WH-MOVEMENT AND THE THEORY OF FEATURE-CHECKING
This book was originally selected and revised to be included in the Wo...
15 downloads
413 Views
1021KB Size
Report
This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!
Report copyright / DMCA form
WH-MOVEMENT AND THE THEORY OF FEATURE-CHECKING
This book was originally selected and revised to be included in the World Theses Series (Holland Academic Graphics, The Hague), edited by Lisa L.-S. Cheng.
WH-MOVEMENT AND THE THEORY OF FEATURE-CHECKING
ANDREW SIMPSON SOAS, University of London
JOHN BENJAMINS PUBLISHING COMPANY PHILADELPHIA/AMSTERDAM
8
TM
The paper used in this publication meets the minimum requirements of American National Standard for Information Sciences — Permanence of Paper for Printed Library Materials, ANSI Z39.48-1984.
Library of Congress Cataloging-in-Publication Data
Simpson, Andrew, 1962Wh-movement and the theory of feature-checking / Andrew Simpson. p. cm. Includes bibliographical references and index. 1. Grammar, Comparative and general--Syntax. 2. Generative grammar. I. Title. P295.S499 2000 415--dc21 ISBN 90 272 2562 1 (Eur.) / 1 55619 856 6 (US) (Hb; alk. paper)
99-462346 CIP
© 2000 – John Benjamins B.V. No part of this book may be reproduced in any form, by print, photoprint, microfilm, or any other means, without written permission from the publisher. John Benjamins Publishing Co. • P.O.Box 75577 • 1070 AN Amsterdam • The Netherlands John Benjamins North America • P.O.Box 27519 • Philadelphia PA 19118-0519 • USA
Table of contents
Foreword and Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . .
ix
Chapter 1 W h in situ and the LF Movement Hypothesis . . . . . . . . . . . . . . . . . . . 1 1.1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 1.2. The phenomena under investigation: LF full category movement . . . 3 1.3. Arguments for LF wh-movement . . . . . . . . . . . . . . . . . . . . . . . . . . . 8 1.3.1. Wh-movement as the construction of an input form to interpretation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8 1.3.2. Wh in situ and strong crossover . . . . . . . . . . . . . . . . . . . . . . . . 9 1.3.3. Superiorit y . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10 1.3.4. Pesetsky (1987) and d-linking . . . . . . . . . . . . . . . . . . . . . . . . . . 10 1.3.5. Wh-adjuncts and extraction islands . . . . . . . . . . . . . . . . . . . . . 11 1.4. Against LF wh-movement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12 1.4.1. Non-parallelism with regard to movement — localit y problems . 13 1.4.2. Non-parallelism with regard to interpretative possibilities — ‘only’ and wh in situ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17 1.4.3. LF raising and conditions on S-structure: parasitic gaps . . . . . . 18 1.4.4. Antecedent contained deletion (1) . . . . . . . . . . . . . . . . . . . . . . 19 1.4.5. Quechua, Chinese, the ECP and LF movement . . . . . . . . . . . . . 23 1.4.6. Scrambling and wh in situ . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29 1.4.7. Anaphor-antecedent relations . . . . . . . . . . . . . . . . . . . . . . . . . . 30
vi
Table of contents
1.4.8. Weak crossover . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.4.9. Bahasa Indonesia . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.4.10. Superiorit y licensing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.4.11. Antecedent contained deletion (2) . . . . . . . . . . . . . . . . . . . . . 1.5. Arguments for LF movement re-interpreted . . . . . . . . . . . . . . . . . . 1.5.1. Wh-adjuncts in situ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.5.2. Strong crossover and wh in situ . . . . . . . . . . . . . . . . . . . . . . . . 1.5.3. D-linking and wh in situ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.5.4. Superiorit y and wh in situ . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.6. Selection, absorption and scope . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.6.1. Selection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.6.2. Absorption . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.6.3. Wh-raising and the representation of scope . . . . . . . . . . . . . . . 1.7. Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
31 33 36 37 39 39 41 43 46 49 49 51 54 66
Chapter 2 W h-movement and feature-checking . . . . . . . . . . . . . . . . . . . . . . . . . 69 2.1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69 2.2. Iraqi Arabic . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71 2.2.1. wh in situ, tense and movement . . . . . . . . . . . . . . . . . . . . . . . . 73 2.3. Hindi . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82 2.3.1. Localit y constraints and (non-)movement dependencies. . . . . . . 85 2.3.2. Mahajan (1990) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91 2.3.3. Brody (1995), O’Neil & Groat (1995) . . . . . . . . . . . . . . . . . . . 95 2.4. Multiple-wh questions in English . . . . . . . . . . . . . . . . . . . . . . . . . . 97 2.4.1. The Triggering Hypothesis . . . . . . . . . . . . . . . . . . . . . . . . . . 103 2.5. Extensions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112 2.5.1. Focus . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112 2.5.2. English full wh in situ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115 2.5.3. Wh-island violations; superiorit y . . . . . . . . . . . . . . . . . . . . . . 117 2.6. Japanese and wh-scrambling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121 2.7. Concluding remarks — the status of LF . . . . . . . . . . . . . . . . . . . . . 133 2.7.1. Non-wh checking dependencies in a model without LF. . . . . . . 138 Chapter 3 Partial Movement and wh-expletive questions . . . . . . . . . . . . . . . . . 3.1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2.1. Basic properties of Partial wh Movement . . . . . . . . . . . . . . . . 3.2.2. Theoretical implications of Partial Movement structures . . . . .
147 147 148 148 153
Table of contents 3.3. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.3.1. Possible accounts of Partial Movement . . . . . . . . . . . . . . . . . . 3.3.2. Direct dependencies and wh-chains . . . . . . . . . . . . . . . . . . . . 3.3.3. Indirect dependencies and clausal pied piping . . . . . . . . . . . . 3.4. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.4.1. Partial Movement as focus movement . . . . . . . . . . . . . . . . . . . 3.4.2. Aside: overt clausal pied piping — a possible explanation . . . . 3.5. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.5.1. Towards a solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.5.2. Partial Movement and tense . . . . . . . . . . . . . . . . . . . . . . . . . . 3.5.3. Wh-expletive questions in Iraqi Arabic and Hindi . . . . . . . . . . 3.5.4. Wh-expletives and Partial Movement . . . . . . . . . . . . . . . . . . . 3.6. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.6.1. Further issues: antilocalit y in wh-expletive structures . . . . . . . 3.6.2. The clause-boundedness of wh-expletives . . . . . . . . . . . . . . . . 3.6.3. Further variation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.7. Summary and concluding remarks . . . . . . . . . . . . . . . . . . . . . . . .
vii 155 155 156 163 172 172 178 181 181 184 188 195 204 204 211 221 225
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 233 Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 240
Foreword and Acknowledgments
This book is a substantially re-written version of the 1995 doctoral dissertation I produced in the linguistics department of the School of Oriental and African Studies (SOAS), Universit y of London. This new version has been updated to take into account a number of works which have appeared in the three years since completion of the original. Chapter four of the 1995 thesis which was primarily on N-word patterns in French, Italian and Flemish has been cut out in order to allow for expanded discussion of wh-expletive questions in Hindi and Iraqi Arabic among various other additions and also in order to keep the focus firmly centred on wh-constructions. Many parts of the original thesis have furthermore now been significantly re-worked, especially in chapters t wo and three, and the result is hopefully a clearer and more tightly constrained set of proposals than was first argued for in the 1995 version. Here I am tremendously indebted to Lisa Cheng, the series editor, who very carefully went through all the thesis pointing to where improvements could be made, raising critical questions to be answered in the re-written version, and generally making many, many useful suggestions as to how I could best incorporate revisions throughout the book. Lisa is an extremely positive critic and here has the editor’s gift of being someone who is able to make one really appreciate the benefits available from reworking one’s earlier writings. Since the completion of this volume, Chomsky (1998) in Minimalist Enquiries: the framework has introduced several new ideas concerning the licensing of feature-checking dependencies which interestingly might seem to parallel certain
x
Foreword and Acknowledgments
suggestions made here in Chapters 2 and 3. Although time does now not allow for the incorporation of any serious comparison of the ideas in both works, I would nevertheless like to mention here (very briefly) some of the similarities bet ween the t wo approaches and how the present work may also provide support for certain of the suggestions put forward in Minimalist Enquiries. As will be seen in Chapter 2, evidence from a variet y of wh-construction t ypes leads to the conclusion that certain instances of wh-feature-checking may be satisfied non-locally and that there is effectively feature-checking ‘at a distance’. This idea developed on the basis of an examination of A’-dependencies interestingly seems to mirror suggestions resulting from a consideration of A-dependencies in Minimalist Enquiries (notably quirky Case phenomena in Icelandic) and results in the same essential suggestion that ‘Agreement’ (feature-checking) may be a long-distance relation (p.39). Further wh-patterns considered in Chapter 2 then lead to the claim that wh-movement essentially takes place for t wo reasons: (a) in some circumstances to identify or activate an underspecified licensing head, and (b) elsewhere in order that the wh-element occurs in the licensing/checking domain of a +wh+Q C. The former motivation for wh-movement resembles in part the idea of a generalized EPP here stated over C0, while the latter arguably corresponds to the new Agreement relation proposed in Minimalist Enquiries. In Chapter 3 it is argued that Partial Wh Movement takes place in order that a substantive wh-phrase becomes “visible” to a higher +wh+Q C checking head at the edge of a lower tensed CP. Such an idea would seem to relate in a very direct way to the ‘phase impenetrabilit y condition’ (p.22) suggested in Minimalist Enquiries; in wh-expletive questions a higher +wh+Q C is unable to look inside a lower tensed CP phase to find its appropriate Goal, and a wh-phrase is therefore forced to raise to the Spec of this tensed CP to become visible to the higher +Q C. The Partial Movement paradigms also interestingly seem to add good support to the Minimalist Inquiries’ characterization of a clausal phase as being ‘a full clause including tense and force’ (p.20), as the patterning found in Partial Movement structures critically distinguishes tensed from non-tensed CPs and shows that wh-elements need to move to the edge/Specifier of a fully (and independently) tensed CP in order to become visible to a higher +Q head. Such structures furthermore provide clear empirical evidence that wh-expletives are moved to the Spec of CP/the edge of the clausal phase and not merged there. This is interesting because Chomsky (1998) argues that the Spec of a phase can only be created via movement and not via simple merge, and that this propert y crucially distinguishes the non-phase head T from the phase heads C and v. With T it would seem possible to simply merge an expletive directly into its Spec; with C however, it is found that wh-expletives must be moved there. This
Foreword and Acknowledgments
xi
requirement that the C-Spec be created via movement also contrasts importantly with the fact that the head of the phase (i.e. the C-position) can in fact be directly filled with a question-particle via simple merge. The Minimalist Inquiries’ predictions as to what may or may not be phases and how phases differ from other projections therefore receives interesting substantiating evidence. Another significant point which emerges in Chapter 3 on Partial Movement is the claim that wh-phrases may potentially be specified for t wo partially independent properties: (i) wh-features, and (ii) a wh-operator. The suggestion that wh-expletives lack the latter propert y then allows for an explanation of the Partial Movement paradigms. Such a distinction in wh-related properties motivated by the patterning in wh-expletive structures might again seem to have a close new parallel in Minimalist Inquiries with Chomsky’s suggestion that wh-phrases may have both ‘an uninterpretable feature wh- and an interpretable feature Q which matches the uninterpretable probe Q of a (+Q) C’ (p.44). Finally, back on the topic of phase-impenetrabilit y, the present work suggests that cross-linguistic variation in wh-paradigms may in part frequently be due to the propert y that tensed CPs are opaque in some languages but not others, depending on the degree to which embedding verbs have become grammaticalized into the structure. Such opacit y effects with tensed CP nodes can be straightforwardly understood as a case of (and for) phase-impenetrabilit y in the new Minimalist Enquiries approach. Consequently, although much of the book presents criticisms of the 1993 and 1995 Minimalist models and argues for changes in the model, the general conclusions interestingly now in fact show certain greater convergence with those Minimalist ideas more recently proposed. Returning now to important acknowledgements, in addition to Lisa Cheng, I would also like to thank many other colleagues and friends for their support, criticism and friendship over the last three years. In London many, many thanks go to the following in particular: Wynn Chao, Monik Charette, Jonathan Kaye, Ruth Kempson, Shalom Lappin, and Jamal Ouhalla. In Frankfurt I would like to thank my good friends Günther Grewendorf, Kathernia Hartmann, Joachim Sabel and Jochen Zeller, and in Los Angeles among the many people I am indebted to I would especially like to thank Joseph Aoun, Jim Huang, Hilda Koopman, Ed Keenan, Audrey Li, Anoop Mahajan, Dominique Sportiche, Tim Stowell, and last but far from least Xiu-Zhi Zoe Wu. Finally I would also like to thank my dear, loving parents again for their non-stop encouragement and for continuing to give me the greatest feelings of securit y and optimism in life.
Chapter 1 Wh in situ and the LF Movement Hypothesis
1.1. Introduction The central theme of this work is a re-examination of the syntactic licensing conditions affecting wh-phrases cross-linguistically and how assumptions concerning the encoding of such dependencies may licitly be used to motivate other more general properties of the internal organisation of the language facult y. Within the Principles and Parameters (P&P) approach to language initiated in Chomsky (1981) it has long been suggested that wh-phrases are licensed via a relation to some clause-peripheral X0 position in which the interrogative nature of a clause is specified, standardly referred to as a +Q Comp/C0. In many languages this t ype of obligatory dependency is analysed as giving rise to movement bet ween the position in which a wh-phrase is understood to be basegenerated and the Specifier of the +Q C0, as for example in English: (1)
Whati did you see ti?
For a variet y of reasons and theoretical argumentation it has further been claimed that in languages where wh-phrases may occur audibly in situ in non-SpecCP positions, the dependency bet ween such wh-elements and a +Q Comp will also result in movement, this taking place covertly and without its effects receiving phonetic interpretation. Thus Chinese (2) and the English multiple wh-question (4) will at some point in their syntactic derivation give rise to a representation as indicated in (3) and (5):
2 (2)
(3) (4) (5)
W h-movement and the theory of feature-checking Ta shuo shenme? he say what ‘What did he say?’ [shenmei [ta shuo ti]] Whoi did he give ti what? [whatk whoi did [he give ti tk]]
Such proposals have important consequences for the construction of potential models of natural language and have resulted in the claim that there exists a level of syntactic representation (LF) beyond that which might be argued to characterise audible forms such as (2) and (4). Chapter 1 of this book sets out to argue at some length that dependencies bet ween wh-elements and a +Q Comp position need not in fact necessarily be established via any movement operation and that wh-phrases occurring in situ, as in (2) and (4), may remain and be interpreted in such positions throughout the derivation without the need to undergo covert LF raising to a +Q C0. In subsequent chapters it will further be suggested that in a broad range of languages there is actually good evidence to indicate that wh-phrases must indeed be licensed by the point of Spell-Out, hence often in non-SpecCP positions. A strong consequence of such conclusions is that wh in situ phenomena can significantly no longer be taken as supportive of and calling for the recognition of a discrete syntactic level of LF derived by post-Spell-Out movement. Given the centrally important and crucial role the case of wh in situ has had in the motivation of LF, the existence of such a level of syntax together with the general possibilit y of post-Spell-Out movement is seriously questioned and a Spell-Out-as-LF approach is suggested to have greater plausibilit y. The organisation of Chapter 1 is briefly as follows. Section 2 first identifies the object of study as (hypothetical) LF category movement. Section 3 then reviews the t ype of theoretical reasoning and evidence which has standardly been given as justification for the claim that wh-phrases in situ at PF undergo covert LF raising to a +Q Comp. Section 4 subsequently counters this with arguments for a non-covert movement analysis of wh in situ presenting a wide variet y of crosslinguistic evidence as support for such an approach. Finally, Section 5 returns to the original motivations for LF wh-raising noted in Section 3 and shows how the phenomena relevant to these arguments may in fact receive explanation in ways which do not presume any covert wh-movement and so are consistent with the general proposals outlined in Section 3. The chapter as a whole provides a theoretical back-drop for the development of a particular view of the nature and localit y of wh-licensing presented in Chapters 2 and 3 where various wh-paradigms problematic for current Minimalist models are given an in depth consideration.
W h in situ and the LF Movement Hypothesis
3
1.2. The phenomena under investigation: LF full category movement Before we begin with this investigation, a few words are in order concerning the actual object of study here. As noted above, a greater stated goal of the work is to understand the extent to which wh-dependencies provide evidence for an LF-based model of syntax, and such an interest has a necessary influence on the way the investigation is carried out, as will become clear below. Wh-dependency phenomena have continually been shown to be the case par excellence for covert movement and LF. Although there have been many analyses of (for example) long-distance anaphora and relative scope phenomena which have been argued to provide support for a level of LF, it is the richness and variet y of wh-paradigms cross-linguistically which has resulted in the most convincing body of evidence for LF. The crucial advantage held by wh-dependencies over the majorit y of other syntactic relations suggested to be encoded by LF movement is the simple fact that wh-movement/displacement is clearly visible in the overt syntax.1 This provides a simple means of comparison with hypothetical LF-formed dependencies, both within a single language such as English where a single wh-phrase undergoes overt raising and secondary wh-phrases remain in situ, and between languages, comparing those with overt wh-movement with full in situ ‘LF-movement’ languages such as Chinese, Thai etc. If an in situ wh-phrase displays properties which can be shown to be t ypical of overtly-moved wh-elements elsewhere, either within the same language or in other languages, then this has standardly been argued to be evidence of the clearest kind in favour of a covert movement analysis. Underlying all such work has been the guiding assumption that the strongest arguments in favour of LF and LF movement are those phenomena which seem to indicate that applications of movement take place at LF which are fully parallel to operations of overt movement and therefore subject to the same set of purely syntactic constraints (such as the ECP etc). Such a rather natural methodological position can be informally referred to as the LF Movement Heuristic:
1
Although it is true that overt Quantifier Raising (QR) has been reported for Hungarian and certain other languages, it is certainly less common than overt wh-movement, and overt longdistance anaphora movement does not seem to have been attested in any language. As for instances of A-movement such as EPP-checking or Case-licensing, its very local clause-bound nature generally makes it less of a potentially useful tool to investigate the hypothetical properties of LF than either wh-movement or other quantificational A′-dependencies.
4 (6)
W h-movement and the theory of feature-checking The LF Movement Heuristic An analysis of LF-movement relating an element A to a higher position B is justified if either (i) or (ii) hold: (i) The interpretation of A is similar to elements which occur overtly in positions such as B. (ii) The relation of A to B is subject to constraints which otherwise characterise overtly-formed syntactic (movement) dependencies.
In attempting to argue for LF and LF movement it has then commonly been assumed that one needs to show signs of a clear parallelism bet ween overt and covert dependencies and that the most convincing arguments for LF movement are those which are able to demonstrate that a hypothetical covert dependency has all the apparent properties of an overtly-formed movement relation. This quite general assumption I will call the LF Parallelism Assumption: (7)
The LF Parallelism Assumption LF-formed movement dependencies are essentially parallel to those created in overt syntax.
Although early pioneering work in Huang (1982) arrived at the conclusion that applications of LF movement are not in fact subject to the identical set of syntactic constraints which restrict overt pre-LF movement, the strength of his argumentation for LF actually lay in the notable parallels pointed out bet ween overt and covert movement. Subsequent investigators such as Nishigauchi (1990) and Pesetsky (1987) have furthermore sought to eradicate the differences posited by Huang and argued that subjacency is indeed a constraint on both overt and covert movement relations, so the drive to find uniformit y and parallelism bet ween overt and covert syntax has continually been strong. Most recently this position was expressed in Chomsky (1993) with the declaration that a syntactic derivation from the point of lexical insertion through to LF should be taken to be subject to uniform computational constraints, and that there should be no syntactically significant level of S-structure distinguishing overt from covert movement. Otherwise put, the difference bet ween overt and covert movement should reduce simply to the audible reflexes of overt movement which occur in the PF output form. Parallelism arguments were also used as essential means to justify and defend a covert movement approach in the face of criticisms that other non-movement means of representation such as binding and co-indexation might be able to model in situ wh-dependencies equally well. Huang (1982) suggests that covert movement is to be preferred to any suggested co-indexation equivalent precisely
W h in situ and the LF Movement Hypothesis
5
because wh in situ dependencies display classic properties of overt movement relations, i.e. the apparent parallelism with overt movement is again what really motivates a covert movement analysis (and ultimately LF as well). Substantially later on, in Chomsky (1995) (following on from certain brief remarks in Chomsky 1993) a rather different general conception of covert LF movement has been seen to emerge which basically rejects the fundamental and key notion of parallelism with overt movement, and the idea that in situ wh-phrases undergo LF movement. This dramatic shift in position is due to a variet y of interacting factors introduced in Chomsky (1995). First of all, the suggestion that there is a significant ±interpretable divide amongst morphological features together with the claim that strong features may only be introduced in the pre-Spell-Out portion of the derivation conspires to rule out the possibilit y of any LF wh-movement. Wh-features are taken to be +interpretable and are suggested to require checking only if strong on a +Q C0, hence only in the overt part of a derivation. Secondly, movement taking place bet ween the point of SpellOut and LF is suggested to be movement of only formal features (ff-movement) and not of any syntactic category containing such features. ‘Category’ movement is argued to take place overtly solely for reasons of PF Convergence and is therefore unnecessary in the covert syntax. Semantic features are similarly taken to be stranded by ff-movement. Concerning the first point, in Chapter 2 it will be argued that wider crosslinguistic evidence indicates that the idea that wh-features are in need of checking only on a +Q C0 (if strong) is simply incorrect and that there are good reasons to believe that all wh-phrases do indeed have a (wh)-checking requirement (hence that the standard pre-Minimalist view that all wh-phrases require licensing in some way is indeed right). LF wh-movement or some alternative mechanism must then be assumed to encode this licensing process. The second rather dramatic innovation noted above, the switch to LF ff-movement, was made in an attempt to add some intuitive content to the economy principle of Procrastinate and account for why movement made at LF might be “cheaper” and more economical than overt pre-Spell-Out movement. The suggestion that for PF reasons overt movement is forced to carry along more than is really needed for simple feature-checking and that LF-movement is not subject to such a constraint and so only raises feature-sets does indeed achieve this aim, but also at a hidden high cost. A critical problem created by the idea of a covert operation of ff-movement which by inherent definition has no parallel in overt syntax is that it is no longer possible to provide any strong empirical justification for such an operation as movement. If ff-movement is taken to be fully different from overt movement stranding both semantic features and any
6
W h-movement and the theory of feature-checking
containing syntactic categories, then one can no longer expect to find supporting evidence either of t ype (6i) or of t ype (6ii) to motivate the idea that movement is involved.2 The rejection of parallelism with overt movement, which was always the central convincing argument-base for earlier analyses of LF-movement, has now left the formally vague idea of ff-movement without any means to justify itself as movement as opposed to some other non-movement co-indexation operation which might indeed be effected without any need for a discrete syntactic level of LF. This latter point is rather important and so will be emphasised again. In order for arguments for a level of LF and for LF movement to be convincing and stronger than any non-movement co-indexation alternative, they basically do indeed need to show parallels with overt movement. If one now suggests that covert movement is in fact radically different from regular overt raising, this possibilit y of motivating LF and LF movement via a comparison with overt movement is automatically lost. Connected with this, it must also be borne in mind that the very existence of LF and LF movement depends directly upon evidence established under the assumption that overt and covert movement do largely parallel each other. If this assumption is now withdrawn, the evidence built upon it also disappears and there is no longer any substantiating motivation for a level of LF. Consequently it is far from obvious if it may be simply announced that operations of covert ff-movement will lead to a further level of representation, LF; such a claim/position has little to support it empirically and basically denies all prior justification of the level of derivation whose existence it
2
As ff-movement is suggested to take place to an X0 head position to check against other matching features, one might imagine that ff-movement could be thought of as the equivalent of (overt) syntactic head-movement. However, whether they are fully phrasal or simply X0 heads, syntactic categories are only ever carried along for PF convergence and so ff-movement cannot be taken to raise up on a syntactic head category — ff-movement is ‘movement’ of a bundle of features which has no phrase-structural status from within an X0 and is therefore quite different from the movement of a syntactic category (such as an X0-head). Furthermore, the hypothesised ff-movement would also not seem to have the expected properties of regular head-movement. For example, in English the head Neg0/not is generally taken to block head-movement of verbs to T0 and so require do-insertion in negative sentences; however subject ff-movement to T0 does not appear to be blocked by Neg0 which is unexpected if it is indeed an application of head-movement similar to V-to-I(-to-C): (i) There does not appear to be a problem. (ii) It does not appear [that he is here]. In Chapter 2 it will also be argued that all wh-phrases require wh-feature-checking and that this is possible long-distance over all kinds of intervening heads in a way which is quite unexpected on a simple head-movement account.
W h in situ and the LF Movement Hypothesis
7
ironically needs to assume.3 Indeed, once the evidence for LF and LF movement which makes use of the idea of parallelism is factored out, it might seem that the only remaining motivation for assuming that ff-movement is in fact a movement operation is the assumption that feature-checking relations must be effected within a strict and uniform notion of localit y, hence that there must be movement of features to a checking head. In Chapter 2 it will be argued that the strict localit y conditions assumed for feature-checking relations are in fact not universal as a constraint and so even this support for a movement approach disappears. For the above reasons, the general investigation carried out here will focus on approaches to LF/wh-movement which are rather more standard than more recent suggestions of ff-movement. A central question we wish to provide some kind of answer to is whether there does indeed exist any kind of LF wh-movement and therefore to determine the actual degree of support that in situ wh-patterns give for LF-based models of syntax. In order for the answer to such a question to enjoy the greatest chance of success and be positive, our approach will be to consider the strongest most convincing arguments to date, and these arguments all do expect that at least a certain amount of parallelism bet ween overt and covert LF movement should exist. Were we to base our investigation on the ff-movement assumption that covert movement should not necessarily be expected to exhibit parallels with overt movement, it would soon become apparent that the evidence, strength of argumentation and justification one could make for a syntactic level of LF and ff-movement as a process of movement rather than some other co-indexation relation would really be very little indeed and far from convincing. We will therefore begin with a number of classic arguments which focus on certain parallelisms bet ween overt and covert wh-dependencies and review suggestions that the patterning observed may best be captured by assuming the existence of covert applications of movement at a level of LF. In Section 3 we will then however turn to a wide variet y of further cases involving wh in situ dependencies where anticipated parallelism effects do not in fact show up and argue that this provides a strong set of arguments against LF wh-movement. Returning to the original array of parallelism facts examined in Section 2, it will be shown that there are highly plausible alternative accounts of the relevant phenomena which do not rely on any covert movement
3
Noting that in order to justify an analysis of movement, it is not sufficient simply to find that a posited ff-movement relation may be subject to certain restrictions and not entirely free. A number of non-movement dependencies such as the licensing of bound-variable pronouns and NPIs are observed to be constrained by localit y-like effects. The identification of a restriction on a licensing relation therefore does not automatically mean that movement is necessarily involved.
8
W h-movement and the theory of feature-checking
operation and hence that the case for LF wh-movement and for LF itself is severely weakened. This will then lead us naturally into Chapter 2 where the role of movement in the licensing of wh-dependencies is given thorough re-examination and a rather different notion of parametrised localit y is argued to characterise the licensing (checking) of wh-features.
1.3. Arguments for LF wh-movement 1.3.1. W h-movement as the construction of an input form to interpretation The occurrence of overt wh-movement in many languages has often been argued to be a consequence of certain inherent logical properties of wh-phrases interacting with constraints on acceptable input forms to interpretation. Noting that wh-questions such as (8) below may appear to bear striking resemblance to the logical representation they could be given within Predicate Calculus, as e.g. in (9) or (10), it is suggested that partial logico-semantic representations which will serve as direct inputs to general processes of interpretation need already be constructed within the syntactic component: (8) (9) (10)
[Which car]i did John buy ti? For which x, x is a car, is it the case that John bought x? ?x [x a car] John bought x i i i
It is argued that raising of the wh-phrase in (8) will give rise to an operatorvariable structure in which the wh-element ‘which’ functions as an operator, the N′ (or NP) car encodes its restriction, and the trace left behind by movement receives interpretation as a variable ranging over car-valued entities just as in the corresponding logical representations. In such a view then, it is proposed that tokens of language will be accepted by central reasoning processes only when presented in specific formats, with all operator-variable relations made explicit among other information. If overt wh-raising therefore takes place essentially in order to build a structure of a t ype necessary for interpretation, it is argued that such an operation should be taken to affect all wh-phrases across all languages at some point within any derivation; assuming that constraints imposed by central cognitive processes on acceptable input forms are not subject to variation and that wh-phrases constitute a single logical t ype with the same properties regardless of whatever language is considered, this should always force wh-raising to a +Q Comp position prior to
W h in situ and the LF Movement Hypothesis
9
interpretation. Where overt wh-movement is not observed to take place (as e.g. in Chinese) it is suggested one must conclude that such raising occurs covertly (yet still within the syntactic component), and before a string is fed off for interpretation, hence for sentences (2) and (4) this resulting in the representations (3) and (5). Such a hypothesis is however clearly in need of further independent support before it can be accepted as conclusive of covert wh-movement. One could convincingly argue that overt wh-movement is in fact triggered for reasons quite other than those suggested above, perhaps as a functional assist in parsing to identify a clause as a wh-question as suggested in Cheng (1997). There are also other instances of raising (such as e.g. Focus) where it cannot be argued that movement to a clause-initial position takes place in order to build an operatorvariable structure necessary for interpretation, so there must be other possible motivations available for this kind of movement. Furthermore, were there to be a constraint on input forms to interpretation that the scope of all logical operators need be made explicit in some format parallel to Predicate Calculus, one might expect that elements such as tense and (sentential) negation would raise to clause-peripheral positions in similar fashion, yet this does not appear to be the case. We therefore now turn to consider what other t ypes of primary evidence and argumentation have been put forward as support and motivation for the LF movement hypothesis, this including superiorit y, crossover and various localit y phenomena. In all these cases it can be suggested that the assumption that there is covert wh-movement will automatically allow one to explain a range of ungrammatical examples via principles and constraints already justified and necessary for quite independent phenomena.
1.3.2. W h in situ and strong crossover It has been argued that the unacceptabilit y of “strong crossover” sentences such as (11) below may be accounted for in terms of the Binding Theory. If an empt y category trace left by wh-movement is taken to be a null R-expression, c-command of such a trace by a co-referential NP will result in a principle C violation: (11)
*Whoi did hei say ti had bought the Porsche?
Similar unacceptabilit y is also observed to be present in cases where no overt wh-movement takes place and an in situ wh-phrase is c-commanded by a coreferential NP:
10 (12)
W h-movement and the theory of feature-checking *When did hei say Mary helped whoi?
In order to account for the impossibilit y of co-reference bet ween the pronoun and the wh-phrase in (12) it is proposed that covert wh-movement will apply to (12) resulting in a [ p, a] empt y category subject to principle C, which hence disallows c-command by the co-indexed pronoun he. Sentences such as (11) and (12) are thus both accounted for in a parallel way and by means of the independently-justified Binding Theory.
1.3.3. Superiority A second argument which has been standardly taken to indicate that there is LF movement of wh-phrases in situ relates to those ungrammatical sentence t ypes known as superiorit y violations, illustrated in (13)–(15): (13) (14) (15)
Whoi ti hid what? *Whati did who hide ti? *Whati did Mary fix ti how?
It has been suggested that if one makes the assumption that in situ wh-phrases undergo LF raising (and also that a Comp to which multiple wh-phrases have raised may only be indexed in a certain way) then the unacceptabilit y of (14) and (15) can be straightforwardly ruled out in terms of the ECP, hence again via a principle already motivated on the basis of other phenomena. After LF raising in (14) and (15), the subject and adjunct wh-phrases will not be able to c-command and antecedent-govern their traces and so give rise to an ECP violation; in (13) by way of contrast, the trace of the object wh-phrase raised at LF will be lexicallygoverned by the verb and no violation will arise.
1.3.4. Pesetsky (1987) and d-linking Pesetsky (1987) suggests that certain restrictions on the possible interpretation of wh-phrases occurring in situ within wh-islands may be accounted for if one assumes that at least a sub-set of wh-elements are forced to undergo raising to a +Q Comp at LF, with this raising being constrained by subjacency in the same way that overt wh-movement is. It is first argued that the in situ object wh-phrases in examples such as (16) and (17) may receive interpretation as being directly questioned only if they are understood as questioning the reference of elements of a restricted set whose full membership is known to both speaker and hearer — d(iscourse)-linked in Pesetsky’s terms:
W h in situ and the LF Movement Hypothesis (16) (17)
11
Who remembers where we bought what/which book? Who wants to know where Bush talked about what/which point?
Pesetsky proposes that d-linked wh-phrases need not undergo raising for their interpretation but may be bound in situ by any c-commanding +Q Comp. Such non-movement binding of a wh-phrase will not be subject to constraints on movement and so when the object wh-phrases in (16) and (17) are taken to be d-linked they consequently may be bound by the +Q Comp in the higher clause (despite this lying exterior to the lower wh-island CP), resulting in an interpretation of the wh-phrases being directly questioned. Because the interrogative scope of a non-d-linked what in (16)–(17) does however seem to be restricted to the +Q Comp of the lower clause, this is taken to indicate that non-d-linked wh-phrases are indeed required to raise to a +Q Comp for interpretation; LF movement of a non-d-linked what to the higher +Q Comp in (16)–(17) will not be possible as this would violate subjacency, just as overt extraction from wh-islands does: (18)
*Whati does he want to know where Bush said ti?
Non-d-linked what in (16)–(17) may therefore only raise to the lower +Q Comp and be interpreted as indirectly questioned. Contrasts in the scope of in situ d-linked and non-d-linked wh-phrases relative to higher +Q Comps are thus accounted for in terms of subjacency constraining the obligatory application of LF movement to certain t ypes of wh-phrase.
1.3.5. W h-adjuncts and extraction islands A fourth argument often presented in favour of an LF-movement approach to wh-phrases occurring in situ at PF concerns the distribution of wh-adjuncts and extraction islands. Huang (1982) notes that elements such as weishenme ‘why’ and zenme ‘how’ in Chinese may not occur within island configurations when they relate to a +Q Comp exterior to the containing island, as for example in (19) where weishenme is inside a complex NP: (19)
*[[Ta weishenme xie] de shu] zui you-yisi ne? [[he why write REL book most interesting Q Intended: What is the reason x, such that a book that he wrote for x is the most interesting?
Huang suggests that the fact that wh-adjuncts may not occur in such environments (with scope at a higher +Q C) is good indication that wh-phrases in Chinese must undergo covert LF raising to a +Q Comp. In examples like (19)
12
W h-movement and the theory of feature-checking
this will give rise to a violation of the ECP when antecedent-government of the trace left by extraction is blocked by the barrierhood of the CNP. On the basis of this kind of localit y phenomena and a variet y of other theoretical argumentation such as that presented immediately above it has therefore been proposed that wh-phrases occurring in situ at PF must establish some relation to a +Q Comp, and that importantly the creation of such a dependency would appear to bear the hallmarks of parallel dependencies established via overt wh-movement. A range of restrictions on the distribution of in situ wh-phrases may receive direct and simple explanation in terms of principles already claimed to constrain applications of movement if it is presumed that the relation of an in situ wh-phrase to a +Q Comp is indeed the result of such a movement operation taking place, rather than just co-indexation of the wh-phrase and Comp. As such hypothetical raising-to-Comp is not perceivable in the PF form of a string, it must consequently be assumed to take place after the point of S-structure and therefore that the syntactic derivation of a string does not necessarily terminate at S-structure but may continue on until some further stage. One is thus led to posit a level of LF in addition to any other derivational levels assumed, LF having the properties of being formed within the syntactic component via applications of movement fully parallel to those occurring in the overt syntax and being constrained by clear syntactic principles such as the ECP and the Binding Theory. There is hence both theoretical and empirical motivation in support of the LF Wh-Movement Hypothesis. In what follows though, it will be argued that a view in which wh-phrases occurring in situ at PF do not undergo any form of covert raising is actually to be preferred. It will be shown that a whole array of theoretical argumentation and empirical data relating to a number of unrelated languages provides strong evidence that where wh-phrases do not raise to a +Q Comp by S-structure/Spell-Out, they are not forced to do so (and indeed may not do so) at any point in the syntax. Having claimed that wh-phrases in certain languages may remain in situ throughout a derivation we will then in Section 1.5 return to the arguments given for LF wh-movement reviewed above and suggest that these and other arguments relating to scope, selection and absorption may be accounted for without the need to posit covert wh-raising.
1.4. Against LF wh-movement If wh-phrases in situ at PF were to be subject to an obligatory raising operation taking place at LF then a certain parallelism in behaviour may be expected
W h in situ and the LF Movement Hypothesis
13
bet ween these in situ wh-phrases and other elements which are observed to undergo overt raising. If such parallelism does not exist, or if it cannot be motivated by independent properties of PF (assuming now the Minimalist view that constraints may not be stated relative to any level of S-structure), then this would seem to constitute direct evidence against the LF raising approach and for an analysis in which wh-phrases appear in their in situ positions throughout a derivation. The arguments and data presented below fall into three basic t ypes: (a) instances where the actual movement operation of pre- and post-Spell-Out elements would appear to be constrained in different ways, (b) cases where the interpretative possibilities open to in situ and overtly moved wh-items are not the same, and (c) instances where an LF movement analysis would have to admit that there are conditions on an independent level of S-structure (which again must be avoided in a Minimalist approach). Three well-known examples representing each of the above three t ypes will first be reviewed, followed by a variet y of other similar cases all pointing to the same conclusions concerning LF movement.
1.4.1. Non-parallelism with regard to movement — locality problems The first case we consider here is a highly important one relating directly back to Section 1.3.5 and the argument that LF wh-movement may be motivated on the grounds that the distribution of wh-phrases appears sensitive to extraction islands. Although Huang (1982) has argued that the unacceptabilit y of wh-adjuncts in island configurations may be accounted for by assuming LF wh-movement is constrained by the ECP, serious problems for the LF movement hypothesis arise when one considers the status of subjacency relative to such movement. The general pre-Minimalist empirical justification for assuming LF raising of wh-phrases in situ at PF is that such movement and the configurations it would give rise to can be shown to be subject to the same syntactic principles that constrain applications of movement and their output forms in the overt syntax (e.g. the ECP and Binding Theory).4 LF is then conceived of as a purely syntactic level of derivation, the result of a continuous derivational process taking place between a point at which items are inserted from the lexicon and that at which the derivation is fed off for interpretation. If covert applications of movement are
4
As noted earlier, the existence of LF and covert post-Spell-Out movement is simply assumed in the Minimalist Program. It is pre-Minimalist work making the rather natural LF Parallelism Assumption in (7) which provided the central argumentation for LF as a syntactic level.
14
W h-movement and the theory of feature-checking
instances of the same syntactic operation that affects items moved prior to SpellOut (or S-structure in pre-Minimalist models) then one should expect for it to be constrained in the same way as overt movement. However, as Huang himself has shown, this does not appear to be the case; argument wh-phrases in Chinese (and other languages, e.g Japanese, English multiple wh-questions) may licitly appear in situ in positions from which overt extraction is quite unacceptable. Examples (20) and (21) below show that relative clauses in Chinese constitute islands for both topicalisation and relativisation, arguably due to movement giving rise to such constructions; this contrasts directly with the fact that the occurrence of argument wh-phrases within such configurations is perfectly acceptable: (20)
(21)
(22)
*Zhangsani, wo mai-le [[ti xie] de shu]. Zhangsan I buy-ASP write REL book ‘Zhangsan, I bought the book that (he) wrote.’ *[Wo mai-le [[ti xie] de shu] de neige-reni] lai-le. I buy-ASP write REL book REL that-person come-ASP ‘The person who I bought the book that (he) wrote came.’ Ni mai-le [[shei xie] de shu]? you buy-ASP [[who write REL book ‘Who is the x such that you bought books that x wrote?’
The same contrast between topicalisation/relativisation and the possibilit y of wh in situ is seen with sentential subjects: (23)
(24)
(25)
??Nei-ge-ren ,
[[Lisi da-le ti] shi wo hen bu gaoxing] that-person [[Lisi hit-ASP make I very not happy ‘That person, that Lisi hit (him) made me not too happy.’ ??[[Lisi da-le t ] shi wo hen bu gaoxing] de nei-ge-ren i i [[Lisi hit-ASP make I very not happy REL that person ‘The man that [Lisi hitting (him)] made me not too happy’ [[Lisi da-le shei] shi ni hen bu gaoxing? [[Lisi hit-ASP who make you very not happy ‘Who did that Lisi hit (him) make you not too happy?’ i
Furthermore, although Chinese may seem to allow a certain amount of wh-topicalisation as in (26) (from Tsai 1994), such topicalisation is not possible at all when it would take place out of an island (thus compare movement of a whphrase in (27) with wh in situ in (22) above):
W h in situ and the LF Movement Hypothesis (26)
(27)
15
Sheii, ni renwei ti zui xihuan Lisi? who you think most like Lisi ‘Who do you think most likes Lisi?’ *Sheii ni mai-le [[ti xie] de shu]? who you buy-ASP write REL book ‘Who is such that you read a book that he wrote?’
If in situ wh-phrases in Chinese were to undergo raising to a +Q Comp at a level of LF this would clearly not appear to be subject to the same localit y constraints that affect other overt movement relations. Despite Huang’s analysis of the restrictions on wh-adjuncts in extraction islands as requiring the assumption of LF wh-movement constrained by the ECP, the overwhelming general observation repeatedly made across a large number of languages is that the distribution of wh-elements in situ does not appear to be constrained by any strict notion of localit y. It therefore may seem unlikely that the relation between an in situ wh-phrase and a +Q Comp is actually one of movement. Later we will also argue that there is reason to be rather cautious in what one attempts to conclude from the wh-adjunct data; first of all, as pointed out by Tsai (1994), the noted ban on occurrence in situ in islands does not in fact generalise to all wh-adjuncts but essentially is a restriction just on weishenme ‘why’ and zenme manner (but not means) ‘how’, and secondly there is evidence that it cannot relate to the ECP, hence that is not necessarily a restriction on any extraction process. Faced with the conflict that certain evidence may seem to point towards LF movement of wh in situ elements but that considerations of subjacency appear to indicate otherwise, two t ypes of position have been adopted. The first, taken up by Huang and various others, is to suggest that certain constraints on movement such as subjacency do not in fact apply uniformly throughout a derivation — Huang (1982) simply proposes that while movement taking place prior to S-structure is constrained by subjacency, movement occurring between S-structure and LF is not. In various work carried out prior to the advent of the Minimalist framework it has not been uncommon to claim that certain syntactic principles may apply discretely at particular points/levels within a derivation, e.g. the Binding Theory, satisfaction of the Case Filter etc. However, there would seem to be no obvious reason why subjacency should only apply to those applications of movement taking place before a certain point within a single derivation; if subjacency is a general syntactic constraint applying “blindly” to any input form then it should not matter where in a derivation such an input form is presented. The problem becomes increasingly more acute when considered from a Minimalist perspective where reference to S-structure as an independent representational level should not
16
W h-movement and the theory of feature-checking
be possible at all. In Minimalist approaches the attempt is made to justify the suggestion that certain operations must apply at particular points within a derivation in terms of properties of the interface levels PF and LF. However, there seems to be no plausible propert y of PF which would explain why applications of move(a) occurring prior to Spell-Out should have to obey strict localit y principles while those taking place post-Spell-Out need not, given also that other localit y principles such as Shortest Move are argued to apply quite uniformly throughout the derivation and the explicit declaration made in Chomsky (1993, p. 8) that: “computational principles are uniform throughout (the derivation to LF)”.5 A second possibilit y is to suggest that LF movement of in situ wh-phrases may be different from overt wh-raising in certain critical ways, either that it may be effected in various indirect ways as e.g. in Nishigauchi’s (1990) pied piping proposal, or that it may make use of options only available at LF, as in Fiengo et al. (1988) where the potential for wh-phrases to extract from islands at LF is linked to and dependent upon QR operations taking place at this level. It is not our purpose here to present in-depth criticisms of all such proposals for reasons of limited space, but earlier critiques have indeed shown there to be severely problematic aspects inherent in each (see for example Fiengo et al. 1988 on LF pied piping, Simpson 1993 on Fiengo et al. 1988). Arguably a far simpler and more rewarding line of approach avoiding the problems and complicated LF mechanics seemingly necessary in any analysis which attempts to reconcile the occurrence of wh in situ in islands with LF wh-movement is instead to assume that LF wh-movement is not necessary, and then attempt to provide alternative accounts of those phenomena which originally motivated such proposals, especially if other supporting evidence can be found to indicate that no wh-raising does in fact occur (as will be presented below). In Chapter 2 it will be suggested that wh-movement where attested overtly does not in fact take place in order to build operator-variable structures universally necessary for the interpretation of wh-phrases, that such elements may indeed be interpreted in situ, and hence that the general lack of localit y effects with wh in situ is simply a function of movement to a +Q Comp not being forced with wh-phrases in languages like Chinese at any level of derivation.
5
Again, if an element has any kind of phrase-structural status, the presence or absence of an accompanying phonetic matrix should be quite irrelevant and not affect the application of purely syntactic constraints. If what is moved is simply features with no phrase-structural status, the problem returns that it is almost impossible to show that this is in fact syntactic movement rather than some other non-movement co-indexation relation.
W h in situ and the LF Movement Hypothesis
17
1.4.2. Non-parallelism with regard to interpretative possibilities — ‘only’ and wh in situ Section 1.4.1 above has pointed out that unless post-Spell-Out applications of movement are subject to a set of locality principles quite different in nature from those constraining overt raising, it may be argued that wh-phrases occurring in situ at PF do not undergo LF raising. There is also evidence that the interpretative possibilities open to wh-phrases in situ do not always mirror those available where overt raising has taken place, this providing further argumentation against any LF movement analysis. One such piece of evidence is provided in Aoun & Li (1993) (relating to earlier work carried out by Tancredi 1990) and concerns the potential scopal interactions of only (and its equivalent in Mandarin Chinese zhi) with wh-phrases both in situ and raised. It is noted that only/zhi may only be associated with a lexical element in its c-command domain and not the trace of an item which has undergone raising out of this domain: (28) (29)
He only likes Mary. Maryi, he only likes ti.
Sentence (28) is ambiguous in that there is one possible reading in which the quantificational force of only is associated with Mary: It is only Mary that he likes, and a second interpretation where only quantificationally restricts (just) the verb like: His relationship to Mary is only that he likes her (i.e. he doesn’t love her). However, where topicalisation takes place to raise the NP object higher than the adverb as in (29), only the second reading is possible. Such contrasts led Tancredi to propose a Principle of Lexical Association (PLA): (30)
The Principle of Lexical Association An operator like only must be associated with a lexical constituent in its c-command domain [i.e. not with the trace of any element].
The PLA holds, as expected, also where overt wh-movement has occurred: (31)
Whoi does Mary only like ti?
(31) may not have the interpretation: ‘Which person is such that he/she is the only person that Mary likes?’ but only one in which only associates with the verb like. Aoun and Li then argue that it is however possible for the quantificational force of only/zhi to be associated with wh-phrases occurring in situ: (32)
Which girl said she only liked what?
18
W h-movement and the theory of feature-checking
(33)
Ta zhi xihuan shei? he only like who ‘Which person is such that he only likes that person (and not others)?’
(32) may be interpreted as asking: ‘Which girl said of which thing that that thing was the only thing that she liked?’ and (33) that of the gloss indicated. This then suggests that in situ wh-phrases do not undergo raising at a level of LF. Were they to do so, leaving just a trace behind in the c-command domain of only/zhi there would be no way to distinguish such an LF representation from one in which wh-raising had taken place prior to Spell-Out, and the same lack of ambiguit y as observed in (31) would be predicted. As the only-wh interaction relates to the scopal interpretation of one element relative to another, it is natural to assume that this is resolved at LF, hence that it is indeed the LF representation of only relative to the wh-phrase which is critical and not any prior PF/S-structure relation.6
1.4.3. LF raising and conditions on S-structure: parasitic gaps The third t ype of case arguing against an analysis of LF wh-movement is where in situ and overtly moved wh-phrases appear to differ in their abilit y to licence various structure t ypes. This will later also be shown in Sections 1.4.10–11 with superiorit y licensing and antecedent contained deletion, but first of all here with parasitic gaps. Critically whereas overt movement of a wh-phrase to Comp licences parasitic gaps in English (providing that the anti-c-command constraint is not violated), the occurrence of a wh-phrase in situ apparently will not: (34) (35)
Whati did John send off without having copied ei? *Whoi did John give ti whatk without having copied ek?
It has commonly been stated that parasitic gaps are licensed by A′-chains resulting from overt/S-structure movement. If one does assume that wh-phrases occurring in situ at PF undergo raising at LF and so form A′-chains at this level, one might expect that parasitic gap structures would also be licensed by wh-phrases in situ in appropriate configurations. However as this turns out not to be the case, it has to remain as pure stipulation and without any obvious explanation that it is specifically and only A′-chains formed prior to Spell-Out that will allow for parasitic gaps. Apart from it not being clear what should give rise to this particular
6
See also Cheng (1997) for further discussion of the only/zhi paradigm.
W h in situ and the LF Movement Hypothesis
19
S-Structure/Spell-Out propert y of parasitic gap constructions, in a Minimalist model of syntax it is also argued that syntactic constraints may not be stated over any level of S-Structure (or the point of Spell-Out) but in terms of the interface levels alone, PF and LF. If other data and theoretical arguments lead one to the claim that wh-phrases in situ at PF do not undergo any raising at LF then this problematic aspect of the licensing of parasitic gaps automatically disappears. It can straightforwardly be stated that parasitic gaps are licensed by A′-chains formed anywhere within a derivation and that the reason why wh-phrases in situ do not allow for parasitic gaps is simply because they never give rise to A′-chains, remaining in their in situ positions throughout until LF/interpretation.
1.4.4. Antecedent contained deletion (1) We now continue with three further cases which again illustrate the problem of non-parallelism with regard to movement, the first of these relating to antecedent contained deletion (ACD) structures. In Section 1.4.1 it was noted that in order to maintain an LF raising analysis of in situ wh-phrases, Huang (1982) and Watanabe (1991) have suggested that subjacency might in fact be a constraint only on pre-S-structure/pre-Spell-Out instances of movement. Here it will be argued that there is good evidence from ACD which indicates that such a position cannot be maintained and therefore that the localit y issue noted earlier really does remain a serious problem for any LF movement approach to wh in situ. Specifically, standard island effects noted to occur with ACD indicate that subjacency constrains not only pure movement relations but also other dependencies encoded at LF which cannot be the result of (pre-Spell-Out) movement. This leads to the conclusion that subjacency does indeed constrain LF represented dependencies and therefore should constrain also those LF dependencies which result from hypothetical LF wh-movement (contra suggestions in Huang/Watanabe). May (1985) suggests that expressions involving ellipsis are interpreted at LF via a reconstruction process in which elided material may be copied from material and structure linguistically present elsewhere in a sentence. In the case of VP ellipsis, it is claimed that a phonetically spelt-out VP will be copied into a second empt y VP position: (36)
S-structure: John has [VP gone], and Bill has [VP e] too. LF: John has [VP gone], and Bill has [VP gone] too.
VP ellipsis structures do however appear to be subject to a well-formedness constraint that neither verb may c-command the other:
20 (37)
W h-movement and the theory of feature-checking *John [VP likes the man who Bill does [VP]].
May has argued that in such instances of Antecedent Contained Deletion (ACD) copying of a potential antecedent VP into the elided VP position will ultimately lead to an infinite interpretative regress, with the result that ellipsis resolution will not in fact be possible. Despite this, sentences parallel to (37) are quite interpretable where the antecedent VP contains a quantificational object, as in (38) and (39): (38) (39)
John [VP likes everyone who Bill does [VP]]. Joan [VP read all the books that Sue did [VP]].
May suggests that if such quantificational NPs induce QR at LF, then the infinite regress facing reconstruction of the elided VP will actually be avoided. After QR of the NP, an antecedent VP will be available for copying which technically no longer contains the elided VP itself, but just a verb and the trace of the QR’ed NP: (40)
[everyone who Bill does [VP2]]i [IPJohn [VP1likes ti]]
From (40) VP1 will be copied into the elided position of VP2, and give rise to a form that may be successfully interpreted: (41)
[everyone who Bill does [VP like ti]i [John likes ti]
Critically relevant to the localit y issue under discussion is that ACD structures have been noted in Haik (1987) to be subject to the same island constraints that affect overt movement: (42) (43) (44) (45)
*John read everything which Mary believes the report that he did [VP]. *John read everything which Mary wonders why he did [VP]. *John read everything whichi Mary believes the report that he read ti *John read everything whichi Mary wonders why he read ti
However, under standard analyses, no movement actually takes place in elliptical structures of this t ype, the dependency between the relative pronoun and an embedded co-indexed variable arising only at LF after reconstruction/copying of a VP antecedent into the empt y VP position has occurred. Therefore, in order to rule out (42) and (43) as subjacency violations parallel to (44) and (45) one is forced to assume that subjacency may also be a constraint on LF representations. If subjacency does then constrain operator-variable dependencies formed at LF, it is clearly predicted that those resulting from LF wh-movement of wh-phrases in situ as hypothesized by Huang (1982) etc should also be subject to subjacency, and in the case of dependencies formed between a +Q Comp and a position within an island configuration this should result in ungrammaticalit y. The fact that argu-
W h in situ and the LF Movement Hypothesis
21
ment wh-phrases may freely occur in situ in islands in Chinese, Japanese etc therefore suggests that no operator-variable dependency of the movement t ype is in fact formed at LF. Furthermore it seems that subjacency should actually be taken as a wider constraint, applying to all dependencies where an element is structurally displaced from the position in which it is base-generated or interpreted/ construed, this perhaps irrespective of the level or way the dependency arises. However, other relations between structurally discrete positions within a tree which do not involve such direct displacement of elements from their place of interpretation perhaps may not be constrained by intervening structure in the same way, as for example with certain co-indexation relations (e.g interpretation of pronouns as bound variables). Despite the arguments above, it might be objected that if one adopts a somewhat different analysis of ACD it may still be possible to maintain that subjacency is a constraint on purely overt movement relations and therefore that wh-phrases may licitly occur within islands due to covert wh-movement not being constrained by such localit y. One could suggest that movement does in fact take place prior to Spell-Out in ACD structures, perhaps with deletion of elements within VP at PF as opposed to LF VP-copying. Lappin (1992) has proposed that in cases of ACD such as (45) regular pre-Spell-Out movement does indeed occur between the object position and SpecCP of the relative clause, this being constrained by subjacency as all (overt) movement is. At some point prior to PF, operator raising will be followed by simple deletion of the phonetic content of the verb in the lower VP: (46)
John read everything [Oi that Mary did [VP e ti]]
However if such an approach is adopted, one then is faced with the problem of why it is apparently not possible for a resumptive pronoun to occur in place of the trace left by relativisation, especially when this potentially might rescue ACD structures which involve island violations. Although resumptive pronouns in English do not sit well in questions and relative clauses, examples can be constructed in which it should be possible to have a resumptive pronoun instead of the hypothesized trace. Parasitic gap structures in English do allow for the lexicalisation of the trace which is presumed to arise from movement of an empt y operator, and such a strategy can save otherwise unacceptable island violations, as in (47): (47)
This is the book whichi Max read ti [Oi before hearing the claim that Lucy read *ti/iti].
22
W h-movement and the theory of feature-checking
However, whereas ACD is indeed possible with parasitic gap constructions (example (48), island violations may not be saved by the use of a resumptive pronoun (49): (48) (49)
This is the book whichi Max read ti [Oi before finding out that Sue did ti] This is the book whichi Max read ti [Oi before hearing the claim that Sue did *ti/*iti].
Hence just where one should be able to base-generate a resumptive pronoun and co-index it with an operator in Comp, this appears not to be possible. Such a result is both unexpected and unaccounted for in a movement analysis of ACD and VP ellipsis in general; if all that is base-generated empt y or deleted in the ‘elided’ VP is the V0 position then it should be possible to base-generate a (resumptive) pronoun in its object position. However, if the entire VP is basegenerated empt y and only reconstructed at LF, there clearly is a principled reason why resumptive pronouns may not appear, access to the lexicon no longer being available at this point.7 Thus in sum it must be admitted that subjacency constrains both S-structure and LF dependencies which involve some kind of structural displacement of elements from the positions in which they are interpreted. Consequently LF raising of in situ wh-phrases is also expected to be subject to subjacency; as such elements may generally occur in all island environments, this strongly seems to suggest they do not undergo covert LF movement to Comp.
7
Lappin, furthermore assumes that other cases of VP ellipsis which do not involve an operator are resolved in essentially the same way as ACD; that is, rather than base-generating an entirely empt y VP constituent and reconstructing it from some antecedent VP at LF, the empt y VP has internal structure, with empt y verb and argument positions: (i) John read Ulysses, and Bill did [VP [V′ [V e1] e2]] too. Thus in (i) e1 represents an empt y verb and e2 an empt y NP object. If empt y terminal nodes corresponding to all the various sub-elements of the VP which will be copied/reconstructed are discretely base-generated (rather than base-generating a single empt y VP node) it should be possible to lexicalise an argument position with a pronoun rather than generate it empt y. In this case the pronoun would not be a resumptive pronoun because it is not associated with a coindexed binding operator; it therefore should not give rise to any of the awkwardness which may arise with operator-resumptive pronouns links in English. However, it is not possible to have a lexicalised form here either, again suggesting that the VP is base-generated empt y as one unit without separate empt y terminal nodes: (ii) *John read Ulysses, and Bill did [VP [V′ [V e] it]] too. (iii) *Bob read War and Peace, but Sam didn’t it. (iv) *I might read Barriers, and so might she it.
W h in situ and the LF Movement Hypothesis
23
1.4.5. Quechua, Chinese, the ECP and LF movement In addition to the above arguments concerning subjacency, data taken from Ancash Quechua and Chinese provide ECP-related evidence that in situ wh-phrases do not undergo any LF raising operations, contra suggestions in Huang (1982). It will be remembered that the critical localit y examples Huang uses to motivate LF wh-movement are cases where wh-adjuncts may not occur in island configurations. On the basis of a variet y of theoretical arguments concerning scope, selection and absorption among others (to be considered in a later section) Huang suggests that all wh-phrases will need to raise to some +Q Comp prior to LF; as subjacency is claimed not to constrain LF dependencies (though this is now very much in question) and the traces of argument wh-phrases will always satisfy the ECP in Chinese, such elements may freely occur in situ in island environments raising to Comp at LF.8 Where LF extraction of wh-adjuncts takes place however this will violate the ECP, antecedent-government of the wh-adjunct trace being blocked by the barrierhood of the island. Thus differences in the distribution of in situ wh-arguments and wh-adjuncts in Chinese (and other languages) are accounted for if it assumed: (a) that all wh-elements raise to Comp at LF, (b) the ECP uniformly constrains both overt and LF movement operations, and (c) subjacency is only applicable to pre-S-structure/Spell-Out movement. Such an account is seriously cast into doubt when one considers the patterning of wh-phrases in Ancash Quechua (AQ) and further data from Chinese. In wh-questions in AQ, as reported by Cole & Hermon (1994), both overt wh-movement and an in situ strategy are attested: (50)
(51)
May-man-taqi Jose munan [Maria ti aywanan-ta]? where-to-Q Jose wants [Maria will-go-ACC ‘Where does Jose want Maria to go?’ Jose munan [Maria may-nan aywanan-ta]? Jose wants [Maria where-to will-go-ACC ‘Where does Jose want Maria to go?’
As with Chinese, wh-phrases are also fully acceptable in situ in islands such as CNPs, whereas overt wh-movement from such positions results in t ypical subjacency-like violations:
8
Huang argues that both objects and subjects are always properly-governed in Chinese, objects by the verb and subjects by Infl.
24
W h-movement and the theory of feature-checking
(52)
(53)
(Qam) kuya-nki ima-ta suwaq nuna-ta? you love-2PL what-ACC steal man-ACC ‘What is x, such that you love the man who stole x?’ *Ima-ta-taqi (qam) kuya-nki ti suwaq nuna-ta? what-ACC-Q you love-2PL steal man-ACC ‘What do you love the man who stole?’
It might therefore be suggested that wh-phrases in AQ need undergo movement to Comp at some derivational point, but that only overt movement is subject to subjacency. However, there is other good evidence in AQ that in situ wh elements do not in fact undergo any kind of LF raising. It has been observed that there exists in AQ a subject-object asymmetry with regard to extraction that is highly reminiscent of the that-trace paradigm in English. Whereas object wh-phrases may raise to the Comp of a higher clause from a position within an embedded CP, this option is not open to subject wh-phrases: (54)
(55)
Ima-ta-taqi Fuan musyan [Rosa ti ruranqan-ta]? what-ACC-Q Juan knows [Rosa made-ACC ‘What does Juan know that Rosa made?’ *Pi-taqi Fuan musyan [ti tanta-ta ruranqan-ta]? who-Q Juan knows bread-ACC made-ACC ‘Who does Juan know that made bread?’
Cole & Hermon suggest that (55) should be analysed as a straightforward ECP violation, the subject position not being properly-governed, just as in illicit cases of subject extraction where an overt complementizer occurs in English:9 (56) (57)
*Whoi did you say that ti came? Whoi did you say that you saw ti?
Despite extraction from subject positions being ill-formed it is nevertheless found that wh-phrases may occur in situ in embedded clause subject positions: (58)
9
Fuan musyan [pi tanta-ta ruranqan-ta]? Juan knows [who bread-ACC made-ACC ‘Who does Juan know made bread?’
It cannot be a subjacency violation as object extraction from embedded CPs is fine.
W h in situ and the LF Movement Hypothesis
25
This crucial piece of data would seem to be irreconcilable with Huang’s position on wh in situ. It is at once suggested that in situ wh-phrases must undergo LF raising in order to satisfy quite general cross-linguistic properties of selection, scope and absorption and also that all LF movement is subject to the ECP. In (58) LF raising of the wh-phrase subject should give rise to an ECP violation parallel to (55), yet such examples are perfectly well-formed. It can therefore be concluded either that wh-phrases in AQ do not undergo LF raising to Comp and hence that general properties of selection, scope etc do not necessitate LF raising of in situ wh-phrases, or that the ECP does not in fact constrain LF movement. The former conclusion would have for effect that the motivation for LF raising in Chinese would then reduce solely to the empirical wh-adjunct localit y facts and there would no longer be any general theoretical reasons to trigger and explain the need for such raising. The latter conclusion if adopted would mean that restrictions on the distribution of in situ wh-adjuncts in Chinese can no longer be taken as indication of movement — if the ECP does not in fact constrain LF raising then the unacceptabilit y of wh-adjuncts in islands in Chinese cannot be ascribed to any ECP violation and would have to be attributed to some other non-movement constraint on wh in situ. Supposing one were in fact to suggest that neither of the principles taken to constrain movement operations (subjacency and the ECP) were actually to be operative at LF there would then no longer seem to be any legitimate or justifiable reason for considering the wh in situ to Comp dependency as one of movement, as it would no longer exhibit any of the key identifying properties of a movement relation. This is indeed the conclusion which arguably ultimately needs to be made; if a dependency between a wh-phrase and a +Q Comp must be established in the course of a derivation then it can be suggested that in a large number of languages this may be effected without the need for movement to relate the wh-phrase to the +Q Comp. Where a wh-phrase is phonetically interpreted in situ as in the AQ examples above the evidence would strongly appear to indicate that no movement operation to Comp takes place at any point and that wh-phrases may therefore generally be interpreted without any raising to Comp.10, 11
10
Lisa Cheng (p.c.) notes that similar arguments to those put forward here can also be presented on the basis of English as observed in Lasnik & Saito (1992). If the wh-subject who were to undergo LF raising to Comp in (ii) below, it might be expected to give rise to an ECP violation as in (i), yet examples such as (ii) are perfectly acceptable: (i) *Whoi did John think that ti came? (ii) Who thought that who came?
11
For the overt raising of wh-phrases in AQ see the discussion related to Basque in Chapter 3.
26
W h-movement and the theory of feature-checking
It should also be noted that the general dilemma here cannot be resolved by suggesting that the ECP is somehow a constraint on PF rather than LF (as proposed in Aoun, Hornstein, Lightfoot & Weinberg 1987). If one were to attempt to argue that all wh-phrases in AQ do undergo LF raising and that the contrast between the subject wh cases (55) and (58) is due to overt movement being later constrained by the ECP as a filter on PF (and so not constraining any LF extraction) then one clearly loses the Chinese wh-adjunct data again, as the “ECP” violations in Chinese would indeed only occur at LF. Further arguments against taking the unacceptabilit y of Chinese wh-adjuncts in situ in islands as indicative of ECP violations (and hence LF movement) can be given from within Chinese too. Tsai (1994) reports that the t ypes of wh-element which do give rise to unacceptabilit y when occurring in situ in islands are actually quite limited and reduce to just weishenme ‘why’ and zenme(-yang) ‘how’ when the latter has a manner reading (but not when understood as questioning means). Other wh-adjuncts may in fact occur quite licitly in island configurations, as in (59) and (60) below:12 (59)
(60)
Ni bijiao xihuan [[ta zenmeyang zhu] de cai]? you more like [[he how cook REL food ‘What is the means x, such that you prefer the dishes which he cooks by x?’ [[Tamen zenmeyang chuli zhe-bi-qian] de shuofa] bijiao kexin? [[they how handle this-CL-money REL story more believable ‘What is the means x, such that [the story [that they handled this money by x]] is more believable?’
Tsai also shows that Chinese allows for certain overt topicalisation/fronting of wh-phrases: (61)
(62)
Sheii ni renwei [ti zui xihuan Lisi]? who you believe most like Lisi ‘Who do you think most likes Lisi?’ Shenmei ni renwei [Lisi zui xihuan ti]? what you think [Lisi most like ‘What do you think Lisi likes most?’
However this kind of wh-fronting is impossible with any t ype of wh-adjunct — neither weishenme ‘why’, zenme(-yang) ‘how’ nor other wh-adjuncts such as shenme-
12
This is also true for Japanese.
W h in situ and the LF Movement Hypothesis
27
shihou ‘when’, zai-nali ‘where’, zenme(-yang) ‘means-how’, even where this would not involve any island violation: (63)
(64)
(65)
(66)
*Zai-nalii ni renwei [ta ti gongzuo]? where you think [he work Intended: ‘Where do you think that he works?’ *Shenme-shihoui ni renwei [ta ti shui-jiao]? when you think [he sleep Intended: ‘When do you think he sleeps?’ *Zenmeyangi ni renwei [Lisi ti yinggai chuli zhe-jian-shi]? how you think [Lisi should handle this-CL-thing Intended:‘What is the means x, such that you think Lisi should handle this thing by x?’ *Weishenmei ni renwei [ti Lisi cizhi]? why you think Lisi resign Intended: ‘Why do you think that Lisi resigned?’
As this movement cannot violate subjacency (and is good with argument wh-phrases) the unacceptabilit y of (63)–(66) is attributed to violations of the ECP, the extraction-sites of the adjuncts not being properly-governed (assuming a Rizzi 1990-t ype view of the ECP). If this is true, and there seems to be no other reason why fronting of wh-adjuncts should not be possible if wh-argument fronting is acceptable, then the extraction and movement of wh-adjuncts in Chinese should not be possible anywhere at any level — an ECP violation should always occur wherever an adjunct is extracted, due to its extraction-site not being properlygoverned. Given then that all adjunct types nevertheless may occur in situ in simple clauses without giving rise to any unacceptability (67)–(68), it can justifiably be argued that the +Q Comp-wh-phrase dependency in these good cases is satisfied in some other way which critically does not involve movement and extraction:13 (67)
(68)
13
Ni renwei [Lisi weishenme cizhi]? you think [Lisi why resign ‘Why do you think Lisi resigned?’ Ni renwei [ta zai-nali gongzuo]? you think [he where work ‘Where do you think he works?’
Here the conclusions are different from those in Tsai (1994) where it is assumed that wh-adjuncts such as weishenme ‘why’ and zenme manner ‘how’ are in fact associated with some kind of movement and extraction.
28
W h-movement and the theory of feature-checking
The unacceptabilit y of (63)–(66) compared with (59)–(60) also strongly argues against Huang’s original conclusion that the ill-formedness of weishenme ‘why’ and zenme(yang) ‘how’ in situ in islands is to be ascribed to the ECP and obligatory LF movement. The raising of any wh-adjunct has been argued above to give rise to an ECP violation (63)–(66) yet the occurrence of wh-adjuncts such as zainali ‘where’, shenme-shihou ‘when’ and means zenme(-yang) ‘how’ in situ within islands is not ill-formed at all (59)–(60). It must therefore be assumed that in these environments they are licensed in situ and without any extraction/movement (as otherwise there would be violations of the ECP). The fact that weishenme ‘why’ and manner zenme(-yang) ‘how’ are however quite unacceptable when occurring in situ in similar islands can consequently not be attributed to any forced LF raising of (all) wh-phrases interacting with the ECP, as then not only weishenme/zenmeyang but all other adjuncts should be ill-formed in such configurations. It should again be noted that invoking the ECP as a filter just on PF to rule out the illicit cases of overt wh-adjunct fronting (but permitting such elements to raise at LF, even from within islands) will not allow one to maintain a coherent LF movement approach either, as then it would no longer be possible to claim that the ill-formedness of weishenme/zenme(-yang) in islands might be due to the ECP constraining LF Movement (i.e. Huang’s original proposal). Ultimately then it can be shown in Quechua and Chinese that the relation between an in situ wh-phrase and a +Q Comp is constrained neither by subjacency nor the ECP and hence would not appear to be a dependency resulting from movement to a +Q Comp. Chinese does exhibit certain restrictions on the distribution of elements such as weishenme and zenme(-yang) yet we have argued that these cannot be reduced to a violation of principles which constrain operations of movement or extraction (the ECP). In Chapter 2 we will present further evidence from a variet y of other languages which clearly indicates that a notion of non-movement localit y constraining the licensing of wh-phrases in situ is indeed necessary. Regarding the apparent exceptional sensitivit y of weishenme/zenmeyang to certain t ypes of containing structure, we suggest that this may also best be analysed as an instance where the licensing of a (wh-)dependency is subject to localit y restrictions although no actual movement occurs in establishing the wh-Comp dependency (this being similar to other non-movement dependencies such as clitic left dislocation in Italian which also appear to be constrained by strict localit y conditions — see Cinque 1990 and Chapter 2 for discussion).
W h in situ and the LF Movement Hypothesis
29
1.4.6. Scrambling and wh in situ Having considered the status of subjacency and the ECP with regard to the distribution of wh-phrases in situ and concluded that overt raising operations would seem to be constrained in ways quite different from any that might be hypothesised to take place at LF, we now look more briefly at a last case where there is a potentially significant lack of parallelism between overt movement and hypothetical wh-raising at LF. This concerns certain contrasts observed in Japanese scrambling data such as (69) and (70) quoted from Saito (1986): (69)
(70)
*[[Mary-ga ti yonda to]k [sono-hon-o]i [John-ga tk itta]]] [[Mary-NOM read that [that-book-ACC [John-NOM said ‘John said that Mary read that book.’ ??[[John-ga dono-hon-o toshokan-kara karidashita to]i [Mary-ga [[John-NOM which book-ACC library from borrowed that Mary-NOM [minna-ga ti omotte-iru ka] shiritagatte-iru koto [everyone-NOM thinking-be Q want-know-be thing ‘Mary wants to know which book everyone thinks John took out from the library.’
In (70) just one application of scrambling has taken place to raise the fronted CP out of a lower embedded clause. The example is poor yet not as fully unacceptable as (69) where scrambling has occurred twice, first to raise the object of the embedded clause sono-hon-o ’that book’ forward, and then to raise the whole lower CP to sentence-initial position. Both such operations of scrambling should in fact be licit, so Saito ascribes the strong unacceptabilit y of (69) to the fact that in the resulting configuration the NP sono-hon-o will not c-command its trace contained within the CP scrambled to a higher position. The situation is comparable to English (71) below (from Barss 1984); although in both (71) and (72) some constraint on extraction is violated, in (71) as opposed to (72) the raised wh-phrase who will fail to c-command its trace and the result is that the example is completely unintelligible. In (72) who does c-command its trace and although the example is seriously degraded it is nevertheless possible to assign it an interpretation: (71) (72)
*[Which picture of ti]k do you wonder [who]i John likes tk? i do you wonder [which picture of ti]k John likes tk ?
??Who
Turning back to (70), if one assumes LF movement to Comp of the wh-phrase dono-hon-o, this should result in a configuration with the same essential properties
30
W h-movement and the theory of feature-checking
as (69) — the trace left behind by movement of the wh-phrase in the scrambled CP will not be c-commanded by the wh-phrase once it has had to lower down to the +Q Comp of the intermediate CP.14 As a result (70) should be as equally unacceptable as (69). The contrast in acceptabilit y between (69) where overt movement takes place and (70) where a wh-element occurs in situ is therefore left unexplained if it is assumed that the wh-phrase must undergo movement at LF, and hence may be taken as an additional argument for assuming that in situ wh-phrases do not raise to Comp at LF.15
1.4.7. Anaphor-antecedent relations Now we may consider three further cases where there is a clear non-parallelism between overt and covert wh-dependencies with regard to interpretative possibilities, similar to the only case in Section 1.4.2. The first of these relates to anaphorantecedent relations and the observation that the interpretative possibilities which are available to items contained within a wh-phrase would also seem to vary depending upon whether the wh-phrase is raised or occurs in situ (that is, if a particular language has wh-phrases both raised and in situ at PF, as e.g. English). This is clearly seen in the following examples noted in Brody (1995): (73) (74)
Johni wondered [which pictures of himselfi/k] Billk liked t. *John wondered when Mary saw [which pictures of himself].
In (73) the anaphor himself may have as antecedent either the subject of the lower CP Bill or that of the matrix clause John. Overt movement of the wh-phrase to SpecCP of the lower clause has for effect that co-reference of an anaphor contained within the wh-phrase and an NP in the immediately dominating clause becomes possible in some way (see Chomsky 1993 for suggestions on how this co-reference possibilit y may technically be explained), while the link with its trace position also allows for the anaphor to take the lower clause subject as a potential antecedent (this occurring either from an LF reconstruction operation or perhaps in virtue of a full copy of the raised wh-phrase being present in its trace position,
14
It would seem that one cannot assume full reconstruction of the scrambled elements prior to LF wh-movement because then (69) should be acceptable as it would not contain an unbound trace.
15
It should be noted that it would not seem possible to invoke any pied piping analysis or the suggestion that subjacency does not constrain post-Spell-Out movement to explain this data either (i.e. those explanations commonly offered to account for differences between overt movement and the distribution of wh in situ).
W h in situ and the LF Movement Hypothesis
31
again see Chomsky 1993 for details). Given that within the Minimalist model of syntax there is no level of representation corresponding to S-structure over which constraints such as the Binding Theory might be stated, and that it would not seem possible to justify stating such interpretative constraints on PF, it must be the case that the Binding Theory applies directly to LF representations and hence not to PF/surface forms such as (74). Were wh-phrases occurring in situ at PF to undergo covert raising at LF, such an operation should bring the wh-phrase in (74) into a position in which co-reference between himself and the matrix subject John should be available, just as in (73): (75)
LF?: John wondered [[which pictures of himself]i [when] Mary saw ti]
The unacceptabilit y of (74) consequently indicates that such examples do not give rise to LF forms like (75), and so again suggests that wh-phrases occurring in situ at PF remain in such positions throughout the course of a derivation.
1.4.8. Weak crossover Weak crossover phenomena can be shown to provide arguments against LF wh-movement of a kind similar to those above, in that possibilities of co-reference between elements within a sentence (in this case between a pronominal element and a wh-phrase) may vary depending upon whether a wh-phrase is overtly raised to Comp or appears in situ at PF. In the past crossover phenomena have actually been taken as general arguments in support of LF movement, in particular as evidence for quantifier raising. It has been suggested that the impossibility of co-reference between the universally quantified NP and the pronoun in (76) should be explained in the same way that co-reference between the pronoun and the wh-phrase in (77) is unavailable: (76) (77)
*Hisi mother likes everyonei. *Whoi does hisi mother like ti?
Koopman & Sportiche (1982) have proposed that the intended co-reference in (77) is not possible because raising of the wh-phrase will result in a configuration in which an operator in SpecCP attempts to bind more than one element interpreted as a variable — the pronoun and the wh-trace, this (it is claimed) violating a bi-uniqueness constraint on operator-variable relations (the Bijection Principle). Noting that co-reference between the pronoun and the QP is equally impossible in (76), it is suggested that the QP everyone undergoes covert LF raising (QR) to a position from which it will c-command both the pronoun and its own
32
W h-movement and the theory of feature-checking
trace, hence giving rise to the same t ype of crossover configuration as observed in the case of wh-movement in (77). The Bijection Principle will then rule out (76) in a way fully parallel to (77): (78)
LF: [everyonei [hisi mother likes ti]]
Recently it has been noted that many cases which might be expected to give rise to weak crossover violations are in fact quite acceptable, contra the predictions of the Bijection Principle; in particular topic structures and non-restrictive relative clauses which exhibit the same basic configurational properties as (76) do seem to allow for co-reference between an operator in SpecCP and a pronoun that has been “crossed-over” in the course of raising to Comp (as long as the pronoun does not also c-command the trace of movement, i.e. strong crossover). This has led to various attempted reformulations of the constraints relevant to weak crossover, in Postal (1993) these being recast partly in terms of the relevant referentiality of an operator, distinguishing topic and non-restrictive relative clause operators from wh and restrictive relative clause operators (with only the latter two t ypes of operator giving rise to crossover violations). However, even with wh-questions there are occasions where an interpretation of co-reference predicted to be unavailable is actually licensed, where certain focus-related elements such as even, only and the adjective (x’s) own occur: (79) (80) (81)
Which mani did even hisi children dislike ti? Which man could only hisi children tolerate ti when hei got mad? Whoi did hisi own children betray ti to the enemy?
As (79)–(81) do allow for co-reference between the raised wh-phrase and the coindexed pronoun, this suggests there is something missing from approaches which seek to account for weak crossover phenomena purely in terms of the structural properties of the configuration and the referentialit y of the raised XP. I would like to suggest that co-reference between the pronoun and the wh-phrase even in cases like (77) is not blocked by any structural properties present in the configuration, just as in topic and non-restrictive clauses, but that pragmatically the strong tendency will be to assign independent reference to pronoun and wh-phrase. The use of even/only/(x’s) own will then make salient an interpretative possibility that is potentially always licensed in such configurations. This contrasts markedly with strong crossover sentences which never allow for co-reference of pronoun and wh-phrase even where the above focusing elements are employed, so that the impossibilit y of co-reference in such examples can indeed be attributed to particular structural properties of the relevant configurations and is not just a question of preference of interpretation:
W h in situ and the LF Movement Hypothesis (82)
33
*Whoi did even hei say ti was incompetent?
Considering now the possibilit y of co-reference between in situ elements and NPs in configurations of weak crossover, it can be noted that the use of even/only/(x’s) own does not in fact licence co-reference in the way that is possible when an element has undergone overt raising. This is found both with quantifier phrases and wh-in-situ: (83) (84) (85) (86) (87)
*Even/only hisi children dislike every mani *Hisi own children dislike every mani *Why did only hisi wife think that you could tolerate which mani? *Why did hisi own children think that Bill had betrayed which mani to the authorities? *When did even hisi children say that Mary disliked which mani?
If the wh-phrases in situ in (85)–(87) were to undergo raising at LF then the relevant configurational properties of (85)–(87) and those of the sentences with overtly raised wh-phrases would be identical at LF. This should then lead one to predict that co-reference in (85)–(87) should also be licensed where even/only/(x’s) own occur with the pronoun (again on the natural assumption that co-referential properties are computed at LF rather than PF). That co-reference is not at all possible can then be taken as further indication that wh-phrases occurring in situ at PF (and QPs it would seem) remain in situ for their interpretation at LF and are not subject to any post-Spell-Out raising operation.
1.4.9. Bahasa Indonesia A last case that will be cited here to illustrate that the interpretative possibilities exhibited by overtly-moved wh-phrases may not mirror those of wh-phrases in situ comes from Bahasa Indonesia as described in Saddy (1991). This is another language similar to Ancash Quechua in which wh-phrases may optionally occur either in situ or raised overtly to a clause-initial position: (88)
16
Sally men-cintai siapa?16 Sally TRANS-love who ‘Who does Sally love?’
The transitive prefix -men appears when there is an overt object occurring after the verb. When there is movement from object position, as in (89) the transitive affix does not occur.
34
W h-movement and the theory of feature-checking
(89)
Siapai yang Sally cintai ti? who yang Sally love ‘Who does Sally love?’
Arguably the raising in (89) may actually be for focus rather than wh-related reasons as non-wh elements are also able to front in this way and the raising of a wh-phrase does not always establish its +interrogative scope at the position moved to:17 (90) Either: Or:
Bill tahu siapai yang Tom cintai Bill know who yang Tom love ‘Bill knows who Tom loves.’ ‘Who does Bill know that Tom loves?’
An X0 element yang occurs to the immediate right of any t ype of fronted element and as this may be preceded by overt complementizers (in embedded CPs) it can be suggested that yang is in fact the head of a F(ocus)P. What is important to note here is that the overt raising of a wh-phrase to a clause-initial position results in possibilities of interpretation that are quite different from sentences with in situ wh-phrases. In (91) below the in situ whphrase may only be interpreted as having wide scope with respect to the c-commanding universally-quantified NP:18 (91)
Setiap orang men-cintai siapa? every person TRANS-love who ‘Who does everyone love?’
Being unambiguous, (91) can only be answered with a single value for siapa ‘who’ rather than pairs of values such as: ‘Mary loves John, Sue loves Bill etc’. However, where overt raising has taken place, as in (92), the sentence becomes ambiguous, there being a second reading available with the universal QP having wide scope over the wh-phrase (and hence allowing for the above pair-t ype answers): (92)
Siapai yang setiap orang cintai ti? who yang every person love ‘Who does everyone love?’
17
See also Cheng (1997) and Chapter 3 here for further discussion of wh-fronting in Bahasa Indonesia and possible connections to topicalisation, clefting and focus.
18
Huang (1982) suggests that similar restrictions occur in Chinese in parallel forms such as (i) with the wh-phrase shei ‘who’ unambiguously taking wide scope: (i) mei-ge-xuesheng dou xihuan shei? every-CL-student all like who ‘Who does every student like?’
W h in situ and the LF Movement Hypothesis
35
If this ambiguit y is a function of raising the wh-phrase to some clause-initial position from which the it will c-command the universal QP and the latter will in turn c-command a trace left behind by the movement of the wh-phrase, then any analysis in which in situ wh-phrases undergo obligatory LF raising to Comp predicts that the ambiguit y attested in (92) should also be present with wh in situ structures such as (91). LF raising of the wh-phrase in (91) should result in a configuration with the same essential properties as (92), the wh-phrase c-commanding the QP and this c-commanding a trace of the wh-phrase. As (91) is in fact unambiguous and one must assume that the scopal interaction of quantificational elements such as QPs and wh-phrases is only resolved at LF it seems one has to conclude that in situ wh-phrases in Bahasa Indonesia do not in fact undergo any covert LF raising. Again we find obvious and significant differences between the overt raising of a wh-phrase (for whatever reasons) and their occurrence in situ, differences which for all relevant purposes would disappear in any LF representation of in situ wh-phrases raised into a +Q Comp and which therefore suggest that the licensing of such elements does not involve (LF) movement.19 Consideration of the minimal pair (91)–(92) also gives rise to an interesting and important secondary conclusion. If as argued the wh-phrase in (91) does not undergo any LF movement to a position c-commanding the subject QP but is nevertheless interpreted as taking wide scope relative to it, then it can be assumed that the possibilit y of such wide scope readings need not depend on the wh-phrase necessarily having to occur in a position outside the c-command domain of the QP. This significant point will be returned to in Section 1.6.3 when we consider arguments for LF raising based on relative scope phenomena.
19
It should be stressed that the argumentation here is valid even if overt raising is taken to be focusmovement to the Spec of a FocusPhrase as assumed here (or clefted as in Cheng 1997), rather than being movement driven by wh-features. The point is that a wh-phrase-trace dependency is somehow created in which the wh-phrase c-commands the quantifier and the quantifier c-commands the trace of the wh-phrase. When this happens overtly in (92) (for whatever reason/trigger), ambiguit y results. Were the in situ wh-phrase in (91) to undergo LF raising to a +Q C c-commanding the quantifier in subject position, the relative c-command relations between wh-phrase, quantifier and trace of the wh-phrase established by this should be the same as in (92) and so there should again be ambiguit y. As sentences such as (91) are unambiguous, this clearly indicates a lack of LF raising.
36
W h-movement and the theory of feature-checking
1.4.10. Superiority licensing The final two cases against LF movement presented now relate to instances where overtly-moved and in situ wh-phrases behave differently with regard to the licensing of certain structures, in a way similar to the differences observed earlier in the abilit y of moved and in situ wh-phrases to license parasitic gap structures. The first of the two cases considered relates to multiple wh-questions and (potential) superiorit y violations. In wh-questions where both subject and object are wh-phrases it has frequently been observed that it is the subject rather than the object which must be overtly raised to Comp as seen in (93) and (94) below: (93) (94)
Whoi ti hid what? *Whati did who hide ti?
(94) has traditionally been analysed as an ECP violation, it being suggested that when who undergoes LF raising to Comp it will not be able to antecedent-govern its trace. In an alternative re-formulation of the ECP in Rizzi (1990) and Cinque (1990) it is proposed that the extraction-site in any movement chain must be properly-head-governed in order to satisfy the ECP. In (94) C0 will not be able to carry the agreement specification that would licence it as a head-governor for SpecIP as SpecCP is occupied by another wh-phrase what already by S-structure (and consequently C0 will carry agreement relating to what due to Spec–Head agreement). However one formally attempts to account for the ill-formedness of examples such as (94) it seems that it is the subject wh-phrase which is unlicensed in some way — were the subject not to be a wh-phrase then the example would be fine, vis: What did Mary hide? Considering such examples, Kayne (1984) made the interesting observation that the additional presence of a third wh-phrase in situ in such cases somehow results in licensing of the subject wh-phrase: (95)
Whati did who hide ti where?
Kayne suggests that licensing of the wh-phrase subject here is parasitic on a ‘gprojection’ created by the extra wh-element linking it to the +Q Comp. What is significant to note is that such a licensing g-projection will only be established by a wh-phrase in situ and will not be created by movement of a wh-phrase. If the relation of a trace of movement to Comp were to establish a path to the Comp with the same properties as those Kayne hypothesizes for the wh in situ then overt raising of the wh-object in (94) should technically result in a g-projection which would allow for parasitic licensing of the subject (yet (94) is quite unacceptable). One is faced with two alternative possibilities. One could attempt to maintain that all in situ wh-phrases undergo raising to Comp and suggest that it is g-projections
W h in situ and the LF Movement Hypothesis
37
established and present at S-structure which will induce the relevant licensing (hence overtly raised wh-phrases will not, as in (94)). In (95) both who and where will raise at LF but where will have played its important licensing role at S-structure being in situ at that point. However, such an approach has as direct consequence that one must allow for constraints applying to S-structure/SpellOut, a possibilit y not permitted within the Minimalist Framework.20 A second option is to suggest that in situ wh-phrases do not undergo covert raising to Comp at any derivational point. (94) and (95) will then be critically distinct at LF in that in the latter a full wh-phrase will occupy a position which may give rise to a licensing g-projection while in the former only a wh-trace will occur. Whatever mechanism one ultimately may make use of to describe the parasitic licensing effect observed in (95), this distinction between wh-phrases in situ and traces of wh-movement must still be recognised, and if licensing constraints may not be stated over S-structure one must assume that at LF the dependency of an in situ wh-phrase to a +Q Comp is encoded in some way other than via movement.
1.4.11. Antecedent contained deletion (2) The final argument we present in this section arguing against any LF raising analysis of in situ wh-phrases relates once more to Antecedent Contained Deletion. As mentioned earlier, May (1985) suggests that the infinite interpretative regress facing unacceptable cases of VP-ellipsis such as (96) below may be overcome in examples like (97) due to an operation of QR, raising the NP containing the elided VP-site out of its own VP to some higher position at LF prior to ellipsis resolution: (96) (97)
*John [VP likes the man who Bill does [VP]]. John likes everyone who Bill does [VP]. → LF: [everyone who Bill does [VP like ti]i [John likes ti]
It is therefore an (LF) raising operation triggered by the quantificational nature of the NP object in (97) which essentially distinguishes (96) and (97) and which ultimately is responsible for the well-formedness of the latter. If wh-phrases occurring in situ at PF undergo covert raising at LF, they are as a result predicted
20
Note that the constraint could not be stated over PF either, as complex syntactic structure such as g-projections should not be visible/interpretable here; the only relations that should be visible at the phonological interface are simple linear adjacency relations, remembering that PF is by definition not a level of syntax.
38
W h-movement and the theory of feature-checking
to license ACD in the same way; raising to Comp of a wh-phrase object containing an elided VP should result in a configuration similar in all relevant aspects to that arising with QP-objects after quantifier raising, and hence allow for VP-ellipsis to be resolved even in cases of ACD. However, as has been noted in Stroik (1992), ACD is not possible with wh in situ, as (99), (101) and (103) show: (98) (99) (100) (101) (102) (103)
John reviewed every student’s paper that Bill did [VP]. *Who reviewed whose paper that Bill did [VP]? Jane used every argument that she could [VP]. *Who used which argument that he could [VP]? Joan criticized everything that Mary did [VP]. Intended meaning: ‘Joan criticized everything that Mary criticized.’ *Who criticized which course that Mary did [VP]? Intended meaning: ‘Who criticized which course that Mary criticized?’
Consequently, given that the hypothetical LF raising of in situ wh-phrases should in principle create the t ype of LF structures allowing for successful VP ellipsis resolution but ACD is found to be quite unacceptable with wh in situ, the obvious conclusion which may again be drawn is that such elements do not in fact undergo any covert LF raising to Comp. Summarising now, this chapter began with a brief presentation of various theoretical arguments taken as central motivation for the LF wh-movement hypothesis. It was first seen that the assumption that wh-phrases in situ at PF undergo covert raising to Comp at a level of LF may indeed allow for potential explanation of a range of linguistic phenomena including crossover and superiorit y violations. Then however it was argued that the striking absence of localit y effects with hypothetical LF movement is strong reason to doubt a covert raising account of wh in situ. If constraints on movement operations such as subjacency are taken to apply in a uniform way throughout any derivation, as is both natural to assume and theoretically required from a Minimalist standpoint, then the fact that wh in situ to Comp dependencies may consistently violate such localit y restrictions would seem to be reasonable indication that such dependencies are not in fact established via movement. Following this, it was noted that other good evidence against the LF wh-movement hypothesis could be found, and arguments and data were presented from a variet y of languages in support of the view that wh-phrases in situ at PF are not subject to covert LF raising. If such elements may therefore remain in situ for their interpretation, contra earlier suggestions, then the evidence and theoretical argumentation originally taken as supportive of the LF (wh-)movement hypothesis appear in need of re-examination. Section 1.5 will therefore now attempt to show that there are available quite coherent ways of
W h in situ and the LF Movement Hypothesis
39
accounting for crossover, superiorit y and various other phenomena which critically do not depend on any notion of covert wh-raising.
1.5. Arguments for LF movement re-interpreted 1.5.1. W h-adjuncts in situ One important set of data which an LF raising approach claims to account for, it will be recalled, relates to localit y constraints on the distribution of wh-adjuncts in situ. As detailed in Section 1.3.5 Huang (1982) suggests that the unacceptabilit y of (certain) wh-adjuncts occurring in situ within island configurations is to be explained in terms of the ECP directly constraining LF raising of in situ wh-phrases. Section 1.4.5 did in fact already reconsider the validit y of such arguments and concluded that restrictions on the occurrence of adjunct wh-phrases in island environments can not be attributed to the ECP (and covert movement). Following Tsai (1994), it was first noted that the set of elements subject to island-t ype localit y conditions is actually very limited, reducing to just (the equivalents of) why and manner how. It was then pointed out that while all wh-adjuncts in Chinese (and subject wh-phrases in Quechua) give rise to ECP violations when overtly raised, non-why/how wh-adjuncts are nevertheless fully acceptable in situ in extraction-islands. The conclusions drawn from this was that either all in situ wh-phrases do undergo raising at LF and the ECP does not constrain such LF movement, or that there simply is no LF wh-raising at all, so that the ECP and subjacency can in fact be taken to apply uniformly to all applications of movement in a derivation. In either case it becomes necessary to posit additional non-movement locality constraints to rule out the occurrence of why/how adjuncts in islands, as if there were to be LF movement then the ECP invoked as a constraint to rule out why/how adjuncts in islands would also incorrectly predict non-why/how adjuncts to be unacceptable in such environments (and therefore one would need a non-ECP/non-movement account of why/how in an LF movement approach), and if there is not in fact any LF movement, then restrictions on why/how can obviously not be attributed to (LF) raising or extraction. Restrictions on the distribution of why/how adjuncts which do not parallel those of other wh-phrases and which again would not seem reducible to constraints on movement are also found elsewhere and not only relative to island phenomena (hence indicating that some special treatment of why/how adjuncts is indeed
40
W h-movement and the theory of feature-checking
necessary). In Japanese multiple wh-questions naze ‘why’ may not occur as the linearly first wh-element in any string, this being illustrated in (105) and (107): (104)
(105)
(106)
(107)
John-ga naze sono-hon-o kaimashita ka John-NOM why that-book-ACC bought Q ‘Why did John buy what?’ *naze dare-ga hon-o kaimashita ka why who-NOM book-ACC bought Q ‘Why did who buy a book?’ dare-ga naze nani-o kaimashita ka who-NOM why what-ACC bought Q ‘Why did who buy what?’ *John-ga naze nani-o kaimashita ka John-NOM why what-ACC bought Q ‘Why did John buy what?’
Neither subjacency nor the ECP is being violated in (105) and (107), as no islands are present to block (hypothetical) LF movement to Comp. (106) further shows that naze may occur licitly in multiple wh questions, just so long as it is not the very first wh-phrase in linear terms. These restrictions on the patterning of naze would seem to be difficult to capture with reference to movement and so additional nonmovement constraints regulating the exceptional distribution of elements such as why-adjuncts are indeed called for in other non-island instances too.21 Further to this, it has been noted that naze may not occur in embedded CPs, when relating to the +Q Comp of a higher clause if this latter clause contains a second wh-phrase: (108)
(109)
Mary-wa [John-ga naze konakatta to] omoimashita ka Mary-TOP [John-NOM why not-came that thought Q ‘What is the reason x, such that Mary thought John did not come because of x?’ *dare-ga [John-ga naze konakatta to] omoimashita ka who-NOM [John-NOM why not-came that thought Q Intended: Who is the person x, and what is the reason y, such that x thought that John did not come because of y?
Again it would not appear that any constraints on movement are being violated
21
See A. Watanabe (1991), S. Watanabe (1994), Hornstein (1995) and Grewendorf & Sabel (1999) for various attempted syntactic treatments though.
W h in situ and the LF Movement Hypothesis
41
here, nor can it be suggested that naze may not co-occur with other wh-phrases, as (106) above is perfectly well-formed. It must therefore be concluded that a variet y of factors not relating to movement may affect the distribution of elements such as why and how adjuncts distinguishing them from other wh-adjuncts and that, taken with the argumentation put forward in Section 1.4.5, the restriction of such elements to non-island positions does not in fact constitute convincing evidence for covert LF wh-raising.
1.5.2. Strong crossover and wh in situ In Section 1.3.2 it was mentioned that crossover phenomena have been used to support and add further motivation to the LF wh-movement hypothesis. In both (110) and (111) co-reference between the wh-phrase who and the subject NP is equally impossible: (110) (111)
*Whoi did Janei see ti yesterday? *When did Janei see whoi?
It is commonly argued that the unacceptabilit y of the intended co-reference relation in cases such as (110) is ultimately due to a Binding Theory violation. Raising of the wh-phrase will leave behind an empt y category trace which is functionally determined as a [ p, a] variable subject to principle C. The trace may therefore not be bound by any element other than the raised wh-phrase itself. As co-reference between the subject NP and the wh-phrase in situ in (111) is also not possible, it is proposed that wh-in-situ cases should be accounted for in a fully parallel way, with LF raising of the wh-phrase giving rise to a [ p, a] variable which will then in turn be subject to principle C. I would like to suggest that examples such as (111) may however be ruled out without the need to posit (covert) movement. First of all, the relation serving as critical input to principle C is actually that between the co-indexed NP and the base-generated position of the wh-phrase, not any position to which the wh-element actually raises; movement of the wh-phrase is essential in standard accounts of strong crossover only in so far as it is argued to provide an empt y category trace of a t ype that will be constrained by principle C. Although the “crossing-over” of the co-indexed c-commanding NP by the wh-phrase in examples like (110) is often highlighted in the description of such examples, this perhaps drawing attention to the movement aspect present, such crossing-over is in fact not necessary for co-reference violations to occur. In (112) below, the wh-phrase does not raise over the co-indexed NP subject but co-reference is equally as
42
W h-movement and the theory of feature-checking
impossible as where there is real crossover: (112)
*Janei wanted to know whoi Mary had seen ti
If neither crossing-over of the co-referential NP nor occurrence of the wh-phrase in a SpecCP position is then of any particular relevance, and it is the base position of the wh-phrase which is centrally important, it can be suggested that the impossibilit y of co-reference in examples like (111) may be accounted for by simply allowing principle C of the Binding Theory to constrain full wh-phrases directly in in situ positions rather than just the traces of wh-elements (avoiding the need to posit covert LF raising to explain the ban on co-reference noted). Just as other overt NP-t ypes are variously subject to principles A, B and C, it does not seem odd to suggest that wh-phrases as overt NPs are also constrained by the Binding Theory. If it is objected that this would then involve classing wh-phrases together with R(eferring) Expressions (as both being subject to principle C) and that wh-phrases unlike the latter do not directly pick out entities in the discourse, one can counter that the traces of wh-phrases (argued to be subject to principle C) cannot be claimed to be directly referential in any more obvious way either. The wh-trace account of strong crossover as initially proposed can also be shown to be insufficient and unable to directly explain a number of cases such as (113)–(115). If wh-movement results in an empt y category trace co-indexed with the constituent that has undergone raising, and it is this empt y category and its index which is subject to the Binding Theory, then principle C will not immediately explain the unavailabilit y of co-reference in the examples below, as the [ p, a] trace is not in fact co-indexed with the c-commanding subject NP: (113) (114) (115)
*[To whomi]j did hei give the book tj? *[Whose gossip about which womani]k did Janei fervently deny tk? *[Whosei book]k did Johni borrow tk?
Either some kind of reconstruction or a full trace copy approach (as per Chomsky 1993) will be necessary to correctly rule out such cases. However then it is no longer an empt y category NP resulting from (wh-)movement and determined as a variable that is taken as direct input to the Binding Theory but rather sub-parts of it which are not themselves exactly wh-traces. It seems that instead of insisting that a particular t ype of empt y category be created for the Binding Theory to apply to (via movement of the element) one should simply allow for principle C to constrain the semantic content of a wh-phrase in its base-generated position however it is formally represented — as a trace (or represented within a trace-copy) or also if phonetically present in situ. If wh-phrases are interpreted as novel and unidentified elements it is quite natural that they may not be co-indexed and
W h in situ and the LF Movement Hypothesis
43
bound by any other NP whose reference must be assumed as fixed, this again independent of the method of representation of the wh-phrase.
1.5.3. D-linking and wh in situ A third argument considered earlier in favour of (certain) LF wh-movement related to Pesetsky’s (1987) proposals on d-linking and wh-islands (Section 1.3.4). Claiming that where d-linked and non-d-linked wh-phrases occur in situ in whislands only the former may be interpreted as having (possible) scope at a +Q Comp exterior to the wh-island (as in 116), Pesetsky suggests that this may be explained if the latter but not the former are forced to undergo LF raising to a +Q Comp for their interpretation, and that this movement is constrained by subjacency: (116)
Who remembers where we bought what/which book?
An explanation of some kind is indeed required if the possibilities of interpretation available to wh-phrases in situ in wh-islands do vary according to the propert y of d-linking, but it is not clear that this should necessarily be reflected/stated in terms of movement. Several problems might appear to afflict Pesetsky’s analysis. First of all, if wh-phrases undergo raising to Comp in order to build an operatorvariable quantificational structure (as Pesetsky assumes is the case with non-dlinked wh-phrases), then it is not obvious why d-linked wh-phrases should in fact be able to remain in situ for their interpretation; with d-linked wh-phrases a speaker is also asking for a value from some set, and one might expect that this should then necessitate a logical encoding at LF parallel to that with non-d-linked wh-phrases, with the wh-operator and restriction in clause-external position binding a variable in the nuclear scope, e.g: (117)
Which novel won the Booker Prize this year t? i [xi a novel] xi won the Booker Prize this year
?x
Although Pesetsky suggests that d-linked wh-phrases are non-quantificational, it seems that d-linked wh-phrases do have the same need as non-d-linked phrases to quantify over a set, allowing as possible answer values any of the individuals from that set. Any difference which exists between questions involving d-linked and non-d-linked wh-phrases is rather a matter of degree–the degree of the restriction placed on the set from which possible answer-values may be drawn; with a non-dlinked who the restriction is all those person-valued entities in the world, whereas with d-linked which person some subset of this entire set is assumed. It seems
44
W h-movement and the theory of feature-checking
somewhat unlikely that a question of degree should necessarily cause such a dramatic difference in the way a phrase may receive interpretation.22 A second problematic aspect of Pesetsky’s analysis is that it leads to the expectation that d-linked wh-phrases should always remain in situ for interpretation, yet in questions containing just a single wh-phrase, this must in fact undergo raising whether d-linked or not: (118) ??You saw which film?23 A third problem for the suggestion that d-linked and non-d-linked wh-phrases are syntactically distinguished in that the former may remain in situ for their interpretation while the latter obligatorily raise comes from a consideration of wh-phrases in situ in other non-wh-islands. Since Pesetsky argues that the lack of subjacency violations observed with wh-phrases in situ taking scope outside of wh-islands is a direct function of their being d-linked (and hence not raising at LF), one would predict that only those wh-phrases that are d-linked would be able to occur in situ in other island configurations. However, as Fiengo et al. (1988) point out, non-dlinked wh-phrases regularly occur in situ in English in all island t ypes, e.g. adjunct clauses, complex NPs, subject islands etc. In Pesetsky’s view such non-d-linked wh-phrases should be forced to undergo raising at LF resulting in violations of subjacency in each case, yet the examples are consistently acceptable.24 One may therefore either conclude that such wh-elements do undergo LF movement but that this is not constrained by subjacency, so that the wh-island cases must be subject to some other non-movement filtering condition, or that subjacency is indeed a constraint on movement occurring throughout the derivation and that no LF movement of wh-phrases takes place; again some other non-movementrelated constraint will consequently need to be invoked to account for the relevant patterning with regard to wh-islands. What this generally seems to indicate is that d-linking may play a role in the
22
It should also be noted that if which-NPs are taken to be inherently d-linked then the set containing possible answer values may in fact be very large, in the case of (117) this essentially being any novel in the world perhaps written within the time-frame of the preceding year).
23
Pesetsky cites Polish as a language in which d-linked wh-phrases may exceptionally remain in situ even in non-multiple-wh questions, while non-d-linked wh-phrases are always subject to raising. However, the overwhelmingly general cross-linguistic observation is that if a language does have obligatory wh-movement, then it treats wh-phrases of both d-linked and non-d-linked t ype in the same way and does not allow the former an in situ option.
24
Fiengo et al. also give a variet y of convincing evidence against any LF pied piping explanation of these cases.
W h in situ and the LF Movement Hypothesis
45
licensing of wh-island violations, but that this does not in fact relate to whether the wh-phrase actually undergoes movement or not. Such a conclusion is indeed argued for in Comorovski (1989) on the basis of data in Romanian. In this language there is a d-linked/non-d-linked distinction with regard to wh-island violations which is essentially parallel to that observed in English — a d-linked wh-phrase base-generated within a wh-island ultimately may have scope at a higher +Q Comp whereas non-d-linked wh-phrases may not. However, what is significantly different to English and Pesetsky’s general proposals concerning such high scope possibilities out of wh-islands is that d-linked wh-phrases in Romanian do in fact undergo movement to attain this scope. As (119) and (120) illustrate, extraction of a fully d-linked wh-phrase from a wh-island is fine, whereas that of a wh-phrase interpreted as non-d-linked is quite unacceptable: (119)
(120)
[La care senatori stii [ce scriitorik tk au apelat ti]]? to which senator you-know [what writers have appealed ‘Which senator x is such that you know what writers called upon x?’ *[La cinei stii [ce scriitorik tk au apelat ti]? [to whom you-know [what writers have appealed Intended: ‘Who is the person x, such that you know what writers have called upon x?’
The possibilit y of wh-island violations with d-linked wh-phrases is therefore seen not to be dependent upon a different mode of interpretation being available with such elements, i.e. in situ binding vs. movement. Where d-linked and non-dlinked wh-phrases in Romanian raise out of wh-islands and are interpreted with scope at some higher +Q Comp we do find the same critical distinction as in English, that only the former t ype are acceptable relating to such wh-islandexternal Comp positions. However, here this is clearly not a function of movement vs. non-movement. One might then be tempted to suggest that this is actually an argument in favour of LF raising of d-linked wh-phrases in English, that all wh-phrases regardless of being d-linked or not are subject to LF raising, just as occurs overtly (with all wh-phrases) in Romanian, and that it is d-linking which will allow for LF violation of wh-islands in English in the same way that it does with overt movement in Romanian. However, such a view cannot be correct as overt extraction of a d-linked wh-phrase from a wh-island in English is still quite unacceptable: (121)
*/??Which senatork do you know which writersi ti appealed to tk?
Contrasting strongly with Romanian (119) where parallel extraction is not illformed at all, this indicates that d-linking does not in fact licence a movement
46
W h-movement and the theory of feature-checking
dependency crossing wh-islands in English. One therefore must return to the assumption that (in situ) d-linked wh-phrases in English do not in fact undergo raising out of a containing wh-island for scope at a higher +Q Comp. If a unitary analysis of what appears to be the same t ype of phenomenon in both Romanian and English is desired, it seems that the notion of movement must be divorced from d-linking, and one should perhaps attempt to account for wh-island violations with d-linked wh-phrases in terms of some general semantic propert y of d-linking allowing for high scope of wh-phrases in these cases rather than suggesting that such interpretative possibilities are somehow a function of the occurrence or non-occurrence of movement. In Romanian movement out of wh-islands plainly does occur with d-linked wh-phrases, in English it arguably cannot. The end result, that in both languages it is only wh-phrases with a d-linked interpretation that may have scope at a +Q Comp external to the wh-islands is nevertheless the same, and hence should be ascribed to general (perhaps semantic) properties of the resulting configuration rather than linking it to the application (or not) of syntactic raising operations. In sum then, Romanian shows here that it is not appropriate to attempt to account for the d-linked/non-d-linked distinction with regard to apparent wh-island violations by wh-in-situ in English with the suggestion that in situ non-dlinked wh-phrases in English are forced to raise to a +Q Comp at LF while d-linked wh-elements may remain in situ. The same d-linked/non-d-linked distinction is also seen to surface with wh-elements that do in fact undergo movement. Consequently the data noted in Pesetsky ultimately cannot be taken as strong evidence for assuming any LF raising of (non-d-linked) wh-phrases and d-linked-related constraints on interpretation must be accounted for quite independently of movement.
1.5.4. Superiority and wh in situ A fourth argument reviewed earlier in Section 1.3.3 that has been often presented as support and motivation for the LF wh-movement hypothesis relates to cases of so-called superiorit y violations. It was suggested that the unacceptabilit y of sentences such as (123) and (124) reveals a subject/adjunct vs. object asymmetry characteristic of ECP violations, and hence might be explained in terms of such a principle governing extraction and movement if it is also assumed that in situ wh-phrases undergo raising at LF: (122) (123)
Whoi ti hid what? *Whatk did who hide tk?
W h in situ and the LF Movement Hypothesis (124)
47
*Whati did Mary fix ti how?
After LF raising to Comp it is proposed that the wh-phrase subject and adjunct in (123) and (124) will fail to c-command and antecedent-govern the traces at their respective extraction-sites, this giving rise to straightforward violations of the ECP as may occur with overt raising of wh-phrase subjects and adjuncts:25 (125) (126)
*Whyi did John wonder who left ti? *Whoi did John claim that ti had gone?
Although an analysis of examples like (122)–(124) based on LF wh-raising and the ECP does allow for an account of superiorit y phenomena in terms of a principle already justified and necessary as a constraint governing extraction/movement, on further inspection there is reason to doubt that the unacceptabilit y of such multiple-wh questions can be attributed to the ECP constraining hypothetical LF wh-movement. The ECP account has been argued to make false predictions in a number of cases, e.g: (127) (128) (129) (130) (131)
Whati did you give ti Mary? *Whati did you give who ti? Whoi did you send the report to ti? ??Who did you send what to t ? i i What did who send where?
In (128) the inner object of the verb who should be lexically (and hence properly) governed by give so LF extraction should not give rise to any ECP violation yet the example is ill-formed. Similarly in (130) what is lexically governed by the verb so again raising at LF is not expected to violate the ECP. Finally in (131) covert extraction of the subject who should result in an ECP violation parallel to (123) yet the example is quite acceptable ((131) being similar to Kayne’s 1984 examples considered in Section 1.4.10 above). If the ECP will not account for such cases there clearly must be other factors constraining the positioning of wh-phrases relative to each other in multiple-wh questions. Comorovski (1989) suggests that what may be relevant here is not any constraint on movement or extraction but rather the surface linear order of wh-phrases interacting with the pragmatics and presuppositions of multiple-wh questions. In Romanian multiple-wh questions all wh-elements undergo overt raising to a clause-initial Comp position; on certain occasions this is shown to 25
Such an approach also requires the further assumption that only the first (overtly) raised wh-phrase may c-command into IP from Comp.
48
W h-movement and the theory of feature-checking
give rise to superiorit y-t ype violations parallel to English, yet in other instances multiple raising of wh-subjects, adjuncts and objects to the same Comp may occur quite licitly and give rise to any permutation in surface linear order, e.g. /subjectobject-adjunct/, /adjunct-object-subject/, /subject-adjunct-object/ etc. What Comorovski shows to be of critical importance here is that the wh-element occurring first in the string must be (interpreted as being) d-linked for any multiple-wh question to be acceptable. As long as this condition is fulfilled all possible variation in terms of linear ordering of the wh-phrases will in fact be acceptable. Comorovski suggests that the relevance of d-linking to multiple-wh questions and superiorit y lies in what may be taken to be an appropriate answer-form to such questions. She points out that multiple-wh questions normally give rise to and require pair-list-t ype answer forms, as e.g: (132)
Which table ordered what? A: Table 1 ordered a steak, Table 2 a souffle and Table 3 the fish.
She then claims that a felicitous answer-form to a multiple wh-question must exhaustively give values for the entire range/membership of the wh-phrase occurring first in linear order. In (132) an acceptable answer should therefore provide a set of pairs of values such that a value for every table which ordered something is present (in one of the pairs). With further illustration Comorovski goes on to argue that there is no expectation that values also be given for every member of the range of other wh-phrases occurring after the first (so that in (132) the range of what could be the entire menu of the restaurant, but not all items on the menu will necessarily appear in an answer to (132)). If this is true, then it can be shown to follow that the linearly first wh-phrase in a multiple-wh question must be d-linked but that other following wh-phrases need not necessarily be so. D-linking of a wh-phrase essentially entails that the full membership of the range of a wh-phrase is assumed to be known to both speaker and hearer. If a multiplewh question requires that values for the entire range of the first (but not any following) wh-phrase appear in its answer-form then this will obviously only be possible if this membership is known, hence d-linked. Comorovski thus attempts to provide an explanation of superiorit y effects which does not reduce to the ECP and extraction/movement. As noted, in Romanian all wh-phrases do undergo movement, and (subject to the d-linking requirement being satisfied) may give rise to any ordering of subject, adjunct and object wh-phrase. It is hence not possible to attribute any superiorit y-t ype violations that are observed to the suggestion that only a single wh-phrase in Comp will be able to c-command and antecedent-govern its trace. Were this to be so
W h in situ and the LF Movement Hypothesis
49
then wh-adjuncts and subjects would (most probably) not be able to co-occur raised to Comp and any wh-phrase in need of antecedent-governing its trace would always have to occur in some fixed position within a wh-string. If there are then plausible accounts of superiorit y relating to linear ordering and constraints on interpretation which do not connect such violations to the ECP, there are also alternative purely syntactic ways of explaining cases of superiorit y, which likewise do not imply or require LF wh-raising in languages like English. One such possibilit y (returned to in Chapter 2 where additional supporting data is presented) is simply to invoke the economy principle of Shortest Move to account for the difference in acceptabilit y between examples such as (122) and (123) repeated below: (122) (123)
Whoi ti hid what? *Whatk did who hide tk?
Movement of the subject wh-phrase to Comp (122) will be shorter than raising of the object (123) and so should be preferred (and therefore necessary) on grounds of economy. This will then rule out examples like (123) quite without reference to any LF movement of wh in situ. In fact if in situ wh-phrases did have to raise at LF it is not obvious that notions of economy would still make any distinction between (122) and (123) — in both cases both the shorter movement of the subject to Comp and that of the object would have to occur at some point, and so the net expenditure in terms of economy should be the same in both derivations. Consequently cases of superiorit y can (ironically) actually be turned into arguments against the LF wh-movement hypothesis.
1.6. Selection, absorption and scope In the last section of this chapter we consider three final theoretical arguments for covert LF wh-raising put forward in Huang (1982) and elsewhere, relating to the notions of selection, absorption and scope.
1.6.1. Selection It has been suggested that overt wh-raising in languages such as English may be triggered by a need to satisfy selectional requirements of a predicate subcategorizing for a question. Movement of a wh-phrase to SpecCP may result in identification of a clause as a question, perhaps via percolation of a +wh-feature from the
50
W h-movement and the theory of feature-checking
Specifier position directly to CP or else via Spec–Head agreement between SpecCP and C0 and subsequent percolation of the +wh-feature from C0 to its phrasal projection; such identification of the clause as +wh-interrogative may then be taken to satisfy +interrogative requirements of the embedding predicate. As wh-raising is forced to take place overtly in English-t ype languages, it is presumed that selectional requirements require satisfaction prior to S-structure/Spell-Out in such languages (though why this should be so is unclear, if a verb such as wonder requires a CP that is +interrogative to result in a well-formed interpretation, this should be a requirement that needs satisfaction only at LF rather than S-structure/Spell-Out): (133) (134)
John wonders whati Jane bought ti. *John wonders Jane bought what.
If it is then further naturally assumed that predicates such as wonder and ask have parallel selectional requirements across all languages and that such selectional requirements are uniquely met via wh-feature percolation from a Comp position to CP, in languages such as Chinese where no overt wh-movement is attested it can be argued that covert LF raising must take place to bring a wh-phrase to Comp: (135)
(136)
Ta xiang-zhidao ni xihuan shei. he want-know you like who ‘He wants to know who you like.’ LF: Ta xiang-zhidao [CPsheii ni xihuan ti]
There are however two good reasons to doubt the strength of such a parallelism t ype argument. First of all, it is not clear that one can correctly attribute overt wh-raising to the satisfaction of selectional requirements. In English (and other similar languages), pre-Spell-Out raising of a single wh-phrase to Comp is forced to take place not only in embedded +interrogative CPs but also in matrix clauses where this raising will not satisfy the selectional requirements of any predicate: (137)
Whoi did John see ti?
If one does attempt to analyse wh-raising as a mechanism to identify a clause as +interrogative, as indeed argued in Cheng’s (1997) Clausal Typing Hypothesis, then this would clearly seem to be driven by clause-internal identification requirements rather than any imposed from outside — in (137) nothing other than properties of the CP itself would necessitate wh-raising for identification of the CP as +interrogative. If some explanation of wh-raising (perhaps clausal-identification) is therefore anyway required for cases such as (137), quite independently of
W h in situ and the LF Movement Hypothesis
51
whether a +interrogative CP occurs embedded or not, it seems over-redundant to suggest that the same operation of wh-raising is also triggered by a second principle (selection) in just a subset of cases (i.e. where the +interrogative CP is embedded).26 If one supposes that +interrogative CPs are indeed subject to a requirement that they be identified as such (following Cheng), a second objection one might raise against the argument for LF wh-raising from selection is that there are ways other than wh-movement that plausibly might result in this kind of clausal identification. Cheng in fact suggests that in languages like Chinese +interrogative clausal-t yping is effected by means of wh-question particles base-generated in Comp. If such particles may then percolate (+wh-)interrogative features from their position in C0 to CP, there would obviously be no need for clausal identification to be duplicated via (covert) wh-raising to SpecCP, and for reasons of economy none should therefore take place.
1.6.2. Absorption Another argument given in Huang (1982) in favour of an LF-movement analysis for in situ wh-phrases relates to the interpretation of multiple-wh questions like (138): (138)
Who bought what?
As discussed earlier, such a question may normally receive an answer form which consists in a set of pairs, matching values from the range of the raised wh-phrase with values from that of the wh-phrase in situ. It is argued by Huang (and others) that a natural logical representation accounting for the phenomena of paired answer forms in questions like (138) would be one in which the in situ wh-phrase occurs raised into the Comp position occupied by the wh-phrase moved at S-structure. At LF all wh-phrases in the same +Q Comp would then absorb together to form a single (but internally complex) quantificational element: (139) (140)
[[whati [whok]][IP tk bought ti]] → absorbed form: [[whati whok][IP tk bought ti]
Although (140) is indeed a logically plausible representation of how such multiple-wh questions may receive interpretation, it is not however a necessary representation which forces one to assume that the only way in which paired
26
Similar arguments that selection cannot be the trigger for wh-movement also appear in Cheng (1997).
52
W h-movement and the theory of feature-checking
readings may arise is due to an operation of absorption applying to quantificational elements appearing at a single structural node. There are other instances of paired answer forms, resulting from the interaction of wh-phrases and universallyquantified NPs, in which it is not standardly assumed that the wh-phrase and the universally-quantified NP undergo any parallel absorption process: (141)
What did everyone buy t? A: Mary bought a pen, John bought a book, etc
The formal operation of absorption outlined in Higginbotham & May (1981) and adopted by Huang and others assumes that quantificational elements must occur in essentially the same (operator-like) position in order for their mutual quantificational force or commonly shared operators to be collapsed into one complex quantificational unit. Supposing the QP everyone in (140) undergoes QR at LF, this will bring it to an IP-adjoined position (following May 1977/85): (142)
[CP Whatk [C did [IP everyonei [IP ti buy tk]
At LF the wh-phrase and the universal QP will therefore not in fact occur raised together at a single node in the structure, yet such a condition is taken as necessary for absorption to apply, providing the critical argument for why an in situ wh-phrase must actually raise to Comp at LF (as in (138)–(140)). In (142) the wh-phrase and QP neither occur at one common node nor are they adjacent (and perhaps may be claimed to be separated from each other by a tense operator did). In spite of this, possibly it might still be argued that configurations such as that in (142) somehow do satisfy the structural requirements relevant for absorption. May (1985) suggests that ambiguit y in relative scope relations will be present where quantificational elements are dominated by the same maximal projections (the Scope Principle — see fn.27 below), and that domination can be defined in such a way that the wh-phrase and QP in (142) will both be dominated only by CP (with the result that either element may take wide scope over the other).27 One could attempt to suggest that it is this same t ype of structural relation which is in fact relevant for absorption (and not adjacency/occurrence at a single node), so
27
May’s Scope Principle essentially has the form in (i) and depends on the definitions of c-command and strong domination in (ii) and (iii) (from Hornstein 1995, p. 20): (i) Scope Rule: If A asymmetrically c-commands B, then A has scope over B (ii) C-command: A c-commands B iff every node that strongly dominates A strongly dominates B (iii) Strong Domination: A node N strongly dominates B iff every segment of N dominates B
W h in situ and the LF Movement Hypothesis
53
that what and everyone in (142) would be able to absorb together.28 However, this will still not account for the possibilit y of paired answer-forms in examples like (143) below: (143)
What do you imagine that everyone bought t? A: I would guess that Mary bought a pen, John bought a book, etc
As QR is assumed to be a clause-bound operation, its application in (143) to the universal QP will result in this element being adjoined to the IP of the lower clause. Given that the wh-phrase is raised to SpecCP of the matrix clause, the QP and the wh-phrase will not be dominated by the same maximal projections at LF. As paired answer-forms are nevertheless possible here they can plainly not be attributed to an operation of absorption applying only to elements that co-occur in certain close structural relations. So, if such wh-QP examples show that the general possibilit y of a wh-question answer-form consisting in pairs of values taken from the range of two discrete quantificational elements does not necessarily depend on these elements having to undergo absorption at a single position in syntactic structure (or alternatively when they occur dominated by the same maximal projections), then it is not unreasonable to suggest that the paired answer-forms possible with multiple-wh questions may also arise without the need for all wh-elements to undergo absorption at a single syntactic node. Consequently, one is not forced to assume LF wh-raising in examples like (137) to explain the occurrence of pair-list answers — such t ypes of answer may be argued to result from the same t ype of non-local process of interpretation that must indeed be available to give rise to paired answer-forms in wh-QP questions.
28
Also it is not clear that quantificational elements containing operators of different t ypes, e.g. a wh-operator and a universal quantifier, could technically give rise to absorption. In the case of wh-absorption it is suggested that wh-operators assumed present in all relevant wh-phrases will be “factored out” (see Higginbotham & May 1981) with the result that a single wh-operator will then apply commonly to the restrictions of all wh-elements in Comp: (i) [[which mani] which girlk tk saw ti? → [[which x,y [x a man, y a girl]] y saw x] However in wh-questions potentially resulting in paired answer-forms where a universallyquantified NP is present, it is not possible to factor-out any operator common to both the wh-phrase and QP. Consequently it may be reasonably argued that paired answer-forms may generally arise without the need for “absorption”.
54
W h-movement and the theory of feature-checking
1.6.3. W h-raising and the representation of scope Finally I would like to consider the extent to which arguments relating to phenomena of scope may be taken as support for the LF wh-movement hypothesis. It has often been suggested that the overt raising of wh-phrases in languages like English is closely associated with, and in some sense even triggered by a requirement that the scopal properties of such elements be represented explicitly in the structure in which they occur. If this is indeed a general requirement on wh-phrases (and perhaps other quantificational elements too) and one which appears to be satisfied via movement to a +Q Comp in certain languages, then it may be argued that all wh-phrases which occur in situ at PF will need to undergo raising to +Q Comp at some point prior to LF and interpretation. Essentially there are two interconnected notions of wh-scope relevant to raising here. One t ype is the scope which a wh-phrase has relative to one particular +Q Comp rather than another, hence in (144) the in situ wh-phrase may for certain speakers be interpreted as being either directly or indirectly questioned depending on which +Q Comp it is associated with: (144)
Who remembers where we bought which book?
In the former interpretation which book is said to take scope at the higher matrix +Q Comp, in the latter at that of the embedded CP. The overtly-raised wh-phrases who and where by way of contrast may only take scope at the +Q Comps they actually occur in at PF and are interpreted as being directly and indirectly questioned respectively. In this sense the scope of a wh-phrase relative to a +Q Comp is fixed via any movement which the wh-phrase undergoes. The issue of how the scope of in situ wh-phrases may be established relative to a +Q Comp, via movement or some other operation, was partially considered above in Section 1.5.3 on d-linking (and will be taken up again in Chapter 2). A second notion of wh-scope which we turn to now is that of a wh-phrase relative to other quantificational elements in its surrounding structure. May (1977/85) suggests that where a wh-phrase occurs in Comp as a result of overt movement its scope relative to other quantificational elements will be computed with direct reference to the raised position it occupies. Huang (1982) then argues that if in situ wh-phrases exhibit the same relative scopal relations as (overtly-)raised wh-elements, this may be taken to result from LF raising to Comp and the computation of the scope of a wh-phrase from its (LF) raised position in a fully parallel way. The questions which will be approached here are (145) and (146):
W h in situ and the LF Movement Hypothesis (145) (146)
55
Are the relative scopal properties of overtly-moved wh-phrases really a result of the raised position which they occupy? To what extent is it necessary to assume LF movement to Comp in order to explain relative scope phenomena?
If it can be shown that wh relative scope is not necessarily a function of any raised position, then the motivation for assuming covert LF raising to Comp to account for similar scopal relations in the case of wh in situ will clearly no longer valid. It is commonly admitted that wh-questions such as (147) containing a universally-quantified NP subject and a wh-phrase object exhibit an ambiguit y not present where the wh-phrase is instead the subject and a universal QP the object: (147) (148)
Whati did everyone buy ti? Whoi ti bought everything?
In (147) either the QP or the wh-phrase may take wide scope relative to the other, so that an answer-form may consist in a single value for the wh-phrase (wide scope of the wh-phrase), or a set of pairs (wide scope of the QP). This latter wide scope interpretation of the universal QP is however not available in (148) and an answer form may only consist in a single value corresponding to the wh-phrase. Such examples have received considerable attention in the literature (see for example May 1985; Williams 1986; Garrett 1996), the question being how the scopal ambiguit y arising in sentences like (147) but not (148) may be linguistically encoded. As mentioned briefly in Section 1.6.2 above, May (1977/1986) suggests that quantificational elements such as everyone undergo an LF operation of quantifier raising, which in the case of (147) will raise the QP to an IP-adjoined position, and that the Scope Principle will then allow for either what or everyone to take wide scope relative to the other in virtue of arguably being dominated by the same maximal projections. Regarding (148), May claims that QR adjoining the object QP to IP will not be possible as in this case it would cause an ECP violation of the subject wh-phrase trace (see May 1985 for details); consequently the QP will adjoin to VP instead. In such a position it will not be dominated by the same maximal projections as the wh-phrase in SpecCP and as a result will only have narrow scope relative to who. May thus sees scopal relations between quantificational elements as resulting from the structural positions they occupy within a syntactic tree (at the level of LF), and crucially that it is with regard to the raised position of a wh-phrase in Comp that its scope is computed. Turning now to Chinese and examples involving quantificational phrases and (in situ) wh-phrases, Huang argues that the latter consistently take wide scope over the former, even in cases where a wh-phrase is c-commanded by a QP at
56
W h-movement and the theory of feature-checking
S-structure: (149)
mei-ge-ren dou mai-le shenme? everyone all buy-ASP what ‘What did everyone buy?’
In this respect wh-phrases are argued to differ quite significantly from other quantificational elements in Chinese, as the structural relation of c-command obtaining between other t ypes of quantificational items at S-structure does in fact encode the scopal relation holding between them; that is, where one QP c-commands a second QP at S-structure the former will necessarily also have wide scope over the latter. In order to explain the apparent wide scope propert y of whphrases even when c-commanded by other QPs at S-structure, Huang argues they must be analysed as undergoing LF-raising to a Comp position from which they will c-command (but critically no longer be c-commanded by) any other quantifier phrases occurring in a clause. He suggests that because overt wh-movement may be taken to give rise to and allow for the potential wide scope interpretation of wh-phrases in languages like English, it is plausible to argue that the wide scope interpretation of wh-phrases in Chinese is also a result of the same t ype of raising. Hence, like May, Huang suggests that the scopal properties of a wh-phrase relate to the actual position a wh-phrase occurs in at LF, and that this will be a +Q Comp in Chinese just as in English (and by hypothesis all languages). Huang does concede that there might be ways to encode the scope of in situ wh-phrases other than via covert movement, perhaps through co-indexation of such elements with a +Q Comp, but argues that a movement analysis is to be preferred for two basic reasons. Firstly, if movement may be taken to be the means by which wh-scope is established in languages such as English, then a certain cross-linguistic uniformit y can be captured if it is suggested that it is also movement (albeit covert) that gives rise to wh-scope in languages where wh-phrases occur in situ (and no additional mechanisms for the representation of such scope need therefore be posited for wh in situ). Secondly, it is argued that the encoding of wh-scope via hypothetical co-indexation with Comp might in fact be nothing more than a purely notational equivalent to the claim that there is covert raising, and that if movement is already attested to affect wh-phrases in certain languages, then a movement analysis of wh in situ scope representation should indeed be preferred to any other formal equivalent such as co-indexation of an in situ wh-phrase with Comp. Again, if the relative scopal properties of wh-phrases are thus attributed to some raised position in a +Q Comp, it must nevertheless be added that there is no claim that a required wide scope interpretation of wh-phrases (in Chinese or other languages) actually triggers wh-movement, whether overt or at LF, just that this is
W h in situ and the LF Movement Hypothesis
57
a reading which may result from wh-raising. Were the establishment of a certain scope relative to other quantificational elements to be the underlying motivation and trigger for wh-movement, one would predict that there would be no raising of a wh-phrase where a reading of narrow scope relative to some other QP was intended, yet this is not the case. Example (147) above is ambiguous between a wide and a narrow scope reading of the wh-phrase relative to the QP everyone; were wh-raising to be a mechanism solely employed for the indication of wide scope (of a wh-phrase), one would expect that a speaker intending an unambiguously narrow scope reading in such an example would choose to leave the wh-phrase in situ within the c-command domain of the QP subject, but: Everyone bought what? is not a standardly acceptable alternative to (147). Similarly, if wh-movement were to be triggered by a need to establish relative scope in general, then in a sentence containing no other quantificational elements relative to which the wh-phrase might be required to establish a scopal relation one would predict that wh-raising should not (be required to) take place, but such is again not the case, vis: *John saw what? Aoun & Li (1993) propose an account of wh-relative scope relations that is in part significantly different from Huang (1982) and May (1977/85), and which we will argue may be adapted to eliminate the need for covert raising of wh in situ from the assessment of wh-scope. Essentially it is suggested that the scope of a wh-phrase relative to other quantificational elements is computed with reference not only to the raised positions of wh-elements in Comp but also with regard to the position of traces of wh-movement. Aoun & Li arrive at such a proposal in order to provide an account of various wh-scope cases which remain unexplained in the system of relative scope resolution put forward in May (1985). It will be recalled that May (1985) ascribes the possible wide scope of either wh-phrase or universally-quantified subject in examples such as (147) to the Scope Principle applying at LF after operations of quantifier raising have taken place: (150)
Whati did everyone buy ti? LF: [CP whatk did [IP everyonei [IP ti buy tk]]]
As both wh-phrase and QP can technically be argued to be dominated by the same set of maximal projections at LF, it is suggested that either element may take wide scope over the other, this resulting in the two possible interpretations of the sentence. While the Scope Principle may be argued to provide an account of the scopal ambiguity present in examples like (150), it does not however appear able to explain the interpretations available in other more complex sentences such as (151):
58 (151)
W h-movement and the theory of feature-checking Whati did the teacher suggest that everyone read ti?
Given the common assumption that QR is a clause-bound operation, everyone will here only raise to adjoin to the lower clause IP; at LF it will therefore not be dominated by the same set of maximal projections as the wh-phrase in the matrix, and according to the Scope Principle should consequently not be able to take wide scope over the latter. Nevertheless the sentence is ambiguous in the same way that (150) is, allowing for a wide scope reading of either the wh-phrase or the QP. Aoun & Li suggest that this wide scope interpretation of the QP is in fact a function of the c-command relation existing between the QP and the (A′-)trace of the wh-phrase. If c-command of a trace of wh-movement may therefore be taken as a sufficient condition for a QP to take scope over a wh-phrase, then the wide scope reading of the QP in examples like (150) may also be claimed to result from the QP c-commanding a wh-trace. The Scope Principle is therefore argued both to be unable to account for cases like (151) and to be unnecessary for those like (150) once a wh-trace-based approach is adopted. Of particular significance here is the demonstration that it does not appear possible to compute all wh-scope relations solely with reference to the raised position of a wh-phrase and that for a significant number of cases it is actually the trace position of a wh-phrase which is critically important. A further consequence of assuming the above view of QP wide scope is that QR may perhaps no longer seem necessary to establish such wide scope — in both (150) and (151) the QPs will c-command the wh-trace from their S-structure/SpellOut positions before (and without the need for) any covert raising. Unambiguous examples such as (152) may also be given simple explanation on the basis of the S-structure c-command relation existing between the QP and the trace of wh-movement, the former not c-commanding the latter and hence not being able to take scope over the wh-phrase: (152)
Whoi ti bought everything?
If one were however to assume QR in (152), then accounting for the unambiguit y of such cases becomes considerably more complicated. First of all it has to be stipulated that the object QP may only adjoin to VP and not to IP (IP-adjunction would result in the QP c-commanding the trace of the wh-phrase in SpecIP and QP wide scope over the wh-phrase should then be possible). Secondly, if QR of an object NP is indeed restricted to adjunction to VP, one must also add that it is only the traces of wh-movement which will count for relative scope assessment, as from a VP-adjoined position the QP object would c-command an NP-trace of the subject wh-phrase in SpecVP after the latter has undergone raising to SpecIP:
W h in situ and the LF Movement Hypothesis (153)
59
LF: [CP Whoi [IP ti [VP everythingk [VP ti bought tk]]]]
If quantificational phrases are however not subject to any operation of QR (and certain evidence against QR has already been presented in Section 1.4.8) then a simpler account of their potential wide scope interpretations may be possible, suggesting that the occurrence of any element or any t ype of trace in the c-command domain of a QP will allow for the QP to take scope over that element. Aoun & Li also argue, contra Huang (1982), that Chinese sentences like (149) are in fact ambiguous and that there is a reading available in which the universal QP takes scope over the wh-phrase: (149)
mei-ge-ren dou mai-le shenme? everyone all buy-ASP what ‘What did everyone buy?’
Here essentially the same account as outlined above can be invoked to explain the possible wide scope interpretation of the QP; the QP subject will c-command the position occupied by the wh-phrase and so naturally may take scope over it. In all the cases of wide QP scope relative to a wh-phrase reviewed, it can therefore be claimed that it is the base-generated or trace position of a wh-phrase which serves as direct input for relative scope assessment and not any position in a +Q Comp; in wh in situ cases like Chinese (149) it would not seem necessary to posit LF wh-raising in order to explain possible narrow scope interpretations of wh-phrases. Now however, one needs to consider how the wide scope readings of wh-phrases relative to QPs may be explained. Aoun & Li basically assume that the wide scope of one element over another is always a function of the first element c-commanding the second. Where an in situ wh-phrase is interpreted as having wide scope over a QP it does not c-command at S-structure, it therefore must be assumed to c-command the QP at some other derivational point prior to interpretation. Aoun & Li consequently also presume that in situ wh-phrases undergo LF raising to Comp, just as in Huang (1982). I would here like to suggest a somewhat different view of the wide scope interpretation of wh-phrases relative to universal QPs which does not entail the former having to stand in any c-command relation to the latter, hence that LF wh-movement is not in fact necessary to account for wh wide scope readings in examples like (149). This first requires a brief reconsideration of precisely what is understood by the term “wide scope” and will result in the suggestion that the “wide scope” of a QP over a wh-phrase is in essence quite different from that of a wh-phrase relative to a QP. Where a universally-quantified NP is interpreted as taking wide scope over a
60
W h-movement and the theory of feature-checking
wh-phrase as in (154), this will give rise to pair-list answer-forms in which a value is drawn from the range of the wh-phrase for every value in that of the QP. There is therefore some direct interaction between the potential referents of the wh-phrase and those of the QP: (154)
What did everyone buy? A: John bought a book, Jane bought a pen, etc
Chierchia (1992) suggests that a structured trace will be left at the wh-extractionsite in (154) and that the universal QP will then bind into the N′/NP of such an NP/DP trace, this N′/NP constituting the range of the wh-phrase. However one chooses to represent the quantificational interaction of QP and wh-phrase in such QP wide scope readings, it has to be admitted that there is indeed some direct dependency between these elements, a pair-list answer-form being a function of both wh-phrase and QP. This kind of interaction or linking of the wh-phrase and universal QP is however completely absent from the wh wide scope reading in examples like (154). Wide scope here does not result in any quantificational dependency of the two elements but rather a fully independent reading of the wh-phrase. As such it is like the specific independent interpretation of indefinite NPs in sentences like (155) (noting this interpretation of indefinites is also commonly referred to as a wide scope reading): (155)
Everyone saw a (specific) film. Possible interpretation: There is one specific film that everyone saw.
I would like to suggest that the wide scope reading of wh-phrases and the specific independent interpretation of indefinite NPs may arise in essentially the same way, not as a result of movement to a position c-commanding other QPs present, but simply in virtue of selecting an interpretation of the wh-phrase/indefinite in which it is not dependent on any other QP, this being quite possible even where the indefinite/wh-phrase remains within the c-command domain of other QPs. Considering the specific “wide scope” interpretation of indefinite NPs first, there is strong evidence that this cannot result from raising to a position higher than other QPs present. Firstly such raising would have to be analysed as an instance of QR which is assumed to be clause-bound, yet indefinites in embedded clauses may be interpreted as specific and taking wide scope relative to QPs occurring in higher clauses: (156)
Everyone thought that John had bought a specific book.
QR in (156) should clearly not be able to raise the indefinite beyond the first
W h in situ and the LF Movement Hypothesis
61
clause to a position c-commanding the universal QP yet it may have the wide scope interpretation associated with such a configuration. Indefinites may also occur within extraction islands and be interpreted as taking “wide scope” over a QP external to the island: (157) Everyone thought that the fact that a specific doctor had diagnosed Jane insane was quite irrelevant. Again it is implausible to suggest that wide scope here is a result of movement of the indefinite out of the containing island to a position c-commanding everyone. A third related case which can be shown to strongly endorse the claim that movement is not responsible for wide independent scope is illustrated in (158), taken from Cormack & Kempson (1991): (158)
Every teacher has sent in two potentially damaging reports that a specific student of mine has been cheating.
In (158) the indefinite a specific student of mine may have wide scope relative to the subject QP ‘every teacher’ at the same time that the latter is interpreted as taking wide scope over the CNP two reports.. which contains the indefinite NP. Here it can not be said that the indefinite extracts for wide scope over the subject QP, as its containing CNP is an island for movement, nor may one propose that the indefinite achieves such wide scope perhaps via some pied piping operation in which the whole CNP would raise above the subject QP (thus attempting to avoid the island/extraction problem) as the subject QP may in fact take scope over the CNP (hence the latter must remain in situ c-commanded by the subject). Cormack & Kempson (1991) further show that specific indefinites do not always have total wide scope with regard to all other quantificational elements in a sentence in the way that proper names do; if they always did have such “widest scope” then it might be possible to argue that this would somehow be a nonquantificational reading of the NP and therefore not a function of QR apparently being non-clause-bound and violating island constraints. Specific indefinites may indeed have wide scope relative to one element in a sentence but also narrow scope relative to another, hence indicating that such specific readings are in fact quantificational in nature. In (158) there is an interpretation available in which the indefinite is construed as specific and taking wide scope relative to its containing island but narrow scope relative to the subject QP: ‘For each of the teachers there is some specific student of mine about which he has sent in two reports.’ The possibilit y of such readings shows that specific NPs do interact scopally with other quantificational elements and cannot simply be claimed to be name-like in their properties.
62
W h-movement and the theory of feature-checking
Thus it may reasonably be concluded that the specific “wide scope” interpretation of indefinite NPs is not a result of any movement taking place to raise such elements out of the c-command domain of other QPs, but is instead available where the indefinite remains in situ (and hence may not c-command those QPs over which it has wide scope). Such wide scope readings can therefore be seen as different in nature from the wide scope interpretation of universal QPs as this latter t ype of wide scope does in fact require that the QP c-command any element over which it is understood as taking scope. Comparing (159) with (156) and (160) with (158) it can be seen that in certain instances where an indefinite may have wide scope over a higher element (the matrix subject in (156) and the CNP in (158)) it is not possible for a universal QP to take wide scope over elements in the same structural positions (as the QPs will never c-command these elements): (159) (160)
A student thought that John had helped everyone. John sent in a report that every student of mine had been cheating.
In (159) the QP everyone may not take scope over a student, and in (160) every student may not have wide scope over its containing CNP. In some sense it is clearly confusing to apply the same cover term of wide scope to the two differing t ypes of interpretation which may result when a universal QP c-commands an indefinite NP as this may be taken to imply that possible wide scope readings of the universal QP and of the indefinite NP should in both cases be analysed as resulting from the same mechanisms and necessary relation of c-command (requiring raising of the indefinite to a position c-commanding the universal when it is understood to take wide scope). While it may be suggested that a universal QP understood as taking wide scope over an indefinite must indeed c-command this NP in order to distribute over it and hence allow for the reference of the indefinite NP to be taken as dependent on that of the universal, there is no sense in which an indefinite NP with wide scope interpretation should need to c-command a universal QP in order to quantify over its reference in any parallel way. The specific reading of an indefinite NP is essentially one where the indefinite may be interpreted as quantificationally independent of other QPs present in its immediate structure; although it could perhaps be argued that an indefinite might need to occur in some position outside the c-command domain of other QPs in order to be interpreted as referentially-independent of these QPs, the data reviewed above clearly suggest that this is not the case, and that an indefinite NP c-commanded by another QP may either be taken as dependent on this QP (its “narrow scope” reading) or specific and independent (wide scope)
W h in situ and the LF Movement Hypothesis
63
without any raising to a position higher than the QP.29 I would now like to suggest that the wide scope reading of wh-phrases relative to universal QPs arises in the same way as that of specific indefinite NPs. Where a universal QP takes wide scope over a wh-phrase this results in pair-list answerforms, as discussed above, and the reference of the wh-phrase is in part dependent on that of the universal, just as the reference of an indefinite NP may be dependent on that of a universal QP where the latter is interpreted as taking wide scope. However, the wide scope reading of a wh-phrase over a universal QP is one in which its reference is not determined by or a function of values in the range of the QP and hence is an “independent” interpretation like that of a specific wide scope indefinite NP. As with the specific interpretation of indefinites it can be suggested that the wide scope independent reading of wh-phrases may arise without the need for a wh-phrase to occur raised in a position where it would c-command those QPs it is understood to take scope over. Thus in the case of wh-in-situ such as (149) repeated below, there will be no requirement that the wh-phrase undergo covert LF raising to Comp in order to receive interpretation as quantificationally independent of the universal QP: (149)
mei-ge-ren dou mai-le shenme? everyone all buy-ASP what ‘What did everyone buy?’
Where overt wh-raising takes place as in English (154), a wh wide scope reading need not be attributed to the position of the wh-phrase in Comp c-commanding the clause, but can instead be argued to relate to its trace position, suggesting that a trace-copy left in the wh-extraction-site may be interpreted as functionally-independent of any c-commanding QP just as wh-phrases in situ or indefinite NPs may. If the narrow and wide scope readings of wh-phrases relative to universal QPs may be argued to parallel those potentially occurring with indefinite NPs and universal QPs, being interpretations of the wh-phrase/indefinite that are either
29
Working in the framework of Labelled Deductive Systems (LDS) Kempson (1996) suggests that indefinites are variables which in fact must always be dependent on some other element (hence they are not technically “independent” when interpreted as specific). In the narrow scope reading of the indefinite NP in (i) its interpretation will be directly dependent upon the universal QP: (i) Everyone read some book. In its wide scope reading Kempson suggests that it will be bound by or dependent for its reference upon the equivalent of a spatio-temporal operator relating the token of a book to the time and place of the action. Within a GB/Minimalist framework one might perhaps translate this into binding by a Davidsonian event operator associated with an (event) variable in Infl.
64
W h-movement and the theory of feature-checking
functionally-dependent on the QP or conversely determined independently of the QP, and if it can be shown that the wide scope independent readings of indefinites must indeed be possible with these elements occurring in situ within the c-command domain of a universal QP (and not result from any covert LF raising operations), then one has reasonable grounds to suggest that wide scope independent readings of wh-phrases may also arise without the need for any extraction from the c-command domain of a universal QP. In addition to this parallelismt ype argument there is however other good evidence for assuming and justifying a non-movement approach to the wide scope readings of wh-phrases in situ. This concerns the potential scopal interaction of multiple quantificational elements in examples such as (161) taken from Fiengo et al. (1988), and is evidence against a raising account of wh wide scope of the same t ype that was earlier presented in connection with indefinite NPs and their possible wide scope readings (see example (158)): (161)
mei-ge-ren dou mai-le [yi-ben [shei xie] de shu] everyone all buy-ASP [one-CL [who write REL book ‘Everyone bought a book that who wrote?’
Fiengo et al. claim that (161) allows for the following possible interpretation: (162)
Who is the x, such that everyone bought one book or another that x wrote?
In this reading the universally-quantified NP subject has wide scope relative to the indefinite/numerically-quantified CNP, but the wh-phrase contained within this CNP has wide scope relative to the subject, thus giving the relative scopal ordering in (163): (163)
who>everyone>a book that…
The critical question here is how the wh-phrase may be understood as taking scope over the subject NP. As it occurs in an extraction-island it is implausible to claim that the wh-phrase undergoes direct movement to the +Q Comp of the sentence (from where it would c-command the subject) as this should violate subjacency. It is also not possible to suggest that raising to Comp is effected indirectly via some pied piping operation (as per Nishigauchi 1990) in which the entire CNP island would raise to SpecCP (with the result that the wh-phrase would remain inside the island throughout the derivation and subjacency would not violated) because the subject is interpreted as having wide scope over the
W h in situ and the LF Movement Hypothesis
65
object CNP (hence the object must be c-commanded by the subject at LF).30 Therefore if movement cannot be claimed to give rise to the wide scope reading of shei ‘who’ in (160), it seems one must concede that such wide scope independent interpretations are possible even where a wh-phrase occurs (and remains) in situ within the c-command domain of a universal QP, just as has been argued is the case with indefinite NPs and specific wide scope readings. Finally certain evidence from Bahasa Indonesia already discussed in Section 1.4.9 may be taken as further support for the view that wide scope readings of wh-phrases are not necessarily a function of raising and occurrence of a wh-phrase in Comp. In Section 1.4.9 it was noted that examples such as (164) with a wh-phrase raised to a position that c-commands a universal QP are ambiguous, with the wh-phrase potentially taking either wide or narrow scope relative to the QP, but that sentences like (165) with a wh-phrase in situ and c-commanded by a universal QP are not ambiguous, there only being an interpretation of wh wide scope: (164)
(165)
Siapai yang setiap orang cintai ti? who yang every person love ‘Who does everyone love?’ Setiap orang men-cintai siapa? every person TRANS-love who ‘Who does everyone love?’
It was argued that if the in situ wh-phrase in (165) were to undergo LF raising to a position equivalent to that occupied by the wh-phrase in (164) then one should expect that (165) would also exhibit the ambiguit y and dual readings found possible in (164). As (165) is however unambiguous, it must therefore be assumed that no such LF raising takes place. Given that the wh-phrase in (165) is interpreted as taking (obligatory) wide scope over the universal QP, but arguably does not raise at LF, it can be concluded that such a wide scope independent reading here cannot be a function of the wh-phrase occurring in a position outside the c-command domain of the QP (as a result of covert movement) and must arise with the wh-phrase remaining in situ in its PF position. Consequently the general possibilit y of a wide scope interpretation of wh-phrases must be allowed for even where such elements remain c-commanded by other QPs at LF. This proposal that wh-phrases may be interpreted in situ as having wide scope
30
Fiengo et al. present cases such as (161/162) among various others as strong arguments against the pied piping hypothesis account of wh-in-situ in extraction islands.
66
W h-movement and the theory of feature-checking
relative to a c-commanding QP has three main results. First one is obviously not forced to assume any covert LF raising to Comp to account for such wide scope; secondly one may capture the intuition that the wide scope interpretations of wh-phrases and indefinite NPs are essentially of the same “independent” t ype and arise in the same way (without any movement being necessary); thirdly it is possible to relate all relative scopal readings of wh-phrases to their in situ or trace positions — earlier in this section it was argued, following Aoun & Li (1993), that narrow scope interpretations may naturally be accounted for on the basis of wh-trace or wh in situ positions; now it has been suggested that wh wide scope readings may also be solely related to these rather than any raised positions. If such a position can be maintained, then arguments relating to relative scope phenomena in general do not provide strong support for the LF Wh-Movement Hypothesis, and a coherent account of wh-scope relations may be offered without the need to assume such covert raising operations.
1.7. Summary In this chapter it has been vigorously argued that wh-phrases occurring in situ at PF remain in such positions throughout a derivation until interpretation, contra proposals made frequently in the literature. Although it was seen that a number of theoretical arguments may indeed be presented as justification and support for an analysis of LF wh-movement, an LF-raising approach is also known to give rise to serious problems of localit y and appears quite irreconcilable with the potential distribution of wh-phrases in situ in islands. We therefore attempted to see if further evidence of other t ypes might suggest that the relation of an in situ wh-phrase to a +Q Comp is not in fact one of (LF) movement, and provided a range of cross-linguistic empirical evidence and theoretical argumentation from a variet y of construction t ypes strongly indicative of just such a conclusion. Those arguments which have been taken as primary motivation for the LF Wh-Movement Hypothesis were subsequently reconsidered and it was shown that in all such cases there exist alternative, plausible means available to account for the linguistic phenomena on which such arguments are based without the need to assume covert wh-movement. Consequently, if LF wh-raising can be taken both to be theoretically unnecessary and (virtually) impossible to maintain in certain instances, the interpretation of wh-phrases in in situ positions must be allowed for as a fully general possibilit y in language; significantly the phenomenon of overt wh raising where attested can therefore not be attributed to properties of language
W h in situ and the LF Movement Hypothesis
67
invariably requiring the construction of a certain t ype of input form to interpretation as has been suggested in the past. Why such overt raising must however take place in certain languages and how the licensing of wh-elements may relate to movement are questions which will now be taken up and closely examined in chapters 2 and 3.
Chapter 2 Wh-movement and feature-checking
2.1. Introduction In Chapter 1 it has been argued that wh-phrases occurring in situ at PF/Spell-Out do not undergo any covert raising to Comp at LF (at least in all the many languages under consideration there) and may receive interpretation in their in situ positions, contra assumptions and claims consistently made within the Government and Binding framework. Chapter 2 now examines why it is that overt raising of wh-phrases is nevertheless forced to take place in various languages, despite the clear expectation that the interpretation of such elements should also theoretically be possible in situ in non-Comp positions without the need for any movement. Considering the issue primarily from a Minimalist point of view and in line with the general approach set out in Chomsky (1993/95) it will indeed be concluded that cross-linguistic variation with regard to wh-raising should be accounted for in terms of purely formal licensing conditions on wh-elements — the checking of wh-features. However, a close examination of the patterning of movement in wh-questions across a range of languages will also be shown to challenge a number of commonly-held central Minimalist assumptions concerning the nature of feature-checking and movement as outlined in Chomsky (1993/95) and lead to a significantly revised conception of the localit y constraining feature-checking relations. Whereas Chomsky (1995) suggests that wh-features will only require checking
70
W h-movement and the theory of feature-checking
when strong and introduced on a +Q C0, evidence from Iraqi Arabic, Hindi and various East European languages presented throughout the chapter will be argued to indicate that wh is in fact a feature-specification which requires licensing on individual wh-phrases rather than on a +Q C0, and that all wh-phrases carry such a requirement. Such a conclusion goes against assumptions made in Chomsky (1993/95) that only a single wh-phrase in any multiple wh-question will have wh-operator-features to be checked against a +Q Comp and suggests that wh is instead more akin to Case in its need to be licensed on all the DPs which bear it. The cross-linguistic patterning found in wh-questions in the languages considered is then argued to lead to the further conclusion that wh-movement takes place for t wo essential reasons: (a) in order that a wh-phrase occurs in the checking domain of a +wh+Q Comp, and (b) so that a wh-phrase can trigger an “ambiguous” underspecified C0-head as +wh — the “Triggering Hypothesis”. Concerning (a) it is argued that feature-checking is a relation which is not in fact confined to the universal Spec–Head/Head-adjoined localit y proposed in Chomsky (1993/95) and that the checking domain relevant for the licensing of wh-feaures is actually subject to certain limited parametric variation, very much in the spirit of Manzini and Wexler’s (1987) treatment of variation in the localit y constraining the relation of (NP) anaphors to their antecedents across languages. Three possible values for the wh-checking domain are specifically identified, all taken to be associated with different lexical instantiations of a +wh+Q C0: the sentence as a unit (English, Chinese), the tense domain immediately containing a +Q Comp (Iraqi Arabic, Hindi), and the +Q Comp position of a clause itself (Romanian, Bulgarian). The strict checking domain of a head as defined in Chomsky (1995) is then taken to be simply an instantiation of one of these possible values available for (wh-)feature checking (the option selected in Romanian-t ype languages). Regarding (b), it is suggested that wh-movement may also be forced to take place in languages where the potential licensing C0-head is ambiguous and underspecified as to the particular t ype of features it will license (in English C0 ranging over +wh+Q, yes/no+Q and (pure) focus). Raising of a wh-phrase to the C0-head will serve to disambiguate it as a licensor for specifically wh-features and allow for all wh-elements in its domain to be successfully licensed. This kind of “triggering” of an underspecified Comp will however not be necessary in languages where C0 may be inherently identified as +wh in virtue of a particular +wh+Q morpheme occurring in the language. Having set out and developed a rather different view of movement and the localit y of feature-checking based on the licensing of wh-phrases in wh-questions, the chapter then attempts to show how such an approach may yield accounts of various other related phenomena, including wh-island violations, superiorit y,
W h-movement and feature-checking
71
focus-raising (and focus in situ), and differences in multiple wh-questions in Chinese and Japanese. We also return to the question of the general existence of a level of LF and covert movement. In Chapter 1 it was noted that what might be considered the strongest and most convincing arguments for assuming a level of LF presented in the past have frequently been those based on analyses of LF wh-movement. The chapter then tried to show that further investigation indicates these arguments to be significantly weaker than originally suggested. In Chapter 2 here it will be pointed out that other remaining potentially good arguments for LF also disappear if certain of the Minimalist assumptions proposed in Chomsky (1995) are adopted, with the result that the primary motivation for assuming a level of LF in a purely Minimalist model largely reduces to what will from now on be referred to as the Checking Uniformity Hypothesis (CUH) — the common assumption that feature-checking relations must be licensed within a uniform Spec–Head or Head-adjoined localit y as claimed in Chomsky (1995). Where such a Spec–Head or Head-adjoined relation bet ween elements needing to effect a checking relation is not established prior to Spell-Out, the CUH effectively requires that such a local relation be created covertly and after SpellOut, hence forcing the assumption of post-Spell-Out syntactic movement and a level of LF. Significantly, the wh-paradigms observed in this chapter will be shown to provide the strongest of empirical challenges to the CUH and indicate that feature-checking relations are not in fact universally subject to any rigidly uniform notion of localit y. A consequence of this is that the CUH can no longer be taken to provide its previous critical justification for a model incorporating LF and it becomes both possible and plausible to pursue an alternative Minimalist approach in which the syntactic derivation is actually taken to terminate at the point of Spell-Out, such a potential Minimalist model having a number of essential properties quite different from those assumed in Chomsky (1995).
2.2. Iraqi Arabic Earlier in Chapter 1 it was observed that various arguments from localit y can be given against the suggestion that wh-phrases undergo LF raising to Comp. As wh-phrases consistently seem able to appear in situ in islands for overt extraction, any analysis of direct LF movement to a +Q Comp would have to admit that preSpell-Out applications of movement may be subject to localit y constraints which do not affect those taking place after Spell-Out, a conclusion to be avoided in the
72
W h-movement and the theory of feature-checking
Minimalist framework which (as noted earlier) assumes that computational principles should apply in a uniform way throughout the derivation.1 Here we consider Iraqi Arabic and localit y-related evidence against the LF Wh Movement Hypothesis of a different and indeed opposite t ype — wh-phrases are shown not to be able to occur in situ in certain environments which freely do allow for overt extraction. If wh-phrases needed only to raise to Comp by LF to fulfill a constraint on interpretation that relevant operator-variable chains be constructed, there is no reason why it should not be possible to satisfy such requirements in these cases, as no barriers to movement are present in the structure. It will therefore be concluded that wh-elements are instead subject to a purely formal licensing requirement — wh-feature-checking — and that all wh-phrases must satisfy this requirement prior to Spell-Out. Movement at LF, though theoretically possible, will not take place as the crucial well-formedness condition on wh-phrases must always be met in the overt syntax. Various strong consequences for Chomsky’s (1995) theory of feature-checking are then seen to follow; in particular, it will be argued that feature-checking must be possible “at a distance” and non-Spec–Head-locally, and also that all wh-phrases must be licensed/feature-checked by a +wh+Q Comp. Subsequent sections relating to other languages will give rise to the same conclusions on the basis of similar evidence.
1
The possibilit y that a crucial difference in the input form to these principles in the pre- and postSpell-Out parts of the derivation might result in certain differences was discussed in Chapter 1. It was argued that if it is only pure features which move at LF and this is taken to be unlike any overt movement, then it actually becomes difficult to provide any evidence to support this being conceived of as an operation of movement. It was also stressed that “ff-movement” should not be confused with overt head-movement. Formal features themselves have no phrase-structural status and occur inside head-categories (additionally there may also be multiple features inside a single head, so there is no one-to-one correspondence relation between features and heads). As it is argued that phrase-structural categories such as X0 heads are pied piped along with formal features only for reasons of PF convergence (which consequently become redundant at LF), it must be assumed that the X0-head containing a feature set is not in fact pied piped along with its set of features when these (are suggested to) move at LF. Consequently ff-movement really has no direct parallel in the overt movement of any phrase-structural t ype. Furthermore in footnote 2 of Chapter 1 it was pointed out that hypothetical ff-movement would seem to be able to violate standard restrictions on occurrences of overt head-movement, again suggesting that the two t ypes of “movement” should not be taken to be equivalent. Despite these reasons to be wary of the general idea of ff-movement, the possibilit y of an ff-movement analysis for the patterns to be presented will be still discussed (and rejected) as we proceed into the data.
W h-movement and feature-checking
73
2.2.1. wh in situ, tense and movement The critical paradigm leading to these conclusions is given in examples (1)–(4) below. In Iraqi Arabic all wh-phrases may occur fully well-formed in situ at SpellOut both in matrix and embedded clauses, there being no requirement that a +Q Comp be filled by a wh-element prior to Spell-Out (unlike English-t ype languages):2 (1)
(2)
Mona shaafat meno? Mona saw whom ‘Who did Mona see?’ Mona raadat [tijbir Su’ad [tisa’ad meno]]? Mona wanted [to-force Su’ad [to-help who ‘Who did Mona want to force Su’ad to help?’
However, although example (2) is fine where the wh-phrase remains in situ in an embedded non-finite clause, a wh-phrase significantly may not occur in situ within an embedded tensed clause when relating to the +Q Comp of a higher CP: (3)
*Mona tsawwarat [CP Ali ishtara sheno] Mona thought Ali bought what Intended: ‘What did Mona think that Ali bought?’
The relevant generalisation appears to be that a wh-phrase must occur in the “tense domain” of a +Q Comp in order to be licensed, where a tense domain may be understood to consist of a tensed/+finite clause and any non-finite clauses dependent on the tensed clause.3 In (3) the bracketed CP constitutes the first tense domain including the wh-phrase sheno ‘what’, but as this +finite CP does not contain a +Q Comp, the result is that the wh-element is not licensed. In (2) as the lower embedded clauses are all -finite, the first CP to count as a tense domain including the wh-phrase meno ‘who’ is actually the entire sentence, and as this may also potentially contain a +Q Comp the sentence is well-formed. Although Huang (1982) and others have noted that (certain t ypes of) wh-adjuncts may be restricted by extraction islands, in the past it has commonly
2
All Iraqi Arabic data here is taken from Wahba (1990) and Ouhalla (1996)
3
It should be noted that tense is not expressed by any overt morpheme in Iraqi Arabic. Those clauses referred to here as being tensed or +finite are those which (a) receive a tensed interpretation, and (b) license overt subjects, and therefore have standard properties of tensed/+finite clauses.
74
W h-movement and the theory of feature-checking
been assumed that the distribution of argument wh-phrases in situ is unconstrained by localit y and that such elements are free to occur in all kinds of environments which may block overt movement. Here however the distribution of in situ wh-phrases in Iraqi Arabic is seen to be restricted in an important way, the occurrence of +tense in a clause clearly resulting in a domain which is opaque for the licensing of wh-phrases. If one presumes that wh-phrases in situ at PF need undergo raising to Comp at LF, one might perhaps attempt to suggest that the ungrammaticalit y of sentences such as (3) is due to this hypothetical movement somehow being blocked and argue that the restrictions on the distribution of wh in situ observed here are simply related to constraints on LF movement. This t ype of position has commonly been adopted to account for other noted restrictions on the occurrence of in situ elements, such as the ban on why/how adjunct t ypes in islands in Chinese (Huang 1982). However, tensed CPs do not appear to constitute barriers to movement in Iraqi Arabic, and not only may the wh-phrase in (3) undergo raising to the matrix from the embedded tensed CP, but when it does do so the result is significantly a fully well-formed question: (4)
shenoi tsawwarit Mona [Ali ishtara ti]? what thought Mona [Ali bought ‘What did Mona think Ali bought?’
Sentences such as (4) thus constitute direct evidence that movement of wh-phrases in Iraqi Arabic is in fact potentially unbounded and the tensedness of a clause does not block movement of elements such as sheno ‘what’. Given the natural Minimalist assumption that derivational constraints should apply in a uniform way both prior to and after the point of Spell-Out (or else recognise Spell-Out as an S-structure-like level of derivation relative to which constraints may be stated), hypothetical LF movement of the wh-phrase in (3) should indeed be permitted, in the same way that pre-Spell-Out raising is licensed.4 As sentences such as (3) are
4
Supporters of ff-“movement” may try to object here that LF ff-movement might be subject to different constraints than overt (full-)category movement. It could be suggested that ff-movement is blocked by tense whereas category movement is not and that this could account for the difference in grammaticalit y in (3) and (4), i.e. it is the input form to the localit y restrictions which critically varies prior to and after Spell-Out rather than the localit y restrictions themselves. However, there are good reasons to believe that even if one does concede the existence of some kind of ff-movement this will not in fact be able to explain the differences observed in (3) and (4). Chomsky (1995) essentially suggests that movement at LF is ff-movement for reasons of economy: at LF it becomes possible to raise just ffs and this is cheaper than movement of any larger containing constituent, hence the option that is automatically selected. However, principles of
W h-movement and feature-checking
75
consistently unacceptable and this arguably cannot be ascribed to any violation of a constraint on movement, some other type of explanation for the ungrammaticalit y of (3) is clearly called for. A possibilit y more compatible with recent Minimalist developments may instead be to suggest that raising of the wh-phrase in (3) would not be blocked from occurring at LF but would then be occurring too late in derivational terms to save the structure from crashing, that some formal licensing propert y related to the wh-phrase critically requires satisfaction prior to Spell-Out and this may therefore not be achieved via post-Spell-Out raising (although such movement would not violate other syntactic principles). Following Chomsky (1993/95), if all operations of movement are triggered by a need to check morphological features, it can be suggested that while overt raising of the wh-phrase in (4) results in successful checking of wh-features between the wh-phrase and the +Q Comp prior to Spell-Out, the ungrammaticalit y of (3) is due to a failure to check wh-features by this derivational point. Assuming such an approach, the paradigm in (1)–(4) leads to the further important conclusion that it is wh-features carried by wh-phrases themselves which are here in need of licensing and not any assumed present on the +Q C0. Were the latter to contain strong wh-operator features, then overt wh-raising should clearly always have to take place as in English, yet examples such (1) and (2) are noted to be fully well-formed without any pre-Spell-Out movement. Wh-movement in Iraqi Arabic instead seems to relate directly to the wh-phrase itself and its position relative to the +Q Comp, being triggered only when the former occurs in
economy are standardly taken to constitute preferences which may be over-ridden for reasons of convergence (and it is precisely for convergence at PF that full categories are pied piped prior to Spell-Out). Consequently, in order to rescue structures where wh-features hypothetically occur “stranded” in an embedded tensed CP, it should indeed be possible to select LF movement of the entire category containing these features (i.e. category pied piping) and so avoid the suggested barrierhood problem. Raising of the DP category containing the relevant wh-features in (3) at LF should violate only those constraints which are violated by overt raising of the same category t ype, and as overt raising of the wh-phrase in (4) is seen to be grammatical, the potential LF raising of the wh-DP should similarly be acceptable. As there is no principled way to disallow category pied piping for convergence at LF (i.e. after the point of Spell-Out) but allow for it before this point without stipulating a constraint over the point of Spell-Out itself, an ff-movement approach to the contrasts here cannot be maintained and a different t ype of solution is clearly necessary. Note that it would simply be blatant stipulation to suggest that category pied piping is only available where the phonetic matrix of a category is also moved, i.e. prior to Spell-Out and not at LF; the computational system should be able to move a category (if necessary) whether it has an attached phonetic matrix or not (and the phonetic matrix of an element should actually be invisible to the computational system, this operating only over syntactic structure. The phonetic features of an element should clearly only be interpretable and visible in the PF branch of the model).
76
W h-movement and the theory of feature-checking
a domain which is opaque for licensing. If movement may again only be assumed to take place for the satisfaction of feature-checking requirements, this clearly indicates that wh-features carried by the wh-phrase itself must be checked by the (obligatory) raising in (4). Such a conclusion would then seem to constitute strong evidence against the suggestion in Chomsky (1995) that +interpretable features (such as wh) will only require checking when strong and present on a functional head — here it is seen that the relevant functional head C0 cannot be argued to carry any strong (operator-)features and it is crucially for the licensing of a wh-phrase (and checking of wh-features carried by it) that movement is necessitated. Consequently wh (and possibly other so-called +interpretable features) may be taken to be a feature specification just as much in need of checking on the XPs which carry it as (for example) the interpretable Case-features present on DPs. A second conclusion to result from a consideration of examples (1)–(4) concerns the locality of feature-checking. Chomsky (1993/95) suggests that feature-checking is a relation which may only be effected within the “strict locality” of Spec–Head and Head-adjoined configurations, this propert y always requiring movement of an element to its checking head. However, the patterning of wh-phrases and movement in Iraqi Arabic as illustrated in sentences (1)–(4) provides evidence that such a claim cannot in fact be (universally) correct, and that feature-checking must also be possible “long distance” within wider domains. (5) below briefly recaps what (1)–(4) have been taken to suggest: (5)
a.
b.
c.
The movement of wh-phrases in Iraqi Arabic as in (4) can only be assumed to take place for the checking of (wh-)features, and will save illicit structures such as (3) from crashing. The relevant features which are checked by movement of a wh-phrase are wh-features on the wh-phrase itself and not any strong operator features on Comp, otherwise wh-raising to Comp would have to take place in all wh-questions (but this is not so, as seen in (1)–(2)). Checking of wh-features on wh-phrases must furthermore take place prior to Spell-Out — hypothetical LF movement of wh-phrases in unacceptable examples such as (3) though not blocked syntactically, would come too late in the derivation to satisfy licensing of the wh-phrase, and indeed is forced to occur overtly, as in (4).
Now critically, although it is concluded that wh-features carried by wh-phrases must be checked prior to Spell-Out, raising to Comp is not forced to take place where a wh-phrase occurs base-generated in the tense domain of the +Q Comp (1)–(2), and a wh-question is fully acceptable when the wh-phrase remains fully in situ. As such wh-elements must however be assumed to carry wh-features in need
W h-movement and feature-checking
77
of pre-Spell-Out checking in the same way that wh-phrases base-generated in embedded tensed CPs do, it can only be concluded that these features are checked prior to Spell-Out on the wh-phrases in their in situ positions. Clearly not being in the Specifier of the checking head (C0) at the point at which wh-checking must be effected, it therefore must be conceded that feature-checking is not in fact always subject to checking within the strict localit y of Spec–Head or Headadjoined positions (i.e. the Checking Uniformit y Hypothesis) and may at least in certain instances also take place within larger domains.5 Although wh-feature-checking in Iraqi Arabic is thus (by hypothesis) not bound to be strictly Spec–Head local, it nevertheless is still constrained by some notion of localit y defined relative to tense — a wh-phrase will only be licensed in the immediate tense domain of the +Q Comp. In addition to this tense-domain restriction, there are in fact also certain other localit y restrictions which appear to correspond more closely to familiar constraints on applications of movement. As illustrated in examples (6) and (7), wh-phrases may not occur in situ either in relative clauses or in wh-islands with scope higher than the +Q Comp of the wh-island itself:
5
Noting it is not possible to suggest that any “empty wh-operator” raises to Comp from the in situ wh-phrases (as Watanabe 1991 proposes is the case in Japanese); if this were to be so, then it should also be possible for such an empt y operator to raise to the +Q Comp from wh-phrases occurring in embedded tensed CPs and examples like (3) would incorrectly be predicted to be well-formed. Furthermore, such empt y operator analyses have basically been proposed and seem appropriate for languages in which the phonetically-overt core of elements functioning as wh-question words is quantificationally under-specified and may be interpreted in various ways depending on the t ype of operator present (see e.g. Nishigauchi 1990 on Japanese, Cheng 1997 on Chinese). As wh-elements in Iraqi Arabic (and Hindi, Section 3) are unambiguously wh-question words, they would not seem obvious candidates for any similar null operator proposal. Finally, where it is indeed possible to compare movement of null operators and phonetically overt elements containing (or equivalent to) such operators within a single construction-type, as in Japanese wh-questions where a null-operator or the phonetically overt core plus the operator may move (see Section 2.6), or relative clauses in English where the relative pronoun may either be phonetically null or overt, there would not seem to be any difference in the localit y constraints on such movement, i.e. movement of the null operator is not more restricted than that of the overt element. Consequently there is no reason to expect that raising of any null operator from a tensed CP in Iraqi Arabic should be more constrained than movement of an overt XP.
78
W h-movement and the theory of feature-checking
(6)
(7)
*Mona ’urfit [il-binti [illii ti ishtarat sheno]]6 Mona knew [the-girl [who bought what ‘Whati did Mona know the girl who bought ti?’ Mona nasat [li-meno tinti sheno] Mona forgot [to-whom to-give what ‘Mona forgot what she should give to whom.’ NOT: What did Mona forget whom to give to?
As such constituents are indeed islands for syntactic extraction, it might then be suggested that an LF movement analysis of in situ wh-phrases should in fact be pursued, despite the above given argumentation to the contrary. In what follows however, it will be argued that the observed island-sensitivit y ultimately does not constitute evidence in favour of an LF movement approach and instead actually adds further support to the claim put forward here that wh-phrases in Iraqi Arabic must be checked prior to Spell-Out. Ouhalla (1996) has noted that if the wh-elements in (6) and (7) are overtly extracted from their containing island environments, the resulting questions are markedly less unacceptable than when the wh-phrases remain in situ inside the islands: 6
Note that in (7) the wh-phrase sheno ‘what’ occurs within a tensed CP that does not also contain the +Q Comp; it might therefore be suggested that the tensedness of the relative clause will in any case block licensing of the wh-phrase by a higher +Q Comp, regardless of the potential islandhood of the CNP. In Iraqi Arabic it is not possible to form non-tensed CNPs of any kind, hence the hypothesis that +tense in the relative clause is blocking licensing of the wh-element cannot be checked in this way. However it is possible to control for the tensed factor in another way. As well as standard in situ wh-questions of the t ype seen in (1),(2),(4), Iraqi Arabic also allows for the use of a particular wh-question particle (QP) strategy to form wh-questions (this strategy is examined further in Chapter 3). An interrogative element sheno often reduced to sh- appears clause-initially and all other wh-phrases may remain in situ. The use of such a QP has the interesting effect of partially overcoming the noted opacit y effect induced by tense on wh in situ, so that a wh-phrase may licitly occur in situ in an embedded tensed CP licensed by the +Q Comp of an immediately higher tense domain (compare (i) below with 3 above): (i) sh-tsawwarit Mona [Ali raah weyn]? QP-thought Mona [Ali went where ‘Where did Mona think that Ali went?’ However, if one attempts to use this QP-strategy to overcome the possible interfering tense effects with wh in situ in CNPs, the resulting questions are still ill-formed: (ii) *Sh-’urfut Mona [il-binti [illii ti ishtarat sheno]]? QP-knew Mona [the-girl [who bought what ‘What did Mona know the girl who bought?’ Consequently it may be concluded that it is not the +finite nature of the CP containing the wh-phrase in the relative clause which is causing a problem for licensing of the wh-phrase here.
W h-movement and feature-checking (8)
(9)
79
??Sheno
i nasat Mona [li-meno tinti ti]? what forgot Mona [to-whom to-give ‘What did Mona forget to whom to give?’ ??Sheno ’urfut Mona [ilbint illi ishtarat t ]? i i what knew Mona [the-girl who bought ‘What did Mona know the girl who bought?’
Whereas (6) and (7) are both completely unacceptable and unintelligible as questions, Ouhalla suggests the reduced acceptability of (8) and (9) is typically that of regular subjacency violations, resulting (simply) from the illicit extraction of an element from within an island configuration. Given that (6) and (7) are significantly worse, it must therefore be concluded they are violating some constraint other than subjacency.7 If one assumes that the licensing of wh-phrases (checking of wh-features) is in some way critical for their interpretation, and further adopts suggestions made above as to how and at what derivational point such licensing/checking must occur, the difference in acceptabilit y between (6)–(7) and (8)–(9) may actually be predicted — in the latter two examples the wh-phrases move to a +Q Comp prior Spell-Out and so are successfully checked by this point, their interpretation as interrogative wh-phrases being licensed; in achieving this however, a pure constraint on movement is violated (subjacency), resulting in the reduced acceptabilit y judgements. In (6) and (7) by way of contrast, the wh-phrases do not appear in a domain where their wh-features can be checked by Spell-Out and so the structures will automatically crash, the wh-elements not being licensed as wh-phrases and hence not allowing for any coherent interpretation. In this sense it can be suggested that sentences such as (6)–(7) are like other instances where morphological features fail to be checked by a certain derivational point and this results in unintelligibilit y, as for example in (10) and (11) where subject D(P) features remain unchecked at Spell-Out. Although no localit y conditions would block the subject DPs from raising at LF, this would simply come too late to check the strong features in AgrS0/T0:
7
Indeed, if the well-formedness requirement on wh-phrases were just to be that they occur raised in a +Q Comp by LF, one might expect that (6)–(7) should perhaps be less unacceptable than (8)–(9), or even fully grammatical, given that LF movement of the wh-phrase might be able to proceed in a way different from that in (8)–(9) where straight and direct extraction from an island environment has occurred. For example, LF raising might be able to make use of island-pied piping operations, as suggested in Nishigauchi (1990), or of the QR-dependent extraction mechanism outlined in Fiengo et al. (1988). That (6)–(7) are actually worse than (8)–(9) then strongly indicates that some propert y of the sentences must crucially be satisfied before Spell-Out.
80
W h-movement and the theory of feature-checking
(10) (11)
*Not John come. *Did not John come. (intended to be a statement)
Thus although the unacceptabilit y of wh-phrases in situ in certain extraction islands might initially prompt one to an LF movement approach, there are good reasons for assuming that LF wh-movement does not in fact take place — both the contrasts in (6)–(7) and (8)–(9) and the unacceptabilit y of wh-phrases in situ in non-island embedded tensed CPs. If this is the case, then the ungrammaticalit y of (6)–(7) with wh-elements in situ in wh- and CNP islands cannot be accounted for in terms of constraints on movement, and it must be conceded that there may also exist island-like localit y constraints on purely non-movement (licensing) relations — a claim which has however previously been argued for by Cinque (1991) (relative to the island-sensitivity of clitic left dislocation structures in Italian) and Bresnan (1976) (on comparative deletion), hence one that is not without independent support.8 Finally a brief consideration of multiple wh-questions can be shown to offer additional and fairly conclusive evidence that it is indeed (wh-features on) wh-phrases themselves which are critically in need of licensing/checking in Iraqi Arabic (rather than any on the +Q Comp) and that all wh-elements may be assumed to carry such a feature-checking requirement. As (12) and (13) show, although multiple wh-questions are permitted in Iraqi Arabic, the distribution of “secondary” wh-phrases is far from free and may directly result in a multiple wh-question being unacceptable: (12)
(13)
shenoi ishtara Ali ti [minshaan yenti li-meno]? what bought Ali [in-order-to give to-whom ‘What did Ali buy to give to whom?’ *meno tsawwar [Ali xaraj weyya meno]? who thought [Ali left with whom ‘Who thought that Ali left with whom?’
In (12) both wh-phrases are straightforwardly licensed by occurring in the same tense domain as the +Q Comp (the lower CP being finite). As the higher wh-phrase in (13) will also be licensed/checked quite regularly in virtue of its position in the tense domain of the +Q Comp, the unacceptabilit y of the sentence can only be ascribed to the presence of the second wh-phrase meno ‘whom’ in the embedded +finite CP, indicating that secondary wh-phrases are constrained by 8
The issue of localit y constraints on non-movement dependencies is returned to briefly in Section 2.3.1.
W h-movement and feature-checking
81
precisely the same factors as “primary” wh-phrases and may not occur in embedded tensed CPs where the licensing +Q Comp is in a higher tense domain. Consequently it must be assumed both that every wh-phrase present in a wh-question carries (wh-) features in need of licensing, and furthermore that all such wh-features must be checked by the same derivational point, namely SpellOut; if this were not the case, then the lower wh-phrase in (13) should indeed be able to raise up to the +Q Comp and check its features at LF (there being no barrier to such movement), yet the sentence is fully ungrammatical.9 To sum up and recap the main points of this section, it has been suggested that wh-movement takes place in order to satisfy certain purely formal licensing requirements on wh-phrases. In Iraqi Arabic it is found that wh-movement is only obligatory for a subset of all wh-phrases, namely those which do not occur basegenerated in the tense domain of a +Q Comp. Movement not being forced on all wh-elements, it cannot be argued that the wh-movement takes place in order to build operator-variable chains or establish scope, as this should clearly be a requirement which would necessitate all wh-phrases to raise to some +Q Comp. Furthermore, it was argued that the raising of wh-phrases base-generated in embedded tensed clauses must take place prior to Spell-Out and cannot be delayed until LF, indicating that the raising occurs to satisfy some mid-derivational constraint. Noting the Minimalist assumption that movement may only occur for purposes of feature-checking, the formal licensing requirement satisfied by wh-movement was naturally identified as a need to check wh-features. It was further argued that the wh-feature-checking requirement relates purely to wh-phrases themselves and not to the +Q Comp (in Iraqi Arabic, at least), and also that all wh-phrases must undergo checking prior to the point of Spell-Out. This then led to the conclusion that wh-feature-checking is in fact not restricted to occurring only in the strict localit y of Spec–Head (or Head-adjoined) configurations and in some instances must be available “long-distance” — if all wh-phrases must be checked by a +Q Comp prior to Spell-Out, then those wh-phrases which are found occurring in situ in the tense domain of the +Q Comp at PF must be licensed non-Spec–Head-locally in these in situ positions. (14)–(17) below repeat these rather important points in summary form, statements which will shortly be
9
As pointed out in an earlier footnote, in an ff-movement approach, there is no reason why at LF the wh-features present on the wh-phrase in the embedded finite clause here should not be able to pied pipe the containing wh-phrase in order to raise (covertly) to the matrix clause, pied piping being available for convergence (and raising of full wh-phrases out of finite clauses is clearly permitted). That (13) is nevertheless ungrammatical then indicates that the relevant wh-features must be checked prior to Spell-Out.
82
W h-movement and the theory of feature-checking
shown to characterise the wh-patterning in Hindi in precisely the same way, so adding strength and support to the basic conclusion that the Checking Uniformit y Hypothesis is not in fact a universal constraint. (14) (15) (16) (17)
All wh-phrases in Iraqi Arabic have a feature-checking requirement. +Q Comps in Iraqi Arabic do not show any signs of having a wh-feature checking requirement. Wh-features on wh-phrases in Iraqi Arabic must be checked prior to Spell-Out. The wh-feature checking localit y in Iraqi Arabic is the tense domain of the +Q Comp. Wh-features may be checked on wh-phrases occurring in any position m-commanded by a +Q Comp within its own immediate tense domain.
2.3. Hindi In Hindi, as in Iraqi Arabic, all wh-phrases may occur in situ at PF and there is no requirement that a +wh+Q Comp be filled by any wh item prior to Spell-Out, unlike English/German etc:10,11 (18)
Raam-ne [Mohan-ko kise dekhne-ke liye] kahaa? Ram-ERG [Mohan-ERG whom to-see for told ‘Who did Ram tell Mohan to look at?’
However, whereas wh-phrases may occur in situ in embedded non-finite CPs, as in (18) above, they may not do so in equivalent tensed clauses:12 (19)
*Raam-ne kahaa [ki kOn aayaa-hE]? Ram-ERG said [that who has-come Intended: ‘Who did Ram say has come?’
10
Most Hindi data here is taken from Mahajan (1990).
11
Mahajan’s own pre-Minimalist analysis of the wh-patterns here will be considered a little later. Here I wish to concentrate on pointing out the strong parallels with Iraqi Arabic and how they must be interpreted from a Minimalist point of view.
12
Unless there is a wh-question-particle in the superordinate clause, this being fully parallel to the use of wh-question particles in Iraqi Arabic (see footnote 5 and remarks there). Chapter 3 returns to this question-particle strategy and compares it with partial movement structures in German and Hungarian.
W h-movement and feature-checking
83
Such tensed CPs nevertheless are not islands for extraction, and as with Iraqi Arabic, not only may a wh-phrase undergo overt raising from a tensed clause, but when this occurs in examples like (19) yielding (20) the result is a perfectly wellformed question: (20)
kOni Raam-ne kahaa ki ti aayaa-hE who Ram-ERG said that has-come ‘Who did Ram say has come?’
Therefore, as with Iraqi Arabic, one is forced to assume that although LF raising to the +Q Comp in (19) must be possible and would furthermore result in a configuration which is well-formed at PF/Spell-Out, such hypothetical raising would come too late to satisfy certain properties of the wh-question. As pre-SpellOut movement of the wh-phrase in (20) will save (19) from otherwise being unacceptable, and as a +wh+Q C0 in Hindi does not always require a wh-element in its Specifier (hence cannot be taken to be generated with strong wh-operator features), it can again only be concluded that wh-raising in (20) takes place to check wh-features carried by the wh-phrase itself prior to Spell-Out. If wh-phrases in Hindi must therefore be feature-checked by Spell-Out, where other wh-phrases are seen to occur quite licitly in situ (as in 18), these wh-phrases must be assumed to be licensed and feature-checked in their in situ positions and consequently not in any strict Spec–Head relation with the checking head C0. Feature-checking is then once more attested to be possible “at a distance”, though again constrained by tense factors and blocked where a wh-phrase occurs in situ in a tense domain which does not contain the +Q Comp (19).13
13
Note that if one were to attempt an ff-movement analysis of the data (despite all the difficulties concerning the idea of ff-movement already mentioned) and suggest that LF ff-movement is sensitive to +tense in a way that overt category-movement is not in order to account for the contrasts observed, one would have to make the assumption that wh-features are weak in Hindi/Iraqi Arabic. If they were to be strong, then the wh-features would have to be checked immediately upon introduction and the wh-movement would have to be overt. However, if the wh-features were to be weak, then as +interpretable features they should not be forced to move and check at all according to Chomsky (1995), so such an explanation should not be possible. Supposing one were then to try to claim that the wh-features on wh-phrases are interpretable somehow, such an assumption should mean that all wh-phrases would have to be checked at some derivational point. This then gives rise to enormous problems for the many languages such as English where wh-phrases occur in situ in all island t ypes. Here the hypothetical LF ff-movement would be violating all kinds of standard localit y constraints and it is highly unlikely that ff-movement conceived of as head-movement could take place out of relative clauses and other strong islands. It would also just be pure descriptive stipulation to suggest that wh-features on wh-phrases in Iraqi Arabic/Hindi are interpretable whereas those on English-t ype languages are
84
W h-movement and the theory of feature-checking
One also finds evidence in Hindi similar to that presented in Iraqi Arabic that all wh-elements require licensing by Spell-Out: (21)
*kOn Raam-ne kahaa ki ti kis-ko maaregaa who Ram-ERG say that who-ACC will-hit Intended: ‘Who did Ram say will hit who?’
In (21) raising to the matrix will effectively licence the first wh-phrase kon ‘who’, but the second wh-phrase kis-ko ‘whom’ also apparently requires licensing by SpellOut, this not being possible here as it remains inside a non-interrogative tensed clause at the point at which feature-checking must take place/Spell-Out. Movement of this wh-phrase out of the tensed CP is both possible and will indeed result in a well-formed question if it takes place prior to Spell-Out, as shown in (22) below: (22)
kOni kis-koj Raam-ne kahaa ki ti tj maaregaa who whom Ram-ERG say that will hit ‘Who did Ram say will hit who?’
Thus again it is seen that all wh-elements must be taken to carry wh-features in need of checking and that checking of all wh-features must take place uniformly by Spell-Out, noting once more that if movement may only be triggered by featurechecking requirements and obviously results in saving an ungrammatical form (22) vs. (21) then the movement of the second wh-phrase in (22) can indeed only be for such (feature-checking) purposes. The patterning observed in Hindi consequently results in the same set of significant conclusions as those made earlier on the basis of Iraqi Arabic. First of all one finds that it is the wh-phrases in a wh-question rather than the +Q Comp which have wh-feature-checking requirements, contra assumptions made in Chomsky (1993/95) that +interpretable features such as wh will only ever require checking on the +Q Comp (when strong). Second, all wh-phrases carry wh-features in need of checking and this needs to be effected prior to Spell-Out, and third and most important of all the feature-checking relation between the +Q Comp and any wh-phrase present need not be Spec–Head local but can be effected between the +Q Comp and any position it m-commands in its own immediate tense domain. Consequently the Checking Uniformit y Hypothesis
+interpretable. As pointed out in footnote 5 wh-words in Iraqi Arabic and Hindi are unambiguously +interrogative so there are no distinctions between wh-words in Iraqi Arabic/Hindi and English which could lead to any plausible ±interpretable partitioning.
W h-movement and feature-checking
85
which largely supports the LF-based architecture of the Minimalist model cannot in fact be a universal and invariable constraint.
2.3.1. Locality constraints and (non-)movement dependencies. The contrasts in (19)–(20) and (21)–(22) above illustrate that wh-feature-checking relations in Hindi are subject to localit y conditions which do not correspond to those on movement (i.e. +tense blocks “long-distance” wh-feature-checking but not movement, just as in Iraqi Arabic). If this licensing relation between the +Q Comp and the wh-phrase does not involve movement at any level of derivation, as indeed argued, then once again we find clear evidence for the existence of localit y constraints on purely non-movement relations. As in Iraqi Arabic, it may be further noted that certain other syntactic environments which constitute islands for movement do also block the hypothesised wh-licensing relation in Hindi, specifically CNPs and wh-islands. This partial overlap with localit y restrictions on actual movement dependencies should not lead one to presume that any movement is necessarily involved in wh-licensing however. As with Iraqi Arabic it can be shown that there is a significant difference in acceptability between extraction and the illicit unlicensed occurrence of wh-elements in situ within such configurations: (23)
(24)
Raam-ne Mohan-se puuchaa ki kis-ne kyaa kEse thiik-kiyaa Ram-ERG Mohan-to asked that who-ERG what how fixed ‘Ram asked Mohan who fixed what how.’ NOT: ‘What did Ram ask Mohan who fixed how?’ ??kOn sii tiim Raam-ne puuchaa ki kis-ko i k Mohan soctaa-hE which team Ram-ERG asked that who-ACC Mohan thinks ki ti tk haraa-degii that will-defeat ‘Which teami did Ram ask whok Mohan thought ti would defeat tk?’
In (23) it is not possible for either of the wh-phrases in the lower wh-island to be interpreted as taking scope at a matrix +Q Comp, hence that licensing of a wh-phrase in a wh-island by a higher +Q Comp is not permitted as an option. However, this does not relate to any impossibilit y of movement — in (24) the wh-phrase kOn sii tiim ‘which team’ is extracted from within the island to the matrix clause and this does result in a coherent interpretation of the wh-phrase as being directly questioned, hence licensed by the matrix +Q Comp. The sentence is not fully acceptable because constraints on movement are thereby violated, any extraction (e.g. of topic phrases) out of wh-islands also resulting in a similar
86
W h-movement and the theory of feature-checking
degraded status. If wh-licensing were always to involve movement and could be effected by LF raising of the wh-phrase kyaa ‘what’ to the matrix +Q Comp in (23), one should expect both (23) and (24) to be equally acceptable. However, (23) is markedly worse than (24) (and indeed impossible) on the attempted interpretation, indicating that it is not a localit y restriction on LF movement which is responsible for the contrast. The contrast may instead be explained by assuming that wh-phrases in Hindi must be licensed by Spell-Out and that this is achieved in (24) but blocked (relative to a matrix +Q Comp) in (23), perhaps because a wh-phrase in Hindi must be licensed by the nearest c-commanding +wh+Q Comp, Relativized Minimalit y then acting as a constraint on non-movement relations.14 There is consequently reason to believe that the CNP/wh-island-t ype effects here are not in fact due to any LF movement occurring and once again one needs to concede that various non-movement dependencies may also be constrained by a distinct notion of localit y (which may nevertheless overlap with constraints on dependencies resulting from extraction). The suggestion that certain islands for syntactic movement may also constrain and block other dependency t ypes which cannot be analysed as resulting from movement has indeed been made in a number of works in the past. One wellknown case is that of clitic left dislocation (CLLD) examined in Cinque (1991). In Italian the relation between a left-dislocated XP and a resumptive clitic in Italian is shown to be sensitive to strong islands and thus might appear to be formed from a topicalisation-fronting operation. Cinque however convincingly argues that the clause-initial XPs in CLLD constructions are base-generated in their PF positions and have not undergone raising, justifying this conclusion with observations such as the following: (a) CLLD is not even mildly sensitive to other Weak Islands which do constrain movement (e.g. wh-islands and Inner-Neg islands), (b) the “gap” in CLLD constructions can only be that of the clitic raised to Infl and not of the initial XP as no clitic-doubling is possible elsewhere in Italian, hence the left-dislocated XP must be base-generated in its clause-initial position and not moved there, (c) ne-cliticization facts also indicate no movement of the initial XP, CLLD contrasting with parallel focus constructions where ne-cliticization phenomena do indicate movement, and (d) CLLD is bad where the “gap” is the subject of a clause selected by an Exceptional Case-Marking verb, whereas movement from such a position is always acceptable (e.g. focus/wh-
14
See also Li (1992) for a view of Relativized Minimalit y as a non-movement-related localit y condition.
W h-movement and feature-checking
87
movement).15 There are thus many independent reasons to reject a movement analysis despite the strong island sensitivit y of CLLD and strong islands are therefore ultimately held to be a constraint on certain non-movement dependencies too. Bresnan (1976) also shows that localit y constraints previously associated only with movement relations appear to affect constructions in which there is arguably no movement. Discussing comparative constructions which Chomsky (1977) analyses as involving raising of a null operator, she provides good evidence that deletion rather than movement is involved. What is “deleted” may often correspond to a non-constituent sub-part of a larger phrase that normally may not undergo movement without pied piping of the remainder of the containing phrase; this is illustrated in (25)–(27) below where the quantificational modifier of a DP can be deleted but not moved/extracted: (25) (26) (27)
She has as many boyfriends as she has [[books]]. *[How many]i does she have [ti [books]]? *[So many]i does she have [ti [books] that…
Bresnan notes in this respect that there are other cases where linear sequences which do not comprise constituents may be affected by deletion rules, as for example in gapping (and therefore deletion rules contrast with the operation of movement which may only apply to strings that form constituents): (28)
Jill tried to hit Jack, and Jack [ ] Jill
Furthermore, certain elements that appear to be deleted in comparative constructions may also be omitted in other environments, even when there is no comparison (and hence no possibilit y of a null operator and movement): (29) (30)
He’s as good as a singer of lieder as he was [[ ] of pop songs]. Are there many nuggets of gold in the jar? There certainly don’t seem to be [many [ ] of pyrite].
Comparative constructions are then shown to be sensitive to island constraints, as shown in (31) and (32) below. Because for the other reasons given above these constructions should be analysed as involving deletion rather than movement, the conclusion made by Bresnan is that island-sensitivit y cannot be taken as an unfailing diagnostic of movement:
15
It is argued that Case is exceptionally assigned to SpecCP rather than SpecIP, so that any element which undergoes movement through SpecCP will successfully be Case-marked. CLLD in these instances is bad as a pro base-generated in SpecIP will not receive Case.
88
W h-movement and the theory of feature-checking
(31) (32)
*Therefore they can hire more men than I met a woman who has [[] boyfriends]. *We ordered more warheads built than we expected the announcement that they had [[ ] missiles].
Similar conclusions may also be reached relative to antecedent-contained deletion (ACD) as indicated in Chapter 1. Such constructions appear to be constrained by subjacency/the CED yet no actual movement can have taken place for the various reasons given there: (33) (34)
*John read everything which Mary believes [the report that he did [ ]]. *John read everything which Mary wonders why he did [ ].
In another set of cases, a licensing relation between two elements appears subject to certain island-like localit y, yet it is implausible to suggest that movement relates licensor and licensee in the dependency as there does not seem to be any morphological feature common to both in need of checking (and movement is assumed to take place solely for such feature-matching/checking). This is the case with the licensing of NPIs in various languages, which seems to show sensitivit y to the CNPC: (35)
*I didn’t find a man who knew anything.
If one attempts to relate this localit y restriction to hypothetical LF movement triggered between the NPI ‘anything’ and the licensing negation, one would have to argue that negation checks some inherent neg-feature carried by the NPI. However, it is well-known that NPIs are licensed by a variet y of different elements (conditionals, comparatives, yes/no questions, negation etc); given that similar localit y effects show up also with these other licensing elements, one would have to suggest that NPIs carry an array of different features each one of which might trigger raising for checking against a licensor, an undesirable claim.16 Indeed, it
16
Undesirable as none of the licensors require an NPI (hence the licensors cannot carry strong features relating to an NPI); if it is claimed that it is features on the NPI which trigger raising, then only one such feature of the set an NPI must hypothetically carry will be checked in any instance — e.g. in (35) only its neg-features will be checked but none of the features that relate to other licensing elements. Such unchecked features should result in crashing every time an NPI is employed. It also seems unlikely that a single feature-t ype could trigger raising to all of the different licensors. In contrast to standard NPIs, n-words in certain languages do seem to show various clear characteristics of feature-checking dependencies, being licensed by only one functional head t ype (Negation) and also undergoing overt mid-derivational movement for licensing in a number of
W h-movement and feature-checking
89
is not commonly assumed that NPI licensing involves any movement operation, partially perhaps for these very reasons, but also because NPI-licensing does not seem to be sensitive to other island environments (so that the localit y evidence is ultimately not even strong either). It must then be admitted that there are certain syntactic configurations such as CNPs which ostensibly block various nonmovement licensing relations too. This is further seen with the licensing of “indefinite” wh expressions in Chinese (Li 1992). In Chinese the elements which function as wh-interrogative words may also receive interpretation as indefinite NPs when licensed in a variet y of environments, some of which are similar to those licensing NPIs, e.g. conditionals, negation, yes/no questions. This licensing illustrated in (36) would however appear to be sensitive to a form of wh-island Relativized Minimalit y. Example (37) shows that an element such as shenme may be licensed as an indefinite NP ‘something/anything’ by the propositional attitude verb yiwei ‘think’. When the same verb yiwei occurs in example (38) however, it is unable to license the indefinite NP interpretation on shenme in the lower clause because the lower clause in (38) is a wh-island. The result is that shenme in (38) may only be interpreted as a wh-interrogative expression ‘what’: (36)
(37)
ta xihuan shenme ma? he like what Q ‘Does he like something (anything)?’ ta yiwei wo xihuan shenme. he think I like what ‘He thinks I like something.’
languages. Interestingly the Iraqi Arabic/Hindi wh-tense paradigm under discussion in this chapter is repeated entirely in Japanese and Korean with n-word licensing/checking of neg-features. Nishiyama, Whitman & Yi (1996) report that although an n-word in an embedded tensed clause in both Japanese and Korean may not be licensed by a higher clause negation (Japanese (i) below), if the n-word undergoes overt movement to the higher clause such sentences become fully acceptable (ii). This patterning, as with Iraqi Arabic and Hindi, ultimately leads to the same conclusions that neg-feature checking in Japanese/Korean may take place non-Spec–Head-locally when an n-word occurs in the immediate tense domain of Neg0 (see Simpson, to appear, for further discussion): (i) *Taroo-ga [Hanako-ga nanimo taberu to] iwanakatta Taroo-NOM [Hanako-NOM nothing eat that say-NEG-PAST Intended: ‘Taroo didn’t say that Hanako eats anything.’ (ii) nanimoi Taroo-ga [Hanako-ga ti taberu to] iwanakatta nothing Taroo-NOM [Hanako-NOM eat that say-NEG-PAST ‘Taroo didn’t say that Hanako eats anything.’
90
W h-movement and the theory of feature-checking
(38)
Zhangsan yiwei ta xiang-zhidao shei xihuan shenme. Zhang think he wonder who like what ‘Zhang thought he was wondering who liked what.’ NOT: ‘Zhang thought he was wondering who anything/something.’
liked
Although it might be suggested that LF movement of the indefinite wh NP out of the wh-island to its potential licensor in the higher clause would result in a subjacency violation and hence that the lack of an indefinite ‘anything/something’ interpretation is due to a simple movement violation, this would not in fact seem to be an appropriate way to capture the observed localit y effect. If indefinite wh expressions were to raise to their licensors, then movement of a wh-expression to the +Q Comp of a yes/no question (as in 36) might lead to the incorrect prediction that they be licensed as wh-interrogative by the +Q Comp rather than wh-indefinite.17 Consequently wh-islands must again be admitted as constraining both movement and (certain) non-movement dependencies. In sum then it is found that there are a variet y of constructions which may appear to offer even strong motivation for assuming a covert movement analysis on the grounds of localit y phenomena, but where such a movement analysis is ultimately not possible due to the presence of other negative evidence. It therefore must be accepted that there do exist certain localit y conditions which apply both to applications of movement and to licensing relations which are not established via movement. The cases briefly reviewed above also indicate that in many instances there is indeed a valid distinction between claiming that a dependency is mentally represented via co-indexation/binding as opposed to being the result of covert movement, and hence that co-indexation does not just instantiate a formal purely notational equivalent to an analysis of covert movement, a charge often levelled against such ways of representing scope/dependencies.18
17
It is also unclear what morphological feature carried by verbs such as yiwei could trigger raising of indefinite wh NPs. Instead this seems to be a licensing relation similar to that with NPIs.
18
In certain respects, the approach to feature-checking dependencies in Iraqi Arabic and Hindi suggested here does indeed liken them to (non-movement) analyses of the licensing of NPIs, anaphora (Progovac 1991a; Manzini & Wexler 1987) and wh-indefinites in Chinese (Li 1992), in the sense that while an element to be licensed must indeed occur in some local relation to its licensor, this localit y need not be that of a Spec–Head configuration and simple occurrence in any position within the relevant localit y/domain will indeed be sufficient for licensing. Featurechecking may however still be taken to be essentially distinct from these more general instances of licensing in virtue of four important properties. First, feature-checking operations invariably involve the matching of a parallel feature specification shared by both licensor and licensee; this
W h-movement and feature-checking
91
2.3.2. Mahajan (1990) Before continuing on to examine some of the consequences which the data in Iraqi Arabic and Hindi have for wh-questions and wh-feature-checking in other languages, I will briefly pause to consider certain potential alternative analyses of the basic paradigm. The first of these is Mahajan’s (1990) original LF-raising treatment of the wh-patterning in Hindi in the Government and Binding framework; this will be followed in Section 2.3.3 with an examination of how two more recent approaches in models without a separate syntactic level of LF might be able to account for the data, Brody’s (1995) Lexico-Logical Form and O’Neil & Groat (1995). Writing prior to the advent of the Minimalist Program, Mahajan (1990) suggests that all wh-phrases in Hindi must appear in a position governed by a +Q Comp by LF, so that all those wh-phrases not in such a position at SpellOut/S-structure will have to raise to Comp at LF. Mahajan attempts to account for the tense restrictions on the distribution of wh-phrases in situ in Hindi by first proposing that all LF wh-raising (cross-linguistically) is actually QR, and then relating this to a further proposal that tensed CPs in Hindi are extraposed during the course of a derivation resulting in such constituents becoming islands for LF extraction. It is suggested that where wh-phrases occur licitly raised out of tensed complement clauses as in (20), the wh-phrase will be extracted from the lower CP before the CP undergoes extraposition, at a point when it is L-marked by V0 and does not constitute any barrier for movement. Once extraposed however, the CP will no longer be L-marked and will block extraction. At LF a wh-phrase in situ in such a tensed CP is argued to first undergo QR to the IP-node which dominates it. This triggers further QR of the embedded IP clause to the matrix IP-node and a position where the (IP-adjoined) wh-phrase could technically be governed by the +Q Comp. In this sequence of QR operations, the movement of the lower IP to
is not the case in NPI or wh-indefinite licensing, where licensing elements may be varied and numerous in t ype and may not seem to share any obvious common propert y with their licensees (as indeed pointed out in the text). Second, while feature-checking heads would seem to require the presence of an element with matching features to license, the licensors of anaphors, NPIs and wh-indefinites do not require the necessary presence of any of the latter. Thirdly, feature-checking requirements may trigger movement of an element to its licensing domain, whereas wh-indefinites and true NPIs would not seem to have this abilit y to raise to an appropriate domain for licensing. Finally, morphological features may in certain instances critically require licensing “mid-way” in a derivation and prior to Spell-Out/PF, whereas elements such as NPIs, anaphors and wh-indefinites would only ever seem to need to be in a licensing configuration actually at LF and not necessarily before this point.
92
W h-movement and the theory of feature-checking
the matrix IP will be blocked from crossing the extraposition island, and so sentences such as (19) will be ill-formed. There are a variet y of reasons why such an account based on extraposition and LF wh-movement as QR would not seem to be attractive however. First of all there is the purely theoretical question of why it is that LF wh-movement should have to be QR. If wh-phrases may undergo unbounded long-movement prior to SpellOut this should also be available as an option after Spell-Out (if movement can indeed be motivated at this point). If the derivation and the computational principles affecting structure (Merge and Move) are uniform from lexical insertion through to LF, there seems to be no way to claim that movement may proceed in one way prior to Spell-Out but not in the same way at LF; the suggestion that all post-Spell-Out raising is QR also effectively recognizes S-structure as a significant derivational point/level relative to which certain syntactic phenomena may be stated (i.e. movement would be constrained to occur in different ways before and after this point). A second general point is that the QR/extraposition account does not extend cross-linguistically as it is predicted to. Mahajan suggests that wh-phrases may occur in situ in relative clauses in Chinese and Japanese but not in Hindi because in the former languages the CP in the relative clause is not extraposed from the N/D-head, whereas it is in Hindi (hence resulting in a barrier to LF-extraction). However, such an extraposition account predicts that wh-phrases should be possible in situ in tensed clauses in Iraqi Arabic as the CP will not be extraposed, yet wh-phrases are not permitted to occur in such environments. Mahajan also seeks to use the extraposition hypothesis to explain the fact that wh-phrases in situ in wh-islands in Chinese but not in Hindi may have scope higher than the wh-island, arguing that the CP wh-island in Chinese is not extraposed, therefore L-marked by V0 and hence not a barrier for LF raising of the IP to which the wh-phrase has been QR-ed. However, this incorrectly predicts that any wh element in a wh-island should be able to achieve higher scope outside the wh-island in Chinese whereas it has been seen that wh-adjuncts such weishenme/ zenme ‘why/how’ may not. If L-marking and extraposition really were the critical factors involved, it is also predicted that wh-phrases in situ in V-initial Iraqi Arabic should be free to take scope higher than a containing (non-finite) wh-island, which is not the case (see example 7). The success of Mahajan’s QR account also depends on a QR-ed IP being adjunct-like for purposes of extraction. Extraction of an argument from an extraposed CP will only result in a subjacency strength violation, as Mahajan shows with other data. However wh-phrases in situ in extraposed CPs are completely unacceptable, paralleling the ungrammaticality resulting from extraction of adjuncts from extraposed CPs. The ECP as a Rizzi/Cinque-t ype condition on
W h-movement and feature-checking
93
extraction-sites will not be violated by QR raising of the embedded IP because the IP will be properly-head-governed by C0 (which may be lexical, and which must be assumed to properly-head-govern IP in grammatical cases of QR IP-raising from non-extraposed non-finite CPs). Therefore ungrammaticalit y must result from failure of the IP to create a Government Chain through to its trace, this being blocked by the weak islandhood of the extraposed CP. However, if IP-raising is indeed sensitive to weak islands then it should not be possible for such constituents to QR out of wh-islands in languages like Chinese and English; establishment of a Government Chain between the raised IP and its trace within the wh-island CP should similarly be blocked by Relativized Minimalit y. If an account in terms of QR and extraposition therefore faces a number of technical and theoretical difficulties, a “bare” extraposition account (i.e. not taking LF wh-movement to be QR) will also not work here either. That is, even if one follows Mahajan’s essentially derivational approach to the critical sequence of extraction and (claimed) extraposition but abandons the idea that post-Spell-Out wh-movement must be different from that occurring prior to Spell-Out, one still arrives at clear predictions which are not borne out by the data. As outlined briefly above, Mahajan suggests that licit overt wh-extraction from a tensed CP (as in 20) takes place when the CP is L-marked by V0 before it undergoes extraposition, whereas (hypothetical) LF wh-movement to Comp from within an extraposed CP will involve extraction from a non-L-marked constituent (hence the occurrence of wh-phrases in situ in such environments is ungrammatical). While the underlying simple logic of such an account might initially seem plausible, it will not however explain the fact that wh-phrases which are arguments are completely ungrammatical when they occur in situ in tensed CPs. Overt extraction of arguments from extraposed CPs results in subjacency-strength violations, as illustrated below in (39) and (40). In these two cases Mahajan argues that the presence of a question-marker kyaa in the matrix clause can be taken as indication that the CP is base-generated in its PF “extraposed” position and not moved there during the course of the derivation (see Mahajan for relevant details/motivation here). As a result the topic-phrase and wh-phrase could not have undergone extraction at any point when the CP was L-marked (compare (40) with other fully acceptable cases of wh-extraction from tensed CPs, assumed to have occurred prior to CP-extraposition, e.g. (20)):19
19
When example (40) is embedded under another higher clause containing a kyaa question particle, Mahajan actually marks it as quite grammatical:
94
W h-movement and the theory of feature-checking
(39)
(40)
??vah
kuttaai Raam-ne kyaa kahaa ki kis-ne ti dekhaa-thaa that dog Ram-ERG QP said that who-ERG saw ‘That dog, who did Ram say saw?’ ??kOn saa gem Raam-ne kyaa kahaa ki hamaarii tiim kis din ti i which game Ram-ERG QP said that our team which day khelegii will-play ‘Which game did Ram say our team would play on which day?’
Thus if extraction from an extraposed CP maximally results in a subjacencystrength violation for arguments (as is generally noted to be the case crosslinguistically, extraposed clauses constituting weak islands only) there is no explanation why the occurrence of wh-arguments in situ in such tensed CPs should be so totally unacceptable, as in (19) repeated here: (19)
*Raam-ne kahaa [ki kOn aayaa hE]? Ram-ERG said who has come ‘Who did Ram say has come?’
Therefore, although extraposed CPs might well appear to be weak islands for extraction (as in (39)–(40)), and one could in fact adopt Mahajan’s derivational approach in part to account for the fact that overt extraction in example (20) (repeated below) is better than in (40), movement in (20) taking place before extraposition creates any islandhood, the full unacceptabilit y of (19) cannot be due to any attempted LF raising of the wh-phrase as this should not result in ungrammaticalit y of this strength and no coherent-intelligible reading but rather be mildly ungrammatical and on a par with (40). Instead, it can be suggested as before that examples such as (19) are ill-formed for the simple reason that featurechecking of the wh-phrase has not taken place by Spell-Out, causing the structure to crash (completely):
(i)
Ravi-ne kyaa socaa ki [kOn saa gemi Raam-ne kyaa kahaa ki hamaarii-tiim Ravi-ERG QP thought that [which game Ram-ERG QP said that our-team which-day will-play kis-din ti khelegii]? ‘Which game did Ravi think Ram said our team will play on which day?’ It is not clear why (i) should be better than (40) and the judgements may be subtle here. Importantly though it indicates that extraction of wh-arguments from extraposed CPs is felt to be either simply acceptable (as in (i)) or somewhat degraded as in (40), but never fully unacceptable.
W h-movement and feature-checking (20)
95
kOni Raam-ne kahaa ki ti aayaa-hE who Ram-ERG said that has-come ‘Who did Ram say has come?’
2.3.3. Brody (1995), O’Neil & Groat (1995) Having seen that the Iraqi Arabic/Hindi paradigm is problematic for the set of Minimalist assumptions set out in Chomsky (1995) and also difficult to capture in more traditional GB approaches such as Mahajan’s QR/extraposition treatment, one might ask whether the critical data could possibly be captured in other non-standard Minimalist accounts without the need to admit that feature-checking need not necessarily be Spec–Head local. Here I will briefly examine two alternative approaches — Brody (1995) and O’Neil & Groat (1995) — and argue that the patterning of wh-phrases in Iraqi Arabic/Hindi is in fact equally as problematic for such frameworks as it is for Chomsky (1995), suggesting that the paradigm can indeed only be successfully handled in an analysis which does allow for the possibilit y of long-distance checking relations. In Brody (1995) it is proposed that chains composed of a contentive element and a co-indexed set of empt y category copies are formed pre-syntactically in the lexicon and inserted directly into LF, so that there is no movement and no syntactic derivation as such. The actual position of the phonetically spelt-out contentive member of the chain is said to be dictated by morphology, and will occur as high in the chain as allowed by the morphology of the language in question. Considering Iraqi Arabic (1) repeated below as (41) it would be assumed that an empt y category copy of the in situ wh-phrase occurs in SpecCP forming a chain with the lower contentive (possibly with other ec links in additional intermediate checking positions not represented here): (41)
eci Mona shaafat menoi? eci Mona saw whom ‘Who did Mona see?’
Examples (4) and (3) should result in the representations in (42) and (43) below with chains linking the +Q Comp to the theta-position of the wh-phrase in the lower clause: (42)
shenoi tsawwarit Mona [eci2 Ali ishtara eci1]? what thought Mona Ali bought ‘What did Mona think Ali bought?’
96 (43)
W h-movement and the theory of feature-checking *eci2 Mona tsawwarat [eci1 Ali istara shenoi] eci2 Mona thought Ali bought what
Ignoring the problem that the wh-contentive (legitimately) occurs as the lowest link of the chain in (41) while the morphology would appear to allow it to occur higher (as in (42), the grammaticalit y of (42) must be taken to indicate that it is indeed possible to form a wh-chain with links crossing tensed clauses. This being so, the unacceptabilit y of (43) is fully unexpected — if (41) shows that a wh-contentive may remain in a lower in situ position within a wh-chain, then it should be possible to form the chain <eci2, eci1, shenoi> with link <eci2, eci1> crossing a tensed CP in (43), just as the link <shenoi , eci2> does so in (42). The essential difficult y in this purely representational system with no movement component is that all the links of a chain result from the same form-chain operation irrespective of where the actual contentive appears, and there is consequently no means to distinguish chain links above the contentive from those occurring below it (with the result that all links should be subject to the same localit y restrictions). The approach put forward here however suggests that there are in fact two discrete notions of localit y relevant in such cases — one which constrains movement and allows the wh-phrase in (42) to raise out of a tensed CP, and a second notion of localit y constraining wh-checking relations restricting the licensing relation between a +Q Comp and a wh-phrase to positions within the same tensed domain, thus allowing (42) and (41) but ruling out (43). While a (purely) representational approach of the kind suggested in Brody (1995) thus does not seem able to capture the patterning in (41)–(43) and some kind of derivational account appears necessary, a second alternative Minimalist approach outlined in O’Neil & Groat (1995) (which is indeed derivational in nature) also fails to predict and allow for the contrasts in (41)–(43). In O’Neil & Groat (1995) it is proposed that all feature-checking is carried out prior to SpellOut, via the movement of an element through all its relevant checking positions. Which particular chain-internal link is phonetically interpreted at PF is then taken to be a function of feature-strength, the highest position in the chain associated with strong features always being selected for phonetic interpretation, or otherwise the lowest link in the case that no strong features have been checked in the chain. Considering (41) above, the account would assume that movement has occurred to the +Q Comp and that the wh-phrase is simply spelt-out in the lowest chain position. Given that (42) indicates movement of a wh-phrase from an embedded tensed CP to a higher clause is indeed permitted, examples such as (43) are then incorrectly predicted to be acceptable, with phonetically empt y links of a movement chain occupying the positions of eci2 and eci1. The problem here again is
W h-movement and feature-checking
97
that licensing/checking positions both above and below a contentive element are conceived of as comprising a uniform chain created by a single operation t ype, in this case movement; the various links of such chains are therefore expected to be essentially identical in nature and critically conform to the same localit y constraints. As this does not appear to be true, there is clear reason to reject any uniform-chain-based approach and instead adopt an account which distinguishes two qualitatively different linking relations. What seems to be required is indeed both the derivational notion of movement with an associated set of localit y restrictions constraining the linking of an element to a lower trace position, and a second notion of localit y relevant where feature-checking dependencies are not forced to be Spec–Head-local. How this idea of potentially non-local featurechecking may apply to wh-relations in languages other than Iraqi Arabic and Hindi and what degree of variation in checking localit y might seem to occur across languages will now be given examination in sections 2.4 and 2.5.
2.4. Multiple-wh wh questions in English Having shown how the patterning of wh-elements in Iraqi Arabic and Hindi would seem to force serious revisions on the Minimalist idea of an invariable and universal checking-localit y, we now turn to the phenomena of multiple-wh questions in English and examine the problems for checking theory created by other languages which exhibit mixed raising and in situ patterns. In English (and other similar languages) the fact that a single “primary” wh-phrase is forced to undergo overt raising to Comp while all other “secondary” wh-phrases obligatorily remain in situ will be shown to lead to the same conclusions concerning checking localit y as were previously made on the basis of Iraqi Arabic and Hindi, namely that the Checking Uniformit y Hypothesis is not a universal constraint and features may in fact be checked outside of the Spec–Head/Head-adjoined checking localit y standardly assumed. It will also (later) call for a re-assessment and reinterpretation of the motivations underlying wh-movement in English-t ype languages, this carried out in Section 2.4.2. Multiple-wh questions in English such as (43) below can be shown to be severely problematic for both the (1993) and (1995) Minimalist approaches: (44)
Whoi did John give ti what?
In the original (1993) Minimalist paper, the difficulties (to be outlined below) at first sight appear to stem from the proposal that movement may only take place
98
W h-movement and the theory of feature-checking
in order to satisfy properties of the element moved (the principle of Greed) interacting with the suggestion that morphological features of a certain strength must uniformly be checked by a specific point in the derivation. In Chomsky (1995) this principle of Greed is actually replaced by a different set of assumptions; however, it will be shown that the fundamental underlying problems created by English-type multiple-wh questions still remain on unsolved and call for essentially the same conclusions. In multiple-wh questions such (44), it is the significantly differing behaviour of primary and secondary wh-phrases which is critically in need of explanation., i.e. the patterning that a single primary wh-phrase is forced to raise to the +Q Comp, but all other secondary wh-phrases present must remain in situ. In Chomsky (1993) it is argued that movement takes place only for the benefit of the element which undergoes the movement and not to satisfy any needs of the element moved to — the principle of Greed: Move-a applies to an element a only if morphological properties of a itself are not otherwise satisfied. The operation cannot apply to a to enable some different element b to satisfy its properties. Last Resort, then, is always “self-serving”: benefiting other elements is not allowed. Alongside of Procrastinate, then, we have a principle of Greed: selfserving Last Resort. (1993, p. 47) Consequently, if overt wh-raising of a primary wh-phrase is necessary, this must be taken to be triggered by a need to satisfy direct requirements of the wh-phrase itself rather than any of the +Q Comp), and hence that wh-features present on the wh-phrase must be checked. As the movement of the primary wh-phrase has to take place prior to Spell-Out, it can only be concluded that the wh-features present on the wh-phrase triggering this movement must be strong. This conclusion now becomes problematic when secondary wh-phrases are taken into consideration. If it is assumed that all elements of a certain t ype (e.g. wh-phrases, DPs etc) carry the same sets of features in need of checking, it is expected that secondary wh-phrases such as what in (44) will also have a strong wh-feature specification. If all strong features must be checked prior to Spell-Out, it is then clearly expected that secondary wh-phrases should also be checked in the overt syntax. However, as feature-checking is claimed to be possible only in the specifier position of a checking head (or adjoined to such a head as in the case of e.g. verb-movement) it would seem that the wh-features on secondary in situ wh-phrases such as what in (44) will not be checked and deleted prior to Spell-Out. Multiple wh-questions of this sort should consequently always crash at PF. The fact that they are perfectly well-formed indicates that something in the proposed account cannot be correct.
W h-movement and feature-checking
99
The position which Chomsky seems to adopt in the first Minimalist paper (though it is not made fully explicit) is to assume that for any multiple wh-question there will only be a single set of wh-features present. If only a single wh-phrase carries such wh-features, then only one wh-phrase will be expected to undergo raising. Such an assumption however might seem to be somewhat implausible. Intuitively it might be rather odd to suggest that a morphological propert y which characterises and identifies a particular class of elements (notably wh-phrases here) is only present on a single member of that group when more than one of these is present in a single sentence. When pre-Spell-Out raising is observed to occur (in English) with subject but not object DPs the possibilit y is not entertained that only the former but not the latter have DP-features in need of checking. Furthermore there is critically strong evidence from a number of languages that wh-features are indeed present on all wh-phrases and that every wh-phrase has a checking requirement. This has already been shown for Iraqi Arabic and Hindi in Section 2.2–2.3, where it was noted that quite generally the raising of a wh-phrase can only be ascribed to requirements of the wh-phrase itself and not those of the +Q Comp, and that in multiple wh-questions all wh-phrases must appear in positions where they can be feature-checked by Spell-Out. In addition to Iraqi Arabic and Hindi there is also direct and indisputable evidence of a related kind from various other East European languages that all wh-phrases do indeed carry wh-features to be checked and that it is not just a single wh-phrase which comes specified for wh-features. As noted in Rudin (1988) in many languages of Eastern Europe all wh-phrases are observed to undergo obligatory overt raising: (45)
(46)
(47)
Koj kogo vizda? (Bulgarian) who whom sees ‘Who sees whom?’ Cine cu ce merge? (Romanian) who with what goes ‘Who goes by what (i.e. means of transportation)?’ Ko koga vidi? (Serbo-Croatian) who whom see ‘Who sees whom?’
If it can then be shown for various unrelated languages that all members of the set of wh-phrases present carry wh-features in need of checking, it may justifiably be concluded that this is indeed a general property of wh-phrases cross-linguistically, just as the observation of morphologically-overt DP-Case and DP-raising in a subset of languages has led to the assumption that DPs in all languages carry
100
W h-movement and the theory of feature-checking
Case-features to be checked by some functional head. This being the case, English t ype mixed raising and in situ languages do present a serious problem for Chomsky’s (1993) feature-checking account when combined with the principle of Greed (also even without the principle of Greed in fact, as will be shown below). If all wh-phrases carry wh-features and these features are uniformly of the same strength, then all wh-phrases should require checking by the same derivational point. The obligatory overt raising of the primary wh-phrase can only be interpreted as taking place to check wh-features on the wh-phrase itself (due to Greed) and these must consequently be assumed to be strong, as the economy principle of Procrastinate should otherwise force raising to be covert. Wh-features present on all wh-phrases present should therefore also all be strong and require checking prior to Spell-Out. Given the strict Spec–Head localit y argued to constrain feature-checking relations all wh-phrases should then be expected to undergo overt raising to Comp, but it is clear that they do not. It would therefore seem that one (minimally) either has to abandon the principle of Greed or modify the localit y conditions governing feature-checking (if other general claims on such feature-checking are to be maintained). As there is now independent evidence (presented in Section 2.2–2.3) already arguing for a different view of the localit y of feature-checking, it might seem that this is indeed the natural route to follow and suggest that feature-checking relations may indeed potentially occur within larger domains than previously imagined. Specifically here in English-t ype languages it can be suggested that all wh-phrases are in fact checked prior to Spell-Out, but that the checking relation is established between the +Q Comp and the wh-phrases in their in situ positions. This is obviously (again) a non-trivial conclusion raising many serious questions about movement and general constraints on the availabilit y of “non-local” feature-checking. One might perhaps wonder whether the principle of Greed should not be rejected instead given the problems that its introduction would seem to bring here. However, shortly it will be shown that simply abandoning the principle of Greed will actually not solve the underlying difficulties created by multiple wh-questions in English and that it is critically the localit y of checking relations which is need of reconsideration. In addition to this, and quite independent of the interaction of Greed and multiple wh-questions in English, there is indeed this converging evidence from other sources (i.e. Iraqi Arabic and Hindi) that the original account of the localit y of feature-checking is in need of modification. I will therefore take this as indication that a successful solution to the general problems raised by multiple-wh questions in English may in fact better be sought by developing the idea of feature-checking as a relation which can also be effected “at a distance” and that this is the aspect of the problem which requires some (significant) revision.
W h-movement and feature-checking
101
In Chomsky (1995) the principle of Greed is in fact abandoned and replaced with a different approach to the legitimisation of movement. Instead of movement of an XP/X0 being triggered by a need to check its own features, Chomsky (1995) suggests that XPs/X0s are in fact attracted by functional heads to check features of these functional heads. Another innovation is that features which are strong must be checked and eliminated immediately rather than simply before SpellOut/PF; this has as consequence that strong features can only be generated on functional heads — if they were generated on XPs it would often not be possible to check them immediately as the relevant checking functional head might not be built into the structure until later on in the derivation. Chomsky also introduces a significant ±interpretable distinction in feature t ypes. Interpretable features such as Case will always have to be checked at some point in the derivation (as they are by inherent nature uninterpretable), whereas +interpretable features will only ever require checking if strong (if they are weak and left unchecked they will not cause a derivation to crash). Wh as a +interpretable feature is consequently predicted to require checking only when strong and on C0, so wh-phrases will never be generated with strong wh-features. If their wh-features are consequently weak, because they are +interpretable they may remain unchecked throughout the derivation. The English pattern of multiple-wh questions therefore seems to be captured — a single wh-phrase will raise to check strong wh-features on the +Q C0 and all other secondary wh-phrases will remain in situ, not requiring any checking. However, on closer inspection, such a re-orientation of the basic system ultimately does not succeed in avoiding the fundamental problems of English multiple wh-questions once additional information about other languages is factored in. Although the set of assumptions will allow for the English wh-patterning in isolation from other languages, once one turns to languages with rather different basic patterns, it soon becomes clear that Greed in itself was not the essential problem. A cornerstone of the Chomsky (1995) account (as in Chomsky 1993 as well) is the assumption that wh-phrases have no inherent checking requirement, being classed as +interpretable. The switch to a system where strong features can only be generated on functional heads formally enforces the view that wh-phrases will never be found to have checking requirements of their own. However, empirically it has repeatedly been shown in sections 2.2 with Iraqi Arabic, in 2.3 with Hindi and here in 2.4 with a variet y of East European languages that wh-phrases most definitely do have feature-checking requirements of their own which must be satisfied or cause a derivation to crash. This hard evidence from a range of unrelated languages cannot simply be ignored, and it cannot be accounted for within the set of assumptions in Chomsky (1995). What the patterns observed would seem to clearly indicate is that wh is in fact a feature
102
W h-movement and the theory of feature-checking
which cross-linguistically requires checking on (all) wh-phrases, and so is actually similar in this to the interpretable Case features assumed to be carried by argument DPs across (all) languages. The conclusion is then that all wh-phrases in English will then have a need to be feature-checked just as in other languages such as Hindi, Romanian etc.20 Reflecting back now on the patterning of multiple-wh questions in English, if one now assumes that not only primary but also secondary wh-phrases require wh-feature-checking by a +Q Comp, and if it is assumed that feature-checking relations may only be effected within the strict localit y of Spec–Head/Headadjoined configurations (i.e. the standard Minimalist position), then all wh-phrases are expected to raise to the +Q Comp at some point in the derivation. As this obviously does not take place overtly in multiple wh-questions, it has to be assumed that this occurs covertly at LF. However, throughout Chapter 1 a whole range of evidence including binding theoretic facts, localit y phenomena, weak crossover, antecedent-contained deletion and other phenomena was presented indicating that such elements do not occur raised to the licensing +Q Comp at any point in the derivation. Taking this to be so, the important conclusion which automatically follows on from this is therefore essentially the same as made earlier regarding wh-feature-checking in Iraqi Arabic and Hindi — (secondary) wh-phrases in English will be feature-checked by a +Q Comp in their in situ positions, and so, as with wh-phrases in Iraqi Arabic and Hindi, not necessarily within the strict Spec–Head localit y argued for in Chomsky (1993–95) but potentially “longdistance” within significantly wider domains.21
20
As already noted once above, it should be emphasised again that the chain of reasoning here is entirely parallel to that used in other comparative work where overt evidence from one set of languages leads to conclusions about other language-t ypes. For example, on the basis of English and a variet y of other languages it is concluded that argument DPs in all languages carry Case features in need of checking, even if those languages manifest no overt morphological Case nor any obvious evidence of checking requirements.
21
Here both the chain of reasoning and the general ensuing analysis of a parametrised wh-checking domain (developed further below) closely ressemble Progovac’s (1991a) treatment of anaphora in Chinese. Noting that the simplex anaphor ziji ‘self’ may licitly occur in relative clauses relating to an antecedent exterior to the island, Progovac rejects any LF head-movement analysis of the anaphor to a higher Infl node (as e.g. in Cole, Hermon & Sung 1990) and instead suggests that anaphors of all t ypes are simply bound in their PF in situ positions. Differences in the localit y relating simplex and complex anaphors to their antecedents are then argued not to result from (LF) movement of the former, but from the binding domains of the former and the latter having different values. Here it will be suggested that differences in the distribution of wh-elements across languages relate to cross-linguistic variation in the localit y of wh-checking domains.
W h-movement and feature-checking
103
The replacement of Greed with various other assumptions then ultimately does not allow Chomsky (1995) to successfully explain the patterning in multiple-wh questions (in English), and instead it seems that what crucially must be changed is the assumption that feature-checking is universally confined to occurring solely within the strict localit y of Spec–Head/Head-adjoined configurations. Observing that wh-phrases in English appear able to occur in situ in any kind of environment, it can be suggested that the wh-checking domain for English is simply the (entire) sentence containing the +Q Comp, and therefore that a wh-phrase may be (wh-)feature-checked in any in situ position m-commanded by a +Q Comp, as noted in (48): (48)
The wh-checking domain in English includes all sentence-internal positions m-commanded by the +Q Comp. wh-checking domain value in English: the sentence
This contrasts with Hindi and Iraqi Arabic where the wh-checking domain corresponds to the immediate tense domain of a +Q Comp: (49)
wh-checking domain value in Iraqi Arabic/Hindi: the tense domain of the +Q Comp
2.4.1. The Triggering Hypothesis The conclusions in 2.4 above now raise an obvious question concerning motivation of overt wh-movement in English. If (secondary) wh-phrases in multiple wh-questions in English may be feature-checked in their in situ positions, we seem to be left without an explanation for why overt raising of a single primary wh-phrase must take place — i.e. if licensing in situ is potentially available, then why should any wh-phrase have to raise to Comp? One might perhaps attempt to suggest that overt raising of a primary wh-phrase does not in fact take place to check wh-features on the wh-phrase, but to check strong operator features on the +Q Comp, much as Chomsky (1995) proposes. However, if it is possible for a wh-checking relation to be established between a +Q Comp and a wh-phrase fully in situ to check wh-features on the latter (as argued here), there is no obvious reason why such a “long-distance” relation should not also allow reciprocal checking of wh-operator features on the former, so some other kind of explanation would seem to be required. A relevant description of the patterning observed may be said to be that once a single wh-phrase has been raised to a +Q Comp then all instances of wh-phrases are licensed, as in (51):
104 (50) (51)
W h-movement and the theory of feature-checking *Did John give what to who? What did John give to who?
Raising of the wh-phrase in (51) essentially makes the clause into a wh-question and requires that values for all/both wh-phrases in the CP be given in any answerform. I would now like to suggest that this movement is necessary in order to “trigger” C0 as an appropriate licensor for (all) wh-elements present in its scope, that C0 is critically ambiguous prior to wh-movement in ranging over a variet y of potential values — focus, +wh+Q, yes/no+Q etc — and that wh-movement into Spec of C0 will function to disambiguate C0, activating it as a licensor for (specifically) wh-t ype elements: (52)
The Triggering Hypothesis Wh-movement to an ambiguous C0-head is necessary to trigger C0 as a licensor for specifically wh-elements.
The specifier position raised to by the wh-phrase is thus posited to be that of a general polarit y-oriented head, much as suggested in Culicover (1992), hosting a variet y of elements other than wh-phrases: (53) (54) (55)
[That film]i I really didn’t like ti. [Not only Hastings]i will I banish ti , I shall also exile Lord Smythe. [So upset]i was she ti that she broke down and cried.
Prior to any raising, C0 may be taken to be crucially underdetermined with respect to its precise “polarit y” setting, so that movement of an element of a certain t ype into SpecCP will be necessary to disambiguate its licensing value. Once disambiguated and triggered in a particular way, the C0 will then be able to function as a licensor for all elements of the relevant t ype, whether such an element has been raised to its Spec or occurs in situ in its licensing domain. As only a single element (e.g. a single wh-phrase) need raise in order to effect such triggering of C0, other elements of the same t ype are indeed free to remain in situ for licensing (providing they occur within the licensing domain of C0).22
22
The position referred to here as C0 and SpecCP may in fact be that of a focus-polarit y-t ype head occurring after C0. Culicover notes that focus-movement of the t ype illustrated in (53)–(55) may occur in embedded clauses following an overt complementizer: (i) John claimed that [War and Peace]i he had never read ti. (ii) John said that [not only Bill]i had Sue deceived ti, she had also taken in Jo. Whether the position moved to is Spec of a post-CP Focus/Polarit y-phrase or SpecCP itself is in fact not really important here. What I do wish to claim, in the general spirit of Culicover, is that all such movement including wh-raising is to the same Spec position. For ease of exposition, I will
W h-movement and feature-checking
105
In a certain way the above proposal may reflect a general idea put forward in Cheng (1997) that wh-movement occurs to “t ype” a clause as +wh, though here the ultimate motivation for such movement is seen to be a formal morphological requirement on wh-phrases themselves that they be licensed prior to Spell-Out, rather than movement satisfying a constraint on CPs that they be identified as (+wh)-interrogative. Following on from this, adapting and making use of another suggestion in Cheng, it can be argued that in some languages C0 is not ambiguous in nature, or rather that there exists an alternative way to disambiguate it, via the direct insertion of question particles, such as ne in Mandarin Chinese. If C0 can be disambiguated and triggered in this way, then no raising of wh-elements need take place and all wh-phrases may remain and be licensed in situ (providing again they occur within the licensing domain of the C0). Thus again in English the suggestion is that raising of a single wh-phrase takes place to SpecCP in order to disambiguate and activate C0 as a licensor for all wh-elements in its domain. Secondary occurrences of wh-phrases may remain in situ and are not required to undergo such movement as raising of a single wh-phrase will suffice to trigger C0. In an apparently wh in situ language such as Iraqi Arabic where wh-movement is not always required in wh-questions, it can be assumed that C0 may be unambiguously +wh+Q due to the existence of a (possibly null) lexical instantiation of C0 which is +wh. This possible lexical choice of C0 will be selected in the numeration and base-generated in C0, serving to render it unambiguously +wh.23 However, movement of a wh-phrase may be forced to take place in order to bring the wh-phrase into a position where it can be licensed by the +wh C0-head — for example when licensing of a wh-phrase in situ is blocked by +finite tense in a lower CP. In this respect it should (importantly) be noted that “wh-movement” in languages like Iraqi Arabic and Hindi is actually unlike wh-movement in English and does nothing to alter the essential
continue to refer to this as SpecCP/Comp, but this is simply intended to identify a single functional projection which is the locus of wh, yes/no+Q and focus licensing, it actually may not be the position where embedding complementizers are inserted. In fact many languages clearly show that wh-movement does target a position which immediately follows and is distinct from that of overt embedding complementisers, see for example Hungarian (Chapter 3 here), Hindi, Indonesian etc. Furthermore Japanese, Burmese and various other languages show a Q-morpheme question particle occurring in a separate head-position lower than the head position occupied by embedding complementisers. For general discussion of the “splitting” of the complementiser functional domain into a variet y of discrete heads, see Rizzi (1995), Koopman (1996). 23
Otherwise put, languages such as Iraqi Arabic will have the equivalent of null wh-question particles.
106
W h-movement and the theory of feature-checking
licensing potential of a Comp; if two wh-phrases both illicitly occur in a lower opaque tensed CP in Iraqi Arabic/Hindi, movement of one of these to the matrix +Q Comp will not result in the second being licensed through some activation of Comp as a licensor — both wh-phrases must move into the licensing domain of the +Q Comp as seen in the contrast in Hindi (21)–(22). Wh-movement/ movement of wh-phrases then takes place for two distinct (though related) reasons: (56) a. b.
Motivations for wh-movement: wh Wh-movement may be necessary to trigger an ambiguous C0 as a licensor for +wh elements.24 Wh-movement may be necessary to bring a wh-phrase into the checking domain of an (unambiguous) +wh+Q Comp.
Given this analysis of wh-raising in English, further explanation is now required of languages such as Romanian and Bulgarian where all wh-phrases in multiple wh-questions undergo raising to Comp. Clearly as only one of these should need appear in the +Q Comp for triggering purposes if C0 is ambiguous, raising of the others would seem redundant, and therefore should not take place. Here I would like to propose that movement is actually forced to take place as a direct result of the localit y of the wh-checking domain being more restricted in these languages. In sections 2.2 -2.4.1 of the chapter the essential thrust of argumentation has been an attempt to establish that feature-checking relations are ultimately not subject to any universal Spec–Head/Head-adjoined localit y condition and to suggest that the checking domain of a functional head may in fact be subject to certain parametric variation across languages. This approach towards the localit y of feature-checking is in essence similar to Manzini & Wexler’s (1987) treatment of Binding Theory localit y where it is suggested that the local domain relevant for the licensing of anaphors may be subject to different parametric settings across
24
This kind of raising might again initially appear similar in certain ways to an idea in Cheng (1997) that elements in various Australian languages undergo raising to a +Q Comp to be licensed as wh-elements. As the relevant elements in these languages may have interpretations other than wh-question words, raising to the +Q Comp essentially functions to disambiguate them as +wh. Here however, the wh-phrases under discussion (in English, for example) are all unambiguously +wh in advance of any raising. The reason for raising to Comp is in order to activate and trigger a potential licensor for their wh-features and not to determine one particular interpretative option for the wh-phrases rather than another. The situation can be compared to Case features on a DP which may be quite unambiguously Case-features but nevertheless still necessitate licensing/checking. With wh in English, part of the process of successful checking requires the C0-head to be triggered in an appropriate way.
W h-movement and feature-checking
107
languages, and that an anaphor must simply occur c-commanded by its licensor somewhere within the relevant local domain in order to be successfully licensed. Here the proposal is that a wh-element must similarly occur in some parametrised local domain m-commanded by a potential licensor (a +wh-Comp) in order for its (wh)features to be checked. In one respect such suggestions are simply an extension of the general idea formalised in Chomsky (1995) that elements with features to be matched and checked against each other must co-occur locally; however, whereas Chomsky assumes a critical and invariant notion of localit y constraining all feature-checking relations, the present account suggests that the notion of localit y may be subject to variation and hence in certain instances correspond to broader domains than Chomsky’s strict checking domain. In the case of English, the wh-checking domain is suggested to be equivalent to the sentence as a unit, in Iraqi Arabic/Hindi it is the first tense domain immediately containing a +wh-Comp. Returning to Romanian and Bulgarian, it can now be suggested that the obligatory raising of all wh-phrases to Comp is indeed simply a reflection of a narrower parametric setting of the wh-checking domain in such languages, and that the wh-checking domain here in fact corresponds to Chomsky’s strict checking domain, basically SpecCP (and possibly C-adjoined positions). Consequently all wh-phrases will be forced to raise to the +Q Comp, either targeting multiple Specifier positions or possibly raising via some kind of “wh-cluster” movement to a single Spec of C0 (as suggested in Grewendorf 1999). Multiple wh-raising in Romanian/Bulgarian is then in essence suggested to be entirely parallel to instances in Iraqi Arabic and Hindi where a wh-phrase is forced to raise from a lower tensed clause to the tense domain of a +Q Comp — in both cases wh-movement must bring a wh-phrase into the relevant parametrised wh-checking domain of Comp. (57)
Wh-checking domain value in Romanian, Bulgarian etc: SpecCP of a +Q C0
Parametrisation of the wh-checking domain in this way can be argued to be determined by the choice or t ype of +wh+Q C0 available in a language, hence a lexical propert y relating to variation in the t ype of a particular functional head. A +wh+Q C0 in Iraqi Arabic will be a different lexical element from the +wh+Q C0 in English and this will be directly responsible for the differences in patterning observed among elements dependent on that head. The cross-linguistic variation suggested to exist here can consequently be seen as very much in line with current assumptions that variation in syntax is caused by simple differences in the lexical
108
W h-movement and the theory of feature-checking
t ypes instantiating functional heads.25 All +wh+Q C0-heads will necessarily be associated with some particular (licensing) localit y and will only be able to license (feature-check) a relevant element inside this localit y. For example, the localit y associated with the +wh+Q C0-t ype in Romanian is essentially just its Specifier position(s) and only wh-elements in such positions will be “visible” to the +wh+Q C0. In Iraqi Arabic, the licensing localit y of the +wh+Q C0 is the tense domain immediately containing it and only wh-phrases present inside this domain will be visible to the +wh+Q C0 for checking. This association of a certain local domain with a functional head can also be suggested to be seen in the cross-linguistic variation found with NPI licensing by a Neg0-head — in various languages Neg0 can only license NPIs in a very local domain and so will not “see” NPIs outside this localit y (e.g. in CNPs and lower tensed clauses), whereas in other languages the Neg0 may quite possibly be able to license into wider domains. The functional head Neg0 then, like C0, will also be specified for a licensing localit y which is subject to certain cross-linguistic variation. Furthermore, if the variation attested here is indeed due to lexical choice and instantiation of the +wh+Q C0, one might expect to find not just variation across languages in the behaviour of wh-elements relative to the +wh+Q C0, but possibly also language-internal variation if a language had available more than a single potential choice for a +wh+Q C0. This is precisely what I will argue is the case in Partial Movement and wh-expletive structures in German, Hungarian, Iraqi Arabic and Hindi in Chapter 3. Selection of a different t ype of +wh+Q C0 will be suggested to have clear automatic consequences on the licensing localit y/wh-checking domain associated with the +Q Comp. If one now pauses to consider the degree of parametric variation in wh-checking domains which might be permitted, it would appear that only three basic t ypes have in fact been identified:
25
Note that in English-t ype languages where C0 is suggested to be ambiguous, it will still be the case that some lexical item with various intrinsic properties is inserted into C0. It is critically the polarit y orientation of the C0 element which is externally determined (from a highly restricted set of possibilities) and not any other possible attributes. Consequently the “open-polarit y” C0 available in the lexicon in English may indeed be positiviely specified for a particular licensing localit y value, and simply remain underspecified for the t ype of element its licensing potential will be directed towards (prior to disambiguation in the syntax). In this view one might then expect that all t ypes of licensing dependent on such a C0 should be associated with the same basic localit y (which would be part of the invariant lexical specification). Clear parallels in the patterning of multiple-focus sentences and multiple-wh questions discussed in the next section would in fact seem to bear this expectation out.
W h-movement and feature-checking (58)
Wh-checking domain t ypes: Type A: Romanian, Bulgarian etc: Type B: Iraqi Arabic, Hindi: Type C: English:
109
SpecCP26 the immediate tense domain of the +Q C0 the sentence27
What this indicates first of all is that there is not simply unlimited variation, but quite possibly just three regular variants which may be noted to re-occur across languages.28 Secondly it may be argued that this particular division is actually not unnatural and may indeed correspond to natural divides in the propositional/ operator structure of a sentence interacting with different stages of grammati-
26
It is possible that Italian might also be analysed as belonging to this group, allowing one to explain the fact that Italian does not seem to tolerate multiple-wh questions. If the wh-checking domain in Italian is limited to SpecCP and Italian requires pre-Spell-Out wh-checking but does not permit either multiple SpecCP positions or any wh-cluster-formation/-movement (unlike Romanian and Bulgarian), then only a single wh-phrase will be able to be licensed (featurechecked) by raising to SpecCP prior to Spell-Out. The occurrence of other wh-phrases in situ and unlicensed at Spell-Out/PF will consequently cause attempted multiple-wh questions to crash.
27
Garrett (1996) identifies himself as a speaker who does not tolerate secondary wh-phrases in situ in embedded tensed clauses in English (at least not with any pair-list readings). This might seem to indicate that there are also speakers of English who have the Iraqi Arabic/Hindi value for wh-checking domain, i.e. the tense domain immediately containing the +wh+Q C0. One then might expect that secondary wh-phrases illicitly occurring in situ in lower tensed CPs would raise up to the +Q Comp of the higher clause to be licensed, as seen in Hindi example (22). However, this will only be possible if multiple SpecCP positions may be projected in a language, as in Romanian but not in Italian (footnote 25). As English may be taken to be a language which does not allow for multiple specifiers of C0 or other heads, such raising will in fact not be possible as a rescue device, this paralleling the situation in Italian.
28
Rudin (1988) suggests that in Serbo-Croat, Czech and Polish a primary wh-phrase raises to SpecCP but other secondary wh-phrases move to adjoin to IP. If this is so, then it might seem to indicate that a fourth type of wh-patterning needs to be recognised distinct from Type A. However, even if secondary wh-phrases do indeed move to IP-adjoined positions rather than SpecCP, this is still fundamentally movement of Type A to a position outside the propositional core and into the operator-functional domain (as presently discussed in the text). Also it is actually quite possible that the evidence for Rudin’s suggestions might be given a different interpretation. The reason for asserting that secondary wh-phrases occur IP-adjoined rather than in SpecCP is primarily that elements such as parentheticals may optionally be placed between the first wh-phrase and those following it. Given that parentheticals are not really fully integrated into the syntactic structure of a sentence, this kind of placement of parentheticals could quite plausibly be explained in terms of PF and prosodic factors, allowing one to maintain that all wh-phrases are in fact in SpecCP (i.e. one could suggest that PF rules allow the insertion of a parenthetical element in second position, even if this interrupts a sequence of wh-phrases in SpecCP, similar to the breaking-up of syntactic constituents by P2 clitics elsewhere in Serbo-Croat).
110
W h-movement and the theory of feature-checking
calisation of propositional attitude verbs into sentential structure. In languages of Type A, wh-elements will simply not be visible to the logical +Q operator if they remain down in the propositional part of the sentence and so are forced to occur up in the operator domain (and SpecCP). In Type B languages, the first tense domain associated with the logical +Q operator position is the propositional core which is open and visible to the the +Q C0 and wh-elements must occur in this domain in order to be correctly interpreted by the interrogative operator (consequently Type B languages differ from Type A languages in the abilit y of the +Q operator to actually see into the propositional structure; in Type A languages the +Q operator will only be able to locate wh-elements in the operator domain external to the propositional structure). In languages of Type C, it can be suggested that propositional attitude verbs have in some sense become integrated into the general functional super-structure dominating a proposition so that they and the tense which they carry are effectively invisible to the +Q operator, and the +Q operator can accordingly license wh-elements in lower embedded tensed CPs. Otherwise put, higher clause declaratives such as: “John said that…” will not be taken as discrete propositions/events but more as operators on a lower proposition. This more advanced grammaticalised status can then be suggested to allow the +Q operator to “look down” further into the lower proposition and license wh-elements there. If such an underlying intuition can be maintained, the basic parameter of variation would then reduce to whether the logical operator Q is able to see into the propositional structure or not, and how that propositional structure is defined — i.e. whether higher clause declaratives are grammaticalised into the structure or not.29 The suggestion made here that the checking domain of a certain t ype of head may be subject to variation across languages, or perhaps that certain localit y constraints have for effect that the checking domain of a head is reduced and (perhaps severely) restricted in one language but not another may seem to go
29
What is expressed here is an attempt to provide some very general understanding of what might underlie the variation found across wh-systems. In technical terms however, we have talked of purely formal mid-derivational licensing of wh-features and so direct reference to notions such as “interpretation” by a logical operator should not really be permitted. Nevertheless, it is not unreasonable to assume that all formal licensing requirements do indeed originate as the grammaticalisation of some aspect of informational or interpretative structure, and so the speculations here can be suggested to refer to possibly earlier interpretative distinctions which have now formally manifested themselves in the feature-checking mechanism. Alternatively, if one adopts the position that Spell-Out forms really are syntactically isomorphic with the input forms to interpretation (i.e. LF), as will be suggested towards the end of this chapter, then the suggestions made here may in fact legitimately make direct reference to aspects of interpretation.
W h-movement and feature-checking
111
against the endeavour to see all syntactic relations as being very local, yet nothing in principal excludes feature-checking from taking place within larger domains than those proposed in Chomsky (1995). Other dependencies in which one element requires licensing by another are not restricted to occurring within the localit y of the checking-domain as defined in Chomsky (1995), e.g. the licensing of NPIs, so there must exist other “non-local” relations within language. Similarly movement (or attraction, as in Chomsky 1995) may also clearly target and “see” a landing-site/element in some potentially non-local part of syntactic structure. In addition to this, feature-checking as outlined in Chomsky (1993) may also occur in either of two position-t ypes — in a Spec–Head configuration or where a head is adjoined to another head, so a certain difference of localit y is admitted even here (by necessit y), a Head-adjoined position arguably being “more local” to the checking head than its Spec position. Furthermore it might seem somewhat odd that if feature-checking may generally occur within the checking-domain of a head, checking is taken not to be possible in the complement-of-head position (complement positions being within the checking domain of a head as defined in Chomsky 1993). In this case, the rationale for excluding the complement-of-head from being a possible checking site in Chomsky (1993) was the attempt to see a significant uniformit y in the licensing of all maximal projections, and the suggestion that both subject and object DPs (and indeed all XPs with features to be licensed) should be checked in the same position-t ype, the specifier of a functional-head. However, in Chomsky (1995) with the switch to LF ff-movement, this uniformit y argument is lost as objects and other XPs which are featurechecked at LF will not raise to specifier positions; instead their formal features are argued to raise directly to the checking functional head. Consequently it is no longer really clear why one should want to exclude the complement-of-head from being a checking position (i.e. the uniformit y argument no longer holds). In Chomsky (1995) it is suggested that argument DPs must always move to be feature-checked and cannot be base-generated in feature-checking positions (hence the complement-of-verb position cannot be a position where features can be checked on an object), yet such a suggestion would not seem to have any obvious motivation or explanation. Therefore it may be argued that checking positions and checking localit y are simply defined to capture what are taken to be relevant empirical patterns (and certain local positions such as complement-of-head may be excluded as checking positions for no particularly clear reasons). If this is so, and the definition of what may constitute a checking position and a checking domain is ultimately driven by a basic need to capture observed empirical patterns, then the wh-paradigms presented in this chapter taken together with arguments firmly based on Minimalist assumptions concerning the relation of movement to feature-
112
W h-movement and the theory of feature-checking
checking would indeed seem to indicate that the strong position put forward in Chomsky (1995) is in need of critical modification, and that feature-checking relations must somehow be possible within wider domains than originally suggested. The table in (59) below now provides a brief summary of what is suggested to motivate wh-movement in the various language-t ypes under consideration, along with the values of their respective wh-checking domains: (59)
W h-movement and wh-checking-domains wh
Language types:
Reason for wh-movement
wh-checking domain value
English
to trigger C0 as +wh
the sentence
Iraqi Arabic, Hindi
so that all wh-phrases occur in the wh-checking domain of the +wh+Q C0
the immediate tense domain containing the +wh+Q C0
Romanian, Bulgarian Italian (?)
so that all wh-phrases occur in the wh-checking domain of the +wh+Q C0
SpecCP
2.5. Extensions Having presented and developed an approach to wh-checking and the crosslinguistic motivations for movement in sections 2.2–2.4 above, I will now consider various possible extensions of the basic account in the areas of focus, English full wh in situ, wh-islands and finally Japanese wh-scrambling (in Section 2.6).
2.5.1. Focus This first section examines focus and suggests that if focused DPs carry a focusfeature in need of checking/licensing by C0 (as in fact proposed in Chomsky 1993) then conclusions similar to those drawn with regard to wh-licensing, triggering and the locality of feature-checking can (and perhaps must) be arrived at. If the position is adopted that all syntactic movement must be triggered by a need to check morphological features, in cases of focus movement it can therefore be assumed that the focused XP is base-generated with focus-features which require checking by C0. Because raising in (60) is seen to be overt it must also be
W h-movement and feature-checking
113
assumed that checking of focus-features must occur prior to Spell-Out (just as with wh-features): (60)
THAT BOOK I never read.
Alongside single focus sentences such as (60), one also finds sentences containing multiple focused DPs as in (61): (61)
THAT BOOK I gave to JOHN.
If it is assumed that the DP JOHN also carries focus-features, focus-features being common to the class of elements which are focused (similar to wh in multiple-wh questions) and focus-features must be checked prior to Spell-Out, then just as with multiple wh-questions one is led to the conclusion that this second focused DP must be feature-checked in its in situ position and not in a Spec–Head configuration with C0. Again if such “checking at a distance” is possible we are led to ask why the first focused DP raises to SpecCP (as it might be expected that it too could be checked in an in situ position) and once more it can be concluded that this is to disambiguate C0 as a licensor for all focused elements in its domain. It may also be argued that in multiple focus sentences such as (61) a t ype of “paired interpretation” arises similar to absorption in multiple wh-questions — in (61) it seems to be heavily implied that other books were given to other people, and one might naturally expect a follow-on to (61) something like: THIS BOOK I gave to MARY; if pair-list interpretations here result from a similar kind of association with a licensing functional head as with multiple wh-questions, this may be taken as indication that both foci are licensed relative to the same C0. It can further be noted that the occurrence of a focused DP in a wh-question is quite unnatural (unless as a rhetorical question): (62)
??Who
saw JUR ASSIC PARK?
Such restrictions have also been observed for Italian (i.e. that focused and wh-phrases may not co-occur (see Cinque 1990)). This might seem rather unexpected given that single in situ focused DPs are possible in English (to which we return shortly). However, if both wh and focus elements require licensing by a (single) C0 which may be disambiguated and triggered as either +wh or +(pure) focus, then it is clear that in (62) one of the two elements will not be successfully licensed (presumably the focused DP) — a single C0 cannot simultaneously be triggered in two ways (this would then also seem to constitute evidence for the assumption that not only raised but also in situ focused DPs are licensed by C0 and therefore that in situ and raised focus are not unrelated and distinct phenomena). Against the objection that there are well-formed occurrences of a single focused
114
W h-movement and the theory of feature-checking
DP in situ (and hence that such in situ occurrences do not carry a focus-feature in need of checking), I would like to suggest that in English a null focus equivalent to the wh-question particles found in other languages such as Japanese, Chinese etc may optionally occur in the initial numeration of a focus sentence: (63)
[FØ] Mary said that JOHN did it.
Such a null particle like question particles in wh in situ languages will disambiguate and trigger C0 as +focus, and so licence the occurrence of the focused DP in situ, hence no raising need take place. Because no movement to trigger C0 is required, the focused DP may occur in islands just as secondary wh-phrases may in multiple wh-questions.30 Multiple focus sentences also seem to show superiorit y effects similar to those in multiple wh-questions: (64) (65)
PAUL hid THE BOOK. BOOKi PAUL hid ti.
??/*THE
This again is indication that in situ focused DPs are licensed by the same C0 that fronted focus-phrases are — if in situ focus were to be independent of focusmovement and licensed in an essentially different way, one would not expect to find any such interaction with moved/fronted focus-phrases. Superiorit y effects may well be explained in terms of the notion of Shortest Move as Chomsky (1995) and Reinhart (1998) suggest (see below), basically that overt movement to SpecCP must take place in the most economical (here shortest) way, and so from the subject rather than object position. This implies that movement of the in situ focused subject in (65) must be considered as a potential option to movement of the object, hence that in situ focusing is not independent of focusing of other elements via movement to SpecCP.
30
Some occurrences of in situ focus appear in fact to be a little odd, especially when not embedded in a context: ?Mary saw JOHN. (i) If such examples are acceptable, the listener must build a context and take the focused DP as contrastive with some other focused DP present in the discourse, e.g: (ii) When we went to Newport it was BILL I saw. Mary saw JOHN. Perhaps in these cases a null Focus particle is licensed in C0 by the presence of the preceding discourse and the focused DP (BILL), much in the way that Huang (1989) suggests that a “zerotopic” operator is licensed in discourse by a salient DP, this zero-topic then licensing occurrences of object pro in Chinese. If instances of in situ focus such as (63) above are better, where the focused DP occurs embedded under an propositional attitude verb or a verb of reporting, then perhaps this predicate is responsible for licensing the null focus particle/operator.
W h-movement and feature-checking
115
So, if all focused elements must be licensed by a single C0 and overt raising is observed to take place alongside licit occurrences of in situ focus phrases, it has to be concluded that the focus-features on the latter are licensed/checked in situ and not necessarily in SpecCP. As with wh-licensing one may also notice certain parallels relating to the localit y of the licensing domain — in English an in situ focused DP may occur in all kinds of islands just as in situ wh-phrases may, whereas in other languages the licensing of focus may be more restricted — in Serbo-Croat one finds instances of multiply-fronted focus-phrases just as all wh-phrases are seen to undergo obligatory raising for reasons of localit y on their licensing, and in Italian where licensing of wh-phrases is restricted to occurring only in SpecCP (not even adjoined to SpecCP) with the result that only a single wh-phrase may occur per +wh+Q C 0, it is found that multiple focus sentences are also not permitted.31
2.5.2. English full wh in situ The suggestion that movement of wh-phrases in languages such as English takes place in order to trigger and disambiguate C0 might seem to be questioned by the existence of certain st ylistic question forms. In British English (at least) instances
31
The occurrence of multiply-fronted focus-phrases in languages such as Serbo-Croat also indicates that focus is a feature which is present and in need of checking on all DPs which carry it, and that this must uniformly be checked on all focused-DPs by Spell-Out (also that focus-raising cannot be argued to take place solely in order to check any strong focus-features on C0, if this were to be the case then one would not find any multiple focus-raising). If one then (reasonably) concludes that focus is cross-linguistically a feature which always requires checking on (all the) XPs which carry it, despite being +interpretable, one is led to the conclusion that in situ focused DPs such as THE BOOK in (64) are checked non-locally and in situ. Overt raising of the focused DP PAUL indicates that C0 must be triggered as a licensor for focused elements by Spell-Out, therefore focusfeature-checking on all DPs must be assumed to be necessary by this point; as THE BOOK does not however occur in SpecCP at Spell-Out, its checking must be non-local. The fact that focused DPs may freely occur in extraction-islands together with binding theoretic evidence similar to that constructed with in situ wh-phrases also shows that in situ focused DPs can at no derivational point be taken to occur raised in SpecCP, and hence that their checking must be non-local, i.e. in short the same conclusions drawn from wh-data can also be made on the basis of the patterning of focus-phrases.
116
W h-movement and the theory of feature-checking
of wh-questions with no wh-raising are often attested (non-echo, also non-d-linked here):32 (66) (67) (68)
So having arrived there, you did what exactly? (And after she entered the room,) you believe that Mary took what (exactly)? (So, when you reveal all this,) you think that Mary will do what? Please tell me.
If this is the case, one must ask how the wh-phrase can be licensed, as it has been explicitly argued that a C0 needs to be triggered in order to licence wh-elements in English. I would like to suggest that similar to the case of single in situ focus in English outlined above, under certain situations a null equivalent to the +wh question particle in languages like Japanese/Chinese is licensed here. Such a null particle, licensed perhaps by factors of st yle/tone, will perform the same functions as wh-raising, triggering/disambiguating C0 as a licensor for wh-elements so that the relevant wh-phrases may remain in situ and need not undergo any raising. Given that the wh-phrase in the British English case is not necessarily d-linked, it cannot be argued that such a wh-phrase escapes the raising requirement because it is somehow “non-quantificational” (as generally suggested in Pesetsky 1987); rather it would seem to be a purely formal licensing requirement which is fulfilled and provided in some other way in these instances. One might then wonder why wh-raising is always required in embedded questions in English and French: (69)
*John wondered Mary saw who.
It might be expected that question-embedding predicates such as wonder/ask would select and licence just such a null question particle in the C0 of their clausal complement, so that wh-raising would not be necessary. However, note that although these verbs do indeed select questions, there still remains the ambiguit y as to whether such interrogative clauses are wh- or yes/no questions, thus the C0 is still ambiguous. Wh-raising to SpecCP always takes place to disambiguate a C0, even if the C0 does appear to be unambiguously +Q due to auxiliary I0-to-C0-movement: (70) (71)
32
*Didk John tk see who? Whoi didk John tk see ti?
This is indeed quite dissimilar from the cases of optional wh in situ in Polish when a wh-phrase is understood as being d-linked (reported in Pesetsky 1987). Here there is no necessary d-linking interpretation at all.
W h-movement and feature-checking
117
Therefore raising of a wh-phrase will still be necessary even when a +Q Comp is selected by a higher verb. However, we might note that the C0 of the clause selected by a verb such as wonder has been disambiguated to the extent that it is +interrogative and the C0 is +Q; therefore no subject-auxiliary inversion is required to take place — selecting a +Q C0, this satisfies/automatically checks the any X0 +interrogative features on C0 itself. Where no such (partial) disambiguation and selection of a t ype of C0 occurs, as in matrix questions, the +interrogative X0 features on C0 need to be checked via raising of a (verbal) head to C0; along with Aoun & Li (1993) it may be assumed that in these cases the +interrogative X0 features are base-generated in Infl and raise with the verb to Comp. Substantiating this general approach is the case of So-/Neg-inversion, as in: (72) (73) (75)
So tired was John that he fell asleep in class. Never had John seen such a sight before. Mary said that never had John seen such a sight.
The verb say does not select any t ype of +focus C0, so auxiliary-raising will always be required to check (So-adj/Neg)-X0-focus features on the C0 whether in matrix or embedded contexts. Generally then, certain apparent counter-examples to the suggestion that wh-raising takes place in order to trigger a C0 can be explained if a null wh-question particle is optionally licensed to appear in the numeration under certain special circumstances (i.e. English would have a second +wh+Q C0 equivalent to the Chinese/Burmese +wh+Q C0 available in the lexicon for use in certain contexts).
2.5.3. W h-island violations; superiority Two other wh-related cases can also be re-considered now given assumptions and claims which have been made regarding wh-licensing and the motivations for wh-raising. The first is that of wh-island “violations”, where a wh-phrase moves from within an embedded question to the +Q Comp of a higher clause and this is not so unacceptable when the lower clause is untensed, e.g: (76)
?Which
booksi did he want to know where to put ti?
If wh-movement ultimately takes place in order that a wh-phrase may be licensed, and is not triggered by any hypothetical requirements of a C0 itself, one might predict that there should be no wh-island violation cases such as (76). The wh-phrase could be successfully licensed by the lower +wh+Q Comp (triggered as
118
W h-movement and the theory of feature-checking
+wh by raising of where) and so could stay in situ in the lower CP.33 Sentences such as (76) can however be accounted for if the need for a particular interpretation of a wh-phrase may necessitate movement. If the wh-phrase which books remains in the embedded CP it will only be interpreted as being indirectly questioned with scope delimited to the lower CP; in order for it to be interpreted as being directly questioned, some wh-phrase must occur in the matrix C0 to trigger it as a +wh+Q licensor. Therefore if such an interpretation is necessary, the wh-phrase which books must raise up to this Comp position. The fact that such raising is actually possible also allows one to conclude that a wh element does not in fact necessarily need to be licensed/feature-checked in the first position where such licensing/checking technically becomes available (i.e. its base position in the embedded wh-question), but may raise to a higher potential checking-position if this movement is driven by other requirements (i.e. the need for a +direct interpretation here). Further related to this, the relative acceptabilit y of wh-island violations such as (76) indicates that strong features must simply be checked at some point prior to Spell-Out in order that they will not cause the derivation to crash at PF and that there is no necessit y that wh-features on wh-phrases be checked as early as possible (i.e. here by the closest available licensing +wh+Q Comp).34 Example (76) above may also be compared with others such as (77) below where the in situ wh-phrase may take scope at the higher +Q Comp without movement: (77)
Who remembers where we bought what?
33
Although the approach to movement and feature-checking outlined here has not committed itself either to maintaining (or abandoning) the principle of Greed as a general derivational constraint (the proposals made here would technically be compatible with either position), it nevertheless has been assumed that wh-raising always takes place to satisfy licensing requirements of wh-phrases themselves — either to disambiguate and trigger a +Q Comp as a licensor for all wh-elements in its domain, or to bring a wh-phrase into the wh-licensing domain of a +Q C0. If “classic” wh-movement in languages like English is then primarily triggered by the need to activate Comp as a licensor of a particular t ype for the wh-phrases present in its domain, it may seem unnecessary and perhaps redundant to suggest that this raising is also triggered by a second requirement — to check wh-features on the +Q C0. Consequently it may not be possible to argue that in wh-island cases such as (76) the wh-phrase which books raises to the matrix Comp to satisfy wh-featurechecking requirements of this Comp (and therefore such wh-island violations do indeed require some other explanation).
34
Possibly one could formally implement the scope requirements of wh-phrases by the inclusion of a +direct (question) feature addition to the wh-features of a wh-element.
W h-movement and feature-checking
119
In (77) the higher C0 is triggered as +wh by who and so may licence the lowest wh-phrase, resulting in a direct question interpretation for what. The full acceptabilit y of (77) with this interpretation clearly contrasts with (76) or (78) below, (78) being the overtly-extracted counterpart to (77) and being quite degraded; this full acceptabilit y can be attributed to the assumption that no LF raising of the wh-phrase what is necessary — the higher clause C0 has been triggered as +wh+Q by who and so what may simply remain in its in situ position and be licensed without any need to raise: (78)
??What
do you remember where we bought t?
Sentences like (77) also raise again another (old) question concerning the scope of wh-phrases, namely why is it that a wh-phrase which is overtly raised to a +Q Comp must obligatorily take scope at that +Q Comp? In other words, why is it that where in (77) cannot have scope at the matrix +Q Comp? It could be argued that although movement of where to the lower +Q Comp in (77) will trigger it as +wh, if such a +wh Comp simply has a requirement that it must bind some wh-phrase, this requirement could be fulfilled by the lower wh-phrase what. Where could then take scope at the higher +Q Comp, which has been triggered as a licensor for wh-elements by raising of who. Such an interpretation would be as in (79): (79)
For which x, x a place, and for which y, y a person, does y remember what z, z a thing we bought in x?
Original formulations of this problem in fact asked why LF wh-movement (to a higher +Q Comp) may not be initiated from A′-positions in general in languages like English, thus why sentences such as (80) are bad: (80)
*Who thinks what Mary bought? Intended interpretation: For which x, x a person and for what y, y a thing, x thinks Mary bought y?
Cases such as (80) can be straightforwardly accounted for under assumptions made here — if wh-movement takes place only in order to trigger a C0 as +wh, and only a single wh-element need appear in SpecCP to effect this triggering, then there is no motivation for the second wh-phrase in (80) to move at all. By claim/assumption it does not move at LF, and so the partial movement in (80) to an intermediate position is justified in no way, and therefore should not take place (by economy). The interpretation in (79) for cases such as (77) can also be quite easily ruled out if one makes the assumption that a principle of Greed does in fact constrain
120
W h-movement and the theory of feature-checking
all operations of movement. If wh-movement to a (potential) +Q Comp takes place in order that C0 is triggered as a licensor for wh-phrases, then such a +Q C0 must end up licensing and binding the wh-element which has been raised into its Spec — according to Chomsky’s (1993) original formulation of Greed, all movement takes place for the direct benefit of the element which undergoes movement, therefore a wh-phrase may not raise to trigger a C0 as +wh unless the wh-phrase itself benefits (directly) from this movement and is licensed by the +wh C0. In (77) where may not raise to the intermediate Comp unless it becomes licensed by this as a result; given that where could be licensed by the matrix +wh C0 in situ without raising, it may not raise just so that another element, the in situ what, becomes licensed.35 Finally we need to check to see whether the suggested account may handle cases of superiorit y. An attractive purely syntactic account of superiorit y phenomena may be based on considerations of Shortest Move, as proposed in Chomsky (1993) and Reinhart (1998) (drawing also on data observed in Lasnik & Saito 1992). The relevant examples are given below in (81)–(83): (81) (82) (83)
I know whoi ti bought what. */?I know whati who bought ti . Who knows whati who bought ti ? /? a. * For which x, x knows for which , 〈z, y〉, y bought z b. For which 〈x, y〉, x knows what y bought
The basic suggestion is that movement of the subject who to the Comp of the lower CP in (81) is a shorter and hence more economical move than the attempted movement of the object in (82). Therefore movement of a wh-subject rather than an object wh-phrase in multiple wh-questions should always take place when possible, by economy. The interesting case noted by Lasnik and Saito is that in (83) — this example is acceptable when the lower subject who is interpreted as having scope at the matrix Comp, but is poor if interpreted at the lower +Q Comp. Following Reinhart (1998) it can be suggested that in the licit interpreta-
35
Note that the set of assumptions proposed in Chomsky (1995) perhaps may not allow one to rule out the illicit interpretation of (77). If wh-movement takes place only to check wh-operator features on a +Q C0 prior to Spell-Out and not for requirements of the wh-phrase itself, there seems to be no reason why the raised wh-phrase where should have to be bound by the particular intermediate +Q Comp it has moved to. If a +wh+Q Comp must simply bind some wh-phrase at LF in order to satisfy the Principle of Full Interpretation (and all wh-phrases must similarly be bound by some +wh+Q Comp at LF), it should be possible for the intermediate +Q Comp to bind (just) the in situ what and for where to be bound by the higher +Q C0.
W h-movement and feature-checking
121
tion of (83) there is no more economical way in which the sentence could be formed, hence that although the lower CP structurally resembles the superiorit y violation in (82), crucially economy will not have been violated in the good interpretation of (83), in the following way. Movement to the matrix Comp by the higher subject who will by hypothesis trigger this Comp as +wh, thus as a licensor for all wh-elements in its domain; if the lower clause who is interpreted with scope at the higher +Q Comp, there will therefore be no motivation for this wh-phrase to move anywhere, it will be licensed in situ and by economy should not undergo any raising. In order for what in the lower CP be licensed as indirectly questioned, what must raise to trigger the lower Comp. Here in considering potential wh-phrases which could be raised to trigger the lower +Q Comp there is no comparison in terms of Shortest Move between who and what, because if who were to raise to the lower +Q Comp it would end up being bound by this Comp (as detailed above) and would therefore fail to get the interpretation of being directly questioned (i.e. having scope at the matrix +Q C0). Examples such as (83) can thus receive neat explanation under the assumption that wh-phrases in situ at PF do not undergo further raising post-Spell-Out but are licensed in their in situ positions, rather than attempting to invoke some kind of ECP account based on LF wh-movement and the configurations it would give rise to.36, 37
2.6. Japanese and wh-scrambling wh This section now turns to wh-questions in Japanese, a language with overt question-particles, and shows how the Triggering Hypothesis and other assump-
36
A standard ECP account does not allow for the existence of the licit interpretation in (83). In a standard ECP-t ype approach there would be an LF representation of the acceptable interpretation of (83) with two subject wh-phrases raised in the same (matrix) +Q Comp. It would consequently be expected that either one or the other would not be able to c-command and antecedent-govern its trace.
37
If wh-feature-checking requirements relate solely to a +Q C0 and not to wh-phrases themselves (as assumed in Chomsky 1995), then it is hard to see how recent Minimalist approaches could allow for the licit interpretation of (83). The structurally closest wh-phrase to a +Q Comp (hence the lower clause who rather than what in (83) should be attracted to this Comp in all cases, to satisfy simple unselective requirements of the +Q C0 that its wh-features be checked by some wh-phrase (i.e. it should not be possible for a +Q Comp to select attraction of one wh-phrase rather than another). If however raising relates directly to licensing of wh-phrase themselves, then it is plausible that a wh-phrase would not raise to a +Q Comp which would not licence its particular intended interpretation.
122
W h-movement and the theory of feature-checking
tions made here can be used to account for the complex patterns of wh-scrambling and wh in situ observed in Takahashi (1993) and Watanabe (1991). In Japanese, unlike English, no overt wh-movement to a +Q Comp is forced to take place. If overt movement of a primary wh-phrase in English-t ype languages takes place in order to trigger the +Q C0 as +wh as suggested, one might then initially presume that +Q Comps in Japanese do not require this kind of triggering, and that instead of wh-movement the presence of the question particle ka (or no) in C0 will automatically activate it as a licensor for wh-elements. However, another possibilit y is that wh-features in Japanese might be weak, and that raising of a wh-element to Comp (for triggering of C0) might instead take place covertly at LF. Data presented in Watanabe (1991) do seem to indicate that some t ype of wh-movement does indeed take place in Japanese: (84)
?John-wa
[Mary-ga nani-o katta kadooka] Tom-ni tazuneta John-TOP [Mary-NOM what-ACC bought whether Tom-DAT asked no? Q
(85)
‘What did John ask whether Mary bought?’ John-wa [Mary-ga nani-o katta kadooka] dare-ni tazuneta John-TOP [Mary-NOM what-ACC bought whether whom-DAT asked no? Q
‘Who did John ask whether Mary bought what?’ Watanabe claims that the contrast between (84) and (85) above suggests that some kind of wh-element must undergo movement to a +Q Comp. In (84) there is only a single wh-phrase present and this occurs in a wh-‘whether’ island. Movement from such a position to the matrix +Q Comp should violate subjacency, so if one assumes that (some kind of) wh-movement does indeed take place, this will explain the the partially degraded status of (84) which is t ypical of weak subjacency violations. In (85) there is an additional wh-phrase present in the matrix clause; hypothetical movement between this indirect object position and the matrix +Q Comp should not violate any localit y constraints and (85) is in fact found to be perfectly acceptable. Writing prior to the Minimalist Program, Watanabe suggests that movement of some kind of wh-element must actually take place prior to S-structure in order to fulfil a condition that all (+wh)+Q Comps contain a wh-element in their Spec by this point (in Japanese). As nothing is visibly observed to undergo movement to Comp prior to S-structure, Watanabe hypothesises that a phonetically-null determiner-like subpart of a wh-phrase separates off from the rest of the (overtly realized) wh-phrase and moves to Comp.
W h-movement and feature-checking
123
Following Huang (1982), Watanabe suggests that all pre-S-structure movement is constrained by subjacency, but that movement occurring at LF is not. Assuming that all wh-phrases need occur in a +Q Comp by LF for reasons of scope/quantification, Watanabe suggests that movement of the second wh-phrase (or its null wh-determiner) in the wh-island in (85) will only take place at LF and hence will not result in any subjacency violation. The crucial difference between (84) and (85) would then be that wh-movement in (84) is forced to take place from within an island at a derivational point where subjacency does apply (prior to S-Structure), but that in (85) this obligatory pre-S-structure movement may be initiated from a position which is zero-subjacent to Comp, hence no violation occurs. This patterning has to be interpreted somewhat differently if one adopts the Minimalist assumption that localit y constraints such as subjacency should apply in a uniform way throughout the derivation, and there are in a fact a number of possible ways to account for the contrasts observed assuming the basic set of proposals put forward in this chapter. Examples such as (83)–(84) do indeed seem to indicate that some t ype of wh-movement to Comp must be involved, but it is not obvious exactly when in the derivation this need occur, nor what it is that actually undergoes movement. Given the dual assumptions made here that subjacency constrains applications of movement throughout the derivation and that only a single wh-element need ever raise to Comp in order to trigger it as a licensor for wh-phrases, it could be that there is just movement of a single full wh-phrase taking place at LF. This movement to trigger the +Q C0 will violate subjacency in (83) because it is forced to take place from within an island configuration. In (84) by way of contrast, the wh-phrase in the matrix clause which is zero-subjacent to the +Q Comp could raise at LF (without violating subjacency) triggering the C0 as a licensor and the C0 would then in turn licence both this wh-phrase and the wh-phrase in the wh-island (wh-licensing being subject to localit y constraints different from those affecting movement).38 However, there are also two other possibilities. First, and in line with Watanabe, it could be suggested that it is actually some phonetically-null subpart of the wh-phrase which undergoes movement to Comp and that this takes place (from a single wh-phrase) prior to Spell-Out. A second alternative is that such a null wh-determiner element
38
Note that it cannot be suggested that the wh-phrase in the wh-island in (83) simply violates nonmovement localit y constraints on wh-licensing, hence that no movement at all takes place in (83). If this were so then licensing of the lower wh-phrase in (84) should also violate such localit y constraints, yet (84) is perfectly acceptable.
124
W h-movement and the theory of feature-checking
moves (from a single wh-phrase) after Spell-Out, triggering of the C0 and licensing of all wh-phrases being only necessary at some point prior to LF. It is thus not possible to decide the issue of what moves when solely on the basis of the data presented in Watanabe (1991). However, there is certain other data relating to scrambling of wh-phrases discussed in Takahashi (1993) which will allow one to reach a conclusion concerning the t ype of wh-movement that occurs in Japanese and when this is forced to take place. Before this is examined, it is necessary to reflect once more on the motivation for this hypothetical wh-movement in Japanese. In the account put forward here it has been suggested that wh-movement occurs to trigger a Comp as a licensor for wh-phrases, checking the wh-features they carry, and that triggering of such a Comp is necessary when C0 is essentially ambiguous and under-specified. In Japanese if a ka (or no) question particle appears in C0, it might be claimed that this in itself should be taken to perform the critical disambiguating function, resulting in a ka-clause only being interpretable as a question (“t yped” in Cheng’s sense), and therefore that wh-movement to trigger C0 should not in fact be necessary. However, although a ka particle does indeed indicate that a CP is +interrogative, such clauses are still crucially ambiguous between being wh- or yes/no questions and the C0 will actually not have been specifically triggered as a wh-licensor — ka is a general question particle and not used only with wh-questions. In English it was argued that apparent disambiguation of C0 as +Q via raising of an auxiliary verb with interrogative features is not sufficient to licence wh-phrases, viz: *Did you see what?; I-to-C movement of an Infl carrying a +Q-feature may perhaps determine C0 as +Q, but raising of a single wh-phrase is still necessary to trigger C0 as a +wh+Q licensor. Thus in Japanese it can justifiably be argued that Comp is still ambiguous even where a ka particle appears, and some kind of wh-movement is in fact required to trigger the +Q C as +wh. Such conclusions would also seem to be supported by a consideration of (Mandarin) Chinese, another wh in situ language with question particles. In Mandarin Chinese question particles are not ambiguous in the way they are in Japanese — ma is exclusively for use with yes/no questions, while ne is used with wh-questions. If a +Q Comp may be fully disambiguated by such a question particle alone (whether null or overt) one might expect that wh-movement should not be necessary to trigger C0 as wh. Considering direct parallels of Watanabe’s Japanese cases, one finds that the crucial contrasts observed by Watanabe to exist in Japanese are in fact significantly absent from Chinese as noted in Huang (1982); (86) shows a wh-phrase freely occurring in a ‘whether’-wh-island and (87)
W h-movement and feature-checking
125
in a regular wh-island, both wh-phrases being interpreted with scope at the matrixclause +Q Comp (hence no movement which would violate subjacency can be taken to occur here, unlike in Japanese): (86)
(87)
ni xiang-zhidao [shei xi-bu-xihuan ni] you wonder [who like-not-like you ‘Who is the person x, such that you wonder whether x likes you or not?’ ni xiang-zhidao [shei mai-le shenme] ne? you wonder [who buy-ASP what Q ‘Who is the person x such that you wonder what x bought?’ ‘What is the y such that you wonder who bought y?’
Hypothetical movement to Comp in Japanese may therefore be ascribed the same motivation it is given in English, occurring to disambiguate C0 as a licensor for all wh-phrases in a certain domain, and no parallel wh-movement would appear to be necessary in Chinese due to the existence of unambiguous wh- and yes/no question particles.39
39
This does not exclude the possibilit y that a language may have unambiguous wh question particles yet still show certain general localit y effects in the distributional patterning of its wh-phrases (in situ). Restrictions on the occurrence of wh elements in situ may not necessarily result from movement having to take place to C0 for triggering purposes, but be due to localit y effects on the licensing relation between a (triggered) C0 and wh-phrases in situ. In Iraqi Arabic it has been noted that wh-phrases may not occur in wh-islands; when additional wh-phrases occur zerosubjacent to a higher +Q Comp outside the island this does nothing to improve the status of wh-phrases within the wh-island, that is, the improvement in acceptabilit y noted in Watanabe’s Japanese examples is not present in Iraqi Arabic. This indicates that when a (single) wh-phrase occurs in an island in Iraqi Arabic it is not any movement to Comp for triggering of C0 which causes a violation, but rather licensing of the wh-phrase which is blocked (one can also remember that wh-phrases in situ are ill-formed in certain other constituents that are not islands for any movement). What one might not expect to find is contrasts of the particular sort seen in (83)–(84) occurring in a language with fully unambiguous wh question particles (i.e. an asymmetry in the localit y conditions affecting primary and secondary wh-elements). Certain speakers of Chinese do in fact report judgements on data such as (83)–(84) different from those given in Huang (1982) and do not allow for high scope of wh-phrases out of wh-islands (see for example Shi 1994). However, such dialects do not show any asymmetry in the patterning of primary and secondary wh-phrases. Consequently it can be assumed that the localit y restrictions these speakers have are constraints on the licensing relation between the +Q C0 and (all) in situ wh-phrases and are actually not indication that covert raising for triggering of C0 is taking place (and importantly if an additional wh-phrase is added to the matrix clause of a sentence such as (87) this will still not license high scope on wh-phrases in the wh-island for speakers of this dialect, hence the ban on such high scope really does not relate to any C0 triggering).
126
W h-movement and the theory of feature-checking
The question remains as to what actually undergoes movement to Comp in Japanese and when this takes place. I would now like to suggest that an answer may be found in an examination of the wh-scrambling phenomena reported in Takahashi (1993), where it is argued that certain instances of what appear to be simple scrambling of wh-phrases in Japanese are actually cases of English-t ype full wh-movement. When worked through and interpreted within Minimalist assumptions and the general proposals made here, the data lead one to conclude that in Japanese some t ype of wh-movement to Comp does in fact have to take place prior to Spell-Out, and therefore that what actually moves will (often) be phonetically null. Example (88) below shows a standard wh-question with the wh-phrase occurring in situ in its base-generated position: (88)
John-wa [Mary-ga nani-o tabeta ka] siritagatteiru no? John-TOP [Mary-NOM what-ACC ate Q want-to-know Q either: ‘Does John want to know what Mary ate?’ or: ‘What does J. want to know whether Mary ate?’
As the possible translations show, the scope of the wh-phrase is ambiguous — it may receive interpretation as being either directly or indirectly questioned; if the former interpretation is selected, then the lower clause is understood as an embedded yes/no ‘whether’ question, if the latter, then the matrix is interpreted as a yes/no question.40 In (89) the wh-phrase appears scrambled/moved to an A′position in the matrix clause, SpecCP according to Takahashi, and only one interpretation is possible, that of a direct wh-question: (89)
nani-oi John-wa [Mary-ga ti tabeta ka] siritagatteiru no? what-ACC John-TOP [Mary-NOM ate Q want-to-know Q ‘What does John want to know whether Mary ate?’ NOT: ‘Does John want to know what Mary ate?’
The critical question here is why an indirect wh-question interpretation is no longer possible after such movement. The translation of example (90) below shows that when the Comp of the clause to which the wh-phrase is moved is not +Q interrogative, scrambled wh-phrases may indeed be reconstructed for binding by a lower +Q Comp:
40
It should be noted that certain stress on the wh-phrase is often necessary for wh-phrases in such wh-islands to take scope at a higher +Q Comp. High scope out of ka-dooka-‘whether’-wh-islands, as in (84), generally appears to be much more easy. Judgements on the data here all consistently come from Takahashi (1993).
W h-movement and feature-checking (90)
127
nani-oi John-ga [Mary-ga ti katta ka] sitteiru what-ACC John-NOM [Mary-NOM bought Q want-to-know ‘John wants to know what Mary bought.’
Scrambling of the wh-phrase in (89) then obviously does seem to give rise to certain effects which significantly differentiate it from the scrambling seen in (90). (89) appears to mirror instances of overt wh-movement in English, as for example in (91): (91)
Which booki do you want to know whok to give ti to tk?
In (91) the wh-phrase ‘which book’ raised to the matrix Comp can only have scope at this position and no interpretation with ‘which book’ reconstructed and bound by the lower +wh+Q Comp is possible, i.e. reconstruction giving the interpretation in (92) below: (92)
‘Do you want to know who to give which book to?’
In this significant way scrambling of the wh-phrase in Japanese (89) thus seems to bear a strong resemblance to wh-movement in English. The account which has been proposed here for English wh-movement is that raising of a wh-phrase to Comp is forced to take place in order to trigger C0 as an appropriate licensor for checking of the wh-features carried by wh-phrases in its domain (and that such triggering must occur by Spell-Out for reasons already detailed). Initially it might seem that a directly parallel account could not be correct for Japanese for the simple reason that overt wh-movement appears to be fully optional in this language. However, adapting ideas in Watanabe, it can be argued that the above scrambling data brought to light by Takahashi can only be explained within current Minimalist approaches if it is assumed that there is also covert pre-SpellOut movement to Comp of some phonetically uninterpreted wh-element in the fully in situ cases such as (88). Before expanding on this idea, the principal alternative to positing obligatory pre-Spell-Out movement will be examined, namely that movement to Comp need only occur by LF. So far the data in Watanabe (1991) have led to the following two conclusions: (a) some t ype of wh-movement to Comp in Japanese is indeed taking place at some derivational point, and (b) only one member of any set of wh-phrases interpreted at a single +Q Comp undergoes movement to that Comp (this explaining the contrasts in localit y restrictions on primary and secondary wh-phrases seen in (84)–(85) and the assumption that subjacency is a constraint applying uniformly throughout a derivation). If this movement is required to take place only by LF, then example (89) above is problematic — pre-Spell-Out scrambling/
128
W h-movement and the theory of feature-checking
movement of the wh-phrase to a +Q Comp seems to restrictively establish scope of the wh-phrase at this particular derivational point. Example (93) below also shows in a somewhat different way the same observation that the scope of a wh-phrase overtly moved to a +Q Comp is set by the position this wh-phrase occurs in at Spell-Out; in this example the wh-phrase may only have scope at the +Q Comp of the intermediate clause (to which it has been moved) and not at that of the higher matrix clause, despite the fact that a wh-phrase in a non+Q Comp position normally may have scope at a higher +Q Comp — see example (88): (93)
kimi-wa [nani-o John-ga [Mary-ga t tabeta to] omotteiru ka] you-TOP [what-ACC John-NOM [Mary-NOM ate that be-thinking Q asked Q kikimashita ka? only: ‘Did you ask what John thought that Mary ate?’ not: ‘What did you ask whether John thought that Mary was eating?’
Thus a wh-phrase overtly moved to a +Q Comp at Spell-Out may not take scope at any +Q Comp higher than the one in which it occurs at Spell-Out (93), nor may it take scope at any lower +Q Comp (89). Were wh-movement to Comp to be required to take place only by LF, then scrambling of a wh-phrase to a +Q Comp prior to Spell-Out should not have any significant effects precisely at SpellOut. That is, in order to capture the observed patterning one would have to simply stipulate that a wh-phrase be obligatorily bound at LF by the +Q Comp it occurs in at Spell-Out (this also effectively recognising the S-structure of preMinimalist GB as a significant and real level of representation relative to which certain constraints may be stated). Thus it would seem that the data may not be accounted for (in any way consistent with recent Minimalist assumptions) if it is suggested that wh-movement in Japanese is only forced to take place by LF (i.e. that wh-features are weak and their checking should therefore by economy only occur at LF). The basic paradigm may however be captured if one argues that movement of some wh-element to Comp must in fact occur prior to Spell-Out. Specifically I propose following and adapting suggestions made in Watanabe that a phonetically null but morphologically discrete wh-determiner/specifier occurs with wh-phrases in Japanese, and that such an element may optionally detach itself and move independently from the rest of the wh-phrase. Much discussion has already been made of the indefinite, variable nature of those lexical items functioning as wh-phrases in Japanese (and many other languages). Cheng (1997) suggests that a null wh-determiner-element will function to add wh-quantificational force to the essentially indefinite variable core of items such as dare, nani etc and give rise to
W h-movement and feature-checking
129
their interpretation as ‘who’, ‘what’ in the same way that suffixation of a -ka morpheme to the same items will add existential force and result in their interpretation as ‘someone/anyone’, ‘something/anything’ (dareka/nanika). The further suggestion that such a (null) wh-determiner-like element may actually separate off from the indefinite NP core and move independently to a +Q Comp is also not without other independent justification. In Serbo-Croat such wh-determinermovement can in fact be seen overtly; as examples (94) and (95) below show, a wh-determiner may either move to Comp on its own or pied pipe the remainder of the wh-phrase with it: (94)
(95)
Cijui si (ti) vidio [ti zenu]? whose did (you) see wife? ‘Whose wife did you see?’ [Ciju zenu]i si (ti) vidio ti? [whose wife did (you) see t? ‘Whose wife did you see?’
Furthermore it is well-known that arguably determiner-like classifier-phrases in Japanese may occur scrambled and separated from the NPs they quantify over, as e.g. in (96)–(98): (96)
(97)
(98)
san-nin-no-gakusei-ga kita three-CL-GEN-student-NOM came ‘Three students came.’ gakusei-ga kyoo san-nin kita student-NOM yesterday three-CL came ‘Three students came yesterday.’ san-mai kodomo-ga sara-o watta three-CL child-NOM plate-ACC broke ‘The child broke three plates.’
I will therefore make the assumption that the wh-features of a wh-phrase in Japanese are carried on such a phonetically null element, and that this element may either move independently or together with the lexically overt core of the wh-phrase. This now allows for the further suggestion that, as with English, a +Q C0 in Japanese must be triggered for licensing/checking of wh-elements prior to Spell-Out and that triggering of C0 is effected by movement to SpecCP of either the null wh-determiner or a full wh-phrase. Such proposals can now be shown to account for the crucial patterning noted in Takahashi (1993). In (88) movement of the null wh-determiner may take place prior to Spell-Out either to the intermediate +Q Comp or directly to the matrix
130
W h-movement and the theory of feature-checking
+Q Comp, triggering and being bound by whichever C0 the wh-determiner is raised to. Movement to the higher +Q Comp is similar to the wh-island violation cases in English considered earlier, where a wh-phrase is able to raise over one +Q Comp to another higher one in order to obtain scope there, as in: ?Which booki do you want to know who to give to ti? In (89) by way of contrast, the only possible interpretation, that of a direct wh-question, simply results from overt movement of the entire wh-phrase to the matrix +Q Comp together with the null wh-determiner (why it is not possible for the wh-phrase to take scope at the lower +Q Comp will shortly be returned to). In (90) the wh-phrase and its null wh-determiner first moves to the lower Comp which is +Q. Because triggering and subsequent licensing/checking of wh-features must take place by Spell-Out, the wh-determiner triggers the +Q Comp, resulting in binding and licensing of the wh-phrase by this Comp. Further pure scrambling then takes the wh-phrase to a higher (-Q) Comp.41 Feature-checking being an operation which simply has to take place at some point during the course of a derivation, the wh-phrase need not remain in the specifier of the C0 which has checked it after checking has taken place.42 Whatever motivates and licences scrambling in Japanese will allow the wh-phrase to move on further having successfully checked its wh-features (unlike English where scrambling, whatever it reduces to, is not licensed). Alternatively it might be the case that the wh-determiner carrying the wh-features necessary to trigger the +Q C0 remains in the specifier of this C0, being separable from the lexically overt NP core of the wh-phrase, and the core raises/scrambles higher (unlike English where the pure wh element is not separable from the core and so the entire wh-phrase necessarily must remain in the +Q Comp it triggers). Now we can return to (89) again to consider the problem of why the wh-phrase in this example may not be interpreted as having scope at the lower +Q Comp. Given that sentences such as (90) indicate that a scrambled wh-phrase may have scope at a lower +Q Comp, it might be expected that this should be possible in (89) too. It could be argued that the wh-phrase should be able to move first to the lower +Q Comp, trigger and be licensed by this Comp and then scramble on further to the higher +Q Comp. Concerning examples like (89), the obvious
41
Here I will not attempt to go into the challenging problem of what motivation there might be for scrambling. For a recent feature-driven approach, see Grewendorf and Sabel (1999).
42
A similar clear example of an element undergoing checking of one set of features and then raising to a higher position is in V-to-T-to-C movement; the verb first checks features against T0 and then raises higher to C0.
W h-movement and feature-checking
131
intuition which seems to be in need of capturing in order to account for restrictions on their possible interpretation is that a wh-phrase may not occur at SpellOut in the Spec of a +Q Comp that it does not take scope at, or within proposals put forward here, that it does not trigger as +wh. In the impossible interpretation of (89) as embedding an indirect wh-question, it is clear that a wh-phrase would occur (at Spell-Out) in the Spec of what would be a yes/no+Q es/no Comp (the higher Comp). I would like to suggest that such a configuration simply results in a conflicting and incompatible feature combination which the grammar does not tolerate. A +Q Comp may be either +wh+Q or yes/no+Q and must be disambiguated as such; as nani-o ‘what’ in (89) is an element which could potentially perform a triggering and disambiguation function for the higher +Q Comp, it can be suggested that it cannot remain in this +Q Comp without disambiguating it as +wh. If the wh-element occurs here in the Spec of a yes/no+Q Comp there will not just be a lack of Spec–Head agreement relating to the +wh/yes-no feature setting, but an actual conflict, the Spec being +wh+Q, its head yes/no+Q. Such a conflict of features is significantly absent in (90) — the matrix Comp to which the wh-phrase has been scrambled is not potentially ambiguous between a +wh and a yes/no +Q setting (i.e. it is not a +Q Comp at all, there being no question particle here) and movement of the wh-phrase into this Comp could not disambiguate it in any way. Therefore, although there is no feature agreement between the +wh Spec and the Q C0, there is also not the relevant conflict of features seen in (89) — in (90) the Comp will tolerate a +wh element in its Spec because such a +wh element does not have the potential to interact with it in any way. The suggested intolerance of the syntax for potential “feature conflicts” of this kind can indeed be seen elsewhere in a variety of constructions. In English, for example, a DP carrying a focus feature may not occur in the Spec of a (yes/no) +Q Comp: (99)
*??THAT BOOK must you buy?
The only possibilit y allowed here is a Left Dislocation construction which does not involve the DP occurring in the Spec of the +Q Comp: (100)
That book, must you buy it?
Similarly, although the first (SpecCP) position in German matrix clauses must normally be filled by an XP carrying either a topic or a wh feature, when the matrix is a yes/no question no XP may occur in this first position: (101)
Hat er den Karl gesehen? has he the Karl seen ‘Has he seen Karl?’
132 (102)
W h-movement and the theory of feature-checking *Den Karl hat er gesehen? the Karl has he seen
Again there would effectively arise here the same problem as in (89) — the fronted focus-topic DP has the potential to trigger and disambiguate the Comp as +focus; it therefore may not occur in the Spec of this C0 if the C0 is to be interpreted some other way, namely as yes/no+Q. In sum then it is proposed that wh-movement of some t ype must take place by Spell-Out in Japanese, this in order to trigger an ambiguous +Q C as +wh so that the wh-features on wh-phrases may be checked by Spell-Out. This essentially accounts for the fact that apparent instances of pure scrambling may not be “undone” at LF — the licensing of wh-phrases must occur by Spell-Out and this necessarily involves movement of some wh-element to a +Q Comp; triggering of such a +Q Comp via movement into its Spec position will result in the wh-element being licensed and obligatorily bound by that C0 (under the Greedlike assumption made in Section 2.5.2 that movement may only take place for the direct benefit of the item moved, hence a wh-element may not move to and trigger a +Q C if this does not result in the wh-element itself being licensed). The fact that wh-movement prior to Spell-Out does not appear to be obligatory is claimed to be due to the Watanabe-inspired proposal that it is a phonetically null wh-determiner which carries the wh-features of wh-phrases in Japanese and that this element may move independently from the indefinite quantificational core (just as seen overtly in Serbo-Croat and with movement of classifier-phrases in Japanese itself). The scrambling data from Takahashi show instances of the wh-determiner optionally pied piping the entire wh-phrase to Comp, again a possibilit y also realised overtly in Serbo-Croat (and with classifier-phrases in Japanese). Concerning its need to trigger an underspecified C0 as +wh prior to Spell-Out, Japanese is taken to pattern basically just like English; the significant surface difference between the two languages is that in Japanese wh-features are carried on a null whdeterminer which may move independently to Comp, whereas, following Cheng (1997) it can be suggested that in English the wh-features are morphologically incorporated into wh-phrases already in the lexicon and may therefore not extract from the DP and raise on their own to Comp. Finally it was suggested that in Chinese a +Q Comp is not ambiguous in the way that it is in Japanese and English, question particles in Mandarin Chinese being either +wh+Q or yes/no+Q, and that as a result of this no wh-raising for disambiguation of Comp is required, this in turn accounting for the lack of contrast in the distribution of
W h-movement and feature-checking
133
primary and secondary wh-phrases in wh-islands that Watanabe has observed exists in Japanese.43
2.7. Concluding remarks — the status of LF I will now conclude this chapter with a brief summary of the main claims put forward, adding also a number of extended remarks on certain broader implications of the approach developed. Chapter 1 argued at some length against the assumption that all wh-phrases occurring in situ at Spell-Out/PF automatically undergo raising to a +Q Comp at LF. Having concluded that it must be possible for wh-elements to remain and be interpreted in situ, Chapter 2 consequently set out to explore why any wh-movement might in fact be required. An examination of the patterning of wh-elements in Iraqi Arabic and Hindi in sections 2.2–2.3 resulted in the conclusion that wh-movement, where attested, takes place in order to fulfill a purely formal mid-derivational licensing requirement on wh-phrases and not to build any necessary semantic configurations. Identifying this licensing requirement as the Minimalist notion of feature-checking primarily due to the observation that it clearly triggers and (is satisfied by) operations of movement, it was then shown that the relevant wh-feature-checking requirement is in fact carried by wh-phrases themselves rather than the +Q C0 (contra regular Minimalist assump-
43
Although the analysis of wh-movement outlined here for Japanese draws significant inspiration from Watanabe (1991), the present account is actually forced to draw similar conclusions about the existence of a null wh-determiner element moving to Comp prior to Spell-Out for reasons quite different to those in Watanabe. Watanabe assumes that subjacency constrains only preS-structure (pre-Spell-Out) applications of movement, and so to account for the contrasts noted has to posit movement of some null wh element. Secondary instances of wh-phrases/wh-operators will raise at LF and this movement is taken not to be subject to subjacency. In the Minimalist model assumed here, subjacency is a uniform constraint applying throughout the derivation. Movement of only a single wh-element to Comp is required because this will suffice to trigger Comp as a potential licensor for all wh-phrases in its domain; and so other wh-phrases need not undergo raising at any point. Given the data presented in Watanabe and these assumptions one could actually still argue that the entire wh-phrase is what undergoes raising to Comp at LF (i.e. for triggering of C0 at this level). It is crucially the scrambling data in Takahashi which force one first to conclude that checking of wh-features must take place prior to Spell-Out, and then, observing there (often) to be no visible movement of any overt wh-phrase prior to Spell-Out, to make the second step of assuming movement of some phonetically null wh-element, such an idea being supported both by language-internal phenomena (classifier-phrase movement) and evidence from other languages (wh-determiner raising in Serbo-Croat).
134
W h-movement and the theory of feature-checking
tions) and that all wh-phrases furthermore have such a requirement. This then led to a second important conclusion that wh-feature-checking in Iraqi Arabic and Hindi is not restricted to occurring in strict Spec–Head/Head-adjoined localit y but may also effected between the +Q C0 and any position m-commanded by the +Q C0 in its immediate tense domain. Ultimately this resulted in the development of a fresh cross-linguistic account of wh-dependencies and feature-checking and the proposal that the wh-checking domain of a +wh+Q C0 may actually take on any of three different values depending on the instantiation of C0 in a language. Each t ype of C0 was argued to be associated with a certain licensing localit y which broadly corresponds to the degree of opacit y present between the operator-functional domain and the descriptive/propositional core of a sentence and the extent to which verbs of communication and propositional attitude are mentally analysed as being grammaticalised into the syntactic structure. A consideration of the asymmetrical patterning of primary and secondary wh-phrases in multiple-wh questions in English also led to the Triggering Hypothesis and the claim that wh-movement may actually take place for two different reasons relating to the general goal of feature-checking. In languages where it is not possible to directly base-generate an unambiguous wh-specification in C0, movement of a wh-phrase to Comp becomes necessary in order that C0 is determined as +wh and able to license (feature-check) wh-elements in its checking domain. Combined with the claim that a +wh+Q C0 potentially may instantiate any of the three discrete values identified, this allowed for a general characterisation of the wh-patterns found across languages, languages being described in terms of (a) whether there exists a lexical element (+wh question particle) which can be basegenerated in C0 and activate it as a licensor for wh or whether wh-movement for triggering is necessary, and (b) what the wh-checking domain value of the +wh+Q C0 is in a language. Such a characterisation was argued to allow for a full account of the range of surface variation observed across languages in their wh-systems. On a more general level the paradigms investigated indicate that the Checking Uniformit y Hypothesis is actually not a universal constraint and feature-checking relations may indeed be effected outside of the strict Spec–Head/Head-adjoined localit y suggested in Chomsky (1993–95). This particular conclusion is of considerable importance as it will now be shown that the Checking Uniformit y Hypothesis really is the principal theoretical motivation for an LF-based model of syntax in Minimalist approaches and that the basic architecture of the Minimalist model depends to a large extent on the existence of the Checking Uniformit y Hypothesis as an invariable constraint. In order to see that this is indeed the case requires here a brief consideration of how the level of LF has been justified prior to the advent of Minimalism and the realisation that earlier arguments for LF
W h-movement and feature-checking
135
essentially all lose their validit y when current Minimalist assumptions are adopted. In Government and Binding Theory (GB; Chomsky 1981, 1982, 1986a,b) the existence of a level of syntactic representation beyond that of S-structure was largely justified by the reasoning that such an extension of the theory would allow one to explain various linguistic phenomena by means of syntactic constraints already assumed in the theory. Accounts of superiorit y, strong crossover, localit y restrictions on wh-adjuncts and aspects of the interpretation of quantifier phrases were all given accounts in terms of known syntactic principles interacting with hypothetical LF raising. As noted in Chapter 1, covert movement in the GB period was generally taken to be fully equivalent to observable overt occurrences of movement, and so could simply be motivated by showing how constraints on overt syntactic processes would also seem to apply to similar dependencies hypothetically formed at LF. In the switch from GB to a Minimalist model of language, Chomsky (1993) subsequently argued both for a fully uniform derivation in terms of computational constraints from the point of lexical insertion through to LF, and also a significant uniformit y in the structural licensing of maximal projections — all XPs with features to be checked were suggested to effect this checking only in the Specifier position of an associated functional head. Additionally, movement was argued to take place only if forced by feature-checking requirements (the principle of Last Resort). This latter proposal had the immediate LF-relevant consequence that GB-st yle QR operations which did not appear to be driven by any clear morphological features could no longer be posited.44 In addition to the QR t ype of evidence for LF therefore being “lost” in Chomsky (1993), many other key arguments motivating the level of LF and covert movement also lose their validit y. Departing from standard GB views, Chomsky (1993) assumed that no LF wh-movement takes place; consequently earlier arguments for LF relating to superiorit y effects, localit y constraints on wh-adjuncts in situ and crossover phenomena may no longer be invoked as motivation for LF as they critically refer to wh-movement at LF. When considered with some care, the level of LF assumed in Chomsky (1993) can in fact largely be said to be motivated by considerations of uniformity and analogy alone, rather than by the range of more empirical argumentation present in the earlier GB model. The assumption of uniformit y is 44
An interesting feature-based theory of QR is suggested in Beghelli (1997) and Beghelli & Stowell (1997). However this treatment does not provide the kind of evidence for LF found in May (1977, 1985) where scopal ambiguit y amongst quantificational elements is directly constrained by syntactic principles also shown to constrain overt applications of movement (e.g. the ECP).
136
W h-movement and the theory of feature-checking
in large part the idea that the licensing of many syntactic dependencies is restricted to occurring in parallel highly local configurations across all languages. Such an assumption leads directly to the necessit y of a level of LF for the simple reason that if certain elements are taken to require licensing in a uniform way/positiont ype, then whenever such a licensing configuration is not overtly created, it must be assumed to be met at a post-Spell-Out point, and so via covert movement/LF. The situation is similar with the assumption of analogy — if an element of a certain t ype shows evidence of raising to be licensed in one language, then by analogy it is assumed that elements of the same t ype will require parallel licensing in other languages. If in these other languages raising does not occur overtly, it will again be taken to (have to) occur in some covert continuation of the derivation. As commented on at various stages earlier, in Chomsky (1995) it is proposed that what undergoes movement in the post-Spell-Out portion of the derivation is actually only the formal features (ffs) associated with a lexical element, where this may include (for example) case, agreement and categorial features but not semantic features. A result of the further distinction made between + and -interpretable features and the suggestion that +interpretable features will only ever require checking when strong on a functional head is that all wh-movement must indeed take place in the overt syntax and not at any level of LF. Similar reasoning is also expected to apply to QR — as quantificational features of QPs are naturally +interpretable, it is predicted that either strong “Quant”-features will trigger overt QR or that all QPs will remain in situ throughout the derivation. Alternative accounts of QR developed in Hornstein (1995) and Kitahara (1992) where the scopal interaction of QPs results from LF raising for Case-checking combined with the Copy Theory of traces may not be invoked in the Chomsky (1995) system as semantic features are taken to be stranded in situ when the Case and agreement features of DPs raise at LF. Consequently then, while constituting the major motivating evidence for the existence of LF in the GB model, neither wh-dependencies nor QR provide positive support for LF in the (1995) Minimalist model. If one asks then what justification empirical or otherwise does remain in Chomsky (1995) for assuming an abstract level of LF formed by covert movement, the answer is that LF and ff-movement are, as in the 1993 model, once more principally motivated by assumptions of analogy and uniformit y, the suggestion that certain dependencies will require licensing in all languages and that this licensing must crucially be effected in a uniform strict localit y. It can therefore be seen both (a) how strongly the Minimalist model relies on the Checking Uniformity Hypothesis as a mainstay support for a model incorporating
W h-movement and feature-checking
137
LF, and (b) how precarious the status of LF actually is in the Minimalist Program, being supported largely by this assumption of universal licensing localit y rather than the broader argument base established in GB but subsequently lost once the set of Minimalist assumptions proposed in Chomsky (1995) are adopted. This chapter has now provided a substantial amount of argumentation based on clear empirical patterns indicating that the Checking Uniformit y Hypothesis is not a universal constraint and that the assumption of a universal Spec–Head/Headadjoined localit y constraining all relations of feature-checking relations is actually not supported. In the case of wh dependencies it has been seen that the feature wh can indeed be licensed in significantly wider domains than standardly assumed and often in positions which are not strictly local to the checking head C0. Consequently, if the mainstay and key supporting assumption necessitating a level of LF is now seriously challenged, arguably to the point where it can no longer be maintained, and if the investigation carried out in Chapter 1 has shown that other arguments in favour of LF in earlier generative approaches are in fact ultimately also unconvincing, it can now be suggested that the most natural step forward in the development of a truly Minimalist model is indeed to assume a model simply without LF, hence a model where the syntactic derivation terminates at Spell-Out and Spell-Out forms are essentially the basic syntactic input forms to interpretation. Such a step though clearly necessitating a different understanding of a number of phenomena would seem to be the inevitable outcome of a rigorous consideration of the evidence and constitute a positive move forward in the development of a psychologically plausible model of language. It is also a readjustment and a streamlining of the model which would certainly bring with it a considerable amount of instant simplification too, one obvious example being that one would no longer need to assume any economy principle of Procrastinate to account for why covert movement would be commonly preferred to overt raising. Clearly a considerable amount of careful thought and serious attention is required to fully develop a robust alternative to current LF-dependent Minimalist approaches, and although Section 2.7.1 below will indeed indicate a variet y of possible directions one might proceed in, it is not the central purpose of this work to attempt such a task in real earnest. Rather at this point the intention is to simply signal the wider global consequences and conclusions which an honest assessment of the full set of patterning in wh paradigms would seem to lead one to.
138
W h-movement and the theory of feature-checking
2.7.1. Non-wh wh checking dependencies in a model without LF. If the above remarks are justified and LF as a discrete level of syntax no longer appears to be really well-motivated and if the Checking Uniformit y Hypothesis is not a universal constraint which necessarily applies to all feature-checking dependencies, a natural question to ask how one should now view other non-wh wh dependencies whose licensing has previously been taken to require LF movement. The licensing of features such as Case, agreement and tense in many languages has commonly been assumed to require covert LF raising to a relevant functional head, so how can one suggest that these relations are now satisfied if there actually is no syntactic extension of the derivation to a level of LF? Here I will offer just a few speculations on the subject and indicate a number of possible ways in which the account developed on the basis of wh-dependencies could in principle be extended to cover other feature-checking relations. There might in fact appear to be two basic ways of approaching this general issue. The first is to question whether the licensing of these attributes really does require a level of LF in any case, independently of what has been argued for in this chapter. Initially in Chomsky (1993) it was assumed that specifications such as Case and agreement required feature-checking and that if this did not take place overtly, then because of the uniform locality suggested to restrict all checking relations this would have to be effected via covert LF movement. However, this interpretation of the patterns which motivate these claims was significantly shaped by other assumptions made in Chomsky (1993), in particular the principle of Greed, which forces raising to take place for the direct benefit of the element moved. Because of Greed, overt subject DP raising could not be analysed as occurring to satisfy just properties of T0, but had to be seen as triggered by a need to check licensing requirements in the DP itself, hence Case. If Case then seemed to require checking on subjects DPs, it was naturally assumed that object DPs would also require parallel licensing of Case, and that this in many languages would be effected covertly between Spell-Out and LF. However, in Chomsky (1995) with Greed replaced by other assumptions, functional heads may now in fact attract other elements such as DPs to satisfy checking requirements of the head alone, and indeed it is suggested that overt subject raising is actually triggered by strong D0-features on T0 (the EPP) and not by any pre-Spell-Out Casechecking requirement of the subject. Consequently subject DP-movement importantly no longer constitutes evidence that Case-features need to be checked via movement. If one now considers object DP-movement in languages such as Icelandic, the optional nature of this raising is again commonly attributed to the
W h-movement and feature-checking
139
occurrence of strong D-features optionally base-generated on v0 and not to any Case/agreement-checking requirement, so again the classic paradigms do not provide evidence that Case/agreement features need checking. Certainly it has continued to be assumed that Case does require checking, but the real overt evidence for this has effectively been lost due to a re-organisation and re-orientation of the general Minimalist system. Furthermore this re-orientation has also classified agreement-features on DPs as technically no longer in need of any licensing. Overall then, given that there is now no longer any overt movement driven by the need to check Case or agreement specifications of DPs, it would actually not be unreasonable to suggest that neither Case nor agreement actually do necessitate checking (at any level). If this may be maintained, then it is clear that the “loss” of LF causes no great problems here as LF covert movement for checking would not be required in any case. Further reflection might well lead to similar conclusions concerning other feature-checking dependencies and possibly suggest that all checking relations are either effected overtly or not at all/never. Related to this general line of thinking, there are two other possible ways of thinking about Case and agreement features on DPs. The raising of subject DPs to SpecTP triggered by the EPP for checking of D0-categorial features is taken to allow for the Case/agreement features present on subjects to be checked as “freeriders” (if indeed they do need checking) and it is essentially the hypothetical checking of Case on object DPs which consistently forces the assumption of LF-raising in standard Minimalist models. Here one possibilit y (to avoid assuming a level of LF) would be to return to the pre-Minimalist GB position and suggest that object DPs (and their Case specification) are simply licensed/checked in complement positions without any movement. As commented on in an earlier section, there would not seem to be any good a priori reason to exclude the complement position from being a possible checking position and the statement made in Chomsky (1995) that argument DPs (as opposed to expletives) must always undergo some kind of movement in order to be able to check their features is really no more than simple stipulation. If one were then to allow (back) in licensing of maximal projections in their complement positions, both subject and object DPs would in fact be overtly feature-checked and the regular in situ occurrence of object DPs would actually not be a reason to call for any level of LF. Another rather general approach to the problem might also be simply to reject the idea that Case, agreement and tense features/morphology necessarily require checking in all languages and suggest that only certain instantiations of morphology are syntactically “visible” and “live”. Concerning the differences between French and English-t ype languages with regard to verb-raising and tense/agreement morphology, it has been very well noted that overt verb-move-
140
W h-movement and the theory of feature-checking
ment seems to coincide with “more” overt agreement morphology, and that languages such as English with very little or zero agreement morphology consistently seem to have no overt verb-movement to T/AgrS; it could therefore be suggested that where a language’s verbal morphology has gone into decay, as in English, and is clearly on the way to disappearing, it may in fact become computationally “invisible” in some sense and either not require checking, no longer being “active” morphology, or allow for simple deletion instead of checking. It can be noted that Chomsky (1993) actually makes a similar kind of suggestion concerning PF — in Chomsky (1993) weak features were argued to be “invisible” at the PF interface and so not in need of checking by that point. Here one can suggest that in languages such as English where the tense/agreement morphology has gone beyond some critical point of decay, this morphology is invisible at LF (or deletes) and so does not trigger any violation of Full Interpretation.45 It would then only be languages/paradigms with “live” morphology that would require checking, this triggering overt movement to the checking head as in French finite clauses. Such an approach would again have for effect that one would not need to posit a level of LF simply to allow for the licensing of features apparently unchecked at the point of Spell-Out; these features would just be deleted and all genuinely necessary checking would take place (by raising) prior to Spell-Out. The above ideas all attempt to suggest that there really is no need to posit an LF extension of the derivation to allow for the checking of certain features on in situ elements because these features quite possibly do not require any checking at all. Supposing however one were to retain the more orthodox view that Case, agreement and tense do require checking, one might ask whether, to avoid the necessit y of positing a level of LF simply to accommodate this checking, it would not be possible to extend the mode of explanation used with wh to capture the relevant cross-linguistic contrasts in some way. Staying with the differences in French and English verb-movement as a representative example of the t ype of contrast which is in need of explanation, what a Spell-Out-as-LF approach would need to explain is how it could be possible for tensed verbs in English but not French to be checked in situ by a higher T0, i.e. why “checking at a distance” might be available in English but not in French. Reflecting back on the account proposed for wh, it was suggested that
45
Tense-features being +interpretable would anyway not cause any violation if left un-checked at LF, and indeed according to Chomsky (1995) should only require checking if actually strong (as with other +interpretable features).
W h-movement and feature-checking
141
overt raising may take place for either of two reasons. One of these was the idea that a potential licensing head may need to be triggered in some way in order to function as an appropriate licensor (the Triggering Hypothesis). This was cast in clear terms of “disambiguation” of C0 as +wh+Q rather than +yes/no+Q or focus etc. The notion of “disambiguation” could however conceivably be thought of as just one instance of a rather more general process of “activation” or “identification” of a head as a licensor able to checking features of a certain sort. In the examination of wh, it was suggested that the C0-head could be disambiguated either via base-generation of a +wh specification directly in C0 (if a language has a wh-question particle t ype element available in its lexicon), or otherwise triggered via the “carrying-in” of a +wh specification to C0 in the form of wh-movement. Simple +Q features were also taken to be either base-generated directly in C0 in the form of a covert or overt Q-element (as in Japanese) or carried-in if originally base-generated on Infl (following Aoun & Li 1991). Using this now to think of the difference between French and English in the behaviour of their tensed verbs, it could be suggested that T0 like C0 requires some general activation in order to check tense-features on the verb and that this may critically be effected in either of two ways — if a language allows independent base-generation of Tense features directly from the lexicon into T0, then T0 can be activated in this way, whereas if this base-generation option does not exist in a language, movement of the verb to T0 may be the only way to activate it. Possibly the observation that tense/agreement morphology in English may occur directly in T0 on the expletive verb ‘do’ in negative sentences can be taken as indication that a tense specification may indeed be directly base-generated in T0 and hence that T0 can be activated as a licensor via simple base-generation (and no verb-movement is necessary). The obligatory overt verb-raising in French by way of contrast would indicate that tense-features cannot be directly base-generated in T0 and must be carried-in to activate it. In this sense English would be the equivalent of a “question-particle wh in situ language” in its tense-licensing, whereas French would be a “wh-movement language”. Such an approach extended across other feature-checking dependencies would permit an account of the differences observed across languages with regard to raising of a whole range of elements. Where an element (a tensed verb etc) does not undergo raising it would be assumed that the relevant licensing head is lexically activated via base-generation of a feature specification in the head and the element with features is consequently checked “at a distance” in its in situ position
142
W h-movement and the theory of feature-checking
without the need for LF raising.46 A second motivation for movement identified with wh dependencies was the need for a wh-phrase to escape localit y restrictions on the checking relation. For example, in Iraqi Arabic, wh-phrases base-generated in non-interrogative tensed CPs were seen to be forced to raise out of this CP to the tense domain of the +Q Comp. One may naturally wonder whether a similar kind of motivation might not also cause movement in other checking dependencies. Although there would not appear to be any obvious syntactic islands interrupting the relation of T0 to V0 or SpecTP to SpecVP, it is in fact possible that Relativised Minimality plays something of a potential blocking role here. Two significant properties of Relativised Minimalit y have been pointed out in the course of the chapter here. The first of these is that Relativised Minimalit y appears to constrain both
46
Such an approach will in fact also allow one to accommodate and encode all the valuable insights resulting from earlier and current work on QR which have actually been taken to necessitate a level of LF. If one adopts in large part the approach to quantificational scope developed in Beghelli (1997) and Beghelli & Stowell (1997), where the scope of quantifiers is licensed by associated functional heads, it can be suggested that wherever QR/covert LF movement of a QP to its licensing head has previously been assumed, one can instead suppose that it is possible to base-generated the necessary licensing features directly in the relevant head and no raising of the QP for triggering/activation of the head will be required. In other words, where no overt QR occurs it can be suggested that there is lexically available the means to activate and identify the relevant quantificational head and a discrete (most commonly covert) X0-level lexical element (similar to a null wh-question particle) with the necessary features can be inserted directly into the potential checking head, and there will be no need for any movement of the QP at any point, the QP being licensed at a distance by the activated head. Where however overt QR is observed to take place (as has actually been noted in an increasing number of languages), it can be suggested that it is not possible to base-generate the feature-specification necessary for identification of the licensing head independently from the QP to be licensed, and raising of the QP together with its feature-specification will therefore be required (this paralleling the need for overt wh-movement in languages like English). Consequently, the general model argued for here without a further syntactic level of LF in fact does not necessarily lose any of the expressive power present in more standard models incorporating a covert continuation of the syntactic derivation, but rather views the creation of such dependencies as being established in a rather different way, linking the movement/non-movement distinction in any dependency to the availabilit y in a language of lexical items which may serve to identify functional heads as legitimate licensing heads. Such a model therefore still permits one to maintain the assumption that cross-linguistically the input forms to interpretation will in fact be essentially uniform with an underlying universal design, links being established between functional heads and substantive elements carrying relevant feature specifications in all language t ypes. The difference between languages will simply lie in where the non-head element is phonetically realised in any head-substantive dependency chain, this being dictated by properties of the lexicon as outlined here and opacit y/island effects which will also be subject to certain cross-linguistic parameterisation.
W h-movement and feature-checking
143
licensing dependencies and relations of movement, but in rather different ways. For example, in Iraqi Arabic and Hindi it was noted that wh-islands appear to block the licensing of a wh-phrase contained inside a wh-island by the +Q Comp of a higher clause (examples 7 and 23). This barrier-t ype effect was argued not to relate to any LF movement but to be a constraint on a non-movement relation, and importantly it was shown that movement of wh-phrases out of wh-islands was clearly more acceptable than the attempted high-clause-scope interpretation of an unmoved wh-phrase remaining inside the wh-island (examples 8 and 24). The rather clear conclusion from this is that the Relativised Minimalit y effect induced by wh-islands restricts movement and licensing relations in different ways, movement being less severely affected by the wh-island than the attempted licensing relation. The second point noted in footnote 39 was that Relativised Minimalit y effects vary across languages and even amongst speakers of a “single” language. It was noted that while certain speakers of Chinese (e.g. Huang 1982; Li 1992) allow for a wh-phrase in a wh-island to be interpreted as taking scope at a higher clause +Q Comp (hence to be licensed by that Comp), other speakers of Chinese (such as Shi 1994) do not. The effects of Relativised Minimalit y are then subject to certain variation across dialects/languages. These conclusions can now potentially be used to account for observed differences in verb-movement again. It may be suggested that certain heads such as Neg0 intervening between T0 and V0 can in various languages constitute Relativised Minimality-like barriers to the licensing relation between a T0 and a V0 position.47 If such Relativised Minimalit y barriers may sometimes block licensing relations but allow movement, then in languages such as French, movement of the verb to T0 may be forced (and allowed) in order to result in licensing of tense features on the verb. In other languages (e.g. English) it can be suggested that similar heads will not constitute Relativised Minimalit y barriers and a longchecking relation between T0 and V0 will be possible (and therefore no raising of the verb will be necessary). Putting it in other words, in languages such as French, T0 will not be able to “see” the V0 position due to the interference of an interven-
47
Following ideas in Cinque (1997), it may be suggested that functional projections are always present in clausal structure even if not signalled by any overt morpheme. Cinque also proposes that when no overt head occurs, a default value of the head may be assumed. In the case of Negation, it can therefore be suggested that there is always some head/phrase present in the clause with a default positive value and that this is simply switched to the opposite negative value by the insertion of an overt neg morpheme (similar to the switch between finite and +finite in T0). The presence of such a head can then be argued to always block “long-licensing” in French (but permit it in English).
144
W h-movement and the theory of feature-checking
ing head, just as the +wh+Q C in Iraqi Arabic/Hindi cannot “see” a wh-phrase in an opaque lower tensed CP and raising of the wh-phrase is forced. In English-t ype languages however, T0 will be able to see and license a tensed verb lower down in V0, in the same way that for many speakers of Chinese and English a wh-phrase in a lower wh-island is visible to the licensing +wh+Q C0 of a higher interrogative clause. Although the above collection of remarks are essentially programmatic in nature, they are intended to show that there are indeed a number of plausible ways to approach the licensing of non-wh dependencies in a model without LF, and that the realistic development of such an approach may consequently not prove to be difficult in any major way. As pointed out in footnote 46, it is possible that many functional heads may be inherently activated as licensors via the base-generation of a feature-specification directly into the head, this being suggested to occur when a language has the equivalent to a null wh-question particle lexically available for other non-wh functional heads as a vehicle to transport the relevant set of features from the lexicon into the head (and otherwise such features will have to be attached to a substantive element and raised to activate/identify the licensing head). The incorporation and full extension of such an idea will in effect allow one to maintain the assumption present in the general Principles and Parameters approach that “LF” or the input form to interpretation is in fact basically uniform across languages, consisting in a universal hierarchically ordered set of functional heads in active dependencies with associated substantive elements. The simple difference between a Spell-Out-as-LF approach as advocated here and more standard models assuming a discrete level of syntax created by further applications of movement after the point of Spell-Out is that the present approach suggests that the encoding of all information necessary for interpretation is in fact already in place by the point of Spell-Out in the form of licensing dependencies between heads and substantive elements. The actual position of the substantive element relative to the head is taken to be relatively insignificant so long as a licensing relation is established, and the substantive element consequently need not always occur raised up to the immediate location of the head. Given then that such a model will allow for the expression of all the essential information present in LF-based models, albeit in slightly different format, the switch to such a model without LF can in fact arguably be implemented with surprisingly little in the way of serious re-adjustment. Such very general reflections aside, the core of this particular chapter has been an attempt to argue for a specific approach towards feature-checking and wh-movement, focusing on the primary wh-movement and in situ patterns commonly found across languages. Chapter 3 will now go on to examine another
W h-movement and feature-checking
145
wh-paradigm which can be shown to be highly problematic for regular Minimalist assumptions, the phenomenon of Partial Movement found in a number of languages such as German, Hungarian, child English and Hindi. It will be argued that a fairly simple and natural extension of the basic account set out here in Chapter 2 is indeed able to provide a successful treatment of the complex patterning observed in a way that is unavailable to more standard Minimalist approaches.
Chapter 3 Partial Movement and wh-expletive questions
3.1. Introduction This chapter takes as its focus the syntactic properties of wh-questions which result from the use of a “Partial Movement” strategy, forms in which a wh-phrase raised to the Q Comp of a subordinate clause is interpreted as taking scope at the +Q Comp of a higher clause which itself is occupied by an uninterpreted wh-expletive element, as e.g. in German (1) below: (1)
Was glaubst du, weni er ti gesehen hat? what believe you whom he seen has ‘Who do you believe he has seen?’
Such structures again pose serious problems for Minimalist checking theory with regard to the motivation for movement and the localit y of feature-checking because the interpreted “real” wh-phrase in Partial Movement questions is forced to undergo pre-Spell-Out wh-movement to a Q Comp position in which its wh-features cannot in fact be checked if feature-checking is indeed restricted to occurring solely within Spec–Head and Head-adjoined configurations. This chapter will argue that the existence of Partial Wh-Movement (henceforth PM) constitutes further strong evidence in support of proposals made in Chapter 2 that feature-checking is actually not subject to a universally rigid strict localit y and in certain instances may also take place within larger domains. It will be claimed
148
W h-movement and the theory of feature-checking
that the wh-features present on partially-moved wh-phrases are in fact checked in the position in which the wh-phrase occurs at Spell-Out (hence not in any Spec–Head relation with the higher checking +Q C), that no further LF raising to the +Q Comp takes place, and that PM is ultimately triggered for the same reasons as regular long wh-movement to a +Q Comp — to determine such a Comp as a licensor for wh-elements. It will suggested that the role played by the wh-expletive is critically to induce a different localit y value in the +Q C and alter the necessary triggering relation holding bet ween the +Q C and the substantive wh-phrase, evidence being provided that a highly similar role is also played by wh-expletives in Iraqi Arabic and Hindi. The chapter begins with an outline of certain general properties of PM and the theoretical problems they give rise to. It then considers quite generally how the relation of the substantive/real wh-phrase to the +Q Comp might be encoded and rejects the possibilit y that there is LF raising to the +Q Comp of any t ype, either direct or indirectly via clausal pied piping as suggested in Horvath (1997). Evidence is also provided that PM importantly cannot be analysed as movement for the pre-Spell-Out checking of any non-wh related features in the Q C0 such as focus or D-/categorial features. A solution to the problems of PM is then offered which draws from the approach to feature-checking developed in Chapter 2, and suggests that introduction of the wh-expletive into Comp in PM questions critically functions to identify a lexical instantiation of +wh+Q C0 in Comp with rather different properties than those of the regular +wh+Q C0 in German and Hungarian. Finally the chapter considers how various other properties of PM may be explained in the light of the proposed analysis, and concludes with remarks on the cross-linguistic t ypology of wh-expletives, suggesting that those in PM questions in German and Hungarian are wh equivalents to English it, while Japanese ka and Chinese ne are wh equivalents to there-t ype expletives.
3.2. 3.2.1. Basic properties of Partial wh Movement In simple bi-clausal PM structures such as (2) below an uninterpreted wh-element, commonly homophonous with the word for ‘what’ in a language, appears in the +Q Comp of a wh-question, and a wh-phrase which is interpreted/questioned (henceforth the “real” or “substantive” wh-phrase) occurs raised
Partial Movement and wh-expletive wh questions
149
into the Q Comp of the lower clause. PM will essentially be illustrated with data from German and Hungarian, much of which is taken from McDaniel (1989) and Horvath (1997): (2)
Was glaubst du, mit wemi Johann ti gesprochen hat? believe you with whom Johann spoken has ‘With whom do you believe Johann has spoken?’ WH
Was in (2) does not request any answer value, has no intrinsic semantic content and will therefore be referred to as a wh-expletive (as in Horvath 1997 and works by various other authors). The position occupied by such wh-expletives in German is arguably SpecCP, occurring to the left of the verb raised into C0 in matrix clauses. This would consequently also seem to indicate that the wh-expletives here are full phrasal categories rather than heads. In Hungarian similar wh-expletives are furthermore clearly inflected for Case, which again points towards an XP rather than X0 status. The formation of wh-questions via a partial movement strategy in any language is generally an option which exists alongside long wh-movement of the substantive wh-phrase, thus (3) with the wh-phrase mit wem ‘with whom’ raised higher to the matrix +Q Comp is a possible question-form with the same interpretation as (2): (3)
Mit wemi glaubst du, dass Johann ti gesprochen hat? with whom believe you that Johann spoken has ‘With whom do you believe that Johann has spoken?’
However, it cannot be argued that there is actually any optionality involved bet ween PM and regular wh-questions; (2) and (3) above result from different numerations, the numeration for (2) containing a wh-expletive which is not present in the numeration of (3). PM and long wh-movement questions are therefore not to be compared in terms of economy.1
1
Wh-expletives can be taken to be part of the numeration of a sentence in the same way that other non-wh expletives such as ‘it’ and ‘there’ are; i.e. Chomsky (1995) assumes that sentences such as (i) and (ii) below are not to be compared in terms of economy (and that there is therefore no real optionalit y here) because the numeration for (i) contains an expletive not present in the numeration of (ii): (i) There arrived a man. (ii) A man arrived. See however Cheng (to appear) for a rather different view on the relation of wh-expletives to the numeration and Castillo, Drury & Grohmann (1999) for an interesting general approach to expletives and the numeration not explored here.
150
W h-movement and the theory of feature-checking
In addition to simple bi-clausal PM structures, multi-clausal forms are also attested in which more than one wh-expletive may occur. In (4) below in addition to the sentence-initial was, a second uninterpreted was occurs in the intermediate clause: (4)
Was glaubst du [was Hans meint [mit wemi Jakob ti WH believe you [WH Hans says [with whom Jakob gesprochen hat]]? spoken has ‘Who do you think Hans says Jakob has spoken with?’
From this it may be concluded that the wh-expletive is not strictly/solely a scopemarker for the real wh-phrase — whereas in bi-clausal structures such as (1) and (2) the wh-expletive can be argued to mark the +Q Comp relative to which the interrogative force of the real wh-phrase is computed, in more complex structures such as (4) a wh-expletive element may occur in an intermediate Q Comp and hence not perform any scope-marking function. In this respect the wh-expletives in PM structures may be argued to be different from those in languages such as Japanese (ka) which are only found in +Q Comps, directly indicating and delimiting the scope of whatever wh-phrases are present.2 A certain apparent optionalit y is also attested in complex multi-clausal PM questions. It is found that either the real wh-phrase undergoes movement to the SpecCP position of the clause in which it is base-generated (as in (4)) and a wh-expletive occurs in the intermediate Comp, or the real wh-phrase may raise higher to the SpecCP of the intermediate clause (when no second wh-expletive is present): (5)
Was glaubst du, [mit wemi Hans meint [ti dass Jakob ti believe you [with whom Hans says that Jakob gesprochen hat]]? spoken has ‘Who do you think Hans says Jakob has spoken with?’ WH
Again, the t wo alternate forms in (4) and (5) are not to be assessed against each other in terms of economy as they derive from different numerations, (4) containing t wo wh-expletives, (5) only one. What crucially has been observed in multi-clausal examples such as (4) and (5)
2
Later however it will be argued that the differences in distribution between was and ka t ype elements reduces largely to considerations of Case.
Partial Movement and wh-expletive wh questions
151
is that for the majority of speakers every SpecCP position bet ween the +Q Comp and the clause in which the real wh-phrase originates must be filled by some wh-element, either a wh-expletive, the real wh-phrase, or the trace of the real wh-phrase. (6) below is therefore unacceptable for most speakers of German who allow for PM questions because the intermediate ( Q) Comp does not contain any of these elements: (6)
*Was glaubst du, dass Hans meint, mit wemi Jakob ti WH believe you that Hans says with whom Jakob gesprochen hat? spoken has
Informally then it appears that some “linking” of the +Q Comp with the real wh-phrase must be established via wh-elements which appear in those (SpecCP) positions through which successive cyclic wh-movement might otherwise take the wh-phrase on its way to the +Q Comp in non-PM question forms. Concerning those speakers of German who do find forms such as (6) acceptable, it will later be argued that this results from optional PF deletion of the intermediate expletive and that the underlying syntax of the construction is essentially the same for all speakers.3 In order to distinguish bet ween those who do permit such deletion and those who do not, the latter majorit y dialect who find (6) unacceptable will be referred to as Dialect A, and those who allow for intermediate-was deletion as Dialect B. Generally statements made in the chapter will be directed towards Dialect A as this allows for easier illustration of the problems created by PM structures. Regarding the wh-expletive elements found with PM, Horvath (1997) provides convincing evidence that these are not base-generated in their PF Comp positions, but appear there as the result of raising from some clause-internal position. Horvath shows both that wh-expletives are overtly Case-marked in Hungarian and that the Case which a wh-expletive carries is not any default Case which might be associated with base-generation in clause-initial position. Rather, the Case on the expletive appears linked to the argument status of the clause in which the partially-moved wh-phrase occurs — if this clause is a selected internal argument, then the expletive will bear accusative Case (7) or some other inherent objective Case where accusative is not assigned (8); if the clause is an external
3
Hungarian will also be suggested to allow for intermediate wh-expletive deletion, with evidence of various kinds supporting such a view.
152
W h-movement and the theory of feature-checking
argument then the Case on the expletive will be nominative (9):4 (7)
(8)
(9)
Mit mondtal, hogy mire szamitanak a gyerekek? what-ACC said-INDEF-2SG that what-AL count-INDEF-3PL the kids-NOM ‘What did you say that the kids expected?’ Mire szamitasz, hogy mit fognak mondani a what-AL count-INDEF-2SG that what-ACC will-3PL say-INF the gyerekek? kids-NOM ‘What do you expect that the kids will say?’ Mi zavarja Marit, hogy hogy beszelnek a what-NOM bother-DEF-3SG Mary that how speak-INDEF-3PL the gyerekek? kids-NOM ‘How does it bother Mary that the kids speak?’
It is therefore natural to assume that the wh-expletive is base-generated in the specifier of a functional head such as AgrO or AgrS, where its Case is checked, and then raised to the position it occurs in at PF. Additional evidence for this comes from the fact that a particular t ype of “indefinite” agreement may be triggered/appear on the verb in the clause in which a wh-expletive occurs. This indefinite agreement otherwise only appears when a (non-d-linked) object wh-phrase undergoes overt raising in the clause of the verb, hence in (10) but not in (11) or (12) where a non-wh object and a d-linked wh object are raised: (10)
(11)
4
Mit mondott Mari? What-ACC said-INDEF-3SG Mary-NOM ‘What did Mary say?’ (Azt) Mondta [hogy eljonnek a gyerekek]? (It-ACC) said-DEF-3SG [that away-come-3PL the kids-NOM ‘She said that the kids would come.’
Note that PM in embedded clauses in Hungarian takes place to a position which linearly follows the embedding complementizer hogy ‘that’, and hence technically not to SpecCP if there is only a single C position. Following Horvath, I will nevertheless continue to refer to the landing-site of PM in Hungarian and German in the same way as Comp/SpecCP. While this is something of an abstraction for Hungarian, ultimately it does not impact seriously on any of the arguments presented in the chapter. Furthermore it is also possible that clause-initial embedding complementizers such as hogy constitute just one part of a more complex series of functional heads commonly referred to with the single label “Comp”, as indeed outlined in Rizzi (1997).
Partial Movement and wh-expletive wh questions (12)
153
Tudjak hogy melyik fiut szereted know-DEF-3SG that which-boy-ACC like-DEF-2SG ‘They know which boy you like.’
Note now that in (9) above the clause containing the partially-moved wh-phrase is an external argument and the occurrence of the wh-expletive does not trigger indefinite agreement on the verb in the matrix. The patterning here follows neatly if it is assumed that the wh-expletive is base-generated in SpecAgrO when the clause selected by the matrix verb is an internal argument, triggering/ checking indefinite agreement on the verb before raising to its PF position, and in SpecAgrS when the clause is an external argument. In this latter case the wh-expletive will not trigger indefinite agreement as this is a feature specification associated with AgrO rather than AgrS. While German does not exhibit the same kind of Case-alternation evidence indicating where the wh-expletive is base-generated, it has been observed that: (a) was is generally only possible with those verbs which also allow a non-wh-expletive es ‘it’, and (b) was and es may not occur together. This would seem to suggest that was and es are base-generated in and compete for the same position. As the non-wh expletive es clearly may occur in a lower object-like position, it is natural to assume that such a SpecAgrO position is where both es and was are base-generated, with was then obligatorily raising to SpecCP exactly as in Hungarian.
3.2.2. Theoretical implications of Partial Movement structures The existence of partial movement structures as outlined above gives rise to a number of serious problems for Minimalist assumptions concerning movement and its motivations. Chomsky (1995) argues that movement operations may only take place for the direct and immediate checking of morphological features and that feature-checking operations are restricted to occurring within the strict localit y of Spec–Head or Head-adjunction configurations. Raising for featurechecking occurs prior to Spell-Out where a checking head carries strong features but is otherwise delayed until LF due to the economy principle of Procrastinate. Furthermore every application of movement must result in some feature of the element moved engaging in a checking relation; thus it is not permissable for an element to move through/to a position if none of the features of the element are actually checked in this position. Though initially justified with a variet y of data and argumentation, it actually seems impossible to maintain without modification the above set of Minimalist
154
W h-movement and the theory of feature-checking
assumptions in light of the basic observed properties of Partial Movement question forms. Such structures appear to provide hard evidence against a number of basic tenets of checking theory as detailed in Chomsky (1995). Specifically, the following problems are raised by Partial Movement: I. How can the obligatory pre-Spell-Out Partial Movement of the real wh-phrase into a Q Comp position be motivated? According to Chomsky (1995) it should only be possible for its wh-features to be checked against a +Q C0 when occurring in the Spec of such a C0, hence not in the Q Comp position it is forced to raise to in PM structures. As no wh-feature-checking results from PM, it is therefore predicted: (a) not to be possible to move wh-elements to such Q Comp positions, and (b) certainly not to be obligatory.5 II. If a wh-expletive obligatorily raises to and occurs in the Spec of the +Q Comp prior to Spell-Out, under standard accounts this must be in order to check a strong operator-feature on the +Q C0. If such a strong wh-operator feature is thereby checked on the +Q C0, there should then be no need for any other wh-element (the real partially-moved wh-phrase) that is related to this +Q C0 to undergo any raising prior to Spell-Out (or perhaps at any level). Movement of a single wh-element to the +Q Comp should be all that is required for satisfaction/checking of its operator-feature and one should not attest any t ype of secondary raising. However, additional movement of a secondary wh-element is also absolutely obligatory. III. In the case of complex multi-clausal PM constructions, how can the movement of intermediate wh-expletives be motivated? Again, in Chomsky (1995) all movement must be justified as resulting in some morphological feature carried by the element moved entering a checking relation with corresponding features on a functional head. However, unlike raising of an initial wh-expletive to a +Q Comp, intermediate wh-expletives must raise to and occur in what are plainly (the specifier positions of) Q Comps, where no direct checking relation can be established between a wh-feature specification on the wh-expletive and the C0 head. These important aspects of Partial Movement structures clearly call for explanation and pose the strongest of challenges to the claim that all movement operations directly result in and are solely motivated by the need for a strictly local feature-checking relation to be established, where such relations further exclusively
5
It will also strongly be argued in a later section that PM cannot possibly be analysed as involving checking of any other non-wh features either.
Partial Movement and wh-expletive wh questions
155
instantiate isolated, bi-unique relations between two elements in abstraction from other elements of a linguistic expression.
3.3. 3.3.1. Possible accounts of Partial Movement Approaching the above problems I will first review what general analyses may be offered for PM structures. Any analysis of such question t ypes would seem to need to offer some account of how the real wh-phrase relates to the +Q Comp at which it is understood to take scope, thus how wen ‘whom’ in (13) is interpreted as being directly questioned: (13)
Was glaubst du, weni er ti gesehen hat? what believe you whom he seen has ‘Who do you believe he has seen?’
There are commonly argued to be two basic ways in which the wen-(matrix)+Q C relation could be established. One possibilit y is that the linking between the partially-moved wh-phrase and the higher Comp is fully direct, with a chain of some sort being established between the two positions, perhaps the result of LF raising of the wh-phrase. Such a general direct dependency approach is pursued in McDaniel (1989), Beck & Berman (1996), Sabel (1996) and various other works. The second possibilit y, suggested in Mahajan (1990), Srivastav (1991), Fanselow & Mahajan (1996) and Horvath (1997), is that the relation of substantive wh-phrase to the +Q Comp should in fact be seen as indirect and mediated via some other linking (an indirect dependency), essentially that between the CP clause in which the wh-phrase is partially-moved and the wh-expletive in the higher SpecCP. Below I will consider each of these general approaches in turn and argue that neither will in fact allow for any effective solution to the problems detailed above, and that what appears to be necessary instead is again a reappraisal of the role of localit y in the licensing of feature-checking relations. Before turning to this however, it should be noted for the sake of clarit y that an assumption accompanying the analysis here is that Partial Movement in German and Hungarian is essentially a uniform phenomenon and that the same set of basic syntactic properties are responsible for the highly striking combination of wh-expletives and Partial wh Movement in both languages. Such a view contrasts with that in Horvath (1997) where the author points to and stresses certain apparent differ-
156
W h-movement and the theory of feature-checking
ences in German and Hungarian PM structures, and suggests that because of these, PM question forms may have a fundamentally different underlying syntax and derivation in the two languages. Discussion of the relevant differences suggested to exist in Hungarian and German PM questions will mostly be postponed until a later section. There it will be seen that these in fact largely reduce just to the availabilit y of PM with factive and certain other non-bridge verbs in Hungarian but not in German. As Beck & Berman (1996) have shown that it is possible to relate this restriction in German to other wh-movement constructions in the language in a way which suggests it should be viewed as a language-particular constraint characterising various constructions in German rather than as a propert y which particularly distinguishes PM in German from PM in Hungarian, I will proceed with the assumption that the tremendous surface similarities in PM questions in Hungarian and German do indeed reflect a broadly parallel underlying syntax.
3.3.2. Direct dependencies and wh-chains A fully direct dependency relation between the partially-moved wh-phrase and the wh-expletive element in the +Q Comp could conceivably be established in a variety of ways. One fairly straightforward possibility adopted in McDaniel (1989), Beck & Berman (1996) and other works is to assume that the partially-moved wh-phrase undergoes further LF raising to the +Q Comp where its scope is established. Such an approach might liken this LF raising to the commonly assumed LF movement of a DP associate to replace a non-wh-expletive in sentences like (14): (14)
There arrived a man. LF: [a man]i arrived ti
One might suggest that the substantive partially-moved wh-phrase in PM questions is a wh-DP associate to the wh-expletive in the +Q Comp and so must raise to replace the latter at LF. A second direct dependency account proposed in Anyadi & Tamrazian (1993) suggests that the surface differences between PM structures and more regular wh-questions do not result from any different underlying syntax, but rather from the way in which a wh-question is phonetically spelled-out. Anyadi and Tamrazian suggest that a fully parallel direct dependency is formed both in wh-questions where a substantive wh-phrase occurs in the +Q Comp (as in (3)) and in questions where a dummy wh-element such as was ‘what’ is found there (examples
Partial Movement and wh-expletive wh questions
157
such as (2)). They propose that in the PM case what happens is that the substantive wh-phrase is simply given phonetic interpretation in a lower chain-internal position than in other wh-questions and that occurrences of was ‘what’ higher than the substantive wh-phrase are phonetic reflexes of the chain formed to the +Q Comp. The chains created in PM questions are thus taken to be syntactically identical to those in other non-PM wh-questions, being subject to regular conditions on chain formation such as subjacency, the ECP etc, and all that sets the two question t ypes apart is the way in which the phonetic component interprets the wh-chain.6 A third foreseeable direct dependency account of PM might be to make use of Chomsky’s (1995) proposals concerning feature-movement/attraction and argue that PM questions involve some kind of pure feature-movement.7 Chomsky (1995) suggests that well-formedness constraints on PF dictate that any pre-SpellOut raising of morphological features to effect a checking relation necessitates the pied piping of a phonetically-interpreted host element containing these features. It is argued that pre-Spell-Out movement of features alone would cause a derivation to crash at PF, bare features being uninterpretable at the PF interface. PostSpell-Out movement, by way of contrast, will not result in structures requiring any phonetic interpretation, and the features necessary for checking relations may therefore raise/be attracted without pied piping of the additional host material necessary for PF convergence. Considering PM structures, it might be suggested that something approaching this ideal pure feature movement actually takes place in the pre-Spell-Out portion of the derivation. One could suppose that a wh-phrase containing wh-features in need of checking initially raises to some ( Q) Comp position, and then instead of raising successive-cyclically further, somehow projects its wh-features off from the substantive wh-phrase host and up to the higher +Q Comp. Wh-“expletives” would then be nothing other than the pure instantiation of the wh-features associated with a wh-phrase, and languages might accordingly vary as to whether features of a particular t ype allow for phonetic interpretation in isolation from their normal host elements (with English do possibly being the pure instantiation of tense/agreement features).
6
Note that Anyadi and Tamrazian adopt Brody’s (1995) framework in which chains are argued to be formed pre-syntactically and then inserted into syntactic structure in a single instance of “create structure”. Although no movement is assumed in such a framework, all chains are nevertheless imputed the same basic properties as those taken to arise via movement in other approaches.
7
An approach along the lines sketched here has in fact recently been taken up and argued for in Cheng (to appear), though making use of proposals in Chomsky (1998) rather than the somewhat different set of assumptions made in Chomsky (1995).
158
W h-movement and the theory of feature-checking
However, all such approaches where a chain connecting the partially-moved wh-phrase and the +Q Comp is formed by movement (or hypothesised to have properties equivalent to movement) are ultimately very difficult to maintain in the light of a number of clear differences and mismatches between readily observable long wh-movement and the PM chains. In any direct dependency movement account it is clearly expected that the dependency between the partially-moved wh-phrase and the +Q Comp should be constrained in the same way that movement to such a position is. This expectation is however not borne out and one finds that there are significant differences between PM structures and related long wh-movement questions suggesting that the former do not in fact result from the same t ype of direct raising to Comp. These mismatches basically occur in two directions. On the one hand one finds that the relation of the partially-moved wh-phrase to the +Q Comp is sometimes more restricted than wh-movement to such a position. Elsewhere the opposite appears to be true and the relation between the partially-moved wh-phrase and the +Q Comp is actually less restricted than the linking of these positions via regular wh-movement. Rizzi (1992) was perhaps the first to observe the non-parallelism between partial and long wh-movement structures with regard to localit y. The case of inner negation islands he presents shows that PM in German is not allowed where long wh-movement is fine:8 (15)
(16)
Mit wemi glaubst du nicht dass Maria ti gesprochen hat? with whom believe you not that Maria spoken has ‘With whom do you not believe that Maria has spoken?’ *Was glaubst du nicht, mit wemi Maria ti gesprochen hat? WH believe you not with whom Maria spoken has Intended: ‘With whom do you not believe that Maria has spoken?’
This would seem to indicate that the relation of the partially-moved wh-phrase to the +Q Comp is not in fact established via simple wh-movement as wh-movement is indeed seen to be possible over negation. A second important set of cases concerns certain clause-adjacency facts found in PM structures in the majorit y Dialect A in German (and by hypothesis in speakers of German Dialect B and Hungarian, though here masked by intermediate wh-expletive deletion). As mentioned in Section 3.2.1, speakers of Dialect A find sentences such as (17)
8
The status of PM and Neg islands in Hungarian is somewhat unclear. Horvath (1997) suggests that PM may be combined with matrix Negation in Hungarian if the CP in which PM has taken place can be interpreted as d-linked but is not possible when the CP is not d-linked.
Partial Movement and wh-expletive wh questions
159
unacceptable because no wh-element occurs in the intermediate Q Comp. If this position is filled with a second wh-expletive as in (18), or the substantive wh-phrase is raised to the intermediate Comp as in (19) then the structure is judged to be well-formed: (17)
(18)
(19)
*Was glaubst du, dass Hans meint, mit wemi Jakob ti WH believe you that Hans says with whom Jakob gesprochen hat? spoken has Intended: ‘Who do you think Hans says Jakob has spoken with?’ Was glaubst du [was Hans meint [mit wemi Jakob ti WH believe you [WH Hans says [with whom Jakob gesprochen hat]]? spoken has ‘Who do you think Hans says Jakob has spoken with?’ Was glaubst du, [mit wemi Hans meint [ti dass Jakob ti WH believe you [with whom Hans says that Jakob gesprochen hat]]? spoken has ‘Who do you think Hans says Jakob has spoken with?’
Sentence (17) should now also be compared with (20) where long wh-movement successfully takes place to the matrix +Q Comp: (20)
Mit wemi glaubst du, dass Hans meint, dass Jakob ti gesprochen hat? with whom believe you that Hans says that Jakob spoken has ‘Who do you think Hans says Jakob has spoken with?’
What example (20) shows is that wh-movement is indeed possible between the lowest clause of a sentence like (17) and its matrix +Q Comp. If the relation of a partiallymoved wh-phrase to a higher +Q Comp were to be a result of LF wh-movement one would then expect that PM structures such as (17) should be acceptable, as nothing should block (LF) movement of mit wem ‘with whom’ to the matrix +Q C and this would result in a well-formed representation paralleling (20). This again suggests that the dependency between a partially-moved wh-phrase and a +Q Comp which represents its scope is not established via any direct LF wh-movement.9 9
To avoid this conclusion one might perhaps suggest that LF raising of the partially-moved wh-phrase may somehow be different from overt wh-movement because what undergoes movement at LF is only the set of formal features of the wh-phrase. Such ff-movement could then be claimed to be blocked by tense and so cause (17) to be unacceptable. However, such a counter-
160
W h-movement and the theory of feature-checking
A patterning diametrically opposed to that above is also found, and Horvath (1997) points out that PM structures in Hungarian are possible in many cases where long wh-movement is not, for example with wh-adjunct phrases and factive clauses: (21)
(22)
Mit sajnalsz hogy hogy viselkedtek a gyerekek? what-ACC regret-2SG that how behaved-3PL the kids-NOM ‘How do you regret that the kids behaved?’ *hogy sajnalod hogy viselkedtek a gyerekek? how regret-2SG that behaved-3PL the kids-NOM ‘How do you regret that the kids behaved?’
Similarly where wh-extraction from subject and adjunct CPs is bad, PM structures are fine: (23)
(24)
(25)
(26)
*?Kineki zavarta Marit, [hogy telefonaltal ti]? who-DAT disturbed Mary-ACC [that phoned-2SG ‘To whom did that you phoned disturb Mary?’ Mi zavarta Marit, [hogyi kinek telefonaltal ti]? what-NOM disturbed Mary-ACC [that who-DAT phoned-2SG ‘To whom did that you phoned disturb Mary?’ *Kivel vagy duhos [merti talalkoztal ti]? who-with be-2SG angry [because met-2SG ‘Who are you angry because you met?’ Miert vagy duhos [mert kiveli talalkoztal ti]? why be-2SG angry [because who-with met-2SG ‘Who are you angry because you met?’10
argument cannot be maintained, for the same reasons discussed in Chapter 2 with regard to Iraqi Arabic and Hindi. LF movement in Chomsky (1995) is suggested to be pure ff-movement because pied piping of a larger host is only required prior to Spell-Out to satisfy PF constraints. Being costly in economy terms and unnecessary at LF, pied piping is therefore avoided after Spell-Out and formal features are assumed to raise quite unsupported to their relevant checking heads. Now, because the avoidance of pied piping has the status of an economy-driven constraint, it should clearly be possible to over-ride this where necessary for convergence. As it is found here in (20) that overt pied piping of the PP containing the wh-features does allow movement to take place across tensed clauses to the +Q Comp, it must be assumed that similar LF pied piping should indeed be available in (17) to allow for checking of wh-features on the higher +Q C, the pied pied PP carrying the wh-features up to the matrix +Q C. As sentences such as (17) are nevertheless unacceptable for the majorit y of speakers, the conclusion remains that the dependency between the position of the partially-moved wh-phrase and the +Q Comp is not mediated via LF (or overt) movement. 10
Note the use of miert ‘why’ here as an (uninterpreted) wh-expletive with adjunct CPs.
Partial Movement and wh-expletive wh questions
161
In these examples one finds that PM dependencies interestingly appear to be less restricted than movement. Because PM is therefore both more restricted than movement in certain environments and also less restricted in others, it would seem unlikely that the linking of a partially-moved wh-phrase with the higher +Q Comp is effected via any direct (LF) raising. When examined closely, direct movement chains and PM dependencies appear to be subject to quite different sets of restrictions. There are other reasons too which suggest that the linking of the real wh-phrase and the wh-expletive in PM structures cannot be established by any movement operation parallel to that occurring in regular wh-movement or with the LF raising of an associate DP to an expletive. One of these concerns the observation of certain anti-localit y effects with wh-expletive questions — where such elements are employed, the real wh-phrase may not occur in the same clause as the expletive: (27)
*Was hast du mit wem gesprochen? WH have you with whom spoken ‘Who did you speak with?’
Non-wh-expletive-DP-associate pairs are not subject to any such anti-localit y and it is not immediately obvious how an LF expletive-replacement analysis of PM would predict that sentences such as (27) are ill-formed. Also if one attempts to somehow equate PM structures with regular wh-movement chains, suggesting that the wh-“expletive” element either instantiates the wh-features of a wh-phrase moved to Comp, or is simply the higher member of a chain whose contentive element is spelled-out in some other chain-internal position, the ill-formedness of such sentences is equally unexpected — regular wh-movement may quite normally link the positions of the two wh-elements in (27) indicating that they are indeed legitimate chain-internal positions for a wh-dependency:11 (28)
Mit wemi hast du ti gesprochen? with whom have you spoken ‘Who did you speak with?’
Horvath (1997) further points out that a chain account directly linking the wh-phrase and the wh-expletive cannot be maintained in view of the fact that both wh-phrase and expletive in Hungarian are independently Case-marked, whereas
11
Though see Cheng (to appear) for a suggestion on how a pure-feature movement account making use of the more intricate “split” movement mechanisms in Chomsky (1998) might attempt to approach this anti-localit y problem.
162
W h-movement and the theory of feature-checking
movement chains are commonly assumed to contain a unique Case-marked position. The Case on the expletive is not a default Case (which could be assigned to an element in SpecCP, as in left dislocation structures in certain languages) but has been shown to correspond to the Case assigned by the predicate of the clause in whose Comp the expletive occurs; consequently it may often not coincide with that of the wh-phrase: (29)
(30)
Mit/*mire mondtal, hogy mire szamitanak a gyerekek? what-ACC/what-AL said-2SG that what-AL count-3PL the kids-NOM ‘What did you say that the kids expected?’ Mire/*mit szamitasz, hogy mit fognak mondani a what-AL/what-ACC count-2SG that what-ACC will-3PL say the gyerekek? kids-NOM ‘What do you expect that the kids will say?’
Such observations seem to discount the possibilit y of any analysis in which the wh-“expletives” are simply instantiations of the wh-features of a wh-phrase raised successive-cyclically through higher Comp positions. They also argue against analysing the wh-expletives as standard links in a wh-chain whose contentive element is phonetically spelled-out in a chain-internal position. In sum then there appear to be a variet y of good reasons for not adopting an analysis which directly links the wh-phrase to the +Q Comp via a chain attributed the properties of movement. Standard wh-chains and expletive-DP-associate pairings appear to have many properties quite different from those observed in wh-expletive-PM structures.12 12
Another argument against the chain approach can be added from German. Alongside Partial Movement and full wh-movement, certain dialects of German also allow for a wh-“copy” strategy, e.g: (i) Weni glaubst du weni er ti gesehen hat? who-ACC believe you who-ACC he seen has ‘Who do you believe he saw?’ Such wh-copy questions might seem to have (more of) the properties of regular wh-chain formation, and it may be suggested that they result from the spelling-out of copies of a wh-phrase in intermediate (-Q) Comp positions when a wh-phrase undergoes successive cyclic raising to a higher +Q Comp. Significantly the PM strategy is not permitted in certain cases where a wh-copy question is wellformed: PM questions are not possible where the tense on the verb in the subordinate clause is determined by that in the higher CP which contains the +Q Comp (such a clause then being “tense-dependent” on the higher clause), as with verbs like wollen ‘to want’:
Partial Movement and wh-expletive wh questions
163
3.3.3. Indirect dependencies and clausal pied piping Considerations such as the above led Horvath (1997) to propose an account in which the link between the real wh-phrase and the +Q Comp is established indirectly via clausal pied piping very much in the spirit of earlier pied piping accounts proposed in Mahajan (1990) and Srivastav (1991).13 The essence of the analysis is as follows: overt raising of the wh-phrase to a ( Q) SpecCP position results in the wh-phrase percolating its wh-features to the CP-node, thereby identifying the clause as a wh-phrase. The wh-expletive element, independently base-generated in the higher clause, also undergoes overt raising, from a position in which its Case is checked to the +Q SpecCP. At LF the entire CP in whose Spec the partially-moved wh-phrase occurs then raises to the expletive in a process of expletive-clausal-associate replacement. The real wh-phrase thus remains within the clause in which it is partially-moved throughout the derivation, this accounting for the lack of CED effects when adjunct wh-phrases occur partially-moved within subject, adjunct and factive CPs.14 Horvath suggests that the analysis she proposes also receives justification from the fact that overt clausal pied piping has been observed to take place in certain languages, for example Basque as described by Urbina (1990) and Quechua in
(ii)
*Was willst du weni Jakob ti besticht? WH want you who-ACC Jakob bribes ‘Who do you want Jakob to bribe?’ Wh-copy questions are however fine in these environments: (iii) Weni willst du, weni Jakob ti besticht? WH want you who-ACC Jakob bribes ‘Who do you want Jakob to bribe?’ The fact that partially-moved wh-phrases may not occur in all those Comp positions through which regular wh-chain formation may be argued to take place (leaving phonetically spelled-out copies) would therefore seem to indicate again that wh-expletive-PM questions are not formed by any movement operation parallel to that observed in syntax.
13
Considering primarily wh-expletive questions in Hindi, Srivastav (1991) suggests that a semantic process of interpretation takes the CP in which a wh-phrase occurs and substitutes it as a questionvalue into the position of a wh-expletive in an immediately dominating clause. In such an account there is consequently no LF movement of the wh-phrase out of its clause, explaining the observed lack of localit y effects and other non-parallelism with movement chains. Mahajan (1990) argues for a syntactic account of the data in Hindi, again with LF clausal raising, the approach being updated in Mahajan (1996).
14
In this sense the analysis is close to the LF pied-piping account of extraction islands containing in situ wh-phrases proposed by Nishigauchi (1986) where a mechanism of wh-feature-percolation is argued to allow for entire island constituents to be pied piped to Comp.
164
W h-movement and the theory of feature-checking
Hermon (1984). In Basque alongside regular long wh-movement examples such as (31), raising of a wh-phrase to the Comp of a lower CP may trigger movement of the whole CP to the +Q Comp of a higher clause, as shown in (32): (31)
(32)
Nor uste duzu ikusi duela Peruk? who think AUX seen has-that Peter ‘Who do you think that Peter has seen?’ [Nor etorriko d-ela] esan du Mirenek uste du-ela Peruk? [who come AUX-that said has Miren think AUX-that Peter ‘Who did Mary say that Peter thinks will come?’
Urbina’s analysis of clausal pied piping in Basque is just as Horvath proposes for Hungarian PM structures — movement of the wh-phrase to the Spec position of a (-Q) CP will allow for wh-feature-percolation to the CP node and subsequent raising of the entire CP identified as a wh-phrase. Such an indirect dependency clausal pied piping approach to PM structures clearly provides a means to account for the noted lack of localit y effects which a direct dependency chain analysis fails to predict. The Hungarian data revealed in Horvath (1997) also provides far greater insight than before into the nature and origin of the uninterpreted wh-expletives appearing in PM questions. However, there are also a number of potentially serious problems present in the analysis which ultimately suggest that a different approach is required. Significantly the important theoretical questions raised by PM structures for Minimalism remain unaddressed as Horvath chooses to cast her analysis in pre-Minimalist GB terms and it is difficult to imagine how the analysis could in fact be satisfactorily translated into the Minimalist model. In addition to this, the GB-st yle analysis would also seem to be beset by the same basic “derivational timing” difficulties which create problems for analyses of PM in a Minimalist framework. Below I will indicate first the range of difficulties facing a general indirect dependency account of PM in Hungarian and German and then continue on to show how a clausal pied piping approach will not solve the problems raised by PM constructions for the Minimalist model in particular. Three distinct problems face an indirect dependency approach to PM such as Horvath (1997) set in pre-Minimalist terms. Among these, the first which has to do with the derivational point at which wh-movement takes place is certainly the most serious and is the same aspect of PM which results in difficulties for descriptions of PM in a Minimalist approach. In order to motivate wh-movement in general, Horvath invokes Rizzi’s (1994) Wh-Criterion:
Partial Movement and wh-expletive wh questions (33)
165
THE WH-CRITERION a. A Wh-operator must be in a Spec–Head configuration with an X0 [+Wh] b. An X0 [+Wh] must be in a Spec–Head configuration with a Wh-operator.
The Wh-Criterion applying to S-structure will clearly allow for and motivate movement of a wh-expletive to the +Q Comp of a PM question. Difficulties arise however when one reflects on the movement of (a) the substantive wh-phrase, and (b) any intermediate wh-expletive. In particular one needs to ask what drives the movement of these elements and why this movement is forced to be overt. Concentrating first on the movement of the substantive wh-phrase, this is seen to raise to a Q Comp: (34)
Mitk mondtal, tk hogy [mire]i szamitanak ti a gyerekek? what-ACC said-INDEF-2SG that what-AL count-INDEF-3PL the kids-NOM ‘What did you say that the kids expected?’
Obviously the trigger for this movement cannot be the Wh-Criterion as the Wh-Criterion only allows for movement to a [+WH] Comp, so this would seem to be a problem. Horvath might suggest that in a GB model one could possibly license this movement via reference to its role in wh-feature percolation, that the movement is driven by the need for the wh-phrase to percolate its wh-features onto the CP and so allow for LF clausal pied piping. However, even if one allows that wh-movement might be permitted for reasons other than the Wh-Criterion, a second serious problem now appears, namely why it is that this movement has to be overt and prior to S-structure. Raising of the embedded clause to replace the wh-expletive in the +Q C only takes place at LF in such an analysis so it is unclear why the PM necessary to percolate the wh-phrase’s wh-features to CP is forced to occur prior to S-structure. Most naturally one would expect that the initial wh-expletive would raise overtly to the +Q Comp to satisfy the Wh-Criterion and that the substantive wh-phrase would remain in situ at S-structure, raising to the lower Comp for feature-percolation and clausal pied piping only when necessary at LF. This is essentially the most critical problem which faces all accounts of PM and it is significantly present even if one assumes a pre-Minimalist approach. Here in an indirect dependency analysis there seems to be no way to account for the observation that PM has to be overt if it is simply part of an expletive associate process which takes place only at LF. Closely related to the issue above is the problem raised by the overt raising of intermediate wh-expletives. In structures containing more than one wh-expletive,
166
W h-movement and the theory of feature-checking
an expletive in an intermediate clause is forced to undergo overt movement to a Q Comp: (35)
Mitk hitt Mari tk hogy miti akartal, ti hogy what-ACC believed Mary-NOM that what-ACC wanted-2SG that kinek telefonaljunk? who-DAT phone-1PL ‘To whom did Mary think that you wanted that we phone?’
This raising cannot be triggered by the Wh-Criterion because the intermediate Comp is not +Q/Wh and it is not obvious at all how this obligatory movement could be forced. Supposing one were again to allow that movement might be permitted in order for the wh-expletive to percolate wh-features to its containing clause as part of a successive cyclic LF clausal replacement process, it is again completely unexpected that this movement of the expletive should have to be overt. If clausal pied piping to the higher +Q Comp occurs only at LF, then the expletive movement for wh-feature percolation should also only take place at LF. Furthermore, it is also highly questionable that movement should indeed be licensed by the possibilit y of some feature-percolation operation occurring after such movement (as an LF clausal pied piping account of PM in a GB framework has to assume). If the movement of a wh-element generally could be legitimised either via the Wh-Criterion or in order to percolate features to the CP node dominating a Q C0 then it is incorrectly predicted that all languages with overt wh-movement should also show overt wh-CP raising. For example, if wh-phrases in English could raise simply to percolate their wh-features to a CP node, this should result in such CPs being identified as +Wh and allow for raising of the CP to a higher +Q Comp. However, forms such as (36) derived as in (37) are quite unacceptable. This would seem to indicate that wh-elements may not raise simply to percolate their wh-features and that the Partial Movement of wh-phrases in Horvath’s account therefore remains problematic: (36) (37)
*[Which booki Mary has bought ti]k has Jane said tk? D-Structure: Jane has said [CP [IP Mary has bought which book] → Step 1: raising to percolate wh-features: Jane has said [CP [which book]i [IP Mary has bought ti] → Step 2: wh-features percolated to CP identify CP as a wh-phrase (the CP raises to the matrix +Q Comp: [CP [Which book]i [IPMary has bought ti]]k has Jane said tk?
Partial Movement and wh-expletive wh questions
167
Thus, in short a pre-Minimalist GB account invoking LF clausal pied piping faces serious difficulties in motivating both the Partial Movement of the substantive wh-phrase and other intermediate wh-expletives and also explaining why this movement is forced to take place prior to S-structure. A second problem for the LF pied piping approach (which in fact also relates to the non-occurrence of forms such as (36) in English) is that such an account leads one to expect that overt clausal pied piping should occur in those languages with overt Partial Movement in addition to LF CP-raising. The simple reason for this is as follows — if one sets aside the problem of how to formally legitimise the Partial Movement of the real wh-phrase, and allows for it to percolate its features to the dominating CP-node, such a wh-CP should be able to undergo raising to a higher +Q Comp if there were to be no wh-expletive in this position. The lower CP would be identified as a wh-phrase and so should be forced to raise to the +Q Comp to satisfy the Wh-Criterion. Nevertheless, such overt raising of a CP containing a partially-moved wh-phrase is not allowed in Hungarian: (38)
*[(hogy) mire szamitanak a gyerekek] hitt Mari? [(that what-AL count-INDEF-3PL the kids-NOM believed Mary-NOM Intended: ‘What did Mary believe that the kids expected?’
Horvath suggests that forms such as (38) are unacceptable for PF reasons, that CP-elements are never permitted in clause-initial Comp positions in Hungarian because they are phonologically too heavy. While such an objection may be true of Hungarian, it certainly does not hold of German. In German not only may non-wh CPs occur fronted in initial position, for certain speakers it is also possible to front a wh-infinitive as in (40): (39)
(40)
[CP dass er nach Berlin fahren wollte]k hat er gar nicht [CP that he to Berlin drive wanted has he absolutely not gesagt tk said ‘He didn’t say that he wanted to drive to Berlin.’ [Wen ihr vorzustellen]k wäre äuserst unklug tk? [whom her-DAT introduce would-be extremely unwise ‘Who would it be extremely unwise to introduce to her?’
Nevertheless raising of a clause in which PM has taken place is not permitted: (41)
*[Mit wemi Johann ti gesprochen hat]k glaubt Hans tk? with whom Johann spoken has believes Hans Intended: ‘Who does Hans believe that Johann spoke with?’
168
W h-movement and the theory of feature-checking
Although it is true that Horvath develops her account based on Hungarian and argues that PM structures in German may in fact result from a quite different syntax, the goal here is to investigate whether a possible explanation of PM may allow for a general solution to the problems posed by PM. Here it can therefore be concluded that an LF clausal pied piping account is unable to explain why overt wh-CP raising is unavailable in languages which are not restricted by a PF filter of the sort existing in Hungarian and one must admit that there are languages where a clausal pied piping approach will fail to allow for an explanation of Partial Movement phenomena. A third argument against the indirect dependency proposals set out in Horvath relates to the possible combination in Hungarian of PM with a yes/no question inside a single CP. Horvath observes that when PM takes place within clauses selected by question-embedding verbs and a wh-expletive occurs in a higher +Q Comp the result is that the wh-phrase is interpreted as taking scope at this latter +Q Comp, and the embedded clause is interpreted as a yes/no question: (42)
Mit kerdeztek, hogy kivel talalkoztal-e mar? what-ACC asked-3PL that who-with met-2SG-Q already ‘With whom did they ask whether you had met?’
This seems quite unexpected under the proposed account — wh-feature percolation after PM is essentially claimed to identify a CP as a clausal wh-phrase, yet intuitively this should conflict with its interpretation as a yes/no question — any interrogative clause should be identified either as a +wh-CP or as a yes/no question but not as both simultaneously, and it would seem impossible for the same CP to receive interpretation as +interrogative at two independent +Q Comp positions (i.e the matrix +Q Comp after LF raising and the embedded +Q Comp).15 Strong evidence that this is so comes from a consideration of how overt clausal pied piping in Basque interacts with yes/no questions, wh and focus. In addition to the possibilit y of clausal raising being triggered by wh-movement to the Comp of a lower clause, Basque also allows for focus movement to trigger such pied-piping (perhaps as a result of focus feature percolation to CP):
15
Note also that if a clause identified as a wh-CP in virtue of the hypothesized feature-percolation undergoes LF raising to a higher +Q C, the wh-trace left behind by such movement should not allow for satisfaction of the selectional requirements of the question-embedding predicate at LF, just as the wh-trace left by successive cyclic movement in (i) does not satisfy parallel requirements for the verb ‘wonder’: (i) *Whoi did you wonder [ti arrived]?
Partial Movement and wh-expletive wh questions (43)
169
[JON etorriko d-ela bihar] esan diot Mireni Jon come AUX-that tomorrow said AUX Mary ‘That it is JON that will come tomorrow, I have told Mary.’
Significantly clausal pied piping with yes/no questions is possible when a focused phrase triggers the pied piping, but totally unacceptable when attempted with clause-internal raising of a wh-phrase (despite wh-clausal pied piping generally being available as a result of such wh-raising): (44)
(45)
[JON d-enentz jin] galdetu dut. Jon has-COMP come asked have ‘[Whether it is JON that has come] I have asked.’ *[Nor etorriko d-en] galdetu duzu? who come AUX-COMP asked AUX Literally: [Who will come] have you asked? Intended interpretation: Who is the x, such that you have asked whether x will come?
Basque thus clearly shows that what is suggested to take place at LF in Hungarian cannot in fact be a legitimate possibilit y — if a wh-phrase percolates its features to a CP and thereby potentially triggers raising, the CP will be fully identified as a wh-phrase and this will conflict with any other intended interpretation as a yes/no question. Pure focus feature percolation on the other hand will not give rise to any such conflict and so may allow for pied piping of yes/no question CPs. It therefore seems unlikely that there is LF clausal pied piping in Hungarian examples such as (42), and that where a wh-phrase is partially-moved in embedded yes/no questions (and indeed in all t ypes of subordinate clause) its method of interpretation is not a result of LF CP raising. Instead it might appear that the wh-phrase must establish a linking to the +Q Comp which is in some way direct — yet at the same time not involving movement or a chain attributed the properties of such.16
16
A fourth (less critical) reason to be suspicious of the GB-cast LF pied piping analysis of PM structures proposed in Horvath concerns the forms that answers to PM questions commonly take. In his influential work on wh in situ in Japanese, Nishigauchi (1986) suggests that where a sequence containing a wh-phrase is pied pied to a +Q Comp at LF, a natural answer-form will consist in a repetition of all the material in the raised XP together with a value for the questioned wh-element. If PM structures do indeed give rise to LF CP-raising, this might lead one to anticipate that a normal answer-form to such questions would consist in a repetition of the entire CP along with a value for the wh-phrase, especially where the CP is a subject, adjunct or factive island. However, answer-forms of this t ype as in (ii) below are certainly not forced with PM
170
W h-movement and the theory of feature-checking
Having considered the general difficulties which the patterns in Partial Movement create for a purely GB-st yle LF pied piping analysis and concluded that various of the key identifying properties of PM remain quite unaccounted for in such an approach, I will now point out how these aspects of PM are also extremely problematic for any Minimalist account based on the idea of an indirect dependency. In a footnote in her paper, Horvath hints that the basic analysis she develops might be translated into a Minimalist account with certain modification: The claims and analysis presented in this paper have been formulated largely prior to and independently of Chomsky’s (1995, ch. 4) Minimalist Program, and some of the mechanisms assumed in it are unavailable within the latter framework, such as feature-percolation and covert CP-movement (i.e. LF pied-piping). But what is important to note is that the major points and conclusions reached here hold regardless of whether this analysis is recast in strictly minimalist terms… (Horvath 1997, p. 513, fn. 3) However, it is actually very difficult to imagine how the analysis could be recast in Minimalist terms as PM has already been noted to pose a significant theoretical challenge to basic assumptions underlying movement and its motivations in the Minimalist Program and an indirect dependency approach of the kind suggested here fares no better in allowing a solution to these problems. It may also be noted that earlier circulated drafts of Horvath’s paper did in fact spend considerable time and effort in attempting to find a way to reconcile the LF pied piping analysis with Minimalist assumptions. That Horvath eventually relinquished such attempts and re-set the account in a GB framework can therefore be taken as indication of the severe difficulties that PM constructions pose for Minimalism and that an indirect dependency approach is not the appropriate vehicle for accommodating PM with Minimalism. The critical problem which PM poses for a Minimalist approach is, as mentioned in Section 3.2.2, quite simply how to account for the Partial Movement of
questions such as (i) and in fact sound rather odd: (i) Was glaubst du, wen er gesehen hat? WH believe you whom he seen has Who do you think he saw? (ii) Er hat Johann gesehen. He has seen Johann. This is clearly rather unexpected under a view in which the whole CP containing the partiallymoved wh-phrase appears raised in the +Q Comp at LF.
Partial Movement and wh-expletive wh questions
171
the real wh-phrase (and any intermediate wh-expletive). Although it may be assumed that the movement of an initial wh-expletive to the +Q Comp is triggered by the need to check strong wh-features on the +Q C0, there is no apparent way to allow for obligatory overt Partial Movement of a wh-phrase to a Q Comp within current Minimalist assumptions. Chomsky (1995) explicitly argues that movement may only take place for the immediate checking of some featurespecification and that feature-checking relations may only be effected within the strict locality of Spec–Head or Head-adjoined relations. Here in PM constructions the forced raising of the substantive wh-phrase should result in the checking of a set of features in the position which the wh-phrase moves to; however this position is a Q Comp and so will not allow for the checking of any wh-features carried by the wh-phrase. The problem is essentially very simple but nonetheless extremely important, and nothing in an indirect dependency analysis will allow for an explanation of this troublesome signature propert y of PM questions. Supposing nevertheless one were somehow to ignore this major problem of PM and consider the remainder of an LF CP-raising analysis from a Minimalist perspective, it soon becomes apparent that once again one is faced with the recurring problem of derivational timing. If one somehow allows for raising of the real wh-phrase to the Q Comp for percolation of its wh-features and subsequent LF raising of the CP, the question arises as to why the initial raising of the wh-phrase for feature-percolation should have to take place prior to Spell-Out. Given that any wh-features possibly in need of checking on the wh-phrase would have to be weak (as the wh-phrase does not occur in any standard checking relation with the +Q Comp at Spell-Out) and only checked by the hypothetical CP-raising, PM to percolate wh-features to the CP-node should also only take place after Spell-Out.17 Essentially all ways of viewing PM structures consistently seem to indicate that some kind of licensing must take place in the pre-Spell-Out portion of the derivation. How this might actually be effected in a somewhat modified Minimalist model we will shortly consider in Section 3.5.1. First however, having discounted both standard Direct and Indirect Dependency accounts as untenable for various reasons, a final means to save an orthodox
17
Added to this is of course the problem recognised by Horvath that pied piping is essentially viewed as a PF-driven phenomenon in Chomsky (1995) and one might not anticipate large-scale LF CP raising in current Minimalist approaches. An exception to this might be in cases where LF pied piping is forced to take place for reasons of convergence, as for example with Nishigauchi’s island cases. However, Partial Movement, which for Horvath indicates LF clausal pied piping, is clearly found in instances where there are also no barriers to movement; here it is therefore not possible to invoke any considerations of convergence for justification of a pied piping analysis.
172
W h-movement and the theory of feature-checking
Minimalist treatment of PM will be examined — the possibilit y that features other than wh are checked by the raising of the wh-phrase in PM constructions.
3.4. 3.4.1. Partial Movement as focus movement The critical problem raised by PM structures for a Minimalist approach is as already stated that movement brings a wh-phrase to an intermediate Q Comp position in which it is impossible for it to check its wh-features if feature-checking is restricted to the strict locality outlined in Chomsky (1995). One seems therefore to be faced with an instance where movement is not driven by a need to check features, contra Chomsky’s (1995) principle of Last Resort. However, such a conclusion is made under the assumption that it is significantly wh-features which are failing to be checked by the PM of the real wh-phrase (because the relevant checking head, the +Q C0, is in a higher clause). Supposing there were however to be other morphological features carried by the wh-phrase which could reasonably be claimed to engage in a checking relation with the intermediate Q C0, then it might be possible to avoid the conclusion that principle of Last Resort is violated. One could simply state that PM takes place for other non-wh reasons and a fully regular local checking relation is indeed established between the intermediate C0 and the wh-element in its Spec position. One obvious possibilit y to explore here is that PM of a wh-phrase might take place to check focus features present in the C0-head, while another route explored in Fanselow and Mahajan (1996) specifically for German is that the wh-phrase checks categorial D-features present in C0. As both possibilities essentially have to confront the same basic counter-arguments and data, in what follows I will primarily just consider the case for focus-feature checking and note that D-feature checking proposals effectively pattern in a largely parallel way. Looking at both Hungarian and German one finds that wh-phrases consistently undergo raising to a clause-initial position into which other t ypes of focused elements may also move; wh and focused phrases thus appear to be attracted to a single functional head. Given this common patterning, it might be suggested that the Partial wh Movement which is found in embedded contexts is actually a case where a wh-phrase is attracted to an embedded focus head to check focus-features, similar to the attraction of non-wh focused DPs to the same embedded position in Hungarian. If such a hypothesis could be maintained it would clearly solve one
Partial Movement and wh-expletive wh questions
173
of the most difficult puzzles of PM structures and allow for a number of fairly straightforward accounts all compatible with general Minimalist principles. Considering only German, Fanselow and Mahajan (1996) propose a similar non- wh-related solution for Partial wh Movement; noting that the Comp position in German must always be filled by some overt element, either in the C0-head position or its Spec, they suggest that a wh-phrase is raised to Comp in PM structures simply to fulfil this condition and that formally the C0-head is attracting an element with D0-categorial features not wh-features. In such an account it is then essentially coincidental that what happens to satisfy this D0-feature checking in C0 is a wh-phrase, and other elements with D0-features could also have carried out the necessary D0-checking. Although such hypothetical accounts do have an initial plausibility, there are good reasons to reject any potential explanations of PM which suggest that the movement of the real wh-phrase is disconnected from its wh-status and triggered by other non-wh-related factors. The first of these relates to the fact that in the majority Dialect A of German one finds that some kind of linking via wh elements of all intermediate Comp positions between the partially-moved wh-phrase and the actual +Q Comp must be effected by Spell-Out, and sentences where a non-wh element is found in such an intermediate Comp position are considered to be ungrammatical. This clearly indicates that Partial wh-Movement is indeed wh-related movement and that a specifically wh-linking to the +Q Comp must be established when PM takes place. Were PM simply to take place for focus reasons or to check D0-features in C0 one should expect that sentences such as (46) would be well-formed. This is not so and they are quite unacceptable for speakers of Dialect A: (46)
*Was glaubst du [dass Hans meint [mit wemi Jakob ti WH believe you [that Hans says [with whom Jakob gesprochen hat]]? spoken has ‘Who do you believe that Hans says that Jakob spoke with?’
Significantly, an acceptable form of (46) can only result from either the inclusion of a second wh-expletive in the intermediate Comp as in (47), or from further movement of the wh-phrase higher to this position as in (48): (47)
Was glaubst du [was Hans meint [mit wemi Jakob ti WH believe you [WH Hans says [with whom Jakob gesprochen hat]]? spoken has ‘Who do you believe that Hans says that Jakob spoke with?’
174 (48)
W h-movement and the theory of feature-checking Was glaubst du [mit wemi Hans meint [ti dass Jakob ti WH believe you [with whom Hans says that Jakob gesprochen hat]]? spoken has ‘Who do you believe that Hans says that Jakob spoke with?’
Supposing that PM were to take place to check focus-features present on a wh-phrase, it should be possible for these focus-features to be checked in any potentially focus-related C0 head and so in the SpecCP position of the lowest clause (noting that a focus account of PM has to assume that focus-features are indeed checked in this position by the wh-phrase in (47)). However, in the absence of an intermediate wh-expletive, the real wh-phrase is forced to move to the intermediate Comp position itself. The necessit y of this further PM has no explanation if focus-features could be checked in the lower Comp position, i.e. there is no reason why focus-checking should be forced to occur in the intermediate Comp in (48) when it should be legitimately possible in the lowest Comp. Considering (47), the fact that a second intermediate wh-expletive is necessary when the wh-phrase is raised only to the lower clausal Comp similarly indicates that Partial Movement of the wh-phrase is not for focus-checking. Hypothetical focus-checking of the wh-phrase should be satisfied by raising to the lower Comp alone and this should not call for any “additions” to other parts of the structure; crucially that it is the inclusion of an additional wh-element which renders (46) acceptable (in (48)) is clear evidence that it is rather some wh-related licensing condition which is in need of satisfaction when Partial Movement of a wh-phrase occurs, noting further that no such intermediate wh-elements are necessary where Partial Movement does not take place and a wh-phrase remains in situ in regular multiple-wh questions: (49)
Wer glaubt, dass Hans meint, dass Jakob mit wem gesprochen who believes that Hans thinks that Jakob with whom spoken hat? has ‘Who believes Hans thinks Jakob spoke with who?’
Thus Partial Movement is essentially wh-oriented, requiring a “successive cyclic” wh-linking to the +Q Comp after its application, and the Partial Movement of a real wh-phrase is itself directly driven to a particular Comp by the need to establish
Partial Movement and wh-expletive wh questions
175
such a linking. A focus-raising analysis simply cannot account for such facts.18 The focus account of PM also fails for a second very basic reason. If wh-phrases may undergo pre-Spell-Out raising to a Q Comp for checking of focus-features in certain instances, notably those observed where a wh-expletive occurs in the +Q Comp (by hypothesis satisfying a strong wh-operator feature in C0), then such focus-raising of wh-phrases should also generally be available where other relevant conditions are satisfied, such as (non-expletive) multiple wh-questions. Considering multiple wh-questions in Hungarian in a standard Minimalist approach, movement of a single wh-phrase to the +Q Comp can be taken to satisfy strong wh-features in Comp, with the result that secondary wh-phrases can remain in situ: (50)
Kinek mondta [hogy Janos talalkozott melyik lannyal]? who-NOM said [that Janos met which girl-with ‘Who said that Janos met with which girl?’
This obviously indicates that (secondary) wh-phrases do not always need to undergo any (pre-Spell-Out) “focus”-movement. However, the focus account does clearly predict that such focus-movement, if only optional, should at least be possible for the secondary wh-phrase, taking it to the Partially-Moved Comp/focus position of the lower clause. Contra expectation this is apparently not a possibility: (51)
*Kinek mondta [hogy melyik lannyali talalkozott Janos ti]? who-NOM said [that which girl-with met Janos
Were PM not to be triggered by some wh-licensing strategy but to take place purely for focus reasons, there is no reason why it should not also be possible in sentences such as (51). The generalisation is that PM can only occur in Hungarian when a specific t ype of wh-element appears in the +Q Comp (a wh-expletive), hence again that PM is indeed wh-related (in some way) and does not take place for wh-independent focus reasons.19
18
A D0-feature-checking analysis will also not be able to account for the fact that sentences such as (46) are unacceptable for speakers of Dialect A. The element dass ‘that’ in the intermediate Comp in (46) should serve to satisfy the D0-feature-checking requirements of this position and the sentence is expected to be fully grammatical. In their treatment of PM in German, Fanselow and Mahajan (1996) only consider data from Dialect B speakers and so leave the more problematic Dialect A forms unaccounted for.
19
The same facts also hold for the majorit y of German-speakers. However, McDaniel (1989) suggests that certain speakers do also allow for PM to occur where a full wh-phrase rather than a wh-expletive appears in the +Q Comp:
176
W h-movement and the theory of feature-checking
Also one should add that not only may PM take place when a wh-expletive occurs in the +Q Comp, it is also forced to take place in these instances, and sentences such as (52) and (53) are fully unacceptable for all speakers of German and Hungarian (i.e. both Dialects A and B of German): (52)
(53)
*Mit gondolsz, hogy Janos latott kit? WH think-3SG that John-NOM saw-3SG who-ACC Intended: ‘Who do you think that John saw?’ *Was glaubst du, dass Johann mit wem gesprochen hat? WH believe you that Johann with whom spoken has Intended: ‘Who do you think that Johnann talked to?’
Given the fact that secondary wh-phrases may occur fully in situ in both Hungarian and German, hence that such elements need not always raise for focuschecking (under a focus account of PM), one would then have to state that focuschecking (or D0-feature-checking) is nevertheless forced to occur just in those cases where a wh-expletive appears, a stipulation which it would seem impossible to justify. Finally it can be added that a focus-raising account of PM would also be at odds with other facts of focus in German. Although the matrix SpecCP position in German is indeed arguably a position in which general focus-features may be checked, it is actually not possible to focus non-wh elements in embedded Comp positions:
(i)
Weri ti glaubt [wenk Hans tk bestochen hat]? who believes [whom Hans bribed has ‘Who believes Hans bribed whom?’ This might not seem to be focus-movement either though, as the partially-moved wh-phrase must necessarily raise to a Comp position in which it is one clause down from the +Q Comp or linked to this +Q Comp by wh-elements in all intervening Comps, i.e. PM is not free to occur to any position where focus-features could hypothetically be checked but is constrained by the need to establish a wh-linking to the +Q Comp: (ii) *Weri ti glaubt [dass ich meinte [mit wemk Jakob tk gesprochen hat]]? who believes [that I thought [with whom Jakob spoken has (iii) Weri ti glaubt [was ich meinte [mit wemk Jakob tk gesprochen hat]]? who believes [WH I thought [with whom Jakob spoken has ‘Who believes I thought that Jakob spoke to whom?’ (iv) Weri ti glaubt [mit wemk ich meinte [tk dass Jakob tk gesprochen hat]]? who believes [with whom I thought that Jakob spoken has ‘Who believes I thought that Jakob spoke to whom?’ It must however be added that the grammaticalit y of even sentences such as (i) has been questioned and their precise status and interpretation is rather unclear.
Partial Movement and wh-expletive wh questions (54)
177
*Johann glaubt, Martin wir gesehen haben Johann believes Martin we seen have Intended: ‘Johann believes that it is Martin that we saw.’
If Partial wh-Movement to an embedded Q Comp occurs for general focuschecking, it is very unclear why it should only be wh elements which allow for this kind of focus-checking. Again it seems clear that it is precisely the wh specification of wh-phrases and not any general focus or D0-features which licenses and triggers their raising in wh-expletive questions. Such simple observations as the above then force one to discount any possibilit y of accommodating Partial Movement into a standard Minimalist model with the claim that PM does in fact lead to an occurrence of strictly-local non-wh featurechecking between the wh-phrase and the Q C0. Instead it seems evident that some other mode of explanation needs to be invoked and a modification of existing Minimalist assumptions would appear to be required. In concluding this sub-section I would also like to add a few further words on the dialect split in German referred to in the text, as the first argument in this section clearly refers to speakers of Dialect A who do not accept examples such as (46). As mentioned earlier, the assumption made here (and supported by a variet y of argumentation in a later section) is that Dialects A and B are in fact only superficially different, Dialect B (and also Hungarian) simply allowing for intermediate wh-expletive deletion. If this is correct, then arguments referring specifically to Dialect A are actually also relevant to Dialect B and to Hungarian. Dialect A is used here for illustration because it is practically less complicated to prove the relevant points with speakers who do not allow for deletion of intermediate expletives. Supposing however that the assumption that Dialects A and B are basically the same could somehow be proved to be incorrect and Dialect B could be shown not to involve intermediate was-deletion, this might make a Minimalist-compatible focus/D-feature account more accessible for Dialect B (though the rest of the argumentation presented does not rely on the dialect split evidence and also argues strongly against a focus/D-feature analysis). Such a focus/D-feature account of Dialect B will however still leave the patterning found with the large number of Dialect A speakers fully unexplained and severely problematic for a regular Minimalist approach. Thus even if one were to be able to provide an account of Dialect B taking it to be fundamentally different from Dialect A in some way, this significantly will not solve the general theoretical problems outlined here in the chapter.20
20
Furthermore, if an account of the more problematic Dialect A can be successfully developed, this should automatically and easily extend to cover Dialect B and Hungarian.
178
W h-movement and the theory of feature-checking
3.4.2. Aside: overt clausal pied piping — a possible explanation Before attempting to develop an approach to PM which will incorporate all of the patterning examined so far, I would like to briefly re-consider why it is that certain languages but not others allow for overt pied piping of wh-clauses and suggest that this may in fact be accounted for in terms of focus-checking. What is essentially in need of explanation is why Basque (55) is possible whereas English (56) is not: (55)
(56)
[Nor etorriko d-ela bihar] esan diozu Mireni? [who come AUX-that tomorrow said AUX Mary ‘Who did Mary say will come tomorrow?’ *[Whati did John buy ti]k has Bill claimed tk?
Webelhuth (1989) points out that while wh-feature percolation upwards from a specifier to a dominating maximal projection is a general possibilit y across category t ypes such as N, A and Adv, giving rise to pied piping-t ype wh-movement to Comp as in (57)–(59), for some reason this would not seem to be possible when a wh-phrase occurs in the Spec of C0, as indicated by the ungrammaticalit y of (56): (57) (58) (59)
[Which book] did you buy? [How big] was the part y? [How quickly] did he run?
Basque (and Quechua) nevertheless do appear to allow for just this possibilit y, so it clearly should not be ruled out by any invariant universal principle, in addition to which there seems to be no obvious reason why wh-percolation from SpecCP should not be permitted. I would like to suggest that what critically distinguishes Basque and Quechua from English and German etc is that in Basque-t ype languages wh-phrases may carry focus and wh-features which are formally distinct from each other. While wh-phrases cross-linguistically are generally argued to constitute a t ype of focused information, it is quite possible that in many/most languages this does not result or translate into there being two separate sets of features carried by wh-phrases corresponding independently to wh and focus. Were there to be languages however in which the wh-phrases did contain both focus and wh-features formally independent of each other, it is possible that such features might allow checking in different loci, this giving rise to a potentially different patterning from languages in which wh and (its associated) focus form a single composite feature-set. If one supposes that this is indeed the case in Basque and Quechua it may be possible to explain cross-linguistic variation with regard to clausal pied piping in
Partial Movement and wh-expletive wh questions
179
the following way. In Basque a wh-phrase may raise to Comp to check focus features against C0 without the wh-features of the wh-phrase also necessarily having to enter into a checking relation with any potential +interrogative features on C0. From the Spec of C0 position, wh-feature percolation to the dominating CP may then take place resulting in the CP being identified as a wh-phrase, this in turn triggering raising of the CP to a higher +Q Comp (i.e. overt clausal pied piping). In English/German focus-feature checking may not take place independently of wh-feature checking as focus/wh-features comprise a single feature set; movement of a wh-phrase to a Comp position therefore cannot be triggered by checking of a focus specification if the wh-features are not also simultaneously checked. As a result, English/German will not allow wh-raising to the Spec of a Q Comp position from which wh-percolation to CP theoretically could apply, and wh-clausal pied piping will not occur.21, 22 If such an account is headed in the right direction, one would expect that wh-phrases in Basque might raise to a Q Comp without this necessarily triggering pied piping (to a higher +Q Comp), i.e. if the focus-features of wh-phrases in Basque may be checked independently of wh-features and if a strong operator feature in the +Q Comp is checked by some other wh-phrase, it should be possible that a secondary wh-phrase undergoes raising to a lower Q Comp for focusfeature checking alone. This prediction is indeed borne out as (60) shows: (60)
Zein kazetaria esan du [zein legea apurtu du ela Major]? which journalist say-AUX [which law broke AUX Major ‘Which journalist said Major broke which law?’
21
Wh-movement to a Q Comp may however occur if wh-features on the wh-phrase are checked in this position, as will shortly be argued is the case in PM structures in German, and if the wh-features are indeed checked in such a position, there will consequently be no trigger for clausal pied piping (further wh-raising for feature-checking), so none will be attested.
22
For some speakers of English a limited amount of wh-clausal raising is in fact possible, as in (i) below: #[Who left] do you think? (i) Significantly this might seem to be restricted to clauses in which the wh-phrase is a subject (compare (i) with (56) above where the wh-phrase is a raised object). In (i) the wh-phrase will have been raised to the SpecIP position to satisfy the EPP; speakers who accept such forms may then allow for the subject’s wh-features either to percolate from here to the higher dominating CP-node (these positions not being separated by any lexical material) or alternatively allow for wh-IP raising as a result of simple wh-feature percolation from SpecIP to the immediately dominating IP-node. This kind of natural feature-percolation from a high clausal position will not be available with wh-objects base-generated in lower clause-internal positions, and there is nothing in a higher position to legitimately attract them and feed the process of feature-percolation to IP/CP.
180
W h-movement and the theory of feature-checking
Basque thus shows evidence of allowing for the pure focus-feature checking of wh-phrases which may not occur in Hungarian (as argued above relative to examples such as (51)).23 Webelhuth’s original generalisation concerning pied piping was that featurepercolation is broadly possible from a Spec to its immediately dominating XP. The “gap in the paradigm”, that a wh-phrase in a Q SpecCP seemingly may not percolate wh-features upwards to CP resulting in clausal pied piping, can now be explained by the fact that raising to such a SpecCP position is not permitted or triggered in the first place when the wh-phrase does not check features in this position. In all other instances of wh-percolation and pied piping in English and German, the wh-element occurs base-generated in the relevant Spec position, so raising to this position does not have to be justified/triggered.24 Thus ultimately it may in fact be possible to maintain in full the generalisation that (wh-)features may be percolated from Spec positions (and allow for pied piping of the dominating XP), explaining the apparent lack of wh-CP raising in English/German in terms of the lack of a trigger for the necessary initial movement to a Q Comp.
23
Bahasa Indonesia is another language which arguably allows for focus-features on a wh-phrase to be checked independently of wh-features — see Saddy (1991) and Chapter 1 for examples in which a wh-phrase raises to a focus position which is not a +Q Comp. Note that although one might expect that this should allow for overt wh-CP raising as in Basque, this is not however found for the simple reason that the wh-features in a +Q Comp in Bahasa Indonesia are not strong and so do not attract any wh-element overtly.
24
Note that in Hungarian DP-internal movement of a wh-element to SpecDP may result in pied piping of the DP to a +Q Comp: (i) [Kineki a ti feleseget]k lattad tk? [who-DAT the wife-ACC saw-2SG ‘Whose wife did you see?’ Again it may be argued, as with Basque “wh-movement” to a Q Comp, that the initial movement to SpecDP is triggered by a non-wh feature-checking relation being satisfied between the element in SpecDP and D0, such movement also occurring DP-internally with non-wh elements. Having raised to SpecDP, kinek ‘who’ may then percolate its wh-features to DP resulting in pied piping to Comp.
Partial Movement and wh-expletive wh questions
181
3.5. 3.5.1. Towards a solution In the preceding sections the following conclusions have been reached: I. There are significant problems with direct dependency approaches to PM which assume LF movement of the substantive wh-phrase (or an equivalence to movement). II. There are also significant problems with indirect dependency approaches which assume LF clausal pied piping. III. Partial Movement is related to the licensing of specifically wh-elements and cannot be explained in terms of the checking of any non-wh feature specification. A further important conclusion which can be made from the patterning found is that the licensing effect of Partial Movement (whatever it is and however it is related to wh) critically must be established prior to Spell-Out. This results from two related fairly simple observations. The first of these is that PM is obligatory overt movement and the real wh-phrase cannot remain in situ in wh-expletive questions: (61)
*Was glaubst du, dass Johann mit wem gesprochen hat? WH believe you that Johann with whom spoken has Intended: Who do you think that Johan spoke to?
For Chomsky (1995) this should therefore lead to the immediate pre-Spell-Out checking of some set of features on the partially-moved wh-phrase. Indeed for any approach the fact that PM is forced to be overt should be tied to some pre-SpellOut/pre-S-structure licensing requirement. A second relevant observation concerns the contrast in German Dialect A between multi-clausal forms such as (48) and (46) repeated here below: (48)
Was glaubst du [mit wemi Hans meint [ti dass Jakob ti WH believe you [with whom Hans says that Jakob gesprochen hat]]? spoken has ‘Who do you believe that Hans says that Jakob spoke with?’
182 (46)
W h-movement and the theory of feature-checking *Was glaubst du [dass Hans meint [mit wemi Jakob ti WH believe you [that Hans says [with whom Jakob gesprochen hat]]? spoken has ‘Who do you believe that Hans says that Jakob spoke with?’
In example (48), PM of the substantive wh-phrase to the intermediate Comp results in a well-formed configuration which is fully grammatical. If the licensing effect created by PM were to relate to a propert y in need of satisfaction only by LF, one might expect that the wh-phrase in example (46) could raise to the intermediate Comp after Spell-Out and so give rise to an acceptable form at LF paralleling (48). Noting that a wh-phrase such as mit wem can indeed raise overtly between these positions (as in (48)) and no obvious barrier exists to block similar movement at LF; from the unacceptabilit y of (46) it would therefore seem that raising of the wh-phrase to the intermediate Comp importantly has to be made prior to Spell-Out.25 Combining the conclusion that PM results in some pre-Spell-Out licensing with the observation that operations of movement clearly are involved leads to the assumption that PM results in an application of obligatory pre-Spell-Out feature-checking. Building in propert y III above that it is specifically wh-elements which are necessarily involved in PM questions now allows for the further important conclusion that Partial Movement takes place in order that wh-features are checked prior to Spell-Out: (62)
Partial wh Movement takes place for the obligatory pre-Spell-Out checking of wh-features.
This unavoidable conclusion forced by the patterning found in PM questions has as a consequence that the wh-phrase undergoing PM must in fact be establishing a wh-feature checking relation by its overt movement. As this movement however brings it to a Q Comp, and as the only head which is able to check its wh-features is located in a higher clause, one appears forced to concede precisely as argued in Chapter 2 that a feature-checking relation may indeed exist between
25
As pointed out already in footnote 9, it is not possible to suggest that the tensedness of the intermediate clause in (46) somehow blocks LF movement because what moves at LF is simply features and not full categories. Pied piping of the PP category containing the wh-features should be licensed at LF by the need for convergence and so permit the effective construction of a form parallel to (48) from (46). That examples such as (46) are not acceptable clearly indicates that Partial Movement takes place to satisfy some pre-Spell-Out licensing requirement.
Partial Movement and wh-expletive wh questions
183
two elements which do not occur in the strict localit y of a Spec–Head relation, and that the Checking Uniformit y Hypothesis is therefore not a universal constraint. Here in PM questions an overt (wh-)checking relation must be taken to somehow be effected between the partially-moved wh-phrase in an intermediate Q Comp and the +wh+Q Comp of a higher dominating clause. Reflecting briefly back on Chapter 2, there it was suggested that in “Englisht ype” languages such as German where there exists a difference in behaviour between primary and secondary wh-phrases, the wh-checking localit y essentially corresponds to the entire sentence-internal m-command domain of the +wh+Q Comp. This propert y which results from the instantiation of C0 in English, German and other similar languages allows for secondary wh-phrases to remain in situ and be licensed in these positions. This being so, the conclusion that the substantive wh-phrase in PM constructions has its wh-features checked “at a distance” by the +wh+Q Comp in the dominating clause is not in itself a problem in any way, and the system developed so far indeed allows for a non-local checking relation of this t ype to exist (i.e. wh-licensing at a distance is indeed the norm in languages like German). The fact that PM constructions quite independently lead to the same conclusion that a checking relation may be effected between two non-local positions would in fact seem to add strong support to the proposals and claims laid out in Chapter 2. Consequently, although there is more which remains to be added and explained here, one very important problematic aspect of Partial Movement is actually not a problem for the current approach. While there appears to be no way that a standard Minimalist approach can accommodate the necessary conclusion that PM results in pre-Spell-Out checking of wh-features on the wh-phrase in a Q Comp, the general account of checking relations developed earlier in Chapter 2 is able to incorporate this patterning with very little difficult y. What is different and in need of explanation in wh-expletive questions is why there should in fact be any need for the substantive wh-phrase to undergo partial movement. One might naturally expect that raising of the wh-expletive to the initial Comp would succeed in triggering the C0 as +wh+Q and the real wh-phrase could be wh-feature-checked fully in situ just like other secondary wh-elements in multiple wh-questions. The unacceptabilit y of examples like (61) above indicates however that this is not so and therefore demands some kind of additional explanation.
184
W h-movement and the theory of feature-checking
3.5.2. Partial Movement and tense In order to get closer to some understanding of why it is that the real wh-phrase undergoes PM in wh-expletive questions (and only in such question types), I would first like to look more closely at the interaction of tense and PM, and at certain tense-related restrictions on where the real wh-phrase must occur relative to the +Q Comp. Partial Movement structures in German can be shown to be significantly sensitive to the notion of a “tense domain” in a way which is similar to the tense domain constraint on wh-licensing in Iraqi Arabic and Hindi. Initially in the introduction to general properties of PM presented in Section 3.2.1 it was stated that in PM structures some wh-element (a wh-expletive or wh-trace) must occupy each and every Comp position intervening between the wh-expletive in the +Q Comp and the partially-moved wh-phrase. In German this is actually not strictly true and the facts are somewhat more complicated. Essentially it is the Comp of every tense domain intervening between the highest tense domain containing the +Q Comp and the lowest clause containing the real wh-phrase that must be filled by some wh-element. In German a tense domain comprises a tensed clause and any additional dependent clauses (complement clauses selected by the tensed verb) that are either non-finite or “tense-dependent”, where a tense-dependent clause is one in which the tense of the verb in this clause is dependent or conditioned/ determined by the tense of the verb in the higher clause; verbs such as wollen ‘want’ induce tense-dependency, while glauben ‘believe’, sagen ‘say’ etc do not. In (63) below, the intermediate clause CP2 will form a single tense domain with the matrix CP1, the tense on sage ‘say’ being dictated by the verb in CP1; CP3 by way of contrast will constitute a tense domain independent of CP1/CP2, tense on gekommen ist ‘has come’ not being uniquely determined by the verb which selects this clause (sagen-say): (63)
[CP1 Was will Hans [CP2 dass ich sage [CP3 warum er zu spät [CP1 WH wants Hans [CP2 that I say [CP3 why he too late gekommen ist]]] come has ‘Why does Hans want me to say that he came too late?’
In a structure of this sort, where CP2 comprises a single tense domain with CP1, it has been observed in McDaniel (1989) that an intermediate wh-expletive need not fill the CP2 Comp position linking the wh-expletive in CP1 to the partiallymoved wh-phrase in CP3, though, as shown earlier, where CP2 is “tense indepen-
Partial Movement and wh-expletive wh questions
185
dent” of CP1 a wh-expletive must be present (for Dialect A speakers). Furthermore, not only need such a wh-expletive not occur in SpecCP2 of (63), this Comp may not be occupied by a wh-expletive either: (64)
*?[CP1 Was will Hans [CP2 was ich sage [CP3 warumi er ti zu spät [CP1 WH wants Hans [CP2 WH I say why he too late gekommen ist]]]? come has
Thus the positions in which a partially-moved wh-phrase and a wh-expletive may occur are observed to be directly dependent on this notion of tense domain.26 The interaction of tense and PM can further be seen below in different form; where an embedded CP is tense-independent of the embedding predicate, the PM + wh-expletive strategy may be employed (65), but where the lower CP is tensedependent (66) (or an infinitive) it may not: (65)
(66)
Was glaubst du, wasi er ti gekauft hat? WH believe you what he bought has ‘What do you believe he bought?’ *Was willst du, weni er ti besticht? WH want you whom he bribes Intended: ‘Who do you want him to bribe?’
There is therefore some kind of anti-localit y effect here which again relates to the notions of tense-dependency and tense domain. (66) shows that PM may not cooccur with a wh-expletive within a single tense domain, while (63) and (64) indicate that a wh-expletive may only occur when actually linking genuine adjacent tense domains. A related tense-sensitivit y is also seen in (46) repeated here again for convenience. For speakers of Dialect A a wh-element must occur in the Comp of every
26
Although Hungarian does not appear to make any distinction between tense-dependent and tenseindependent finite verbs, PM in this language can still be argued to be sensitive to the notion of tense domain, as in German. A tense domain in Hungarian will comprise a tensed clause and any dependent non-finite clauses, with all the basic patterning observed in German also being present relative to these tense domains. For example, the Comp of every tense domain between that containing the +Q Comp and the tense domain where the real wh-phrase occurs partially-moved is commonly filled by a wh-expletive prior to Spell-Out (though this is sometimes partially masked by optional PF deletion as discussed in a later section), and wh-expletives need and may not occur in the Comp of intermediate non-finite clauses, just as they may not occur in the Comp of tensedependent clauses in German (despite the fact that wh-phrases elsewhere may indeed occur in the +Q Comp of non-finite clauses in Hungarian).
186
W h-movement and the theory of feature-checking
tense domain between the +Q Comp and the tense domain containing the partially-moved real wh-phrase. Whereas examples such as (63) show that the Comp position of intermediate tense-dependent clauses need and must not be occupied by a wh-expletive, when such an intermediate clause is independently tensed and so an independent tense domain, its Comp must be filled with a second wh-expletive as in (47): (46)
(47)
*Was glaubst du [dass Hans meint [mit wemi Jakob ti WH believe you [that Hans says [with whom Jakob gesprochen hat]]? spoken has Intended: ‘Who do you believe that Hans says that Jakob spoke with?’ Was glaubst du [was Hans meint [mit wemi Jakob ti WH believe you [WH Hans says [with whom Jakob gesprochen hat]]? spoken has ‘Who do you believe that Hans says that Jakob spoke with?’
This indicates that the wh-dependency between the +wh+Q Comp and the partially-moved wh-phrase cannot cross an intervening independent tense domain unless there is an additional wh-expletive present there to assist in this. Whexpletives then essentially seem to perform some kind of tense-related linking function here, between a wh-phrase partially-moved to the Comp of one tense domain through the Comp(s) of any intermediate tense domain(s) to the +Q Comp in the highest tense domain. This tense domain sensitivit y found in PM questions naturally recalls tense constraints on wh-questions in Iraqi Arabic and Hindi. Common to both PM and wh-questions in Iraqi Arabic/Hindi are the following properties: (a) the necessary relation of the wh-phrase to the +wh+Q Comp is directly defined in terms of a tense domain (in Iraqi Arabic and Hindi wh-phrases having to occur within the tense domain of the +Q Comp), (b) wh-features in all cases require checking prior to Spell-Out, (c) checking of wh-features takes place “at a distance” and not in any strictly-local Spec–Head configuration, and (d) overt movement of a wh-phrase which occurs too distant from the +Q Comp in tense domain terms will rescue an ill-formed structure.27 The existence of such a set of commonly-
27
So, for example, a wh-phrase base-generated in an embedded tensed clause in Iraqi Arabic and Hindi will raise up to the tense domain of the +Q Comp. Similarly in German the substantive wh-phrase in a wh-expletive question separated from the +Q Comp by an intervening tense
Partial Movement and wh-expletive wh questions
187
shared properties raises the possibilit y that the general mode of explanation invoked in the Iraqi Arabic and Hindi cases might also be appropriate here for Partial Movement and lead to an insight into the tense-sensitive movement involved in the wh-expletive questions examined here. Reflecting back on the account offered for Iraqi Arabic and Hindi in Chapter 2, it was suggested that the patterning found in these languages essentially resulted from the particular lexical instantiation of a +wh+Q C0 in Iraqi Arabic and Hindi, and that the +wh+Q C0 available and inserted from the lexicon in Iraqi Arabic and Hindi projects a specific tense-defined licensing localit y. This +wh+Q C0 was then argued to contrast with two other potential lexical instantiations of a +wh+Q C0 with different associated licensing domains found cross-linguistically in English- and Romanian-t ype languages.28 The proposal to account for Iraqi Arabic and Hindi and the wh/tense patterns found there with the suggestion that there is certain limited variation in the possible lexical instantiations of a +wh+Q Comp can now be suggested to offer an appropriate means to approach the similar wh-tense interactions in PM constructions. One way of describing a relevant parallelism between Iraqi Arabic/Hindi and PM structures in German and Hungarian is as follows. In regular non-wh-expletive questions in Iraqi Arabic and Hindi the relation of a wh-phrase to a +wh+Q Comp is forced to be more local than the relation of secondary wh-phrases to a +wh+Q Comp in other languages such as English. This is suggested to be because the +wh+Q Comp in Iraqi Arabic/Hindi is a different lexical item from that in English and enforces a particular tense-related localit y on the licensing of wh-phrases causing them to raise overtly into this localit y when base-generated in lower positions. Turning now to PM questions and comparing them with regular non-wh-expletive questions in German, one finds that the relation of the substantive wh-phrase to the +wh+Q Comp in PM questions is forced to be more local than the regular relation of secondary wh-phrases to a +wh+Q Comp in non-whexpletive questions, and as in Iraqi Arabic and Hindi a wh-phrase is seen to have to raise into a narrower tense-defined linking with the +wh+Q Comp. Quite
domain CP, as in examples such as (46), may raise up to the intermediate Comp position and restore grammaticalit y as in (48) repeated below as (i): (i) Was glaubst du [mit wemi Hans meint [ti dass Jakob ti gesprochen hat]]? WH believe you [with whom Hans says that Jakob spoken has ‘Who do you believe that Hans says that Jakob spoke with?’ 28
In English-t ype languages the +wh+Q value of C0 was argued to be triggered by raising of a wh-phrase to Comp. In this sense it should be thought of as a potential lexical value of C0 requiring identification.
188
W h-movement and the theory of feature-checking
possibly an explanation of this contrast internal to German might then be that the patterning results from two different lexical instantiations of +wh+Q C0, that a rather different +wh+Q C0 occurs in PM questions induced by the wh-expletive and that such a +wh+Q C0 is critically associated with some kind of a reduced tense-defined licensing localit y similar in ways to the +wh+Q C0 in Iraqi Arabic/Hindi, consequently forcing Partial Movement of the substantive wh-phrase to satisfy this narrower localit y requirement. Such an attempted route of explanation would essentially be entertaining the possibilit y that languages might in fact allow for differently-valued instantiations of a single functional head. Although initially this might seem somewhat exotic as an idea, if such multiple potential values are taken to be induced by the different lexical items which a language has at its disposal, then in fact a proposal of this general kind is actually rather natural and worth pursuing. Furthermore there is also compelling evidence from Iraqi Arabic and Hindi that something along these lines must indeed be possible, as a relevant (but somewhat lengthy) detour through wh-expletive questions in these languages will show below, supporting the credibilit y of an account of Partial Movement which will soon attribute its central defining properties to the t ype of +Q head induced in the structure.
3.5.3. W h-expletive questions in Iraqi Arabic and Hindi In Chapter 2 it was noted that wh-phrases in Iraqi Arabic and Hindi may not occur in non-interrogative embedded tensed CPs, and that pre-Spell-Out wh-raising to the tense domain of a higher +Q Comp is forced to take place to license such structures: (67)
(68)
(69)
(70)
*Mona tsawwarat [CP Ali ishtara sheno] Mona thought Ali bought what Intended: ‘What did Mona think that Ali bought?’ Shenoi tsawwarit Mona [CP Ali ishtara ti]? what thought Mona Ali bought ‘What did Mona think Ali bought?’ *Raam-ne kahaa [ki kOn aayaa hE]? Ram-ERG said [that who come has ‘Who did Ram say has come?’ kOni Raam-ne kahaa [ki ti aayaa hE]? who Ram-ERG said [that come has Who did Ram say has come?
Partial Movement and wh-expletive wh questions
189
In addition to movement however there is another way (resulting from a different numeration) in which the meanings of (68) and (70) may grammatically be expressed. If a wh-expletive element occurs in the higher clause, the wh-phrase can in fact licitly remain in situ in the lower tensed CP. In both IA and Hindi the wh-expletive derives from the word for ‘what’ in the language, this paralleling the similar source of wh-expletives in German and Hungarian:29 (71)
(72)
sh-tsawwarat Mona [CP Ali ishtara sheno] WH-thought Mona Ali bought what ‘What did Mona think that Ali bought?’ Raam-ne kyaa kahaa [CP ki kOn aayaa hE]? Ram-erg WH said that who come has ‘Who did Ram say has come?’
What the introduction of the wh-expletive here would seem to do is to alter and extend the licensing domain of the +Q Comp. Whereas a wh-phrase normally must occur in the immediate tense domain of the +Q C0, the presence of the wh-expletive allows for a substantive wh-phrase to occur in an embedded tensed clause adjacent to the clause containing the +Comp. Note also that in Iraqi Arabic the wh-expletive element occurs raised into the +Q Comp. Here it is important to bear in mind that a +wh+Q Comp in Iraqi Arabic (and Hindi) otherwise does not contain strong wh-features in need of checking and a +Q Comp need not be filled by any wh-element; consequently the raising of the wh-expletive to the +Q Comp cannot be analysed as taking place to satisfy any strong wh-features present there. Instead the introduction of the wh-expletive into the +Q Comp would seem to take place for the explicit purpose of allowing a real wh-phrase to remain in an in situ position within an adjacent tensed clause. Earlier in Chapter 2 it was argued that a +wh+Q Comp in Iraqi Arabic is a lexical item associated with a licensing domain corresponding to its immediate tense domain and that this specification accounts for the patterning of wh-phrases in the language. In wh-expletive questions it can now be suggested that the localit y value of the regular default +wh+Q Comp in Iraqi Arabic is significantly replaced with another value triggered by the wh-expletive and that the introduction of the wh-expletive into the +Q Comp functions to induce a second instantiation of +wh+Q C0 lexically available in Iraqi Arabic which projects a wider wh-licensing domain critically extending to the adjacent tense domain. In Hindi the wh-expletive kyaa does not occur in C0 but in some pre-verbal
29
In Iraqi Arabic the wh-expletive is regularly reduced from sheno just to sh-.
190
W h-movement and the theory of feature-checking
position cliticised to the verb. Here I would like to speculate that this position is actually a low focus-t ype head dominating the VP just as in Hungarian and that this position as in Hungarian is indeed the locus of +Q/focus-t ype features in the clause. In the introduction to this chapter it was noted that wh-movement in Hungarian is to a +Q/focus-head which is lower than the embedding complemetiser hogy ‘that’ and also often lower than the subject position.30 The term “Comp” was therefore used, as in Horvath (1997), in a rather abstract way as a general cover term for the functional head which encodes +Q and focus specifications in Hungarian, it being understood that the position of embedding complementisers is technically higher than this +Q/focus head and that there might also be other intervening functional heads present between the embedding complementiser head and the focus/+Q head, as proposed in Szabolcsi (1997). Having now suggested that the wh-expletive kyaa in Hindi may also occur in a pre-verbal +Q/focus position which is lower than the position of embedding complementisers and subjects, I will take the same license as before when using the term Comp/C0 to refer to the position of raised wh-phrases in Hungarian and henceforth also use the term “Comp” and +Q C0 to refer to the putative preverbal focus/Q position in Hindi.31 The suggestion is then that both in Iraqi Arabic and in Hindi the possibilit y for a wh-phrase to occur in an embedded tensed clause is due to the wh-expletive effectively inducing a different licensing localit y value in Comp. Technically one may imagine either that the wh-expletive as a lexical element brings with it into Comp this different localit y value and transmits it to the Comp head via Spec–Head agreement, or that the wh-expletive simply identifies the t ype of +wh+Q C0 inserted from the lexicon, i.e. the presence of the wh-expletive in Comp signals the +wh+Q C0 as being the lexical variant with the extended licensing localit y. This extended licensing localit y induced by the presence of the wh-expletive is not unrestricted and it is found that a wh-phrase must occur in the tense domain
30
See Szabolcsi (1997) for an articulated analysis of clausal structure in Hungarian and the position of focus and subjects.
31
It is possible that the locus of +Q in Iraqi Arabic is also actually somewhat lower than the initial embedding complementiser position. Wahba (1991) notes that the wh-expletive element may also surface in an unreduced form sheno in a position which precedes the subject but follows the verb raised to the C0 position. This is a non-canonical position for an object, being higher than the VP and the subject, and it may correspond to the same +Q/focus-head suggested for Hungarian and (now) Hindi. In connection with this, see also Schütze (1996) for discussion of low focus projections in Korean, and Jayaseelan (1996) on pre-verbal focus and wh in Malayalam.
Partial Movement and wh-expletive wh questions
191
immediately adjacent to and dominated by the tense domain containing the +Q Comp (and the wh-expletive) in order to be licensed. While this restriction is satisfied in (71) and (72) above, in (73) and (74) the tense domain immediately adjacent to the tense domain containing the +Q Comp/wh-expletive is CP2, and as the wh-phrase does not occur in this tense domain but in a still lower tense domain (CP3), the structures are ill-formed: (73)
(74)
*[CP1 raam-ne kyaa socaa [CP2 ki ravii-ne kahaa [CP3 ki kOnsaa [CP1 Ram-ERG WH thought that Ravi-ERG said that which aadmii aayaathaa]]]? man came Intended: ‘Which man did Ram think that Ravi said came?’ *[CP1 sh-i’tiqdit Mona [CP2 Ahmed tsawwar [CP3 Ali sa’ad meno]]]? [CP1 WH-believed Mona Ahmed thought Ali helped who Intended: ‘Who did Mona believe Ahmed thought Ali helped?’
Such ill-formed structures significantly can be rescued via movement of the wh-phrase from the lowest tensed clause into the intermediate tensed CP, as (75) shows: (75)
[CP1 raam-ne kyaa socaa [CP2 ki kOnsaa aadmiii ravii-ne kahaa [CP1 Ram-ERG WH thought that which man Ravi-ERG said [CP3 ki ti aayaathaa]]]? [CP3 that came ‘Which man did Ram think that Ravi said came?’
The acceptabilit y of examples such as (75) importantly confirms much of what has been suggested above. First of all, the ungrammaticalit y of (73) and (74) indicates clearly that it is not simply any wh-features which might occur on the +Q Comp which require checking in wh-expletive questions; the +Q Comp is occupied by a wh-expletive which under standard Minimalist assumptions might be expected to check any strong wh-operator features on Comp yet the sentences are still unacceptable. Hindi (75) shows that movement of the real wh-phrase to the intermediate clause will rescue the structure; this then indicates that (a) it is the real wh-phrase itself which is in need of licensing (because when it is moved up the structure becomes fully acceptable), (b) this licensing is effected via obligatory movement and so can be understood as a wh-feature checking requirement, and (c) the checking of wh-features on the wh-phrase must be effected prior to SpellOut, just as is the case with wh-phrases in other non-wh-expletive questions (otherwise the raising of the wh-phrase should not be forced to take place overtly). All of this taken together then indicates that the checking localit y of the +wh+Q Comp has indeed undergone a significant change. Critically the obligatory
192
W h-movement and the theory of feature-checking
movement of the wh-phrase for pre-Spell-Out checking takes it not into the tense domain of the +Q Comp as in regular non-wh-expletive questions but instead into the adjacent tense domain where wh-phrases may not normally occur in the absence of a wh-expletive. It would seem that the introduction of a wh-expletive into Comp extends the checking localit y beyond the normal immediate tense domain of the +Q Comp to include the tense domain immediately dominated by the tensed clause containing the +Q Comp. A wh-phrase base-generated in a position lower than this then overtly targets the intermediate tense domain and is licensed by raising into this domain. Further data show that all wh-phrases which occur base-generated in a tense domain not immediately adjacent to the tense domain containing the +Q Comp (and the wh-expletive) must undergo this raising in order to be licensed. (76) below is unacceptable due to the occurrence of the wh-phrase kis-ko ‘whom’ in the lowest tense domain CP3 and raising of this element to the intermediate tense domain CP2 in (77) immediately restores grammaticalit y: (76)
(77)
*[CP1 raam-ne kyaa socaa [CP2 ki kOni ravii-ne kahaa [CP3 ki ti [CP1 Ram-ERG WH thought that who Ravi-ERG said that kis-ko maaregaa]]]? whom will-hit Intended: ‘Who did Ram think that Ravi said will hit who?’ [CP1 raam-ne kyaa socaa [CP2 ki kOni kis-kok ravii-ne kahaa [CP1 Ram-ERG WH thought that who whom Ravi-ERG said [CP3 ki ti tk maaregaa]]]? [CP3 that will-hit ‘Who did Ram think that Ravi said will hit who?’
This confirms again that it is a pre-Spell-Out feature-checking requirement carried by (all) wh-phrases rather than the +Q Comp which triggers the movement found in wh-expletive questions just as in non-wh-expletive forms. It is quite important to stress the significance of the fact observed here that a wh-phrase base-generated in a lower tensed clause such as CP3 in (73)–(75) or (76)–(77) is forced to raise to the intermediate tensed clause CP2 and not necessarily into CP1, the clause containing the +Q Comp. In non-wh-expletive questions it was found that a wh-phrase base-generated in a non-interrogative tensed CP has to raise to the clause where the +Q Comp is located and despite the arguments presented in Chapter 2 that this is simply raising into the tense domain of the +Q Comp, it might have been suggested that this movement is actually raising to the +Q Comp itself. Here in wh-expletive questions however it is fully clear that the movement which results in grammaticalit y can in no way be
Partial Movement and wh-expletive wh questions
193
analysed as being movement to the +Q Comp and instead is only necessary to the tensed clause dominated by the +Q Comp. The configuration necessarily created by Spell-Out linking the wh-phrase to the +Q Comp therefore resists all analysis as instantiation of a Spec–Head relation and one is forced to concede that an obligatory pre-Spell-Out checking relation which triggers overt movement is indeed effected “at a distance” between the +wh+Q Comp and a wh-phrase located in the adjacent tensed clause. It is also important to stress that the obligatory overt nature of the movement attested in examples such as (75) and (77) rules out the possibilit y of any LF-raising account. Mahajan (1990, 1996) suggests an LF clausal-raising analysis of wh-expletive structures in Hindi similar to the proposals in Horvath (1997) for Hungarian. In such an approach the clause containing the real wh-phrase is argued to raise to the wh-expletive and the +Q Comp at LF in a process of expletive-associate CP replacement, effectively bringing the wh-phrase into the strict checking domain of the +Q Comp at LF.32 However, although such a proposal might be tenable for simple bi-clausal examples such as (72), it cannot account for the fact that movement out of a more deeply embedded position in examples such as (75) and (77) is forced to be overt. If the clausal replacement operation takes place only at LF, there is no reason why any wh-movement such as that seen in (75) and (77) should have to take place prior to Spell-Out. If this movement were to be a necessary part of an LF clausal raising process, by Procrastinate it should clearly also be delayed until LF. Its necessary overt nature indicates instead that licensing of the wh-phrase must be effected in the overt syntax and therefore not via any covert clausal pied piping. Reflecting back on the alternations in (73) and (75), and (76) and (77) once again, it can be seen that there are very close parallels with alternations in multiclausal PM questions such as (46) and (48) repeated below. In both the Hindi and the German examples one finds wh-expletive questions which are ill-formed because a substantive wh-phrase is located in a clause which is separated from the clause containing the +wh+Q Comp by an intervening independent tense domain. In both languages if the wh-phrase is raised overtly into this intermediate tense domain adjacent to the clause containing the +Q Comp the examples immediately become acceptable:
32
Srivastav (1994) also suggests a similar analysis in which the clause containing the wh-phrase replaces the wh-expletive, but takes this to be a semantic rather than a syntactic operation.
194 (46)
(48)
W h-movement and the theory of feature-checking *Was glaubst du [dass Hans meint [mit wemi Jakob ti WH believe you [that Hans says [with whom Jakob gesprochen hat]]? spoken has ‘Who do you believe that Hans says that Jakob spoke with?’ Was glaubst du [mit wemi Hans meint [ti dass Jakob ti WH believe you [with whom Hans says that Jakob gesprochen hat]]? spoken has Who do you believe that Hans says that Jakob spoke with?
The parallelism also extends further. In German it was noted that a second way in which a substantive wh-phrase might occur in the lowest tense domain in such a multi-clause structure was if the intermediate tense domain contained a second intermediate wh-expletive in its Comp, example (47): (47)
Was glaubst du [was Hans meint [mit wemi Jakob ti WH believe you [WH Hans says [with whom Jakob gesprochen hat]]? spoken has ‘Who do you believe that Hans says that Jakob spoke with?’
Interestingly this is paralleled in Hindi by entirely similar forms, and an unacceptable example such as (73) may also be rendered grammatical if a second wh-expletive occurs in the intermediate clause/tense domain, as in (78): (73)
(78)
*[CP1 raam-ne kyaa socaa [CP2 ki ravii-ne kahaa [CP3 ki kOnsaa [CP1 Ram-ERG WH thought that Ravi-ERG said that which aadmii aayaathaa]]]? man came Intended: ‘Which man did Ram think that Ravi said came?’ [CP1 raam-ne kyaa socaa [CP2 ki ravii-ne kyaa kahaa [CP3 ki [CP1 Ram-ERG WH thought that Ravi-ERG WH said that kOn aayaathaa]]]? who came ‘Who did Ram think that Ravi said came?’
The existence of such strong parallels between wh-expletive questions in Iraqi Arabic, Hindi and German suggests that the syntax of Partial Movement structures must in some way be related to the wh-expletive strategy in Iraqi Arabic and Hindi and that a common t ype of explanation should ideally be invoked to cover
Partial Movement and wh-expletive wh questions
195
both sets of cases. Considering Iraqi Arabic and Hindi, the patterning observed leads to the conclusion that the wh-expletive is critically involved in altering the licensing localit y value of the +wh+Q Comp. As Iraqi Arabic and Hindi are both languages which do not require any wh-element overtly raised to a +Q Comp, the use of the wh-expletive cannot be attributed to any immediate requirements of the +Q Comp and instead would seem to be employed solely to allow for substantive wh-elements with features in need of pre-Spell-Out checking to be licensed in a wider localit y than in non-wh-expletive questions. In some instances this involved necessary overt movement of a wh-phrase into the redefined checking domain, and in all cases the different patterning of the substantive wh-elements present was clearly a direct function of the presence of the wh-expletive. It was therefore suggested that the introduction of the wh-expletive into Comp induces or otherwise identifies a +wh+Q C0 which is actually a different lexical element from that occurring in non-wh-expletive questions, one with an associated licensing localit y value which is less restricted than the default +wh+Q C0 in these languages. Turning now to Partial Movement questions, again one finds that the behaviour of wh-elements is different in the presence of a wh-expletive element occurring raised in the +Q Comp position, and a substantive wh-phrase is forced to raise to establish a pre-Spell-Out linking to the +wh+Q Comp which significantly is defined in terms of tense and the notion of tense domain. Just as in Iraqi Arabic and Hindi the substantive wh-phrase may not be separated from the tense domain immediately containing the +Q Comp by any other intervening independently tensed clause. It can therefore be suggested that the differences in wh-behaviour observed in wh-expletive questions in German are similarly a result of the same essential syntactic process and that a wh-expletive raised to the Comp of a Partial Movement structure functions to identify a different lexical instantiation of a +Q C0, one which has an associated localit y value which directly results in Partial Movement of the real wh-phrase. Exactly why this Partial Movement is rendered necessary and the existence of an important difference between the German and Iraqi Arabic/Hindi constructions will now be considered in Section 3.5.4.
3.5.4. W h-expletives and Partial Movement Above it has been suggested that Partial wh-Movement is triggered by the need for a substantive wh-phrase to establish some kind of local relation with the +Q C0 of a wh-expletive question, and that this additional wh-movement to a Q C which is not required in regular wh-questions may be a direct result of the wh-expletive inducing a different +Q C0 in the structure. In Iraqi Arabic and Hindi it was
196
W h-movement and the theory of feature-checking
argued that the different +wh+Q C0 identified by the wh-expletive projects a wh-checking domain which is wider than the regular default +wh+Q C0, allowing wh-elements in these question forms to be licensed in embedded tensed clauses adjacent to the tense domain containing the +Q Comp. In German it might therefore be imagined that there is a different +wh+Q C0 identified in wh-expletive questions and that this +wh+Q C0 brings with it a different wh-licensing localit y, in this case actually more restricted. Such a reduced wh-licensing localit y would then cause the substantive wh-phrase to raise up overtly into a tense-defined local relation with the +Q Comp for licensing by the +wh+Q C0. However, such a view cannot in fact be fully maintained and although there are indeed very strong parallels between Iraqi Arabic and Hindi and wh-expletive questions in German and Hungarian, there is also an important difference. In Iraqi Arabic and Hindi it was clearly seen that all wh-elements were equally affected by the presence of the wh-expletive and were forced to occur in the tense domain adjacent to the tense domain containing the +Q C0 by Spell-Out, confirming suggestions that use of the wh-expletive results in a modification of the wh-checking localit y. In German however, it is found that only a single wh-phrase is actually forced to undergo Partial Movement in wh-expletive questions, and other substantive wh-phrases are free to remain in situ as seen in example (79): (79)
Was glaubst du [wanni Hans ti an welcher Universität studiert hat]? believe you [when Hans at which university studied has ‘When do you believe that Hans studied at which universit y?’ WH
Given the conclusion made in Chapter 2 that all wh-phrases carry wh-features in need of checking, if it were to be the case that wh-expletives result in a reduced wh-licensing domain in German and so trigger Partial Movement into this domain, it is clearly expected that all wh-phrases should have to raise into this more local relation with the +Q C0. As only a single wh-phrase is required to undergo Partial Movement and secondary wh-phrases may remain in situ, the role of the wh-expletive in German consequently needs to be re-assessed. Re-considering the patterning with Partial Movement structures, wh-expletive questions in German (and Hungarian) appear to be significantly similar to regular, non expletive wh-questions in that in both question t ypes it is just a single substantive (i.e. non-expletive) wh-phrase that is required to undergo (some form of) movement. This movement in wh-expletive questions cannot be motivated by the need to check any non-wh focus-t ype features and must bring the wh-phrase into a specific position where some relation with other structurally higher wh-elements is established, creating a linking to the +Q Comp. The fact that
Partial Movement and wh-expletive wh questions
197
Partial Movement is therefore necessitated by the need to form a relation with a +Q Comp (hence is related to and motivated by wh-interrogative features) taken together with the observation that PM affects only a single substantive wh-phrase in multiple wh-questions, as in non-wh-expletive structures, strongly suggests that PM actually takes place for the same reasons that full wh-movement to a +Q Comp occurs in non-wh-expletive questions, in the case of languages like German and English in order to establish a clause as a wh-question and trigger Comp as a licensor for all wh-phrases in its domain. If this is indeed the correct assessment of Partial wh Movement and the underlying primary goal of PM is in fact to determine a higher Comp as +wh+Q, the mechanism by which such a Comp is triggered as a wh-licensor must nevertheless be somewhat different from the mechanism in non-wh-expletive questions as PM clearly does not bring the real wh-phrase directly into the specifier of the +Q C. Initially one might naturally expect that raising of the wh-expletive alone to a +Q Comp should result in the triggering of this C0 as a licensor for other wh-elements. However, the need for some secondary Partial Movement of a real wh-phrase suggests that this cannot be so, and wh-expletive questions in German and Hungarian might only seem to be licensed as wh-questions through the combination of wh-expletive raising and PM of the substantive wh-phrase. Such a conclusion is something of a surprise, as it might be anticipated that the wh-expletive should have the necessary wh-specification for determination of the C0 as +wh without the need for any other wh-raising, yet it appears that the wh-expletive in isolation is actually unable to perform the regular triggering function carried out by contentful wh-phrases. Though certainly puzzling to begin with, a plausible and satisfactory explanation of this problematic aspect of PM questions can be offered when one reflects a little more on the general nature and function of expletives and compares this with the situation in PM questions. In PM questions the suggestion that the wh-expletive requires the support of a nonexpletive wh-phrase (in a certain local relation) to trigger the higher Comp as +wh+Q is essentially equivalent to claiming that the wh-expletive is critically defective and unable to perform in the way of other wh-elements. Such a broad assumption is in fact not unnatural as the propert y of having some kind of formal deficiency can be said to be an important defining characteristic of a number of expletive-t ype elements and expletives are often found to be place-holder elements which substitute into a position with a reduced subset of the properties normally carried by their non-expletive equivalents. Due to the under-specification and partially “defective” nature of such expletives, an associate element which does carry the properties lacking in the expletive must often be present in some local relation to the expletive to satisfy relevant requirements of the structure in which
198
W h-movement and the theory of feature-checking
the expletive is embedded; in many approaches it is suggested that the associate raises to the expletive at LF, and this movement operation supplies the position occupied by the expletive with the full set of necessary properties/features. To illustrate this with an example, Chomsky (1995) suggests that the expletive element ‘there’ is effectively a deficient DP which is inserted into the subject position simply to check D-features in T0 (the EPP). The deficiency attributed to ‘there’ is that it lacks Case and agreement features and so is unable to check the Case and agreement features associated with the subject position. As a consequence of this an associate NP/DP is required to be locally present in the structure to supply these specifications, as in example (80). Example (81) can be argued to be ill-formed precisely because there is no associate NP/DP to satisfy the Case/agreement checking requirements in T0/V0: (80) (81)
There arrived a man. *There rained.
The patterning observed with there contrasts with the second English DP-expletive it. It is assumed to carry Case and agreement features in addition to the D0-features which will satisfy the EPP. Consequently with it no associate is necessary in simple weather predicate sentences, and it alone is able to check all D0, Case and agreement features in the subject position: (82)
It rained.
Returning to PM constructions I would like to suggest that the hypothesised inabilit y of the wh-expletive to trigger C0 as +wh+Q and the requirement that it be supported by a substantive wh-phrase “associate” can be accounted for by positing a formal deficiency in was (and its equivalents in Hungarian) somewhat similar to that suggested for English there. What needs to be captured somehow is the observation that unlike substantive wh-phrases, the wh-expletive was cannot identify a Comp as +wh even though morphologically it is a +wh element with wh-features. The problem is essentially very much like that presented by the unacceptabilit y of (81) — even though the expletive there is assumed to be a DP and DPs might naturally be expected to carry Case and agreement specifications as part of their inherent nature as DPs, the patterning found indicates instead that there is defective and rather exceptionally does not carry these characteristic DP features (and consequently a substantive DP which can provide these properties must occur to support there as its associate). Consequently, parallel to the way in which examples such as (81) may be explained by assuming there to be a DP lacking certain important t ypical DP-features, I would now like to suggest that PM constructions may largely be accounted for by positing that wh-expletives such as
Partial Movement and wh-expletive wh questions
199
German was critically lack some formal propert y which is otherwise present in other substantive wh-phrases and which allows for the triggering of a Comp as +wh. In a refinement of earlier proposals, I would now like to propose that the triggering of a C0 as +wh+Q is not in fact simply sensitive to purely morphological features, but that a wh-element must also contain a wh-operator element in order to be able to identify a C0 as +wh+Q, and that this is precisely the formal attribute lacking from wh-expletives in German and Hungarian. Just as there is a defective DP-expletive which lacks certain regular DP-specifications, possibly as a result of an ongoing process of grammaticalisation, similarly was can be taken to be a defective wh-element lacking the wh-operator which potentially occurs in other contentful wh-expressions. Following general ideas in Tsai (1994), it can be suggested that where a wh-operator may be base-generated is indeed a parameter of variation across languages and lexical items, and that substantive interpreted wh-phrases in German and Hungarian potentially will carry wh-operators whereas wh-expletives in these languages may not.33 This suggested deficiency of wh-expletives in German and Hungarian will now allow for the development of an account of PM structures which explains the behaviour of the substantive wh-phrase, i.e. the motivation behind its Partial Movement. It can be suggested that because the wh-expletive is unable to trigger the relevant Comp as +wh+Q, a substantive wh-phrase “associate” must be present to supply the wh-operator which will result in triggering of the Comp. Furthermore, the wh-phrase with the wh-operator must occur sufficiently local to the targeted Comp prior to Spell-Out in order for this to be identified as +wh, and therefore Partial Movement of the wh-phrase is forced, in order to bring the wh-phrase into the tense-defined local relation apparently necessary for it and its wh-operator to be visible to the Comp.34 The role played by the wh-expletive in the triggering process is essentially similar to that found with many other expletive elements. Being typically defective,
33
An interesting question is whether or not all wh-expletives lack a wh-operator. Certainly there are elements such as Burmese leh which are in fact able to unambiguously establish a clause as +wh+Q. Comparing the situation to that found with non-wh-expletives, just as English contains a DP expletive ‘it’ which exhibits the full range of DP properties and a second DP-expletive which lacks certain of these, so it might be natural to expect that there would be some wh-expletives which would indeed contain wh-operators and other which might not.
34
Note that although it might possibly be objected that a wh-operator is a semantic element not relevant to pre-LF syntax (though the term “wh-operator” nevertheless is used to drive overt pre-LF wh-movement in Chomsky 1993, 1995), the model argued for here is one in which the forms created by Spell-Out are in fact essentially the syntactic input to interpretation. Consequently it is indeed legitimate to suggest that certain semantic objects may trigger overt syntactic movement.
200
W h-movement and the theory of feature-checking
the wh-expletive is unable to directly satisfy requirements of the relevant head C0, but crucially is able to allow for these properties to be provided by some second element found in a certain local relation to the head. Similar to the use of wh-expletives in Iraqi Arabic and Hindi, the effect of introducing the wh-expletive into Comp in German and Hungarian is to substitute in a different localit y requirement in the head-position. In Iraqi Arabic and Hindi this results from a full identification of a variant +wh+Q C0 lexically available in these languages and allows for wh-elements to be licensed in a wider local domain than in non-wh-expletive questions. In German and Hungarian however, it can be suggested that the under-specified wh-expletive is only able to partially identify properties of the Comp, failing to induce a fully-specified +wh+Q C0 but succeeding in identifying a localit y value which allows for a substantive wh-phrase with the necessary wh-operator to occur further away and still trigger the Comp as +wh+Q.35 Effectively the presence of the wh-expletive in German/Hungarian prevents the C0 from defaulting to a purely declarative value and allows the Comp to be fully identified by a wh-element in a lower position. Critically it would seem that the substantive wh-phrase cannot trigger the Comp as +wh from an in situ position (hence example (61) is unacceptable) and triggering must be effected prior to Spell-Out via raising of the wh-phrase to a position which is closer to the Comp requiring identification as +wh. In earlier sections it was shown that the necessary Spell-Out localit y linking a partially-moved wh-phrase to a +Q Comp in acceptable cases in German Dialect A (and also assumed underlyingly for speakers of Dialect B and Hungarian) needed to be defined in terms of tense and the notion of a tense domain. It was found that the substantive wh-phrase could not be separated from the clause containing the +Q Comp by any intervening independently tensed clause, and if such a configuration occurred in a derivation, the wh-phrase was forced to raise up higher into this intermediate tense domain. It was also shown that Partial Movement does not result in wh-phrases occurring in all of those Comp positions which might be moved through by any long successive cyclic wh-movement; specifically, PM does not take place to the Comp of a tense-dependent clause as such a clause comprises part of a higher tense domain. The relevant generalisation argued for was that the substantive wh-phrase must occur in the tense domain adjacent to that containing the Comp to be triggered as +wh, or be related to this position by a
35
The term “triggering” can essentially be understood to be the same as “identification”. Raising of a wh-phrase to a SpecCP position will normally identify the C0 as the/a +wh+Q instantiation of C0 in a language.
Partial Movement and wh-expletive wh questions
201
chain of intermediate wh-expletives linking all intervening tense domains. The different localit y enforced by the introduction of the wh-expletive into the Comp to be identified as +wh can therefore be taken to be that the substantive wh-phrase with the wh-operator must necessarily be located in an adjacent tense domain in order for its wh-operator to be visible to the higher Comp and identify it as +wh+Q. In addition to this the wh-phrase must also occur raised into the Comp of this tense domain, i.e. undergo Partial Movement to the Comp of this clause. This can possibly be understood as a necessary further requirement on making the wh-operator visible to the higher Comp. Noting that a tense domain is defined in terms of tense and hence most probably TP, possibly Partial Movement to the SpecCP position higher than TP brings a substantive wh-phrase minimally out of one tense domain and into a higher one. Alternatively this raising to Comp does not take the wh-phrase out of its containing tense domain, but in its clauseperipheral position simply makes the wh-operator visible to the Comp of the dominating tense domain in a way similar to patterns found with anaphors. When anaphors are raised to a Comp position within wh-phrases it has been noted that they automatically become visible to potential antecedents in the dominating clause and so may be bound there: (83)
Johni wondered [CP [which pictures of himselfi]k Mary had discarded tk]
Whichever descriptive form is chosen, I would like to assume that Partial Movement either within a single clause (when the clause is adjacent to the tense domain containing the Comp to be triggered as +wh) or from one clause to a higher intermediate clause in more complex structures in both cases essentially takes place for the same basic reason of satisfying a tense-sensitive localit y requirement on triggering induced by the presence of the wh-expletive: in order to fully identify the Comp as +wh+Q the wh-operator carried by the wh-phrase must be made visible to the higher C0 prior to Spell-Out, and such visibilit y here is achieved by occurring in the Comp position of an adjacent tense domain.36
36
In well-formed multiple expletive structures such as (i) below, the substantive wh-phrase does not raise to the Comp position of the tense domain adjacent to the clause containing the potential +Q Comp, and instead a second wh-expletive occurs in this position: (i) Was glaubst du, was er meint, wen Hans gesehen hat? WH believe you WH he thinks whom Hans seen has ‘Whom do you believe he thinks Hans saw?’ This intermediate wh-expletive will not be able to identify the higher Comp as +wh as wh-expletives do not carry wh-operators. However, as with the wh-expletive raised to the initial Comp, it can be argued that the occurrence of an intermediate expletive in a Q Comp has a
202
W h-movement and the theory of feature-checking
Briefly summarising now what has been suggested here as an explanation of certain of the behaviour of wh-phrase and wh-expletive in Partial Movement questions, it has been argued that wh-expletives in German and Hungarian are best understood as being t ypically defective expletive elements which do not carry all of the properties required by a particular head but permit these to be supplied by a second associate element located further away. Specifically it was suggested that wh-expletives in German and Hungarian do not have the wh-operator necessary to fully identify a Comp as +wh+Q, and that only substantive wh-phrases are base-generated with wh-operators in these languages. Consequently a real wh-phrase with the required wh-operator is forced to appear in a certain local relation to the C0/wh-expletive to trigger it as +wh+Q. The function of the wh-expletive is to substitute in/identify a different localit y requirement on the triggering of the C0, allowing for the wh-phrase containing the wh-operator to occur further down in the structure rather than directly in SpecCP. Due to this modified localit y requirement the real wh-phrase is also forced to undergo raising (i.e. Partial Movement) to a Comp position in a tense domain adjacent to the clause containing the relevant C0 in order to become visible to the latter. In this sense the wh-expletive is argued to function in a way very similar to wh-expletives in Iraqi Arabic and Hindi and indeed expletives in general, allowing a syntactic relation to be satisfied in a wider localit y than otherwise in the absence of the expletive, so apparently changing the localit y restrictions imposed by the relevant head on such a relation. In Iraqi Arabic and Hindi, wh-phrases may be licensed in more deeply-embedded positions in wh-expletive questions than in their non-wh-expletive counterparts as the licensing/checking domain of the +wh+Q C0 is extended by the introduction of the wh-expletive. In German a certain specification necessary to the C0-head itself is provided by an element which is also located further away than in non-wh-expletive questions — whereas the real wh-phrase with the wh-operator would normally have to occur in the Specifier of the +Q C0 in order to trigger it as +wh+Q, in wh-expletive questions the wh-phrase is permitted to occur further down in the structure and still fulfil its triggering role.
(further) effect on the necessary triggering localit y, extending it again to the following adjacent tense domain and allowing for a wh-phrase located there to identify the higher Comp as +wh. The movement of all wh-expletives present can be assumed to take place for their own selfserving licensing requirements. Being morphologically wh and thus having wh-features to be checked, a wh-expletive will only be licensed if it occurs in the wh-checking domain of a +Q Comp triggered as +wh and such a wh-checking domain will only be projected if a wh-expletive present in the numeration raises to extend the triggering localit y of Comp, thereby allowing it to be identified as +wh+Q by a lower contentful wh-phrase.
Partial Movement and wh-expletive wh questions
203
If it is indeed plausible to approach the wh-expletive in PM questions in this way and view it as an under-specified element requiring support from a real wh-phrase in a particular local relation thus causing Partial Movement, such an approach offers a number of clear advantages over other possible accounts. First of all it is clearly able to capture the necessary overt nature of Partial Movement, the propert y established earlier that PM critically takes place in order that some wh-relation wh is satisfied overtly and prior to Spell-Out. This aspect of PM questions so highly problematic for standard Minimalist assumptions and often largely ignored is straightforwardly accommodated in the present system which suggests that the wh-checking domain regularly projected in German allows for pre-SpellOut checking of wh-features carried by wh-phrases in any clause-internal position m-commanded by the +wh+Q C0. The fact that the partially-moved wh-phrase must therefore be checked prior to Spell-Out but does not occur in a +Q Comp at this point is therefore unproblematic. Secondly, the “expletive-associate” account proposed here is essentially a direct dependency account which does not posit any LF movement. Consequently the account avoids the problems created by the need to assume LF movement linking the partially-moved wh-phrase to the +Q Comp. The account also clearly avoids all the problems associated with indirect dependency approaches. Thirdly, in Chapter 2 it was suggested that crosslinguistic differences in the patterning of non-wh-expletive questions may be accounted for if one assumes a certain limited range of variation in the lexical instantiation of a +wh+Q C0. If different languages might have differing lexical values available for a +wh+Q C0 position, the expectation was then created that a single language might also have available in its lexicon more than one potential instantiation of a +wh+Q C0. In Iraqi Arabic, Hindi and PM structures it has here been argued that wh-expletives are indeed able to identify variant +wh+Q C0 forms which occur alongside the default +wh+Q C0 induced/inserted in these languages and that the tentative prediction made in Chapter 2 is therefore borne out in certain languages. Finally, if the account sketched out here is essentially on the right lines, it might seem to allow for an interesting conclusion about expletive-associate links in general. Commonly it is assumed that an associate element (or its formal features) undergoes raising to an expletive at LF in order to provide the position occupied by the expletive with certain necessary properties or features. Here in PM constructions however there is evidence that the properties lacking in the expletive must be provided locally prior to Spell-Out but not necessarily in any direct Spec–Head local relation with the relevant head. This would then seem to indicate that the function of the expletive is not to allow for the properties missing in the expletive to be substituted into its position later in the derivation, but instead for these simply to be located in a wider localit y.
204
W h-movement and the theory of feature-checking
Consequently the critical function of expletives might appear to be to change the localit y requirements of heads (basically by identifying different lexical instantiations of these heads) rather than simply allowing for a delay in weak-featurechecking until a later stage in the derivation.
3.6. 3.6.1. Further issues: antilocality in wh-expletive wh structures Having discussed how the various movements of wh-expletive and wh-phrase may be motivated in a way which is consistent with general properties and requirements of wh-licensing as proposed in Chapter 2, I will now consider how various other restrictions on the formation of wh-expletive questions may be explained. It has been claimed above that Partial Movement of a wh-phrase is necessary to bring such an element into a position in which it will be visible to a Comp containing a wh-expletive; only if a wh-phrase occurs within a certain localit y relative to the +Q Comp will it be able to trigger the latter as a licensor for all wh-elements in its domain. However, in addition to the requirement that the substantive wh-phrase occur within a certain tense-defined local domain, various other anti-locality effects have also been observed to obtain in wh-expletive questions in Hungarian and German, and it might appear that a substantive wh-phrase actually may not occur too close to the +Q Comp occupied by the wh-expletive. For example, a wh-phrase may not appear in situ in a clause whose Comp is occupied by a wh-expletive (84), nor either in situ or partially-moved in an infinitival or tense-dependent CP complement of a clause whose Comp is filled by an expletive (85)–(88): (84)
(85)
(86)
*Was glaubst du was? WH believe you what Intended: ‘What do you believe?’ *Was glaubst du was gesagt zu haben? WH believe you what said to have Intended: ‘What do you believe to have said?’ *Mit akarsz kivel beszelni? WH want who-with talk-INF Intended: ‘With whom do you want him to talk?’
Partial Movement and wh-expletive wh questions (87)
(88)
205
*Was willst du dass er wen besticht? WH want you that he whom bribe Intended: ‘Whom do you want that he bribe?’ *Was willst du, wen er besticht? WH want you whom he bribe Intended: ‘Whom do you want that he bribe?’
Such facts are clearly in need of explanation if the general account of PM proposed above is to be successful. Examples such as (86) and (88) may in fact be quite easily accounted for, but the ungrammaticalit y of (84), (85) and (87) is somewhat unexpected. In (86)–(68) the wh-phrases should be visible within the relevant tense-defined domain of the +Q Comp without the need for any PM and so none should occur (by economy). However, there is no immediately obvious reason why (84), (85) and (87) should be ill-formed — although no PM takes place, none is predicted to be necessary, as the wh-phrases in their in situ positions all occur within the same tense domain as the (+Q) Comp and so should indeed be sufficiently local to it to effect its triggering as +wh.37 One possible approach to these patterns might be to suggest that introduction of the wh-expletive into Comp not only changes the triggering localit y of this Comp, requiring that a substantive wh-phrase be found visible within an adjacent tense domain, but also brings with it a certain anti-localit y, so that a wh-phrase may not in fact be found within the same tense domain as the (+Q) Comp. Similar “pure antilocalit y” effects have indeed been attested in various other syntactic relations. Ladusaw (1980), for example, notes that Positive Polarit y Items in English may not occur in the scope of clausemate negation (with narrow scope) but may be c-commanded by negation if occurring in some lower clause:38 (89) (90)
#John
did not see someone. John did not claim that he loves someone.
In a similar vein, Iatridou (1986) points out an antilocalit y effect obtaining with
37
(84), (85) and (87) are also difficult to explain for any LF raising approach, there being no localit y restrictions to bar movement to Comp. They would also seem to be highly problematic for any feature-movement approach. Were was simply to be the instantiation of a wh-phrase’s wh-features raised to a higher Comp position, one would naturally expect that was/these wh-features should be able to raise to a +Q Comp from a wh-phrase in its base-generated in situ position, stranding it in such a position, yet (84), (85) and (87) are quite ungrammatical. Cheng (to appear) attempts to resolve this clear problem in a feature-movement account with the help of the rather different assumptions about feature-movement made in Chomsky (1998).
38
Parallel patterns are reported for Serbo-Croat in Progovac (1991b).
206
W h-movement and the theory of feature-checking
the Greek anaphor o idhios, which must be free in its governing category, but bound within the sentence it occurs in: (91)
(92)
O Yanis pistevi oti o idhios tha kerdhisi John believes that himself will win ‘Johni believes that hei/*k will win.’ *O Yanis aghapa ton idhio John loves himself
However, it is not clear at all what actually causes such apparent anti-localit y effects, and in German and Hungarian rather than simply stipulate that the wh-expletive induces anti-localit y, it would obviously be more satisfying to identify some principled reason for the ban on substantive wh-phrases occurring in the tense domain of the +Q Comp. Arguably an explanation for the patterns observed may be given in terms of Case. Horvath (1997) points out that the occurrence of PM and wh-expletives in Hungarian seems to be possible only in those cases where parallel non-whexpletive structures are also well-formed, e.g. PM is not possible in infinitives with a wh-expletive raised to Comp (as in example (86) above), and the pairing of a non-wh expletive with an infinitival CP associate is also found to be unacceptable: (93)
*Szamitok ra [beszelni a gyerekekkel]? count-1SG it-AL [talk-INF the kids-with Intended: ‘I count on it to talk to the kids.’
When the embedded complement CP is tensed however, both wh- and non-wh expletive structures are possible. In a similar way, although DP-internal wh-movement to SpecDP is possible in Hungarian (as indeed earlier observed in footnote 24), this allowing the entire DP to raise to Comp as in (94), it is not possible to construct PM-type structures in which a wh-phrase raises DP-internally just to SpecDP and Comp is filled by a wh-expletive, as in example (95). This seems to parallel the impossibilit y of pairing DPs with non-wh expletives, as seen in (96): (94)
(95)
[DP Kineki a ti feleseget]k lattad tk? [DP who-DAT the wife-ACC saw-2SG ‘Whose wife did you see?’ *Mit lattad [DP kineki a ti feleseget]? WH-ACC saw-2SG who-DAT the wife Intended: ‘Whose wife did you see?’
Partial Movement and wh-expletive wh questions (96)
207
*Azt mondta [a maga velemenvet]. It-ACC said-3SG the self opinion Intended: ‘He told (it) his own opinion.’
An essentially similar patterning has been reported for German (see for example Fanselow & Mahajan 1996) and the wh-expletive was would basically only seem to be possible in structures where a non-wh expletive es may be used (in corresponding declarative sentences). Consequently, bearing in mind the observation that wh-expletives in Hungarian are clearly Case-marked with non-default Case (non-wh expletives also showing parallel inflections), it can be argued that the unacceptabilit y of these ungrammatical forms reduces to the simple failure of an element to be Case-marked/checked. In examples such as (95)–(96) and (84) repeated below, both expletive and DP will require Case-marking/checking but only a single objective Case(-checking position) is available in the clause: (84)
*Was glaubst du was? WH believe you what Intended: ‘What do you believe?’
Examples such (84) may then be ruled out without the need to assume any irreducible anti-localit y induced by the wh-expletive. The possibilit y of accounting for examples such as (84) in German with a simple explanation in terms of Case would also seem to provide a strong argument in favour of the earlier suggestion that German was is indeed first base-generated in a low Case-checking position and then later raised to SpecCP, rather than being directly base-generated in Comp, i.e. that German really is parallel to Hungarian in this respect. It is not at all clear how the “anti-localit y” reviewed here could otherwise be explained in an approach which takes was to be base-generated in Comp.39 If the unacceptabilit y of examples like (84) is due to Case reasons and not a
39
A further good reason to believe that German was is base-generated in SpecAgrO and not Comp is the fact that was and the non-wh-expletive es ‘it’ may not co-occur in wh-questions with PM in an object clause such as (i) below: (i) *Was glaubst du es [mit wem Johann gesprochen hat]? WH believe you it [with whom Johann spoken has Intended: ‘Who do you believe Johnann spoke with?’ This co-occurrence restriction is simply explained if was and es both need to check Case in the same position — SpecAgrO (both also possibly being base-generated in SpecAgrO). It should also be remembered that in matrix clauses was clearly occurs preceding the inflected verb which has been moved to C0, indicating that was must have the status of an XP and cannot be assumed to be an X0 base-generated in C0.
208
W h-movement and the theory of feature-checking
general ban on wh-expletive and wh-phrase occurring within the same tense domain, one might expect that a subject wh-phrase should be able to co-occur in the same clause as a wh-expletive bearing objective Case, as the occurrence of a subject wh-phrase checking Case in SpecAgrS should not interfere with the Casechecking of a wh-expletive in SpecAgrO. For many speakers such structures are indeed quite acceptable, adding good support to the Case-related account. Example (97) is from Müller & Sternefeld (1996) and (98) from Fanselow & Mahajan (1996): (97)
(98)
Was meint wer, mit wem sie gesprochen hat? WH believe who-NOM with whom she spoken has ‘Who believes he went where?’ Was glaubt wer, wen sie liebt? WH believe who-NOM whom she loves ‘Who believes she loves whom?’
Certain other speakers nevertheless find apparently similar examples unacceptable. Example (99) is from Beck & Berman (1996): (99)
*Was hat wer gedacht, wen wir anrufen sollten? WH has who-NOM thought whom we telephone should Intended: ‘Who thought we should call whom?’
All three examples are however superiorit y violation configurations — the subject wh-phrase rather than the object wh-expletive should move to Comp (movement from SpecAgrS being closer than from SpecAgrO). It is therefore possible that those speakers who reject such examples do so simply for reasons of superiorit y, and the unacceptabilit y of such forms (for certain speakers) consequently is not evidence against a Case-based explanation of the “anti-locality” patterns. Given the unclear status of superiorit y for many speakers of German, it is also not surprising that other speakers will in fact accept such structures.40 Additional support for the hypothesis that wh-expletives generally require Case-
40
An alternative configuration in which a superiorit y violation is avoided, with movement to Comp of the subject as in (i) below, also proves to be ungrammatical, significantly for all speakers. Here however, it will again be possible to argue that such structures are ill-formed for (other) Caseindependent reasons (see Section 3.6.2 below), allowing for a Case-based account of (84) to be maintained: (i) *Wer glaubt was, wo er hingegangen ist? who-NOM believe WH where he went has Intended: ‘Who believes he went where?’
Partial Movement and wh-expletive wh questions
209
checking comes from the observation of certain other wh-expletive structures in Hungarian. Where a predicate does not check/assign any objective Case in Hungarian, PM structures involving an adjunct CP are still possible via the use of a non-Case bearing wh-expletive miert ‘why’: (100)
Miert vagy duhos [mert kiveli talalkoztal ti]? why be-2SG angry [because who-with met-2SG ‘Who is the person such that you are angry because you met that person?’
In (100) it is importantly not possible to substitute a wh-expletive such as mit ‘what’ for miert ‘why’. This fact is simply explained if it is assumed that an object wh-expletive such as mit crucially requires Case-checking, whereas miert ‘why’ does not — inserted in (100) mit would fail to check its objective-Case. A Case-based approach will also account for the fact that PM is not possible within CNPs, although being quite acceptable in other t ypical island configurations (e.g. SubjectCPs, adjunct CPs): (101)
*Mit hallottal [a hirt hogy kiveli talalkozott Mari ti]? WH-ACC heard-2SG [the news-ACC that with-whom met Mary Intended: ‘Who is the person x, such that you heard the news that Mary met with x?’
(101) is arguably ill-formed not because of any localit y violation linking the wh-phrase and the +Q Comp caused by the island structure, but simply because either the wh-expletive or the (C)NP will not be Case-checked, AgrO only being able to check a single accusative Case. Case-checking requirements of wh-expletives in German/Hungarian further explain why a wh-expletive present in the numeration is not in fact merged into the lower Q Comp position which the substantive wh-phrase moves to, i.e. why one does not find examples such as (102) or (103): (102)
(103)
*Was glaubst du, [CP was Hans wen gesehen hat]? WH believe you WH Hans whom seen has Intended: ‘Whom do you believe Hans saw?’ *Wasi glaubst du [CP ti Hans wen gesehen hat]? WH believe you Hans whom seen has
Chomsky (1995) suggests that the operation Merge is economically cheaper than Move. Consequently in a bi-clausal structure such as (104/105), economy will prefer the merging of an expletive into the embedded SpecTP position to satisfy the EPP rather than movement of the lower DP to this position:
210 (104) (105)
W h-movement and the theory of feature-checking Therei seems [ti to be a man in the garden]. *There seems [[a man]i to be ti in the garden].
Although merging of a wh-expletive into the lower Q Comp in (102) and (103) should then be more economical than movement of the wh-phrase to this position, (102) and (103) are actually unacceptable (in contrast to (104)). Examples such as these may however be excluded if it is assumed that before any raising to Comp the expletive first needs to be inserted in a position where it can check its Case, and it cannot be merged directly into SpecCP. In (102/103) it will not be possible to base-generate the wh-expletive in any case-checking position in the lower clause as there are other DPs present which need to check Case against these positions.41 Such considerations of Case also allow one to establish a partial t ypology of wh-expletives which accounts for certain cross-linguistic differences. The interrogative element occurring in a +Q Comp in languages like Japanese (ka) and Chinese (ne) may be considered to be of the same basic t ype as was in German or mit in Hungarian — an uninterpreted functional expletive element. However, in these languages no anti-localit y effects of the t ype seen to occur in Hungarian/German are attested, so that an expletive may freely co-occur with a wh-phrase in the same clause/tense domain: (106)
Tanaka-san-wa nani-o kaimashita ka? Tanaka-HON-TOP what-ACC bought Q ‘What did Mr. Tanaka buy?’
Wh-phrases may also occur in CNPs with the interrogative expletive in Comp of the same clause (vs (101) above): (107)
Tanaka-san-wa [dare-ga kaita hon]-o kaimashita ka? Tanaka-HON-TOP [who-NOM wrote book-ACC bought Q ‘Mr. Tanaka bought the book that who wrote?’
This basic difference between Japanese/Chinese t ype interrogative expletives and those in Hungarian/German may be given a simple explanation if it is assumed that the former do not carry Case-features (in need of checking) while the latter do. In (106)–(107) it is only the wh-phrase nani-o and the CNP-object that will
41
Note again here that the abilit y of a Case-based approach to account for the otherwise unexpected ungrammaticalit y of examples such as (102) and (103) provides further good support for the view that was is base-generated in SpecAgrO and then raised to SpecCP. The patterning in (102/103) is once again difficult to explain if one assumes was to be base-generated in Comp.
Partial Movement and wh-expletive wh questions
211
require Case-checking in SpecAgrO and the Caseless expletive will not have to raise through this position as well, possibly being base-generated directly in Comp; consequently the interrogative expletive may co-occur with these DP elements in the same clause and there is no German/Hungarian anti-localit y. In this sense ka and ne would be wh-equivalents to there-t ype expletives in English, while German was and Hungarian mit would parallel English it — Chomsky (1995) arguing that a cluster of properties differentiating it and there expletive constructions (and analogues in other languages) may be accounted for under the assumption that it but not there requires Case-checking. Finally the impossibilit y of wh-expletives occurring with wh-phrases in situ within non-finite and tense-dependent clauses such as examples (108) and (88) below can be accounted for with the same basic reference to failure of Casechecking. Essentially following Chomsky (1995, Section 5.6) it can be assumed that there are two lexical entries for embedding verbs such as wollen ‘want’ which t ypically occur here, and that these two forms critically differ in their Caselicensing potential; whereas one form of wollen will select a DP and have Case to assign to it, the second entry for wollen will be as a verb selecting a clause and not having any Case to assign. Where an infinitive clause or a tense-dependent clause occurs there will therefore simply be no Case-checking possibilit y for a wh-expletive and this will cause sentences such as (108)–(88) to crash. As can also be seen from (109) it is not possible to have a non-wh-expletive with verbs like wollen either, indicating that the problem here concerns the licensing of the expletive rather than any other wh-related anti-localit y effect: (108)
(88)
(89)
*Was willst du [morgen was kaufen]? WH want you [tomorrow what buy-INF Intended: ‘What do you want to buy?’ *Was willst du dass er wen besticht? WH want you that he whom bribe Intended: ‘Who do you want him to bribe?’ *Ich will es, dass er Johann besticht. I want it that he Johann bribes Intended: ‘I want him to bribe Johann.’
3.6.2. The clause-boundedness of wh-expletives wh In addition to the patterns examined above, there are certain other non-trivial restrictions on Partial Movement constructions which also still remain to be explained. Essentially there are three questions relating to the position and
212
W h-movement and the theory of feature-checking
occurrence of wh-expletives in German and Hungarian which need to be examined, the first of these relating directly to German Dialect A. In Dialect A it has been noted that an overt wh-expletive must occur in the Comp of each tense domain between the +Q Comp and the tense domain containing the partiallymoved substantive wh-phrase. Consequently (110) below is fine if the intermediate clause contains a second wh-expletive in its Comp, but is quite unacceptable to speakers of Dialect A if no wh-expletive occurs here, as shown in (111):42 (110)
Was glaubst du, was er meint, wo Hans hingegangen ist? believe you WH he thinks where Hans gone has ‘Where do you believe he thinks Hans went?’ *Was glaubst du, (dass) er meint, wo Hans hingegangen ist? WH believe you (that) he thinks where Hans gone has WH
(111)
The question which needs to answered is why is it apparently not possible for a single wh-expletive to be base-generated in the intermediate clause and raise up to the matrix +Q through the Comp of the intermediate clause so resulting in a surface form such as (111)? It has been argued that German was is an XP rather than an X0 head and so one might anticipate that long unbounded movement should be possible. The fact that examples such as (111) are ungrammatical however suggests that long movement of the wh-expletive from an intermediate clause to the higher +Q Comp is not a possibilit y for speakers of Dialect A. A second related question concerns parallel multi-clausal structures in German Dialect B and Hungarian. In both Dialect B and Hungarian it seems to be (optionally) possible for an intermediate tense domain to occur without any second wh-expletive in its Q Comp position, i.e. example (111) is acceptable to speakers of German dialect B and Horvath (1997) suggests that similar structures are also acceptable to speakers of Hungarian. The question then is obviously what allows for this difference between German Dialect A on the one hand and Dialect B and Hungarian on the other and is the underlying syntax of PM questions for Dialect B and Hungarian speakers really significantly different from German Dialect A or not? A third issue left unresolved from Section 3.6.1 is the question why it appears that a wh-expletive must occur raised in some Comp and always results in unacceptable forms if left in an in situ position, i.e. why attempted PM questions such as example (i) from footnote (40) repeated below as (112) is unacceptable for
42
As noted earlier, it is also normally possible to rescue structures such as (110) by raising the substantive wh-phrase to the intermediate Comp position.
Partial Movement and wh-expletive wh questions
213
speakers of both Dialects A and B (similar forms being unacceptable in Hungarian too): (112)
*Wer glaubt was, wo er hingegangen ist? who-NOM believe WH where he gone has Intended: ‘Who believes he went where?’
Considering example (112) first, one could argue that if a full substantive wh-phrase such as wer ‘who’ in (112) is raised to the +Q Comp triggering this as +wh, there should in fact be no reason for Partial Movement of the secondary substantive wh-phrase wo ‘where’ to take place in the lower clause and that this is the source of the ungrammticalit y. Partial Movement is only necessary where a wh-expletive wh occurs in a +Q Comp, a lower wh-phrase having to raise to a position where it is visible to the +Q Comp occupied by the wh-expletive to determine it as +wh. In (112) full triggering of the +Q Comp is effected by the higher substantive wh-phrase wer and so no PM is required, hence none should take place (by economy). If this were to be correct however, one would then expect that a variant form of (112) with no Partial Movement and the secondary substantive wh-phrase occurring fully in situ should be well-formed. Nevertheless such forms are equally unacceptable, as (113) shows: (113)
*Wer glaubt was, dass er wo hingegangen ist? who-NOM believes WH that he where gone has Intended: ‘Who believes he went where?’
Possibly one might suggest that examples such as (113) are ill-formed because the wh-expletive is redundantly present in such structures, performing no function at all and therefore unlicensed, violating the Principle of Full Interpretation. One might assume that wh-expletives must fulfil their role of altering the triggering localit y of a +Q Comp in order to be licensed as legitimate syntactic objects. An alternative and more promising explanation however, I believe, may be to suggest that wh-expletives in German are (wh-)clitics which need to attach to a Comp-host for PF-related reasons. They will therefore always be forced to undergo raising (to Comp) and may not occur in situ, as in (112) and (113). In general support of such a view, there is actually good phonological evidence in Iraqi Arabic that wh-expletives may indeed be clitic-like. In Iraqi Arabic the wh-expletive equivalent to German was is derived from the word for ‘what’ sheno, and optionally reduces to sh- and cliticizes to the verb (which one may assume to be raised to Comp):
214 (114)
W h-movement and the theory of feature-checking Sh- tsawwarit Mona [Ali raah weyn]? WH thought Mona [Ali went where ‘Where did Mona think that Ali went?’
As noted in an earlier section, according to Wahba (1991) sequences such as (114) may alternate with forms in which the wh-expletive appears in an unreduced form as sheno and occurs between the verb and the subject. It would therefore seem that there is a clear ongoing process of cliticisation taking place with the wh-expletive in Iraqi Arabic. Interestingly it would also appear that the wh-expletive in Hindi is similarly becoming a wh-clitic. Mahajan (1996) points out that the wh-expletive kyaa (again homophonous with the word meaning ‘what’ in Hindi) cannot be separated from the verb by any material at all and seems to be turning into an element which is proclitic on the verb, just as seen above in Iraqi Arabic. If this may also be true of German as suggested, it will allow for an explanation of (112)–(113), which otherwise remain rather puzzling — was in both cases needs to undergo cliticisation (to a Comp host) and is not acceptable left stranded in situ. If wh-expletives are thus taken to be clitics which attach to a ±Q Comp, the unexpected boundedness of wh-expletive movement is also easily accounted for — when raised to the Comp of an intermediate clause (as in (110)) the expletive will obligatorily cliticise to its Comp host and will not be able to raise any higher. Consequently, independent overt wh-expletives will need to occur in the Comps of all tense domains between the +Q C and the Comp occupied by the partiallymoved wh-phrase in order to extend the triggering domain as suggested.43 Turning to Hungarian now, one finds that all speakers accept multi-clausal examples similar to those in German Dialect A, with a wh-expletive in both the +Q Comp and also in the Q Comp of an intermediate clause, as in (115):
43
Mahajan (1996) tentatively suggests a similar conclusion for Hindi. As in German Dialect A, all tense domains between the +Q Comp and the tense domain containing a substantive wh-phrase must contain a wh-expletive: (i) raam-ne kyaa socaa [ki ravii-ne *(kyaa) kahaa [ki kOn aayaathaa]] Ram-ERG WH thought [that Ravi-ERG *(WH said [that who came ‘Who did Ram think that Ravi said came?’ It would therefore seem that kyaa in Hindi also cannot undergo long-movement from an embedded clause to a higher clause containing the +Q Comp, unlike other wh-DPs in the language. As with German and Iraqi Arabic one can again suggest with Mahajan that this is due to the clitic-like nature of the wh-expletive.
Partial Movement and wh-expletive wh questions (115)
215
Mit mondtal [hogy mit tudnak [hogy melyik WH-ACC said-INDEF-2SG [that WH-ACC know-INDEF-3PL [that which fiut szereted]? boy-ACC like-DEF-2SG ‘Which boy did you say they know you like?’
In addition to multi-wh-expletive structures, Horvath (1997) suggests that forms in which there is no intermediate wh-expletive are also acceptable, as illustrated in (116): (116)
Mit mondtal [hogy tudnak [hogy melyik fiut WH-ACC said-INDEF-2SG [that know-INDEF-3PL [that which boy-ACC szeretted]? like-DEF-2SG ‘Which boy did you say they know you like?’
In fact it appears that such forms are actually not acceptable to all speakers and there would seem to be a dialect split in Hungarian similar to that in German. As in German we can therefore refer to those who do not accept structures such as (116) as being speakers of Dialect A, and to those who do accept such forms as speaking Dialect B. Considering (116) now in Dialect B (hence as acceptable for this dialect’s speakers), although it might initially be assumed that the intermediate clause in examples such as (116) does not contain any wh-element and that the wh-expletive mit is simply base-generated in the higher clause, Horvath (1997) suggests that this is actually not the case and there is evidence that some form of wh-element has in fact been present at some point in the intermediate clause. Forms such (116) are shown to contrast with other similar multi-clausal PM structures with no intermediate wh-expletive which speakers of both dialects find unacceptable: (117)
*Mit mondtal [hogy tudjak [hogy melyik fiut WH-ACC said-INDEF-2SG [that know-DEF-3PL [that which boy-ACC szereted]? like-DEF-2SG ‘Which boy did you say they know you like?’
The critical minimal difference between (116) and (117) is the presence of indefinite agreement on the verb in (116) but not (117). Where indefinite agreement occurs it appears that a structure with no overt intermediate wh-expletive is acceptable, but when it is absent in (117) the question is ill-formed. As this indefinite agreement otherwise only occurs where an accusative-marked
216
W h-movement and the theory of feature-checking
wh-object has undergone movement (as noted in Section 3.2.1), its necessary presence in (116) would seem to indicate that a wh-element has indeed undergone raising in this intermediate clause. Significantly this suggests that in Hungarian there must indeed somehow be a linking of the +Q Comp through all intervening tense domains to the clause in which the real wh-phrase is located, just as in German Dialect A, and wh-expletive question forms in Iraqi Arabic and Hindi. Horvath therefore suggests that the wh-expletive which occurs in the matrix +Q Comp is actually base-generated in the intermediate clause and then raised successive-cyclically to its Spell-Out position, its movement through the intermediate clause SpecAgrO first triggering indefinite agreement on the verb in this clause before raising higher still and causing similar agreement on the verb in the matrix clause. Although a successive-cyclic movement analysis will indeed clearly account for the occurrence of indefinite agreement in the intermediate clause in (116), there are other reasons to doubt whether this might be the correct interpretation of the patterning and I would like to suggest instead a rather different conclusion, namely that wh-expletives are clause-bound clitics in Hungarian too, with the difference between (116) and German Dialect A being that Hungarian Dialect B allows for optional phonetic deletion of intermediate wh-expletives whereas German Dialect A does not. One simple reason for preferring such a PF deletion account to a long-movement analysis is that theoretically one would not expect it to be possible for the wh-expletive mit in (116) to be able to trigger indefinite agreement in both matrix and intermediate clauses. As noted in Section 3.2.1, indefinite agreement is strongly associated with accusative Case and only overt accusative-marked object wh-movement gives rise to indefinite agreement. It was therefore concluded that accusative Case and indefinite agreement must be checked by a wh-object in the same position, SpecAgrO, and that only accusativeCase-triggered overt movement through SpecAgrO will result in the occurrence of indefinite agreement. Considering example (116) again, in order for a single wh-expletive to check/trigger indefinite agreement in both intermediate and matrix clauses, such a wh-expletive would have to raise through two (accusative) Case positions, which should not be possible. If a wh-expletive inserted in the intermediate clause has its case checked in SpecAgrO of this clause, it should not be able to raise to check such Case-features again in a higher SpecAgrO, as Case-features being interpretable allow for checking only in a single position. The situation here is parallel to the impossibilit y of raising a subject DP from a position where it has already checked nominative Case to a higher +Case or Case-less subject position to check D0-features on T0 (the EPP):
Partial Movement and wh-expletive wh questions (118) (119)
217
*Johni seemed ti left. [Johni to seem [ti has left] is suspicious.
In (119) the need for (just) D0-features to be checked on T0 is not strong enough for T0 to be able to attract the DP [John] once this DP has checked Case in the lower SpecTP, and it therefore would not seem to be possible for a DP to undergo further raising from a position where it has successfully checked Case to another A-position, even if there are features in the A-head which could in fact be checked by the DP. Further confirmation that object-associated verbal agreement may only be checked by an object DP in a single clause comes from well-known patterns in French where it has been observed that a relativised or wh-raised object may only trigger/check object agreement in the clause in which it is base-generated and not in any higher clause: (120)
la fille qu’il a dit(*e) que Jean a vue the girl who he has said(-FEMSG) that John has seen-FEMSG ‘the girl who he said John saw’
Therefore it can be concluded that a wh-expletive which checks accusative Case and indefinite agreement in an intermediate clause in Hungarian should not be able to raise to the SpecAgrO position of a higher clause and check indefinite agreement also in this clause.44 If there are indeed two occurrences of indefinite agreement in examples such as (116), it would seem that these must instead be caused by the occurrence of two independent wh-expletives (which it is known can occur overtly for all speakers, as in (115)), with simple deletion of the intermediate wh-expletive being permitted after the expletive has raised to the intermediate Comp. In addition to these theoretical arguments, it is also possible to actually test empirically whether a wh-expletive is able to undergo long movement from an intermediate clause to a position in a higher clause. In a multi-clausal structure it is possible to select verbs in the matrix and intermediate clauses which will critically assign different cases to any wh-expletive base-generated in their clause. In example (121) the particular choice of matrix predicate will normally give rise to nominative Case occurring on a wh-expletive base-generated in the higher clause, while the verb in the intermediate clause regularly assigns accusative Case to a wh-expletive originating there. If it were to be possible for a wh-expletive to
44
Noting also that if a wh-expletive base-generated in the intermediate clause has to raise to the matrix through the intermediate SpecCP position (i.e. the position where an overt intermediate wh-expletive is always found), then raising from this position to the matrix SpecAgrO should clearly be a case of Improper Movement.
218
W h-movement and the theory of feature-checking
undergo long-movement out of an embedded intermediate clause up to a higher Comp position, in a structure such as (121) one should expect to find that an accusative-marked wh-expletive should be able to occur in the matrix +Q Comp, having raised to this position from the intermediate clause. Furthermore, if the possibilit y of multi-clausal wh-expletive questions such as (116) with no overt wh-expletive in the intermediate clause can only result from successive-cyclic movement of a wh-expletive from the intermediate clause, then the only t ype of wh-expletive expected to be possible in the matrix +Q Comp in (121) is one with accusative Case. Contrary to these clear predications, and to the extent that multiclausal structures with no overt intermediate expletive are acceptable at all (i.e. to the extent that Dialect B really does exist) it is found that the only possible wh-expletive which can occur in the matrix clause in (121) is actually one with nominative Case: (121)
(?)Mi/*mit
zavarja Marit, hogy hitt bother-DEF-3SG Mary-ACC that thought-INDEF-3SG Janos, hogy kivel beszelt? John-NOM that who-with talked-3SG ‘With whom does it bother Mary that John thought that she spoke?’ WH-NOM/WH-ACC
Similar results are achieved when the matrix predicate assigns accusative Case and the intermediate clause verb would assign allative Case to a wh-expletive basegenerated in its clause. If long successive-cyclic raising of an intermediate-clause wh-expletive were to be available, one would again expect that the only possible Case t ype on a wh-expletive in such a multi-clausal structure with no overt intermediate wh-expletive would be allative. However, as (122) shows, this turns out not to be possible and only an accusative-marked expletive reflecting the case assigned by the matrix predicate is actually permitted for those who do allow the intermediate clause Comp not to be filled by an overt wh-expletive (i.e. Dialect B speakers): (122)
Mit/*mire hitt Janos, hogy szamitasz, WH-ACC/WH-AL thought—INDEF-3SG John-NOM that count-2SG hogy mit fognak mondani a gyerekek? that what will-3PL say-INF the kids-NOM ‘What did John think that you expect that the kids will say?’
What this indicates is that long movement of a wh-expletive would not in fact seem to be possible and therefore the difference between structures such as (115) where a wh-expletive is overtly present in the intermediate clause and parallel forms such as (116) where there is no second wh-expletive should instead be
Partial Movement and wh-expletive wh questions
219
taken to be due to simple PF deletion of the intermediate expletive. Turning now to re-consider German once more, earlier we divided German into two basic dialect groups on the basis of how speakers react to multi-clausal PM questions such as (123) in which no wh-element occurs in the intermediate Comp position: (123)
%Was glaubst du, dass Hans meint, mit wemi Jakob ti WH believe you that Hans says with whom Jakob gesprochen hat? spoken has Intended: ‘Who do you believe that Hans thinks that Jakob spoke with?’
The majorit y of speakers, it might seem, reject such forms and require either that the substantive wh-phrase be raised to the intermediate Comp or that a second wh-expletive occur in this position, as in (124). This group was subsequently referred to as Dialect A: (124)
Was glaubst du, was Hans meint, mit wemi Jakob ti gesprochen believe you WH Hans says with whom Jakob spoken hat? has ‘Who do you believe that Hans thinks that Jakob spoke with?’
WH
Other speakers however do appear to accept forms such as (123) without any intermediate wh-expletive, and were therefore referred to separately as being speakers of Dialect B, without any further examination of what might actually underlie the dialect split. Significantly though, Dialect B here essentially manifests the same patterns optionally found in Hungarian Dialect B and one might therefore imagine that this similarlity in German Dialect B and Hungarian Dialect B should be accounted for in a uniform kind of way (i.e. if one is able to understand how Dialect A and Dialect B t ype forms are related within Hungarian, it should provide a useful means to better understand the parallel differences found in German). Above it has been argued that there is critical evidence in Hungarian in favour of a PF deletion account of structures which do not contain wh-expletives in intermediate Comp positions. A natural assumption to make would then be that what distinguishes German Dialect A from Dialect B really is a similar surface deletion phenomenon and that Dialect B speakers simply allow for the PF deletion of intermediate wh-expletives. If this is so, then there would not be any great syntactic differences between the two dialects and the dialect split would relate solely to a fairly low-level PF Spell-Out rule. A clear alternative to such a PF deletion account might be to assume instead that the single wh-expletive which
220
W h-movement and the theory of feature-checking
occurs in the matrix +Q Comp in sentences such as (123) in Dialect B is first generated in the intermediate clause and then raised to the matrix, undergoing long successive-cyclic wh-movement to the +Q Comp. However, there is reason to believe that this cannot in fact be an appropriate analysis of the pattern found in Dialect B and a PF deletion account is therefore to be preferred. In Fanselow and Mahajan (1996) two important observations are made which are highly relevant here. The first of these is that non-wh-expletives may not undergo long movement from a position within an embedded clause to the Comp of a higher clause, as shown in (125) and (126) below: (125)
(126)
*Esi denke ich dass ti regnen könnte it think I that rain could Intended: ‘I think that it could rain.’ *Esi glaubt Fritz ti kommt der Pfarrer. it believes Fritz comes the priest Intended: ‘Fritz thinks that the priest is coming.’
If non-wh-expletives are unable to undergo this kind of unbounded raising to a higher clause Comp then it is unlikely that wh-equivalents should be able to do so. Secondly, it is noted that speakers of Dialect B would not seem to allow for long unbounded movement of even regular non-expletive wh-DPs. If no other t ype of wh-DP and no other expletive is in fact able to raise between clauses in this way, it would therefore seem extremely improbable that a wh-expletive should be able to effect this kind of movement, and consequently it might seem that structures such as (123) are indeed created by the deletion of an intermediate wh-expletive just as suggested for Hungarian. If such a view can be maintained then it clearly suggests that certain of the apparent differences between the dialects and languages manifesting Partial Movement structures may in fact only be rather superficial and that underneath such surface variation lies a fundamentally uniform syntactic construction. The various complex restrictions on the occurrence of wh-expletives and perceived antilocalit y effects relative to substantive wh-phrases examined here may also ultimately all be accounted for without the need for additional assumptions and via means already justified by other independent syntactic phenomena. Although initially puzzling and apparently complex, the set of constraints which result in the phenomenon of Partial Movement questions might actually seem to be rather familiar and fully recurrent in other syntactic patterns.
Partial Movement and wh-expletive wh questions
221
3.6.3. Further variation Before closing this chapter with an overview and summary of the principal argumentation and proposals put forward, I would finally like to consider just a little further the issue of cross-linguistic uniformit y referred to above and review the motivation behind certain suggestions in the literature that Partial Movement might actually be a significantly different syntactic phenomenon in different languages. Horvath (1997) in particular suggests that PM structures in Hungarian are distinct from those in German with respect to important properties and that Partial Movement may therefore arise from a non-unitary source with quite a different underlying syntax in the two languages. Such a possibilit y would seem to be rather surprising. Throughout the patterning of PM questions in both Hungarian and German one continually finds so many strong and striking similarities that the most natural assumption to entertain is that the syntax of PM questions in German and Hungarian is fundamentally the same and that the tremendous similarities observed are not just the product of pure coincidence. Consequently it might seem more reasonable to attempt (at least initially) to explain any differences found in terms of either language-particular surface properties, as in Section 3.6.2 above, or some other less catastrophic parameter of cross-linguistic variation rather than simply abandon the assumption that the obvious parallels in German and Hungarian do indeed relate to a basically uniform syntax. Despite the obvious attraction of being able to account for Partial Movement phenomena in German and Hungarian in a more or less uniform way, suggestions have however been made that one should possibly provide quite different treatments of Partial Movement in the two languages because of the existence of certain observed non-parallelism in behaviour. What these differences might actually be is considered in Horvath (1997) and actually reduces to just two potential areas which might seem to be significant — the existence of differences in PM questions containing negation or negative quantifiers, and differences with predicates which embed factive clauses. With regard to negation, it was originally believed that there was an important asymmetry between German and Hungarian. In German it has been welldocumented since Rizzi (1992) that PM questions may not co-occur with negation higher than the partially-moved wh-phrase: (127)
*Was glaubst du nicht, mit wem er gesprochen hat? WH believe you not with whom he spoken has Intended: ‘Who don’t you think he spoke with?’
222
W h-movement and the theory of feature-checking
In contrast to German, in Hungarian it might appear that negation does not in fact constrain PM dependencies: (128)
Mit nem ismertbe Janos hogy WH-ACC not admitted-INDEF-3SG John-NOM that hanyszor hamisitotta az alairasodat? how-many-times forged-DEF-3SG the signature.2SG-POSS-ACC Lit:‘How many times is it that John didn’t admit that he had forged your signature that many times?’
However, Horvath (1997) notes that if one controls for the choice of predicate in the matrix clause which contains the negation and uses equivalents to the verbs which commonly occur in PM questions in German (such as ‘think’, ‘say’ ‘believe’ etc), then Hungarian PM questions actually do show the same sensitivit y to negation as found in German: (129)
*Mit nem gondolsz hogy kivel beszelt Mari? WH-ACC not think-INDEF-2SG that who-with spoke-3SG Mary-NOM Intended: ‘Who don’t you think Mary spoke with?’
Consequently further investigation reveals that the area of negation is not in fact a significant locus of variation between German and Hungarian and so ultimately no reason to posit underlying differences in the syntax of their PM questions. An interesting incidental observation made by Beck & Berman (1996) concerning the interaction of negation with Partial Movement structures in German is that negation or the presence of negative quantifiers would also seem to constrain the occurrence of certain other dependencies including wh in situ. In (130) below the presence of the negative quantifier in subject position c-commanding the wh-phrase in situ results in the multiple-wh question being severely degraded, this in the same way that negation/a negative quantifier will cause unacceptabilit y in PM questions: (130)
??Wen
i hat niemand wo ti gesehen? whom has no-one where seen Intended: ‘Who did no-one see where?’
(130) contrasts with similar forms which are actually quite acceptable where the wh-phrase is overtly raised across the negative quantifier: (131)
Woi hat niemand ti Karl gesehen? where has no-one Karl seen ‘Where did no-one see Karl?’
Partial Movement and wh-expletive wh questions
223
Beck & Berman suggest that the in situ wh-phrase in (130) will undergo raising at LF and that the relevant constraint on questions containing negation/negative quantifiers is that a negative element may not c-command a trace which has been created by LF movement. The same constraint will extend to cover the unacceptable PM case in (127) above on the assumption that the partially-moved wh-phrase mit wem undergoes similar LF movement to the +Q Comp leaving behind an illicit LF trace c-commanded by the sentential negation. Such a proposal is difficult to accept however as it suggests that the traces created by LF movement are critically different from those left by pre-Spell-Out raising, which seems to introduce an unjustified asymmetry into the model.45 In a standard Minimalist approach there is also no motivation for the in situ wh-phrase to move at LF in (130) as its wh-features are interpretable and so should not be in need of any checking. Instead it would seem that what the wh in situ and the PM cases in fact have in common is that neither the wh in situ element nor the partially-moved wh-phrase are related to the higher +Q Comp via movement, and it is this which critically distinguishes these wh-dependencies from actual overt wh-movement. It can be suggested that whereas negation does not block movement of a wh-phrase to a higher +Q Comp, it does block the licensing relation which the current account argues must necessarily exist prior to Spell-Out between the +wh+Q Comp and all wh-phrases present in a wh-question, i.e. the checking of wh-features on all these elements. Interestingly the fact that negation constrains not only PM structures but also wh in situ appears to confirm the suggestion made in this chapter that the relation of a partially-moved wh-phrase to the +Q Comp is actually no different from that of an in situ wh-phrase to a +Q Comp and that both require feature-checking in the same basic way. The second area of potential difference between German and Hungarian concerns the interaction of factive verbs and Partial Movement. In German it has been noted that Partial Movement questions are incompatible with factive and certain other non-bridge verbs occurring as the predicate in the matrix clause containing the +Q Comp: (132)
45
*Was bedauerst du, wen sie liebt? WH regret you whom she loves Intended: ‘Who do you regret that she loves?’
It must again be remembered that full category-movement should be a possibilit y at LF if this will result in a convergent derivation and therefore that the posited LF-movement in these cases should not be forced to leave traces which are in any way different from those left by overt raising.
224 (133)
W h-movement and the theory of feature-checking *Was hat er zugeflüstert, wen sie liebt? WH has he whispered whom she loves Intended: ‘Who did he whisper that she loves?’
This contrasts with Hungarian where parallel forms appear to be quite acceptable: (134)
Mit sajnalsz hogy hogy viselkedtek a gyerekek? WH-ACC regret-INDEF-2SG that how behaved-3PL the kids-NOM ‘In what way do you regret that the kids behaved in that way?’
Here then there would seem to be some real variation. However, Fanselow & Mahajan (1996) note that the situation in German is a little complex. Factive verbs would also appear to constrain extraction from V2-complement clauses such as (135), though movement out of regular factive complements as in (136) is generally much less degraded: (135)
(136)
*Weni bedauerst du, liebt der Mann ti? whom regret you loves the man Intended: ‘Who do you regret that the man loves?’ (?)Wen bedauerst du, dass er t liebt? i i whom regret you that he loves ‘Who do you regret that he loves?’
It is far from clear how to relate PM constructions with extraction from V2-clauses and why extraction of the same (referential) object wh-phrase is bad from V2-factive complements but more or less acceptable from regular factive complements. As there is consequently much unexplained in general about the patterning of wh-dependencies with factive verbs, it is therefore not easy to assess the significance of the factive verb-Partial Movement patterning in isolation and difficult to be sure of what one may really conclude from this data. Nevertheless, even if one does concede a genuine difference here between German and Hungarian, this actually remains the sole significant difference between the two languages pointed to in Horvath’s conclusions and is arguably also the main piece of evidence which has led both Horvath and others to suggest that PM phenomena in German and Hungarian may in fact be quite different syntactic constructions. Given the tremendous wealth of other common patterns and constraints shared by both languages and the relative lack of major differences, it would seem that the importance of the factive-verb distinction has been rather over-exaggerated and it is unexpected that one would choose to ignore the overwhelming similarities in German and Hungarian and favour a non-uniform approach to Partial Movement on the basis of such comparatively limited distinc-
Partial Movement and wh-expletive wh questions
225
tions. Instead it would appear that an analysis which will apply to both German and Hungarian in a largely parallel way should be pursued as vigourously as possible, and even if one were in fact to find other areas where the patterning of two languages is different in some way, these differences should be shown to be both highly significant and irreducible to any possible lexical or surface-type parametrisation before abandoning the clearly natural goal of a uniform analysis.46
3.7. Summary and concluding remarks I would now like to conclude with a brief summary of the account developed here in Chapter 3, and then follow this with some very general closing remarks addressing the broader orientation and goals of all three chapters of the work. Chapter 3 started off identifying the central problem posed by Partial Movement constructions for all potential Minimalist (and GB) accounts of such structures, namely the observation that in wh-expletive questions in German and
46
Horvath also compares German and Hungarian with Hindi on the grounds that both Mahajan (1990, 1996) and Srivastav (1991) have suggested a clausal pied piping account of Hindi wh-expletive questions which is similar to Horvath’s analysis of Hungarian. Horvath draws attention to the fact that Hindi is different from German and Hungarian in that the expletive element kyaa actually licenses and permits embedded yes/no questions as well as embedded wh-elements. This is certainly true, but it is possible that there are in fact now two homophonous and separate lexical items in Hindi pronounced as kyaa, one with an associated wh value and the other with a yes/no specification. The evidence which might lead to such a conclusion is that a kyaa which licenses an embedded wh-element absolutely must occur cliticised to the front of the verb and the kyaa-verb sequence cannot be interrupted by any other element, whereas when kyaa licenses a yes/no question, the kyaa-verb sequence may be interrupted as seen in (ii), this observation being made by Mahajan (1996): (i) siitaa-ne kyaa (*abhii) socaa ki ravii-ne kisko dekhaa Sita-ERG WH (*just-now thought that Ravi-ERG who saw ‘Who did Sita think (just now) that Ravi saw?’ (ii) siitaa-ne kyaa kal tumhe dekhaa thaa? Sita-ERG yes/no+Q yesterday you-DAT-saw saw be-past ‘Did Sita see you yesterday?’ This suggests that kyaa as a yes/no element is possibly rather different from wh-kyaa and may in fact have a discrete lexical entry. Whatever is the correct characterisation here, the current chapter does not take Hindi to have the same underlying syntax as German and Hungarian for the simple reason that there are indeed highly significant differences between wh-expletive constructions in these languages as examined in sections 3.5.3 and 3.5.4, and furthermore there is no forced Partial Movement of the t ype found in German and Hungarian.
226
W h-movement and the theory of feature-checking
Hungarian there is obligatory overt movement of a wh-phrase to a position in which in any standard approach it actually cannot check its wh-features (or satisfy the Wh Criterion). Presenting general arguments against both direct and indirect dependency analyses assuming LF movement as a means to account for PM, three important characteristics of PM constructions were singled out as requiring explanation in any potential approach. First of all the Partial Movement of the real wh-phrase is forced to take place prior to Spell-Out. Second, this raising must establish a clear linking to some higher +Q Comp. Third, it cannot be suggested that any non-wh features such as focus or non-wh D0 features may be the trigger for this movement of the wh-phrase and it has to be assumed that it is critically the wh-specification common to wh-phrases which licenses and forces PM. The conclusion drawn from a consideration of these three properties taken together was that Partial Movement takes place in order that wh-features carried by the wh-phrase are licensed prior to Spell-Out, and no other interpretation of the patterning found was argued to be available. Because the partially-moved wh-phrase occurs lower than the licensing +Q C0 in a Q Comp at Spell-Out this conclusion was noted to be highly problematic for the Checking Uniformit y Hypothesis which requires strict Spec–Head/Head-adjoined localit y for any feature-checking relation to be effected. The approach developed in Chapter 2 on the basis of quite independent patterns however significantly does permit just this kind of licensing at a distance. In Chapter 2 it was furthermore also argued that the wh-checking domain in languages such as German is indeed the sentential m-command domain of the +Q Comp. Consequently, satisfaction of the posited pre-Spell-Out checking requirement of the wh-phrase in a non-Spec–Head-local relation with its checking head was shown to be unproblematic and actually a source of clear additional support for the approach initiated in Chapter 2. The important issue which did remain in need of addressing however was why PM of the wh-phrase should be necessary at all given that full in situ licensing of wh-phrases is suggested to be possible in German. In attempting to answer this question, it was pointed out that that the linking of the partially-moved wh-phrase to the +Q Comp in PM constructions appears to be constrained by the occurrence of tense in a way which closely resembles the tense domain constraint on the licensing of wh-dependencies in Iraqi Arabic and Hindi considered in Chapter 2. This then led to an examination of wh-expletive questions in Iraqi Arabic and Hindi and the discovery that such constructions are tense-sensitive as well but in a way which is different to non-wh-expletive questions. It was seen that the introduction of a wh-expletive into a +Q Comp in these languages has the clear effect that the wh-licensing domain of the +Q Comp is increased to include the adjacent tense domain. Once again the licensing of wh-phrases was found to
Partial Movement and wh-expletive wh questions
227
be necessary prior to Spell-Out and this requirement forced all wh-phrases located in deeply embedded positions to raise overtly into the tense domain adjacent to the +Q Comp for pre-Spell-Out checking at a distance by the higher +Q C0. Technically it was suggested that raising of the wh-expletive to Comp in such question forms resulted in identification of a (phonetically null) lexical instantiation of a +wh+Q C0 with different properties and localit y values than the default +wh+Q C0. Such a view then satisfied an expectation created in Chapter 2 when it was originally suggested that surface variation found in wh-questions across languages directly results from the occurrence of differently-valued +wh+Q C0 elements in the lexicon of different languages. There it was argued that if the t ype of +wh+Q C could possibly vary across languages, one might also find that an individual language could have available in its lexicon more than just a single instantiation of a +wh+Q C. This prediction now seemed to be borne out by the Iraqi Arabic and Hindi wh-expletive patterns analysed in the way suggested. Armed with the insights provided by the wh-expletive patterns in Iraqi Arabic and Hindi, we then returned to German and Hungarian to see if the similar tense-related wh-patterns in Partial Movement constructions might be explained by some parallel function of the wh-expletive in these instances too. A significant difference was however found between the German/Hungarian cases on the one hand and Iraqi Arabic/Hindi wh-expletive questions on the other. In the latter it is clear that introduction of the wh-expletive into Comp results in a change of the pre-Spell-Out licensing localit y affecting all wh-phrases whereas raising of a wh-expletive to a +Q Comp in German and Hungarian seems to result in a localit y change which directly affects just a single substantive wh-phrase in multiple-wh questions, forcing only one of the wh-phrases present to undergo Partial Movement from its in situ position. Noting that this clearly seems to parallel the pattern found in regular multiple-wh questions where only a single wh-phrase is actually forced to undergo pre-Spell-Out raising, it was suggested that the Partial Movement of a wh-phrase in wh-expletive questions may be concluded to take place for essentially the same reason that raising of a single wh-phrase occurs in non-expletive wh-questions, to result in the triggering of some higher C0 as +wh+Q. Attempting to understand how this might be the case in wh-expletive questions, it was then suggested that the wh-expletive which might otherwise be expected to be able to trigger a C0 as +wh+Q may be assumed to be critically defective in a way very similar to other expletive types. Like certain other expletives which are found to lack some purely formal propert y and require that this propert y is supplied on/by a local associate element, in German and Hungarian it was argued that despite being morphologically +wh, wh-expletives in these languages actually lack the necessary potential to successfully trigger a C0-head as
228
W h-movement and the theory of feature-checking
+wh+Q and that the presence of a wh-operator in a wh-element is importantly what is required to effect the triggering of Comp. Following Tsai (1994) it was suggested that where a wh-operator may be base-generated is a locus of crosslinguistic lexical variation and that such an operator is crucially absent from wh-expletives in German and Hungarian. As a direct consequence of this deficiency, Partial Movement of a substantive wh-phrase carrying a wh-operator was argued to become necessary to bring the wh-operator into a local relation with the Comp to be determined as +wh, and raising of the wh-phrase to the Comp of a lower clause was suggested to result in the wh-operator carried by the wh-phrase it and its operator being sufficiently visible to the C0 in the higher tense domain for triggering to be effected. The role of the wh-expletive in establishing this preSpell-Out triggering of Comp was argued to be that the wh-expletive identifies a different triggering localit y in the higher C0-position. Whereas a wh-operator would normally have to be located immediately in the Spec or head position of a C0 to be determined as +wh in order to effect triggering, when the wh-expletive occurs raised to Comp it critically alters this localit y requirement and allows for a wh-operator made visible within an adjacent tense domain to successfully identify the higher Comp as +wh+Q. Following this basic characterisation of the interaction of wh-expletives and Partial Movement, the chapter also offered various suggestions as to how various other properties and restrictions of wh-expletive constructions could be explained and argued that PM constructions in German and Hungarian should be taken to be an essentially uniform phenomenon. Generally, in addition to providing a principled account of PM questions which firmly recognises the important pre-Spell-Out nature of Partial Movement and argues that this significant propert y cannot be satisfactorily explained by means of any LF/covert movement analysis, the chapter also resulted in a number of broader conclusions. First of all, the paradigms examined in PM structures once again strongly confirmed the need to acknowledge occurrences of “checking at a distance” in certain instances and so presented further supporting arguments against the Checking Uniformit y Hypothesis as a universal constraint. Secondly, a consideration of wh-expletive questions in Iraqi Arabic and Hindi was argued to indicate that the localit y of a checking domain associated with a single functional head may actually show certain variation depending on the particular lexical instantiation of that head in any token sentence. Thirdly, reflecting on the often deficient nature of expletives and the requirement that any formal properties lacking in an expletive must be supplied by a lexical associate, it was suggested that these properties are not (necessarily) supplied via any LF raising and replacement operation, but rather simply have to be located in some local relation to the
Partial Movement and wh-expletive wh questions
229
relevant functional head. Expletives are characteristically under-specified elements which may be used to partially identify properties of a functional head, and one of these may be a localit y requirement on the location of other necessary properties not present in the expletive, allowing these properties to be found in a wider local relation. A fourth point to note and re-emphasise here is that all information necessary for the licensing of wh-dependencies in PM and other wh-expletive questions in Iraqi Arabic and Hindi would clearly seem to be explicitly encoded in the pre-Spell-Out forms of such questions, adding further reason to believe that wh-licensing must in fact always be effected in the overt syntax and not at any further covert level of LF. Finally, I would like to close this chapter with some thoughts on the wider issues under discussion in all three chapters of the book. On a very general level, the major issue of this book has been an attempt to confront and investigate the issue of uniformity in language and how one should attempt to balance out assumptions of underlying uniformit y in language against the clear variation that one actually does find across languages on the surface. Adopting a broad Chomskyean approach to the problems posed by language acquisition and assuming that some significant underlying uniformit y (i.e. UG) must indeed be responsible for the tremendous success and speed of language acquisition, important questions arise as to where exactly in language this uniformit y is located, what shape the uniformit y takes and how one can accommodate any surface variation with the assumption of a uniform and universal syntax. Put in other terms, one needs to ask which aspects of the grammar should be taken to be fully uniform and invariant and where there may be genuine variation due to certain lexical parametrisation. To probe the issue in some depth, the chapters here have concentrated almost exclusively on wh paradigms as a general testing ground, for the reason that such dependencies appear to show significant variation across languages and therefore offer a great potential to investigate how one should best try to reconcile assumptions of uniformit y with apparent variation. It was noted that in standard Minimalist approaches and also late GB, a common way to address the uniformit y-variation issue has been to suggest that if certain surface variation is perceived across languages but an underlying uniformit y is assumed, then this uniformit y simply must be present at some more abstract level of derivation. The result of such reasoning has therefore been to posit a level of LF where the apparent variation across languages can be suggested to be largely eradicated and languages will present uniform structures to general processes of
230
W h-movement and the theory of feature-checking
interpretation.47 In Minimalism itself the development of a theory of featurechecking with its accompanying rigid and uniform licensing localit y has been particularly important as a mechanism to ensure uniformit y at the LF level and so considerable attention has been given here to the justification of such a principle and its necessary LF output. The general question specifically under consideration in this study was whether or not the localit y constraining the licensing of many syntactic dependencies should indeed be taken to be part of the invariant underlying uniformit y which we take to characterise human language and whether the projection of a level of LF is actually necessary for the encoding of putative universal forms. Considering the LF issue and wh-dependencies, in Chapter 1 it was argued that careful re-examination of a wide variet y of arguments for LF movement based on patterns found in wh in situ constructions reveals that such constructions ultimately do not in fact provide convincing evidence in favour of LF, and that all the wh patterns taken to support a level of LF may actually be accounted for without the need to posit covert LF movement. In Chapter 2 we then considered a wh paradigm found in both Iraqi Arabic and Hindi which strongly seemed to challenge the assumption that a uniform strict localit y might govern all featurechecking relations and argued that the problematic patterning may not be explained away simply by invoking a further covert level of derivation. Rather it was suggested that the paradigm forces one to assume that feature-checking relations are not universally subject to a fully rigid licensing localit y and that there is consequently no absolute uniformity in the way that many movement-associated syntactic dependencies are formally established. The necessit y of a level of LF to allow for and encode hypothetical strict localit y between functional heads and the features licensed by such heads in feature-checking dependencies was therefore shown to be severely challenged. Chapter 2 also showed that other general support for a uniform level of LF earlier provided in the GB period is furthermore lost when certain standard Minimalist assumptions are adopted. Finally in Chapter 3 we have seen additional patterns indicating again that the Checking Uniformit y Hypothesis cannot be maintained as a universal restriction and that
47
In fact, it may be noted that current Minimalist models do not in fact assume that full uniformit y is encoded at LF, despite the suggestion that feature-checking relations must be established in a uniform way by this level. At least with A-chains, at LF there will be differences between languages depending upon whether A-movement is overt and pied pipes the semantic features of a DP or whether it is covert and strands the DP’s semantic features. This clear non-uniformit y caused by PF-factors is then not cancelled out by the assumption of a covert level of syntax encoding crosslinguistic uniformit y.
Partial Movement and wh-expletive wh questions
231
a certain limited amount of variation must be conceded in the localit y constraining the licensing of feature-checking relations. The global conclusions to result from all of this are that: (a) the localit y constraining feature-checking relations is in fact not part of the invariant uniformit y taken to underlie language, (b) there is consequently no necessit y to assume a level of LF simply to express this hypothesised uniform licensing localit y, and (c) there is quite generally considerable justification for a Minimalist model without a discrete level of LF and an approach in which the forms produced by Spell-Out are taken to be the essential input forms required by general processes of interpretation.48 Such conclusions now return one to the uniformit y issue again and the question of whether the adoption of these assumptions actually results in any real loss of uniformit y in the Minimalist model of language supposed. The answer to this can be argued to be “no”, I believe. Faced with the apparent confusion of large-scale cross-linguistic surface variation one cannot be sure ahead of time which aspects of language will actually make up the underlying rigid uniformit y assumed to exist across all languages and careful examination is clearly necessary before one can assert with confidence the full universalit y of any phenomenon. Here a detailed investigation of a wide variet y of wh-dependencies has ultimately led to the conclusion that the localit y of feature-checking is an aspect of syntax which is in fact subject to certain variation triggered by the properties of lexical items instantiating various functional heads. What now remains as significantly universal in the model is the assumption that there exists a fixed array of hierarchically ordered functional projections across all languages establishing dependencies with associated lexical items. Critically it is the realisation of the mapping operation between the set of lexical and functional elements which is argued to be potentially satisfied in (somewhat) different local relations and therefore not be part of any invariant universal underlying design. Reflecting once again on the general notion of uniformit y in language it can be suggested that the above conclusions ultimately do not give rise to any genuine loss of uniformit y, but simply result in a finer understanding of what in fact comprises rigid uniformit y. Uniformit y which can be taken as highly significant is assumed to be present in the cross-linguistic organisation of functional structure and the requirement that
48
As noted at the end of Chapter 2, if one adopts the Beghelli-Stowell (1997) view that the functional domain includes an array of quantificational projections, it can be suggested that quantificational elements are generally licensed in situ by one of these heads and that the licensing head then represents the scope of the in situ QP. This is entirely similar to the suggestion that the scope of an in situ wh-phrase is represented by the +Q Comp which licenses it, and allows for a fully general account of relative scope without the need for any covert raising.
232
W h-movement and the theory of feature-checking
functional heads successfully establish licensing relations with corresponding elements generated in lower lexical domains. Relatively low-level and insignificant variation is here argued to be found in the way that such dependencies are constrained to be physically effected, this due to a number of largely morphological and lexical factors which may vary across languages, some of which were detailed in the summary to Chapter 2. The final result is then suggested to be a streamlined Minimalist approach which is actually compatible with much of the fundamentally derivational view of syntax argued for in recent years, and also now possibly just a little clearer as to where the important divide lies between full underlying uniformit y and limited parametric variation.
Bibliography
Anyadi, S. & A. Tamrazian. (1993). Wh-movement in Armenian and Ruhr German. UCL working papers in linguistics 5, edited by J. Harris, 1–22, Universit y College London. Aoun, J, N. Hornstein, D. Lightfoot, & A. Weinberg. (1987). Two t ypes of localit y. Linguistic inquiry 18, 537–577. Aoun, J. & Y-H. A. Li. (1991). The interaction of operators. Principles and parameters of comparative grammar, edited by R. Freidin, 163–181 Cambridge, Mass.: MIT Press. Aoun, J. & Y-H. A. Li. (1993). Wh-elements in situ: Syntax or LF. Linguistic inquiry 24, 199-238. Barss, A. (1984). Chain binding. Manuscript, Cambridge: Massachusetts Institute of Technology. Beck, S. (1996). Quantified structures as barriers for LF movement. Natural Language Semantics 4, 1–56. Beck, S. & S. Berman. (1996). Wh-scope marking: Direct vs. indirect dependency. Papers on wh-scope marking, Arbeitspapiere des Sonderforschungsbereichs 340, edited by U. Lutz & G. Müller, 59–83. Universit y of Stuttgart & Universit y of Tübingen. Beghelli, F. (1997). The syntax of distributivit y and pair-list readings. Ways of scope taking, edited by A. Szabolcsi, 349–408. Dordrecht: Kluwer. Bresnan, J. (1976). Evidence for a theory of unbounded transformations. Linguistic analysis 2, 1–51.
234
Bibliography
Brody, M. (1990). Some remarks on focus field in Hungarian. UCL working papers in linguistics 5, edited by J. Harris, 201–225, Universit y College London. Brody, M. (1995). Lexico-logical form: A radically minimalist theory. Cambridge, Mass.: MIT Press. Castillo, J. C., J. E. Drury & K. K. Grohmann. (1999). Tensed domains and expletive constructions. Paper presented at the Linguistics Societ y of America 73rd Annual Meeting, Los Angeles, 9th January. Cheng, L. L.-S. (1997). On the typology of wh-questions. New York: Garland. Cheng, L. L.-S. (to appear). Moving just the feature. Wh-scope marking, edited by U. Lutz, G. Müller & A. von Stechow. Amsterdam: John Benjamins. Chierchia, G. (1992). Questions with quantifiers. Natural language semantics 1,181–234. Chomsky, N. (1981). Lectures on government and binding. Dordrecht: Foris. Chomsky, N. (1973). Conditions on transformations. A Festschrift for Morris Halle, edited by S. Anderson & P. Kiparsky, 232–86. New York: Holt, Rinehart & Winston. Chomsky, N. (1977). On wh-movement. Formal syntax, edited by A. Akmajian, P. Culicover & T. Wasow, 71–132. New York: Academic Press. Chomsky, N. (1981). Lectures on government and binding. Dordrecht: Foris. Chomsky, N. (1982). Some concepts and consequences of the theory of Government and Binding. Cambridge, Mass.: MIT Press. Chomsky, N. (1986a). Knowledge of language: Its nature, origin and use. New York: Praeger. Chomsky, N. (1986b). Barriers. Cambridge, Mass.: MIT Press. Chomsky, N. (1993). A minimalist program for linguistic theory. The view from building 20: Essays in linguistics in honor of Sylvain Bromberger, 1–52. Cambridge, Mass.: MIT Press. Chomsky, N. (1995). The minimalist program. Cambridge, Mass.: MIT Press. Chomsky, N. (1998). Minimalist enquiries: The framework. Manuscript, Cambridge: Massachusetts Institute of Technology. Cinque, G. (1991). Types of A′-dependencies. Cambridge, Mass.: MIT Press. Cinque, G. (1997). Adverbs and the universal hierarchy of functional projections. Manuscript: Universit y of Venice. Cole, P., G. Hermon & L.-M. Sung. (1990). Principles and parameters of longdistance reflexives. Linguistic inquiry 21, 1–22. Cole, P. & G. Hermon. (1994). Is there LF movement? Linguistic inquiry 25, 239–263.
Bibliography
235
Comorovski, E. (1989). Discourse and the syntax of multiple constituent questions. Doctoral dissertation, Cornell Universit y. UMI Dissertation Services, Ann Arbor, Michigan Cormack, A. & R. Kempson. (1991). On Specificit y. Meaning and truth, edited by J. Garfield, & M. Kiteley, 546–581. New York: Paragon House. Culicover, P. (1992). Topicalization, inversion, and complementizers in English. OTS working papers in linguistics, edited by D. Delfitto, 54–87, Utrecht Institute of Linguistics. Fanselow, G. & A. Mahajan. (1996). Partial movement and successive cyclicit y. Papers on wh-scope marking, Arbeitspapiere des Sonderforschungsbereichs 340, edited by U. Lutz & G. Müller, 131–161. Universit y of Stuttgart & Universit y of Tübingen. Fiengo, R, C.-T. J. Huang, H. Lasnik, & T. Reinhart. (1988). The syntax of wh in situ.Proceedings of the seventh west coast conference on formal linguistics 7, edited by H. Borer, 81–98. Stanford: the Center for the Study of Language and Information. Garrett, E. (1996). Wh in situ and the syntax of distributivit y. Syntax at Sunset, UCLA working papers in linguistics, edited by E. Garrett & F. Lee, 129–145. Universit y of California at Los Angeles. Grewendorf, G. (1999). The additional-wh effect and multiple-wh fronting. Specifiers, edited by D. Adger, S. Pintzuk, B. Plunkett & G. Tsoulas, 146–162. Oxford Universit y Press. Grewendorf, G. & J. Sabel. (1999). Scrambling in German and Japanese: Adjunction versus multiple specifiers. Natural language and linguistic theory 17, 1–65. Haik, I. (1987). Bound variables that need to be. Linguistics and philosophy 11, 503–530. Hermon, G. (1984). Syntactic modularity. Dordrecht: Foris. Higginbotham, J. & R. May. (1981). Questions, quantifiers, and crossing. Linguistic review 1, 41–80 Hornstein, N. (1995). The grammar of logical form: from GB to Minimalism. Oxford: Blackwell. Horvath, J. (1986). FOCUS in the theory of grammar and the syntax of Hungarian. Dordrecht: Foris. Horvath, J. (1997). The status of ‘wh-expletives’ and the partial wh-movement construction of Hungarian. Natural language and linguistic theory 15, 509–572. Huang, C.-T. J. (1982). Logical relations in Chinese and the theory of grammar. Doctoral Dissertation, MIT, Cambridge, Mass. Huang, C.-T. J. (1989). Pro-drop in Chinese: A generalized control theory. The null subject parameter. edited by O. Jaeggli & E. Safir, 185–214. Dordrecht: Kluwer.
236
Bibliography
Iatridou, S. (1986). An anaphor not bound in its governing category. Linguistic inquiry 7, 766–772. Jayaseelan, K. A. (1996). Question-word movement to focus and scrambling in Malayalam. Linguistic analysis 26, 63–83. Kayne, R. (1984). Connectedness and binary branching. Dordrecht: Foris. Kempson, R. (1995). Semantics, pragmatics and natural language interpretation. Handbook of contemporary semantic theory, edited by S. Lappin, 561–598. Oxford: Blackwell. Kitahara, H. (1992). Checking theory and scope interpretation without quantifier raising. Harvard working papers in linguistics 1, 51–71. Cambridge. Mass.: Harvard Universit y. Koopman, H. & D. Sportiche. (1982). Variables and the Bijection Principle. Linguistic review 2, 139–160. Koopman, H. (1996). The spec-head configuration. Syntax at Sunset, UCLA working papers in linguistics, edited by E. Garrett & F. Lee, 30–45. Universit y of California at Los Angeles. Ladusaw, W. A. (1980). Polarity sensitivity as inherent scope relations. New York: Garland. Lappin, S. (1992). The syntactic basis of ellipsis resolution. Proceedings of the Stuttgart workshop on ellipsis, Edited by S. Berman & A. Hestvik, 45–80. Stuttgart: Universit y of Stuttgart. Lasnik, H. & M. Saito. (1992). Move a. Cambridge, Mass.: MIT Press. Li, Y-H. A. (1992). Indefinite wh in Mandarin Chinese. Journal of east Asian linguistics 1, 125–155. Mahajan, A. (1990). The A/A-bar distinction and movement theory. Doctoral dissertation, Cambridge: Massachusetts Institute of Technology. Mahajan, A. (1996). Wh-expletives and the syntax of partial movement. Papers on wh-scope marking, Arbeitspapiere des Sonderforschungsbereichs 340, edited by U. Lutz & G. Müller, 163–177. Universit y of Stuttgart & Universit y of Tübingen. Manzini, R. & K. Wexler. (1987). Parameters, Binding Theory and learnabilit y. Linguistic inquiry 18, 413–414 May, R. (1977). The grammar of quantification. Doctoral dissertation, Cambridge: Massachusetts Institute of Technology. May, R. (1985). Logical form. Cambridge, Mass.: MIT Press. McDaniel, D. (1989). Partial and multiple wh-movement. Natural language and linguistic theory 7, 565–604. Müller, G & W. Sternefeld. (1996). A′-chain formation and economy of derivation. Linguistic Inquiry 27, 480–511. Nishigauchi, T. (1990). Quantification in syntax. Dordrecht: Reidel.
Bibliography
237
Nishiyama, K, J. Whitman, & E-Y. Yi. (1996). Syntactic movement of overt whphrases in Japanese and Korean. Japanese/Korean Linguistics 5, edited by N. Akatsuka, S. Iwasaki & S. Strauss, 337–352. Stanford: the Center for the Study of Language and Information. O’Neil, J. & Groat, E. (1995) Unifying Spell-Out and the LF interface in the minimalist framework. Paper presented at GLOW, Tromso, Norway. Ouhalla, J. (1996). Remarks on the binding properties of wh-pronouns. Linguistic inquiry 27, 676–707. Pesetsky, D. (1987). wh in situ: Movement and unselective binding. The representation of (in-)definiteness. edited by E. J. Reuland, & A. ter Meulen, 98–129. Cambridge, Mass.: MIT Press. Postal, D. (1993). Remarks on weak crossover effects. Linguistic inquiry 24, 539–56. Progovac, L. (1991a). Relativised SUBJECT, long-distance reflexives, and accessibility. Manuscript, Indiana Universit y. Progovac, L. (1991b). Polarit y in Serbo-Croatian: Anaphoric NPIs and pronominal PPIs. Linguistic inquiry 22, 567–572. Reinhart, T. (1994). Interface economy. Talk given at SOAS, Universit y of London. Reinhart, T. (1998). wh in situ in the framework of the minimalist program. Natural language semantics 6, 29–56. Rizzi, L. (1982). Issues in Italian syntax. Dordrecht: Foris. Rizzi, L. (1990). Relativized minimality. MIT Press, Cambridge, Mass. Rizzi, L. (1992). Argument/adjunct (a)ssymetries. Proceedings of the North East Linguistics Society, 22, 365–381. Amherst: Graduate Linguistic Student Association, Universit y of Massachusetts. Rizzi, L. (1994). Residual verb second and the Wh-criterion. Parameters and functional heads: Essays in comparative syntax. edited by A. Belletti & L. Rizzi, 78–102. Oxford Universit y Press. Rizzi, L. (1997). The fine structure of the left periphery. Elements of grammar, edited by L. Haegeman, 281–337. Dordrecht: Kluwer. Rudin, C. (1988). On multiple questions and multiple wh fronting. Natural language and linguistic theory 6, 445–501. Sabel, J. (1996). Asymmetries in Partial Movement. Papers on wh-scope marking, Arbeitspapiere des Sonderforschungsbereichs 340, edited by U. Lutz & G. Müller, 289–315. Universit y of Stuttgart & Universit y of Tübingen. Saddy, D. (1991). Wh scope mechanisms in Bahasa Indonesia. MIT working papers in linguistics 15, edited by L. L.-S. Cheng & H. Demirdash, 183–218. Cambridge: Massachusetts Institute of Technology.
238
Bibliography
Saito, M. (1986). Scrambling as semantically vacuous A′-movement. Alternative conceptions of phrase structure, edited by M. Baltin & A. Kroch, 182–200. Universit y of Chicago Press. Schütze, C. (1996). Korean “case stacking” isn’t: Unifying noncase uses of case particles. Proceedings of the North East Linguistic Society 26, edited by K. Kusumoto, 351–366. Amherst: Graduate Linguistic Student Association, Universit y of Massachusetts. Shi, D. (1994). The nature of Chinese wh-questions. Natural language and linguistic theory 12, 301–333. Simpson, A. (1993) LF wh-movement from Huang (1982) to Aoun & Li (1993). Manuscript, SOAS, Universit y of London. Simpson, A. (1994). Licencing of wh: a non-movement approach to the syntax of wh in situ. SOAS working papers in linguistics 4, edited by M. Cobb & S. Jensen, 191–232. SOAS, Universit y of London. Simpson, A. (1995). Partial wh-movement and the localit y of feature-checking. SOAS working papers in linguistics 5, edited by S. Jensen, 211–231. SOAS, Universit y of London. Simpson, A. (to appear). On covert movement and LF. The Minimalist Parameter. Amsterdam: John Benjamins. Srivastav, V. (1991). Wh dependencies in Hindi and the theory of grammar. Doctoral dissertation, Ithaca: Cornell Universit y. Stowell, T. & F. Beghelli, F. (1997). Distributivit y and negation: The syntax of ‘each’ and ‘every’. Ways of scope taking. edited by A. Szabolcsi, 71–107. Dordrecht: Kluwer. Stroik, T. (1992). English wh in situ constructions. Linguistic analysis 22, 133–153. Szabolcsi, A. (1997). Strategies for scope taking. Ways of scope taking, edited by Anna Szabolcsi, 109–154. Dordrecht: Kluwer. Takahashi, D. (1993). Movement of wh-phrases in Japanese. Natural language and linguistic theory 11, 655–678. Tancredi, C. (1990). Not only EVEN, but even ONLY. Manuscript, Cambridge: Massachusetts Institute of Technology. Tsai, W-T. D. (1994). On nominal islands and LF extraction in Chinese. Natural language and linguistic review 12, 121–175. Tsai, W-T. D. (1994). On economizing the theory of A-bar dependencies. Doctoral dissertation, Cambridge: Massachusetts Institute of Technology. Tsimpli, I. M. & A. Roussou. (1993). Polarit y items in modern Greek: Their distribution and interpretation. UCL working papers in linguistics 5, edited by J. Harris, 129–162. Universit y College London.
Bibliography
239
Urbina, J. O. de. (1990). Operator feature percolation and clausal Pied Piping. Papers on wh movement, edited by L. L.-S. Cheng & H. Demirdash, 193–208. Cambridge: Massachusetts Institute of Technology. Wahba, W.A-F. B. (1991). LF movement in Iraqi Arabic. Logical structure and linguistic structure, edited by C.-T. J. Huang & R. May, 253–276. Dordrecht: Kluwer. Watanabe, A. (1991). Wh in situ, subjacency, and chain formation. Manuscript, Cambridge: Massachusetts Institute of Technology. Watanabe, S. (1994). (Anti-)superiority as weak crossover. Manuscript, Universit y of Southern California, Los Angeles. Webelhuth, G. (1992). Principles and parameters of syntactic saturation. Oxford Universit y Press. Williams, E. (1984). There-insertion. Linguistic Inquiry 15, 131–153. Williams, E. (1986). The reassignment of the functions of LF. Linguistic inquiry 17, 265–99.
Index
A absorption 51–53, 113 agreement 138, 139, 198 antecedent contained deletion (ACD) 19–22, 37, 38, 88 anaphors 30, 31, 90 n. 18, 102 n. 21, 206 Anyadi, S. 156, 157 n. 6 Aoun, J. 17, 26, 57–59, 66, 117, 141 B Barss, A. 29 Basque 163, 164, 168, 169, 178–180 Beck, S. 155, 156, 208, 222, 223 Beghelli, F. 135 n. 44, 142 n. 46, 231 n. 48 Berman, S. 155, 156, 208, 222, 223 Bijection Principle 31, 32 Binding Theory 31, 42, 106 principle C 9, 10, 41, 42, Bresnan, J. 80, 87 Brody, M. 30, 95–97, 157 n. 6 Bulgarian 70, 99, 106–109 Burmese 199
C Case 99, 100–102 n. 20, 106 n. 24, 138, 139, 162, 198, 206–211, 216–218 Castillo, J. 149 n. 1 Checking Uniformit y Hypothesis (CUH) 71, 77, 82, 84, 97, 134, 137, 138, 183 Cheng, L. 9, 18 n. 6, 34 n. 17, 35 n. 35, 50, 51 n. 26, 77 n. 5, 105, 106 n. 24, 128, 132, 149 n. 1, 157 n. 7, 161 n. 11, 205 n. 37 Chierchia, G. 60 Chinese 1, 2, 9, 11, 14–17, 21, 23, 25–28, 34(fn18), 50, 55, 56, 59, 63, 64, 70, 89, 92, 102 n. 21, 105, 124–125, 148, 210 Chomsky, N. 1, 4, 5, 30, 31, 42, 69, 70, 72, 74 n. 4, 75, 76, 83 n. 13, 84, 87, 98, 100–103, 107, 111, 112, 120, 135–140, 153, 154, 157, 160 n. 9, 171 n. 17, 172, 181, 198, 199, 205 n. 37, 210 Cinque, G. 36, 80, 86, 113, 143 Clausal Typing Hypothesis 50, 105
242 clitic left dislocation (CLLD) 86 Cole, P. 23, 24, 102 n. 21 Comorovski, E. 45, 47, 48 comparative deletion 87 Cormack, A. 61 crossover strong 9, 10, 41, 42, weak 31–33 Culicover, P. 104 Czech 109 n. 28 D d-linking 10, 11, 43–46, 116 n. 32, 152, 158 n. 8 Drury, J. 149 n. 1 E economy 49, 51, 74 n. 4, 114, 119, 120–121, 137, 149, 150, 205, 209, 210, 213 ECP 11, 23–27, 36, 39, 46–49, 55, 93, 121 ellipsis 19, 20, 22, 37, 38 F Fanselow, G. 155, 172, 173, 175 n. 18, 207, 208, 220, 224 features ±interpretable features 5, 76, 83 n. 13, 84, 101, 102 ff-movement 5, 6 n. 2, 7 n. 3, 7, 72 n. 1, 74 n. 4, 81 n. 9, 83 n. 13, 111, 136, 157, 158, 159 n. 9, 205 n. 37 Fiengo, R. 16, 44, 64, 65 n. 30, 79 n. 7, focus 9, 104, 112–115, 168, 172–180 French 117, 139, 140, 143, 217 Full Interpretation (the Principle of) 213 G Garrett, E. 55, 109 n. 27 German 131–132 and throughout Chapter 3 g-projections 36
Index Greed (the Principle of) 98, 100, 101, 118 n. 33, 119, 120, 138 Greek 206 Grewendorf, G. 40 n. 21, 107 Groat, E. 96–97 Grohmann, K. 149 n. 1 H Haik, I. 20 Hermon, G. 23, 24, 102 n. 21, 164 Higginbotham, J. 52, 53 n. 28 Hindi 70, 82–86, 91–95, 105–107, 109, 144, 148, 186–196, 200, 202, 203, 214, 216, 225 n. 46 Hornstein, N. 26, 52 n. 27, 136 Horvath, J. 148, 149, 151, 155, 158 n. 8, 161, 163–171, 190, 193, 206, 212, 215, 216, 221, 222, 224, 225 n. 46 Huang, C.-T.J. 4, 11, 13–15, 19, 20, 23, 25, 28, 34 n. 18, 39, 49, 51–59, 73, 74, 114 n. 30, 123, 124, 143 Hungarian 148 I Iatridou, S. 205 Icelandic 138 identification 200 n. 35, 201, 202 indefinite wh 89–90, 90 n. 18, Indonesian 33–35, 65, 180 n. 23 inner negation islands 158 Iraqi Arabic 70, 71–90, 105–107, 109, 144, 148, 186–190, 194–196, 200, 202, 203, 213, 214, 216 Italian 86, 115 J Japanese 26, 29–30, 39, 77 n. 5, 88 n. 16, 92, 121–133, 148, 150, 210 Jayaseelan, K. 190 n. 31 K Kayne, R. 36, 47 Kempson, R. 61, 63 n. 29 Kitahara, H. 136
243
Index Koopman, H. 31, 105 n. 22 Korean 88 n. 16, 190 n. 31 L Labelled Deductive Systems (LDS) 63 n. 29 Ladusaw, W. 205 Lappin, S. 21, 22 Lasnik, H. 25 n. 10, 120 Last Resort (the Principle of) 98, 135, 172 Lexical Association (the Principle of) 17 the LF Movement Heuristic 4 the LF Parallelism Assumption 4 Li, Y-H.A. 17, 57, 58, 59, 66, 86 n. 14, 89, 90 n. 18, 117, 141, 143 Lightfoot, D. 26 M Mahajan, A. 82 n. 10, 11, 91–94, 155, 163, 172, 173, 175 n. 18, 193, 207, 208, 214, 220, 224, 225 n. 46 Malayalam 190 n. 31 Manzini, R. 90 n. 18, 106 May, R. 19, 20, 37, 53 n. 28, 54–57 McDaniel, D. 149, 155, 156, 175 n. 19, 184–186 Müller, G. 208 N negation 143, 221–223 negative polarit y items (NPIs) 88, 90 n. 18, 108 n-words 88 n. 16 Nishigauchi, T. 4, 16, 64, 77 n. 5, 79 n. 7, 163, 169, 171 n. 17 Nishiyama, K. 88 n. 16 numeration 149, 150 O O’Neil, J. 96–97 Ouhalla, J. 73 n. 2, 78, 79
P parasitic gaps 18–19, 21, 22 Pesetsky, D. 4, 10, 11, 43–46, 116, pied piping 16, 30 n. 15, 64, 65 n. 30, 72 n. 1, 74 n. 4, 79 n. 7, 81 n. 9, 129, 148, 157, 160 n. 9, 163, 165–171, 178–180, 182 n. 25, 193 Polish 44 n. 23, 109 n. 28, 116 n. 32 positive polarit y items 205 Postal, D. 32 Procrastinate 5, 100, 137, 193 Progovac, L. 90 n. 18, 102 n. 21, 205 n. 38 Q QR 16, 20, 31–33, 37, 52, 53, 55–61, 79 n. 7, 91–93, 135, 136, 142 n. 46 Quechua 23–25, 33, 163 question particles 105, 116–117, 122, 124, 132 R reconstruction 20, 29, 30, 42 Reinhart, T. 114, 120 Relativized Minimalit y 86, 89, 93, 142, 143 resumptive pronouns 21, 22 Rizzi, L. 27, 36, 105 n. 22, 152 n. 4, 158, 164, 221 Romanian 45–48, 70, 99, 106–109 Rudin, C. 99 S Sabel, J. 40 n. 21, 155 Saddy, D. 33, 180 n. 23 Saito, M. 25 n. 10, 29, 120 scope 54–67, 119 Schütze, C. 190 n. 31 scrambling 29, 126, 127, 129, 130 selection 49–51 Serbo-Croat 99, 109 n. 28, 115, 129, 205 n. 38
244 Shi, D. 125, 143 Shortest Move 16, 49, 114, 120–121 Sportiche, D. 31 Srivastav, V. 155, 163, 193 n. 32, 225 n. 46 Sternefeld, W. 208 Stowell, T. 135 n. 44, 142 n. 46, 231 n. 48 Stroik, T. 38 subjacency 4, 10, 15, 19–23, 39, 43, 44, 79, 93, 122, 123, 133 n. 43 Sung, L. 102 n. 21 superiorit y 10, 36, 37, 46–49, 120–121, 208 Szabolcsi, A. 190 T Takahashi, D. 122, 124, 126, 129 Tamrazian, A. 156, 157 n. 6 Tancredi, C. 17 tense 140, 141 tense domain 73, 77, 80–83, 103, 184–186, 189–192, 196, 200–201, 205 triggering 106, 116, 120, 122, 124, 127, 130, 197, 199–200, 202 the Triggering Hypothesis 104, 121, 134, 141
Index Tsai, W-T.D. 14, 15, 26, 27, 39, 199, 228 U Urbina, J.O. de 163, 164 V visibilit y 201–202 W Wahba, W. 73 n. 2, 214 Watanabe, A. 19, 40 n. 21, 77 n. 5, 122–124, 127, 128, 132–133 Watanabe, S. 40 n. 21 Webelhuth, G. 178 Weinberg, A. 26 Wexler, K. 90 n. 18, 106 wh-checking domain 103, 107, 109, 112, 122–123 wh-copy movement 162 n. 12 the Wh-Criterion 165, 166 wh-islands 10, 44–46, 77, 78, 90, 117–120, 125–128, 143 Whitman, J. 88 n. 16 Williams, E. 55 Y Yi, E-Y. 88 n. 16