the other hand, is slightly problematic. The fact that it may precede negation and cannot be placed in between the verb and the particle is expected and follows when we assume that V-prt reanalysis does not occur; just as in Danish the pronoun then undergoes both short and regular object shift. However, we would also expect that V-prt reanalysis need not occur, which wrongly predicts that the pronoun should also be able to appear in the position following the particle. (122) a. b.

Pétur henti Pétur threw Pétur henti Pétur threw

ekki út . the carpet not away ekki út . it not away

In order to account for the judgments on the Icelandic example in (122b), we must assume that for some reason V-prt reanalysis is blocked when the direct object is a pronoun. This is not an isolated case since the examples in (123) suggest that the same holds for English. (123) a. b.

John looked up the information/*it. John looked the information/it up.

Den Dikken (1995a: section 2.4.5) accounted for the fact that a pronominal object blocks V-prt reanalysis by postulating the ad hoc stipulation that, contrary to non-pronominal DPs, pronouns cannot be assigned case in their base position. In the remainder of this subsection I will try to develop a more principled account of this fact (which will still leave several questions open). However, before I can do that I have to address another question: Why does the lack of short object shift in the V-prt construction not give rise to a fatal violation of EPP(ij) in English, given that we can always use the construction without V-prt reanalysis in which this violation does not

The restrictions on regular object shift 195

arise? I want to solve this problem by adopting the claim, which will be extensively motivated in Chapter 5, that when the verb selects a small clause complement, EPP(ij) cannot only be satisfied by short object shift of the subject of the small clause, but also by moving the small clause as a whole. Consider the examples in (124). (124) a. *The jeweler put carefully the diamonds into the box. b. The jeweler put the diamonds carefully into the box. c. The jeweler put the diamonds into the box carefully. d. *The jeweler put into the box carefully the diamonds.

Example (124a) is excluded by EPP(ij), given that nothing has moved into the local domain of V. Example (124b) is acceptable since the small clause subject has undergone short object shift, and (124c) is acceptable because EPP(ij) is satisfied by shifting the complete small clause. As I will show in Chapter 5 the choice between (124b) and (124c) is related to the information structure of clause: in (124b) the predicative PP is part of the focus, whereas in (124c) it is part of the presupposition of the clause. Example (124d), finally, shows that movement of the small clause predicate into the local domain of V must pied pipe its subject. Now consider the examples in (125). On December 23, 2007, I performed a Google search on the strings within straight brackets, and obtained the results given between parentheses. 44 Now assume with Den Dikken (1995a) that particles like up are ergative predicates: the DP the information is therefore the complement (internal argument) of the particle up, so that the underlying word order is as given as in (125a). (125) a. *Peter will look [quickly up the information] b. *Peter will look [the information quickly up] c. Peter will look [up the information quickly] d. *Peter will look [up quickly the information]

(0) (0) (176.000) (0)

The fact that the string in (125a) is not found is not surprising, of course, since it is excluded for the same reason as (124a): this order induces a fatal violation of EPP(ij) since neither the small clause subject nor the small 44

Two examples of (125a) were found, but these were part of a text on linguistics where the sentences containing these word orders were marked as ungrammatical. Furthermore I ignored one case of (125b) and two cases of (125d) given that they were part of non-English texts. There were furthermore two cases of (125d), in which the DP the information was not part of the same clause as the particle and the adverb. These kind of cases, of course, also occur in the result for the string in (125c), so that the actual number of relevant cases will be lower.

196 Chapter 3: Regular object shift

clause headed by up has moved into the local domain of V. The constraint EPP(ij) is satisfied in example (125b) by short object shift of the small clause subject and in (125c) by shifting the complete small clause. Example (125b) is nevertheless not attested (cf. also Den Dikken 1995a:40), which can be accounted for under the present set of assumptions by pointing out that verbal particles normally do not convey “new” information; the particle up is therefore part of the presupposition of the clause, so that by hypothesis movement of the full small clause will be preferred. Example (125d) is excluded for the same reason as (124d): movement of the small clause into the local domain of V cannot strand the small clause subject. So far we have only discussed four out of the six possible orders. Now, consider the search results on the two remaining word orders in (126), in which the DP precedes the particle. (126) a. *Peter will look [quickly the information up] b. Peter will look [the information up quickly]

(0) (524)

That example (126a) is not found is again not surprising, since it is excluded for the same reason as (124a) and (125a): this order induces a fatal violation of EPP(ij) since neither the object nor the small clause headed by up has moved into the local domain of V. Although the order in (126b) is clearly grammatical, it is apparently hardly ever used (only about 0.3% of the time). This may seem surprising given that the V-O-prt order can readily be found when the adverb is absent: however, also in this case it is less frequent than the V-prt-O order, which occurs at least 11 times as often:45 (127) a. b.

Peter will look [the information up] Peter will look [up the information]

(63.200) (702.000)

Perhaps we may account for the unpopularity of the order in (126b) by adopting the standard assumption that the object is normally placed in front of the particle by means of (short) object shift. If so, the word order in (126b) could normally be derived only, when short object shift is followed by remnant movement of the small clause. This is, however, impossible for several reasons. First, since movement of the small clause is not necessary to satisfy EPP(ij), which is already satisfied by short object shift of the small 45

The result for the string “the information up” was severely polluted. I found 138.000 occurences of this string, but 74.800 instances of these were part of the larger string “the information up to date”, which clearly does not involve the particle up. Of course, the results may be distorted in other ways as well.

The restrictions on regular object shift 197

clause subject, it is excluded either by the Last Resort Condition on movement or (when we assume that a single probe can enter into an agreement relation with multiple goals) the economy constraint *MOVE. Secondly, this derivation would only derive the desired word order if the remnant small clause were not merged at the root of the structure but instead be “tucked in” below the shifted object. An alternative and more feasible way of deriving the order in (126b) would be to assume that the DP is first moved into some higher small clause-internal position, after which the complete small clause undergoes movement into the local domain of the verbal root V. The difference in frequency between (125c) and (126b), repeated here as the prime-less examples in (128), show that this small clause-internal movement of the DP is apparently not a favored option for non-pronominal DPs. (128) a. a′. b. b′.

Peter will look [up the information quickly] Peter will look [SC up the information] quickly tSC Peter will look [the information up quickly] Peter will look [SC the informationi up ti] quickly tSC

(176.000) (524)

Now contrast the examples in (128a&b) with a non-pronominal small clause subject to those in (129a&b) involving a pronominal one. What we observe is that in (129) the pronoun must precede the particle, that is, the order in (129b) seems ungrammatical: the Google search had 510 results, but I scanned the first 100 of these and in the great majority of cases the three words in the string were not even part of the same clause. Of course, we cannot derive the word order in (129a) by first applying short object shift to the pronoun and subsequent remnant movement of the small clause for the same reasons why this derivation cannot derive the word order in (126b). Consequently, we must assume that the pronoun is normally moved into some higher small clause-internal position before the full small clause is moved into the local domain of V. The fact that (129b), with the structure in (129b′), is ungrammatical shows that this small clause-internal movement of the pronoun is obligatory. (129) a. Peter will look [it up quickly] a′. Peter will look [SC iti up ti] quickly tSC b. *Peter will look [up it quickly] b′. Peter will look [SC up it] quickly tSC

(868.000) (< 25)

The discussion above therefore suggests the observational generalizations in (130): for the discussion to follow, it is especially the first part of the generalization that is relevant.

198 Chapter 3: Regular object shift (130) English: a pronominal internal argument of the particle must undergo small clause-internal movement, whereas a non-pronominal internal argument normally does not undergo this movement.

If this generalization is correct, the fact that V-prt reanalysis is blocked when the internal argument of the particle is a pronoun can be derived by appealing to the independently established fact that V-prt reanalysis requires adjacency of the verb and the particle. Consider the examples in (131), taken from Den Dikken (1995a:106), which show that particle modifiers like right force the object to undergo short object shift. (131) a. *They looked right up the information. b. They looked the information right up

Since short object shift is only allowed when V-prt reanalysis does not take place, the examples in (131) can be used to argue that V-prt reanalysis requires adjacency of the verb and the particle. Now, if small clauseinternal movement of the pronoun is obligatory in English, the pronoun will always intervene between the verb and the particle, so that V-prt reanalysis cannot take place in this case either. Now, we can finally return to the problem that in the Icelandic example in (122b) the particle cannot precede the pronoun. The Icelandic facts in (120c) and (122) suggest that Icelandic is similar to English in allowing short object shift across the particle and requiring the small clause-internal movement of a pronoun. If so, the small clause-internal shift of the pronoun will block V-prt reanalysis, just as in English. Consequently, the strong ranking of EPP(ij) in Icelandic will force regular object shift of the pronoun, and this account for the remaining problem correctly predicts that in (122) the pronominal object cannot remain in its base position following the particle. If the line of argumentation given above is feasible, the distinctions in (119) can be reformulated along the lines in (132). (132) a. b. c.

Swedish: the internal argument of the particle cannot undergo small clause-internal movement Ÿ V-prt reanalysis always applies. Danish: the internal argument of the particle must undergo small clause-internal movement Ÿ V-prt reanalysis never applies. Icelandic: a pronominal internal argument of the particle must undergo small clause-internal movement, whereas a nonpronominal internal argument cannot undergo this movement, unless this is needed as an intermediate step in the derivation of short object shift Ÿ V-prt reanalysis may apply with nonpronominal arguments only.

The restrictions on regular object shift 199

It will be clear that generalizations of this sort are eligible for an optimalitytheoretic account comparable to the one given for regular object shift in this chapter. However, given that providing such an account requires an indepth investigation of the properties and the internal structure of small clauses, I will not develop it any further here but leave it for future research. There is still one final question that must be raised concerning the impossibility of object shift in Swedish verb-particle construction like (114). Of course, if (132a) holds in full force this immediately accounts for the fact that the derivation that leads to (133b) is excluded, but it is perhaps less clear why short object shift could not simply skip the subject position of the small clause, as in (133b′). The same questions can be raised for the representations in (134b&b′) with regular object shift, which feature short object shift as an intermediate step. (133) a.

Peter har inte kastat [SC bort mattan]. Peter has not thrown away the-carpet b. *Peter har inte kastat mattan [SC t′ mattan bort tmattan] b′. *Peter har inte kastat mattan [SC bort tmattan]

(134) a.

Peter kastade inte [SC bort den]. Peter threw not away it b. *Peter kastade den inte t′′den [SC t′den bort tden]. b′. *Peter kastade den inte t′′den [SC bort tden].

The unacceptability of the representations in (133/134b′) follows without further ado when we adopt the plausible assumption that small clauses constitute phases; cf. for example Den Dikken (2006). If so, the subject of the small clause can only be extracted from the small clause via the phase edge: skipping the subject position of the small clause is excluded since it would lead to a violation of Chomsky’s (2001) Phase Impenetrability Condition (PIC); cf. section 4.3.1 for more discussion of the PIC and section 5.2.3 for the claim that small clauses constitute phases. 3.3. Summary This chapter has discussed regular object shift. Section 3.1 first established on the basis of an extensive discussion of Holmberg’s Generalization that the restrictions on Scandinavian regular object shift cannot be attributed to properties of the generator (=CHL). Given the D&E framework adopted here, this implies that these restrictions must be due to the evaluator. Section 3.2 therefore developed an optimality-theoretic account of the restrictions in Table 2, repeated on the next page.

200 Chapter 3: Regular object shift

Section 3.2.1 has shown that, as far as regular object shift is concerned, we can divide languages into three basic types by making use of the three constraints in (66), repeated here as (135): languages of type 1 do not have regular object shift, those of type 2 allow regular object shift both with pronouns and non-pronominal DPs, and those of type 3 have regular object shift with pronouns only. (135) • Constraints needed for deriving the basic typology w.r.t. object shift: a. EPP(case): an unvalued case-feature attracts its goal. b. *MOVE: *t (do not move). c. D-PRONOUN: a weak/definite pronoun must be VP-external. Table 2: Restrictions on object shift and scrambling

FinnishSwedish English Swedish Norwegian Danish Icelandic Dutch German

weak pronouns —

DPs

— optional optional obligatory obligatory obligatory obligatory

— — — — optional optional optional

—

across across IO V n.a. n.a. n.a. — — — — — +

n.a. — — — — + +

across nonverbal particles n.a. n.a. — + + + + +

In addition to the constraints in (135) we have used the three constraints in (136) in order to derive the more fine-grained distinctions. (136) • Constraints needed for fine-tuning the analysis: a. ALIGNFOCUS: the prosodically unmarked focus is the rightmost constituent in its clause. b. RELMIN: if the foot of X-chain Į c-commands the foot of X-chain ȕ, the head of Į c-commands the head of ȕ (X-chain is A-, A′-, or head chain). c. HEAD-COMPLEMENT ( H-COMPL): a head precedes all terminals originally dominated by its complement.

The constraint ALIGNFOCUS was used to account for the fact that in Icelandic and Dutch object shift of full noun phrases is possible only when the noun phrase belongs to the presupposition of the clause. RELMIN was used to account for the fact that the Scandinavian languages and Dutch do

Summary 201

not allow object shift to cross the indirect object. H-COMPL, finally, was used to account for the fact that Scandinavian object shift cannot cross the main verb, or the verbal particle after V-prt reanalysis has taken place. When we include the two constraints in (137), which were already introduced in Chapter 2, our analysis of short and regular object shift involves eight constraints in total. Below I will summarize the ranking of these constraints for the languages we have discussed. (137) a. b.

EPP(ij): unvalued ij-features attract their goal. *STRAY FEATURE: Attach functional features to

the root they are

associated with.

We have discussed only one type 1 language, Finnish-Swedish. The ranking of this language is given in (138). The fact that *MOVE outranks both EPP(case) and D-PRONOUN predicts that it has no regular object shift, whereas the subranking EPP(ij) >> ALIGNFOCUS >> *MOVE predicts that Finnish-Swedish has obligatory short object shift. Note that if it turns out that verbal particles block short object shift in this language, just as in Swedish, this would show that EPP(ij) is outranked by H-COMPL; I have no data available that establish this, though. (138) Finnish-Swedish (type 1): EPP(ij) >> ALIGNFOCUS >> *MOVE >> {D-PRONOUN,

EPP(case)}

We have discussed three type 2 languages: Icelandic, Dutch, and German. We have seen in section 3.2.1.2 that EPP(case) is outranked by ALIGNFOCUS in these languages and that this accounts for the fact that regular object shift only takes place when the object is part of the presupposition of the clause. Since short object shift is not sensitive to the information structure of the clause, EPP(ij) must outrank ALIGNFOCUS. The ranking of all these languages is therefore as given in (139). (139) Germanic full object shift languages (type 2): EPP(ij) >> ALIGNFOCUS >> EPP(case) >> *MOVE

Icelandic, on the one hand, and Dutch and German, on the other, differ in that the former is a VO-language, whereas the latter are OV-languages. We have seen in section 2.4 that this can be accounted for by assuming that Icelandic has obligatory V-to-v, whereas Dutch and German do not. One way to account for this is to assume that whereas Icelandic has a strong ranking of *STRAY FEATURE, Dutch and German have a weak ranking of this constraint. Since this will be spelled out in more detail in the following

202 Chapter 3: Regular object shift

chapter, I will not yet include this constraint in the rankings below. We have further seen in section 3.2.3 that regular obect shift is not allowed in complex tense construction in Icelandic, whereas this is allowed in their Dutch/German counterparts. From this we concluded that in Icelandic the constraint H-COMPL outranks both EPP(ij) and EPP(case), whereas in Dutch/German the ranking is the opposite. (140) a. b.

Icelandic: H-COMPL >> EPP(ij) >> ALIGNFOCUS >> EPP(case) >> *MOVE Dutch/German: EPP(ij) >> ALIGNFOCUS >> EPP(case) >> *MOVE >> H-COMPL

Section 3.2.2 has shown that regular object shift across the indirect object is blocked in Icelandic and Dutch, but allowed in German. This can be accounted for by assuming that RELMIN outranks ALIGNFOCUS in Icelandic and Dutch, but not in German.46 We have further seen that two strategies are used to satisfy RELMIN in Icelandic and Dutch. When the direct object is a non-pronominal DP, an indirect object that is in focus of the clause will block regular object shift, whereas a pronominal direct object will force regular object shift of the indirect object. We have seen that this can be accounted for by assuming that in these languages ALIGNFOCUS is outranked by D-PRONOUN. Finally, the fact that Icelandic perfect tense constructions do not have regular object shift of pronouns shows that D-PRONOUN is outranked by H-COMPL in this language. This gives rise to the rankings in (141). (141) a. b. c.

Icelandic: H-COMPL >> {EPP(ij), D-PRONOUN, RELMIN} >> ALIGNFOCUS >> EPP(case) >> *MOVE Dutch: {EPP(ij), D-PRONOUN, RELMIN} >> ALIGNFOCUS >> EPP(case) >> *MOVE >> H-COMPL German: {EPP(ij), D-PRONOUN } >> ALIGNFOCUS >> RELMIN >> EPP(case) >> *MOVE >> H-COMPL

Danish was used as an illustration of the third type of language, which has regular object shift of pronominal objects only. This was accounted for by assuming that Danish has the subranking in (142a). Since short object shift is possible with all noun types and is not sensitive to the information of the clause, it follows that Danish also has the subranking in (142b). Just 46 Recall from section 3.2.2.1 that we do not have direct evidence for the precise ranking of RELMIN in German, but in order to keep the difference with Dutch and Icelandic as small as possible we have provisionally assumed in (141c) that RELMIN is ranked in between ALIGNFOCUS and EPP(case).

Summary 203

as in Icelandic, object shift may not cross the main verb or an indirect object in Danish; the former shows that H-COMPL outranks both EPP(ij) and D-PRONOUN, as in (142c), and the latter that RELMIN outranks both D-PRONOUN and ALIGNFOCUS, as in (142d). Finally, we have seen that there are two strategies to avoid a violation of RELMIN: an indirect object that is part of the focus of the clause will block object shift of a pronominal direct object, whereas a pronominal direct object will force object shift of an indirect object that is part of the presupposition of the clause. Section 3.2.2.2 has shown that this follows when we assume that D-PRONOUN is outranked by ALIGNFOCUS, as in (142e). When we put all these bits and pieces together, we get the ranking in (142f). (142) a. b. c. d. e. f.

D-PRONOUN >> *MOVE >> EPP(case) EPP(ij) >> ALIGNFOCUS >> *MOVE H-COMPL >> {EPP(ij), D-PRONOUN} RELMIN >> {ALIGNFOCUS, D-PRONOUN} ALIGNFOCUS >> D-PRONOUN Danish (type 3): H-COMPL >> EPP(ij) >> RELMIN >> ALIGNFOCUS >> D-PRONOUN >> *MOVE >> EPP(case)

Finally we have seen that Swedish is of a mixed type and that it mainly differs from Danish in that D-PRONOUN and *MOVE are in a tie, as a result of which pronominal object shift is truly optional; cf. (143). (143) Swedish (mixed type 1/3): H-COMPL >> EPP(ij) >> RELMIN >> ALIGNFOCUS >> D-PRONOUN *MOVE >> EPP(case)

In section 4.1, we will see that Icelandic and the other Germanic languages discussed here differ in their ranking of *STRAY FEATURE: Icelandic has the strong ranking *STRAY FEATURE >> *MOVE, as a result of which it has V-to-I both in main and in embedded clauses, whereas the other languages have the weak ranking *MOVE >> *STRAY FEATURE as a result of which V-to-I normally does not apply in embedded clauses.

Chapter 4 Object shift and other movement types

The previous chapter has given an account of the data that are normally discussed in the literature on object shift. The proposal itself, however, raises a number of new questions that will be discussed in this section. For example, we have seen in section 2.4.2.5 that there are reasons to assume that in Danish and Swedish short object shift “pushes up” the verbal root V into v/Asp, whereas we have seen in section 3.2.1 that regular object shift is blocked in these languages when V-to-I does not apply. This raises the question why regular object shift is unable to push the verb up into I in this case. This question will be discussed in section 4.1. Recall that the analysis in the previous chapter is based on the assumption that short and regular object shift are triggered by the strong ranking of EPP(ij) and EPP(case). Since the ij- and case features are also involved in triggering movement of the subject, this raises the question what the rankings we have established so far predict for this movement. This question will be discussed in section 4.2. Finally, we have seen in Chapter 3 that in some cases object shift seems to interact with A′-movement: so far I have put this issue aside, but section 4.3 will take it up and discuss it at some length. This section thus provides a discussion of the relation between object shift and the three movement types that are normally distinguished: headmovement, A-movement, and A′-movement. Since this topic is hardly discussed in the literature, this chapter should be seen as no more than a first inventory of the questions that could be raised in this context. I believe that the study of the relation between object shift and the other movement types constitutes an important area for future research that may add many new insights to our current understanding of core grammar. 4.1. Object shift and verb movement Section 2.4.2.5 has shown that there are reasons to assume that in Danish and Swedish short object shift “pushes up” the verbal root V into v/Asp in order to satisfy H-COMPL. Section 3.2.1, on the other hand, has shown that H-COMPL blocks regular object shift in these languages when V-to-I does not apply. This gives rise to the question why the push-up strategy is

206 Chapter 4: Object shift and other movement types

possible with short, but not with regular, object shift. This section will discuss this question. 4.1.1.

The Danish/Swedish verb movement puzzle

In the D&E framework verb movement can be accounted for by assuming that the functional heads in the extended projection of the verbal root V contain unvalued formal features that must be valued by V. Consequently verb movement will be forced if the EPP constraints associated with these features outrank *MOVE. For example, a strong ranking of EPP(v) will force movement of the verbal root V into the functional head v, a strong ranking of EPP(asp) will force movement of the verbal root V into Asp, and a strong ranking of EPP(tense) will force movement of the finite verb (the V+v complex) into the inflectional head I. Section 2.4.2.5 has suggested that these three EPP constraints can be unified into a single constraint *STRAY FEATURE, which requires that the functional features of the functional heads be amalgamated with the root they are associated with. When this is indeed possible, the strong ranking *STRAY FEATURE >> *MOVE defines languages in which the verbal root V obligatorily moves into v/Asp and the finite verb obligatorily moves into I (unless some other constraint outranking *STRAY FEATURE blocks these movements). An example of such a language is Icelandic, which has obligatory V-to-I in both main and embedded clauses. This is illustrated by the prime-less examples in (1): the fact that the finite verb precedes the shifted object is clear evidence for this movement, as is shown in the simplified structures in the primed examples in (1). (1)

a. a′ b. b′.

Jón las ekki . Jón read this book not [IP S V+v+I [O tV+v ekki [t′O tV tO]]] að Jón las ekki . that Jón read this book not [CP að [IP S V+v+I [O tV+v ekki [t′O tV tO]]]]

The weak ranking *MOVE >> *STRAY FEATURE defines languages in which movement of the verbal root V into v/Asp or V-to-I normally does not apply. The Germanic OV-languages were claimed to be examples of this language type: since the subranking *MOVE >> *STRAY FEATURE normally blocks V-to-v/Asp and V-to-I, short object shift results in the OVorder found in perfect tense constructions and embedded clauses. This is illustrated in, respectively, (2a) and (2b). Example (2c) shows that in main clauses the finite verb does undergo V-to-I, which is clear from the fact that

Object shift and verb movement 207

it must precede the scrambled object. This must, of course, be attributed to some constraint outranking *MOVE that favors verb second. (2)

a. b. c.

Jan heeft waarschijnlijk verkocht . Jan has his car probably sold dat Peter waarschijnlijk verkoopt . that Peter his car probably sells Peter verkoopt waarschijnlijk . Peter sells his car probably

The fact that the verb follows the sentential/negative adverbs in (3) shows that V-to-I does not apply in embedded clauses in Danish and Swedish either. This suggests that in these languages *STRAY FEATURE is also outranked by *MOVE. (3)

a. b.

Jeg I att that

spurgte asked Ulf inte Ulf not

hvorfor why köpte bought

Peter aldrig læste den. Peter never read it boken the book

(Danish) (Swedish)

A potential problem for this claim, however, is that it wrongly predicts that V-to-v/Asp is also blocked in Danish and Swedish, so that short object shift should result in inversion of the verb and the object: the examples in (3) show that the verb also precedes the object in embedded clauses. Section 2.4.2.5 suggested that this problem can be solved by adopting the constraint H-COMPL: the subranking H-COMPL >> EPP(ij) >> *MOVE forces V-to-v/Asp when short object shift crosses the verb. For convenience, the constraint rankings that were proposed in 2.4.2.5 are repeated here as (4). (4)

a.

b.

Germanic VO-languages: (i) Icelandic: H-COMPL >> EPP(ij) >> *STRAY FEATURE >> *MOVE (ii) Swedish/Danish: H-COMPL >> EPP(ij) >> *MOVE >> *STRAY FEATURE Germanic OV-languages: EPP(ij) >> *MOVE >> {*STRAY FEATURE, H-COMPL}

Putting the verb second phenomenon aside for the moment, the rankings in (4) account nicely for the data in (1) to (3), although a new and serious problem arises in the case of Danish and Swedish, which I will illustrate below for Danish. Since we have already established in Chapter 3, ex. (143), that the constraint D-PRONOUN is ranked in between EPP(ij) and *MOVE, we

208 Chapter 4: Object shift and other movement types

are dealing with the Danish ranking in (5); constraints irrelevant for or present discussion are omitted. (5)

Danish: H-COMPL >> EPP(ij) >> D-PRONOUN FEATURE, EPP(case)}

>> *MOVE >> {*STRAY

The subranking H-COMPL >> D-PRONOUN >> *MOVE predicts that regular object shift of definite pronouns should also force the V+v complex to move into I, just like the subranking H-COMPL >> EPP(ij) >> *MOVE predicts that short object shift forces the verbal root V complex to move into the light verb v. Consequently we do not derive the attested pattern in (6), but wrongly predict Danish to exhibit the pattern in (7), where V-to-I applies in embedded clauses when the object is a weak pronoun, but not when it is a non-pronominal DP. (6)

(7)

• Attested Danish pattern: a. Jeg spurgte hvorfor I asked why b. Jeg spurgte hvorfor I asked why

Peter aldrig Peter never Peter aldrig Peter never

læste read læste read

den her bog this book den. it

• Predictions for Danish on the basis of the constraint ranking in (5): a. Jeg spurgte hvorfor Peter aldrig læste den her bog (correct) I asked why Peter never read this book b. Jeg spurgte hvorfor Peter læste den aldrig. (incorrect) I asked why Peter read it never

That we predict the pattern in (7) is shown in Tableaux 1 and 2. The relevant candidates are given in (8): the a-set are cases in which the root V remains in its base position, the b-set has V-to-v, and the c-set has V-to-I. The prime-less examples have no object shift, the single primed examples have short object shift, and the doubly primed examples have short and regular object shift. The abbreviation C in these candidates stands for the lexical complementizer, which happens to be phonetically empty in (6) and (7) due to the presence of a wh-word in SpecCP; cf. the remark in fn.47 on p.210. Note further that I have given the violations of *MOVE that arise from object shift and verb movement in separate rows; this is of no theoretical significance but for convenience only. (8)

a. C ... I ... v ... V O/pronoun a′. C ... I ... v ... O/pronoun V tO a′′. C ... I ... O/pronoun v ... t′O V tO

Object shift and verb movement 209 b. b′. b′′. c. c′. c′′.

C ... I ... V+v ... tV O/pronoun C ... I ... V+v ... O/pronoun tV tO C... I ... O/pronoun V+v ... t′O tV tO C ... V+v+I ... tv tV O/pronoun C ... V+v+I ... tv O/pronoun tV tO C ... V+v+I ... O/pronoun tv ... t′O V tO

Tableau 1: V-to-I in Danish embedded clauses (object ≠ pronoun) HCOMPL

(8a) (8a′) (8a′′) (8b) (8b′) ) (8b′′) (8c) (8c′) (8c′′)

EPP

(ij) *!

DPRONOUN

*! *!

*MOVE (t0) (tV/tv) * **

*! * **

*! *!

* **!

* * * ** **! **!

*STRAY F ** ** ** * * *

EPP

(case) * * * * * *

Tableau 2: V-to-I in Danish embedded clauses (Object = pronoun)

Object = pronoun (8a) (8a′) (8a′′) (8b) (8b′) (8b′′) (8c) (8c′) (8c′′) /

HCOMPL

EPP

(ij) *!

*! *! *!

DPRONOUN

* * * *!

*! *!

* *!

*MOVE (t0) (tV/tv) * ** * ** * **

* * * ** ** **

*STRAY F ** ** ** * * *

EPP

(case) * * * * * *

The acceptability of (6b) and the unacceptability of (7b) show that in the case of Danish pronoun shift the expected push-up strategy is excluded when it comes to V-to-I; only the blocking strategy in (3) is possible. We must therefore account for the fact that short and regular object shift differ by assuming that the former favors the push-up strategy, whereas the latter favors the blocking strategy. Below I will argue that this follows from the

210 Chapter 4: Object shift and other movement types

interaction of these movements with V-to-I. For this reason, I will start by discussing V-to-I in section 4.1.2; however, since investigating in detail the many intricacies involved in V-to-I in the Germanic languages would lead me too far astray, I will confine myself to establishing the main tendencies. Section 4.1.3 will conclude with providing a solution to the Danish/Swedish verb movement puzzle outlined above. 4.1.2.

V-to-I in the Germanic languages

This section will provide an analysis of verb movement in the Germanic languages; since we have already discussed V-to-v/Asp in section 2.4, emphasis will be on V-to-I. The constraints that will be featured in the analysis are given in (9). (9)

a. b. c.

*STRAY FEATURE: amalgamate formal features of the functional heads with the root they are associated with. LEXICALLY FILL TOP F (LFTF): the highest head position in an extended projection must be lexically filled. NOLEXM: a ș-role assigning head remains in its ș-domain (a main verb does not move).

The constraint *STRAY FEATURE was already discussed earlier: it is the unification of the EPP constraints EPP(v), EPP(Asp) and EPP(tense), and its ranking relative to *MOVE will determine whether the verbal root V will move into the higher functional heads (v/Asp and I) in its extended projection. The constraint LEXICALLY FILL TOP F can be held responsible for the verb second phenomena in languages like Danish/Swedish and the Germanic OV-languages: LFTF >> *MOVE >> *STRAY FEATURE predicts that in the absence of a complementizer the verb must be moved into the highest functional projection in the extended projection of the verb.47 When we further assume that complementizers must be selected, and can 47 Of course, the notion “lexically filled” in (9b) must be distinguished from the notion “phonetically realized”. More specifically, it may be the case that a complementizer that is lexically present is phonetically empty as a result of the constraint TEL; see section 1.1.2 for discussion. The constraint LEXICALLY FILL TOP F is very similar to the following principle proposed in Zwart (2001): “Let Ȗ be a chain of F-related elements (Į, ..., ȕ), where Į c-commands ȕ. Then Į must contain LEX-features, and ȕ is spelled out in the highest positions of Ȗ containing the LEXfeatures of ȕ”, where “Į and ȕ are F-related if Į is involved in a feature valuation involving F, where F is a formal feature of ȕ”. The intended effect is as follows: if Į has no LEX-features of its own, the LEX-features of ȕ are moved to Į, so that ȕ itself is also spelled out in Į; if Į does have LEX-features of its own, movement of the LEX-features of ȕ is exluded and ȕ is spelled out in its original position.

Object shift and verb movement 211

therefore only appear in embedded clauses, it will follow that verb second must apply in main clauses (and in embedded clauses which are complementizer-less for one reason or other). The third constraint, NOLEXM, is adapted from Grimshaw (1997) and goes back at least to Pollock (1989): its role is to block movement of main verbs, while still allowing movement of auxiliary and modal verbs, and it is used to account for the fact that in English V-to-I is restricted to modal and auxiliary verbs, that is, cannot apply to main verbs. The three constraints in (9) in tandem with the general economy constraint *MOVE suffice to provide a basic typology of V-to-I. These four constraints define 24 different languages of six typologically different kinds. Table 1 shows that three of these types are attested in my sample from the Germanic language family. Table 1: V-to-I in the Germanic languages RANKING

1 *MOVE >> {*STRAY FEATURE, LFTF}; ranking NOLEXM not relevant 2 LFTF >> *MOVE >> *STRAY FEATURE; LFTF >> NOLEXM 3 NOLEXM >> LFTF >> *MOVE >> *STRAY FEATURE 4 *STRAY FEATURE >> {*MOVE, NOLEXM}; ranking LFTF not relevant 5 NOLEXM >> *STRAY FEATURE >> *MOVE; NOLEXM >> LFTF 6 LFTF >> NOLEXM >> *STRAY FEATURE >> *MOVE

# 24 8

MAIN

EMBEDDED

CLAUSES

CLAUSES

LANGUAGE

—

—

not in sample

3

+

—

1

aux only

—

Dutch/German Danish/Swedish not in sample

8

+

+

Icelandic

3

aux only

aux only

English

1

+

aux only

not in sample

The first type arises when *MOVE outranks the two constraints that favor V-to-I, *STRAY FEATURE and LEXICALLY FILL TOP F, and is characterized by having no V-to-I at all. To my knowledge there are no languages of this type within the Germanic language family. Probably, languages like Japanese that exhibit a strict OV-order are of this type. The second and the third types are languages with verb second in main clauses only. The weak ranking of *STRAY FEATURE blocks V-to-I in embedded clauses, whereas the subranking LFTF >> *MOVE forces V-to-I in main clauses. The second and third type differ in the relative ranking of NOLEXM and LFTF. The third type has the subranking NOLEXM >> LFTF and therefore allows verb second with auxiliary and modal verbs only; to my knowledge there are no languages of this type within present-day Germanic, but Tonya Kim Dewey (p.c.) informs me that reconstructed Proto-

212 Chapter 4: Object shift and other movement types

Germanic is of this type; cf. Fortson (2004) and Dewey (2006). The second type has the subranking LFTF >> NOLEXM, and allows verb second in main clauses with all verbs. The Germanic OV-languages and Swedish/Danish are examples of this type, as is illustrated by the examples in (10) and (11). The primed examples show that V-to-I is categorically exluded in embedded clauses, whereas the prime-less examples show that in main clauses V-to-I applies both to lexical and auxiliary verbs. (10)

(11)

• Dutch a. Jan verkoopt zijn auto waarschijnlijk. Jan sells his car probably a′. dat Jan zijn auto waarschijnlijk verkoopt. that Jan his car probably sells b. Jan heeft zijn auto waarschijnlijk verkocht. Jan has his car probably sold b′. dat Jan zijn auto waarschijnlijk heeft verkocht. that Jan his car probably has sold

(V-to-I) (No V-to-I) (aux-to-I) (no aux-to-I)

• Swedish (Holmberg & Platzack 1995:71-2, Sells 2001:9) a. Ulf köpte inte boken. (V-to-I) Ulf bought not the book a′. att Ulf inte köpte boken (no V-to-I) that Ulf not bought the book b. Han har inte köpt boken. (aux-to-I) he has not bought the book b′. att han inte har köpt boken. (no aux-to-I) that he not has bought the book

The fourth type arises when *STRAY FEATURE outranks the constraints that may block movement, NOLEXM and *MOVE, and is defined by allowing V-to-I with all verbs in all contexts. This type is instantiated by Icelandic. (12)

• Icelandic (Holmberg & Platzack 1995:71, Jónsson 1996:9-10) a. Jón las ekki þessa bók. (V-to-I) Jón read not this book a′. að Jón las ekki þessa bók. (V-to-I) that Jón read not this book b. Jón hefur ekki lesið þessa bók. (aux-to-I) Jón has not read this book b′. að Jón hefur ekki lesið þessa bók. (aux-to-I) that Jón has not read this book

Object shift and verb movement 213

The fifth and the sixth type arise when we have a strong ranking of *STRAY FEATURE, but NOLEXM outranks *STRAY FEATURE. This results in languages that normally have V-to-I with auxiliary/modal verbs only. The two types differ in the relative ranking of NOLEXM and LEXICALLY FILL TOP F. When the former outranks the latter, V-to-I is restricted to auxiliary/model verbs in all contexts. English is an example of this type, which is clear from the examples in (13), where the dichotomy between main and auxiliary verbs is clear from their position relative to the sentential adverb probably. (13)

• English a. John probably bought the book. a′. that John probably bought the book. b. John has probably bought the book. b′. that John has probably bought the book.(aux-to-I)

(no V-to-I) (no V-to-I) (aux-to-I)

On the subranking LEXICALLY FILL TOP F >> NOLEXM, V-to-I of main verbs is restricted to main clauses, whereas aux-to-I may apply in all contexts. To my knowledge there are currently no Germanic languages of this type, but Tonya Kim Dewey (p.c.) again informs me that reconstructed Proto-Indo-European exhibits this behavior; cf. Fortson (2004) and Dewey (2006). 4.1.3.

Solving the Danish/Swedish verb movement puzzle

Now that we have discussed the constraints underlying V-to-I in the Germanic languages, we can start to solve the Danish/Swedish verb movement puzzle: why does short object shift force V-to-v/Asp, whereas regular object shift does not force V-to-I? Let us analyze this problem step by step. First, we have seen in the previous subsection that the subranking in (14a) predicts that in embedded clauses V-to-v/Asp and V-to-I are blocked, unless forced by some other constraint outranking *MOVE. Secondly, we have seen that the subranking in (14b) forces application of short object shift and regular object shift of pronouns, unless this is blocked by some other constraint outranking EPP(ij) and D-PRONOUN. Thirdly, Section 2.4.2.5 has shown that subranking (14c) will force V-to-v/Asp when short object shift applies. Subranking (14d) therefore wrongly predicts that regular pronoun shift also forces V-to-I.

214 Chapter 4: Object shift and other movement types (14)

a. b. c. d.

LFTF >> *MOVE >> *STRAY FEATURE EPP(ij) >> D-PRONOUN >> *MOVE >> EPP(case) H-COMPL >> EPP(ij) >> *MOVE H-COMPL >> D-PRONOUN >> *MOVE

In principle, we can take two different routes to solve this problem. The first route is to find a way to block regular object shift in the relevant contexts, while still allowing short object shift. As far as I can see, this requires the postulation of some new constraint: taking this route should therefore be dismissed as ad hoc, unless independent evidence can be given for this new constraint. The second route makes use of the already independently established fact that EPP(ij) outranks D-PRONOUN in Danish/Swedish; cf. (142/143) in Chapter 3. When we postulate a constraint C that blocks verb movement and rank this constraint in between EPP(ij) and D-PRONOUN, the desired effect will be derived. What we have to do then is to identify this constraint C, which, of course, should preferably be independently motivated. The question therefore is whether there is a constraint that could block V-to-I in example (3b′), repeated here as (15b′). (15)

a.

Jeg I b. Jeg I b′. *Jeg I

spurgte asked spurgte asked spurgte asked

hvorfor Peter aldrig why Peter never why Peter aldrig why Peter never hvorfor Peter læste why Peter read

læste den her bog. read this book læste den. read it den aldrig. it never

This question can be answered in the affirmative: V-to-I of the main verb læste can in principle be blocked by the constraint NOLEXM. Tableau 3 shows the evaluation that arises when we rank this constraint in between EPP(ij) and D-PRONOUN. The relevant candidates in (8) are repeated here as (16). The candidate with short object shift and V-to-v/Asp, but no regular object shift and V-to-I is now correctly selected as the optimal candidate. Note that none of the violations of D-PRONOUN is fatal. This correctly predicts that when we replace the pronominal object by a full noun phrase the same candidate will be selected as the optimal one; cf. (15a). (16)

a. a′. a′′. b. b′. b′′.

C ... I ... v ... V pronoun C ... I ... v ... pronoun V tO C ... I ... pronoun v ... t′O V tO C ... I ... V+v ... tV pronoun C ... I ... V+v ... pronoun tV tO C... I ... pronoun V+v ... t′O tV tO

Object shift and verb movement 215 c. C ... V+v+I ... tv tV pronoun c′. C ... V+v+I ... tv pronoun tV tO c′′. C ... V+v+I ... pronoun tv ... t′O V tO

* *

*! *!

* * * ** **! **!

* * * *

*MOVE (tV/tv)

*MOVE (t0) * ** * ** * **

* * * ** ** **

EPP(case)

*!

DPRONOUN

NOLEXM

EPP(ij)

*! *! *!

*STRAYF

(16a) (16a′) (16a′′) (16b) (16b′) ) (16b′′) (16c) (16c′) (16c′′)

LFTF

HCOMPL

Tableau 3: V-to-I and regular object shift in Danish embedded clauses

** ** ** * * *

* * * * * *

Tableau 4 gives the evaluation of the main clause counterpart of (15b) in (18b). The candidates in (17) are similar to those in (16) in most respects, but crucially differ from them in that the complementizer is lacking. The evaluation shows that as a result all examples in which V-to-I fails to apply are now excluded by LEXICALLY FILL TOP F (in tandem with H-COMPL). This leaves us with three remaining candidates that violate NOLEXM to the same extent, so that EPP(ij) and D-PRONOUN get the final say: the optimal candidate is the one in which both short and regular object shift of the pronominal object have applied. (17)

a. a′. a′′. b. b′. b′′. c. c′. c′′.

... I ... v ... V pronoun ... I ... v ... pronoun V tO ... I ... pronoun v ... t′O V tO ... I ... V+v ... tV pronoun ... I ... V+v ... pronoun tV tO ... I ... pronoun V+v ... t′O tV tO ... V+v+I ... tv tV pronoun ... V+v+I ... tv pronoun tV tO ... V+v+I ... pronoun tv ... t′O V tO

216 Chapter 4: Object shift and other movement types

*>

*! *< *< *! *!

*>

*STRAYF

(17a) (17a′) (17a′′) (17b) (17b′) (17b′′) (17c) (17c′) (17c′′) )

LFTF

HEADCOMPL

Tableau 4: V-to-I and pronoun shift in Danish main clauses

** ** ** * * *

* * * * * *

Note, finally, that when the object is a non-pronominal DP, as in (18a), D-PRONOUN will not be violated, so regular object shift will be blocked by *MOVE. (18)

a. b.

Hvorfor Why Hvorfor Why

læste read læste read

Peter aldrig . Peter this book never Peter aldrig . Peter it never

This concludes our discussion of the Danish/Swedish verb movement puzzle. The puzzle is solved by ranking the independently motivated economy constraint NOLEXM in between EPP(ij) and D-PRONOUN: the subranking EPP(ij) >> NOLEXM accounts for the fact that short object shift forces V-to-v; the subranking NOLEXM >> D-PRONOUN accounts for the fact that regular object shift of pronouns is blocked when V-to-I of the main verb does not apply for independent reasons. 4.2. Object shift and subject shift (A-movement) Chapter 2 and Chapter 3 have argued that short and regular object shift are triggered respectively by the ij- and case features on the verb. When the constraints EPP(ij) and EPP(case) have a strong ranking, object shift must apply unless these constraints are outranked by some other constraint that disfavors object shift. Given that movement of the subject is generally assumed to be triggered by the unvalued ij- and case features on the

Object shift and subject shift (A-movement) 217

functional head I, and given that the constraints EPP(ij) and EPP(case) simply require that unvalued ij- and case features trigger movement of their goal and thus do not distinguish between subjects and objects, it follows that the proposed ranking of EPP(ij) and EPP(case) should not only enter the description of object shift, but also that of movement of the subject into SpecIP (henceforth: subject shift). We concluded earlier that all Germanic languages have the ranking EPP(ij)>> ALIGNFOCUS>>*MOVE on the basis of the fact that short object shift is not sensitive to the information structure of the clause, and is therefore obligatory (unless it is blocked by H-COMPL). Consequently, if the claim that I has unvalued ij-features is correct, this predicts that subject shift is likewise obligatory. This prediction is indeed correct for the Germanic VO-languages (although some complications arise in Icelandic due to the existence of quirky subjects), but the Dutch examples in (19) show that it is clearly false for the Germanic OV-languages: the fact that the sentential adverb waarschijnlijk ‘probably’ can be either preceded or followed by the subject mijn vader ‘my father’ suggests that subject shift is in fact optional. (19)

a.

b.

Morgen komt waarschijnlijk mijn vader op bezoek. Tomorrow comes probably my father on visit ‘Probably, my father will visit me tomorrow.’ Morgen komt mijn vader waarschijnlijk op bezoek.

Nevertheless, the restrictions on subject shift in examples like (19) are similar to those on regular object shift (cf. Diesing 1992, De Hoop 1992, and many others): section 4.2.1 will show that the subject exhibits the behavior that can be expected on the basis of the subranking ALIGNFOCUS>>EPP(case)>>*MOVE. This suggests that subject and object shift are indeed subject to similar restrictions, but that some yet unknown factor interferes with the predictions that follow from the subranking EPP(ij)>> ALIGNFOCUS>>*MOVE. Section 4.2.2 will argue that this factor is related to differences in the agreement relations between the object and the verb, on the one hand, and the subject and I, on the other. It proposes that the constraint EPP(ij) should be split up into the constraints EPP(gender), EPP(number), and EPP(person), and that the dichotomy between short object shift and subject shift follows from the ranking EPP(gender)>> ALIGNFOCUS>>EPP(person) >> *MOVE. Section 4.2.3 will show that splitting up EPP(ij) is also necessary to account for the fact that short object shift cannot cross the main verb in the Germanic VO-languages, while subject shift can: this will follow when we assume that these languages

218 Chapter 4: Object shift and other movement types

have the subranking EPP(person) >> H-COMPL >> EPP(gender) >> *MOVE. Section 4.2.4 will conclude this section with a discussion of the effects of the constraint RELMIN on subject shift. 4.2.1.

Subject shift and the constraint EPP(case)

This section will show that the restrictions on subject shift in the Germanic OV-languages resemble those on regular object shift. Section 3.2.1 has shown that the subranking in (20) predicts that regular object shift is obligatory in the Germanic OV-languages, unless the object is part of the focus of the clause. (20)

Germanic OV-languages: ALIGNFOCUS>> EPP(case) >> *MOVE

From this it follows that regular object shift of a definite noun phrase is apparently optional: object shift is forced by the strong ranking of EPP(case) when the object is part of the presupposition of the clause, but is blocked by ALIGNFOCUS when it is part of the focus of the clause. This is illustrated again in (21). The underlining in these examples indicates sentence accent, which is normally assigned to the focus of the clause. (21)

a.

b.

dat Jan waarschijnlijk mijn huis zal kopen. that Jan probably my house will buy ‘that Jan will probably buy my house.’ dat Jan mijn huis waarschijnlijk zal kopen.

Note in passing that (21a) and the other a-examples below are also acceptable when sentence accent is placed on the main verb, in which case the focus of the clause consists of both the object and the verb. Such cases differ from example (21b) with object shift in that the latter is only acceptable when the object is construed as part of the presupposition of the clause. Since non-specific, indefinite noun phrases introduce a new discourse referent, they are by definition part of the focus of the clause. The constraint ranking in (20) therefore correctly predicts that they cannot undergo object shift. (22)

a.

dat Jan waarschijnlijk een huis zal kopen. that Jan probably a house will buy b. *dat Jan een huis waarschijnlijk zal kopen.

Specific indefinite objects behave just like definite objects: they only shift when they are part of the presupposition of the clause. Consequently, nonshifted indefinite objects like twee artikelen in (23a) are ambiguous and can

Object shift and subject shift (A-movement) 219

have either a specific or a non-specific reading; the shifted object in (23b), on the other hand, has a specific reading only. (23)

a.

b.

dat Jan waarschijnlijk twee artikelen zal lezen. that Jan probably two (of the) articles will read Ambiguous: ‘that Jan will probably read two (of the) articles.’ dat Jan twee artikelen waarschijnlijk zal lezen. that Jan two of the articles probably will read Specific only: ‘that Jan will probably read two of the articles.’

The notions of non-specific and specific indefinite noun phrase are normally described as “unknown to speaker and hearer” and “unknown to hearer only”, respectively. Specific indefinite NPs can often be paraphrased by means of a partitive phrase: under its specific reading twee artikelen in (23) is construed as “two of the articles”. Finally, the examples in (24) show that weak definite pronouns obligatorily shift, which is due to the fact that they are never part of the focus of the clause. The strong ranking of EPP(case) therefore correctly predicts regular object shift to be obligatory. (24)

a.

*dat Jan waarschijnlijk het koopt. that Jan probably it buys b. dat Jan het waarschijnlijk koopt.

For completeness’ sake example (25b) shows that contrastively stressed pronouns need not shift. In Dutch, contrastive stressing of third person pronouns is possible with masculine and feminine (but not neuter) pronouns, provided that they refer to a [+animate] entity (Broekhuis et al. 2003). Contrastive stress is indicated by means of small capitals. (25)

Je hebt waarschijnlijk HEM/HAAR/*HET you have probably HIM/HER/IT ‘You probably saw HIM/HER/IT.’

gezien. seen

Now, consider the corresponding examples involving a subject. The examples in (26) show that a definite noun phrase acting as the subject need not be moved into the “regular” subject position adjacent to the complementizer, but may remain in a lower position (presumably its base position, SpecvP). These examples also show that, as in the case of object shift, the application of subject shift depends on the information structure of the clause: the subject can only remain in its base position when it is part of

220 Chapter 4: Object shift and other movement types

the focus of the clause, but when the subject is part of the presupposition of the clause, subject shift is obligatory. (26)

a.

b.

dat waarschijnlijk de buurman Jans huis zal kopen. that probably the neighbor Jan’s house will buy ‘that the neighbor will probably buy Jan’s house.’ dat de buurman waarschijnlijk Jans huis zal kopen.

The examples in (27) show that non-specific, indefinite noun phrases do not shift into the regular subject position. Just like the object in (23), the subject may get either a specific or a non-specific interpretation when it follows the sentential adverb, but must be interpreted specifically when it precedes the adverb (cf., e.g., De Hoop & Swart 1990).48 (27)

a.

b.

dat waarschijnlijk twee studenten Jans huis zullen kopen. that probably two (of the) students Jan’s house will buy Ambiguous: ‘that two (of the) students will probably buy Jan’s house.’ dat twee studenten waarschijnlijk Jans huis zullen kopen. that two students probably Jan’s house will buy Specific only: ‘that two of the students will probably buy Jan’s house.’

The examples in (28), finally, show that definite pronouns must shift into the regular subject position (provided that they are not assigned emphatic accent; cf. the discussion of (25)). (28)

a.

b.

dat waarschijnlijk ZIJ/*ze zijn huis koopt. that probably she his house buys ‘that she will probably buy his house.’ dat ze waarschijnlijk zijn huis koopt.

For completeness’ sake, note that the unacceptability of (28a) with a weak pronoun shows that we cannot account for (26) and (27) by assuming that the sentential adverb can optionally be adjoined to TP, as was suggested by one of the reviewers of Broekhuis (2007a); this wrongly predicts (28a) to be acceptable with a weak pronoun (unless it could be shown that weak 48 Placing the expletive er in front of the adverbial phrase disambiguates example (27a): in that case, only the non-specific reading survives. I have little to say here about the expletive, although I want to note that its distribution is not restricted by properties of the subject alone (see, e.g., Bennis 1986, and Haeseryn et al. 1997: section 8.6.3.3, for extensive discussion). Further it must be stressed that the presence of er is not motivated by the EPP (after all, then we would also predict an expletive to appear in examples like (26a)), but rather signals that the clause does not contain any presupposed material.

Object shift and subject shift (A-movement) 221

pronouns move into a position higher than TP, for which Dutch provides no evidence, since, as we will see in section 4.2.4, object pronouns normally do not precede non-pronminal subjects). The discussion above shows that, as predicted by the subranking ALIGNFOCUS>> EPP(case) >> *MOVE, the movement behavior of objects and subjects is similar when they have the form of a definite, an indefinite or a pronominal NP. The same holds when they are of some other noun phrase type. I will give two more examples here (see Broekhuis et al. 2003 for a more complete overview). First consider the examples in (29). The object in (29a) has an existential interpretation: Jan is reading something that is probably a bestseller (or, alternatively, Jan is doing something, which is probably reading a bestseller). The object in (29a′), on the other hand, has a generic reading: it expresses that bestsellers are likely to be read by Jan. The same is perhaps even clearer in (29b&b′): (29b) expresses that Jan generally reads bestsellers, whereas (29b′) expresses that most bestsellers are read by Jan. The examples in (29) therefore show that generic NPs must undergo object shift. (29)

a. a′. b. b′.

dat Jan waarschijnlijk een bestseller leest. that Jan probably a bestseller reads dat Jan een bestseller waarschijnlijk leest. dat Jan meestal een bestseller leest. that Jan generally a bestseller reads dat Jan een bestseller meestal leest.

Let us provisionally assume that this restriction follows from the fact that generic noun phrases do not refer to entities and are thus not able to introduce new entities into the domain of discourse, so that they inherently are not part of the focus of the clause (alternatively, one may appeal to some version of Diesing’s 1992 Mapping Hypothesis; I leave this option open here). Since the constraint ranking in (20) does not explicitly refer to objects, we now expect that generic subjects must also shift. The examples in (30) show that this expectation is indeed borne out both for singular and for plural indefinite NPs: in the prime-less examples the non-shifted subject receives an existential interpretation, whereas in the primed examples the shifted subject is construed generically. (30)

a. a′.

dat that dat that

er meestal there generally een nijlpaard a hippopotamus

een nijlpaard in het water ligt. a hippopotamus in the water lies meestal in het water ligt. generally in the water lies

222 Chapter 4: Object shift and other movement types b. b′.

dat that dat that

er meestal nijlpaarden in het water liggen. there generally hippopotami in the water lie nijlpaarden meestal in het water liggen. hippopotami generally in the water lie

Object and subject also behave similarly when they are quantified. Consider the examples in (31a&b), which differ with respect to the relative scope of the adverb vaak ‘often’ and the universally quantified object alle boeken ‘all books’. Example (31a) expresses that it holds for each of the books individually that they are often taken along by Jan; the contention can be true if the number of books is ten and Jan never takes more than three books with him. Example (31b), on the other hand, expresses that John often takes along the full set of books; the contention is false if the number of all books is ten and Jan never takes more than three books with him.49 (31)

a. b.

dat Jan alle boeken vaak meeneemt. that Jan all books often takes-along dat Jan vaak alle boeken meeneemt.

(∀ > vaak) (vaak > ∀)

To a large extent, scope relations between XPs in the middle field can be read off of the linear order of the clause in Dutch. This could be accounted for by assuming the constraint SCOPE in (32a), which was already briefly mentioned in section 1.3.3.2, and which requires that relative scope relations correspond to hierarchical (thus linear) order. When we assume the partial ranking in (32b), we correctly predict object shift to be blocked under the intended reading in (31b). (32)

a. b.

SCOPE:

relative scope of quantifiers corresponds to the hierarchical order of their topmost A-positions. Dutch: SCOPE >> EPP(case) >> *MOVE

What is crucial here is that this ranking does not only block object shift in (31b), but also subject shift in (33b), as expected.

49 The interpretations given here are at least the highly preferred ones; one of the reviewers of Broekhuis (2007a) claims to also have an inverted scope reading for example (31a). For the speakers I have consulted (including myself) this is marginally possible at best and requires that the DP be contrastively focused, in which case the word order probably results from reconstructable A′-movement into SpecFocusP; cf. the remark below example (34).

Object shift and subject shift (A-movement) 223 (33)

a. b.

dat alle studenten vaak hier zijn. that all students often here are dat vaak alle studenten hier zijn.

(∀ > vaak) (vaak > ∀)

Note in passing that the examples in (31) and (33) show again that it is not the numeration that determines the candidate set; cf. the discussion in section 1.3.3. The a- and b-examples are based on the same numeration, so that MP wrongly predicts that the a-examples should be blocked by the b-examples as the less economical derivationsņin OT-terminology, it involves one additional violation of *MOVE. By assuming that the candidates in the candidate set must have the same meaning, however, the a- and b-examples are not part of the same candidate set, so that they cannot be compared, which correctly predicts that they are all possible word orders. To conclude the discussion of examples (31) and (33), it is important to note that postulation of the constraint SCOPE does not imply that relative scope can always be read off of linear order. For example, in languages where *MOVE outranks SCOPE (English might be such a language), the optimal candidate may violate SCOPE. This allows us to handle scope ambiguities without having to resort to Quantifier Raising.50 In fact it is not even true that the ranking in (32) implies that relative scope can always be read off from linear order: since the movement that gives rise to the desired order must satisfy the Last Resort Condition, we predict that scope ambiguities do also arise in Dutch when the universally quantified DP is the complement of, e.g., a predicative PP. The examples in (34), adapted from Den Besten & Broekhuis (1992), show that this prediction is indeed correct. Examples (34a) and (34b) are structurally identical, but differ in the preferred scope relations for extra-linguistic reasons: in (34a) the existential quantifier is preferably construed as taking wide scope (Peter is dipping the same finger in all the sauces), whereas in (34b) the universal quantifier is preferably construed as taking wide scope (all vases will end up with a rose in them). This suggests that the examples in (34) are actually ambiguous, due to the fact that PPs do not undergo A-movement. 50 This comes close to Williams’ (2003: section 9.1.1) claim that the relation between syntactic and semantic representations does not necessary involve a fully compositional translation, but can also be a matching translation. He suggests that syntactic and semantic representations are “two languages that are being matched up as definable independent of one another, according to laws of form that might differ”, with the result that “there might not be an isomorphic match” between the two representations. This implies that “if there is no isomorphic [semantic] structure, the matching relation might pick the “nearest” structure as the translation”.

224 Chapter 4: Object shift and other movement types (34)

a.

b.

dat Peter een vinger in elke saus dipte. (preferably ∃ > ∀) that Peter a finger in every sauce dipped ‘that Peter dipped a finger in every sauce.’ dat Peter een bloem in elke vaas zette. (preferably ∀ > ∃) that Peter a flower in every vase put ‘that Peter put a flower in every vase.’

In this way we derive in the domain of quantifier scope what Williams (2003) refers to as PƗnҚini’s Principle: there is only isomorphism between the semantic and the syntactic representation if syntax makes an isomorphic structure available; see also fn. 50. For completeness’ sake, note that the isomorphism holds between the relative scope of quantifiers and the hierarchical relation between their topmost A-positions in the syntactic representation, and not their topmost A′-positions. This is clear from the fact that it is also possible to say dat Peter in elke vaas een bloem zet and dat Peter in elke saus een vinger dipte without changing the preferred scope readings in (34). In these examples the preposed predicative PP, which is assigned contrastive focus accent, is arguably A′-moved into SpecFocusP. The fact that this movement does not affect the scope reading follows if derived A′-positions do not count for scope purposes, that is, if A′-movement is “reconstructable”. This subsection has argued that regular object shift and subject shift are subject to similar restrictions in Dutch, which shows that, as expected, these movements are sensitive to the same syntactic constraints. This means that we can continue with discussing the more problematic aspect of the analysis, namely the fact that the strong ranking of EPP(ij) does not seem to force the application of subject shift in Dutch. 4.2.2.

Subject shift and the constraint EPP(ij)

We have seen that the partial ranking EPP(ij) >> ALIGNFOCUS >> *MOVE predicts that the ij-features on V force short object shift (unless this is blocked by H-COMPL, as is the case with short object shift across the main verb into the local domain of the auxiliary in complex verb constructions in the Scandinavian languages, and across verbal particles in Swedish). Given that V-to-v does not apply in the Germanic OV-languages (unless forced by LEXICALLY FILL TOP F), this correctly predicts that the object normally precedes the verb in these languages. Now, when we assume that the functional head I also has unvalued ij-features, we expect that these will also force application of subject shift. Section 4.2.1 has shown, however, that this expectation is not borne out. This suggests that something special

Object shift and subject shift (A-movement) 225

is going on in the case of the subject that we have overlooked so far. If so, there are in principle two possibilities: the subject differs in some way from the object as a result of which it cannot be probed by the unvalued ijfeatures on I, or, alternatively, the unvalued ij-features on I differ in some way from those on V as a result of which subject shift need not apply. It is easy to show that the non-obligatoriness of subject shift is not due to some difference between the subject and the object, and must therefore be due to some difference in the unvalued ij-features on I and V. Let us begin with considering the examples in (35), which contain the dyadic unaccusative verb overkomen ‘to happen’, and in which the definite subject de ergste rampen ‘the most terrible disasters’ can either follow the indirect object or be placed into the “regular” subject position, SpecIP; cf., e.g., Lenerz (1977), Koster (1978), Den Besten (1985), and Broekhuis (1992). (35)

a.

b.

dat het meisjedat. waarschijnlijk de ergste rampennom. overkwamen. that the girl probably the worst disasters happened ‘that the most terrible disasters probably happened to the girl.’ dat de ergste rampennom. het meisjedat. overkwamen.

That the unvalued case feature on I does not force subject shift follows from the discussion in section 4.2.1: when the definite subject belongs to the focus of the clause, as in (35a), movement is blocked by ALIGNFOCUS; when it is part of the presupposition of the clause, as in (35b), movement must apply. The examples in (36) show that, as expected, non-specific indefinite subjects do not shift; (36b) is degraded (although perhaps marginally acceptable under a specific reading of the subject, which I have indicated by means of the double cross). (36)

a.

b.

dat het meisjedat. waarschijnlijk erge rampennom. overkwamen. that the girl probably terrible disasters happened ‘that terrible disasters probably happened to the girl.’ # dat erge rampennom het meisjedat. overkwamen.

The examples in (37) show that pronominal subjects must shift, which is due to the fact that they cannot be construed as part of the focus of the clause. (37)

a.

*dat het meisje zij overkwamen. that the girl they happened ‘that they happened to the girl.’ b. dat zij het meisje overkwamen.

226 Chapter 4: Object shift and other movement types

The important thing to note at this point is that the subject must precede the verb in clause-final position in all cases. Since the subject is an internal argument of the verb, it can only end up in this position when it has moved from its postverbal base position into the preverbal position. Since the light verb v associated with unaccusative verbs does not have an accusative case feature, this movement must have been triggered by the unvalued ijfeatures on V. This implies that the subject occupies a vP-internal position in (35a) and (36a). That this is indeed the case is also suggested by the fact that the subject must follow the sentential adverb waarschijnlijk; inverting the order gives rise to a highly marked result in (35a) and leads to unacceptability in (36a).51 This is shown in (38). (38)

a.

??

dat het meisjedat. de ergste rampennom. waarschijnlijk overkwamen. that the girl the worst disasters probably happened b. *dat het meisjedat. erge rampennom. waarschijnlijk overkwamen. that the girl terrible disasters probably happened

The fact that the ij-features on V force shift of the subject into preverbal position whereas the ij-features on I do not force subject shift suggests that the problem under discussion is not caused by some difference between subjects and objects, but by some difference between I and V. The next question therefore is: In what respect are V and I different? I will argue below that the difference lies in the fact that I and V have a different set of unvalued ij-features. This implies that the proposal so far has been based on too simple a view of ij-features; we may therefore hope that once we have a better understanding of these features, all will fall into place. There is a general tendency in the literature to look upon the ij-features as a semantic unit: ij-features are the gender, number and person features of noun phrases. And in syntax, too, the ij-features are normally treated as a complex object, a feature bundle. There are, however, good reasons not to adopt such a view of the ij-features. Note that ij-features of nouns fall into two categories: the features [number] and [person] are features with clear semantic import, whereas [gender] is rather a formal class marker, which 51

The same is noted for German in Sabel (2005:295-6), who attributes this observation to Webelhuth (1986), which I have not been able to consult. In section 3.1 of his paper, Sabel gives a nice review of several other arguments that show that the nominative argument occupies a VP-internal preverbal position in examples like (35a) and (36a). For completeness’ sake, note that the proposal in the main text does not imply that the nominative arguments must be preceded by the VP-adverbs: the object is placed in preverbal position by means of short object shift, which may (but need not) cross the VP-adverbs in Dutch/German (cf. the discussion in section 2.4.1.3).

Object shift and subject shift (A-movement) 227

may but need not have semantic import. I will assume that this distinction is reflected in the way these features are introduced into the computation: [gender] is clearly a lexical property of the noun, so that it must be introduced into the derivation by the noun itself; the features [number] and [person], on the other hand, are not inherent features of the noun but semantically relevant features that are introduced by the functional heads NUM and D, respectively. This distinction is, of course, reminiscent of Chomsky’s (1995a:231) distinction between intrinsic and optional features. The fact that we tend to think of ij-features as a feature bundle is not because they inherently constitute a unit of some sort, but because they are normally combined within a single DP. It is, however, not a logical necessity that the three always co-occur. They clearly do in (39a), where we are dealing with a definite noun phrase headed by a count noun, but it may also be the case that one or more of the functional heads (hence ij-features) is missing: the definite noun phrase in (39b) headed by a non-count noun may lack the number feature, the indefinite noun phrase headed by a count noun in (39c) may lack a person feature, and the indefinite noun phrase headed by a non-count noun in (39d) may have no functional head at all and, hence, have only a gender feature.52 a.

• Structure [D[person] [NUM[number] [N[gender]]]]

b.

[D[person] [N[gender]]]

c.

[NUM[number] [N[gender]]]

d.

[N[gender]]

(39)

• Example de vier/∅ mannen the four/∅ men de wijn the wine vier/∅ mannen four/∅ men wijn wine

If it is indeed true that the ij-features on noun phrases are introduced into the derivation by separate (functional) heads, there is no reason to assume that the unvalued ij-features that trigger movement of the noun phrase always make up a complete ij-feature bundle. Consequently, we 52

This proposal is, of course, entirely in line with a Bare Phrase Structure approach along the lines of Chomsky (1995b). Note that I do not claim that ij-features must be phonologically supported; features may also be inserted into the structure by abstract functional heads. The behavior of generic noun phrases (e.g., the fact that they may not occur in expletive constructions: *there is red wine healthy) suggests that they contain a [person] feature, that is, an empty D; cf. Longobardi (1994) and later work for further discussion.

228 Chapter 4: Object shift and other movement types

cannot exclude that V contains a different subset of the ij-features than I. That something like this may indeed be the case is suggested by the overt morphology that is used to express object and subject agreement. In the Romance languages, for instance, object agreement typically involves gender (and number), whereas subject agreement involves person (and number). This is illustrated in (40), taken from Burzio (1986): the object clitic la agrees in gender and number with the participle, whereas the subject Giovanni agrees with the finite verb in person and number. (40)

Giovanni la ha accusata. Giovanni her has3SG accusedSG.fem ‘Giovanni has accused her.’

In what follows I will therefore adopt the two statements in (41). When we may identify the agreement on the participle in (40) to that on predicative adjectives, the discussion in Baker (to appear: ch.1) suggests that (41a) is a universal property of natural language. Baker further claims that, although (41b) is a property of the Western European languages, it is subject to parameterization. Note in passing that Chomsky’s (2001) discussion of case assignment exploits claim (41a), but crucially rejects claim (41b): I is, what he calls, ij-complete. Since I have already rejected in section 3.1.2.4.4 the claim that case assignment and agreement are related, this is innocuous. (41)

a. b.

V has the unvalued ij-features [gender] and [number] I has the unvalued ij-features [person] and [number]

The solution to the problem posed by the asymmetry between short object shift and subject shift now suggests itself. Apparently, I was wrong in assuming a single constraint EPP(ij) for all cases of agreement: instead we must assume three separate constraints, one for each ij-feature, as in (42). Since EPP(number) will not play a role in the discussion below, I will ignore it in what follows, and concentrate on EPP(person) and EPP(gender). (42)

a. b. c.

EPP(person): [person] attracts its goal. EPP(number): [number] attracts its goal. EPP(gender): [gender] attracts its goal.

The desired distinction can now be made by replacing the ranking in (20) by the one in (43). The subranking EPP(gender) >> ALIGNFOCUS >> *MOVE expresses that the gender feature on V obligatorily triggers movement of the postverbal internal argument into preverbal position. As before, the subranking ALIGNFOCUS >> EPP(case) >> *MOVE expresses that the case

Object shift and subject shift (A-movement) 229

feature on v/I triggers movement of the object/subject provided that it is part of the presupposition of the clause. Since the subranking ALIGNFOCUS >> EPP(person) >> *MOVE also allows the subject to remain in its base position when it is part of the focus of the clause, we now correctly predict that subject shift is optional. Note that the precise ranking of EPP(case) and EPP(person) cannot be determined on the basis of the data under discussion; this is again expressed by placing the constraints between curly brackets. (43)

Germanic OV-languages: EPP(gender) >> ALIGNFOCUS >> {EPP(case), EPP(person)} >> *MOVE

This concludes our discussion of the claim that object shift and subject shift are triggered by similar features. This claim implies that object shift and subject shift are subject to the same constraints, so that we expect that these movements behave in similar ways. Our investigation of Dutch has shown that this expectation is borne out only partially. Section 4.2.1 has shown that regular object shift and subject shift are subject to the same information structural condition, namely that the shifted argument must be part of the presupposition of the clause. This section, on the other hand, has shown that short object shift and subject shift differ in a non-trivial manner: whereas the former is obligatory, the latter is not. This has led us to the conclusion that these movements are not triggered by the full ij-feature set, but by individual ij-features: bearing on the observation that object agreement involves agreement in gender (and number) but not in person, I proposed that short object shift is triggered by the person feature on V; bearing on the observation that subject-verb agreement involves agreement in person (and number) but not in gender, I proposed that subject shift is triggered by the person feature on I. As a result, the dichotomy between short object shift and subject shift can be accounted for by assuming that the constraints EPP(gender) and EPP(person) differ in their ranking relative to ALIGNFOCUS: by assuming that EPP(gender) outranks ALIGNFOCUS, we predict that short object shift is not sensitive to the information structure of the clause, whereas by assuming that ALIGNFOCUS outranks EPP(person) we predict that subject shift is sensitive to it. 4.2.3.

Subject shift and the constraint H-COMPL

The fact that the constraints entering the evaluation are universal of course implies that the EPP constraints in (42) also enter the evaluation in the Germanic VO-languages. The fact that the Germanic VO-languages have both obligatory short object shift and obligatory subject shift suggests that in these languages EPP(gender) and EPP(person) both outrank ALIGNFOCUS,

230 Chapter 4: Object shift and other movement types

so that this doesn’t seem to affect the analyses given so far. We will see in this subsection, however, that distinguishing EPP(gender) and EPP(person) may help us to account for certain differences between object shift and subject shift in these languages as well. First, recall from section 2.4.2.2 that Dutch examples like (44a) show that both the participle and the auxiliary are endowed with unvalued ij-features, and that the strong ranking of EPP(ij) therefore predicts that the object must precede the auxiliary verb. Despite the fact that the Germanic VO-languages also have a strong ranking of EPP(ij), section 3.2.3 has shown that the object cannot move into the local domain of the auxiliary verb in these languages, which is illustrated by (44b) for Icelandic. This was accounted for by assuming that EPP(ij) is outranked by H-COMPL in the Scandinavian languages: short object shift of the object into the local domain of the auxiliary crosses the main verb and therefore fatally violates H-COMPL. (44)

a.

b.

dat Jan waarschijnlijk heeft gelezen . that Jan probably this book has read ‘that Jan has probably read this book.’ Jón hefur ekki taux keypt . Jón has not the book bought ‘Jón has not bought the book.’

The problem that arises for the Germanic VO-languages is that the assumption that H-COMPL outranks EPP(ij) will also block subject shift when the subject is an internal argument of the verb, that is, in unaccusative and passive constructions. That this is clearly an undesirable result is illustrated by means of the Icelandic passive and unaccusative constructions in (45), taken from Jónsson (1996:132). (45)

a. b.

Tölvurnar voru keyptar ttölvurnar fyrir starfsfókia. the computers were bought for the staff Bækurnar höfðu horfið tbækurnar úr hillunni. the books had disappeared from the shelf

This problem can be straightforwardly solved by adopting the three EPP constraints in (42) and by assuming that the Germanic VO-languages have the subranking EPP(person) >> H-COMPL >> EPP(gender) >> *MOVE: the subranking EPP(person) >> H-COMPL >> *MOVE correctly allows subject shift across the main verb, while the subranking H-COMPL >> EPP(gender) >> *MOVE still disallows short object shift from doing this. Tableau 5 provides the evaluation of the examples in (45). For convenience, I have

Object shift and subject shift (A-movement) 231

left out the candidates in which the required verb movements did not apply; these will be excluded for the reasons discussed in section 4.1. (46)

a. b. c. d. e. f.

I+v+aux …ekki tv ... taux... Asp+V ... tV S I+v+aux …ekki tv ... taux... Asp+V ... S tV tS I+v+aux … ekki tv... S taux... Asp+V ... tS tV tS S I+v+aux … ekki tv... tS taux... Asp+V ... tS tV tS S I+v+aux … ekki tv... taux... Asp+V ... tS tV tS S I+v+aux … ekki tv... taux... Asp+V ... tV tS

Tableau 5: Subject shift in Icelandic passive/unaccusative constructions EPP

(46a) (46b) (46c) (46d) ) (46e) (46f)

(person) *! *! *!

H-COMPL

* ** ** **

EPP

EPP

(gender) ** *

(case) * * *

*MOVE (tO only) * **

*! *!*

It is important to note that the tableau shows that the subject must move successive cyclically through all intervening A-positions; more specifically, the unacceptability of the fifth candidate shows that the subject must undergo short object shift into the local domain of the auxiliary, which is excluded in other Scandinavian complex verb constructions (being an unaccusative verb, the auxiliary, of course, does not trigger regular object shift). This shows that some form of SHORTEST STEPS, which was proposed by Chomsky (1995a: ch.3), follows from the current system without further ado. We will return to this issue in section 4.3.1. Although this in fact concludes our argument, two remarks are still in order. First, the discussion of the examples in (45) presupposes that the movement of the subject starts out from the postverbal position, that is, the position in which indefinite subjects may also appear in existential constructions like (47a), taken from Jónsson (1996:131). However, one might object to this claim by pointing out that in examples like (47b), also taken from Jónsson (1996:202), the indefinite subject of the expletive construction appears in front of the main verb: since examples like these might be taken to be instances of an adjectival passive, we might also claim that the movement of the subject in examples like (45a) starts out from the preverbal position and hence does not violate H-COMPL.

232 Chapter 4: Object shift and other movement types (47)

a. b.

Það there Það there

höfðu verið keyptar have been bought voru Þrjár tölvur were three computers

tölvur fyrir starfsfókia. computers for the staff keyptar fyrir starfsfókia. bought for the staff

There is, however, reason to assume that example (47b) is not an adjectival passive but derived by moving the subject from postverbal position into its surface position. This movement is, however, not A- but A′-movement. In order to see this, consider the transitive examples in (48), taken from (Svenonius 2000:(13)). (48)

a. b.

Strákarnir höfðu hent miklu grjóti í bílana. the boys had thrown much rock in the cars Strákarnir höfðu miklu gróti hent í bílana. the boys had much rock thrown in the cars ‘The boys had thrown many rocks at the car.’

Example (48b) shows that movement of the object across the verb is also allowed. This option is, however, restricted to quantified and negative DPs. In the former case, the movement is optional (depending on the interpretation of the object), and in the latter case it is obligatory, as is shown by the examples in (49). (49)

a.

*Strákarnir höfðu hent engu grjóti í bílana. the boys had thrown no rock in the cars b. Strákarnir höfðu engu gróti hent í bílana. the boys had no rock thrown in the cars ‘The boys had thrown no rocks at the car.’

Since we have already seen in section 1.3.3.2 that negative phrases must be placed into SpecNegP, we may safely assume that the object movement in (49) is not triggered by the ij- or case features of the auxiliary, but rather by some unvalued semantic [+affect] feature on some higher functional head. In other words, the order in (49b) is the result of A′-movement and not of A-movement. Note that this correctly predicts that Neg-movement is also obligatory in the expletive construction in (50). (50)

a.

*Það there b. Það there

var was var was

keypt engin tölva. bought no computer engin tölva keypt. no computer bought

(Jónsson 1996:184/5)

Object shift and subject shift (A-movement) 233

This suggests that the placement of the quantified DPs in preverbal position in (47b) and (48b) is likewise the result of A′-movement triggered by some unvalued [+affect] feature which triggers movement of quantified/negative phrases (cf., e.g., Brody 1990a and É. Kiss 2002: ch.5&6, who show that Hungarian has a similar movement of quantified and negative phrases). Since the subject in (45) is a definite DP and definite DPs are not attracted by this unvalued [+affect] feature, we cannot take recourse to this feature in our account of the fact that subject shift may cross the main verb. Consequently our earlier conclusion that H-COMPL must be outranked by 53 EPP(person) still stands. The second remark concerns the impossibility of passivization of Swedish verb-particle constructions (cf., e.g., Bobaljik 2002 and Koeneman 2006); the examples in (51) show that a verbal particle does not only block object shift but also subject shift. This is a potential problem for the analysis proposed above, given that the subranking EPP(person) >> H-COMPL in principle allows for subject shift across the participle and the verbal particle. (51)

a.

Peter har inte kastat bort . Peter has not thrown the-carpet away b. Peter kastade inte bort . Peter threw it not away c. *Skräpet måste bli kastat ut. Scrap-the must be thrown out

Recall from section 3.2.4, however, that the ungrammaticality of object shift is due, not to some constraint violation, but to a violation of the inviolable Phase Impenetrability Condition (PIC): since the small clause constitutes a phase and movement of the small clause subject into the edge of the small clause is excluded, the only option is that object shift proceeds 53

English expletive constructions perhaps also support the proposal that EPP(ij) must be replaced by the three more specific constraints in (42). Chomsky (1995a: ch.3) has claimed that expletives like there are D-elements adorned with a [person] feature, and that merging there with I satisfies the latter’s “EPP-property”. However, if subject shift is triggered by a strong ranking of EPP(ij), merging of the expletive does not suffice to satisfy this constraint: at least the unvalued number feature and, on Chomsky’s assumption that I is ij-complete, also the unvalued gender feature on I would still cause a violation of this constraint. On the assumption that subject shift is triggered by the strong ranking of EPP(person), on the other hand, merging of the expletive would suffice to satisfy EPP(person). Recall that I have argued in section 1.3.3.1 that the expletive construction is actually due to locative inversion, but since I will argue in Chapter 5 that predicate movement is also triggered by unvalued ij-features, the argument given here might be retained in some form or other.

234 Chapter 4: Object shift and other movement types

in one fell swoop, but this is excluded by the PIC. Given that subject shift into SpecIP in the passive construction in (51c) is preceded by short “object” shift, passivization is also excluded by the PIC. For completeness’ sake, note that Swedish does allow for passivization when the particle precedes the verb, as in (52a). I will follow Koeneman (2006) in assuming that prt+V combinations are actually compound verbs, which is clear from the fact that these may undergo V-to-I as a whole, as illustrated by Koeneman’s example in (52b). Consequently (52a) does not contain a small clause, so passivization is allowed by the PIC. (52)

a. b.

4.2.4.

Skräpet måste bli Scrap-the must be Jag avbryter inte I offbreak not

utkasted. out-thrown samtalet. conversation-the

Subject shift and the constraint RELMIN

Section 3.2.2 has discussed the fact that the constraint RELMIN blocks output representations in which a goal G2 has moved into the local domain of a probe P across another potential goal G1 of P, as in (53b). We have further seen that this problem can be solved by applying either the blocking strategy in (53a) or the push-up strategy in (53c). (53)

a. [P [… G1 … [… G2 …]]] b. *[PB G2 P1 [… G1 … [ … tG2 … ]]] c. [G1 P2 [PB G2 P1 [… tG1 … [… tG2 … ]]]]

In our discussion of regular object shift we have seen that the blocking strategy is preferred by the constraint ALIGNFOCUS: when an indirect object is focused it will block object shift of a direct object that is part of the presupposition of the clause. The push-up strategy, on the other hand, is favored by the constraint D-PRONOUN. Now, when D-PRONOUN outranks ALIGNFOCUS, as in Icelandic and Dutch, a pronominal direct object will push-up the indirect object into its shifted position, irrespective of whether it belongs to the focus or to the presupposition of the clause. When ALIGNFOCUS outranks D-PRONOUN, as in Danish and Swedish, the indirect object will only be pushed-up when it is part of the presupposition of the clause. This section briefly discusses the consequences of assuming RELMIN for subject shift. Given that regular object shift places the object into a position preceding the base position of the subject, this gives rise to a

Object shift and subject shift (A-movement) 235

violation of RELMIN when the subject remains in its base position. Since we have seen that the subject obligatorily moves into SpecIP in the Scandinavian languages, this situation will not arise there. We have seen in section 4.2.1, however, that the subject in Dutch may remain in its base position when it is part of the focus of the clause, which raises the question how the violations of RELMIN are solved in this language. Section 3.2.2.1 and 4.2.2 have established subrankings (54a) and (54b) for Dutch; when we combine the two we get the ranking in (54c). (54)

a. b. c.

{D-PRONOUN, RELMIN} >> ALIGNFOCUS >> EPP(case) >> *MOVE EPP(gender) >> ALIGNFOCUS >> {EPP(case), EPP(person)} >> *MOVE {D-PRONOUN, RELMIN, EPP(gender)} >> ALIGNFOCUS >> {EPP(case), EPP(person)} >> *MOVE

As we have seen in section 3.2.2.1, the subranking ALIGNFOCUS >> EPP(case) correctly predicts that an indirect object that is part of the focus of the clause blocks regular object shift of a non-pronominal direct object. From this it follows that we also predict that a subject that is part of the focus of the clause will block object shift of a non-pronominal object that is part of the presupposition of the clause. The best way to test this is by considering the interpretation of indefinite subjects. Since the examples in (23) and (27) above have shown that non-specific indefinite noun phrases are always part of the focus of the clause, we get the following prediction: when the subject appears to the right of the sentential adverb waarschijnlijk in (55), it can have either a non-specific or a specific interpretation, whereas it can only have a specific interpretation when it occurs to the left of waarschijnlijk. The examples in (55) show that this prediction is indeed correct. Observe that, as predicted, the interpretation of the subject in (55b&b′) does not depend on the placement of the object, which may either follow or precede the sentential adverb depending on whether it is part of the focus or the presupposition of the clause. (55)

a.

b.

b′.

dat waarschijnlijk twee jongens dit boek gelezen hebben. that probably two (of the) boys this book read have Ambiguous: ‘that two (of the) boys have probably read this book.’ dat twee jongens waarschijnlijk dit boek gelezen hebben. that two of the boy probably this book read have Specific only: ‘that two of the boys have probably read this book.’ dat twee jongens dit boek waarschijnlijk gelezen hebben. that two of the boy this book probably read have Specific only: ‘that two of the boys have probably read this book.’

236 Chapter 4: Object shift and other movement types

Section 3.2.2.1 has also shown that the subranking {D-PRONOUN, RELMIN} >> ALIGNFOCUS >> EPP(case) correctly predicts that a definite pronominal object will push-up the indirect object, irrespective of whether the latter is part of the presupposition or the focus of the clause. The expectation is now that a pronominal object will also push-up the subject. Although judgments are somewhat subtle, the examples in (56) suggest that this expectation is indeed borne out: (56a) shows that placement of an indefinite subject in the position following the sentential adverb waarschijnlijk indeed gives rise to a marked result, and (56b) shows that when the subject is placed to the left of the adverb, it can be construed either as specific or non-specific. (56)

a.

b′.

??

dat waarschijnlijk twee jongens het gelezen hebben. that probably two (of the) boys it read have ‘that two (of the) boys have probably read this book.’ dat twee jongens het waarschijnlijk gelezen hebben. that two of the boy it probably read have Ambiguous: ‘that two (of the) boys have probably read this book.’

Nevertheless, the fact that (56a) is possible might pose a problem for our proposal. Note that the examples in (57) show that we find more or less the same pattern when the subject is definite. (57)

a.

b.

??

dat waarschijnlijk de jongens het that probably the boys it ‘that the boys probably have read it.’ dat de jongens het waarschijnlijk that the boys it probably ‘that the boys probably have read it.’

gelezen hebben. read have gelezen hebben. read have

I believe that the relative acceptability of (56a) and (57a) is due to the fact that it is not always easy to distinguish sentence and contrastive focus accent. As is illustrated in (58), contrastively focused and negative subjects preferably follow sentential adverbs like waarschijnlijk ‘probably’, which results in blocking regular object shift of the object pronoun. I therefore believe that (56a) and (57a) are only possible when the subject twee/de jongens ‘two/the boys’ is contrastively focused. (58)

a.

dat waarschijnlijk zelfs JAN het gelezen heeft. that probably even Jan it read have ‘that probably even Jan has read it.’ a′. ??dat zelfs JAN het waarschijnlijk gelezen heeft

Object shift and subject shift (A-movement) 237 b.

dat waarschijnlijk niemand het gelezen heeft. that probably no one it read has ‘that probably no one has read it.’ b′. ??dat niemand het waarschijnlijk gelezen heeft.

German is like Dutch in that it allows the subject to remain in its base position when it is part of the focus of the clause. Section 3.2.2.1 has shown, however, that German differs from Dutch in that it does allow regular object shift across the indirect object. Consequently, we predict that regular object shift across the subject is allowed as well. The (slightly adapted) examples in (59) from Vikner (1994:511) show that this prediction is indeed correct. (59)

a.

b.

dass diesen Mann (der) Peter nie früher gesehen that this manacc thenom Peter never before seen ‘that Peter has never seen this man before.’ Warum hat es (der) Jan gekauft? why has it thenom Jan bought ‘Why did Jan buy it?’

hat. has

I want to conclude this section with a brief note on dyadic unaccusative (and similar passive) examples like (35) repeated here as (60). At first sight, the discussion above seems to imply that subject shift across the indirect object should be blocked, so that example (60b) is wrongly predicted to be ungrammatical. (60)

a.

b.

dat het meisjedat. waarschijnlijk de ergste rampennom. overkwamen. that the girl probably the worst disasters happened ‘that the most terrible disasters happened to the girl.’ dat de ergste rampennom het meisjedat. overkwamen.

This problem can be solved by following the conclusion from section 3.1.2.4 that intervention effects only arise when the shift occurs across an active goal. Since the subject is still active (= has an unvalued case feature) when regular object shift applies, object shift across the subject results in a violation of RELMIN, whereas subject shift in (60) applies after the indirect object has been inactivated by the valuing of its case feature, and can therefore apply without violating RELMIN. 4.2.5.

Conclusion

Section 4.2.1 has shown that the subranking in (61a) correctly predicts that in the OV-languages case-driven movement of the object and the subject

238 Chapter 4: Object shift and other movement types

are both sensitive to the information structure of the clause: it is possible only when the object/subject is part of the presupposition of the clause. Section 4.2.2, on the other hand, showed that the subranking in (61b) incorrectly predicts that agreement-driven movement is not only obligatory for objects but also for the subject. This problem was solved by splitting the constraint EPP(ij) into three separate constraints EPP(person), EPP(number) and EPP(gender). By assuming that the ij-feature sets on V and I differ in that V has an unvalued gender feature, whereas I has an unvalued person feature, we could make the desired distinction by assuming the ranking in (61c): this ranking predicts that short object shift is insensitive to the information structure of the clause, whereas subject shift is sensitive to it. (61)

a. b. c.

ALIGNFOCUS>> EPP(case) >> *MOVE EPP(ij) >> ALIGNFOCUS >> *MOVE EPP(gender) >> ALIGNFOCUS >> {EPP(case), EPP(person)} >> *MOVE

Section 4.2.3 continued to show that splitting up EPP(ij) also enables us to solve the problem that in the Scandinavian languages regular object shift is blocked by the main verb, whereas subject shift is not. Section 4.2.4, finally, has shown that RELMIN does not only affect regular object shift but also subject shift, and that it is correctly predicted that, in Dutch, subjects that are part of the focus of the clause block regular object shift of nonpronominal objects, but are forced to undergo subject shift when the object is a pronoun. It is also correctly predicted that German, due to the low ranking of RELMIN in this language, lacks these relativized minimality effects. 4.3. Object shift and A′-movement Although the earlier literature on relativized minimality effects gives rise to the expectation that there will be no interaction between object shift and A′-movement, sections 4.3.1 and 4.3.2 will show that there are at least two cases where such interaction does seem to occur: (i) at least in Dutch and Icelandic, wh-movement of the direct object forces regular object shift of the indirect object; (ii) in the Scandinavian languages, leftward movement of a quantified/negative direct object may likewise force the indirect object to shift. This section will argue that these facts only seemingly involve interaction of A- and A′-movement, and will reanalyze them in a way that fully accords with the intuition underlying Rizzi’s (1990) Relativized Minimality. We will see that the new analysis will also enable us to account for a number of yet unsolved problems that we came across in our earlier discussion of regular object shift in Chapter 3. Section 4.3.3, finally, will

Object shift and A′-movement 239

take up a final remaining problem concerning A′-movement, which we also had to put aside in Chapter 3, namely that topicalization of the main verb (or, in Swedish, the particle) may feed regular object shift. 4.3.1.

Wh-movement (“shortest steps”)

Haegeman (1991) and Den Dikken (1995a) have observed that in West Flemish and Dutch wh-movement of a direct object across a dative object has a degrading effect when the latter follows a sentential adverb like waarschijnlijk ‘probably’, that is, when it has not undergone regular object shift (although the examples are more or less acceptable when the indirect object is contrastively focused). This degrading effect is, however, completely absent when the dative object is shifted into the position in front of the sentential adverb. This contrast is illustrated by the examples in (62), which are taken from Den Dikken (1995a:198). (62)

a.

*?

Wat zal Jan waarschijnlijk Marie tDO geven? (Dutch) what will Jan probably Marie give a′. Wat zal Jan Marie waarschijnlijk tIO tDO geven? b. *?Wa goa Valère vandoage Marie tDO geven? (West Flemish) what goes Valère today Marie give b′. Wa goa Valère Marie vandoage tIO tDO geven?

Although I have not found similar data from the Scandinavian languages in the literature, Halldór Sigurðsson (p.c.) informs me that the judgments in Icelandic go in the same direction. According to him, neutral questions require regular object shift of the indirect object: example (63b) is therefore the neutral way of asking what María did not give to Ólof. When the indirect object Ólof has not shifted, the prominent reading is the one in which it is contrastively focused, which is clear from the paraphrase of example (63a) provided by Sigurðsson: “What was it that Mary did not give to Olaf (but to somebody else)”. (63)

a. b.

#

Hvað gaf María ekki Ólof tDO? whatacc. gave Maríanom. not Olofdat. Hvað gaf María Ólof ekki tIO tDO?

The data in (62) and (63) are not expected under the assumptions adopted so far. For example, the two constraints inducing intervention effects introduced earlier are related to head- and A-movement only: H-COMPL forbids object shift across the main verb and RELMIN forbids object shift to cross another A-position when this position is the head of an A-chain in the

240 Chapter 4: Object shift and other movement types

output representation. The seeming fact illustrated in (62) and (63) that A′-movement can sometimes be blocked by an element in an A-position could, of course, be accounted for by introducing a new constraint that blocks A′-movement of an argument across another argument in a VPinternal position, but since a constraint of this sort has no precedent in the earlier stages of the theory, this has a clear ad hoc flavor and should therefore be avoided if possible. A more promising alternative is to adopt a “shortest steps” approach to the problem by assuming that A′-movement of arguments cannot skip A-positions that could in principle be used as intermediate landing sites, which implies that in examples like (62) and (63) wh-movement of the direct object is preceded by regular object shift. This gives rise to the abstract representations in (64). (64)

a. b.

DOWh Vfin subject t′DO Adverb IO tDO DOWh Vfin subject IO t′DO Adverb tIO tDO

*?

When we further adopt the standard assumption that the head of an A-chain is also visible when it has undergone subsequent A′-movement, the primeless examples in (62) and example (63a) are correctly ruled out by RELMIN: since the wh-trace t′DO, which is the head of the A-chain of the direct object, precedes the indirect object in the output representation in (64a), RELMIN is violated. The primed examples in (62) and example (63), on the other hand, are correctly ruled in: since the wh-trace t′DO in the representation in (64b) is preceded by the head of the A-chain created by regular object shift of the indirect object, RELMIN is satisfied. From this account of the Dutch and Icelandic data, it follows that the German counterparts of the prime-less examples in (62) are acceptable: due to the low ranking of RELMIN, German does not exhibit minimality effects, so the output representation in (64a) is allowed in this language. That this prediction is correct is illustrated by the examples in (65), again taken from Den Dikken (1995a:198). (65)

a. b.

Was wird der Hans wahrscheinlich der Maria geben? what will the Hans probably the Maria give Was wird der Hans der Maria wahrscheinlich geben?

The “shortest steps” approach to the examples in (62) and (63) has various predecessors in the earlier stages of the theory. For example, Chomsky (1995a: ch.3) adopted Shortest Steps (alongside Fewest Steps) as a global condition, and later attributed it to the operation ATTRACT/MOVE by assuming the condition in (66); cf. Chomsky (1995a: ch.4, (304)).

Object shift and A′-movement 241 (66)

Only the head of a chain CH enters into the operation ATTRACT/MOVE (trace is immobile).

Furthermore, there is independent empirical evidence in favor of this approach. Consider the French examples in (67), which show that participial agreement does not occur with a direct object in post-verbal position, but does show up when the object is wh-moved or placed in a clitic position (cf. for example Kayne 1985;1989, Burzio 1986, and Belletti 2006). This may be accounted for when we assume that this agreement is the morphological reflex of the fact that A′-movement (wh- and clitic movement) forces the object to move through certain A-positions that remain empty when A′-movement does not apply (cf. also the discussion of Arabic subject-verb agreement in Chomsky 1995a: ch.3). (67)

a. b. c.

Paul a repeint/*repeintes Paul has repainted Paul les a repeintes. Paul them has repainted Je sais combien de tables I know how many tables

les tables. the tables

il a he has

(French)

repeintes. repainted

Although examples like (67) unequivocally support the “shortest steps” approach, it is not a priori clear whether the “shortest steps” effects should be attributed to the generator or to the evaluator. Let us start by discussing the former option, which would clearly be favored by the current minimalist approaches. These approaches may try to derive the fact that A′-movement of the object triggers past participle agreement by appealing to some version of phase theory, more specifically, Chomsky’s (2001) inviolable Phase Impenetrability Condition (PIC) that requires that A′movement of the object proceed through the edge of v*P, and must therefore be preceded by regular object shift. There are, however, two problems with this proposal. The first problem is theory-internal; although the PIC indeed requires that the object be moved through the specifier of the head containing the unvalued case feature, specv*P, it does not force movement of the object through the specifier of the head containing the unvalued ij-features, SpecVP. This proposal thus has to stipulate that Amovement cannot skip an intervening accessible A-position, which is essentially what (66) does, but in effect this maintains Shortest Steps as a primitive notion. The second problem is that A′-movement of objects in fact does not obligatory trigger participle agreement in French (cf. Kayne 1985;1989). Rather, the question whether participle agreement shows up

242 Chapter 4: Object shift and other movement types

depends on semantic factors; agreement requires that the wh-phrase be D-linked (the answer must select entities from some presupposed set; see Belletti, 2006: section 4.3, and the references cited there for discussion). The optionality of participle agreement in French and the way it is conditioned suggests that at least these “shortest steps” effects should be attributed to the evaluator. Assuming this is entirely consistent with our earlier discussion below Tableau 5 on p.231, where it was shown that the constraint EPP(ij) requires that subject shift in Icelandic passive and unaccusative constructions satisfies “shortest steps” by first undergoing “short object shift”. If so, we may account for the optional agreement in French by appealing to the interaction of EPP(ij) and other constraints of a more semantic nature. Since developing a full-fledged analysis of French participle agreement is clearly beyond the scope of this study, I will not work out the suggested approach in detail here but instead leave this to researchers that are better equipped for this task. Note that the discussion above does not imply that the PIC is superfluous in accounting for all “shortest steps” effects; it only shows that the PIC cannot immediately be held responsible for the fact that A′-movement of objects must proceed through SpecVP, but leaves open the possibility that the PIC is instrumental in accounting for the fact that this movement must proceed through Specv*P. If so, we must conclude that the generator and the evaluator actually conspire in deriving the “shortest steps” requirement: the PIC forces A′-movement of object to be preceded by regular object shift, whereas EPP(ij) forces the application of short object shift in that case (provided, at least, that it is not outranked by some other constraint that disfavors this movement; cf. the discussion of the examples in (70) below). This stresses once more the underlying intuition of this study that a descriptively and explanatorily adequate theory must appeal to both the generator and the evaluator. Assuming that the suggested approach to deriving “shortest steps” effects is essentially correct, I will conclude this section with discussing two problems that we have put aside in our earlier discussion in Chapter 3. The first problem was mentioned in section 3.2.2.4 and involves examples such as (68). (68)

a. b.

VemIO gav duS denDO inte tIO tDO? (Swedish) who gave you it not MariaIO fortalte duS detDO forhåbentlig ikke tIO tDO. (Danish) Maria told you it hopefully not ‘I hope you did not tell MARIA.’

Object shift and A′-movement 243

Holmberg (1999) and Vikner (2006) claim that examples like (68) show that object shift of the direct object is possible across an A′-trace left by wh-movement or topicalization of the indirect object. Consequently, if traces of A′-movement are visible, as was assumed above, we wrongly predict these examples to be excluded by RELMIN. However, this problem does not arise in the “shortest steps” approach: the structures in (68) are excluded as movement in one fell swoop into SpecCP crosses a potential landing site of the indirect object. More specifically, the PIC requires the indirect object to undergo regular object shift before it can be A′-moved into SpecCP. This results in the configurations in (69), in which regular object shift of the direct object does not cross the head of the A-chain formed by object shift of the indirect object; RELMIN is therefore not violated, so the output representations are correctly predicted to be grammatical. (69)

a. b.

VemIO gav duS t′IO denDO inte tIO tDO MariaIO fortalte duS t′IO detDO forhåbentlig ikke tIO tDO

The “shortest steps” approach potentially runs afoul of the second problem, which we had put aside in section 3.1.2.4, but we will see that it actually follows from our earlier conclusion that (the lack of) certain “shortest steps” effects follow from the interaction of EPP(ij) with the interface constraints. First consider the Icelandic examples in (70), and recall that these examples show that a quirky subject blocks agreement of the finite verb and the nominative argument when two conditions are met: (i) the contrast between (70a) and (70b) shows that the quirky subject and the nominative argument must not be co-arguments, and (ii) the contrast between (70b) and (70c) shows that the quirky subject must intervene between the finite verb and the nominative argument; when the intervener is the A-movement trace of a quirky subject, agreement is allowed. (70)

a.

b.

c.

það líkuðu einhverjum þessir sokkar. there likedpl somebodydative these socksnom/pl ‘Somebody liked these socks.’ það finnst/*finnast einhverjum stúdent tölvurnar ljótar. some studentdative the computersnom/pl ugly there findsg/findpl ‘Some student considers the computers ugly.’ Mér finnst/finnast tölvurnar ljótar. medative findsg/findpl the computersnom/pl ugly ‘I consider the computers ugly.’

244 Chapter 4: Object shift and other movement types

Now consider example (71). Holmberg & Hróarsdóttir (2004:655) argued on the basis of this example that agreement is still disallowed when the intervening quirky subject is a wh-trace. (71)

Hvaða stúdent veist þú ... which studentdative know you ... ... [t′wh að finnst/*finnast twh tölvurnar ljótar]? ... that findsg/findpl the computers ugly ‘Which student do you know considers the computers ugly?’

At first sight this seems at odds with our “shortest steps” approach. Since wh-movement of the quirky subject crosses a potential landing site, viz. SpecIP, the “shortest steps” approach at first sight predicts that whmovement cannot take place in one fell swoop: the quirky subject is expected to first move into SpecIP, as in (72), so that the trace of the indirect object in between the verb and the nominative argument does not head an A-chain and therefore does not count as an intervener. This, in turn, wrongly predicts that agreement between the finite verb and the nominative tölvurnar ‘the computers’ should be possible. (72)

Hvaða student veist þú [t′wh að [IP twh finnst/*finnast tdative tölvurnar ljótar]]?

It must be noted, however, that in this case the “shortest steps” derivation is not forced by the PIC, as movement of the dative from the position indicated by tdative into SpecCP does not cross a phase boundary given that tdative can be assumed to occupy its shifted position at the edge of v*P. Movement of the quirky subject into the subject position of the embedded clause should therefore be due to the effect of EPP(ij); cf. section 4.2. This, in turn, implies that movement of the dative argument into the subject position can in principle be blocked if there is some constraint that disfavors this movement and outranks EPP(ij). Is there reason to assume that there is such a constraint? I believe there is. In order to see this, it is important to note that the derivation in (72) leads to a COMP-trace configuration: the wh-trace twh immediately follows the complementizer að. The problem posed by (71) can now be solved by assuming that there is some constraint, which I will not try to identify, that disfavors this configuration.54 When this constraint outranks EPP(ij), move54 Our conclusion would of course remain the same if we assume that the COMPtrace configuration is excluded by some inviolable principle like the ECP. Dekkers (1999) has, however, convincingly argued on empirical grounds that COMP-trace effects can be better accounted for in an optimality-theoretic manner.

Object shift and A′-movement 245

ment of the dative phrase Hvaða student into the subject position is blocked and wh-movement must apply in one fell swoop from its position intervening between the verb and the nominative into SpecCP of the embedded clause, as in (73). (73)

Hvaða student veist þú [t′wh að finnst/*finnast twh tölvurnar ljótar]?

This leads to the conclusion that Holmberg & Hróarsdóttir were actually right in claiming that in example (71) the trace intervening between the finite verb and the nominative argument is a wh-trace; since this also implies that this trace is the head of an A-chain, it will count as an intervener so that we correctly predict agreement to be blocked. This section has discussed a number of unexpected and sometimes even paradoxal intervention effects. I solved these problems by showing that in many but not all cases derivations obey the “shortest steps” requirement, first proposed in Chomsky (1995a: ch.3), according to which A′-movement of DP-arguments must be preceded by the available A-movement steps. It was further argued that the generator and the evaluator conspire in deriving this “shortest steps” requirement. Firstly, the inviolable Phase Impenetrability Condition (PIC) forces A′-movement of objects to be preceded by regular object shift. This accounts for the seeming fact illustrated in (62) and (63) that A′-movement of objects can be blocked by an element in an A-position by claiming that the intervening element actually blocks the regular object shift that must precede the A′-movement. It also accounts for the seeming fact illustrated in (68) that regular object movement is possible across an A′-movement trace by showing that this trace is actually an A-movement trace, which is not expected to block A-movement. Secondly, the constraint EPP(ij) favors “shortest steps” by requiring that A′-movement of objects and subjects be preceded by the A-movements licensed by the unvalued ij-features on V and I, that is, short object shift and subject shift. In this case, however, the “shortest steps” requirement can be overruled by a more highly ranked interface constraint disfavoring these A-movements. This makes it possible to maintain the claim that the trace intervening between the verb and the nominative in (71) is a wh-trace by assuming that that subject shift of the quirky subject is blocked by some constraint that disfavors the COMP-trace configuration that would result from the subsequent wh-movement. As a result of this we correctly predict the attested intervention effect.

246 Chapter 4: Object shift and other movement types

The present proposal therefore enables us to eliminate the contradictory statement that the A′-movement trace invokes an intervention effect in (71) but not in (68). Instead we can simply maintain clause (40c) of the definition of closeness in Chapter 3, according to which the intervention effect occurs only when the intervening element is active (=heads an Achain and contains an unvalued feature). 4.3.2.

Quantifier movement and Neg-movement

Section 3.2.3 has extensively discussed the fact that regular object shift cannot cross the main verb in the Scandinavian languages. Nevertheless, section 4.2.3 has shown that there are certain cases in which an object may precede the main verb. Some examples from Icelandic are given in (74). (74)

• Icelandic QP-movement and Neg-movement a. Jón hefur lesið . (QP-mvt) Jón has many books read b. Jón hefur lesið . (Neg-mvt) Jón has no book read

The leftward movement of the object across the main verb is semantically restricted, and is only possible with quantified or negative noun phrases. When the object is quantified the leftward movement is optional, albeit that the two surface orders result in different interpretations (Christensen 2005). When the object is a negative noun phrase the movement is obligatory. Henceforth I will refer to the two movements as QP-movement and Negmovement, respectively; in the discussion below I will mainly make use of examples involving Neg-movement. The Icelandic examples in (75), taken from Jónsson (1996:23/34), show that A′-movement may freely apply across the main verb. This means that the acceptability of the examples in (74) follows when we assume that these are not cases of object shift, but semantically motivated A′-movement. (75)

a.

b.

Hvaða bók ætli Pétur hafi lesið? which book wonder Pétur has read ‘Which book has Peter read?’ Maríu hef ég aldrei hitt. Maríu have I never met

That Neg/QP-movement in examples like (74) is different from regular object shift and should be considered instances of A′-movement has been extensively argued by Svenonius (2000:263-6). For instance, the examples

Object shift and A′-movement 247

in (76) clearly show that, contrary to regular object shift, Neg/QPmovement is not triggered by the case feature on v: the case of the shifted DP in (76b) is assigned by the preposition, and the shifted PP in (76c) does not require case at all. Other indications that we are dealing with A′movement are that Neg/QP-movement licenses parasitic gaps, causes weak crossover violations and does not extend the binding possibilities; I refer to Svenonius (2000) for an extensive discussion of this. (76)

a.

*Jón hefur tolað við engan. Jón has spoken with no-one b. Jón hefur engani tolað við ti. c. ?Jón hefur við engani tolað ti.

The fact that the case feature on the light verb v does not trigger Neg/QPmovement is further supported by the fact that the landing site of this movement is actually too low in the structure for that. First consider the Danish examples in (77), taken from Christensen (2005). Example (77a) shows that the landing site of Neg-movement is higher than the base position of the auxiliary verb. However, the fact illustrated in (77b&b′) that the negative object must follow the floated quantifier alle ‘all’ associated with the subject suggests that the landing site is internal to the v*P headed by the auxiliary.55 (77)

a.

at jeg ingen bøger havde læst. that I no books have read b. *De har ingenting alle fået. they have nothing all gotten b′. Har de alle ingenting fået? have they all nothing gotten ‘Did they all get nothing?’

(Danish)

The order restriction in (77b&b′) shows that Neg/QP-movement must be distinguished from regular object shift since that the latter must cross floating quantifiers associated with the subject. This is illustrated by means of the Swedish example in (78), taken from Holmberg (1986:165); the nonshifted pronoun follows, whereas the shifted pronoun precedes the floating quantifier (as well as negation). 55

The discussion of (77b&b′) in Christensen (2005:111) confuses me somewhat, since he suggests that the preverbal placement of the negative object is actually not the result of Neg-movement. How this order could arise otherwise is unclear to me. The main point at this moment is, however, that the landing site is internal to the v*P headed by the auxiliary, to the right of the base-position of the subject, and this seems consistent with Christensen’s data.

248 Chapter 4: Object shift and other movement types (78)

Varför läste studenterna inte why read the students it not ‘Why didn’t all the students read it?’

alla ? all

(Swedish)

Recall that section 1.3.2.1 has argued that A′-movement may be necessary for reasons of Full Interpretation and is therefore forced by the need to arrive at a converging derivation. This implies that there is no need for EPP-constraints that favor this type of movement, and hence that it is a universal property of language that H-COMPL does not block A′-movement. If this assumption is indeed on the right track, we expect that Neg/QPmovement will also differ from regular object shift in that it cannot be blocked by RELMIN. This predicts that Neg/QP-movement of the direct object may not only cross the main verb but also the indirect object. The contrast between the regular object shift examples in (79) and the Negmovement examples in (80) shows that this prediction is indeed correct. (79)

(80)

• Icelandic object shift in double object constructions a. Pétur sýndi Maríu bókina . Pétur showed often Maríu the book b. *Pétur sýndi bókinai oft Maríu ti. Pétur showed book-the often Maríu

(

IO-DO) (*DO-IO)

• Icelandic Neg-movement in double object constructions (complex tense) a. *Jón hefur sagt Sveini ekkert. (*IO-DO) Jón has said Sveinn nothing ( DO-IO) b. Jón hefur ekkerti sagt Sveini ti.

However, Neg/QP-movement seems to behave differently in simple tense constructions: the examples in (81) show that, contrary to the case in (80), Neg-movement of the direct object across the indirect object is excluded (cf., e.g., Jónsson 1996:93-4 and Svenonius 2000:273ff.). (81)

• Icelandic Neg-movement in double object constructions (simple tense) a. Jón sagði Sveini ekkert. ( IO-DO) Jón said Sveinn nothing b. *Jón sagði ekkert Sveini. (*DO-IO)

This leads to the problem in (82), which I will henceforth refer to as the Neg-movement puzzle and which was extensively discussed in different terms in several contributions to the thematic issue of Theoretical linguistics 31 on object shift in reply to Fox & Pesetsky’s (2005) article on Cyclic Linearization.

Object shift and A′-movement 249 (82)

• The Neg-movement puzzle: a. Neg-movement of a negative phrase obligatorily applies (unlike object shift) in order to satisfy Full Interpretation. b. Neg-movement of the direct object must cross the indirect object if it crosses the main verb, e.g., in complex tense construction (unlike object shift). c. Neg-movement of the direct object cannot cross the indirect object when it does not cross the main verb as well, e.g., in Icelandic simple tense constructions (same as object shift).

The answer to the question of why (82b) holds seems relatively straightforward: object shift of the indirect object across the main verb is blocked by H-COMPL, and consequently (82a) forces Neg-movement of the direct object to cross both the main verb and the indirect object. The remainder of this subsection will therefore focus on the question: Why does (82c) hold? Svenonius (2000: section 1.5) has claimed that Neg/QP-movement in simple tense constructions is not A′- but A-movement: more specifically, he suggests that in simple tense constructions these movements target the same position in which case is assigned. He claims that this accounts (among other things) for the fact that the examples in (83b&b′) are degraded since the moved elements are not in need of case checking. (83)

a.

Ég I b. ?Ég I b′. *?Ég I

hef engan talað við. have no one spoken with talaði engan við. spoke nobody with talaði við engan. spoke with nobody

If Svenonius’ claim that Neg/QP-movement in simple tense constructions is A-movement is correct, the fact that it cannot cross the indirect object follows immediately, as this would result in a fatal violation of RELMIN. However, we have seen above that the Danish data in (77) suggest that at least in complex verb constructions the landing site of Neg/QP-movement is v*P-internal, and thus cannot be regular object shift. In fact, Svenonius (2000:277) himself provides evidence that in simple tense constructions, too, the landing sites of object shift and Neg/QP-movement are different: the examples in (84) show that the former crosses sentential adverbs while the latter does not.

250 Chapter 4: Object shift and other movement types (84)

a.

b.

Hann les bókina ennþá/víst/sennilega. he reads the book yet/apparently/probably ‘he still/apparently/probably reads the book.’ Hann les ennþá/víst/sennilega . he reads few books yet/apparently/probably ‘He still/apparently/probably reads few books.’

Svenonius’ claim that we are dealing with A-movement in simple tense constructions also incorrectly predicts that the examples in (83b&b′) are completely excluded instead of just degraded. Furthermore, the fact established earlier that auxiliaries have unvalued case features that trigger regular object shift in Dutch and German raises the question why these case features cannot trigger Neg-movement in the Scandinavian complex tense constructions as well. Since these considerations cast serious doubts on Svenonius’ proposal, let us investigate other ways of accounting for the Neg- movement puzzle. First note that the discussion of wh-movement in the preceding subsection suggests that Neg/QP-movement must be preceded by Amovement. Since we have seen that the landing site of Neg/QP-movement is arguably v*P-internal, EPP(ij) will ensure that these movements are preceded by short object shift. This of course has no effect for main verbs, which obligatorily trigger short object shift anyway, but it does for the auxiliary verbs which normally do not trigger short object shift as this would lead to a violation of H-COMPL. Now consider the crucial Icelandic examples again, repeated here as (85). (85)

a. b.

Jón hefur Jón has nothing Jón sagði Jón said nothing

sagt Sveini . said Sveinn Sveini . Sveinn

(*IO-DO) ( IO-DO)

Fist consider example (85a), which shows that Neg-movement may cross the indirect object despite the fact that this movement is preceded by short object shift triggered by the ij-features on the auxiliary, and thus gives rise to a violation of RELMIN. The fact that this is allowed follows from the fact established earlier that RELMIN is outranked by H-COMPL, which blocks the short object of the indirect object across the main verb that would undo this violation of RELMIN. Now consider example (85b): the more or less standard assumption is that the inversion of the direct and the indirect object that results from Neg-movement can and therefore must be made undone in this case without violating Holmberg’s Generalization (our constraint H-COMPL) by applying regular object shift to the direct object.

Object shift and A′-movement 251

Although the suggested analysis makes intuitive sense, there is still one important problem from our present perspective: since the inversion of the direct and the indirect object is the result of Neg-movement (=A′movement), this inversion will not invoke a violation of RELMIN. This means that regular object shift cannot be forced by this constraint and will thus still be predicted to be “optional”, that is, subject to the normal information structure conditions. This suggests that we cannot solve the Negmovement puzzle by taking recourse to the shape conservation constraints. I would therefore like to propose an entirely different approach to this problem that adopts as its starting point the assumption that the [+affect] feature that triggers Neg/QP-movement is not introduced by some independent functional head, but is actually part of the verbal root of the finite verb. As a result, short object shift triggered by the finite verb will result in simultaneously satisfying EPP(ij) and creating the operator configuration required by Full Interpretation. From this it follows straightforwardly that Neg/QP-movement of the direct object will cross the indirect object in the complex tense construction only. First, consider the somewhat simplified derivation of the simple tense construction in (86). In this case the [+neg] feature is situated on the root of the main verb V, so that short object shift of the direct object has the concomitant effect of moving the direct object into its scope position; since the indirect object also undergoes obligatory short object shift, the inversion caused by Neg-movement (= short object shift) of the direct object will be undone. (86)

a. b. c. d.

[VP Vij/neg IO DOneg] [VP IO DOneg Vij/neg [VP tV tIO tDO]] [VP S v+V [VP IO DOneg tV [VP tV tIO tDO]]] etc.

In the perfect tense construction, the [+neg] feature is situated on the root of the auxiliary verb so that Full Interpretation will force short object shift of the direct object into the local domain of the root of the auxiliary despite the fact that this will invoke violations of H-COMPL and RELMIN. The inversion of the indirect and the direct object that results from Negmovement in this case cannot be made undone by shifting the indirect object into the local domain of the root of the auxiliary verb due to the subranking H-COMPL >> RELMIN: applying short object shift would satisfy the lower ranked constraint RELMIN by invoking a fatal violation of the higher ranked constraint H-COMPL. The global and somewhat simplified representation of the perfect tense construction is given in (87).

252 Chapter 4: Object shift and other movement types (87)

a. b. c. d. e. f. g.

[VP Vij IO DOneg] [VP IO DOneg Vij [VP tV tIO tDO]] [AspP Asp+V [VP IO DOneg t′V [VP tV tIO tDO]]] auxneg [AspP Asp+V [VP IO DOneg t′V [VP tV tIO tDO]]] [auxP DOneg auxneg [AspP Asp+V [VP IO t′DO t′V [VP tV tIO tDO]]]] [v*P S v*+aux [auxP DOneg taux [AspP Asp+V [VP IO t′DO t′V [VP tV tIO tDO]]]]] etc.

Recall that the verbal traces in (86/87b) are due to the fact that we have adopted Grimshaw’s (1997) theory of extended projections, which was discussed in section 3.1.2.1. Adopting this theory is crucial for the analysis sketched above, since this will allow us to create a “NegP” on top of VP by remerger (movement) of V in order to allow Neg-movement of non-objects. For example, the base structure in (88a) allows two extended projections of V in order to facilitate the movements triggered by the unvalued features on the verbal root: in (88b) an extended projection is created to host the shifted object, and in (88c) to enable Neg-movement of XP. (88)

a. b. c.

[VP Vij/neg DPij XPneg] [VP DPij Vij/neg [VP tDP XPneg]] [VP XPneg Vneg [VP DPij tV [VP tDP XPneg]]]

This kind of analysis can readily be extended to other types of A′-movement, and may even receive support from English do-support. First, note that do-support in questions is related to the presence of the CP-level. If CP is an extended projection of IP created by remerging the inflectional head I to IP, this may follow from the suggestion in fn.8 on p.36 that dosupport is required by NOVACM in order to make I-to-C movement (=remerger of I) visible as subject-aux inversion; see the structure in (89b) below. Secondly, assume that the unvalued feature that triggers wh-movement, which I will henceforth refer to as [wh], need not be expressed on C, but can be expressed on one of the lower heads in the extended projection of the verb. Evidence in favor of the second assumption is provided by, e.g., the Slavic languages and Hungarian, in which wh-movement targets some clause-medial position to the right of the complementizer (if present). Finally, assume that in English the wh-feature can be assigned to I (Rizzi 1996). If so, movement of a wh-subject into SpecIP will be sufficient both for satisfying the constraint EPP(person) and for satisfying Full Interpretation (Rizzi’s Wh-criterion). Consequently, no CP-level need be created and do-support will not apply. This results in the IP-structure in (89a). However, when we are dealing with a wh-object, an

Object shift and A′-movement 253

additional CP-layer must be created by remerger of I in order to enable wh-movement of the object, with the concomitant effect of do-support in (89b). (89)

a. b.

[IP Who I[wh] [came twho]] [CP what did[wh] [IP you tI [tyou see twhat]]]

The remainder of this section will discuss the consequences of the sketched proposal for the description of Neg-movement in the Mainland Scandinavian languages. The discussion will be illustrated by means of Danish data taken from Christensen (2005:163-5). Consider the examples in (90). The judgments on these examples are as predicted: (90a) is excluded by Full Interpretation because Neg-movement has not taken place; the object shift of the indirect object pronoun (90b) is excluded since it fatally violates H-COMPL; (85c) is acceptable since it satisfies both H-COMPL and Full Interpretation. (90)

a.

*Jeg I b. *Jeg c. Jeg

har lånt hende ingen bøger. have lent her no books har hende ingen bøger lånt tIO tDO. har ingen bøger lånt hende tDO.

Now, consider the simple tense constructions in (91). Again the judgments are as predicted: like (90a), (91a) is excluded by Full Interpretation; the contrast between (91b) and (91c) follows from the fact that regular object shift of the pronominal indirect object is required by the high ranking of D-PRONOUN in Danish. Observe again that the landing sites of regular object shift and Neg-movement differ. The former precedes the adverb faktisch, whereas the latter follows it. (91)

a.

*Jeg lånte faktisch hende ingen bøger. I lent actually her no books b. Jeg lånte hende faktisch ingen bøger tIO tDO. c. *Jeg lånte faktisch ingen bøger hende tDO.

Observe that the order given in (91c) implicitly shows that Christensen adopts the standard assumption that in the absence of regular object shift, Neg-movement results in inversion of the two objects. Now consider the examples in (92), in which the indirect object is a non-pronominal DP. The ungrammaticality of example (92c), in which the indirect object has undergone regular object shift, is of course expected as non-pronominal DPs normally do not undergo this movement in Danish.

254 Chapter 4: Object shift and other movement types

This leaves us with the two orders in (92a&b). The fact that (92b) is ungrammatical is unexpected under the standard assumption that Negmovement inverts the order of the indirect and the direct object. Since we have seen just now that Christensen implicitly adopts this assumption, it does not come as a surprise that he concludes from (92a&b) that Negmovement does not apply in this case. (92)

a.

Jeg I b. *Jeg I c. *Jeg I

gav gave gav gave gav gave

jo barnet ingenting. prt. the-child nothing jo ingenting barnet tDO. prt. nothing the-child barnet ingenting jo tIO tDO. the-child nothing prt

Christensen’s conclusion is, of course, at odds with our leading assumption in (82a) that Neg-movement is necessary in order to satisfy Full Interpretation (Haegeman’s 1995 Neg-criterion): not applying Negmovement leads to a crashing derivation. However, the new proposal, according to which the [+affect] feature that triggers Neg/QP-movement is part of the verbal root of the finite verb, does not predict that Negmovement in simple clauses results in inversion of the two objects: short object shift of the direct object simultaneously results in satisfying Full Interpretation, and the resulting inversion of the direct and indirect object is subsequently undone by obligatory short object shift of the latter. Therefore, if one wants to maintain the claim that not applying Neg-movement results in a crashing derivation, the judgments on the examples in (92) provide strong evidence for this proposal. 4.3.3.

Topicalization

This subsection concludes our discussion of the interaction of object shift and A′-movement by discussing the fact already mentioned in section 3.1.3 that topicalization of the main verb may feed object shift; see Engels & Vikner (2006) for data that show that this is a property of a wider range of topicalization constructions. First, consider the Swedish examples in (93), taken from Holmberg (1999). As we have in section 2.4.2.3, the shifted order in (93a) can in principle be derived by moving the object Marit into the local domain of the verbal root of the auxiliary verb, followed by movement of the auxiliary to I. The fact that this order does not surface in the Scandinavian languages is due to the fact that this would result in a fatal violation of H-COMPL, which prohibits that the object precedes the main verb. The acceptability of

Object shift and A′-movement 255

example (93b) shows, however, that this violation of H-COMPL can be voided by topicalization of the participle. (93)

a. b.

Jag har inte kysst I have not Marit kissed ? Kysst har jag inte Marit Kissed have I not Marit

Note in passing that the claim that the derivation of (93b) involves short object shift of the object into the local domain of the auxiliary makes it possible to assume in accordance with the Emonds (1976) structure preservation constraint that the participle is placed into first position as the result of remnant VP-topicalization. Holmberg (1999), on the other hand, has to claim that this is the result of V-topicalization, which violates this constraint. I will return to Holmberg’s claim later in this section. Given that topicalization of the participle makes short object shift into the local domain of the verbal root of the auxiliary possible, it does not come as a surprise that the same holds for regular object shift of a pronoun into the local domain of the light verb v associated with the auxiliary. This is shown in the Swedish examples in (94), also taken from Holmberg (1999). From now on, the discussion will only consider the interaction of topicalization and regular object shift. (94)

a.

b.

Jag har inte kysst . I have her not kissed ‘I haven’t kissed her.’ Kysst har jag henne inte. Kissed have I her not

We have seen in section 3.1.3 that Holmberg (1999: section 3) has argued that examples like (94b) cannot be accounted for if we assume that regular object shift is a syntactic operation. The argument is repeated in somewhat more detail here. Example (94a) shows that object shift cannot apply across the main verb when the latter is in situ. Holmberg claims that this is a property of the derivation and not of the resulting representation, as is stated in (95). (95)

A violation of [Holmberg’s Generalization] cannot be repaired by subsequent operations (Holmberg 1999:9).

Statement (95) is motivated by the examples in (96) and (97). First, (96) shows that regular object shift of exceptionally case-marked subjects (of

256 Chapter 4: Object shift and other movement types

infinitival clauses and small clauses) is possible, as long as it does not cross the matrix verb. (96)

a.

Jag I b. *Jag I

hörde henne heard her har henne have her

inte [thenne hålla föredrag] not give talk inte [VP hört [thenne hålla föredrag]] not heard give talk

If HG could be repaired by subsequent operations, we would wrongly predict that (97a) is acceptable, as the matrix verb hört now precedes the shifted pronoun henne ‘her’ as a result of VP-topicalization. Since (97b) shows that VP-topicalization of this type is possible as such, Holmberg concludes that something like the restriction in (95) must hold. (97)

a. *[VP hört [thenne hålla föredrag]] jag har henne inte b. [VP hört [henne hålla föredrag]] jag har inte

This conclusion subsequently led Holmberg to the conclusion that (94b) could only be derived by first moving the participle into SpecCP, as in (98a), and subsequently applying object shift, as in (98b). The derivation in (98) is, however, counter-cyclic and thus violates Chomsky’s (1995a: ch.3) extension condition on movement. (98)

a. b.

[CP kysst [C′ har jag inte [VP tkysst henne]]] [CP kysst [C′ har jag henne inte [VP tkysst thenne]]]

Holmberg solves this problem by assuming that object shift is not a syntactic but a post-syntactic operation, which he assumes not to be subject to the extension condition. The contrast in (94) can now be accounted for by assuming that object shift is subject to a general condition that it cannot apply across phonologically visible VP-internal material (except adjuncts): the post-syntactic representation of (94a) prior to object shift is (99a), and hence object shift in (99a′) is blocked; the representation of (94b) prior to object shift is (99b), and hence object shift in (99b′) is allowed. (99)

a. jag har inte [VP kysst henne] a′. *jag har henne inte [VP kysst henne] b. kysst har jag inte [VP kysst henne] b′. kysst har jag henne inte [VP kysst henne]

Holmberg’s argument above is crucially based on the condition in (95) that object shift cannot cross the main verb during the syntactic derivation.

Object shift and A′-movement 257

This claim is, however, incompatible with the current version of phase theory. For example, I argued in the preceding subsections in favor of the following standard assumptions: (i) the Phase Impenetrability Condition (PIC) forces wh-movement of the object to proceed via the edge of v*P, and (ii) the intermediate movement is triggered by the case-feature on v*. This implies that wh-movement of the object is obligatory preceded by regular object shift, which crosses the verb (v*+V) during the derivation. The claim in (95) therefore wrongly predicts wh-movement of objects to be categorically excluded in the Scandinavian languages. We therefore have to choose between dropping the PIC or dropping Holmberg’s assumption in (95). Since dropping the PIC would be at odds with the evidence in favor of the “fewest steps” approach to A′-movement from the preceding subsections, I conclude that it is (95) that should be dropped. A theory-internal argument in favor of the same conclusion is that condition (95) has no natural place in the D&E framework adopted in this study. According to this framework, HG must be attributed either to some condition on the operations of the generator, or to the evaluation of the output representation. Since it was extensively argued in section 3.1 that the former cannot be maintained, the latter must be the case. Consequently, we must conclude that the derivation that leads to (94b) is immaterial, so that the claim in (95) cannot be correct. The only thing that matters is whether H-COMPL is violated in the output representation. Finally, I believe that Holmberg’s proposal runs counter to the minimalist spirit given that it states HG twice in the grammar: once in the form of condition (95) on syntactic derivations and once as a condition on the postulated postsyntactic rule of object shift. This is, of course, undesirable unless one is willing to say that HG is a general condition on displacement, irrespective of whether it is the result of a synactic or a postsyntactic process. But this would raise the question whether these processes are truly different, or whether we are simply saying that there are two syntactic cycles. Since I have seen no conclusive evidence for the claim that these syntactic and postsyntactic displacement processes are truly different, I conclude that we are actually introducing an additional syntactic cycle. I believe this to be an unwelcome result in view of the minimalist goals we have been setting over the last decade. For the reasons above I conclude that we must drop (95). This has the important advantage that we can also drop the–in my view–dubious underlying assumption that topicalization of the past participle in (98) is the result of head-movement of the participle into SpecCP, and assume a runof-the-mill remnant VP-topicalization analysis instead; cf. Engels & Vikner

258 Chapter 4: Object shift and other movement types

(2006) for the same conclusion and some empirical arguments against head-movement. A slightly simplified derivation of (94b) is given in (100). (100) a. b. c. d. e.

har kysst henne henne har t′DO kysst tDO henne inte t′′DO har t′ DO kysst tDO Jag har henne inte t′′DO taux t′ DO kysst tDO [CP [PartP kysst tDO] har [IP jag t′ aux henne inte t′′DO taux t′ DO tPartP]]

From the structure in (100a), (100b) is derived by the two applications of short object shift triggered by the ij-features on, respectively, the past participle and the auxiliary. As we have seen in our discussion of the perfect constructions in Dutch, these movements are allowed by the generator. The structure in (100c) is derived by merging the negative adverb inte ‘not’ and application of regular object shift, and (100d) by the subsequent application of V-to-I to the finite auxiliary; the subject jag ‘I’ is of course moved from some lower position, but this is ignored here. The surface structure in (100e) is derived by moving the auxiliary into the Cposition and topicalization of a projection of the participle (indicated by PartP) that includes at least the trace in the original position of the direct object. Since the surface order satisfies H-COMPL and the derivation is in full accordance with Chomsky’s extension condition on movement, this is the representation that will be selected as the optimal candidate. It must be acknowledged, of course, that rejecting (95) leaves the ungrammaticality of ECM examples like (97a) as an unsolved problem, but this is certainly not the only one concerning VP-topicalization. For example, the fact noted in Den Besten & Webelhuth (1990) that stranded prepositions cannot be pied piped under VP-topicalization has not received a satisfying solution either (and is mostly ignored in the literature).56 The unacceptability of Dutch examples like (101b) might lead to the idea that subextraction from a constituent contained in a topicalized VP is excluded, which would also effectively rule out the Swedish example in (97a). (101) a.

Jan wil daari niet [VP [PP ti op] wachten] Jan wants there not for wait ‘Jan doesn’t want to wait for it.’ b. *[VP [PP ti op] wachten] wil Jan daar niet

56

See the discussion in section 2.4.2.5. Recall that the fact that VP-topicalization can leave stranded prepositions behind in the right periphery of the clause does follow if we adopt the analysis proposed there.

Object shift and A′-movement 259

It is not entirely clear whether this proposal is tenable since it predicts that examples comparable to (97a) are also ungrammatical in Dutch. I give some examples in (102), which are actually the Dutch renderings of the crucial Swedish examples provided by Holmberg (1999) in favor of his proposal (his (19)-(21)). The status of the Dutch examples is unclear given that native speakers provide varying judgments; although these examples are clearly more marked than the more simple cases comparable to the Swedish example in (94b), I would be willing to accept them as grammatical, but some of the speakers I have consulted flatly reject them. (102) a.

%

Een praatje horen geven heb ik haar nooit. a talk hear give have I her never b. %Verward gemaakt heeft het me altijd. confused made has it me always c. %Naar de bodem laten zinken heeft hij het nooit. to the bottom let sink has he it never.

If these examples are grammatical, we must conclude that my suggestion based on the ungrammaticality of examples like (101b) cannot be upheld. However, if they are ungrammatical, we should conclude (again) that it is Holmberg’s account of the unacceptability of (97a) that should be dropped. In order to see this, first note that Holmberg’s proposal cannot be extended to the Dutch case, because it crucially relies on to HG, reinterpreted as a condition on the derivation. Since Dutch is not subject to HG, the ungrammaticality of the Dutch examples in (102) must receive an alternative account for this by, e.g., maintaining the earlier suggestion that there is a ban on subextraction from a constituent contained in a topicalized VP. Since this account will also rule out the relevant Swedish examples, Holmberg’s account becomes superfluous. This subsection has shown that Holmberg’s claim in (95) must be dropped both on empirical and on conceptual grounds. Further, I have argued that (95) predicts that the Dutch renderings of the Swedish examples that Holmberg used in support of this claim are grammatical in Dutch: this is due to the fact that (95) excludes the relevant Swedish examples by taking recourse to HG, which Dutch is not subject to. Unfortunately, this prediction is difficult to test since judgments on the pertinent Dutch examples vary from “marked” to “unacceptable”. I believe that this variation in judgments should make us wary, and stop us from jumping to conclusions without a more thorough investigation of the grammatical status of the

260 Chapter 4: Object shift and other movement types

examples involved. 57 Consequently, we must conclude that, for the time being at least, the topicalization data do not constitute reliable evidence in favor of or against any particular theory: at best, we can let the theory decide. 4.4. Conclusion This section has discussed the interaction of object shift with the three movement types that are normally distinguished: head-movement, Amovement, and A′-movement, as well as a number of problems that were put aside in the earlier chapters. Section 4.1 discussed the following dichotomy between short and regular object shift in Danish and Swedish: whereas the former pushes the verbal root up into v/Asp, the latter does not push-up the verb into I. This was accounted for by assuming that the constraint NOLEXM, which disfavors movement of the main verb, is ranked in between EPP(ij), which favors short object shift, and D-PRONOUN, which favors regular object shift of pronouns—the subranking H-COMPL >> EPP(ij) >> NOLEXM favors short object shift over non-application of verb movement, whereas the subranking H-COMPL >> NOLEXM >> D-PRONOUN prefers non-application of verb movement to application of regular object shift of pronouns. Section 4.2 discussed the consequences of our analysis of object shift for subject shift in Dutch, and tried to solve the problem that the subranking EPP(ij) >> ALIGNFOCUS >> *MOVE incorrectly predicts that subject shift is obligatory in Dutch. This led to the hypothesis that there are actually separate EPP constraints for the [gender], [number], and [person] features. The fact that short object shift is obligatory in Dutch whereas subject shift is optional can now be accounted for by assuming that these movements are triggered by different features: since short object shift is triggered by the [gender] feature on V, whereas subject shift is triggered by the [person] feature on I, the desired distinction can be derived by assuming the subranking EPP(gender) >> ALIGNFOCUS >> EPP(person) >> *MOVE. It was further shown that distinguishing EPP(gender) and EPP(person) is also necessary to account for the fact that Scandinavian subject and object shift differ in that only the former may cross the main verb. Finally, section 4.3 argued that there is no direct interaction between object shift and A′-movement; those interactions that seemingly arise are due to the fact that, as a result of the (inviolable) Phase Impenetrability 57 Perhaps even more so, since Holmberg (1999:11) points out himself that ECMconstructions are also problematic for his proposal that preposing of the participle involves V-topicalization: it wrongly predicts that the infinitival clause can be stranded by topicalization of the participle.

Conclusion 261

Condition and the (violable) constraint EPP(ij), A′-movement normally does not skip A-positions that could in principle be used as intermediate landing sites. This section concluded with a discussion of the fact that topicalization of the main verb may feed regular object shift. It was argued that this follows from the fact that, in the Scandinavian language as well, H-COMPL allows object shift to cross the verb as long as the original order is restored in the output representation. This also enabled us to drop Holmberg’s (1999) dubious assumption that topicalization of the verb involves head-movement, and replace it by a run-of-the-mill remnant VP-movement analysis.

Chapter 5 Predicate movement

The discussion in the previous chapters has provided a detailed analysis of short and regular object shift, while on the way also providing an analysis of subject shift (movement of the subject into SpecIP) and verb movement (V-to-v/Asp and V-to-I). The analysis took the derivation-and-evaluation (D&E) framework as its point of departure, according to which the representations that are generated by the computational system CHL are evaluated in an optimality-theoretic manner on the basis of a limited set of constraints. The set of constraints invoked in the account of object/subject shift and verb movement are given in (1). (1)

a.

b.

CHL constraints: (i) Economy constraints: *MOVE, NOLEXM, NOVACM (ii) EPP constraints: EPP(ij), EPP(case), *STRAY FEATURE Interface constraints: (i) PF constraints: RELMIN, H-COMPL (ii) LF constraints: ALIGNFOCUS, D-PRONOUN

As was discussed in Chapter 1, the OT-evaluation postulated in the D&E framework can be seen as an alternative for the output filters assumed in the various versions of the P&P framework. Although these P&P filters may express useful insights, they are problematic since they normally appear in the guise of complex statements. The OT formalism, on the other hand, derives these statements from more primitive notions of the theory, such as the constraints in (1). This enables us to study the filter component in its own right: whereas the credibility of the proposed filters is often very hard to judge due to their complexity, judging the credibility of the postulated constraints is much easier given that these take the form of simple and general statements. Another problematic aspect of filters is that they are inherently ad hoc in nature given that they are formulated to rule out a restricted set of constructions/examples, and can therefore never be independently supported. This is different from the simpler constraints in (1): their simplicity and generality makes it possible to apply them to a wider range of cases. This chapter, which is of a more speculative nature than the previous ones, aims at illustrating this by claiming that EPP(ij) does not only trigger short

264 Chapter 5: Predicate movement

object shift, but also predicate movement. The assumptions underlying this claim will be spelled out in section 5.1. Sections 5.2-5.5 will explore the consequences of the claim for several constructions in different languages: locative inversion in English, predicate movement in Dutch, movement of the so-called verbal modifiers in Hungarian, and l-participle fronting in Bulgarian. 5.1. The underlying assumptions This chapter takes Hoekstra & Mulder’s (1990) analysis of the locative inversion construction in (2b), which will henceforth be referred to as H&M, as its point of departure and aims at reformulating it in more current terms, while extending the empirical scope of the original proposal to a larger set of constructions involving displacement of small clause predicates and VPs in English, Dutch, Hungarian, and Bulgarian. (2)

a. b.

The baby carriage rolled down the hill. Down the hill rolled the baby carriage.

The new proposal will stress H&M’s basic insight (based on earlier work by Moro 1997) that locative inversion in (2b) is licensed due to the fact that the predicative phrase down the hill and the DP the baby carriage are in an agreement relation; cf. section 5.2.1 for the details. This agreement relation between a predicate and its argument is sometimes morphologically reflected, as in the case of adjectival agreement in the Italian copular construction in (3a). Furthermore, it is formally identical to object agreement in (3b) in the sense that it involves agreement in gender and number. The examples in (3) are taken from Burzio (1986). (3)

a. b.

Maria è malata. Maria is ill3SG.fem Maria è arrivata. Maria has arrived3SG.fem

This chapter assumes that the predicative agreement relation is not restricted to APs but holds cross-categorially. This means that the predicative PP in (2) also agrees with the DP in ij-features, even though in this case the agreement relation is not morphologically expressed. Further, it is assumed that the predicative agreement relation as well as the object agreement relation is present cross-linguistically, despite the fact that this agreement relation has a morphological reflex in a limited number of languages only. In short, I will adopt the statements in (4) as axioms.

The underlying assumptions 265 (4)

a. b.

Axiom I: A predicate agrees in ij-features with the DP it is predicated of. Axiom II: V agrees in ij-features with its internal argument or the subject of its internal argument (when the internal argument is a small clause or an infinitival complement of the Subject Raising or the ECM-type).

When we adopt the axioms in (4), H&M’s basic insight can be rephrased in more current terms as in (5), which obviously constitutes the null hypothesis from the perspective of the Last Resort Condition on movement. Given axiom (4a) and hypothesis (5), it is now correctly predicted that in (2) either the DP the baby carriage or the predicative PP down the hill can value, and hence be attracted by, the unvalued ij-features on I. (5)

Null hypothesis: If A and B agree in ij-features, both A and B can be the goal of unvalued ij-features on some higher head H, and, consequently, be a candidate for internal merge with H.

The two axioms in (4) differ considerably in status, in the present-day versions of the principles-and-parameters framework. Axiom II seems widely accepted: it constitutes the core of most discussions within MP that deal with object movement, and has also been adopted in the previous chapters. Axiom I is generally accepted as far as predicative adjectives are concerned (see, e.g., Baker, to appear, who suggests that this is a universal property of languages), but much less so in the case of prepositional or nominal predicates: the former, for example, never exhibit morphological agreement, and (as is of course expected) the latter only exhibit agreement insofar as the predicative noun phrase is not inherently marked for ij-features. The examples in (6), for example, show that although the nominal predicate does agree in number with the subject, the two do not agree in person. (6)

a. b.

Jij yousg. Jullie youpl.

bent are zijn are

de beste kandidaat. the best candidate de beste kandidaten. the best candidates

It is not clear whether this type of disagreement matters much for what will follow: the intended result of Axiom I is to ensure that a predicate is endowed with ij-features which can subsequently be probed by unvalued features on the higher functional heads; cf. (5). If nominal predicates do have ij-features of their own, they will be potential goals for these features

266 Chapter 5: Predicate movement

by definition, and nothing more need be said (cf. section 5.2.3.4 for relevant discussion). This means that it is especially my claim that predicative PPs are endowed with ij-features that is controversial. It can only be indirectly supported, and this is what I intend to do in this chapter. In assessing Axiom I, it is also important to realize that it replaces H&M’s implicit claim that predicative PPs are assigned case, which also features prominently in Koster’s (1994) analysis of predicate movement in Dutch; cf. section 5.3.2.1. Since predicative PPs do not show case agreement with their subject, this claim is of course equally hard to prove. Den Dikken (2006) assumes in his introduction to section 4.3.2 that, in accordance with the Last Resort Condition on movement, locative inversion implies some Agree relation between I and the inverted PP but he does not commit himself to any claim about the nature of this agreement relation. When we want to retain the Last Resort Condition, we have two options: either we postulate some new unvalued feature on I that enters into an Agree relation with the predicative PP and try to find independent evidence for this feature, or we appeal to one of the independently motivated features on I. The discussion below explores the latter option and will argue on empirical grounds that appealing to ij-features is a better option than taking recourse to case features. From axiom (4a) and hypothesis (5), the two statements in (7) follow as a corollary. Since the small clause predicate is predicated of the subject of an unaccusative construction, and the subject can be a goal of the ij-features on both the verbal root V and I, the predicate can be a goal of these heads as well; as a result, movement may involve either the subject or the predicate, as stated in (7a). Similarly, since the predicate is predicated of the object of a transitive construction, and since the object can be the goal of the ij-features on the verbal root V, the predicate can be a goal of the verbal root as well; as a result, movement may involve either the object or the predicate, as stated in (7b). Note that the formulation in (7) leaves open the question under what conditions predicate movement applies instead of DP movement, but this will be discussed extensively later in this chapter. (7)

Predicate movement may target any position in the local domain of a head H with unvalued ij-features that is normally targeted by: a. the nominative DP in an unaccusative construction (=SpecVP or SpecIP); b. the accusative DP in a transitive construction (=SpecVP).

Given axiom (4b) and hypothesis (5), we may conclude that a verb phrase may target the same positions as its internal argument, unless the pertinent position is locally related to the head of the VP itself, since, of course, a VP

The underlying assumptions 267

cannot value the features on its own head. This means that if we phrase the corollaries from the perspective of the attracting head, we get the corollaries in (8): (8a) is the standard assumption, (8b) is a reformulation of the two corollaries in (7), and (8c) is the corollary that follows from (4b) and (5). (8)

Unvalued ij-features on H may attract: a. argument DPs b. small clauses c. verb phrases

Corollaries (8b&c) show that adopting the axioms in (4) and hypothesis (5) extends the empirical scope of H&M’s original insight considerably. This chapter argues that the constructions in (9) behave in accordance with the corollaries in (7) and (8) and thus provide indirect support for (4) and (5). (9)

a.

b. c. d.

English: Locative inversion and Preposing around Be (e.g., Emonds 1976, Hoekstra & Mulder 1990, Rochemont & Culicover 1990, Den Dikken & Næss 1993, Bresnan 1994, Den Dikken 2006) Dutch: Predicative movement (e.g., Zwart 1993,1997, Koster 1994) Hungarian: Movement of verbal modifiers (e.g., Brody 1990b, É.Kiss 2002,2006) Bulgarian: Fronting of l-participles (e.g., Lema & Rivero 1989, Broekhuis & Migdalski 2003)

The remainder of this chapter is organized as follows. Section 5.2 develops in more detail the alternative analysis for locative inversion outlined above, and will show that locative inversion is an instantiation of (7a). Section 5.3 will turn to Dutch and show that this language has no locative inversion of the English type. It has been argued, however, that Dutch does have predicate movement into PredP, and I will show that this movement shares several properties with locative inversion. I will therefore reformulate this movement along the lines suggested above by claiming that it is an instantiation of (7b). Section 5.4 will continue with É. Kiss’ (2006) suggestion that the so-called verbal modifiers in Hungarian are actually predicative phrases, which must therefore be moved into PredP. I will argue that, in this case also, we can reanalyze the data along the lines indicated above. Section 5.5 will finally take up the discussion of l-participle fronting in Bulgarian, and suggest that in this language both types of predicate movement in (7) occur. Before I start, a warning flag is needed. This chapter does not provide a fully developed theory about the constructions investigated here but should rather be seen as a program for future research. I mainly aim at providing

268 Chapter 5: Predicate movement

empirical support for the axioms in (4) and hypothesis (5), and therefore I occasionally only indicate the direction the analysis should take or sidestep questions to which I do not have a full-fledged answer yet. Nevertheless, I believe that the discussion will provide a better understanding of the construction types that are investigated here. 5.2. Locative inversion in English This section develops an alternative to H&M’s influential analysis of locative inversion that will be based on hypothesis (5). Section 5.2.1 starts with a brief review of H&M’s proposal, continues with arguing that the introduction of the operation Agree requires an alternative formulation of their proposal, and concludes with providing a sketch of what the analysis might look like in the Minimalist Inquiry framework. Sections 5.2.2 and 5.2.3 provide my own proposal within the D&E framework, and section 5.2.4 shows that this proposal can readily be extended to cases involving verbal projections. Section 5.2.5 ends the discussion of locative inversion with three concluding remarks on the analysis of developed in this section. 5.2.1.

Hoekstra & Mulder’s analysis from an MP perspective

H&M have argued that in the locative inversion construction in (10b) the prepositional phrase down the hill occupies the same position as the subject in (10a). They claim that the motivation for the movement into SpecIP is the same in both cases: this movement is necessary to satisfy the Case Filter, that is, to assign nominative case to the subject. (10)

a. b.

The baby carriage rolled down the hill. Down the hill rolled the baby carriage.

Since the PP is predicated of the subject the baby carriage, H&M claim that the two are generated as part of a small clause. This implies that we are actually dealing with an unaccusative construction, so the surface structures of the two examples in (10) are approximately as given in (11a) and (11b), respectively. That the subject can be marked with nominative case in (11a) is obvious as it is situated in the designated position where, under the assumptions of the 1980s, this case is assigned. That the subject can be assigned nominative case in (11b) is less clear, though. (11)

a. b.

[IP DP I [VP V [SC tDP Pred ]]] [IP Pred I [VP V [SC DP tPred ]]]

Locative inversion in English 269

H&M solve this by proposing that the movement of the predicative PP into SpecIP makes it possible to transfer nominative case to the subject in its base position. Their account rests on the three assumptions in (12). They also assume that each element has a unique index, from which it follows that co-indexing is transitive: if A is co-indexed with B and B with C, then A is also co-indexed with C. (12)

a. b. c.

A moved phrase is co-indexed with its trace. Finite I assigns nominative case to its specifier; nominative case assignment involves co-indexing of I and the element receiving case. Predication relations involve co-indexing of the predicate and the DP it is predicated of.

According to the assumptions in (12), the co-indexing in example (10a) is as given in (13a). Given that the DP is assigned the indices i and j under (12a) and (12b), and the DP trace is assigned the indices i and k under (12a) and (12c), we can conclude from the unique index requirement that i=j=k. For this reason (13a) is equivalent to (13a′). As the reader can verify him/herself, co-indexing of the phrases in (10b) is as given in (13b), which is equivalent to (13b′). (13)

a. a′. b. b′.

[IP DPi/j Ij [VP V [SC ti/k Predk ]]] Ÿ [IP DPi Ii [VP V [SC ti Predi ]]] [IP Predj/k Ij [VP V [SC DPi tk/i ]]] Ÿ [IP Predi Ii [VP V [SC DPi ti ]]]

In (13a) nominative case is assigned directly to the DP in SpecIP, but in (13b) this case is, in a sense, transferred to the DP in its base position via the chain of indices. Since I is co-indexed with the predicative PP in SpecIP under (12b), the PP is co-indexed with its trace under (12a), and the PP trace is co-indexed with the subject DP under (12c), it follows from the transitivity of co-indexing that I is also co-indexed with the postverbal DP, and, as a result, assigns nominative case to it. H&M’s proposal predates the minimalist program. Among other things, this is clear from the fact that their analysis crucially relies on co-indexing, which violates Chomsky’s (1995a: ch.3) inclusiveness condition. If this condition is to be respected, a reformulation of the proposal is called for. Their proposal is in accordance with the Last Resort Condition on movement, since it claims that the movement of the subject/predicative PP is motivated by the Case Filter. However, this proposal cannot be maintained in the current Minimalist Inquiry framework, given the assumption that the case feature of the subject can be valued under Agree. The obligatory movement

270 Chapter 5: Predicate movement

of the subject/PP should therefore follow from an EPP-feature on I. Note that assuming this also voids the problem that H&M’s proposal poses for the inclusiveness condition; since there is no “transfer” of nominative case to the subject via a chain, the indices in (13) are not needed. However, this still leaves us two questions that need to be answered. First, postulating an EPP-feature on I is not sufficient to trigger movement of the predicative PP, as the Last Resort Condition on movement only allows this when the PP is probed by I. This leads to the question in (14a), which in effect asks what unvalued feature on I triggers the movement of the predicative PP. Secondly, we want to know what determines the choice between (10a) and (10b). Rochemont & Culicover (1990) and Bresnan (1994) have argued that the choice between the two examples is related to the information structure of the clause: example (10b) is possible only if the subject belongs to the focus (new information) of the clause. So this observation leads to the question in (14b). (14)

a. b.

Which unvalued feature on I probes the predicative PP? How can we account for the focus restriction on locative inversion?

Let us start with question (14a) concerning the trigger for the movement of the predicative PP. The answer to this question seems implicitly given by the intuition underlying H&M’s proposal, namely that the predicate and the DP it is predicated of agree, which in current terms means that they have the same ij-features. Given that I has unvalued ij-features, we may assume that the ij-features on the predicate make locative inversion possible. If this is on the right track, this gives rise to hypothesis (5): if A and B agree in ij-features, both A and B can be the goal of the unvalued ij-features on some higher head H, and, consequently, be a candidate for internal merge with H. It is this hypothesis that I want to investigate in the remainder of this chapter. The fact that H&M did not opt for ij-features but for the case feature on I is of course related to the fact that the former were not yet considered potential triggers of A-movement at the time that H&M put forth their proposal; the Case Filter singled out case as the only potential trigger for this type of movement. One of the reasons to select the ij-features rather than case for capturing H&M’s intuition is that this solves the empirical problem for their original proposal that in some languages predicates can be assigned a case different from the case assigned to the DP they are predicated of. This is illustrated by means of the Hungarian example in (15), taken from É.Kiss (to appear), where the predicate is dative, whereas the DP it is predicated of is accusative.

Locative inversion in English 271 (15)

János-t okos-nak tartott-ák. Johnacc cleverdat regarded3PL ‘They regarded John clever.’

Now consider question (14b) concerning the focus restriction on locative inversion: why is locative inversion only possible when the subject is part of the focus of the clause? In the Minimalist Inquiry framework, this should be attributed to a filter comparable to the one postulated by Chomsky (2001:31) in his account of Holmberg’s Generalization. This filter might require that a subject that is part of the focus of the clause remain vP-internal if a predicative PP can be used as an alternative to satisfy the EPP-feature on I. The conditional clause is of course needed to avoid the incorrect prediction that examples like (16b) are grammatical when the DP John is part of the focus of the clause. (16)

a. John/JOHN died. b. *died JOHN.

Although Chomsky’s (2001) discussion of Icelandic regular object shift reviewed in section 1.1.1 shows that it is not impossible to formalize this kind of conditional descriptive generalization within the Minimalist Inquiry framework, I prefer to develop an analysis within the D&E framework for the reasons already outlined in Chapter 1. 5.2.2.

Locative inversion in the D&E framework

We have seen in section 2.3.4 that whereas English does not have regular object shift, triggered by the unvalued case feature on v, it does have obligatory short object shift, triggered by the unvalued gender feature on V. This can be accounted for by assuming that English has (i) the subranking EPP(gender) >> ALIGNFOCUS >> *MOVE, which expresses that short object shift normally applies and cannot be blocked by considerations of information structure, and (ii) the subranking *MOVE >> EPP(case), which expresses that regular object shift is “normally” blocked. From the weak ranking of EPP(case) it also follows that the unvalued case feature on I cannot be held responsible for the EPP-property of English, that is, the fact that the subject position is obligatorily filled. This implies that the EPPproperty must follow from the fact that EPP(person), just like EPP(gender), outranks both ALIGNFOCUS and *MOVE. The English ranking is therefore as given in (17); since the relative ranking of EPP(person) and EPP(gender) will not be important in what follows, I will henceforth use EPP(ij) as an abbreviation for {EPP(gender), EPP(person)} for simplicity.

272 Chapter 5: Predicate movement (17)

English: EPP(ij) >> {ALIGNFOCUS, *MOVE} >> EPP(case)

It is worth pointing out that the independently established ranking in (17) supports my earlier claim that movement of the predicative PP in the locative inversion construction cannot be triggered by the case feature on I. Consequently, if one wants to maintain the Last Resort Condition on movement, it also supports my reformulation of H&M’s proposal, which holds the ij-features responsible for triggering locative inversion. An even more interesting property of the ranking in (17) is that it provides us with a more or less straightforward analysis of the locative inversion constructions in (10), repeated here as (18). (18)

a. b.

The baby carriage rolled down the hill. Down the hill rolled the baby carriage.

Let us first investigate what (17) predicts when the subject is part of the presupposition of the clause, that is, for examples like (18a). Consider the evaluation in Tableau 1. Since the subject is not part of the focus of the clause, the constraint ALIGNFOCUS is not relevant here. Leaving SpecIP empty is excluded since this leads to a fatal violation of the highest ranked constraint EPP(ij): this blocks the third candidate. Because the first and second candidate score equally well with respect to *MOVE, EPP(case) gets the last say: movement of the PP violates both *MOVE and EPP(case), whereas movement of the subject violates *MOVE only, and therefore the latter is preferred. We thus correctly predict that the first candidate is the optimal one. Tableau 1: DP & Pred ⊄ focus

example (18a) DP I ... V tDP Pred Pred I ... V DP tPred e I ... V DP Pred

EPP(ij)

) *!

AF

*MOVE * *

EPP(case)

*! *

If the subject is part of the focus of the clause, as in (18b), the evaluation proceeds as indicated in Tableau 2, where the focused phrase is given in italics. As was also the case in Tableau 1, leaving SpecIP empty leads to a fatal violation of EPP(ij). Moving the subject into SpecIP results in a structure violating ALIGNFOCUS, which can be avoided by moving the predicative PP instead. The second candidate with locative inversion is therefore the optimal one.

Locative inversion in English 273 Tableau 2: DP ⊂ focus; Pred ⊄ focus

example (18b) DP I ... V tDP Pred Pred I ... V DP tPred e I ... V DP Pred

EPP(ij)

AF

*!* ) *!

*MOVE * *

*

EPP(case)

* *

Two remarks on Tableau 2 are in order. First, I want to note that I am aware of the fact that there are good reasons to be suspicious about the conclusion that the first candidate is not optimal and therefore unacceptable; I will return to this after I have completed the discussion of the predictions that follow from the ranking in (17). Secondly, I want to repeat that the constraint ALIGNFOCUS is gradient, which means that each constituent that follows the focus of the clause will invoke a violation of this constraint. This accounts for the fact that the first and third candidates in Tableau 2 differ in the number of violations: in the first candidate, the DP in focus is followed by the verb and the predicative phrase, whereas in the third candidate it is only followed by the predicate. In the present evaluation, the number of violations of ALIGNFOCUS does not play a role, but it will be crucial in the evaluations provided by Tableaux 6 and 8 below. When both the DP and the predicative phrase are part of the focus of the clause, it is again the DP that must be moved into SpecIP. This is shown in Tableau 3; as in Tableau 1, the constraint EPP(case) is decisive. For obvious reasons, movement of the subject DP is also forced when only the predicative phrase is part of the focus of the clause; for completeness’ sake, this is shown by the evaluation in Tableau 4. Tableau 3: DP & Pred ⊂ focus

example (18a) DP I ... V tDP Pred Pred I ... V DP tPred e I ... V DP Pred

EPP(ij)

AF

*!

** ** *

EPP(ij)

AF

)

*MOVE * *

EPP(case)

*MOVE * *

EPP(case)

*! *

Tableau 4: DP ⊄ focus; Pred ⊂ focus

example (18a) DP I ... V tDP Pred Pred I ... V DP tPred e I ... V DP Pred

) *!* *!

* *

274 Chapter 5: Predicate movement

The discussion above has shown that the English ranking in (17) solves the problem that we formulated in (14b) without further ado. It straightforwardly accounts for the fact that movement of the subject into SpecIP is blocked only when (i) the subject is part of the focus of the clause, and (ii) a predicative phrase is present. From the proposal it further follows that a third condition must be satisfied, namely that the predicative PP is part of the presupposition of the clause: when the PP is also part of the focus of the clause, it is again the subject that must move (cf. Tableau 3). As was suggested above, there still may be a problem with the first candidate in Tableau 2, which seems acceptable despite the fact that it is not selected as the optimal candidate. This suggests that there is still something missing in the present analysis. This may be related to the fact that so far I have ignored Den Dikken & Næss’ (1993) claim that locative inversion is obligatorily followed by topicalization of the fronted predicative PP. There are various ways to implement this claim. One could, for example, postulate some constraint outranking EPP(ij) that disfavors the occurrence of non-nominal material in SpecIP (cf. Stowell’s 1981 Case Resistance Principle); as a result of this, SpecIP would only be available as an intermediate landing site of a topicalized (or at least A′-moved) predicative PP, which implies that if topicalization of the predicative PP is excluded for some reason, locative inversion will be blocked as well. The result of this is that there is actually a wider range of constructions available to the speaker than suggested by the evaluations discussion above: the problematic candidate in Tableau 2 with the non-presuppositional subject in subject position and the presuppositional predicative PP in postverbal position will be selected as the optimal one when topicalization of the predicative PP does not apply. For simplicity’s sake, I will henceforth ignore considerations related to the topichood of the predicative PP until I return to the issue in section 5.2.5.2. 5.2.3.

Fine-tuning the analysis

The analysis of locative inversion given in the previous section was slightly simplified by ignoring the fact that the strong ranking of EPP(ij) in (17) requires that locative inversion be preceded by short “object shift” of the subject of the small clause into the local domain of V. This obligatory movement has consequences for the analysis of locative inversion that will be spelled out in more detail in this section. Consider again the relevant examples, repeated here as (19). (19)

a. b.

The baby carriage rolled down the hill. Down the hill rolled the baby carriage.

Locative inversion in English 275

I will start by discussing the regular construction in (19a). The earlier discussion assumed that the subject of the small clause moves into SpecIP in one fell swoop, as indicated in (11a): [IPDPI[VP V [SC tDP Pred]]]. However, the strong ranking of EPP(ij) forces the subject of the small clause to move successive cyclically via the local domain of V, so that we need the additional movement in (20b). After the merger of v and I in (20c), the DP can move directly into SpecIP: the unaccusative light verb v does not assign case and the unaccusative vP does not constitute a phase, so that the DP need not (in fact, cannot) move via the specifier of vP. (20)

a. b. c. d.

[VP V [SC DP Pred]] short “object shift” [VP DP V [SC tDP Pred]] I [vP v+V [VP DP tV [SC tDP Pred]]] merging of v, V-to-v, and merger of I DP movement [IP DP I [vP v+V [VP tDP tV [SC tDP Pred]]]]

Since the three steps in (20a-c) are also part of the derivation of the locative inversion construction in (19b), it is clear that there are in principle two ways in which the surface order in (19b) could be derived from the structure in (20c). This is illustrated in (21): (21a) is derived by moving only the predicate of the small clause into SpecIP, whereas (21b) is derived by moving the complete small clause into this position.58 (21)

a. b.

[IP Pred I [vP v+V [VP DP tV [SC tDP tPred]]]] [IP [SC tDP Pred] I [vP v+V [VP DP tV tSC]]]

movement of Pred movement of full SC

The structure in (21a) remains closest to the one adopted by H&M. However, given that the predicate heads the small clause, it cannot be ruled out a priori that attraction of the features of the predicate requires pied piping of the remainder of the small clause, as in (21b). The following subsections will argue that there are actually reasons to assume that the structure in (21a) is not well-formed, and that the derivation must proceed as indicated in (21b). Given that Den Dikken (2006) argues in favor of a slightly more sophisticated version of (21a), his proposal will also be briefly discussed in this section. 5.2.3.1. The Minimal Link Condition One reason to reject the derivation in (21a) is that it is not in accordance with the locality theory we have assumed so far, according to which the 58 There is yet another plausible derivation of locative inversion, in which the DP remains in its base position and the predicate moves into the local domain of V before moving into SpecIP: [IP Pred I [vP v+V [VP t′Pred tV [SC DP tPred]]]]. In section 5.2.3.1, example (27), I will discuss a more sophisticated version of this derivation.

276 Chapter 5: Predicate movement

ij-features on I must attract the closest potential goal, where closeness is defined as in (22); cf. section 3.1.2.4. (22)

γ is closer to probe P than β in [… P [… γ … β]], iff: (a) γ c-commands β, (b) γ and β are not in the same minimal domain, and (c) γ is active.

Consider again the structure in (20c). According to (22), the DP in SpecVP is closer to I than the predicate of the small clause given that the latter is not in the minimal domain of (the trace of) V. Consequently, (21a) is ungrammatical, as movement of Pred across the DP into SpecIP violates the Minimal Link Condition (MLC) on movement. Example (21b), on the other hand, is grammatical, as the DP and the remnant of the small clause are both in the minimal domain of V and, consequently, equidistant from I. The present proposal therefore only allows movement of the remnant of the SC. In this respect it differs sharply from the proposal in Den Dikken (2006), which maintains that it is the predicate and not the full small clause that undergoes movement in the locative inversion construction. I will therefore briefly discuss this proposal. Den Dikken (2006:115ff.) assumes a slightly more layered structure of the small clause. He assumes that the subject and the predicate of the small clause are dependents of a functional head of the RELATOR (REL) type. The basic structure of small clauses is therefore as given in (23). (23)

[RP SUBJECT [REL [PREDICATE]]]

Den Dikken further argues that extraction of the predicate from the RELATORP (RP) is prohibited on two counts. First, assuming the original formulation of equidistance from Chomsky (1995a: ch.3), it will violate the MLC.59 Secondly, on the assumption that RP constitutes a phase, extraction 59

This actually does not follow from the definition of closeness adopted in Den Dikken (2006:114): ȕ is closer to K than Į unless ȕ is in the same minimal domain as (a) the target position IJ or (b) Į (Chomsky 1995a:178). Since the subject and the predicate are both in the minimal domain of the RELATOR, they are equidistant from any probe external to RP according to clause (b), which is essentially equivalent to (22b) in the main text. Consequently, movement of the predicate across its subject DP should be admissible as far as the MLC is concerned. What is needed therefore is rather Chomsky’s (1995a: ch.3) original formulation of the notion of equidistance; Den Dikken (p.c.) confirms that this is indeed what was intended. But perhaps this is all immaterial: Den Dikken’s appeal to the PIC may be sufficient to block the relevant movement, and therefore it may not be needed to take recourse to the MLC at all.

Locative inversion in English 277

of the predicate will also violate Chomsky’s (2001) Phase Impenetrability Condition (PIC); since the predicate is not at the edge of RP, the PIC excludes extraction of the predicate from this phase. Den Dikken solves the problem with the MLC by assuming that predicate inversion involves movement of the RELATOR to a so-called LINKER F (which may but need not be a functional head). After this headmovement, the subject and the predicate are in the minimal domain of the chain (RELATOR, t), so the predicate may move across the subject into the specifier of FP without violating the MLC.60 (24)

[FP PREDICATE F+ REL [RP SUBJECT [tREL tPRED ]]]

The problem with the PIC is also solved by postulating that the movement of the RELATOR to the LINKER F extends the phase to FP: “Movement of the head of a phase to a higher F extends the phase to FP” (p.115). Den Dikken represents this as in (25), in which ĭ indicates the phase boundary: in the small clause structure in (25a), RP constitutes a phase by definition; the movement of the RELATOR to the LINKER F in (25b) extends this phase to FP; the MLC now allows movement of the predicate into SpecFP, that is, into the new edge of the phase, from which it is accessible for probes external to FP.61 (25)

a. b. c.

[RP SUBJECT [REL [PREDICATE]]] ĭ [FP F+ REL [RP SUBJECT [tREL PREDICATE]]] ĭ ( ) [FP PREDICATE F+ REL [RP SUBJECT [tREL tPRED ]]] ĭ

60 Den Dikken (2006: section 4.3.2.2) assumes that there is a second way of avoiding a violation of the MLC, namely by moving head H of the predicate to the RELATOR: “Movement of the head H of a phrase HP embedded in a phase ĭ to the head of a phase makes both H and its maximal projection visible to probes outside the phase” (p.115). This proposal is put to use in Den Dikken’s analysis of (i) socalled “beheaded’ locative inversion constructions that can be found in Norwegian and (ii) Dative Shift. I have little to say about this proposal, although it might be interesting to point out that this proposal has the consequence that v*P is never a barrier for the complement of V in languages with V-to-v movement (like English and the Scandinavian languages). This will have wide ramifications for phase theory, which deserve further investigation. 61 Note that this implies that verb-second will extend the v*P-phase to IP or even CP, which will again have wide ramifications for phase theory, but in this case the proposal has a predecessor in Broekhuis (1992) who has argued something along similar lines within the Barriers framework by claiming that V-to-I enables I to l-mark, and thus to void the barrierhood of VP.

278 Chapter 5: Predicate movement

From this it follows that it is the predicate in isolation and not the small clause as a whole that is moved in the locative inversion construction. I will illustrate this here by assuming that the movement into SpecIP is preceded by “short object shift”, but nothing much would change if one assumes (as Den Dikken does) that the movement applies in one fell swoop. First assume that V takes the structure in (25a) as its complement, as in (26a). Movement of the predicate into the local domain of V is excluded because this would violate both (i) the MLC (in Chomsky’s orginal formulation) as the subject is closer to V than the predicate, and (ii) the PIC. The only option is therefore to move the subject of the small clause, as in (26b). After the merger of v, V-to-v and the merger of I, the structure is as indicated in (26c). Movement of the predicate into SpecIP is again excluded by the MLC and the PIC, so that moving the subject, as in (26d), is again the only option. (26)

a. b. c. d.

V [RP SUBJECT [REL [PREDICATE]]] [SUBJECT V [RP tSUBJECT [REL [PREDICATE]]]] I ... v+V ... [SUBJECT tV [RP tSUBJECT [REL [PREDICATE]]]] [SUBJECT I ... v+V ... [t′SUBJECT tV [RP tSUBJECT [REL [PREDICATE]]]]]

Now assume that V takes (25c) as its complement, as in (27a). Now it is the movement of the subject that gives rise to a violation of the MLC and the PIC, so that it is the predicate that must be moved into the local domain of V, as in (27b), and ultimately into SpecIP. Movement of the entire small clause must of course be excluded in one way or another since this would wrongly predict that the subject is pied piped. (27)

a. b. c. d.

V [FP PRED F+ REL [RP SUBJECT [tREL tPRED ]]] [PRED V [FP t′PRED F+ REL [RP SUBJECT [tREL tPRED ]]]] I ... v+V ... [PRED V [FP t′PRED F+ REL [RP SUBJECT [tREL tPRED]]]] [PRED I ... v+V ... [t′′PRED V [FP t′PRED F+ REL [RP SUBJECT [tREL tPRED]]]]]

We will see in the following subsection, however, that there are empirical reasons to reject the claim that the small clause predicate can be moved in isolation. From this I will conclude that small clause-internal subject-predicate inversion is not necessary to make locative inversion possible. Note that the argument below does not show that the small clauses structure proposed by Den Dikken is wrong: I will therefore remain agnostic in this respect. 5.2.3.2. Resultative constructions This subsection will show that resultative constructions provide empirical evidence in support of the claim that movement of the small clause

Locative inversion in English 279

predicate in isolation is excluded. Since we have seen that the ij-features on V may attract the subject of the small clause, hypothesis (5) implies that they can also attract the small clause itself; cf. (8b) above. The resultative constructions in (28) suggest that this prediction is indeed correct.62 (28)

a. *The jeweler put carefully the diamonds into the box. b. The jeweler put the diamonds carefully into the box. c. The jeweler put the diamonds into the box carefully. d. *The jeweler put into the box carefully the diamonds.

Example (28a) is of course unacceptable because the strong ranking of EPP(ij) requires that the goal of the ij-features on V be moved into the local domain of V: on our earlier assumptions, this requirement is not met since both the DP the diamonds and the predicate into the box are in their base position following the VP-adverb carefully, as is shown in (29a). The requirement can be satisfied by movement of the DP in (28b), with the representation in (29b). (29)

a. *[vP S v+V [adv tV [SC DP Pred]]] b. [vP S v+V [VP DP adv tV [SC tDP Pred]]]

Now consider the examples in (28c&d), which show that the predicate can also precede the VP-adverb, but not when the DP is stranded. Given that inversion of the subject of the small clause and the predicate is possible in the locative inversion construction, this seems somewhat surprising at first sight. However, the contrast between (28c) and (28d) follows immediately when we assume that movement of the predicate obligatorily pied pipes the remainder of the small clause, as in (29c); independent movement of the predicate, as in (29d), is excluded. (29)

c. [vP S v+V [VP [SC DP Pred] adv tV tSC]] d. *[vP S v+V [VP Pred adv tV [SC DP tPred]]]

The ungrammaticality of (28d) does not follow from Den Dikken’s proposal. Given that the verb can take the structure in (25c) as its complement, we can derive the impossible order without violating the MLC (in Chomsky’s original formulation) or the PIC. The resulting structure would be as given in (30), and the derivation differs only from the one 62 Note that manner adverbs like carefully may also precede the main verb: The jeweler carefully put the diamonds into the box. Since I don’t have anything interesting to say about this fact, I will restrict my attention to manner adverbs in postverbal position.

280 Chapter 5: Predicate movement

indicated in (27a-c) in that the light verb v* introduces an external argument. In order to account for the ungrammaticality of the word order in (28d), we must therefore reject the small clause-internal subject-predicate inversion postulated by Den Dikken in (25b-c). (30)

[S v*+V [VP PRED adv tV [FP t′PRED F+ REL [RP DP [tREL tPRED ]]]]]

In my view, the rejection of small clause-internal subject-predicate inversion is a welcome result given that the movements in (25b) and (25c) seem to violate the Last Resort Condition on movement; at least, I am not aware of any independently motivated unvalued formal features that may trigger these movements. Now, let us consider the consequences for the analysis of locative inversion. The conclusion that clause-internal subject-predicate inversion does not exist also excludes a derivation of locative inversion along the lines sketched earlier in footnote 58, according to which the DP remains in its base position and the predicate is moved into SpecIP via the local domain of V. If we adopt Den Dikken’s small clause structure in (23), the derivation would be as indicated in (31). (31)

a. b. c.

V [RP SUBJECT [REL PRED]] [VP PRED V [RP SUBJECT [REL tPRED]]] [IP PRED I [vP v+V [VP t′PRED tV [RP SUBJECT [REL tPRED]]]]]

Since SUBJECT and PRED are base-generated in the minimal domain of REL, the movement of PRED in (31b) is of course allowed by the version of the MLC adopted in this study. However, given that I rejected the small clauseinternal subject-predicate inversion in (25), movement of PRED across the RP-boundary will be excluded by the PIC under Den Dikken’s (plausible) assumption that RP constitutes a phase; cf. Van Riemsdijk’s (1978) claim that the maximal projections of all categories constitute bounding nodes for movement. Since PRED cannot be extracted from RP, movement of it must pied pipe all RP-internal elements, including the subject. From this we must conclude that locative inversion can only be derived by first extracting the subject from RP by means of short “object shift” followed by remnant movement of RP into SpecIP, as indicated in (21b) above. Let me conclude this section with a discussion of the following question raised by Den Dikken (p.c.): Can the order in (28c) be the result of VP-movement across the manner adverb? Although I argue elsewhere that certain instances of VP-movement are possible, I believe the answer to be negative. VP-movement must be triggered by some unvalued feature on

Locative inversion in English 281

some higher head external to VP. However, given that (i) the strong ranking of EPP(ij) forces the application of short object shift, (ii) NOVACM requires short object shift to be made visible by crossing the VP-adverbs, and (iii) short object shift must take place before a new head can be merged, the surface order of the object and the adverbs is already fixed at the time that VP-movement may come into play. Next to this theory-internal reason to answer Den Dikken’s question in the negative, there is also empirical evidence that indicates that VP-movement is not able to cross VP-adverbs. Consider the examples in (32). (32)

a. b.

dat Jan de diamanten in de kist legt . that Jan carefully the diamonds in the box put dat Jan de diamanten in de kist legt . that Jan with care the diamonds in the box put

Recall our earlier assumption from section 2.4.2.5 that Extraposition is the overt reflex of VP-movement. If this assumption is indeed correct, the fact that the manner adverbs voorzichtig ‘carefully’ and met zorg ‘with care’ cannot be “extraposed” in languages like Dutch, which do allow extraposition of higher adverbs, shows that VP-movement across VP-adverbs is not an available option. I therefore conclude that the word order in (28c) must be derived as indicated in (29c). 5.2.3.3. Copular inversion constructions The claim that movement of the predicate must pied pipe the remainder of the small clause can also be motivated on the basis of predicate inversion in the complement of the verb to consider. The examples in (33b&c) show that this inversion is possible only when the copular verb to be is present. I will start by discussing my own proposal, which is then followed by a discussion of Den Dikken’s (2006) analysis. (33)

a. I consider John (to be) the best candidate. b. *I consider the best candidate John. c. I consider the best candidate to be John.

Under our current proposal the ungrammaticality of (33b) follows from the fact that the word order in this example can only be derived by movement of the small clause predicate in isolation into the local domain of the verbal root consider, as in (34b): if this step is illicit, we will never arrive at the structures in (34c&d).

282 Chapter 5: Predicate movement (34)

a. [VP consider [SC DP Pred]] b. *[VP Pred consider [SC DP tPred]] c. *[vP S v+consider [Pred tV [SC DP tPred]]] d. *[IP S I [vP tS v+consider [Pred tV [SC DP tPred]]]]

The word order of (33c), on the other hand, will be obtained by the derivation in (35). The verbal root be provides a landing site for the subject of the small clause, as in (35b), after which the complete small clause may move across the subject into the subject position of the infinitival clause, which results in predicate inversion. This is shown in (35d): movement of the small clause across the raised small clause subject is allowed because they are both part of the minimal domain of (the trace of) the verbal root be, and are thus equidistant from I according to our definition in (22). Subsequently, the verbal root V of the matrix clause is added and the small clause must now move into V’s local domain in order to satisfy EPP(ij), as in (35e&f). The steps in (35g&h) complete the derivation.63 (35)

a. b. c. d. e. f. g. h.

[VP be [SC DP Pred]] [VP DP be [SC tDP Pred]] [vP v+be [VP DP tbe [SC tDP Pred]]] [IP [SC tDP Pred] I [vP v+be [VP DP tbe tSC]]] [VP consider [IP [SC tDP Pred] I [vP v+be [VP DP tbe tSC]]]] [VP [SC tDP Pred] consider [IP t′SC I [vP v+be [VP DP tbe tSC]]]] [vP S v+consider [VP [SC tDP Pred] tV [IP t′SC I [vP v+be [VP DP tbe tSC]]]]] [IP S I [vP tS v+consider [VP [SC tDP Pred] tV [IP t′SC I [vP v+be [VP DP tbe tSC]]]]]]

Since the small clause and its DP subject are no longer in the same minimal domain in (35e), the definition of closeness in (22) also predicts that movement of the latter across the small clause in SpecIP, as in (35f′), will result in a violation of the MLC. This accounts for the fact that examples like (36a) and (36b) are ungrammatical as they both involve (35f′) as an intermediate step. (35)

e. [VP consider [IP [SC tDP Pred] I [vP v+be [VP DP tbe tSC]]]] f′. *[VP DP consider [IP [SC tDP Pred] I [vP v+be [VP t′DP tbe tSC]]]]

63 The motivation given in the main text for the presence of the copula is similar to the one given in Moro (1997) in that it provides additional landing sites that enable locative inversion. The two proposals differ in that in my proposal the additional landing sites facilitate movement of the subject of the small clause followed by remnant movement of the small clause, whereas in Moro’s proposal it is only the predicate of the small clause that moves.

Locative inversion in English 283 (36)

a. *I consider John the best candidate to be. b. *John is considered the best candidate to be.

The set of facts in (33) and (36) were put forward by Den Dikken as crucial evidence in favor of his analysis, but the discussion above shows that we can straightforwardly derive them by calling upon the same assumptions that underlie our earlier account of locative inversion. In order to show that the two proposals make the same predictions, I will briefly review Den Dikken’s (2006) analysis. Consider again the examples in (33b&c). Den Dikken claims that the absence or presence of the copula be is related to the application or non-application of movement of the RELATOR to the postulated LINKER F. Consider the structures in (37): the absence of the copular in (37a) signals that the RELATOR remains in place, so that movement of the predicate violates the MLC and PIC; the presence of be in (37b) signals that the RELATOR did move to F, so that movement of the predicate is now in accordance with the MLC and the PIC. (37)

a. *I consider [FP the best candidate F [RP John REL tthe best candidate]] b. I consider [FP the best candidate REL+F (= be) [RP John tREL tthe best candidate]]

The ungrammaticality of the examples in (36) also follows from Den Dikken’s proposal. Since movement of the RELATOR to the LINKER F results in phase extension, the subject of the small clause is no longer situated at the edge of the (extended) phase FP in (38). As a result the subject cannot be attracted by the probe, and is therefore frozen in place (unless the DP can be moved via an edge position, which Den Dikken, 2006:123, excludes by stipulation). (38)

Probe … [FP the best candidate REL+F (= be) [RP John tREL tthebest candidate]]

So far the two proposals seem to make the same predictions. A potential problem for our new proposal is that it cannot account for the fact that the passive counterpart of (33c), shown in (39a), also requires that the copula be be present — the ungrammatical be-less form can be derived as indicated in (39b): since the small clause subject and the SC are both in the domain of (the trace of) the verbal root consider, they are equidistant to any probe external to VP, so that the step that leads to (39b′′′) is licit.

284 Chapter 5: Predicate movement (39)

a. b. b′. b′′. b′′′.

The best candidate is considered *(to be) John. [VP consider [SC DP Pred]] [VP DP consider [SC tDP Pred]] [vP v+consider [DP tV [SC tDP Pred]]] [IP [SC tDP Pred] I [vP v+consider [DP tV tSC ]]]

In this respect Den Dikken’s proposal fares better (see his discussion in section 4.5.1.1), but this is immediately counterbalanced by the fact that the acceptability of examples like The best solution remains/becomes instant retreat follows from the present proposal without further ado, whereas Den Dikken (2006: section 4.5.1.2) has to postulate that the RELATOR must move into the head of an aspectual phrase that acts as an inaudible LINKER, so that there is no overt reflex of this movement.64 It therefore seems that the two proposals are in a tie in this respect. The discussion above suggests that the two proposals fare equally well in the description of the data discussed here. The fact that Den Dikken’s proposal requires the introduction of a new assumption, namely the notion of phase extension, suggests that our proposal is to be preferred unless the need for the notion of phase extension can be independently motivated. The following subsection argues that this is probably not the case. 5.2.3.4. Copular inversion and agreement Den Dikken (2006:117) shows that his proposal can readily explain the puzzling agreement facts in the copular inversion construction in (40), in which the copular agrees with the predicate of the small clause and the subject of the small clause is assigned accusative case.

64 Something similar holds for resultative constructions like if Bill has an alibi for 6 p.m., that makes the murderer John, for which Den Dikken postulates that the verb make selects an AspP, in which the aspectual head functions as an inaudible LINKER. In this case, Den Dikken indirectly motivates the aspectual head by assuming that it can be filled with or spelled out as a particle, in which case raising of the RELATOR will be blocked. This accounts nicely for the fact that, when a particle is present, predicate inversion is only possible with an overt copula, that is, when there is an additional LINKER: They made the murderer out *(to be) John. Note that the first example is also a problem for my proposal: I provisionally assume that make is a composite verb consisting of a causative root CAUSE and the copular be: [.. v [.. CAUSE [.. BE [sc John the murderer]]] with CAUSE and BE both endowed with unvalued ij-features: BE can then attract the subject John, thus enabling remnant movement of the small clause into the local domain of the causative root. The verb in the make out construction, on the other hand, is a noncausative, hence non-composite verb, so that an additional copular is needed to license predicate inversion.

Locative inversion in English 285 (40)

a. b.

The biggest problem is/*are the children. The best candidate is me/her/him.

This follows straightforwardly from Den Dikken’s claim that raising of the RELATOR to F results in phase extension. First consider the intermediate representation in (41a) of example (40a) after predicate inversion. In this structure, the small clause subject is no longer at the edge of the phase FP, and is therefore invisible for the unvalued ij-features on the inflectional node I. As a result, the ij-features on I can only enter into an Agree relation with the inverted predicate, which is reflected by the fact that the verb in (40a) agrees with the singular predicate the biggest problem and not with the plural postverbal subject. For similar reasons, the postverbal subject pronouns in (40b) cannot be the goal of the unvalued case-features on I, as a result of which they surface with default accusative case. (41)

a. b.

I .. [FP the biggest problem REL+F (= be) [RP the children tREL tthe biggest problem]] I .. [FP the best candidate REL+F (= be) [RP me/her/him tREL tthe best candidate]]

At first sight, this seems to provide strong evidence in favor of Den Dikken’s notion of phase extension, especially since this effect cannot be replicated in my proposal by appealing to locality theory. Following our earlier assumption the derivation proceeds as in (42). (42)

a. b. c. d. e. e′.

[VP be [SC DP Pred]] [VP DP be [SC tDP Pred]] [vP v+be [VP DP tbe [SC tDP Pred]]] I … [vP v+be [VP DP tbe [SC tDP Pred]]] [IP DP I … [vP v+be [VP tDP tbe [SC tDP Pred]]]] [IP [SC tDP Pred] I … [vP v+be [VP DP tbe tSC]]]

Since in (42d) the DP and the small clause are part of the minimal domain of the trace of the verbal root be, they are equidistant from I, which can therefore enter into an agreement relation with both/either the DP and/or the predicate: this is also clear from the fact that we can derive either the regular order DP be Pred in (42e) or the inverted order Pred be DP in (42e′) from this structure. However, Den Dikken (2006:118) himself offers us a way out by introducing the notion of promiscuity, which states that “whenever T [in our terms: the inflectional head I] can establish all of its feature agreement relationships with one and the same constituent, it must do so” (emphasis given by Den Dikken); cf. also Chomsky’s (2001:15) principle “Maximize

286 Chapter 5: Predicate movement

Matching Effects”. Given that I must enter into an agreement relation with the predicate in order to trigger predicate inversion, the agreement facts in (40) immediately follow from promiscuity without there being any need to appeal to the notion of phase extension. Consequently, (40) does not provide independent evidence for this notion. To conclude this section I want to note that Den Dikken (2006:280, fn.43) correctly points out that Dutch and English behave crucially different in this respect: the Dutch examples in (43) show that the subject of the small clause determines the agreement on the copular and must be assigned nominative case. This suggests that Dutch differs from English in not having the copular inversion construction, so that the examples in (43) are derived by Topicalization. I perfectly agree with that and refer the reader to section 5.3.1 for a similar conclusion about locative inversion. (43)

a. b.

Het grootste probleem zijn/*is de kinderen. The biggest problem are/is the children De beste kandidaat ben ik/ben jij/is hij. the best candidate am I/are you/is he

5.2.3.5. Predicate inversion and A′-movement of the postverbal subject The examples in (36) have shown that the postverbal subject cannot be Amoved across the inverted predicate. The examples in (44) show that the same holds for A′-movement (there are some potentially problematic data; cf. Den Dikken, 2006:281, fn.47, for discussion). I will not discuss the reason for this at great length, but simply follow Den Dikken (2006: section 4.4.3) in assuming that this is related to focus. (44)

a. *Which boy did this girl say that her greatest love was twhich boy? b. *Which horse did out of the barn run twhich horse?

Den Dikken (2006:125) correctly observes that the subject of a predicate inversion construction is interpreted as a focus, and he claims that as a result of this, it is frozen in place. First consider an example like *Brian, Imogen thinks the best candidate is, in which the (inverted) subject Brian is topicalized. Den Dikken suggests that this example can be excluded as a pragmatic anomaly: the subject Brian cannot simultaneously be construed as a focus of the embedded clause and as a topic of the main clause. A problem with this proposal is, however, that wh-phrases are normally considered to constitute the focus of the clause. Den Dikken himself actually appeals to this in his discussion of the unacceptability of *How good a candidate does Imogen think is Brian. He claims that this example is unacceptable because there can only be a single focus per clause. Since the subject Brian must be

Locative inversion in English 287

a focus in order to trigger inversion, the wh-phrase how good a candidate cannot be a focus, which results in an anomalous reading. If wh-phrases are indeed a focus, examples like (44) have a perfectly coherent informationalstructural make up, and consequently some other account is called for. Although I agree perfectly with Den Dikken that the impossibility of A′-movement of the subject is related to focus, we need not exclude it as a pragmatic anomaly because it follows straightforwardly from our proposal outlined above. Recall that predicate inversion is mainly motivated by ALIGNFOCUS in that it enables the subject of the small clause to be in the right periphery of the clause when it is part of the focus of the clause. Now if the postverbal subject is, for example, wh-moved into a position in the left periphery, ALIGNFOCUS will be violated irrespective of whether predicate inversion does or does not apply; hence predicate inversion will serve no purpose and consequently will be blocked by the economy condition *MOVE. This accounts not only for the fact that wh-movement across an inverted predicate is impossible, but also for the fact that an insitu wh-phrase is fully compatible with locative inversion (cf. Coopmans 1989 and Den Dikken 2006:131). (45)

a. b.

Which girl said that her greatest love was which boy? Out of which barn ran which horse?

5.2.3.6. Conclusion This section has argued that extraction of the predicate from a small clause is excluded, so that predicate inversion always involves movement of a complete small clause. Given that Den Dikken’s (2006) analyses of locative and copular inversion crucially adopt the assumption that extraction of the predicate part of the small clause is possible, I have taken some of the core data that he has put forward in favor of his proposal and showed that these can be handled equally well by assuming that it is the full small clause that is displaced. We have further seen that the ungrammaticality of example (28d) provides empirical evidence for the claim that predicates cannot be extracted from their small clause. Consider again the representations in (21), repeated here as (46). (46)

a. b.

[IP Pred I [vP v+V [VP DP tV [SC tDP tPred]]]] [IP [SC tDP Pred] I [vP v+V [VP DP tV tSC]]]

movement of Pred movement of full SC

I have claimed that the representation in (46a) is excluded due to the fact that extraction of the predicate from the small clause violates the MLC as formulated in (22) (as well as the PIC). Consequently, predicate movement

288 Chapter 5: Predicate movement

must involve movement of the complete small clause, which, in turn, implies that locative inversion is the result of remnant movement, as in (46b): first the DP is moved out of the small clause into the local domain of the verbal root V, and, subsequently, the remnant small clause containing the trace of the DP is moved into SpecIP across the landing site of the DP. The fact that the shifted subject of the small clause does not act as an intervener for the Agree relation between I and the remnant small clause follows from clause (22b): the subject of the small clause and the small clause remnant are both in the minimal domain of (the trace of) the verbal root V, and are therefore equidistant from I, so that no intervention effect is expected to arise. 5.2.4.

Extending the analysis: VP-movement

Given that the subject of the small clause agrees not only with its predicate, but also with the verbal root V (cf. Axiom II), we expect that I can attract not only the small clause but also the projection of the verbal root, VP. There are indeed reasons to assume that this expectation is borne out. Consider the examples in (47), taken from Rochemont & Culicover (1990). (47)

a. b. c.

John walked into the room nude. Into the room walked John nude. Into the room nude walked John.

Example (47a) is the way to express the intended contention in a neutral context, for example, as an answer to the question What happened? The examples in (47b&c), on the other hand, are only possible when the constituents in italics are part of the focus of the clause. Under the assumption that the material preceding the verb occupies SpecIP and therefore constitutes a single phrase, we must conclude that phrases of different size can be preposed: (47b) can be derived by preposing the small clause, while (47c) is derived by preposing a verbal projection that includes VP-adjoined material, in this case the supplementive nude.65 So far we have assumed that short object shift in the Germanic VOlanguages is derived as in (48): the ij-features on the verbal root V probe the object and the strong ranking of EPP(ij) forces movement of the object across the VP-adverb adv into the local domain of V. The inversion of the 65

One of my U.S. informants claims that (47c) is only acceptable when the supplementive nude is set off by pauses: Into the room, nude, walked John. At this moment I cannot tell whether this observation has any bearing on the claim that this example is derived by VP-movement.

Locative inversion in English 289

object and the verbal root that is the result of short object shift is undone by subsequent V-to-v. (48)

a. b. c.

[VP (adv) [V O ]] [VP O (adv) [V tO]] [vP S v+V [VP O (adv) [ tV tO]]]

However, assuming a derivation of this sort makes it hard to derive examples like (47c) by appealing to VP-movement. Since we have argued in the previous section that movement of a small clause predicate obligatorily pied pipes its specifier, we have no alternative than to assume the same in the case of VP-movement. Now, consider again the derivation of short object shift in constructions with a small clause predicate. (49)

a. b. c.

[VP (adv) [V [SC DP Pred]]] [VP DP (adv) [V [SC tDP Pred]]] [vP v+V [VP DP (adv) [tcopula [SC tDP Pred]]]]

Given the assumptions so far, the shifted object is still part of the VP, so that movement of the VP into SpecIP should pied pipe the object and (47c) cannot be derived. This problem can be readily solved, however, since we have adopted the theory of extended projections outlined in section 3.1.2.1; cf. the discussion of structure in (14). According to this proposal, short object shift involves movement not only of the object but also of the verbal root V itself, which leads to the creation of an extended projection of V, the higher VP in (50b). When we assume that the lower VP is a possible goal of the unvalued ij-features of I, we can derive examples like (47c) in the way indicated in (50c&d). (50)

a. b. c. d.

[VP adv [V [SC DP Pred]]] [VP DP V [VP adv [tV [SC tDP Pred]]]] [vP v+V [VP DP tV [VP adv [tV [SC tDP Pred]]]]] [IP [VP adv [tV [SC tDP Pred]]] I [vP v+V [VP DP tV tVP]]]

The choice between the three alternative word orders in (47) can be attributed to the constraint ALIGNFOCUS. First, consider example (47a), in which the subject can be part of the presupposition of the clause. Since movement of the subject into SpecIP satisfies both EPP(ij) and EPP(case), it is preferred to moving the small clause or the VP, since the latter options only satisfy EPP(ij). The evaluation is given in Tableau 5: note that the traces that result from, and the violations of *MOVE that are invoked by obligatory short object shift are not indicated because the candidates do not differ in this respect.

290 Chapter 5: Predicate movement Tableau 5: subject ⊄ focus EPP

e walked John into the room nude John walked tDP into the room nude ) [SC Into the room] walked John tSC nude [VP Into the room nude] walked John tVP

AF

*MOVE

(ij) *!

EPP

(case) * * * *

*! *!

When the subject is part of the focus of the clause, ALIGNFOCUS favors locative inversion. The choice between (47b) and (47c) depends on whetherthe supplementive is part of the focus or of the presupposition of the clause, and is therefore also determined by ALIGNFOCUS. When the supplementive is also part of the focus of the clause, the evaluation is as indicated in Tableau 6 (again, the traces that result from, and the violations of *MOVE that are invoked by obligatory short object shift are not indicated respect). Tableau 6: DP & supplementive ⊂ focus EPP

e walked John into the room nude John walked tDP into the room nude [SC Into the room] walked John tSC nude ) [VP Into the room nude] walked John tVP

(ij) *!

AF

** **!* * **!

*MOVE

EPP

(case) * * * *

* *

The first candidate is unacceptable due to the strong ranking of epp(ij). The second candidate is excluded because the subject is placed too far to the left: it is followed by three constituents and hence induces three violations of the gradient constraint ALIGNFOCUS. The third and the fourth candidate differ in that the former induces only a single violation, whereas the latter induces two violations of ALIGNFOCUS. Movement of the small clause is therefore preferred. Tableau 7 provides the evaluation for the case in which only the subject is part of the focus of the clause. As we can see, it is correctly predicted that in this case the supplementive cannot be stranded, since this would give rise to a violation of ALIGNFOCUS, which means that the complete VP must be moved.

Locative inversion in English 291 Tableau 7: DP ⊂ focus; supplementive ⊄ focus EPP

e walked John into the room nude John walked tDP into the room nude [SC Into the room] walked John tSC nude [VP Into the room nude] walked John tVP )

(ij) *!

AF

** *!** *!

*MOVE

EPP

(case) * * * *

* *

For completeness’ sake, Tableau 8 shows that the subject will move into SpecIP when the predicate is part of the focus of the clause, which shows that the word order in (47a) is compatible with (at least) two readings. Tableau 8: Pred ⊂ focus EPP

e walked John into the room nude John walked tDP into the room nude ) [SC Into the room] walked John tSC nude [VP Into the room nude] walked John tVP

(ij) *!

AF

* * **!* **!*

*MOVE

EPP

(case) * * * *

* *

Finally, note that it is not possible to extract the small clause from the VP and then apply VP-movement to the remnant VP, thus deriving the ungrammatical word order *Nude walked John into the room. This derivation is blocked due to the (inviolable) requirement that the supplementive be c-commanded by the DP at some stage in the derivation in order to make secondary predication possible (cf. Williams 1980). This requires that the DP be extracted from the small clause by short object shift and moved across the supplementive nude, as in (51b), with the result that the small clause must remain VP-internal, and is consequently pied piped by movement of the lower VP in (51d). (51)

5.2.5.

a. b. c. d.

[VP nude [V [SC DP Pred]]] [VP DP V [VP nude [tV [SC tDP Pred]]]] I [vP v+V [VP DP t′V [VP nude [tV [SC tDP Pred]]]]] [IP [VP nude [tV [SC tDP Pred]]] I [vP v+V [VP DP t′V tVP]]]

Conclusion: three brief remarks

Subsections 5.2.3 and 5.2.4 have argued against Hoekstra & Mulder’s (1990) claim that locative inversion is derived by moving the predicate of

292 Chapter 5: Predicate movement

the small clause into SpecIP, which was also adopted in a somewhat different form by Den Dikken (2006), and in favor of the claim that it involves remnant movement of the full small clause or even the most embedded (=lexical) projection of the verbal root V. The latter claim is completely in line with Rochemont & Culicover (1990), who likewise argued that examples like (47c) are the result of VP-movement. The proposed analysis can be readily applied to other constructions that were extensively discussed in Emonds (1976) and Rochemont & Culicover (1990) under the general designation of Preposing around be. That examples like (52a&b) can be analyzed along the lines of locative inversion seems obvious: we are dealing with copular constructions here so that the PP/AP is clearly predicative. When the subject DP is part of the focus of the clause, the examples in (52) receive the evaluation in Tableau 2. However, it seems that the analysis cannot be straightforwardly applied to examples in the present progressive tense like (52c). I will therefore briefly discuss examples like these in section 5.2.5.1. (52)

a. b. c.

At the entrance to the park was an old statue. Happiest to see her was her mother. Accompanying the magazine are cents-off coupons.

I concluded section 5.2.2 with a brief discussion of Den Dikken & Næss’ (1993) claim that locative inversion must be followed by topicalization of the fronted predicate; I will discuss this claim a bit more extensively in section 5.2.5.2. Section 5.2.5.3 concludes this section on predicate movement in English with a brief comment on Den Dikken’s (2006:87) rejection of “exploiting focus as the rationale for Locative Inversion”: since this is precisely what we do by appealing to the constraint ALIGNFOCUS, we must show that his objections to appealing to the informational structural properties of the clause in the analysis of locative inversion are flawed in the sense that they do not apply to the present proposal. 5.2.5.1. Present progressive tense (be V-ing) This section will briefly discuss locative inversion in present progressive tense constructions like (52c), taken from Birner (1995) and repeated below as (53a), and show that it is not possible to straightforwardly extend the proposed analysis to these constructions because of an interfering factor: the problem will be solved by claiming that phrases headed by a present participle (V-ing) contain a designated subject position. Let us start by analyzing (53a) along the lines assumed earlier and see where this leads us. Since the verb phrase accompanying the magazine is predicated of the

Locative inversion in English 293

subject, we might hypothesize that in this case also, the two agree in ij-features, so that, as a result, the ij-features on I can attract either the DP or the VP. If this is so, the structure of (53a) should look like the one given in (53b), where the phrase headed by the present participle is considered the predicate of a small clause, just like the PP and AP predicates in (52). A problem with this representation is that we have seen earlier that the predicate of a small clause cannot be moved independently. Moving the complete small clause, on the other hand, is impossible since this would result in pied piping of the DP cents-off coupons. Hence, in order to remain consistent, we must postulate a functional head F that attracts this DP, which is subsequently followed by remnant movement of the small clause, as in (53c). (53)

a. b. c.

Accompanying the magazine are cents-off coupons. [VP accompanying the magazine] are [SC [DP cents-off coupons] tVP] [SC tDP [VP accompanying the magazine]] are [FP [DP cents-off coupons] F tSC ]

That movement of the subject into SpecFP is needed can be established independently on the basis of the examples in (54), taken from Rochemont & Culicover (1990). Given that the PP in front of her is predicated of the DP Mary, we must conclude that this DP is base-generated as the subject of the VP-internal small clause, as in (54b). In order to derive (54a), the DP Mary must move to some position external to VP, as in (54b′), since otherwise it would be pied pied by the fronting of the VP and end up in between the present participle and the small clause predicate. After movement of Mary into SpecFP, however, the desired order can be derived by moving the VP into the subject position. This is shown in (54b′′). (54)

a. b. b′. b′′.

Standing in front of her was Mary. [VP standing [SC Mary in front of her]] [FP Mary F [VP standing [SC tMary in front of her]]] [VP standing [SC tMary in front of her]] was Mary F tVP

The question that must be raised now is what the nature of the head F is, and what triggers the movement of the DP into its specifier. An answer to the latter question that immediately leaps to mind is that the movement is triggered by the ij-features on the present participle, but this cannot be the correct solution for at least two reasons. First, under the current assumptions, the ij-features on the present participle trigger movement not only of the DP but also of the present participle itself; cf. the discussion of (50).

294 Chapter 5: Predicate movement

The configuration that results from that is therefore as indicated in (55), so that preposing of the present participle would involve movement of the higher VP and therefore cause pied piping of the DP Mary. (55)

[VP Mary standing [VP tstanding [SC tMary in front of her]]]

Secondly, as is illustrated in (56a), present participles are able to assign accusative case. Under standard assumptions this implies that present participles merge with a light verb v, so that the reordering of the present participle and the DP that results from short object shift in (56c) is subsequently made undone by the application of V-to-v in (56d). (56)

a. b. c. d.

John was putting a book onto the shelves. [VP putting [SC a book onto the shelves]] [VP a book putting [VP tputting [SC ta book onto the shelves]]] [vP v+putting [VP a book t′putting [VP tputting [SC ta book onto the shelves]]]]

This suggests that (54a) cannot be derived from (55), repeated below as (57a), by merging of I and movement of the phrase headed by the present participle into SpecIP. Instead, we first have to derive (57b) by merging of v and applying V-to-v. In order to be able to derive (54a), we now first have to extract the DP Mary by merging of some functional head F and movement of Mary into SpecFP, and only then can (54a) be derived by remnant movement of the vP triggered by the ij-features on I. (57)

a. b. c. d.

[VP Mary standing [VP tstanding [SC tMary in front of her]]] [vP v+standing [VP Mary t′standing [VP tstanding [SC tMary in front of her]]]] [FP Mary F [vP v+standing [VP t′Mary t′standing [VP tstanding [SC tMary in front of her]]]]] [IP [vP v+standing [VP t′Mary t′standing [VP tstanding [SC tMary in front of her]]]] [I was] [FP Mary F tvP ]]

The discussion above has established that the movement that extracts the DP from the fronted phrase headed by the present participle is not triggered by the ij-features on the present participle, but must be triggered by some independent functional head F. The question of what feature of F triggers the additional movement of the DP is clearly related to the question of what the proper structure of the phrase headed by the present participle is. What I want to suggest here is that the DP is moved into a position comparable to that of Mary in nominalized present participle phrases like (58a), or in “absolute” present participle constructions like (58b).

Locative inversion in English 295 (58)

a. b.

Mary’s standing in front of her was provocative. He counted on Mary standing in front of her.

If so, SpecFP cannot be defined as the landing site of an internal argument of the verb or a small clause subject, but must be defined instead as the landing site of the DP with the grammatical function of subject. This is needed for independent reasons, because under the former definition we wrongly predict that in examples like (59a&b) there is no difference in the placement of the small clause subjects a man and a book. Under the latter definition, on the other hand, the required distinction can be made: in (59a), the DP a man functions as the subject of the present participle phrase and can therefore be placed in SpecFP and precede the present participle, whereas in (59b) the DP a book does not function as the subject of the present participle phrase (the DP John has this function), and consequently cannot be placed in SpecFP but must remain in its lower position following the present participle; cf. (56). (59)

a. b.

There was standing in front of her. John was < *a book> putting onto the shelves.

That phrases headed by a present participle have a designated subject position (SpecFP) is also shown by the minimal pair in (60), adapted from Hegarty (2005): the angle brackets indicate alternative placements of the DP. There is no reason to assume that the two examples differ with respect to short object shift, but nevertheless the examples differ in placement of the DP: in (60a) it follows the verb be, whereas in (60b) it precedes it. This follows when we adopt the claim that there is a designated subject position in the extended projection of the present participle being but not in that of the past participle been. (60)

a. b.

There has been solved in this room. There was being solved in this room.

The examples in (59) and (60) therefore clearly show that the FP projection that we had to postulate in order to account for locative inversion construction in (54a) is needed for independent reasons. 5.2.5.2. Locative inversion and topicalization In the discussion of locative inversion so far, I have more or less ignored the claim in Den Dikken & Næss (1993) that locative inversion in English and Norwegian is obligatorily followed by topicalization of the fronted predicate. This claim is based on the contrast between the examples in (61),

296 Chapter 5: Predicate movement

taken from Bresnan (1994:108-9): example (61a) is unacceptable because the fronted PP remains in the subject position of the embedded infinitival clause; (61b) is acceptable because the PP is moved into the initial discourse topic position of the main clause. (61)

a. *I expect on this wall to be hung a portrait of our founder. b. On this wall I expect to be hung a portrait of our founder.

Den Dikken (2006:99) further argues that locative inversion exhibits other properties of topicalization; see also Haegeman (2004) for useful discussion. The examples in (62), for example, show that topics do not invert with an auxiliary and that the same holds for preposed locative PPs and auxiliaries; in fact, the preposed PP creates an absolute island in the sense that it never allows movement across it. (62)

a. To Imogen, Brian never gives presents a′. *Does, to Imogen, Brian never give presents? b. On this wall seems to have hung a picture of Brian b′. *Does on this wall seem to have hung a picture of Brian?

Den Dikken (2006:101) further notes, referring to Culicover & Levine (2001), that preposed locative phrases cannot be moved themselves, not even if the movement is very local, as in (63b), in which the trace occupies SpecTP. (63)

a. *Into the room I claim/believe tInto the room walked/will walk Brian b. *This is the city where twhere lives Brian.

Den Dikken concludes from these facts that locative inversion does not involve movement of a predicative PP, but rather involves movement of a phonetically empty “pro-predicate” into the subject position, with the preposed PP “base generated in a topic position strictly local to the landing site of the pro-predicate in SpecTP”. This is shown in the simplified structure in (64), which does not take into account the internal structure of the small clause. (64)

[PPi [TP pro-PREDi T [V [SC DP tpro-PRED]]]]

In order to account for the obligatoriness of movement of the pro-predicate, Den Dikken takes recourse to Rizzi’s (1986) theory of pro-licensing, which states that pro is in need of formal and content licensing; in (64), this is obtained by placing the pro-predicate in a position where it can be formally licensed by T (presumably under spec-head agreement) and contentlicensed by the base-generated PP.

Locative inversion in English 297

Den Dikken leaves the notion “strictly local” used in the quote above undefined. At first sight, it appears to mean something like “adjacent to” or “in the same clause/extended projection as”, but this interpretation runs afoul of examples like (61b), where the base-generated PP-predicate is in the topic position of the main clause, and the pro-predicate is situated in the subject position of the embedded infinitival clause — in order for the suggested analysis to go through, we therefore have to assume that a preposed pro-predicate in the subject position of an infinitival clause is sufficiently local to the PP base-generated in the topic position of the main clause for content licensing, but that this does not hold for a preposed propredicates in the subject position of an embedded finite clause like (63a). Anyway, given that “strictly local” is not that local after all, it is not at all clear whether the base-generation approach is indeed superior to the original movement approach in Den Dikken & Næss (1993). Section 5.2.2 above has shown that locative inversion is possible only if the following two conditions are met: (i) the subject is part of the focus of the clause, and (ii) the PP-predicate is part of the presupposition of the clause. The discussion in this subsection has shown that there are strong empirical reasons to strengthen the latter condition by saying that the PPpredicate must be a discourse topic. Although base-generation might be a viable way of accounting for this condition, the movement approach is to be preferred from our present perspective. First, I have argued earlier that predicate inversion involves movement of the full small clause, not just the small clause predicate, and it is not immediately clear whether this can be made compatible with Den Dikken’s base-generation approach which crucially postulates an empty pro-predicate. Secondly, this chapter will argue later in, e.g., our discussion of Dutch and Bulgarian that there are cases of predicate movement that clearly target an A-position as the final destination of the predicate. The fact that Den Dikken’s base-generation approach cannot be extended to these cases suggests that we must adopt the original movement approach to locative inversion according to which the predicative PP is topicalized via SpecIP. There are various ways to implement the movement analysis (one possibility was already sketched in the conclusion of section 5.2.2), but since developing a full-fledged theory will lead me into empirical domains that I have not yet sufficiently explored, I will leave this for future research. Before I conclude this subsection, I want to point out that I believe that there is good reason to doubt that there is an absolute ban in the locative inversion construction on having (lexical) predicative phrases in the subject position. If expletive there-constructions are also instances of locative inversion, as assumed in earlier studies like Moro (1997), we must

298 Chapter 5: Predicate movement

conclude that examples like (65a) constitute counterexamples to the claim that locative inversion must be followed by topicalization, or, alternatively, involves a structure like (64). Similarly, the examples in (65b&c) show that inversion of there and the auxiliary is allowed in interrogative expletive constructions, and, hence, that there does not occupy a designated topic position superior to the subject position.66 (65)

a. b. c.

We expect [IP there to be several prizes awarded] Does there appear to be support for your hypothesis? Why does there appear to be support for a very high exemption […] of the estate tax?

That the there-construction does not involve topicalization was already recognized in Den Dikken & Næss (1993:fn.6), who accounted for this fact by stipulating that the expletive there, contrary to predictive PPs, can occupy a case position; it therefore need not topicalize and, as a result, can remain in SpecIP. An alternative and, in my view, better explanation might be that the expletive there simply does not have the proper semantic make up for being a topic and can therefore remain in the subject position. Finally, it must be noted that examples like (65) are problematic for Den Dikken’s (2006) theory, since we cannot say that there is similar to the empty pro-predicate in that it must undergo movement in order to get content licensed given that there is no base-generated topic that would be able to do this. Furthermore it seems unlikely that there must be formally licensed, given that it is a lexical element. It therefore seems that the movement of there is purely altruistic: it occurs only in order to allow the subject to remain in its base position (i.e., to avoid a violation of ALIGNFOCUS) by satisfying the EPP, that is, the constraint EPP(ij). This casts serious doubt on the idea that topic-hood of the fronted PP is a necessary condition for locative inversion, and suggests that (at least in some cases) it might be sufficient that the subject (but not the fronted PP) is part of the focus of the clause.

66 Examples (65b&c) were taken from the internet. A Google search has shown that the sequence “… does there …” is very frequent (over one million tokens), and a quick survey of a small set of cases has made clear that the majority of these cases involve the expletive there construction. In passing note that there are differences between there-constructions insofar as wh-movement of the subject across the expletive is involved: those with the copular be do allow it whereas those with other main verbs do not; cf. Hartmann (to appear) for an overview of the data and useful discussion.

Locative inversion in English 299

5.2.5.3. Locative inversion and focus The analysis of locative inversion proposed in this section crucially takes recourse to the constraint ALIGNFOCUS, and is therefore similar to Bresnan’s (1994) in assuming in that locative inversion is dependent on the information structure of the clause. Den Dikken (2006: section 4.2.1.1) takes issue with this and argues that there are serious problems with considering “focus as the rationale for Locative Inversion”. Instead of invoking considerations of information structure, he develops a proposal in section 4.2.2.2 that takes recourse to Rizzi’s theory of pro-licensing; cf. the discussion of (64) above. In this section I will show, however, that Den Dikken’s arguments against the information structural approach do not apply to the D&E analysis developed in this section, which is therefore not affected by them. Den Dikken’s first objection against an information structural approach to locative inversion is that focus cannot be the trigger for the inversion: There is no denying, obviously, that presentational [§ new information] focus plays a role in the analysis of Locative Inversion — but it seems plain that focus on the theme [= small clause subject] is not the trigger for the inversion, precisely because there are other ways to allow the theme to be focussed. (Den Dikken 2006:88; the parts between square brackets are mine)

The D&E analysis proposed earlier is completely in line with Den Dikken by claiming that it is not new information focus (or the information structure of the clause) that triggers locative inversion. Rather, the inversion is triggered by the ij-features on the inflectional head I. One of the problems in Den Dikken’s discussion of this issue is that he incorrectly uses the notions of motive and trigger as synonyms (see the prepenultimate line of his introduction to section 4.2). This is, however, not justified: triggers in the technical sense of the minimalist program refer to unvalued formal features that enter into an agreement relation with a goal and thus make movement of this goal possible, whereas the motive (which can be formalized in the form of a condition, a filter or a constraint ranking) rather determines whether or not this movement actually takes place. Actually, Den Dikken’s own proposal can be viewed in the same way: contrary to what he suggests in section 4.2, the movement of the empty propredicate is not triggered by the licensing conditions on empty categories, but by some unvalued formal feature on I (as is implicitly acknowledged by Den Dikken in his introduction to section 4.3.2 on p.112); the licensing conditions on the empty pro-predicate are not what make locative inversion possible, but what make it obligatory in the light of the fact that failure to apply it would lead to a crashing derivation.

300 Chapter 5: Predicate movement

A second objection to theories that appeal to the information structure of the clause is that: … locative inversion is by no means “the only syntactic means available to meet the functional requirement of presentational focus” [Bresnan 1994]: one may also exploit expletives to keep the theme low and in languages like English it is not even impossible to realize the presentation focus as a subject, by exploiting intonation [..]. (Den Dikken 2006:87-88; the reference between square brackets is added by me)

The proposal put forth in this section does not imply, however, that it is a sufficient condition for locative inversion that the subject is part of the focus of the clause; it is just a necessary one. For example, we have already seen in section 5.2.2 that locative inversion also depends on the status of the predicate: it is possible only when the predicate is part of the presupposition of the clause. Furthermore we have seen that it is readily possible to provide an account for the fact established in section 5.2.5.2 that the predicative PP must be a discourse topic by postulating some constraint that disfavors predicative PPs to occupy SpecIP and outranks ALIGNFOCUS. The most important objection raised by Den Dikken against the informational structural approaches to locative inversion seems to be that they go against the postulate of the autonomy of syntax by requiring “that discourse analysis be prior to syntactic analysis” (Den Dikken 2006:90). However, this objection does not apply to the D&E approach defended here: the informational structural considerations only enter the grammar during the evaluation of the output of the computational system, after application of Den Dikken’s “syntactic analysis. The present proposal is therefore completely in line with the postulate of the autonomy of syntax. The discussion above will have made clear that I do not take Den Dikken’s objections “to be sufficient discouragement for an attempt at deriving the rationale for Locative Inversion (and Predicate Inversion more generally) from information-structural notions such as focus and topic” (Den Dikken 2006:90). On the contrary, I have argued throughout this book that such notions are at the heart of the matter and I believe that we will only obtain a deeper understanding of phenomena involving predicate inversion (and word order in general) by studying more intensively the effects of information structure on the surface realization of clauses. 5.3. Predicate movement in Dutch This section discusses the consequences of hypothesis (5) for Dutch. Section 5.3.1 starts with showing that locative inversion does not apply in Dutch and provides a preliminary account for this. Section 5.3.2 will argue,

Predicate movement in Dutch 301

however, that there is some kind of predicate movement in Dutch as well. On the basis of the fact that this movement has various properties in common with locative inversion, it will be analyzed as movement triggered by the ij-features of the verbal root V. Adopting the analysis in this section will enable us to solve several problems that Dutch poses for Kayne’s (1994) version of the universal base hypothesis, adopted earlier, according to which all languages have the linear order specifier-head-complement (cf. the discussion in section 2.3). 5.3.1.

Locative inversion in Dutch?

The question whether Dutch has locative inversion seems pretty hard to answer. This is due to the verb second nature of Dutch main clauses. Although example (66b) looks similar to locative inversion, it is normally not analyzed as involving movement of the predicative PP into the regular subject position, but as involving the independently motivated process of topicalization; the finite verb is assumed to occupy the C position so that the predicative PP occupies SpecCP, not SpecIP. That the traditional analysis is available is clear from the fact that the order is also possible when the subject is a weak pronoun, which must occupy SpecIP; cf. section 4.2. 67 (66)

a. b. c.

Het jongetje ging naar de speeltuin. the little boy went to the playground Naar de speeltuin ging het jongetje. to the playground went the little boy Naar de speeltuin ging ie. to the playground went heweak

In order to determine whether locative inversion is possible in Dutch, we therefore have to exclude the interference of topicalization. One way of doing that is to restrict our attention to embedded clauses, in which topicalization is not allowed. However, if Den Dikken & Næss (1993) are correct in claiming that locative inversion is obligatorily followed by topicalization, we immediately predict that locative inversion cannot occur in embedded clauses like (67b). 67

My examples in (66)-(70) all contain directional PPs headed by naar ‘to’ because these are compatible with a predicative reading only; locational PPs like in dat het jongetje in het zwembad sprong ‘that the little boy jumped in/into the swimming pool’ can be ambiguous between a predicative and an adverbial reading. Under the adverbial reading such locational PPs can precede the DP arguments in the clause: dat in het zwembad het jongetje sprong ‘that the little boy jumped in/*into the swimming pool’.

302 Chapter 5: Predicate movement (67)

a.

dat het jongetje naar de speeltuin ging. that the little boy to the playground went b. ??dat naar de SPEELtuin het jongetje ging ... that to the playground the little boy went ... (en naar de biblioTHEEK het leergierige meisje). and to the library the studious girl

There are indeed several reasons to assume that locative inversion is not involved in (67b). First, the example is pretty marginal to begin with, and requires a strong, contrastive focus accent on the predicative PP naar de speeltuin, which suggests that we are dealing with focus movement of the PP. This is supported by the fact that inversion of a predicative PP and the DP it is predicated of can also be observed in transitive constructions of the type in (68). Again, (68b) is marginally acceptable at best, and requires that the PP be assigned contrastive focus accent. (68)

a.

dat Jan het jongetje naar de speeltuin bracht. that Jan the little boy to the playground brought b. ??dat Jan naar de SPEELtuin het jongetje bracht ... that Jan to the playground the little boy brought ... (en naar de biblioTHEEK het leergierige meisje). and to the library the studious girl

A final and conclusive argument for rejecting a locative inversion analysis of this example is that the predicative PP need not be adjacent to the complementizer, which unambiguously shows that it does not occupy the regular subject position, SpecIP; example (69), in which the PP follows the adverbial phrase gisteren, has more or less the same status as (67b). (69)

??

dat gisteren naar de SPEELtuin het jongetje ging ... that yesterday to the playground the little boy went ... (en naar de biblioTHEEK het leergierige meisje). and to the library the studious girl

The judgments on the predicate-subject inversion in main clauses with some other constituent in clause-initial position are similar to, or perhaps even worse than those on (67b). This is shown in (70). (70)

*?

Gisteren ging naar de SPEELtuin het jongetje ... yesterday went to the playground the little boy ... (en naar de biblioTHEEK het leergierige meisje). and to the library the studious girl

Predicate movement in Dutch 303

The conclusion we have to draw from the discussion above is that we can only establish in an indirect way whether locative inversion is possible in Dutch. One attempt to do that can be found in Zwart (1992), who argues in favor of locative inversion in Dutch on the basis of existential examples like (71). He claims that under the assumption that expletive er occupies SpecIP, the fact that er may be absent in the existential construction (71c) follows if the predicative PP in de tuin in (71c) has moved into SpecIP, that is, if this is a locative inversion construction.68 (71)

a.

b. c.

dat er veel mensen in de tuin zijn. that there many people in the garden are ‘that there are many people in the garden.’ dat er in de tuin veel mensen zijn. dat in de tuin veel mensen zijn.

There are various reasons to cast a skeptical eye on this argument. First of all, the assumption that the predicative PP in (71c) occupies SpecIP is not compatible with Den Dikken & Næss’ (1993) claim that inverted PPpredicates cannot remain in subject position but must be topicalized; this claim in tandem with Zwart’s proposal wrongly predicts that er-drop is only possible in main clauses with the PP in first position. Of course, one may propose that Dutch locative inversion differs from English locative inversion in that topicalization is not required, but then the examples in (67)-(70) should be accepted as evidence against locative inversion in Dutch. Secondly, (72b) shows that the expletive er can also be dropped when it is followed by, e.g., an adverbial phrase of place or time; the rules that govern er-drop are extremely complex and it is certainly not true, as claimed by Zwart (1992) with reference to a manuscript by Eric Reuland, “that er can remain unexpressed if and only if a locative PP is fronted”. I refer to Haeseryn et al. (1997: section 8.6.3.3) for counterexamples of various types. Note that (72b) also shows that the phenomenon of er-drop can occur in transitive constructions, for which a locative inversion analysis is impossible by definition. 68 Note that the expletive can also be dropped in (71a), but then the existential reading is replaced by a partitive one: “many of the people”. This is crucially not the case in (71c), which in fact disallows this partitive reading. A second argument in favor of locative inversion provided by Zwart is that the subject must remain in its base position when the expletive is present. We have seen in section 4.2, however, that this is a general property of subjects that are part of the focus of the clause, so that Zwart’s observation cannot be used as an argument in favor of locative inversion either.

304 Chapter 5: Predicate movement (72)

a.

b.

dat er in de tuin/na school veel kinderen voetbal spelen. that there in the garden/after school many children soccer play ‘that many children are playing soccer in the garden/after school.’ dat in de tuin/na school veel kinderen voetbal spelen.

Finally, there is strong reason to assume that the PPs in (71) are actually not predicative PPs, but, just like the PPs in (72), adverbial phrases (recall from fn.67 on p.67) that locative PPs like in de tuin ‘in/into the garden’ can be used both as predicates and as adverbial phrases). If true, this entirely rules out an analysis based on locative inversion. The first reason for assuming that the PPs in (71) are adverbial phrases is that in existential constructions with the verb be the PP is often optional: there are many people who believe that ...; cf. Hartmann, to appear, for extensive discussion. Of course, leaving out the PP in examples like (71b) may give rise to a less felicitous result, but this may simply be due to pragmatics: when there is no appropriate context, a sentence like there are many people is simply not informative and thus violates Grice’s (1975) Maxim of Quantity. Another, in my view conclusive, reason for assuming that the PP in (71b) is adverbial is that this explains why the word order in this example is fully acceptable, contrary to the word orders in (67)-(69), which are undeniably derived by leftward movement of a PP-predicate and which are marginally acceptable at best. Since, as far as I can see, there is no compelling reason to assume that the PP in (71c) is different from the one in (71b), we may conclude that the alternation in (71b) and (71c) can receive an account similar to the alternation between the examples in (72), which indisputably contain adverbial PPs. Since there is neither direct nor indirect evidence for locative inversion in Dutch, I will simply apply Ockham’s razor and conclude that it does not occur in Dutch. The next section will show that this also follows from the analysis of another type of predicate inversion that does occur in Dutch: following the methodological guideline that in unclear cases the theory decides (Chomsky 1957:14; Newmeyer 1983: section 2.2.2) therefore leads again to the conclusion that Dutch has no locative inversion. 5.3.2.

Predicate movement in Dutch

According to Kayne’s (1994) version of the universal base hypothesis all languages have the underlying order specifier-head-complement, and therefore also an underlying VO-order. Section 2.4 has shown that we can account for the obligatory OV surface order found in languages like Dutch by assuming the subranking EPP(gender) >> ALIGNFOCUS >> *MOVE. The strong ranking of EPP(gender) forces movement of the object into the local

Predicate movement in Dutch 305

domain of V, and the fact that EPP(gender) outranks ALIGNFOCUS expresses that this movement cannot be blocked by considerations of the information structure of the clause. However, this proposal solves only one of a larger set of problems that Dutch poses for the universal base hypothesis. The following subsections will discuss two of these problems, and show how these can be solved by taking recourse to hypothesis (5). 5.3.2.1. The placement of small clause predicates Small clause predicates constitute a similar problem for Kayne’s universal base hypothesis as objects: they are base-generated in complement position, that is, in a position to the right of the verb, but obligatorily precede the clausefinal verb(s) in Dutch. This is illustrated in (73), where the angle brackets indicate alternative placements of the predicative PP uit bed ‘out of bed’. (73)

a. b.

dat that dat that

de baby viel . the baby out of bed fell Jan de baby duwde . Jan the baby out of bed pushed

In order to account for the placement of the predicate, Zwart (1993) proposed that the clause contains a PredP between V and AGRO, and that the small clause obligatorily moves into the specifier of this projection in Dutch. According to Koster (1994), SpecPredP may contain either predicative phrases or DPs. With respect to DPs, Koster follows De Hoop (1992) in claiming that DPs occupying SpecPredP must be part of a complex predicate: de afwas doen ‘to do the dishes’, een klap geven ‘to give a blow/to hit’, etc. In order to make the proposal in accordance with the Last Resort Condition on movement, Koster claims that “small clause predicates have Case-like N-features that can only be checked by overt movement to [Spec, PredP]”. This section will argue that this proposal is on the right track but will reanalyze it in line with the underlying assumptions discussed in section 5.1. 5.3.2.1.1. The trigger If Kayne’s universal base hypothesis is correct, a proposal along the lines of Zwart (1993;1997) and Koster (1994) seems unavoidable. I therefore adopt the proposal with only two alterations. The first alteration is that I do not assume that the movement of small clauses is triggered by case-like N-features; instead, I assume in accordance with hypothesis (5) that this movement is triggered by the unvalued ij-features on V. An important empirical reason for making this revision is the fact that small clauses must also move in unaccusative constructions like (73a) above; unaccusative

306 Chapter 5: Predicate movement

verbs like vallen are normally assumed not to assign case, and we therefore have no trigger for the obligatory movement of the predicative PP. A more theory-internal reason for making this revision is that I am adopting here our earlier assumption that the case features are situated on the light verb v, and therefore situated too high in the structure. The second alteration is directly related to the previous one: the idea that predicate movement is triggered by the agreement features on V of course voids the need for postulating the functional head Pred. Let us now start by considering what the derivation of a transitive construction like (73b) might look like. Assume that the verbal root V takes a small clause complement as in (74a). Since we have seen that the unvalued ij-features on V can probe either the DP or the small clause predicate, the strong ranking of EPP(gender) requires that either the DP or the small clause move into the local domain of V. Given that the small clause must precede V, apparently only the option of moving the small clause is available, as in (74b). The next step is to add v and the external argument, as in (74c), and from this point onwards the derivation may proceed in the same fashion as regular transitive constructions: the case features on v attract the DP (provided, at least, that the latter is part of the presupposition of the clause), after which I is merged and the subject is moved into SpecIP. Recall that the OV order in Dutch follows from the fact that V-to-v only applies in main clauses; cf. section 2.3. (74)

a. b. c.

[VP V [SC DP Pred]] [VP [SC DP Pred] V [VP tV tSC]] [vP S v [VP [SC DP Pred] V [VP tV tSC]]]

The first two steps in the derivation of the unaccusative example in (73a) are identical to those of the transitive construction in (74a-b), and are given as (75a-b). From this point the two derivations diverge: in the transitive case the vP contains an external argument (cf. (74c)), whereas in the unaccusative case there is no external argument (cf. (75c)). After this stage, the derivation of the unaccusative example is rather straightforward: I is added and the DP moves on to SpecIP in order to satisfy EPP(case) and EPP(person), as in (75d). (75)

a. b. c. d.

[VP V [SC DP Pred]] [VP [SC DP Pred] V [VP tV tSC]]] [vP v [VP [SC DP Pred] V [VP tV tSC]]] [IP DP I [vP v [VP [SC tDP Pred] V [VP tV tSC]]]]

Predicate movement in Dutch 307

Again, the movement of the subject into SpecIP only takes place when the subject is part of the presupposition of the clause, since otherwise ALIGNFOCUS would block this movement. Observe that in (75d), it is the DP that must move into SpecIP and not the entire small clause; this follows from the fact that movement of the DP results in satisfaction of both EPP(case) and EPP(person), whereas movement of the small clause will result in satisfaction of EPP(person) only. 5.3.2.1.2. The constraint A-OVER-A When we compare the first two steps of the derivations of the Dutch (transitive/unaccusative) examples in (74/75) to those of the English (unaccusative) locative inversion construction in (76), we see that the main difference is that whereas the full small clause must be moved in Dutch, in English it is only the DP that is moved. (76)

a. b. c. d.

[VP V [SC DP Pred]] [VP DP V [VP tV [SC tDP Pred]]] [vP v+V [VP DP tV [VP tV [SC tDP Pred]]]] [IP [SC tDP Pred] I [vP v+V [VP DP tV [VP tV tSC ]]]]

In order to account for the difference between Dutch and English we must provide some account for the following generalization. (77)

Observational Generalization: In the structure [VP V [SC DP Pred]], the unvalued ij-features on V attract the small clause in Dutch; in the English locative inversion construction these features attract the DP.

It is not too hard to formulate a constraint that accounts for the attraction of the full small clause in Dutch. I will provisionally assume that there is a constraint A-OVER-A that favors movement of the highest category that can in principle satisfy EPP(ij). Since this constraint reduces the length of the movement path, we may consider it an economy constraint. It is crucial to note that A-OVER-A cannot be an inviolable condition on a par with the MLC; this would wrongly predict that the subject of a small clause can never be attracted by V, so that English locative inversion would be nonexistent. Now assume that in Dutch A-OVER-A is ranked higher than ALIGNFOCUS, as in (78b). (78)

a. b.

A-OVER-A: *[FP X(P) P ... [YP ... tXP ...]] if YP and X(P) are both potential goals of probe P. Dutch ranking: {A-OVER-A, EPP(gender)} >> ALIGNFOCUS >> {EPP(case), EPP(person)} >> *MOVE

308 Chapter 5: Predicate movement

The claim that A-OVER-A outranks ALIGNFOCUS predicts that the gender features on V will attract the small clause, irrespective of the information structure of the clause. This is illustrated by the evaluations in Tableaux 9 to 11. Tableau 9: Object/predicate movement in Dutch

DP & Pred ⊄ focus [VP V [SC DP Pred]] [VP DP V [VP tV [SC tDP Pred]]] [VP [SC DP Pred] V [VP tV tSC]] )

A-

EPP

OVER-A

(gender) *!

AF

*!

*MOVE

** **

Tableau 10: Object/predicate movement in Dutch

DP ⊂ focus; Pred ⊄ focus [VP V [SC DP Pred]] [VP DP V [VP tV [SC tDP Pred]]] [VP [SC DP Pred] V [VP tV tSC]] )

A-

EPP

OVER-A

(gender) *!

*!

AF

*MOVE

* ** **

** **

AF

*MOVE

*

** **

Tableau 11: Object/predicate movement in Dutch

DP ⊄ focus; Pred ⊂ focus [VP V [SC DP Pred]] [VP DP V [VP tV [SC tDP Pred]]] [VP [SC DP Pred] V [VP tV tSC]] )

A-

EPP

OVER-A

(gender) *!

*!

After the derivational stage depicted in the tableaux above is reached, the full small clause cannot undergo any other A-movement (although A′-movement of the small clause of course remains possible). In the case of transitive constructions the reason for this is trivial. The small clause cannot be probed by the case feature of v*, so the small clause subject can be extracted from the small clause in order to satisfy EPP(case) without violating A-OVER-A. The reason for this is different in unaccusative constructions. Given that the unaccusative light verb v has no case feature, it cannot attract the small clause subject, so movement of the small clause into SpecIP will necessarily pied pipe the small clause subject. When the DP is not part of the focus of the clause, movement of the DP is preferred to movement of the full small clause as the former satisfies both EPP(case) and EPP(ij), whereas the latter would only satisfy EPP(ij). When the subject

Predicate movement in Dutch 309

is part of the focus of the clause, movement of the DP is still preferred to movement of the small clause as the latter pied pipes the subject and does therefore not contribute to reducing the number of violations of the higher ranked constraint ALIGNFOCUS. In the English locative inversion construction, satisfaction of the constraint A-OVER-A must be blocked by some higher ranked constraint. The reason why it is the DP that is moved into the local domain of V is that only in this way can the locative inversion construction contribute to satisfying ALIGNFOCUS; when the subject does not leave the small clause, movement of the small clause into SpecIP pied pipes the DP, so that ALIGNFOCUS would still be violated (actually, it would be violated twice instead of once). By ranking ALIGNFOCUS higher than A-OVER-A in English, the desired distinction is derived. Given the ranking EPP(ij) >> {ALIGNFOCUS, *MOVE} >> EPP(case) established earlier (cf. (17)), this leads to the ranking in (79a) or (79b). Since there is at present no empirical reason to prefer one over the other, I will arbitrarily assume that the ranking is as given in (79a), just for concreteness’ sake. (79)

• English ranking: a. EPP(ij) >> {ALIGNFOCUS, *MOVE} >> {A-OVER-A, EPP(case)} b. EPP(ij) >> ALIGNFOCUS >> {*MOVE, A-OVER-A} >> EPP(case)

Now let us consider the predictions that follow from (79a) in more detail. In transitive constructions, ranking (79a) predicts that movement of the full small clause would be blocked by ALIGNFOCUS when the predicative phrase is part of the focus of the clause, so that the DP must move in isolation. In other cases, movement of the small clause is preferred. Of course, since English has obligatory V-to-v, which moves V across the preposed phrase, the difference between the two representations can only be observed by considering the position of the DP and the predicate relative to the VPadverbs. The structures we have to consider are given in (80). (80)

a. *[IP Subj I [vP tSubj v+V [adv [VP tV [SC DP Pred]]]]] b. [IP Subj I [vP tSubj v+V [VP DP tV adv [VP tV [SC tDP Pred]]]]] c. [IP Subj I [vP tSubj v+V [VP [SC DP Pred] tV adv [VP tV tSC]]]]

Now consider the evaluations in Tableaux 12 and 13. For convenience, the tableaux only contain candidates that are not excluded for independent reasons, that is, they only contain candidates with subject shift and V-to-v, and without regular object shift. Furthermore, the tableaux only give those constraint violations that are directly related to short object shift of the DP/small clause into the local domain of V (and the concomitant movement

310 Chapter 5: Predicate movement

of the verbal root V). The evaluations show that (80a) is excluded since it violates the highest ranked constraint EPP(ij), and that both (80b) and (80c) can be realized depending on the information structure of the clause: when the predicate is part of the presupposition of the clause the constraint Aover-A forces movement of the complete small clause; when the predicate is part of the focus of the clause, the subranking ALIGNFOCUS >> A-OVERA forces movement of the subject of the small clause. Note that this does not depend on the status of the subject of the small clause: when it is part of the focus of the clause, we must add the stars within parentheses in the column ALIGNFOCUS, but this does not affect the result of the evaluation. Tableau 12: Transitive predicative constructions in English

Pred ⊄ focus structure (80a) structure (80b) structure (80c) )

EPP(ij)

AF

*MOVE

A-OVER-A

*!

(*) (**) (**)

** **

*!

EPP(case)

Tableau 13: Transitive predicative constructions in English

Pred ⊂ focus structure (80a) structure (80b) ) structure (80c)

EPP(ij)

AF

*MOVE

A-OVER-A

*!

(*) (**) (**)*!

** **

*

EPP(case)

The crucial examples were actually already discussed in section 5.2.3, where it was shown that the difference between examples (28b) and (28c), repeated here as (81a&b), can be accounted for by assuming that the unvalued ij-features on V attract the object in the former case, and the full small clause in the latter. According to my U.S. informants, the semantic difference between the two examples is subtle but goes in the expected direction: (81a) is used in a neutral context (e.g., as an answer to What did the jeweler do?), whereas (81b) is marked in that it is most natural when the adverb (and not the predicative PP) is part of the focus of the clause. (81)

a. a′. b. b′.

The jeweler put the diamonds carefully into the box. ... [VP [DP the diamonds] t′put [carefully [VP tput [SC tDP into the box]]]] The jeweler put the diamonds into the box carefully. ... [VP [SC the diamonds into the box] t′put [carefully [VP tput tSC]]]

When we are dealing with an unaccusative example, the situation is slightly more complex, since then not only the status of the predicate but

Predicate movement in Dutch 311

also the status of the DP becomes relevant. First consider the representations in (82): (82a) results from subsequently moving the DP into the local domain of V and I; (82b) is the result of first moving the small clause into the local domain of V and subsequent movement of the DP into the local domain of I; (82c), finally, is the locative inversion construction, which is derived by first moving the DP into the local domain of V, and subsequent remnant movement of the small clause into the local domain of I (cf. the derivation in (76)). (82)

a. b. c.

[IP DP I [vP v+V [VP t′DP t′V [adv [VP tV [SC tDP Pred]]]]]] [IP DP I [vP v+V [VP [SC tDP Pred] t′V [adv [VP tV tSC]]]]] [IP [SC tDP Pred] I [vP v+V [VP DP t′V [adv [VP tV tSC ]]]]]

The three structures in (82) are the result of differences in the information structure of the clause. This is shown by the evaluations in Tableaux 14 to 17. Again, the tableaux only contain candidates that are not excluded for independent reasons, that is, candidates with subject shift and V-to-v. Furthermore, I ignored the two verb movements that take place in all examples: the indicated violations of *MOVE are therefore due to movement of the DP and/or the small clause. Tableau 14: Unaccusative predicative constructions in English

DP & Pred ⊄ focus Structure (82a) Structure (82b) Structure (82c)

AF

)

*MOVE ** ** **

A-OVER-A *!

EPP(case)

*!

*

Tableau 15: Unaccusative predicative constructions in English

DP ⊄ focus; Pred ⊂ focus Structure (82a) ) Structure (82b) Structure (82c)

AF

*! *!**

*MOVE ** ** **

A-OVER-A *

EPP(case)

*

*

Tableau 16: Unaccusative predicative constructions in English

DP ⊂ focus; Pred ⊄ focus Structure (82a) Structure (82b) Structure (82c) )

AF

*!** *!** *

*MOVE ** ** **

A-OVER-A *

EPP(case)

*

*

312 Chapter 5: Predicate movement Tableau 17: Unaccusative predicative constructions in English

DP & Pred ⊂ focus Structure (82a) Structure (82b) Structure (82c)

AF

)

*** ****! ****!

*MOVE ** ** **

A-OVER-A *

EPP(case)

*

*

The differences between the structures in (82a) and (82b) can only be observed by taking the position of the adverbs into consideration: in the former the predicate follows the VP-adverb, whereas in the latter it precedes it. That these orders both occur is clear from the examples in (83). In (83a) the predicative part of the small clause follows the VP-adverb gracefully, which indicates that the derivation is as given in (83a′); it is the DP that is moved into the local domain of V to satisfy EPP(ij), after which the DP moves on into SpecIP. In (83b), on the other hand, the predicative part of the small clause precedes the VP-adverb. This suggests that the derivation is as given in (83b′); it is the whole small clause that is moved into the local domain of V, after which the DP is extracted from the small clause and moved into SpecIP. As predicted by the ranking in (79a) the difference between (83a) and (83b) is again a matter of information structure: in the former case the small clause predicate is part of the focus of the clause, in the latter it is instead the adverb that is part of the focus of the clause. (83)

a. a′.

b. b′.

The queen walked gracefully down the stairs. [IP The queen I [vP v+walked [VP tDP tV [gracefully [VP tV [SC tDP down the stairs]]]]]] The queen walked down the stairs gracefully. [IP The queen I [vP v+walked [VP [SC tDP down the stairs] tV [gracefully [VP tV tSC]]]]]

As we have seen before, the locative inversion construction is derived by first moving the subject of the small clause into the local domain of V and subsequent remnant movement of the small clause into SpecIP. We simplified the discussion of the locative inversion construction by not taking into consideration the possibility that the VP is moved into SpecIP (see the discussion in section 5.2.4), an option that is clearly available given the acceptability of (84a), taken from Rochemont & Culicover (1990). When we now compare the examples in (84a) and (84b) we see that pied piping of the adverb requires that the predicate follows the adverb.

Predicate movement in Dutch 313 (84)

a. Gracefully down the stairs walked the queen. b. *Down the stairs gracefully walked the queen.

This fact can be used to support our earlier conclusion that locative inversion requires that ij-features on V target the DP, and cannot involve movement of the complete small clause into the local domain of V. The order in (84a) can be derived as in (85a) by first moving the DP into the local domain of V followed by remnant movement of the lower VP with pied piping of the VP-adverb. The ungrammatical order in (84b), on the other hand, simply cannot be derived. The predicate can only precede the adverb if we move the complete small clause into the local domain of V, as in (85b). Consequently locative inversion with pied piping of the VPadverb would now require that the higher VP is moved, but this would wrongly pied pipe the DP as well. (85)

a. b′.

[vP v+walked [VP the queen twalked [gracefully [VP twalked [SC tthe queen down the stairs]]]]] [vP v+walked [VP [SC the queen down the stairs] twalked [gracefully [VP twalked tSC ]]]]

Observe that the present analysis of locative inversion reveals a very interesting fact about the constraint ALIGNFOCUS. This constraint was originally introduced to block movement of (subject and object) DPs that are part of the focus of the clause. In the present analysis of locative inversion, things take an unexpected turn in that ALIGNFOCUS forces movement of the DP that is part of the focus, since only then is locative inversion possible. So, depending on the syntactic context, ALIGNFOCUS may either block or force movement of a DP that is part of the focus of the clause. 5.3.2.1.3. Why locative inversion is blocked in Dutch In English, the subranking ALIGNFOCUS >> A-OVER-A allows violation of A-OVER-A in order to satisfy ALIGNFOCUS. This is not possible in Dutch, which has the subranking A-OVER-A >> ALIGNFOCUS. The evaluation in Tableau 18 shows that this subranking consequently prohibits locative inversion in Dutch; the candidate set in (86) does not include candidates with verb movement, which will be excluded by the weak ranking of *STRAY FEATURE. The candidates in (86a-c) have the predicate in a position following the verb in clause final position and these are correctly excluded as the result of either a fatal violation of either EPP(gender) or A-OVER-A. Candidate (86d) is the locative inversion construction, which is

314 Chapter 5: Predicate movement

also excluded due to a fatal violation of A-OVER-A. Candidates (86e&f), which can be distinguished by considering the relative placement of the DP and the sentential adverb indicated by ADV, are both possible surface forms in Dutch. The fifth candidate arises when the small clause subject is part of the focus of the clause, as in the pertinent evaluation. The sixth candidate will be selected when the small clause subject is part of the presupposition of the clause, as not applying subject shift would then result in a fatal violation of EPP(case) and EPP(person). (86)

a. b. c. d. e. f.

[IP I ADV [vP v [VP V [SC DP Pred]]]] [IP I ADV [vP v [VP DP V [VP tV [SC tDP Pred]]]]] [IP DP I ADV [vP v [VP tDP V [VP tV [SC tDP Pred]]]]] [IP [SC tDP Pred] I ADV … [vP v [VP DP V [VP tV tSC]]]] [IP I ADV [vP v [VP [SC DP Pred] V [VP tV tSC]]]] [IP DP I ADV [vP v [VP [SC tDP Pred] V [VP tV tSC]]]]

Tableau 18: No locative inversion in Dutch

DP ⊂ focus; Pred ⊄ focus (86a) (86b) (86c) (86d) (86e) (86f)

A-OVERA

EPP

*! *! *! *! )

AF

EPP(case) EPP(person)

* ** *** * ** ***!

** **

(gender)

* **

*MOVE

** *** *** ** ***

5.3.2.1.4. Comparison of the placement of DPs and predicative phrases According to the proposal above, the preverbal placement of predicative phrases in the OV-languages is the result of short “object shift”, that is, triggered by the ij-features on the verbal root V. From this it should follow that there are certain similarities between the placement of objects and predicative phrases. Of course, the distribution of objects and predicative phrases is not identical given that objects have case features, which may be probed by the case features on v, that are lacking on predicative phrases. This accounts for the fact that whereas objects may be placed in front of the sentential adverbs as a result of regular object shift, predicates must follow these adverbs, as in (87).

Predicate movement in Dutch 315 (87)

a.

b.

dat Jan waarschijnlijk zal lezen. that Jan that book probably will read ‘that Jan will probably read that book.’ dat Jan dat hek waarschijnlijk zal verven. That Jan that gate green probably will paint ‘that Jan will probably paint that gate green.’

We have seen in section 2.4.1.3 that the Germanic VO- and OV-languages differ in that in the former short object shift must cross the VP-adverbs, whereas in the latter this is only optionally the case. This was attributed to the constraint NOVACM, which requires that movement is visible in the output. Given that in the OV-languages short object shift crosses the clausefinal verb, movement of the object across the VP-adverbs is not necessary in order to make the shift visible. This accounts for the fact that both orders in (88a) can be derived by short object shift (although we have not yet established what determines the choice between the two options). If the predicate is moved into preverbal position by means of short object shift, we expect that the predicate may also precede the VP-adverbs. Example (88b) shows, however, that this expectation is not borne out. (88)

a.

b.

dat Jan waarschijnlijk nauwkeurig zal lezen. that Jan probably that book accurately will read that Jan will probably read that book accurately dat Jan het hek waarschijnlijk zorgvuldig that Jan the gate probably green rigorously zal verven. will paint that Jan will probably paint the gate green rigorously.

The fact that predicative phrases normally occupy a position left-adjacent to the clause-final verb cluster may therefore cast some doubt on the claim that it occupies this position as the result of short object shift. Note, however, that predicative phrases are not the only elements that must be left-adjacent to the clause-final verbal sequence. Consider the examples in (89). (89)

a.

b.

dat Jan net met een verfspuit geverfd heeft. that Jan just the gate with a spray gun painted has ‘that Jan just painted the gate with a spray gun.’ dat Jan daar gisteren mee geverfd heeft. that Jan there yesterday the gate with painted has ‘that Jan painted the gate with it yesterday.’

316 Chapter 5: Predicate movement

These examples show that whereas an instrumental PP may be either preceded or followed by the object, the stranded preposition mee of an instrumental PP must be left-adjacent to the verb cluster and therefore cannot be followed by the object. This shows that certain elements are privileged in the sense that they are preferred in the position left-adjacent to the verb. Van Riemsdijk (1997) has convincingly argued that the grammatical order in (89b) cannot be derived by means of incorporation of the stranded preposition into the main verb, on the basis of the fact that it cannot precede the main verb when the latter follows the auxiliary in the verb cluster. In this respect, stranded prepositions differ from verbal particles, which can be incorporated. (90)

a.

b.

dat Jan daar gisteren het hek heeft geverfd. that Jan there yesterday the gate with has painted ‘that Jan painted the gate with it yesterday.’ dat Jan de was heeft gehangen. hung that Jan the laundry PRT. has ‘that Jan hung up the laundry.’

From this we must conclude that the privileged status of stranded preposition forces short object shift to cross the stranded preposition, with the result in (89b). In a similar way, we could now argue that the privileged status of predicative phrases blocks movement of these phrases across the VP-adverbs, with the result in (88b). Given that both predicative phrases and stranded prepositions are privileged, the question arises what happens if they occur in the same clause. Consider the examples in (91). The pattern in (91a) is of course as expected: given that the predicative phrase is privileged, movement of the predicative phrase into preverbal position cannot cross the instrumental phrase. The fact that both word orders in (91b) are acceptable in tandem with Van Riemsdijk’s conclusion that stranded prepositions cannot incorporate into the main verb shows that predicative phrases may cross the stranded preposition of an instrumental phrase. This follows if we assume that predicative phrases and stranded prepositions are equally privileged. (91)

a.

b.

dat Jan het hek met een verfspuit verft tgroen. that Jan the gate green with a spray gun paints ‘that Jan is painting the gate green with a spray gun.’ dat Jan daar het hek mee verft tgroen. that Jan there the gate green with paints ‘that Jan is painting the gate green with it.’

Predicate movement in Dutch 317

This shows that, just like object shift, the postulated leftward movement of predicative phrases actually can cross VP-adjuncts, albeit that this is restricted to those cases in which the adjunct is a stranded preposition.69 The discussion above, of course, has left open several questions. For example, we have not discussed what brings about the privileged status of predicative phrases and stranded prepositions, let alone given a formal account of it. Furthermore, we have not yet established what determines the choice between the two options in (91b); see Van Riemsdijk (1997) for some tentative suggestions. I will leave these issues to future research. 5.3.2.1.5. A final question In the previous subsections we have seen that in clauses containing a predicative phrase the constraint A-OVER-A forces leftward shift of the predicative phrase: example (92a) reflects the base order and is ungrammatical because it violates EPP(ij); (92b) satisfies EPP(ij), but is unacceptable since it violates A-OVER-A; example (92c), finally, satisfies both EPP(ij) and A-OVER-A and is therefore selected as the optimal candidate. (92)

a.

*dat Jan met een verfspuit verft [SC het hek groen] that Jan with a spray gun paints the gate green b. *dat Jan met een verfspuit het hek verft [SC thet hek groen] c. dat Jan met een verfspuit [SC het hek groen] verft tSC

Given that the ij-features of the verbal root are valued in (92c), we expect that the object het hek cannot undergo short object shift anymore. When this object shifts, this must therefore be a case of regular object shift, that is, 69

Van Riemsdijk (1997) claims that predicative PPs cannot cross stranded prepositions. Two of the examples he uses to illustrate this point are problematic: his (41a) involves a prepositional indirect object and not a predicative PP, and his (43) involves a stranded preposition of a PP embedded in the predicative PP (and, to my ear, is actually not as bad as he claims). His example (42), repeated here in a slightly different form as (ia), is the only case that involves a predicative PP and the stranded preposition of a VP-adjunct, and in this case the order predicative PP stranded preposition sounds quite acceptable (slightly marked at worst) according to some of my informants, including myself. Another example is given in (ib). (i)

a.

Je kunt er moeilijk mee gaan. you can there difficultly to the doctor with go

b.

Jan heeft er de boeken mee < uit de kast>gehaald. Jan has there the books out-of the bookcase with taken

‘You can’t very well go see a doctor with a thing like that. ‘Jan has taken the books out of the bookcase with it.

318 Chapter 5: Predicate movement

triggered to the case features on v. Since object shift normally crosses the sentential adverbs we expect that this object shift cannot target a position in between the VP-adverbs and the sentential adverbs. As is shown in (93), this prediction is not borne out: (93a) is the structure that results from leftward movement of the small clause; the order in (93c) is expected to result from regular object shift; the order in (93b) is wrongly expected not to arise. (93)

a. b. c.

dat Jan waarschijnlijk met een verfspuit [SC het hek groen] verft. that Jan probably with a spray gun the gate green paints dat Jan waarschijnlijk het hek met een verfspuit [SC thet hek groen] verft. dat Jan het hek waarschijnlijk met een verfspuit [SC thet hek groen] verft.

Why is it that, contrary to expectations, (93b) is acceptable? The reason for this is that we were wrong in assuming that regular object shift obligatorily crosses the sentential adverb. Rather, we must assume that, in principle, regular object shift can choose whether it crosses the sentential adverb or not. That crossing the sentential adverbial phrase nevertheless appears to be obligatory is due to the fact that the constraint NOVACM will be violated when regular object shift applies string vacuously. Now, consider again example (93b). As can be seen there, regular object shift does not apply string vacuously but crosses the VP-adverb met een verfspuit. As a result of this, NOVACM is satisfied, and crossing the sentential adverb is therefore not necessary. This account of the acceptability of (93b) shows that short and regular object shift are similar in not requiring that adverbs (of, respectively, the VP and sentential type) be crossed. The (misconceived) impression that this is an inherent property of these movements is due to the effect of the constraint NOVACM: only when it is needed to void a fatal violation of this constraint is object shift required to cross adverbs of the relevant type. 5.3.2.1.6. Conclusion This section has established that in Dutch small clause constructions EPP(ij) can only be satisfied by moving the small clauses into preverbal position, because moving the external argument of the small clause in isolation is blocked by a constraint that I provisionally called A-OVER-A. In this way we account for the fact that small clause predicates, despite being basegenerated in postverbal position, obligatorily precede the verbs in clausefinal position. I have shown that in neutral contexts movement of the complete small clause is also preferred in English, but that this effect of A-OVER-A can be blocked due to the fact that this constraint is outranked by ALIGNFOCUS: when the predicate is part of the focus of the clause,

Predicate movement in Dutch 319

blocks movement of the predicate so that movement of the DP in isolation is required. In the locative inversion construction ALIGNFOCUS also forces movement of the DP that is part of the focus of the clause: this DP can only be placed in the right periphery of the clause when we first extract it from the small clause, and proceed by applying remnant movement of the small clause/VP into SpecIP across the moved DP. This section has also discussed certain surprising differences between the placement of DPs and predicative phrases. ALIGNFOCUS

5.3.2.2. Extending the analysis: VP-movement The introduction to section 5.2.5 claimed that the proposal concerning English locative inversion constructions can be readily extended to the constructions illustrated in (52) to (54), which are known under the general designation of Preposing around be. Similarly the preverbal position in Dutch is not reserved to predicative PPs: all predicative complements must occupy this position, irrespective of whether they are PPs, APs, or NPs. I will assume without illustrating this any further that the examples in (94) can all be accounted for in essentially the same way. (94)

a. b. c.

dat that dat that dat that

Jan de baby duwde . Jan the baby out of bed pushed Jan de muur verfde . Jan the wall yellow painted Jan Peter vindt . Jan Peter a good painter considers

This section will focus on VP-movement in Dutch perfect tense constructions like (95). I will argue in section 5.3.2.2.1 that the participle-auxiliary order in this example is derived from the underlying auxiliary-participle order by movement of the VP het boek gelezen ‘read the book’ to the left of the auxiliary, and propose that this VP-movement is triggered by the unvalued ij-features on the verbal root of the auxiliary. Section 5.3.2.2.2 will conclude the discussion by exploring some consequences of this analysis for the description of the perfect tense construction in the VO-languages. (95)

dat Jan het boek gelezen heeft. that Jan the book read has ‘that Jan has read the book.’

5.3.2.2.1. VP-movement in Dutch perfect tense constructions Consider the Dutch examples in (96). The placement of the direct object in (96a) follows from the subranking EPP(gender) >> ALIGNFOCUS >> *MOVE,

320 Chapter 5: Predicate movement

which forces movement of the object into the local domain of both the participle and the auxiliary; cf. the discussion in section 2.4.2.2. What we have not discussed so far, however, is that the grammatical order in (96a) alternates with the order in (96b), in which the participle also precedes the auxiliary. That example (96b) is possible follows from our assumption that the participle agrees in ij-features with its internal argument, and is consequently a potential probe for the unvalued ij-features of the auxiliary. (96)

a.

b.

dat Jan heeft gelezen . that Jan the boek has read ‘that Jan has read the book.’ dat Jan het boek gelezen heeft.

If this proposal is on the right track the derivations of the examples in (96) are as indicated in (97). In both cases, the strong ranking of EPP(ij) forces the object to move into the local domain of the participle. The unvalued features on the auxiliary, on the other hand, may attract either the direct object or the VP, which gives rise to, respectively, (97a) and (97b). (97)

a. b.

dat Jan dat boek heeft [VP t′dat boek gelezen tdat boek] dat Jan [VP dat boek gelezen tdat boek] heeft tVP

Although the alternation between (96a) and (96b) follows straightforwardly from the proposal developed above, a problem seems to arise with the constraint A-OVER-A. In the remainder of this section I will suggest a tentative solution to this problem. Section 5.3.2.1 has argued that the constraint A-OVER-A can be held responsible for the fact that in Dutch small clause constructions, the complete small clause must be moved into the preverbal position. However, this constraint leads us to expect that EPP(gender) always forces movement of the more encompassing phrase into the local domain of the auxiliary. We therefore wrongly predict (97a) to be ungrammatical, because A-OVER-A requires that the unvalued ij-features on the auxiliary trigger movement of the VP. Example (98) shows that a similar problem arises in perfect constructions with a small clause complement. Postulating the constraint A-OVER-A again leads to the wrong prediction that only (98b) is acceptable. (98)

a. b.

dat Jan de muur helblauw heeft geschilderd. that Jan the wall bright blue has painted dat Jan de muur helblauw geschilderd heeft.

Let us consider the derivation of the examples in (98) in more detail. The derivation starts out with the VP structure in (99a). Since in Dutch the

Predicate movement in Dutch 321

unvalued gender feature on V forces movement of its probe, either the small clause or its DP subject must move into the local domain of V; given the fact that the small clause precedes the main verb, the constraint AOVER-A correctly predicts movement of the full small clause, as in (99b). In (99c) the auxiliary heeft is merged, whose gender feature likewise forces movement of its probe. Now the option is between moving the DP, the small clause or the VP. It is clear that A-OVER-A favors movement of the VP, as in (99d), and example (98b) shows that this is indeed an acceptable option. However, example (98a) shows that it is also possible to subextract the small clause, as in (99d′), despite the fact that this is prohibited by A-OVER-A. Note that A-OVER-A does correctly predict that it is impossible to extract the DP from the VP and the small clause, as in (99d′′): *dat Jan de muur heeft helblauw geschilderd.70 (99)

a. b. c. d. d′. d′′.

[VP geschilderd [SC de muur helblauw]] [VP [SC de muur helblauw ] geschilderd tSC ] [auxP heeft [VP [SC de muur helblauw] geschilderd tSC ]] [auxP [VP [SC de muur helblauw] geschilderd] heeft tVP]] [auxP [SC de muur helblauw] heeft [VP tSC geschilderd tSC ]] *[auxP de muur heeft [VP [SC tde muur helblauw] geschilderd tSC ]]

There are several ways in which we could approach the unexpected acceptability of the examples in (96a) and (98a). One option is to assume that the word order aux-participle found in these examples does not result from core grammar. There are a number of good reasons to take this position. First, Dutch seems to be the only Germanic language that allows this order. Secondly, the order aux-participle was promoted by the prescriptive grammarians, who condemned the participle-aux order as a “Germanism”, that is, unwanted influence from German; as a matter of fact, the aux-participle is still promoted as the more elevated style in some current instruction books on writing. Thirdly, Zuckermann (2001: ch.4) has shown that younger children have a preference for the participle-aux order, even if the parents prefer the aux-participle order. Finally, De Sutter (2005: section 7.3.2) has argued there there is psycholinguistic evidence for the claim that aux-participle order is marked given that even for adults the participle-aux order is easier to produce than the aux-participle order: when the computational load or the complexity of interaction increases, adult 70

The order DP-aux-Adj-Vmain is sometimes found, but is subject to various additional restrictions: for example the adjective in this order must be monosyllabic and cannot be accompanied by a modifier or a complement. For this reason, I will not discuss cases like this here.

322 Chapter 5: Predicate movement

speakers tend to use to the participle-aux order more often. This all suggests that the participle-aux order is the basic order in Dutch, and that the availability of the aux-participle order might be due to explicit instruction. Of course, claiming that the aux-participle order is not part of core syntax does not imply that nothing more need be said. More specifically, I believe that the order in (96a) and (98b) must still be syntactically derived since it is simply too productive to assume it is “learned” on an item-byitem basis. What I want to suggest here is that, despite appearances, the derivation of (96/98a) does involve VP-movement. One argument in favor of this claim concerns the position of the VP-adverbs. Since VP-adverbs modify the projection of the main verb, the null hypothesis is that they are adjoined to the projection of the main verb: this implies that they are generated as part of the complement of, and thus to the right of, the auxiliary verb. In the surface realization of the Dutch clause, however, the VP-adverbs must precede the auxiliary; the examples in (100) further show that their placement does not depend on the position of the main verb. (100) a. b.

dat Jan nauwkeurig dat boek gelezen heeft. that Jan meticulously that book read has dat Jan dat boek heeft gelezen.

Of course, example (100a) can be readily derived if the VP-adverb is pied piped by movement of the VP; cf. the derivation of the English example in (84a). If (100b) does not involve VP-movement, the word order in this example can only be derived by moving the adverb in isolation. This, however, seems to go against the Last Resort Condition on movement since there does not seem to be any trigger that motivates this movement. This leads to the conclusion that (100b) is derived by VP-movement of a remnant VP. The derivation of (100b) should therefore be roughly as indicated in (101); I have ignored the subject for simplicity. (101) a. b. c. d. e. f.

[gelezen dat boek] [VP dat boek gelezen [tgelezen tdat boek]] [nauwkeurig [VP dat boek gelezen [tgelezen tdat boek]]] [XP X+gelezen [VP nauwkeurig [VP dat boek t′gelezen [tgelezen tdat boek]]]] [auxP heeft [XP X+gelezen [VP nauwkeurig [VP dat boek t′gelezen [tgelezen tdat boek]]]]] [auxP [VP nauwkeurig [VP dat boek t′gelezen [tgelezen tdat boek]]] heeft [auxP theeft [XP X+gelezen tVP]]]

Predicate movement in Dutch 323

The fact that Standard Dutch is the only Germanic OV-language allowing the order aux-participle can now be accounted for by assuming that the more formal register of Standard Dutch is exceptional in allowing movement of the participle into the head position of X. Given the acquisition and psycholinguistic data summarized earlier, we may assume that this exceptional part of the syntax of Standard Dutch is the result of explicit traning. The derivation in (101) of course raises the following urgent question: what does XP stands for? I want to suggest that XP is the functional projection AspP, which was already introduced in 2.4.2. Although we have seen in section 2.4.2.2 that V-to-Asp normally does not take place in the Germanic OV-languages, we may assume that it can exceptionally take place in Standard Dutch in examples like (96a) and (98a). The fact that this is possible is of course surprising given our earlier conclusion that *STRAY FEATURE is outranked by *MOVE in Standard Dutch, so future research will have to show whether the order aux-participle is the result of some other yet unidentified constraint, or whether we should simply accept that these orders are not part of core grammar. It is clear that in order to settle this issue we will also have to take into account non-perfective complex verb constructions like (102), since in these examples the modal verb(s) normally precede the main verb in Dutch as well (cf. Den Besten & Broekhuis 1989). (102) a.

b.

dat Jan een boek zal willen lezen/*lezen willen zal that Jan a book will want read ‘that Jan wants to read a book.’ dat Jan de muur helblauw wil verven/*verven wil that Jan the wall bright blue wants paint

This section has shown that the fact that participles tend to precede the auxiliary in the Germanic OV-languages can be considered a natural consequence of Axiom II in (4b), according to which the main verb agrees in ij-features with its internal argument: due to the constraint A-OVER-A the unvalued gender feature on the auxiliary will trigger movement of the verb phrase instead of short object shift. The aux-participle order that is found in Standard Dutch can be readily derived by assuming that VP-movement is preceded by V-to-Asp, although we still have to account for the fact that the latter movement, which is normally blocked in the Germanic OV-languages, can exceptionally take place in the derivation of this order.

324 Chapter 5: Predicate movement

5.3.2.2.2. The analysis of the perfect tense construction in the VO-languages If the analysis of the perfect tense construction in the OV-languages proposed in the previous subsection is on the right track, this may also have consequences for the analysis of the perfect tense construction in the VOlanguages. Recall that section 2.4.2.3 has argued that short object shift into the local domain of the verbal root of the auxiliary is excluded in the VO-languages due to the fact that this gives rise to a fatal violation of H-COMPL. Consider the derivation of the ungrammatical English perfect tense example in (103a). In (103a) the unvalued ij-features on the verbal root read trigger short object shift of the direct object, as in (103b). The resulting inversion of the verb and the object can be made undone by subsequent V-to-Asp movement of read, as in (103c), so that H-COMPL is respected in the structure so far. This is different when the unvalued features of the verbal root of the auxiliary verb have in (103d) attract the object. This results in the structure in (103e) which fatally violates H-COMPL: although the inversion of the object and the verb have resulting from short object shift into the local domain of the auxiliary can be made undone by V-to-v, the inversion of the object and the main verb cannot. (103) a. b. c. d. e. f. g h.

[read the book] [the book read [tread tthe book]] [Asp+ read [the book t′read [tread tthe book]]] have [Asp+ read [the book t′read [tread tthe book]]] [the book have [Asp+read [t′the book t′read [tread tthe book]]]] v [the book have [Asp+read [t′the book t′read [tread tthe book]]]] [they v+have [the book thas [Asp+read [t′the book t′read [tread tthe book]]]]] etc.

Section 2.4.2.3 therefore concluded that short object shift into the local domain of the auxiliary does not take place. However, we now know that as a result of object agreement the verbal projection of the participle is also a potential goal of the unvalued ij-features of the auxiliary; it therefore may be the case that this projection will move into the local domain of the verbal root of the auxiliary. This would certainly be the case if the ij-features of the participle percolate up to the level of AspP: the ij-features on the verbal root of the auxiliary will then attract the complete AspP, and the inversion of the auxiliary and the AspP in (103e′) will be made undone by subsequent V-to-v of the verbal root of the auxiliary; the resulting structure in (103g′) satisfies H-COMPL.

Predicate movement in Dutch 325 (103) d. e′. f′. g′. h′.

have [Asp+ read [the book t′read [tread tthe book]]] [[Asp+ read [the book t′read [tread tthe book]]] have tAspP] v [[Asp+ read [the book t′read [tread tthe book]]] have tAspP] [they v+have [[Asp+ read [the book t′read [tread tthe book]]] thave tAspP]] etc.

Things would be different if the ij-features of the participle percolate up no further than the extended projection in (103b), which I will refer to as PartP; short object shift of PartP is excluded then, since it will place the object to the left of the participle thus invoking a fatal violation of H-COMPL, just like leftward movement of the object. This is shown in the derivation below. (103) d. e′′. f′′. g′′. h′′.

have [Asp+ read [the book t′read [tread tthe book]]] [[the book t′read [tread tthe book]] have [Asp+read tPartP]] v [[the book t′read [tread tthe book]] have [Asp+read tPartP]] [they v+have [[the book t′read [tread tthe book]] thave [Asp+read tPartP]]] etc.

Since I have suggested for Dutch that the auxiliary-participle order is derived by remnant movement of PartP, it is tempting to claim that the derivation in (103e′′-g′′) is the correct one. If so, we must conclude that the unvalued ij-features normally do not trigger VP-movement in the VOlanguages. We may, however, still expect to find short object shift of PartP in topicalization constructions of the type discussed in section 4.3.3: topicalization of AspP will restore the original order of the participle and the direct object, and thus void the fatal violation of H-COMPL in (103g′′). Since I did not study this possibility, I must leave it to future research. 5.4. Verbal modifiers in Hungarian This section will argue that the analysis of the English and Dutch data in the previous sections can be extended to account for the distribution of the so-called verbal modifiers (VMs) in Hungarian. Section 5.4.1 starts with a discussion of some of the core data and the analysis of these data provided by É. Kiss (2002; 2006). É. Kiss’ analysis is built on the assumption that VMs and focused phrases compete for the same position. Section 5.4.2 argues that this cannot be correct, and that the more traditional approach according to which the two types of movement are triggered by different features is more to the point. More precisely, I will argue that movement of the VM can be considered another illustration of the hypothesis in (5) from section 5.1, in that it is triggered by the ij-features on V, which are of

326 Chapter 5: Predicate movement

course not plausible triggers for movement of focused phrases. Section 5.4.3 will provide an account of the complementary distribution of VMs and focused phrases. 5.4.1.

Core data and previous analysis

Consider the examples in (104). Example (104a) shows that in neutral sentences the VM el ‘away’ must precede the finite verb. However, if one of the constituents in the clause is focused, that phrase must be placed immediately in front of the finite verb, whereas the VM must follow it. This is shown in (104b), where (identificational) focus is indicated by means of small capitals. (104) a. b.

János ment . János away went JÁNOS ment .

A more or less standard way of accounting for the examples in (104), which É. Kiss (2002: section 4.2) partly attributes to Brody (1990b;1995), is to assume that at least two functional projections can be found on top of VP: Asp(ect)P and FocusP. SpecAspP is considered the landing site for VMs, which are often aspectual in nature, and SpecFocusP is the landing site for constituents acting as identificational focus. Since the verb is assumed to move into the head positions of these phrases, the derivation of the examples in (104) should be as indicated in (105). (105) a. b.

János [AspP el ment [VP ... tment ... tel ...]] [FocusP JÁNOS ment [AspP el tment [VP ... tment ... tel ...]]]

É. Kiss (2002) dismisses this type of analysis by pointing out that it wrongly predicts that the VM must be right-adjacent to the finite verb in the focus construction;71 the examples in (106b&c) show that the VM be can be placed much more to the right. É. Kiss therefore suggests that the focused phrase and the VM compete for the same position.

71

This actually does not follow from Brody’s (1990b;1995) proposals but only from the specific implementation of some of his ideas by É. Kiss, especially the idea that the VM moves into SpecAspP.

Verbal modifiers in Hungarian 327 (106) a.

János be mutatta Pétert Marinak. János VM introduced Péteracc Maridat ‘János introduced Peter to Mary.’ b. PÉTERT mutatta János be Marinak. c. (?)PÉTERT mutatta János Marinak be.

É. Kiss (2006) gives shape to this suggestion by assuming that both VMs and focused constituents target Koster’s (1994) PredP, and, indeed, the presupposition that VMs are predicates doesn’t seem to be too far-fetched. First, it has often been argued that particles are actually small clause predicates (cf., e.g., Den Dikken 1995a). Secondly, a cursory inspection of the other constituents that can be placed into the position preceding the finite verb immediately shows that this position is normally occupied by predicative elements: (107) to (109) give a number of representative examples, taken from Komlósy (1985). In (107) the element preceding the finite verb is a predicative locational phrase, 72 in (108) an adjectival predicate, and in (109) a nominal predicate. (107) a.

b.

(108) a.

b.

Péter laposra verte Jánost. Péter flat-on beat Johnacc ‘Péter beat John to pulp.’ János az asztalra tette a könyvet. János the table-onto put the bookacc ‘János put the book on the table.’ Mari tegnap beteg volt. Mari yesterday ill was ‘Mari was ill yesterday.’ János okosnak tartja Pétert. János cleverdat consider Peteracc ‘János considers Peter clever.’

(109) János tavaly katona volt. János last year soldier was ‘János was a soldier last year.’

The fact that bare main verbs can also occupy the position preceding the finite verb is, of course, also in line with the idea that this position is designated to predicative elements. 72

Synchronically seen, the locational phrases in (107a&b) are case-marked DPs, but in other constructions they can also be postpositional phrases; cf. example (119b) below.

328 Chapter 5: Predicate movement (110) Mari venni akar egy autót. Mari buyinf want a caracc. ‘Mari wants to buy a car.’

The examples in (111), which are again taken from Komlósy (1985), show that bare nouns may also function as VMs, that is, target the same position as the predicative elements discussed above. This is a potential problem for É. Kiss’ proposal: although from a semantic point of view bare nouns are indeed predicates, it is obvious that at the level of the clause they do not function as such syntactically — they function, rather, as regular (non-specific) arguments in the sense that they carry a thematic role. Consequently, it is not clear what motivates their movement into the specifier of a PredP. Another drawback of É. Kiss’ claim that bare nouns function as predicates at the level of the clause is that it blurs the distinction between arguments and predicates. (111) a.

b.

János újságot olvas a kertben. János newspaperacc read the garden-in ‘John is reading a newspaper/newspapers in the garden.’ Péternek víz ment a szemébe. Péterdat water went the eye-poss-into ‘Water got into Peter’s eye(s).’

The examples above have shown that the set of VMs (= the set of elements that can occupy the position preceding the finite verb in neutral sentences) consists of the elements listed in (112). Section 5.4.2 will argue that what these elements have in common is that they carry (at least) a gender feature that can be probed by the unvalued gender feature on the verbal root V. In this way, the Hungarian data discussed in this section will receive an account similar to the data from English and Dutch discussed earlier. (112) a. b. c.

bare nouns (111) particles (104) and other predicative phrases (107) - (109) bare main verbs (110)

However, let us first complete our discussion of É. Kiss’ (2006) proposal. The assumption that focused constituents are also placed in SpecPredP leads É. Kiss to the conclusion that these elements are predicates as well, and much of her paper is devoted to providing a (semantic) motivation for this claim. Since wh-phrases arguably occupy the focus position in Hungarian, É. Kiss correctly predicts that they also block movement of the VM; cf. (113b).

Verbal modifiers in Hungarian 329 (113) a.

b.

JÁNOS ment . János away went ‘It was JÁNOS who went away.’ Ki ment ? who away went ‘Who went away?’

Example (114) shows that negation also blocks movement of the VM, and É. Kiss accounts for this fact by claiming that in examples like these the verb must be interpreted as an information focus (which I take to refer to the new information in the clause), and that this voids the need for moving the VM into SpecPredP. (114) János János away

nem ment . not went

There are at least two objections to this account of (114). The first is that information focus on the verb does not void the need for VM movement in affirmative clauses; apparently the presence of negation is crucial for blocking movement of the VM. The second objection is related to the claim that VMs, focused constituents, and wh-phrases are all predicative phrases. If this is indeed true, it becomes mysterious why information focus on the verb in negative clauses blocks movement of VMs into SpecPredP, while still allowing focus and wh-movement in examples like (115). (115) a.

b.

JÁNOS nem jött el. János not come VM ‘It was János that didn’t come.’ Ki nem jött el. who not come VM ‘Who didn’t come?’

Section 5.4.3 will show that it is not necessary to assume that focused constituents and wh-phrases are predicates by providing an analysis that straightforwardly accounts for the complementary distribution of these phrases and the VMs in the position preceding the finite verb. From this analysis the fact that negation blocks movement of the VM will follow as well. But in order to be able to introduce the proposal, I will first have to discuss the question of what triggers the movement of the VMs in (112) in the first place.

330 Chapter 5: Predicate movement

5.4.2.

What do the verbal modifiers have in common?

Section 5.2 has investigated the set of elements that can be attracted in English by the ij-features of I, and concluded that this set consists of the subject of the clause, small clause predicates, and VPs predicated of the subject. Section 5.3 has shown that the set of elements that can be attracted by the ij-features of V in Dutch are the object, small clause predicates predicated of the object, and verb phrases that agree with the object. Now, compare these two sets with the set of elements that can be used as VMs in (112). Again we are dealing with nominal phrases, small clause predicates and verb phrases. This suggests that the movement process that, in neutral sentences, leads to the placement of the VM in the position preceding the finite verb falls into the same class as the English and Dutch movement processes discussed in sections 5.2 and 5.3. In short, we should be able to handle the Hungarian data by calling upon hypothesis (5). An important question is, of course, whether the Hungarian movement is of the English or of the Dutch type, that is, whether movement of the VM is triggered by the ij-features on I or by those on V. The fact that the VM can be a direct object unambiguously shows that the latter is the case. Example (111a), repeated here as (116), illustrates this. (116) János újságot olvas a kertben. János newspaperacc read the garden-in ‘János is reading a newspaper/newspapers in the garden.’

The conclusion that it is the ij-features on V that trigger movement of the VM makes a straightforward prediction: a VM can also be the (bare noun) subject of the clause provided that it is an internal argument. In order words, the subject can be a VM when the verb is unaccusative, but not when the verb is unergative. This prediction is confirmed by the facts in (117). In (117a), which was given earlier as (111b), the verb is unaccusative and the nominative bare noun phrase víz ‘water’ can be placed in the position preceding the finite verb. In example (117b), constructed by my co-author Vera HegedĦs, the verb is intransitive and the bare noun cannot precede the finite verb; the subject must be preceded by an article, and must either follow the finite verb, as in (117c), or precede it as a topic, as in (117c′).

Verbal modifiers in Hungarian 331 (117) a.

Péternek víz ment a szemébe. Péterdat water went the eye-poss-into “Water got into Péter’s eye(s).” b. *Fiú nevetett. boy laughed c. Nevetett egy/a fiú. laughed a/the boy c′. Egy/A fiú nevetett. a/the boy laughed

The fact that verbal particles like el in (104a) and small clauses like those given in (107) to (109) can be placed into the same position as the nominal arguments follows from axiom (4a), according to which these elements have the same ij-features as the noun phrase they are predicated of. According to axiom (4b) the same holds for example (110), where a verb phrase is placed in front of the finite verb. There are still many questions to answer. For example, one may ask why only a bare noun object/subject can act as a VM, that is, why only bare nouns can be moved into the position preceding the finite verb. If the unvalued ij-features on V trigger this movement, there must be some independent reason why (118b) is unacceptable under its neutral reading when the article is present. (118) a. b.

János adott Marinak *(egy) könyvet. Jánosnom. gave3SG Maridat. a bookacc. János (*egy) könyvet adott Marinak.

Similarly, we have to ask why only bare verbs can act as verbal modifiers (cf. (110)). Note that these two questions cannot be answered by saying that we are dealing with head movement, since it is clear that the pertinent position can be filled by complex phrases as well. This is illustrated in (119). In (119a) a full locational DP is placed in the position of the VM, so that this example shows that there is no general ban on having non-bare noun phrases in this position. That a more complex phrase can be used to fill this position is also shown by (119b&c) which involve a PP and a modified AP, respectively. 73 73 Of course, the examples in (119) can also be uttered with focus accent on the predicative phrase, in which case the predicates are in the designated focus position. According to one of the reviewers of Broekhuis & HegedĦs (to appear), this is the only option for (119c). However, according to Veronika HegedĦs, this example can be uttered without triggering the exhaustive focus reading. Even if the reviewer were right, the examples in (119a&b) would still prove our point.

332 Chapter 5: Predicate movement (119) a.

b.

c.

János az asztalra tette a könyvet. János the table-on put the book-acc ‘János put the book on the table.’ János az asztal alá tette a széket. János the table under put the chair-acc ‘János put the chair under the table.’ Mari tegnap nagyon beteg volt. Mari yesterday very ill was ‘Mari was ill yesterday.’

The fact that predicative DPs/PPs/APs can be used as VMs suggests that a bare verb functioning as VM must also be analyzed as a full VP. This, in turn, implies that the complements of the verb are obligatorily removed from the VP before the phrase is moved into the position in front of the finite VP, a conclusion that was also reached by Koopman & Szabolcsi (2000:39ff.). The fact that only bare verbs and, especially, bare nouns can function as VMs, of course, remains in itself as mysterious as ever; on this issue I only have some tentative remarks at best. First, we can note that the infinitive can be a transitive verb; cf. (110). If we adopt the standard assumption that it is the infinitive that is responsible for accusative case assignment, we must conclude that the extended projection of the infinitive phrase must at least contain a v*, which may in principle attract the direct object, so that extraction of the object from VP is in accordance with the Last Resort Condition on movement. Secondly, we must note that there is a more general tendency to reduce the size of the VM as much as possible. This is illustrated by example (120a), which involves an adjectival predicate containing the head elégedett ‘satisfied’ taking a PP-complement az eredménnyel ‘with the result’: when the adjectival phrase is placed into the preverbal position, the complement is obligatorily stranded. Example (120a) shows that the same holds for noun phrases, which must strand postnominal (but not prenominal) modifiers. The data in (120) were again provided by Veronika HegedĦs. (120) a.

b.

Anna elégedett-nek tĦnt . the resultinstr seemed Anna satisfieddat ‘Anna seemed satisfied with the result.’ János könyvet olvasott . János bookacc the rain.forestsdel read ‘John read a book on rain forests.’

The obligatory stranding of the complement and modifier in the examples in (120) suggests that there is some constraint favoring stranding of such

Verbal modifiers in Hungarian 333

material, which may also account for the fact that in the case of verbal projections remnant movement is preferred. Of course this all hardly scratches the surface of the problems involved, but this does not affect the conclusion that, at least potentially, the Hungarian data can be fruitfully approached by taking recourse to hypothesis (5). 5.4.3.

The distribution of verb modifiers and focused phrases

One of the long-standing problems with Hungarian is the complementary distribution in the position preceding the finite verb of VMs and focused phrases/wh-phrases/negation. É. Kiss (2006) tries to account for the complementary distribution of VMs and focused constituents by claiming that they both should be considered predicative elements, which therefore compete for the same position, SpecPredP. According to this logic, whphrases should also count as predicative elements. This leads us to a definition of predication that is pretty remote from everyday scientific use and blurs the distinction between arguments and predicates. This, I think, is undesirable and should therefore only be pursued as a last resort, that is, if there is really no other option. Here I will show that there is such an option and, hence, that stretching the definition of the notion of predication is not necessary. The proposal I will give here builds on earlier proposals by SzendrĘi (2004) and Csirmaz (2004) that certain types of verbs, which they dub stress-avoiding verbs, normally do not take neutral, main sentential stress. The general idea is that since stress is assigned “to the leftmost phonological word in the leftmost phonological phrase of the intonational phrase”, the finite verb will be assigned stress if it is not preceded by some other constituent in its intonational phrase (see SzendrĘi 2004 for a more detailed discussion). Here I will assume as a working hypothesis that in principle all finite verbs are stress-avoiding. In other words, there is a constraint NOSTRESS-Vfinite that forbids the assignment of neutral, main sentential stress to finite verbs; see the appendix, sub IIc, for a discussion that places this constraint in a wider cross-linguistic context. (121)

NO-STRESS-Vfinite:

do not assign neutral, main sentential stress to finite verbs.

The constraint NO-STRESS-Vfinite favors placement of a VM, a focused constituent, a wh-phrase, or the negative adverb nem in front of the finite verb; if this does not happen the finite verb is in the position to which stress is assigned so that NO-STRESS-Vfinite would be violated. Note in passing that topic movement does not help to avoid a violation of NO-STRESS-Vfinite since the constituents in the Hungarian topic field are not part of the same

334 Chapter 5: Predicate movement

intonational phrase as the verb, which is clear from the fact that topics can be followed by an intonation break. It is important to stress that assuming NO-STRESS-Vfinite, of course, does not imply that neutral, main sentential stress is never assigned to a finite verb; after all, we are dealing with a violable constraint. When there is nothing that can be placed in the position preceding the finite verb, this verb can be assigned stress. This was illustrated in (117), repeated here as (122); since the ij-features on the subject of an unergative verb cannot value the unvalued features on V, it cannot function as a VM and the verb nevetett ‘laughed’ must be assignedmain sentential stress in this example, because it is the first element in its intonational phrase. Something similar arises with transitive verbs that do not take a verbal particle and cannot take a bare noun as its complement like utál ‘hate’ or sajnál ‘regret’; cf. Komlósy’s (1989) class of stress-requiring verbs. (122) a.

Nevetett egy/a fiú. laughed a/the boy b. *Fiú nevetett. boy laughed

(123) a.

János utálja a spenótot. János hates the spinachacc ‘János hates (the) spinach.’ b. *János spenótot utál. János spinachacc hates

In passing, note that the stress-avoiding verbs discussed by SzendrĘi (2004) are only special in that they are not main verbs; this implies that these verbs are always accompanied by a VM, namely the infinitival main verb, so that it is never necessary to assign main sentential stress to these verbs. The constraint NO-STRESS-Vfinite can be assumed to interact with the two other constraints that are relevant here: EPP(ij), which favors movement of the goal of the ij-features on V, and the economy constraint *MOVE. Since non-bare noun objects cannot undergo this movement, we can safely assume that we are dealing with a weak ranking of EPP(ij) for Hungarian. Suppose the ranking is as given in (124). (124) Hungarian: NO-STRESS-Vfinite >> *MOVE >> EPP(ij)

This ranking predicts that movement of the VMs (which can all value the ij-features on V) is only allowed when it is needed to avoid a violation of NO-STRESS-Vfinite. This, in turn, is only the case when the finite verb is not preceded by a focused constituent, a wh-constituent, or negation. From this

Verbal modifiers in Hungarian 335

the complementarity in distribution between these elements and the VMs follows without further ado.74 What we still need to account for, however, is why it is the focused or interrogative phrase that precedes the finite verb, and not the VM. A key observation in this respect is that the presence of negation does not block focus and wh-movement (cf. (115)), which unambiguously shows that these two movement types are independently needed and not solely motivated by satisfaction of the constraint NO-STRESS-Vfinite. Focus and wh-movement are simply obligatory, which follows from our earlier claim that Full Interpretation forces application of all semantically motivated movements (§ A′-movement); cf. section 1.3.2.1. That negation must precede the finite verb is of course due to the fact that the head of NegP has scope over VP. Since the obligatory focus or wh-movement will satisfy NO-STRESS-Vfinite, it follows that movement of the VM will be blocked by the weak ranking of EPP(ij) in such contexts. Another question is why, for example, wh-movement blocks focus movement of some other constituent. Here I will not discuss questions like this, but will simply conclude that the ranking in (124) provides an adequate solution for the more limited problem we have set out to solve, namely the fact that movement of the VM is possible only if the verb would otherwise end up as the first element in its intonational phrase. 5.4.4.

Conclusion

This section has addressed a limited number of questions pertaining to the distribution of the VMs in Hungarian. I have argued that what these VMs have in common is that they are suitable goals of the ij-features on the verbal root V. The fact that the VMs can only appear in the position preceding the finite verb when this verb is not preceded by a focused constituent, a wh-phrase, or negation follows from the constraint ranking in (124). Since we are dealing with the weak ranking *MOVE >> EPP(ij), movement of the VMs applies only when this avoids a violation of the constraint NO-STRESSVfinite, which outranks *MOVE. This is only the case when the clause does not contain a focused constituent, a wh-phrase, or negation. 74 This proposal does not yet account for the fact that nonverbal VMs must precede the infinitival main verb in complex verb constructions when NO-STRESS-Vfinite is satisfied by negation or focus/wh-movement. In this respect, Hungarian resembles the so-called Verb Projection languages like West Flemish where objects and predicates may permeate the clause-final verb cluster; see Koopman & Szabolcsi (2000), É.Kiss (2004), and, especially, SzendrĘi & Tóth (2004) for extensive discussion of the possible word orders in the Hungarian verb cluster. I leave this issue for future research.

336 Chapter 5: Predicate movement

5.5. L-participle fronting in Bulgarian This section will show that so-called l-participle fronting can also be fruitfully approached by means of the theory developed in this chapter. L-participle fronting refers to the phenomenon in Slavic that the l-participle can, and sometimes must, be placed in front of the auxiliary. This phenomenon is illustrated by means of the Bulgarian examples in (125). (125) a.

b.

Paulina e proþela knigata. Paulina be3SG readSG.fem book-the ‘Paulina has read the book.’ Proþela e knigata. readSG.fem be3SG book-the ‘She has read the book.’

The discussion of l-participle fronting will be organized as follows. Section 5.5.1 will first briefly point out some conspicuous properties of the l-participle. Section 5.5.2 will start the discussion of l-participle fronting by reviewing and slightly adapting Broekhuis & Migdalski (2003) and Migdalski’s (2006) proposal that, contrary to what is normally assumed, l-participle fronting involves XP- and not head-movement. If l-participle fronting is indeed XP-movement, example (125b) must be derived by means of remnant movement; before the VP headed by the l-participle is fronted, the arguments must have been extracted from this VP, in order to avoid their being pied piped by l-participle fronting. If this proposal is on the right track, the word order in (125a) is somewhat surprising: if the object is obligatorily extracted from the VP, we expect it to precede the l-participle. The conclusion that can be drawn from this is that in (125a) the VP headed by the l-participle has also moved leftward, which is in fact independently supported by the fact that the l-participle and the object can be split by sentential adverbs. This implies that the analysis of the examples in (125) is as given in (126). (126) a. b.

Pauline e [VP procela tknigata] knigata tVP [VP Procela tknigata] e t′VP knigata tVP

Section 5.5.3 will suggest that both the short l-participle movement in (126a) and the long l-participle movement in (126b) can be accounted for by adopting the hypothesis in (5). The two types of VP-movements can be seen as instantiations of the two types of movement predicted by the corollary in (7); short l-participle fronting in (124a) is triggered by the unvalued ij-features on the auxiliary e ‘is’, whereas long l-participle fronting in (126b) is triggered by the unvalued ij-features on I.

L-participle fronting in Bulgarian 337

5.5.1.

Some properties of l-participles

Before I start discussing l-participle fronting, I want to review some of the properties of the so-called l-participles (see Migdalski, 2006: ch.2, for a more extensive discussion). First, the main verb in (125) is called an “l-participle” because (just like the past participle) it combines with an auxiliary and its stem ends in /l/: proþel ‘read’. A property of l-participles that will be crucial for the analysis of l-participle fronting to be developed in the following subsection is that l-participles morphologically agree in number and gender with their external argument. The Bulgarian examples in (127) illustrate this for gender; cf. Migdalski (2006:62) for the complete inflectional paradigms in Bulgarian and Serbo-Croatian. (127) a. b.

Ivan e proþel-Ø IvanSG.masc. be3SG readSG.masc. Paulina e proþel-a PaulinaSG.fem. be3SG readSG.fem.

knigata. book-the knigata. book-the

It is important to note that the agreement facts in (127) show that l-participles and past participles are different: l-participles agree with their external argument, whereas we have seen earlier that past participles agree with their internal argument. No doubt, this difference concerning agreement is related to another difference between l-participles and past participles concerning auxiliary selection: l-participles invariably take the auxiliary BE, whereas past participles instead select HAVE (provided they are not unaccusative). Given our earlier assumption that BE, unlike HAVE, is not able to introduce an external argument, this implies that l-participles and past participles differ in their thematic properties. Since BE cannot introduce an external argument, the internal and the external argument must both be base-generated within the extended projection of the l-participle. Since the function of HAVE is to introduce the external argument (as well as to assign accusative case), the extended projection of the Germanic past participle contains only an internal argument; cf. the discussion in section 3.1.2.3. 5.5.2.

Long l-participle fronting

Lema & Rivero (1989) have analyzed l-participle fronting as headmovement. The main reason for this is that the l-participle is not able to pied pipe the direct object or any other VP-internal material. The landing site of the l-participle is claimed to be C. Broekhuis & Migdalski (2003) and, especially, Migdalski (2006) have shown that there are at least three problems with this proposal. First, this analysis implies that head-

338 Chapter 5: Predicate movement

movement crosses the head position occupied by the auxiliary and therefore violates the HMC: in order to allow this violation, Lema & Rivero proposed to derive the HMC-effects from a slightly complicated version of the ECP. The second, and more serious, problem is that this proposal wrongly predicts that the subject can be placed in between the l-participle in the C-position and the auxiliary e in the I-position (or some lower position) of the clause. (128) *[CP [C′ Proþela [IP Paulina [I′ e [ ... knigata]]]]]

The third problem is that this proposal wrongly predicts that l-participle fronting is restricted to main clauses. This is indeed what is claimed by Lema & Rivero, but it appears that there is some variation here, since Embick & Izvorski (1994) give example (129a) as acceptable; another example, taken from Lambova (2002), is given in (129b). Examples like these are also accepted by many of the informants consulted by Broekhuis & Migdalski (2003) and Migdalski (2006). (129) a.

b.

Rasbrah þe proþel beše knigata. understood that readSG.masc. be2SG.past book-the ‘I understood you had read the book.’ Decata kazvat þe gledali sa filma. kids-the say that watchedPL. be3PL.pres movie ‘The kids say that they have seen the movie.’

An alternative head-movement analysis of l-participle fronting is given by Boškoviü (1997). According to him, the l-participle has not moved to C, but adjoined to the auxiliary in I. This proposal has no problems with the HMC, and also correctly predicts the judgments on the examples in (128) and (129). The problem with this account, however, is that it wrongly predicts that the subject may precede the l-participle. (130) *[IP Paulina [I′ proþela+e [ .... knigata]]]

Given these problems it seems reasonable to ask whether it is possible to analyze l-participle fronting as XP-movement. Of course, if so, the problem noted by Lema & Rivero (1989) that the l-participle cannot pied pipe any material still holds, but this is certainly not a unique property of Bulgarian. We have seen in the previous section that something similar holds for the verbal VM in Hungarian: despite the fact that the position preceding the finite verb can be occupied by full phrases, the verbal VM cannot pied pipe any VP-internal material either. Apparently both in

L-participle fronting in Bulgarian 339

Hungarian and in Bulgarian all material (apart from the verb) that is generated VP-internally is obligatorily moved into some VP-external position. Saying that the problem in Bulgarian is far from unique is, of course, not the same as saying that this problem is easy to solve; see the conclusion of section 5.4.2 for some tentative suggestions. Hypothesis (5) predicts l-participle fronting to be possible given that the l-participle enters into an agreement relation with the ij-features of the subject (cf. (127)): the unvalued ij-features on I may therefore attract either the subject or the participle phrase. Furthermore, the claim that l-participle fronting involves movement of the VP headed by the l-participle (henceforth: participle phrase) into SpecIP immediately accounts for the fact that the subject cannot precede the auxiliary. Actually, the subject is normally not morphologically expressed (must be pro) when l-participle fronting applies, a fact to with I will return shortly. Despite the fact that movement of the l-participle in Bulgarian and the VM in Hungarian target different positions (SpecIP and SpecVP, respectively), l-participle fronting has certain features in common with the movement of the VM. Consider the examples in (131), which intend to show that, in some cases at least, l-participle fronting is not truly optional but can only apply when the l-participle has a contrastive/emphatic focus reading: the example must be pronounced with a marked pitch fall following the fronted l-participle (Lambova 2002). (131) a.

b.

PROýELA beše knigata. readSG.fem be3SG.past book-the ‘She had READ the book.’ Beše proþela knigata. be3SG.past readSG.fem book-the ‘She had read the book.’

The fact that (131b) is acceptable shows that l-participle fronting need not apply. This strongly suggests that we are dealing with the weak ranking *MOVE >> EPP(person), which is confirmed by the fact that overtly realized subjects need not be placed in the position preceding the past auxiliary either. This, in turn, suggests that l-participle fronting in (131a) is forced by semantic considerations. This can be formalized by means of the evaluations in Tableaux 19 and 20, which are, however, slightly misleading. For ease of presentation I provisionally assumed the constraint FOCUS, which requires that a contrastively focused phrase be clause-initial. Note, however, that, given our earlier assumption that semantically motivated movement must apply, this constraint is actually not needed as the non-

340 Chapter 5: Predicate movement

optimal candidate in Tableau 20 crashes due to a violation of Full Interpretation, and is therefore actually not part of the candidate set; to indicate this I shaded the row containing this candidate. Note further that since object movement from the participle phrase takes place in all examples, I have not indicated the violation of *MOVE that results from that movement, for convenience. Tableau 19: L-participle fronting with the past tense (non-clitic) auxiliary beše

l-participle ⊄ focus a. Proþelai beše knigata ti b. Beše proþela knigata

FOCUS

*MOVE *!

)

EPP(person)

*

Tableau 20: L-participle fronting with the past tense (non-clitic) auxiliary beše

l-participle ⊂ focus a. PROýELAi beše ti knigata b. Beše PROýELA knigata

FOCUS

) *!

*MOVE *

EPP(person)

*

The examples in (132) show that (in the absence of a phonetically realized subject) l-participle fronting must apply when we are dealing with a present tense auxiliary: these auxiliaries are special in that they cannot be the first element in the clause, but must be phonologically supported by something on their left, for which reason they are often called clitic auxiliaries. In this respect, present auxiliaries like e differ from their past tense counterparts like beše in (131), which are not clitics and can be the first element of their clause. (132) a.

Proþela e knigata. readSG.fem be3SG.pres. book-the ‘She has read the book.’ b. *E proþela knigata.

The fact that l-participle fronting in (132) is obligatory can therefore be attributed to the fact that the clitic auxiliaries must be phonologically supported from the left, which can be formally expressed by postulating the interface constraint ENCLITIC in (133a); see the appendix, sub IIc, for a discussion that places this constraint in a wider cross-linguistic context. When we assume the ranking in (133b), the evaluations of the examples in (132) are as given in Tableaux 21 and 22. As before, the second candidate in Tableau 22 is shaded because it may simply violate Full Interpretation and therefore not be part of the candidate set.

L-participle fronting in Bulgarian 341 (133) a. b.

ENCLITIC: an enclitic element is phonologically supported on the left. Bulgarian: {ENCLITIC, FOCUS} >> *MOVE >> EPP(person)

Tableau 21: L-participle fronting with the present tense (clitic) auxiliary e

l-participle ⊄ focus Proþelai e ti knigata ) E proþela knigata

ENCLITIC

FOCUS

*MOVE *

EPP(person)

*!

*

Tableau 22: L-participle fronting with the present tense (clitic) auxiliary e

l-participle ⊂ focus PROýELAi e knigata ti ) E PROýELA knigata

ENCLITIC

FOCUS

*>

knigata. think1SG that read beAUX.3SG.pres. book-the ‘I think that he had READ the book.’ Mislja þe PROýEL e knigata.

Tableau 24: L-participle fronting with a clitic auxiliary preceded by phonological material

l-participle ⊄ focus þe proþeli e ti knigata þe e proþel knigata )

ENCLITIC

FOCUS

*MOVE *!

EPP(person)

*

Tableau 25: L-participle fronting with a clitic auxiliary preceded by phonological material

l-participle ⊂ focus þe PROýELi e ti knigata ) þe e PROýEL knigata

ENCLITIC

FOCUS

*!

*MOVE *

EPP(person)

*

75 For unknown reasons, negation not only voids the need for participle fronting in (134), but actually blocks it: the examples in (134b&b′) are also unacceptable under the marked intonation pattern. The fact that negation also blocks English locative inversion suggests that we are dealing with a more general phenomenon, which might be related to the fact that sentence negation creates weak islands; cf., e.g., Cinque (1990) and Rizzi (1990).

344 Chapter 5: Predicate movement

5.5.3.

Some speculations on short l-participle fronting

Section 5.5.2 has argued that l-participle fronting involves XP-movement. This implies that example (136b) is derived by means of remnant movement of the participle phrase: if the object does not evacuate the participle phrase before l-participle fronting applies, it would be wrongly predicted to be pied piped. Given this proposal, the word order in (136a) is perhaps somewhat surprising: if the object is extracted from the participle phrase, we expect it to precede the l-participle. (136) a.

b.

Paulina e proþela knigata. Paulina be3SG readSG.fem. book-the ‘Paulina has read the book.’ Proþela e knigata. readSG.fem. be3SG book-the ‘She has read the book.’

One possible way to account for the unexpected word order in (136a) is to assume that this example is derived by a “short” version of l-participle fronting, which moves the participle phrase across the shifted object into the position following the auxiliary, as in (137a). Evidence in favor of short l-participle fronting is provided by the fact that the l-participle and the object can be split by high (sentential) adverbs: since these adverbs are normally assumed to occupy a position external to vP, it is not likely that the l-participle can cross these adverbs by head-movement; this is normally restricted to finite verbs that move to T. The fact that the auxiliary and the l-participle are always adjacent when the former precedes the latter further suggests that short l-participle movement is obligatory in Bulgarian. If so, this implies that the structure of (136b) is given as in (137b): long l-participle fronting does not proceed in one fell swoop but in two movement steps. (137) a. b.

Pauline e [PartP proþela tknigata] knigata tPartP [PartP Proþela tknigata] e t′ PartP knigata tPartP

The next question that we have to ask is what the landing site of short l-participle fronting is. Given that the auxiliary normally occupies the second position of the clause, it seems rather safe to assume that the auxiliary has undergone V-to-I. This makes it possible to assume that short l-participle fronting is triggered by the ij-features on the auxiliary, and thus targets the specifier of auxP. In order to derive example (137a), we must assume that the inversion of the auxiliary and the PartP that results from

L-participle fronting in Bulgarian 345

short l-participle fronting is subsequently undone by V-to-I of the auxiliary. This gives rise to the representations in (138).76 (138) a. b.

Pauline I+e [auxP [PartP proþela tknigata] taux [... knigata tPartP]] [PartP Proþela tknigata] I+e [auxP t′ PartP taux [... knigata tPartP]]

Although this will lead me even further into the realm of speculation, let me try to make the derivation a bit more precise. First note that the analysis suggested in (138), according to which short l-participle fronting is triggered by the ij-features on the auxiliary, is only possible due to the fact established earlier that the l-participle agrees with the subject of the clause. Hypothesis (5) therefore predicts that PartP can also be attracted by the ij-features on the auxiliary (cf. the discussion of VP-movement in Dutch in section 5.3.2.2) as well as the ij-features on the inflectional head I; cf. the corollary in (7a). From the fact that the l-participle agrees with the subject of the clause we can also conclude that the subject is generated within the extended projection of the l-participle, which must therefore include some light verb v. If the main verb is transitive, we can safely assume that the light verb has a case feature that enters into an Agree relation with the object of the participle. When we further assume that this case feature on v* actually triggers movement of the object, the structure, after merger of the auxiliary e, will be as given in (139a). Given this structure, we can account for the fact that short l-participle fronting strands the object by assuming that lparticiple fronting involves a lower extended projection of the participle, that is, PartP or the lower segment of the v*P. (139) a.

[auxP e [v*P knigata v* [vP Paulina/pro tv [PartP proþela tknigata]]]]

Since the subject and the participle phrase constitute some sort of small clause, and we have seen in our discussion of English locative inversion that predicate movement cannot strand the subject of the small clause, the most plausible assumption is that it is the lower v*P segment that is moved. This has the additional advantage that we can also account for the fact that we cannot satisfy EPP(gender) by moving the subject by calling upon 76

Alternatively, we may assume that the auxiliary does not move into I, but that the original order between the auxiliary and the l-participle is restored by moving the verbal root of the auxiliary into v: this may give us a better understanding of the fact that the auxiliary and the l-participle must be adjacent and cannot be split by the sentential adverbs which can be assumed to be adjoined to vP (or higher). Krzysztof Migdalski (p.c.) must be acknowledged here for bringing various potential problems involving adverb placement to my attention.

346 Chapter 5: Predicate movement A-OVER-A; if this constraint outranks ALIGNFOCUS, movement of the participle phrase will always be preferred. From this it will follow that we derive the structure in (139b).

(139) b. [auxP [vP Paulina/pro tv [PartP proþela tknigata]] e [vP knigata v* tvP]]

Given that l-participle fronting is obligatory, we may conclude that EPP(gender) outranks *MOVE. However, the fact that subject shift and long l-participle fronting are optional indicates that we have a weak ranking of EPP(person) and EPP(case). This weak ranking is overruled in the case of the clitic auxiliaries by the constraint ENCLITIC (and in the case of the nonclitic auxiliaries by considerations of convergence, that is, Full Interpretation). This gives rise to the ranking in (140). (140) Bulgarian: {ENCLITIC, EPP(gender)} >> *MOVE >> {EPP(person), EPP(case)}

This ranking, of course, creates at least one important problem, because the weak ranking of EPP(case) goes against our earlier assumption that v* obligatory attracts the object. We can perhaps solve this problem by assuming that, just as in Hungarian, this weak ranking is overruled by some constraint that requires that the moved participle phrase contain as little lexical material as possible. Let us assume that this is the case. The derivation is continued by adding the inflectional head I, and V-toI of the auxiliary; cf. the discussion of (138) above. This results in the structure in (139c). (139) c. [IP I+e [auxP [vP Paulina/pro tv [PartP proþela tknigata]] taux [knigata v* tvP]]]

From here on the continuation of the derivation will depend on various factors. When the clitic auxiliary is supported from the left by other overt material, movement of the subject/PartP into SpecIP is not needed and is therefore blocked by the weak ranking of EPP(person) and EPP(case). This results in the structure in (139d). When this is not the case, either the subject or the participle phrase must move into SpecIP in order to avoid a violation of ENCLITIC. If the subject is phonetically realized, as in (139d′), it can be attracted by the case feature on I without invoking a violation of A-OVER-A given that the extended projection of the l-participle is not a potential goal for this feature: this option will furthermore be preferred, given that movement of the subject will satisfy more EPP constraints than the extended projection of the l-participle. Finally, when the subject is pro, as in (139d′′), ENCLITIC can only be satisfied by applying long l-participle fronting.

L-participle fronting in Bulgarian 347 (139) d. X(P) [IP I+e [auxP [vP Paulina/pro tv [PartP proþela tknigata]] taux [knigata v* tvP]]] d′. [IP Paulina I+e [auxP [vP tPaulina tv [PartP proþela tknigata]] taux [knigata v* tvP]]] d′′. [IP [vP pro tv [PartP proþela tknigata]] I+e t′vP [auxP taux [knigata v* tvP]]]

The analysis suggested above has the important advantage that it treats both long and short l-participle fronting as A-movement since they are both triggered by ij-features (on I and aux, respectively). This is an important result because if short l-participle movement were the result of A′movement, one would expect long l-participle fronting to be blocked as a case of improper movement. If the analysis suggested above is on the right track, this shows that Bulgarian has both types of predicate movement in (7), and thus provides indirect support for Axiom I in (4a). 5.6. Conclusion This chapter has explored the two axioms in (4) and hypothesis (5) and has shown how these may help to account for a variety of constructions: locative inversion in English, predicate movement and verb raising in Dutch, movement of verbal modifiers in Hungarian, and l-participle fronting in Bulgarian. If the axioms in (4) and hypothesis (5) are on the right track, they will no doubt play a role in a much larger set of phenomena. One case that readily comes to mind, but which I have not investigated, concerns the fact that the set of French/Italian clitics is so peculiarly constituted. French clitics, for example, are virtually all nominal in nature: French has a full set of pronominal clitics like le/lui ‘him/her’ and the partitive clitic en, which is also nominal in nature. There is, however, one exception: the locational clitic y meaning “there”, “in it” “on it”, etc. This raises the question as to why there is a locational clitic, but not, e.g., a temporal one. Hypothesis (5) now suggests where to look for an answer: both pronominal and locational clitics are characterized by the fact that they have ij-features, which in turn suggests that clitic movement is triggered by the ij-features on V or I. Given Moro’s 1997 claim that expletive constructions involve locative inversion, this may also shed new light on the French expletive il y a ... ‘there is ...’ construction. This example shows that this section has by no means exhaustively discussed all ramifications of the axioms in (4) and hypothesis (5).

Chapter 6 Summary and conclusion

This study has provided an extensive investigation of object shift in the Germanic languages within the derivation-and-evaluation (D&E) framework outlined in Chapter 1. The core idea embodied in this framework is that, in order to develop an explanatorily and descriptively adequate theory of syntax, restrictions must be formulated on both syntactic derivations and the resulting syntactic representations. This is achieved by assuming that representations created by some version of the computational system of human language CHL from the minimalist program are evaluated in an optimality-theoretic fashion, as indicated in Figure 1. Input

Generator

Output representations

Evaluator

Optimal output

Figure 1: The derivation-and-evaluation model

In Chapter 2 it was claimed that there are in fact two types of object shift: short and regular. Short object shift is triggered by the ij-features on the verbal root V and may cross the VP-adverbials adv, as in (1). It was further claimed that this type of object shift applies obligatorily in all Germanic languages, and that the distinction between the OV- and VO-languages can be accounted for by assuming that in the latter, the inversion of the verbal root V and its complement OBJ that results from short object shift is undone by subsequent V-to-v. (1)

• Short object shift a. [... v[case] [OBJ ...adv V tOBJ]] b. [...V+v[case] [OBJ ... adv tV tOBJ]]

(OV-languages) (VO-languages)

Chapter 3 provided an analysis of regular object shift. This movement is triggered by the case features on the light verb v and may cross the sentential adverbs ADV, as in (2). In the VO-languages regular object shift cannot cross the main verb, which implies that this verb must be moved into the I-position for regular object shift to be possible. This fact is normally referred to as Holmberg’s Generalization (HG).

350 Chapter 6: Summary and conclusion (2)

• Regular object shift a. [OBJi ... ADV v [t′OBJ ... V tOBJ]] b. [OBJi ... ADV V+v [t′OBJ ... tV tOBJ]] b′. [...V+v+I [OBJi ..tv [t′OBJ ... tV tOBJ]]]

(OV-languages) (VO-languages)

Section 3.1 began with an extensive discussion of HG, and argued that this generalization cannot be attributed to some inviolable condition on the operations of CHL. Section 3.2 therefore developed an optimality-theoretic account of the restrictions on regular object shift in Table 1. Table 1: Restrictions on the application of regular object shift

FinnishSwedish English Swedish Norwegian Danish Icelandic Dutch German

weak pronouns —

DPs

— optional optional obligatory obligatory obligatory obligatory

— — — — optional optional optional

—

across across IO V n.a. n.a. n.a. — — — — — +

n.a. — — — — + +

across nonverbal particles n.a. n.a. — + + + + +

The constraints that enter the analyses of short and regular object shift, as well as those of subject shift and verb movement, are given in (3). In the summary below I will focus on the six constraints that are essential in accounting for the facts indicated in Table 1: *MOVE, EPP(case), D-PRONOUN, ALIGNFOCUS, RELMIN and H-COMPL. (3)

a.

b.

CHL constraints: (i) Economy constraints: *MOVE, NOLEXM, NOVACM (ii) EPP constraints: EPP(ij), EPP(case), *STRAY FEATURE Interface constraints: (i) PF constraints: RELMIN, H-COMPL (ii) LF constraints: ALIGNFOCUS, D-PRONOUN

The relative ranking of *MOVE and EPP(case) distinguishes object shift languages from non-object shift languages: ranking (4a) expresses that the unvalued case feature on v (normally) does not trigger regular object shift; ranking (4b), on the other hand, expresses that this feature (normally) does trigger regular object shift.

Summary and conclusion 351 (4)

a. b.

*MOVE >> EPP(case): object shift is (normally) blocked. EPP(case) >> *MOVE: object shift (normally) applies.

The languages defined by the ranking in (4a) can be subdivided by means of the relative ranking of D-PRONOUN and *MOVE: when the former outranks the latter, the language allows object shift of weak pronouns, whereas languages that have the opposite ranking prohibit it. (5)

a. b.

D-PRONOUN >> *MOVE: object shift of pronouns is (normally) allowed. *MOVE >> D-PRONOUN: object shift of pronouns is (normally) blocked.

The rankings in (4) and (5) can be seen as the formalization of two macroparameters that distinguish the language types in Figure 2.

*MOVE >> EPP(case)

No full object shift

EPP(case) >> *MOVE Full object shift: Icelandic

D-PRONOUN >>*MOVE

*MOVE >>D-PRONOUN

Pronoun shift: Danish

No object shift: Romance

Figure 2: Macro-parameterization of languages with respect to regular object shift

The constraint ALIGNFOCUS accounts for the fact that in the languages with full object shift, object shift may be “optional”: when we assume that this constraint outranks EPP(case), we predict that object shift is restricted to objects that are part of the presupposition of the clause. The constraints RELMIN and H-COMPL will block object shift across respectively the indirect object and the main verb provided that they are ranked above the constraints that favor regular object shift, EPP(CASE) and D-PRONOUN. H-COMPL also enters the analysis of the fact that in Swedish object shift cannot cross the verbal particle. The analyses of (short and regular) object shift raise various new questions, which are addressed in Chapter 4. Section 4.1, for example, observes that there are in principle two ways to satisfy Holmberg’s Generalization (=the constraint H-COMPL) in languages that normally do not have V-to-I in embedded clauses: the first way is to block object shift in these cases, and the second is for object shift to force V-to-I. The question must therefore be raised as to why it is the first option that is chosen in languages like Danish and Swedish. Section 4.2 observes that the analyses of object shift make explicit predictions about movement of the subject, and

352 Chapter 6: Summary and conclusion

investigates these. Section 4.3 discusses the fact that it is occasionally claimed that object shift interacts in intricate ways with A′-movement, and argues that these interactions are only apparent and follow from the fact that for various reasons A′-movement is generally preceded by the A-movement operations that can potentially apply. Chapter 5 is more speculative in nature and aims at exploiting the constraints involved in the analysis of object shift a bit further by claiming that EPP(ij) not only triggers short object shift, but also triggers predicate movement. This chapter is programmatic in nature and illustrates a number of new avenues of research that are opened up by the D&E approach developed in this study. The fact that we are able to account for all the core facts concerning object shift in the Germanic languages by means of a framework that supplements the computational system CHL with a small number of OTconstraints (which all find their origin in the P&P framework) in my view strongly supports the D&E claim that an explanatorily and descriptively adequate theory of syntax can only be obtained by formulating restrictions on both the syntactic derivations and the resulting syntactic representations.

Appendix: Constraints used in this study

Section 1.3.2 has discussed the claim that the OT-evaluator in the D&E framework is a formalization of the interface conditions postulated in MP. If so, we expect the syntactic constraints in CON to be somehow related to the three components involved: the computational system CHL, which creates the relevant syntactic representations in the candidate set, and the two interpretive systems that interpret them, namely the articulatoryperceptual and the conceptual-intentional components. It has therefore been assumed that the syntactic constraints in CON can be divided into the basic types shown in (1a) and (1b). (1)

• The syntactic constraints in CON are of two basic types: a. CHL constraints (i) EPP constraints (ii) Economy constraints b. Interface (LF and PF) constraints

I. The CHL constraints The constraints related to the computational system in (1a) come in two subtypes: the EPP constraints, which favor the application of the movement operations that can be triggered by the unvalued formal features, and the economy constraints that disfavor these movements. The relative ranking of the EPP and the economy constraints determine whether the movement does or does not apply under “normal” circumstances: when an EPP constraint outranks the economy constraint *MOVE the ranking is “strong” in the sense that the movement favored by the EPP constraint normally applies; when *MOVE outranks the EPP constraint the ranking is “weak” in the sense that the movement favored by the EPP constraint does not normally apply. A. The EPP constraints The EPP constraints favor application of movements that satisfy the Last Resort Condition. In principle we could assume that there are as many EPP constraints as there are unvalued features, but I have argued in section 1.3.2.1 that [+affect] features like [wh] or [neg] do not have corresponding EPP constraints, given that wh- or neg-movement may simply be universally required for reasons of LF-legibility: output structures in which these

354 Appendix: Constraints used in this study

movements did not apply may simply fail to receive an interpretation in the conceptual-intentional component. If this is indeed the case, it leads to the conclusion that only the uninterpretable formal features have corresponding EPP constraints. The EPP constraints that are used in this study are given in (2). (2)

a.

b. c.

EPP(ij): unvalued ij -features attract their goal. (i) EPP(gender) (ii) EPP(person) (iii) EPP(number) EPP(case): unvalued ij-features attract their goal. *STRAY FEATURE: amalgamate formal features of the functional heads with the root they are associated with.

EPP(ij).

If this constraint outranks the economy constraint *MOVE, we predict that the ij-features on the verbal root V trigger short object shift, and that the ij-features on I trigger movement of the subject. Since languages like Dutch arguably have obligatory short object shift but optional subject movement, it was argued in section 4.2 that EPP(ij) must actually be divided into the three separate constraints in (2a,i-iii): by following the more or less standard assumption that the verbal root V has a gender but not a person feature and that I has a person but not a gender feature, the behavior of Dutch can be accounted for by assuming the subranking EPP(gender) >> ALIGNFOCUS >> EPP(person) >> *MOVE; see the discussion of ALIGNFOCUS below. EPP(case). If this constraint outranks the economy constraint *MOVE, we predict that the case feature on the light verb v normally triggers regular object shift, and that the case feature on I normally triggers movement of the subject.

*STRAY FEATURE. This constraint requires, among other things, that the verbal features on the functional heads in the extended projection of the verb attach to the verb. If this constraint outranks the economy constraint *MOVE, we predict that the verbal features trigger V-to-v/Asp and v-to-I. The findings in this study did not indicate a need to divide this constraint into more specific constraints that refer to the individual features on the functional heads that are normally distinguished, so we may hope that it is not necessary to postulate separate constraints like EPP(v), EPP(aspect) or EPP(tense).

Constraints used in this study 355

B. The economy constraints The two basic economy constraints are *SELECT and *MOVE, which block external and internal merge respectively. (3)

a. b.

*MOVE (STAY): don’t apply internal merge. *SELECT: don’t apply external merge.

*MOVE: This constraint is used to derive the effects of the MP economy condition that prohibits movement if no EPP/edge-feature is present. The relative ranking of *MOVE and the EPP constraints determines whether a certain movement that satisfies the Last Resort Condition does or does not apply in the “normal” case. *SELECT. Contrary to *MOVE, *SELECT does not seem to interact with the EPP constraints, due to the fact that external merge is normally required for reasons of LF-legibility/interpretability. Section 1.3.3 has argued, however, that *SELECT is nevertheless needed in order to account for the fact that languages like English seem to favor internal over external merge, whereas languages like Dutch seem to favor external over internal merge. This study has shown that it is necessary to assume at least the following three more specific economy constraints next to *MOVE. These constraints only have an effect on the selection of the optimal candidate when they outrank *MOVE. They may, for example, override a strong ranking of the EPP constraints by blocking movements of certain designated elements (NOLEXM) or when the movement results in a disfavored configuration (NOVACM and A-OVER-A). (4)

a. b. c.

NO LEXICAL MOVEMENT (NOLEXM): a ș-role assigning head remains in its ș-domain (a main verb does not move). NO VACUOUS MOVEMENT (NOVACM): don’t apply internal merge when it is not visible in the output. A-OVER-A: *[FP X(P) P ... [YP ... tXP ...]] if YP and X(P) are both potential goals of probe P.

(NOLEXM). This constraint was proposed by Grimshaw (1997) and goes back at least to Pollock (1989); its role is to block movement of main verbs while still allowing movement of auxiliary and modal verbs, and it is used to account for the fact that in English V-to-I is restricted to modal and auxiliary verbs, that is, cannot apply to main verbs. It is not clear to me whether this constraint has a counterpart in

NO LEXICAL MOVEMENT

356 Appendix: Constraints used in this study

current MP, because I am not aware of any recent MP accounts of this dichotomy between main and auxiliary/modal verbs. NO VACUOUS MOVEMENT (NOVACM). This

constraint is a direct descendant of Chomsky’s (1986) Vacuous Movement Hypothesis, and is akin to the more recent minimalist economy condition that the application of movement must have an effect on the outcome of the derivation (see Chomsky 1995a:294, Chomsky 2001, and Sabel 2005).

A-OVER-A. This constraint is a descendant of the A-over-A condition on transformations, and favors attraction of the least embedded potential goal; it can be seen as an economy constraint since it favors shorter over longer paths. A-OVER-A is introduced in Chapter 5 in order to account for the fact that in some cases the ij-features on V/I attract a predicative phrase rather than the subject of that phrase. II. Interface constraints The typical effect of highly ranked interface constraints is that they override strong and weak rankings of the EPP constraints. For convenience, I will distinguish three groups. The first two groups are central to our analysis of object shift, whereas the third group was proposed in order to solve more specific problems that arose along the way. A. Semantically/phonologically motivated constraints That highly ranked interface constraints may override strong and weak rankings of the EPP constraints clearly holds for the interface constraints in (5), which are crucial for our account of regular object shift and V-to-I. (5)

a. b. c.

ALIGNFOCUS (AF):

the prosodically unmarked focus is the rightmost constituent in its clause. D-PRONOUN: a weak/definite pronoun must be vP-external: *[vP ... pron[+def] ...]. LEXICALLY FILL TOP F (LFTF): the highest head position in an extended projection must be lexically filled.

ALIGNFOCUS (AF). This constraint can override a strong ranking of, e.g., EPP(ij) or EPP(case) with the result that the movement triggered by the ij- or

case features becomes sensitive to the information structure of the clause: the movement is blocked when the goal of the probe is part of the focus of the clause. This constraint makes filters of the type proposed in Chomsky (2001) to account for Icelandic object shift superfluous. It is important to

Constraints used in this study 357

keep in mind that ALIGNFOCUS is a so-called gradient constraint: a violation of this constraint results for each constituent that follows the element in focus. D-PRONOUN. This constraint can override a weak ranking of, e.g., EPP(ij) or EPP(case) with the result that the movement triggered by the ij- and case

features become sensitive to the type of object involved: the movement is allowed when the goal of the probe is a weak/definite pronoun, but not when it is a non-pronominal DP. LEXICALLY FILL TOP F (LFTF). This constraint can override a weak ranking of *STRAY FEATURE. In this study this constraint is used to derive

the asymmetry in languages like Dutch and Danish between embedded and main clauses with respect to verb second: LEXICALLY FILL TOP F favors verb movement when no lexical complementizer is present, as is normally the case in main clauses. B. Order preservation constraints The constraints in (6) favor preservation of the underlying order of arguments subject-indirectobject-direct object, and the underlying headcomplement order. (6)

a.

b.

RELATIVIZED MINIMALITY (RELMIN): if the foot of X-chain Į c-commands the foot of X-chain ȕ, the head Į c-commands the head of ȕ (X-chain is A-, A′-, or head chain). HEAD-COMPLEMENT (H-COMPL): a head precedes all terminals originally dominated by its complement.

RELATIVIZED MINIMALITY (RELMIN). The constraint RELMIN reinterprets Rizzi’s (1990) Relativized Minimality as a constraint on output representations: movement of a potential goal G2 into the local domain of a probe P across another potential goal G1 of P is disfavored only when this leads to a reordering of G1 and G2 in the output representation. HEAD-COMPLEMENT (H-COMPL). The

constraint H-COMPL disfavors output representations in which the underlying order head-complement is inverted. Given Kayne’s (1994) Linear Correspondence Axiom, one might expect that there is also a constraint SPEC-H that favors output representations in which heads follow all terminals dominated by their specifier. This was not investigated in this study, but if this is indeed the case, the three order preservation constraints SPEC-H, H-COMPL, and RELMIN may conspire such

358 Appendix: Constraints used in this study

that, among other things, they derive the effects of Koeneman’s (2006) principle of Thematic Shape Conservation (TSC), which requires that within the thematic domain of the clause the underlying order of the thematic categories (=ș-role assigners and assignees) is maintained. As Koeneman points out, the TSC (and therefore also the present approach) immediately accounts for the fact that OS can be blocked by thematic categories, but not by adverbs. This voids the argument found in Holmberg (1999) and Bobaljik (2002) in favor of the claim that adverbs are PFinvisible, and for this reason do not block regular object shift. Section 3.2.2 and section 4.1 have shown that there are in principle two strategies that can be used to avoid potential violations of H-COMPL and RELMIN in the case of object shift: (i) the higher co-argument/main verb may block object shift, or (ii) object shift pushes the higher coargument/main verb up into some higher position (provided that this is allowed by the Last Resort Condition on movement). Which strategy is chosen depends on the relative ranking of the constraints ALIGNFOCUS and D-PRONOUN. C. More peripheral constraints The remaining constraints that are used in this study are proposed to solve certain smaller problems that arose along the way. Given this, there is a fair chance that these constraints will turn out to be of a more or less ad hoc nature. Nevertheless, I have the impression that these constraints, once they are properly phrased, may have a much wider application. For example, the two constraints in (7), which are proposed in the discussion of Hungarian and Bulgarian in Chapter 5, have an effect similar to the constraints in IIA above. They are special, however, in that they favor movement for altruistic reasons: it is some special property of the probe that motivates the overriding of a weak ranking of the EPP constraints. (7)

a. b.

NO-STRESS-Vfinite:

do not assign neutral, main sentential stress to finite verbs. ENCLITIC: an enclitic element is phonologically supported on the left.

Since the two constraints in (7) were both used in order to account for the fact that a finite verb must be preceded by some element, I hope that at some point these constraints can be unified, and used to give a more insightful account of verb second and other second position phenomena. Because I have not studied this possibility, I will not speculate on this here

Constraints used in this study 359

but will instead confine myself to referring to Anderson (2000) for similar ideas. Unfortunately, Anderson did not elaborate these ideas and, to my mind, gave up the project of deriving verb second by taking recourse to the stress properties of finite verbs too quickly. The constraint SCOPE in (8) was proposed in section 2.4.1.3 in order to account for certain word order restrictions between scope bearing objects and VP-adverbs. (8)

SCOPE: relative scope of quantifiers corresponds to the hierarchical order of their topmost A-positions.

The fact that SCOPE is violable enables us to account for the fact that there need not be isomorphism between semantic and syntactic representations. Isomorphism arises only when syntax is able to derive the relevant syntactic structure. In this way we may be able to do away with Quantifier Raising and derive in the domain of quantifier scope what Williams (2003) refers to as PƗnҚini’s Principle. Other interface constraints that were briefly mentioned in Chapter 1 are LEFT EDGE (CP), TELEGRAPH, and RECOVERABILITY. Given that these constraints do not play a role in this study I refer to Pesetsky (1997;1998) and Broekhuis & Dekkers (2000) for relevant discussion. Two constraints that are normally considered to be at the heart of OT-phonology, and which are conspicuously absent from the review above, are the faithfulness constraints MAX(X) and DEP(X). Section 1.3.3 has argued that these constraints are not part of OT-syntax, and suggested that this difference between OT-phonology and OT-syntax is due to the fact that, whereas the phonological representations are directly related to the articulatoryperceptual system, there is no direct link between the syntactic representations and the performance systems.

Bibliography

Ackema, Peter, Ad Neeleman and Fred Weerman 1992 Deriving Functional Projections. In Proceedings of the North East Linguistic Society 23, ed. Amy J. Schafer, 17-31, University of Ottawa. Anagnostopoulou, Elena 2003 The syntax of ditransitives. Berlin/New York: Mouton de Gruyter. 2005 Holmberg's Generalization and Cyclic Linearization. Remarks on Fox and Pesetsky. Theoretical Linguistics 31: 95-110. Anderson, Stephen R. 2000 Towards an optimal account of second-position phenomena. In Optimality Theory: phonology, syntax and acquisition, eds. Joost Dekkers et al. Oxford, Oxford University Press. Archangeli, Diana 1997 Optimality Theory. An introduction to linguistics in the 1990s. In Optimality theory: an overview, eds. Diana Archangeli & Terence Langendoen, 134-170. Malden/Oxford, Blackwell. Baker, Mark to appear The Syntax of Agreement and Concord. Cambridge: Cambridge University Press. Baltin, Mark R. 1987 Do antecedent contained deletions exist? Linguistic Inquiry 18: 579-595. Barbiers, Sjef 1995 The syntax of interpretation, HIL dissertations 14, University of Leiden: Ph.D. dissertation. 2000 The right periphery in SOV languages: English and Dutch. In The derivation of VO and OV, ed. Peter Svenonius, 45-67. Amsterdam/ Philadelphia, John Benjamins. 2005 Word-order variation in three-verb clusters and the division of labour between generative linguistics and sociolinguistics. In Syntax and variation. Reconciling the biological and the social, eds. Leonie Cornips & Karen P. Corrigan, 233-264. Amsterdam/Philadelphia, John Benjamins. Belletti, Adriana 2006 (Past) Participle Agreement. In The Blackwell companion to syntax, eds. Martin Everaert & Henk van Riemsdijk. Malden (US)/Oxford (UK), Blackwell.

362 Bibliography Bennis, Hans 1986 Gaps and dummies. Dordrecht: Foris. Bennis, Hans and Teun Hoekstra 1984 Gaps and parasitic gaps. Linguistic Review 4: 29-87. Birner, Betty 1995 Pragmatic constraints on the verb in English inversion. Lingua 97: 233-356. Bobaljik, Jonathan David 1995 Morphosyntax: the syntax of verbal inflection, MIT: Ph.D. Dissertation. 2002 A-chains at the PF-interface: copies and 'covert' movement. Natural Language & Linguistic Theory 20: 197-267. 2003 Floating Quantifiers: handle with care. In The second Glot International state-of-the article book. The latest in Linguistics, eds. Lisa Cheng & Rint Sybesma, 107-148. Berlin/New York, Mouton de Gruyter. Boeckx, Cedric 2000 Quirky Agreement. Studia Linguistica 54: 354-380. Boeckx, Cedric and Sandra Stjepanovic 2001 Head-ing toward PF. Linguistic Inquiry 32: 345-355. Boškovic, Željko 1997 The syntax of non-finite complementation. An economy approach. Cambridge (Mass.): MIT Press. Bresnan, Joan 1994 Locative Inversion and the architecture of Universal Grammar. Language 70: 72-131. 2000 Optimal Syntax. In Optimality Theory: phonology, syntax and acquisition, eds. Joost Dekkers et al. Oxford, Oxford University Press. Brody, Michael 1990a Some remarks on the focus field in Hungarian. In UCL working papers in Linguistics 2. London, University College of London. 1990b Remarks on the order of elements in the Hungarian focus field. In Approaches to Hungarian, vol.3: structures and arguments, ed. István Kenesei. Szeged, JATE. 1995 Focus and checking theory. In Approaches to Hungarian 5, ed. István Kenesei. Szeged, JATE. Broekhuis, Hans 1992 Chain-government: issues in Dutch syntax, HIL dissertations 2, University of Amsterdam/HIL: Ph.D. dissertation. 1996 On heads and the Linear Correspondence Axiom. In Linguistics in the Netherlands 1996, eds. Crit Cremers & Marcel Den Dikken, 25-36. Amsterdam/Philadelphia, John Benjamins. 1997a Nogmaals Verb Projection Raising. Tabu 27: 1-27. 1997b Recente ontwikkelingen in de generatieve syntaxis I-III. Gramma/ TTT 6: 13-39, 129-150, 219-242.

Bibliography 363 1999

Theoretische aspecten van de generatieve syntaxis I-II. Gramma/TTT 7: 3-23, 89-116. 2000 Against feature strength: the case of Scandinavian object shift. Natural Language and Linguistic Theory 18: 673-721. 2002 Adpositions and adpositional phrases. Tilburg: University of Tilburg, Models of Grammar. 2005 Locative Inversion in English. In Linguistics in the Netherlands 2005, eds. Jenny Doetjes & Jeroen Van der Weijer, 49-60. Amsterdam/ Philadelphia, John Benjamins. 2006a Derivations (MP) and Evaluations (OT). In Optimality Theory and Minimalism: A Possible Convergence? Linguistics in Potsdam 25, eds. Hans Broekhuis & Ralf Vogel, 137–193. University of Postdam: http://www.ling.uni-potsdam.de/lip. 2006b The universal base hypothesis: VO or OV? In Linguistics in the Netherlands 2006 eds. Jeroen Van de Weijer & Bettelou Los, 28-39. Amsterdam/Philadelphia, John Benjamins. 2007a Subject shift and object shift. Journal of Comparative Germanic Linguistics 10: 109-141. 2007b Does Defective Intervention Exist? In Linguistics in the Netherlands 2007, eds. Bettelou Los & Marjo Van Koppen, 49-61. Amsterdam/ Philadelphia, John Benjamins. Broekhuis, Hans and Leonie Cornips 1994 Undative constructions. Linguistics 32: 173-190. Broekhuis, Hans and Joost Dekkers 2000 The minimalist program and optimality theory: derivations and evaluations. In Optimality Theory: phonology, syntax and acquisition, eds. Joost Dekkers et al., 386-422. Oxford/New York, Oxford University Press. Broekhuis, Hans and Veronika HegedĦs to appear Predicate movement in Dutch and Hungarian. Lingua. Broekhuis, Hans, Evelien Keizer and Marcel Den Dikken 2003 Nouns and noun phrases. Unpublished Ms., Modern grammar of Dutch. Occasional papers 4. Tilburg. Broekhuis, Hans and Wim Klooster 2001 On Merge and Move/Attract. In Progress in Grammar. Articles at the 20th Anniversary of the Comparison of Grammatical Models Group in Tilburg, eds. Marc van Oostendorp & Elena Anagnostopoulou. Electronic publication: http://www.meertens.knaw.nl/books/ progressingrammar. to appear Merge and Move as costly operations. Groninger Arbeiten zur germanistischen Linguistik. Broekhuis, Hans and Krzysztof Migdalski 2003 Participle fronting in Bulgarian as XP-movement. In Linguistics in the Netherlands, eds. Leonie Cornips & Paula Fikkert, 1-12. Amsterdam/Philadelphia, John Benjamins.

364 Bibliography Broekhuis, Hans and Kees Van Dijk 1995 The syntactic function of the auxiliary of time. In Linguistics in the Netherlands 1995, eds. Marcel Den Dikken & Kees Hengeveld. Amsterdam/Philadelphia, John Benjamins. Burzio, Luigi 1986 Italian syntax: a government-binding approach. Dordrecht/Boston/ Lancaster/Tokyo: Reidel. Caink, Andrew 1999 Against 'long' head movement: lexical insertion and the Bulgarian auxiliary 'be'. In Topics in South Slavic Syntax and Semantics, eds. Mila Dimitrova-Vulchanova & Lars Hellen, 91-123. Amsterdam, John Benjamins. Chomsky, Noam 1957 Syntactic structures. The Hague: Mouton. 1973 Conditions on transformations. In A Festschrift for Morris Halle, eds. Stephen Anderson & Paul Kiparsky, 71-132. New York, Holt, Rinehart, and Winston. 1981 Lectures on government and binding. Dordrecht: Foris. 1986 Barriers. Cambridge, Mass.: MIT Press. 1991 Some notes on economy of derivation and representation. In Principles and parameters in comparative syntax, ed. Robert Freidin. Cambridge (Mass.), MIT Press. 1995a The minimalist program. Cambridge (Mass.): MIT Press. 1995b Bare phrase structure. In Government and binding theory and the minimalist program, ed. Gert Webelhuth, 383-439. Oxford/ Cambridge (Mass.), Blackwell. 2000 Minimalist inquiries: the framework. In Step by step. Essays on minimalist syntax in honor of Howard Lasnik, eds. Roger Martin et al., 89-155. Cambridge (Mass.), MIT Press. 2001 Derivation by phase. In Ken Hale. A life in Language, ed. Michael Kenstowicz, 1-52. Cambridge (Mass.), MIT Press. 2004 Beyond explanatory adequacy. In Structures and beyond, ed. Adriana Belletti, 104-131. Oxford, Oxford University Press. 2005 On phases. Unpublished Ms. MIT. Chomsky, Noam and Howard Lasnik 1977 Filters and control. Linguistic inquiry 8: 425-504 [Quoted from Howard Lasnik. 1990. Essays on Restrictiveness and Learnability. Dordrecht, Kluwer]. Christensen, Ken Ramshøj 2005 Interfaces. Negation – syntax – brain, University of Aarhus: Ph.D. dissertation. Cinque, Guglielmo 1990 Types of A'-dependencies. Cambridge (Mass.)/London. 1999 Adverbs and functional heads: a cross-linguistic perspective. New York/Oxford: Oxford University Press.

Bibliography 365 Collins, Chris 1997 Local economy. Cambridge (Mass.)/London: MIT Press. Coopmans, Peter 1989 Where stylistic and syntactic processes meet: Locative inversion in English. Language 65: 728-751. Corver, Norbert and Henk van Riemsdijk 1994 Studies of scrambling: movement and non-movement approaches to free word-order phenomena. Berlin/New York: Mouton de Gruyter. Costa, João 1998 Word order variation. A constraint-based approach, LOT Dissertations 14, University of Leiden: Ph.D. dissertation. Csirmaz, Anikó 2004 Particles and phonologically defective predicates. In Verb clusters. A study of Hungarian, German and Dutch, eds. Katalin É. Kiss & Henk van Riemsdijk, 225-252. Amsterdam/Philadelphia, John Benjamins. Culicover, Peter and Robert Levine 2001 Stylistic Inversion and the that-t effect: a reconsideration. Natural Language and Linguistic Theory 19: 283-310. De Hoop, Helen 1992 Case configuration and noun phrase interpretation, University of Groningen: Ph.D. dissertation. De Hoop, Helen and Henriëtte De Swart 1990 Indefinite objects. In Linguistics in the Netherlands 1990, eds. Reineke Bok-Bennema & Peter Coopmans, 91-100. Dordrecht, Foris. De Hoop, Helen and Wim Kosmeijer 1991 Case and scrambling: D-structure versus S-structure. In Studies in comparative Germanic syntax, eds. Hubert Haider et al. Dordrecht/ Boston/London, Kluwer Academic Publishers. De Sutter, Gert 2005 Rood, groen, corpus! `Een taalgebruiksgebaseerde analyse van woordvolgordevariatie in tweeledige werkwoordelijke eindgroepen. Katholieke Universiteit Leuven: http://webs.hogent.be/gertdesutter/ publications/phd.pdf. Dekkers, Joost 1999 Derivations & evaluations. On the syntax of subjects and complementizers, HIL dissertations 21: Ph.D. dissertation. Den Besten, Hans 1985 The ergative hypothesis and free word order in Dutch and German. In Studies in German Grammar, ed. JindĜich Toman, 23-65. Dordrecht/Cinnaminson, Foris. Den Besten, Hans and Hans Broekhuis 1989 Woordvolgorde in de werkwoordelijke eindreeks. Glot 12: 79-138. 1992 Verb Projection Raising in het Nederlands. Spektator 21: 21-34.

366 Bibliography Den Besten, Hans and Gert Webelhuth 1990 Stranding. In Scrambling and barriers, eds. Günter Grewendorf & Wolfgang Sternefeld. Amsterdam/Philadelphia, John Benjamins. Den Dikken, Marcel 1995a Particles: on the syntax of verb-particle, triadic, and causative constructions. New York/Oxford: Oxford University Press. 1995b Extraposition as intraposition, and the syntax of English tag questions. Unpublished ms. Free University (Amsterdam). 2006 Relators and Linkers. The syntax of predication, predicate inversion, and copulas. Cambridge (Mass.)/London: MIT Press. Den Dikken, Marcel and Alma Næss 1993 Case Dependencies: the case of predicate inversion. The Linguistic Review 10: 303-336. Déprez, Viviane 1989 On the Typology of Syntactic Positions and the Nature of Chains: Move-Alpha to the Specifier of Functional Projections, MIT: Ph.D. dissertation. Dewey, Tonya Kim 2006 The Origins and Development of Germanic V2, University of california, Berkeley: Ph.D. dissertation. Di Sciullo, Anna-Maria and Edwin Williams 1987 On the definition of the word. Cambridge (Mass.): MIT Press. Diesing, Molly 1992 Indefinites. Cambridge (Mass.): MIT Press. 1997 Yiddish VP order and the typology of object movement in Germanic. Natural Language and Linguistic Theory 17: 369-427. Doetjes, Jenny 1997 Quantifiers and selection: on the distribution of quantifying expressions in French, Dutch and English, HIL dissertation 32, University of Leiden: Ph.D. dissertation. É. Kiss, Katalin 2002 The syntax of Hungarian. Cambridge (UK): Cambridge University Press. 2004 Parrallel strategies of verbal complex verb formation in Hungarian and West-Germanic. In Verb clusters; A study of Hungarian, German and Dutch, eds. Katalin É. Kiss & Henk van Riemsdijk, 335-358. Amsterdam/Philadelphia, John Benjamins. 2006 Focusing as predication. In The Architecture of Focus, eds. Valeria Molnár & Susanne Winkler. Berlin/New York, Mouton de Gruyter. to appear First steps towards a theory of the verbal particle. In Approaches to Hungarian 9, eds. István Kenesei et al. Budapest, Akadémia Kiadó. Eisner, Jason 1999 Doing OT in a straitjacket: UCLA.

Bibliography 367 Embick, David and Roumanya Izvorski 1994 On long head movement. In Proceedings of ESCOL, ed. J. M. Fuller et al., 104-115. Emonds, Joseph 1976 A transformational approach to English syntax: root, structurepreserving, and local transformations. New York: Academic Press. Engels, Eva and Sten Vikner 2006 An Optimality-Theoretic Analysis of Scandinavian Object Shift and Remnant VP-Topicalisation. In Optimality Theory and Minimalism: a Possible Convergence? Linguistics in Potsdam 25, eds. Hans Broekhuis & Ralf Vogel, 195–231. Potsdam: http://www.ling.unipotsdam.de/lip/. Fortson , Benjamin W. IV. 2004 Indo-European language and culture: An introduction. Oxford: Blackwell. Fox, Danny and David Pesetsky 2005 Cyclic Linearization of Syntactic Structure. Theoretical Linguistics 31: 1-45. Grice, H.P. 1975 Logic and Conversation. In Speech Acts: Syntax and Semantics 3, eds. P. Cole & J. Morgan, 41-58. New York, Academic Press. Grimshaw, Jane 1991 Extended projections. Unpublished Ms. Brandeis University, Walthamm (Mass.). 1997 Projection, heads and optimality. Linguistic Inquiry 28: 373-422. Groat, Erich and John O'Neil 1996 Spell-Out at the LF interface. In Minimal Ideas. Syntactic studies in the minimalist framework, eds. Werner Abraham et al., 113-139. Amsterdam/Philadelphia, John Benjamins. Haeberli, Eric 2002 Features, categories and the syntax of A-positions. Dordrecht/ Boston/London: Kluwer. Haegeman, Liliane 1986 The double object construction in West Flemish. The Linguistic Review 5: 281-300. 1991 Scrambling, Clitic Placement and Agr Recursion in West Flemish. Unpublished manuscript University of Geneva. 1992 Theory and description in generative syntax: a case study in West Flemish. Cambridge: Cambridge University Press. 1995 The syntax of negation. Cambridge: Cambridge University Press. 1998 Verb movement in embedded clauses in West-Flemish. Linguistic Inquiry 29: 631-656. 2004 Speculations on subject positions, predication and predicate inversion in English. Unpublished Ms. Unpublished draft paper: STL Lille III.

368 Bibliography Haegeman, Liliane and Henk van Riemsdijk 1986 Verb projection raising, scope, and the typology of rules affecting verbs. Linguistic Inquiry 17: 417-466. Haeseryn, Walter, Kirsten Romijn, Guido Geerts, Jaap De Rooij and Maarten C. Van den Toorn 1997 Algemene Nederlandse spraakkunst. Groningen: Nijhoff. Haider, Hubert 1997a Precedence among predicates. Journal of Comparative Germanic Linguistics 1: 3-41. 1997b Extraposition. In Rightward Movement, eds. Dorothee Beermann et al., 115-151. Amsterdam/Philadelphia, John Benjamins. 2000 OV is more basic than VO. In The derivation of VO and OV, ed. Peter Svenonius, 45-67. Amsterdam/Philadelphia, John Benjamins. 2003 V-clustering and clause union: causes and effects. In Verb constructions in German and Dutch, eds. Pieter A.M. Seuren & Gerard Kempen, 91-126. Amsterdam/Philadelphia, John Benjamins. 2005 How to turn German into Icelandic - and derive the OV-VO contrasts. Journal of Comparative Germanic Linguistics 8: 1-53. Haider, Hubert, Susan Olsen and Sten Vikner 1995 Introduction. In Studies in comparative Germanic syntax, eds. Hubert Haider et al. Dordrecht/Boston/London. Hale, Ken and Samuel Keyser 1993 On argument structure and the lexical expression of syntactic relations. In The view from building 20: essays in linguistics in honor of Sylvain Bromberger, eds. Ken Hale & Samuel Keyser. Cambridge, MA/London, MIT Press. Hartmann, Jutta M. to appear Expletives in existenstials: English there and German da, Tilburg University: Ph.D. dissertation. Haslinger, Irene 2007 The syntactic location of events. Aspects of verbal complementation in Dutch, University of Tilburg: Ph.D. dissertation. Hegarty, Michael 2005 Expletive subjects and feature checking in Germanic existential constructions. Unpublished Ms., 20th comparative Germanic syntax Workshop. Tilburg. Handout. Hoekstra, Teun 1984 Transitivity. Grammatical relations in government-binding theory. Dordrecht/Cinnaminson: Foris. Hoekstra, Teun and René Mulder 1990 Unergatives as copular verbs; locational and existential predication. The Linguistic Review 7: 1-79.

Bibliography 369 Holmberg, Anders 1986 Word order and syntactic features in the Scandinavian languages and English, University of Stockholm, Department of General Linguistics: Ph.D. dissertation. 1999 Remarks on Holmberg's generalization. Studia Linguistica 53: 1-39. Holmberg, Anders and Thorbjörg Hróarsdóttir 2004 Agreement and movement in Icelandic raising constructions. Lingua 114: 651-673. Holmberg, Anders and Christer Platzack 1995 The role of inflection in Scandinavian syntax. New York/Oxford: Oxford University Press. Holmberg, Anders and Halldór Ármann Sigurðsson 2006 Icelandic Dative Intervention: Person and Number are separate probes. Unpublished Ms., http://ling.auf.net/lingbuzz/000371. Hornstein, Norbert 1984 Logic as grammar. Cambridge, Mass.: MIT Press. 1995 Logical Form: from GB to minimalism. Oxford: Blackwell. 2001 Move! A minimalist theory of construal. Malden (Mass.)/Oxford (UK): Blackwell Publishers. Huybregts, Riny 1991 Clitics. In Grammatische analyse, ed. Jan Model. Dordrecht, ICG Printing. Jaegli, Osvaldo 1986 Passive. Linguistic Inquiry 17: 587-622. Johnson, Kyle 1991 Object positions. Natural language and Linguistic Theory 9: 577636. Jonas, Dianne and Jonathan Bobaljik 1993 Specs for Subjects; the Role of TP in Icelandic. In Papers on Case and Agreement I. MIT Working Papers in Linguistics 18, eds. Jonathan Bobaljik & Colin Phillips, 59-98. Cambridge (USA), MIT. Jónsson, Jóhannes G. 1996 Clausal architecture and Case in Icelandic, GLSA: University of Massachusetts: Ph.D. dissertation. Josefsson, Gunlög 1992 Object shift and weak pronominals in Swedish. Working Papers in Scandinavian Syntax 49: 59-92. Kayne, Richard S. 1985 L'accord du participe passé en Français et en Italien. Modèles Linguistiques VII: 73-89. 1989 Facets of Romance past participle agreement. In Dialect variation and the theory of grammar, ed. P. Benincà. Dordrecht, Foris. 1994 The antisymmetry of syntax. Cambridge (Mass.): MIT Press. 2000 Parameters and universals. New York: Oxford University Press. 2005 Movement and Silence. Oxford: Oxford University Press.

370 Bibliography Kennedy, Christopher 1997 Antecedent-Contained Deletion and the syntax of quantification. Linguistic Inquiry 28: 662-688. Kitahara, Hisatsugu 1997 Elementary operations and optimal derivations. Cambridge (Mass.)/London: MIT Press. Koeneman, Olaf 2006 Shape conservation, Holmberg's generalization and predication. In Comparatieve studies in Germanic syntax, eds. Jutta M. Hartmann & László Molnárfi. Amsterdam/Philadelphia, John Benjamins. Koizumi, Masatoshi 1993 Object agreement phrases and the split VP hypothesis. In Papers on case and agreement I, eds. Jonathan Bobaljik & Colin Phillips, 99-148, MIT. Komlósy, András 1985 Predicate complementation. In Approaches to Hungarian 1: Data and descriptions, ed. István Kenesei. Szeged, JATE. 1989 Fókuszban az igék. Általános Nyelvészeti Tanulmányok XVII: 171182. Koopman, Hilda and Anna Szabolcsi 2000 Verbal complexes. Cambridge (Mass.)/London: MIT Press. Koster, Jan 1978 Locality principles in syntax. Dordrecht: Foris. 1994 Predicate incorporation and the word order of Dutch. In Paths towards Universal Grammar. Studies in Honor of Richard S. Kayne, eds. G. Cinque et al., 255-277. Washington D.C., Georgetown University Press. Kroch, Anthony 1989 Reflexes of grammar in patterns of language change. Linguistic Variation and Change 1: 199-244. 1994 Morphosyntactic variation. In Papers from the 30th Regional Meeting of the Chicago Linguistics Society: Parasession on Variation and Linguistic Theory. Kuþerová, Ivona to appear Agreement in Icelandic: an argument for derivational theory of intervention effects. In WECOL 2006 LaCharité, Darlene and Carole Paradis 2000 Derivational residue: hidden rules in Optimality Theory. In Optimality Theory: phonology, syntax and acquisition, eds. Joost Dekkers et al., 386-422. Oxford/New York, Oxford University Press. Lambova, Mariana 2002 Is head movement syntactic: Evidence from Bulgarian. In Proceedings of HUMIT 2001. MIT Working Papers in Linguistics 43, 91-104.

Bibliography 371 Lasnik, Howard 1991 On the necessity of binding conditions. In Principles and parameters in comparative grammar, ed. Robert Freidin. Cambridge MA/ London, MIT Press. 1999a Minimalist analysis. Oxford: Blackwell. 1999b Chains of arguments. In Working minimalism, eds. Samuel David Epstein & Norbert Hornstein. Cambridge (Mass.)/London, MIT Press. Lechner, Winfried 2005 Interpretive effects of head movement. Unpublished Ms. semanticsarchive, http://semanticsarchive.net/Archive/mJlYmI2O/ Lechner%20HM%202005.pdf. Legendre, Géraldine 2000 Morphological and prosodic alignment of Bulgarian clitics. In Optimality Theory, Phonology, Syntax and acquisition, eds. Joost Dekkers et al., 423-462. Oxford, Oxford University Press. Lema, José and María Luisa Rivero 1989 Long Head Movement: ECP vs. HMC. NELS 20: 333-347. Lenerz, Jürgen 1977 Zur Abfolge nominaler Satzglieder im Deutschen. Studien zur deutschen Grammatik 5. Tübingen: Narr. Longobardi, Giuseppe 1994 Reference and proper names: a theory of N-movement in syntax and Logical Form. Linguistic Inquiry 25: 609-665. Mahajan, Anoop 1990 The A/A-bar distinction and movement theory, MIT: Ph.D. dissertation. May, Robert 1985 Logical form: its structure and derivation. Cambridge, MA: MIT Press. McCarthy, John J. and Allen Prince 1993 Prosodic Morphology: Constraint Interaction and Satisfaction. Unpublished Ms. University of Massachusetts, Amherst/Rutgers University, http://roa.rutgers.edu/files/482-1201/482-1201-mccarthy0-1.pdf. Migdalski, Krzysztof 2006 The syntax of compound tenses in Slavic: Ph.D. dissertation. Moro, Andrea 1997 The raising of predicates. Predicative noun phrases and the theory of clause structure. Cambridge: Cambridge University Press. 2000 Dynamic antisymmetry. Cambridge (Mass.)/London: MIT Press. Müller, Gereon 2000 Shape conservation and remnant movement. In Proceedings of NELS 30, eds. A. Hirotani et al., 525-539. Amherst (Mass.), GLSA. 2001 Order preservation, parallel movement, and the emergence of the unmarked. In Optimality-theoretic syntax, eds. Géraldine Legendre et al., 113-142. Cambridge (Mass.)/London, MIT Press/MITWPL.

372 Bibliography Nash, Léa and Alain Rouveret 1997 Proxy categories in phrase structure theory. NELS 27: 287-304. Neeleman, Ad 1994a Scrambling as a D-structure phenomenon. In Studies on Scrambling. Movement and non-movement approaches to free word-order phenomena, eds. Norbert Corver & Henk van Riemsdijk, 387-429. Berlin/New York, Mouton de Gruyter. 1994b Complex predicates, OTS dissertation series, LEd: Ph.D. dissertation. Neeleman, Ad and Hans Van de Koot to appear The nature of discourse templates. Journal of Comparative Germanic Linguistics. Newmeyer, Frederick J. 1983 Grammatical Theory. Its limits and possibilities. Chicago/London: University of Chicago Press. Nilsen, Øystein 2003 Eliminating positions, University of Utrecht: Ph.D. dissertation. 2007 Object shift by Remnant Movement (handout). Paper presented at Workshop “Object Positions - Formal and Functional Approaches” (January 18-19, 2007), Aarhus, Denmark. Nunes, Jairo 1999 Linearization of chains and phonetic realization of chain links. In Working Minimalism, eds. Samuel David Epstein & Norbert Hornstein. Cambridge (Mass.)/London, MIT Press. 2004 Linearization of chains and sideward movement. Cambridge (Mass.)/London: MIT Press. Pesetsky, David 1997 Optimality theory and syntax: movement and pronunciation. In Optimality theory, eds. Diana Archangeli & Terence Langendoen. Malden/Oxford, Blackwell. 1998 Some optimality principles of sentence pronunciation. In Is the best good enough?, eds. Pilar Barbosa et al., 337-383. Cambridge (Mass.)/London, MIT Press/MITWPL. Pollock, Jean-Yves 1989 Verb movement, Universal Grammar and the structure of IP. Linguistic Inquiry 20: 365-424. Rivero, María-Luisa 1991 Long head-movement and negation: Serbo-Croatian vs. Slovak and Czech. The Linguistic Review 8: 319-351. Rizzi, Luigi 1986 Null objects in Italian and the theory of pro. Linguistic Inquiry 17: 501-557. 1990 Relativized Minimality. Cambridge (Mass.)/London: MIT Press. 1996 Residual Verb Second and the Wh Criterion. In Parameters and functional heads. Essays in comparative syntax, eds. Adriana Belletti & Luigi Rizzi. Oxford/New York, Oxford University Press.

Bibliography 373 Roberts, Ian 1987 The representation of implicit and dethematized subjects. Dordrecht: Foris. Rochemont, Michael and Peter Culicover 1990 English focus constructions and the theory of grammar. Cambridge: Cambridge University Press. Sabel, Joachim 2005 String-vacuous scrambling and the Effect on Output Condition. In The free word order phenomenon. Its syntactic sources and diversity, eds. Joachim Sabel & Mamoru Saito. Berlin/New York, Mouton de Gruyter. Schütze, Carson Theodore Robert 1997 INFL in child and adult language: agreement, case and licensing, MIT: Ph.D. dissertation. Sells, Peter 2001 Structure Alignment and optimality in Swedish. Stanford: CSLI Publications. Sigurðsson, Halldór Ármann 1996 Icelandic finite verb agreement. Working Papers in Scandinavian Syntax 57: 1-46. 2003 Case: abstract vs. morphological In New perspectives on Case Theory, eds. Ellen Brandner & Heike Zinsmeister, 223-267. Stanford, CSLI Publications. Sigurðsson, Halldór Ármann and Anders Holmberg 2006 Icelandic Intervention effect: person and number are separate probes. Unpublished Ms. Lund University and Newcastle University. Slioussar, Natalia 2007 Grammar and Information Structure. A study with reference to Russian, University of Utrecht/LOT: Ph.D. dissertation. Sportiche, Dominique 1988 A theory of floating quantifiers and its corollaries for constituent structure. Linguistic Inquiry 19: 425-449. Stowell, Tim 1981 Origins of Phrase Structure, MIT: Ph.D. dissertation. Svenonius, Peter 1996 The Optionality of Particle Shift. Working Papers in Scandinavian Syntax 57: 47-75. 2000 Quantifier Movement in Icelandic. In The derivation of VO and OV, ed. Peter Svenonius, 255-292. Amsterdam, John Benjamins. SzendrĘi, Kriszta 2004 A stress-based appraoch to climbing. In Verb clusters. A study of Hungarian, German and Dutch, eds. Katalin É. Kiss & Henk van Riemsdijk, 205-223. Amsterdam/Philadelphia, John Benjamins.

374 Bibliography SzendrĘi, Kriszta and Ildikó Tóth 2004 Hungarian verbal clusters: results of a questionnaire survey. In Verb clusters: A study of Hungarian, German and Dutch, eds. Katalin É. Kiss & Henk van Riemsdijk, 87-119. Amsterdam/Philadephia, John Benjamins. Taraldsen, Knut Tarald 1995 On agreement and nominative subjects in Icelandic. In Studies in comparative Germanic syntax, eds. Hubert Haider et al. Dordrecht/Boston/London, Kluwer Academic Publishers. Thráinsson, Höskuldur 2001 Object shift and scrambling. In Contemporary Syntactic Theory, eds. Mark Baltin & Chris Collins. Malden, Blackwell Publishing. Van den Toorn, M.C. 1971 Enkele opmerkingen over het indirect object. Levende Talen: 32-71. Van Koppen, Marjo 2005 One probe - two goals: aspects of agreement in Dutch dialects, LOT dissertations 105, University of Leiden: Ph.D. dissertation. Van Riemsdijk, Henk 1978 A case study in syntactic markedness: the binding nature of prepositional phrases. Lisse: Peter de Ridder Press. 1997 Push chains and drag chains: complex predicate split in Dutch. In Scrambling, ed. Shigeo Tonoike. Tokyo, Kurosio Publishers. Vanden Wyngaerd, Guido 1988 Raising-to-object in English and Dutch. Unpublished Ms., Dutch Working Papers in English Language and Linguistics. 1989b Object shift as an A-movement rule. Unpublished Ms., MIT Working Papers in Linguistics. Vangsnes, Øystein Alexander 2002 Icelandic expletive constructions and the distribution of subject types. In Subjects, expletives and the EPP, ed. Peter Svenonius. Oxford/New York, Oxford University Press. Verhagen, Arie 1986 Linguistic theory and the function of word order in Dutch. A study on interpretive aspects of the order of adverbials and noun phrases. Dordrecht: Foris. Vikner, Sten 1989 Object shift and double objects in Danish. Working Papers in Scandinavian Syntax 44: 141-155. 1990 Verb movement and Licensing of NP positions in the Germanic languages: Ph.D. dissertation [draft version]. 1994 Scandinavian Object Shift and West Germanic Scrambling. In Studies on Scrambling. Movement and Non-movement Approaches to free Word-Order Phenomena, eds. Norbert Corver & Henk van Riemsdijk, 487-517. Berlin/New York, Mouton de Gruyter.

Bibliography 375 2006

Object Shift. In The Blackwell companion to Syntax, eds. Martin Everaert & Henk van Riemsdijk. Oxford, Blackwell.

Vogel, Ralf to appear Weak Function Word Shift. Linguistics. http://roa.rutgers.edu/files/ 663-0504/663-vogel-0-0.pdf. Wagner, Michael 2007 A Semantic Motivation for Focus and Givenness Movement. Paper presented at The Second Brussels Conference on Generative Linguistics, Alternatives to Cartography. Watanabe, Akira 1991 Wh-in situ, Subjacency and Chain-formation, MIT: Ph.D. dissertation. Webelhuth, Gert 1986 On passive and the notion subject. In Massachusetts occasional papers in linguistics 10, eds. David Lebeaux & Armin Mester, 115136. Amherst. 1989 Syntactic saturation phenomena and the modern Germanic languages: Ph.D. dissertation. 1992 Principles and Parameters of Syntactic Saturation. New York: Oxford University Press. Williams, Edwin 1980 Predication. Linguistic inquiry 11: 203-238. 2003 Representation theory. Cambridge (Mass.)/London: MIT Press. Woolford, Ellen to appear Case locality: pure domains and object shift. Lingua. Zanuttini, R. 1991 Syntactic properties of sentential Negation: a comparative study of Romance languages, University of Pennsylvania: Ph.D. dissertation. Zuckerman, Shalom 2001 The acquisition of "optional" movement. Ph.D. dissertation University of Groningen: http://irs.ub.rug.nl/ppn/228016754. Zwart, Jan Wouter 1992 Dutch expletives and small clause predicate raising. In North East Linguistic Society 22, ed. K. Broderick, 477-491. 1993 Dutch syntax. A minimalist approach, Groningen Dissertations in Linguistics, University of Groningen: PhD. Dissertation. 1997 Morphosyntax of verb movement. A minimalist approach to the syntax of Dutch. Dordrecht: Kluwer. 2001 Syntactic and phonological verb movement. Syntax 4: 34-62.

Subject index

A Adjunction — of adverbial phrases to VP/vP, 75 VP- —, 25 Adverb — placement, 98, 101, 104 Object shift across —, 97, 119, 358 Sentence —, 75, 119 VP- —, 75, 97, 119 Agree, 10, 269 Locality conditions on —. See Locality theory Agreement, 217, 263–347 — in copular inversion constructions, 284 — in Icelandic quirky subject constructions, 138 Complementizer —, 130 Multiple —, 107 Object —, 107, 228, 319 Past Participle —, 241 Subject-verb —, 130, 228 Antecedent-contained Deletion, 93 Auxiliary. See Auxiliary verb

B Binding, 63, 64, 91, 104 Blocking. See Blocking strategy Branching Constraint, 76

C Cartography, 76 Case — Filter, 122, 268 — Theory, 122 Morphological —, 122 Structural —, 167

Chain Operator-variable —, 39 Closeness, 124, 140, 141, 276 Complementizer, 211 Complementizer trace configuration, 5, 245 Component Phrase structure —, 5 Transformational —, 5 Condition, 24 Bare output —s. See Interface conditions Economy —, 9 Shortest Steps, 231, 239, 240, 307, See also Economy constraints Effect on Output —, 13, 15, 37, 101 Extension —, 154 Global —, 9 Interface —, 4, 10 Last Resort —, 8, 124, 319 Minimal Link —, 121–57, 276 Strict cyclicity —, 153, 256 Consider construction, 281 Constraint — ranking, 17, 28 Strong —, 40 Weak —, 40 Alignment —, 47, 156 CHL —, 35 Clash —, 35 Gradient —, 45 Implication —, 35 LF —, 45, See also Interface constraints PF —, 46, See also Interface constraints —s versus filters, 17, 47, 263 Tied —s, 19, 20, 163 Types of syntactic —, 35, 353 Universal set of —s CON, 17

378 Subject index Constraints, 17, 352–59 Economy —, 35 *MOVE, 36, 37, 43, 159–65 *SELECT, 37 A-OVER-A, 300–325, 346 NOLEXM, 36, 37, 205–16 NOVACM, 37, 101 EPP —, 39 *STRAY FEATURE, 113, 183 EPP(asp), 104 EPP(case), 39, 354, passim EPP(gender), 228 EPP(number), 228 EPP(person), 228 EPP(tense), 39, 113 EPP(v), 96–117 EPP(ij), 39, 354, passim Faithfulness — DEP(X), 49 MAX(X), 49 Interface —, 35 ALIGNFOCUS, 44, 158, 162, 180 D-PRONOUN, 43, 159–65, 172, 177 ENCLITIC, 340 HEAD-COMPL, 46, 109, 135, 158, 183–90, 206, 207, 230, 240 LEFT EDGE (CP), 18 LEXICALLY FILL TOP F, 132, 205–16 NO-STRESS-Vfinite, 333, 340 RELMIN, 46, 158, 234, 240 SCOPE, 45, 222 TELEGRAPH, 18 Other — HEAD-LEFT, 47 RECOVERABILITY, 18

D Defective intervention, 136 Deletion, 6, 32 Derivation — -and-evaluation model, 1–60, passim Converging —, 9, 48 Syntactic —, 1 Domain Minimal —, 124

do-support, 36, 253 Double object construction, 165–83, 248

E Equidistance, 64, 136, 276 Evaluation, 19 Evaluator, 3, 17, 22, 27–30, 34–48 Expletive — Construction, 231 — subject raising constructions, 50 — there, 50 Extraposition, 25, 99, 115, 281

F Feature — movement, 10 — strength, 28, 32, 42 [+affect], 40, 232 Case —, 12, 62, 71, 119–203 EPP- —, 13, 28, 30, 32 Formal — Uninterpretable —. See Unvalued formal feature Unvalued —, 8 Strong —, 11, 28 Weak —, 11, 28 ij- —s, 62, 71, 90–117, 119 Gender, 227 Intrinsic —, 227 Number, 227 Optional —, 227 Person, 227 Filter, 4–16, 15, 34 Doubly Filled COMP —, 6 Language-specific —, 5 —s versus constraints, 17, 47, 263 That-trace —, 7, See also complementizer trace configuration Focus, 236 — movement, 63, 326 Contrastive —, 44 Exhaustive —, 44 Identificational —, 326 New information —, 12, 44, passim Full Interpretation, 9, 48

Subject index 379

Information structure. passim Input. See Syntactic Input Interpretive complex INT, 12, 29, 41 Inversion Dative-nominative —, 225, 237 Locative —, 52, 264–300 Predicate —. See also Predicate Movement

Icelandic, 11, 44, 62, 70, 138, 161, 169, 171, 187, 191, 193, 206, 212, 230, 231, 239, 243, 246, 248 Italian, 107, 228, 264 Norwegian, 70, 163, 179, 296 OV- —, 71–96, 100 Romance —, 15, 161 Swedish, 66, 97, 134, 153, 155, 163, 181, 190, 191, 193, 212, 242, 255 Tegelen Dutch, 131 Verb (projection) raising —, 82 VO- —, 71–96, 98, 105, 108 West Flemish, 82, 168, 239 Wh-in situ —, 39 Last Resort. See Last Resort Condition Lexical Integrity Hypothesis, 48 Lexicon, 28 Light verb. See Light verb Linear Correspondence Axiom, 73 Linearization, 71–96 Cyclic —, 248 Locality theory, 121–57 Locational proform, 65, 176 Locative Inversion. See Locative Inversion

L

M

Languages Bulgarian, 335–47 Danish, 29, 42, 65, 67, 162, 177, 183, 188, 193, 207, 208, 214, 242, 253 Dutch, 14, 53, 63, 65, 67, 79, 83, 101, 105, 121, 123, 129, 132, 133, 142, 155, 161, 165, 167, 171, 176, 185, 207, 212, 216–38, 258, 265, 300– 325 English, 6, 18, 36, 50, 53, 79, 98, 101, 104, 161, 185, 191, 213, 253, 264, 268–300 Finnish-Swedish, 160, 185, 201 French, 18, 36, 241 German, 165, 173, 185, 237, 240 Hungarian, 270, 325–35

Merge, 24 External —. See Select Internal —. See Move Minimal Link Condition. See Minimal Link Condition Minimalist program, 1–60 Move, 8, 24, 38, 50 Movement — of a locational proform, 65, 177 — of contrastively focused phrases. See Focus movement — of quantified phrases. See QPmovement A- —, 11, 62, 216–38, 216–38 A′- —, 11, 63, 122, 238–54 Extension condition on —. See Extension Condition External motivation for —, 11, 299

G Generator, 3, 16, 24–27, 32–34 Goal Active/Inactive —, 141 Grammar — competition, 163 Architecture of —, 1–60 Core —, 1, 3, 12, 26, 59, 321 Universal —, 6

H Harmonic boundedness, 17, 22 Holmberg’s Generalization, 12, 111, 121–57, 183–90

I

380 Subject index Feature —, 129 Focus —. See Focus movement Locality conditions on —. See Locality theory and Minimal Link Condition Optional —, 28 Phonological —, 12 Predicate —, 79, 83, 88, 263–347 — in Bulgarian, 335–47 — in Dutch, 300–325 — in English, 264–300 — in Hungarian, 325–35 Remnant —, 115, 196, 255, 276, 280, 293, 311, 344 Verb —, 36, See Verb movement VP- —, 81, 319 Wh- —, 39, 239

— across the main verb. See Holmberg's Generalization — and morpho-phonological constraints on verb inflection, 155 — as a post-syntactic operation, 153 — of locational proforms?, 176 The interaction of — and A′movement, 238–54 The interaction of — and Amovement, 216–38 The interaction of — and verb movement, 205–16 Typology of —languages, 160 Short —, 72, 90–117, 119, 274 Optimality theory, 1–60 Output. See Syntactic Output

N

P

Negation Constituent —, 54 Sentential —, 53 Negative polarity item, 53, 92 Neg-movement, 63, 232, 246 — puzzle, 248 Noun phrase Definite —, 218, 225 Generic —, 221 Indefinite —, 225 Non-specific —, 218, 220, 235 Specific —, 219, 235 Partitive —, 219 Pronominal —, 219, 220, 225, See also Pronoun Quantified —, 222 NP-movement. See Subject shift Numeration, 16, 50–56

Parameter, 40, 42 Directionality —, 77 Macro- —, 41 Parasitic gap, 63 Participle L- — fronting, 154, 255, 335–47 Passive —, 104, 133 Past —, 133 Present — (V-ing), 292 Particle Verbal —, 66, 86, 91, 190, 234 Passive, 230, 233 Regular —, 167 Semi-—, 168 Periphery, 1, 3, 12, 26, 59, 321 Phase — Impenetrability Condition (PIC), 9, 234, 241, 245, 257, 277 — theory, 33, 71, 234, 241, 257, 277 Phonological — border, 14 — edge, 13 — movement, 12, 26 Predicate — inversion. See Predicate Inversion — movement. See Predicate movement

O Object shift. Passim — as A-movement, 62 Regular —, 11, 28, 41, 72, 119–203 — across a verbal particle, 190–99 — across adverb, 97, 119, 358 — across the indirect object, 165– 83

Subject index 381 Preposition Stranding. See Rextraction Presupposition, 12, passim Principle Empty Category — (ECP), 5, 7, 244 Extended Projection — (EPP), 50 Probe P as —, 114 Procrastinate, 11, 28, 42 Pronoun — shift. See Regular object shift A′-movement of —, 176 R- —. See also Locational proform Stressed —, 219, 220 Push-up. See Push-up strategy

Q QP-movement, 231, 246 Quantifier — movement, 231 — raising, 93, 223 Floating —, 134, 247 Quantifiers Floating —, 95 Quirky — case, 139 — subject, 138

R Raising Quantifier —. See Quantifier raising Right-node —. See Right-node raising Subject —. See Subject raising Recoverability, 6 Relative — Clause, 6 — pronoun, 6 Relativized Minimality, 124, 166 Representation Syntactic —, 1 R-extraction, 66, 88, 316 Right-node raising, 25

S Scrambling, 25, 62 — of PP, 67 — of the A′-movement type, 63, 67 — of the A-movement type, 63, See also Regular object shift Base-generation of —, 67 Theoretical approaches to —, 63 Select, 8, 24, 50 Set Candidate —, 17, 32 Reference —, 9 Spell-Out, 10, 11, 130 Strategy Blocking —, 112, 166, 171, 209, 234 Push-up —, 112, 166, 171, 209, 234 Structure preservation constraint, 255 Subject — raising, 50 — shift, 216–38 Exceptionally case-marked —, 91 Icelandic quirky —. See Quirky subject Supplementive, 102 Syntactic — input, 16, 50–52 — output, 24–27 System Articulatory-perceptual —, 10, 35, 49 Computational — of human language (CHL), 8, 24, See also Generator Conceptual-intentional —, 10, 35, 38, 95

T Tableau, 19 Tense Perfect — construction, 104, 132, 155, 186, 188, 190, 231, 248, 319 Simple — construction, 97, 248 Tied constraints, 19, 20, 163 Topicalization VP- —, 255, See also L-participle Fronting

382 Subject index

U Universal base hypothesis, 71–96

V Vacuous Movement Hypothesis, 101 Verb — (projection) raising. See Complex verb construction — movement, 210 — and object shift, 205–16 Basic Typology, 211 Danish/Swedish — movement puzzle, 205–16 V-to-Asp, 104, 186, 323 V-to-I, 72, 119, 129 V-to-v, 72, 91, 90–117, 119, 129, 206, 207

— second, 26, 86, 210 Auxiliary — Case features on —, 133 ij-features on —, 107, 133 Complex — construction, 80–90, 323, See also perfect tense construction Copular —, 281 Light — v, 12, 61, 96, 119, 126 Unaccusative —, 225, 230, 237 Verbal Modifiers, 326 Verbal root V, 61, 96, 119

W Webelhuth’s paradox, 63

X X-bar-theory, 25