You are on page 1of 306

Sluicing

Oxford Studies in Theoretical Linguistics


General editors: David Adger, Queen Mary, University of London; Hagit Borer, University of South-
ern California

Advisory editors: Stephen Anderson, Yale University; Daniel Büring, University of California, Los
Angeles; Nomi Erteschik-Shir, Ben-Gurion University; Donka Farkas, University of California,
Santa Cruz; Angelika Kratzer, University of Massachusetts, Amherst; Andrew Nevins, University
College London; Christopher Potts, Stanford University, Amherst; Barry Schein, University of
Southern California; Peter Svenonius, University of Tromsø; Moira Yip, University College London

Recent titles
 InterPhases
Phase-Theoretic Investigations of Linguistic Interfaces
edited by Kleanthes Grohmann
 Negation in Gapping
by Sophie Repp
 A Derivational Syntax for Information Structure
by Luis López
 Quantification, Definiteness, and Nominalization
edited by Anastasia Giannakidou and Monika Rathert
 The Syntax of Sentential Stress
by Arsalan Kahnemuyipour
 Tense, Aspect, and Indexicality
by James Higginbotham
 Lexical Semantics, Syntax, and Event Structure
edited by Malka Rappaport Hovav, Edit Doron, and Ivy Sichel
 About the Speaker Towards a Syntax of Indexicality
by Alessandra Giorgi
 The Sound Patterns of Syntax
edited by Nomi Erteschik-Shir and Lisa Rochman
 The Complementizer Phase
edited by E. Phoevos Panagiotidis
 Interfaces in Linguistics
New Research Perspectives
edited by Raffaella Folli and Christiane Ulbrich
 Negative Indefinites
by Doris Penka
 Events, Phrases, and Questions
by Robert Truswell
 Dissolving Binding Theory
by Johan Rooryck and Guido Vanden Wyngaerd
 The Logic of Pronominal Resumption
by Ash Asudeh
 Modals and Conditionals
by Angelika Kratzer
 The Theta System
Argument Structure at the Interface
edited by Martin Everaert, Marijana Marelj, and Tal Siloni
 Sluicing
Cross-Linguistic Perspectives
edited by Jason Merchant and Andrew Simpson
 Telicity, Change, and State
A Cross-Categorial View of Event Structure
edited by Violeta Demonte and Louise McNally
 Ways of Structure Building
edited by Myriam Uribe-Etxebarria and Vidal Valmala
 The Morphology and Phonology of Exponence
edited by Jochen Trommer
For a complete list of titles published and in preparation for the series, see pp –.
Sluicing:
Cross-Linguistic
Perspectives

Edited by
JAS ON M E RC HA N T A N D A N DR EW SI M P S ON

1
3
Great Clarendon Street, Oxford ox2 6dp
Oxford University Press is a department of the University of Oxford.
It furthers the University’s objective of excellence in research, scholarship,
and education by publishing worldwide in
Oxford New York
Auckland Cape Town Dar es Salaam Hong Kong Karachi
Kuala Lumpur Madrid Melbourne Mexico City Nairobi
New Delhi Shanghai Taipei Toronto
With offices in
Argentina Austria Brazil Chile Czech Republic France Greece
Guatemala Hungary Italy Japan Poland Portugal Singapore
South Korea Switzerland Thailand Turkey Ukraine Vietnam
Oxford is a registered trade mark of Oxford University Press
in the UK and in certain other countries
Published in the United States
by Oxford University Press Inc., New York
© editorial matter and organization Jason Merchant and Andrew Simpson 
© the chapters their several authors 
The moral rights of the authors have been asserted
Database right Oxford University Press (maker)
First published 
All rights reserved. No part of this publication may be reproduced,
stored in a retrieval system, or transmitted, in any form or by any means,
without the prior permission in writing of Oxford University Press,
or as expressly permitted by law, or under terms agreed with the appropriate
reprographics rights organization. Enquiries concerning reproduction
outside the scope of the above should be sent to the Rights Department,
Oxford University Press, at the address above
You must not circulate this book in any other binding or cover
and you must impose the same condition on any acquirer
British Library Cataloguing in Publication Data
Data available
Library of Congress Cataloging in Publication Data
Data available
Typeset by SPI Publisher Services, Pondicherry, India
Printed in Great Britain
on acid-free paper by
MPG Books Group, Bodmin and King’s Lynn

ISBN –––– (Hbk)


–––– (Pbk)

         
Contents
General Preface vi
The Contributors vii
List of Abbreviations xi

. Introduction 
Jason Merchant and Andrew Simpson
. Guess who? 
John Robert Ross
. How do you sluice when there is more than one CP? 
Jeroen van Craenenbroeck
. Two cases of violation repair under sluicing 
Sandra Stjepanović
. How many kinds of sluicing, and why? Single and multiple sluicing in
Romanian, English, and Japanese 
Frederick Hoyt and Alexandra Teodorescu
. Case morphology and island repair 
Masanori Nakamura
. Island-sensitivity in Japanese sluicing and some implications 
Teruhiko Fukaya
. Sluicing without wh-movement in Malagasy 
Ileana Paul and Eric Potsdam
. Sluicing in Indo-Aryan: An investigation of Bangla and Hindi 
Tanmoy Bhattacharya and Andrew Simpson
. Sluicing in Mandarin Chinese: An instance of pseudo-sluicing 
Perng Wang Adams and Satoshi Tomioka
. Sluicing in Turkish 
Atakan İnce

References 
Author Index 
Subject Index 
General Preface
The theoretical focus of this series is on the interfaces between subcomponents of the
human grammatical system and the closely related area of the interfaces between the
different subdisciplines of linguistics. The notion of “interface” has become central
in grammatical theory (for instance, in Chomsky’s recent Minimalist Program) and
in linguistic practice: Work on the interfaces between syntax and semantics, syntax
and morphology, phonology and phonetics, etc. has led to a deeper understanding of
particular linguistic phenomena, and of the architecture of the linguistic component
of the mind/brain.
The series covers interfaces between core components of grammar, includ-
ing syntax/morphology, syntax/semantics, syntax/phonology, syntax/pragmatics,
morphology/phonology, phonology/phonetics, phonetics/speech processing, seman-
tics/pragmatics, and intonation/discourse structure, as well as issues in the way that
the systems of grammar involving these interface areas are acquired and deployed in
use (including language acquisition, language dysfunction, and language processing).
It demonstrates, we hope, that proper understanding of particular linguistic phenom-
ena, languages, language groups, and inter-language variations all require reference to
interfaces.
The series is open to work by linguists of all theoretical persuasions and schools of
thought. A main requirement is that authors should write so as to be understood by
colleagues in related subfields of linguistics and by scholars in cognate disciplines.
David Adger
Hagit Borer
The Contributors
Perng Wang Adams received her first master’s degree in the English Lan-
guage/Linguistics Program at the University of Arizona in  and her second mas-
ter’s degree in Linguistics at the University of Delaware in . She has presented
her work at conferences such as the UPenn Linguistics Colloquium and the Western
Conference on Linguistics. Her primary research interests are comparative syntax,
language acquisition, and linguistics field methods.
Tanmoy Bhattacharya is Associate Professor in Linguistics at the University of Delhi.
He has Ph.D.s in linguistics from the University of Hyderabad, India and University
College London, UK, the latter as a Commonwealth Scholar. His interests in syntax
are the syntax of nominal structures and wh-constructions in South Asian languages,
and the syntax of Indian Sign Language. He has published articles co-authored with
Andrew Simpson in journals such as Linguistic Inquiry and Lingua, and has published
in a range of edited volumes, including Linguistic Theory and South Asian Languages,
Argument Structure (John Benjamins) and Yearbook of South Asian Languages and
Linguistics (Mouton).
Jeroen van Craenenbroeck is Professor of Dutch Linguistics at the Center for
Research in Syntax, Semantics, and Phonology (CRISSP) in the Hogeschool-
Universiteit Brussel. He received his Ph.D. from Leiden University in the Netherlands
in . His main research interests are ellipsis, pronominal doubling, and the syntax
of the left periphery. In  he published an Oxford University Press monograph
entitled ‘The syntax of ellipsis. Evidence from Dutch dialects’. His journal publications
include articles in Linguistic Inquiry, Syntax, and Lingua, and he is general editor of
the Linguistic Variation Yearbook (John Benjamins).
Teruhiko Fukaya is Professor at Gunma Prefectural Women’s University in Japan. He
received his Ph.D. in linguistics from the University of Southern California in .
His main research interests are in ellipsis phenomena, such as sluicing, stripping, and
fragment answers, as well as in comparative syntax of Japanese and English in general.
His publications include papers on sloppy-identity readings and island-sensitivity in
sluicing and stripping in the proceedings of the West Coast Conference on Formal
Linguistics.
Frederick Hoyt studied at the University of New Hampshire, Cornell University,
and the University of Texas at Austin, where he received his Ph.D. in linguistics in
. His research interests include the syntax and semantics of Arabic, negation and
negative concord, and the syntax and semantics of information structure. He received
viii The Contributors

a Fulbright-Hayes fellowship for field research in Jordan and has presented at many
conferences on both general linguistics and Arabic linguistics.
Atakan İnce is a language instructor and materials developer at MultiLingual Solu-
tions in Maryland, USA. He received his Ph.D. in linguistics from the University of
Maryland, College Park in . His main research interests are ellipsis, Turkish syn-
tax, and agreement. He has published in conference proceedings volumes including
the Proceedings of the Workshop on Altaic Formal Linguistics and the Proceedings of
the North East Linguistics Society, as well as working papers such as the University of
Maryland Working Papers in Linguistics.
Jason Merchant is Professor of Linguistics at the University of Chicago. He has written
extensively on ellipsis, including on sluicing, swiping, fragment answers, verb phrase
ellipsis, antecedent-contained ellipsis, comparative ellipsis, and nominal ellipsis. His
other interests are in case, split ergativity, locality, islands, agreement, and topics in
the syntax–semantics interface. His primary language areas are in Germanic, Greek,
and Romance. He studied at Yale, Tübingen, Utrecht, and the University of California
at Santa Cruz, where he received his Ph.D. in .
Masanori Nakamura is Professor at Senshu University, Japan. He received his Ph.D.
in linguistics from McGill University, Canada in . His main research interests
are in comparative syntax, locality principles, and grammatical functions. He has
published in journals such as Linguistic Inquiry, The Linguistic Review, and Lingua, as
well as in refereed conference proceedings volumes including Is the Best Good Enough?
Optimality and Competition in Syntax (MIT Press and MITWPL), The Minimalist
Parameter (John Benjamins), Dimensions of Movement: From Features to Remnants
(John Benjamins), and InterPhases: Phase-Theoretic Investigations of Linguistic Inter-
faces (Oxford University Press).
Ileana Paul is currently Associate Professor at the University of Western Ontario
(Canada), and she also holds the Canada Research Chair in Linguistics. She received
her Ph.D. from McGill University, Canada in . Her research focuses on the syntax
of Malagasy, the language spoken in Madagascar. She has published several book
chapters and journal articles (in journals such as Syntax, Lingua, and Oceanic Linguis-
tics) on a range of topics, including existentials, nominal structure and interpretation,
and clefts. She recently co-edited Determiners: Universals and Variation, published by
John Benjamins.
Eric Potsdam is Associate Professor of Linguistics at the University of Florida. He
received his Ph.D. in Linguistics from the University of California, Santa Cruz in .
He works primarily on Austronesian languages, especially Malagasy, and his research
interests focus on syntactic issues related to clause structure, raising and control,
The Contributors ix

and wh-question strategies cross-linguistically. He has published in such journals as


Language, Linguistic Inquiry, and Natural Language and Linguistic Theory.
John Robert “Haj” Ross is Professor in the Department of Linguistics and Technical
Communication at the University of North Texas. He studied at Yale, Bonn, Berlin,
UPenn, and MIT, where he received his Ph.D. in , having completed what has
proven to be one of the most influential dissertations in linguistics, ‘Constraints on
variables in syntax’, in which he documented his discovery of island effects. He was on
the faculty at MIT for almost two decades, and has made fundamental contributions
in syntax, semantics, pragmatics, morphology, phonology, and phonetic symbolism.
He coined the term “sluicing”, and described and analyzed the construction for the
first time in his seminal paper, ‘Guess Who?’, published in .
Andrew Simpson is Professor of Linguistics and East Asian Languages and Cultures
at the University of Southern California. His research is focused on the compar-
ative syntax of East, Southeast, and South Asian languages, in particular Chinese,
Japanese, Korean, Thai, Vietnamese, Burmese, Bangla, and Hindi. He is the author
of Wh-Movement and the Theory of Feature Checking (John Benjamins), the editor of
Language and National Identity in Asia and Language and National Identity in Africa
(Oxford University Press), and joint general editor of the Journal of East Asian Linguis-
tics. He has published articles in Linguistic Inquiry, Language, Lingua, the Journal of
East Asian Linguistics, Studies in Language, Language and Linguistics, and the Journal
of the South East Asian Linguistics Society.
Sandra Stjepanović is Associate Professor at West Virginia University, USA. She
received her Ph.D. in Linguistics from the University of Connecticut, USA in .
Her main research interests are in comparative syntax, free word order, and ellipsis
phenomena. She has published in journals such as Linguistic Inquiry, the Journal of
Slavic Linguistics, and The Linguistic Variation Yearbook, as well as in refereed volumes
such as The Copy Theory of Movement on the PF Side (John Benjamins) and Multiple
Wh-Fronting (John Benjamins).
Alexandra Teodorescu received her Ph.D from The University of Texas at Austin and
is currently a Visiting Scholar at New York University. Her primary research interests
include the syntax and semantics of Romanian, noun phrase modification, posses-
sives, negative polarity/negative concord, sluicing, agreement, and the structure of the
verbal domain. She has presented her work at various conferences including the West
Coast Conference on Formal Linguistics, Linguistic Symposium on Romance Languages,
UPenn Linguistics Colloquium, and The Georgetown University Round Table. She has
published a book chapter and several papers in refereed conference proceedings, and
co-edited a CSLI volume.
x The Contributors

Satoshi Tomioka is Associate Professor of Linguistics and Cognitive Science at the


University of Delaware. He received his Ph.D. from the University of Massachusetts
at Amherst in . His main research interests are semantics, pragmatics, syntax,
and their interfaces, and his recent research topics include wh-interrogatives, ellipsis,
plurality and distributivity, contrastiveness, and scalar implicatures. He has published
in such journals as Natural Language Semantics, the Journal of East Asian Linguistics,
the Journal of Pragmatics, and Lingua. His articles have also appeared as book chap-
ters in Topics in Ellipsis (Cambridge University Press), Information Structure (Oxford
University Press), and Interfaces (John Benjamins).
List of Abbreviations

, ,  st, nd, rd person


abl ablative case
acc accusative case
Adj adjective
AP adjective phrase
asp aspect
at agent topic marker
Aux, aux auxiliary
ba object marker (Chinese)
C complementizer
CED Condition on Extraction Domains
cl classifier
CLD contrastive left dislocation
CLLD clitic left dislocation
com commitative case
COMP, comp complementizer
cond conditional
COORD coordinate phrase
cop copula
CP complementizer phrase
CR Constituent Raising
D determiner
dat dative case
de marker of nominal modification (Chinese)
dem demonstrative
Det determiner
DP determiner phrase
ECP empty category principle
erg ergative case
f feminine
Foc, foc focus
FocP, FP focus phrase
fut future tense
gen genitive case
xii List of Abbreviations

GQ genitive of quantification
IICF Inverse Inherent Case Filter
Imp imperative
impers impersonal
IP inflection phrase
IPEC IP-ellipsis condition
LBC Left Branch Condition
LF logical form
loc locative case
m masculine
Neg, neg negation
nmlz nominalizer
nom nominative case
NP noun phrase
OBL obligatory transformation
obl oblique case
Op operator
OPT optional transformation
P preposition
pass passive
PF phonological form
pl plural
pol politeness marker
poss possessive
PP preposition phrase
prep preposition
Pres, pres present tense
prog progressive aspect
pron pronoun
Prt, prt particle
pst past tense
rc relative clause
red reduplication
refl reflexive
rel relative
Q question
QP question phrase
S sentence
S sentence-bar
sg, sing singular
List of Abbreviations xiii

SLC sluicing-like construction


SOV subject–object–verb typology
SVO subject–verb–object typology
SpecXP specifier of XP (etc.)
subj subject
T, Tns tense
TP tense phrase
top topic
TP tense phrase
tt theme topic marker
VOS verb–object–subject typology
VSO verb–subject–object typology
vP “little-v” phrase
V verb
VP verb phrase
This page intentionally left blank

Introduction
JAS ON M E RC HA N T A N D A N DR EW SI M P S ON

“Sluicing” refers to the elliptical phenomenon illustrated in (a), which in most con-
texts can be taken as equivalent to (b):
() a. You called someone, but I don’t know who.
b. You called someone, but I don’t know who you called.
Haj Ross, in his seminal  CLS paper ‘Guess Who?’ (Ross (), reprinted in this
volume for the first time), identified the alternation in (a) and (b) and proposed
to analyze (a) as the result of a transformation he named sluicing applying to (b).
Sluicing, on his formulation, took as part of its input an interrogative S into whose
COMP position a wh-phrase had moved, and deleted the highest S node out of which
wh-movement had occurred. In the decades since Ross first described these facts, the
term sluicing has been generalized to apply to any apparently elliptical construction
that has the sense of a question but has the form merely of an interrogative phrase
(typically a wh-phrase used in otherwise uncontroversial constituent questions). 
The phenomenon of sluicing has attracted sustained interest over the years, as
it appears to involve two well-studied linguistic phenomena: unbounded syntactic
dependencies (wh-movement structures) and ellipsis.
The first phenomenon represents one of the major discoveries in linguistics of
the th century: traditional grammars were silent on the fact that some syntactic
dependencies (the relation between an interrogative pronoun and its gap, and mutatis
mutandis for relative pronouns, topics, comparatives, etc.) are apparently unbounded.
Furthermore, such dependencies are subject to constraints on their position and kind:
certain structures cannot host a gap associated with an element outside of them—
these structures, following Ross (), who discovered and named them, are known
as “islands”. Taken together, these empirical discoveries, rooted in formal theories of
 It has also become common to use the verb “sluice” as a hyponym of “elide”, as in “The TP node has
been sluiced” (equivalent to “The TP node has been elided”), “the sluiced material” (equivalent to “the elided
material”), etc. This usage derives directly from the name Ross gave to his deletion transformation, though
modern authors may not mean to endorse that particular transformational approach to the phenomenon.
2 Jason Merchant & Andrew Simpson

syntax, rank among the most interesting scientific discoveries of the past century,
and are certainly some of the most important in linguistics (phrase structure and
the phoneme being comparable). As such, these structures have garnered enormous
attention from theorists, who have studied their properties as a window into the
workings of the linguistic system.
The second phenomenon concerns another central issue of debate and inquiry
in theoretical linguistics: the putative existence and nature of syntactic objects that
do not correspond directly to elements in the speech (or gestural) signal. Ellipsis
provides a particularly important and fertile empirical domain for addressing these
issues, because it involves a mapping between an arbitrarily complex meaning and the
complete absence of a phonological signal.
Traditional generative approaches to ellipsis have assumed that this mapping is
mediated by unpronounced syntactic representations, for primarily two reasons. The
first reason arises from a guiding theoretical intuition that in general, identity of
meaning indicates identity of form; the semantic correspondence between an elided
constituent and some antecedent is therefore taken to indicate the presence of a
syntactic representation that is not pronounced but forms the input to interpretation.
The second reason is specific to ellipsis: A large proportion, perhaps a majority, of
analyses of ellipsis developed over the past forty years have assumed that an elided
constituent is identical to some syntactic antecedent, that is, that ellipsis is licensed by
a syntactic identity condition.
Recent work has challenged both these underlying arguments for unpronounced
syntactic structure in ellipsis. On the one hand, a diversity of approaches to the
syntax–semantics interface have been developed that reject the hypothesis that iden-
tity of meaning entails identity of form, instead deriving identity of meaning from
features of the interpretive system. On the other hand, a great deal of evidence
has accumulated which has been taken to indicate that the identity condition in
ellipsis should be stated over meanings, not syntactic representations. We think it
is thus now possible and indeed necessary to re-evaluate the arguments for and
against representationalist accounts of ellipsis with a more nuanced eye, and to
directly address the question of whether the crucial facts can be explained just as well
within purely interpretationalist approaches, which eschew syntactic representations
in ellipsis.
This book brings together researchers with an interest and expertise in this domain
to provide a state-of-the-art re-evaluation of these fundamental questions, concen-
trating on sluicing. Specifically, the papers in this volume focus on the empirical and
analytical arguments for positing or not positing unpronounced syntactic structures
in sluicing, in an effort both to sharpen our understanding of the mechanisms that
handle ellipsis and to further develop our understanding of the syntax–semantics
interface. A broader purpose of the book is to bring various conflicting claims in the
literature about the nature of the data under close scrutiny, with the goal of establish-
Introduction 3

ing agreement about what facts a theory of sluicing needs to explain, independent of
framework-specific assumptions.
Though Ross’s original paper contained a few crucial examples from German, most
of the claims in the literature of the following thirty years about sluicing, and about
syntactic and semantic architectures based on them, have rested on a close examina-
tion of facts from just one language, English. This volume aims to subject analyses
of sluicing to the light of an expanded empirical domain, and so to test the mettle of
these analyses against more of the diversity of the world’s languages, particularly given
the much better understanding currently available of the differing strategies languages
make use of in forming questions. To this end, we have gathered papers that examine
sluicing and sluicing-like phenomena from a cross-linguistic perspective. Such work
is especially valuable as the kind of close analytical examination that is necessary to
make sound theoretical claims. There is at present no prospect for doing more surface-
oriented typological work on sluicing, as sluicing is not a construction that is typically
mentioned in grammars or in the kinds of surveys that syntactic typologists used to
working with genetically diverse languages rely on.
It is not the purpose of this introduction to provide a general overview of ellipsis
(for recent summaries, see Lappin (), Johnson (), Winkler (), Reich (to
appear), Merchant (to appear a, b), van Craenenbroeck and Merchant (to appear)) or
even of sluicing in particular (see Merchant () for one such). Our goal is more
modest: to set the stage for the papers. These papers represent the state of the art
in investigating whether sluicing or sluicing-like structures exist in a wider variety
of languages, and if so, what properties these constructions have, and how these
properties can inform our understanding of unbounded dependencies and of ellipsis.
Following Ross’s original () paper, reprinted as Chapter , the volume contains
nine chapters on a broad variety of languages. Each chapter either provides a broad
characterization of sluicing/sluicing-like constructions in the languages being con-
sidered, or focuses in on specific aspects of sluicing in those languages as a means to
expand our general theoretical understanding of sluicing and broader morphosyn-
tactic phenomena, which may be manifested in sluicing constructions in particularly
revealing ways.
In Chapter , ‘How do you sluice when there is more than one CP?’, Jeroen van
Craenenbroeck examines the interaction of sluicing with the split-CP hypothesis, the
idea that there may be more than one CP projection in the C-domain (Rizzi ()
and others). Considering the possibility that wh-movement might actually target
either a higher or a lower SpecCP position, as suggested by comparisons of languages
such as Hungarian and English, and that the clausal constituent deleted in sluicing is
the complement of the C-head whose specifier wh-phrases raise to, a clear range of
variation is predicted to be theoretically possible in sluicing constructions. Van Crae-
nenbroeck observes that if wh-movement targets a lower position in the C-domain,
as assumed in Hungarian, the clausal constituent which is deleted in sluicing may be
4 Jason Merchant & Andrew Simpson

TP, whereas if wh-phrases raise up to a higher SpecCP position, it may in principal


be either TP or some lower CP-projection that is deleted in sluicing constructions.
With regard to Dutch, it has been argued that there are two complementizer positions
in the left periphery, and with data from Strijen Dutch, van Craenenbroeck shows
that wh-movement can indeed target a specifier position between these two head
positions. The analysis developed by van Craenenbroeck is that the higher CP-layer
CP (in Dutch, and by hypothesis across other languages) is responsible for clausal
typing, and the lower one, CP , is a position where operator features are checked.
A crucial and important claim made by van Craenenbroeck is that there is a sig-
nificant difference between simple and complex wh-phrases in their base position
and movement within the C-domain. Complex wh-phrases (e.g. equivalents to which
man/book) are argued to be base-generated in the higher SpecCP , whereas simple
wh-phrases (e.g. equivalents to who, what) are taken to move from an IP-internal theta
position through SpecCP to SpecCP . Van Craenenbroeck provides support for this
proposal with a range of empirical patterns involving two complementizer patterns
in Strijen Dutch, wh-copying constructions in German, and Superiority phenomena.
The chapter subsequently shows how the simple/complex wh-phrase distinction and
the split CP hypothesis can be used to account for various interesting and complex
patterns found in sluicing constructions in Germanic.
The first of these is the occurrence of “swiping”, in which the regular sequencing
of preposition and wh-phrase is inverted in sluices (e.g. Peter went to the movies, but
I don’t know who with). Van Craenenbroeck shows that swiping can be analyzed as
resulting from the stranding of a preposition in SpecCP  as the wh-phrase transits
this position en route to SpecCP . The observed restriction of swiping to simple wh-
phrases is a natural consequence of this analysis, as complex wh-phrases never raise
through the lower SpecCP position (hence the ungrammaticality of ∗ Lois was talking,
but I don’t know which person to). The approach is also able to explain why swiping is
not possible in the absence of sluicing (e.g. ∗ Who to was Lois talking?). It is suggested
that deletion of the TP constituent in sluicing repairs a non-uniform chain consisting
of both PP and DP links.
The second, special pattern of sluicing investigated further in the chapter is “spad-
ing” (van Craenenbroeck ), elliptical forms in which a sluiced wh-phrase is imme-
diately followed by a demonstrative, as is possible in Wambeek Dutch:
() A: Jef eid iemand gezien. B: Wou da?
Jeff has someone seen who that
‘A: Jeff saw someone. B: Who?’
Van Craenenbroeck presents convincing case-related arguments that the structure
underlying spading forms of sluicing is in fact a cleft construction with a wh-pivot. He
also suggests that the demonstrative pronoun undergoes focus movement to SpecCP
prior to deletion of the TP complement of C . Such an analysis makes the prediction
Introduction 5

that spading should not be possible with complex wh-phrases, and the prediction
is indeed borne out. Finally, the chapter looks at instances in which both spading
and swiping co-occur in the same sluicing construction, as is found to be possible in
Frisian. The complex linear output of [simple wh-phrase > demonstrative pronoun >
stranded preposition] is shown to be fully as expected given the analysis developed in
the chapter, which provides a wealth of empirical patterns and insights into sluicing
from varieties of north Germanic.
Moving from the Germanic branch of Indo-European to a member of the Slavic
group, Chapter , ‘Two cases of violation repair under sluicing’, by Sandra Stjepanović,
considers two instances in Serbo-Croatian in which structures that would be ill-
formed in the absence of sluicing are acceptable when the ellipsis effect of sluicing
occurs. The first of these concerns the optional omission of a preposition when a wh-
phrase that is a PP undergoes movement in sluicing constructions, shown in ():
() Petar je glasao protiv nečega, ali ne znam (protiv)
Petar is voted against something-gen but not I-know against
čega.
what-gen
‘Petar voted against something, but I don’t know what.’
Such patterns may appear to suggest that Serbo-Croatian is a language optionally
permitting preposition-stranding, and that the occurrence of the preposition-less
form čega in () results from P-stranding and subsequent TP-deletion. However,
P-stranding is not possible in regular Serbo-Croatian wh-questions. Stjepanović
observes that if such a mechanism were to be exceptionally licensed in sluicing
contexts, the acceptability of forms such as () would pose a potential threat to
Merchant’s () otherwise robust P-stranding generalization—that languages only
permit P-stranding in sluicing if they also allow the stranding of such elements
in other, non-elliptical, contexts of wh-movement. Using evidence from the co-
ordination of wh PP-remnants in instances of multiple sluicing, Stjepanović shows
that a P-stranding analysis of preposition omission in sluicing cannot in fact be main-
tained, and that a different post-syntactic mechanism of P-drop fed by sluicing must
be assumed to license the phenomenon of preposition omission. Such a PF process
of deletion is noted to occur only where the content of a preposition is recoverable
from the antecedent clause in sluices, and may not occur where the preposition heads
a contrastively focused phrase—conditions which are elsewhere required by opera-
tions of de-accenting and deletion. Merchant’s P-stranding generalization is therefore
argued not to be violated by the patterns found in Serbo-Croatian.
In a second instance of “violation repair” licensed as a result of sluicing in
Serbo-Croatian, Stjepanović discusses the patterning of the genitive of quantification
(GQ), in which higher numerals assign a genitive case to their NP complements. It
is well-documented that such numerically-quantified phrases cannot occur as the
6 Jason Merchant & Andrew Simpson

complements of verbs which assign inherent case to their objects. However, intrigu-
ingly this restriction appears to be lifted in sluicing constructions, and a wh-phrase
that is a numeral phrase with GQ can occur as the object remnant of an elided TP
whose verb would normally assign inherent case. Sluicing thus again appears to be
able to repair a violation, in this instance relating to case-marking. Stjepanović argues
that the patterning found in sluicing indicates that inherent case assignment imposes
a special PF-related visibility requirement that this case be overtly realized wherever
appropriate conditions occur. Commonly, a morphological deficiency in numeral
phrases with GQ prohibits the latter from overtly manifesting inherent case and
therefore blocks their occurrence with inherent case assigning verbs. However, where
such verbs are elided in contexts of sluicing, it is suggested that their undischarged
case features are also deleted, permitting the exceptional occurrence of complement
numeral phrases with GQ as wh-remnants. The interaction of sluicing with GQ phe-
nomena in Serbo-Croatian consequently leads to broader conclusions about licensing
conditions associated with inherent case and the derivational point at which such
licensing is imposed and checked (PF).
Chapter , by Frederick Hoyt and Alexandra Teodorescu, compares sluicing con-
structions in the mixed grouping of Romanian, English, and Japanese, and asks:
“How many kinds of sluicing (are there), and why?” It is noted that the three lan-
guages being investigated regularly form their wh-questions in quite different ways.
Romanian is a Romance language that has developed multiple overt wh-movement
as a result of long-term influence from neighboring Slavic languages; Japanese is a
prototypical wh-in-situ language, with no forced overt wh-movement; and English
shows both overt wh-movement and wh-in-situ in multiple wh-questions, with a
single wh-phrase undergoing overt movement to SpecCP, and other, secondary wh-
phrases remaining in situ. Hoyt and Teodorescu point out that, because the properties
of sluicing constructions are often thought to derive directly from the kinds of wh-
constructions a language has available, it might be expected that Romanian, Japanese,
and English would all show markedly different patterns of sluicing. The chapter care-
fully shows how there are indeed both interesting similarities and differences in the
sluicing constructions found in the three languages, and that two fundamental forms
of sluicing occur in the set of three languages: (i) ellipsis of an IP/TP constituent
following overt wh-movement to the left periphery, and (ii) ellipsis of a CP constituent
in underlying cleft constructions. English is commonly taken to exhibit the former
pattern, and Japanese has often been assumed to have sluices of the second type. One
of the primary questions that Hoyt and Teodorescu set out to answer is what type of
sluicing Romanian may be assumed to follow.
In certain ways, Romanian and Japanese are seen to pattern in a very similar
way with regard to sluicing, and differently from English. Both languages permit
(a) multiple wh-remnants, (b) the potential occurrence of aggressively non-d-linked
wh-remnants equivalent to what the hell, why the devil, (c) overt complementizers with
Introduction 7

wh-remnants, and (d) non-wh-remnants of a variety of types, including referential,


quantificational, and free-choice NPs. Despite such similarities, the chapter argues
that sluicing in Romanian should be analyzed like English sluicing as IP-ellipsis, and
that the similarities between Romanian and Japanese can be shown to follow from
independent syntactic properties of the two languages, for example, multiple sluicing
resulting from the occurrence of multiple wh-movement in Romanian, and multiple
(wh) foci pivots in clefts in Japanese. The chapter also shows that Romanian multiple
wh-sluices cannot be analyzed as instances of gapping. Finally, the chapter provides
a partial typology of sluicing constructions based on two parameters of variation,
and highlights its central conclusion that the term “sluicing” does not refer to a
single syntactic structure, and instead refers to a “correlation between certain ellipsis
configurations, the forms of which vary in different languages”.
Chapters  and  of the volume also have a focus on Japanese and certain important
observed differences (and similarities) in the patterning of sluicing in Japanese and
English. Chapter , ‘Case morphology and island repair’, by Masanori Nakamura,
investigates aspects of the derivational licensing of case in Japanese and English,
and instances of sluicing-dependent island “repair”. The puzzle Nakamura sets out
to address is why the sluicing of case-marked NPs in Japanese appears to be island-
sensitive and not subject to repair by ellipsis, whereas parallel locality violations in
English sluicing into islands can be repaired by TP deletion. Nakamura shows that
the full patterning of sluicing involving island constituents in Japanese is actually quite
complex, and that the acceptability of a wh-remnant relating to a position within an
island in a sluiced clause crucially depends on the type of element that undergoes
movement as the wh-remnant. If the latter is a case-marked NP, the result is ungram-
maticality, whereas if the wh-remnant is an argument PP, sluicing structures contain-
ing islands are in fact acceptable. Additionally, it is shown that not all PP wh-remnants
are immune to island configurations, and adjunct PP remnants cannot be related to
extraction-sites in underlying island constituents. Nakamura takes this global pat-
terning as evidence that (a) movement does occur from fully projected underlying
syntactic structures in all instances of sluicing, and (b) where such movement involves
extraction of a wh-phrase from an island constituent, this automatically causes a
locality violation, which, syntactically, can be repaired with argument wh-phrases by
sluicing deletion of the island in Japanese, just as in English. Nakamura then suggests
that the actual ungrammaticality of case-marked wh-remnants in sluices involving
islands in Japanese is not due to any inability of ellipsis to repair syntactic violations
in sluices, which might be parametrized differently in Japanese and English, but due
to differences in the way case is licensed in the two languages.
Building on proposals in Harada () and Fukui and Sakai () that case-
marking in Japanese is a PF phenomenon but in English is a syntax-internal process,
Nakamura suggests that case on Japanese NPs is licensed by a post-Spell-Out mecha-
nism of Morphology. In order for this case-licensing to be successful, NPs must retain
8 Jason Merchant & Andrew Simpson

appropriate connections with their base positions, as it is the tail of an argument


chain that is assumed to be case-marked, case features being transferred from the
tail to each member of the chain and realized phonetically on the head of the chain.
Where island constituents interrupt the links of an argument chain at any point, it
is suggested that this will block the critical process of case transfer, and the head
of the chain will fail to be case-licensed at PF. It is therefore the island-sensitivity
of processes of Morphology that is argued to cause the ungrammaticality of certain
instances of sluicing with case-marked NPs in Japanese, and while the deletion of a
clause containing an island constituent will repair syntactic movement violations in
sluicing constructions, no such repair effect is available for island violations occurring
with operations of Morphology. With such suggestions, Nakamura is able to maintain
that cross-linguistically there is a fully uniform process of syntactic island repair in
sluicing constructions, despite the apparent exceptions reported in Japanese, and does
this in a way which attributes the “problematic” Japanese patterns to variations in
morphology and parametrized differences in the ways that case may be licensed in
different languages. Nakamura’s chapter, like Stjepanović’s, thus shows how evidence
from sluicing can shed interesting light on the possible PF-related properties that
case may have in certain instances, and how the derivational licensing of case may
be subject to variation both across case type (structural vs inherent case) and across
languages (PF-licensed in Japanese, syntactically-licensed in English).
In Chapter , ‘Island-sensitivity in Japanese sluicing and some implications’,
Teruhiko Fukaya investigates the issue of island effects in Japanese and English sluic-
ing from a novel angle. Considering both instances of case-marked sluicing, where
the wh-remnant bears a case-marker or postposition, and non-case-marked sluicing,
where the wh-remnant appears without such elements in Japanese, Fukaya examines
the potential presence of quantificational, dependent scope readings referred to as
“covariant” and “multiple-event” readings. Presenting a range of carefully controlled
data in Japanese, Fukaya argues that such readings are available in both case-marked
and non-case-marked sluicing in examples where no island configuration occurs
in the antecedent clause, but are systematically absent in case-marked sluicing—
though not in non-case-marked sluicing—where relative clause and adjunct islands
are present and contain a “non-indefinite” correlate to the wh-remnant. Fukaya takes
such patterns as a rather different kind of confirmation that case-marked sluicing
in Japanese is island-sensitive, whereas non-case-marked sluicing is not, as indeed
initially asserted in Fukaya and Hoji () on the grounds of other kinds of data.
Fukaya then shows how the copy theory of ellipsis resolution proposed in Fukaya and
Hoji (), together with a version of Merchant’s () proposal for local movement
in instances of sluicing in propositional island contexts, can be used to capture the
patterns found in Japanese case-marked sluicing.
Turning to non-case-marked sluicing in Japanese, Fukaya notes that the covari-
ant and multiple-event readings reported to be unavailable with case-marked wh-
Introduction 9

remnants in the presence of relative clause and adjunct islands are legitimate interpre-
tations in the same configurations when no case-marker occurs on the wh-remnant.
Fukaya proposes modeling this patterning by means of a quite different copula and
pro structure for non-case-marked sluicing, similar in underlying form to English
sentences such as I know John talked to someone, but I can’t remember who it was, in
which it is interpreted as referring to the property denoted in the first conjunct of the
antecedent-sluiced clause pair.
Fukaya then reconsiders English, and suggests that just as Japanese may be assumed
to permit two different structures for sluicing, linked to the presence/absence of
case-marking of the wh-remnant, so too English may allow for two different sluicing
structures, dependent on preposition pied-piping with wh-remnants. Fukaya suggests
that if contrast sluicing (Merchant (): She met RINGO, but I don’t know who else)
is examined for the availability of the kinds of covariant and multiple-event readings
noted in Japanese, and the property of preposition pied-piping is manipulated in
island contexts, the same basic conclusions reached for Japanese can actually also be
reached for English, namely that one structural form of sluicing in English is island-
sensitive, while a second form is not. Fukaya shows how this conclusion about English
poses a challenge to recent assumptions that relative clauses are PF representational
islands that can be ameliorated by PF deletion (Fox and Lasnik ).
The topic of Chapter  is ‘Sluicing without wh-movement in Malagasy’, by Ileana
Paul and Eric Potsdam. The Austronesian language Malagasy, spoken in Madagas-
car, often gives rise to interesting linguistic questions in connection with its cross-
linguistically rare but robust VOS word order. With regard to sluicing, Paul and
Potsdam note that Malagasy has a construction that appears to show all the hallmarks
of sluicing in other languages, yet Malagasy is a language that can be demonstrated
not to have wh-movement, making use of either an in situ or a pseudo-cleft strategy to
form wh-questions. In investigating the Malagasy sluicing construction, the authors
first consider the possibility that wh-movement might exceptionally occur in Mala-
gasy just in sluicing constructions and be licensed by deletion of the TP constituent
containing the trace of wh-movement, a “Deletion Repair” analysis. However, the
authors show that there are several serious problems for such an analysis in Malagasy,
and so reject it in favor of a second hypothesis, that what appears to be regular
sluicing in Malagasy in fact makes use of a pseudo-cleft construction to generate forms
that are similar in surface appearance to sluices created by wh-movement in other
languages.
The specific proposal made by Paul and Potsdam for “sluicing-like constructions”
(SLCs) in Malagasy is that the wh-phrase in SLCs is the predicate remnant of a pseudo-
cleft construction, and occurs in clause-initial position, preceding TP, as a result of
a regular operation of predicate-fronting present in the language. The authors point
out that a common analysis of VOS word order in Austronesian languages has been to
attribute it to the raising of vP constituents to the specifier of a projection above TP. In
10 Jason Merchant & Andrew Simpson

pseudo-cleft constructions, this operation of predicate-fronting is shown to provide


the necessary input to feed TP deletion and create surface sequences with simple
wh-remnants, which closely resemble sluices in other languages. With reference to
a range of language-internal patterns, Paul and Potsdam confirm that the set of wh-
elements that can occur in SLCs is identical to the set of wh-phrases that can appear as
predicates in pseudo-cleft constructions, adding further support to their proposal. As
the authors note, the chapter contributes to the typology of ways in which wh-in-situ
languages can form sluicing-like surface structures, and provides additional support
for a non-unified analysis of SLCs in wh-in-situ languages. It is emphasized that the
strategy a language makes use of to form SLCs/sluices is directly dependent upon
the syntactic mechanisms independently available in the language, and that different
underlying mechanisms may actually result in similar surface patterns.
Chapter  of the volume visits ‘Sluicing in Indo-Aryan: An investigation of Bangla
and Hindi’, by Tanmoy Bhattacharya and Andrew Simpson. This chapter builds up a
profile of sluicing constructions in two widely-spoken languages of South Asia: Bangla
and Hindi. Although these languages have commonly been described as wh-in-situ
languages, evidence and argumentation has also been presented suggesting that both
Bangla and Hindi are actually languages with disguised overt wh-movement (Simp-
son and Bhattacharya ). Such assumptions create the expectation that Bangla
and Hindi should permit sluicing by wh-movement and PF clausal deletion as in
English. The chapter therefore attempts to determine the degree to which sluicing
constructions in Bangla/Hindi parallel or show differences from sluicing in English-
type languages, and also how they relate to patterns of sluicing in typologically closer
languages such as Japanese, where sluicing has often been analyzed as having a quite
different syntactic derivation from English. The first part of the chapter considers
both a wh-movement analysis of sluices in Bangla/Hindi and alternative reduced
cleft/copula analyses, and argues that the former is much better supported by the
range of patterns found in the two languages.
In the second part of the chapter, Bhattacharya and Simpson investigate the poten-
tial effects of movement-associated constraints on sluicing in Bangla/Hindi, in par-
ticular Superiority and island constraints. The former phenomenon is shown to be
complex in its patterning in Bangla. While Superiority effects are generally not found
to occur in regular multiple wh-questions in Bangla, they do appear more rigidly
in contexts of multiple wh-sluicing. However, it is observed that the unacceptability
of certain multiple wh sequences can be alleviated by certain strategies that involve
scrambling/repositioning the correlate antecedents of the wh-phrases in the clause
preceding the sluice, and by topicalization of the CP-remnant containing the sluiced
wh-phrases. The authors note that a Shortest Move analysis of Superiority seems to
be inappropriate to capture the patterns found, as it incorrectly predicts ungrammat-
icality in all instances of wh object > subject sequences, and does not permit for
the Superiority “repair effects” that occur in sluicing constructions. The chapter also
Introduction 11

considers the interaction of sluicing and island configurations in Bangla and Hindi,
and notes that there appear to be challenging differences between the two languages
in the acceptability of sluicing with islands. Whereas Bangla generally allows for the
kind of island repair effects suggested to characterize sluicing in English and many
other languages, Hindi is more restrictive and shows a significantly lower tolerance
for sluicing involving island constituents, in such a sense being more akin to what has
been reported for case-marked sluicing in Japanese. The authors then speculate on
whether a case-theoretic analysis such as that suggested by Nakamura for Japanese in
Chapter  might be appropriate to capture the apparent differences between Bangla
and Hindi.
The broad theme of sluicing in putative wh-in-situ languages that extends across
Chapters – is continued in the final two chapters of the volume, with studies of Chi-
nese and Turkish. Chapter , by Perng Wang Adams and Satoshi Tomioka, considers
Mandarin Chinese and defends a pseudo-sluicing analysis of sluicing-like construc-
tions in the language, as initially proposed by one of the authors, Adams (), and
also Wei (). This approach argues that “sluiced” clauses in Chinese simply consist
of a pro subject, which may refer to individuals, events or propositions present in the
discourse, the copula shi, and a wh-phrase, essentially similar to English forms such
as Who/when/why was that?, in which no clausal ellipsis is assumed. The pro-form
analysis is able to account for a range of properties in Chinese sluicing constructions
that are otherwise difficult to accommodate in a movement and deletion approach.
First is the presence of the copula shi, which is obligatory when the simplex argument
wh-phrases shei ‘who’ and shenme ‘what’ occur, and optional with other wh-phrases.
Second, the approach naturally accounts for the observation that an overt demon-
strative na ‘that’ can replace the null pronoun in all of the sluicing environments
in Mandarin. Third, it is possible to explain an argument–adjunct asymmetry that
exists in sluicing in Chinese but not in English—unlike English, Chinese sluicing with
an argument wh-phrase requires an overt antecedent to be present in the preceding
sentence/clause. This is argued to relate to a generalization established in Heim ()
that implicit/unexpressed arguments are not salient enough to license the use of
pronouns in subsequent discourse. When adjuncts are involved in sluicing, however,
the pro is taken to be an instance of sentential anaphora referring to the whole of a
preceding sentence/clause as an event or proposition, and hence a null pronominal
is licensed and identified in such contexts. Fourth, the pseudo-sluicing analysis is
able to make sense of a restriction on the occurrence of means/manner zenmeyang
‘how’ in sluices—this is shown to be similar to the unacceptability of means/manner
how in parallel English copula-plus-pronoun forms such as ∗ How was that?, though
means/manner how is fine in sluices involving movement and deletion: We do know he
did it, but we don’t know how he did it how. Finally, the total lack of island-sensitivity
in Chinese sluicing constructions (including a lack of Left Branch Condition effects,
as noted by Wei (), in clear contrast to the presence of such effects in English)
12 Jason Merchant & Andrew Simpson

is a natural consequence of the lack of movement assumed in the pseudo-sluicing


approach for Chinese.
Adams and Tomioka also discuss two patterns which initially appear to be
challenges for the pro-form analysis of Chinese sluicing: the availability of sloppy
interpretations in sluicing and the possibility of multiple wh-remnants. The former
property has been presented as evidence for a movement-and-deletion-type approach
to Chinese sluicing by Wang and Wu (), but Adams and Tomioka argue that the
distribution of sloppy identity in Chinese sluices is actually better predicted under
their proposal. They also give empirical arguments relating to conditions on the use
of the copula shi that multiple sluicing can be very naturally analyzed in terms of the
pro-form analysis; hence such patterns are also not damaging to the pseudo-sluicing
approach. Finally, the chapter considers two alternative analyses of Chinese sluicing—
a movement and clausal deletion account, and a reduced cleft analysis—and points
out how both alternative approaches face problems in accounting for the particular
properties of sluicing constructions found in Mandarin Chinese.
The last chapter of the volume is an investigation of ‘Sluicing in Turkish’, by Atakan
Ince. Turkish is a case-rich, agglutinating SOV language that is assumed not to have
wh-movement. In his analysis of sluicing-like constructions in Turkish, Ince argues
that these are derived in essentially the same movement and TP-ellipsis way as in
English, except that it is focus features that drive the movement of wh-phrases to the
left periphery in Turkish sluices rather than wh-features. Such a proposal is noted to
go against suggestions in Kuwabara () and Kizu () that sluicing constructions
in wh-in-situ languages like Turkish are elliptical cleft constructions, and Ince devotes
much of the first half of the chapter to showing why a cleft analysis is not appropriate
to model Turkish sluices. It is pointed out that Turkish exhibits the case-matching
requirement of sluices found in many languages, in which the case of the wh-remnant
must match that of the correlate in the antecedent clause, and can therefore be any
of a number of cases in Turkish, whereas clefted NPs in Turkish only ever surface
with nominative case. In a similar vein, postpositions appear pied-piped with wh-
remnants in sluicing constructions, but can never occur with pivot NPs in clefts.
Other properties distinguishing sluices and clefts in Turkish are (a) the possible occur-
rence of adjuncts as wh-remnants in sluices but not as pivots in cleft constructions,
(b) the regular occurrence of a copula in clefts with both argument and adjunct NPs,
but the impossibility of a copula with wh-remnants that are adjuncts in clefts, and
(c) the possible occurrence of multiple wh-remnants in sluicing contrasting with the
unacceptability of multiple elements as pivots in cleft constructions. On the grounds
of this range of differences between clefts and sluicing constructions, Ince concludes
that the latter are not formed via the reduction of a cleft construction, and must instead
arise via movement of the sluiced wh-phrase to the C-domain, attributing this to focus
movement, which also results in a special emphatic intonation pattern being applied
to the wh-remnant.
Introduction 13

In the second part of the chapter, Ince investigates what appears to be a puzzling
exception to the case-matching requirement on wh-remnants in sluicing construc-
tions: the nominative case which occurs on a wh-phrase that has as its correlate the
subject of an embedded clause. Embedded clauses are nominalized in Turkish, and
their subjects regularly surface with genitive case. However, sluiced wh-phrases with
embedded subject antecedents can only occur in nominative case, and not genitive
case. Ince shows that the nominative case cannot be attributed to any underlying
cleft structure with sluices involving embedded clause subjects, as multiple sluicing
is possible with such subjects, and this also requires that the subject be in nominative
case (and, as noted, multiple clefting is not acceptable in Turkish). Ince then goes
on to develop an interesting analysis of the phenomenon which makes use of and
modifies Hiraiwa’s () account of nominative–genitive conversion in Japanese and
the Minimalist assumption that there is a process of cyclic Spell-Out which applies
to phase constituents (Uriagereka ; Chomsky , ). Ince argues that in
Turkish sluicing structures, an embedded wh-subject cannot be licensed with genitive
case because this requires agreement between C and a complex T-v-V head created in
the syntax, but in sluicing constructions the latter is deleted before it can agree with the
phase head C . The sluiced wh-phrase therefore has to bear the nominative case that is
assigned by a T-v-V amalgamate failing to agree with C . Sluicing, in conjunction with
cyclic Spell-Out is thus held to block the transmission of information that would result
in a surface case form, and in this sense Ince’s conclusions offer a further potential
insight into the general issue of PF-related case-licensing discussed in Chapters 
(Stjepanović) and  (Nakamura).

Guess who?
JOH N ROBE RT RO S S

.
I will be concerned in this paper with describing the operation of a rule which I will
refer to as “Sluicing”.  This rule converts sentences like those in () to the correspond-
ing sentences in ().
() a. Somebody just left—guess who just left.
b. Ralph is going to invite somebody from Kankakee to the party, but they don’t
know who he’s going to invite to the party.
c. He is writing (something), but you can’t imagine
⎧ ⎫

⎪ what ⎪


⎪ where ⎪


⎪ ⎪


⎪ ⎪


⎪ why ⎪


⎨ ⎪

how (fast)
⎧ ⎫ he is writing.

⎪ ⎪ to ⎪ ⎪

⎪⎨
⎪ ⎪ ⎪
⎬ ⎪


⎪ with ⎪


⎪ whom ⎪


⎪ ⎪ for ⎪ ⎪


⎩⎩⎪ ⎪
⎭ ⎪

etc.
() a. Somebody just left—guess who.
b. Ralph is going to invite somebody from Kankakee to the party, but they don’t
know who.

 The research on which this paper was based was supported in part by NIMH Grant MH-. The
paper was originally published in Papers from the th Regional Meeting of the Chicago Linguistic Society,
edited by Robert I. Binnick, Alice Davison, Georgia M. Green, and Jerry L. Morgan, –. I would like
to express my thanks to Joan Bresnan and George Williams for many helpful comments; to Bruce Fraser,
Morris Halle, and David Perlmutter for reading and making many improvements on an earlier version;
and especially to George Lakoff, whose penetrating criticisms and insightful suggestions have led directly
to some of the major conclusions of the paper.
Guess who? 15

⎧ ⎫

⎪ what ⎪


⎪ ⎪


⎪ where ⎪


⎪ ⎪


⎪ why ⎪


⎨ ⎪

how (fast)
c. He is writing something, but you can’t imagine ⎧ ⎫ .

⎪ ⎪ to ⎪ ⎪


⎪ ⎪
⎨ ⎪
⎬ ⎪


⎪ with ⎪


⎪ whom ⎪


⎪ ⎪
⎪ for ⎪
⎪ ⎪


⎩⎩ ⎭ ⎪

etc.
This rule has the effect of deleting everything but the preposed constituent of
an embedded question, under the condition that the remainder of the question is
identical to some other part of the sentence, or of a preceding sentence.
In Section ., I will present three arguments to the effect that the sentences in
() and () must be related, and that the sentences in () are less basic than those in
()—that in fact the remainders of the embedded questions in () must be “sluiced” to
produce the sentences in (). In Section ., I will argue that the rule in question must
apply after the rule of Question Formation has preposed the various “question-words”
to the beginning of the embedded clauses in (). Finally, in Section ., I will discuss
briefly the other, broader implications for the theory of language which will follow
from the adoption of Sluicing as a grammatical deletion transformation.

.
In this section, I will argue against a man that is presently made of straw, since no other
grammarian, to the best of my knowledge, has discussed such sentences as those in
() in detail. No one, therefore, has yet proposed an alternative analysis which runs
directly counter to the derivation I have suggested—that the sentences of () result
from the application of a rule of deletion to the sentences of (). However, recent work
by Akmajian (), Chomsky (), Dougherty (), Jackendoff (), and others
has suggested the logical possibility of a different source for the second clauses of
the sentences of ()—namely, essentially their surface structures. That is, it might be
argued that the second clause of (b) should be derived not directly from the second
clause of (b), but rather directly from the approximate structure shown in ().
() S

Neg NP Aux VP

they Tns V NP

Pres know who


16 John Robert Ross

The fact that (b) is synonymous with (b) would, in such an analysis, not be accounted
for by deriving both from the same deep structure, which is the account of this fact
that I would give, but rather by proposing an interpretive semantic rule, which would
inspect the structures in the environment of () and determine that the NP who must
bear the grammatical relation of object to the verb invite. Let me refer, in the following
discussion, to any theory of the sentences in () which does not analyze the question-
words there as being the last remnants of full question clauses as an “interpretive”
theory.

..
... The first argument against an interpretive theory of the facts of () was sug-
gested by facts pointed out to me by George Williams. He observed that in German,
where sentences paralleling those in () are to be found, the question-word must,
under certain conditions, agree in case with some NP in a preceding clause. Thus in
(), only, wem (dative) ‘whom’ is possible, because the verb schmeicheln ‘to flatter’
takes only dative objects.

() Er will jemandem schmeicheln, aber sie wissen nicht


he wants someone.dat flatter but they know not

wem
∗ wen .

who.dat/∗ who.acc
‘He wants to flatter someone, but they don’t know who.’

In (), by contrast, only wen (accusative) ‘whom’ is possible, because the verb loben
‘to praise’ takes only accusative objects.

() Er will jemanden loben, aber sie wissen nicht


he wants someone.acc praise but they know not

wen
∗ wem .

who.acc/∗ who.dat
‘He wants to praise someone, but they don’t know who.’

Similar facts can be adduced for that dialect of English which distinguishes between
who and whom. For this dialect, whom would appear in (b) and (b), but who in ().

() a. Somebody from Kankakee is going to be invited to the party by Ralph, but

who
they don’t know ∗ is going to be invited to the party by Ralph.
whom
b. Somebody from Kankakee is going to be invited to the party by Ralph, but

who
they don’t know ∗
whom.
Guess who? 17

Excluding for the moment the facts of (b) and (b), the English rule of Case
Marking would seem to be approximately that stated in (). 
() Case Marking 
X – NP – Y
obl
   ⇒
 [+objective] 
Condition:  is not dominated immediately by S.  , 
But now observe that the facts of (b) and (b) cause grave difficulties with respect
to Case Marking, for an interpretive theory, at least. For in such a theory, the deep
structure of the second clause of (b) is identical to the surface structure of the second
clause of (b), namely (). How can such a theory account for the fact that rule () must
apply to () in (b)—and, to complicate matters, in the derivation of (),
() Whom don’t they know?
—but not in the derivation of (b)?
At least part of the answer to this question is evident: rule () would have to apply
to the output of the interpretive semantic rule. That is, whether or not who or whom
was appropriate could not be established until the interpretive rule had determined
how the question-word functioned in the preceding clause.
But now note that the output of the interpretive rule cannot be the semantic repre-
sentation itself, for the question-word is a semantic direct object of invite, regardless of
whether it shows up in the nominative or in the objective case. Thus the interpretive
rule must produce an output structure which is in the form of a passive, for it is only
to such a structure that Case Marking can be correctly applied.
 I say “approximately” because of the existence of people who profess to say such bizarre strings as It is
we and the like. If such people do exist, rule () can be modified in a trivial way.
 Note that the existence of such variants as Big O is taller than I/me is not to be attributed to different
statements of Case Marking, but rather to the way this rule is ordered with respect to the rule which deletes
elements in than-clauses. That is, since all speakers agree in accepting Big O is taller than I am and rejecting
∗ Big O is taller than me am, dialects which have nominative pronouns after than can be accounted for by
ordering Case Marking before the rule which deletes am in the than-clause. On the other hand, dialects
which exhibit oblique pronouns after than can be accounted for by reversing the order of these two rules,
for when am has been deleted, the node S which used to dominate the than-clause will be pruned (for a
discussion of this operation on trees, cf. Ross (: Chapter , especially Section ..)), and the subject of
the than-phrase, being no longer dominated immediately by S, will not be protected from being converted
into an oblique pronoun by the condition on Case Marking.
 This condition is not stateable in the present theory of grammar, which only allows Boolean conditions
on analyzability in the statement of language-particular transformational rules, unless the inventory of basic
predicates in which such conditions can be stated is expanded to include A ≥ B (“A immediately dominates
B”) in addition to A > B (“A (weakly) dominates B”) and A = B (“A is identical to B”). I submit, therefore,
that the correctness of rule () argues strongly for the need of adding this predicate to the list of basic
predicates.
 The condition is more complicated, due to the fact that conjoined subject NPs, even though they are
not immediately dominated by S, must remain in the nominative case: God and I are similar. I will not go
into the changes of rule () that follow from this fact.
18 John Robert Ross

Unfortunately, however, if the output of the first interpretive semantic rule has the
form of a passive, then a second interpretive rule, Anti-Passive, will be necessary to
convert this intermediate form to its active form, so that other interpretive semantic
rules will be able to apply to the output of Anti-Passive in the desired fashion. I
am doubtful that the prospect of postulating a grammatical rule of Passive and a
semantic rule of Anti-Passive, or, alternatively, of only postulating the latter rule,
and of generating passives as deep structures, would be embraced with relish by any
interpretivist. And, of course, this argument generalizes easily. For example, if there
is a rule of Tough Movement,  which converts (a) into (b),
() a. It was tough to get the marbles away from one of your kid sisters.
b. One of your kid sisters was tough to get the marbles away from.
then the fact that Guess whom? can follow (a), but not (b), indicates the need for
an interpretive rule of Anti-Tough Movement. In fact, if any interpretive theory of the
sentences in () is correct, either there are no transformational rules which produce
derived subjects, or for each such rule there is an equal and opposite interpretive rule
which reverses its effects, after Case Marking has applied.
To be sure, no interpretive theory of the sentences in () exists, so I may be beating
a dead horse, but I think it is perhaps of some value to catalog the additional assump-
tions that proponents of any such theory would have to make.
... Consider next sentences such as that in ().
() He’s going to give us one old problem for the test, but which problem isn’t clear.
An interpretive theory would postulate, as the deep structure of the second clause of
(), the structure in ().
() S

Neg NP Aux VP

Det N Tns V Adj

+N Pres be clear
+sing

which problem
But what of such sentences as ()?
() He’s going to give us some old problems for the test, but which problems isn’t
clear.
 This term is Paul Postal’s (Postal ). Postal expresses doubt, and I agree with him, as to whether
such a rule can be assumed to exist.
Guess who? 19

The importance of this sentence, as far as any interpretive theory is concerned, is of


course the fact that it manifests singular number agreement, despite the fact that
its subject is superficially plural. Since the deep structure of the second clause of
(), under such a view, would be identical to (), except that [−sing] would replace
[+sing], two problems will arise. First of all, some way must be found to distinguish
between the second clause of () and the unrelated, though superficially highly sim-
ilar (), which manifests plural number agreement.

() Which problems aren’t clear and easy to do?

Secondly, it is not sufficient for the interpretive rule to look at () and determine
that the NP which problems must bear the grammatical relation of direct object to the
verb give: In addition, it must somehow provide an output that the rule of Number
Agreement can apply to to yield the desired singular form of the verb. I can see only
two possibilities for this, neither of them being very palatable.
The first is for the interpretive rule to mark any question-word it operates on with
some ad hoc feature, and to state the rule of Number Agreement in such a way as to
make reference to this ad hoc feature and to treat any NP bearing this feature as a
singular NP, irrespective of the value of the feature [±sing] which the NP bears.
The major defect in this proposal should be obvious: it makes accidental the fact
that the Number Agreement rule chooses to interpret the ad hoc feature as singular,
instead of plural. But this fact should follow automatically, in any correct analysis,
from the fact that complement sentences generally behave like singular NPs, as the
sentences in () attest.

is
() a. That Bill left ∗ tragic.
are

is
b. Why he did it ∗ a puzzle.
are

gives
c. Being drunk ∗ me a kick.
give

The second possible way of accounting for the singular agreement in () would be
for the interpretive rule not to merely plug in (somehow) which problems as the direct
object of give in the left half of (), but for it to reconstruct, by the inverse operation of
Sluicing, in effect, an embedded question in subject position in the right-hand clause
of (). The rule of Number Agreement could then be stated in a general form, but
a major new problem would have arisen: what to do with this new subject, from a
semantic point of view? If it must be interpreted, then, since it can be of any desired
degree of transformational complexity, interpretive rules which undo the effects of
virtually all transformations will have to be added to the semantic component, as far
as I can see. It may be, of course, that I have not been able to imagine a resourceful
enough interpretivist as a straw man, but I would be surprised if the twin difficulties
20 John Robert Ross

posed by sentence () and by sentence () could be satisfactorily overcome within
an interpretive theory.
... A third argument that what is involved in the generation of such sentences
as those in () is a deletion transformation, and not an interpretive rule, is provided
by such sentences as those in ().
() a. She says she’s inviting some men—I wonder how many men?
⎧ ⎫
⎨ those old men ⎬
b. ∗ I wonder the centerfielder for the Cardiac Kids ?
⎩ ⎭
your uncle Casimir
If phrase structure rules are to generate as a well-formed deep structure the second
clause of (a), I can see no natural way of preventing the same phrase structure rules
from generating the ungrammatical strings of (b). Therefore, any linguist wishing
to maintain an interpretive theory here will be forced to give up a rather natural
characterization of the possible objects of the verb wonder—namely, that this verb
must be followed in deep structure by an embedded question,  a restriction which
correctly excludes the non-sentences of (b). Instead, such a linguist, in order to
generate the second clause of (a), will have to generate the sentences of (b) as
syntactically well-formed, but semantically anomalous.
But, it might be countered, selectional violations are indeed semantic violations, 
so there is nothing wrong with this claim. Unfortunately, however, the linguist willing
to give up this bath-water will also have to jettison the baby, for he will not only have to
claim that (a) and (b) are both syntactically well-formed, the latter being seman-
tically deviant, but also that (c) is syntactically well-formed, and only semantically
deviant.
() a. I polished my tongue.
b. ∗ I polished my liberty.
c. ∗ I polished valid.
To see why this is so, consider the sentences in ().

 I ignore here the homophonous verb wonder that appears in such sentences as I wonder that you
survived. This verb, though intuitively closely related to the wonder of (a), must apparently differ from it
in at least some features, for while the wonder of (a) can appear with the progressive auxiliary, this wonder
cannot. Also, whereas Bill didn’t wonder whether you had left is perfectly natural, ?∗ Bill didn’t wonder that
you had left strikes me as decidedly peculiar. I also do not intend the above characterization of possible
objects of wonder to apply to wonder about, which seems to differ from both of the other wonders, despite
the strong intuition that I have that we are dealing here with variants of the same lexical entity. I can see
no way of capturing this intuition at present.
 This position, with which I agree, was first defended by Jackendoff (see Jackendoff ()), and then,
independently, by McCawley (see McCawley ()).
Guess who? 21

() a. He says that her objections are valid, but I wonder how valid?
b. ∗ I wonder valid.

Since in all sets of phrase structure rules proposed to date, adverbs of degree (and
how in (a) is one) are optional constituents of adjective phrases, there is no non-
ad hoc way of allowing the second clause of (a) as a well-formed deep structure,
but excluding (b). Thus one advocating an interpretive theory must abandon the
claim made in Chomsky () that while choosing incorrect subcategories of the
same lexical category cannot lead to ungrammaticality, in general (see Chomsky (:
Chapter ) for a more precise exposition of this claim), choosing incorrect lexical
categories always results in ungrammatical strings.
It is not my purpose here to argue for or against Chomsky’s claim. I regard the
issues as open. Rather, I merely wish to point out that anyone wishing to maintain an
interpretive theory of the sentences in () must reject Chomsky’s claim, and defend
the claim that all the sentences in () are only semantically deviant.

() a. I wonder valid.


b. Hoarse wonder valid.
c. Hoarse obscure valid.

..
Let us now consider the sentences in () and (), which would also pose serious
problems for an interpretive theory in which the second clause of (b) is derived
from ().

() a. We know that he was eating, but what isn’t clear.


b. We know that he was eating, but with whom isn’t clear.
c. We know that he was eating, but how rapidly isn’t clear.

() a. We know that he was eating, but it isn’t clear what.


b. We know that he was eating, but it isn’t clear with whom.
c. We know that he was eating, but it isn’t clear how rapidly.

Intuitively, it is obvious that the second clauses of () should be derived from the
corresponding second clauses of () by the rule of Extraposition, which I have stated
approximately in (). (See Rosenbaum () and Ross (: Chapter ), for discus-
sion of this rule).

() Extraposition
X − [it − S] − Y
opt
    ⇒
   +
Condition:  = Poss NP Ing X
22 John Robert Ross

However, if the deep structures of the second clauses of () parallel (), in which
there is no NP of the form [it S]NP , obviously Extraposition will not be able to apply
to (), unless the structural description of this rule is complicated in some ad hoc way,
so that the rule will not only extrapose a sentence following an it, but also a question-
word, where this latter branch of the rule will also have to insert the pronoun it under
the NP which used to dominate the question-word.
Unfortunately, the difficulties in accounting for the sentences of () under an
interpretive theory will still not be avoided, even with this ad hoc complication of
Extraposition. For because the question-words which remain after Sluicing applies
can be homophonous with question-words which did not appear as constituents of
embedded clauses (thus the second clause of () is ambiguous), the extended rule
of Extraposition would have to be restricted in some way from converting (a),
whose subject cannot be interpreted as a constituent of an embedded question (cf.
the ungrammaticality of ()), into the ungrammatical (b).
() a. Who is hoarse?
b. ∗ It is hoarse who?
⎧ ⎫

⎪ Who left ⎪

⎨ ⎬
∗ Why he did it
() is hoarse.

⎪ How long the stew cooked ⎪

⎩ ⎭
Whether the moon’s a balloon

Thus, a condition would have to be placed on the extended rule of Extraposition which
restricted the extraposing of question-words to those sentences whose main verb or
adjective allowed embedded questions in deep structure, an obviously ad hoc and
repetitious “solution” to the problem posed by ().
Confronted with such a problem, one wishing to deny the existence of a rule of
Sluicing, and to maintain a version of an interpretive theory, might propose () as
the deep structure of the second clause of (a).
() S

Neg NP1 Aux VP

N S2 Tns be Adj

it Q NP2 Pres clear

what
To this structure, the rule of Extraposition stated in () could apply, without ad
hoc extensions being necessary. An additional advantage of () is that the problem
Guess who? 23

of getting singular number agreement in sentences like () would be solved. The
node NP in () would determine number agreement, and it would be singular
by virtue of the generalization underlying (). It would probably also be possible
to avoid the arguments in Section ... above, which derived from the deviance
of such sentences as those in (b), (b), (c), and (), for an interpretivist could
require that every constituent appearing in place of NP in () would have to be a
question-word.
The argument in Section ..., however, is still valid, as far as I can see: the
impossibility of whom in (b) could not be accounted for with the general rule of
Case Marking which is given in (). In addition, the deep structure shown in (),
which implies the existence of a phrase structure rule like that shown in (),
⎧ ⎫
() a. ⎪
⎪Q NP/it — ⎪


⎨ ⎪

S→


⎪ Q ⎪


⎩ (Neg) NP Aux VP⎪

b. Imp

raises additional problems, for the rule in (a) can be iterated. Thus the second clause
in (a) could be derived from (), or from (a), or from (b).
S b. S
() a.

Neg NP Aux VP Neg NP Aux VP

N S Tns be Adj N S Tns be Adj

it Q NP Pres clear it Q NP Pres clear

N S N S

it Q NP it Q NP

what what
In fact, every sentence containing a sluiced question would be derivable from an
infinite number of distinct deep structures. This undesirable result could only be
avoided by some ad hoc restriction which prohibited such deep structures as those in
(). The ad-hoc-ness of any such restriction, coupled with the fact that the argument
in Section ... appears to remain valid, indicates that () cannot be regarded as a
viable alternative to an analysis incorporating a transformational rule of Sluicing. In
addition, the arguments to be given in Section . below, which suggest that Sluicing
must be ordered after the rule of Question Formation, pose equally grave problems
for an interpretive theory, whether based on () or on ().
24 John Robert Ross

.
The three phenomena that I will discuss in this section are not only further arguments
for the existence of a transformational rule of Sluicing—they also constitute evidence
that the rule of Sluicing must follow the rule of Question Formation.

..
Consider first the sentences in ().
⎧ ⎫

⎪ a picture of somebody, but I don’t know ⎪


⎪ ⎧ ⎫ ⎪


⎨ ⎨ who ⎬⎪ ⎬
() I know he has of whom .

⎪ ⎩∗ ⎭⎪ ⎪

⎪ a picture of whom ⎪ ⎪
⎩ somebody’s picture, but I don’t know whose picture ⎪
⎪ ⎭

An interpretive theory must treat the fact that NPs like who, of whom,  and whose
picture may appear as the deep objects of verbs like know, wonder, guess, etc., while
the NP a picture of whom may not, as an accidental fact. However, an analysis which
derives the sentences in () from those in () by a deletion rule can make use of the
fact that precisely the same set of NPs can appear at the head of embedded questions,
as () shows.
⎧ ⎫
⎪ who he has a picture of ⎪
⎪ ⎪
⎨ ⎬
of whom he has a picture
() I don’t know ∗ .

⎪ a picture of whom he has ⎪

⎩ ⎭
whose picture he has
However, the facts of () and () do not merely support the derivation of the
sentences in () by a deletion rule, as opposed to an interpretive rule. In addition,
they argue strongly against a third logically possible source for such sentences. That
is, it might be argued that (b), for instance, derives not from (b), but rather from
(), a stage derivationally prior to (b).
() Ralph is going to invite somebody from Kankakee to the party, but they don’t
know S [Q Ralph is going to invite WH + somebody to the party]S .
The rule effecting this conversion might be stated somewhat as in ().
() Z − [X − NP − Y]S − Z − [Q − X − NP − Y]S − Z
opt
          ⇒
         
Condition:  = ,  = ,  > WH + some
 I here accept the identification, at the level of deep structure, of prepositional phrases and noun
phrases, which was suggested first by Paul Postal, in classes at MIT and at the Linguistic Institute at the
University of Indiana in  (see also Postal ()) and later proposed, on independent grounds, by
Charles Fillmore (see Fillmore ()).
Guess who? 25

In applying this rule to the structures underlying the sentences of (), the variables
in terms  and  could be chosen in such a way as to make the NP of term  be
somebody, of somebody, a picture of somebody, or somebody’s picture. Since the third
of these choices of variables would lead to the ungrammatical sentence in (), some
further ad hoc restriction would have to be imposed upon rule ().
While I do not wish to argue that () is basically incorrect, it does seem to me
that it should be reformulated to follow the rule of Question Formation, which can be
stated in a preliminary fashion as in (), for reasons which will appear shortly.
() Q − X − NP − Y
   
OBLIG
    ⇒
Condition:  = WH + some + X
As was discussed in detail in Ross (, Section .), by the universal convention
of Pied-Piping, when some NP which is mentioned in the structural description of
a rule is flanked by variables, if the structural change of the rule specifies that this
NP is to be reordered, then, when carrying out this reordering, either the mentioned
NP or any higher NP meeting certain additional conditions can be moved. If it is
a higher NP that moves, the constituents of this NP are said to “pied-pipe” with
the movement of the mentioned NP. Thus although the mentioned NP in Question
Formation is the one that starts with WH (i.e. who(m)), by Pied-Piping, the higher
NP NP [of NP [whom]NP ]NP may instead be preposed.
Exactly what types of constituents may pied-pipe varies from language to language,
from rule to rule, and even from dialect to dialect. In particular, while it is possible to
pied-pipe NPs of the form a picture of whom in forming relative clauses, as (c) shows,
() a. J. Edgar Hoover, who I have a picture of in my locket, is a cutie.
b. J. Edgar Hoover, of whom I have a picture in my locket, is a cutie.
c. J. Edgar Hoover, a picture of whom I have in my locket, is a cutie.
this type of NP must be prevented from pied-piping in the rule of Question Formation,
as the ungrammatical sentence of () shows. 
The conclusion is obvious: if the rule of Sluicing, appropriately revised, follows
the rule of Question Formation, then the ad hoc restriction on this latter rule which
prevents the pied-piping of NPs like a picture of whom need not be repeated, as would
be the case if the reverse ordering obtained. Therefore, I claim, the sentences in () are
to be derived by a deletion transformation, not from such more basic strings as that

 Joan Bresnan has called to my attention such sentences as He has a picture of somebody, but a picture
of whom (∗ he has) I don’t know, which are obvious counterexamples to this generalization. I confess to being
totally baffled by such sentences, which may well totally invalidate this first argument. I can do nothing but
call the attention of future researchers to this problem, in the hope that they will be able to solve it.
26 John Robert Ross

in (), but rather from such strings as those in (), which are the output of the rule
of Question Formation.

..
A second argument for this ordering can be derived from the following considera-
tions. Normally, the pied-piping of a preposition, when its NP object is questioned, is
optional, as the variants in () suggest. However, there are conditions under which
prepositions may not pied-pipe, and must be stranded by any rule which moves the
NP which follows the preposition (see Ross (: Section .) for some discussion of
this phenomenon). Thus, in my speech (there is much dialectal variation with respect
to these facts), while the “a” sentences in (–) are grammatical, the “b” sentences,
in which the prepositions have pied-piped, are not.
() a. Who are you going to do away with?
b. ∗ With whom are you going to do away?
() a. What will we have to make do with?
b. ∗ With what will we have to make do?
() a. Which plot did the FBI get wind of first?
b. ∗ Of which plot did the FBI get wind first?
The restriction I proposed (Ross ), namely that all prepositions must be
stranded when following an idiomatic sequence VA, where A is any single constituent
(e.g. particle, verb, adjective, etc.), may or may not stand up under further investiga-
tion, but whatever the correct restriction to exclude the “b” sentences above turns out
to be, if the rule of Sluicing follows Question Formation, the latter rule’s operation
being subject to this restriction, then the contrast between the “a” sentences and the
“b” sentences in (–) will follow automatically.
() a. Bill’s planning on doing away with one of his in-laws, but I don’t know which.
b. ∗ Bill’s planning on doing away with one of his in-laws, but I don’t know
with which.
() a. We’ll have to make do with some kind of vile . beer for our punch, but I
don’t know exactly what kind.
b. ∗ We’ll have to make do with some kind of vile . beer for our punch, but I
don’t know with exactly what kind.
() a. The FBI got wind of one of the many plots to smoke draft cards, but I can’t
remember which.
b. ∗ The FBI got wind of one of the many plots to smoke draft cards, but I can’t
remember of which.
It is not obvious to me how any interpretive theory of sluiced questions can make
use of a restriction on pied-piping to account for such contrasts as those in (–), for
Guess who? 27

in such theories, the second clauses of these sentences do not undergo any movement
rules in their derivations. Thus I regard these sentences as constituting strong evidence
for the existence of a transformational rule of Sluicing.

..
A third argument, which exactly complements the one just discussed, can be adduced
from sentences of a type not considered so far—(b).
() a. She was dancing, but I don’t know with whom.
b. She was dancing, but I don’t know who with.
Under certain conditions, which I will not go into in detail here, it is possible to
delete everything in an embedded question except the question-word and a stranded
preposition. This kind of Sluicing is the source of the elliptical question What for?,
and of many other sentences like (b).
One of the conditions under which this kind of Sluicing is not possible is when the
stranded preposition introduces any one of a number of types of adverbial preposi-
tional phrases.  An example of this restriction is the impossibility of (b).
() a. He would report me under some circumstances, but I can only guess
under which.
b. ∗ He would report me (under some circumstances), but I can only guess
which under.
If Sluicing follows Question Formation, the ungrammaticality of (b) can be made
a consequence of the fact that the prepositions of prepositional phrases of condition
which begin with under must pied-pipe—such sentences as (b) are impossible.
() a. Under what circumstances will the moon implode?
b. ∗ What circumstances will the moon implode under?
Of course, if these two rules appear in the reverse order, the fact that just those prepo-
sitions which cannot be stranded by Question Formation cannot follow the question-
word in a sluiced embedded question will have to be mentioned in the statement of
the rule of Sluicing, as well as being mentioned as a constraint on Pied-Piping, where
it is independently motivated. Thus the facts of () and () again argue strongly for
the ordering Question Formation—Sluicing.
Obviously, these same facts also constitute strong counter-evidence for any inter-
pretive theory of the sentences in (), since any such theory would have to mention
the idiosyncratic behavior of such prepositions as under in the rule which “interprets”
the question-word as a full embedded clause.

 This restriction was first observed by Yuki Kuroda (see Kuroda ()).
28 John Robert Ross

Actually, however, the existence of such sentences as (b) constitutes a far graver
problem for any interpretive theory, for such a theory will have to add either the
malodorous phrase structure rule () to the grammar of English, or the equally
unappetizing reordering transformation in (), in order to generate the surface con-
stituents NP and preposition, in that order.
() NP → NP + P
() X – [P – NP]NP – Y
opt
    ⇒
  + 
Condition:  = WH + X
It is apparent that the greatest difficulty with these rules is limiting their usefulness.
Without severe and ad hoc restrictions, strings like those in (a) and (b) will be
converted, by () and (), respectively, into (a) and (b).
() a. NP Aux V NP NP
b. Who talked to whom about what?
() a. ∗ [Joe of]NP will talk [Bill to]NP [his investments about]NP
b. ∗ Who talked whom to what about?
Summing up, I think that the three arguments in Section ., and the three in the
present section, demonstrate conclusively that the deep structure of such sentences
as those of () are not identical to their surface structures, but rather that they derive
from sentences such as those in (), by a transformational rule of Sluicing, which,
furthermore, must be ordered after Question Formation. In the next section, I will
sketch some of the consequences of these conclusions.

.
..
A preliminary formulation of the rule of Sluicing appears in (). 

 That the formulation in () is inadequate should be obvious enough not to require extensive
comment. The major difficulty with () is term , which is necessary so that sentences like (b) can
be generated. But clearly, in any correct statement of Sluicing, no such term should appear. The fact that a
preposition can follow the question-word in a sluiced question should follow automatically from the fact
that prepositions can be stranded in English. In French, German, and Russian, and in many other languages,
where prepositions cannot be stranded, sentences like (b) do not exist. Thus to state term  in a rule of
Sluicing is to miss a generalization. I will attempt a more adequate formal statement of this rule in the
revised version of this paper which will appear in Steinberg and Jakobovits [published in ].
Guess who? 29

() Sluicing
W − [X − ([−Def])NP − Y]S − Z − [S NP − [S X − (P) − YS ]S ] − R
opt
          ⇒
         
Conditions:  = 
=
[] is an embedded question
The first thing to note about this rule is that it applies to non-constituents, a prop-
erty distinguishing it from most other transformational rules. However, it may be only
necessary to weaken the natural and desirable constraint that all transformations can
only affect constituents to the extent specified in (b). 
() a. Only nodes may be transformationally adjoined or substituted.
b. Terms which are transformationally adjoined to or substituted for (deletion
being a special case of the latter) must be either nodes or variables.
The second fact of importance about rule () is that the variable in term  must be
able to range over sentence boundaries, if this rule is to convert sentence sequences
like (a) into sequences like (b).
() a. We’re going somewhere today. Who knows where we’re going today?
b. We’re going somewhere today. Who knows where?

 Two recent papers, Burt () and Pope (), present additional evidence that rules must be able
to delete non-constituents. The former paper argues that sentences like After Bill, John spoke to Mary about
sloppiness must derive from the fuller structure underlying After John spoke to Bill about sloppiness, John
spoke to Mary about sloppiness.

(i) Although I don’t know who, I know he wants to see someone.


(ii) Although I know he wants to see someone, I don’t know who.
(iii) I know he wants to see someone, although I don’t know who.
(iv) ?∗ I don’t know who, although I know that he wants to see someone.
This last sentence is decidedly strange, in my speech, unless the subordinate clause is separated by an
extra-long pause. Thus Sluicing would appear to have to be formulated as a bidirectional rule of deletion
under identity. The fact that it obeys the basic pronominalization constraint (that pronouns cannot both
precede and command the NP they refer to), as can be seen by the four sentences above, thus provides one
more piece of evidence that such sentences as () are derived by a rule of deletion; if a transformational
process of deletion is involved, the fact that it obeys the basic pronominalization constraint is an automatic
consequence of the generalization stated in Section . of Ross (), which specifies that all deletions
are subject to the pronominalization constraint. It is not apparent how any interpretive theory of sluiced
sentences could make use of this restriction. Pope () argues that the difference in the intonation of
the phrase Yes, happily (namely, falling or rising pitch), when used as a short answer to the question Is
Bluebeard married?, is to be accounted for by allowing a transformation which can delete variables to apply
after the output of the intonation rules that produce the differences in pitch between the sentences Yes, John
is happily married (falling pitch) and Yes, John is married, happily (rising pitch).
30 John Robert Ross

..
A third fact of importance has to do with the ambiguity of such sentences as ().
() Harold scratched his arm and so did I.
Most speakers find that () can mean either that the speaker scratched Harold’s arm,
or that the speaker scratched his own arm. That is, () seems to have to be derivable
from either (a) or (b).
() a. Haroldi scratched hisi arm and Ij scratched hisi arm too.
b. Haroldi scratched hisi arm and Ij scratched my j arm too.
The problem is, of course, that the two verb phrases in (b) are not strictly identical,
so any theory in which () can be derived from (b) must contain a definition of
“sloppy” identity, in which it is specified exactly what differences can be disregarded.
A very preliminary definition was given in Ross (: Section .), in which it was
suggested that two otherwise identical strings which differed only in commanded
pronouns (as is the case with scratch hisi arm and scratch myj arm in (b)) could
be regarded as identical for the purposes of deletion. 
Recently, an alternative to this solution, involving interpretive rules for pro-forms
like do so, do it, so do, etc., has been proposed (Akmajian ). Essentially, Akmajian
suggests that () should be derived from a structure very close to its surface structure,
and that the interpretive rule for so did should either “plug in” the NP I into both the
subject and possessive pronoun positions of the clause on the left, or only into the
subject position, thus accounting for both possible meanings. Akmajian extends his
proposal to deal with more complicated examples, such as (),

 This definition may provide a necessary condition for deletion under sloppy identity, but it is far from
providing a sufficient one, as cases like the following show.
⎧ ⎫
⎨ you ⎬
() I told you that you would be famous, and Jack i told Betty j (that shej would be famous).
⎩ ⎭
hei
As the parentheses suggest, the that-clause in the second half of the above sentence can be deleted. If
the underlying subject was you, the definition of sloppy identity is not called into play, for the string that
you would be famous is repeated in both clauses. However, the string that shej would be famous can also be
deleted, for its only difference from the former string is the difference between the pronouns you and she,
both of which are commanded by their antecedents.
The problem is that the proposed definition would also allow the deletion of the string that hei would
be famous, for the pronoun he is also commanded by its antecedent. Since the sentence can never have
this meaning when the that-clause is deleted, it is obvious that further restrictions, probably having to do
with the notion “corresponding” command pronoun, must be incorporated into the definition of sloppy
identity. The problem of improving upon the proposed definition appears to be staggeringly complex, and
it has thus far been unresolved.
Guess who? 31

() Bellwether hoisted Furbelow up into a pine tree and poured paint on him, and
I want to do it to Peapod, using a cactus and catsup.
where he claims that the pro-form do it must stand for the discontinuous chunk . . .
hoist . . . up into . . . and poured . . . on . . . . While I do not disagree with Akmajian’s
intuition as to the meaning of this pro-form, I would reject the claim that the pronoun
it must refer back to a non-constituent. For if the underlying structure of all activity
sentences, such as (a), contains a higher verb do, as in (b), an analysis which is
supported on many independent grounds, 
() a. Wolves whistle.
b. S

NP VP

wolves V NP

do it S

NP VP

wolves V

whistle
then the structure immediately underlying () will be that shown in ().

 See Lakoff and Ross (in preparation) for discussion of this analysis.
32 John Robert Ross

S0
()
S1 S3

NP VP and S4 ?

Bellwether V NP I want to do NP NP using NP and NP

did it S2 it S5 P NP a cactusk

to Peapodj catsupm

VP

VP VP

V NP Prt NP and VP

hoist himj up P NP V NP NP
VP
into itk pour itm P NP
VP VP
on himj
V NP Prt NP and VP

hoist up P NP V NP NP

Furbelow i into pour paint P NP

a pine tree on himi

Under the proposed definition of sloppy identity, S and S are identical in (), so
the independently motivated rule of S Deletion, which converts sentences like (a)
into (b), can apply to delete the constituent S .
() a. Jack believes maple syrup is fattening, but I don’t believe maple syrup is
fattening.
b. Jack believes maple syrup is fattening, but I don’t believe it.
Thus, such sentences as () do not force the abandonment of the important limitation
on pronominalization operations stated in (b): In converting () to (), only a
constituent is deleted. 

 Exactly the same is true of sentences like Jack failed French, but it wouldn’t have happened with Span-
ish, which Chomsky suggests as a counterexample to the claim that only constituents can be pronominalized
(Chomsky ). Given the definition of sloppy identity, and the following structure for the second clause
of the above sentence,
Guess who? 33

Thus it appears that such sentences as () and () could in principle be generated
either by some kind of interpretive rule, or by incorporating the notion of sloppy
identity into linguistic theory. It would be desirable if outside evidence as to which of
these alternatives is correct could be brought to bear on this issue, so that the theory
will not have to contain both of these highly complex mechanisms.
Sluiced questions provide just such evidence. Consider the sentences in ().
⎧ ⎫ ⎧ ⎫
() ⎪ ⎪ ⎪ ⎪
⎪ how
⎪ ⎪
⎪ ⎪ how
⎪ ⎪


⎪ ⎪
⎪ ⎪
⎪ ⎪

⎨ when ⎬ ⎨ when ⎬
Bob knows where to crane his neck, but I don’t know where .

⎪ ⎪
⎪ ⎪
⎪ ⎪

⎪ how long ⎪
⎪ ⎪ ⎪ how long ⎪
⎪ ⎪

⎩ ∗ why ⎪
⎭ ⎪
⎩ ∗ why ⎪

My claim is that () has undergone Sluicing. Another possible source, suggested to
me by Barbara Hall Partee, might appear to be VP Deletion, the rule that converts
() into (). This rule would convert sentences like (a) into (b). A second rule
would then produce (c), which has the form of a sluiced question.
() a. Jack knows how to [play chess]VP , but I don’t know how to [play chess]VP .
b. Jack knows how to play chess, but I don’t know how to.
c. Jack knows how to play chess, but I don’t know how.
While I do not contest the derivation of (b) from (a), I see no reason to assume the
existence of the rule of To Deletion that would be necessary to convert (b) to (c).
Several facts argue against such a rule. First, the rule of Sluicing can never operate on
embedded whether-clauses, as the sentences in () show.

(i) S1

NP VP

it S2 wouldn’t have happened with NP

NP VP Spanishi

Jack V NP

failed iti

it can readily be seen that only the constituent S need be deleted under identity.
34 John Robert Ross

⎧ ⎫
() ⎪
⎪ when ⎪


⎪ ⎪


⎨ why ⎪

Ralph knows that I went, but his wife doesn’t know where .

⎪ ⎪


⎪ how ⎪

⎩ ∗ whether ⎪
⎪ ⎭

If a rule of To Deletion existed, it would have to be constrained not to convert (a)


into (b), a restriction which obviously would duplicate the one necessary to exclude
the ungrammatical sentence in ().

() a. Buff doesn’t know whether to punt or not, and I don’t know whether to,
either.
b. ∗ Buff doesn’t know whether to punt or not, and I don’t know whether, either.

The second reason, which is to my mind far more compelling than the first, is that
the putative rule of To Deletion would be totally superfluous, if Sluicing is in the
grammar. That is, the sentences in () and (c) would automatically be generated by
the simplest formulation of Sluicing, so there are no sentences that only To Deletion
can generate. I conclude that no rule of To Deletion should appear in the grammar.
Let us suppose, however, that my arguments against To Deletion can be shown to
be invalid. Sloppy identity will still be required, in order to convert (a) to (b), the
latter being the input to To Deletion.

() a. I know how to [crane my neck]VP , but you don’t know how to [crane your
neck]VP .
b. I know how to crane my neck, but you don’t know how to.

Observe that the VP crane my neck here deletes the VP crane your neck, which can only
be identical if some definition of sloppy identity appears in the theory of grammar.
A last-ditch attempt at keeping sloppy identity out of the theory of grammar could
be made, if one were to deny the existence of a rule of VP Deletion. Such sentences as
those in () indicate the scope of the problems confronting any such analysis.

are
() a. Some people think there are no such rules, but there ∗ .
is
b. He knows how to dress, but I don’t know how to.
c. ∗ He knows how to get high, but he doesn’t know why to.

am
d. Paul Anderson’s fat, and I ∗ too.
’m
If there is no rule of VP Deletion, but rather some such interpretive rule as Akmajian
proposes, then such VP-less sentences as the underlined phrases in () will have to be
generated as deep structures. Particularly troublesome will be the problem of limiting
the deep structure position of the NP expansion there (as in (a)) in such a way that
Guess who? 35

the (presumably) necessary transformational rule of There Insertion is not duplicated


in essence by the necessary phrase structure rules. Also, number agreement will have
to be by an interpretive convention (see (a)), a theoretical choice which will entail
the unusual assertion that ∗ he are fat is only semantically anomalous.
In (b), some rules like those proposed in (a) above will have to be devised, with
provision being made for the complementizer to to follow the NP in (a). Of course,
no analysis containing such a revised phrase structure rule can explain why it is to that
appears in this rule, rather than some other morpheme, such as ing or Winnebago.
If there is a rule of VP Deletion, the impossibility of (c) is an automatic conse-
quence of the strange fact, pointed out to me by Joseph Emonds, that in embedded
questions consisting of a questioned constituent, and an infinitive, all question words
are possible except why. 
⎧ ⎫
() ⎪ what ⎪

⎪ ⎪


⎪ ⎪


⎪ when ⎪

⎨ ⎬
where
I wondered to eat.

⎪ how long ⎪


⎪ ⎪


⎪ whether ⎪⎪


⎩∗ ⎭
why

If phrase structure rules generate how to directly, as they would seem to have to, if
(b) is to be produced, they must be constrained so as not to generate why to. Any
such constraint will duplicate the one which excludes the ungrammatical sentence
in ().
The most telling argument for a rule of VP Deletion has to do with the impossibility
of contraction evidenced in (d). Harold King has pointed out  that the general fact
about contraction in English is that the auxiliary verbs can never contract when their
object has been deleted, unless the auxiliary is later inverted. Some examples of this
restriction are shown in ().

is
() a. Tall though Lew ∗ , we can beat the Bears.
’s

is
b. I am taller than Lew ∗ .
’s

is
c. I wonder how much wine there ∗ .
’s

am
d. Ready I ∗ to help you.
’m

 And, of course, its underlying source for what reason/purpose etc.


 See King (). The consequences of King’s observation for the theory of grammar are elaborated
in Lakoff ().
36 John Robert Ross

The generalization linking the sentences in () to (d) is clear: In all cases, it is a
transformational rule of deletion that makes subsequent contraction impossible. This
fact, in conjunction with those having to do with (a–c), indicates strongly that there
is a rule of VP Deletion.
Thus whether it is argued that the sentence in () is to be derived via Sluicing, or
via VP Deletion and To Deletion, it is obvious that the theory of grammar must con-
tain some definition of sloppy identity, for either derivational route for () involves
deletion under identity. But equally obviously, the deleted phrase must be to crane my
neck, because of the impossibility of ().
() ∗ I don’t know how to crane his neck.

The only way I can see to avoid the argument for sloppy identity provided by () is
to claim that the relevant VP in () is crane neck until some very late, post-deletion,
stage of the derivation, thus necessitating no theory of sloppy identity. But even this
apparent way out can be closed off by a consideration of such sentences as ().
() I know how to say I’m sorry, and Bill knows how, too.
If the second clause of () has the meaning of (), and for the present, let us
disregard the possibility of any other meanings, if any in fact exist,
() Bill knows how to say he’s sorry.
then it is obviously unlikely that the subject of be sorry does not appear in deep
structure, being filled in only later.
To summarize, my conclusions from the above are:
() a. There is no rule of To Deletion.
b. There is a rule of VP Deletion.
c. The theory of grammar must contain some definition of the notion of sloppy
identity.
d. No evidence exists that such interpretive rules for pro-forms as those pro-
posed by Akmajian are necessary, above and beyond the definition of sloppy
identity. Therefore no such rules exist.

..
Consider next such sentences as those in ().

who
() Bill seems to be doing away with someone, but he won’t say ∗ with whom .

The impossibility of with whom in () clearly indicates that Sluicing is at work here
(recall the argument in Section ..). But if Sluicing is at work, then something must
be deleting under identity. Semantically, it is clear that the source for the second clause
of () must be (a), not (b).
Guess who? 37

() a. Bill won’t say who he is doing away with.


b. ∗ Bill won’t say who he seems to be doing away with.
But this means that the tensed VP is doing away with is deleting under identity with
the infinitival VP to be doing away with.  This suggests that these phrases should
be absolutely identical at some point in the derivation, unless sloppy identity is to be
defined in such a way that some grammatical morphemes, like the infinitive marker
to, can also be disregarded in checking for identity. In either case, it suggests a closer
relationship between non-finite clauses and finite clauses than has previously been
demonstrable.
A similar close relationship between derived nominals (see Chomsky () for dis-
cussion of this term) and finite clauses can be adduced, on the basis of such sentences
as those in ().
() Bill mentioned his plans to do away with someone, but he didn’t mention

who  .
∗ with whom

 In addition, of course, it suggests that the rule variously called Pronoun Replacement (Rosenbaum
), It Replacement, Subject Raising, etc. must be formulated as a copying rule, not as a chopping rule, as
previously believed, because it is the entire clause he is doing away with in (a) that must be deleted under
identity with a previous clause. That is, the derived constituent structure of the first clause of (), at the
time Sluicing applies, must be as shown below.

(i) S1

NP VP

Bill V NP

seems S2

for NP to VP
+ Doom
be doing away with someone
Bill
The feature [+Doom], first proposed by Postal (), will cause the subject of S to delete at some later
point. The advisability of making Subject Raising a copying rule, where the original NP is doomed, is
suggested by such sentences as:

(ii) John is said to be smart, but I don’t believe it.


 This phrase may be grammatical if construed as a modifier of plan, a reading which I will disre-
gard here.
38 John Robert Ross

Clearly, Sluicing is involved here, and equally clearly, the source for the second clause
of () must be ().
() Bill didn’t mention who he plans to do away with.
Thus Sluicing provides evidence against any theory of grammar in which his plans
to do away with someone is not identical, at some stage, to he plans to do away with
someone.

..
The most interesting consequence of Sluicing has to do with the sentences in ().
() a. ∗ Irv and someone were dancing together, but I don’t know who Irv and were
dancing together.
b. ??Irv and someone were dancing together, but I don’t know who.
(a) is ungrammatical because the Coordinate Structure Constraint (see Ross (:
Section .) for some discussion of this constraint) has been violated: A conjunct has
been questioned. The interesting sentence is (b). Most speakers find this sentence
ungrammatical to some extent, and in the most interesting dialects, it is perceived to
be less ungrammatical than (a).
What this means is the following: The constraints on chopping rules proposed in
Ross () cannot be locally defined—rather, they must be derivational constraints,
in the sense of Lakoff (). That is, whether or not Sluicing, an optional rule
which applies after Question Formation (see Section . above), applies to a string
in which the Coordinate Structure Constraint has been violated, affects the degree of
ungrammaticality of the string. The sequence of rules Question Formation—Sluicing
produces a less deviant sentence than Question Formation alone does, assuming, in
each case, that Question Formation has applied in some way that violates one of the
constraints on chopping rules. That other constraints are affected can be seen from
the sentences in (–).
() The Complex NP Constraint (Ross : Section .)
a. ∗ She kissed a man who bit one of my friends, but Tom doesn’t realize which
one of my friends she kissed a man who bit.
b. ?She kissed a man who bit one of my friends, but Tom doesn’t realize which
one of my friends. 
c. I believe the
 claim that he bit someone, but they don’t know who I believe
∗ the claim that he bit.

d. I believe (??the claim) that he bit someone, but they don’t know who.
() The Sentential Subject Constraint (Ross : Section .)
a. It is possible that he’ll hire someone, but I won’t divulge who (it is possible
that he’ll hire).
Guess who? 39

b. That he’ll hire someone is possible, but I won’t divulge


who that he’ll hire is possible


.
??who
() The Left Branch Condition (Ross : Section .)
a. ∗ I know that he must be proud of it, but I don’t know how he must be proud
of it.
b. ∗ I know that he must be proud of it, but I don’t know how. 
The facts of (–) provide additional evidence for there being a rule of Sluicing
involved in the derivation of sentences like (), for it is not obvious how an interpretive
theory can make use of the constraints on variables in excluding the ungrammatical
sentences in (b), (b), (d), (b), and (b). In addition, however, they provide
evidence of the strongest kind that the theoretical power of derivational constraints
is needed in linguistic theory, for it appears that the way the constraints proposed in
Ross () should be restated is, informally, something like ().
() If a node is moved out of its island,  an ungrammatical sentence will result.
If the island-forming node does not appear in surface structure, violations of
lesser severity will (in general) ensue.
This statement makes it clear that the degree of ungrammaticality which is atten-
dant on the violation of a constraint on variables cannot be ascertained except by
inspection of the whole subsequent derivation. In short, ungrammaticality is a prop-
erty not of merely deep or surface structures, or of pairs of trees which are related by
rules, but rather of derivations.

..
In summary, I have argued in Section . that there must be a transformational rule of
deletion, Sluicing, involved in the derivation of such sentences as (). In Section ., I
presented evidence that Sluicing must follow Question Formation, and in Section .
I have tried to show that if Sluicing is a rule, then transformations must be able to
apply across sentence boundaries, and they must be able to delete variables under
identity. Furthermore, sloppy identity must be defined in the theory of grammar, and
finite clauses appear to be related more closely to various non-finite constructions
than has been previously argued for. Finally, all constraints on chopping rules appear
to be derivational constraints.

 The fact that Johnny stole someone’s wallet, but I forget whose (wallet) is fully grammatical with or
without wallet suggests to me not that the Left Branch Condition is not operative in this sentence, but rather
that wallet has become one by a general rule of pronominalization, and that the resulting ungrammatical
sequence ∗ whose one is then obligatorily converted to whose.
 See Ross (: Section .) for a definition of this term.

How do you sluice when there


is more than one CP?
J EROEN VA N CR AE N E N B ROE CK

. Introduction
As is well known, sluicing refers to the phenomenon whereby an interrogative clause
is reduced to a mere wh-phrase. Consider a basic example in ().
() Ed saw someone, but I don’t know who.
In this sentence, the IP part of the embedded clause who Ed saw is deleted, leaving
only the wh-phrase. Schematically, the structure of this clause can be represented as
in () (see Merchant : ).
() CP

who C´
[+wh]
C0 IP
[+Q]

In this partial derivation, the wh-phrase who first moves to the specifier position of the
embedded CP, and at PF, the phonological content of the IP is deleted. (See Merchant
(, to appear b), van Craenenbroeck and Merchant (to appear) for ample argumen-
tation in favor of a PF-deletion analysis of sluicing.) What this abstract representation
does not take into account, however, is that what was traditionally conceived of as a
single projection, i.e. CP, might in fact constitute a conglomerate of more than one
functional projection (see in particular Rizzi () and much literature in its wake).
Needless to say, such a view would necessitate a further refinement of the analysis
schematically represented in (). For example, if wh-movement targets a low CP-
projection, material that surfaces in higher CP-layers (either through movement or
as the result of base generation) should in principle be able to survive the deletion
process that is part of sluicing. On the other hand, if the wh-phrase surfaces in a high
Sluicing with more than one CP 41

CP-layer, it might be a low CP-projection that is deleted in sluicing, rather than IP.
In order to make these issues more concrete, consider the simplified and partial tree
structures in (). In these representations, the unitary CP of the analysis in () has
been split up into two separate projections (neutrally labeled CP and CP here), and
the combination of the two parameters mentioned above (the position targeted by
the wh-phrase on the one hand and the projection undergoing deletion on the other)
yields four logical possibilities.

() a. wh in SpecCP1, deletion of CP2 b. wh in SpecCP1, deletion of IP


CP1 CP1

wh wh
Þ ELLIPSIS
CP2 CP2
Þ ELLIPSIS
IP IP

c. wh in SpecCP2, deletion of CP2 d. wh in SpecCP2, deletion of IP


CP1 CP1

Þ ELLIPSIS
CP2 CP2
wh wh Þ ELLIPSIS
IP IP

Arguably, the structure represented in (c) is not a well-formed instance of sluicing.


Here, the wh-phrase is contained inside the projection that is deleted (i.e. CP ), and as
a result it does not show up overtly. Given that the presence of an overt remnant (wh
or otherwise; see van Craenenbroeck and Lipták (, , to appear)) is a defining
characteristic of sluicing, this structure is not a licit instantiation of this construction.
Moreover, under the uncontroversial assumption that ellipsis sites cannot contain any
focus-marked material (Merchant : Chapter ), this structure is illicit regardless
of whether it represents an instance of sluicing or not. Given that a sluiced wh-phrase
is invariably focus-marked (Hartman ), it cannot be part of a structure that is
elided, and the partial derivation depicted in (c) should crash. This leaves the three
options in (a), (b), and (d) as possible structural representations of sluicing in a
split CP.
As discussed by van Craenenbroeck and Lipták (, , to appear), the con-
figuration in (d) is attested in many languages. As a way of illustrating this, consider
the Hungarian example in () (Merchant : –).
42 Jeroen van Craenenbroeck

() A gyerekek találkoztak valakive de nem emlékszem, hogy


the children met someone.with but not I.remember that
kivel.
who.with
‘The children met someone, but I don’t know who.’ [Hungarian]
It is well established that in Hungarian, wh-movement targets a low CP-projection,
traditionally referred to as FocP (see for example Lipták () and references cited
there). Given that the CP-layer hosting the declarative complementizer hogy ‘that’
dominates FocP, it is not contained in the ellipsis site and should be able to survive
deletion. The presence of hogy to the left of the sluiced wh-phrase in () shows that this
is indeed the case. Moreover, given that topicalization targets a position in between
hogy and FocP in Hungarian, it too should be immune to sluicing. The example in
() (van Craenenbroeck and Lipták to appear: –) shows that this prediction is
borne out. 
() ?Tudom, hogy a diákok es a tanárok is meghívtak valakit,
know-sg that the students and the teachers also invited someone
de nem tudom, hogy a diákok kit.
but not know-sg that the students whom
‘I know that the students and the teachers each invited someone, but I don’t know
who the students invited.’ [Hungarian]
What () and () illustrate, then, is that the configuration in (d) represents one pos-
sible form the interaction between the syntax of sluicing and the split-CP hypothesis
can take: When the wh-phrase targets a low CP-projection and IP is deleted, the higher
left peripheral layers remain available and can contain overt material. In this paper I
focus on the other two configurations the taxonomy in () yields, i.e. the structures in
(a) and (b). I show that even if a moved wh-phrase ends up in the highest available
CP-layer, there is still variation as to which part of the clausal structure undergoes
deletion, i.e. IP in some cases and a low CP-projection in others.
This paper is organized as follows. In the next section I introduce and defend a
particular instantiation of the split-CP hypothesis. I show that—at least in Dutch,
German, Frisian, and English—there are good reasons to assume that the derivation
of a wh-question involving a complex wh-phrase like which boy differs substantially
from one featuring a simple wh-phrase such as who or what. In Section ., I combine
this view on the CP-domain with the syntax of sluicing, and argue that the interaction
between the two leads to the conclusion that sluicing does not always delete the same
 In this example several variables have to be controlled for. On the one hand, the fronted element cannot
represent new information, for then it would be a focus and hence in complementary distribution with the
wh-phrase in SpecFocP. On the other hand, it cannot be entirely old information either, because then it
would undergo deletion with the rest of IP. This delicate balance probably explains the mild deviance of the
example in ().
Sluicing with more than one CP 43

part of the clausal structure. In Section ., I discuss two subtypes of sluicing that cor-
roborate this view. Specifically, I focus on the English construction Merchant ()
has dubbed swiping, and on the phenomenon found in a number of Dutch dialects
and Frisian whereby a sluiced wh-phrase is followed by a demonstrative pronoun
(Hoekstra , van Craenenbroeck ). Section . sums up and concludes.

. Splitting up CP: Simple versus complex wh-phrases


This section is organized as follows. In Section .. I outline my version of the split-
CP hypothesis and point out to what extent it is related to or compatible with existing
accounts. Section .. further explores various aspects of the analysis and shows that
the proposal is both well grounded in the generative research tradition and supported
by a variety of data. Section .. sums up and briefly discusses two remaining issues.

.. Outlining the proposal


The crux of my analysis is that it assigns a different CP-domain-internal syntax to
simple wh-phrases like who, what, and why (and PPs containing them) on the one
hand, and complex wh-phrases such as which boy, whose mother, and what kind of
pasta on the other. Schematically, the differences can be represented as in ().

() a. simple wh b. complex wh


CP1 CP1

whi C1⬘ whi C1⬘

C01 CP2 C01 CP2


[+Q] [+Q]
ti C2⬘ Opi C2⬘

C20 IP C2 IP
[+Op] [+Op]
…ti… …ti…

These partial tree structures are to be interpreted as follows. I assume that the CP-
domain should be split up into (at least) two separate functional projections (which
I will continue to label CP and CP for simplicity’s sake). The higher CP-layer is
the one responsible for clause typing (in the sense of Cheng ()), while the lower
one marks the position where operator/variable-dependencies are created (i.e. where
operator features are checked). This distinction, I want to argue, has considerable
consequences for the syntax of wh-movement. Specifically, while simple wh-phrases
44 Jeroen van Craenenbroeck

move from their IP-internal base position through SpecCP (where they check an
operator feature) onto SpecCP (to check a clause typing feature), complex wh-phrases
are base-generated in SpecCP (and check their clause typing feature in situ), while
an empty operator moves from the IP-internal base position to SpecCP (to check the
operator feature).
The proposal just sketched bears some resemblance to existing accounts of the
CP-domain and/or the syntax of wh-movement. Most notably, it fits into the line of
research exemplified by Munaro (), Poletto and Pollock (), Zanuttini and
Portner (), and others, which argues that there is a correlation between the inter-
nal complexity of a wh-phrase and its left-peripheral syntax, i.e. simple and complex
wh-phrases target different positions in the left periphery. Moreover, the idea that
(a particular group of) wh-phrases can be base-generated in SpecCP is one that can
be found in the literature on Irish (see in particular McCloskey (, )) and on
wh-adverbials (Reinhart , Rizzi , Culicover ). Finally, the combination
of base-generating a wh-phrase in a left-peripheral position and concomitant empty
operator movement is reminiscent of languages that employ a cleft strategy to form
wh-questions (see Cheng () for discussion). In spite of these parallelisms, how-
ever, it is clear that the specific proposal in () differs substantially from the accounts
mentioned. For example, although base-generating a wh-phrase in the left periphery
is not a new idea, no one has to my knowledge proposed that only complex wh-phrases
make use of this strategy. Similarly, while the different left-peripheral syntax of simple
and complex wh-phrases is not a new topic, it has so far not been applied to the set
of languages under discussion here (Dutch, Frisian, German, and English). Summing
up, it is clear that the current proposal as it stands is in need of some empirical support.
This is what I turn to in the next section.

.. Backing up the proposal


In this section I review five sets of data, each of which provides empirical support for a
particular aspect of the analysis introduced in the previous section. These arguments
should have the combined effect of strengthening the proposal as a whole. All the
examples will be drawn from Germanic, more specifically from Dutch, German,
Frisian, and English.

... Two head positions Although most of the existing proposals for splitting up
CP concentrate on non-Germanic languages such as Italian or Hungarian, a language
like Dutch also contains clear indications that CP is not the unitary projection it
was once believed to be. As was pointed out by Hoekstra and Zwart (, )
and Bennis (, ), the fact that Dutch allows the interrogative complementizer
of ‘if ’ and the declarative complementizer dat ‘that’ to co-occur both in embedded
wh-questions and in embedded yes/no-questions suggests that there are two head
Sluicing with more than one CP 45

positions present in the left periphery in this language. Consider in this respect the
examples in ().
() a. Ik weet niet of dat Jan gaat komen.
I know not if that John goes come
‘I don’t know if John will come.’
b. Ik vraag me af wie of dat je zoekt.
I ask me prt who if that you seek
‘I wonder who you’re looking for.’ [colloquial Dutch]
Following Hoekstra and Zwart and Bennis, I assume that of ‘if ’ occupies the head
position of CP , while dat ‘that’ heads CP .  As the presence of more than one func-
tional head is still one of the most convincing arguments in favor of splitting up CP, the
data in () provide strong support for the proposal outlined in the previous section.
However, given that the Hoekstra/Zwart/Bennis analysis of () is not uncontested—
Sturm () suggests that the sequence of dat ‘if that’ represents one single morpho-
logically complex complementizer heading a single CP—I want to present an extra
argument in its favor. This will have the added effect of strengthening another aspect
of the analysis presented above, namely the claim that complex wh-phrases do not
leave a trace in SpecCP .
The data I want to discuss concern the fact (first noted by Hoekstra ()) that in
a small number of Dutch dialects the complementizer of ‘if ’ can precede rather than
follow the wh-phrase in an embedded wh-question. As shown in (), the dialect of
Strijen is one of them. 
() Ik weet niet of met wie Jan oan et proate was.
I know not if with who John on it talk was
‘I don’t know who John was talking to.’ [Strijen Dutch]
The prediction raised by such data for the lexical status of the string of dat ‘if that’ is
clear. If this sequence represents one single complementizer, wh-phrases should be
unable to occur in between of ‘if ’ and dat ‘that’ (given uncontroversial assumptions
about lexical integrity). If, on the other hand, of ‘if ’ and dat ‘that’ head different
projections, wh-phrases might in principle be able to land in the specifier position

 Note that my analysis diverges markedly from that of Hoekstra and Zwart when it comes to the
identification of these two CPs. They argue that the high CP-projection is targeted exclusively by wh-
phrases, while the lower one only hosts topicalised phrases. The Strijen Dutch data in () form a clear
counterexample to this claim. My account is more akin to that of Bennis. He claims that CP is the projection
specialized in clause typing. We disagree, however, when it comes to CP . Bennis suggests it is a projection
in which subordination is marked (and he calls it SubP), while I will argue that it is the projection where
operator features are checked.
 This phenomenon appears to be dying out in the dialects of Dutch: Of the four Strijen Dutch speakers
I consulted, only two allowed the complementizer to precede the wh-phrase (while all allowed it to follow),
and for the dialect of Amsterdam, where the phenomenon has also been reported, I was unable to find a
speaker allowing the pattern.
46 Jeroen van Craenenbroeck

of the projection headed by dat ‘that’, in which case they would appear in between the
two complementizers. As the example in () illustrates, it is the second of these two
predictions that is borne out.
() Ik weet niet of met wie dat Jan oan et proate was.
I know not if with who that John on it talk was
‘I don’t know who John was talking to.’ [Strijen Dutch]
That fact that met wie ‘with whom’ can occur in between of ‘if ’ and dat ‘that’ is a clear
indication that these two complementizers head different projections, pace Sturm
(). Moreover, the examples in () and () are relevant for another aspect of my
analysis as well. They seem to suggest that in the dialect of Strijen, wh-phrases have the
option of landing in the specifier position of CP without moving on to SpecCP .  As
such, these data make a very strong prediction with respect to complex wh-phrases.
If they are base-generated in SpecCP , they should be unable to be preceded by
of ‘if ’. Complex wh-phrases cannot optionally land in SpecCP because they never
move through that projection in the first place. As is shown in (), this prediction is
borne out.
() Ik vroag me af < ∗ of > welke jonge < of > die maisjes gistere
I ask me off if which boy if the girls yesterday
gezien hebbe.
seen have
‘I wonder which boy the girls saw yesterday.’ [Strijen Dutch]
Summing up, the Strijen Dutch data I have looked at in this section turn out to provide
strong support for two independent aspects of the analysis of the CP-domain intro-
duced in the previous section. On the one hand, the fact that wh-phrases can occur in
between of ‘if ’ and dat ‘that’ shows that of dat ‘if that’ is not a single, morphologically
complex complementizer. On the other hand, the absence of complex wh-phrases to
the right of of ‘if ’ suggests that they never land in SpecCP .

... The operator/non-operator status of wh-phrases The analysis in () suggests


that there is a substantial featural difference between simple and complex wh-phrases.
While the former move through SpecCP to check an operator feature, the latter are
base-generated in SpecCP and only check a clause-typing feature. In other words,
simple wh-phrases are syntactic operators, but complex ones are not. Interestingly,
precisely this conclusion was reached on entirely independent grounds in much syn-
tactic literature published in the eighties and early nineties. For example, much of the

 I will have nothing to say about what causes the variation between the dialect of Strijen and, say,
standard Dutch. For lack of a more insightful explanation, one could claim that the clause-typing feature
on C is optionally strong in this dialect. It is clear that more research is needed on this topic. See also van
Craenenbroeck (: –), who shows that Frisian is similar to Strijen Dutch in this respect.
Sluicing with more than one CP 47

data that was discussed in the context of English wh-in-situ (in multiple wh-questions)
is amenable to a similar analysis (Reinhart , , Pesetsky , Hornstein and
Weinberg , Guéron and May ). The most well-known piece of data in this
respect concerns the contrast in ().
() a. ∗ What did who buy?
b. What did which boy buy?
While simple wh-phrases are subject to Superiority, complex ones are not.  This
contrast receives a straightforward account under the assumption that simple wh-
phrases, unlike their complex counterparts, are syntactic operators. Specifically, in
(a) the operator feature on C has failed to attract the closest bearer of a matching
feature (i.e. who), and as a result the derivation is ruled out as a violation of Attract
Closest (or some other comparable locality principle). The complex wh-phrase which
boy in (b) on the other hand does not bear an operator feature (it is not a syntactic
operator), and so the derivation converges. Summing up, Superiority contrasts like
the one in () support the assumption that complex wh-phrases, unlike simple ones,
are not syntactic operators (see Reinhart () for similar reasoning based on weak
crossover violations at LF). 
Another set of data that leads to this conclusion comes from contrastive left dislo-
cation (CLD) in Dutch. Consider the examples in ().
() a. Die jongensi , diei ken ik niet t die .
those boys dem know I not
‘Those boys, I don’t know.’
b. ∗ Iedereeni , diei ken ik niet t die .
everybody dem know I not [Dutch]
The sentence in (a) represents a typical instantiation of CLD: A phrase (in this case
the DP die jongens ‘those boys’) is merged in the left periphery of the clause, while
a coreferential demonstrative pronoun moves from the IP-internal base position to
the V-position of the clause (see Hoekstra and Zwart () and Hoekstra () for
arguments in favor of this analysis for Dutch and Frisian CLD, and see Grohmann
() for general discussion of CLD in Germanic). What the example in (b) illus-
trates is that the CLD-ed phrase cannot be a bare quantifier (see Cinque () for a

 Note that I am following Hornstein and Weinberg (), Reinhart (), Guéron and May (),
Aoun et al. (), and Aoun and Li () in assuming the relevant factor distinguishing (a) from (b)
to be syntactic complexity rather than d-linking (as in Pesetsky (, ), Comorovski (), and Dayal
()). See Aoun and Li (: Chapter ) and van Craenenbroeck (: –) for relevant discussion.
 In this short discussion I cannot do justice to the vast literature on Superiority, nor is it my ambition
to do so. Rather, what I wanted to show is that the basic facts in () follow naturally from the present
proposal under an Attract Closest account of Superiority (cf. Chomsky (: )). See Pesetsky (),
Dayal (), Fanselow (), Barss (), Aoun and Li (), and references cited there for further
discussion of Superiority.
48 Jeroen van Craenenbroeck

similar claim about Italian clitic left dislocation). Now consider from this perspective
the contrast in ().
() a. ??Welke jongens die ken je niet?
which boys dem know you not

b. Wie die ken je niet?
who dem know you not
Not surprisingly, when a wh-phrase occurs in the left-peripheral CLD position, the
result is less than perfect. A CLD-ed constituent should be fully referential, which
wh-phrases clearly are not. What is interesting, however, is that there is once again a
contrast between simple and complex wh-phrases. While the latter yield a relatively
deviant result, the former are categorically excluded from CLD. Given the view on
the CP-domain developed in this paper, the cause for the ungrammaticality of (b)
is the same as that in (b), i.e. CLD disallows bare quantifiers. The fact that there is
a contrast with complex wh-phrases is an indication that these do not function as a
syntactic operator.
Summing up, the operator/non-operator distinction between simple and complex
wh-phrases is both well-grounded in the generative research tradition and supported
by a variety of facts.

... Spelling out intermediate copies The proposal outlined in the previous sec-
tion postulates a difference in derivational history between simple and complex wh-
phrases. While the former are merged in the IP-internal base position and move
successive-cyclically to their ultimate left-peripheral landing site, the latter are base-
generated in that landing site and do not move at all. This implies that if there is
a construction in which the derivational history of a phrase can be made visible,
a contrast should emerge between the two types of wh-phrase. Consider from this
perspective the data in () (Nunes : , , n).
() a. Wen glaubt Hans wen Jakob gesehen hat?
who thinks Hans who Jakob seen has
‘Who does Hans think that Jakob saw?’
b. Mit wem glaubst du mit wem Hans spricht?
with whom think you with whom Hans speaks
‘Who do you think Hans is talking to?’
c. ∗ Wessen Buch glaubst du wessen Buch Hans liest?
which book think you which book Hans reads
[colloquial German]
These sentences exemplify the construction known as wh-copying. Although they
each represent a single wh-question, they contain more than one wh-phrase. Fol-
lowing a long research tradition, Nunes () proposes to analyze wh-copying as
Sluicing with more than one CP 49

involving movement chains in which more than one copy has been spelled out (see
also Hiemstra (), du Plessis (), McDaniel (), Höhle (), Fanselow and
Mahajan (), and Fanselow and Ćavar () for earlier, comparable accounts).
Thus, in the example in (a), the wh-phrase wen ‘who’ has undergone run-of-the-
mill successive-cyclic movement, but instead of spelling out only the highest copy
in this movement chain (as is standard), the PF-component chooses—for whatever
reason and through whatever mechanism—to also spell out the intermediate copy
in the SpecCP of the embedded clause. What is interesting from the present per-
spective, however, is that not all types of wh-phrase can partake in this construction.
Specifically, while simple wh-phrases (a) and PPs containing them (b) are allowed,
complex wh-phrases like wessen Buch ‘whose book’ (c) are systematically excluded.
This is precisely what one would expect from the point of view of the theory developed
here. Complex wh-phrases are base-generated in the left periphery of the clause. As
a result, they leave no intermediate copies, and their non-occurrence in wh-copying
follows straightforwardly.

... The empty operator As den Dikken () pointed out in his discussion of
operator movement in Dutch imperatives, there are various ways of detecting the pres-
ence of an empty operator in Dutch. One of them concerns preposition stranding. As
is well known, Dutch is what one could call a partial preposition stranding language,
in that prepositions can only be stranded in a very restricted set of contexts, i.e. when
the element stranding the preposition is either an empty operator or a so-called R-
pronoun (van Riemsdijk b). Consider two representative examples in ().

() a. Die sleutel is te klein [Opi om het slot mee t i open te maken.]
that key is too small for the lock with open to make
‘That key is too small to open the lock with.’
b. Waari heb jij dat slot mee t i open gemaakt?
where have you that lock with open made
‘What did you open that lock with?’ [Dutch]

The example in (a) is an instance of tough-movement (one of the prototypical


empty operator constructions), while in the question in (b), the R-pronoun waar
‘where’ has been wh-moved to SpecCP. In both cases, a preposition has successfully
been stranded by the moving element, thus illustrating the generalization presented
above. A fact that has gone unnoticed so far, however, is that there is a contrast
between simple and complex wh-phrases when it comes to preposition stranding. It
is illustrated in ().

() a. ∗ Wie wil je niet mee samenwerken?


who want you not with cooperate
‘Who won’t you cooperate with?’
50 Jeroen van Craenenbroeck

b. ?Welke jongen wil je niet mee samenwerken?


which boy want you not with cooperate
‘Which boy won’t you cooperate with?’ [Dutch]
While simple wh-phrases clearly cannot strand a preposition, their complex coun-
terparts fare much better, in many cases even yielding a fully grammatical result.  , 
In light of these data, one could argue that complex wh-phrases are a third type of
element—besides empty operators and R-pronouns—that can strand a preposition in
Dutch, but it would be difficult to find a common characteristic that distinguishes
empty operators, R-pronouns, and complex wh-phrases on the one hand from simple
non-R wh-phrases on the other. However, if the derivation of wh-questions with
complex wh-phrases involves empty operator movement, the contrast in () is pre-
cisely what one would expect. This conclusion is further corroborated by the contrast
in ().
() a. ∗ Met wie wil je niet mee samenwerken?
with who want you not with cooperate
‘Who won’t you cooperate with?’
b. ?Met welke jongen wil je niet mee samenwerken?
with which boy want you not with cooperate
‘Which boy won’t you cooperate with?’ [Dutch]
Not only can complex wh-phrases strand a preposition, but they can also be merged
as a PP in spite of the fact that the IP-internal gap is nominal. Once again, the theory
developed here allows for a straightforward explanation. Given that it is an empty
operator that strands the preposition, the left-peripheral wh-phrase is not required
to be of the same categorial type as the IP-internal gap. The fact that this option is
not available in (a) suggests that simple wh-phrases do not make use of this empty
operator strategy.

... Truncating the CP-domain The final aspect of my analysis for which I want
to present supporting evidence concerns the CP-domain-internal hierarchical differ-
ence between simple and complex wh-phrases. While the latter invariably occupy the
highest specifier available, the former first target the lower of the two CP-layers. This

 There is some variability in the judgments here. Not all speakers find examples like (b) fully accept-
able. All the speakers I have consulted share the intuition that there is a contrast between (a) and (b),
though. I suspect normative judgments are blurring the picture somewhat. See also Merchant (: –,
n) for some remarks concerning idiolectal variation with respect to preposition stranding in Dutch.
 Interestingly, Takami () reports similar contrasts even for a fully-fledged preposition-stranding
language like English. In contexts where preposition stranding is traditionally argued to be less felicitous
(e.g. from adjunct PPs), complex wh-phrases are more acceptable than minimal ones. Consider these two
examples (Takami : ):
() a. ??What did you feel dizzy after?
b. Which brand of cigarette did you feel dizzy after?
Sluicing with more than one CP 51

predicts that if CP is missing, complex wh-phrases can no longer occur, while simple
ones might. Moreover, given that of ‘if ’ occupies the head position of CP , the absence
of complex wh-phrases should correlate with the absence of this complementizer. I
want to propose that (one type of) free relatives instantiate precisely this hypothetical
truncation scenario. Consider the examples in ().
() a. Wat op tafel ligt is voor jou.
what on table lies is for you
‘What lies on the table is for you.’
b. ∗ Welk boek op tafel ligt is voor jou.
which book on table lies is for you
c. Wat dat op tafel ligt is voor jou.
what that on table lies is for you
‘What lies on the table is for you.’
d. ∗ Wat of op tafel ligt is voor jou.
what if on table lies is for you [Dutch]
These sentences exemplify the so-called definite usage of free relatives, i.e. those free
relatives that can be paraphrased by means of a definite description (see Grosu and
Landman () for discussion). The contrast between (a) and (b) shows that in
this type of free relative, complex wh-phrases are disallowed, while simple ones can
freely occur. (see Groos and van Riemsdijk (: –) for the original observation
and similar data from German, and see Grosu () for more general discussion
of these “anti-pied-piping effects”.) Moreover, as the minimal pair in (c–d) shows,
while the complementizer dat ‘that’ can occur in this construction, its interrogative
counterpart of ‘if ’ cannot (see also Hoekstra () for similar remarks about Frisian).
As was pointed out above, these judgments are exactly what one would expect from
the point of view of the theory developed in this paper. Complex wh-phrases and the
interrogative complementizer pattern together because both of them only occur in
CP . Hence, when this projection goes missing, they cannot show up.

.. Summary and two outstanding issues


In Section .. I discussed a variety of data in support of the proposal made earlier,
and showed that the various assumptions making up that proposal can be backed
up by independent evidence. Before returning to sluicing, I want to briefly highlight
two issues that were left unresolved by my treatment of the CP-domain and wh-
movement.
The first concerns the precise dividing line between simple and complex wh-
phrases and its relation to operatorhood. In particular, what determines whether a
wh-phrase belongs to one or the other category, and is there any inherent reason
why phrases in the first category are syntactic operators, while those in the second
one are not? I want to suggest that it is the presence of a nominal restriction (or N-
restriction for short) that makes complex wh-phrases complex. A DP such as which
52 Jeroen van Craenenbroeck

book is a prototypical instance of a complex wh-phrase (book being the N-restriction),


while a wh-adverb like why is on the opposite end of the complexity scale. Using
the presence of an N-restriction as the defining criterion for complexity has two
advantages. First of all, it assigns an intermediate status to the wh-elements who and
what. On the one hand, they can be treated as bare wh-pronouns that have no N-
restriction, while on the other, they can also be analyzed as involving an implicit
restriction (say, ‘person’ in the case of who and ‘thing’ in the case of what). This double
analysis is reminiscent of the distinction between “bare” personal pronouns, such as
us, and pronouns that are combined with a nominal, such as us linguists (see Noguchi
(: –) for discussion). This is a desirable result, because it is well known that
who and what differ from wh-adverbs like how and why in that they sometimes display
non-operator behavior (Superiority being the prime example; see in this respect also
note  above). If the presence of an N-restriction is what determines how complex
a wh-phrase is, the dual behavior of these wh-pronouns follows naturally. A second
advantage of this approach concerns the definition of operatorhood. Wh-phrases that
have an N-restriction denote a set and can therefore be interpreted in situ (e.g. through
choice functions or as the result of unselective binding). Assuming that such an in situ
strategy is more economical than movement (see Tsai () for an explicit implemen-
tation of this assumption), the operator/non-operator status of simple versus complex
wh-phrases follows naturally: Only complex wh-phrases denote a set, and as a result
only complex wh-phrases can—and therefore, by Economy, must—be interpreted as
a non-operator.  Although the finer details of this proposal need to be worked out
further, using the presence of an N-restriction as the defining criterion for complexity
seems to yield promising results.
A second issue that requires some attention concerns reconstruction. If complex
wh-phrases are base-generated in the left periphery of the clause, one would expect
them to be unable to reconstruct into the IP-internal base position, contrary to fact.
There are two possible ways of dealing with this discrepancy. One would be to slightly
modify the proposal, and to have complex wh-phrases move from the IP-internal base
position to SpecCP in one fell swoop, i.e. without having them stop over in SpecCP .
Seeing as they are non-operators, they cannot check the operator feature on C  , and
as a result, there is no reason for them to move to (or through) this projection. In
such a scenario, an empty operator would have to be merged in SpecCP to check
the operator feature of C  . The problem with this alternative is that it is not straight-
forwardly compatible with some of the evidence presented in Section .. (e.g. the
preposition-stranding data and the facts concerning wh-copying). An alternative way

 The formulation here seems to suggest that languages like English and Dutch should allow complex
wh-phrases to stay in situ in simple wh-questions, which is not the case. Note, however, that apart from the
fact that an operator/variable-dependency needs to be formed, wh-questions are also subject to a clause-
typing requirement. In a wh-in-situ language like Chinese, there are clause-typing particles to take care of
this (Cheng ), but in Dutch and English, this requirement forces the wh-phrase to occur clause-initially.
Sluicing with more than one CP 53

of approaching the reconstruction problem would be to give up the assumption that


all cases of reconstruction have to be analyzed as the result of activating a lower copy
in a movement chain. In particular, there is a list of constructions (clefts, pseudo-
clefts, Italian CLLD, Dutch CLD, etc.) that clearly display connectivity effects, but
seem to resist an analysis in terms of syntactic movement. It might well be that for
these constructions one has to resort to alternative (e.g. semantic) reconstruction
mechanisms. If that turns out to be the case, wh-questions with complex wh-phrases
would constitute another construction that needs to be added to this list. 
All in all, then, it seems fair to say that the proposal outlined in Section .. is not
only well supported by a variety of data from a variety of languages, but the problems
it encounters and the questions it raises also receive a fairly natural account. In the
next section I turn to the interaction between this proposal and the syntax of sluicing.

. The split CP-domain meets sluicing


This section examines the interaction between the split CP-system outlined above and
the syntax of sluicing. Merchant (: –, ) proposes that the deletion process
characteristic of sluicing should be implemented by means of a syntactic feature,
which he dubs [E]. This feature is merged with the C -head whose complement is to
be elided, and it combines the syntactic, semantic, and phonological effect of ellipsis.
Put differently, the [E] feature represents all the relevant information distinguishing
elliptical—in this case sluiced—clauses from their non-elliptical counterparts. Con-
sider the representations in () (Merchant : –).
() a. the syntax of [E]: E[uwh∗ , uQ∗ ]
b. the phonology of [E]: φIP → ∅/E___
c. the semantics of [E]: E = λ p: e-given(p) [p]
These formulas indicate what the syntactic, phonological, and semantic contribution
of the [E] feature is to the derivation. For example, the rule in (b) states that the
phonological representation of IP is reduced to null (i.e. elided) when it is in the
complement position of (a head bearing) the [E] feature (though see Gengel () for
a different approach). The representation in (c) contains the recoverability require-
ment on ellipsis: The semantic composition of a clause containing the [E] feature can
only proceed (i.e. a clause can only be elided) when there is a suitable antecedent
available (for in-depth discussion of the notion of e-givenness, see Merchant (:
Chapter )). More relevant from the present perspective, however, are the syntactic
licensing requirements on [E] represented in (a). As was pointed out by Lobeck

 As an aside, it is worth pointing out that several of the constructions mentioned in the main text
are assumed to involve operator movement, i.e. it might be the case that moving an operator from the
base position into the left periphery is as effective a reconstruction strategy as activating a lower copy in a
movement chain. I leave this issue as a topic for further research.
54 Jeroen van Craenenbroeck

(), only the null C of constituent questions allows its complement to be elided
by sluicing. Merchant’s theory captures this observation by assuming that the [E]
feature is itself endowed with [+wh, +Q] features. Moreover, these features are unin-
terpretable (and hence in need of checking) and strong (indicated by the star in (a)),
which implies that they have to be checked in a local relationship, not as the result of a
long-distance agreement mechanism such as Agree (though see Aelbrecht () for
an implementation in terms of Agree). This ensures that [E] is only syntactically licit
when merged with the null C of constituent questions, which in turn implies that
only the complement of this type of C can be sluiced.
As was already pointed out in Section ., Merchant presents his analysis from the
point of view of a single, unsplit CP. Specifically, the [+Q] feature and the [+wh]
feature are situated on one and the same head. The previous section has argued for a
more refined view, in which the clause-typing feature (i.e. [+Q]) is checked in CP ,
while the operator feature (which I call [+Op]) is situated on C  . In order to see what
consequences this has for the syntactic licensing requirements of the [E] feature, I will
now go through an abstract sample derivation of a wh-question. Assume that C  has
just been merged on top of IP, and that it is marked with the [E] feature. In this local
configuration, the operator feature of [E] can be checked against that of C  . This is
shown in ().
() C2⬘

C02 IP
[+Op]
[E] …
+Op,+Q

Next, C  is merged (abstracting away momentarily from possible phrasal movement


to SpecCP , but see below). It attracts [E], which can then check its [+Q] feature. This
is represented in ().
() C1⬘
CP2
C10
[+Q] … C2⬘
[E]
+Op,+Q C02 IP
[+Op] …
[E]
+Op,+Q

At this point in the derivation, the [E] feature is syntactically fully licensed. This means
that it is now in a position from which it can trigger deletion. Given that the [E]
feature always deletes the complement of the head on which it resides, the abstract
Sluicing with more than one CP 55

derivation depicted in (–) leads to the conclusion that it is the lower CP-layer,
rather than IP, that is deleted when sluicing targets a split CP-system. As it stands,
however, this conclusion is too general, as we have not yet taken into account the fact
that the derivation might also involve phrasal movement to SpecCP . The remainder
of this section shows that this movement in some cases blocks the [E] feature from
moving to C  , causing sluicing to delete IP instead of CP .
The first type of derivation we need to consider is one involving a complex wh-
phrase. Recall that complex wh-phrases are base-generated in SpecCP and involve
empty operator movement to SpecCP . This latter step is represented in ().

() CP2

Opi C⬘2
[+Op]

C02 IP
[+Op]

[E] ... ti ...


+Op,+Q

At this point, the [E] feature is still not fully licensed (its [+Q] feature has not been
checked). This means that it has to move to C  , just as in the abstract representation in
(). Subsequently, the complex wh-phrase is merged in SpecCP , and the derivation
of the sluiced clause is complete: The wh-phrase resides in SpecCP , and the [E] feature
triggers deletion of CP . This is shown in ().

() CP 1

whi C1⬘
[+Q]
Þ ELLIPSIS
0
C1 CP2
[+Q]

[E ] Opi C⬘2
+Op,+Q [+Op]

C02 IP
[+Op]

[E ] … ti ...
+Op,+Q

Now consider a derivation involving a simple wh-phrase. Recall that in this case, there
is no empty operator. It is the wh-phrase itself which moves from the IP-internal base
position through SpecCP onto SpecCP . At the level of CP , this yields the following
representation.
56 Jeroen van Craenenbroeck

() CP 2

whi C2 ⬘
[+Op, +Q]
0
C2 IP
[+Op]
[E ] ... ti ...
+Op,+Q

Note that in this structure there is an alternative way for the [E] feature to become
syntactically fully licensed. Unlike the empty operator in the structure in (), the
simple wh-phrase is endowed not only with an operator feature, but also with a
clause-typing feature. This means that [E] can now check its [+Q] feature in a local
(spec/head-)relation against the [+Q] feature of the wh-phrase. As a result, it no
longer needs to move to C  in order to be licensed, and it triggers deletion of IP.
After the wh-phrase has moved on to SpecCP , the derivation can be represented as
in ().
() CP 1

whi C 1⬘
[+Op,+Q]
0
C1 CP2
[+Q]

ti C2 ⬘
Þ ELLIPSIS
C02 IP
[+Op]

[E ] … ti …
+Op,+Q

Summing up, the derivational difference introduced in the previous section between
simple and complex wh-phrases turns out to have considerable repercussions for the
syntax of sluicing in that it is the internal complexity of the sluiced wh-phrase that
determines which part of the clausal structure is elided: the lower CP-layer in the
case of complex wh-phrases and IP in the case of simple wh-phrases. This conclusion
follows straightforwardly from the assumption that the feature responsible for ellipsis
triggers deletion as soon as it is syntactically fully licensed. In the next section I explore
some of the empirical consequences of this finding.

. Corroborating evidence: Stranding to the right of a sluiced wh-phrase


The previous section has argued that sluicing with complex wh-phrases deletes CP ,
while in sluicing with simple wh-phrases IP is elided. This predicts that material that
Sluicing with more than one CP 57

surfaces in SpecCP should be able to survive sluicing when the sluiced wh-phrase is
simple, but not when it is complex. In this section I show that this prediction is borne
out. I discuss two subtypes of sluicing in which overt material occurs to the right of a
sluiced, simple wh-phrase. In the first case a preposition has been stranded in SpecCP
by the moving wh-phrase, while in the second case a demonstrative pronoun moves
independently of the wh-phrase to this specifier position. The fact that both these
constructions are disallowed in sluicing with complex wh-phrases will provide strong
support for the theory argued for in this paper.
This section is organized as follows. In Section .. I focus on the construction
dubbed swiping by Merchant (). Swiping receives a very straightforward and
natural analysis under the assumptions outlined above. In Section .. I turn to
a construction attested in various dialects of Dutch and in Frisian and show that
it involves focus movement of a demonstrative pronoun from an underlying cleft
construction into SpecCP . Section .. sums up and concludes.

.. Swiping
“Swiping” is an acronym that stands for sluiced wh-word inversion with prepositions
in Northern Germanic (Merchant : ) (see also Ross (), Rosen (), van
Riemsdijk (a), Lobeck (), Chung et al. (), Kim (), Culicover (),
Richards (, ), Culicover and Jackendoff (), Hasegawa (), Hartman
and Ai (), Hartman (), van Craenenbroeck (), and Larson ()).  An
example is given in ().

() Peter went to the movies, but I don’t know who with.

In this sentence, the normal order of preposition and wh-phrase (in which the former
precedes the latter: with who) has been reversed. As a result, it looks like the prepo-
sition has been stranded to the right of the sluiced wh-phrase. Interestingly, not all
wh-phrases can partake in swiping. Consider the contrast in ().

() a. Lois was talking, but I don’t know who to.


b. ∗ Lois was talking, but I don’t know which person to.

These examples show that while swiping is perfectly well-formed with simple wh-
phrases such as who or what, it is systematically excluded with complex ones like which
person. Under a traditional approach to wh-movement that treats simple and complex
wh-phrases alike, this distinction cannot be straightforwardly accommodated. From
the point of view of the theory developed in the preceding two sections, however,
the contrast in () is precisely what one would expect. In order to see why this is the
case, I now go through a step-by-step derivation of the swiped clause in (a), starting

 For reasons of space I do not provide a comparison between existing accounts of swiping and the one
developed in this section. See Merchant : – and van Craenenbroeck : –.
58 Jeroen van Craenenbroeck

from the point at which C  is merged into the structure. The relevant representation
is given in ().
() [C  C  [+Op, [E]+Op,+Q] [IP Lois was talking [PP to who[+Op, +Q] ]]]
Recall that C  is endowed with an operator feature targeting the (operator feature
of the) wh-phrase who. Moreover, given that this is the derivation of an elliptical
(i.e. sluiced) clause, the [E] feature also resides on C  (see the previous section for
discussion). This feature is itself marked [+Op, +Q], and given that it is in a local
relation with C  , its [+Op] feature can be checked in this configuration. The next
step in this derivation involves the movement of the wh-phrase who to SpecCP in
order to check the operator feature of C  . In so doing, who pied-pipes the preposition
to, so that the entire PP to who surfaces in SpecCP . This is illustrated in ().
() [CP [PP to who[+Op, +Q] ] [C  C  [+Op, [E]+Op,+Q] [IP Lois was talking
[PP to who[+Op, +Q] ]]]]
Recall that the wh-phrase in this configuration checks not only the operator feature
of C  , but also the [+Q] feature of [E]. This means that [E] has to move no further
in order to become syntactically fully licensed, and that it will later (i.e. at PF) trigger
deletion of IP (i.e. the complement of the head on which it resides).
Next, C  is merged. It bears a strong clause-typing feature (see ()), and it attracts
the wh-phrase who to its specifier, stranding the preposition to in SpecCP . This is
shown in ().
() [C  C  [+Q] [CP [PP to who[+Op, +Q] ] [C  C  [+Op, [E]+Op,+Q] [IP Lois was
talking [PP to who[+Op,+Q] ]]]]]
() [CP  who[+Op,+Q] [C   C  [+Q] [CP [PP to who[+Op, +Q] ] [C   C  [+Op,
[E]+Op,+Q] [IP Lois was talking [PP to who[+Op, +Q] ]]]]]]

Finally, at PF, [E] triggers deletion of IP, and the lower copy in the movement chain of
who is also deleted. This yields the representation in ().
() [CP who[+Op, +Q] [C   C  [+Q] [CP [PP to who[+Op, +Q] ] [C   C  [+Op,
[E]+Op,+Q] [IP Lois was talking [PP to who[+Op,+Q] ]]]]]]

What remains, then, is the wh-phrase who in SpecCP , together with the stranded
preposition to in SpecCP . In other words, the derivation just sketched has success-
fully yielded the swiped clause in (a). At the same time, it is also clear why the
example in (b) is not well-formed. In sluiced clauses involving a complex wh-
phrase, it is CP rather than IP that is deleted. This means that any material that occurs
in SpecCP is contained in the ellipsis site and hence cannot show up to the right of
the sluiced wh-phrase.
Summing up, the contrast in () can be straightforwardly accounted for under
the theory advocated in this paper. As such, swiping represents a first illustration of
Sluicing with more than one CP 59

the interaction between the split CP-system outlined in Section . and the syntax
of sluicing. Before moving on to the second subtype of sluicing, however, I want to
briefly highlight another aspect of my analysis. It concerns the movement operation
represented in (), where the wh-phrase who strands the preposition to in SpecCP
on its way to SpecCP . At first glance, this looks like a violation of the ban on preposi-
tion stranding in intermediate positions (Postal ). Consider an illustration of this
principle in ().
() ∗ Who do you think [PP for t i ]k she bought a present t k ?
i

In this sentence, the entire PP first moves to the embedded SpecCP, where the wh-
phrase who subsequently strands the preposition for on its way to the SpecCP of the
matrix clause. The fact that this example is not well-formed is taken by Postal ()
to be an indication that preposition stranding is not allowed in intermediate positions
of successive-cyclic A -movement. This suggests that the movement operation in
() should be as ungrammatical as the one in (). In both cases a preposition is
stranded in the intermediate step of a successive-cyclic wh-movement operation. I
want to suggest that the representation in () indeed violates the above-mentioned
principle, but that this violation is undone by the PF-deletion of IP. Assume that what
is wrong with the example in () is the fact that it contains a non-uniform chain at
PF. Specifically, the wh-movement chain of (for) who contains (at least) one DP-link
and two PP-links. PF cannot parse (i.e. linearize) such a non-uniform chain, and as
a result the derivation crashes. Now note what happens in the representation in ().
Here, the [E] feature ensures that the entire IP is deleted at PF. This means that all but
the highest PP-link in the wh-movement chain become invisible to PF. As a result, all
that remains is the movement chain of who from SpecCP to SpecCP , which consists
entirely of DP-links and can easily be parsed by PF. In other words, by deleting part of
the wh-movement chain, ellipsis (in this case sluicing) allows swiping to circumvent
the ban on preposition stranding in intermediate positions. This line of reasoning
has two advantages. First of all, it fits into a growing body of literature arguing that
ellipsis can be used to rescue what would otherwise be an illegitimate derivation or
representation (see Merchant () for an overview). Secondly, it accounts not only
for the ungrammaticality of () and the well-formedness of (a), but also for the
fact that the word order characteristic of swiping is disallowed in non-elliptical wh-
questions (a fact also noted by Merchant (: –)). Consider a representative
example in ().
() ∗ Who to was Lois talking?

At first sight, it is unclear what rules out this example. It can be derived exactly as in
(–) except for the deletion of IP: First the entire PP to who moves to SpecCP ,
and then the wh-phrase strands the preposition on its way to SpecCP . I propose that
the ungrammaticality of this example is due to the ban on preposition stranding in
60 Jeroen van Craenenbroeck

intermediate positions. That is, just like the sentence in (), the one in () contains
a non-uniform chain which causes the derivation to crash at PF. Given that in this
case ellipsis does not come to the rescue to undo this violation, the example is ruled
out. This means that the analysis of swiping I have proposed accounts not only for
the contrast between simple and complex wh-phrases, but also for more general
properties of swiping. I take this to be a further indication that the approach pursued
in this paper is on the right track.

.. Spading
The second subtype of sluicing has received much less attention in the literature than
swiping. It was first briefly discussed for Frisian by Hoekstra () and later more
extensively for Frisian and dialectal Dutch by van Craenenbroeck (). Consider a
representative example from the dialect of Wambeek in ().

() A: Jef eid iemand gezien. B: Wou da?


Jeff has someone seen who that
‘A: Jeff saw someone. B: Who?’ [Wambeek Dutch]

In B’s reply in this dialogue, the sluiced wh-phrase wou ‘who’ is followed by the
demonstrative pronoun da ‘that’. I will henceforth refer to this construction as
“spading”, which is an acronym for sluicing plus a demonstrative in non-insular
Germanic (van Craenenbroeck : ).  ,  I argue that spading represents a second
illustration of the interaction between sluicing and the split CP-system outlined in
Section ..
This section is organized as follows. In Section ... I explore the basic spading
data somewhat further. I argue that the structure underlying this construction is
that of a cleft with a wh-pivot. Section ... contains the analysis of spading. I
demonstrate that the basic properties of this construction follow straightforwardly
from the assumptions presented in the preceding sections. Finally, in Section ...,
I show that swiping and spading can co-occur in one and the same language (Frisian)
and that the properties of this combined construction are entirely as expected.

 Just as in English, the distal demonstrative pronoun in the dialect of Wambeek is homophonous with
the declarative complementizer. This means that it is not a priori clear whether the element following
the wh-phrase in () is a demonstrative pronoun or a complementizer. Note, however, that in the latter
case, this sentence would constitute an example of a complementizer occurring in a matrix wh-question,
a constellation otherwise unattested in this dialect. Moreover, dialects which morphologically distinguish
between the distal demonstrative pronoun and the declarative complementizer invariably use the former
in spading. See van Craenenbroeck (: –) for in-depth discussion of the categorial status of da ‘that’.
 Spading is attested in various dialects of Dutch, Frisian, Eastern Norwegian, and certain varieties
of German. Moreover, a construction similar to spading can be found in French, Serbo-Croatian, and
Czech. However, in these languages, the occurrence of a demonstrative pronoun to the right of a wh-
phrase is not restricted to sluicing contexts. I leave a cross-linguistic unification of spading and these related
constructions as a topic for further research. See van Craenenbroeck (: –) for some discussion
of the French construction.
Sluicing with more than one CP 61

... Background: Spading stems from a cleft with a wh-pivot In this section I argue
that the structure underlying a spading example such as B’s reply in () (repeated
below as ()) is not that of the regular wh-question in (), but rather that of the cleft
in (). 
() Wou da?
who that
‘Who?’
() Wou ei Jef gezien?
who has Jeff seen
‘Who did Jeff see?’
() Wou is da da Jef gezien eit?
who is that that Jeff seen has
‘Who is it that Jeff saw?’ [Wambeek Dutch]
The arguments in support of this claim all have the same logical structure. I first iden-
tify a property with respect to which clefts with a wh-pivot and “regular” sluicing—
which I assume to be derived from a non-cleft wh-question; see Merchant (:
–) for discussion—display diverging behavior, and then show that spading pat-
terns with clefts rather than with regular sluicing. For reasons of brevity, I only present
two such arguments here. For a more fully worked out argumentation, I refer the
reader to van Craenenbroeck (: –).
The first argument concerns the case of sluiced and spaded wh-pronouns. For obvi-
ous reasons, this can only be demonstrated for those dialects that morphologically
mark case on wh-pronouns. As is illustrated in () and (), the dialect of Waubach
is one of these (see Hinskens (: Section ..) for a more elaborate discussion of
the Waubach case system).
() A: ’t Kumt murrege inne noa ’t fees.
it comes tomorrow someone to the party
B: Wea (kemp noa ’t fees)?
who-nom comes to the party
‘A: Someone is coming to the party tomorrow. B: Who (is coming to the
party)?’ [Waubach Dutch]
() A: Ich han inne gezieë.
I have someone seen
B: Wem (has-te gezieë)?
who-acc have-you seen
‘A: I saw someone. B: Who (did you see)?’ [Waubach Dutch]
 Note that the element occupying the matrix SpecIP-position in the example in () is the demonstrative
pronoun da ‘that’ and not—as in English—the personal pronoun (he)t ‘it’. See van Craenenbroeck :
–, –, n, n for further discussion of the two types of clefts and their relation to spading.
62 Jeroen van Craenenbroeck

What these examples show is that in both non-elliptical and elliptical (i.e. sluiced)
wh-questions, subject wh-phrases are marked nominative in the dialect of Waubach,
while their object counterparts surface in the accusative form. This distinction breaks
down, however, in clefts with a wh-pivot. There, both object and subject wh-phrases
are marked nominative. This is shown in () and (). 
() Wea is dat dea noa ’t fees kemp?
who-nom is that rel to the party comes
‘Who is it that is coming to the party?’ [Waubach Dutch]
() Wea is dat dea-s-te gezieë has?
who-nom is that rel-ca-you seen have
‘Who is it that you saw?’ [Waubach Dutch]
This means that the case of a spaded wh-phrase provides a first testing ground for
the central claim made in this section. If spading derives from an underlying cleft,
both subject and object wh-phrases should surface in the nominative form, unlike in
“regular” sluicing. As is shown in () and (), this prediction is borne out. 
() A: ’t Kumt murrege inne noa ’t fees. B: Wea dat?
it comes tomorrow someone to the party who-nom that
‘A: Someone is coming to the party tomorrow. B: Who?’ [Waubach Dutch]
() A: Ich han inne gezieë. B: Wea dat?
I have someone seen who-nom that
‘A: I saw someone. B: Who?’ [Waubach Dutch]
The second argument concerns the fact that sluiced wh-phrases can be modified by
nog ‘else’, while cleft pivots cannot (see in this respect also Merchant (: –) on
so-called ‘contrast sluices’). This holds regardless of whether nog ‘else’ is pied-piped
by the wh-pivot or not. Consider some representative examples in () and ().
() A: Jef ei nie alliejn Lewie gezien. B: Nieje? Wou nog?
Jeff has not just Louis seen no who else
‘A: Jeff hasn’t just seen Louis. B: No? Who else?’ [Wambeek Dutch]
() ∗ Wou < nog > was da < nog > da Jeff gezien eit?
who else was that else that Jeff seen has
[Wambeek Dutch]
Again, spading patterns with clefts and not with regular sluicing in disallowing the
wh-phrase to be modified by nog ‘else’, irrespective of its linear order vis-à-vis the
demonstrative pronoun. This is shown in ().
 In the gloss in (), rel stands for ‘relative pronoun’ and ca for ‘complementizer agreement’.
 As pointed out in van Craenenbroeck () the case facts are more subtle than is suggested in the
main text. As this refinement does not affect the point made here, I leave it undiscussed.
Sluicing with more than one CP 63

() A: Jef ei nie alliejn Lewie gezien. B: Nieje? Wou < ∗ da > nog
Jeff has not just Louis seen no who that else
< ∗ da >?
that
‘A: Jeff hasn’t just seen Louis. B: No? Who else?’ [Wambeek Dutch]
Summing up, in this section I have established that spading is a subtype of sluicing
that differs from regular instances of sluicing in that it is derived from an underlying
cleft with a wh-pivot, rather than from a regular wh-question.
... The analysis of spading: Focus movement of the demonstrative to SpecCP 
The previous section has shown that spaded clauses stem from an underlying cleft
with a wh-pivot. This means that the spading example in () can now be schemati-
cally represented as in () (where the use of strikethrough signals PF-deletion).
() A: Jef eid iemand gezien. B: Wou is da da Jef gezien eit?
Jeff has someone seen who is that that Jeff seen has
‘A: Jeff saw someone. B: Who?’ [Wambeek Dutch]
As it stands, this representation faces a problem. It seems to suggest that spading
involves non-constituent deletion, a theoretically unappealing move. For this reason
I propose that the demonstrative pronoun in spading undergoes (focus) movement
to the left periphery of the clause. The representation in () can then be replaced by
the one in ().
() Wouk dai t i is t k da Jef gezien eit?
who that is that Jeff seen has
‘Who?’ [Wambeek Dutch]
The hypothesis that da ‘that’ undergoes focus movement to the left periphery is cor-
roborated by two other properties of spading. Firstly, in this construction it is always
the demonstrative pronoun that bears main stress, and never the wh-phrase—as is
the case in regular sluicing. Given the close correlation between focus (movement)
and stress, this is precisely what one would expect from the representation in ().
Secondly, unlike regular sluicing, spaded clauses carry a negative presupposition. For
example, by adding the demonstrative pronoun to the sluiced wh-phrase, speaker B
in the dialogue in () indicates that he did not expect Jeff to have seen someone (e.g.
because he knows that Jeff stayed home alone all day). Given that in the cleft under-
lying B’s reply, the demonstrative pronoun refers back to the preceding statement in
the discourse (see van Craenenbroeck (: –) for discussion), this reading is
entirely predictable. Specifically, by focusing da ‘that’, the sentence ‘Jeff saw someone’
is singled out from all the possible activities Jeff could have been engaged in, and is
given high salience. In other words, it is the focus on the demonstrative pronoun that
creates the negative-presupposition reading.
64 Jeroen van Craenenbroeck

This leaves the question of the landing site of the postulated movement operation.
That is, where does the demonstrative pronoun move to? Given the view on the CP-
domain developed in Section ., the most natural place for the demonstrative to land
in would be SpecCP . Recall that CP is the projection in which clause-typing features
are checked, which makes it ill-suited as a landing site for focus movement (focus not
being a clause type). Moreover, in their exploration of the left periphery in Italian and
its dialects, Benincà and Poletto () show that foci systematically target the lower
half of the CP-domain, which in my proposal would translate as CP . Thirdly, focus
constructions are typically characterized as operator/variable-dependencies (see e.g.
Rizzi ()). That too would suggest that C  is the most natural head for the focus
feature. Accordingly, I propose that the demonstrative in a spading example under-
goes focus movement to SpecCP triggered by a matching feature on C  . In order
to make this more concrete, I now go through the derivation of the example in (),
starting from the point at which the matrix C  is merged. This is shown in ().

() [C  C  [+Op, +F, [E]+Op,+Q] [IP da[+F] is wou[+Op, +Q] da Jef gezien
that is who that Jeff seen
eit]]
has
C  is endowed with both an operator feature and a focus feature (abbreviated here
as [+F]). The former targets (the operator feature of) the wh-phrase wou ‘who’, while
the latter probes the focus feature of the demonstrative pronoun. Moreover, given
that spading is a subtype of sluicing, C  also hosts the [E] feature, which checks its
operator feature against that of C  . Next, two movement operations take place. First
the demonstrative pronoun moves to SpecCP to check its focus feature, and then the
wh-phrase moves to the inner specifier of C  to check its operator feature (i.e. it tucks
in; see Richards (: Chapter )). This is represented in ().

() [CP da[+F] [CP wou[+Op, +Q] [C  C  [+Op,+F, [E]+Op,+Q] [IP da[+F] is
wou[+Op, +Q] da Jef gezien eit]]]]

Recall that in this local configuration, [E] can check its [+Q] feature against that of
the wh-phrase in SpecCP . It is now syntactically fully licensed, and at PF will trigger
the deletion of IP. The next step in the derivation involves the merger of C  . This head
is endowed with a clause-typing feature targeting that of the wh-phrase. Accordingly,
wou ‘who’ moves to SpecCP . This is shown in ().

() [CP wou[+Op,+Q] [C  C  [+Q] [CP da[+F] [CP wou[+Op,+Q] [C  C  [+Op,+F,
[E]+Op,+Q] [IP da[+F] is wou[+Op, +Q] da Jef gezien eit]]]]]]

Finally, at PF, [E] triggers the deletion of IP, and the copy of wou ‘who’ in SpecCP is
deleted as well. This yields the representation in ().
Sluicing with more than one CP 65

() [CP wou[+Op,+Q] [C  C  [+Q] [CP da[+F] [CP wou[+Op,+Q] [C  C  [+Op,+F,
[E]+Op,+Q] [IP da[+F] is wou[+Op,+Q] da Jef gezien eit]]]]]]
What remains is the wh-phrase wou ‘who’ in SpecCP and the focus-moved demon-
strative pronoun in SpecCP , i.e. the derivation has successfully yielded an instance
of spading. Moreover, the analysis just outlined predicts that spading should be
disallowed with complex wh-phrases. Recall that sluicing with complex wh-phrases
deletes CP rather than IP. This implies that any material that occurs in CP (either
as the result of base-generation or through movement) should be unable to surface
to the right of a sluiced complex wh-phrase. As shown in (), this prediction is
borne out.
() A: Jef eid ne student gezien. B: Welke student (∗ da)?
Jeff has a student seen which student that
‘A: Jeff saw someone. B: Who?’ [Wambeek Dutch]
As was first observed for Frisian by Hoekstra (: –), complex wh-phrases are
excluded from spading. This construction thus constitutes a second illustration of the
interaction between the split CP-domain outlined in Section . and the syntax of
sluicing.
... Spading meets swiping: The case of Frisian Nothing in the analyses of swiping
and spading prevents the two phenomena from co-occurring. In particular, if the
underlying structure of a sluiced clause is a cleft with a PP-contained simple wh-
phrase as pivot, that wh-phrase should be able to strand its preposition in SpecCP in
addition to the demonstrative pronoun moving to that position. As illustrated in (),
Frisian instantiates precisely such a scenario.
() A: Jan hat juster in praatsje holden.
John has yesterday a talk held
B: Wêr dat oer?
where that about
‘A: John gave a talk yesterday. B: What about?’ [Frisian]
In B’s reply in this dialogue, a sluiced wh-phrase (in this case the R-pronoun wêr
‘where’) is followed by the demonstrative pronoun dat ‘that’, which is in turn followed
by the stranded preposition oer ‘about’. As such, this example combines swiping with
spading. This conclusion is further corroborated by the fact that complex wh-phrases
are excluded from the pattern illustrated in (). This is shown in ().
() A: Jan hat juster in praatsje holden.
John has yesterday a talk held
B: ∗ Hokker boek dat oer?
which book that about
‘A: John gave a talk yesterday. B: About which book?’ [Frisian]
66 Jeroen van Craenenbroeck

In order to see how the derivation of () proceeds, consider the tree structure
in ().
() CP 1

wê r C 1⬘
[+Op,+Q]

C01 CP2
[+Q]

dat CP2
[+F]

[PP twêr oer ] C2 ⬘ Þ ELLIPSIS

0
C2 IP
[+Op,+F]
[E ] tdat wie tPP dat Jan juster in praatsje holden hat
+Op,+Q

The IP in this structure contains a cleft which is roughly the equivalent of ‘What is it
that John gave a talk about yesterday?’. When C  is merged on top of this structure,
the demonstrative pronoun dat ‘that’ first moves to SpecCP to check the focus feature
on C  . Then, the wh-PP moves to check the operator feature of C  , and it tucks in
beneath the demonstrative pronoun. Finally, C  is merged. It bears a strong clause-
typing feature attracting the wh-phrase wêr ‘where’ to its specifier. As a result of this
movement, the preposition oer ‘about’ is stranded in the inner specifier of CP . On
top of all this, C  is also endowed with the [E] feature. Given that [E] can check both
its [+Op] and [+Q] features in a local environment, it remains on C  throughout
the derivation. This means that at PF, it triggers the deletion of IP (the complement
of the head on which it resides). What this derivation yields, then, is an elliptical
clause consisting of a wh-phrase followed by a demonstrative pronoun followed by
a stranded preposition, i.e. it has successfully derived the combination of swiping and
spading.

.. Summary
In Section . I have examined one of the empirical predictions made by the theo-
retical proposal outlined in Section .. If sluicing with complex wh-phrases differs
from sluicing with simple wh-phrases in the amount of structure that is elided (CP
in the former case, IP in the latter), material that appears in SpecCP should be
able to survive the deletion process in one case but not the other. I have discussed
two subtypes of sluicing that display precisely the expected pattern. In swiping, a
preposition is stranded in SpecCP by the wh-phrase on its way to SpecCP , while
in spading, a demonstrative pronoun (focus-)moves to this position independently
of the wh-phrase.
Sluicing with more than one CP 67

. Conclusion
This paper has explored the interaction between the split-CP hypothesis and the
syntax of sluicing. I have shown that when there is more than one CP, there is also
variation as to which part of the clausal structure is deleted by sluicing. When wh-
movement targets a low CP-layer (as in Hungarian), IP is deleted, but when it targets
a high CP-layer, it can be either IP or a low CP-projection that is deleted.
The analysis of the CP-domain in Section . postulates an important difference
between simple and complex wh-phrases. While the former move from their IP-
internal base position through the lower CP-projection (CP ) and onto the higher one
(CP ), the latter are base-generated in SpecCP and involve empty operator movement
from the IP-internal base position to SpecCP . This has non-trivial consequences for
the syntax of sluicing (Section .). While simple wh-phrases are able to license the
feature responsible for ellipsis in situ (i.e. on C  ), the empty operator involved in wh-
movement with complex wh-phrases is not. As a result, the ellipsis feature has to move
on to C  . Under the assumption that it always triggers deletion of the complement
of the head on which it resides, this line of reasoning leads to the conclusion that in
sluicing with complex wh-phrases, CP is deleted, while in sluicing with simple wh-
phrases, IP is deleted.
This conclusion was put to the test in Section ., which examined two subtypes
of sluicing in which overt material is stranded to the right of the sluiced wh-phrase.
The first one, dubbed “swiping” by Merchant (), involves preposition strand-
ing in SpecCP by the moving wh-phrase on its way to SpecCP , while the second
one, “spading”, features a demonstrative pronoun that undergoes focus movement to
SpecCP . The fact that both swiping and spading are illicit with complex wh-phrases
provides strong additional support for the proposal.

Two cases of violation repair


under sluicing
S A N DR A STJ E PA NOV IĆ

. Introduction
It has been observed that sluicing, a type of ellipsis, can repair certain types of
grammatical violation (see Ross (), Lasnik (), Merchant (, ), Fox
and Lasnik (), and Boeckx and Lasnik (), among others). In other words,
certain structures that are deviant without sluicing become grammatical with sluicing.
Merchant () provides a taxonomy of structures repairable by ellipsis, dividing
them into three classes:

() a. Class I: An operation O (generally movement) that is required in non-


elliptical context is not found under ellipsis: ellipsis bleeds O.
b. Class II: An operation O (generally movement) that is impossible in non-
elliptical contexts is found under ellipsis: ellipsis “feeds” O
c. Class III: Apparent miscellanea (Merchant )

In this paper I discuss two cases of repair under sluicing in Serbo-Croatian that
fall into Classes I and II. The first case concerns the loss of the preposition in the
wh-phrase remnants of sluicing, also discussed in Stjepanović (). At first sight
such cases look like typical cases of preposition-stranding (P-stranding) violation
repair: Serbo-Croatian seems to allow P-stranding under sluicing, even though it does
not allow it under wh-movement without sluicing. However, upon closer scrutiny,
it becomes apparent that the loss of the preposition under sluicing is not condi-
tioned by P-stranding. The preposition is pied-piped in overt syntax together with the
wh-phrase as usual. It is then deleted at PF, but not through sluicing (i.e. TP deletion),
although sluicing provides a necessary condition for it to apply. I call the operation
that deletes the preposition in the presence of sluicing “P-drop”. Since P-drop is not
possible without sluicing, and is fed by it, it is a Class II operation.
Violation repair under sluicing 69

The second case of violation repair under sluicing discussed in the paper concerns
the repair of Inverse Case Filter violations for inherent case, which I will refer to
as the “Inverse Inherent Case Filter” (IICF). This type of violation repair is found
in Serbo-Croatian sentences involving numeral phrases, such as pet djevojaka ‘five
girls’, in which the numeral quantifier pet ‘five’ assigns genitive of quantification to
its nominal complement djevojaka ‘girls’. As discussed in Franks () and Bošković
() among others, these phrases cannot occur as complements of inherent-case-
marking verbs. Bošković () proposes that such sentences are ungrammatical
owing to an LF version of the Inverse Case Filter. I show that such violations are
repaired under sluicing, and that the Inverse Case Filter for inherent case therefore
cannot be subsumed solely under LF mechanisms. Rather, I propose that the IICF
has a morphological requirement that demands inherent case to be morphologically
realized, if it can be. I show that the repair of violations of this condition is due to
the fact that an offending case feature of the verb, which fails to be assigned to its
complement, is no longer present after the deletion of TP occurs at PF. Since the
assignment of this case feature is necessary in structures without sluicing, but is not
required with sluicing, the repair of the IICF then falls under the Class I type of
elliptical repair.
The chapter is organized as follows: In Section ., I briefly summarize the findings
of Stjepanović () that concern the loss of preposition in structures with sluicing.
I show why such cases cannot be considered cases of P-stranding repair, but are
instances of P-drop. In Section ., I turn to the repair of violations of the IIFC under
sluicing. I show why these cases force us to conclude that the IICF is a constraint
that requires morphological case to be realized, if it can be. Section . concludes the
paper.

. P-drop under sluicing


.. The core facts
As shown in Stjepanović (), the preposition that heads a wh-PP in Serbo-
Croatian can be omitted if the wh-PP is a remnant of sluicing:

() a. Petar je glasao protiv nečega, ali ne znam (protiv)


Petar is voted against something-gen but not I-know against
čega.
what-gen
‘Petar voted against something, but I don’t know what.’
b. ∗ (Protiv) čega je Petar glasao?
against what-gen is Petar voted
‘What did Petar vote against?’
70 Sandra Stjepanović

At first sight, sentences like (a) with a missing preposition could be taken to be cases
of the repair of a P-stranding violation under sluicing. Note that P-stranding is not
possible in Serbo-Croatian under regular overt wh-movement:
() ∗ Čega je Petar glasao protiv?
what-gen is Petar voted against
Under the assumption that the second conjunct of (a) involves the repair of
P-stranding and that sluicing involves overt wh-movement followed by TP deletion
at PF (Merchant (), among others), the second conjunct in (a) would receive the
derivation in ().
() a. . . . ali ne znam [CP čegai [TP je Petar glasao protiv t i ]
but not I-know what-gen is Petar voted against
(overt wh-movement)
b. . . . ali ne znam [CP čegai [TP je Petar glasao protiv t i ]
‘. . . but I don’t know what’
(TP deletion at PF)
In (a), the wh-phrase undergoes overt movement, stranding the preposition. This
illicit structure with the stranded preposition is then deleted at PF when sluicing
applies, as in (b). With the illicit structure gone, the violation is repaired, and the
result is a grammatical sentence.
While in Stjepanović () I assume that sluicing does involve overt wh-
movement followed by TP deletion at PF, I show that the analysis in () cannot be
maintained for Serbo-Croatian. Thus, Serbo-Croatian does not threaten Merchant’s
() P-stranding generalization in (). In the next section, I discuss evidence for
this claim.
() Form–identity generalization II: P-stranding
A language L will allow preposition stranding under sluicing iff L also allows
preposition stranding under wh-movement. (Merchant : )

.. Is it P-stranding?
Even though Serbo-Croatian does allow Ps to be omitted in sluicing remnants, as
shown in (), in Stjepanović () I show that such sentences do not involve the
repair of P-stranding violations. The evidence for this claim is based on the examples
in (), which involve the omission of P in coordinated sluicing remnants.
() a. Petar je sakrio igračku ispod jedne stolice i pored jednog
Petar is hid toy under one chair-gen and beside one
zida, ali ne znam (ispod) koje stolice i (pored)
wall-gen but not I-know under which chair-gen and beside
kojeg zida.
which wall-gen
Violation repair under sluicing 71

‘Petar hid the toy under a chair and next to a wall, but I don’t know which chair
and which wall.’
b. Petar je glasao za nešto i protiv nečega,
Petar is voted for something-acc and against something-gen
ali ne znam (za) šta i (protiv) čega.
but not I-know for what-acc and against what-gen
‘Petar voted for something and against something, but I don’t know for
what and against what.’
In (), both the antecedent and the remnant of sluicing are coordinated phrases with
two PP conjuncts headed by two different Ps. Note that P-omission is possible in the
remnant conjuncts. Note also that the remnant wh-phrases do not need to have the
same morphological case. Now, if the remnant wh-phrases in () when P is omitted
were remnants of sluicing, these data would provide a strong argument that the P-
loss in the remnants is not due to P-stranding. This is because under no current
theory of movement can the coordinated remnant phrase move as a constituent, while
stranding the two prepositions. So the question is how this coordinated remnant is
derived and what its structure is. In Stjepanović (), I consider three possibilities—
either it is a base-generated fragment (along the lines of van Riemsdijk (a)), or it
involves pseudo-sluicing (a cleft-clause strategy), or it is a result of sluicing—and I
show that the last option is correct. First, based on data such as (–), I show that
the remnants in () cannot be base-generated fragments, nor can they be derived by
pseudo-sluicing.
() Petari je sjeo pored nekog svogi prijatelja i ispod nekog drveta,
Petar is sat beside some self friend-gen and under some tree-gen
ali ne znam (pored) kojeg svogi prijatelja i (ispod) kojeg
but not I-know beside which self friend-gen and under which
drveta.
tree-gen
‘Petari sat beside a friend of hisi and under a tree, but I don’t know beside which
friend of hisi and under which tree.’
() a. Petar je sakrio igračku ispod jedne stolice i pored jednog
Petar is hid toy under one chair-gen and beside one
zida, ali ne znam koja stolica i koji zid ∗ (su

wall-gen but not I-know which chair-nom and which wall-nom are
to bili ispod koje i pored kojeg je on
that been-PL under which-gen.f and beside which-gen.m is he
sakrio igračku).
hid toy
‘Petar hid the toy under a chair and next to a wall, but I don’t know which
chair and which wall it was under which and next to which he hid the toy.’
72 Sandra Stjepanović

b. Petari je sjeo pored neke svojei prijateljice, ali ne znam koja


Petar is sat beside some self friend-gen but not I-know which
∗ svoja/njegova /Petrova prijateljica ∗ (je to bila pored koje je
i i
self/his/Petar’s friend-nom is it been beside which is
on sjeo).
he sat
‘Petari sat beside a friend of hisi , but I don’t know which friend of hisi it was
beside which he sat.’
Example () shows that the remnants exhibit A-binding connectivity effects, since a
reflexive is possible in the remnants.  The only way that the reflexive in the remnants
can be bound in these examples is if we posit that the remnants are a part of a full TP
at some point, with the subjects Petar c-commanding the reflexive. Under a base-
generated fragment approach, it would be extremely difficult to account for how
binding occurs here. Given this, I conclude that it is not the correct approach. Example
() shows that the coordinated remnant in () is also not derived by pseudo-sluicing
of the type illustrated in () and discussed in Merchant (: ). In (), sluicing
deletes the cleft clause, leaving the wh-phrase as a remnant.
() Someone just left—guess who (it was_that just left)
As shown in (), sluicing a cleft clause in Serbo-Croatian is not possible. While the
second conjunct in () sounds awkward when it contains a full cleft clause, it is much
better than when the cleft clause is sluiced. Thus, in Stjepanović (), I conclude
that we cannot assume that in cases like (), the remnants are derived by pseudo-
sluicing. But even if sluicing the cleft clause were possible, we would still have evidence
against the pseudo-sluicing approach to examples like (), based on case and binding
connectivity effects. P-less remnants display case connectivity effects, having exactly
the case that is assigned by the missing Ps. The remnants in (a), for instance, are
genitive, because these cases are assigned by ispod ‘under’ and pored ‘beside’. On
the other hand, cleft-clause pivots, such as those in (), are always nominative. The
contrast between () and (b) shows that P-less remnants and the pivots of cleft
clauses behave differently with respect to A-binding connectivity effects as well. While
P-less remnants exhibit A-binding connectivity effects, as illustrated in (), cleft-clause
pivots in Serbo-Croatian do not. Example (b) shows that a reflexive pronoun in the
pivot is not permitted, while a co-indexed pronoun or an R-expression is. Given these
facts, I conclude that P-less remnants are not derived by pseudo-sluicing, despite the
fact that this strategy seems to be productive in other languages that do not allow
P-stranding (see for example Kuwabara (), Merchant (), and Rodrigues et al.
()). On the other hand, if we assume that P-less remnants are remnants of regular
sluicing, we straightforwardly account for both types of connectivity effect.
 For evidence that reflexives in Serbo-Croatian need a clausemate antecedent, see Zlatić ().
Violation repair under sluicing 73

.. The structure of P-less sluices


In the previous section we have seen that P-less remnants in Serbo-Croatian examples
like () are not base-generated fragments, nor are they pseudo-sluicing remnants.
Rather, they are true sluices. The question is: What are their structures? Given that
their correlates are coordinated PPs, it is reasonable to expect that the sluices are also
coordinated PPs, with their Ps lost in the course of derivation. However, in Stjepanović
(), I show that we also have to worry about another possibility. A reasonable
hypothesis is that they might also be coordinated CPs, with each PP undergoing
wh-movement within its CP, and losing its P at some point in the derivation. Note
that this state of affairs would not exclude the possibility of the P-loss being due
to P-stranding under sluicing. In Stjepanović (), I show that if data are suffi-
ciently controlled, the P-loss under sluicing is possible in coordinated PPs. When,
for example, we consider the meaning of (a), it can be understood as involving a
single place (under a chair and next to a wall) where the subject Petar hid the toy.
This is confirmed by the fact that the sentence E, da mi je znati gdje je to mjesto! ‘Eh,
I’d really like to know where that place is!’ can naturally follow (a). The fact that
this interpretation is obtained indicates that the remnant in (a) can consist of two
conjoined PPs. Coordinating two CPs without sluicing gives rise to an interpretation
in which two places are implied:

() . . . ali ne znam ispod koje stolice je on sakrio igračku i


but not I-know under which chair-gen is he hid toy and
pored kojeg zida je on sakrio igračku.
beside which wall-gen is he hid toy
‘. . . but I don’t know under which chair he hid the toy, and next to which wall
he hid the toy.’

Example () clearly implies two places, since it is felicitous to follow it with a sentence
such as E, da mi je znati gdje su ta mjesta! ‘Eh, I’d really like to know where those places
are!’, but not with a sentence such as E, da mi je znati gdje je to mjesto! ‘Eh, I’d really like
to know where that place is!’. If the remnants in (a) were coordinated CPs that had
lost their TPs as a result of sluicing, (a) should only have the interpretation exhibited
by its non-sluiced counterpart in (). Since this is not true, I conclude in Stjepanović
() that the remnants in (a) may be conjoined PPs that have lost their Ps.
Here I provide another argument that P-loss is possible in coordinated PP rem-
nants. Note that P-loss is possible with PPs that are modifiers of nouns within NPs.
Consider ().

() On voli jabuke iz jedne zemlje i sa jedne planine,


He likes apples from one country-gen and on one mountain-gen
ali ne znam koje zemlje i koje planine.
but not I-know which country-gen and which mountain-gen
74 Sandra Stjepanović

‘He likes apples from a certain country and a certain mountain, but I don’t know
which country and which mountain.’
In (), the antecedent of sluicing is a coordinated phrase with two PP conjuncts
headed by two different Ps. The coordinated PP is an adjunct of the noun jabuke
‘apples’. The sluicing remnant in the second conjunct of () is a coordinated phrase
consisting of what on the surface look like two coordinated wh-NPs that correspond to
the indefinite complements of Ps in the antecedent. The coordinated remnant is P-less,
while the antecedent does contain Ps. It is reasonable to assume that the remnant is a
coordinated PP, just as its antecedent is. The question of whether the remnant could
be derived from two coordinated CPs of the sort in (a) can be raised in this case
as well, which would not preclude the possibility of having the repair of P-stranding
violations, as illustrated in (b). However, the interpretation of sentences like () and
() tells us that PP is a clear underlying option, just as it was the case above. 
() a. . . . ali ne znam iz koje zemlje on voli jabuke i
but not I-know from which country-gen he loves apples and
sa koje planine on voli jabuke.
on which mountain-gen he loves apples
‘. . . but I don’t know apples from which country he likes and apples from
which mountain he likes.’
b. . . . ali ne znam [CP [koje zemlje]i [TP on voli [NP jabuke
but not I-know which country-gen he loves apples
[PP iz [NP t i ]]]] i [CP [koje planine]j [TP on voli
from and which mountain-gen he loves
[NP jabuke [PP sa [NP t j ]]]].
apples on
‘. . .but I don’t know which country and which mountain.’

Example () is interpreted as involving the same apples that grow in a particular
country and on a mountain in that country, while (a), with two coordinated CPs
that are the source of sluicing in (b), is not: (a) can only be interpreted as involving
some apples that grow in a particular country and different apples that grow on a
particular mountain. Therefore, I conclude that () involves coordinated PPs in the
remnant that have lost their Ps. And if it is possible to delete Ps in these data without

 Note also that, unlike in English, in Serbo-Croatian, adjunct PP extraction out of NPs is possible, as
shown by Stjepanovic ():

(i) Iz koje zemlje on voli jabuke?


from which country he likes apples
‘Apples from which country does he like?’
Thus, sentences like () do not involve the repair of a violation caused by PP adjunct extraction out
of NPs.
Violation repair under sluicing 75

P-stranding occurring first, then the P-loss in data like () need not be conditioned
by P-stranding either.
Since it is possible to have missing Ps in the coordinate PP remnants of sluicing, it is
clear that the target clause undergoing sluicing cannot involve syntactic P-stranding
in the standard sense, if sluicing involves wh-movement in narrow syntax followed
by TP deletion at PF. The “non-movement” theories of sluicing proposed in Lobeck
() and Chung et al. (), among others, might at first sight be more successful
in accounting for these data. Under the approach of Chung et al., a P-less NP remnant
is base-generated in the SpecCP, and the complement of C is a null TP. At LF, this
null TP is replaced by copying the antecedent TP, with the remnant binding a variable
provided by the indefinite in the copied TP, as in ().

() a. Petar je glasao protiv nekoga, ali ne znam


Petar is voted against someone-gen but not I-know
[CP koga C [TP e]] ⇒ at Spell-Out
who-gen
‘Petar voted against somebody, but I don’t know who.’
b. Petar je glasao protiv nekoga, ali ne znam
Petar is voted against someone-gen but not I-know
[CP kogax C [TP Petar je glasao protiv nekogax ]]. ⇒ at LF
who-gen Petar is voted against someone
The derivation of the second conjunct of () with a coordinated PP as a remnant under
this approach is given in ().

() . . . ali ne znam [CP koje stolicex i kojeg ziday


but not know which-gen chair-gen and which wall-gen
C [TP Petar je sakrio igračku ispod [jedne stolice]x i pored
Petar is hid toy under one chair-gen and beside
[jednog zida]y ] ⇒ at LF
one wall-gen
Depending on whether both variables in () can be considered to be bound, this
approach might be able to account for data like (). It can also easily account for the
case and binding connectivity effects in sluicing remnants discussed above. However,
although it may look successful at first sight, the account faces problems. It predicts
that all non-P-stranding languages should allow P-omission under sluicing, which is
not the case (see Merchant ()). As I show in Stjepanović (), even in Serbo-
Croatian there are types of ellipsis in which Ps cannot be omitted, such as in VP-
ellipsis contexts () and (), comparative ellipsis (), gapping (), and also in some
constructions that do involve sluicing, such as sluicing under sprouting (), multiple
sluicing (), and contrast sluicing ().
76 Sandra Stjepanović

() Protiv koga će oni glasati, a ∗ (protiv) koga ćete vi?


against whom-gen will they vote and against whom-gen will you
‘Against whom will they vote, and against whom will you?’
() O kojoj djevojci je on više mislio nego što je ∗ (o) kojem
about which girl-loc is he more thought than what is about which
momku?
guy-loc
‘About which girl did he think more than he did about which guy?’
() Protiv koga će oni glasati, a ?∗ (protiv) koga vi?
against whom-gen will they vote, and against who-gen you
‘Against who will they vote, and against who will you vote?’
() Znam protiv koga je on glasao, ali ne znam ∗ (protiv) koga je
I-know against whom is he voted, but not I-know against whom is
Marija.
Marija
‘I know who he voted against, but I don’t know who Marija voted against.’
() Petar je glasao, ali ne znam ∗ (protiv) čega.
Petar is voted but not I-know against what-gen
‘Petar voted but I don’t know against what.’

() Neko je glasao protiv nečega, ali ne znam ko


someone is voted against something but not I-know who-nom
∗ (protiv) čega.

against what-gen
‘Someone voted against something, but I don’t know who against what.’

() Znam da je glasao protiv Petra. Znaš li ∗ (protiv) koga


I-know that he voted against Petar You-know Q against who
drugog?
else
‘I know that he voted against Petar. Do you know who else (he voted against)?’
Given that such an approach over-generates considerably, I conclude that we cannot
rely on the non-movement approach to sluicing to account for the data discussed in
this paper.
Since the base-generation accounts as well as the pseudo-sluicing account of P-less
remnants can arguably be ruled out, and since P-stranding is not a necessary condition

 For evidence that P-loss under sprouting is also not allowed in English and the reasons behind it, see
Chung ().
 For evidence that cases like these involve multiple sluicing rather than gapping, see Stjepanović ().
Violation repair under sluicing 77

for P-loss, the loss of P does not seem to be conditioned by anything in the narrow
syntactic computation. Rather, this state of affairs suggests that P-loss in sluicing rem-
nants is a post-syntactic phenomenon. Assuming that sluicing involves wh-movement
in narrow syntax followed by TP deletion at PF, the computation of sentences involv-
ing P-less remnants within narrow syntax proceeds in the same way as for their non-
sluiced counterparts. In other words, wh-PP undergoes overt wh-movement. At PF,
TP is deleted. At this point, the right conditions for the deletion of P are created,
and P is deleted. I call this operation that deletes the preposition “P-drop”. While the
exact reasons for the occurrence of P-drop are still somewhat mysterious, it is not
unreasonable to assume that it occurs at PF, given the characteristics it has as observed
in the data discussed so far:

() a. P-drop occurs only in the presence of simple sluicing.


Non-sluiced counterparts of sentences involving P-less remnants are
ungrammatical. P-drop is also not possible with other ellipsis phenomena,
as shown in (–).
b. P-drop occurs only if the preposition is recoverable, i.e. if it is present in the
antecedent. Recall that P-drop under sprouting is not allowed.
c. The preposition undergoing P-drop cannot be part of a contrastively focused
phrase.
In the contrast-sluicing example in (), the remnant is in clear contrast
with its correlate in the antecedent. Bošković () and Stjepanović (),
among others, argue that in multiple sluicing in Serbo-Croatian, wh-phrases
are also contrastively focused. In gapping and pseudo-gapping, the remnants
are in contrast with their correlates in the antecedent. Recall that none of
these constructions allow P-less sluices.

Thus, in order to be dropped, a preposition needs to be recoverable, and it cannot


head a contrastively focused phrase. De-accenting and deletion require exactly such
conditions. Why must the deletion of a preposition occur only when the remnant
precedes the gap created by sluicing? This remains to be seen. However, what we
have identified here is an operation that is found under sluicing, but is impossible in
non-sluicing contexts. In other words, sluicing “feeds” P-drop. As a result, these con-
structions fall into Class II in the taxonomy of elliptical repair proposed by Merchant
().
In the rest of the paper I discuss another type of violation repair under sluicing,
which has not been noticed so far. This type of repair is found in Serbo-Croatian
sentences containing numeral phrases, such as pet muškaraca ‘five men’ in which the
numeral (pet ‘five’) assigns the so called genitive of quantification (GQ) to its nominal
complement (muškaraca ‘men’).
78 Sandra Stjepanović

. Inverse Inherent Case Filter violation repair


.. Genitive of quantification
The following sentences illustrate the use of GQ in Serbo-Croatian:
() a. Koliko muškaraca će on pozvati?
how-many men-gen will he invite
‘How many men will he invite?’
b. On će pozvati pet muškaraca.
He will invite five men-gen
‘He will invite five men.’
() a. ∗ Koliko muškaraca se on zahvalio?
how-many men-gen refl he thanked
‘How many men did he thank?’
b. ∗ On se zahvalio pet muškaraca.
he refl thanked five men-gen
‘He thanked five men.’
Koliko muškaraca ‘how many men’ and pet muškaraca ‘five men’ are numeral phrases
with GQ. In (), both phrases occur as complements of the structural case-marking
verb pozvati ‘to invite’, which assigns accusative case to its complement. In (),
on the other hand, they are complements of the verb zahvaliti se ‘to thank’, which
assigns dative, an inherent case in Serbo-Croatian. As the contrast between () and
() shows, these numeral phrases are not felicitous as complements of inherent
case-marking verbs, while they are fully acceptable as complements of structural
case-marking verbs. Contrasts like these have received considerable attention in the
literature on Slavic GQ (see, among others, Franks (, ), Bošković (,
)).
Bošković (), who incorporates many insights of Franks (, ), who in
turn builds on Babby (), among others, attributes the ungrammaticality of ()
to a violation caused by the impossibility of the verb discharging its inherent case
onto its numeral phrase complement. More precisely, he argues that morphologically
undeclinable forms of higher numerals like pet ‘five’ or their wh counterparts, such
as koliko ‘how many’, are caseless in Serbo-Croatian. Since the head of the numeral
phrase is caseless, the verb cannot check its case against it. I will refer to this as a
violation of the Inverse Inherent Case Filter (IICF). Furthermore, following Chom-
sky’s (b) approach to inherent case, Bošković assumes that a verb that assigns
inherent case will theta-mark its object iff it assigns it the inherent case in question.
Given this assumption, according to Bošković, having a numeral phrase like those in
() as an object of an inherent-case-assigning verb in Serbo-Croatian then inevitably
causes a theta-criterion violation. I will refer to this analysis as a violation of an LF
version of the IICF. On the other hand, examples like () are grammatical because
Violation repair under sluicing 79

structural case is not associated with theta-role assignment in Chomsky’s (b) case
system. As a result, even if the verb cannot check its structural-case feature against
its numeral-phrase complement, the derivation does not crash as there is no theta-
criterion violation, unlike in (). In other words, under Bošković’s analysis there is
no Inverse Case Filter for structural case. In the next section I discuss an interesting
behavioral pattern of these constructions in instances of sluicing.

.. Sluicing and GQ


Interestingly, what has not been noticed so far with respect to GQ constructions in
Serbo-Croatian is that sluicing repairs violations of the IICF, as shown in ().
() a. Znam da će se zahvaliti jednoj ženi, ali ne
I-know that will refl thank one-dat woman-dat but not
znam koliko muškaraca.
I-know how-many men-gen
‘I know that he will thank one woman, but I don’t know how many men.’
b. A: Vidio sam da se približio jednom lavu.
seen am that refl approached one-dat lion-dat
B: A koliko tigrova?
and how-many tigers-gen
‘A: I saw that he approached one lion. B: And how many tigers?’
c. ∗ Koliko tigrova se približio?
how-many tigers refl approached
‘How many tigers did he approach?’
Example (a) shows that the same numeral phrase koliko muškaraca ‘how many men’
that cannot occur as a complement of the inherent case-marking verb zahvaliti se ‘to
thank’ is possible if the TP in which the phrase originates is elided by sluicing. Sluicing
occurs in an embedded clause here. Example (b) shows that the numeral phrase
koliko tigrova ‘how many tigers’ is a felicitous remnant of matrix sluicing. Note that
the verb whose complement the numeral phrase is understood to be is the inherent
case-marking verb približiti se ‘to approach’, which assigns dative to its complement.
Example (c) shows that the same numeral phrase cannot occur with this verb in a
non-sluiced sentence. The question is how to account for this contrast between the
non-sluiced examples and their sluiced counterparts.
Recall that Bošković () attributes the ungrammaticality of examples like ()
and (c) to an LF version of the IICF. However, given that examples like (a) and
(b) are grammatical, it is clear that the ungrammaticality of these examples cannot
be subsumed solely under LF mechanisms. This is because the LFs of (a) and
(c) on the one hand, and the embedded clause of the second conjunct in (a)
and (b) on the other, are the same (under both LF and PF approaches to sluicing).
Thus, if (a) and (c) were to be ungrammatical due to a violation of an LF version
80 Sandra Stjepanović

of the IICF, one would then expect (a) and (b) to be just as ungrammatical,
counter to fact.
The contrasts observed here can be straightforwardly accounted for, however, if we
assume that sluicing is a PF process (as in Merchant (, ), Lasnik (), Fox
and Lasnik (), Boeckx and Lasnik (), and Bošković (), among others),
and that the inadequacy exhibited in examples like () and (c) is rooted in PF
properties, and can be eliminated by deleting the offending structures when sluicing
applies at PF. I would therefore like to propose that the impossibility of numeral
phrases like koliko tigrova ‘how many tigers’ and pet tigrova ‘five tigers’ occurring in
inherent-case contexts in Serbo-Croatian is attributed to a morphological inadequacy.
I propose that the IICF has a morphological requirement, which should be formulated
along the following lines:
() Inherent case must be morphologically realized, if it can be.
A verb lexically specified for inherent case must assign its morphological case
feature to the NP that checks its abstract case feature against the verb in syntax.
The formulation in () implies that an inherent-case-marking verb is lexically spec-
ified with a morphological case feature, which it must assign to its complement. The
Case Filter, as a licensing condition on NPs, is also present in syntax, i.e. NPs regularly
have unvalued/unchecked abstract case features that they must check/value in the
standard way.  Such an approach is akin to proposals by Legate () and Pesetsky
(), who argue for the necessity of both abstract and morphological case, where
morphological case is a post-syntactic realization of abstract case.
Given such assumptions, examples like () and (c) will both be well-formed with
respect to the Case Filter—the relevant NPs are licensed in the course of the deriva-
tion, and nothing goes wrong in the syntactic portion of the derivation, there being
no theta-criterion violation, unlike in the analysis of Bošković (). However, ()
and (c) both violate the PF-related requirement on the overt realization of inherent
case morphology in (), because the verb does not discharge its morphological case
feature onto its complement—this numeral NP is a morphologically caseless form.
On the other hand, if at PF the structure that contains the verb with the undischarged
feature is deleted by sluicing, the offending morphological case (affix) is eliminated,
and the sentence is rescued, as in (a–b).
In contrast to the situation with inherent case, it can be assumed that there is no
Inverse Case Filter for structural cases. If the verb is not specified with a morpholog-
ical case feature (i.e. it is not an inherent case-marking verb), then a morphological
reflex of structural case can be realized through PF rules that spell out accusative or
nominative, depending on the head that the structural case feature was associated
 There still should be a distinction between abstract structural and inherent case in narrow syntax, in
that with inherent cases, an NP must check its abstract case feature against an element that assigns it a
theta-role. I take it that in such cases an NP is marked with abstract inherent case. For more discussion, see
footnote .
Violation repair under sluicing 81

with in syntax. The rules may not enforce necessary overt realization of such case
features. As a result, a morphologically caseless form is possible in structural case
environments, as in (). Alternatively, the numeral forms such as pet ‘five’ and their
wh-counterpart koliko ‘how many’ may be ambiguous between caseless forms and
syncretic nominative/accusative forms,  as suggested in Bošković () and Franks
() with respect to similar examples in Russian. In the case of (), the acc spell-
out rule would apply, with koliko muškaraca ‘how many men’ and pet muškaraca ‘five
men’ thus being accusative. 

 If numeral phrases like pet tigrova ‘five tigers’ or koliko tigrova ‘how many tigers’ are ambiguous
between caseless and nominative/accusative forms, the question is whether the remnants in cases like ()
can be nominative, which might lead one to ask whether they can then be the remnants of pseudo-sluicing.
Recall, however, that pseudo-sluicing is not possible in such cases in Serbo-Croatian. First, there cannot be
nominative remnants whose correlates in the antecedent bear a different case:
(i) Petar je prišao nekome, ali ne znam ∗ ko / kome
Petar is approached someone-dat, but not I-know who-nom / whom-dat
‘Petar approached someone, but I don’t know who.’
Second, recall that if the wh-phrase in the second conjunct of (i) is nominative, but its correlate in the
antecedent bears a different case, there has to be a cleft clause. In other words, deleting the cleft clause is
not possible:
(ii) Petar je prišao nekome, ali ne znam ko ∗ (je to bio)
Petar is approached someone-dat, but not I-know who-nom is that been
‘Petar approached someone, but I don’t know who it was.’
However, examples like (i) give rise to a further question regarding the analysis proposed in this paper.
The question is: What prevents a derivation in which the verb in the second conjunct of (i) does not assign
a morphological dative feature to its wh-complement, the wh-phrase moves to SpecCP, and TP is deleted
at PF under sluicing together with the offending dative feature on the verb? In such a derivation, one might
expect that the wh-phrase could be spelled out as nominative or accusative by the accusative/nominative PF
spell-out rules, yet this is not possible and results in an ungrammatical output. As mentioned in footnote ,
and discussed in more detail in Stjepanović (in preparation), I still assume that there is a critical difference
between abstract structural and inherent case. In (i), the wh-phrase is marked with abstract inherent case.
I believe it is reasonable to assume that the nominative/accusative spell-out rules are incompatible with
abstract inherent case.
 Interestingly, as shown by Franks () and Bošković (), among others, numeral phrases such
as pet lavova ‘five lions’ and their wh-counterparts such as koliko lavova ‘how many lions’ are possible as
complements of prepositions that assign inherent case:
(i) prema jednom lavu / pet lavova / koliko lavova
towards one lion-dat / five lions-gen / how-many lions-gen
‘towards one lion/five lions/how many lions’
While there is currently no explanation of this difference in the behavior of numeral phrases as comple-
ments of inherent-case-marking verbs and prepositions, under the analysis proposed here, the difference
could be derived if prepositions themselves can be taken to be morphological realizations of case, as it has
sometimes been claimed for certain Ps cross-linguistically, such as English of (Chomsky b), Serbo-
Croatian sa ‘with’, which realizes instrumental case (Bošković ), and prepositions in Amharic (Baker
and Kramer ). If this is true, then under the analysis proposed here, NP complements of Ps would
satisfy the Case Filter in syntax by checking/valuing their abstract case feature against P, but since P realizes
case morphologically, the morphological case feature does not need to be passed onto N. As a result,
morphologically caseless NPs would be possible as complements of Ps.
82 Sandra Stjepanović

To sum up, assuming () allows one to account for the contrasts between examples
like () and (a). Since () is a requirement observed in non-sluiced contexts, but
is not necessary under sluicing, constructions like (a) fall into Class I of structures
subject to elliptical repair.

. Conclusion
In this chapter I have discussed two cases of violation repair under sluicing in Serbo-
Croatian, and have shown that they involve operations or constraints that are either
fed or bled by sluicing. One operation is P-drop, which may occur only in the presence
of sluicing. The other is the IICF constraint, which is formulated as a requirement that
inherent case be realized morphologically, if it can be. I have shown that this require-
ment must be obeyed in structures without sluicing, but may be obviated in structures
with sluicing. Thus, two new cases can be added to the taxonomy of elliptical repair
developed by Merchant (): the repair of the IICF in Serbo-Croatian as an instance
of Class I, and P-drop as an instance of Class II. In addition to documenting these two
interesting cases of violation repair under sluicing, the paper also has consequences
for the analysis of higher numeral phrases in Serbo-Croatian that contain the genitive
of quantification. It was shown that the impossibility of these phrases in inherent case
contexts is due to a morphological requirement of inherent case.

How many kinds of sluicing, and


why? Single and multiple sluicing in
Romanian, English, and Japanese
F R E DE R IC K HOY T A N D A L E X A N DR A T E OD OR E S C U

. Introduction
Romanian is a language which has “multiple-sluicing” constructions.  “Sluicing”, a
term coined by Ross (), refers to sentences in which the clausal sub-constituent
of a question is elided, leaving a “floating” wh-phrase (or “remnant”). For example,
in () the remnant is English who or its Romanian equivalent cine, and the ellipse is
understood as meaning ate my cookies:
() a. Someone ate my cookies, but I don’t know who.
b. Cineva mi-a mâncat prăjiturile, dar nu ştiu cine.
someone cl.sg-pst.sg eaten cookies-the but not know.sg who
‘Someone ate my cookies, but I don’t know who.’
Multiple-sluicing constructions have two or more remnants. While these are odd at
best in English, they are perfectly acceptable in Romanian:
() a. *Someone kissed someone, but I don’t know who whom.
b. Cineva a sărutat pe cineva, dar nu ştiu cine pe
someone pst.sg kissed acc someone but not know.sg who acc
cine.
who
‘Someone kissed someone, but I don’t know who (kissed) whom.’
 We thank the editors and other contributors to this volume, and in particular Jason Merchant and
Andrew Simpson, for their feedback. We also thank Bernhard Schwarz, Rajesh Bhatt, Danny Fox, Lisa
Green, Junko Shimoyama, Steve Wechsler, Jason Merchant, Virginia Hill, and various participants of
LSRL  for their comments on different stages of this work. Thanks also to Masa Deguchi, Hitoshi Hirioshi,
Makiko Irie, and Tomoko Sakuma for their help with Japanese data, and to Dan Tecuci for his assistance
with Romanian.
84 Frederick Hoyt & Alexandra Teodorescu

It has also been noted that Japanese allows multiple sluicing (Takahashi ,
Shimoyama , Merchant , Nishigauchi , Hiraiwa and Ishihara ):
() a. Taroo-ga dareka-ni nanika-o ageta rasii ga boku-wa
Taro-nom someone-dat something-acc gave heard but I-top
dare-ni nani-o da ka wakara-nai.
who-dat what-acc is Q know-not
‘I heard that Taro gave someone something, but I don’t know who what.’
This raises the question of whether Romanian and Japanese might have some syntactic
similarities which allow them both to generate multiple sluicing, a property which
English generally seems to lack.
English sluicing has been analyzed as ellipsis of the IP-constituent of a clause,
leaving a CP-projection containing a remnant (Lobeck , Merchant , ):
() a. Ann is marrying someone you know. Guess who.
b. . . . [CP who [IP she is marrying]].
Shimoyama (), Merchant (), and Hiraiwa and Ishihara () have argued
that while Japanese examples like () appear to resemble English sluicing examples
(a), they actually contain ellipsis of the CP-constituent in a cleft construction, rather
than of the IP-node of a matrix clause (b):
() a. Bill-ga nanika-o nusunda rasii kedo, . . .
Bill-nom something-acc stole seem but
‘It seems that Bill stole something, but . . . ’
b. . . . watashi-wa [[CP Bill-ga nusunda no] nani-o (da) ka] siranai.
I-top Bill-nom stole comp what-acc is Q know-not
‘. . . I don’t know what (it is that Bill stole).’
In addition to allowing multiple remnants, Romanian and Japanese sluicing have
other similarities that are unlike English: They both allow non-wh-remnants as well
as overt complementizers in the sluice. Given these similarities, it is natural to ask
whether Romanian and Japanese sluicing might have similar analyses.
The topic of this chapter is whether Romanian sluices like () are like English or
Japanese sluices in terms of their structure; in other words, whether (a) or (b) is a
more appropriate structural analysis for Romanian:
() a. Guess [QUESTION who [IP just left]].
b. Guess [CLEFT who [CP it was that just left]].
We argue that despite the superficial similarities between Romanian and Japanese
sluicing, an IP-ellipsis analysis of Romanian sluicing as in (a) is to be preferred.
We show that the similarities between Romanian and Japanese sluicing are epiphe-
nomenal and follow from independent syntactic properties of the two languages.
How many kinds of sluicing, and why? 85

An IP-ellipsis analysis, similar to ones that have been proposed for English, imme-
diately accounts for the key properties of Romanian sluicing. The differences between
Romanian and English sluicing follow from the presence of a richer structure in the
left periphery of embedded clauses in Romanian, a property of Romanian that has
been argued for independently.
The implication of our results is that the term “sluicing” as it has been used does
not describe a natural class of syntactic structures. Instead, “sluicing” describes a cor-
relation between certain ellipsis configurations, the forms of which vary in different
languages (consider () or () and a semantic interpretation, which is, as far as we
can tell, consistent across languages, as suggested by the glosses given in the exam-
ples above). Given “sluicing” as a general category, we distinguish between “English
sluicing”, “Romanian sluicing”, and “Japanese sluicing”. “Sluicing” therefore implies
nothing about the syntactic analysis for the data. Rather, we describe analyses in terms
of different kinds of ellipsis, such as “IP-ellipsis” or “CP-ellipsis”.
The chapter is organized as follows: In Section . we discuss the superficial similar-
ities between Romanian and Japanese sluicing, and how these superficial similarities
contrast with English sluicing. In Section . we discuss the crucial similarity between
English and Romanian sluicing: island insensitivity. In Section . we compare pos-
sible analyses for Romanian sluicing, and conclude that an IP-ellipsis analysis is the
only option. We provide evidence that supports such an analysis. In Section . we
conclude with a short discussion of the typological implications of our results.

. Similarities between Romanian and Japanese sluices


Romanian and Japanese sluicing sentences share several properties. These include:
multiple wh-remnants (Section ..); aggressively non-d-linked wh-remnants (Sec-
tion ..); overt complementizers (Section ..); and a variety of non-wh rem-
nants (Section ..), including referential, quantificational and free-choice items
(Shimoyama , Merchant ).
These are discussed in turn, and it is shown that English sluices either lack the
properties in question or exhibit them in restricted kinds of examples.

.. Multiple wh-remnants


Romanian (a) and Japanese (b) both allow multiple remnants in the sluice:

() a. Ion a dat cuiva ceva şi vreau să


Ion pst.sg given someone-dat something and want.sg subj
ştiu [cui] [ce].
know.sg whom-dat what
‘Ion gave something to someone, and I want to know whom what.’
86 Frederick Hoyt & Alexandra Teodorescu

b. Taroo-ga dareka-ni nanika-o ageta rasii ga boku-wa


Taro-nom someone-dat something-acc gave heard but I-top
[dare-ni] [nani-o] da ka wakara-nai.
who-dat what-acc is Q know-not
‘I heard that Taro gave someone something, but I don’t know whom what.’
By contrast, in English, multiple remnants are much more restricted and are fre-
quently found to be unacceptable: 
() a. * John gave someone something, and I want to know [who what].
b. ? John gave something to someone, but I don’t know [what to whom].
Multiple sluicing has been noted in a variety of other languages, including Turkish
(Inçe, this volume), Bangla and Hindi (Bhattacharya and Simpson, this volume), and
the Slavic languages (Rudin , Merchant , Grebenyova ).

.. Aggressively non-d-linked wh-remnants


Both Romanian and Japanese allow aggressively non-d-linked wh-words or phrases
(Pesetsky ) as remnants, such as Romanian ce dracu ‘what the devil’ or cine dracu
‘who the devil’ (a), and Japanese ittai nande ‘why the hell’ (b):
() a. Cineva mi-a ascuns cheile şi aş vrea să
someone cl.sg-pst.sg hidden keys-the and opt.sg want subj
ştiu şi eu [cine] [dracu].
know.sg emph I who devil-the
‘Someone hid my keys on me, and I’d like to know who-the-hell.’
b. Minna-ga awateteiru kedo, boku-wa [ittai] [nande] ka sirainai.
everyone-nom panic but I-top hell why Q know-not
‘Everyone is panicking, but I don’t know why-the-hell.’
English, in contrast, does not:
() ?? Someone ate my sandwich, and I would really like to know [who-the-hell].
Den Dikken and Giannakidou () have argued that English what-the-hell phrases
are a kind of negative polarity item. If their arguments extend to Romanian and
Japanese, then the generalization here would be that Romanian and Japanese both
allow negative polarity items as remnants, while English does not.

 Richards () notes that multiple sluicing becomes more acceptable in English if the remnants
are PPs and/or non-argumental. Richards likewise notes that the unacceptability of multiple remnants in
English can be mitigated by addition of a conjunction between the two remnants:

(i) John gave someone something, and I want to know who and what.
(ii) John gave something to someone, but I don’t know what or to whom.
How many kinds of sluicing, and why? 87

.. Overt complementizers


Both Japanese and Romanian tolerate an overt complementizer in the remnant of the
sluice. This complementizer can be either interrogative () or indicative ():

() Overt interrogative complementizer


a. Am aflat că cineva a plecat, dar nu ştiu
pst.sg learned that someone pst.sg left but not know.sg
[dacă] Ion.
if Ion
‘I found out that someone left, but I don’t know if Ion.’
b. John-ga dareka-o kubinisita rasii kedo, boku-wa Bill [ka
John-nom someone-acc fired seem but I-top Bill Q
dooka] siranai.
whether know-not
‘It seems that John fired someone, but I don’t know if Bill.’
() Overt indicative complementizer
a. Dan: Cine crezi că a câştigat premiul întâi?
who think.sg that pst.sg won prize-the first
‘Who do you think won first prize?’
Alex: Ştiam [că] Anca.
knew.sg that Anca
‘I know (that) Anca (did).’
b. John-ga dareka-o kubinisita rasii kedo, boku-wa Bill [to]
John-nom someone-acc fired seem but I-top Bill that
omou.
think
‘It seems that John fired someone, and I think that (it was) Bill.’
(Merchant : )

English in contrast does not tolerate overt complementizers in the remnant of a sluice:

() a. *One of the foreign students won the department fellowship, and I wonder
[whether/if] Louise.
b. One of the foreign students won the department fellowship, and I wonder
[if] it’s Louise.

.. (Multiple) non-wh remnants


Both Japanese and Romanian allow a variety of non-wh remnants, such as referential
NPs (), adverbs (), and PPs ():
88 Frederick Hoyt & Alexandra Teodorescu

() Referential NPs


a. Mi s-a spus că cineva s-a întâlnit cu
Me refl-pst.sg told that someone refl-pst.sg met with
cineva, şi mă întreb dacă [Ion cu Maria].
someone and me-refl wonder.sg if Ion with Maria
‘I was told that someone met with someone, and I wonder if Ion with Maria.’
b. John-ga dareka-o kubinisita rasii kedo, boku-wa [Bill-o]
John-nom someone-acc fired seems but I-top Bill-acc
to omou.
that think
‘It seems that John fired someone, and I think Bill.’
() Adverbs
a. Carmen vrea sa-şi ia maşină, şi suspectez că
Carmen wants subj-cl.sg take car and suspect.sg that
[repede].
quickly
‘Carmen wants to buy herself a car, and I suspect (that) pretty soon.’
b. Toto-wa kuruma-o kaitagatte iru, [suguni] da to omou.
Toto-top car-acc buy-want asp soon is that think
‘Toto wants to buy a car, (and) I suspect that soon.’
() Prepositional phrases
a. Da, am aflat şi eu că Ioana a fugit cu
Yes, pst.sg learned and I that Ioana pst.sg eloped with
cineva, dar n-aş paria că [cu Radu].
somebody but not-opt.sg bet that with Radu
‘Yes, I found out too that Ioana ran off with somebody, but I wouldn’t bet that
with Radu.’
b. Akiko-ga dareka-to kakeochisita to kiita kedo, [Taroo-to]
Akiko-nom someone-with eloped that heard but Taro-with
to-wa omowanakatta.
that-top not-expected
‘I heard that Akiko eloped with someone, but I didn’t expect that (it was)
Taro.’

In addition, Japanese and Romanian both allow a variety of strong quantifica-


tional NPs as remnants. Examples include Romanian toţi and Japanese minna (both
meaning ‘everyone’), and free-choice indefinites like Romanian oricine and Japanese
daremo (both meaning ‘anyone’):
How many kinds of sluicing, and why? 89

() Strong quantifiers (‘everyone’)


a. Da, e adevărat că mulţi au votat pentru Iliescu, dar nu
yes, is.sg true that many pst.pl voted for Iliescu but not
cred că [toţi].
believe.sg that everyone
‘Yes, it is true that many people voted for Iliescu, but I don’t believe that
everyone.’
b. Dareka-ga kono-kuruma-o naoseru to omou kedo, [minna-ga]
someone-nom this-car-acc can-fix that think but everyone
to-wa omowanai.
Q-top think-not
‘Someone can fix this car, but I don’t know if everyone.’
() Free-choice indefinites (‘anyone’)
a. Ştiu că profesorul ajută pe multă lume, dar mă
know.sg that professor-the helps acc many people but me-refl
întreb dacă [pe oricine oricând].
wonder if acc anyone anytime
‘I know that the professor helps many people, but I wonder if he helps anyone
anytime.’
b. Dareka-ga kono-kuruma-o naoseru to omou kedo, [daredemo]
someone-nom this-car-acc can-fix that think but anyone
ka-wa wakaranai.
Q-top know-not
‘Someone can fix your car, but I don’t know if (just) anybody.’

English, on the other hand, allows only wh-remnants. Overt complementizers and
non-wh-remnants of any category are degraded or unacceptable: 

() a. ??I heard that Mary is marrying someone I know; I wonder [if John].
b. ??Yes, I also heard that Ioana has eloped with somebody, but I wouldn’t bet
[that with Radu].

 Bhattacharya and Simpson (this volume) note that non-wh-remnants can be acceptable in Eng-
lish provided that no complementizer is present, and if the remnant is interpreted with contrastive
focus:
(i) It’s true that many people voted for Blair, but I don’t think EVERYONE.
(ii) Sue just left with someone, but I don’t think with YOUR date.
While we have little to say here about why non-wh remnants are less acceptable or at least less widely
used in English, we speculate that it may have to do with differences in how contrastive focus is
marked.
90 Frederick Hoyt & Alexandra Teodorescu

c. ??Hanako wants to buy a car, and I suspect [that soon].


d. ??Luis says that some people from our class cheated on the exam, but I don’t
think [everyone].
e. ??Someone can fix your car, but I don’t think [(just) ANYbody].

. Similarities between Romanian and English sluices


Despite these similarities between sluicing in Japanese and Romanian, the two lan-
guages differ in one crucial respect, namely that Japanese sluices are island-sensitive,
while Romanian ones are not. In this respect, Romanian is like English in allowing
remnants to be extracted out across island boundaries inside elided constituents in
violation of well-known island constraints such as Ross’s () coordinate structure
constraint, complex-NP constraint, relative-clause island constraint, sentential sub-
ject constraints, and adjunct constraint.
Parallel English and Romanian examples are given in () for coordinate-structure
islands, in () for complex-NP islands, in () for relative-clause islands, and in ()
for adjunct islands.

() Coordinate-structure island


a. He invited Akiko and someone else, but I don’t know who.
(cf. *I don’t know whoi he invited [COORD Akiko and t i ].)
b. Dan a invitat-o pe Anca şi pe încă cineva, dar
Dan pst.sg invited-cl.f.sg acc Anca and acc other someone but
nu ştiu pe cine.
not know.sg acc who
(cf. *Nu stiu [pe cine]i Dan a invitat-o
not know.sg acc who Dan pst.sg invited-cl.f.sg
[COORD pe Anca şi t i ].)
acc Anca and
‘Dan invited Anca and someone else, but I don’t know who.’
() Complex-NP island
a. Jerry heard a rumor that someone burnt the archive down, but I don’t know
who.
(cf. *I don’t know whoi Jerry heard [NP a rumor that t i burnt the archive
down].)
b. Emil a împrăştiat zvonul că cineva a dat foc
Emil pst.sg spread rumor-the that someone pst.sg given fire
arhivei şi sunt curioasă cine.
archive-dat and be.sg curious.FS who
How many kinds of sluicing, and why? 91

(cf. *Sunt curioasa cinei Emil a împrăştiat [NP zvonul


be.sg curious who Emil pst.sg spread rumor-the
că t i a dat foc arhivei].)
that pst.sg given fire archive-dat
‘Emil spread the rumor that someone’s set the archive on fire, and I wonder
who.’
() Relative-clause island

a. Ana drives a car that belongs to somebody else, but I don’t know who.
(cf. *I don’t know whoi Ana drives a car [REL that belongs to t i ].)
b. Ana conduce o maşină care este a altcuiva, dar nu
Ana drives a car that is gen somebody-else but not
ştiu a cui.
know.sg gen who
(cf. *Nu ştiu a cuii Ana conduce o maşină [REL care
not know.sg gen who Ana drives a car which
este t i ].)
is
‘Ana drives a car that is somebody else’s, but I don’t know whose.’
() Adjunct island

a. The victim left after one of the linguists, but I don’t know which.
(cf. *I don’t know whichi the victim left [ADJUNCT after t i ].)
b. Victima a plecat după unul dintre lingvişti, dar nu
victim-the pst.sg left after one of-the linguists but not
ştiu după care.
know.sg after which
(cf. *Nu ştiu [după care]i victima a plecat [t i ].)
not know.sg after which victim-the pst.sg left
‘The victim left after one of the linguists, but I don’t know which.’

Romanian sluices with non-wh remnants (–) show the same island insensitivity
as those with wh-remnants, indicating that a similar structure underlies the two
classes of example.

() Coordinate-structure island

Dan a invitat-o pe Anca şi pe înca cineva;


Dan pst.sg invited-cl.f.sg acc Anca and acc other someone
bănui că pe Elena.
suspect.sg that acc Elena
92 Frederick Hoyt & Alexandra Teodorescu

(cf. *Bănui că [pe Elena]i Dan a invitat-o


suspect.sg that acc Elena Dan pst.sg invited-cl.f.sg
[COORD pe Anca şi t i ].)
acc Anca and
‘Dan invited Anca and someone else; I suspect that Elena.’
() Complex-NP island
Emil a împrăştiat zvonul că cineva a dat foc
Emil pst.sg spread rumor-the that someone pst.sg given fire
arhivei, şi suspectez că George.
archive-dat and suspect that George
(cf. *Suspectez că Georgei Emil a împrăştiat [NP zvonul
Suspect that George Emil pst.sg spread rumor-the
că t i a dat foc arhivei]).
that pst.sg given fire archive-dat
‘Emil spread the rumor that someone set the archive on fire, and I suspect
that George.’
() Relative-clause island
Ana conduce o maşină care este a altcuiva, dar nu sunt
Ana drives a car that is gen somebody-else but not be.sg
sigur dacă a lui Şerban
certain if Şerban’s
(cf. *Nu sunt sigur dacă [a lui Şerban]i Ana conduce o maşină
not be.sg certain if Şerban’s Ana drives a car
[REL care este t i ].)
that is
‘Ana drives a car that is somebody else’s, but I’m not sure if Şerban’s.’
() Adjunct island
Victima a plecat după unul dintre lingvişti, dar nu
Victim-the pst.sg left after one of-the linguists but not
mi-e clar dacă după Mirel.
dat.sg-be.sg clear whether after Mirel
(cf. *Nu mi-e clar dacă [după Mirel]i victima
not dat.sg-be.sg clear whether after Mirel victim-the
a plecat [ADJUNCT t i ].)
pst.sg left
‘The victim left after one of the linguists, but it is not clear to me whether
after Mirel.’
How many kinds of sluicing, and why? 93

In contrast, (case-marked) remnants in Japanese sluices obey islands (Takahashi ,


Shimoyama , Merchant , Nishigauchi , Hiraiwa and Ishihara ). 
Japanese examples analogous to the English and Romanian ones in (–) are
degraded or unacceptable:

() Coordinate-structure island

??Taroo-wa Akiko-to dareka-o shootaisiita rasii kedo,


Taro-nom Akiko-and someone-acc invited seem but
watashi-wa dare-o ka siranai.
I-top who-acc Q know-not
‘It seems that Taro invited Akiko and someone, but I don’t know who.’
() Complex-NP island (Merchant )

*Taroo-ga Hanako-ga nanika-o katta toyuu uwasa-o


Taro-top Hanako-nom something-acc bought comp rumor-acc
sinjiteiru ga, watashi-wa nani ka siranai.
believe but I-top what Q know-not
‘Taro believes the rumor that Hanako bought something, but I don’t know
what.’
() Relative-clause island (Shimoyama )

*John-ga dareka-ga kaite-o sagasite iru rasii ga,


John-nom someone-nom painted-acc looking for seem but
boku-wa dare-ga ka siranai.
I-top who-nom Q know-not
‘It seems that John is looking for a picture that somebody painted, but I don’t
know who.’
() Adjunct island

*Taroo-wa dareka-ga gan kamoshirenai to-o kiita


Taro-top someone-nom cancer may-have that-acc hear-pst
naita kara ga, boku-wa dare-ga ka siranai.
because cry-pst but I-top who-nom Q know-not
‘Taro cried because he heard that someone might have cancer, but I don’t
know who.’

 For discussion of non-case-marked remnants in Japanese and island violations, see Fukaya and Hoji
(), Fukaya (this volume), and Nakamura (this volume).
94 Frederick Hoyt & Alexandra Teodorescu

The properties of English, Romanian, and Japanese sluicing are summarized in the
following table:

Aggressively
Overt Multiple Non-d-linked Non-wh Island
Language C wh-remnants wh-words remnants sensitivity

English No No No No No
Romanian Yes Yes Yes Yes No
Japanese Yes Yes Yes Yes Yes

. The syntax of Romanian sluices


We suggest that the following facts have to be accounted for in any analysis of
Romanian sluicing:

() a. Multiple remnants


b. Non-wh remnants
c. Overt complementizers
d. Island insensitivity

In this section we examine possible accounts for (). We review previous analyses of
sluicing in English and Japanese, and show that Romanian sluicing cannot be analyzed
in terms of the latter (Section ..). In Section .. we show that only an IP-ellipsis
analysis accounts for (). The mechanisms behind the analysis are described in
Section .... We discuss the implications for English in Section ...

.. Romanian sluicing as CP-ellipsis


Because (a–c) above are facts that Romanian has in common with Japanese, it seems
natural to try to extend the analysis of Japanese sluicing to Romanian. Shimoyama
(), Merchant (, ), and Hiraiwa and Ishihara () argue that Japanese
sluicing should be analyzed as ellipsis of the CP-constituent of a cleft structure. In
English, clefts consist of an expletive pronoun, a copular verb, an NP (the “focus”),
and a relative-clause-like constituent (the “presupposition”):

() a. It was John [CP who left early].


b. Show me [whati it is t i [CP that you like]].

In Japanese an overt expletive is absent, and the copular verb is only optionally
present:
How many kinds of sluicing, and why? 95

() Japanese clefts


a. John-ga dareka-o kubinisita rasii kedo, boku-wa
John-nom someone-acc fired heard but I-top
[CP John-ga t i kuninisita no] -wa Bill-oi (da) to omou.
John-nom fired comp -top Bill-acc is that think
‘I heard that John fired someone, and I think that it was Bill that he fired.’
b. Taroo-ga dareka-ni nanika-o ageta rasii ga, boku-wa
Taro-nom someone-dat something-acc gave heard but I-top
[CP Taroo-ga t i t j ageta no] -ga dare-nii nani-oj (da)
Taro-nom gave-pst comp -nom who-dat what-acc is
ka siranai.
Q know-not
‘I heard that Taro gave someone something, but I don’t know who what it
was that he gave.’
According to this analysis, the remnants of a Japanese sluice are the foci of the under-
lying cleft, and the elided CP is its presupposition:
() a. John-ga dareka-o kubinisita rasii kedo, boku-wa
John-nom someone-acc fired heard but I-top
[[CP John-ga t i kuninisita no]-wa Bill-oi (da) to] omou.
John-nom fired comp -top Bill-acc is that think
‘I heard that John fired someone, and I think that Bill.’
b. Taroo-ga dareka-ni nanika-o ageta rasii ga, boku-wa
Taro-nom someone-dat something-acc gave heard but I-top
[[CP Taroo-ga t i t j ageta no]-ga dare-nii nani-oj (da) ka]
Taro-nom gave-pst comp -nom who-dat what-acc is Q
siranai.
know-not
‘I heard that Taro gave someone something, but I don’t know who what is
was that he gave.’
As in English clefts, the focus of a Japanese cleft is in an island-sensitive dependency
with a variable within its presupposition. Furthermore, Japanese clefts allow multiple
pivots (Hiraiwa and Ishihara ):
() a. [CP Taroo-ga ageta no] -wa Hanako-ni ringo-o mit-tu da.
Taro-nom gave C -top Hanako-dat apple-acc three-cl cop
‘It is three apples to Hanako that Taro gave.’
b. [CP Hanako-ga sensei-ni [CP tabeta to] iituketa no] -wa
Hanako-nom teacher-dat ate C told C -top
Taroo-ga kono-ringo-o da.
Taro-nom this-apple-acc is
‘It is Taro, this apple that Hanako told the teacher that ate.’
96 Frederick Hoyt & Alexandra Teodorescu

As such, a CP-ellipsis analysis immediately explains the island-sensitivity of Japanese


clefts as well as the availability of multiple remnants.
However, Romanian lacks clefts with multiple pivots, and in fact may lack clefts
altogether (Dobrovie-Sorin , Grosu , Merchant ), so a cleft-reduction
analysis à la Japanese will not account for ().

() a. *E Maria (că) vreau să întâlnesc.


is Maria that want.sg subj meet.sg
‘It is Maria that I want to meet.’
b. *E Ion ce/care a câştigat premiul întâi.
is Ion that/who pst.sg won prize-the first
‘It is Ion that won the first prize.’
c. *E Ion pe care l-am întâlnit ieri.
is Ion acc who cl-pst.sg met yesterday
‘It is Ion who I met yesterday.’
(Grosu , cited in Merchant : )

If Romanian sluicing is not to be analyzed as CP-ellipsis, then it seems that it must


be analyzed as IP-ellipsis, like English sluicing. This would at least account for (d),
island insensitivity. However, as we have seen, Romanian differs from English in terms
of (a–c). In order to apply an IP-ellipsis analysis to Romanian sluicing, we need to
show either that such an analysis can be extended to cover these facts, or that (a–c)
follow from independent properties of Romanian syntax. In what follows, we argue
for the latter conclusion.

.. An IP-ellipsis account


Given that Romanian sluicing cannot be analyzed as CP-ellipsis, the remaining possi-
bility is that it is IP-ellipsis. The basic idea is that Romanian is like English, in that
sluicing is ellipsis of the clausal or propositional sub-constituent of an embedded
question. We assume Merchant’s () analysis of sluicing in English as a starting
point.
Merchant argues that IP-ellipsis is licensed under semantic rather than syntac-
tic identity. He accounts for the apparent island insensitivity of English sluicing by
arguing that English sluices actually contain no (syntactic) islands. He divides island
constraints into three classes, which he proceeds to explain away as being due to
pragmatic, phonological, and semantic constraints, respectively. This allows sluicing
to be uniformly explained as ellipsis of an IP-node with the remnants heading well-
formed A -chains rooted inside the ellipse. Semantic identity is enforced by the Focus
Condition, which requires that the set of alternative propositions presupposed by the
sluice entail its antecedent, and vice versa.
How many kinds of sluicing, and why? 97

A crucial element of this analysis is the argument that a sluice and its antecedent
have nearly-identical LFs, differing only in the form and indexing of the variables
they contain. Merchant assumes that focused constituents, like wh-words, undergo
quantifier raising, leaving traces inside the IP-node in which they originate. Traces
are interpreted as variables or E-type pronouns, which, despite being syntactically
different, can have equivalent interpretations.
For example, the sluice in (b), repeated here as (a), would be (b), with the trace
bound by the wh-word cine. The antecedent would be (c).
() a. [Cinevai [IP t i mi-a mâncat prăjiturile]], dar nu ştiu
someone me-aux.sg eat cookies-the but not know.sg
[cinej [IP t j mi-a mâncat prăjiturile]].
who me-aux.sg eat cookies-the
‘Someone ate my cookies but I don’t know who.’
b. [IP t j mi-a mâncat prăjiturile ]
c. [IP t i mi-a mâncat prăjiturile ]
The sluice in (b) and the antecedent in (c) differ only in the indices on the traces
in their subject positions, allowing the Focus Condition to be satisfied.
... Multiple wh-fronting Under the IP-ellipsis analysis, the availability of mul-
tiple wh-remnants in Romanian sluices follows directly from the fact that Romanian
is a multiple wh-fronting language (Rudin , Comorovski , Dobrovie-Sorin
, Alboiu ), a property it shares with other Balkan languages. Wh-items are
fronted in both main clauses (a) and embedded clauses (b):
() a. Cine pe cine a văzut?
who acc who pst.sg saw
‘Who saw whom?’
b. Chiar mă întrebam cine pe cine a intervievat.
actually refl ask.pst.sg who acc who aux interview
‘I was actually wondering myself who interviewed who.’
IP-ellipsis predicts this without further elaboration, since it would involve ellipsis of
the constituent(s) below the position occupied by the fronted wh-words. For example,
(a) above can be analyzed in terms of IP-ellipsis, assuming an LF-representation as
follows:
() Ion [cuivai cevaj [IP a dat t i t j ]], şi vreau
Ion someone-dat something pst.sg given and want.sg
să ştiu [cuii cej [IP a dat t i t j ]].
subj know.sg whom-dat what pst.sg given
‘Ion gave something to someone, and I want to know what to whom?’
98 Frederick Hoyt & Alexandra Teodorescu

On the other hand, if Romanian lacks clefts with multiple pivots, then one might
suggest that Romanian multiple sluicing constructions are a kind of gapping con-
struction. Like English, Romanian has gapping, and, as in English, it occurs in non-
interrogative clauses and involves multiple, non-wh remnants:

() a. Gabriela a comandat o margarita şi Mihai un Ursus.


Gabriela pst.sg ordered a margarita and Mihai an Ursus
‘Gabriela ordered a margarita and Mihai an Ursus.’
b. L-am văzut pe vărul meu la bibliotecă şi pe
cl.m.sg-pst.sg saw acc cousin-the my at library and acc
nevasta sa la magazin.
wife-the his at store
‘I saw my cousin at the library and his wife at the store.’

While there is still no consensus as to how gapping is to be analyzed, a gapping analysis


of the Romanian data would explain both the multiple remnants and the non-wh
phrases, because gapping applies to indicative clauses and leaves multiple remnants.
However, gapping is found in more restrictive syntactic contexts than sluicing is.
A sluice and its antecedent are both embedded within conjoined matrix clauses, while
gapping only occurs within local conjunctions (Johnson n.d.: , Romero : ):

() a. Andrei a luat cartea şi Marga atlasul.


Andrei pst.sg took book-the and Marga atlas-the
‘Andrei took the book and Marga the atlas.’
b. *(Cred că) Andrei a luat cartea şi cred
believe.sg that Andrei pst.sg took book-the and believe.sg
că Marga atlasul.
that Marga atlas-the
‘(I believe that) Andrei took the book and I believe that Marga the atlas.’
c. Cred că Andrei a luat cartea şi că Marga
believe.sg that Andrei pst.sg took book-the and that Marga
atlasul.
atlas-the
‘I believe that Andrei took the book and that Marga the atlas.’

In sluicing constructions, on the other hand, the sluice and its antecedent are typically
embedded inside other clauses, up to arbitrary levels of embedding:

() a. Cred că cineva a furat ceva, dar n-am


think.sg that someone pst.sg stole something, but not-have.sg
nici o idee cine sau ce.
any an idea who or what
‘?I think [that someone stole something], but I don’t know [who or what].’
How many kinds of sluicing, and why? 99

b. Am auzit [că George a spus [că cineva vrea


pst.sg heard that George pst.sg said that someone wants
să-l înşele pe prietenul lui]], dar nu cred [că
subj-cl.sg cheat acc friend-the his but not think.sg that
ştie [cine]].
know.sg who
‘I heard that George said that someone wants to cheat on his friend, but I
don’t think that he knows who.’

Another difference between gapping and sluicing is that sluicing allows “backwards
ellipsis”, meaning that the sluice precedes its antecedent in linear order:

() a. Nu ştiu CINE cu CINE, dar sunt sigur că toţi


not know.sg who with who, but be.sg sure that everyone
se vor combina cu cineva.
refl fut.sg combined with someone
‘I don’t know WHO with WHO, but I am sure that everyone will get
hooked up with someone.’

Gapping, on the other hand, does not allow backwards ellipsis:

() a. *Andrei cartea şi Marga a luat atlasul.


Andrei book-the and Marga pst.sg took atlas-the
‘Andrei the book and Marga took the atlas.’
b. *(Cred că) Andrei a luat cartea şi cred
believe.sg that Andrei pst.sg took book-the and believe.sg
că Marga atlasul.
that Marga atlas-the
‘(I believe that) Andrei took the book and I believe that Marga the atlas.’

Therefore a gapping analysis will not account for multiple-remnant sluicing in


Romanian.

... The structure of the remnant domain and the left periphery As we saw,
Romanian sluicing allows one or more non-wh remnants. Under an IP-ellipsis analy-
sis, this would follow from the presence of topicalization and focus-fronting in
Romanian main clauses (Dobrovie-Sorin , Motapanyane , Alboiu ,
Cornilescu ), as well as in embedded clauses (b):

() a. Nu s-a stabilit [dacă la Balcescu toţi


not impers-pst.sg established if at Balcescu all
profesorii sunt in grevă].
professors-the are on strike
‘It is not known whether all the professors at Balcescu are on strike.’
100 Frederick Hoyt & Alexandra Teodorescu

b. Nu ştiam [că pe MARIA a ales-o Ion].


not knew.sg that acc MARIA pst.sg chosen-cl.m.sg Ion
‘I didn’t know that Ion chose Maria (rather than Ileana).’
Topics and foci in embedded clauses are subject to the same ordering restrictions
that they are subject to in root clauses. First, topics must precede foci. Topics include
names, definite NPs, d-linked wh-words, and strong quantifiers:
() a. Mă întreb Ion cui o fi dat cartea?
refl.sg wonder.sg Ion who-dat might have given book-the
‘I have no clue who Ion might have given the book to.’
b. *Mă întreb cui Ion o fi dat cartea?
refl.sg wonder.sg who-dat Ion might have given book-the
c. Nu ştiu dacă primarul pe FLORIAN îl vrea.
not know.sg if mayor-the acc Florian cl.m.sg want.sg
‘I don’t know whether the mayor wants Florian (rather than Ion).’
d. *Nu ştiu dacă pe FLORIAN primarul îl vrea.
not know.sg if acc Florian mayor-the cl.m.sg want.sg
Second, in embedded clauses as in root clauses, foci must be immediately left-adjacent
to the tensed verb, and to the right of any topics. Foci include wh-words, polarity sensi-
tive quantifiers, negative polarity items, referential NPs pronounced with contrastive
focus, and aggressively non-d-linked wh-words. These different kinds of foci are in
complementary distribution with each other (), although multiple foci from one
particular class may occur together ():
() a. *Nu ştiu [pe cine nimeni n-a vrut să vad].
not know.sg acc who nobody not-pst.sg wanted subj see
‘I don’t know who nobody wanted to see.’
b. *ştii [cineva pe cine vroia să lovească]?
know.sg someone acc who wanted subj hit.sg
‘Do you know who somebody wanted to hit?’
c. *Mă întreb [unde MARIA trebuie să stea (şi nu
refl.sg ask.sg where Maria must.sg subj stay.sg and not
Ion)].
Ion
‘I don’t know where it is that MARIA has to stay (rather than Ion).’
d. *Nu mă îndoiesc [că MARIA cu nimic nu
not refl.sg doubt.sg that Maria with nothing not
te-a deranjat].
cl.sg-pst.sg bother
‘I don’t doubt that it was MARIA that didn’t bother you with anything.’
(Alboiu , adapted)
How many kinds of sluicing, and why? 101

() a. Ştii [cine ce a mâncat]?


know.sg who what pst.sg eaten
‘Do you know who ate what?’
b. Mă întreb [dacă nimeni cu nimic nu te va
refl.sg ask.sg if nobody with nothing not cl.sg fut.sg
ajuta].
help
‘I wonder if nobody is going to help you with anything.’
c. Nu mă îndoiesc [că cineva ceva va găsi de
not refl.sg doubt.sg that somebody something fut.sg find of
făcut].
done
‘I don’t doubt that somebody will find something to do.’
(Alboiu , adapted)

This shows us that the left-periphery of a Romanian embedded clause parallels the left-
periphery in root clauses. So, as before, an IP-ellipsis analysis of Romanian sluicing
directly predicts the presence of (possibly multiple) non-wh remnants, as these posi-
tions all c-command the ellided constituent itself. For example, (a) above is analyzed
as ():

() Mi s-a spus că [S cinevai [cu cineva]j [S t i


me refl-pst.sg tell that someone with someone
s-a întâlnit t j ]], mă întreb [CP dacă [S Ioni [cu
refl-pst.sg meet me-refl wonder if Ion with
Maria]j [t i s-a întâlnit t j ]]]
Maria refl-pst.sg meet
‘I was told that someone met with someone, and I wonder if Ion with
Maria.’

.. So, what about English?


Given our analysis, the fact that English allows only single wh-remnants can be
explained simply by the fact that English allows fronting of only one wh-word. How-
ever, English does allow topicalization and focus-fronting, if not to the same degree
as Romanian. The question is therefore: Why doesn’t English sluicing allow non-wh
remnants? If sluicing is simply a matter of IP-ellipsis, this should be possible.
One answer would be to follow Merchant () in assuming that Romanian
IP-ellipsis is subject to the Focus Condition as English is, but to parameterize the
syntactic licensing condition, which we call the “IP-ellipsis condition” (IPEC). For
English, the IPEC requires that the ellided constituent be sister to a [+Q, +wh]
102 Frederick Hoyt & Alexandra Teodorescu

complementizer. Romanian would have a more relaxed version of the IPEC, which
would allow IP-ellipsis under sisterhood with any complementizer other than a rela-
tive clause complementizer [+wh, −Q]. We refer to the English-type IPEC as “strong”
IPEC, and to the Romanian-type IPEC as “weak” IPEC:

() a. English:

+Q −Q

+wh ∅ relative-that
−wh if that

b. Romanian:
+Q −Q

+wh ∅ ∅
−wh dacă că

We are not currently aware of any independent motivation for this principle, so for
the time being we simply restate the facts.
Another answer might be that English verbs like know or wonder have more
restrictive semantics than their Romanian counterparts, disallowing topicalized con-
stituents in their complements. This might follow from the fact that although English
allows topics in root clauses, it does not allow them in embedded questions (Hudson
).
As far as Japanese goes, the IP-ellipis condition does not apply, because the rem-
nants in a Japanese sluice are simply the foci in a cleft, and there is no restriction that
we are aware of that requires cleft-foci to be wh-words. This is plain even in English
examples of cleft-reduction:

() a. Someone ate my cookies, and I want to know [CP which one of you it was
[CP who ate my cookies]]!
b. Someone ate my cookies, and I think [CP that it was John [CP who ate my
cookies]].

Example (a) is a reduced cleft within the context of an embedded question, giving
it a sluice-like appearance and semantics, without the actual syntax of a sluice, while
(b) is a reduced cleft in a non-question environment. The two examples indicate
that English clefts do not impose a wh-restriction on cleft-foci. The same seems to be
true in Japanese.
How many kinds of sluicing, and why? 103

. Discussion: A typology of sluicing


According to our discussion, sluicing constructions vary across languages according
to the type of ellipsis involved (CP-ellipsis vs IP-ellipsis), and the type of remnants
allowed in the sluice (weak vs strong ellipsis condition):

Language Ellipsis type Ellipsis condition

Romanian IP Weak
Japanese CP Weak
English IP, CP Strong

We have not included size of remnant set (singleton vs multiple) in this table because
we have observed that whether or not a language allows multiple remnants follows
from independent principles of the grammar. Romanian allows multiple remnants
because it also allows multiple wh-fronting. Japanese, on the other hand, allows mul-
tiple remnants because it also allows clefts with multiple foci. Therefore we conclude
that whether or not Romanian, Japanese, or English allows multiple remnants in a
sluice has nothing to do with properties of sluicing per se. Future research must deter-
mine whether any languages have multiple remnants with a strong IPEC, singleton
remnants with CP-ellipsis, or singleton remnants with a weak IPEC.
Secondly, we conclude that the term “sluicing” does not actually describe a syntactic
configuration at all. Instead, it seems to describe a correlation between the certain
kinds of word strings found in the languages we have looked at and kinds of inter-
pretations associated with those strings. Note that English, Romanian, and Japanese
sluices have comparable word strings in at least some cases (those involving singleton
wh-remnants), and, as far as we can tell, sluices in all three languages have comparable
semantics. However, the three kinds of sluicing differ syntactically, with Romanian
and English sluicing having one general kind of structure (IP-ellipsis) which is disjoint
with the structure of Japanese sluicing (CP-ellipsis). A subject for future research
is how to relate the differences in structural description that we have seen to the
similarities in interpretation.

Case morphology and island repair∗


M A S A NOR I NA KA M U R A

. Introduction
Surveying an impressive range of data from a variety of languages, Merchant ()
argues extensively that ellipsis like sluicing (Ross ) involves syntactic movement
followed by phonological deletion. One of his compelling arguments has to do with
morphological case—more specifically, the cross-linguistically robust requirement
that the sluiced wh-phrase must bear the case that its correlate bears (Merchant
(: ), cf. Ross ()). Merchant demonstrates that such case-matching cannot be
adequately handled by an analysis of the kind proposed by Chung et al. (), which
takes ellipsis to involve copying operations at LF.
It has been suggested in the literature that in certain languages, morphological case
is licensed in PF. For instance, building on Kuroda’s () insight, Harada () and
Fukui and Sakai () argue explicitly that case features in Japanese become visible
only after Spell-Out. If case in Japanese is indeed a PF phenomenon, then we would
expect it to interact with ellipsis, assuming that ellipsis does involve phonological
deletion. Under the LF analysis of ellipsis, we would expect it not to impact on ellipsis
in any way.
The main purpose of this chapter is to show that case-marked NPs behave dif-
ferently from non-case-marked PPs under ellipsis in Japanese. This state of affairs
is explicable if both the PF analysis of ellipsis and the PF analysis of case-marking
are correct. Then the discussion to follow constitutes an argument for these analyses
and against both the copy theory of ellipsis and the syntactic theory of Japanese case,
under which the observed difference remains a mystery.

∗ I would like to thank Moira Alvarez, Lucía Brandani, Marcela Depiante, Laura Kornfeld, and Pablo
Zdrojewski for the time and effort they put in to organize the  Workshop on Morphosyntax in
Buenos Aires, Argentina, where this paper was delivered. Many thanks are also due to the participants
in the workshop, especially Jason Merchant, an anonymous reviewer, and Andrew Simpson for valuable
comments and suggestions. As usual, grammaticality judgments reported here are relative rather than
absolute. Remaining errors are my own.
Case morphology and island repair 105

This paper is organized as follows. Section . lays out some background, on which
the succeeding discussion is based, and contains a brief summary of Merchant’s ()
analysis of locality under ellipsis. Section . introduces Hiraiwa and Ishihara’s ()
analysis, adopted here, of elliptical constructions in Japanese (sluicing and stripping
in particular). Section . presents representative data on locality under ellipsis in
Japanese. The key observation to be explained is that in Japanese island violations
caused by NPs apparently cannot be repaired by ellipsis, but those caused by PPs can.
Section . is an attempt to account for the observation. It is suggested that a mecha-
nism of cyclic case-marking in post-Spell-Out Morphology (Halle and Marantz ,
) along the lines of Kuroda (), coupled with the assumption that Morphology
exhibits the Anti-connectivity Effect—elements extracted out of an island cannot be
“reconstructed” to their original positions—captures the seemingly peculiar behavior
of focus-moved NPs with respect to island repair. Section . considers sentential
complements in Japanese, pointing out that declarative complements seem to pattern
with NPs, and interrogative ones with PPs, in terms of island repair. This discrepancy
is shown to be amenable to the present account. Section . concludes the discussion.

. Background: locality under ellipsis in English


Let us begin with a little bit of background on ellipsis using data from English. Both ()
and () are familiar examples of sluicing, which I assume involves IP deletion in PF
(elided constituents are indicated by strikethrough). 

() Jack bought something, but I don’t know [CP what [IP Jack bought t]].
() I believe that he bit someone, but they don’t know [CP who [IP I believe [that
he bit t]]].

The following example illustrates what has come to be known as “island repair”, where
wh-movement has taken place out of a complex NP island (Ross ), but ellipsis
saves the sentence (see Merchant () and Fox and Lasnik (), among others).

() They want to hire someone who speaks a Balkan language, but I don’t know
[CP which (Balkan language) [IP they want to hire someone who speaks t]].

One way to deal with the grammaticality of () is to say that the sentence does not
involve an island violation in the first place and thus is well-formed. Merchant ()
suggests that that is the case, giving () as a source of ().

() They want to hire someone who speaks a Balkan language, but I don’t know
which (Balkan language) she should speak.

 For discussions of the general licensing conditions on elliptical sites, see Lobeck () and Merchant
(), among others.
106 Masanori Nakamura

However, Lasnik () shows that island repair does take place with complex NP
islands, based on data like ().
() Every linguisti met a philosopher who criticized some of hisi work, but I’m
not sure [CP how much of hisi work [IP every linguisti met a philosopher who
criticized t]].
In () the wh-element contains a variable, his, bound by the subject universal quanti-
fier every linguist. The presence of the bound variable, which must reconstruct into a
position c-commanded by its binder at LF, ensures that the sluicing site contains the
complex NP island. Therefore, one can safely conclude that in () the wh-movement
takes place out of the island, but the violation is nullified by sluicing.
Nonetheless, at an observational level, ellipsis does not always nullify locality
violations because of interfering factors.  For concreteness, let us assume Mer-
chant’s () analysis of island repair. The central idea is that intermediate traces
of island-escaping XPs are PF-defective (cf. Chomsky ()). Given the assumption
that A -movement targets every intermediate maximal projection (Fox (), cf.
Chomsky (a)), it follows that island repair is observed only if ellipsis phonolog-
ically eliminates all the maximal projections with illegitimate traces lying between
the moved element and the island. This analysis can explain not only the grammat-
icality of () but also the ungrammaticality of (b) (see Hankamer (), Merchant
(, )).
() a. Does Abby speak the same Balkan language that Ben speaks?
b. ∗ No, [FP Charlie [CP ∗ t [IP Abby speaks the same Balkan language that t
speaks]]].
Example () is fine because all the defective traces created by the wh-movement are
eliminated in PF. Example (b) is an example of Hankamer’s () “wrong” trans-
formation, where the second speaker makes a correction to some aspect of the first
speaker’s utterance. In () what is corrected is the NP Ben within the island. Merchant
() suggests that in (b) the focused NP moves to what he agnostically calls SpecFP
above CP, and then the sentence undergoes ellipsis. If this is correct, the IP-ellipsis
necessarily leaves behind the CP projection containing a defective trace (indicated
by ∗ t), causing the ungrammaticality of (b). 
With this much of the background in mind, we now turn to Japanese data.

 VP-ellipsis in English is a case in point. See Merchant (, ) and Fox and Lasnik ().
 See Fox and Lasnik () for an interesting alternative analysis of (), which shares certain ideas with
Merchant’s (, ) account but differs from it in detail.
Case morphology and island repair 107

. Derivation of ellipsis in Japanese


First, let us see how ellipsis in Japanese is derived. In () is an example of sluicing in
Japanese (Takahashi () among numerous others).

() Taroo-ga nanika-o katta ga,


Taro-nom something-acc bought but
boku-wa [nani-o (da) ka] sira-nai.
I-top what-acc cop Q know-neg
‘Taro bought something, but I don’t know what.’

In the second conjunct in (), the embedded clause contains only the wh-phrase nani
‘what’ and the Q-marker ka, with the copula da being optional.  I follow Fukaya and
Hoji () in regarding sluicing as a variant of stripping in Japanese, as in ().

() A: Taroo-ga Alfa Romeo-o katta.


Taro-nom Alfa Romeo-acc bought
B: Boku-wa [Fiat-o (da) to] omotteita.
I-top Fiat-acc cop comp thought
‘A: Taro bought an Alfa Romeo. B: I thought that (it was) a Fiat (that he
bought).’

In the second sentence in (), only the focused non-wh-phrase and the complemen-
tizer to along with the optional copula appear in the embedded clause. I also assume
with Hiraiwa and Ishihara () that these constructions are related to the in situ
focus construction involving nominalization, given in (). 

() Taroo-ga Alfa Romeo-o katta no da.


Taro-nom Alfa Romeo-acc bought nmlz cop
‘It is an Alfa Romeo that Taro bought.’

In sluicing, the focused wh-phrase moves to Spec of Focus Phrase (Rizzi ), headed
by the copula, and the nominalized complement of the Focus head undergoes dele-
tion. The phonological realization of the copula is optional. The tree in () illustrates
the relevant portion of the derivation of () (Hiraiwa and Ishihara ). 

 The case-marker on the wh-phrase in the sluiced clause is optional in (). Non-case-marked sluicing
(and its stripping counterpart; see below) will not be discussed here. Fukaya and Hoji () and Hiraiwa
and Ishihara () point out that non-case-marked ellipsis does not exhibit any island effects. Fukaya and
Hoji () suggest that it does not involve any movement (cf. Hiraiwa and Ishihara (: n)). See also
Nishiyama , Kizu  for relevant discussion.
 The possible connection between syntactic focus and nominalization in Japanese is pointed out by
Sakai ().
 See Kim () for an analysis of Japanese (and Korean) sluicing similar to Hiraiwa and Ishihara’s. If
CP counts as a phase in Japanese, the movement in () must proceed via the edge of CP. See below.
108 Masanori Nakamura

() FocP

NP Foc⬘

nani-o CP Foc

IP C (da)

Taro-ga t katta no

The stripping in () involves a derivation analogous to (), as in ().


() Boku-wa [FocP Fiat-o [CP Taroo-ga t katta no] (da)] to omotteita.
As in the case of the wh-phrase in (), the focused phrase in () undergoes movement
into Spec of FocP, which is followed by phonological deletion.
In short, the relevant elliptical constructions in Japanese involve syntactic deriva-
tions similar to those involved in sluicing in English. The category deleted under
sluicing and stripping is CP in Japanese, whereas the one deleted under sluicing and
“wrong” transformation is IP in English. Focus movement is assumed to target Spec
of FocP (Merchant’s () FP) in both languages (possibly in all languages). 

. Locality under ellipsis in Japanese


Now let us consider some representative data on locality under ellipsis in Japanese.
As has been noted in the literature (Fukaya and Hoji (), among others), focus-
moved NPs are not eligible for island repair in Japanese. This is shown in (). 

 Merchant () says in passing that “we may suspect that FP is to be identified with Rizzi’s 
FocusP”.
 In the Japanese examples on island repair to follow, I will use zibun ‘self ’ as a bound variable to make
sure that islands are crossed by focus movement. Recall the above remarks surrounding English ().
An anonymous reviewer says, “the sluice with the nominative -ga is ungrammatical regardless of island
violations (unless it is coupled with other wh-phrases, like the author’s ()).” Thus for the reviewer, the
following is not acceptable.

(i) Dareka-ga sono e-o rakusatusita ga,


someone-nom the painting-acc made a successful bid but
‘Someone made a successful bid for the painting, but . . .’
boku-wa dare-ga (da) ka siranai.
I-top who-nom cop Q know-neg
‘. . . I don’t know who.’
Case morphology and island repair 109

() Daremo-ga [[zibun-no sinseki-ga mita] hito]-o sagasiteiru


everyone-nom self-gen relative-nom saw person-acc looking.for
ga,
but
‘Everyonei is looking for a person who hisi relative saw, but . . .’
∗ boku-wa [[dono zibun-no sinseki-ga] [[[daremo-ga t mita] hito]
I-top which self-gen relative-nom everyone-nom saw person
-o sagasiteiru no] (da)] ka sira-nai.
-acc looking.for nmlz cop Q know-neg
(‘. . . I don’t know which relative of hisi .’)
Above, the subject wh-phrase dono zibun-no sinseki-ga ‘which relative of his’ has been
focus-moved out of the complex NP island headed by hito ‘person.’ Unlike in English
cases such as () and (), ellipsis does not save (). Example () makes the same
point.
() Daremo-ga [[zibun-no sinseki-o mikaketa] hito]-ni atta ga,
everyone-nom self-gen relative-acc came.across person-dat met but
‘Everyonei met a person who came across hisi relative, but . . .’

There are speakers, including myself, who find examples like the one above grammatical (although the
omission of the copula da may lead to slight degradation or awkwardness for reasons unclear to me). I do
not have an account of the variation in judgment (neither do Hiraiwa and Ishihara (), as far as I can see).
The reviewer also points out that there may be an intervening processing factor with respect to the
possibility of island repair. Consider the following example, where both the matrix and embedded verbs
take accusative arguments.

(ii) Daremo-ga [[zibun-no sinseki-o mita] hito] -o sagasiteiru ga,


everyone-nom self-gen relative-acc saw person -acc looking.for but
‘Everyonei is looking for a person who saw hisi relative, but . . .’
∗ boku-wa [[dono zibun-no sinseki-o] [[[daremo-ga t mita] hito] -o
I-top which self-gen relative-acc everyone-nom saw person -acc
sagasiteiru no] (da)] ka sira-nai.
looking.for nmlz cop Q know-neg
(‘. . . I don’t know which relative of hisi .’)
This example, like (), is a case of non-repair. The reviewer mentions that she or he tends to misinterpret
the accusative remnant dono zibun-no sinseki-o ‘which relative of his’ as the argument of the matrix verb
sagasiteiru ‘looking for’, which would lead to semantic anomaly.
In the examples of (non-)repair in Japanese to follow, this potential processing difficulty is controlled
for: The remnant of sluicing or stripping will bear a case or a postposition different from the case born by
the matrix argument.
110 Masanori Nakamura

∗ boku-wa [[dono zibun-no sinseki-o] [[[daremo-ga t mikaketa]


I-top which self-gen relative-acc everyone-nom came.across
hito]-ni atta no] (da)] ka sira-nai.
person-dat met nmlz cop Q know-neg
(‘. . . I don’t know which relative of hisi .’)
Here what has been extracted out of the island is the object wh-phrase, and again
island repair does not seem to apply.
The ungrammaticality of () and (), in sharp contrast with the grammaticality
of their English counterparts, raises the intriguing question of why there should be
a cross-linguistic difference in terms of phonological deletion, which is supposed to
behave in more or less the same way (Fox and Lasnik ).
Faced with () and (), one may think that island repair is never found in Japanese.
Examples like () demonstrate that this line of thinking is incorrect:
() Daremo-ga [[zibun-no iken-to kotonaru] syuchoo]-o hihansita
everyone-nom self-gen opinion-with differ claim-acc criticized
ga,
but
‘Everyonei criticized a claim which contrasted with hisi opinion, but . . .’

boku-wa [[dono zibun-no iken-to] [daremo-ga [[t kotonaru]


I-top which self-gen opinion-with everyone-nom differ
syuchoo]-o hihansita no] (da)] ka sira-nai.
claim-acc criticized nmlz cop Q know-neg
‘. . . I don’t know with which opinion of hisi .’
In () it is the argument PP (postpositional phrase) dono zibun-no iken-to ‘with which
opinion of his’ rather than an NP that has undergone focus movement out of an island
in the second conjunct.  Unlike in () and (), ellipsis does rescue the derivation
from a violation in (). Consider next the minimal pair in ().

 For the sake of concreteness, let us adopt the following notion of argumenthood, taken from Schütze
and Gibson (: ):
“If a phrase P is an argument of a head H, P fills a role in the relation described by H, the presence of which
may be implied by H. P’s contribution to the meaning of the sentence is a function of that role and hence
depends on the particular identity of H”.
See Schütze and Gibson () for some diagnostics for PP arguments in English, though they are not
necessarily applicable to Japanese.
Case morphology and island repair 111

() a. Daremo-ga [[zibun-no bokujyoo-o deta] uma]-ni aitagatteiru


everyone-nom self-gen ranch-acc got.out horse-dat want.to.meet
ga,
but
‘Everyonei wants to meet a horse which got out of hisi ranch, but . . .’
∗ boku-wa [[dono zibun-no bokujyoo-o] [daremo-ga [[t deta]
I-top which self-gen ranch-acc everyone-nom got.out
uma]-ni aitagatteiru no] (da)] ka sira-nai.
horse-dat want.to.meet nmlz cop Q know-neg
(‘. . . I don’t know which ranch of hisi .’)
b. Daremo-ga [[zibun-no bokujyoo-kara deta] uma]-ni
everyone-nom self-gen ranch-from got.out horse-dat
aitagatteiru ga,
want.to.meet but
‘Everyonei wants to meet a horse which got out of hisi ranch, but . . .’
boku-wa [[dono zibun-no bokujyoo-kara] [daremo-ga [[t deta]
I-top which self-gen ranch-from everyone-nom got.out
uma]-ni aitagatteiru no] (da)] ka sira-nai.
horse-dat want.to.meet nmlz cop Q know-neg
‘. . . I don’t know from which ranch of hisi .’

The verb deru ‘get out’ can take either an accusative NP or a source PP as its comple-
ment. Thus the sole difference between (a) and (b) lies in the categorial status of
what has been extracted out of the island: In (a), as in () and (), it is an NP that
has focus-moved, whereas in (b), as in (), it is a PP that has. This seemingly minute
discrepancy influences the results: (a) is ill-formed, whereas (b) is well-formed.
Data like () indicate that the NP–PP distinction is the determining factor. 
The Japanese data in (–) may be reminiscent of the contrast we saw earlier
between grammatical () and ungrammatical () in English. Two possible accounts
of the Japanese data, alluded to in Merchant (), come to mind.
First, one could argue that the NP cases are like those of “wrong” transformation
and are illegitimate for some locality reason, and that the PP cases do not involve
movement at all. Under this account, we would expect that PP focusing never exhibits
island effects. This expectation, however, is not fulfilled, as shown in ().

 I leave out discussion of remnants marked with the dative particle ni, which can be categorially
ambiguous between NP and PP (Sadakane and Koizumi ). We could in principle tease ni-marked NP
and ni-marked PP apart on the basis of proposed diagnostics (see Miyagawa (: Chapter ) for one such),
but I have chosen to concentrate on solid cases where the question of the categorial status of focus-moved
elements will not arise.
112 Masanori Nakamura

() Daremo-ga [[zibun-no hema-de awateta] kaisya]-o yameta


everyone-nom self-gen blunder-for panicked company-acc quit
ga,
but
‘Everyonei quit a company which panicked for hisi blunder, but . . .’
∗ boku-wa [[dono zibun-no hema-de] [daremo-ga [[t awateta]
I-top which self-gen blunder-for everyone-nom panicked
kaisya]-o yameta no] (da)] ka sira-nai.
company-acc quit nmlz cop Q know-neg
(‘. . . I don’t know for which blunder of hisi ,’)
Now, () is similar to () in that a PP has been extracted out of an island, yet it
is not an argument but an adverbial: The focus-moved phrase is a reason adjunct.
Unlike (), () is excluded. If we assume that PPs do undergo focus movement,
a unified treatment becomes available: The contrast between () and () reduces
to the familiar argument–adjunct asymmetry. Extending Lasnik and Saito’s ()
analysis, Lasnik () suggests that adverbial wh-phrases are subject to locality con-
ditions not only in syntax but also at LF (cf. Reinhart ()). Given this suggestion,
() is ruled out because the island intervenes between the head and the tail of the
wh-chain at LF. So the hypothesis that PP cases do not involve movement seems
untenable.
Second, one could maintain that NP focusing and PP focusing have different land-
ing sites. In particular, it may be suggested that PP focusing targets a position higher
than the one targeted by NP focusing. If that is the case, the account of the difference
between sluicing and “wrong” transformation in English would automatically extend
to the Japanese data.
Besides being theoretically undesirable, such an assumption is not empirically
justifiable. If it is true that focus-moved NPs occupy positions higher than those
occupied by focus-moved PPs, we would predict that when an NP and a PP undergo
focus movement simultaneously, the former should always precede the latter. This
prediction is false, as shown in ().
() a. Hanako-ga [dareka-ga dareka-to atta to] itta ga,
Hanako-nom someone-nom someone-with met comp said but
‘Hanako said that someone met with someone, but . . .’
boku-wa [dare-ga dare-to [Hanako-ga [t t atta to] itta no]
I-top who-nom who-with Hanako-nom met comp said nmlz
(da) ka sira-nai.
cop Q know-neg
Lit. ‘. . . I don’t know who with whom.’
Case morphology and island repair 113

b. Hanako-ga [dareka-to dareka-ga atta to] itta ga,


Hanako-nom someone-with someone-nom met comp said but
‘Hanako said that with someone, someone met, but . . .’
boku-wa [dare-to dare-ga [Hanako-ga [t t atta to] itta no]
I-top who-with who-nom Hanako-nom met comp said nmlz
(da) ka sira-nai.
cop Q know-neg
Lit. ‘. . . I don’t know with whom who.’

The first conjunct of (a) contains the canonical word order, in which the nominative
indefinite precedes the comitative one within the embedded clause. In the second
conjunct of (a), both of their corresponding wh-phrases undergo focus movement
and the word order is retained. On the other hand, the antecedent clause of (b)
involves scrambling of the comitative indefinite over the nominative one. In the fol-
lowing elided clause, the two wh-remnants keep the reversed order produced by the
scrambling, with the PP preceding the NP. From the viewpoint of Richards’ ()
“tuck-in” derivation forced by his notion of Shortest, the preservation of word order
strongly suggests that the focused elements move into the same projection, i.e. FocP,
creating multiple specifiers.
To summarize, it has been shown that island repair in Japanese is sensitive to
the distinction between NP focusing and PP focusing: Apparently, island violations
incurred by NPs cannot be repaired, whereas those incurred by PPs can. Two possible
attempts to capture the distinction, based on alleged non-movement in PP focusing
and variation in landing sites, have been considered and eventually rejected.

. Proposal
In light of the hypothesis that there is no cross-linguistic difference in terms of
what ellipsis can do to save otherwise illegitimate derivations (cf. Fox and Las-
nik ()), I assume that in both NP and PP focus movement, repair of island
violations in syntax does take place in Japanese. Then what is ultimately respon-
sible for the ill-formedness of examples such as () and ()? I suggest capi-
talizing on the fundamental difference between NP and PP: (Certain) NPs need
Case/case, whereas PPs do not. The analysis to be presented treats morphological
case in Japanese as the culprit, and hence is in full accordance with principles-and-
parameters practice, which tries to reduce cross-linguistic variations to differences in
morphology.
Kuroda () argues that morphological case-marking in Japanese is determined
by the linear order of NPs reflecting their base positions. His original formulation,
dubbed “Linear Case Marking”, is given in () (Kuroda : ).
114 Masanori Nakamura

() Linear Case Marking:


Attach ga to the first unmarked noun phrase in the sentence; if the sentence still
contains an unmarked noun phrase, attach o to that noun phrase.
Linear Case Marking applies in a cyclic fashion from the most deeply embedded
clause to the root clause. Under Kuroda’s system, NPs are introduced into derivations
“unmarked” or bare. The “first” or leftmost NP is assigned nominative case ga, and if
there is any NP left in the cycle, it is assigned accusative case o. 
Based on Kuroda’s idea, Harada () and Fukui and Sakai () explicitly argue
that case-marking in Japanese takes place in PF.  As a piece of evidence for their
claim, Fukui and Sakai () present the following examples of coordination.
() a. [[Zimintoo-kara gaimudaizin-ni Yamada-si]-to
LDP-from minister.of.foreign.affairs-dat Yamada-Mr./Ms.-and
[Hosyutoo-kara zaimudaizin-ni Suzuki-si](-to)]-ga
CP-from minister.of.finance-dat Suzuki-Mr./Ms.(-and)-nom
syuuninsita.
assumed
Lit. ‘[From the Liberal Democratic Party, Mr./Ms. Yamada (assumed) the
minister of foreign affairs] and [from the Conservative Party, Mr./Ms. Suzuki
assumed the minister of finance].’
b. ∗ [Gaimudaizin-ni Yamada-si-ga Zimintoo]-to
minister.of.foreign.affairs-dat Yamada-Mr./Ms.-nom LDP-and
[zimudaizin-ni Suzuki-si-ga Hosyutoo](-to)-kara
minister.of.finance-dat Suzuki-Mr./Ms.-nom CP-(and)-from
syuuninsita.
assumed
Lit. ‘[Mr./Ms. Yamada (from) the Liberal Democratic Party (assumed) the
minister of foreign affairs] and [Mr./Ms. Suzuki from the Conservative Party
assumed the minister of finance].’
Details aside, Fukui and Sakai show that in well-formed (a) the nominative case-
marker ga attaches to a non-constituent in syntax, which indicates that case-marking
in Japanese is not a matter of syntax but a matter of PF. In marked contrast, (b)

 See Kuroda () and his subsequent works for a full picture of how exactly () interacts with other
conventions to yield observed case-marking patterns in Japanese.
 It should be mentioned that Harada () and Fukui and Sakai () make claims that are incom-
patible with the present work. For example, they hold that Japanese lacks syntactic movement altogether,
contrary to Hiraiwa and Ishihara’s () analysis adopted here. Furthermore, Harada () argues that
her Modified Linear Case Marking targets the root node first and then works its way down to the bottom.
I assume with Kuroda that case-marking is a bottom-up process. For Harada, NPs bear morphological
case features that are only visible in PF. When talking about Japanese case, I use terms like “assignment”,
“marking”, and “realization” interchangeably.
Case morphology and island repair 115

is ruled out because the postposition kara ‘from’ attaches to a non-constituent, but
postpositions can take only syntactic constituents as their complements.
Here I adopt a particular view on (), taking into account certain innovations in
recent linguistic theorizing (see also Harada () for an alternative view). First,
derivations proceed phase by phase (Chomsky , ). The operation Spell-Out
is cyclic in nature and applies to a derivation as soon as each phase is completed. For
present purposes, let us assume that the phase in Japanese is CP (“the sentence” in
()), and when it becomes complete, the complement of the phase head C, namely
IP, is sent to the PF branch of grammar, leaving the material in the edge of the
phase, if any, still syntactically accessible (Hiraiwa (), cf. Chomsky’s (, )
Phase Impenetrability Condition).  The adoption of cyclic Spell-Out guarantees that
in the typical cases under present consideration, case-marking targets NPs in their
θ-positions.
Second, the post-Spell-Out branch of the derivation contains a component called
Morphology (Halle and Marantz , ), where case morphemes, I assume, are
added to NPs in Japanese. Morphology inherits hierarchical structures assembled in
syntax, but may alter them by morphological operations. Importantly, these oper-
ations are known to respect locality principles. Phonological operations including
ellipsis apply to the output of Morphology to yield PF representations.
Given that case-marking in Japanese takes place in Morphology and is essentially
a PF-related phenomenon, it is expected to interact with PF processes such as ellipsis.
What goes wrong in (), (), and (a), I suggest, is case-marking on the focus-
moved NPs. In order for case-marking on focus-moved NPs to be successful, they
must retain appropriate connections with their base positions: The tail of an argument
chain is case-marked (in keeping with ()) and the case feature is transferred to each
member of the chain.  Let us call this operation “case transfer”. As an operation
in Morphology, it should be subject to locality considerations, island constraints in
particular. We know independently from examples like () that islands can disrupt
connections of the relevant kind (Longobardi ).

() a. What does John think that every student bought t?


b. ??What does John believe the claim that every student bought t?

Example (a) involves no island and is ambiguous, with either the quantified expres-
sion every student or the wh-phrase what taking scope over the other. On the other
hand, the degraded (b), where the surface position of the wh-phrase and its under-
lying position are separated by an island, permits only one interpretation, with the

 Harada () claims that TP (=IP) constitutes a phase in Japanese.


 Working within a copy theory of ellipsis, Fukaya and Hoji (: footnote ) express the same kind of
intuition. They remark, “We assume that what makes it necessary for the CM[case-marked]-construction
to have the empty IP structure is the ‘licensing’ of the case-marker on the NP. Briefly put, we assume that
the case-marked NP is interpreted by being ‘connected’ to a position within the θ-domain of a verb.”
116 Masanori Nakamura

wh-phrase taking scope over every student. In other words, the wh-phrase cannot be
“reconstructed” into its thematic position.
Although what the scope facts in () show is that an island-violating NP loses con-
nectivity to its base position in LF, let us hypothesize that it does, too, in Morphology,
where case assignment takes place. Given this hypothesis, we can begin to understand
the difference in grammaticality between () and (), whose rough derivations are
illustrated below. 
() In Syntax and Morphology:
a. [Taroo nani katta] (= ())
Taro what bought
b. nanii [Taroo nanii katta] no
what Taro what bought nmlz (movement to the edge of CP)
c. nanii [Taroo-ga nanii -o katta] no
what Taro-nom what-acc bought nmlz (spell-out and
case-marking)
d. nanii nanii [Taroo-ga nanii -o katta] no (da)
what what Taro-nom what-acc bought nmlz cop (movement to
Spec of FocP)
e. [boku-wa nanii -o nanii -o [Taroo-ga nanii -o katta] no
I-top what-acc what-acc Taro-nom what-acc bought nmlz
(da) ka sira-nai]
cop Q know-neg (spell-out and case transfer)
In PF:
f. [boku-wa nanii -o nanii -o [Taroo-ga nanii -o katta] no (da)
I-top what-acc what-acc Taro-nom what-acc bought nmlz cop
ka sira-nai]
Q know-neg (ellipsis)
() In Syntax and Morphology:

a. [pro dono zibun sinseki mikaketa] (= ())


which self relative came.across
b. dono zibun sinsekii [pro dono zibun sinsekii mikaketa]
which self relative which self relative came.across
(movement to the edge of CP)
c. dono zibun sinsekii [pro-ga dono zibun-no sinsekii -o
which self relative -nom which self-gen relative-acc
mikaketa]
came.across
(spell-out and case-marking)
 Material already sent to Morphology is indicated by shading. It is assumed here that there is a general
rule in Morphology that assigns the genitive marker no to elements such as NPs within an NP. I remain
uncommitted as to the exact nature of wa-marking on topic phrases.
Case morphology and island repair 117

d. [daremo dono zibun sinseki [pro-ga dono zibun-no


everyone which self relative -nom which self-gen
sinseki-o mikaketa] hito-ni atta]
relative-acc came.across person-dat met
e. dono zibun sinsekii [daremo dono zibun sinsekii [pro-ga
which self relative everyone which self relative -nom
dono zibun-no sinsekii -o mikaketa] hito-ni atta] no
which self-gen relative-acc came.across person-dat met nmlz
(movement to the edge of CP)
f. dono zibun sinsekii [daremo-ga dono zibun-no sinsekii -o
which self relative everyone-nom which self-gen relative-acc
[pro-ga dono zibun-no sinsekii -o mikaketa] hito-ni
-nom which self-gen relative-acc came.across person-dat
atta] no
met nmlz
(spell-out, case-marking, and case transfer)
g. dono zibun sinsekii dono zibun sinsekii [daremo-ga dono
which self relative which self relative everyone-nom which
zibun-no sinsekii -o [pro-ga dono zibun-no sinsekii -o
self-gen relative-acc -nom which self-gen relative-acc
mikaketa] hito-ni atta] no (da)
came.across person-dat met nmlz cop (movement to Spec of FocP)
h. [boku-wa dono zibun-no sinsekii dono zibun-no sinsekii
I-top which self-gen relative which self-gen relative
[daremo-ga dono zibun-no sinsekii -o [pro-ga dono
everyone-nom which self-gen relative-acc -nom which
zibun-no sinsekii -o mikaketa] hito-ni atta] no (da) ka
self-gen relative-acc came.across person-dat met nmlz cop Q
sira-nai]
know-neg
(spell-out and failed case transfer)

In PF:

i. ∗ [boku-wa dono zibun-no sinsekii dono zibun-no sinsekii


I-top which self-gen relative which self-gen relative
[daremo-ga dono zibun-no sinsekii -o [pro-ga dono zibun-no
everyone-nom which self-gen relative-acc -nom which self-gen
sinsekii -o mikaketa] hito-ni atta] no (da) ka sira-nai]
relative-acc came.across person-dat met nmlz cop Q know-neg
(ellipsis)
118 Masanori Nakamura

Let us look at () first. The derivation starts out with (a), where the NPs in
the embedded clause lack case morphology. In (b) the wh-phase moves to the
edge of CP. This is immediately followed by the spell-out of IP and case-marking of
the NPs within IP in Morphology, as in (c). In (d) the moved wh-phase bearing
no case remains syntactically active (because it has not been sent to Morphology yet)
and undergoes further movement to Spec of FocP. Then the matrix clause gets spelled
out and the accusative morpheme is added to the moved wh-phase by case transfer in
Morphology, as in (e). Finally, the embedded CP is phonologically deleted in PF to
yield well-formed (f).
Turning now to (), we can see that it shares similar derivational properties with
(), but the crucial difference has to do with (h). Case transfer is banned there
because the focus movement has extracted the wh-phase out of the island. If (h) is
correct, the subsequent ellipsis in PF in (i) can in no way save the derivation, leav-
ing the focus-moved wh-phrase without case morphology. In short, () is doomed,
resulting in a violation of the Japanese version of the Case Filter. 
The well-formedness of () and (b) is expected under the present account. Since
what undergoes movement is a PP, case is never an issue. Although the movement
violates an island constraint in syntax, the violation is nullified by ellipsis. 
Admittedly, it is far from clear why movement in syntax and case transfer in Mor-
phology should differ in terms of island repair. The difference can be highlighted
by comparing (b) and (). Unlike (b), () with sluicing does permit the recon-
struction of the wh-phrase into the island, indicating that anti-connectivity produced
by an island-violating syntactic movement can indeed be lifted by ellipsis. Whatever
the exact reason, we must ascribe it to some special property of Morphology. One
possibility, endorsed above, is that locality constraints are part of the definition of the
operation of case transfer itself. Another possibility is simply that ellipsis cannot save
violations incurred after Spell-Out. I will have to leave this as an open question.

. An extension
The preceding discussion centered around the behavior of NP and PP arguments
with respect to island repair. At this point you may wonder how clausal complements
behave in terms of locality under ellipsis: The literature abounds in arguments for the
parallels between NP and CP, and some researchers, including Lamontagne and Travis
(), claim that case is the nominal counterpart of C.

 Case-markers can sometimes be dropped, but NPs like those used in () must bear a case morpheme
when focus-moved. See () below.
 Obviously, the case-based analysis has implications for the issue of locality of syntactic movement
without ellipsis in Japanese, but I will not delve into it here.
Case morphology and island repair 119

There are certain similarities between case-markers on the one hand and so-
called complementizers on the other in Japanese. One of them has to do with the
(im)possibility of dropping these particles. Observe ().

() a. Taroo-ga hon(-o) katta.


Taro-nom book-acc bought
‘Taro bought a book.’
b. Hon∗ (-o) Taroo-ga katta no da.
book-acc Taro-nom bought nmlz cop
‘(It was) a book (that) Taro bought.’

As shown in (a), the accusative case-marker can be omitted when adjacent to the
verb. However, it cannot be deleted when it undergoes focus movement.
Although the complementizer to in the Tokyo dialect can never be dropped, its
counterpart te in the Osaka dialect can. Consider the following: 

() a. Hanako-ga [Taroo-ga takoyaki-o koota (te)] yuuta.


Hanako-nom Taro-nom octopus ball-acc bought comp said
‘Hanako said (that) Taro bought octopus balls.’
b. [Taroo-ga takoyaki-o koota ∗ (te)] Hanako-ga yuuta-n
Taro-nom octopus ball-acc bought comp Hanako-nom said-nmlz
ya.
cop
Lit. ‘(It was) that Taro bought octopus balls (that) Hanako said.’
(Osaka dialect)

In (a), where the sentential complement is adjacent to the matrix verb, the com-
plementizer is optional. In (b), where the entire sentential complement has focus-
moved, it is obligatory. The parallel between () and () is straightforward. Let us
then adopt () along the lines of ().

() Attach to/te to unmarked clausal complements.

As with case-markers, complementizers are assumed to be realized in the PF com-


ponent. In the Tokyo dialect to must always be phonologically realized on bare non-
interrogative sentential complements.
In interrogative embedded clauses one can never drop the Q-marker ka. This is true
even in the Osaka dialect, as shown below:

() a. Hanako-ga [Taroo-ga takoyaki-o koota ∗ (ka)] kiita.


Hanako-nom Taro-nom octopus ball-acc bought Q asked
‘Hanako asked whether Taro bought octopus balls.’

 Thanks to Takako Kawasaki (personal communication) for her judgments on the Osaka dialect.
120 Masanori Nakamura

b. [Taroo-ga takoyaki-o koota ∗ (ka)] Hanako-ga kiita-n


Taro-nom octopus ball-acc bought Q Hanako-nom asked-nmlz
ya.
cop
Lit. ‘(It was) whether Taro bought octopus balls (that) Hanako asked.’
(Osaka dialect)

The presence of the particle ka is necessary both in (a) with the canonical word
order and in (b) with the fronting of the embedded clause.
Given (), the present analysis predicts that declarative sentential complements
accompanied by complementizers should pattern with NPs accompanied by case-
markers in terms of island repair: they should fail to exhibit amelioration effects. On
the other hand, interrogative complements, like argument PPs, should be eligible for
repair because they involve no PF morphology. Although the contrast between ()
and () involving stripping may be subtle, it seems to go in the expected direction. 

() A: Daremo-ga [[[zibun-no hahaoya-ga doroboo da to]


everyone-nom self-gen mother-nom thief cop comp
kaita] sinbunsya]-o uttaeta.
wrote newspaper-acc sued
‘Everyonei sued a newspaper which wrote [that hisi mother was a thief].’

 An anonymous reviewer claims that the ungrammaticality of () is due to the illegitimate combination
of to and da, citing the following examples of clefting:

(i) a. ∗ [Bob-ga omotteiru no]-wa [yasai-ga takai to] da.


Bob-nom think nmlz-top vegetable-nom expensive comp cop
(Lit. ‘It is that vegetables are expensive that Bob thinks.’)
b. [Bob-ga tazuneta no]-wa [yasai-ga takai ka] da.
Bob-nom asked nmlz-top vegetable-nom expensive Q cop
‘It was whether vegetables are expensive that Bob asked.’

I fully agree with the reviewer’s judgments here, but for speakers like me, the to-da sequence does not
always yield bad results. For instance, (ii) involving stripping is acceptable with the presence of da (in fact,
the version without da containing the to-to sequence sounds markedly unnatural if one does not put a pause
between the two instances of to):

(ii) A: Daremo-ga [[[zibun-no titioya-ga kibisii to] itta.


everyone-nom self-gen father-nom strict comp said
‘Everyonei said hisi father was strict.’
B: Boku-wa [zubun-no hahaoya-ga yasasii to] [daremo-ga t itta no] (da) to
I-top self-gen mother-nom sweet comp everyone-nom said nmlz cop comp
omotteita.
thought
‘I thought (it was) that his mother was sweet.’

This indicates that the ill-formedness of (), especially with da, cannot be ascribed to the alleged general
ban on the to-da sequence.
Case morphology and island repair 121

B: ∗ Boku-wa [[zibun-no titioya-ga yakuza da to]


I-top self-gen father-nom gangster cop comp
[daremo-ga [[t kaita] sinbunsya]-o uttaeta no] (da)] to
everyone-nom wrote newspaper-acc sued nmlz cop comp
omotteita.
thought
Lit. ‘I thought (it was) [that hisi father was a gangster].’
() A: Daremo-ga [[[zibun-no hahaoya-ga doroboo da ka]
everyone-nom self-gen mother-nom thief cop Q
tazuneta] sinbunsya]-o uttaeta.
asked newspaper-acc sued
‘Everyonei sued a newspaper which asked [whether hisi mother was a
thief].’
B: Boku-wa [[zibun-no titioya-ga yakuza da ka] [daremo-ga
I-top self-gen father-nom gangster cop Q everyone-nom
[[t tazuneta] sinbunsya]-o uttaeta no] (da)] to omotteita.
asked newspaper-acc sued nmlz cop comp thought
Lit. ‘I thought (it was) [whether hisi father was a gangster].’
Recall that sluicing and stripping in Japanese are different sides of the same coin: The
only difference is that wh-phrases are focus-moved in the former, whereas non-wh-
phrases are focus-moved in the latter. Thus in both of the second sentences in () and
(), the entire embedded clauses have been extracted out of an island. But it appears
that ellipsis saves only (). As in the case of NPs focus-moved out of an island, it
is assumed that the locality violation in syntax in () is indeed remedied. The ill-
formedness in () results from the impossibility of morphologically realizing to: The
transfer of to to the focus-moved sentential complement is blocked by the island in
Morphology, leading to a “COMP Filter” violation.
To wrap up this section, I have suggested that the analysis of (failure of) ellipsis
repair motivated by the behavior of NP/PP arguments can be extended to capture the
behavior of declarative/interrogative clausal complements.

. Implications and conclusion


Before closing the present discussion, let me mention a few implications of the analysis
offered here for the theory of grammar in general and for the theory of Japanese
grammar in particular.
To the extent that the analysis is on the right track, it lends additional support to
the deletion theory of ellipsis, advocated by Merchant () and others. In order to
come to grips with the apparent impossibility of island repair with focused NPs in
122 Masanori Nakamura

Japanese, it has been crucial to assume that those NPs undergo syntactic movement
out of an island, as argued by Hiraiwa and Ishihara () (see also Kim ()). It
has been suggested that the seeming absence of island repair is due to the failure of
morphological case assignment after Spell-Out. One might try to modify this case-
based analysis in such a way that it will be compatible with the LF copy theory
of ellipsis of the kind advanced by Chung et al. () (see also Fukaya and Hoji
()). The theory would have to say that a problem with case arises after the copying
process, duplicating the relevant structure including the island at LF. This, however,
runs counter to the independently motivated claim that case assignment in Japanese
takes place in the PF branch of the derivation (Harada , Fukui and Sakai ).
Therefore, the deletion theory is superior to the copy theory.
The success of the present analysis counts as an argument for the parametric vari-
ation between English-type languages, where structural Case checking takes place in
syntax, and Japanese-type ones, where morphological case assignment takes place
in Morphology. Unlike in Japanese, wh-arguments bearing Case do exhibit island
repair in English, indicating that Case is checked properly in syntax. The NP–PP
distinction in Japanese ellipsis noted above supports the existence of the variation
in a way that supplements the arguments presented by Harada () and Fukui and
Sakai ().
In addition, the hypothesis that all forms of phonological deletion can in principle
nullify island violations in syntax (Fox and Lasnik ) receives empirical support.
One big problem Japanese poses for the hypothesis, namely, the fact that focus-moved
NPs appear to exhibit no island repair, has been dealt with in terms of the parametric
property of morphological case assignment in the language. As we saw above, PPs
are certainly eligible for island repair even in Japanese, suggesting the universality of
deletion mending violations in syntax.
Starting with the brief discussion of English ellipsis, this paper has addressed the
issue of locality in Japanese elliptical constructions, namely, sluicing and stripping.
It has been shown that there are interesting interlinguistic, as well as intralinguistic,
differences in the relevant empirical domain. An analysis of these differences has
been presented using some tools from the recent developments in the principles-and-
parameters approach. The proposed analysis has the virtue of neither parametrizing
LF properties nor compromising the hypothesis that deletion of any sort has the
potential to repair island violations, a welcome result from the perspective of language
acquisition; after all, children are supposed to have no positive evidence regarding
silence. It is hoped that this work has contributed to the ongoing discussion of syntax–
phonology interaction, which has now become one of the most fruitful areas of lin-
guistic research.

Island-sensitivity in Japanese sluicing


and some implications∗
T E RU H I KO F U KAYA

. Introduction
Ross () was the first to bring to light the construction that he referred to as
“sluicing”, as in the second conjunct in (b).
() a. Somebody just left—guess who just left.
b. Somebody just left—guess who. Ross (: )
For ease of reference, I will refer to the wh-phrase in sluicing as the “remnant” and to
the element in the first conjunct that corresponds to the remnant as the “correlate”. In
(b), for example, somebody is the correlate, and who is the remnant.
Ross (: –) observed that the island effects are weaker in sluicing than in
its non-elliptical counterpart, as in (–).
() [Complex NP Constraint (relative clause)]
a. ∗ She kissed a man who bit one of my friends, but Tom doesn’t realize which
one of his friends she kissed a man who bit.
b. ?She kissed a man who bit one of my friends, but Tom doesn’t realize which
one of his friends. (Ross : (a–b))

∗ This chapter is based on Chapters  and  of Fukaya , which are extensions of Fukaya . I would
like to thank Hajime Hoji for his extensive comments and suggestions at various stages of this chapter.
I am also grateful to Joseph Aoun, Hagit Borer, Kiyoko Kataoka, Audrey Li, Jason Merchant, Yasuhiko
Miura, Emi Mukai, Chris Potts, Barry Schein, Yukinori Takubo, and Ayumi Ueyama for their feedback
on (portions of) its earlier versions. I would also like to thank Mark Freiermuth for his help with stylistic
improvement. All remaining errors are of course my own.
124 Teruhiko Fukaya

() [Sentential Subject Constraint]


a. ∗ That he’ll hire someone is possible, but I won’t divulge who that he’ll hire is
possible.
b. ??That he’ll hire someone is possible, but I won’t divulge who.
(Ross : (b))
() [Coordinate Structure Constraint]
a. ∗ Irv and someone were dancing together, but I don’t know who Irv and were
dancing together.
b. ??Irv and someone were dancing together, but I don’t know who.
(Ross : ())
Observing these facts, Ross gave the informal statement regarding island effects in
sluicing in ().
() If a node is moved out of its island, an ungrammatical sentence will result. If the
island-forming node does not appear in surface structure, violations of lesser
severity will (in general) ensue. (Ross : ())
Although Ross maintained that English sluicing still exhibits some island effects, it
has been the general consensus in the literature that island effects are not observed
in English sluicing (see Chomsky (), Levin (), Chung et al. (), Merchant
(, ), Lasnik (), and Fox and Lasnik (), among others). Examples ()
and () are taken from Merchant’s works.
() [Complex NP Constraint (relative clause)]
a. ∗ They want to hire someone who speaks a Balkan language, but I don’t
remember which Balkan language they want to hire [someone who speaks_].
b. They want to hire someone who speaks a Balkan language, but I don’t remem-
ber which. (Merchant : Chapter  ())
() [Adjunct Clause Constraint]
a. ∗ Ben will be mad if Abby talks to one of the teachers, but she couldn’t remem-
ber which (of the teachers) Ben will be mad [if she talks to _].
b. Ben will be mad if Abby talks to one of the teachers, but she couldn’t remem-
ber which. (Merchant : (a–b))
Schematically, if the structure in () is obtained in the first conjunct and sluicing is
acceptable, it has been considered as evidence that there are no island effects.
() . . . [ISLAND . . . correlate . . .] . . .
Following Ross () many researchers have maintained that the remnant has under-
gone regular wh-movement.
Island-sensitivity in Japanese sluicing 125

The issue of why island effects are nullified in sluicing has been one of the prominent
topics in the literature on sluicing. Chung et al. () propose an analysis where
there is no movement involved in the derivation of sluicing; Merchant () pursues
an account in which only non-island-violating local movement is involved in the
derivation of sluicing in island contexts; Fox and Lasnik () and Merchant ()
propose that syntactic islands are PF phenomena and that deletion of the violating
structure ameliorates the island violation. Kimura () argues that the remnant
wh-phrase stays in situ in some cases of sluicing.
In this paper, I will first examine case-marked sluicing in Japanese, in which the
remnant is marked with a case-marker or a postposition, and demonstrate that despite
appearing otherwise, it is in fact sensitive to the relative clause and the adjunct islands.
I will then show that the copy theory of ellipsis resolution proposed in Fukaya and Hoji
() combined with a version of Merchant’s (: Chapter ) proposal for local
movement in sluicing in propositional island contexts can account for what appears
to be the peculiar behavior of Japanese sluicing.  I will also investigate non-case-
marked sluicing in Japanese, where the remnant is not marked with a case-marker or
a postposition, and argue that its apparent island-insensitivity is due to the availability
of a copula structure with an empty pro as the subject. I will then revisit English
sluicing and demonstrate that it also exhibits relative-clause island effects and suggest
that the relative-clause island is not a PF representational island in English, contra the
recent view. To account for the new facts of English sluicing discussed in this paper, it
will be suggested that two types of sluicing be distinguished in English, corresponding
to case-marked and non-case-marked sluicing in Japanese.
The paper is organized as follows. Section . observes the instances of island-
sensitivity in case-marked sluicing in Japanese. Section . introduces Fukaya and
Hoji’s () theory of ellipsis resolution adopted in this paper. Section . examines
how the facts observed in Section . are accounted for under the theory of ellipsis
resolution in Section . combined with a version of Merchant’s () proposal for
local movement. Section . discusses island-sensitivity in non-case-marked sluicing
in Japanese and shows how the facts are accounted for under the copula analysis.
Section . re-examines English sluicing from the perspective gained from the exam-
ination of Japanese sluicing. Section . concludes the paper.

. Island-sensitivity in Japanese sluicing


In this section I will show that Japanese sluicing respects syntactic islands, although
in some cases it appears to be insensitive to them as in the case of English sluicing.
I will examine two types of islands for that purpose: the relative-clause island and the

 Merchant () refers to the types of island that involve a proposition (for example, the relative clause
and the adjunct clause) as “propositional islands”.
126 Teruhiko Fukaya

adjunct island. Before discussing island effects, let us first give an overview of sluicing
in Japanese with respect to the types of elements that are allowed as the correlate.

.. Japanese sluicing


Inoue (, ) first observed that Japanese has a construction similar to English
sluicing as in (); (a) is a simplex sentence example, and (b) is an embedded sen-
tence example. 

() a. John-wa dareka-o suisensita ga, boku-wa [dare-o ka]


John-top someone-acc recommended but I-top who-acc Q
siranai.
know.not
‘John recommended someone, but I don’t know who <John recom-
mended _>.’
b. John-wa Mary-ga dareka-o suisensita to itteita ga,
John-top Mary-nom someone-acc recommended that said but
boku-wa dare-o ka oboeteinai.
I-top who-acc Q remember.not
‘John said that Mary recommended someone, but I don’t remember who
<John said that Mary recommended _>.’ (Inoue : )

These are cases where the correlate is an indefinite; however, a definite can also be
the correlate in Japanese sluicing, as in the case of what Merchant (: , )
refers to as contrast sluicing in English. In (a) we have an example where a simplex
sentence appears to be missing, while (b) is an example where a complex sentence
appears to be missing. 

() a. boku-wa [[Yamada Kensetu]-ga [Tanaka giin] -ni


I-top Yamada Construction Co.-nom Tanaka rep. -to
wairo-o okutta] no-wa sitteiru ga, [[hoka-no dono giin] -ni
bribe-acc gave that-top know but other-gen which rep. -to
ka] wa siranai.
Q top know.not

 The underline indicates that the phrase is the correlate, and the portion within angle brackets in the
translation indicates the intended interpretation of the ellipsis.
 In (b) and (b), I use remnants marked with a postposition to ensure that the remnants correspond
to elements within the embedded clauses, because Japanese allows so-called Major Objects as the under-
lined element in (i); see Ueyama (: Appendix A.) for relevant discussion.

(i) Susan-wa John-o [tensai da to] omotteiru.


Susan-top John-acc genius cop that think
‘Susan thinks of John that he is a genius.’
Island-sensitivity in Japanese sluicing 127

‘I know that the Yamada Construction Company gave a bribe to represen-


tative Tanaka, but I don’t know to which other representative <the Yamada
Construction Company gave a bribe _>.’
b. boku-wa [keisatu-ga [[Yamada Kensetu]-ga [Tanaka giin] -ni
I-top police-nom Yamada Construction Co.-nom Tanaka rep. -to
wairo-o okutta] to happyoosita no] -wa sitteiru ga, [[hoka-no
bribe-acc gave that announced that -top know but other-gen
dono giin] -ni ka] wa siranai.
which rep. -to Q top know.not
‘I know that the police announced that the Yamada Construction Company
gave a bribe to representative Tanaka, but I don’t know to which other rep-
resentative <the police announced that the Yamada Construction Company
gave a bribe _>.’

In addition to indefinite and definite NPs, in situ wh-phrases can also be the correlate
in Japanese sluicing as in (). In (a) we have a case where a simplex sentence appears
to be missing, while (b) is a case where a complex sentence appears to be missing.

() a. boku-wa [Suzuki-ga [zimintoo-no dono giin] -ni wairo-o okutta


I-top Suzuki-nom L.D.P.-gen which rep. -to bribe-acc gave
ka] sitteiru ga [[minsyutoo-no dono giin] -ni ka] wa siranai.
Q know but D.P.-gen which rep. -to Q top know.not
‘I know to which representative of the Liberal Democratic Party Suzuki gave
a bribe, but I don’t know to which representative of the Democratic Party
<Suzuki gave a bribe _>.’
b. boku-wa [keisatu-ga [Suzuki-ga [zimintoo-no dono giin] -ni
I-top police-nom Suzuki-nom L.D.P.-gen which rep. -to
wairo-o okutta] to happyoosita ka] sitteiru ga, [[minsyutoo-no
bribe-acc gave that announced Q know but D.P.-gen
dono giin] -ni ka] wa siranai.
which rep. -to Q top know.not
‘I know to which representative of the Liberal Democratic Party the police
announced that Suzuki gave a bribe, but I don’t know to which represen-
tative of the Democratic Party <the police announced that Suzuki gave a
bribe _>.’

The above has illustrated that (i) an indefinite, (ii) a definite, and (iii) an in situ
wh-phrase can be a correlate in Japanese sluicing. Next, we turn to island-sensitivity
in Japanese sluicing. In the following discussion, correlates of the types (ii) and (iii)
will play a crucial role because only with these correlate types can the existence of
island effects be demonstrated.
128 Teruhiko Fukaya

.. The relative-clause island


Let us begin with the relative-clause island. Prior to discussing sluicing, however, I will
first demonstrate that relative clauses are indeed syntactic islands in Japanese. First,
consider ().
() [[kanari-no kazu]-no nikkei kigyoo]i -ga [seihu-ga
a.large.number-gen Japanese company -nom the.government-nom
[sokoi -no kogaisya] -ni aturyoku-o kaketeiru] to happyoo sita.
that.place-gen subsidiary -to pressure-acc is.putting that announced
Lit. ‘[A large number of Japanese companies]i announced that the government
is putting pressure on itsi subsidiary.’
The QP in the matrix subject position ([[kanari-no kazu]-no nikkei kigyoo]-ga
‘a large number of Japanese companies’) can enter into bound-variable anaphora
with the dependent term (soko ‘it’) within the embedded object, giving rise to
the covariant reading “For each of the large number of Japanese companies x, x
announced that the government is putting pressure on x’s subsidiary.” As illustrated
in (), the NP [sokoi -no kogaisya]-ni ‘to its subsidiary’ within the embedded clause
in () can be moved to the front of the matrix subject, retaining the covariant
reading.
() [sokoi -no kogaisya] -ni [[kanari-no kazu]-no nikkei kigyoo]i
that.place-gen subsidiary -to a.large.number-gen Japanese company
-ga [seihu-ga _ aturyoku-o kaketeiru] to happyoo sita.
-nom the.government pressure-acc is:putting that announced
Lit. ‘On itsi subsidiary [a large number of Japanese companies]i announced that
the government is putting pressure _.’
Next, consider (), where the QP ([[kanari-no kazu]-no nikkei kigyoo]-ga ‘a large
number of Japanese companies’) resides in the matrix subject position and the phrase
containing the dependent term (soko ‘it’) sits within the matrix object complex NP.
() [[kanari-no kazu]-no nikkei kigyoo]i -ga
a.large.number -gen Japanese company -nom
[COMPLEX NP [sokoi -no kogaisya] -ni [kekkan buhin]-o
that.place-gen subsidiary -to defective parts-acc
noonyuu siteita meekaa] -o uttaeta.
was.supplying maker -acc sued
Lit. ‘[A large number of Japanese companies]i sued a manufacturer that had
been supplying defective parts to itsi subsidiary.’
In this case as well, the QP and the dependent term can enter into bound-variable
anaphora, giving rise to the covariant reading given in ().
Island-sensitivity in Japanese sluicing 129

() For each of the large number of Japanese companies x, x sued a manufacturer
who had been supplying defective parts to x’s subsidiary.
In contrast to (), the NP [sokoi -no kogaisya]-ni ‘to its subsidiary’ within the object
complex NP in () cannot be moved to the front of the matrix-clause subject and still
retain the covariant reading in (). This is illustrated in ().
() ∗ [soko
i -no kogaisya] -ni [[kanari-no kazu]-no
that.place-gen subsidiary -to a.large.number-gen
nikkei kigyoo]i -ga [COMPLEX NP _ kekkan buhin-o
Japanese company -nom defective parts-acc
noonyuu siteita meekaa] -o uttaeta.
was.supplying maker -acc sued
Lit. ‘To itsi subsidiary, [a large number of Japanese companies]i sued a manu-
facturer that had been supplying defective parts _.’
The unacceptability of () with the intended covariant reading contrasts with the
acceptability of (). The difference between () and () is that the original site of
the fronted NP is within the embedded clause in the former and within the relative
clause in the latter. This shows that the dislocated element cannot be related to the
underscored position within the relative clause, and I claim that this indicates that
the relative clause is indeed a syntactic island in Japanese. Note that () does not
simply involve an island that contains the “gap” associated with the “dislocated” phrase
outside it: It involves reconstruction effects of binding. This is to ensure that the
dislocated element is related by means of movement to the position marked by an
underscore that is within the relative clause. 
Now let us return to sluicing. If we follow the criterion for island-insensitivity
in sluicing mentioned in Section ., i.e. whether sluicing is possible when the first
conjunct is of the structure in (), sluicing in Japanese appears not to exhibit island
effects. Consider ().
() keisatu-wa [ISLAND [proi Los Angeles-de [aru yuumee zin] -ni mayaku-o
police-top Los Angeles-in a celebrity -to drug-acc
utta] otokoi ] -o taihosita rasii ga, boku-wa [dare-ni ka] siranai.
sold man -acc arrested seem but I-top who-to Q know.not
‘I heard the police arrested a man who had sold drugs to a celebrity in LA, but
I don’t know to whom.’
In (), the correlate is located within a relative-clause island; nevertheless, sluicing
is possible. This indicates that Japanese sluicing with an indefinite correlate does not
 The complication is invoked in light of the observation made by Emi Mukai in her unpublished
paper (Mukai ) that island effects tend not to show up very clearly in the absence of reconstruc-
tion effects of binding, scope dependency, etc., which has also been supported by on-line experimental
results.
130 Teruhiko Fukaya

exhibit island effects.  In sluicing with a definite correlate, island effects do not seem
to be observed either, as can be seen in () and ().
() [keisatu-wa [ISLAND [proi [Tanaka giin]-ni wairo-o okutta] otokoi ] -o
police-top Tanaka rep.-to bribe-acc gave man -acc
taihosita ga, Bill-wa [dono giin-ni ka] siranakatta rasii.
arrested but Bill-top which rep.-to Q knew.not seem
‘The police arrested the man who had given a bribe to representative Tanaka,
but it seems that Bill didn’t know to which representative.’
() boku-wa keisatu-ga [ISLAND [proi [Tanaka giin]-ni wairo-o okutta]
I-top police-nom Tanaka rep.-to bribe-acc gave
otokoi ] -o taihosita no-wa sitteiru ga, [[hoka-no dono giin] -ni
man -acc arrested that-top know but other-gen which rep. -to
ka] -wa siranai.
Q -top know.not
‘I know that the police arrested the man who had given a bribe to representative
Tanaka, but I don’t know to which other representative.’
Sluicing with a wh-phrase correlate does not seem to exhibit island effects either, as is
observed in () and ().
() boku-wa [[ISLAND [sensyuu proi [dono giin] -o hihansita]
I-top last.week which congressman -acc criticized
sinbunsyai ] -ga boikotto sareteiru ka] sitteiru ga, Bill-wa [[dono
newspaper -nom is.being.boycotted Q know but Bill-top which
giin] -o ka] siranai rasii.
congressman -acc Q know.not seem
Lit. ‘I know which congressman is such that the newspaper that criticized
him last week has been boycotted, but it seems that Bill doesn’t know which
congressman.’
() boku-wa [[ISLAND [sensyuu proi [minsyutoo-no dono giin]
I-top last.week Dem. Party-gen which congressman
-o hihansita] sinbunsyai ] -ga boikotto sareteiru ka] sitteiru ga,
-acc criticized newspaper -nom is.being.boycotted Q know but
[[kyoowatoo-no dono giin] -o ka] (-wa) siranai.
Repub. Party-gen which congressman -acc Q (-top) know.not

 Takahashi () and Fukaya and Hoji () argue that examples like () are unacceptable. I suggest
that the judgments reported therein arose because we attempted to obtain the non-local reading through
non-local resolution, using my terminology here. See the discussion in Section .. As I will claim in
that section, following Merchant (: Chapter ), the non-local reading can also arise from the local
resolution, and I would like to suggest that this is the reason speakers in fact tend to accept such examples,
contrary to what is reported in Takahashi () and Fukaya and Hoji ().
Island-sensitivity in Japanese sluicing 131

Lit. ‘I know which congressman of the Democratic Party is such that the news-
paper that criticized him last week has been boycotted, but I don’t know which
congressman of the Republican Party.’
Note that the second conjuncts in (), (), and () seem to give rise to the
readings in (a), (b), and (c), respectively.
() a. I don’t know who is such that the police arrested the man who sold drugs
to him.
b. It seems that Bill didn’t know which representative is such that the police
arrested the man who had given a bribe to him.
c. It seems that Bill doesn’t know which congressman is such that the newspa-
per that criticized him last week has been boycotted.
If we take a closer look at the readings available in the examples in () and (),
however, a different picture emerges. The reading available in (), where the rem-
nant is attached with hoka-no ‘other-gen’, is given in () (let us call it the “non-
covariant reading”). This reading corresponds to that of the non-elliptical sentence
in ().
() I know that the police arrested the man who had given a bribe to representative
Tanaka, but I don’t know which other representative he (= the man who the
police arrested) had given a bribe to. [the “same briber for different politicians”
(non-covariant) reading]
() . . . [sono otoko-ga [hoka-no dono giin] -ni wairo-o okutta ka] -wa
that.man-nom other-gen which rep. -to bribe-acc gave Q -top
siranai.
know.not
‘(I know that the police arrested [the man who had given a bribe to represen-
tative Tanaka]i , but) I don’t know which other representative hei had given a
bribe to.’
What is puzzling is that the reading available in the non-elliptical counterpart of (),
given in (), is missing in ().
() . . . [keisatu-ga [ISLAND [proi [hoka-no dono giin] -ni wairo-o
police-nom other-gen which rep. -to bribe-acc
okutta] otokoi ] -o taihosita ka] -wa siranai.
gave man -acc arrested Q -top know.not
‘. . . I don’t know which other representative is such that the police arrested the
man who had given a bribe to him.’
Example () gives rise to the reading in () (let us call it the “covariant reading”),
while () does not.
132 Teruhiko Fukaya

() (I know that the police arrested the man who had given a bribe to representative
Tanaka, but) I don’t know which other representative is such that the police
arrested the man who had given a bribe to him. [the “different bribers for
different politicians” (covariant) reading]
On this reading, “the man who had given a bribe to another representative (the
identity of whom the speaker does not know)” denotes an individual distinct from
“the man who had given a bribe to representative Tanaka.” Thus, more than one briber
is involved.
Now let us turn to sluicing with a wh-correlate in (). The reading available in ()
is given in ().
() I know which congressman of the Democratic Party is such that the newspa-
per that criticized him last week has been boycotted, but I don’t know which
congressman of the Republican Party it (= the paper that has been boycotted)
criticized last week. [the “same newspaper for different congressmen” (non-
covariant) reading]
This is the reading that is available in the non-elliptical sentence in ().
() . . . [sono sinbunsya-ga sensyuu [kyoowatoo-no
that.newspaper-nom last.week Repub. Party-gen
dono giin] -o hihansita ka] (-wa) siranai.
which congressman -acc criticized Q (-top) know.not
‘. . . I don’t know which congressman of the Republican Party that newspaper
criticized last week.’
Note that there is only one newspaper involved in this case.
What is puzzling again is that the sluicing sentence lacks the reading available in its
non-elliptical counterpart given in ().
() . . . [[[ISLAND [sensyuu proi [kyoowatoo-no dono giin] -o
last.week Repub. Party-gen which congressman -acc
hihansita] sinbunsyai ] -ga boikotto sareteiru] ka] (-wa) siranai.
criticized newspaper -nom is.being.boycotted Q (-top) know.not
‘. . . I don’t know which congressman of the Republican Party is such that the
newspaper that criticized him last week has been boycotted.’
Example () yields the reading in (), which its sluicing counterpart lacks.
() (I know which congressman of the Democratic Party is such that the newspaper
that criticized him last week has been boycotted, but) I don’t know which
congressman of the Republican Party is such that the newspaper that criticized
him last week has been boycotted. [the “different newspapers for different con-
gressmen” (covariant) reading]
Island-sensitivity in Japanese sluicing 133

We have seen in this subsection that only the non-covariant reading is available
in sluicing examples in relative-clause island contexts when the remnant is modified
with hoka-no ‘other-gen’ or when the wh-remnant and the wh-correlate are made
contrastive, as illustrated in () and (). This has been shown to be in sharp contrast
to the case of their non-elliptical counterparts in () and (), which readily yield
the covariant reading.

.. The adjunct island


Now let us turn to the second type of island, the adjunct island. First, consider ().
() [[kanari-no kazu]-no kigyoo]i -ga [keisatu-ga [sokoi -no
a.large.number-gen company -nom police-nom that.place-gen
kanbu] -ni [zizyoo tyoosyu] -o okonatteiru] to happyoo sita
executive -dat interview -acc is.doing that announced
Lit. ‘[A large number of companies]i announced that the police are interviewing
itsi executive.’
The QP in the matrix subject position ([[kanari-no kazu]-no kigyoo]-ga ‘a large num-
ber of companies’) and the dependent term (soko ‘it’) in the embedded object can enter
into bound-variable anaphora, yielding the covariant reading “For each of the large
number of companies x, x announced that the police are interviewing x’s executive.”
Observe (), where the NP containing the dependent term in the embedded clause
([sokoi -no kanbu]-ni ‘its executive’) has been moved to the front of the matrix-clause
subject. In this case, the covariant reading is retained.
() [sokoi -no kanbu] -ni [[kanari-no kazu]-no kigyoo]i -ga
that.place-gen executive -dat a.large.number-gen company -nom
[keisatu-ga _ [zizyoo tyoosyu]-o okonatteiru] to happyoo sita
police-nom interview-acc is.doing that announced
Lit. ‘Itsi executive, [a large number of companies]i announced that the police
are interviewing _.’
Now consider (), where the NP containing the dependent term resides within an
adjunct clause. In this case as well, the QP in the matrix subject position ([[kanari-
no kazu]-no kigyoo]-ga ‘a large number of companies’) and the dependent term (soko
‘it’) within the adjunct clause can enter into bound-variable anaphora, yielding the
covariant reading in ().
() [[kanari-no kazu]-no kigyoo]i -ga [ADJUNCT CLAUSE keisatu-ga
a.large.number-gen company -nom police-nom
[sokoi -no kanbu] -ni [zizyoo tyoosyu]-o okonatta] -kara
that.place-gen executive -dat interview-acc did because
kisyakaiken-o hirakazaru-o enakunatta
press.conference-acc had.to.hold
134 Teruhiko Fukaya

Lit. ‘[A large number of companies]i had to hold a press conference because the
police interviewed itsi executive.’

() For each of the large number of companies x, x had to hold a press conference
because the police interviewed x’s executive.

In contrast to (), the NP containing the dependent term cannot be moved to the
front of the matrix-clause subject while retaining the covariant reading, as illustrated
in ().

() ∗ [soko
i -no kanbu] -ni [[kanari-no kazu]-no kigyoo]i -ga
that.place-gen executive -dat a.large.number-gen company -nom
[ADJUNCT CLAUSE keisatu-ga _ [zizyoo tyoosyu]-o okonatta] -kara
police-nom interview-acc did because
kisyakaiken-o hirakazaru-o enakunatta
press.conference-acc had.to.hold
Lit. ‘[A large number of companies]i had to hold a press conference because the
police interviewed itsi executive.’

The unacceptability of () with the intended covariant reading, in contrast with the
acceptability of (), indicates that the adjunct clause is indeed a syntactic island in
Japanese. Note that as in the case of the relative-clause island, I used cases involving
bound-variable anaphora to ensure that the dislocated element is related by means of
movement to the position marked by an underscore that is within the adjunct.
Let us now turn to sluicing. Following the criterion for island-insensitivity in sluic-
ing mentioned in Section ., i.e. whether sluicing is possible when the first conjunct is
of the structure in (), we observe that sluicing in Japanese does not appear to exhibit
adjunct island effects either. In (), the correlate is situated within an adjunct clause,
but sluicing is acceptable, indicating that there are no adjunct island effects.

() Taroo-wa izen [ISLAND [aru giin]-ni wairo-o okutta] -kara


Taro-top before a rep.-to bribe-acc gave because
taiho sareta rasii ga, boku-wa [dono giin] -ni ka siranai.
was.arrested seem but I-top which rep. -to Q know.not
‘Taro has been arrested before because he gave a bribe to a representative, but
I don’t know to which representative.’

By the same criterion, sluicing examples with a definite and a wh-phrase correlate do
not seem to exhibit adjunct island effects either, as can be seen in () and ().

() Taroo-wa izen [ISLAND [Tanaka giin] -ni wairo-o okutta] -kara
Taro-top before Tanaka rep. -to bribe-acc gave because
taiho sareta ga, Yamada-wa [dono giin] -ni ka siranai.
was.arrested but Yamada-top which rep. -to Q know.not
Island-sensitivity in Japanese sluicing 135

‘Taro has been arrested before because he gave a bribe to representative Tanaka,
but Yamada doesn’t know to which representative.’

() boku-wa [Taroo-ga izen [ISLAND [dono giin] -ni wairo-o


I-top Taro-nom before which rep. -to bribe-acc
okutta] -kara taiho sareta ka] sitteiru ga, Yamada-wa
gave because was.arrested Q know but Yamada-top
[dono giin] -ni ka siranai.
which rep. -to Q know.not
‘I know which representative is such that Taro has been arrested before
because he (= Taro) gave a bribe to him, but Yamada doesn’t know to which
representative.’

Examples (–) appear to give rise to the readings in (a–c), respectively.

() a. Taro has been arrested before because he gave a bribe to a representative, but
I don’t know which representative is such that Taro has been arrested before
because he (= Taro) gave a bribe to him.
b. Taro has been arrested before because he gave a bribe to representative
Tanaka, but Yamada doesn’t know which representative is such that Taro has
been arrested before because he (= Taro) gave a bribe to him.
c. I know, but Yamada doesn’t know, which representative is such that Taro has
been arrested before because he (= Taro) gave a bribe to him.

Now consider the cases where the remnant is modified with hoka-no ‘other-gen’,
as in ().

() boku-wa [Taroo-ga izen [ISLAND [Tanaka giin] -ni wairo-o


I-top Taro-nom before Tanaka rep. -to bribe-acc
okutta] -kara taiho sareta no] -wa sitteiru ga,
gave because was.arrested that -top know but
[hoka-no dono giin] -ni ka wa siranai.
other-gen which rep. -to Q top know.not
‘I know that Taro has been arrested before because he gave a bribe to represen-
tative Tanaka, but I don’t know to which other representative.’

The reading available in () is given in (). Note that there is a single arrest for two
bribery cases (let us call it the “single-event reading”).

() Taro has been arrested once before because he gave bribes to two different
representatives. I know that one of them is representative Tanaka, but I don’t
know who the other is. [the single-event reading]

This is the reading that is available in the non-elliptical sentence in ().


136 Teruhiko Fukaya

() . . .[Taroo-ga [hoka-no dono giin] -ni wairo-o okutta ka] siranai.
Taro-nom other-gen which rep. -to bribe-acc gave Q know.not
‘. . . I don’t know to which other representative Taro gave a bribe.’
In the non-elliptical counterpart of (), given in (), a different reading is avail-
able, in which there are two distinct events where Taro was arrested. In one of
the events he was arrested because he gave a bribe to representative Tanaka, and in
the other he was arrested because he gave a bribe to another representative, but the
speaker does not know the identity of the second representative. This reading given
in () (let us call it the “multiple-event reading”) is not available in ().
() . . .[Taroo-ga izen [[hoka-no dono giin] -ni wairo-o
Taro-nom before other-gen which rep. -to bribe-acc
okutta] -kara taiho sareta ka wa siranai.
gave because was.arrested Q top know.not
‘. . . I don’t know [which other representative]i Taro has been arrested before
because he gave a bribe to himi .’
() Taro has been arrested because he gave a bribe to a representative, and
Taro has been arrested on a different occasion because he gave a bribe to
another representative. I know that one of the representatives is representative
Tanaka, but I don’t know who the other representative is. [the multiple-event
reading]
Let us turn to sluicing with a wh-phrase correlate where the correlate and the rem-
nant are made contrastive by being modified with different phrases. Consider ().
() boku-wa [Taroo-ga izen [ISLAND [zimintoo-no dono giin] -ni
I-top Taro-nom before L.D.P.-gen which rep. -to
wairo-o okutta] -kara taiho sareta ka] sitteiru ga, [minsyutoo-no
bribe-acc gave because was.arrested Q know but D.P.-gen
dono giin] -ni ka wa siranai.
which rep. -to Q top know.not
‘I know which representative of the Liberal Democratic Party is such that Taro
has been arrested before because he (= Taro) gave a bribe to him, but I don’t
know to which representative of the Democratic Party.’
This sentence yields the reading in (a) but not the one in (b).
() a. Taro has been arrested once before because he gave bribes to a representative
of the Liberal Democratic Party and a representative of the Democratic
Party. I know who the representative of the Liberal Democratic Party is, but
I don’t know who the representative of the Democratic Party is. [the single-
event reading]
Island-sensitivity in Japanese sluicing 137

b. Taro has been arrested because he gave a bribe to a representative of the


Liberal Democratic Party, and Taro has been arrested on a different occa-
sion because he gave a bribe to a representative of the Democratic Party.
I know who the representative of the Liberal Democratic Party is, but I don’t
know who the representative of the Democratic Party is. [the multiple-event
reading]
Note that there is a single arrest for two bribery cases in (a) (hence the single-event
reading), and that there is more than one arrest in (b) (hence the multiple-event
reading). The single-event reading in (a) is what is available in the non-elliptical
sentence in ().
() . . .[minsyutoo-no dono giin] -ni wairo-o okutta ka siranai.
D.P.-gen which rep. -to bribe-acc gave Q know.not
‘. . . I don’t know to which representative of the Democratic Party Taro gave a
bribe.’
In the non-elliptical counterpart of (), given in (), on the other hand, the
multiple-event reading in (b) is readily available.
() . . .[Taroo-ga izen [ISLAND [minsyutoo-no dono giin] -ni wairo-o
Taro-nom before D.P.-gen which rep. -to bribe-acc
okutta] -kara taiho sareta ka] wa siranai.
gave because was.arrested Q top know.not
‘. . . I don’t know which representative of the Democratic Party is such that Taro
has been arrested before because he (= Taro) gave a bribe to him.’
We have seen in this subsection that only the single-event reading is avail-
able in sluicing examples in adjunct island contexts when the remnant is modi-
fied with hoka-no ‘other-gen’ or when the wh-remnant and the wh-correlate are
made contrastive, as illustrated in () and (). This is in sharp contrast to the
availability of the multiple-event reading in their non-elliptical counterparts in ()
and ().

.. Summary
In the previous subsections, we have examined Japanese sluicing in island con-
texts (the relative clause and the adjunct islands) and demonstrated that there
are restrictions on the availability of readings in the definite-correlate sluicing
with the remnant modified with hoka-no ‘else/other’ and also in the wh-phrase-
correlate sluicing where the correlate and the remnant are made contrastive by
being modified with different phrases. The findings in this section are summarized
in ().
138 Teruhiko Fukaya

()
Type of the island Available reading Missing reading

relative-clause island non-covariant covariant


adjunct island single-event multiple-event

In the following discussion, I will use as cover terms “local reading” for the non-
covariant and single-event readings and “non-local reading” for the covariant and
multiple-event readings when it is not necessary to make the relevant distinctions.
Now three questions arise regarding the availability of these readings.
() Question : How can the non-elliptical versions yield the non-local readings?
Question : Why do the sluicing versions lack the non-local readings?
Question : How can the sluicing versions yield the local readings?
In Section ., we will address the questions in (), but preceding this, I will give an
overview of the theory of ellipsis resolution adopted in this paper.

. Ellipsis resolution in Japanese sluicing


Following Fukaya and Hoji (), who in turn draw on Hoji (: Chapter ),
I assume a copy theory of ellipsis resolution for Japanese sluicing, as summarized
in ().  , 
() a. The remnant in Japanese sluicing is base-generated in a position adjoined to
an empty TP.
b. In order for the remnant to be interpreted, a TP available in the discourse is
copied into the empty TP at LF.
c. The copied TP must have an empty slot within it so that the remnant can be
syntactically related to this position.
d. A constituent within a TP can optionally undergo the LF operation Con-
stituent Raising (CR), which raises and adjoins the constituent to a TP that
dominates it (cf. Reinhart ()). As a result, a TP with an empty slot is
created. In the case of sluicing, the correlate undergoes CR.
e. CR is sensitive to syntactic islands (à la Reinhart ()).

 See Kizu (), Kuwabara (), and Nishiyama et al. (), among others, for an alternative view
that sluicing structures are reduced clefts in Japanese. See Fukaya (: Section .) for an argument
against this view.
 The choice of the copy analysis over a deletion analysis is based solely on conceptual considerations.
In the view of Japanese grammar we are pursuing, overt feature-driven syntactic movement, which is
necessary for the deletion analysis, is completely absent in this language, and Japanese is assumed to have
only PF and LF movement (see Fukaya (: Chapters  and )). We attribute this to the presence in
English-type languages and the absence in Japanese-type languages of the uninterpretable inflectional
features that trigger overt movement, one example of which is wh-movement. See Fukui () and his
subsequent works.
Island-sensitivity in Japanese sluicing 139

Let us illustrate how this mechanism works, using the diagrams in () and ().
() a. 1st Conjunct at Spell-Out: b. 1st Conjunct at LF:
TP TP

NP T' NP TP
<correlate>
VP T NP T'

. . . NP. . . VP T
<correlate>
... t ...

() a. 2nd Conjunct at Spell-Out: b. 2nd Conjunct at LF:


… CP… … CP …

TP C TP C
| |
VP T ka VP T ka

CP V CP V

TP C (da) TP C (da)

NP [TP Ø] <== TP is copied here. NP TP


<remnant> <remnant>
NP T'

VP T

…t…

In the first conjunct the correlate moves and adjoins to a TP that dominates it, as in
(b), and as a result, a TP with an empty slot is generated. Then, the boxed lower TP
in (b) is copied into the empty TP in (a), which is base-generated in the sluicing
site, and the remnant is given an interpretation by being associated with the empty slot
in the copied TP as in (b).  Note that it is assumed that CR can adjoin a constituent
to any TP that dominates it, as indicated in (d), as long as it does not cross syntactic
islands, as postulated in (e). This then makes it possible for the correlate to raise

 In this structure, I put extra layers of CP, VP, and TP directly above the lower TP because it has been
pointed out that the copula da can optionally appear in Japanese sluicing. See Nishiyama et al. (), Kizu
() and Nishigauchi (: Section ..), among others, for some relevant discussion. In the following
exposition, I will mostly ignore the possibility of having the copula in sluicing.
140 Teruhiko Fukaya

within an island, as will be illustrated in Section .. This derivation has the same
effect as Merchant’s (: Chapter ) proposal for local movement in propositional
island contexts. For cases like (a), Merchant proposes the structure in (b). He
thus claims that only local wh-movement is involved in the derivation of the sluicing
cases that appear to be island-insensitive.
() a. They hired someone who speaks a Balkan language—guess which!
(Merchant : Chapter  (a))
b. Guess whichi [she speaks t i ]!
(Merchant : Chapter  (a))
In the next section, I will show how the facts we have seen in Section . can be
accounted for under this theory.

. Local vs non-local resolutions


In this section, I address the questions raised at the end of Section ., and, in so
doing, I will show how the facts in Japanese sluicing can be accounted for with
the mechanism of ellipsis resolution summarized in Section .. More specifically,
I will argue that the availability of the non-local readings is due to the wh-phrase
taking scope over the island, and that this is not possible in sluicing because of the
restriction on CR in (e). I will then claim, following Merchant (: Chapter ), that
apparent island-insensitivity in Japanese sluicing results if the local resolution yields
an interpretation indistinguishable from the interpretation that would be obtained by
the non-local resolution.
Some clarification of the terminology is in order here. In the context of our dis-
cussion a “resolution” refers to an operation (copying in the theory pursued here)
that gives a structure to a missing part, and a “reading” refers to an interpreta-
tion obtained as a result of a resolution. A “local resolution” is then an operation
of copying onto the ellipsis site a TP resulting from CR-ing the correlate within
an island in the first conjunct. On the other hand, a “non-local resolution” would
be an operation of copying onto an ellipsis site a TP that would result from CR-
ing the correlate across an island in the first conjunct if such movement were
possible.

.. Cases where the local reading is not equivalent to the non-local reading
... Scope of a wh-phrase Now let us address the first question in ()—how can
the non-elliptical versions yield the non-local readings?—starting with the covariant
reading in the case of the relative clause. As we have seen above, (), which has
the non-elliptical structure in () in the second conjunct, gives rise to the covariant
reading in ().
Island-sensitivity in Japanese sluicing 141

() . . .[keisatu-ga [ISLAND [proi [hoka-no dono giin] -ni wairo-o


police-nom other-gen which rep. -to bribe-acc
okutta] otokoi ]-o taihosita ka] -wa siranai.
gave man -acc arrested Q -top know.not
‘. . . I don’t know which other representative is such that the police arrested the
man who had given a bribe to him.’
() (I know that the police arrested the man who had given a bribe to representative
Tanaka, but) I don’t know which other representative is such that the police
arrested the man who had given a bribe to him. [the “different bribers for
different politicians” (covariant) reading]
() nd conjunct:
I CP-TOP know.not

TP C

NP T⬘ ka

the police VP T

NP V

TP NP arrested

... which other rep. ... man

I assume (), following Baker ().


() Wh-in-situ takes scope at the position of the Q-morpheme that binds it. 
Compare the following two examples to see the point in ().
() a. Taroo-wa [Hanako-ga dare-to atta ka] itteimasita ka
Taro-top Hanako-nom who-with met Q said Q
‘Did Taro say who Hanako had met?’
b. Taroo-wa [Hanako-ga dare-to atta to] itteimasita ka
Taro-top Hanako-nom who-with met that said Q
‘Who did Taro say Hanako had met?’
The difference between (a) and (b) is whether or not the Q-morpheme ka
resides in the embedded clause. In (a) the wh-phrase takes scope at the embed-
ded clause, and hence the entire sentence is a yes–no question with an embedded

 See also Pesetsky (), Cheng (), and Li (), among others, for a similar proposal.
142 Teruhiko Fukaya

wh-interrogative. In (b), on the other hand, the wh-phrase takes scope at the matrix
clause, and hence the entire sentence is a wh-interrogative. 
In (), the wh-in-situ [hoka-no dono giin]-ni ‘to which other representative’ takes
scope at the position of the Q-morpheme ka. Since the Q-morpheme ka is located in
the C position of the clause higher than the complex NP, the wh-phrase takes scope
over the latter, allowing hoka-no ‘other-gen’ to take scope over the complex NP. I claim
that it is this scope relation that gives rise to covariant readings in the relative-clause
island cases.
The same holds for the wh-correlate case. Recall that () yields the covariant
reading in (). Since the wh-in-situ is bound by the Q-morpheme located in the C
position that is higher than the complex NP in this case as well, [kyoowatoo-no dono
giin]-o ‘which congressman of the Republican Party’ takes scope over the complex NP,
thereby giving rise to the covariant reading.
() . . .[[[ISLAND [sensyuu proi [kyoowatoo-no dono giin] -o
last.week Repub. Party-gen which congressman -acc
hihansita] sinbunsyai ] -ga boikotto sareteiru] ka] (-wa) siranai.
criticized newspaper -nom is.being.boycotted Q (-top) know.not
‘. . . I don’t know which congressman of the Republican Party is such that the
newspaper that criticized him last week has been boycotted.’
() (I know which congressman of the Democratic Party is such that the newspaper
that criticized him last week has been boycotted, but) I don’t know which
congressman of the Republican Party is such that the newspaper that criticized
him last week has been boycotted. [the “different newspapers for different con-
gressmen” (covariant) reading]
The relevant assumption on covariant readings is as follows.
() The necessary condition for the covariant reading:
The covariant reading is obtained structurally only if the wh-phrase takes scope
over the complex NP.
Let us turn to the adjunct island cases. We have seen that (), which has the non-
elliptical structure in (), yields the multiple-event reading in ().
() . . .[Taroo-ga izen [[hoka-no dono giin] -ni wairo-o
Taro-nom before other-gen which rep. -to bribe-acc
okutta] -kara taiho sareta ka wa siranai.
gave because was.arrested Q top know.not
‘. . . I don’t know [which other representative]i Taro has been arrested before
because he gave a bribe to himi .’
 Note that in (a) the wh-phrase cannot take matrix scope, although there is a Q-morpheme ka
residing in the matrix clause. This indicates that the interpretation of a wh-phrase is constrained by some
kind of locality, i.e. it is interpreted at the closest Q-marker. See Harada () for discussion on this issue.
Island-sensitivity in Japanese sluicing 143

() Taro has been arrested because he gave a bribe to a representative, and Taro
has been arrested on a different occasion because he gave a bribe to another
representative. I know that one of the representatives is representative Tanaka,
but I don’t know who the other representative is. [the multiple-event reading]
() nd conjunct:
I CP know.not

TP C

NP T´ ka

Taro VP T

PP VP

TP P was arrested

...which other rep.... because


In this case, the wh-phrase [hoka-no dono giin]-ni ‘to which other representative’ is
bound by the Q-morpheme situated in the C position of the CP in the complement
of the matrix VP (C in ()). It thus takes scope at that position, allowing hoka-no
‘other-gen’ to take scope there. I assume that the wh-phrase [hoka-no dono giin]-ni
taking scope over the adjunct clause gives rise to the multiple-event reading. The same
line of reasoning also applies to wh-correlate cases, as in ().
() . . .[Taroo-ga izen [ISLAND [minsyutoo-no dono giin] -ni wairo-o
Taro-nom before D.P.-gen which rep. -to bribe-acc
okutta] -kara taiho sareta ka] wa siranai.
gave because was.arrested Q top know.not
‘. . .I don’t know which representative of the Democratic Party is such that Taro
has been arrested before because he (= Taro) gave a bribe to him.’
Since the wh-phrase [minsyutoo-no dono giin]-ni ‘to which representative of the
Democratic Party’ is bound by the Q-morpheme in the C position above the adjunct
clause, the wh-phrase takes scope over it. This gives rise to the multiple-event reading
in (b).
() b. Taro has been arrested because he gave a bribe to a representative of the
Liberal Democratic Party, and Taro has been arrested on a different occa-
sion because he gave a bribe to a representative of the Democratic Party. I
know who the representative of the Liberal Democratic Party is, but I don’t
know who the representative of the Democratic Party is. [the multiple-event
reading]
144 Teruhiko Fukaya

The assumption for multiple event readings is as follows.


() The necessary condition for the multiple-event reading:
The multiple-event reading is obtained structurally only if the wh-phrase takes
scope over the adjunct clause.
Combining () and (), we obtain the generalized condition regarding the non-local
reading.
() The necessary condition for the non-local reading:
The non-local reading is obtained structurally only if the wh-phrase takes scope
over an island.
... Lack of the non-local reading Let us now turn to the second question in ():
“Why do the sluicing versions lack the non-local readings?” According to (), the
wh-phrase must take scope over the island to obtain the non-local reading. There
would be two ways to establish this. One way would be for the wh-phrase within
an island to be bound by a Q-morpheme that is higher than the island. This option,
however, is not available in sluicing because the wh-remnant (the boxed NP in ()) is
base-generated in the position adjoined to an empty TP and cannot reside within an
island.
() nd conjunct at Spell-Out:
... CP ...

TP C

T´ ka

VP T

CP V
|
TP C (da)

NP [TP Ø ]

wh-phrase

The other way would be for the wh-remnant to be associated with a position within
an island after a discourse-available TP is copied, as in (b). In order for such an
association to be established, the correlate must move across the island in the first
conjunct as in (a).
Island-sensitivity in Japanese sluicing 145

() a. 1st conjunct: b. 2nd conjunct:


TP ... TP ...

NP TP NP TP
<correlate> <remnant>
... ISLAND ... ... ISLAND ...

... t ... ... t ...

This movement, however, is blocked by the constraint in (e). With both options
unavailable, the remnant wh-phrase cannot take scope over the island. This makes
the non-local reading unavailable because the necessary condition for the non-local
reading in () is not met. In this way, the unavailability of the non-local read-
ing in the sluicing cases in question is accounted for under the theory of ellipsis
resolution assumed here along with the necessary condition for the non-local reading
in ().
... The local reading Let us address the third question in ()—how can the
sluicing versions yield the local readings?—using () as an example.
() boku-wa keisatu-ga [ISLAND [proi [Tanaka giin]-ni wairo-o okutta]
I-top police-nom Tanaka rep.-to bribe-acc gave
otokoi ]-o taihosita no-wa sitteiru ga, [[hoka-no dono giin] -ni
man -acc arrested that-top know but other-gen which rep. -to
ka] -wa siranai.
Q -top know.not
‘I know that the police arrested the man who had given a bribe to representative
Tanaka, but I don’t know to which other representative.’
In the first conjunct, the NP [Tanaka giin]-ni ‘to representative Tanaka’ can raise and
adjoin to the TP within the island, as in (a), because CR can raise and adjoin a
constituent to a TP that dominates it as long as no syntactic islands are crossed. Then
the lower TP can be copied into the empty TP in the second conjunct, yielding the
structure in (b). 

 I assume here that the Japanese relative clause, unlike its English counterpart, has an empty prono-
minal element within it instead of a trace of a relative operator and that this empty pronominal element is
co-indexed with the head noun of the relative clause via predication (cf. Kuno : Chapter , Haig ,
among others).
146 Teruhiko Fukaya

() a. 1st conjunct: b. 2nd conjunct:


NP = ISLAND CP know.not
3 3
TP NPi TP C
3 4 3 |
[rep. T]-DAT TP man [which other TP ka
<correlate> 3 rep.] -DAT 3
NP T⬘ <remnant> NP T⬘
4 3 4 3
proi VP T proi VP T
6 6
... t ... ... t ...

I assume, adopting Merchant’s (: Chapter ) analysis, that pro in the second
conjunct can function as an E-type pronoun, yielding the interpretation “the man
who the police arrested”. Then the structure in (b) gives rise to the reading in ().
() I know that the police arrested the man who had given a bribe to representative
Tanaka, but I don’t know which other representative he (= the man who the
police arrested) had given a bribe to. [the “same briber for different politicians”
(non-covariant) reading]
The same reasoning holds for wh-correlate sluicing as in (), where the correlate and
the remnant are made contrastive by being modified with different phrases. With the
pro in the copied TP interpreted as “the paper that has been boycotted”, () gives rise
to the reading in ().
() boku-wa [[ISLAND [sensyuu proi [minsyutoo-no dono giin] -o
I-top last.week Dem. Party-gen which congressman -acc
hihansita] sinbunsyai ] -ga boikotto sareteiru ka] sitteiru ga,
criticized newspaper -nom is.being.boycotted Q know but
[[kyoowatoo-no dono giin] -o ka] (-wa) siranai.
Repub. Party-gen which congressman -acc Q (-top) know.not
Lit. ‘I know which congressman of the Democratic Party is such that the news-
paper that criticized him last week has been boycotted, but I don’t know which
congressman of the Republican Party.’
() I know which congressman of the Democratic Party is such that the newspa-
per that criticized him last week has been boycotted, but I don’t know which
congressman of the Republican Party it (= the paper that has been boycotted)
criticized last week. [the “same newspaper for different congressmen” (non-
covariant) reading]
Next, let us turn to the adjunct island cases as in ().
Island-sensitivity in Japanese sluicing 147

() boku-wa [Taroo-ga izen [ISLAND [Tanaka giin] -ni wairo-o


I-top Taro-nom before Tanaka rep. -to bribe-acc
okutta] -kara taiho sareta no] -wa sitteiru ga,
gave because was.arrested that -top know but
[hoka-no dono giin] -ni ka wa siranai.
other-gen which rep. -to Q top know.not
‘I know that Taro has been arrested before because he gave a bribe to represen-
tative Tanaka, but I don’t know to which other representative.’

If CR raises the correlate within the adjunct clause in the first conjunct as in (), and
the boxed lower TP is copied into the second conjunct, we obtain ().

() Within the adjunct clause in the st conjunct:


... PP

TP P

‘to rep. Tanaka’ TP ‘because’

... t ...

() [TP [hoka-no dono giin] -ni [TP pro t wairo-o okutta]] ka
other-gen which rep. -to bribe-acc gave Q
siranai.
know.not
Lit. ‘I don’t know which other representative pro gave a bribe to.’

With the pro interpreted as Taroo, () yields the reading “I don’t know to which
other representative Taro gave a bribe (on that occasion)” (the single-event reading).
Note that, since the wh-phrase does not scope out of the adjunct, the multiple-event
reading (“Taro was arrested more than once before because he gave a bribe to different
representatives”) is not available, as we have seen in Section .... The same line
of reasoning holds for cases where the wh-correlate and the wh-remnant are made
contrastive by being modified with different phrases—as in (), which gives rise to
the reading in (a).

() boku-wa [Taroo-ga izen [ISLAND [zimintoo-no dono giin] -ni


I-top Taro-nom before L.D.P.-gen which rep. -to
wairo-o okutta] -kara taiho sareta ka] sitteiru ga, [minsyutoo-no
bribe-acc gave because was.arrested Q know but D.P.-gen
dono giin] -ni ka wa siranai.
which rep. -to Q top know.not
148 Teruhiko Fukaya

‘I know which representative of the Liberal Democratic Party is such that Taro
has been arrested before because he (= Taro) gave a bribe to him, but I don’t
know to which representative of the Democratic Party.’
() a. Taro has been arrested once before because he gave bribes to a representative
of the Liberal Democratic Party and a representative of the Democratic
Party. I know who the representative of the Liberal Democratic Party is, but
I don’t know who the representative of the Democratic Party is. [the single-
event reading]
... Summary In the previous subsections, I have provided answers to the three
questions raised at the end of Section .. They are summarized here.
() Question : How can the non-elliptical versions yield the non-local
readings?
Answer : By being bound by a Q-morpheme higher than the island, the
wh-phrase can take scope over the island. This scope relation gives
rise to the non-local readings.
Question : Why do the sluicing versions lack the non-local readings?
Answer : CR is constrained by island conditions and cannot yield the struc-
ture necessary for the remnant to be associated with a position
within an island.
Question : How can the sluicing versions yield the local readings?
Answer : CR can raise the correlate within the island, and the TP within the
island is copied into the second conjunct.

.. Cases where the local reading is equivalent to the non-local reading
Next, let us turn to the apparent island-repair examples. I will follow Merchant (:
Chapter ) and show that apparent island-insensitivity in Japanese sluicing results if
the local resolution yields an interpretation indistinguishable from the interpretation
that would be obtained by the non-local resolution. Recall that apparent island repair
is observed in (i) indefinite-correlate cases (e.g. () and ()), (ii) definite-correlate
cases where the remnant is not modified with hoka-no ‘other-gen’ (e.g. () and ()),
and (iii) wh-correlate cases where the correlate and the remnant are not made con-
trastive by being modified with different phrases (e.g. () and ()). First we consider
the relative-clause island cases. Recall that the correlate within a relative clause in the
first conjunct can raise and adjoin to a TP within the relative clause. This movement
is legitimate because it does not cross a syntactic island. In (), for example, CR can
raise the NP [Tanaka giin]-ni ‘to representative Tanaka’ and adjoin it to the TP within
the relative clause, as in ().
Island-sensitivity in Japanese sluicing 149

() [keisatu-wa [ISLAND [proi [Tanaka giin]-ni wairo-o okutta] otokoi ] -o


police-top Tanaka rep.-to bribe-acc gave man -acc
taihosita ga, Bill-wa [dono giin-ni ka] siranakatta rasii.
arrested but Bill-top which rep.-to Q knew.not seem
‘The police arrested the man who had given a bribe to representative Tanaka,
but it seems that Bill didn’t know to which representative.’
() NP =ISLAND

TP NPi

[rep. Tanaka]-DATj TP man


<correlate>
NP T⬘

proi VP T

... tj ...

Then the lower TP, i.e. [TP proi t j wairo-o okutta] ‘proi had given a bribe t j ’, is copied
into the empty TP in the second conjunct, giving the structure in ().
() Bill CP know.not

TP C
|
[‘which other rep.’]-DAT TP ka
<remnant>
NP T⬘

proi VP T

... tj ...

With the pro interpreted as “the man who the police arrested”, along the lines of
Merchant (: Section .), () yields the interpretation in ().
() It seems that Bill didn’t know which representative he (= the man who the
police arrested) had given a bribe to.
150 Teruhiko Fukaya

On the other hand, () would result if CR could raise the correlate across the island
and the TP obtained by CR in the first conjunct could be copied into the empty TP in
the second conjunct, as in ().
() . . . Bill-wa [CP [TP dono giin-ni [TP keisatu-ga [ISLAND [proi t j
Bill-top which rep.-to police-nom
wairo-o okutta] otokoi ] -o taihosita]] ka] siranakatta rasii.
bribe-acc gave man -acc arrested Q knew.not seem
‘. . . it seems that Bill didn’t know which representative the police arrested the
man who had given a bribe to him.’
() a. 1st conjunct: b. 2nd conjunct:
TP ... TP ...

NPj TP NP TP
<correlate> <remnant>
... ISLAND ... ... ISLAND ...

... tj ... ... tj ...

Example () would then yield the interpretation in ().


() Bill didn’t know which representative is such that the police arrested the man
who had given a bribe to him.
Since CR obeys island constraints, () is not an available option for (). Note, how-
ever, that the interpretations in () and () are not distinguishable. Although what
we have obtained syntactically is the structure via the local resolution as in (), we
“feel” that we have obtained the non-local reading in () because the local resolution
gives rise to a reading indistinguishable from that yielded by the non-local resolution.
The same line of reasoning holds for the indefinite-correlate and wh-correlate cases
in () and (), respectively.
Next, let us consider the adjunct island cases as in ().
() Taroo-wa izen [ISLAND [aru giin]-ni wairo-o okutta] -kara
Taro-top before a rep.-to bribe-acc gave because
taiho sareta rasii ga, boku-wa [dono giin] -ni ka siranai.
was.arrested seem but I-top which rep. -to Q know.not
‘Taro has been arrested before because he gave a bribe to a representative, but I
don’t know to which representative.’
In this case as well, since CR can raise an element to any TP dominating it provided
that it does not cross an island, it can raise the correlate [aru giin]-ni ‘to a represen-
tative’ and adjoin it to the TP within the adjunct. Thus, we obtain (b) from (a) in
the first conjunct.
Island-sensitivity in Japanese sluicing 151

() a. . . . [TP pro [aru giin]-ni wairo-o okutta] -kara . . .


a rep.-to bribe-acc gave because
‘. . . because he gave a bribe to a representative . . .’
b. . . . [TP [aru giin]-nii [TP pro t i wairo-o okutta]] -kara . . .
a rep.-to bribe-acc gave because
Lit.‘ . . . because to a representative he gave a bribe . . .’
The lower TP in (b) is copied into the empty TP in the second conjunct, and we
obtain ().
() boku-wa [TP [dono giin] -ni [TP pro t i wairo-o okutta]] ka
I-top which rep. -to bribe-acc gave Q
siranai.
know.not
‘I don’t know to which representative he gave a bribe.’
This structure gives rise to the reading in (a), which is indistinguishable from the
one in (b), which the non-local resolution would yield if it were possible.
() a. I don’t know to which representative he (= Taro) had given a bribe on that
occasion (= when he was arrested).
b. I don’t know which representative is such that Taro has been arrested before
because he (= Taro) gave a bribe to him.
Since we are looking at a single event in which Taro was arrested, there is no way
to distinguish between the local reading and the non-local reading. Thus, we “feel”
that the non-local reading is available, although the local resolution is the only option
available. The same line of reasoning holds also for definite-correlate cases where the
remnant is not modified with hoka-no ‘other’, as in (), and for wh-correlate cases
where the correlate and the remnant are not made contrastive by being modified with
different phrases, as in (). Note that in such cases as well, there is no means to
distinguish between the local and non-local readings because it is always the case
that a single event is involved. This is the impetus behind our crucial reliance on
the sluicing examples with other types of correlates and remnants to show island-
sensitivity in Japanese sluicing.

.. Summary
In the previous subsections we have shown how the rather intricate properties of
Japanese sluicing observed in Section . can be accounted for, given the analysis in
Fukaya and Hoji (), combined with a version of Merchant’s (: Chapter )
proposal for local movement in sluicing. Thus far, we have investigated case-marked
sluicing only; in the next section, I will turn to non-case-marked sluicing, which has
properties quite distinct from case-marked sluicing.
152 Teruhiko Fukaya

. Non-case-marked sluicing in Japanese


In the preceding sections, we have investigated case-marked sluicing, where the rem-
nant wh-phrase is case-marked. Fukaya and Hoji () claimed that case-marked
sluicing should be distinguished from non-case-marked sluicing, where the remnant
wh-phrase is not case-marked. Note that what we refer to here as “case-markers”
include postpositions as well as nominative, accusative, and dative markers. We pro-
posed that non-case-marked sluicing can have the structure schematized in (). 
() [TP NP [VP [CP [TP pro [VP < remnant > (da)]] Q] V]]
The non-case-marked sluicing in (a), for example, can be represented as in (b).
() a. [John-wa kinoo dareka-ni atta] rasii ga, boku-wa [dare ka]
John-top yesterday someone-dat met seem but I-top who Q
siranai.
know.not
‘John seems to have met someone yesterday, but I don’t know who.’
b. . . . [TP boku-wa [VP [CP [TP pro dare (da)] ka] siranai]].
I-top who cop Q know.not
We thus claimed that non-case-marked sluicing can be represented as a kind of copula
construction like John-wa gakusei da ‘John is a student’. The subject of the sentence is
pro, which is comparable to sore ‘that’. It functions as a deep anaphor in the sense of
Hankamer and Sag (), and can be interpreted as referring to the property denoted
in the first conjunct. Thus, (b), for example, is interpreted analogously to I don’t
know who that is, where that is interpreted as the property of being the person such
that John met him yesterday. One crucial aspect of this type of sluicing is that since
no copying of a TP is necessary in the second conjunct, no movement of the correlate
needs to take place in the first conjunct.
We have seen above that the case-marked sluicing examples in (), (), (),
and () do not give rise to the non-local readings in (), (), (), and (b),
respectively. Their non-case-marked counterparts, which can be obtained by omitting
the case-markers on the remnant wh-phrases, by contrast, appear to give rise to such
non-local readings. This is expected in Fukaya and Hoji’s () analysis of non-case-
marked sluicing. According to our account, the non-case-marked counterparts of (),
(), (), and () can have the structures in (a–d), respectively.

 Note in this connection that nothing prevents non-case-marked sluicing from having the same struc-
ture as its case-marked counterpart in principle. Thus, non-case-marked sluicing is structurally ambiguous,
although there may be a bias toward the structure in ().
Island-sensitivity in Japanese sluicing 153

() a. (the non-case-marked counterpart of ())


. . . [CP pro [hoka-no dono giin] ka] -wa siranai.
other-gen which rep. Q -top know.not
‘. . . I don’t know which other representative.’
b. (the non-case-marked counterpart of ())
. . . [CP pro [kyoowatoo-no dono giin] ka] (-wa) siranai.
Repub.Party-gen which rep. Q (-top) know.not
‘. . . I don’t know which representative of the Republican Party.’
c. (the non-case-marked counterpart of ())
. . . [CP pro [hoka-no dono giin] ka] -wa siranai.
other-gen which rep. Q -top know.not
‘. . . I don’t know which other representative.’
d. (the non-case-marked counterpart of ())
. . . [CP pro [minsyutoo-no dono giin] ka] -wa siranai.
Dem. Party-gen which rep. Q -top know.not
‘. . . I don’t know which representative of the Democratic Party.’
Since the pro in non-case-marked sluicing can be interpreted as referring to the prop-
erty denoted in the first conjunct, the pros in (a–d) can be interpreted as referring
to the properties in (a–d), respectively.
() a. The property of being the person such that the man who had given a bribe
to him was arrested by the police
b. The property of being the person such that the newspaper that criticized him
last week has been boycotted
c. The property of being the person such that Taro has been arrested before
because he (= Taro) gave a bribe to him
d. The property of being the person such that Taro has been arrested before
because he (= Taro) gave a bribe to him
Structures (a–d) thus give rise to the interpretations in (a–d), respectively.
() a. I don’t know which other representative has the property in (a).
b. I don’t know which representative of the Republican Party has the property
in (b).
c. I don’t know which other representative has the property in (c).
d. I don’t know which representative of the Democratic Party has the property
in (d).
Notice that the interpretations in (a–d) are equivalent to the non-local readings in
(), (), (), and (b), respectively. Hence, the non-case-marked sluicing cases
appear to yield the non-local readings that are not obtained in their case-marked
counterparts.
154 Teruhiko Fukaya

. Implications for English sluicing


Recall that it has been the general consensus that regular sluicing in English, where
the correlate is an indefinite, is not sensitive to islands, as briefly summarized in
Section .. The island-insensitivity in such cases has led researchers to develop a
theory in which island violations are ameliorated in sluicing. Note in this connection
that island effects are apparently not observed in cases where the remnant carries a
preposition either, as in ().
() a. I remember Abby hired someone who works on a Balkan language, but
I don’t remember on which Balkan language.
b. I know that the police arrested a man who had sold drugs to a celebrity in
LA, but I don’t know to whom/to which celebrity.
In this section, I will re-examine island-sensitivity in English sluicing in light of the
data in Japanese sluicing observed in the preceding sections and demonstrate that
English sluicing in fact exhibits island effects, contrary to the widely accepted gener-
alization. More specifically, I will argue that the violation of the relative-clause island
in English is not ameliorated by deletion, contra Fox and Lasnik () and Merchant
(), among others. The new evidence concerning English sluicing observed in this
section is in fact a challenge to any theory that is based on the assumption that island
violations in general are nullified under sluicing. Based upon this new evidence, I will
suggest that two types of sluicing be distinguished in English, corresponding to case-
marked and non-case-marked sluicing in Japanese as illustrated in Sections . and ..

.. Contrast sluicing


Let us begin by examining the properties of what Merchant (: –) calls “contrast
sluicing”, because we rely crucially on it for relevant data. In () we see examples of
contrast sluicing.
() a. Abby speaks GREEK, but I don’t remember what OTHER languages.
b. She met RINGO, but I don’t know who else.
(Merchant (: ()); cf. Merchant (: ))
Merchant () claims that contrast sluicing indeed obeys island constraints, giving
the examples in (). (The portions in the angle brackets are not phonetically realized.)
() a. ∗ Abby wants to hire someone who speaks GREEK, but I don’t remember
what OTHER languages <she wants to hire someone who speaks _>. 
 For Barry Schein (personal communication, March ), this example improves by changing
languages to language. He also pointed out that (i) is acceptable for him.

(i) Abby wants to hire someone who speaks three Romance languages, but I don’t remember how many
Germanic languages.

Some kind of singular/plural matching seems to be at play here, but I have to leave the issue for future
research.
Island-sensitivity in Japanese sluicing 155

b. ∗ The radio played a song that RINGO wrote, but I don’t know who else <the
radio played a song that _ wrote>.
(based on Merchant (: ()))

In an attempt to account for the unacceptability of () while maintaining his account
of the acceptability of (b) and (b), Merchant claims that the degraded status of
these examples is due to the nature of focus movement in the first conjunct. He
stipulates that “island-escaping focus-movement cannot target the highest IP [= TP]”
(Merchant : ). In (), according to his account, the position of the wh-phrase
remnant in the second conjunct requires that its correlate take scope over the entire
clause in the antecedent site because of the requirement on parallelism; however, the
focus movement can only move the correlate to the matrix VP, not to the matrix TP.
It is then expected that if there is an extra TP on top of the first conjunct, the
resulting structure will improve because this should allow focus movement in the first
conjunct to target the TP that corresponds to the TP complement of the [+wh] C in
the second conjunct. That is, focus movement should be able to target the TP Abby
wants to hire someone who speaks GREEK and the TP the radio played a song that
RINGO wrote in (a) and (b), respectively. Nevertheless, as the unacceptability of
() shows, merely having an extra TP layer (underlined in ()) in the first conjunct
does not seem to help.

() a. ∗ I know Abby wants to hire someone who speaks GREEK, but JOHN doesn’t
know what language.
b. ∗ I know the radio played a song that RINGO wrote, but JOHN doesn’t
know who.

The examples in () improve, however, if some further adjustments are made, such
as giving the first and second conjuncts a parallel matrix clause and the same subject,
adjusting the singular/plural mismatch, etc. The examples in () with such adjust-
ments are acceptable (adjustments are underlined).

() a. I remember Abby wants to hire someone who speaks GREEK, but
I don’t remember which OTHER language.
b. I know the radio played a song that RINGO wrote, but I don’t know
who ELSE.

Although the exact nature of contrast sluicing just observed is not clear, the contrast
between () and () and what is reported in footnote  seem to suggest that the
acceptability of these contrast-sluicing examples is not contingent upon (LF-) syntac-
tic restrictions, such as a restriction on focus movement, as Merchant () claims.
Note that (a) seems to give rise to the reading where different people speak differ-
ent languages, and that (b) appears to yield the reading where different musicians
wrote different songs, as indicated in ().
156 Teruhiko Fukaya

() a. [for (a)] I remember Abby wants to hire someone who speaks Greek, but I
don’t remember which other language is such that Abby wants to hire some-
one who speaks it. [the “different people for different languages” reading]
b. [for (b)] I know the radio played a song that Ringo wrote, but I don’t know
who else is such that the radio played a song that he wrote. [the “different
musicians for different songs” reading]
Let us call these types of readings “covariant readings”, as in Section .. If we assume,
as we did in Section ., that the covariant reading in contrast sluicing arises only if the
remnant wh-phrase takes scope over the complex NP, the availability of these readings
seems to indicate that the wh-phrase actually moves out of the island in the second
conjunct, as indicated in (), and that the island violation is ameliorated in English
sluicing, as has been widely accepted in the literature. 
() nd conjunct:
... CP

DP C⬘
<remnant>
C TP

... COMPLEX NP ...

... t ...

If we investigate the phenomena more closely, however, a different picture emerges.

.. Contrast sluicing with a prepositional remnant


Consider () and (), where the remnant occurs with a preposition. The “a” exam-
ples are simplex-clause cases, while the “b” examples are embedded-clause cases.
() a. I remember Abby works on GREEK, but I don’t remember on which OTHER
language. [intended to mean: “I don’t remember on which OTHER language
Abby works”]
b. I remember Bill said that Abby worked on GREEK, but I don’t remember on
which OTHER language. [intended to mean: “I don’t remember on which
OTHER language Bill said Abby worked”]

 In what follows, I will adopt without discussion the widely accepted assumption that overt
wh-movement is involved in the derivation of English sluicing. See Chung et al. () for a claim that the
wh-remnant is base-generated in SpecCP. See also Kimura () for a claim that the remnant wh-phrase
can stay in situ in some cases of sluicing.
Island-sensitivity in Japanese sluicing 157

() a. I remember that Bill sold drugs to JOHN, but I don’t remember to who(m)
ELSE. [intended to mean: “I don’t remember to who(m) else Bill sold drugs”]
b. I remember that Mary said Bill sold drugs to JOHN, but I don’t remember to
who(m) ELSE. [intended to mean: “I don’t remember to who(m) else Mary
said Bill sold drugs”]
Speakers who accept pied-piping of a preposition in wh-questions find these sluicing
examples acceptable.
Now if we combine the pied-piped remnant and modification by else/other, island
effects emerge, just as in the analogous cases of Japanese sluicing observed in Sec-
tion .. Compare (a) and (b), which differ from each other only with respect to
the presence vs absence of the preposition on the wh-phrase.
() a. I remember Abby wants to hire someone who works on GREEK, but I don’t
remember which OTHER language.
b. I remember Abby wants to hire someone who works on GREEK, but I don’t
remember on which OTHER language.
While the covariant reading in (), where different people work on different lan-
guages, appears to be available in (a), it is not in (b).
() I remember Abby wants to hire someone who works on Greek, but I don’t
remember which other language is such that Abby wants to hire someone who
works on it. [the “different people for different languages” (covariant) reading]
Likewise, while (a) seems to give rise to the covariant reading in (), (b) does
not yield such a reading.
() a. I know that the police arrested a man who had sold drugs to JOHN in LA,
but I don’t know who ELSE.
b. I know that the police arrested a man who had sold drugs to JOHN in LA,
but I don’t know to who(m) ELSE.
() I know that the police arrested a man who had sold drugs to John in LA, but
I don’t know who else is such that the police arrested a man who had sold drugs
to him. [the “different dealers for different people” (covariant) reading]
In () is a descriptive generalization that emerges regarding the availability of the
covariant reading in English sluicing in island contexts.
() The covariant reading is available in contrast sluicing with a non-prepositional
wh-phrase remnant, while it is not available in contrast sluicing with a prepo-
sitional wh-phrase remnant.
One complication here is that there are speakers who do not readily accept
wh-questions with a pied-piped preposition. For such speakers, the judgments on
158 Teruhiko Fukaya

() and () seem to be blurred. In order to obtain clearer judgments, it is necessary
to test with cases where the pied-piping of the preposition is acceptable even for
such speakers. As Rodman (), cited in Postal (: –), points out, there are
instances in English in which pied-piping of a preposition is obligatory.  Consider
the examples in () and ().
() a. ∗ Whose sake did you take the stand for?
b. For whose sake did you take the stand?
() a. ∗ What way does John do syntax in?
b. In what way does John do syntax?
Pied-piping of the preposition is obligatory in these examples.
Now consider examples of sluicing with a prepositional remnant in () and ().
() a. I remember Bill took the stand for JOHN’s sake, but I don’t remember for
who ELSE’s sake. [intended to mean: “I don’t remember for who else’s sake
Bill took the stand”]
b. I remember Abby said Bill took the stand for JOHN’s sake, but I don’t
remember for who ELSE’s sake. [intended to mean: “I don’t remember for
who else’s sake Abby said Bill took the stand”]
() a. I know that John did syntax in (the) OPTIMALITY-THEORETICAL way,
but I don’t remember in what OTHER way. [intended to mean: “I don’t
remember in what other way John did syntax”]
b. I know that Abby said John did syntax in (the) OPTIMALITY-
THEORETICAL way, but I don’t remember in what OTHER way. [intended
to mean: “I don’t remember in what other way Abby said John did syntax”]
The acceptability of these examples indicates that for who else’s sake and in what other
way are usable in sluicing. Having established that, we can now use these expressions
in our experiments. Consider ().
() I remember Abby wants to talk to a person who took the stand for JOHN’s
sake, but I don’t remember for who ELSE’s sake.
As in the case of (b) and (b), the covariant reading in (b) is not available in
(). The reading available in () is the single-witness reading in (a).
() a. . . . I don’t remember for who else’s sake he (= the person who Abby wants
to talk to) took the stand. [the single-witness reading]
b. . . . I don’t remember who else is such that Abby wants to talk to a person
who took the stand for his sake. [the multiple-witness (covariant) reading]

 I am grateful to Chris Potts (personal communication, March ) for drawing my attention to
such cases.
Island-sensitivity in Japanese sluicing 159

Likewise, () does not give rise to the covariant reading in (b). The reading
available in () is the single-linguist reading in (a).
() I know USC hired someone who did syntax in the OPTIMALITY-
THEORETICAL way, but I don’t remember in what OTHER way.
() a. I don’t remember in what other way she or he (= the person who USC hired)
did syntax [the single-linguist reading]
b. I don’t remember what other way is such that USC hired someone who did
syntax in it [the multiple-linguist (covariant) reading]
It thus seems reasonable to conclude that we have established the descriptive gener-
alization in ().

.. Two types of sluicing in English


The table in () summarizes the data of contrast sluicing that we observed in the
preceding subsections.
()
Remnant type Covariant Examples
reading

a. non-prepositional wh-phrase available (), (a), (a)


b. prepositional wh-phrase unavailable (b), (b), (), ()

If we assume that the unavailability of covariant readings indicates the presence of


island effects, as we did in the case of Japanese sluicing, it follows that contrast sluicing
with a prepositional remnant is sensitive to islands. The fact that covariant readings
are available in contrast sluicing with a non-prepositional remnant, on the other hand,
seems to indicate that it is not sensitive to islands.
Now let us consider how the facts summarized in () can be captured under
such an assumption. Let us start with (b). I assume that the wh-phrase in sluicing
undergoes regular wh-movement, and that this movement is sensitive to islands, as
schematically shown in ().
() ... CP

DPi C⬘
<remnant>
C TP

... COMPLEX NP ...

... ti ...
160 Teruhiko Fukaya

This is why the covariant reading is not available in (b). Note that we cannot
attribute the island-sensitivity in (b) to the restriction on the movement of the
correlate as Merchant () does, because the extra TP layer in the first conjunct in
examples like (b), (b), (), and () should enable the correlate to move to the
position parallel to that of the wh-phrase remnant in the second conjunct, and also
because (a) and (b) have the same type of correlate, namely, a definite DP. Note
also that it is imperative to further assume that the island is not a PF island. If it were,
the island violation would be ameliorated under deletion, and the type of sluicing
in (b) would be expected to yield the covariant reading. This conclusion runs
counter to the widely accepted assumption that island effects are ameliorated under
sluicing.
Let us turn to (a). Since we must now consider that the island violation under
discussion is not ameliorated under sluicing, as we have seen in the case of (b),
sluicing with a non-prepositional remnant would be expected to exhibit an island
effect as well if the only derivation available were the same derivation as sluicing
with a prepositional remnant, as in (b). Since sluicing with a non-prepositional
remnant does not exhibit island effects, however, we are led to conclude that it must
have a representation distinct from the one for sluicing with a prepositional remnant.
Considering that this type of sluicing in English exhibits properties similar to Japanese
non-case-marked sluicing, as observed in Section ., I wish to suggest that it has the
same type of representation as Japanese non-case-marked sluicing. That is, I suggest
that it can have a representation as indicated in () with that referring to a property
denoted in the first conjunct.

() [CP [wh-phrase] C [TP that is]]

For example, the second conjuncts in (a) and (a) can have the structures in (a)
and (b), respectively.

() a. [for (a)] I don’t remember [CP which other language [TP that is]]
b. [for (a)] I don’t know [CP who else [TP that is]]

In (a), that is interpreted roughly as “the property of being the language such that
Abby wants to hire someone who works on it”. In (b), that is interpreted as “the
property of being the person such that the police arrested a man who had sold drugs
to him in LA”. These interpretations give rise to the apparent covariant reading as
in ().

() a. [for (a)] I don’t remember which other language has the property of
being the language such that Abby wants to hire someone who works on it.
b. [for (b)] I don’t know who else has the property of being the person such
that the police arrested a man who had sold drugs to him in LA.
Island-sensitivity in Japanese sluicing 161

I thus claim that two types of sluicing be identified in English, analogous to the two
types of sluicing in Japanese (case-marked sluicing and non-case-marked sluicing),
as in (). (The material in the angle brackets is not phonetically realized.)
() a. Prepositional-remnant sluicing:
[CP [PP P [DP wh-phrase]]i C < [TP . . . [VP V . . . t i . . .]] >]
b. Non-prepositional-remnant sluicing:
(i) [CP [DP wh-phrase]i C < [TP . . . [VP V . . . t i . . .]] >]
(ii) [CP [DP wh-phrase]i C <[TP that is t i ]>]
Prepositional-remnant sluicing has the same derivation as the non-elliptical
wh-question sentence except that the material other than the wh-phrase remnant
is deleted. Non-prepositional-remnant sluicing, on the other hand, is ambiguous
between two representations, as in the case of Japanese non-case-marked sluicing:
a structure that is equivalent to that of prepositional-remnant sluicing and a copular
structure that is radically different from that of prepositional-remnant sluicing, as
schematized in (b-ii).  Prepositional-remnant sluicing cannot have the represen-
tation equivalent to (b-ii), presumably because () is not possible.
() a. ∗ on which other language that is
b. ∗ to whom else that is
Given (i) that prepositional-remnant sluicing with an indefinite correlate as in ()
does not seem to exhibit island effects, (ii) that prepositional-remnant sluicing cannot
have a representation analogous to (b-ii), and (iii) that the relative-clause island
is not a PF island that is ameliorated by deletion, it follows that the local-resolution
analysis of sluicing originally proposed by Merchant (: Chapter ) (or at least
some version of it) should still be an option available for English sluicing as well. That
is, (a), for example, should have the structure in () before deletion, as Merchant
(: Chapter ) originally claimed.
() [CP [PP on [DP which Balkan language]]i C [TP she works ti ]]
With she interpreted as “the person who Abby hired”, the second conjunct of (a)
gives rise to the reading “I don’t remember which Balkan language the person Abby
 The same distinction has been suggested for English stripping by Hoji (: Chapter ) and Fukaya
and Hoji (: footnote ). It was observed there that non-prepositional remnant stripping gives rise to
the sloppy reading in the weak crossover context as in (i-B) while such a reading is not available in the case
of prepositional remnant stripping, as in (i-B ).
(i) A: Hisi students often talk to Johni .
B: Bill, too. /Well, Bill, too.
(≈ ‘That is true of Bill, too.’, where ‘That’ ≈ ‘one’s students often talking to him’)
[intended to mean ‘Hisj students often talk to Billj , too.’]
B : ∗ To Bill, too. /∗ Well, to Bill, too.
[intended to mean ‘Hisj students often talk to Billj , too.’]
162 Teruhiko Fukaya

hired works on”, which is equivalent to the reading that would be available if the
movement of the wh-remnant across the island were possible, i.e. the reading “I
don’t remember which Balkan language is such that Abby hired someone who works
on it”.
Note that non-prepositional-remnant sluicing with an indefinite correlate as in (b)
and (b) is also structurally ambiguous. Example (b), for example, is ambiguous
between the two representations in (), both of which give rise to apparent covariant
readings with she interpreted as “the person who they want to hire” in (a), and with
that interpreted as “the property of being the language such that they want to hire
someone who speaks it” in (b).

() b. They want to hire someone who speaks a Balkan language, but I don’t remem-
ber which.

() a. [CP [DP which Balkan language]i C [TP she speaks ti ]]


b. [CP [DP which Balkan language]i C [TP that is ti ]]

.. Summary
In the previous subsections, I have examined contrast sluicing in English, and
observed the facts summarized in (). Assuming what we have argued for Japanese
sluicing in the preceding sections, it was claimed that contrast sluicing with a
prepositional remnant as in (b) is sensitive to islands, while contrast sluicing with
a non-prepositional remnant as in (a) appears not to be. It was shown that island-
sensitivity in (b) cannot be attributed to the property of the correlate in the first
conjunct as Merchant () claims. It was then claimed that the violation of the
relative-clause island cannot be ameliorated in sluicing contexts, contra the widely-
held view that islands are PF phenomena. In order to account for the new data, it was
proposed that there should be two types of sluicing in English, analogous to the two
types of sluicing in Japanese.

. Conclusion
In this paper I have investigated island-sensitivity in various types of Japanese sluicing,
and I have discussed some implications that our examination of Japanese sluicing has
for the analysis of English sluicing. I first showed that, although it appears otherwise
at first glance, case-marked sluicing in Japanese is sensitive to syntactic islands such as
relative clauses and adjuncts. I did so by scrutinizing whether certain types of interpre-
tation are available in Japanese case-marked sluicing with various types of correlates
and remnants. I claimed that the lack of non-local readings in certain types of sluicing
is due to the island constraint. I then showed how the copy theory of ellipsis resolution
proposed in Fukaya and Hoji () combined with a version of Merchant’s (:
Island-sensitivity in Japanese sluicing 163

Chapter ) proposal for local movement can account for the intricate properties of
Japanese case-marked sluicing. I also investigated the properties of non-case-marked
sluicing in Japanese and suggested that the apparent availability of the non-local
reading is due to the availability of a kind of copula structure proposed in Fukaya and
Hoji (). In light of what was discovered in Japanese sluicing, I also re-examined
English sluicing and suggested that the relative-clause island in English is not to be
regarded as a PF representational island that can be ameliorated by PF deletion. For
the apparent lack of island effects in some types of English sluicing, I suggested that
it is necessary to assume Merchant’s (: Section .) local-resolution strategy as
well as a radically distinct representation comparable to that for Japanese non-case-
marked sluicing.
Before concluding the paper, I wish to briefly mention some issues to be addressed
in future research. First, I have claimed that island effects in Japanese are brought
about by the restriction on Constituent Raising in (e), which states that it is sensitive
to syntactic islands; however, this is essentially a stipulation, and how to derive the
restriction theoretically still remains an issue. Second, locality effects are observed in a
variety of phenomena in Japanese, such as the case-marked cleft construction, scram-
bling, case-marked stripping, and the case-marked topic construction, in addition
to case-marked sluicing discussed in this paper. Whether or not the effects observed
in these constructions are a uniform phenomenon is another issue that needs to be
addressed.  Third, given the facts regarding English sluicing discussed in this paper,
it may now be difficult to maintain that the relative-clause island in English is a PF
island that can be ameliorated by deletion. Thus, the issue of how to theoretically
re-characterize islands in English (at least the relative-clause island) also needs to be
addressed. It is hoped that we will be able to obtain insight into such issues by paying
close attention to the (un)availability of the non-local readings.

 In some cases the observed locality effects seem to be clause-boundedness, rather than islandhood.
In this connection, it is worth noting that some speakers find examples like (b) and (b) unacceptable
under the multiple-event reading where the event denoted in the matrix clause (“the police announced . . .”
in those cases) takes place more than once. This points to the possibility that the movement involved in
the derivation of case-marked sluicing in Japanese is clause-bounded rather than island-sensitive, and the
source of the unacceptability of examples like (b) and (b) under the multiple-event reading might have
to be investigated. I wish to address the relevant issues in a separate paper.

Sluicing without wh-movement


in Malagasy∗
I L E A NA PAU L A N D E R IC P OT SDA M

. Introduction
Sluicing is the construction illustrated in (a), in which an interrogative clause is
reduced to only a wh-phrase. A standard analysis of sluicing (Ross , Merchant
) is that it is movement of a wh-phrase to the specifier of C , followed by deletion
of the TP below the wh-phrase, as shown in (b).
() a. Somebody left and I know who.
b. Somebody left and I know [CP whoi [C C [wh] [ TP t i left]]].
If wh-movement is a prerequisite for sluicing, the prediction is that wh-in-situ lan-
guages should not have this construction. For one wh-in-situ language, Malagasy, a
Western Austronesian language spoken on the island of Madagascar, this prediction
is apparently incorrect:
() nandoko zavatra i Bao fa hadinoko hoe inona.
paint thing Bao but forget.sg comp what
‘Bao painted something but I forget what.’
The goal of this chapter is to explore how wh-in-situ languages, Malagasy in particular,
can have sluicing without wh-movement. Our primary conclusion is that sluicing is
not in fact a unified syntactic phenomenon and that different languages use different
syntactic means to arrive at the same surface form. To avoid confusion, we will hence-
forth use the term “sluicing-like construction” (SLC) to describe a construction in

∗ The authors would like to thank the following Malagasy speakers for their help with the data: Charlotte
Abel-Ratovo, Tina Boltz, Noro Brady, Annie Rasoanaivo, Hasina Randriamihamina, Voara and Bodo Ran-
drianasolo, and Vololona Rasolofoson. We would also like to thank the audience at the Chicago Linguistics
Society  and two anonymous reviewers for their helpful comments and questions. Any remaining errors
are our own. This research was supported in part by a Canada Research Chair (Tier II) to Ileana Paul and
NSF grant BCS- to Eric Potsdam.
Sluicing in Malagasy 165

which an interrogative clause is realized only as a wh-phrase, regardless of its under-


lying syntactic derivation. We reserve the term “sluicing” for Ross and Merchant’s
analysis of SLCs in which there is wh-movement followed by TP-deletion.
The chapter is organized as follows. Section . presents an overview of analyses
of SLCs in several wh-in-situ languages. It demonstrates that such languages use a
variety of syntactic means to arrive at what looks superficially like English sluicing but
is not, upon closer analysis. It is in this context that we turn to Malagasy. Section .
presents some basic facts about Malagasy word order. Section . turns to question
formation and shows that Malagasy is in fact a wh-in-situ language. This observation
is not uncontroversial because Malagasy appears to have wh-movement. We provide
evidence that such apparent fronting is in fact a base-generated pseudo-cleft structure,
not wh-movement. Section . presents the Malagasy SLC along with two possible
analyses. We reject a sluicing analysis in which the Malagasy SLC involves exceptional
wh-movement that is licensed by deletion of otherwise illicit structure at phonolog-
ical form (PF). In Section . we provide evidence for our own analysis, that the
Malagasy SLC is derived via predicate-fronting of the wh-phrase followed by deletion.
Predicate-fronting has been independently proposed by other researchers as a general
mechanism to derive the predicate-initial (VOS) word order of some Austronesian
languages, so our analysis dovetails well with recent theoretical proposals. Section .
contains conclusions and further issues.

. Typology of sluicing


Wh-in-situ languages are highly relevant to studies of sluicing because the standard
sluicing analysis leads us to expect that such languages will not have sluicing, there
being no wh-movement operation to feed the deletion. Contrary to expectations,
however, it has been documented that many wh-in-situ languages do have a sluicing-
like construction (SLC)—a construction that looks like sluicing. Merchant ()
briefly considers the relevance of wh-in-situ languages for his analysis of sluicing,
and concludes, based on data from Chinese and Japanese, that “what appears to
be sluicing in these languages is the result of operations different from the move-
ment + deletion derivation found in languages with overt wh-movement.” We agree
with Merchant’s conjecture, and in the remainder of this section we briefly review
the diverse alternative strategies that have been proposed for the wh-in-situ lan-
guages Japanese, Chinese, and Javanese. The paper goes on to propose that the wh-
in-situ language Malagasy appeals to yet a different strategy, in line with Merchant’s
proposal.
In Chinese and Japanese, it has been suggested that the SLC resembles a cleft, with
the wh-phrase as a complement to a copular verb. Japanese has been characterized
as deriving its SLC using a reduced cleft with a deleted copula (see Merchant ()
166 Ileana Paul & Eric Potsdam

and references therein; see also Fukaya and Hoji (), Nakamura (this volume), and
Fukaya (this volume)):
() dareka-ga sono hon-o yon-da ga, watashi-wa [CP [TP proexpl
someone-nom that book-acc read-pst but I-top
dare da] ka] wakaranai.
who cop Q know.neg
‘Someone read the book, but I don’t know who (it is).’
In Chinese, it has been proposed that SLCs are formed via a similar copular construc-
tion with a null anaphoric pronoun, not involving any deletion (Adams , Adams
and Tomioka this volume, Wang , Wei ):
() Xiaomei mai le yi-jian liwui , danshi ta bu gaosu wo [TP proi shi
Xiaomei buy asp one-cl present but she not tell sg is
shenme].
what
‘Xiaomei bought a present, but she didn’t tell me what (that was).’
As Merchant concludes, these languages have something that looks like a sluice, but
arguably without wh-movement + deletion. For in-depth discussion of the intricacies
of SLCs in Japanese and Chinese, see Chapters , , and  of this volume.
There is another class of wh-in-situ languages, which adopt a different strategy for
SLCs. In these languages, it has been argued that the wh-phrase moves, but not via
wh-movement. Javanese (Adams , ) has an SLC in which it is suggested that
there is focus movement of a wh-phrase to a clause-initial position followed by TP-
deletion: 
() umpamane Tika lunga ibune kudu ngerti [FocP [neng ngendi]i
if Tika go mother must know loc where
[TP dheweke lunga t i ]].
sg go
‘If Tika goes somewhere, her mother must know where (she goes).’

In what follows, we propose that Malagasy illustrates a similar, but distinct, possibility:
The wh-phrase moves via predicate-fronting, and, as in Javanese and English, the TP
is then deleted.
() nisy olona nihomehy ka nanontany ianao hoe [FP [vP iza]i
exist person laugh and ask you comp who
[TP no nihomehy t i ].
prt laugh
‘Someone laughed and you asked who (the one who laughed was).’
 In fact, as discussed by Adams (), Javanese appears to have three different sluicing strategies,
depending on the nature of the wh-phrase (NP, PP, non-PP adjunct).
Sluicing in Malagasy 167

If the analysis of Malagasy is correct, it strengthens the hypothesis that different


languages may arrive at the same surface form via different syntactic means, and SLCs
do not constitute a unified analytical class. Which strategy (or strategies) a language
uses to derive its SLCs will depend upon the syntactic mechanisms independently
available in the language.  Wh-in-situ languages thus make a valuable contribution
to the study of sluicing phenomena. They support the claim that the English-type
derivation is not the only route to a sluicing-like surface representation. 

. Basic Malagasy word order


Malagasy is well-known for having fairly rigid VOS word order, (). More gener-
ally, the predicate can be any phrasal category, not only VP; so the language can be
described as predicate-initial, (). 

() mividy ny akoho i Bao.


buy the chicken Bao
‘Bao is buying the chicken.’

() a. [vorona ratsy feo]NP ny goaika.


bird bad voice the crow
‘The crow is a bird with an ugly voice.’
b. [faly amin’ ny zanany]AP Rasoa.
proud prep the child.sg Rasoa
‘Rasoa is proud of her children.’
c. [any an-tsena]PP Rakoto.
prep acc-market Rakoto
‘Rakoto is at the market.’

One exception to this relatively fixed word order is that complement CPs are extra-
posed to a clause-final position yielding VSO order, (a). The immediately post-verbal
position for the CP is impossible, (b). 

 See Hoyt and Teodorescu (this volume) for similar conclusions based on the difference between
English, Romanian, and Japanese sluicing.
 A fourth typological possibility is discussed by Bhattacharya and Simpson in Chapter  of this volume,
where it is argued that what appears to be a wh-in-situ language, Bangla/Bengali, is in fact a language with
regular masked wh-movement. SLCs in Bangla/Bengali are therefore formed by regular wh-movement and
TP-deletion as in English and other languages with clearer instances of overt wh-movement.
 There is considerable debate in the literature over the nature of the clause-final DP, whether it is a
subject or an A topic-like element. We continue to refer to it as a subject for convenience, without taking
a stand on the issue. See Pearson () for discussion. As pointed out by the anonymous reviewer, if the
clause-final DP is located in a topic projection within the CP-layer, then the deleted constituent in the
Malagasy SLC must be larger than TP.
 Certain CP-like constituents (control and raising-to-object complements) do not obligatorily
extrapose.
168 Ileana Paul & Eric Potsdam

() a. milaza Rabe [fa nividy ny akoho i Bao].


say Rabe that buy the chicken Bao
‘Rabe says that Bao bought the chicken.’
b. ∗ milaza [fa nividy ny akoho i Bao] Rabe.
say that buy the chicken Bao Rabe

. Questions in Malagasy


.. Two types of wh-question
Malagasy has two strategies for forming information questions. When questioning
non-subjects, wh-in-situ is possible (see Sabel () for discussion), ().

() a. nividy inona i Be? object


buy what Be
‘What did Be buy?’
b. nividy vary taiza i Be? adjunct
buy rice where Be
‘Where did Be buy the rice?’
c. ∗ nividy vary iza? ∗ subject

buy rice who


(‘Who bought the rice?’)

There is no evidence of wh-movement in such examples. Tests for covert movement


show that the wh-phrase does not seem to move even at LF: Wh-in-situ is not sensitive
to islands, (), and does not trigger weak crossover, ().

() namangy ny lehilahy izay nanasa inona i Be?


meet the man rel wash what Be
(Lit. ‘Be met the man who washed what?’)
‘What did Be meet the man who washed?’
() manaja an’iza ny reniny?
respect who.acc the mother.sg
‘Whoi does hisi mother respect?’

The second question strategy is that, for non-complements (subjects and adjuncts),
the wh-phrase appears at the beginning of the clause followed by the particle no  , ().
(See Keenan (), MacLaughlin (), Paul (), Sabel (), and Law ()
for further description.)

 Traditional Malagasy grammarians (Malzac ) have suggested that no is a determiner that is
diachronically related to the determiner ny. Potsdam () suggests that it is a relative-clause comple-
mentizer. Its exact analysis is not directly relevant.
Sluicing in Malagasy 169

() a. ∗ inona no nividy i Be? ∗ object

what prt buy Be


(‘What did Be buy?’)
b. taiza no nividy ny vary i Be? adjunct
where prt buy the rice Be
‘Where did Be buy the rice?’
c. iza no nividy ny vary? subject
who prt buy the rice
‘Who bought the rice?’
While such examples might appear to involve wh-movement with a question
complementizer no immediately following the fronted wh-phrase, we will show
in the following subsection that they are actually pseudo-clefts (see Dahl (),
Paul (), Potsdam (), and Kalin (); but see Law () for a somewhat
different view). As schematized in (), the initial wh-phrase is the predicate of the
clause, also called the focus or pivot. The remaining material is a headless relative in
subject position. The wh-phrase has not actually undergone wh-movement; rather,
it has its clause-initial position by virtue of being a predicate in a predicate-initial
language. The only A -movement in the structure is null-operator movement in the
relative clause, as shown.
() [predicate iza] [subject/headless relative no Opi nihomehy t i ]?
who prt laugh
(Lit. ‘The one who laughed is who?’)
‘Who laughed?’
Embedded wh-questions take the same form, as illustrated in (). They are intro-
duced by the formative hoe, which we assume is a complementizer. They are obliga-
torily extraposed. 
() a. nanontany ianao [hoe iza no nihomehy].
ask you comp who prt laugh
‘You asked who laughed.’
b. tsy fantatro hoe taiza no nividy vary i Be.
neg know.sg comp where prt buy rice Be
‘I don’t know where Be bought the rice.’

.. Evidence for the pseudo-cleft structure


In this section we provide evidence supporting the pseudo-cleft analysis of wh-
questions, repeated in (a). We argue against a wh-movement analysis, schematized
 Hoe is in fact a defective verb meaning ‘say’. We assume that its use here is as a complementizer. It may
introduce either embedded questions, as in (), or direct quotations. It is possible that the two uses are
related, but we take no stand on the issue.
170 Ileana Paul & Eric Potsdam

in (b), in which wh-questions are derived by ordinary wh-movement to the specifier


of CP, as in English. Further details and argumentation can be found in Potsdam
().

() a. [predicate wh-phrase] [subject no Opi . . . t i ]  pseudo-cleft analysis


b. [CP wh-phrasei [C no [TP . . . t i ]]] × wh-movement analysis

Our two arguments in favor of the pseudo-cleft analysis and against the wh-movement
analysis can be summarized as follows. (i) Wh-questions show parallels with the
focus construction, which Paul () analyzes as a pseudo-cleft. The parallels
are immediately accounted for if both constructions have the same structure. (ii) The
initial wh-phrase in wh-questions behaves like a predicate, as is expected under the
pseudo-cleft analysis but not under the wh-movement analysis.
Malagasy has a focus construction, illustrated in (a), that appears similar to wh-
questions. The focus construction is most naturally translated into English with a
cleft or pseudo-cleft. Paul () advances a pseudo-cleft analysis of the construction,
assigning (a) the structure in (b). The initial focused element is the predicate of
the clause, and the subject is a headless relative clause.

() a. Rasoa no nihomehy.


Rasoa prt laugh
‘It was Rasoa who laughed.’
b. [[predicate Rasoai ] [subject/headless relative no Opi nihomehy t i ]].
Rasoa prt laughed
Lit. ‘The one who laughed was Rasoa.’

There are a number of parallels between the focus construction and wh-questions,
suggesting that they should share a single syntactic structure. First, both are formed
by preposing a constituent and following it immediately with the particle no. Second,
the two constructions have a similar focus interpretation of the initial XP: Wh-phrases
indicate a request for new information in the same way that focused XPs supply
new information. Third, the two constructions are subject to an identical fronting
restriction that we already saw above for wh-questions: Only subjects and adjuncts
can be fronted (see Keenan () and others). The same restriction holds of the focus
construction, ().

() a. ∗ ny vary no nividy Rabe. ∗ object

the rice prt buy Rabe


(‘It was the rice that Rabe bought.’)
b. omaly no nividy ny vary Rabe. adjunct
yesterday prt buy the rice Rabe
‘It was yesterday that Rabe bought the rice.’
Sluicing in Malagasy 171

c. Rabe no nividy ny vary. subject


Rabe prt buy the rice
‘It was Rabe who bought the rice.’
Analyzing wh-questions as clefts immediately accounts for these parallels. They are
unexplained or at least accidental under the wh-movement analysis, since the focus
constructions and wh-questions would have very different structures.
The pseudo-cleft analysis is also supported by observations that the initial wh-
phrase behaves like a predicate. There are a number of verbal elements that flank the
predicate in Malagasy and thus help to identify it. For example, the floating quantifiers
daholo ‘all’ and avy ‘each’, and the VP-adverb foana ‘always’, are post-predicate particles
and must immediately follow the predicate in VOS clauses:
() a. namaky ny boky daholo ny ankizy.
read the book all the child
‘All the children read the book.’
b. any an-tsena foana Rakoto.
there acc-market always Rakoto
‘Rakoto is always at the market.’
For concreteness, we assume that these particles are right-adjoined to vP. Under the
pseudo-cleft analysis, then, these particles should immediately follow the wh-phrase,
as illustrated in (a). Under the wh-movement analysis, on the other hand, we expect
to find these particles at the end of the clause, (b). 
() a. CLEFT ANALYSIS b. WH-MOVEMENT ANALYSIS
TP CP

T⬘ DP whi C⬘

T vP no ... C TP
no
vP particle T⬘ ti

wh T vP

vP particle

...

As predicted by the pseudo-cleft analysis, these elements immediately follow a


wh-phrase in questions:

 In these trees, the subject/topic is in a rightward specifier of TP. We will modify this clause structure
below. The purpose of these trees is to illustrate the position of particles—the precise position of the
subject/topic is immaterial at this point.
172 Ileana Paul & Eric Potsdam

() a. iza daholo no namaky ny boky?


who all prt read the book
‘Who all read the book?’
b. iza foana no any an-tsena?
who always prt there acc-market
‘Who is always at the market?’
Other post-predicate particles that behave the same way include the exclamative
element anie and the parenthetical hono ‘so they say’.
Malagasy also has pre-predicate particles. The modal tokony ‘should’ and the
emphatic tena ‘indeed’ must immediately precede the predicate in VOS clauses:
() a. tokony hamangy an-dRakoto Rasoa.
should visit acc-Rakoto Rasoa
‘Rasoa should visit Rakoto.’
b. tena nanapaka bozaka Rabe.
indeed cut grass Rabe
‘Rabe indeed cut the grass.’
The pseudo-cleft analysis correctly predicts that these particles immediately precede
a wh-phrase in wh-questions, (). The fronting analysis cannot correctly account for
the grammaticality of the examples. 
() a. tokony iza no hamangy an-dRakoto?
should who prt visit acc-Rakoto
‘Who should visit Rakoto?’
b. tena iza no nanapaka bozaka?
indeed who prt cut grass
‘Who indeed cut the grass?’
Potsdam (, ) explores such data in more detail, but even at this level of
presentation it can be seen that the data make sense if wh-questions are pseudo-clefts
in which the initial wh-phrase is a predicate, not a fronted element. Under the wh-
movement analysis, the placement of the various elements is unexpected because the
wh-phrase is not a predicate but is very high in the clause structure. Such elements
would have to have special distribution statements for wh-questions, different from
ordinary clauses.
We conclude that Malagasy has no wh-movement. Wh-questions use either an in
situ or a pseudo-cleft strategy. This sets up an analytical challenge because, as we show
in the next section, Malagasy has an SLC.

 Note that both sets of particles can in general appear further to the right in the structure (Potsdam
). This position does not distinguish the two analyses, however, because there is also a vP-adjunction
site within the headless relative for these particles under the pseudo-cleft analysis.
Sluicing in Malagasy 173

. Malagasy sluicing


Before introducing Malagasy SLC examples, recall the English example, repeated
from ():
() a. Somebody left and I know who.
b. Somebody left and I know [CP whoi [C C [wh] [TP t i left]]].
In such examples, we will call the missing material the “sluiced clause”, and indicate
it with strikethrough. The “remnant” is the wh-phrase that remains (who above), and
the “correlate” is the XP corresponding to the wh-phrase (somebody above) in the
“antecedent clause”.
Two examples of the Malagasy SLC are given in ().
() a. nandoko zavatra i Bao fa hadinoko hoe inona.
paint thing Bao but forget.sg comp what
‘Bao painted something but I forget what.’
b. nisy olona nihomehy ka nanontany ianao hoe iza.
exist person laugh and ask you comp who
‘Someone laughed and you asked who.’
If such examples instantiate genuine sluicing, they are surprising, because wh-
movement prior to the deletion would be required. We have just argued, however,
that Malagasy has no wh-movement. If the SLCs are not sluicing, the questions arises
as to how such examples can be derived. In what follows, we propose two analyses.
The first, in Section .., suggests that there actually is wh-movement, despite our
earlier conclusions about the syntactic structure of Malagasy wh-questions. Under
this solution, the Malagasy SLC actually is sluicing. We reject this analysis, and pro-
pose instead, in Section .., that the Malagasy SLC does involve deletion, but that
the input configuration for TP-deletion is derived not by wh-movement but by a
general predicate-fronting operation that exists independently to derive VOS word
order. This analysis is compatible with our conclusions about the structure of wh-
questions above, and supports the claim that languages arrive at SLCs by different
syntactic means.

.. Deletion repair


One analysis of the Malagasy SLC is that, despite appearances, the SLC examples do
involve wh-movement and thus are regular instances of sluicing. Such an analysis of
SLCs in Malagasy might naturally be desirable in that it appeals to an analysis that is
well-motivated for other languages and does not require positing any new syntactic
mechanisms. The derivation of () would be as in (), parallel to the English sluicing
derivation.
 The antecedent clause in this example takes the form of an existential construction because indefinite
subjects are impossible in Malagasy (Keenan ).
174 Ileana Paul & Eric Potsdam

() nisy olona nihomehy ka nanontany ianao hoe [CP izai [TP
exist person laugh and ask you comp who
nihomehy t i ]].
laugh
‘Someone laughed and you asked who’

()
CP

C CP
hoe
DPj C⬘

iza C TP
‘who’ ÞØ
nihomehy tj
‘laughed’

We will call this the Deletion Repair analysis: Malagasy has wh-movement just in case
deletion eliminates the TP containing the trace of wh-movement. We might assume
that Malagasy elsewhere does not show wh-movement because it would violate some
general movement restriction in the language, and that the deletion somehow amelio-
rates the violation, perhaps by eliminating the offending trace. Such an analysis could
receive support from the observation that sluicing apparently does rescue violations
of constraints on movement, notably island constraints (Ross ) (data from Chung
et al. () and Merchant ()):

() a. Complex noun phrase constraint


They want to hire someone who speaks a Balkan language, but I don’t
remember which they want to hire someone who speaks.
b. Wh-island
Sandy was trying to work out which students would be able to solve a certain
problem, but she wouldn’t tell us which one she was trying to work out which
students would be able to solve.
c. COMP-trace effect
It has been determined that someone will be appointed, but I can’t remember
who it has been determined that will be appointed.

Lasnik () and Kennedy and Merchant () propose specific analyses of this
genre in which an illicit movement is rendered licit by PF-deletion.
In what follows, we present four problems with the Deletion Repair analysis.
Further details of the argumentation are discussed in Potsdam (). First, if wh-
movement is to the specifier of CP, as is often assumed in languages with overt
Sluicing in Malagasy 175

wh-movement, such as English, German etc., it is unexpected that the wh-phrase


follows rather than precedes the embedded question complementizer hoe:

() nandoko zavatra i Bao fa . . .


paint thing Bao but
a. . . . hadinoko hoe inona.
forget.sg comp what

b. . . . hadinoko inona (hoe).
forget.sg what comp
‘Bao painted something but I forget what.’

Second, the Deletion Repair analysis predicts that accusative-case wh-phrase rem-
nants should be grammatical, because wh-movement should be able to target any
wh-phrase. This is incorrect, ().

() ∗ nanasa olona Rabe ka nanontany aho hoe an’iza.


invite person Rabe and asked I comp who.acc
(‘Rabe invited someone and I asked whom.’)

Note that we are working within a Deletion Repair analysis whereby deletion is
assumed to remedy any movement constraint violations. Thus, the fact that Malagasy
cannot in actuality question complements, as mentioned in Section .., is irrele-
vant, because the deletion might naturally be expected to repair the violation of this
restriction.
Third, the analysis predicts that if there were a configuration in which sluicing
could not ameliorate the movement constraint violation, such examples would be
ungrammatical. An example is sluicing with implicit correlates. English sluicing with
implicit correlates is illustrated in (). There is no overt correlate in the antecedent
clause to which the wh-remnant corresponds. The correlate is implicit.

() a. She’s reading. I can’t imagine what.


b. They’re baking a cake, but they wouldn’t say for whom.

Implicit correlates are relevant because sluicing with implicit correlates cannot violate
constraints on movement (Chung et al. , Romero , Merchant ). Sluicing
deletion is unable to rescue such derivations—contrast the ungrammatical examples
in () with the grammatical examples in ().

() a. Complex noun phrase constraint


∗ Kim knows the person who was reading but she won’t say what (she knows

the person who was reading).


b. Wh-island
∗ Agnes wondered when John would bake a cake but it’s not clear for whom

(Agnes wondered when John would bake a cake).


176 Ileana Paul & Eric Potsdam

A Deletion Repair analysis predicts that Malagasy sluices with implicit correlates
should likewise be ungrammatical; however, this is incorrect:

() nihira Rasoa fa tsy fantatro hoe inona.


sing Rasoa but neg know.sg comp what
‘Rasoa was singing but I don’t know what.’

Finally, it remains mysterious why there would be wh-movement just in instances


of sluicing. There does not seem to be any language-internal motivation for such
movement. We therefore conclude that the SLC in Malagasy does not involve oth-
erwise unavailable wh-movement. The question remains how Malagasy SLCs are in
fact derived. In the next subsection we offer an alternative solution and analysis.

.. Predicate-fronting
Our proposal, in line with Merchant’s suggestion, is that what looks like sluicing in
Malagasy makes use of some other non-wh-movement syntactic mechanism. In other
words, Malagasy’s SLC is not prototypical sluicing. The elements of our analysis are as
follows: First, SLCs involves embedded questions that are pseudo-clefts, just as root
questions are pseudo-clefts. Example () illustrates an embedded question. It takes
the form of a matrix wh-question introduced by the complementizer hoe.

() nanontany ianao hoe iza ∗ (no) nihomehy.


ask you comp who prt laugh
‘You asked who laughed.’

Second, the wh-predicate of the pseudo-cleft moves out of TP via predicate-fronting.


Recently, there have been a number of proposals in the literature that VOS word order
in Austronesian languages is derived from an underlying SVO order via predicate-
fronting (see Massam and Smallwood (), Rackowski and Travis (), Mas-
sam (), Pearson (), Aldridge (, ), Travis (), and Cole and
Hermon (); see Chung () for critical discussion). The derivation of a basic
VOS clause is as in (), in which an underlying SVO structure is transformed
into VOS by fronting the predicate phrase, vP, to the specifier of a projection FP
above TP. 

() a. mividy ny akoho i Bao.


buy the chicken Bao
‘Bao is buying the chicken.’

 The authors noted here differ in the details of the fronting analysis, in particular, in the final landing
site of the predicate. These details are not relevant for present purposes. An anonymous reviewer of the
chapter asks how tense-marking ends up on the verb. We assume a lexicalist approach to morphology: The
verb is merged into the derivation fully inflected.
Sluicing in Malagasy 177

b. FP

vPi F⬘

mividy ny akoho F TP
‘buy the chicken’
DP T⬘

i Bao T vP
‘Bao’
ti

The simplest assumption is that such predicate-fronting also occurs in (embedded)


wh-questions and as part of the derivation of SLC examples. The wh-phrase predicate
fronts, and then the TP, which no longer contains the predicate, deletes, as shown
in (b).

() a. nisy olona nihomehy ka nanontany ianao hoe iza


exist person laugh and ask you comp who
no nihomehy
prt laugh.
‘Someone laughed and you asked who (laughed).’
b.
CP

C FP
hoe
vP i F⬘

iza F TP
‘who’
DP T⬘
ÞØ
no nihomehy T vP
‘the one who laughed’
ti

The predicate-fronting account of VOS word order is motivated in part by


theory-internal considerations. By invoking predicate-fronting, it is possible to
avoid the assumption that Malagasy has some rightward and some leftward spec-
ifier positions.  Moreover, all movement is strictly leftward, rather than mixed

 For example, it may otherwise be necessary to assume a rightward specifier for the subject in VOS
sequences in the absence of (predicate) fronting of the verb and object.
178 Ileana Paul & Eric Potsdam

rightward and leftward. There are also empirical motivations. As we have already
seen in (), repeated here in (), movement of complements in Malagasy is
blocked.
() a. ∗ inona no nividy i Be? ∗ object

what prt buy Be


(‘What did Be buy?’)
b. taiza no nividy ny vary i Be? adjunct
where prt buy the rice Be
‘Where did Be buy the rice?’
c. iza no nividy ny vary? subject
who prt buy the rice
‘Who bought the rice?’
Assuming that a fronted predicate, being in a specifier, creates an island for
A -movement, and assuming that adjuncts such as taiza ‘where’ are adjoined above
vP, there is a simple explanation of this otherwise unusual restriction on extraction.
Objects are “frozen” within fronted predicates, while (high) adjuncts and subjects are
free for extraction. 
If predicate-fronting is independently part of Malagasy grammar, then it provides
the necessary movement to feed TP-deletion in the SLC examples, as shown in the
derivation in (b). In the next section we provide some evidence that the above
derivation is on the right track. Before turning to this evidence, however, we discuss
some details of the licensing of deletion.  Since at least Lobeck (), it has been
believed that deletion must be licensed by a syntactic head. We propose that it is
the complementizer hoe that licenses TP-deletion in Malagasy.  Since it is not the
complement of the complementizer hoe that deletes, but rather the complement of F
(see the tree in (b)), the head of FP must also play a role in licensing deletion. To see
how this can be implemented formally, consider Merchant’s () discussion of the
licensing conditions on sluicing in English. Merchant (: ) argues that only the
null [+wh, +Q] interrogative C licenses a null TP (i.e. sluicing). Within a Minimalist
checking theory, he claims that a feature E on the T head moves to C to be checked,
but that it can only be checked by the null [+wh, +Q] interrogative C . This feature,
as well as giving the semantics for sluicing, indicates that its sister (i.e. TP) is not to be
pronounced. Adapting this analysis to Malagasy, the feature E can be generated on F ,

 An anonymous reviewer asks if we predict that all VOS languages will show the same extraction
asymmetry. We actually don’t believe that all VOS languages necessarily involve predicate-fronting, and so
not all VOS languages would necessarily be subject to this restriction. We leave this issue open to further
empirical research, but see Chung () for some relevant discussion.
 For discussion of the identity conditions on sluicing, we refer the reader to Potsdam (), who
argues in favor of semantic rather than syntactic identity.
 At this point we take no stand on whether other complementizers also license deletion. Initial data
indicate that they may, but we have not adequately explored the facts.
Sluicing in Malagasy 179

and it enters into an Agree relation with C (hoe) without movement. As in English,
E marks its sister (TP) to be unpronounced at PF. The result is that the TP in (b) is
not pronounced.
The above implementation suggests one source of cross-linguistic variation in the
realization of sluicing. The feature E, which licenses deletion, can be generated on
different heads: T in English (and other languages discussed by Merchant ())
and F , the head above T , in Malagasy. In addition, the feature can be strong in
some languages, forcing movement for checking, and weak in others, being checked
by Agree. This creates the appearance that either the complementizer itself, or the
head of the complement of the complementizer, licenses sluicing. Another language
that seems to illustrate this second state of affairs is Hungarian (Merchant : ), in
which the wh-phrase remnant also follows the embedded complementizer in sluicing:

() a gyerekek találkoztak valakivel de nem emlékszem, (hogy)


the chidren met someone.with but not I.remember that
kivel.
who
‘The children met someone, but I don’t remember who.’

Finally, we note here that Malagasy SLCs, as in Hungarian, do not at first glance
conform to Merchant’s (: ) generalization:

() Sluicing-COMP generalization


In sluicing, no non-operator material may appear in COMP.

As can be seen in the tree in (b), the C head is filled with hoe. Merchant (:
), however, takes the generalization in () to be a prosodic constraint, relat-
ing to complementizers being adjacent to the sluicing site. In Malagasy, hoe is not
adjacent to the elided TP; the wh-phrase intervenes. Therefore it may not be a
true counterexample once the generalization is more precisely formulated. See van
Craenenbroeck (this volume) for related discussion of sluicing with multiple CP
projections.
We now turn to evidence in favor of predicate-fronting.

. Evidence for the predicate-fronting analysis


Our evidence in favor of predicate-fronting plus TP-deletion as the source of Mala-
gasy SLCs consists in showing that the wh-phrase remnant in sluicing is actually a
predicate, as is expected under the proposed derivation.
First, the same elements that can flank predicates in matrix clauses (see Sec-
tion ..) also co-occur with wh-phrase remnants in SLCs. Pre-predicate ele-
ments such as the modal tokony ‘should’ and the emphatic element tena ‘indeed’
180 Ileana Paul & Eric Potsdam

can precede a wh-phrase in an SLC, (), and post-predicate elements such as


the floating quantifier daholo ‘all’ and the VP adverb foana ‘always’ can follow the
wh-remnant, ().
() a. misy olona tokony hamangy an-dRasoa fa tsy fantatro hoe
exist person should visit acc-Rasoa but neg know.sg comp
[pred tokony iza].
should who
‘Someone should visit Rasoa but I don’t know who should.’
b. nisy olona nanapaka bozaka fa tsy tadidiko hoe tena
exist person cut grass but neg remember.sg comp indeed
iza.
who
‘Someone cut the grass but I don’t remember who indeed did.’

() a. nahandro zavatra maro Rasoa fa tsy fantatro hoe inona


cook thing many Rasoa but neg know.sg comp what
daholo.
all
‘Rasoa cooked many things but I don’t know what all.’
b. any an-tsena matetika ny mpivarotra sasany fa tsy fantatro
there acc-market often the merchant some but neg know.sg
hoe iza foana.
comp who always
‘Some merchants are often at the market but I don’t know who always is.’
Second, all and only the wh-phrases that can be predicates can be SLC remnants. We
have already seen that accusative wh-phrases cannot be SLC remnants, (). They also
can not be questioned in a pseudo-cleft, (), because only subjects and some adjuncts
can be questioned with this strategy, as discussed in Section ...
() ∗ nanasa olona Rabe ka nanontany aho hoe an’iza.
invite someone Rabe and ask I comp who.acc
(‘Rabe invited someone and I asked whom.’)
() ∗ an’iza no nanasa Rabe?
who.acc prt invite Rabe
(‘Whom did Rasoa invite?’)
In the same vein, prepositional phrases can be pseudo-clefted and sluiced: 
() tamin’ inona no namonoan-dRasoa ny akoho?
with what prt kill.pass-Rasoa the chicken
‘What did Rasoa kill the chicken with?’
 The verb in () is in what is called the circumstantial voice. It is roughly equivalent to the passive of
an applicative.
Sluicing in Malagasy 181

() namono ny akoho tamin-javatra maranitra Rasoa fa tsy fantatro


kill the chicken with thing sharp Rasoa but neg know.sg
hoe tamin’ inona.
comp with what
‘Rasoa killed the chicken with something sharp but I don’t know with what.’
The set of wh-phrases that can appear in SLCs is therefore identical to the set of wh-
phrases that can be predicates in pseudo-clefts.
In summary, wh-phrase remnants in SLCs are predicates. This observation supports
our claim that SLC examples are derived by predicate-fronting and subsequent TP-
deletion. The derivation is similar to that assumed for English except that the wh-
phrase is fronted by predicate-fronting, not wh-movement.

. Conclusion
In this paper we have provided an analysis of an SLC in the wh-in-situ language
Malagasy. Our analysis of the Malagasy SLC contributes to the typology of ways in
which wh-in-situ languages create a sluicing-like surface structure. In Malagasy, we
have argued, an SLC involves wh-predicate-fronting followed by TP-deletion:
() nisy olona nihomehy ka nanontany ianao hoe [FP [vP iza]i
exist person laugh and ask you comp who
[TP no nihomehy t i ].
prt laugh
‘Someone laughed and you asked who (the one who laughed was).’
Thus Malagasy has an SLC strategy that is distinct from that of other wh-in-situ
languages, such as Japanese, Chinese, and Javanese, discussed in Section .. The
derivation is similar to, yet also distinct from, that of English sluicing.
A consequence of our analysis is that Malagasy provides further support for a
non-unified analytical approach to SLCs. Our proposal and the above languages
highlight the fact that sluicing is not a single, uniform syntactic construction. A
sentence that looks superficially like English sluicing need not have an English-like
derivation, and wh-in-situ languages cease to be counterexamples to the movement-
plus-deletion view of sluicing if their SLCs can be derived by other means. The strategy
(or strategies) a language uses to arrive at a sluice can be assumed to directly depend
upon the syntactic mechanisms independently available in the language. Thus while
Javanese employs focus-fronting, Malagasy exploits predicate-fronting. Chinese and
Japanese, on the other hand, have no movement at all to feed a sluicing derivation,
and rely on ellipsis alone.
182 Ileana Paul & Eric Potsdam

Another consequence of our analysis is that it provides useful evidence for the
general analysis of predicate-fronting in Malagasy. While there is much recent work
espousing predicate-fronting as the mechanism by which verb-initial word order in
Austronesian languages is derived, there is thus far little empirical evidence for this
fronting operation (see Chung () for important discussion) and it is usually
adopted based on theory-internal considerations. Our analysis suggests that Malagasy
must have predicate-fronting if the derivation of SLC examples is to succeed. The
analysis thus has potentially important consequences for theories of Austronesian
clause structure.

Sluicing in Indo-Aryan:
An investigation of Bangla and Hindi
TA NMOY BHAT TACHARYA AND ANDREW SIMPSON

. Introduction
This chapter establishes a profile of sluicing constructions in two widely-spoken
Indo-Aryan languages of South Asia: Bangla and Hindi. Although traditionally
described as being wh-in-situ languages, both Bangla and Hindi have a distribu-
tion of wh-elements that suggests that they are actually languages with overt wh-
movement (Simpson and Bhattacharya ) and so might be expected to permit
sluicing formed by wh-movement and PF clausal deletion, as hypothesized for lan-
guages such as English (Ross , Merchant , Fox and Lasnik ). The chapter
consequently attempts to determine the degree to which sluicing in Bangla/Hindi
may parallel or differ from the production of sluicing in English-type languages, and
also how it may relate to sluicing patterns in typologically closer Japanese, where
sluicing is often assumed to have a rather different syntactic derivation from that in
English (Nishiyama et al. , Fukaya and Hoji , Hiraiwa and Ishihara ).
The structure of the chapter is as follows. Section . first outlines basic properties
of Bangla/Hindi and the formation of wh-questions in these languages. Section .
then introduces sluicing proper, and examines what kind of analysis would seem to
be supported by the patterning observed in Bangla/Hindi, considering with some
care whether a form of the reduced cleft/copula analyses of sluicing in Japanese
and Chinese might be appropriate for Bangla/Hindi. Arriving at the conclusion
that it is wh-movement rather than copula deletion or cleft reduction that under-
lies Bangla/Hindi sluicing, Section . proceeds to investigate the potential effect of
movement-associated constraints on sluicing in Bangla/Hindi, in particular Supe-
riority and Subjacency/the CED. This leads to the observation of an unexpected
difference in patterning in Bangla and Hindi, and an interesting challenge for the
construction of theories on cross-linguistic variation in island-sensitivity in sluices.
After some consideration of the potential causes of the variation, Section . concludes
184 Tanmoy Bhattacharya & Andrew Simpson

the chapter with an outline of the further comparative research into Indo-Aryan (and
other) languages that seems to be required as a result of the present investigation of
Bangla and Hindi.

. Bangla, Hindi, and the syntax of wh-questions


Bangla and Hindi are both languages that have commonly been described as dom-
inantly head-final and neutrally SOV in word order (Dasgupta , Kachru ),
though with regular post-verbal extraposition of tensed complement clauses caus-
ing stylistically neutral SVO sequences (as in German). Bangla and Hindi have also
regularly been referred to as wh-in-situ languages, owing to the lack of obvious wh-
fronting in examples such as () and () below, where the SOV word order of a
declarative equivalent appears to be maintained (Bayer , Srivastav ):
() jOn kon boi-Ta poRlo. Bangla
John which book-cl read
‘Which book did John read?’
() us-ne kOnsii gaaRii kharidii. Hindi
he-erg which car bought
‘Which car did he buy?’
In Simpson and Bhattacharya (), however, this wh-in-situ characterisation of
Bangla/Hindi is questioned, and with a primary focus on Bangla, it is argued that
overt movement of wh-elements to the C-domain does in fact occur, as in many other
languages. It is suggested that such movement is frequently disguised by the higher
occurrence of subjects in clause-initial topic position in Bangla, but becomes clearly
observable in multi-clause structures where a wh-element from an embedded clause
is interpreted with interrogative scope in a higher clause. In Bangla (), for example,
it is ungrammatical for the wh-phrase ke ‘who’ to remain in situ in the embedded
clause, as in (a), and overt movement into the matrix clause is necessary to license
the wh-phrase. This can be achieved either via raising of the wh-phrase alone, as in
(b), or via pied-piping of the entire clause containing ke, as in (c). In both instances
it is assumed that the matrix-clause subject is in a topic position located higher than
the wh-licensing position.
() a. ∗ jOn bhablo [ke cole gEche].
John thought who leave gone?
b. jOn kei bhablo [t i cole gEche].
John who thought leave gone?
 The following symbols are used to represent certain sounds in Bangla and Hindi: T, D, R represent
retroflex /t/, /d/, and /r/, respectively; S is a palato-alveolar/s/; N is a velar nasal; M is nasalization; E is an
open mid front vowel; O is an open mid back vowel.
Sluicing in Bangla and Hindi 185

c. jOn [ke cole gEche]i bhabche t i ?


John who leave gone think?
‘Who did John think has left?’
Similar patterns occur in Hindi, though with two differences worthy of note. First of
all, when wh-phrases undergo overt movement from an embedded clause, they are
frequently positioned before the matrix-clause subject, as in (b) below. Secondly, the
clausal pied-piping option seen in Bangla (c) is not available as an alternative fronting
strategy in Hindi. 
() a. ∗ Raam-ne socaa [ki kOn aayaa hE].
Ram-erg thought that who come has
b. kOni Raam-ne socaa [ki t i aayaa hE].
who Ram-erg thought that come has
‘Who did Ram think has come?’ (adapted from Mahajan ())
In multiple-wh questions, the occurrence of multiple wh-movement is attested.
For example, in () and (), where both wh-phrases originate in the embedded
clause but can only be interpreted with matrix-clause interrogative scope (as the
embedding verb does not permit a question complement), overt movement of both
wh elements to the matrix occurs and results in a fully grammatical output. As the
“c” examples importantly show, raising of just one of the pair of wh-phrases is not
sufficient to license the multiple-wh questions in () and (), and both wh-phrases
are required to undergo movement to the matrix. Bangla and Hindi are therefore
multiple-wh-fronting languages, similar in their wh-syntax to members of the Slavic
group described in Bošković ().
() a. ∗ tumi bhabcho [je Se kOkhon ki kineche]? Bangla
you think that s/he when what bought
(‘When do you think she bought what?’)
b. tumi kOkhoni kik bhabcho [je Se t i t k kineche]?
you when what think that s/he bought
‘When do you think she bought what?’
c. ∗ tumi kOkhoni bhabcho [je Se t i ki kineche]?
you when think that s/he what bought
(‘When do you think she bought what?’) 
 Hindi does, however, allow the construction of wh-questions with wh-expletives as an alternative to
wh-fronting, as in the example below. For analyses of wh-expletive questions in Hindi, see Mahajan (),
Simpson (), Lahiri (), and Manetta (), among others.
Raam-ne kyaa socaa [ki kOn aayaa hE]?
Ram-erg Q thought that who come has
‘Who did Ram think has come?’
 Example (c) is acceptable only if the first wh-phrase (kOkhon ‘when’) is interpreted as originating in
the matrix clause and refers to the time of “thinking” rather than “buying”.
186 Tanmoy Bhattacharya & Andrew Simpson

() a. ∗ Raam-ne socaa [ki kOn kis-ko maaregaa]? Hindi


Ram-erg thought that who who-dat will.hit
(‘Who did Ram think will hit who?’)
b. kOni kis-kok Raam-ne socaa [ki t i t k maaregaa]?
who who-dat Ram-erg thought that will.hit
‘Who did Ram think will hit who?’
c. ∗ kOni Raam-ne socaa [ki t i kis-ko maaregaa]?
who Ram-erg thought that who-dat will.hit
(‘Who did Ram think will hit who?’) (from Mahajan ())

Given the occurrence of wh-movement in Bangla/Hindi as illustrated above, one


might expect that Bangla and Hindi would both permit sluicing constructions, as
sluicing is often suggested to be parasitic on wh-movement, and significant numbers
of languages with overt wh-movement have also been found to allow for sluicing
(Merchant ). Such an initial expectation is indeed borne out, and, as will be
illustrated in some detail in Section ., both Bangla and Hindi show fully productive
patterns of sluicing similar in many ways to those observed in other languages. This
therefore raises the possibility of investigating sluicing as a general cross-linguistic
phenomenon from the particular perspective of Indo-Aryan languages and using
these to achieve further insight into the ways that sluicing constructions may possibly
vary within different language groups. This is the goal of Sections .–., where we
first address the issue of whether sluicing in Bangla/Hindi is most plausibly attributed
to instances of wh-movement and remnant deletion, or perhaps to some other kind of
process of ellipsis, and then examine how various constraints specifically associated
with movement may interact with sluicing in the two languages.

. Sluicing in Bangla and Hindi: Initial properties


.. An introduction to sluicing patterns in Bangla/Hindi and general
hypotheses of sluicing
Sentences () and () below present two initial examples of sluicing in Bangla and
Hindi respectively, confirming that sluicing is indeed a property of these languages. 

() Mini-r dokan-theke keu ek-Ta boi curi-koreche, kintu ami


Mini-gen shop-from someone one-cl book steal done.has, but I
jani na ke.
know not who
‘Someone has stolen a book from Mini’s shop, but I don’t know who.’

 The occurrence of sluicing in Hindi is also noted and illustrated in Merchant () in the presentation
of sluicing patterns from a wide range of languages.
Sluicing in Bangla and Hindi 187

() Raam-ne us-dukaan-se kuch kharidaa, par mujhe nahii maluum


Ram-erg that-shop-from something bought but to.me not known
ki kyaa.
C what
‘Ram bought something from that shop, but I don’t know what.’

A widespread assumption about the syntax of sluices in general is that the wh-
phrase appearing in sentence-final position and corresponding to an indefinite
expression in an earlier part of the sentence is the remnant of a deletion process
applying to (most of) a selected embedded clause containing the wh-phrase (see
Merchant () and references therein). Such a necessarily multi-clausal analysis of
sluicing is in contrast with the possible view that the wh-phrase might be a direct
argument of the (most deeply embedded) overt verb present in sluicing sentences,
e.g. guess in Guess who. As in many other languages, there is simple evidence in Bangla
and Hindi indicating that wh-phrases in sluices are indeed components of some lower,
phonetically unrealized clause rather than being arguments of the verb that precedes
them in the overt, surface output. For example, Bangla ke in () above occurs in a
nominative-case shape rather than the accusative-case form kake, which might be
expected if ‘who’ were to be the complement of the preceding verb ‘know’, and in
Hindi (), there is the clear presence of a complementizer element ki, signalling that
kyaa ‘what’ is part of a semi-deleted lower clause.
If it is reasonably straightforward to assume that the wh-phrases present in sluices
in Bangla/Hindi are the remnants of a process of clausal ellipsis, the next impor-
tant question, as with sluicing in other languages, is what can be concluded about
the identity and underlying shape of the elided clause. Specifically, does this have
a structure that can be assumed to be parallel to that of the overt clause con-
taining the antecedent of the wh-phrase, or are there reasons to believe that the
elided/phonetically unpronounced clause containing the wh-phrase is not isomorphic
to the “antecedent clause”? Quite generally, from a cross-linguistic perspective, there
are two broad types of analysis of the elision of clauses in sluicing that have been
cogently argued for in recent years and should be considered in relation to sluicing
in Indo-Aryan. Each of these approaches can be shown to be supported by a range
of observable patterns in sluicing constructions, and it is possible and also likely
that both types of analysis are in fact correct, though crucially for different sets of
languages.
The first general mode of analysis of sluicing argued for in a considerable number
of works (among them Ross (), Merchant (), Fox and Lasnik (), and
many of the chapters in this volume) is the hypothesis that sluicing regularly results
from the partial deletion of an embedded clause that is largely isomorphic to the
clause containing the indefinite antecedent of the wh-phrase. In such an approach it is
commonly claimed that wh-movement first promotes the wh-phrase from within the
188 Tanmoy Bhattacharya & Andrew Simpson

clause to a clause-peripheral position, and that this is then followed by deletion of the
clausal residue, i.e. deletion of the largest constituent present in the clause that does
not contain the wh-phrase. In the case of English, the suggestion is that movement of
the wh-phrase to SpecCP is followed by ellipsis of the TP complement to C as in ():
() Jane just married someone, but I don’t know [CP who [TP she just married who]].
The assumption that sluicing is derived via wh-movement from a clause that is iso-
morphic to the overt antecedent clause is well supported by a number of arguments,
as frequently noted in the literature. For example, it allows for a simple explanation
of the observation that non-default case-marking present on the sluiced wh-phrase
regularly corresponds to the case that would be assigned to a wh-phrase occupying a
structural position paralleling that of the indefinite antecedent in the overt antecedent
clause. It also captures correlations between the occurrence and properties of sluicing
in a language and the availability and form of wh-movement that a language may have.
For example, English does permit (and in fact requires) wh-movement, but limits this
to movement of a single wh-phrase per +Q COMP, and sluicing is found to have a
similar restriction, regularly allowing just a single wh-phrase to occur in common,
well-formed sluices (e.g. Someone just stole something from our office, and I know who
(∗ what).).  Romanian, by way of contrast, allows multiple overt wh-movement and
is found to permit multiple-wh elements to occur in sluicing constructions (Hoyt
and Teodorescu this volume). Such correlations, and others noted in Ross ()
and Merchant (), can be naturally accounted for if it is assumed that sluiced wh-
phrases undergo regular wh-movement in a clause structurally parallel to the clause
containing their antecedent(s), and are otherwise difficult to capture in a principled
and convincing way.
The second broad approach to sluicing, argued to be appropriate for certain lan-
guages with patterns rather different from those in English, Spanish, German, and
other languages with clear occurrences of overt wh-movement suggests that sluic-
ing may be produced by the reduction of a clause built around a copula, in two
distinguishable ways. The first type of copula-centered analysis of sluices posits the
existence of a clause containing a copula, a pronoun, and a wh-phrase. The former
two elements may frequently be phonetically empty/unpronounced, either in virtue
of a language having null copulas and null pronouns in general and employing these
in sluices, or owing to TP-ellipsis deleting the phonetic matrix of the pronoun and
copula following wh-movement to SpecCP. The latter analysis was in fact originally
suggested for English in Erteschik-Shir (), and was suggested to create underlying
 There have been certain suggestions that some multiple-wh sluicing may be possible in English, but
there is much disagreement on the acceptability of the data. Even in instances where it is argued that
multiple sluicing may be acceptable, it is noted that there are very special restrictions that do not apply
in languages that regularly allow for multiple-wh sluicing (for example, the second wh-phrase must be a
PP and cannot be a regular DP argument, as noted in Lasnik ()). This suggests that structures that
appear to be multiple sluicing in English are actually formed by quite a different process from those in
other languages.
Sluicing in Bangla and Hindi 189

forms such as (), in part justified by the possibility of overt non-elided sequences
such as ().
() Someone from Bristol is coming to dinner. Guess [CP who [TP it is who]].
() Someone from Bristol is coming to dinner. Guess [CP who [TP it is who]].
In a similar vein, though without the assumption of wh-movement, analyses of sluic-
ing in Japanese have been proposed in Nishiyama et al. () and Fukaya and Hoji
(), in which the wh-phrase occurs in a clause with a null pronominal subject and
a frequently null copula as the main verbal element:
() a. Mary-ga nanika-o katta.
Mary-nom something-acc bought
‘Mary bought something.’
b. Boku-wa [_ nani (-o) (dearu) ka] wakaranai.
I-top pro what -acc be Q do.not.know
‘I don’t know what (it is).’ (from Takahashi ())
Again, the existence of fully overt forms such as () is used to support such an
analysis, as in Erteschik-Shir (), together with the observation in Nishiyama et al.
() that sometimes a copula must in fact be present in a sluiced clause.
() Boku-wa [sore-ga nani-o dearu ka] wakaranai.
I-top that-nom what-acc be Q do.not.know
‘I don’t know what it/that is.’
Chinese has also been argued to be best analyzed as having “sluices” arising via the use
of null pronominal subjects and a sometimes necessarily overt copula (Adams ,
Wei , Adams and Tomioka this volume), and this would indeed seem to be the
most plausible analysis of Chinese, given the patterns found in the language.
The second copula-centered analysis of sluicing suggests that sluices may critically
result from the reduction of a cleft structure (see, for example, Fukaya et al. () and
Hiraiwa and Ishihara ()).  In such an approach, sluicing arises from the deletion
of a CP-clause in which movement has occurred to create a cleft structure, as well
as frequent deletion/non-pronunciation of both the copula which embeds this CP-
clause, and the pronominal subject of the copula.
Both wh-movement + IP/TP-deletion and copula-centered analyses of sluicing
are motivated by the need to explain variation in the patterning of sluices in differ-
ent languages (such as, for example, the frequent intrusion of copulas in sluices in
certain languages but not others). Because the two modes of analysis consequently
hypothesize different underlying structures (and processes) in “sluicing”, they give

 Note that Fukaya and Hoji () examine both types of copula-related sluicing analysis mentioned
here, and suggest that both are in fact made use of in Japanese.
190 Tanmoy Bhattacharya & Andrew Simpson

rise to different expectations about the surface effects of such structures, for example
with regard to clause-internal positioning of the wh-phrase, case on the wh-phrase,
single/multiple-wh sluicing restrictions, and various other phenomena. In a copula-
centered analysis of sluicing, for instance, one might naturally expect that the case
on the wh-phrase “remnant” would be licensed by the copula (or its associated func-
tional structure) and hence be invariant, whereas in a wh-movement analysis of sluic-
ing, there is an expectation that wh-phrase remnants will occur in variant/different
case-forms, reflecting the case assigned to the wh-phrase in the deleted underlying
clause prior to wh-movement and TP-deletion. Similarly, if copulas (including in
clefts) are restricted to selecting a single complement XP, one might not expect a
copula/reduced-cleft analysis to be appropriate for a language which permits multiple-
wh sluicing.
Despite the availability of various broad diagnostics bearing on the structure of
sluicing constructions, much care is still needed in the application of default expecta-
tions about case and other patterns in cross-linguistic investigations of sluicing, and
there may not be any fully fixed set of properties that are associated with copula-type
sluices as opposed to sluices involving wh-movement and deletion of an isomorphic
clausal structure. For example, if one considers Japanese as a language where a good
case can be made for a reduced-cleft copula analysis of sluices, it is found that (a) case
is not invariant on the wh-phrases in sluices, and wh-phrases appear marked in
various ways with case and postpositions, and (b) the occurrence of multiple-wh-
phrases is actually possible. Although perhaps unexpected if a unitary cross-linguistic
conception of the behavior of elements in clefts and copula sentences is entertained
(as copulas and clefts do not permit multiple complements with potentially different
case forms in many languages), from the standpoint of Japanese, properties (a) and
(b) are fully consistent with the properties of cleft sentences in the language. Japanese
clefts do allow for clefted elements to occur in case-forms or with postpositions
that corresponds to gap positions in the embedded CP-constituent present in clefts,
and it is also possible for there to be clefting of more than just a single element
in cleft sentences in Japanese. Consequently, although variant case-marking may be
advanced as an argument for a wh-movement + TP-deletion analysis of sluices, it
may not be a fully reliable indicator of such a derivation, as similar patterns may
sometimes also arise in cleft constructions in a language. Similar conclusions may
apply to other “default” expectations used to argue for either a wh-movement + TP-
deletion analysis or alternatively a non-isomorphic copula analysis of sluices. What
is therefore clearly required in each fresh examination of sluicing in a language is
a careful comparison of sluices with other patterns in the same language, in par-
ticular those relating to wh-movement, case, copulas, and cleft constructions. With
this caution in mind, we now turn our attention back to Bangla and Hindi, and
consider what mode of deletion would seem to be operative in sluicing within these
languages.
Sluicing in Bangla and Hindi 191

.. Probing the properties of sluices in Bangla/Hindi


The question to be approached here with regard to Bangla and Hindi is whether
clausal reductions in sluicing sentences result from wh-movement and the ellipsis of
a structure isomorphic to the clause containing the indefinite antecedent of the wh-
phrase (the correlate), or alternatively whether sluiced clauses are underlyingly some
kind of copula structure. Considering sluicing in Bangla first, certain data might seem
to be open to either analysis, e.g. ():
() Mini kichu kinechilo, kintu ami jani na ki.
Mini something bought.had but I know not what
‘Mini had bought something, but I don’t know what.’
The Bangla wh-word ki ‘what’ in () is invariant in its shape in both object and subject
functions, and so its occurrence there could be either as an object form, produced by
the deletion of an underlying structure mirroring that of the clause containing kichu
‘something’, or perhaps as a subject form associated with a null copula. In Bangla
the copula is regularly not expressed in any overt form in the present tense, and
so () might potentially allow for an analysis as a null-copula structure. The use
of wh-expressions other than ki is, however, more revealing, and the equivalent of
‘who’ occurs in distinct nominative, accusative, genitive, and genitive/dative-plural
forms, respectively ke, kake, kar, and kader. Such forms, furthermore, regularly occur
in sluices, as illustrated in (–):
() ami Sunlam je keu jonaki-ke Thokieche, kintu ami jani na
I heard that someone Jonaki-acc cheated.has but I know not
ke.
who.nom
‘I heard that someone had cheated Jonaki, but I don’t know who.’
() ami Sunlam je Raam kau-ke biye-koreche, kintu ami jani
I heard that Ram someone-acc marriage-do.has but I know
na kake.
not whom.acc
‘I heard that Ram has married someone, but I don’t know who.’
() karo rag hoeche, kintu ami jani na kar.
someone.gen anger happen.has but I know not who.gen
‘Someone has become angry, but I don’t know who.’
In all the above examples, the patterns appear to show “case-matching/P-matching”
and marking of the wh-phrase with a case/postposition form that also occurs on
the indefinite antecedent of the wh-phrase. Such a correspondence relation, and the
occurrence of non-default case on the wh-phrase might be assumed to be caused
by deletion of an underlying structure paralleling that of the clause containing the
192 Tanmoy Bhattacharya & Andrew Simpson

antecedent of the wh-phrase, and might be less naturally attributed to an underlying


copular structure. However, as noted in the preceding section, there is a clear need to
be careful here, as in certain languages case- and P-marked DPs are sometimes found
to occur with copulas. In Bangla, whether a null (or deleted) copula might be able
to co-occur with and license DPs in different case-forms and marked with different
postpositions is difficult to establish with present-tense time reference, as Bangla has
no common present-tense copula form that is overt. However, for past-time reference
there is an overt form that can be used to test copula compatibility with DPs marked
by different cases and postpositions—the element chilo. This element can be used to
produce sequences such as ():
() keu jonaki-ke Thokiechilo, kintu ami jani na je Se
someone Jonaki-acc had.cheated but I know not C s/he
ke chilo.
who-nom was
‘I heard that someone had cheated Jonaki, but I don’t know who it was.’
Significantly, although it is possible for a wh-phrase in nominative case to occur in
a past-tense copula clause, it is not possible for chilo to license or co-occur with a
wh-phrase in accusative or genitive case, as seen in () and () below. Instead, it is
necessary for the wh-phrase to appear in nominative case in such examples. 
() ProkaS kau-ke Thokiechilo, kintu ami jani na je Se
Prakash someone-acc had.cheated but I know not C s/he
ke /∗ kake chilo.
who.nom /whom.acc was
‘Prakash cheated someone, but I don’t know whom.’
() karor OSukh korechilo, kintu ami jani na je Se ke
someone.gen sickness had.done but I know not C s/he who.nom
/∗ kar chilo.
/who.gen was
‘Someone was sick, but I don’t know who.’
If the copula chilo is not present, but overt pronoun-forms such as Se ‘s/he’ or Se-Ta
‘it’ occur in addition to a wh-phrase (as they do in () and ()), this will correspond
to a simple (present-tense) null copula structure paralleling that in non-sluiced wh-
questions such as ():

 It should be added that occurrence of the copula chilo in () may be felt to be a little unnecessary, and
a present-tense null copula may even be preferred. However, () is at most a little over-specified and in
strong contrast to () and (), which are clearly ungrammatical and contain case forms that cannot be
produced by the syntactic structure/components of the sentence.
Sluicing in Bangla and Hindi 193

() Se ke?
s/he.nom who.nom
‘Who is s/he?’
When the projection of null copula structures is forced by the presence of pronouns
such as Se and Se-Ta, it is found that a nominative case ke ‘who’ is required, while
accusative kake and genitive kar are again not licensed in the presence of a pronoun:
() karo rag hoeche, kintu ami jani na Se ke
someone-gen anger has.happened but I know not s/he who-nom
/∗ kar.
/who-gen
‘Someone has become angry, but I don’t know who.’
() Raam kau-ke biye-koreche, kintu ami jani na Se
Ram someone-acc marriage-do.has but I know not s/he
ke /∗ kake.
who.nom /whom.acc
‘Ram has married someone, but I don’t know who.’
This indicates that copular structures do not license wh-forms such as kake and kar. As
these elements freely occur in sluice constructions, however, it can be concluded that
they are not licensed by a deleted/null copula structure, but by the elided predicate of a
clause isomorphic to the clause containing the indefinite antecedent of the wh-phrase.
Such a conclusion is further strengthened by the observation that the case-matching
noted to occur in (–) is actually a strict requirement in sluices, where the wh-
remnant is the only element present in the elided clause, and there is no other overt
nominative pronoun present to induce an underlying null copula structure, e.g:
() Raj kau-ke Thokieche, kintu ami jani na kake /∗ ke.
Raj someone-acc cheated.has but I know not who.acc /who.nom
‘Raj has cheated someone, but I don’t know whom.’
() karo gaRi baire park kOra ache, kintu amra jani na
someone.gen car outside park done is but we know not
kar /∗ ke.
who.gen /who.nom
‘Someone’s car is parked outside, but we don’t know whose.’
() Mini karo Songe kOtha bolechilo, kintu ami jani na
Mini someone.gen with word spoke but I know not
[kar Songe] /∗ ke.
who.gen with /who.nom
‘Mini spoke with someone, but I don’t know who.’
The set of patterns presented here indicate fairly clearly that two kinds of output with
closely related meanings but different syntactic structures are possible in Bangla. First
194 Tanmoy Bhattacharya & Andrew Simpson

of all, it is possible to identify structures containing an overt pronoun, an overt/covert


copula form, and a wh-phrase in nominative case (but not in accusative or genitive
case), as in (), (), and (). Secondly, elided structures are possible in which
only a wh-phrase corresponding to a preceding indefinite antecedent is present, and
the wh-phrase necessarily occurs in the same case/postposition-marked form as its
antecedent. This is seen in (–) as well as in (–).  It is this latter set of forms
that constitute sluices proper, derived via the deletion of some large non-copula com-
ponent of the clause, which appears to match the structure of the clause containing
the antecedent of the wh-phrase.
Relative to such deletion structures two further important points can be made.
First, it is found that more than one wh-phrase can be present in sluicing constructions
in Bangla, as shown in (a) and (a) (with case-matching clearly seen on kake
‘whom’ in (a)). As illustrated in (b) and (b), such multiple-wh forms cannot
be the output of any copula construction, null or overt, and any attempt to have an
overt pronoun form (or overt past-tense copula) in addition to the wh-phrases as
indication of a copula structure results in ungrammaticality: 

 Note that in the absence of overt clues indicating a non-elided null/overt copula structure (such
as the presence of pronouns), there seems to be a considerably strong pressure in Bangla for matching
of the covert structure of the embedded clause containing the wh-phrase in (–) with that of the
“antecedent clause”, resulting in the use of a case-form of the wh-phrase that corresponds to the case-
form of the indefinite antecedent DP (i.e. a considerable pressure for isomorphic sluicing). In theory, one
might expect that examples such as (–) would actually be acceptable with a nominative case-form
ke, produced by the combination of a null copula and a null pronoun/pro, in place of an overt Se ‘s/he’.
However, such a hypothetically available form does not seem to be tolerated, and where there is no overt
pronoun present forcing the assumption of an underlying null copula structure, speakers automatically
require case/P-matching and the projection of an isomorphic sluiced structure.
 The multiple-wh sluices in () and () are instances in which the indefinite antecedents of the
wh-phrases both originate in the same clause. In Japanese multiple-wh sluicing it has been suggested
(Takahashi ) that there may be a clausemate condition on multiple-wh sluicing, requiring the indefinite
antecedents of wh-phrases to be base-generated in the same clause. The effects of a similar restriction in
Bangla are quite interesting to observe. In Bangla, a bi-clausal structure such as in (i) below, with an indef-
inite DP/PP in the higher clause and a second indefinite DP/PP in the lower, embedded clause, can only
serve as the antecedent for a single-wh sluice structure, which indicates that a clausemate restriction on
multiple-wh sluicing does indeed hold in Bangla. However, such a restriction can interestingly be overcome
by raising the indefinite in the lower clause to a position in the higher clause adjacent to the first indefinite,
focusing the two indefinites in the same clausal position, as illustrated in (ii).

(i) tumi kawke bolle [je bacca-Ta kichu curi-koreche], kintu ami jani na
you someone.dat told C child-cl something steal-do.has but I know not
kake (∗ ki).
whom what
‘You told someone that the child has stolen something, but I don’t know whom.’
(ii) tumi kawke kichu bolle [je bacca-Ta kichu curi-koreche], kintu ami jani
you someone.dat something told C child-cl steal-do.has but I know
na kake ki.
not whom what
‘You told someone that the child has stolen something, but I don’t know to-whom what.’
Sluicing in Bangla and Hindi 195

() a. Raam kawke kichu diyeche, kintu ami jani na


Ram someone.dat something give.has but I know not
ka-ke ki.
whom-dat what
‘Ram has given someone something, but I don’t know what and to whom.’
b. ∗ Raam kawke kichu diyeche, kintu ami jani na
Ram someone.dat something give.has but I know not
Se-gulo ka-ke ki (chilo).
they whom-dat what was
(‘Ram has given someone something, but I don’t know what and to whom.’)
() a. keu kichu bheNgeche, kintu ami jani na ke ki.
someone something break.has but I know not who what

b. keu kichu bheNgeche, kintu ami jani na Se-gulo ke
someone something break.has but I know not they who
ki (chilo).
what were
(‘Someone has broken something, but I don’t know who and what.’)
Secondly, Bangla can be noted to be a language that does not have cleft constructions.
Consequently, the case-/P-matching that is found in sluices cannot be attributed
to any reduced-cleft form and the occurrence of non-default case/postpositions on
wh-phrases due to hypothetical raising of the latter from positions within the CP
component of a cleft. This absence of cleft constructions in the language, together
with the observation of necessary case-/P-matching and the potential occurrence of
multiple (case-marked) wh-phrases in sluices, suggests that a copula-centered analysis
of sluices is not appropriate for Bangla and can justifiably be discounted.
Similar facts also obtain in Hindi, and are in certain ways clearer, as Hindi is not
a null-copula language. Considering similar patterns to those observed in Bangla, it
can first of all be noted that non-elided, non-sluice forms with copulas as in () are
possible, with a nominative case wh-phrase kaun ‘who’.
() Raam-ne kisi-se baat-kii-thii, par mujhe nahii maluum ki
Ram-erg someone-with speech-do-pst but me.obl not know C
voh kaun hai.
s/he who.nom is
‘Ram spoke with someone, but I don’t know who s/he is.’
As further illustrated in (–) below, in sluicing constructions where ellipsis of part
of the clause appears to have occurred, a variety of other non-nominative case-forms
and postpositions can also occur on wh-phrases.
Note that a wh-phrase in a non-multiple-wh sluice is able to refer back to an indefinite antecedent in
an embedded clause, so it is not the depth of embedding of the second indefinite in (i) that disallows its
participation in sluicing; rather, it is the separation of the two indefinites in different clauses.
196 Tanmoy Bhattacharya & Andrew Simpson

() Raam-ne kisi-ko kitaab dii-thii, par mujhe nahii maluum ki


Ram-erg someone-dat book give-pst but me.obl not know C
kis-ko.
who-dat
‘Ram gave a book to someone, but I don’t know who.’
() Mini-ne kisi-se Saadi-kii-thii, par mujhe nahii maluum ki
Mini-erg someone-with marriage-do-pst but me.obl not know C
kis-se.
who-with
‘Mini married someone, but I don’t know who.’
() kisi-ne meri ghari cori-kii-thii, par mujhe nahii maluum ki
someone-erg my watch stealing-do-pst but me.obl not know C
kis-ne.
who-erg
‘Someone stole my watch, but I don’t know who.’
Considering now whether wh-phrases marked with non-nominative case or with
postpositions can occur in (overt) copula constructions parallel to (), examples such
as () and () show that this is not possible:
() ∗ Raam-ne kisi-ko kitaab dii-thii, par mujhe nahii maluum ki
Ram-erg someone-dat book give-pst but me.obl not know C
voh/yeh kis-ko hai.
s/he/it who-dat is
(‘Ram gave a book to someone, but I don’t know who.’)
() ∗ Mini-ne kisi-se Saadi-kii-thii, par mujhe nahii maluum ki
Mini-erg someone-with marriage-do-pst but me.obl not know C
voh/yeh kis-se hai.
s/he/it who-with is
(‘Mini married someone, but I don’t know who.’)
Because non-nominative and postposition-marked wh-phrases cannot be licensed by
copulas but do regularly occur in sluices, this indicates fairly clearly that copulas are
not involved in the production of sluiced clauses. As Hindi, furthermore, has no cleft
constructions, like Bangla, sluices in Hindi can also not be analyzed as the reduc-
tion of some underlying cleft structure. This therefore leads to the conclusion that
sluicing in Hindi must arguably result from the reduction of a structural copy of the
clause containing the antecedent of the wh-phrase, and that the non-default cases and
postpositions appearing on sluiced wh-phrases are licensed by their base-generation
within such a clause. Concerning the latter case-/postposition-marking, it can be
Sluicing in Bangla and Hindi 197

noted that there is a strict case-/postposition-matching requirement in Hindi sluices,


as in Bangla, and the case-/postposition-marking of a wh-phrase must correspond to
the case/postposition present on its antecedent DP, as illustrated in () and ():

() Raam-ne kisi-ko kitaab dii-thii, par mujhe nahii maluum ki


Ram-erg someone-dat book give-pst but me.obl not know C
kis-ko ∗
/ kaun.
who-dat /who.nom
‘Ram gave a book to someone, but I don’t know who.’
() Mini-ne kisi-se Saadi-kii-thii, par mujhe nahii maluum ki
Mini-erg someone-with marriage-do-pst but me.obl not know C
kis-se/∗ kaun.
who-with/who.nom
‘Mini married someone, but I don’t know who.’

In both Bangla and Hindi there is consequently good evidence indicating that sluices
are not derived from underlying copula structures or clefts, and are instead instances
of deletion of a clausal structure that appears to be isomorphic to the structure of
the antecedent clause containing the correlate to the wh-phrase. As Bangla and Hindi
can furthermore both be argued to be languages in which overt movement of wh-
phrases regularly occurs, the ellipsis of the clause containing the wh-phrase in sluices
can be suggested to be directly facilitated by wh-movement and the evacuation of wh-
phrases from their base-generated positions in the underlying copy of the antecedent
clause. As additional support for such an assumption, one can usefully consider the
derivation of non-sluiced equivalents to sluice constructions, in which no deletion
of material matching that of a preceding clause has occurred. As seen in example
(a) from Bangla below, the wh-phrase ki ‘what’ must necessarily undergo movement
from its base-generated position in the lowest clause to a position in the higher clause
selected by jani ‘know’, where it can be licensed and also satisfy the interrogative
selectional requirements of jani. Example (b) shows that the non-occurrence of
movement of the wh-phrase in the same structure automatically results in ungram-
maticality. Overt wh-movement similar to that in (a) can therefore naturally be
assumed to occur and to be necessary where the PF-deletion of material repeated in
a preceding clause occurs, i.e. in sluicing: 

 If movement of a wh-phrase to the left periphery is what critically permits PF-deletion of the TP part
of a clause in sluicing, it might be expected that the movement of other non-wh elements to similar left-
peripheral positions would also allow for a non-wh equivalent to sluicing—“stripping” (Fukaya and Hoji
). In particular, focus-movement of DPs and PPs might be anticipated to provide a possible input to
TP-deletion and the creation of non-wh “sluices”, as discussed in Hoyt and Teodorescu (this volume). This
does indeed seem to be possible to a certain extent in Bangla and also in English, though there are signs
198 Tanmoy Bhattacharya & Andrew Simpson

() a. Mina bhabche je Raam kichu curi-koreche, kintu amra jani


Mina thinking.is C Ram something steal-do.has but we know
na Mina kii bhabche [je Raam t i curi-koreche].
not Mina what thinking.is C Ram steal-do.has
‘Mina is thinking that Ram has stolen something, but we don’t know what
Mina is thinking that Ram has stolen.’
b. ∗ Mina bhabche je Raam kichu curi-koreche, kintu amra
Mina thinking.is C Ram something steal-do.has but we
jani na Mina bhabche [je Raam ki curi-koreche].
know not Mina thinking.is C Ram what steal-do.has
Ultimately then, the analysis of sluicing in Bangla and Hindi that would seem to be
most clearly supported by patterns observable in the two languages is that sluicing has
essentially the same kind of derivation as regularly suggested for languages such as
English in Ross (), Merchant (), and Fox and Lasnik (), being driven by
a combination of wh-movement and the deletion of a clausal structure paralleling the
antecedent clause. Before we go on to consider how the movement assumed to occur
in Bangla/Hindi sluices potentially interacts with various restrictions on movement in
Section ., the final part of the present section will briefly consider certain differences
between sluicing in Bangla/Hindi and sluicing in English and other similar languages,
and how these differences might be accounted for.

that it is more restricted and less automatic in its application than regular wh sluicing. Examples (i) and (ii)
below show relevant patterns in Bangla with a “sluiced” DP and PP following the verb bOla ‘to say’:

(i) keu Mini-r kOmpiuTar-Ta curi-koreche, ebong ma bollen bola.


someone Mini-gen computer-cl steal-do.has and mother said Prakash
‘Someone has stolen Mini’s computer and mother said (it was) Prakash.’
(ii) eimatro masTar-mOSai karo Songe kOtha-bolchilen. ProkaS bollo Mini-r
just.now teacher-Mr. someone.gen with word-speaking.was Prakash said Mini-gen
Songe.
with
‘Just now the teacher was talking with someone. Prakash said with Mini.
With regard to English, Hoyt and Teodorescu indicate that the attempted sluicing of non-wh phrases is
regularly ungrammatical. However, there are in fact many instances where it can legitimately occur. What
seems to be critically necessary is that there is some potential focus on the remnant XP, and that no overt
complementizer occurs. The examples below are modifications of examples given by Hoyt and Teodorescu,
with deletion of the complementizer present in their examples:
(iii) It’s true that many people voted for Blair, but I don’t think EVERYONE.
(iv) Sue just left with someone, but I don’t think with YOUR date.
(v) Carmen wants to buy herself a car, and I suspect PRETTY SOON.
The no-overt-complementizer restriction may perhaps be understood to be a result of the doubly-filled
COMP filter, if focused XPs in English non-wh sluice constructions undergo raising to SpecCP rather than
a focus position following C (which might otherwise be presumed on the basis of rather stylistically marked
literary English examples such as ?John thought that liberty they would not achieve).
Sluicing in Bangla and Hindi 199

.. Distinctive properties of sluicing in Bangla and Hindi


... Complementizer deletion In sluicing constructions in many languages, the
application of clausal ellipsis following wh-movement to SpecCP results in the wh-
phrase being the sole overt remnant of the elided clause, and it is this occurrence of
across-the-board deletion of material in the sluiced clause that has in large part moti-
vated the analysis of deletion of a single clausal constituent in sluicing (the IP/TP).
Considering sluices in Bangla and Hindi, there is certain variation between the two
languages with regards to full deletion of CP-internal material, specifically relating
to the occurrence of overt complementizers in sluicing. In Hindi, there is a strong
pressure for the complementizer ki ‘that’ to be overt in sluices and precede the wh-
phrase, as in (). In Bangla, by way of contrast, overt use of the complementizer je in
a similar position in sluices is felt to be quite unnatural and even ill-formed. Sluices
are therefore commonly formed without je as in ().
() Raam-ne kuch ciiz cori-kii-thii, par mujhe nahii maluum ∗ (ki)
Ram-erg some thing stealing-do-pst but me.obl not know C
kyaa. Hindi
what
‘Ram had stolen something, but I don’t know what.’
() Mini kau-ke biye-koreche, kintu ami jani na (∗ ??je)
Mini someone-acc marriage-do.has but I know not C
kake. Bangla
whom-acc
‘Mini has married someone, but I don’t know who.’
The obvious question is how to interpret this variation in the occurrence of comple-
mentizers in Bangla and Hindi, and whether to assume that it reveals some difference
in the underlying syntax of sluices in the two languages. We would like to speculate
here that it does, and that it may relate to a difference in the height of the landing-site
of movement in Bangla and Hindi sluices. Considering Hindi first, the regular occur-
rence of ki before wh-phrases in Hindi sluices might seem to indicate that wh-phrases
undergo raising to a lower part of the C-domain than that occupied by embedding
complementizers, and to a wh-licensing position similar in height to the landing-
site of wh-movement in Hungarian (also below an embedding complementizer (Kiss
)).  Turning now to Bangla, because the complementizer je cannot co-occur
with movement of a wh-phrase to the C-domain, we suggest this indicates that wh-
phrases may regularly undergo raising to a higher part of the C-domain in sluices, and

 It might also be possible that wh-movement in Hindi takes place to the vP edge, as suggested in
Manetta (), rather than to a lower part of the C-domain. More precise information on the location of
the subject and certain other pre-wh elements in Hindi is necessary before it can be concluded how high
in the clause wh-phrases are actually raised to.
200 Tanmoy Bhattacharya & Andrew Simpson

disallow the occurrence of je either owing to a doubly-filled-COMP-type violation or


because the occurrence of a wh-phrase in the Spec of a C occupied by je would result in
a clash of opposing features (+interrogative and +declarative). Raising of a wh-phrase
to such a height might furthermore be expected to correspond to a form of topic-
like interpretation being associated with the wh-phrase, and such an assumption is
not implausible for sluice constructions, in which the indefinite antecedent of the
wh-phrase first introduces a new entity in the antecedent clause, and the questioned
reference value of this entity subsequently becomes the link-topic of the following
(sluiced) clause. Because of such a topic-like interpretation being arguably open to
the wh-phrase in sluices, this additionally often licenses and results in movement of
the sluiced CP remnant to a higher topic position preceding the subject of the embed-
ding clause, not only in Bangla, as illustrated in (), but also in languages such as
English and German, which otherwise do not readily permit the topicalization of wh
elements:

() gOto SOpata-e Mini kawke biye-koreche, kintu [CP [DP


last week-loc Mini someone-dat marriage-do.has but
kake]]i amra jani na t i .
whom-dat we know not
‘Last week Mini got married to someone, but we don’t know who.’

() John just got engaged to someone, but [CP [DP who]]i we really don’t know t i .

() Johann hat das Buch jemandem gegeben, aber [CP [DP wem]]i
Johann has the book someone.dat given but whom-dat
wissen wir gar nicht t i .
know we at.all not
‘Johann gave the book to someone, but [CP whom] we really don’t know.’

This topic-fronting of the sluiced CP containing the wh-phrase is also possible in


Hindi, and when it does occur, it is noticeable that the complementizer ki does not
and cannot occur:

() Sita-ne kisi-se Saadi-kii-thii, par [CP (∗ ki) kis-se] (yeh)


Sita-erg someone-with marriage-do-pst but C who-with it
mujhe nahii maluum.
me.obl not know
‘Sita had married someone, but I don’t know who.’

This may therefore suggest that when a wh-phrase is interpreted in a topic-like way
in Hindi sluices, it actually undergoes raising to a higher position in the C-domain
as in Bangla, causing deletion of ki, and fronting of the CP unit to topic position in
Sluicing in Bangla and Hindi 201

the higher clause.  ,  Note that the potential occurrence of the resumptive expletive
yeh should indicate that the entire sluiced clause is fronted here and not just the
wh-phrase, as extraction of elements from within CPs is noted to be unacceptable
when the clausal expletive yeh co-occurs with a CP (Mahajan ). Variation in
the occurrence of complementizers in sluicing in Bangla and Hindi in both post-
verbal and clause-initial positioning of sluiced CPs may consequently be useful and
informative, and potentially allow for a finer picture of wh-raising to unfold (see
also van Craenenbroeck (this volume), for interesting discussion of wh-movement
to different positions within the C-domain).
... Subjects in sluiced clauses A second issue which arises in regard to possible
differences in sluicing in Bangla and Hindi is whether sluicing in these languages
involves the deletion of a single clausal unit (e.g. TP), or the deletion of more than
just a single constituent. A now widely-held assumption about sluicing in English and
many other languages (Merchant () and others) is that wh-movement to SpecCP
makes it possible for the full residue of a clause to be elided in a single occurrence of
deletion. This is because wh-movement in English-type languages commonly raises
wh-phrases to a position higher than other elements in the clause. In Bangla, however,
it is not uncommon for wh-phrases to undergo raising to a position that is lower than
the position of the subject, as illustrated in ():
() JOn kei bhabche [t i cole gEche].
John who thinking.is leave go.has?
‘Who is John thinking has left?’
If such a sequence were assumed to be the input to sluicing (within a larger structure),
the question is whether there might need to be two operations of deletion, one eliding
the clausal constituent below the position of the raised wh-phrase, and the second
deleting the subject. Following on from this, supposing more than one constituent
were to undergo necessary ellipsis in sluicing in certain languages, this would then
call into question whether the occurrence of such obligatory multiple deletion might
potentially weaken the common analysis of across-the-board deletion effects in sluices
as resulting from the deletion of a single clausal constituent.

 The suggested sequence of wh-raising to a high topic-like position within the sluiced CP, followed
by pied-piping of the CP to a higher-clause topic position, is similar to operations of focus-movement and
clausal pied-piping in Basque (Urbina ). In Basque it is frequently found that an element may first
undergo focus-movement to the clause-initial SpecCP focus position in an embedded clause, and then
trigger pied-piping of this CP to the SpecCP focus position of a higher clause. One focus-driven movement
therefore results in a second focus-related movement of a larger clausal constituent. See Urbina () for
examples and further discussion of this sequencing of movement.
 The non-occurrence of the complementizer ki (and Bangla je) here can be noted to be in marked
contrast with the observation that fronting of a CP in English requires the necessary occurrence of a
complementizer that, which is otherwise quite optional in its occurrence:
(i) [∗ (That) John is not coming], I simply don’t believe.
202 Tanmoy Bhattacharya & Andrew Simpson

Concerning the first question, what is actually observed in Bangla (and Hindi) is
that there is indeed necessary deletion of all material in the sluiced CP that would be
a repetition of material in the antecedent clause, with the exception of the wh-phrase.
Hence (non-wh) subjects in sluices must undergo ellipsis in the same way that other
non-wh material does. In (), for example, what is assumed to be present under-
lyingly in the sluiced, bi-clausal part of the structure is the sequence of words Mina
ki bhabche je Raam curi-koreche, literally meaning ‘What does Mini think that Ram
stole?’, when unembedded. In this sequence, the wh-element ki ‘what’ has undergone
movement from its base position as the object of curi-koreche ‘stole’ in the lower clause
to a position following the subject, which is assumed to be in a topic position in the
higher clause. This post-subject/topic position is where the wh-phrase would naturally
move to if the sequence were to serve as a non-sluiced direct or indirect/embedded
question. When sluicing occurs, not only is the clausal material following the raised
wh-phrase eliminated, but also the pre-wh-phrase subject/topic is necessarily left
unpronounced/phonetically deleted, as shown in ():
() Mina bhabche je Raam kichu curi-koreche, kintu amra jani
Mina thinking.is C Ram something steal-do.has but we know
na ∗ (Mina) kii bhabche je Raam t i curi-koreche.
not Mina what thinking.is C Ram steal-do.has
‘Mina is thinking that Ram has stolen something, but we don’t know what.’
It may therefore seem necessary to assume two separate deletion operations in such
cases of sluicing, eliminating not just post-wh but also pre-wh material. In consid-
ering whether it is theoretically undesirable to allow for the occurrence of multiple
operations of deletion within a “single” instance of sluicing, it can be noted that such
an issue is actually not confined to languages like Bangla and Hindi (and Turkish
and Romanian, as described in Ince (this volume) and Hoyt and Teodorescu (this
volume), respectively), but it actually also occurs in English in certain circumstances.
For example, when the wh-element in a sluice is a possessor phrase ‘whose’, there
is fairly automatic deletion of the NP that it qualifies, and retention of the NP is
unnatural: 
() Someone’s car is on the lawn, but we don’t know whose ?(car).
As the NP possessee occurs in SpecCP, its deletion will constitute a second operation
of ellipsis distinct from deletion of the TP clausal unit. Were it therefore necessary to
concede the occurrence of potentially multiple, individual occurrences of ellipsis in
sluicing for languages such as Bangla, Romanian, and other languages where raised
wh-phrases are commonly preceded by certain other elements, this ultimately might

 If the NP is retained, then focal pitch on whose has to be significantly increased, and the NP must be
fully destressed: ‘. . . but we don’t know WHOSE car. However, simple deletion of the NP would seem to be
the much more common strategy.
Sluicing in Bangla and Hindi 203

not be so particularly unusual, as it may perhaps also be necessary to assume multiple


constituent deletion (in certain circumstances) in languages where wh-phrases raise
to the highest, leftmost position in a clause, such as English.
It can also be suggested that there may be other ways of looking at the applica-
tion of ellipsis in Bangla and other similar languages, and that it may be possible to
avoid the conclusion that sluicing involves multiple instances of deletion. This can be
done by challenging the assumption that the input to sluicing is in all ways bound
to resemble the form of a non-sluiced wh-question. In Simpson and Bhattacharya
(), the frequent occurrence of a non-wh subject to the left of a raised wh-phrase
in regular wh-questions is attributed to the subject occurring in a high topic position.
Considering sluicing constructions, it might be reasonable to suggest that because
everything (assumed to be) in the underlying structure of a sluiced clause is a rep-
etition of material from the antecedent clause, it is all equally old material (with the
exception of the wh-phrase), and no single non-wh element is distinguished as a topic.
If this were to be so, it could be assumed that there is no necessary placement of
other non-wh material in topic-like positions preceding the raised wh-phrase, and
the wh-phrase would indeed naturally occur as the highest element in the clause in
Bangla, allowing for single-constituent deletion of all other elements in the residue of
the clause, as is assumed to occur in English sluices. A parallel here can be drawn with
English to illustrate this line of thinking. In spoken, colloquial English it is actually not
uncommon to hear wh-questions produced with an initial non-wh topic-like element
such as a frame-setting adverbial, as in ():

() Yesterday, what did you buy in that shop?

Because this kind of structure is possible, and because the antecedent clause to a sluice
may occur with an adverbial in clause-initial position, one might perhaps be led to
expect that an adverbial could precede a raised wh-phrase in the CP to be sluiced.
This being so, one would in theory be faced with the same problem as in Bangla, that
of explaining why elements assumed to occur in a pre-wh position must also undergo
ellipsis along with post-wh material. Example () below indicates that an adverbial
in pre-wh position is clearly unacceptable in sluices in English:

() Yesterday John bought something, but we don’t know ∗ (yesterday) what.

Here, as proposed above, it can be suggested that the potential occurrence of an


element in pre-wh topic position in regular non-sluiced wh-questions does not neces-
sarily mean that the same element can or necessarily should occur in a topic position
within a sluice construction (where the wh-phrase itself may in fact be the event-
participant most construed in a topic-like way), and it may be perfectly legitimate
for such an element to be base-generated and built in some lower position within
the clause. This theoretical possibility then provides a plausible way of preserving the
204 Tanmoy Bhattacharya & Andrew Simpson

view that sluicing is ellipsis of a single, clausal constituent.  ,  If this proves to be


sustainable as a hypothesis, it may in turn suggest that sluicing in languages such as
Bangla and Hindi is actually parallel to English in resulting from wh-movement and
a single occurrence of across-the-board deletion of all material present within TP.
To sum up now what has been considered in this section as a whole, the aim of
the section has been both to provide an initial characterisation of a range of basic
properties of sluicing in Bangla/Hindi and to attempt to establish what kind of analy-
sis of sluicing constructions would seem to be appropriate for Bangla/Hindi—one
incorporating movement and clausal ellipsis, or alternatively the assumption of some
kind of copula-centered structure with (potentially) no displacement of the remnant
wh-phrase. Having considered a variety of copular and case-related patterns in the
two languages, it was concluded that the observable evidence does not motivate an
underlying copula (or reduced cleft) structure, and instead points towards an analysis
of sluicing in Bangla and Hindi much closer to the wh-movement and IP/TP-deletion
analysis now widely assumed for languages such as English. If this is indeed correct
and the most plausible interpretation of sluicing patterns found in Bangla and Hindi,
a next, natural step in the investigation is to consider how common restrictions
on (wh-)movement might apply to the movement of wh-elements in Bangla/Hindi
sluices. This is now taken up as the subject matter of the following section.

. Sluicing, Superiority, and island effects


The general assumption that sluicing is the result of wh-movement and clausal ellipsis
creates the expectation that sluices should show the effects of restrictions on move-
ment (for example, Subjacency, the ECP etc.), and the issue of island effects within
sluicing constructions has recently become a topic of considerable interest (Chung
et al. , Fukaya and Hoji , Merchant , Lasnik , Fox and Lasnik ).
To a significant extent, the discussion has been centered on English, with reference
to certain other European languages, and Japanese. The goal of the present section is
therefore (in part) to help broaden the areal coverage of descriptions of locality effects
and sluicing, with a consideration of sluicing into island constituents in Bangla and
Hindi. To this we also add an investigation of Superiority and its potential effects in

 Another possibility for languages such as Bangla would be to suggest that the subject is realized as a
pro, and so syntactically present but phonetically null.
 In order to eliminate the assumption that raised possessor-wh cases such as () require a second
application of deletion in sluicing, it may be possible to suggest that the wh-constituent ‘whose’ is extracted
from the DP in its TP-internal position, and this does not result in a Left Branch Condition violation, owing
to island-repair effects from the PF-deletion of offending traces (Lasnik , Fox and Lasnik ). The
NP residue of the DP remnant could then be deleted as part of the TP, and only one instance of constituent
deletion would occur in the sluice.
Sluicing in Bangla and Hindi 205

sluicing constructions, a topic which has not received as much attention as island
phenomena in recent works on sluicing. 

.. Multiple-wh sluicing and Superiority


Ever since Chomsky (), multiple-wh questions have been noted to be subject
to restrictions on the way that raising of their wh-phrases takes place, with certain
sequences of raised and in situ wh-phrases being fully acceptable, and others quite
ill-formed and exhibiting a “Superiority effect”. Although initially a phenomenon
observed in languages such as English, where just a single wh-phrase undergoes
overt movement, Superiority effects have now also been extensively documented
for languages with multiple overt wh-movement, such as Bulgarian and Romanian
(Rudin , Bošković , Richards ). If sluicing constructions are derived via
multiple-wh movement in various languages, there is clearly an expectation that one
might observe Superiority effects in some of these languages, this to some extent
depending on how Superiority is understood to operate as a constraint. A common
intuition present in many analyses of Superiority phenomena is that the acceptabil-
ity of multiple-wh questions requires that the wh-phrase structurally closest to an
interrogative C be attracted to this C before other wh-phrases in its scope. Such an
understanding of Superiority has however been formally stated in two rather different
ways in recent years, and is seen by some as a derivational constraint (Chomsky
) and by others as a constraint on representations (Aoun and Li ). Con-
cerning multiple-wh sluices, these two modes of description can be noted to make
slightly different predictions. If Superiority reduced to the Shortest Move/Minimal
Link Condition is a constraint on derivations, then it is expected that violations
of such a constraint during the syntactic derivation of a sluice construction will
necessarily result in ungrammaticality, and multiple-wh sluices will be expected
to exhibit the same Superiority effects that are otherwise present in non-sluiced
multiple-wh questions. If Superiority is, however, a constraint on representations,
it is possible that “violations” of Superiority might allow for some kind of mitiga-
tion owing to the special reduced form of representation that sluiced clauses have
at Spell-Out/PF because of TP-deletion, as is suggested to occur in the case of vari-
ous islands (Lasnik , Fox and Lasnik ). Superiority effects are therefore not
necessarily predicted to occur in multiple-wh sluices under a fully representational
view.
Turning now to consider multiple-wh sluicing constructions in Bangla, it is found
that Superiority effects do seem to occur, as illustrated in ():

 We are actually not aware of any other studies of Superiority phenomena in sluice constructions,
though it is possible that these exist somewhere in the literature.
206 Tanmoy Bhattacharya & Andrew Simpson

() a. Mini-ke kal rate keu kichu diyeche, kintu ami jani
Mini-acc last night someone something give.has but I know
na [ke ki].
not who what
‘Last night someone gave Mini something, but I don’t know who and what.’
b. ∗ Mini-ke kal rate keu kichu diyeche, kintu ami jani
Mini-acc last night someone something give.has but I know
na [ki ke].
not what who

Such patterns might therefore be analyzed as violations of the Shortest Move/Minimal


Link Condition, with the violation of such constraints not being mitigated by sluicing
and clausal ellipsis, unlike the situation with island violations, as discussed in Lasnik
() and Fox and Lasnik (). However, we suggest that this is actually not the
correct interpretation of the unacceptability of (b) and other similar cases. In Bhat-
tacharya and Simpson (), a general reconsideration of Superiority phenomena
leads to the suggestion that Superiority effects are not the result of a purely hierarchi-
cal/structural restriction such as the Shortest Move Condition enforcing attraction
of the highest wh-phrase to a +Q C before all other wh-phrases, but should instead
be attributed to a representational constraint on the linear sequencing of information,
which requires that the initial (first moved) wh in a multiple-wh question be naturally
open to construal as a focal center of interest with a greater informational prominence
than secondary/additional wh-phrases present in the question (hence essentially be
more “topic-like” than the other wh-phrases present). Where it is difficult to naturally
construe the linearly initial wh-phrase as having such a property, for example when
an inanimate bare patient ‘what’ is promoted over a +human-agent ‘who’, Superiority
effects are clearly perceived, but these are a function of the mismatch between relative
focal prominence of wh-phrases and their positioning in a sequence of wh-phrases,
rather than being due to the distance and timing of movement of the wh-elements
present. When factors such as relative animacy and prosody are carefully controlled
for and speakers are guided to background contexts where (for example) objects
can be credited with a greater natural salience than subjects, then Superiority effects
are suspended from constructions that might be expected to have them on purely
structural grounds. In sum, then, what is (suggested to be) critical for the licensing
of multiple-wh sequences (of either Bulgarian or English types) is that the hearer can
construct a context in which the linearly initial wh-phrase is interpreted naturally as
more of a topic and center of interest than the remaining wh elements.
Multiple-wh sluicing constructions, approached from such a viewpoint, are partic-
ularly interesting, and especially so in those languages referred to as “discourse config-
urational languages” (Kiss ), where free-word-order permutations tend to mirror
the informational status of participants in a clause/event. In sluicing constructions,
Sluicing in Bangla and Hindi 207

wh-phrases regularly have antecedents explicitly represented in a clausal structure


that is assumed to correspond closely to the structure containing the wh-phrases.
Consequently, an informational-structuring and relative-prominence relation may be
regularly imposed on the antecedents of wh-phrases in the antecedent clause, and
guide hearers to assume that the same relative-prominence relation should hold for
the corresponding wh-phrases. In sluicing constructions, hearers will therefore be led
to assume that the relative prominence of DPs/PPs encoded in their linear ordering in
the antecedent clause should normally be mirrored by a parallel ordering in the sluiced
CP, and it may be less easy for hearers to accept a non-parallel sequencing of wh-
elements. These hypothesized properties of sluicing in languages with free word order
now lead to two expectations that can be empirically investigated. The first of these is
that it may be anticipated that Superiority effects will be felt more strongly in sluices
than in non-sluiced multiple-wh questions (which are not embedded in an explicit
context), as the latter allow hearers more freedom to build contexts that will license
wh-phrases in various linear orders, owing to the projection of different prominence
relations among the wh-phrases. This expectation turns out to be generally correct in
Bangla, and Superiority effects in sluices are considerably clearer than in non-sluiced
multiple-wh questions.
The second expectation raised by the above is that the availability of free word
order might be used to build an ordering of elements in the antecedent clause of
a sluice that would help license a (parallel) ordering of wh-phrases that technically
“violates” the Shortest Move Condition. For example, if the unacceptability of the ki >
ke ‘whatobject ’ > ‘whosubject ’ ordering in (b) is due to it being a genuine violation of
Shortest Move, the expectation is that such orderings should always be unacceptable.
However, if the ill-formedness of (b) is instead due to the difficulty of accepting
bare ‘what’ as more focal and topic-like than ‘who’, then it might be expected that a
manipulation of the order of the antecedents to the wh-phrases in the matrix clause
might render a ‘what’ > ‘who’ ordering in the sluice more acceptable. If scrambling
(or some other type of movement) can be used to promote an object kichu ‘something’
over a subject keu ‘someone’ and present it as putatively more topic-like than a subject
keu, this might have effects on the acceptability of a ke > ki ordering in the sluice,
structuring a context that might allow for patient/theme ke ‘what’ to be interpreted as
genuinely more of a center of interest than agent ke ‘who’. The results of investigating
such a possibility are interesting and are described below.
An initial attempt at switching the order of the antecedent DPs in the matrix clause
and trying to copy this linear sequencing into the order of the sluiced wh-phrases
does not appear to allow for the legitimate occurrence of ke > ki ‘what’>‘who’ in the
sluice, and a ‘non-matching’ ki > ke ‘who’>‘what’ order is still necessary:

() a. ∗ kichu keu curi-koreche, kintu ami jani na [ki ke].


something someone steal-do.has, but I know not what who
208 Tanmoy Bhattacharya & Andrew Simpson

b. kichu keu curi-koreche, kintu ami jani na [ke ki].


something someone steal-do.has, but I know not who what
‘Someone has stolen something, but I don’t know who and what.’

The resistance to a whsubject > whobject order therefore continues to be particularly


strong, and might be seen to support a Shortest Move analysis of Superiority in sluices.
However, if the sluiced CP containing the wh-phrases is fronted to the topic position
of the higher embedding clause, it is found that inversion of the sequencing of the
antecedent DPs does have a very clear effect on the sequencing of the wh-phrases in
the sluice, and a ‘what’ > ‘who’ order becomes fully acceptable:

() Mini-ke kichu kal rate keu diyeche, kintu [ki ke] ami
Mini-dat something last night someone give.has but what who I
jani na.
know not
‘Last night someone gave something to Mini, but I don’t know who and what.’

Note that the occurrence of a ‘what’ > ‘who’ order in this pre-subject topic position is
clearly dependent on the inversion of the antecedent DPs in the initial clause, and is
unacceptable as an ordering if there is no re-sequencing of the antecedent ‘something’
to a position before ‘someone’. Hence, it is not true that the occurrence of wh-elements
in the pre-subject position is unconstrained:

() ∗ Mini-ke kal rate keu kichu diyeche, kintu [ki ke] ami
Mini-dat last night someone something give.has but what who I
jani na.
know not
‘Last night someone gave Mini something, but I don’t know who and what.’

Furthermore, if a linear sequencing of ‘something’ > ‘someone’ occurs in


the antecedent clause, this enforces the Superiority-violating ‘what’ > ‘who’ order in
the raised pre-subject sluiced CP, and a non-matching ‘who’ > ‘what’ order is in fact
not permitted:

() ∗ Mini-ke kichu kal rate keu diyeche, kintu [ke ki] ami
Mini-acc something last night someone gave but who what I
jani na.
know not
‘Last night someone gave Mini something, but I don’t know who and what.’

Such patterns therefore clearly support an approach to Superiority that allows for con-
textual prompts and the manipulation of information structure to license whobject >
whsubject configurations, and are unexpected in analyses that attribute Superiority
effects to purely structural conditions such as Shortest Move. Reflecting on why a
Sluicing in Bangla and Hindi 209

realignment of the indefinite antecedents permits a ‘what’ > ‘who’ sequencing when
the sluiced CP is fronted but not when it follows the verb, one possible explanation is
that this relates to the rather different intonation possibilities open to wh-phrases in
the two positions. In post-verbal, final position, wh-phrases in sluices are restricted
in the level of stress they commonly bear, whereas in pre-subject, fronted position,
sluiced wh-phrases must actually occur with a significant degree of stress and also
allow for slight pausing to occur in multiple-wh sequences (which is not a natural
intonation pattern for post-verbal multiple-wh sequences). It is possible that the
“legitimization” of a bare inanimate ‘what’ as more focal and topic-like than a subject
‘who’ may require the use of special intonation patterns in conjunction with the re-
sequencing of the DP antecedents, and such intonation is only naturally produced in
pre-subject position.  , 
Finally, it can be noted that there is a second pragmatic way to influence the accept-
ability of ‘what’ > ‘who’ sequences in sluicing constructions, which further motivates
an approach to Superiority that is not purely based on hierarchical structure. If an
emphatic adverbial such as Sotti ‘really’ is used to qualify the verb of cognition that
embeds the sluiced CP, it is found that either ordering of a bare wh-subject and wh-
object are possible in sluices without movement of the sluice to pre-subject position,
and also without any re-sequencing of the antecedent DPs:

 A similar observation about the effect of prosody on Superiority violations in English is made in
Bhattacharya and Simpson (), and may well be relevant for German too, as noted in footnote .
 Interestingly, a pattern very similar to that described here in Bangla is found in German. German is a
language in which Superiority effects seem to be largely absent. However in multiple-wh sluices, Superiority
effects are actually felt very clearly, as illustrated in (i).

(i) Gestern hat Hans jemandem etwas besonderes gegeben, aber ich weiss nicht
yesterday has Hans someone.dat something special given but I know not
[wem was] / ∗ [was wem].
whom what what whom
‘Yesterday Hans gave someone something, but I don’t know whom and what.’

Here, if topicalization of the indefinite object was ‘what’ is employed to try to promote its interpretation
as more focal and topic-like than wem ‘whom’, this is not sufficient to license a was > wem order in the
post-verbal position (as in Bangla). However, topicalization of the object combined with fronting of the
sluiced CP does allow for such an order to occur:

(ii) Etwas besonderes hat Hans gestern jemandem gegeben, aber [was wem] weiss
something special has Hans yesterday someone.dat given but what whom know
ich gar nicht.
I at.all not
‘Yesterday Hans gave someone something, but I don’t know whom and what.’

This patterning is fully parallel to that described for Bangla, and also seems to be licensed by different
intonational properties of the post-verbal and topicalized position of the sluiced CP. Many thanks to
Joachim Sabel for helpful discussion of the German data here.
210 Tanmoy Bhattacharya & Andrew Simpson

() Mini-ke kal rate keu kichu diyeche, kintu ami Sotti
Mini-dat last night someone something give.has but I really
jani na [ke ki] / [ki ke].
know not what who who what
‘Last night someone gave Mini something, but I really don’t know who and
what.’
This licensing effect of the introduction of the adverbial is interestingly also one that is
sensitive to the linear ordering of the adverbial and the sluiced wh-phrases, something
which may perhaps be a processing effect. If the sluiced CP containing the wh-phrases
is fronted to the topic position of the higher clause, and so precedes the adverbial
Sotti, the licensing effect of the adverbial is lost, and only the order ‘who’ > ‘what’ is
possible, as shown in (). For the special ordering of ‘what’ > ‘who’ to be possible
and accepted by speakers, it therefore seems that Sotti has to be processed before the
Superiority-violating ki > ke sequence: 
() Mini-ke kal rate keu kichu diyeche, kintu [ke ki]
Mini-dat last night someone something give.has but who what
/ ∗ [ki ke] ami Sotti jani na.
what who I really know not
‘Last night someone gave Mini something, but I really don’t know who and
what.’
Such a patterning is again rather unexpected for purely hierarchical analyses of Supe-
riority phenomena, but less surprising for accounts that also accord an importance to
the linear presentation of elements in a structure.
In sum, then, the set of patterns observed above in multiple-wh sluicing provide
useful information and insight into the nature of Superiority, in part because of the
effects that the antecedent clause and its manipulation can have on the sequencing
of wh-phrases, and also because of the ability of the wh-phrases contained in the
sluiced CP to occur in different positions in the sentence and give rise to differ-
ent Superiority effects accordingly (though maintaining the same relative sequenc-
ing of wh elements). Sluicing constructions may also show Superiority effects more
strongly/clearly than non-sluiced multiple-wh questions. Quite generally, the obser-
vation that Superiority effects can be overcome by the control of elements external to
the CP constituent in which wh-movement occurs, and by the positioning of the CP
itself, can be noted to be difficult to reconcile with analyses of Superiority that only

 In terms of its contribution to the meaning of the sentence, the use of Sotti seems to result in a highly
non-d-linked interpretation of the wh-phrases, indicating that the speaker has no idea of the identity of the
item given and the person given to. This is interesting because it is often suggested that an increase in the
d-linked interpretation of wh-phrases can lead to Superiority violations being overcome (Pesetsky ).
Here, however, the opposite effect appears to be attested, and the highlighting of specifically non-d-linked
interpretations of wh-phrases seems to allow them to licitly occur in a Superiority configuration.
Sluicing in Bangla and Hindi 211

focus on the movement of wh-phrases inside a CP, and characterize Superiority effects
as the violation of a purely structural constraint computing the distance (and timing)
of wh-movement. The variety of patterns attested in sluices instead provides impor-
tant support and justification for a view of Superiority as an information-structure
restriction on the output of (wh-)movement, requiring that the linearly initial wh
element in a multiple-wh question be naturally interpretable as the primary focus of
the question. By making use of the particular properties of sluicing constructions, it is
therefore possible to probe the underlying nature of a constraint regularly associated
with movement, and to identify patterns that pose an interesting challenge to estab-
lished views, just as has similarly occurred in relation to island phenomena in recent
years, as will shortly be discussed.

.. Sluicing and islands in Bangla and Hindi


If sluicing constructions are frequently the result of operations of wh-movement
applying within an underlying clausal structure elided at PF, a second natural expec-
tation is that this movement would be island-sensitive and not permitted out of
configurations blocking other pre-Spell-Out extraction. It is therefore an intrigu-
ing and interesting surprise about sluices in English and a variety of other lan-
guages that sluicing often seems to be acceptable in structures assumed to con-
tain islands prior to PF-ellipsis. The reaction to such an unanticipated patterning
in Lasnik () and Fox and Lasnik () has been to suggest that the appar-
ent ability of wh-movement in sluices to violate regular locality restrictions is a
function of the clausal ellipsis operation at PF; sluicing of the TP containing the
path of wh-movement results in deletion of all illicit traces of wh-movement, and
so gives rise to the phenomenon of “PF island repair”. Potentially in conflict with
such a theory and the patterning that supports it, however, is Japanese, where
sluicing has been argued to be island-sensitive, and not permitted when the wh-
phrase relates to a position within an island for movement (at least, when the wh-
phrase is case-marked; see Takahashi () and Fukaya and Hoji ()). This
fully opposite patterning in Japanese is genuinely puzzling. It raises the question
of why there should be a difference in languages with regards to the effects of PF-
ellipsis of island structures, and whether the occurrence of PF island repair might
indeed be open to some kind of parametric variation. If it were, one would cer-
tainly expect to find other languages with island-sensitive sluicing patterns similar
to those described for Japanese, and so it is useful and important to expand the
cross-linguistic depth of coverage of the interaction of sluicing with island config-
urations. As Bangla and Hindi have now both been argued to have sluicing con-
structions resulting from wh-movement and clausal deletion, they clearly have the
potential to contribute to the issue of the (non-)occurrence of island effects in
sluices. Typologically, Bangla and Hindi are also similar to Japanese in a num-
212 Tanmoy Bhattacharya & Andrew Simpson

ber of ways, being pro-drop, case-marking, free-word-order languages, with neutral


SOV–Aux patterns, and so might perhaps be expected to show a similar island-
sensitivity in sluices to that reported for Japanese.
Considering first Bangla, it is found that island restrictions actually do not appear to
affect sluicing constructions, and wh-phrases are free to relate to positions within the
full range of island types in the (hypothesized) underlying structure. We illustrate this
here with examples of relative-clause islands (), coordinated-DP structures (),
complex NPs (), adjunct CPs (), and wh-islands (). As is shown in (), (),
(), (), and (), all such configurations are strong islands for non-sluiced overt
wh-movement/extraction. The patterning of island-related sluices in Bangla would
therefore seem to further support the generalization based on languages other than
Japanese that the effects of islandhood can be voided by PF clausal ellipsis. Note,
furthermore, that case-matching effects are observed in the Bangla island sluices just
as in non-island sluicing (see (), (), (), and ()).
Relative-clause examples:

() ora Emon kaw-ke nite cay je kono Ek-Ta bharotiyo


they such someone-dat take.to want who who one-cl Indian
bhaSa bOle, kintu amar mone nei, kon-Ta.
language speaks but my mind.in not.be which-cl
‘They want to hire someone who speaks an Indian language, but I don’t know
which one.’

() ∗ ora [kon bharotiyo bhaSa]i Emon kaw-ke nite cay je


they which Indian language such someone-dat take.to want who
t i bOle?
speaks
(‘∗ Which Indian language do they want to hire someone who speaks?’)

Coordinated-DP examples:

() ora [jonaki eboN onno ar-Ek-jon chatro-ke] Sahajjo korte


they Jonaki and another and-one-cl student-dat help to.do
ceSTa korlo, kintu amar mone nei ka-ke.
try did but my mind.in not.be who-dat
‘They tried to help Jonaki and some other student, but I don’t know who.’

() ∗ ora [kon chatro-ke]i bollo tumi [jonaki eboN t i ] Sahajjo korte
they which student-dat said you Jonaki and help to.do
ceSTa korlo?
try did
(‘∗ Which student did they say you tried to help Jonaki and?’)
Sluicing in Bangla and Hindi 213

Complex-NP examples:
() ami [Mina-r Sikriti je jonaki kaw-ke bhalobaSe]
I Mina-gen acknowledgement C Jonaki someone-dat loves
biSSaS kori, kintu ami jani na ka-ke.
believe do but I know not who-dat
‘I believe Mina’s acknowledgement that Jonaki loves someone, but I don’t know
whom.’
() ∗ tumi ka-kei [Minar Sikriti je jonaki t i bhalobaSe]
you who-dat Mina-gen acknowledgement C Jonaki loves
biSSaS kOro?
believe do
(‘∗ Whom do you believe Mina’s acknowledgement that Jonaki loves?’)
Adjunct-CP examples:
() jonaki parTi cheRe cole gElo [karon Ek-jon otithi o-ke
Jonaki party leave walk went because one-cl guest him-dat
Opoman korlo], kintu o ama-ke bollo na ke.
insult did but he me-dat said not who.nom
‘Jonaki left the party because one of the guests insulted him, but he didn’t tell
me who.’
() ∗ ora [kon otithi]i bollo je jonaki parTi cheRe cole gElo karon t i
they which guest said C Jonaki party from walk went because
o-ke Opoman korlo.
him-dat insult did
(‘∗ Which guest did they say that Jonaki left the party because (that guest)
insulted him?’)
Wh-island examples:
() jonaki bojha-r ceSTa korche je kon chatro kono Ek
Jonaki understand-gen try is.doing C which student some one
SOmoSSa-r Somadhan jane, kintu amader bolche na
problem-gen solution knows but us is.saying not
kon-Ta-r
which-cl-gen.
‘Jonaki is trying to work out which student knows the solution to some prob-
lem, but he’s not telling us which one [which problem].’
() ∗ jonaki [kon SOmoSSa-r]i bojha-r ceSTa korche je kon
Jonaki which problem-gen understand-gen try is.doing C which
chatro t i Somadhan korte parbe.
student solution to.do will.be.able
(‘∗ Which problem is Jonaki trying to work out which student will be able
to do?’)
214 Tanmoy Bhattacharya & Andrew Simpson

Turning now to Hindi, an investigation of parallel patterns to the above, making use
of the same data set with a range of linguistically sophisticated informants resulted
in a surprising non-uniformity of judgments on the acceptability of sluicing with
island structures, and considerable variation in reaction to the data presented. This
ranged from rejection/non-acceptance of all attempts to build sluices incorporating
island structures, through a middle ground of finding some but not other structures
unacceptable, to a more general acceptance of the majority of island configurations in
sluicing sentences. A subset of the data used to elicit judgments from linguist native
speakers is given below in (–), with a “%” symbol marking examples that were
acceptable to some but not other speakers.

Relative-clause example:

() %vo log kisi ek aadmii-ko nOkrii-denaa-caahte hEN jo koi


those people some one man-dat hiring-giving-want be C some
ek bhaartiya bhaSa boltaa hai, par mujhe nahiiN maalum ki
one Indian language speak be but me.dat not known C
kOnsii.
which
‘Those people want to hire someone who speaks an Indian language, but I don’t
know which one.’

Adjunct-CP example:

() %mE-ne sunaa hE ki Sharma-ji bahut khuS honge agar Raam


I-erg heard be C Sharma-pol very happy be if Ram
unkii kisi baccii-se Saadi-kar-le, par mujhe nahiiN
his some daughter-with marriage-do-cond but me.dat not
pataa ki kis-se.
known C who-with
‘I heard that Mr. Sharma will be very happy if Ram marries one of his daughters,
but I don’t know which one.’

Wh-island example:

() ∗ Raam pataa kar rahaa hai ki kOnsii laRkii uskaa ek kamraa
Ram know do asp be C which girl his one room
kiraaye-par legii, par mujhe nahiiN maalum ki kOnsaa.
rent-loc take.will but me.dat not known C which
(‘Ram is finding out which girl will rent one of his rooms, but I don’t know
which one.’)
Sluicing in Bangla and Hindi 215

Coordinated-DP example:
() %Un logoM-ne Raam Or ek laRkii se madad karne-ko kahaa,
those people-erg Ram and one girl with help doing-acc said
par mujhe nahiiN pataa ki kOnsii.
but me.dat not known C which
‘Those people asked Ram and one girl to help, but I don’t know which (girl).’
Complex-NP example:
() %mE Minaa-kii yah baat maantaa huN ki Raam kisi-se
I Mina-gen this word believe be C Ram someone-with
pyaar-kartaa hE, par mujhe nahiiN pataa ki kis-se.
love-doing be but me.dat not known C who-with
‘I believe Mina’s claim that Ram loves someone, but I don’t know who.’
As indicated by the star in (), sluicing involving a wh-island was felt to be
unacceptable by all consulted. The adjunct-CP and relative-clause examples, () and
(), were accepted or rejected by equal numbers of speakers, and complex-NP and
sentential-subject structures were accepted more often than they were rejected.
At the present time, it is not fully clear how to interpret this variation in reaction to
sluicing with island structures in Hindi. With regards to the unacceptability of similar
forms in Japanese, Nakamura (this volume) raises the interesting possibility that this
may be due to differences in the way that case is assigned in Japanese and English-
type languages. Supposing there were a genuine difference between Bangla and Hindi
in island-related sluicing, one might wonder whether this could then be related to
differences in case-assignment mechanisms in the two languages. However, although
there are indeed certain differences in the case systems present in Bangla and Hindi,
with Hindi case being somewhat more segmentable than Bangla (and hence more like
Japanese in the relevant way), these differences do not seem substantial enough to us
to justify significantly distinctive mechanisms of case assignment. In addition to this,
it is not obvious how internal variation among speakers of Hindi would be naturally
captured in such a case-based approach.
One potential clue to variation in the acceptance of island-sluicing that may be
more relevant and worthy of consideration relates to the physical instantiation of
the wh-phrase in the remnant clause in the Hindi examples—the use of either a
bare wh form, such as ‘who’ or ‘what’, or a ‘which(-NP)’-type phrase. When wh-
phrases of the latter type were used, it was frequently reported that there was a clear
preference for non-deletion of the NP complement, and if the elided NP parts of
such phrases were restored in various of the island violation cases such as () and
(), this resulted in a marked improvement in their acceptability. This suggests that
the unacceptability of sluicing with islands in at least certain instances may be a
parsing-like difficulty some speakers experience in recovery of the reference of the
wh-phrases; when the wh-phrase is more explicitly specified, its link to the antecedent
216 Tanmoy Bhattacharya & Andrew Simpson

is more easily established. Note that ‘which’ in Hindi is actually inflected with gender
features, and Bangla bare ‘which’ (where the NP is elided) regularly occurs with a
classifier attachment. Both of these specifications have the potential to provide (what
one might expect would be) unambiguous clues to the reference of the antecedent of
the wh-phrase. However, the gender specification in Hindi is possibly not perceived
as salient enough when the antecedent of a wh-phrase is embedded in the more
complex surrounding structure of an island configuration, and retention of the NP
complement is required to fully disambiguate the linking. A related preference for
‘which-NP’-forms over bare wh-phrases was also attested in island-related sluicing in
the investigation of Hindi, and is paralleled by similar preferences in English sluicing
with islands, as illustrated in (b), where the antecedent of the wh-phrase is contained
within a relative-clause island:
() a. Bill just criticized one of the teachers, but I don’t know who.
b. I just talked with someone who knows one of your teachers, but I can’t
remember ?who/which teacher.
If there is certain variation among speakers in the way that wh-phrases are naturally
linked to syntactically legitimate antecedents (in Hindi this possibly being related to
the salience with which clues such as gender-marking are perceived), this may allow
for the beginnings of an account of the variation perceived within Hindi in cases of
sluicing with island constituents.  The natural question following on from this is
whether all cases of island-sluices classed as unacceptable might ultimately be reduced
to a recovery problem or not.
Supposing that it were the case, it is then tempting to wonder whether certain other
cross-linguistic cases of apparent island violations with sluicing might also be reduced
to issues of recoverability, with three cases in particular suggesting themselves here.
First of all, there is the recovery of the content of “implicit correlates” (Merchant
), which is noted to be easily made in non-island environments, but not pos-
sible in island configurations.  Secondly, there is the recovery of the attachment
site of wh-adjuncts in the antecedent clauses of sluices in Japanese (and possibly
other languages): Nakamura (this volume) notes that sluiced wh-adjuncts cannot be
interpreted as relating to positions within an island configuration in the antecedent
clause, and this locality restriction is not the result of the case-assignment mechanism
 To some extent, the situation could be likened to the occurrence of pro in a Rizzi ()-like approach,
where there is an important distinction in the licensing and recovery of the content of pro. The sluicing of
island structures might be licensed by syntax and PF-deletion, but be perceived as unacceptable, owing to
a failure to successfully link a wh-phrase with its intended antecedent and so recover the potential content
of the wh-phrase.
 Examples of the use of implicit correlates in sluicing, also given in Paul and Potsdam (this vol-
ume), are:

(i) She’s reading, but I can’t imagine what.


(ii) ∗ Kim knows the person who was reading, but she won’t say what.
Sluicing in Bangla and Hindi 217

proposed in the chapter, which only imposes a locality restriction on wh-arguments.


Thirdly, there is the linking of wh-adjuncts in sluices in Chinese to positions in an
antecedent clause; in Wei () this is described as not being possible to positions
within preceding island structures. However, no wh-movement is assumed to occur in
the construction of Chinese sluice-type constructions, and hence the locality restric-
tion on adjuncts must be given some other kind of explanation.
Furthermore, variation in the acceptance of sluicing with islands such as that
encountered in the investigation of Hindi is actually not so fully absent from other
languages regularly discussed in the literature. Although English is now commonly
described as not showing island effects in sluicing constructions, when sluicing in
English was originally considered in Ross (), it was suggested that it was in
fact constrained by syntactic islands.  Conversely, with regard to Japanese, although
Fukaya and Hoji () present case-marked sluicing in Japanese as being regularly
unacceptable with island structures, Fukaya () notes that various speakers report
different judgments and find island-related sluicing of the type discussed to actually
be quite acceptable.
All the above seems to highlight the need for careful control of the data, and for
further testing to discover whether the unacceptability reported in various instances
of island-sluicing is due to real syntactic differences among languages, or is alterna-
tively a non-syntactic artefact of the data, perhaps arising from added complications
in the information structure present in more complex syntactic configurations. With
regards to the situation in Hindi, this will have to be the target of a future investigation
that focuses specifically on islands, and cannot be resolved here.  We will therefore
have to leave the jury currently out and in need of further evidence before being able
to reach a solid verdict on the sensitivity of Hindi sluicing to syntactic islands, and
whether this would in turn result in a real difference to the patterning in Bangla.
If a rigorous further probing of Hindi confirms that island restrictions are indeed
present in the language but remain absent from Bangla, what would then seem to be
required is a wider, careful survey of Indo-Aryan languages to determine how these
might vary with regard to the presence/absence of locality restrictions in sluicing,
and whether it may be possible to systematically relate this to other properties found
in such languages. With such a future call to arms, we now close the present chapter
with a brief summary of progress made in the course of this study of sluicing in Bangla
and Hindi.

 Sluicing constructions involving islands were, however, characterized as being less ungrammatical
than their non-sluiced counterparts, typically with an unacceptability ranging from “?” for Complex NP
Constraint violations, through “??” for Coordinate Structure and Sentential Subject Constraint violations,
to “∗ ” for Left Branch Condition violations (Ross  (–)).
 However, initial signs are that issues of recoverability will not be the appropriate way to attempt to
explain all cases of perceived unacceptability in Hindi island-sluices, and the clear ill-formedness of sluicing
with wh-islands in particular seems to suggest that syntactic locality principles are being violated in such
configurations.
218 Tanmoy Bhattacharya & Andrew Simpson

. Concluding remarks


The aim of this investigation was to build up an understanding of the way that sluicing
constructions are derived and syntactically structured in Bangla and Hindi, as a step
towards incorporating such languages in wider debates on cross-linguistic properties
of sluicing. Viewed from a traditionalist perspective, Bangla and Hindi are both wh-
in-situ languages and so are not obviously expected to exhibit sluicing, on the assump-
tion that this requires operations of wh-movement. Reconsidered in a rather different
way, however, and following Simpson and Bhattacharya (), Bangla and Hindi can
actually both be argued to exhibit overt wh-movement in regular wh-questions, and
therefore might well be expected to license movement-derived sluice constructions.
The chapter also carefully examined a number of alternative possibilities, suggested
to be instantiated in sluices in other wh-in-situ languages, but demonstrated that such
potential analyses would not be consistent with observable language-internal proper-
ties of Bangla and Hindi. It was therefore concluded that the most plausible analysis
of sluicing in Bangla/Hindi is indeed that it is produced, as in English and many
other languages, by a sequence of wh-movement and PF-ellipsis of a clause broadly
isomorphic to that of the antecedent clause. Such a conclusion then naturally led to an
examination of how various constraints assumed to restrict (wh-)movement interact
with sluicing constructions. Considering the case of Superiority first, it was argued
that Superiority effects are in fact the result of an output constraint on the linear
sequencing of wh-phrases rather than a hierarchical restriction on the attraction of
wh-phrases to COMP, and that various of the properties of sluicing constructions in
particular can be used to demonstrate this, with interesting variation in Superiority
effects appearing in topicalized and post-verbal positions of sluiced CPs. Finally, the
chapter examined how the introduction of island constituents into sluicing construc-
tions may have effects on their well-formedness, and detected intriguing differences in
reactions to island-sluices in Bangla and Hindi. While further exploration now seems
to be necessary to ascertain the real loci of variation within Hindi, this first pass at
Hindi and Bangla suggests that the study of sluicing and islands has the potential to
be a highly interesting area of future comparative research within the Indo-Aryan
group of languages.


Sluicing in Mandarin Chinese:


An instance of pseudo-sluicing
PE R NG WA NG A DA M S A N D S ATO SH I TOM IOKA

. Introduction
The purpose of this chapter is to reveal the nature of apparent cases of sluicing in
Mandarin Chinese (henceforth, Chinese for short) and endorse a non-deletion analy-
sis of them. “Sluicing” refers to an elliptical interrogative sentence where only a wh-
phrase is overtly pronounced. The sentences in () exemplify the sluicing construction
in English.
() a. John bought something for his girlfriend, but he didn’t tell us what.
b. If Mary went somewhere for lunch, her husband must know where.
c. Bill went to Delaware in a hurry, and his parents wondered why.
Chinese also has a construction akin to English sluicing, as shown in (). For the
sake of convenience, we will refer to this construction as “Chinese sluicing”, but this
terminology by no means indicates that the construction in question is syntactically
comparable to its English counterpart.
() a. Lisi chuqu yuehui; ta mama xiang zhidao (shi) zai nali /han
Lisi go-out date sg mother want know cop at where with
shei.
who
‘Lisi went out for a date; her mother wondered where/with whom.’
b. Lisi mai le yiyang dongxi gei mouren, dan wo bu zhidao shi
Lisi buy asp one-cl thing give someone but sg not know cop
shei /shenme.
who what
‘Lisi bought something for someone, but I don’t know who/what.’
In (), even untrained eyes can detect a major difference between Chinese and English,
namely, the appearance of the copula verb shi. This presence is not accidental, and
220 Perng Wang Adams & Satoshi Tomioka

it plays an important role in our proposal. The second issue concerns the typology
of wh-interrogative sentences. English is a wh-movement language, and the overt
wh-movement fits well with our understanding of how deletion and ellipsis works.
It is generally agreed among researchers (or at least there is some kind of desider-
atum shared among them) that the deletion/ellipsis process only targets syntactic
constituents. As illustrated in (), the overt wh-movement sets the sentence up nicely
for the deletion/ellipsis process of the non-wh part of the sentence. The strikethrough
represents the portion affected by the process.
() The derivation of (a):
John bought something for his girlfriend, but he didn’t tell us [CP whati [IP John
bought t i for his girlfriend]].
This characterization of English sluicing naturally raises a question for those lan-
guages that do not have obligatory wh-movement. One of the main research ques-
tions in the syntax of wh-in-situ languages has been on the mobility of in situ wh-
questions: Does covert wh-movement occur at the level of logical form (LF), or are
in situ wh-phrases truly immobile? However, this query does not have direct conse-
quences for the matter of sluicing in wh-in-situ languages. The English-style sluicing
process would not be a viable option for those languages unless movements of wh-
phrases take place at the level of surface syntax. Are there independent motivations
for such movements (e.g. focus movements of wh-phrases)? If there are, is English-
style sluicing a viable option for Chinese? On the other hand, if the process of wh-
movement followed by deletion is unavailable for Chinese, what process yields the
structure in ()?
The first step to answer these questions is to grasp the characteristics of Chinese
sluicing. As will become clear in the following section, Chinese sluicing shows some
features that are shared by its English counterpart, but these similarities are eclipsed by
many striking differences. We argue that these differences call for a radically different
structure for Chinese sluicing. More concretely, we endorse the “pseudo-sluicing”
structure in the sense of Merchant (), as illustrated in ().
() Lisi mai le yiyang dongxi gei Dawu, dan wo bu zhidao [pro shi
Lisi buy asp one-cl thing give Dawu but sg not know cop
shenme].
what
‘Lisi bought something for Dawu, but I don’t know what it/that was.’
The pseudo-sluicing analysis, first proposed by Adams () and Wei (),
hypothesizes that a sluiced clause has a phonologically silent pronominal subject
followed by the copula shi and the wh-remnant. We will present a variety of facts sur-
rounding Chinese sluicing that support this analysis. However, this pro-form analysis
of Chinese sluicing has been challenged, and we will examine some problematic facts
presented by the opponents to the pseudo-sluicing approach. When appropriate, the
Sluicing in Mandarin Chinese 221

current proposal will be compared with such alternatives as the focus-plus-deletion


analysis of Wang and Wu () and the reduced-cleft analysis commonly found
for Japanese sluicing (Shimoyama , Nishiyama , Kizu ). We hope to
show that, even with some puzzles left unaccounted for, the pseudo-sluicing analysis
presents the most likely scenario of what is going on in Chinese sluicing.

. Characteristics of Chinese sluicing


In this section, we list the main properties of Chinese sluicing, and in doing so,
we will make frequent references to its English counterpart. Let us begin with the
common features that the two languages share. First, sluicing requires a pronounced
antecedent IP; in other words, sluicing cannot be uttered when the only clue on the
missing material is provided non-linguistically. The contrast found in () illustrates
this point.
() a. Mouren jie le zheben shu, danshi meiren gaosu wo shi
Someone borrow asp this-cl book but nobody tell sg cop
shei.
who
‘Somebody borrowed this book, but nobody told me who.’
b. #(I was searching for a book of mine.)
Meiren gaosu wo shi shei.
Nobody tell sg cop who
‘Nobody told me who.’
Secondly, in the discourse preceding a sluiced sentence, one often finds a correlate to
the remnant wh-phrase, called the inner antecedent in Chung et al. (). When
there is such a correlate, it must be indefinite. The examples in () exemplify this
restriction.
() a. Universal-quantifier inner antecedent
∗ Lisi
kan dao meige xuesheng, danshi Dawu bu zhidao shi shei.
Lisi see asp every-cl student but Dawu not know cop who
‘∗ Lisi saw every student, but Dawu doesn’t know whom.’
b. Proper-name inner antecedent
∗ Lisi
dao Riben qu le, danshi Dawu bu zhidao shi nali.
Lisi to Japan go asp but Dawu not know cop where
‘Lisi went to Japan, but Dawu doesn’t know where.’ 
 The English version here can mean “Dawu doesn’t know where in Japan Lisi went”, but the Chi-
nese counterpart in (b) does not render this different (unintended) interpretation. Incidently, the
English copular construction Dawu doesn’t know where that is does not have this different meaning
either.
222 Perng Wang Adams & Satoshi Tomioka

Thirdly, the voice of the verb used in the antecedent IP and the voice shown by the
wh-remnant have to be identical. In (), for instance, an active clause is followed by a
sluiced clause in the passive voice, and the resulting sentence is unacceptable.
() Active–passive mismatch
∗ Mouren da ku le Lisi, danshi wo bu zhidao shi bei shei.
Someone hit cry asp Lisi, but sg not know cop pass who
‘∗ Someone hit Lisi and made her cry, but I don’t know by whom.’
Despite these similar licensing environments, Chinese sluicing differs from its English
counterpart in several respects. As these differences form the foundation of our pro-
form analysis, we will discuss them one by one.

.. The presence of the copula shi


The first difference is the presence of the copula, which has already been mentioned in
the previous section. It is required in the sluicing structure with the simplex-argument
wh-phrases shei ‘who’ and shenme ‘what’ in Chinese, as shown in ().
() a. Lisi mai le yijian liwu, danshi ta bu gaosu wo ∗ (shi) shenme.
Lisi buy asp one-cl present but sg not tell sg cop what
‘Lisi bought a present, but she didn’t (want to) tell me what.’
b. Lisi da dianhua gei mouren, danshi wo bu zhidao ∗ (shi) shei.
Lisi hit phone to someone but sg not know cop who
‘Lisi called someone, but I don’t know whom.’
For other wh-phrases such as prepositional phrase (PP) wh-phrases (‘with whom’, ‘to
whom’, ‘to where’, ‘for what’, etc.) and complex-argument wh-phrases (‘what person’,
‘what thing’, ‘what time’, ‘which place’, ‘which one/song’, ‘how-many dogs’, etc.), the
occurrence of the copula is optional, as shown in ().
() a. Lisi congcong qu le Riben, danshi wo bu zhidao (shi)
Lisi hurry-red go asp Japan but sg not know cop
weishenme.
for-what
‘Lisi went to Japan in a hurry, but I don’t know why.’
b. Lisi han mouren qu Riben, danshi wo bu zhidao (shi) han shei.
Lisi with someone go Japan but sg not know cop with who
‘Lisi went to Japan with someone, but I don’t know with whom.’
c. Ruguo Lisi qu le mouge difang, ta mama kending zhidao (shi)
If Lisi go asp some-cl place sg mom surely know cop
nali.
where
‘If Lisi goes to some place, her mother knows for sure where.’
Sluicing in Mandarin Chinese 223

d. Lisi bu xihuan yishou ge, danshi wo bu zhidao (shi)


Lisi not like one-cl song but sg not know cop
neiyishou(ge).
which-one-cl(song)
‘Lisi doesn’t like one song, but I don’t know which one/song.’
e. Lisi yang le yixie gou, danshi wo bu queding (shi)
Lisi raise asp some dog but sg not sure cop
jizhi(gou).
how-many-cl(dog)
‘Lisi kept some dogs, but I am not sure how many dogs.’
The presence of the copula is one of the main obstacles for extending the “deletion”
analysis of English sluicing to Chinese, as pointed out by the original advocates of the
pro-form analysis, Adams () and Wei ().

.. Implicit adjunct and argument correlates


One of the striking features of English sluicing is that the presence of an indefinite
correlate (an inner antecedent) is optional regardless of wh-type. Whether the rem-
nant is an argument (see ()) or an adjunct (see ()), the correlate can be missing in
the antecedent clause.
() a. Maria was reading (something) last night, but we didn’t know what.
b. The host is ready to serve the guests (something), but she hasn’t told us
what.
() a. Anna has already gone (to some place), but no one knows where.
b. Jenny is going out (with someone) tonight, but she won’t tell us with whom.
c. Diana was very upset (for some reason), and she let everyone know why.
In Chinese, however, there is a clear argument–adjunct asymmetry: While the pres-
ence of an antecedent for an adjunct wh-remnant is merely optional, just as is the case
for English, it is required for an argument wh-remnant. The sentences in () and ()
exemplify this contrast.
() Argument-wh
a. ∗ Lisi zai fangjian li yandu ___, danshi wo bu zhidao shi
Lisi at room in study but sg not know cop
shenme.
what
‘Lisi is studying t i in her room, but I don’t know whati .’
b. ∗ Lisi yao zhaodai keren ___, danshi ta meiyou gaoshu wo shi
Lisi want serve guest but sg not-have tell sg cop
shenme.
what
‘Lisi will serve the guests t k , but she hasn’t told me whatk .’
224 Perng Wang Adams & Satoshi Tomioka

() Adjunct-wh
a. Lisi (yinwei moushi) yao lihun, dan wo bu zhidao (shi)
Lisi because something want divorce but sg not know cop
weishenme.
for-what
‘Lisi wants to get a divorce (because of something), but I don’t know why.’
b. Lisi (zai moudi) pongjian ta de laoshi, danshi wo bu
Lisi at somewhere bump-see sg de teacher but sg not
zhidao (shi) zai nali.
know cop at where
‘Lisi ran across her teacher (at some place), but I don’t know where.’

.. Impossibility of sluicing with zenme(yang) ‘how’


Another important difference between English and Chinese is the behavior of the wh-
phrases that quantify over manners or methods. While the English manner/method
wh-phrase how is a perfectly acceptable sluice, as shown in (), the Chinese counter-
part zenme(yang) ‘how’ cannot be used in the sluicing construction, as shown in ().

() a. Bart fixed the car, and we all wonder how.


b. Fred solved the problem, but he wouldn’t tell us how.

() a. ∗ Dawu xiuhao le che; women dou zai cai (shi) zenme(yang).
Dawu fix-good asp car pl all asp guess cop how
‘Dawu fixed the car, and we all wonder how.’
b. ∗ Zhangsan zhaodai keren haixian, dan women bu zhidao (shi)
Zhangsan serve guest seafood but pl not know cop
zenme(yang).
how
‘Zhangsan served the guests seafood, but we don’t know how.’

If zenmeyang ‘how’ used to inquire manners or methods cannot appear in sluicing,


is there a way for Chinese speakers to produce a grammatical sentence that has the
same meaning as its English sluicing counterpart? The sentences in () show us two
strategies without the first conjunct repeated.

() a. . . . women dou zai cai (shi) zenme(yang) xiuhao de.


pl all asp guess cop how fix-good de
‘. . . we all wonder how (it was that he fixed the car).’
b. . . . dan women bu zhidao (shi) yong shenme fangshi/taidu.
but pl not know cop use what method/attitude
‘. . . but we don’t know with what method/attitude.’
Sluicing in Mandarin Chinese 225

The sentence in (a) is derived from the cleft construction with the nominalization
of the matrix verb of the first conjunct, xiuhao ‘fix-good’, which follows zenme(yang)
‘how’. The word de in Chinese is the nominalization marker in relative clauses.
The second strategy in (b) is using a coverb, yong ‘use’, along with a complex-
argument wh-phrase that consists of the wh-phrase shenme ‘what’ and a noun that
has a meaning related to method or manner, such as fangshi ‘method’ or taidu
‘attitude’.

.. Island effects


Wh-questions have been the main domain that showcases a variety of island con-
straints on syntactic movement. In this regard, it has been particularly revealing that
those constraints are often lifted in English sluicing, as exemplified by the contrast
between () and ().

() a. Subject island


∗ Which country did [every student from t ] pass the exam?
i i
b. Adjunct island
∗ Who will Bruce be offended [if you fail to invite t ]?
j j
c. Complex NP island
∗ Which article did the professor pass [most students who read t last year]?
k k

() a. Repair of subject island


Every student from some country passed the exam, and I wonder which
country.
b. Repair of adjunct island
Bruce will be offended if you fail to invite someone, but only his wife
knows who.
c. Repair of complex NP island
Last year the professor passed most students who read some article, and the
students this year are desperate to find out which article.

For the repair of these island effects, Chinese patterns much like English, as shown
in ().

() Repair of subject island


a. Meige cong mouge guojia lai de xuesheng dou tong guo
Every-cl from some-cl country come de student all pass asp
kaoshi, wo xiang zhidao shi na yige guojia.
exam sg want know cop which one-cl country
‘Every student from some country passed the exam, and I wonder which
country.’
226 Perng Wang Adams & Satoshi Tomioka

b. Repair of adjunct island


Aqiu [yinwei Lisi zuotian mai le yijian dongxi] bu gaoxing,
Aqiu because Lisi yesterday buy asp one-cl thing not happy
danshi ta bu shuo shi shenme (dongxi).
but sg not say cop what thing
‘Aqiu was unhappy because Lisi bought something yesterday, but he didn’t
say what.’
c. Repair of complex NP island
Aqiu xiangxin [Lisi song gei shizhang mouyang tebiede liwu de
Aqiu believe Lisi give to mayor some-cl special present de
yaoyan], danshi ta bu zhidao shi shenme.
rumor but sg not know cop what
‘Aqiu believed the rumor that Lisi gave the mayor some special present, but
he does not know what.’
Wei () notes, however, that there is one important asymmetry between the two
languages under sluicing, namely, the Left Branch Condition (LBC). Both languages
impose this condition on overt movements. The examples below, for instance, demon-
strate that the LBC disallows the movement of an adjectival modifier out of a DP.
() a. ∗ How bigi did Andy buy [a t i car]?
b. ∗ Duo da dej Zhangsan mai-le [yibu t j chezi]?
How big de Zhangsan buy-asp one-cl car

‘ How big did Zhangsan buy a car?’
Interestingly, the status of the LBC under sluicing is different in the two languages.
While sluicing does not seem to repair the LBC violations in English, there are no ill
effects due to the LBC in Chinese sluicing, as shown by the contrast in ().
() a. ∗ Andy recently bought a car, but I don’t know how big.
b. Lisi zuijin mai le yiliang chezi, danshi wo bu zhidao (shi)
Lisi recently buy asp one-cl car but sg not know cop
duo da.
how big
‘Lit: Lisi recently bought a car, but I don’t know how big.’
Island effects (or the lack thereof) in sluicing are such complex phenomena that it is
beyond the scope of this paper to give a full, comprehensive analysis of them.  We
nonetheless believe that Wei’s discovery of the disappearance of the LBC in Chinese
sluicing provides an important clue for the proper analysis of Chinese sluicing.
 As have been observed by some (e.g. Wang ()), some of the island effects remain for adjunct wh-
phrases. Wang () regards the lack of island repair with adjunct wh-phrases as a piece of evidence for a
movement analysis for Chinese sluicing. However, we would like to point out that island repair does occur
even for an adjunct wh-phrase if it has an overt indefinite inner antecedent, as exemplified below.
Sluicing in Mandarin Chinese 227

.. Summary
Summing up, we have shown that Chinese sluicing is similar to English sluicing in
the following respects: (i) It cannot be uttered out of the blue; (ii) the antecedent of
the wh-remnant is indefinite; (iii) the voice in the first and second conjuncts has
to be identical; and (iv) implicit adjunct antecedents are possible. However, there
are some striking differences between the two languages. The differences include
the following: (i) It has a copula on the surface, which is required with bare argu-
ment wh-phrases ‘who’ and ‘what’, and optional with other wh-phrases; (ii) the
wh-phrase zenme(yang) ‘how’ is not a possible remnant; (iii) an implicit argu-
ment of a verb cannot serve as an antecedent for the wh-remnant; and (iv) viola-
tions of the LBC are repaired. These characteristics do not constitute the exhaus-
tive list of the relevant properties of Chinese sluicing, but we believe that they
provide a sufficient ground for our next step—to put forth a non-deletion analy-
sis of Chinese sluicing. There are, however, some important issues missing that
are often discussed in connection with sluicing and ellipsis phenomena in general.
Among them are sloppy readings for pronouns contained in the missing mate-
rials and the possibility of multiple wh-remnants in sluicing. These issues will
certainly not be ignored, and we will discuss them after spelling out the main
proposal.

. Chinese sluicing as pseudo-sluicing


.. The proposed analysis
The analysis of Chinese sluicing we endorse here is the pseudo-sluicing analysis,
proposed independently by Adams () and Wei (), in which missing material
in sluicing constructions is analyzed not as a result of ellipsis/deletion but as a phono-
logically silent pronoun (a small pro). A relevant example and illustration is repeated
in ().

(i) Yinwei Dawu de laoban renwei ta zai moushi toulan er kaichu le ta, danshi
because Dawu de boss think sg at sometime be-lazy so fire asp sg but
wo bu zhidao (na) shi shenme shihou.
sg not know that cop what time
‘[[Because Dawu’s boss thought that he was being lazy at some timei ], he fired him t j ], but I don’t
know wheni/j (that was).’
(ii) Lisi pongdao zai mouchu jiao ta xiaotiqin de laoshi, danshi wo bu zhidao
Lisi run-across at some-place teach sg violin de teacher but sg not know
(na) shi zai nali.
that cop at where
‘[Lisi ran across the teacher [who taught her violin at some placei ] t j ], but I don’t know wherei/j
(that was).’

In addition, Wei () presents more counter-examples that show that even antecedent-less adjuncts can
“escape” islands in sluicing if their interpretations are more “entity-like” (e.g. specific degrees or specific
manners).
228 Perng Wang Adams & Satoshi Tomioka

() Lisi mai le yiyang dongxi gei Dawu, dan wo bu zhidao [pro shi
Lisi buy asp one-cl thing give Dawu but sg not know cop
shenme].
what
‘Lisi bought something for Dawu, but I don’t know what it/that was.’ (= ())
Chinese is commonly recognized as a discourse pro-drop language (see Chao (),
Li and Thompson (), and Xu (), among many others), where argument
omissions are frequently observed, and the pseudo-sluicing analysis appeals to this
language-specific property. The distribution of silent arguments has some intricacies,
as discussed extensively by Huang () and, most recently, by Li () and Li and
Wei (to appear). Particularly intriguing are certain subject–object asymmetries that
suggest that not all instances of silent pronouns in Chinese are small pros. The most
controversial seems to be the status of null objects (see Li ()), and in this respect,
it is probably reasonable to assume that the sluicing structure at hand is spared from
this controversy because the missing material hypothesized to be a pro corresponds
to the subject in the _ shi XP structure.
As for the semantics of the silent anaphora, no new inventions are necessary. For
sluicing with an argument wh-remnant, the missing pronoun has meaning akin to a
pronoun that takes as its antecedent an indefinite NP that does not have the pronoun
in its scope. For such a pronoun, a variety of proposals have been made: E-type
pronouns (e.g. Evans (), Cooper (), and Heim ()), unselectively bound
anaphora (see Heim ()), and dynamically bound anaphora (e.g. Chierchia ()),
among others. In this paper, we assume, following Adams () and Wei (),
that the silent subject in Chinese sluicing is an E-type pronoun, but we acknowledge
that it is possible to adopt a different framework without affecting the spirit of the
proposal. For the mechanics of an E-type pronoun, we adopt Heim and Kratzer’s
() rendition: (i) There is an implicit definite determiner, and (ii) it also comes
with indexed anaphora of a predicate type, whose semantic content is pragmatically
recovered. This process is exemplified in (), where g is an assignment function that
supplies the value for the NP.
() a. Dawu gei mouren yishu hua, danshi wo bu zhidao pro
Dawu give some-person one-cl flower but sg not know
shi shei.
cop who
‘Dawu gave someone a bouquet, but I don’t know who (pro was).’
b. The internal structure of pro : [DP [+def] [NP e ]]
c. [DP [+def] [NP e ]]g = the unique x such that [g()](x) = .
Let g:=[ → λy. Dawu gave y a bouquet]. Then
[DP [+def] [NP e ]]g = the unique person Dawu gave a bouquet to.
Sluicing in Mandarin Chinese 229

The silent pronoun with an adjunct wh-remnant cannot be an E-type pronoun of


the kind described above. In Wei (), it is treated as an event-denoting pronoun.
This certainly covers most of the cases of sluicing with adjunct wh-remnants, but a
possible complication arises with respect to why-questions. It is still hotly debated
whether causation is a relation between eventualities or propositions (see Kratzer
()). As far as sentential anaphora is concerned, it can be used as the argument
of a predicate that clearly selects a proposition, whether it is overt or silent, as shown
below.
() a. Sam managed to pass the exam, but Ted still cannot believe it/that.
b. Zhangsan shefa tongguo le ceshi, dan Lisi haishi wufa
Zhangsan manage pass asp exam but Lisi still cannot
xiangxi pro.
believe
‘Zhangsan managed to pass the exam, but Lisi still cannot believe it/that.’
We remain neutral regarding the semantics of the silent pronoun in adjunct-sluicing
sentences, calling it a sentential pro, whose denotation can be either an event or a
proposition.
The pseudo-sluicing analysis of Chinese sluicing is a pro-form analysis that involves
no ellipsis/deletion. It makes a number of empirical predictions that would be unex-
pected under the ellipsis/deletion account, and, as we will see in the next subsection,
they are by and large confirmed by the empirical data.

.. Advantages of the pseudo-sluicing analysis


This subsection examines in detail how the pseudo-sluicing analysis explains the
properties of Chinese sluicing that we reviewed in Section .. First, recall that,
unlike its English counterpart, Chinese sluicing with an argument wh-remnant
requires the presence of an overt inner antecedent. The relevant examples are
repeated below.
() (= ())
a. ∗ Lisi zai fangjian li yandu _, danshi wo bu zhidao shi shenme.
Lisi at room in study but sg not know cop what
‘Lisi is studying t i in her room, but I don’t know whati .’
b. ∗ Lisi yao zhaodai keren _, danshi ta meiyou gaoshu wo shi
Lisi want serve guest but sg not-have tell sg cop
shenme.
what
‘Lisi will serve the guests t k , but she hasn’t told me whatk .’
This fact correlates closely with Heim’s () observation that unexpressed or implicit
arguments are not sufficient to license the use of pronouns in the subsequent dis-
course, as exemplified below.
230 Perng Wang Adams & Satoshi Tomioka

() a. Arthur married recently. And #she is very rich.


b. Bertha graduated with a degree in psychology. #It was a prestigious univer-
sity.
c. Carla is studying hard, and #it is physics.

The same constraint applies to Chinese pronouns, whether they are phonologically
overt or silent, as exemplified in ().

() a. Lisi zai fangjian li yandu _. #(Na) shi yingwen.


Lisi at room in study that cop English
‘Lisi is studying in her room. It is English.’
b. Zhangsan zuijin jiehun le. Caicai #ta duo you qian?
Zhangsan recently marry asp guess sg how have money
‘Zhangsan married recently. Guess how rich is #she?’

This explains not only the ungrammatical examples in () but also the fact that they
are still ungrammatical even when the null subjects are overtly pronounced as the
demonstrative pronoun na ‘that’, as shown below.

() a. ∗ Lisi zai fangjian li yandu _, danshi wo bu zhidao na shi


Lisi at room in study but sg not know that cop
shenme.
what
(‘∗ Lisi is studying t i in her room, but I don’t know whati that/it is.’)
b. ∗ Lisi yao zhaodai keren _, dan ta meiyou gaoshu wo na shi
Lisi want serve guest but sg not-have tell sg that cop
shenme.
what
(‘∗ Lisi will serve the guests, but she hasn’t told me what that/it was.’)

Sluicing with an adjunct wh-remnant, on the other hand, does not require an overt
inner antecedent. This is hardly surprising under the current proposal. In this type
of sluicing, the missing pronoun is an instance of sentential anaphora that refers to
an event or proposition, and any overt sentence, even without any indefinite adjunct,
can be the antecedent of such a pronoun. Some examples in English and Chinese are
shown below.

() Doris got married (some time/somewhere/for some reason). It was last year./It
was in Hong Kong./It was because she needed a green card.

() Lisi (zai moushi /zai mouchu /weile mouge liyou) jiehun le.
Lisi at some-time at somewhere for some-cl reason marry asp
‘Lisi got married (some time/somewhere/for some reason).’
Sluicing in Mandarin Chinese 231

Na shi zai  nian. /Na shi zai Meiguo. /Na shi yinwei ta
That cop at  year that cop at America that cop because sg
xuyao luka.
need green-card
‘That was in ./That was in the USA./That was because she needed a green
card.’
Consequently, Chinese sluicing does not impose the overt-antecedent requirement
on adjunct wh-remnants.
The second advantage of the current proposal is that it gives a very natural expla-
nation for the presence of the copula shi. The complication is that the copula is not
always required, and the condition of optionality or obligatoriness does not come
directly from the pseudo-sluicing structure. For this matter, we concur with Wei
(): The wh-phrases shei ‘who’ and shenme ‘what’ are the only two nominal wh-
phrases that cannot occur as predicates in isolation, so the presence of the copula
shi is required with these two wh-phrases in order to form a predicate following the
null pronoun in the verbless copulative construction. The copula in this structure
acts as an identification marker that has the features of [+verb, −adverb, −noun].
Other wh-phrases can be construed as predicates unsupported by other material,
and so the presence of shi is not obligatory. When it does appear with any of these
wh-phrases, it acts as an emphatic marker that has the features of [−verb, +adverb,
−noun].
Another important asymmetry between English and Chinese is the behav-
ior of manner/method wh-phrases. Unlike its English counterpart, the Chinese
zenme(yang) ‘how’ cannot appear in sluicing sentences, but this peculiarity is no
longer a mystery in a pseudo-sluicing analysis. Adams () observes that this wh-
phrase cannot be used in sluicing with the overt demonstrative pronoun na ‘that’. The
example sentences previously given in () are repeated here with na.
() a. ∗ Dawu xiuhao le che; women dou zai cai na (shi)
Dawu fix-good asp car pl all asp guess that cop
zenme(yang).
how
‘∗ Dawu fixed the car, and we all wonder how that was.’
b. ∗ Zhangsan zhaodai keren haixian, dan women bu zhidao na (shi)
Zhangsan serve guest seafood but pl not know that cop
zenme(yang).
how
‘∗ Zhangsan served the guests seafood, but we don’t know how that was.’
In (), both Chinese sluicing with na and its English pseudo-sluicing counterpart
with that are ungrammatical with the respective wh-remnants zenme(yang) and how.
232 Perng Wang Adams & Satoshi Tomioka

More specifically, the wh-phrase ‘how’ in the copulative clause does not function as
a manner/method modifier. This is also evident in simple conversations, as shown in
() for English and () for Chinese.
() Speaker A: Someone robbed the bank on the corner.
Speaker B: a. Who/when/where/why was that?
b. (#)How was that? [= What is your evaluation of the event?, =
How did he rob the bank?]
() Speaker A: Nadui enaide fuqi da chao le yijia.
that-cl loving couple big quarrel asp one-fight
‘That loving couple had a big fight.’
Speaker B: a. Zhende ma? (Na shi) heshi / zai nali / weishenme?
real Q that cop when at where why
‘Really? When/where/why was that?’
b. Zhende ma? (Na shi) zengme(yang) le?
real Q that cop how asp
‘Really? What happened? / #What was the manner/method of
their fighting?’
These facts seem to suggest that, unlike temporal or locative modifying wh-phrases,
the manner and method modifying wh-phrases cannot be interpreted as event
properties—at least not in the copulative construction. While we are not ready to
present any theoretical accounts for this asymmetry,  we have now found a rather
convincing reason why the manner/method wh-phrase cannot be used in Chinese
sluicing.
Island repair is another phenomenon that the proposed analysis can account for
quite naturally. In a sense, the term “repair” is not really appropriate, because there
is nothing to be repaired. There are no structures that get deleted, and no island-
violating movements are involved. Therefore, we correctly expect the lack of island
effects in Chinese sluicing, including those cases that concern the LBC. Recall that an
attributive wh-remnant with a covert antecedent is grammatical in Chinese sluicing,
whereas the comparable sluicing examples in English are not acceptable.
() (= ())
a. ∗ Andy recently bought a car, but I don’t know how big.
b. Lisi zuijin mai le yiliang chezi, danshi wo bu zhidao (shi)
Lisi recently buy asp one-cl car but sg not know cop
duo da.
how big
‘∗ Lisi recently bought a car, but I don’t know how big.’
 Wei () hypothesizes that the manner/method wh-phrase has a verbal dependency, i.e. requires a
lexical verb, and is therefore an illicit predicate in the proposed structure.
Sluicing in Mandarin Chinese 233

This contrast was puzzling because an overt movement of an attributive adjective


out of a DP is ungrammatical in both languages (see e.g. ()), but, as Wei ()
argues convincingly, it is no longer a mystery. An illicit movement did take place
in the English sluicing example, which, for some reason, cannot be repaired. In the
Chinese sluicing example, on the other hand, there was no movement involved, and
the sentence is correctly predicted to be acceptable. This point is highlighted by
the following grammatical example in English, which is the overt paraphrase of our
proposed Chinese structure.
() Andy recently bought a car, but I don’t know how big it/that was.
In Section ., we saw a number of differences between English and Chinese in their
sluicing strategies. These differences are all well accounted for under the pseudo-
sluicing analysis. 

. Challenges
Although the proposed analysis, which employs no deletion or movement, has many
virtues as we observed in the previous section, it also faces some potentially difficult
challenges. We will discuss two of these, which have previously been presented as
evidence against a pseudo-sluicing analysis of Chinese sluicing, or a pro-form analysis
in general.

.. Sloppy identity


The availability of sloppy interpretations of pronouns under sluicing has been cited as
evidence against a pro-form/pseudo-sluicing analysis (see e.g. Wang and Wu ()
for Chinese and Takahashi () for Japanese). We will show, however, that this
type of criticism is misplaced, and that the distribution of sloppy identity in Chinese
sluicing is better predicted under the current proposal.
Sluicing in English does not yield sloppy readings as often as VP-ellipsis does. As
Merchant (: ) points out, for many native speakers of English, sloppy-identity
readings are “highly inaccessible” in sluicing sentences. Yet, in certain elliptical con-
texts, sloppy readings are available. Here is one such example:
 Wei () presents another advantage of the pseudo-sluicing analysis. Chinese sluicing can have
aggressively non-d-linked wh-phrases, while they cannot appear in English sluices (Merchant ). Chi-
nese sluicing patterns with the pseudo-sluicing structure of English, as shown below.

(i) Someone dented my car, and I will find out who the hell ∗ (it was)!
(ii) Mouren ba wo de che zhuang le yige dong, wo yao diaocha na/Ø daodi
Someone ba sg de car hit asp one-cl hole sg want find-out that the-hell
∗ (shi) shei.
cop who
‘Someone dented my car; I will find out who the hell it was!’
234 Perng Wang Adams & Satoshi Tomioka

() Johni wondered which novel hei should buy, and Billj wondered which dictio-
nary.
⇒ Strict reading: . . . and Billj wondered which dictionary hei (= John)
should buy.
⇒ Sloppy reading: . . . and Billj wondered which dictionary hej (= Bill)
should buy.

However, the Chinese counterpart of () behaves in a dramatically different way. As


(a) shows, directly translating () into Chinese produces an unacceptable sentence.
Thus, neither the sloppy-identity reading nor the strict reading is available. Wei ()
presents an example that is much more acceptable, (b), which still lacks the sloppy
interpretation, contrary to what one might expect from the English sentence ().
() a. ∗ Yuehan xiang zhidao ta yinggai mai na yiben xiaoshuo,
John want know sg should buy which one-cl novel
(erqie) Bier xiang zhidao (shi) na yiben zidian.
and Bill want know cop which one-cl dictionary
(‘∗ John wondered which novel he should buy, and Bill wondered which
dictionary it/that was.’)
b. Yuehan zhidao (ta)ziji you yige jisui de xiaohai,
John know (sg)self have one-cl how-many-age de child
danshi Bier bu zhidao (shi) jisui.
but Bill not know cop how-many-age
‘Johni knows how old a child hei has, but Billj doesn’t know how old (a child
hei/∗j has).’
Adams () concludes that sloppy readings are unavailable in Chinese sluicing,
and further claims that it follows from her pro-form analysis, demonstrating that
the paraphrasing of Chinese sluicing sentences with the overt pronoun na ‘that’ lacks
the sloppy readings as well. However, the availability of sloppy readings in Chinese
sluicing turns out to be far more intricate than portrayed in Adams (). Wei ()
and Wang and Wu (), for instance, report that sluicing with adjunct wh-phrases
such as weishenme ‘why’ and zai nali ‘at where’ can yield sloppy readings, as shown
in (). 
() a. Dawu zhidao ta/ziji zai nali chuchehuo, Aqiu ye zhidao
Dawu know sg/self at where have-car-accident Aqiu also know
(shi) zai nali.
cop at where
‘Dawui knows where hei had a car accident, and Aqiuj also knows where
(hei/j had a car accident).’
 Our informants point out that if the pronoun’s referent is ‘Dawu’, they would use ziji intead of ta. Also,
adding the copula shi makes it more difficult to obtain the sloppy interpretation. Furthermore, they prefer
to end the sentences at the matrix verb, zhidao ‘know’, when sloppy readings are concerned.
Sluicing in Mandarin Chinese 235

b. Dawu zhidao ta/ziji weishenme bei ma, Aqiu ye zhidao (shi)


Dawu know sg/self why pass scold Aqiu also know cop
weishenme.
why
‘Dawui knows why hei was scolded, and Aqiuj also knows why (hei/j was
scolded).’
The emerging picture is rather complex. Sloppy readings are possible in some
instances of Chinese sluicing, but the distribution is actually quite limited.  While
there are some debatable examples that not all native speakers of Chinese agree on,
the following are valid generalizations:
() a. Sloppy readings are easier to obtain with adjunct wh-remnants than argu-
ment wh-remnants.
b. Paraphrasing sluiced sentences with an overt pronoun na ‘that’ rarely pro-
duces sloppy interpretations.
What are the theoretical underpinnings of these generalizations? While they do
present a difficult challenge to our pseudo-sluicing analysis, they are not as ruinous as
portrayed by the “movement-and-deletion” advocates. As a matter of fact, we believe
that (a) and (b) are possibly more problematic for deletion analyses. If one is com-
mitted to the thesis that missing materials can have sloppy readings if and only if what
is missing is the result of deletion/ellipsis, then the natural conclusion is a “mixed”
theory: Chinese sluicing with an adjunct wh-phrase is an instance of deletion/ellipsis,
while argument wh-questions do not feed into a deletion/ellipsis structure. Although
this is a logical possibility, we do not see an independent motivation for such a
dichotomy. It is also not easy to account for the lack of sloppy identity under sluicing
with argument wh-remnants within a deletion/ellipsis account.
Furthermore, while sloppy readings are often used as evidence against a pro-form
analysis for missing constituents, this argument is based on the erroneous assump-
tion that pronouns never yield sloppy readings. The “paycheck pronoun” is the most
famous case where pronouns can have sloppy readings, as shown in ().
() The womani whoi deposited heri paycheck in the bank is wiser than the
womanj whoj deposited it in the Brown University Employees’ Credit Union.
(Jacobson :  ())
The pronoun it is understood as ‘herj paycheck’, where the contained pronoun her is
bound by the local relative operator. Despite this fairly well-known case of an overt

 Wei () points out two important restrictions on the availability of sloppy identity under Chinese
sluicing. First, the antecedent in the first conjunct has to be a lexically identical wh-phrase for the wh-
remnant. Second, the pronoun that renders the sloppy identity has to be properly c-commanded by its
antecedent.
236 Perng Wang Adams & Satoshi Tomioka

pronoun with a sloppy interpretation, the idea of “sloppy reading iff deletion/ellipsis”
is widespread, and we speculate that it may have its root in cases like ().

() a. Ana bought a bouquet for her mother, and Bella bought a box of chocolates
for her. [her = Ana’s mother, = Bella’s mother]
b. Carl thinks his diploma is genuine, but David thinks it is a fake. [it = Carl’s
diploma, = David’s diploma]

The exact condition that governs the availability of sloppy readings is not well under-
stood, but to the extent that typical ellipsis examples have conjunctive structures (with
and or but), the proponents of the ellipsis account may think it reasonable to use
the lack of sloppy readings in () as evidence against the pro-form/pseudo-sluicing
analysis.
We believe, however, that this reasoning fails to capture the two generalizations
stated above. The first generalization (= (a)), which concerns the argument–adjunct
asymmetry, can translate into the following generalization in the context of the pro-
form analysis: Sentential anaphora (either event anaphora or propositional anaphora)
yields sloppy readings more easily than individual anaphora (of type e). Notice that
the examples in () are both individual-denoting pronouns, and the lack of sloppy
readings correlates with the difficulty of eliciting sloppy readings with argument wh-
remnants in sluicing. It is also highly significant that sentential anaphora involving
it or that seems to permit sloppy readings more easily than non-sentential anaphora,
even in conjunctive environments, as attested in the following examples from English
(cf. Jacobson ()).

() a. Katei was certain that heri kids were secretly smoking, but Lucy j , being
oblivious, never suspected it. [it = that heri/j kids were secretly smoking]
b. Both Katei and Lucy j are very understanding, but Katei would still mind
if heri daughter became a single mother while Lucy j wouldn’t mind it/that
so much. (it/that = if heri/j daughter became a single mother)

Therefore, the first generalization is more in line with our pro-form analysis than with
its ellipsis-based competitors.
The second generalization is about the paraphrasing of sluiced sentences with the
demonstrative pronoun na ‘that’. This generalization can also be morphed into a
slightly different form: Silent pronouns are more compatible with sloppy readings than
overt pronouns are. This trend has been noted, independently of the issue at hand, for
Japanese pronouns by Kurafuji (), and Chinese also shows the same tendency, as
shown below.

() a. Zhangsan xihuan tade laoshi. Lisi ye xihuan ta.


Zhangsan like his teacher Lisi also like sg
‘Zhangsan likes his teacher, and Lisi also likes her.’
Sluicing in Mandarin Chinese 237

b. Zhangsan xihuan tade laoshi. Lisi ye xihuan.


Zhangsan like his teacher Lisi also like
‘Zhangsan likes his teacher, and Lisi also does.’
In (a), an overt personal pronoun ta is used as the object of the second clause; this
pronoun can only be interpreted as the object of the first conjunct, ‘Zhangsan’s teacher’,
and not as ‘Lisi’s teacher’. In contrast, (b) employs a null object, and both the strict
reading and the sloppy reading are available.
Thus, the pseudo-sluicing analysis of Chinese sluicing correctly predicts the gen-
eral propensity of the distribution of sloppy readings. Adjunct wh-remnants, which
come with silent sentential pronouns, are more compatible with sloppy readings than
argument wh-remnants are. This is because the empty pronouns with argument wh-
remnants are of type e. While the presence of a sloppy reading in a sluiced sentence
does not mean that the corresponding na ‘that’ paraphrase has the same reading, this
is simply because null pronouns yield sloppy readings more easily than overt ones
do. While we are yet to provide a complete analysis of sloppy readings in Chinese
sluicing, which requires a better understanding of anaphora with sloppy interpreta-
tions in general, we conclude that the current proposal actually fares better, perhaps
surprisingly, than its ellipsis/deletion competitors in dealing with sloppy readings in
Chinese sluicing.

.. Multiple sluicing


The availability of multiple wh-remnants is another phenomenon that has been pre-
sented as a challenge to a pro-form analysis of Chinese sluicing. We argue, how-
ever, that sluicing with multiple wh-remnants in Chinese is an instance of conjoined
pseudo-sluiced sentences.  First of all, the restrictions on the appearance of the copula
shi are identical in sluicing with single and multiple wh-remnants: Its presence is
obligatory for each simplex-argument wh-remnant, but is optional for other wh-
phrases (although it sounds better if shi appears with each wh-phrase), as shown
in ().
() a. Mouren tou le tade yiyang dongxi, wo xiang zhidao ∗ (shi)
someone steal asp his one-cl thing sg want know cop
shei ∗ (shi) shenme.
who cop what
Lit. ‘Someone stole one of his belongings, and I wonder who what.’
b. Laoshi chufa le mouren, wo xiang zhidao ∗ (shi) shei (shi)
teacher punish asp someone sg want know cop who cop
weishenme.
why
Lit. ‘Teacher punished someone, and I wonder who why.’
 This possibility has been mentioned in Wei (: Section ..) as well as in Chiu (: ), who
uses multiple sluicing as evidence for a movement analysis of Chinese sluicing.
238 Perng Wang Adams & Satoshi Tomioka

In Chinese, multiple sentences that are juxtaposed without overt conjunctive markers
can be understood to be conjoined sentences, and such an interpretation is possi-
ble even under embedding with an attitude verb, whether an embedded sentence is
declarative or interrogative.

() a. Wo zhidao Dawu shi Beijing ren, Lisi shi Taipei ren.
sg know Dawu cop Beijing person Lisi cop Taipei person
‘I know Dawu is from Beijing, (and) Lisi is from Taipei.’
b. Wo xiang zhidao naxie xuesheng shi cong Beijing lai de,
sg want know which-some student cop from Beijing come de
naxie xuesheng shi cong Taipei lai de.
which-some student cop from Taipei come de
‘I wonder which students are from Beijing, (and) which students are from
Taipei.’

Therefore it is quite reasonable to consider the embedded sentences in () to be


implicitly conjoined. Not surprisingly, the overt conjunctive marker han/he/yiji ‘and’
can be inserted in any of the sentences in () or (). The pseudo-sluicing analysis
predicts that sluicing with multiple identical wh-remnants would be unacceptable:
Each wh-remnant would be associated with an empty pronoun, and if there are iden-
tical wh-phrases, none of the pronouns could be uniquely identified. This prediction
is borne out, as exemplified below.

() a. ∗ Laoshi shuo mouren da le mouren, wo xiang zhidao shi shei


Teacher say someone hit asp someone sg want know cop who
shi shei.
cop who
(‘∗ Teacher said that someone hit someone, but I wonder who (it was),
who(m) (it was).’)
b. ∗ Laoshi yinwei Lisi shoushang chufa mouren, wo xiang zhidao
Teacher because Lisi get-hurt punish someone sg want know
zai heshi zai heshi.
at when at when
(‘∗ Teacher punished someone because Lisi got hurt, but I wonder when (it
was), when (it was).’)

Chiu (), an advocate of a movement account, lists the following restrictions


on multiple sluicing, which cannot be easily explained under the pseudo-sluicing
analysis.

() (= Chiu (: ()))


a. Multiple sluicing does not allow more than one wh-argument, whereas no
such restriction applies to wh-adjuncts.
Sluicing in Mandarin Chinese 239

b. Wh-adjuncts cannot precede any wh-arguments.


c. The antecedents of any two wh-remnants must be clausemates with each
other.
However, these restrictions are either misanalyzed or far less robust than we normally
expect from grammatical constraints. First, we have already seen that (a) only
applies to two arguments that are identical (= our examples in ()), and our pro-
form account is perfectly capable of explaining this pattern. Chiu’s second restriction
is not completely accurate either: An adjunct wh-remnant with an overt indefinite
antecedent can actually precede an argument wh-remnant, as exemplified below.
() a. Zhangsan zai mouchu da le mouren, wo xiang zhidao (shi)
Zhangsan at some-place hit asp someone sg want know cop
zai nali ∗ (shi) shei.
at where cop who
Lit. ‘Zhangsan hit someone somewhere, I wonder where (it was), whom (it
was).’
b. Zhangsan dao mouge difang chu jie yiwei dashi,
Zhangsan to some-cl place go pick-up one-cl ambassador
danshi wo bu zhidao (shi) dao nali ∗ (shi) na yiwei
but sg not know cop to where cop which one-cl
dashi.
ambassador
Lit. ‘Zhangsan went somewhere to pick up an ambassador, but I don’t know
where (it was), which ambassador (it was).’
c. Zhangsan zai moushi chu mai yiyang ta hen xihuan de
Zhangsan at some-time go buy one-cl sg very like de
dongxi, danshi wo bu zhidao (shi) zai heshi ∗ (shi) shenme
thing but sg not know cop at when cop what
dongxi.
thing
Lit. ‘Zhangsan went to buy something he really liked at some time, but I don’t
know when (it was), what thing (it was).’
In fact, native speakers of Chinese are more willing to accept the pattern in which an
antecedent-less adjunct wh-remnant precedes an argument wh-remnant if the copula
is present in front of the adjunct wh-remnant, as shown below.
() Lisi (weile yige yuanyin) da le mouren, dan wo bu zhidao shi
Lisi for one-cl reason hit asp someone but sg not know cop
weishenme shi shei.
why cop who
Lit. ‘Lisi hit someone (for some reason), but I don’t know why who.’
240 Perng Wang Adams & Satoshi Tomioka

The clausemate condition in (c) can also be violated. Specifically, when the
antecedents of the wh-remnants are made explicit, they need not be clausemates with
each other.
() a. Mouren gaosu Zhangsan [xuexiao mouchu you yichang
someone tell Zhangsan school some-place have one-cl
yanjiang], dan wo bu zhidao ∗ (shi) shei (shi) zai nali.
speech but sg not know cop who cop at where
‘Someone told Zhangsan that there was a speech at some place at school, but
I don’t know who (it/that was), where (that was).’
b. Zhangsan zai moushi tingdao tade taitai song Lisi mouyang
Zhangsan at some-time hear his wife give Lisi some-cl
dongxi, dan wo bu zhidao (shi) zai heshi ∗ (shi) shenme dongxi.
thing but sg not know cop at when cop what thing
Lit. ‘At some (particular) time Zhangsan heard that his wife gave Lisi some-
thing, but I don’t know when (that was), what (it/that was).’
Consequently, Chiu’s observed restrictions on sluicing in Chinese can be reduced
to the following two factors: (i) the presence vs absence of an overt antecedent for
an adjunct wh-remnant, and (ii) the presence vs absence of the copula shi for an
adjunct wh-remnant. While these properties and their effects on the acceptability of
Chinese sluicing present intriguing puzzles, they do not seem to favor either of the
two types of analysis. If anything, the pro-form analysis has a slight edge, because it is
not unreasonable to suppose that the first restriction is related to the salience and/or
recoverability of an antecedent for the empty pronoun.
In sum, we have shown that multiple-sluicing data are, to a large extent, consistent
with the predictions of the pseudo-sluicing account. The phenomenon has been cited
as supporting evidence for a movement analysis, but the generalizations used for
the argument turn out to be inaccurate and are possibly more problematic for the
movement analysis than for the pro-form analysis.

. Comparison with alternative analyses


We have pointed out a number of advantages of the pseudo-sluicing analysis for
Chinese and attempted to account for some challenges that are arguably problem-
atic for the proposal. In this section, we will critically review the two non-pro-form
approaches and argue that neither of them presents itself as a viable alternative to the
pseudo-sluicing analysis.

.. A focus movement account


The closest parallel to the English-style “move-and-delete” account is found in
Wang and Wu (). Since Chinese lacks overt wh-movement, a move-and-delete
Sluicing in Mandarin Chinese 241

approach requires that a comparable surface movement be found, and Wang and Wu
identify it as focus movement—the fronting operation of a focused constituent to the
specifier of a focus phrase (FP). An FP is suggested to be projected above an IP, making
the IP deletable after the movement. This process is illustrated in the example below
(simplified from Wang and Wu (: (–)).

() Lisi yujian yige ren, keshi wo bu zhidao [CP shi


Lisi meet one-cl person but sg not know cop

[FocP sheii Foc [IP Lisi yujian t i ]]].
who Lisi meet
‘Lisi met someone, but I don’t know who.’

In the previous section, we reviewed two key phenomena that have been used to
motivate a move-and-delete approach to Chinese sluicing, and our conclusion was
that they do not actually threaten the pseudo-sluicing analysis in any serious way.
Moreover, a number of troubling aspects of focus movement have been pointed
out, most notably by Li and Wei (to appear). We summarize their crucial points as
follows.

() a. The presence of the copula shi—obligatory with ‘who’ and ‘what’ but optional
with other wh-phrases—is not properly explained.
b. Unlike in English, even when syntactic islands are not involved, focus move-
ment of wh-phrases, especially directional prepositional phrases like dao-
nali ‘to where’ and quantity-denoting phrases like duoshan-qian ‘how much’,
creates ungrammatical or deviant sentences before IP-deletion.
c. The ungrammaticality of the wh-phrase zenme(yang) ‘how’ in Chinese sluic-
ing is not properly explained. 
d. Unlike its English counterpart, Chinese sluicing is not constrained by Max-
Elide (see Merchant ()), which provides that if ellipsis targets an XP
containing an A -trace, XP must not be properly contained in any YP that is
a possible target for deletion. The contrast between the two languages would
be left unaccounted for if they employed the same strategy for sluicing.

 Wang and Wu claim that the movement proposal actually rules out the wh-phrase zenmeyang ‘how’
as a wh-remnant in Chinese sluicing, as fronting is a prerequisite in this analysis, and zenmeyang does
not permit fronting. However, their proposal cannot predict the ungrammaticality of the similar wh-
phrase zenme ‘how’ in sluicing. As shown below, the fronting of this wh-phrase is licit, but zenme is also
ungrammatical in sluicing.
(i) Zenme Laowu xiuru Lisi (ne)?
how Laowu insult Lisi Q
‘How come Laowu insulted Lisi?’

In fact, according to the reasoning in Wang and Wu (), many grammatical sluicing sentences should
be unacceptable, since the fronting of a wh-phrase would actually render such sentences ungrammatical
before deletion.
242 Perng Wang Adams & Satoshi Tomioka

e. In the Chinese passive construction, the head bei takes the whole clause
following it as its complement in the fronting process, not just the nomi-
nal phrase that follows it. The focus-movement approach can be shown to
predict the opposite.
We will not go into the details of Li and Wei’s arguments, to which we adhere. We
recommend the reader to consult Li and Wei (to appear: Section .) for further
discussions and concrete examples.
While we find the evidence presented by Li and Wei convincing enough to discredit
the move-and-delete account of Wang and Wu (), we would like to present an
additional fact that is arguably problematic for the focus-movement approach. Recall
that the distribution of sloppy readings is more limited in Chinese than in English, and
some of the illuminating examples were presented in (). However, as shown in (),
sentence (a) becomes grammatical when a focus-moved structure is pronounced
overtly in the second conjunct. Moreover, the sentence does have the sloppy reading.
As a matter of fact, the sloppy reading seems preferred to the strict reading.
() Yuehan xiang zhidao na yiben xiaoshuo ta yinggai mai, (erqie)
John want know which one-cl novel sg should buy and
Bier xiang zhidao (shi) na yiben zidian ta yinggai mai.
Bill want know cop which one-cl dictionary sg should buy.
‘John wondered which novel he should buy, and Bill wondered which dictionary
he should buy.’
Under the focus-movement analysis, the sentence in () should create the appro-
priate input for the deletion operation, which would result in (a). Thus, the two
sentences are expected to pattern alike, both in terms of their acceptability and the
possibility of a sloppy interpretation, contrary to fact. In addition, the input for the
deletion operation for (b) becomes ungrammatical and nonsensical when we front
the wh-phrase, jisui ‘how-many-age’, as shown in ().
() ∗ Yuehan zhidao jisui (ta)ziji you yige de xiaohai,
John know how-many-age (sg)self have one-cl de child
danshi Bier bu zhidao (shi) jisui (ta)ziji you yige de
but Bill not know cop how-many-age (sg)self have one-cl de
xiaohai.
child
In sum, we believe that enough evidence has been accumulated to conclude that the
move-and-delete approach based on focus movement is untenable.

.. Reduced-cleft analysis


Another alternative we can consider is the elliptical/reduced-cleft analysis that has
garnered tremendous popularity among the Japanese researchers (see Shimoyama
Sluicing in Mandarin Chinese 243

(), Nishiyama (), Kizu (), Fukaya and Hoji (), Hiraiwa and Ishihara
(), and Fukaya (), among many others). Considering the many typological
similarities that exist in the syntax of wh-interrogatives in Japanese and Chinese,
this analysis certainly deserves some close examination. The derivation for a sluiced
sentence in Japanese in a reduced-cleft analysis is shown in ().
() Dareka-ga sono hon-o yon-da ga, watashi-wa [CP Opi
Someone-nom that book-acc read-pst but sg-top
[IP _i sono hon-o yon-da] no] -wa [CP [IP dare (da)] ka]
that book-acc read-pst nmlz -top who cop Q
wakaranai.
know-not
‘Someone read that book, but I don’t know who [IP (it is)[CP that read that
book]].’
However appropriate this analysis may be for the comparable construction in
Japanese, extending it to Chinese is not straightforward. Indeed, it faces a number of
challenges that we believe are very difficult to overcome. The examples in () below
are instances of cleft sentences in Chinese. To form a cleft sentence in Chinese, the
copula shi is inserted in front of the focus phrase, which in (a–c) has the function
of a time adverbial, the subject, and a subject-oriented manner adverbial, respectively.
As seen in (d), it is not possible for shi to immediately precede and focus the object.
() a. ∗ (Shi) [Lisi]Focus zuotian zai shudian li mai le yiben
cop Lisi yesterday at bookstore in buy asp one-cl
xiaoshuo (de).
novel de
‘It is Lisi who bought a novel in the bookstore yesterday.’
b. Lisi ∗ (shi) [zuotian]Focus zai shudian li mai le yiben xiaoshuo
Lisi cop yesterday at bookstore in buy asp one-cl novel
(de).
de
‘It is yesterday that Lisi bought a novel in the bookstore.’
c. Lisi zuotian ∗ (shi) [xinggaocailiede]Focus mai le yiben xiaoshuo
Lisi yesterday cop high-spirited buy asp one-cl novel
(de).
de
‘It is with a high, happy spirit that Lisi bought a novel yesterday.’
d. ∗ Lisi zuotian zai shudian li mai le shi [yiben xiaoshuo]Focus
Lisi yesterday at bookstore in buy asp cop one-cl novel
(de).
de
(‘It is a novel that Lisi bought in the bookstore yesterday.’)
244 Perng Wang Adams & Satoshi Tomioka

One of the most notable aspects of these examples is the obligatory presence of the
copula. While the copula shi in () may give the impression that there is a close
correlation between the sluicing and cleft constructions, its distributions are quite
different in the two environments. As we mentioned earlier, the copula is obligatory in
Chinese sluicing only with bare argument wh-remnants, but is optional with the other
wh-remnants. However, its presence is required in all instances of cleft sentences,
regardless of the categorial type of the focused phrase; for example, the temporal
expression in (b) and the manner/method expression in (c) both require shi. This
discrepancy would be hard to explain if a clefted sentence were indeed the input for a
deletion operation producing sluiced sentences.
The second issue is the focused manner/method expression in (c). This sen-
tence contrasts sharply with attempted sluicing sentences with the manner wh-phrase
zenme(yang) ‘how’ as its remnant, which is uniformly rejected by Chinese speakers.
() (= ())
a. ∗ Dawu xiuhao le che; women dou zai cai (shi) zenme(yang).
Dawu fix-good asp car pl all asp guess cop how
‘Dawu fixed the car, and we all wonder how.’
b. ∗ Zhangsan zhaodai keren haixian, dan women bu zhidao (shi)
Zhangsan serve guest seafood but pl not know cop
zenme(yang).
how
‘Zhangsan served the guests seafood, but we don’t know how.’
A reduced-cleft analysis of sluicing in Chinese does not seem to offer any natural
explanation for this contrast.
Another problem in potentially trying to claim that Chinese sluicing sentences are
derived from cleft sentences is that the object of a verb cannot be clefted in Chinese, 
as shown in (d). This conflicts with the fact that the object of a verb is a legitimate
remnant in sluicing, as shown below.
() Lisi zuotian zai shudian li mai le yiyang dongxi, dan wo bu
Lisi yesterday at bookstore in buy asp one-cl thing but sg not
zhidao shi shenme.
know cop what
‘Lisi bought something in the bookstore yesterday, but I don’t know what (it
was).’
 Wei () also points out that some sluicing sentences cannot be clefted, as shown below.

(i) ∗ Cai yue-lai-yue gui le, dan wo bu qingchu [shi duoshao-qian


vegetables more-and-more pricy asp but sg not clear cop how-much
cai yue-lai-yue gui le].
vegetables more-and-more pricy asp
(‘Vegetables are getting more and more expensive, but I am not clear how much they are.’)
Sluicing in Mandarin Chinese 245

Chinese also has another construction not too dissimilar to the cleft structure, which
is often referred to as a “pseudo-cleft”. This construction has a structure in which the
copula shi connects a predicate and an NP subject modified by a relative clause, as
shown in ().

() [IP [DP Na-ge [NP [RC zuotian mai le xiaoshuo de] ren] shi
that-cl yesterday buy asp novel de person cop
[DP Lisi]].
Lisi
‘The person who bought a novel yesterday was Lisi.’

It can be noted that the idea of identifying a pseudo-cleft sentence as the input
of a sluiced sentence faces exactly the same set of problems as an elliptical cleft
analysis does: (i) The copula shi is obligatory in all cases, including those with wh-
remnants that are not bare arguments; (ii) the manner/method wh-phrase is incor-
rectly expected to be a legitimate sluice, as shown in (a); and (iii) some focused
phrases, such as indirect object PPs, cannot be pseudo-clefted, as shown in (b).

() a. Aqiu han Amei jiaowang de taidu shi hen renzhende.


Aqiu with Amei date de attitude cop very seriously
‘The attitude with which Aqiu and Amei dated each other was very serious.’
b. ∗ Aqiu song yizhi zuanshi jiezhi de ren shi gei Amei.
Aqiu give one-cl diamond ring de person cop to Amei
(‘The person that Aqiu gave a diamond ring to was Amei.’)

Finally, just as was the case with the focus movement analysis, an elliptical pseudo-
cleft approach also makes an incorrect prediction with regard to the availability of
sloppy readings in, and the acceptability of, examples such as (a). A full pseudo-cleft
counterpart of (a) shown below is grammatical and has a sloppy reading, whereas
(a) does not permit a sloppy reading and is actually ungrammatical. This clearly
suggests that pseudo-clefts are not the input to sluicing structures. Note also that the
copula is obligatory in pseudo-clefts such as (), but is optional in slucing involving
similarly complex wh-phrases, a further unexpected contrast for any pseudo-cleft
approach to Chinese sluicing.

() Yuehan xiang zhidao ta yinggai mai de shi na yiben xiaoshuo,


John want know sg should buy de cop which one-cl novel
(erqie) Bier xiang zhidao ta yinggai mai de shi na yiben
and Bill want know sg should buy de cop which one-cl
zidian.
dictionary
‘Johni wondered which novel hei should buy, and Billj wondered which dictio-
nary hei/j should buy.’
246 Perng Wang Adams & Satoshi Tomioka

In sum, then, while the elliptical cleft/pseudo-cleft structure has been the most promi-
nent analysis for sluicing in Japanese, the extension of this analysis to Chinese faces
many problems and cannot be easily implemented.

.. Summary
Prior to this section, we reviewed two problems for the pro-form analysis of Chinese
sluicing, namely the patterning of sloppy interpretations and multiple sluicing. While
competitors of the pseudo-sluicing analysis have claimed that such properties cannot
be accounted for under a pseudo-sluicing analysis that involves no ellipsis, we argue
that these phenomena, though not completely explained, can indeed be accommo-
dated within a pseudo-sluicing analysis, and probably in a more natural way than
in the other approaches. With the disappearance of such apparent advantages, the
competitor analyses to the pseudo-sluicing approach also lose their appeal. Along
with Li and Wei (to appear), we have also added more evidence in pointing out the
major problems that exist for a focus movement analysis. In addition, in line with
Adams () and Wei (, ), we have clarified that Chinese sluicing is unlikely
to be derived from reduced clefts/pseudo-clefts. The list of problems provided here is
not intended to be exhaustive, but we believe that enough evidence has been collected
and presented to cast serious doubts on both focus-movement and (pseudo-)cleft
analyses of Chinese sluices.

. Conclusion
In this chapter, we have investigated apparent cases of sluicing in Chinese and found
further support for the proposals made by Adams () and Wei (). Chinese
sluicing should be analyzed as pseudo-sluicing, in which the missing material is rep-
resented by a phonologically silent pro-form. The conclusions reported here are not
particularly surprising given the typological differences between Chinese and English
in wh-questions. English, a wh-movement language without pro-drop, is expected to
differ from Chinese, a wh-in-situ language with robust pro-drop. While focus move-
ment in Chinese may appear to perform the same function as overt wh-movement in
English, the empirical evidence is not in favor of drawing a direct analogy between
Chinese and English sluicing.
More intriguing is the micro-variation between Japanese and Chinese, both of
which are wh-in-situ and pro-drop languages. The former employs elliptical clefts
to derive apparent sluicing, while the latter derives it through a simpler copula-
tive clause. The reason for this difference is still not clear to us, but Fukaya and
Hoji’s () “mixed” theory shortens the distance between the two languages: They
argue that sluiced sentences with case-marked wh-remnants must have elliptical cleft
Sluicing in Mandarin Chinese 247

structures as their input, while those with “bare” non-case-marked wh-remnants


should be analyzed as pseudo-sluicing cases. However, the interesting question still
remains of why Chinese cannot make use of cleft or pseudo-cleft sentences to build
sluicing structures, and we must wait for a future occasion to tackle this intriguing
puzzle.


Sluicing in Turkish*
ATA KA N İ NC E

. Introduction
This chapter argues that, contrary to what has been suggested in Kuwabara () and
Kizu (), sluicing is possible in wh-in-situ languages such as Turkish. Furthermore,
sluicing structures in Turkish are argued to be derived in the same way as they are
derived in English. However, in addition to wh-features, focus features are operative
in sluicing in Turkish. This view contrasts with proposals in Kuwabara () and
Kizu () that sluicing structures in wh-in-situ languages are really elliptical cleft
constructions.
The chapter also considers the phenomenon of case mismatch in Turkish sluicing
structures, which was first noted by Merchant (). In Turkish, the (wh-/non-wh-)
subject of an embedded clause bears genitive case.  When sluiced, however, the
subject must bear nominative case. The chapter presents an analysis of this phe-
nomenon that utilizes a modified version of Hiraiwa’s () account of nominative–
genitive conversion in Japanese and the Multiple Spell-Out model of Minimalism
(Uriagereka , Chomsky , ). In such a model it is assumed that the basic
operation in narrow syntax is Merge, and that what is sent to the interface levels,
LF and PF, is not the whole structure of a derivation, but rather chunks of structure
that have been formed up to specific derivational stages, the domains of phases (see
Section ..).

. Sluicing
Turkish is an agglutinating wh-in-situ SOV language. As shown in (), the unmarked
SOV word order of Turkish is preserved in questions. The wh-phrase kimle is in
∗ I am grateful to Ilhan Cagri, Howard Lasnik, Jason Merchant, Andrew Simpson, and the two reviewers
for their invaluable assistance and comments.
 Unless the subject is non-specific, in which case there is evidence that it remains in situ and is bare.
See Kennelly (), and especially Cagri ().
Sluicing in Turkish 249

its case-checking position and has not overtly moved to a position preceding the
subject-NP Ahmet. Compare the Turkish word order with its English equivalent,
where the wh-phrase who must precede the subject. 

() Ahmet-∅ kim-le konuş-uyor-∅?


Ahmet-nom who-com talk-prog-sg
‘Who is Ahmet talking to?’

In Kuwabara () and Kizu () it has been argued that sluicing is not possible in
wh-in-situ languages because sluicing is defined as TP-deletion, leaving behind only a
CP remnant where the moved wh-element is pronounced (in SpecCP). Nevertheless,
sluicing structures do regularly occur in Turkish, as illustrated in ():

() Hasan-∅ biri-yle konuş-uyor-∅; ama kim-le bil-mi-yor-um.


Hasan-nom one-com talk-prog-sg but who-com know-neg-prog-sg
‘Hasan is talking to someone; but I don’t know who with.’

The non-sluiced version of () appears in ().

() Hasan-∅ biri-yle konuş-uyor-∅; ama [kim-le konuş-uyor-∅]


Hasan-nom one-com talk-prog-sg but pro who-com talk-prog-sg
bil-mi-yor-um.
know-neg-prog-sg
‘Hasan is talking to someone; but I don’t know who he is talking to.’

The elided phrases in the sluiced version of () are struck through in the representa-
tion ().

() . . . Hasan kimle konuşuyor . . .

The problem here is that no theory of ellipsis can naturally explain how an inter-
mediate position in a string can be pronounced while phrases before and after this
position are elided. For example, in (), if we assume the subject-NP Hasan is in
SpecTP, then TP-deletion would result in deletion of the wh-element kimle along
with the VP. This is illustrated in () and (). In the derivation in (), the subject-
NP Ahmet raises to SpecTP, the verbal amalgam konuşuyor ‘is talking’ moves to T, and
the object wh-phrase kimle ‘with who’ raises to the specifier of vP, where it checks case
features. 
 A wh-phrase can be scrambled to pre-subject position, but this is orthogonal to the account at hand:

(i) Kim-lei Hasan t i konuşuyor?


who-com Hasan is.talking
‘Who is Hasan talking with?’
 In İnce (), I assume that Turkish is an overt-V-raising language, following Kural () and Aygen
().
250 Atakan İnce

() TP
2
Ahmetj 2
vP T-v-is.talkingi
2
with.whok 2
tj 2
VP v-ti⬘
2
NP V
tk ti

In (), both the lexical verb and the subject-NP are in positions higher than the
wh-phrase kimle ‘with who’. They cannot be elided when kimle is in a lower posi-
tion because kimle would be deleted as well, as the unacceptable structure in ()
demonstrates.
() * aP
2
Ahmet bP
2
gP 
konusuyor
2 “is.talking”
kimle
“with.who”

In order to account for the well-formedness of (), we must therefore assume that
kimle raises to a position higher than TP, as in ().

() CP
2
with.whoi C⬘
2
TP C
2
Ahmet T⬘
2
vP T
2 is.talking
2
ti ...

In investigating the general phenomenon of sluicing in Turkish, the following the-


oretical assumptions are made use of here. I adopt the copy theory of movement
(Chomsky and Lasnik (), among others), which takes place in Narrow Syntax,
and there is no post-syntactic LF movement. In this theory a lower element is copied
Sluicing in Turkish 251

and remerged in a higher position in the structure. Both such copies remain in the
structure, but only one is marked for pronunciation. In instances where the lower copy
is pronounced, the output is referred to as “covert movement”, and when the highest
copy is pronounced, the term “overt movement” is applied. Whether a movement is
overt or covert depends on feature strength. If the attracting feature is strong, the
head of the chain formed by copy and remerge is pronounced. On the other hand, if
the attracting feature is weak, the lower copy is pronounced.
Wh-features are uninterpretable. Phrases with wh-features must move to SpecCP
to have their features checked or deleted. In Turkish, wh-features are weak. A chain is
formed by movement (copy and merge) between the wh-element in its case position
and its copy in SpecCP. Wh-features being weak, the copy in SpecCP is marked for
PF-deletion, and the lower copy of the wh-chain is pronounced.
In spite of the fact that wh-expressions do not undergo overt wh-movement in
Turkish, there is evidence for overt movement of the wh-phrase to SpecCP in Turkish
sluicing structures.  The evidence comes from the Case-Matching requirement in ().
The sluiced phrase must obligatorily have the same case as its correlate in the
antecedent clause. A base-generation model of sluicing such as the one proposed by
Chung et al. (), where the sluiced wh-phrase is directly merged in SpecCP, cannot
account for a case that would have had to be assigned in a case-marking position or
merged into the structure in a theta position with lexical case. 
() Case-Matching
The sluiced wh-phrase must bear the case that its correlate bears.
(Merchant : )
Let us look at examples of case-matching in Turkish. In (), the indirect object of the
verb ver- ‘to give’ bears dative case.
() Hasan-∅ Ahmet-e para ver-di-∅.
Hasan-nom Ahmet-dat money give-pst-sg
‘Hasan gave money to Ahmet.’
Now note that the wh-indirect object of ver- must also bear dative case in sluicing
constructions, ().
() Hasan biri-ne para ver-miş-∅; ama kim-e
Hasan-nom one-dat money give-pst-sg but who-dat
bil-mi-yor-um.
know-neg-pres-sg
‘Hasan gave money to someone; but I don’t know who (to).’
 For wh-constructions in Turkish, see Akar (), Özsoy (), Arslan (), and Aygen
(, ).
 In an alternative proposal, Chomsky and Lasnik () assume LF case-checking for English. However,
this idea doesn’t work in overt-case-marking languages where case must be visible at PF as well.
252 Atakan İnce

The same holds for sluiced wh-phrases bearing other cases, which must all match
the case of their non-sluiced counterparts. In (), the antecedent indefinite bears
ablative case, as must the sluiced wh-phrase. In (), the antecedent indefinite in
complement position has accusative case, as must its sluiced wh-phrase counter-
part. This case-matching requirement is accounted for naturally if the wh-phrase is
assigned/checks case in a clause-internal position and is then raised to a periph-
eral position prior to deletion of the remainder of the clause, as illustrated in (b)
and (b).
() a. A: Hasan birin-den borç al-mış-∅.
Hasan-nom one-abl debt take-pst
B: Kim-den?
who-abl
‘A: Hasan borrowed money from someone. B: Who from?’
b. Kimden [Hasan kimden borç almış]?

() a. A: Ali-∅ dün biri-ni ara-dı-∅.


Ali-nom yesterday one-acc called-sg
B: Kim-i acaba?
who-acc I.wonder
‘A: Ali called someone yesterday. B: I wonder who.’
b. Kimi [Ali dün kimi aradı] acaba?

The same case-matching requirement is observed in multiple-sluicing examples as


well. In (), the indirect-object wh-phrase kime has dative case, matching its non-
sluiced correlate, and the non-specific direct-object wh-phrase ne is in its bare form,
corresponding to its bare direct-object correlate. 
() Ahmet-∅ biri-ne bir-şey ver-miş-∅; ama kim-e ne
Ahmet-nom one-dat one-thing give-pst but who-dat what
bil-mi-yor-um.
know-neg-prog-sg
‘Ahmet gave someone something; but I don’t know what to who.’
 Enç () observes that non-specific direct objects in Turkish are bare, as in (a) below, in contrast to
specific direct objects, which must bear an overt accusative-case morpheme, as in (b).

a. Hasan-∅ elma yi-yor-∅.


Hasan-nom apple eat-prog-sg
‘Hasan is eating an apple/apples.’
b. Hasan-∅ elma-yı yi-yor-∅.
Hasan-nom apple-acc eat-prog-sg
‘Hasan is eating the apple.’
Sluicing in Turkish 253

As noted above, in sluicing constructions, the wh-phrase must be in the CP-domain


for deletion of the remainder of the clause to be possible, and we have seen evidence
that in Turkish the sluiced phrase should be assumed to be in SpecCP as well. However,
in Turkish, wh-features are weak, as the language is wh-in-situ, and so it cannot be
wh-features that result in the overt wh-movement of sluicing. I propose, therefore, that
in sluicing structures, what causes and permits the overt raising of wh-expressions to
the CP-domain is the checking of focus rather than wh-features.  , 
As support for a focus-driven view of movement in sluicing, it can be noted that
sluiced wh-phrases must bear focus intonation, and can generally be assumed to be
focused elements. The non-sluiced version of (a) is commonly de-accented with flat
intonation, as denoted by the shading in (b). This intonation is not acceptable for
the sluiced wh-phrase in (c).

() a. John bought something, but I don’t remember what.


b. John bought something, but I don’t remember what he bought.
c. ∗ John bought something, but I don’t remember what.

In Turkish as well, the sluiced wh-phrase bears the marked pronunciation of focus
stress: The wh-phrase kimle ‘with who’ in () must be stressed as in ().

() Hasan biri-yle konuş-uyor-φ ama kim-le bil-mi-yor-um.


Hasan-nom one-comm talk-prog-s but who-comm know-neg-prog-s

Furthermore, whenever there are focus features, the focused phrase is pronounced
in the position where it checks those features. To my knowledge, no language has
been argued to have covert focus-feature checking. Thus, focus features are regu-
larly strong, triggering overt movement and pronunciation of the head of the focus
chain.
In Turkish, with no overt wh-movement, it can therefore be assumed that focus
features that motivate the pronunciation of the wh-phrase in the CP-domain. In
sluicing, the wh-phrase ne ‘what’ copies and merges in SpecCP where it checks both
wh and focus features, as in ().

 Aygen () proposes that a null Q with a focus feature raises to C . It is well-known that focus
in Turkish also has intonational ramifications (see Göksel and Özsoy , and references therein); i.e.,
a focused element must have distinctive intonation. In this sense, an argument for the raising of a null
Q with a focus feature is empirically weak since a null element checking focus features will not express
phonological information such as focus intonation. Therefore, I do not adopt this suggestion.
 The focus-based analysis of sluicing proposed here builds on work initially developed during the
course of the author’s graduate research in the University of Maryland and finalized in İnce (). For
an independently-developed view that sluicing in a wh-in-situ language (Farsi) may be licensed via focus-
movement, see Toosarvandani ().
254 Atakan İnce

() CP
2
ne (‘what’) C⬘
wh-features
focus-features

As shown in (), after movement of the wh-phrase ne to SpecCP, the TP is deleted,


and the higher copy of the wh/focus chain is pronounced since the focus features
present in the wh-phrase are strong.

() CP
2
wh-nei C⬘
2
C
TP
5 deleted
… ti …

In such an approach to sluicing constructions, a natural question is: Why does a


wh-element not raise to the CP-domain overtly in other, non-sluicing wh-questions?
The simple answer that I believe can be given here is that wh-phrases in other non-
sluicing environments commonly lack focus features, and this is what distinguishes
them from wh-phrases in sluices. 

 Evidence for the proposal that wh-phrases are not inherently [+FOCUS] comes from optional-wh-
movement languages such as Malagasy, Tuki, and Bahasa Indonesia (Sabel ). In these languages, a wh-
phrase raises to the C-domain only when there is a focus particle, as shown in (i) and (ii) from Malagasy.
If wh-phrases were inherently focused, they would raise to C with or without the presence of the focus
particle.

(i) a. [CP  [Inona no]i heverin-dRabe [CP  fa novidin-dRakoto t i ]]?


what foc pres.tt.believe-Rabe that pst.tt.buy-Rakoto
‘What does Rabe believe that Rakoto has bought?’
b. [CP  Heverin-dRabe [CP  fa [inona no]i novidin-dRakoto t i ]]?
pres.tt.believe-Rabe that what foc pst.tt.buy-Rakoto
‘What does Rabe believe that Rakoto has bought?’
(ii) a. Mividy (∗ no) inona any amin’ ny magazay Rabe?
pres.at.buy foc what there in the shop Rabe
‘What does Rabe buy in the shop?’
b. Inona ∗ (no) vidin-dRabe any amin’ ny magazay?
what foc pres.tt.buy-Rabe there in the shop
‘What does Rabe buy in the shop?’ (Sabel : –)

Furthermore, for a phrase to be focused, it must be either new information or contrastive with given
information. However, wh-phrases do not introduce new information, nor do they contrast with other
given information; they merely request new information. This is an argument against wh-phrases being
inherently focused.
Sluicing in Turkish 255

. Sluicing in Turkish = elliptical clefts


So far, I have argued that in sluicing, the wh-phrase occurs in SpecCP as the result
of (focus) movement. Now let’s consider an alternative possibility, that the so-called
sluicing structures are actually elliptical cleft structures. Kizu () makes just that
argument for Japanese and Turkish, suggesting that sluicing structures in wh-in-situ
languages are elliptical cleft constructions.
Kizu argues that the occurrence of a copula marker attached to a wh-phrase is
possible in both sluicing and cleft structures, as in (), and concludes the following:
. . . whether it is obligatory or not, it seems that the existence of copula is observed
widely in sluicing constructions of wh-in-situ languages. This fact supports the idea
that sluicing involves a cleft construction because the latter is a certain type of copula
sentence. (Kizu )

Additionally, Göksel and Özsoy () provide evidence in Turkish that wh-phrases are not inherently
focused. Observing that a wh-phrase cannot precede a focused phrase unless it bears focal stress, they
suggest that focus is expressed through stress in Turkish. Note that the focused phrase must precede the
relevant wh-phrase, as in (iv). If wh-phrases in Turkish were inherently focused, they would precede, rather
than follow, the focused phrase.

(iii) a. ∗ Ne zaman OKUL-A gid-ecek-sin?


what time school-dat go-fut-sg
‘When will you go TO SCHOOL?’
b. ∗ Kim-∅ SEN-İ sev-iyor-∅?
who-nom you-acc love-prog-sg
‘Who loves YOU?’
c. ∗ Kim-i SİNEMA-DA gör-ecek-sin?
who-acc cinema-loc see-fut-sg
‘Who will you see AT THE CINEMA?’
(iv) a. OKUL-A ne zaman gid-ecek-sin?
school-dat what time go-fut-sg
‘When will you go TO SCHOOL?’
b. SEN-İ kim-∅ sev-iyor-∅?
you-acc who-nom love-prog-sg
‘Who loves YOU?’
c. SİNEMA-DA kim-i gör-ecek-sin?
cinema-loc who-acc see-fut-sg
‘Who will you see AT THE CINEMA?’
(v) a. NE ZAMAN OKUL-A gid-ecek-sin?
what time school-dat go-fut-sg
‘When will you go to school?’
b. KİM SEN-İ sev-iyor-∅?
who-nom you-acc love-prog-sg
‘Who loves you?’
c. KİM-İ SİNEMA-DA gör-ecek-sin?
who-acc cinema-loc see-fut-sg
‘Who will you see at the cinema?’
256 Atakan İnce

() a. Mehmet bir-şey söyle-di-∅ ama, ne-y-di-∅


Mehmet one-thing tell-pst-sg but what-be-pst-sg
hatirla-mi-yor-um.
remember-neg-prog-sg
‘Mehmet said something, but I don’t remember what (it) was.’
b. Mehmet’in kır-dığ-ı bir vazo-y-du.
Mehmet-gen break-nmlz-poss.sg one vase-be-pst
‘What Mehmet broke was a vase.’ (Kizu :  ())

However, this argument is problematic because in Turkish cleft constructions, a


clefted phrase can only bear nominative case.  In the double-object example in (a),
the subject Hasan has nominative case, the direct object kitab ‘book’ has accusative,
and the indirect object Suzan has dative case. But in all the clefted versions of (a), the
clefted DP can only bear nominative case. A clefted phrase bearing the case it would
have if not clefted is ungrammatical.

() a. Hasan-∅ Suzan-a kitab-ı ver-di-∅.


Hasan-nom Susan-dat book-acc give-pst-sg
‘Hasan gave Susan the book.’
b. [Suzan-a kitab-ı ver-en] Hasan-∅-dı.
Susan-dat book-acc give-comp Hasan-nom-pst
‘It was Hasan who gave the book to Susan.’
c. [Hasan-ın Suzan-a ver-diğ-i] kitab-∅/∗ ı-tı.
Hasan-gen Susan-dat give-comp-poss.sg book-nom/∗ acc-pst
‘It was the book that Hasan gave to Susan.’

 A reviewer notes that (b–d) have been analyzed as free relatives in the literature (by Kornfilt (),
among others). However, according to Caponigro (, , ), free relatives cross-linguistically
must have wh-words in clause-initial/final position (according to the head-parameter) “or a morpho-
logically complex word with a wh-word as its root” (e.g. whoever) (Caponigro : ). Thus, (b–d)
cannot be free relatives, because they lack a wh-word. Furthermore, as shown in (i–iii) below, as min-
imal pairs for (b–d), respectively, a wh-phrase added to a “free relative” yields ungrammaticality in
Turkish.

(i) ∗ Suzan-a kitab-ı ver-en kim-∅ Hasan-∅-dı.


Susan-dat book-acc give-comp who-nom Hasan-nom-pst
(ii) ∗ Hasan-ın Suzan-a ver-diğ-i ne kitap-∅-tı.
Hasan-gen Susan-dat give-comp-poss.sg what book-nom-pst
(iii) ∗ Hasan-ın kitab-ı ver-diğ-i kim-∅ Suzan-∅-dı.
Hasan-gen book-acc give-comp-poss.sg who-nom Susan-nom-pst
Also, whereas both argument and adjunct wh-phrases, as in (iv) and (v) below, respectively, can head free
relatives in, for example, English, Turkish does not have a structure similar to (v). In other words, a locative
adjunct as a free relative is impossible in Turkish.

(iv) I’ll marry [who you choose].


(v) You can’t smoke [where the kids are playing]. (Caponigro : ).
Sluicing in Turkish 257

d. [Hasan-ın kitab-ı ver-diğ-i] Suzan-∅/∗ a-dı.


Hasan-gen book-acc give-comp-poss.sg Susan-nom/∗ dat-pst
‘It was Susan who Hasan gave the book (to).’
Now compare the examples in () with the following sluicing examples. In (a),
the sluiced wh-phrase has dative case, in (b), ablative case, and in (c), accusative
case. In every instance, the sluiced wh-phrase must obligatorily bear the same case as
its non-sluiced correlate.

() a. Ahmet-∅ biri-ne kitap-∅ ver-miş-∅, ama kim-e/∗ -∅


Ahmet-nom one-dat book-nom give-pst-sg but who-dat/∗ -nom
bil-mi-yor-um.
know-neg-pres-sg
‘Ahmet gave the book to someone, but I do not know to who.’
b. Ahmet-∅ birin-den borç-∅ al-mış-∅, ama kim-den/∗ -∅
Ahmet-nom one-abl debt-nom take-pst-sg but who-abl/∗ -nom
bil-mi-yor-um.
know-neg-pres-sg
‘Ahmet borrowed money from someone, but I don’t know who from.’
c. Ahmet-∅ biri-ni döv-müş-∅, ama kim-i/∗ -∅
Ahmet-nom one-acc beat-pst-sg but who-acc/∗ -nom
bil-mi-yor-um.
know-neg-pres-sg
‘Ahmet beat someone, but I don’t know who.’
This disparity in case requirement is evidence that sluicing in Turkish is different from
cleft constructions.
More evidence in support of a sluicing analysis, and against a clefting analysis,
comes from postposition pied-piping.
() Postposition pied-piping
A postposition must be pied-piped with a wh-item in occurrences of sluicing.
 Although the clause preceding the clefted phrase seems on the surface to be identical to a relative
clause, the two structures are distinct. The clefted phrase is not the external head of the preceding clause.
There must be a pause between the clause and the DP as in (i) below. This contrasts with relative clauses,
where there is no pause between the relative clause and its head, as in (ii).
(i) Cleft construction
(∗ Bu kız) dün gör-düğ-üm # gelin-di. (# signifies pause)
(∗ this girl) yesterday see-comp-sg # bride-pst
‘It was a bride I saw yesterday.’

(ii) Relative clause


Bu kız [dün ∅i gör-düğ-üm] gelini -di.
this girl yesterday see-comp-sg bride-pst
‘This girl is [the bride [that I saw t yesterday]].’
258 Atakan İnce

The following examples show that postposition pied-piping is obligatory in sluicing


structures in Turkish.
() Hasan-∅ bir ip al-mış-∅, ama kim-∅ için
Hasan-nom a rope buy-pst-sg but who-nom for
bil-mi-yor-um.
know-neg-pres-sg
‘Hasan bought a rope, but I don’t know for who.’
() Tartış-ıyor-lar-dı, ama ne-∅ hakk-ın-da
discuss-prog-pl-pst but what-nom about-poss.sg-loc
bil-mi-yor-um.
know-neg-pres-sg
‘They were conversing, but I don’t know about what.’
() Bu oyun-∅  yüzyıl-da yaşa-mış-∅ biri-∅ taraf-ın-dan
this play-nom  century-loc live-pst-sg one-nom side-poss.sg-abl
yaz-ıl-mış-∅ ama kim-∅ taraf-ın-dan
write-pass-pst-sg but who-nom side-poss.sg-abl
bil-mi-yor-uz.
know-neg-pres-pl
‘This play was written by someone who lived in the th century, but we don’t
know by whom.’
() Ahmet-∅ Rektörlük-te ol-acak-∅, ama ne zaman-a
Ahmet-nom Chancellor’s.building-loc be-fut-sg but what time-dat
kadar bil-mi-yor-um.
until know-neg-pres-sg
‘Ahmet will be in the Chancellor’s Building, but I don’t know until when.’
Notice, however, that it is not possible to cleft wh-phrases with pied-piped postposi-
tions (the clefted phrases are in bold letters).
() ∗ Yaz-ıl-an-∅ oyun-∅ kim-∅ taraf-ın-dan?
write-pass-rel-sg play-nom who-nom side-poss.sg-abl
(‘By whom was it that the play was written?’)
() ∗ Tartış-tık-ları ne-∅ hakk-ın-da-y-dı?
discuss-comp-poss.pl what-nom about-poss.sg-loc-cop-pst
(‘About what was it that they discussed?’)
The same basic contrast between sluicing and cleft structures holds in Greek as well
(Merchant ). In the sluiced example in (a), the wh-phrase must bear the same
case as its correlate, whereas in the cleft structure in (b), the wh-phrase cannot bear
the accusative of its correlate, and must instead appear with nominative case.
Sluicing in Turkish 259

() I astinomia anekrine enan apo tous Kiprious prota, ala


the police interrogated one.acc from the Cypriots first but
dhen ksero
not I.know
a. {∗ pjos /pjon}.
which.nom which.acc
b. {pjos itan /∗ pjon itan}.
which.nom it.was which.acc it.was
‘The police interrogated one of the Cypriots first, but I don’t know which/which
it was.’ (Merchant : )
Another difference between clefting and sluicing is that (wh-)adjuncts cannot be
clefted in Turkish, () and (). Adjuncts, however, can be sluiced, as in () and ().
() ∗ Ali-nin git-tiğ-i dün. /ne zaman-∅?
Ali-gen go-comp-poss.sg yesterday what time-nom
(‘It’s yesterday that Ali went./When is it that Ali went?’)
() ∗ Ali-nin git-tiğ-i Ankara-∅.
Ali-gen go-comp-poss.sg Ankara-nom
(‘It’s to Ankara that Ali went.’)
() Ali-∅ Ankara-ya git-ti-∅, ama ne zaman bil-mi-yor-um.
Ali-nom Ankara-dat go-pst-sg but what time know-neg-pres-sg
‘Ali went to Ankara, but I don’t know when.’
() Ali-∅ bir yer-e git-ti-∅, ama nere-ye bil-mi-yor-um.
Ali-nom a place-dat go-pst-sg but where-dat know-neg-pres-sg
‘Ali went somewhere, but I don’t know where.’
Furthermore, Turkish cleft constructions can be followed by the existential copula
ol- (‘it is/was’), as in () and (), but sluiced adjunct wh-phrases cannot be, as in
(–).
() Ahmet-in borç-∅ ver-diğ-i-nin Hasan-∅
Ahmet-gen debt-nom give-comp-poss.sg-gen Hasan-nom
ol-duğ-u-nu bil-iyor-um.
be-comp-poss.sg-acc know-pres-sg
‘I know that it is Hasan who Ahmet lent money (to).’
() Hasan-a borç ver-en-in Ahmet-∅ ol-duğ-u-nu
Hasan-dat debt give-comp-gen Ahmet-nom be-comp-poss.sg-acc
bil-iyor-um.
know-pres-sg
‘I know that it is Ahmet who lent money to Hasan.’
260 Atakan İnce

() Araba-yı onar-dı-∅, ama nasıl (∗ ol-duğ-u-nu)


car-acc fix-pst-sg but how be-comp-poss.sg-acc
bil-mi-yor-um.
know-neg-pres-sg
‘He fixed the car, but I don’t know how (∗ it was).’
() Araba-yı onar-dı-∅, ama niye (∗ ol-duğ-u-nu)
car-acc fix-pst-sg but why be-comp-poss.sg-acc
bil-mi-yor-um.
know-neg-pres-sg
‘He fixed the car, but I don’t know why (∗ it was).’
() Araba-yı onar-dı-∅, ama ne zaman (∗ ol-duğ-u-nu)
car-acc fix-pst-sg but what time be-comp-poss.sg-acc
bil-mi-yor-um.
know-neg-pres-sg
‘He fixed the car, but I don’t know when (∗ it was).’
() Mücevherler-i sakla-dı-∅, ama nereye (∗ ol-duğ-u-nu)
jewels-acc hide-pst-sg but where be-comp-poss.sg-acc
bil-mi-yor-um.
know-neg-pres-sg
‘He hid the jewels, but I don’t know where (∗ it is).’
Lastly, as shown in (), multiple sluicing is grammatical in Turkish, whereas, as
demonstrated in (), multiple clefting is not.
() Ahmet birin-den bir-şey al-mış-∅; ama kim-den ne
Ahmet-nom one-abl one-thing borrow-pst-sg but who-abl what
bil-mi-yorum.
know-neg-pres-sg
‘Ahmet borrowed something from someone; but I don’t know what from
whom.’
∗[
() TP [CPAhmet-in t i t j al-dığ-ı] [T Hasan(-dan)i kitap-∅j ]].
Ahmet-gen borrow-rel-sg Hasan(-abl) book-nom.
‘It’s a book from Hasan that Ahmet borrowed.’
In sum, sluicing and cleft structures in Turkish have different properties. In sluicing
structures, the sluiced wh-phrase appears bearing the case that it checks against a
relevant functional head in the elided clause. In addition, not only argument, but
also adjunct, wh-phrases can be sluiced in Turkish. Postposition pied-piping is oblig-
atory in Turkish sluicing structures, and multiple sluicing is permitted. Conversely,
in cleft structures, the clefted phrase can only appear in nominative case and must be
Sluicing in Turkish 261

an argument. Neither postposition pied-piping nor multiple clefting is permitted in


Turkish cleft structures. We can conclude that the clefting account of Turkish sluicing
structures is not viable.

. A case mismatch


In Turkish, embedded clauses are nominalized, and display nominal/possessive agree-
ment. The subject of the embedded clause has genitive case, and the verb has posses-
sive agreement with the genitive case-marked subject.  The entire embedded clause
is case-marked by the matrix verb, as in ().
() Ahmet [kim-in Ankara-ya git-tiğ-i] -ni
Ahmet-nom who-gen Ankara-dat go-comp-poss.sg -acc
san-ıyor-∅?
think-prog-sg
‘Whoi does Ahmet think t i went to Ankara?’
Interestingly, when the embedded clause is sluiced, the subject wh-phrase must bear
nominative case, and not the genitive of its non-sluiced counterpart, as in ().
() Ahmet [biri-nin Ankara-ya git-tiğ-i] -ni söyle-di-∅;
Ahmet-nom one-gen Ankara-dat go-comp-poss.sg -acc tell-pst-sg
ama kim-∅ bil-mi-yor-um.
but who-nom know-neg-pres-sg
‘Ahmet said someone went to Ankara, but I don’t know who.’
As demonstrated in (), genitive case on the sluiced subject wh-phrase is unaccept-
able.
() ∗. . . ama kim-in bil-mi-yor-um.
but who-gen know-neg-pres-sg
One might be inclined to view () as a cleft structure, since the sluiced embedded
subject cannot have the genitive case of its correlate (as would be expected in sluicing),
and instead, must bear nominative case. This idea, though, cannot be right. First, it

 For studies on genitive case in Turkish and Turkic languages, as well as in languages such as Dagur and
Japanese, see Kornfilt (, , , ), Aygen (), Hale (), Krause (), Miyagawa (),
and Ochi (). Miyagawa () and Ochi () propose that an external nominal head is responsible
for genitive case. Kornfilt (, , ) argues that agreement is responsible for genitive case on the
subject NP in Turkish, in that verbal agreement licenses nominative case, and nominal agreement licenses
genitive case. Aygen () proposes that covert phrasal movement to a nominal head is responsible for
genitive case in Turkish. Kornfilt () also proposes that a nominal head n/N is responsible for genitive
case-checking. However, under Aygen’s () and Kornfilt’s () analyses, it would remain a mystery
why the sluiced subject wh-phrase of embedded complement clauses cannot have genitive case, because a
head, but not an amalgam, assigns genitive case to the subject NP. So it should be possible for the nominal
head to assign genitive case to the subject NP before it is elided.
262 Atakan İnce

would be odd for there to be (and it would be theoretically difficult to state) a rule that
bans just sluicing of the subject of an embedded complement clause, while sluicing of
all other arguments and adjuncts is permitted.
Second, multiple sluicing of embedded arguments also requires that the subject be
in nominative case (rather than the embedded genitive), as in (). We have already
seen that multiple clefting is not acceptable in Turkish, so examples such as () must
be instances of sluicing, not clefting.
() Ahmet biri-nin biryer-e git-tiğ-i-ni söyle-di-∅,
Ahmet-nom one-gen one.place-dat go-comp-poss.sg-acc tell-pst-sg
ama kim-∅ nere-ye hatırla-mı-yor-um.
but who-nom where-dat remember-neg-prog-sg
‘Ahmet said that someone went somewhere, but I don’t remember who where.’
As a possible solution, one might suggest that a wh-phrase sluiced out of an embed-
ded clause can only have some kind of default case, nominative perhaps being the
default case in Turkish (Kornfilt ).  However, if this were the case, other sluiced
wh-phrases, whether arguments or adjuncts, would bear default case as well. Exam-
ples () and () show that this cannot be correct, as case-matching is required on
all the sluiced wh-phrases (in bold) except that of the embedded subject, kim, in ().
() Ahmet [Hasan-ın bir yer-den gel-diğ-i] -ni
Ahmet-nom Hasan-gen one place-abl come-comp-poss.sg -acc
söyle-di-∅. Nere-∗ (den) bil-mi-yor-um.
tell-pst-sg where-abl know-neg-pres-sg
‘Ahmet told me that Hasan came from somewhere. I do not know from where.’
() Fatma [biri-nin bir yer-e git-tiğ-i] -ni
Fatma-nom one-gen one place-dat go-comp-poss.sg -acc
duy-muş-∅. Acaba kim-∅ nere-∗ (ye).
hear-pst-sg wonderment who-nom where-dat
‘Fatma heard that someone went somewhere. I wonder who where.’
It is apparent that the case mismatch in () is unique to sluiced wh-subjects of
embedded complement clauses.
I assume that the sluiced subject of embedded complement clauses is derived via
movement and is not base-generated in its Spell-Out position; hence in (), the
wh-subject has raised from the most deeply embedded clause to the SpecCP of the
next-higher clause (the CP-complement of the verb ‘know’). In the next section,
I present evidence of inter-clausal raising of embedded wh-subjects in sluicing. In
Section .., I suggest an explanation for the case mismatch by adopting Hiraiwa’s
() nominative–genitive conversion account and a Multiple Spell-Out model.
 Schütze () also suggests that nominative case is the default case in overt-case-morphology
languages.
Sluicing in Turkish 263

.. The position of the embedded sluiced subject


Let us first determine that a sluiced wh-phrase from an embedded complement clause
does raise out of the latter to a higher CP-domain.
Turkish has a Q-like particle -ki, which attaches to both yes–no and wh-
interrogative clauses, as shown in () and (). 
() a. Ahmet gel-di-∅ mi?
Ahmet-nom come-pst-sg Q
‘Did Ahmet come?’
b. Ahmet gel-di-∅ mi ki?
Ahmet-nom come-pst-sg Q prt
‘Did Ahmet come, then?’
() a. Ahmet ne ye-di-∅?
Ahmet-nom what eat-pst-sg
‘What did Ahmet eat?’
b. Ahmet ne ye-di-∅ ki?
Ahmet-nom what eat-pst-sg prt
‘What did Ahmet eat, then?’
This ki cannot occur in non-interrogative environments ().
() A: Kim-∅ gel-di-∅ ki?
who-nom come-pst-sg prt
B: Ali (∗ ki).
Ali-nom prt
‘A: Who came, then? B: Ali (∗ then).’

 See Besler () for the ki particle. A reviewer suggests that the reason ki cannot occur in embedded
complement clauses is because ki takes TP as its complement, and complement clauses in Turkish are AspPs,
and not TPs. I disagree with this line of reasoning. First, the Q-particle that precedes ki is the Force head;
thus, ki must take CP as a complement and not a lower phrase. Note that the ki particle can take a TP
complement as in (i) below, where it gives a “but . . .” reading, but this is an altogether different function
from that of the ki particle above. (For other functions of ki, see İnce ()).
(i) Ali gel-di-∅ ki.
Ali-nom come-pst-sg prt
‘But, Ali came.’
Second, like ki, the Turkish yes–no Q-particle -mi does not occur in embedded complement clauses either
(ii), but it can follow adjunct clauses (which are also not TPs) (iii). So the restriction against complement
clauses cannot be so simple as the phrasal label of the clause.
(ii) ∗ Ali [Hasan-ın gel-di-ğ-in-i-mi] sor-du-∅.
Ali-nom Hasan-gen come-pst-comp-poss.sg-acc-Q ask-pst-sg
(‘Ali asked whether Hasan came.’)
(iii) [Ali kız-ınca-mı] Hasan gel-di-∅?
Ali-nom get.angry-when-Q Hasan-nom come-pst-sg
‘Is it when Ali got angry that Hasan came?’
In İnce (), I propose that ki merges in Force position in the CP-domain, and that ForceP does not
project in non-matrix CP-domains.
264 Atakan İnce

It can occur in matrix clauses, but cannot occur in embedded clauses ().
() a. Hasan-ın ne ye-diğ-i-ni duy-du-n ki?
Hasan-gen what eat-comp-poss.sg-acc hear-pst-sg prt
‘What did you hear that Hasan ate, then?’
b. ∗ Hasan-ın ne ye-diğ-i-ni ki duy-du-n?
Hasan-gen what eat-comp-poss.sg-acc prt hear-pst-sg
(‘What did you hear that Hasan ate, then?’)
With regard to the clausal location of sluicing, it has been pointed out that there are
two types of sluicing structure: embedded-clause sluicing and matrix-clause sluicing
(Lasnik ). The morpheme ki can appear in matrix sluicing. This is shown in (),
where ki can optionally occur with the sluiced wh-phrase kim ‘who’.
() A: Sen-i biri-∅ ara-dı-∅.
you-acc one-nom call-pst-sg
B: Kim-∅ (ki)?
who-nom prt
‘A: Someone called you. Who (then)?’
The logic, then, is that because ki cannot appear in embedded clauses, if a wh-phrase
remains in the embedded CP-phrase it originates in, it should not be able to co-occur
with ki. Since in () ki is acceptable, we must assume that the wh-phrase kim must be
in the matrix CP.
In (), now, the sluiced wh-phrase originates as the subject of the embedded
clause, and in () the sluiced wh-phrase is base-generated as the indirect object of
the embedded clause. Significantly, both sluiced wh-phrases can co-occur with ki.
() Hasan sen-i biri-nin ara-dığ-ı-nı söyle-di-∅.
Hasan-nom pron.sg-acc one-gen call-comp-poss.sg-acc tell-pst-sg
Kim-∅ ki?
who-nom prt
‘Hasan said that someone called you. Who, then?’
() Hasan Ahmet-in birisi-ne kitap-∅ ver-diğ-i-ni
Hasan-nom Ahmet-gen one-dat book-nom give-comp-poss.sg-acc
söyle-di-∅. Kim-e ki?
tell-pst-sg who-dat prt
‘Hasan said that Ahmet gave a book to someone. Who to, then?’
The non-sluiced versions of () and () appear as () and (), respectively.
() [CP [Hasan sen-i [CP [kim-in ara-dığ-ı]] -nı
Hasan-nom pron.sg-acc who-gen call-comp-poss.sg -acc
söyle-di-∅] ki]?
tell-pst-sg prt
‘Who did Hasan say called you, then?’
Sluicing in Turkish 265

() [CP [Hasan [CP [Ahmet-in kim-e kitap-∅


Hasan-nom Ahmet-gen who-dat book-nom
ver-diğ-i]] -ni söyle-di-∅] ki]?
give-comp-poss.sg -acc tell-pst-sg prt
‘Who did Hasan say that Ahmet gave a book, then?’

In both cases, ki is in the matrix CP-domain. In () and (), it must therefore be
concluded that the sluiced wh-phrases have raised all the way to the matrix SpecCP,
because (a) ki occurs only in matrix clauses, and (b) ki appears clause-finally, which
means it is in the CP-domain. If the sluiced wh-phrases in () and () were to remain
in the embedded clauses where they originate, this would mean that deletion would
have to apply to the matrix subject and verb while retaining ki in the matrix clause,
and deleting everything except the wh-element in the embedded clause. No theory
can account for this kind of discontinuous ellipsis.
We must therefore assume that in these constructions, the sluiced wh-phrase raises
to the matrix SpecCP out of an embedded complement clause. The position of the
sluiced wh-phrase and the elided part in () is as in ().

() [CP Kim-ei [C [TP Hasan [CP Ahmet’in t i kitap-∅


who-dat Hasan-nom Ahmet-gen book-nom
ver-diğ-i] -ni söyle-di-∅] ki]]?
give-comp-poss.sg -acc tell-pst-sg prt

.. A solution to the genitive case mismatch


By adopting a modified version of Hiraiwa’s () account of the nominative–genitive
conversion in the case of subjects in embedded environments, and assuming a Mul-
tiple Spell-Out model, it is possible to explain the case anomaly found with sluiced
subjects of embedded clauses in Turkish.
First, Hiraiwa argues that the nominative–genitive conversion observed in relative
clauses and noun complement clauses in Japanese, like those shown in (), is a result
of a C-T-v-V head amalgam formed via AGREE. Hiraiwa suggests that the phi-features
of the C-T-v-V amalgam check genitive case of the subject NP.

() a. kinoo John-ga katta hon


yesterday John-nom bought book
‘the book which John bought yesterday’
b. kinoo John-no katta hon
yesterday John-gen bought book
‘the book which John bought yesterday (Hiraiwa : )
c. John-wa [CP kinoo Mary-ga kita koto/no] -o siranakatta
John-top yesterday Mary-nom came C -acc knew.not
‘John didn’t know that Mary came yesterday.’
266 Atakan İnce

d. John-wa [CP kinoo Mary-no kita koto/no] -o siranakatta


John-top yesterday Mary-gen came C -acc knew.not
‘John didn’t know that Mary came yesterday.’ (Hiraiwa : )

Second, Uriagereka () and Chomsky (, ) propose a Multiple Spell-
Out model whereby a derived structure is sent to LF and PF not once (Single Output
Syntax), but rather in subclausal units, referred to as “phases”, which have their own
syntactic uniformity. Phases are assumed to result from the construction of vPs and
CPs (and maybe DPs and some PPs). In the Multiple Spell-Out model, the com-
plement of each phase head is known as the “domain” of the phase, and the head
and specifier(s) comprise the “edge” of the phase. Once a phase has been derived, its
domain is sent to LF and PF. The material in the edge of the phase is sent to LF and
PF as part of the subsequent phase.
Sluicing is commonly assumed to involve TP-ellipsis. However, in a Multiple Spell-
Out model, ellipsis cannot be applied only to a TP. After the vP phase is derived, its
domain, the VP, is sent to LF and PF. Then the CP phase is derived, and its domain,
TP, is marked for elision and spelled out. However, if VP is not marked for elision at
the point of Spell-Out, the material within it will be pronounced, even if TP is elided.
Thus, in sluicing structures, the domain of each phase must be marked for elision at
Spell-Out. Let us look at an illustration of why this must be so. Let’s assume there is a
VP with a light manner adverb adjoined to it. At the end of the vP phase, its domain,
the VP, is spelled out, i.e. sent to LF and PF, as shown in (a). Note that the adverb is
sent to PF to be pronounced, while the verb has raised to v. The derivation continues in
(b), when, at the CP phase, its domain, the TP, is marked for elision. Now, however,
as shown in (c), what will be pronounced is the wh-expression of the sluice and
the adverb, which, because it had been spelled out in the previous phase, is not in the
structure when the TP is marked for deletion. As sequences such as (c) are clearly
unacceptable, it needs to be concluded that in sluicing, the domain of all phases must
be marked for elision at Spell-Out.

() a. At the vP phase:


vP

VP v
Adverb
V
VP sent to LF and PF
Sluicing in Turkish 267

b. At the CP phase:
CP

whi
TP C

vP T

v-V
Delete TP
(i.e. mark for elision and Spell-Out)
c. Wh adverb
Example: ∗ Who slowly?

Now let’s see how it is that the sluiced wh-subject lacks genitive case. I adopt Hiraiwa
() in assuming that the phi-features of the C-T-v-V amalgam check genitive case
of the subject in embedded complement clauses.  I further assume that it is the T-v-V
amalgam that needs to agree with C, not vice versa. Turkish is an overt V-raising
language. In sluicing structures, after the vP phase of the embedded clause is formed,
the V head-moves to v, and the VP is marked for elision and sent to LF and PF. At
the CP phase, the subject wh-phrase raises to SpecTP. The v-V amalgam head-moves
to T. At this stage, the phi-features of the T-v-V amalgam check nominative case of
the subject wh-phrase, after which the wh-subject raises to SpecCP to check C’s wh
and FOCUS features. Crucially, the CP phase is complete after movement of the wh-
subject to its Spec. It is then that TP is marked for elision and is spelled out. In sluicing,
however, crucially the C-T-v-V amalgam that assigns/checks genitive case is never
formed.
Two points should be kept in mind. First, the T-v-V amalgam is elided within
TP prior to checking its features against C. Recall that it is the T-v-V amalgam that
must agree with C. When this amalgam is elided with its features still unchecked,
the derivation does not crash, because the amalgam’s uninterpretable features are also
elided (Lasnik ), and C does not have any uninterpretable features that still need
to be checked.
Second, even though the sluiced wh-subject bears nominative case, it must have
raised to SpecCP, because had it remained in SpecTP, it would be elided with TP, and
would not be pronounced at the end of the derivation.

 I assume that the -DIK verbal morpheme of complement clauses is a combination of Tense (-DI-) plus
Complementizer (-K), along the lines of Kural (), rather than a single functional category. That is, it is
composed of separate and distinct functional heads. For other views, see Lees (), Kornfilt (, ),
and Özsoy (), who consider -DIK a nominalizer, while Kennelly () and Aygen () call it Aspect.
268 Atakan İnce

In sum, since it is the C-T-v-V amalgam that checks genitive case on the subject,
and since the T-v-V amalgam is elided in sluicing before it agrees with C, genitive case
cannot be checked on the subject.
In (), repeated as (a) below, the phi-features of the T-v-V amalgam check
nominative case of the wh-phrase kim ‘who’ in SpecTP in the embedded clause as
shown in the tree in (b). The wh-phrase then raises to the matrix SpecCP, with
nominative case checked.
() a. Hasan sen-i biri-nin ara-dığ-ı-nı
Hasan-nom pron.sg-acc one-gen call-comp-poss.sg-acc
söyle-di-∅. Kim-∅ ki?
tell-pst-sg who-nom prt
‘Hasan said that someone called you. Who, then?’
b. CP

kimi C⬘
+NOM
TP C

kimi T⬘
+NOM
Mark TP for elision
vP T-v-V
and Spell Out

ti ...

Note that the movement of the nominative wh-subject to SpecCP cannot be a case
of scrambling. In Turkish a wh-phrase can scramble to the CP phrase.  Critically, in
contrast to sluicing, the scrambled subject of an embedded complement clause must
bear genitive case, as shown in (b) and (), and nominative case on a scrambled
subject is unacceptable, as shown in ().
() a. [TP Hasan [XP [CP Ahmet’in Ankara’ya
Hasan-nom Ahmet-gen Ankara-dat
git-tiğ-i-ni san-ıyor-∅]]].
go-comp-poss.sg-acc think-prog-sg
‘Hasan thinks Ahmet went to Ankara.’
b. Ahmet’ini [TP Hasan [XP [CP t i Ankara’ya gittiğini sanıyor.]]]

 For scrambling in Turkish, see Aygen (), Demircan (), Kornfilt (), and Kural (),
among others. For A-scrambling see Miyagawa () and references therein. For A-scrambling in Turkish,
see Aygen (). See also Saito () and Ueyama () for arguments against an A vs A-bar distinction
in scrambling. Ueyama () argues that non-reconstruction of scrambled elements cannot be evidence
for A-scrambling.
Sluicing in Turkish 269

() Kim-ini [TP Ahmet [XP [CP t i öl-düğ-ü]] -nü


who-gen Ahmet-nom die-comp-poss.sg -acc
san-ıyor-∅].
think-prog-sg
‘Who does Ahmet think died?’
() a. ∗ Ahmeti [TP Hasan [XP [CP t i Ankara’ya gittiğini sanıyor]]].
b. ∗ Kimi [TP Ahmet [XP [CP t i öldüğünü sanıyor]]].

It can therefore be concluded that attempting to explain sluicing of embedded subjects


by means of a scrambling analysis is not tenable.

. Conclusion
Sluicing structures in Turkish cannot be reduced to cleft structures. The evidence
comes from the different case requirements in clefting vs sluicing: Clefting is restricted
to argument phrases; postposition pied-piping is obligatory in sluicing, but impossible
in clefting; and multiple sluicing is grammatical, whereas multiple clefting is not.
Although the sluicing of embedded subjects seems to run counter to the
case-parallelism requirement, it can be explained in terms of a C-T-v-V amalgam
checking genitive case in a Multiple Spell-Out model. In sluicing structures, the
embedded wh-subject cannot occur in and be licensed with genitive case because the
T-v-V amalgam is deleted before agreeing with C . The sluiced wh-phrase is therefore
bound to bear the nominative case assigned by a T-v-V amalgam.
References
Adams, Perng Wang () ‘Sluicing in Chinese: The beauty of silence’. Unpublished manu-
script, University of Delaware.
() ‘Sluicing in Javanese’. Paper presented at the  Western Conference on Linguis-
tics (WECOL), University of Arizona.
() ‘The structure of sluicing in Mandarin Chinese’. Penn Working Papers in Linguistics
: –. [Sudha Arunachalam and Tatjana Scheffler (eds.), Proceedings of the th Pennsyl-
vania Linguistics Colloquium.]
() ‘Sluicing in Javanese: Constrained by its syntax’. Unpublished manuscript, Univer-
sity of Delaware.
Aelbrecht, Lobke () The Syntactic Licensing of Ellipsis. Amsterdam: John Benjamins.
Akar, Didar () ‘Wh-constructions in Turkish’. Unpublished MA thesis, Boǧaziçi University,
Istanbul.
Akmajian, Adrian () ‘An interpretive principle for certain anaphoric expressions’. Unpub-
lished manuscript, MIT, Cambridge, MA.
Alboiu, Gabriela () ‘Focus dependencies: Evidence from Romanian’. Unpublished manu-
script, Research Group in Linguistics, Queen’s University.
() ‘The features of movement in Romanian’. Ph.D. dissertation, University of
Manitoba.
Aldridge, Edith () ‘Nominalization and wh-movement in Seediq and Tagalog’. Language
and Linguistics : –.
() ‘Ergativity and word order in Austronesian languages’. Ph.D. dissertation, Cornell
University.
Aoun, Joseph, Norbert Hornstein, David Lightfoot, and Amy Weinberg () ‘Two types of
locality’. Linguistic Inquiry : –.
Aoun, Joseph and Yen-hui Audrey Li () Essays on the Derivational and Representational
Nature of Grammar. Boston, MA: MIT Press.
Arslan, Z. Ceyda () ‘Approaches to wh-constructions in Turkish’. Unpublished MA thesis,
Boǧaziçi University, Istanbul.
Aygen, Gülşat () ‘Extractability and the nominative case feature on tense’. Unpublished
manuscript, Harvard University.
() ‘Finiteness, case and clausal architecture’. Ph.D. dissertation, Harvard University.
() ‘Q-particle’. Journal of Linguistics and Literature : –.
Babby, Leonard H () ‘Case, prequantifiers, and discontinous agreement in Russian’. Natural
Language and Linguistic Theory : –.
Baker, C. () ‘Notes on the description of English questions: The role of an abstract question
morpheme’. Foundations of Language : –.
Baker, Mark and Ruth Kramer () ‘ “Prepositions” as case morphemes inserted at PF
in Amharic’. Paper presented at the Fifth Brussels Conference on Generative Linguistics
(BCGL): Case at the Interfaces, December –, , Brussels, Belgium.
References 271

Barss, Andrew () ‘Minimalism and asymmetric wh-interpretation’, in Roger Martin, David
Michaels, and Juan Uriagereka (eds.), Step by Step: Essays on Minimalist Syntax in Honor of
Howard Lasnik. Cambridge, MA: MIT Press, –.
Bayer, Josef () Directionality and Logical Form. Dordrecht: Kluwer.
Benincà, Paolo and Cecilia Poletto () ‘Topic, focus and V: Defining the CP-sublayers’,
in L. Rizzi (ed.), The Structure of IP and CP. The Cartography of Syntactic Structures, vol. .
New York, NY: Oxford University Press, –.
Bennis, Hans () ‘Voegwoordvariaties’ [Complementizer variations], in A. van Santen and
M. van der Wal (eds.), Taal in Tijd en Ruimte. Leiden: SNL, –.
() ‘On the interpretation of functional categories’, in Hans Bennis, Martin Everaert,
and Eric Reuland (eds.), Interface Strategies. Amsterdam: KNAW, –.
Besler, Dilek () ‘The question particle and movement in Turkish’. Unpublished MA thesis,
Boǧaziçi University, Istanbul.
Bhattacharya, Tanmoy and Andrew Simpson () ‘Argument prominence and the nature of
Superiority violations’, in Eric Reuland, Tanmoy Bhattacharya, and Giorgos Spathias (eds.),
Argument Structure. Amsterdam: John Benjamins, –.
Boeckx, Cedric and Howard Lasnik () ‘Intervention and repair’. Linguistic Inquiry :
–.
Bošković, Željko () ‘On multiple feature-checking: Multiple wh-fronting and multiple
head movement’, in Samuel Epstein and Norbert Hornstein (eds.), Working Minimalism.
Cambridge, MA: MIT Press, –.
() ‘Multiple wh-fronting’. Linguistic Inquiry : –.
() ‘Case of genitive of quantification in Russian’, in Cedric Boeckx (ed.), Agreement
Systems. Amsterdam: John Benjamins, –.
() ‘A minimalist account of genitive of quantification’, in Luka Szucsich, Gerhild
Zybatow, Uwe Junghanns, and Roland Meyer (eds.), Formal Description of Slavic Languages
. Frankfurt am Main: Peter Lang, –.
() ‘Rescue by PF deletion, traces as (non)interveners, and the that-trace effect’.
Unpublished manuscript, University of Connecticut.
Burt, Kim () ‘Some rules of non-constituent deletion’. Unpublished manuscript, MIT.
Cagri, Ilhan () ‘Minimality and Turkish relative clauses’. Ph.D. dissertation, University of
Maryland.
Caponigro, Ivano () ‘Free relatives as DPs with a silent D and a CP complement’, in
V. Samiian (ed.), Proceedings of the Western Conference on Linguistics . Fresno, CA:
California State University, –.
() ‘Free not to ask: On the semantics of free relatives and wh-words cross-
linguistically’. Ph.D. dissertation, UCLA.
() ‘The semantic contribution of wh-words and type-shifts: Evidence from free rela-
tives cross-linguistically’. Unpublished manuscript, University of Maryland.
Chao, Yuen-Ren () A Grammar of Spoken Chinese. Berkeley and Los Angeles, CA: Univer-
sity of California Press.
Cheng, Lisa () ‘On the typology of wh-questions’. Ph.D. dissertation, MIT.
Chierchia, Gennaro () ‘Anaphora and dynamic binding’. Linguistics and Philosophy :
–.
272 References

Chiu, Liching Livy () ‘A focus-movement account on Chinese multiple sluicing’. Nanzan
Linguistics: Special Issue : –.
Chomsky, Noam () Aspects of the Theory of Syntax. Cambridge, MA: MIT Press.
() ‘Remarks on nominalization’, in Roderick Jacobs and Peter S. Rosenbaum (eds.),
Readings in English Transformational Grammar. Waltham: Blaisdell-Ginn, –.
() ‘Deep structure, surface structure, and semantic interpretation’, in Steinberg and
Jakobovits (eds.) (), –.
() ‘Some empirical issues in the theory of transformational grammar’, in Paul Stanley
Peters (ed.), Goals of Linguistic Theory. Englewood Cliffs: Prentice-Hall, –.
() ‘Conditions on transformations’, in Stephen Anderson and Paul Kiparsky (eds.),
A Festschrift for Morris Halle. New York, NY: Holt, Rinehart, and Winston, –.
(a) Barriers. Cambridge, MA: MIT Press.
(b) Knowledge of Language: Its Nature, Origin, and Use. New York, NY: Praeger.
() The Minimalist Program. Cambridge, MA: MIT Press.
() ‘Derivation by phase’. MIT Occasional Papers in Linguistics, no. . Cambridge, MA:
MITWPL.
() ‘Minimalist inquiries: The framework’, in Roger Martin, David Michaels, and Juan
Uriagereka (eds.), Step by Step: Essays on Minimalist Syntax in Honor of Howard Lasnik.
Cambridge, MA: MIT Press, –.
() ‘Derivation by phase’, in Michael Kenstowicz (ed.), Ken Hale: A Life in Language.
Cambridge, MA: MIT Press, –.
Chomsky, Noam and Howard Lasnik () ‘The theory of Principles and Parameters’, in
J. Jacobs, A. von Stechow, W. Sternefeld, and T. Vennemann (eds.), Syntax: An International
Handbook of Contemporary Research, vol. . Berlin: Walter de Gruyter, –.
Chung, Sandra () ‘What fronts? On the VP raising account of verb-initial order’, in
A. Carnie, S. D. Collberg, and H. Harley (eds.), Verb First: Papers on the Syntax of Verb Initial
Languages. Amsterdam: John Benjamins, –.
Chung, Sandra, William Ladusaw, and James McCloskey () ‘Sluicing and Logical Form’.
Natural Language Semantics : –.
Cinque, Guglielmo () ‘Bare quantifiers, quantified NPs and the notion of operator at
S-structure’. Rivista di Grammatica Generativa : –.
Cole, Peter and Gabriella Hermon () ‘VP raising in a VOS language’. Syntax : –.
Comorovski, Ileana () Interrogative Phrases and the Syntax–Semantics Interface. Boston,
MA: Kluwer.
() ‘Multiple wh-movement in Romanian’. Linguistic Inquiry : –.
Cooper, Robin () ‘The interpretation of pronouns’, in F. Heny and H. S. Schnelle (eds.),
Syntax and Semantics : Selections from the Third Groningen Round Table. New York, NY:
Academic Press, –.
Cornilescu, Alexandra () ‘The double-subject construction in Romanian’, in Virginia
Motapanyane (ed.), Comparative Studies in Romanian Syntax. Dordrecht: Elsevier, –.
van Craenenbroeck, Jeroen () ‘Ellipsis and accommodation. The (morphological) case of
sluicing’. Handout with presentation at MIT’s Linguistics Colloquium, April , .
() The Syntax of Ellipsis: Evidence from Dutch Dialects. New York, NY: Oxford
University Press.
References 273

van Craenenbroeck, Jeroen and Anikó Lipták () ‘Ellipsis in Hungarian and the typology of
sluicing’, in Kiyong Choi and Changguk Yim (eds.), Ellipsis in Minimalism: Proceedings of the
Seventh Seoul International Conference on Generative Grammar. Seoul: Hankook, –.
() ‘The cross-linguistic syntax of sluicing: Evidence from Hungarian relatives’. Syntax
: –.
(to appear) ‘What sluicing can do, what it can’t and in which language: On the cross-
linguistic syntax of ellipsis’, in Lisa Cheng and Norbert Corver (eds.), Diagnosing syntax.
New York, NY: Oxford University Press.
van Craenenbroeck, Jeroen and Jason Merchant (to appear) ‘Elliptical phenomena’, in
Marcel den Dikken (ed.), Cambridge Handbook of Generative Syntax. Cambridge:
Cambridge University Press.
Culicover, Peter () ‘Topicalisation, inversion and complementizers in English’, in D. Delfitto,
M. Everaert, A. Evans, and F. Stuurman (eds.), Going Romance and Beyond. Utrecht: Univer-
sity of Utrecht, –.
() Syntactic Nuts: Hard Cases, Syntactic Theory, and Language Acquisition. New York,
NY: Oxford University Press.
Culicover, Peter and Ray Jackendoff () Simpler Syntax. New York, NY: Oxford Univer-
sity Press.
Dahl, Otto Christian () ‘Focus in Malagasy and Proto-Austronesian’, in P. Geraghty,
L. Carrington and S.A. Wurm (eds.), FOCAL I: Papers from the Fourth International Confer-
ence on Austronesian Linguistics. Pacific Linguistics, –.
Dasgupta, Probal () ‘Questions and relative and complement clauses in a Bangla grammar’.
Ph.D. dissertation, New York University.
Dayal, Vaneeta () ‘Multiple wh-questions’, in Martin Everaert and Henk van Riemsdijk
(eds.), The Syntax Companion, vol. . Malden, MA: Blackwell. –.
Demircan, Ömer () ‘The rules of inversion in Turkish’, in Bengisu Rona (ed.), Current Issues
in Turkish Linguistics: Proceedings of the Fifth International Conference on Turkish Linguistics,
vol. . Ankara: Hitit Yayınevi, –.
den Dikken, Marcel () ‘Empty operator movement in Dutch imperatives’, in D. Gilbers and
S. Looyenga (eds.), Language and Cognition, vol. . Groningen: University of Groningen,
–.
den Dikken, Marcel and Anastasia Giannakidou () ‘From hell to polarity: “Aggressively
non-d-linked” wh-phrases as polarity items’. Linguistic Inquiry : –.
Dobrovie-Sorin, Carmen () ‘Clitic doubling, wh-movement, and quantification in
Romanian’. Linguistic Inquiry : –.
() The Syntax of Romanian. Berlin: Mouton de Gruyter.
Dougherty, Raymond C. () ‘Two theories of pronominalization’. Unpublished manu-
script, MIT.
Enç, Murvet () ‘Semantics of specificity’. Linguistic Inquiry : –.
Erteschik-Shir, Nomi () On the Nature of Island Constraints. Bloomington: Indiana Uni-
versity Linguistics Club.
Evans, Gareth () ‘Pronouns, quantifiers, and relative clauses’. Canadian Journal of Philoso-
phy : –.
274 References

Fanselow, Gisbert () ‘The MLC and derivational economy’, in A. Stepanov, G. Fanselow
and R. Vogel (eds.), Minimality Effects in Syntax. Berlin: Mouton de Gruyter, –.
Fanselow, Gisbert and Damir Ćavar () ‘Remarks on the economy of pronunciation’, in
Gereon Müller and Wolfgang Sternefeld (eds.), Competition in Syntax. Berlin: Mouton de
Gruyter, –.
Fanselow, Gisbert and Anoop Mahajan () ‘Towards a minimalist theory of wh-expletives,
wh-copying and successive cyclicity’, in U. Lutz, G. Müller and A. von Stechow (eds.), Wh-
Scope Marking. Amsterdam: John Benjamins, –.
Fillmore, Charles J. () ‘The case for case’, in Emmon Bach and Robert Harms (eds.), Uni-
versals in Linguistic Theory. New York: Holt, Rinehart, and Winston, –.
Fox, Danny () ‘Reconstruction, binding theory, and the interpretation of chains’. Linguistic
Inquiry : –.
Fox, Danny and Howard Lasnik () ‘Successive cyclic movement and island repair: The
difference between sluicing and VP ellipsis’. Linguistic Inquiry : –.
Franks, Steven () ‘Parametric properties of numeral phrases in Slavic’. Natural Language
and Linguistic Theory : –.
() Parameters of Slavic Morphosyntax. New York, NY: Oxford University Press.
Fukaya, Teruhiko () ‘Island (in)sensitivity in Japanese sluicing and stripping’, in Gina
Garding and Mimu Tsujimura (eds.), Proceedings of the nd West Coast Conference on
Formal Linguistics (WCCFL). Somerville, MA: Cascadilla Press, –.
() ‘Sluicing and stripping in Japanese and some implications’ Ph.D. dissertation, USC.
[Available at http://digitallibrary.usc.edu/search/controller/view/usctheses-m.html]
Fukaya, Teruhiko and Hajime Hoji () ‘Stripping and sluicing in Japanese and some implica-
tions’, in Sonya Bird, Andrew Carnie, Jason Haugen, and Peter Norquest (eds.), Proceedings
of the th West Coast Conference on Formal Linguistics (WCCFL). Somerville, MA: Cas-
cadilla Press, –.
Fukui Naoki () ‘A theory of category projection and its applications’. Ph.D. disserta-
tion, MIT.
Fukui, Naoki and Hiromu Sakai () ‘The visibility guideline for functional categories: Verb
raising in Japanese and related issues’. Lingua : –.
Gengel, Kirsten () ‘Focus and ellipsis: A generative analysis of pseudo-gapping and other
elliptical structures’. Ph.D. dissertation, University of Stuttgart.
Göksel, Aslı and A. Sumru Özsoy () ‘Is there a focus position in Turkish?’, in A. Göksel
and C. Kerslake (eds.), Studies on Turkish: Proceedings of the Ninth International Conference
on Turkish Linguistics. Wiesbaden: Horrossowitz, –.
Grebenyova, Lydia () ‘Multiple interrogatives: Syntax, semantics, and learnability’. Ph.D.
dissertation, University of Maryland.
Grohmann, Kleanthes () ‘Prolific peripheries: A radical view from the left’. Ph.D. disserta-
tion, University of Maryland.
Groos, Anneke and Henk van Riemsdijk () ‘Matching effects in free relatives: A parameter
of core grammar’, in A. Belletti, L. Brandi, and L. Rizzi (eds.), Theory of Markedness in Gener-
ative Grammar: Proceedings of the  GLOW Conference. Pisa: Scuola Normale Superiore,
–.
Grosu, Alexander () Three Studies in Locality and Case. London: Routledge.
References 275

Grosu, Alexander and Fred Landman () ‘Strange relatives of the third kind’. Natural Lan-
guage Semantics : –.
Guéron, Jacqueline and Robert May () ‘Extraposition and Logical Form’. Linguistic Inquiry
: –.
Haig, J.H. () ‘Shadow pronoun deletion in Japanese’. Linguistic Inquiry : –.
Hale, Ken () ‘On the Dagur object relative: Some comparative notes’. Journal of East Asian
Linguistics : –.
Halle, Morris and Alec Marantz () ‘Distributed Morphology and the pieces of inflection’, in
Kenneth Hale and Samuel Jay Keyser (eds.), The View from Building : Essays in Linguistics
in Honor of Sylvain Bromberger. Cambridge, MA: MIT Press, –.
() ‘Some key features of Distributed Morphology’, in Andrew Carnie, Heidi Harley,
Tony Bures (eds.), MIT Working Papers in Linguistics : Papers on Phonology and Morphol-
ogy. Cambridge, MA: MITWPL, –.
Hankamer, Jorge () Deletion in Coordinate Structures. New York, NY: Garland.
Hankamer, Jorge and Ivan Sag () ‘Deep and surface anaphora’. Linguistic Inquiry : –.
Harada, Kazuko () ‘Constraints on WH-Q binding’. Studies in Descriptive and Applied
Linguistics : –.
Harada, Naomi () ‘Licensing PF-visible formal features: A linear algorithm and case-
related phenomena in PF’. Ph.D. dissertation, University of California, Irvine.
Hartman, Jeremy () ‘Focus, deletion, and identity: Investigations of ellipsis in English’. BA
thesis, Harvard University.
Hartman, Jeremy and Ruixi Ressy Ai () ‘A focus account of swiping’. Unpublished manu-
script, Harvard University.
Hasegawa, Hiroshi () ‘Swiping involves preposition stranding, not pied-piping’. Handout
with a presentation to the Thirtieth Annual Meeting of Generative Linguistics of the Old
World, University of Tromso.
Heim, Irene () ‘The semantics of definite and indefinite noun phrases’. Ph.D. dissertation,
University of Massachusetts at Amherst.
() ‘E-type pronouns and donkey anaphora’. Linguistics and Philosophy : –.
Heim, Irene and Angelika Kratzer () Semantics in Generative Grammar. Malden, MA:
Blackwell.
Hiemstra, Inge () ‘Some aspects of wh-questions in Frisian’. NOWELE : –.
Hinskens, Frans () ‘Dialect leveling in Limburg: Structural and sociolinguistic aspects’.
Ph.D. dissertation, University of Nijmegen.
Hiraiwa, Ken () ‘On nominative–genitive conversion’, in Elena Guerzoni and Ora
Matushansky (eds.), MIT Working Papers in Linguistics : A View from Building E. Cam-
bridge, MA: MITWPL, –.
() ‘Eliminating PBC: Multiple spell-out, scrambling and the edge operation’. Penn
Working Papers in Linguistics : –. [Elsi Kaiser and Sudha Arunachalam (eds.), Pro-
ceedings of the th Annual Penn Linguistics Colloquium.]
Hiraiwa, Ken and Shinichiro Ishihara () ‘Missing links: Cleft, sluicing, and “no da” con-
struction in Japanese’, in Tania Ionin, Heejeong Ko, and Andrew Nevins (eds.), MIT Working
Papers in Linguistics : Proceedings of the nd HUMIT Student Conference in Language
Research (HUMIT). Cambridge, MA: MITWPL, –.
276 References

Hoekstra, Eric () ‘Overtollige voegwoorden en de volgorde of + interrogativum/relativum’


[Spurious complementizers and the order if + interrogative/relative pronoun]. De Nieuwe
Taalgids , –.
() ‘On D-pronouns and the movement of topic features’. Folia linguistica : –.
Hoekstra, Eric and C. Jan-Wouter Zwart () ‘De structuur van de CP. Functionele projecties
voor topics en vraagwoorden in het Nederlands’ [The structure of CP. Functional projections
for topics and question words in Dutch]. Spektator : –.
() ‘Weer functionele projecties’ [Functional projections revisited]. Nederlandse
Taalkunde : –.
Hoekstra, Jarich () ‘The split CP hypothesis and the Frisian complementizer system’.
Unpublished manuscript, Frisian Academy.
Höhle, Tilman () ‘German w. . . w-constructions’. Unpublished manuscript, University of
Tübingen.
Hoji, Hajime () ‘Theories of anaphora and aspects of Japanese syntax’. Unpublished man-
uscript, USC.
Hornstein, Norbert and Amy Weinberg () ‘Superiority and generalized binding’, in
J. McDonough and B. Plunkett (eds.), Proceedings of the North Eastern Linguistic Society
. –.
Huang, C.-T. James () ‘On the distribution and reference of empty pronouns’. Linguistic
Inquiry : –.
Hudson, Richard () ‘Trouble on the left periphery.” Lingua : –.
İnce, Atakan () ‘Island-sensitive sluicing in Turkish’, in Leah Bateman and Cherlon Ussery
(eds.), Proceedings of the Thirty-Fifth Annual Meeting of the North East Linguistic Society
(NELS). Amherst, MA: GLSA, –.
() ‘Dimensions of ellipsis: Investigations in Turkish’. Ph.D. Dissertation, University
of Maryland.
Inoue, Kazuko () Henkei Bunpoo to Nihongo [Transformational Grammar and Japanese].
Tokyo: Taishukan.
() Nihongo-no Bunpoo Kisoku [Grammatical Rules in Japanese]. Tokyo: Taishukan.
Jackendoff, Ray S. () ‘A note on selectional restrictions’. Unpublished manuscript, MIT.
Jacobson, Pauline () ‘Paycheck pronouns, Bach-Peters sentences, and variable-free seman-
tics’. Natural Language Semantics : –.
Johnson, Kyle () ‘What VP ellipsis can do, and what it can’t, but not why’, in Mark Baltin
and Chris Collins (eds.) The Handbook of Contemporary Syntactic Theory. Oxford: Black-
well, –.
(n.d.) ‘In search of the English middle field’. Unpublished manuscript, University of
Massachusetts at Amherst.
Kachru, Yamuna () Hindi. Amsterdam: John Benjamins.
Kalin, Laura () ‘News about the no-phrase: Specificational pseudoclefts in Malagasy’, in
S. Chung, D. Finer, I. Paul, and E. Potsdam (eds.), The Proceedings of AFLA , –.
[Available at http://westernlinguistics.ca/afla/]
Keenan, Edward L. () ‘Remarkable subjects in Malagasy’, in Charles Li (ed.), Subject and
Topic. New York, NY: Academic Press, –.
References 277

Kennedy, Christopher and Jason Merchant () ‘Attributive comparative deletion’. Natural
Language and Linguistic Theory : –.
Kennelly, Sarah D. () ‘Turkish subordination: [−CP, −Tense, +Case]’, in A. Konrot (ed.)
Modern Studies in Turkish Linguistics: Proceedings of the th International Conference on
Turkish Linguistics Eskişehir, Turkey: Anadolu University, –.
() ‘Nonspecific external arguments in Turkish’. Dilbilim Araştırmaları, –.
Kim, Jeong-Seok () ‘Syntactic focus movement and ellipsis: A minimalist approach’. Ph.D.
dissertation, University of Connecticut, Storrs.
Kimura, H. () ‘A wh-in-situ strategy for sluicing’. English Linguistics : –.
King, Harold V. () ‘On blocking the rules for contraction in English’. Linguistic Inquiry
: –.
Kiss, Katalin () Discourse Configurational Languages. Oxford: Oxford University Press.
() The Syntax of Hungarian. Cambridge: Cambridge University Press.
Kizu, Mika () ‘Sluicing in wh-in-situ languages’, in Kora Singer, Randall Eggert, and Gre-
gory Anderson (eds.), CLS : Papers from the Main Session. Chicago, IL: Chicago Linguistic
Society, University of Chicago, –.
Kornfilt, Jaklin () ‘Case-marking, agreement and empty categories in Turkish’. Ph.D. dis-
sertation, Harvard University.
() ‘Constraints on free relative clauses in Turkish’, in A. Alexiadou (ed.) FAS Papers in
Linguistics . Berlin: FAS, –.
() ‘Scrambling and incorporation in Turkish’, in A. Alexiadou, N. Fuhrhop, P. Law,
and S. Löhken (eds.), FAS Papers in Linguistics . Berlin: FAS, –.
() ‘Subjects and their case in Turkish/Turkic embeddings’. Paper presented at the
Workshop on Altaic Languages, MIT.
() ‘Subject case in Turkish nominalized clauses’, in U. Junghanns and L. Szucsich
(eds.), Syntactic Structures and Morphological Information. Berlin/New York, NY: Mouton
de Gruyter, –.
() ‘Agr in Turkish as an expression of categorial features’, in M. Kelepir and B. Öztürk
(eds.), MIT Working Papers in Linguistics : Proceedings of the Second Workshop on Altaic
Formal Linguistics. Cambridge, MA: MITWPL, –.
Kratzer, Angelika () ‘Scope or pseudo-scope? Are there wide-scope indefinites?’, in Susan
Rothstein (ed.), Events and Grammar. Dordrecht: Kluwer, –.
Krause, Cornelia () ‘On reduced relatives with genitive subjects’. Ph.D. dissertation, MIT.
Kuno, Susumo () The Structure of the Japanese Language. Cambridge, MA: MIT Press.
Kurafuji, Takeo () ‘Japanese pronouns in dynamic semantics: The null/overt contrast’. Ph.D.
dissertation, State University of New Jersey, New Brunswick.
Kural, Murat () ‘V-to(-I-to)-C in Turkish’, in Felipo Beghelli and Murat Kural (eds.), UCLA
Occasional Papers in Linguistics : Recent Papers in Syntax, Semantics, and Computational
Linguistics, Los Angeles: UCLA, –.
Kuroda, Sige-Yuki () ‘Generative grammatical studies in the Japanese language’. Ph.D.
dissertation, MIT.
() ‘English relativization and certain related problems’. Language : –.
Kuwabara, Kazuki () ‘Multiple wh-phrases in elliptical clauses and some aspects of clefts
with multiple foci’, in Masatoshi Koizumi, Masayuki Oishi, and Uli Sauerland (eds.), MIT
278 References

Working Papers in Linguistics : Formal approaches to Japanese linguistics . Cambridge,


MA: MITWPL, –.
Kuwabara, Kazuki () ‘On the properties of truncated clauses in Japanese’, in K. Inoue
(ed.) Researching and Verifying an Advanced Theory of Human Language: Explanation of the
Human Faculty for Constructing and Computing Sentences on the Basis of Lexical Conceptual
Features. Chiba: Kanda University of International Studies, –.
Lahiri, Utpal () Questions and Answers in Embedded Contexts. Oxford: Oxford University
Press.
Lakoff, George () ‘On derivational constraints’, in Robert Binnick, Alice Davison, Georgia
Green, and Jerry Morgan (eds.), Papers from the th Regional Meeting of the Chicago Linguis-
tic Society. Chicago: CLS, –.
() ‘On generative semantics’, in Steinberg and Jakobovits (eds.) (), –.
Lakoff, George and John Robert Ross (in preparation) ‘Abstract Syntax’.
Lamontagne, Greg and Lisa Travis () ‘The syntax of adjacency’, in Megan Crowhurst (ed.),
Proceedings of the Sixth West Coast Conference on Formal Linguistics (WCCFL). Stanford,
CA: CSLI Publications, –.
Lappin, Shalom () ‘The interpretation of ellipsis’, in Shalom Lappin (ed.) The Handbook of
Contemporary Semantic Theory. Oxford: Blackwell, –.
Larson, Bradley () ‘Swiping and decomposed Merge’. Unpublished manuscript, University
of Maryland.
Lasnik, Howard () ‘On feature strength: Three minimalist approaches to overt movement’.
Linguistic Inquiry : –.
() ‘When can you save a structure by destroying it?’, in Minjoo Kim and Uri Strauss
(eds.), Proceedings of the st Conference of the North East Linguistic Society (NELS), vol. .
Amherst, MA: GLSA, –.
() ‘Movement, ellipsis, and their interaction: Implications for syntactic theory’.
Lectures given at Leiden University.
() ‘Multiple sluices in English?’ Presentation given at UCLA.
Lasnik, Howard and Mamoru Saito () Move α: Conditions on its Application and Output.
Cambridge, Mass.: MIT Press.
Law, Paul () ‘The syntactic structure of the cleft construction in Malagasy’. Natural Lan-
guage and Linguistic Theory : –.
Lees, Robert B. () ‘Turkish nominalization and a problem of ellipsis’. Foundations of Lan-
guage : –.
Legate, Julie () ‘Morphological and abstract case’. Linguistic Inquiry : –.
Levin, Lori () ‘Sluicing: A lexical interpretation procedure’, in Joan Bresnan (ed.) The Mental
Representation of Grammatical Relations. Cambridge: MIT Press, –.
Li, Audrey Y.-H. () ‘Indefinite WH in Chinese’. Journal of East Asian Linguistics : –.
() ‘Born empty’. Paper presented at the th Annual Meeting of the Linguistic Society
of America, Pittsburgh, PA.
Li, Audrey Y.-H. and Ting-Chi Wei (to appear) ‘Ellipsis’, in James C.-T. Huang, Audrey Y.-H.
Li, and Andrew Simpson (eds.), The Handbook of Chinese Linguistics. Oxford: Blackwell.
Li, Charles N. and Sandra A. Thompson (). A Functional Reference Grammar of Mandarin
Chinese. Berkeley and Los Angeles, CA: University of California Press.
References 279

Lipták, Anikó () ‘On the syntax of wh-items in Hungarian’. Ph.D. dissertation, Leiden
University.
Lobeck, Ann () Ellipsis: Functional Heads, Licensing, and Identification. Oxford: Oxford
University Press.
Longobardi, Giuseppe () ‘Extraction from NP and the proper notion of head government’,
in Alessandra Giorgi and Giuseppe Longobardi (eds.), The Syntax of Noun Phrases: Config-
uration, Parameters, and Empty Categories. Cambridge: Cambridge University Press, –.
MacLaughlin, Dawn () ‘Wh-movement in Malagasy: An extraction asymmetry’, in
A. Akinlabi (ed.), Theoretical Approaches to African Linguistics. Trenton, NJ: Africa World
Press, –.
Mahajan, Anoop () ‘The A/A distinction and movement theory’. Ph.D. dissertation, MIT.
Malzac, R. P. () Grammaire Malgache. Paris: Société d’Editions Géographiques, Maritimes
et Coloniales.
Manetta, Emily () ‘Wh-expletives in Hindi-Urdu: the vP phase’. Linguistic Inquiry : –.
Massam, Diane () ‘VSO and VOS: Aspects of Niuean word order’, in Andrew Carnie and
Eithne Guilfoyle (eds.), The Syntax of Verb Initial Languages. Oxford: Oxford University
Press, –.
Massam, Diane and Carolyn Smallwood () ‘Essential features of predication in Niuean and
English’, in K. Kusumoto (ed.), The Proceedings of the th Meeting of the North East Linguistic
Society (NELS). Amherst, MA: GLSA, –.
McCawley, James D. () ‘The role of semantics in grammar’, in Emmon Bach and Robert
Harms (eds.), Universals in Linguistic Theory. New York: Holt, Rinehart, and Winston,
–.
McCloskey, James () Transformational Syntax and Model Theoretic Semantics: A Case Study
in Modern Irish. Dordrecht: Reidel.
() ‘Resumptive pronouns, A -binding, and levels of representation in Irish’, in
R. Hendrick (ed.). The Syntax of the Modern Celtic Languages. Syntax and Semantics . New
York, NY: Academic Press, –.
McDaniel, Dana () ‘Partial and multiple wh-movement’. Natural Language and Linguistic
Theory : –.
Merchant, Jason () ‘Pseudosluicing: Elliptical clefts in English and Japanese’, in
A. Alexiadou N. Fuhrhop, P. Law, and U. Kleinhenz (eds.), ZAS Working Papers in Linguistics
. Berlin: Zentrum fur Allgemeine Sprachwissenschaft –.
() ‘The syntax of silence: sluicing, islands, and identity in ellipsis’. Ph.D. dissertation,
University of California, Santa Cruz.
() The Syntax of Silence: Sluicing, Islands, and the Theory of Ellipsis. Oxford: Oxford
University Press.
() ‘Swiping in Germanic’, in C. Jan-Wouter Zwart and Werner Abraham (eds.), Studies
in Comparative Germanic Syntax. Amsterdam: John Benjamins, –.
() ‘Fragments and ellipsis’. Linguistics and Philosophy : –.
() ‘Sluicing’, in M. Everaert, H. van Riemsdijk, R. Goedemans, and B. Hollebrandse
(eds.), The Blackwell Companion to Syntax. Malden: Blackwell, –.
() ‘Variable island repair under ellipsis’, in Kyle Johnson (ed.) Topics in Ellipsis.
Cambridge: Cambridge University Press, –.
280 References

Merchant, Jason () ‘A taxonomy of elliptical repair’. Handout from SyntaxFest , Indiana
University. [Available at http://www.indiana.edu/∼lingdept/SyntaxFest/iu..ell.repair.pdf]
(to appear a) ‘Diagnosing ellipsis’, in Lisa Cheng and Norbert Corver (eds.) Diagnosing
Syntax. Oxford: Oxford University Press.
(to appear b) ‘Ellipsis’, in Artemis Alexiadou, Tibor Kiss, and Miriam Butt (eds.), Hand-
book of Contemporary Syntax, nd edition. Berlin: Walter de Gruyter.
Miyagawa, Shigeru () Structure and Case Marking in Japanese. New York, NY: Acad-
emic Press.
() ‘LF case-checking and the Minimal Link Condition’, in Colin Phillips (ed.),
MIT Working Papers in Linguistics : Papers on Case and Agreement II. Cambridge, MA:
MITWPL, –.
() ‘The EPP, scrambling, and wh-in-situ’, in Michael Kenstowicz (ed.), Ken Hale:
A Life in Language. Cambridge, MA: MIT Press, –.
Motapanyane, Virginia () ‘Focus, checking theory, and fronting strategies in Romanian’.
Studia Linguistica : –.
Mukai, Emi () ‘Movement in Japanese’. Unpublished manuscript, USC.
Munaro, Nicola () ‘On some differences between exclamative and interrogative wh-phrases
in Bellunese: further evidence for a split-CP hypothesis’. Unpublished manuscript, University
of Padua.
Nishigauchi, Taisuke () ‘ “Multiple sluicing” in Japanese and the functional nature of
wh-phrases’. Journal of East Asian Linguistics : –.
() Ronri Koozoo to Bunpoo Riron—Nitieigo no WH Gensyoo [Logical Structure and the
Theory of Grammar—WH Phenomena in Japanese and English]. Tokyo: Kurosio.
Nishiyama, Kunio () ‘Sluicing without wh-movement’, in Luis López and Natàlia Díaz-
Insensé (eds.), Cornell Working Papers in Linguistics, vol. . New York, NY: Cornell
University, –.
Nishiyama, Kunio, John Whitman, and Eun-Young Yi () ‘Syntactic movement of overt wh-
phrases in Japanese and Korean’, in Noriko Akatsuka, Shoichi Iwasaki, and Susan Strauss
(eds.), Japanese/Korean Linguistics, vol. . Stanford, CA: CSLI Publications, –.
Noguchi, Tohru () ‘The role of syntactic categories in anaphora’. Ph.D. dissertation,
University of Massachusetts at Amherst.
Nunes, Jairo () Linearization of Chains and Sideward Movement. Cambridge, MA:
MIT Press.
Ochi, Masao () ‘Move F and ga/no conversion in Japanese’. Journal of East Asian Linguistics
: –.
Özsoy, A. Sumru () ‘A -dependencies in Turkish’, in B. Rona (ed.), Current Issues in Lin-
guistics: Proceedings of the Fifth International Conference on Turkish Linguistics. Ankara: Hitit
Yayınevi, –.
() ‘On “small” clauses, other “bare” verbal complements and feature checking in Turk-
ish’, in Eser Erguvanlı Taylan (ed.), The Verb in Turkish Amsterdam: John Benjamins, –.
Paul, Ileana () ‘Concealed pseudo-clefts’. Lingua : –.
Pearson, Matt () ‘The clause structure of Malagasy: A minimalist account’. Ph.D. disserta-
tion, UCLA.
References 281

() ‘The Malagasy subject/topic as an A element’. Natural Language and Linguistic


Theory : –.
Pesetsky, David () ‘Wh-in-situ: Movement and unselective binding’, in Eric Reuland
and Alice ter Meulen (eds.), The Representation of (In)definiteness. Cambridge, MA: MIT
Press, –.
() Phrasal Movement and its Kin. Cambridge, MA: MIT Press.
() ‘Russian case morphology and the syntactic categories’. Handout, MIT.
du Plessis, Hans () ‘Wh movement in Afrikaans’. Linguistic Inquiry : –.
Poletto, Cecilia and Jean-Yves Pollock () ‘On the left periphery of Romance interrogatives’.
Abstract for GLOW , April –, , Meertens Institute, Amsterdam.
Pope, Emily () ‘Answers to yes-no questions’. Linguistic Inquiry : –.
Postal, Paul M. () Cross-Over Phenomena: A Study in the Grammar of Coreference. York-
town Heights, NY: IBM Scientific Report.
() ‘On some rules that are not successive-cyclic’. Linguistic Inquiry : –.
() Three Investigations of Extraction. Cambridge, MA: MIT Press.
Potsdam, Eric () ‘The cleft structure of Malagasy wh-questions’, in Hans-Martin Gärtner,
Paul Law, and Joachim Sabel (eds.), Clause Structure and Adjuncts in Austronesian Languages.
Berlin: Mouton de Gruyter, –.
() ‘Malagasy sluicing and its consequences for the identity requirement on ellipsis’.
Natural Language and Linguistic Theory : –.
Rackowski, Andrea and Lisa Travis () ‘V-initial languages: X or XP movement and adver-
bial placement’, in Andrew Carnie and Eithne Guilfoyle (eds.), The Syntax of Verb Initial
Languages. Oxford: Oxford University Press, –.
Reich, Ingo (to appear) ‘Ellipsis’, in Klaus von Heusinger, Claudia Maienborn, and Paul Portner
(eds.) Semantics: An International Handbook of Natural Language Meaning, vol. . Berlin,
Mouton de Gruyter.
Reinhart, Tanya () ‘A second Comp position’, in A. Belletti, L. Brandi, and L. Rizzi (eds.),
Theory of Markedness in Generative Grammar: Proceedings of the  GLOW Conference.
Pisa: Scuola Normale Superiore, –.
() ‘Wh-in-situ: who vs. which N’. Unpublished manuscript, Tel Aviv University.
() ‘Interpreting wh-in-situ’. Unpublished manuscript, Tel Aviv University.
() ‘Elliptic conjunctions—non-quantificational LF’, in Asa Kasher (ed.), The Chom-
skyan Turn, –. Malden, MA: Blackwell.
() ‘Wh-in-situ in the framework of the minimalist program’. Natural Language Seman-
tics : –.
Richards, Norvin () ‘What moves where when in which language?’. Ph.D. dissertation, MIT.
() Movement in Language: Interactions and Architectures. Oxford: Oxford Univer-
sity Press.
van Riemsdijk, Henk (a) A Case Study in Syntactic Markedness: The Binding Nature of
Prepositional Phrases. Dordrecht: Foris Publications.
(b) ‘On the diagnosis of wh movement’, in Samuel J. Keyser (ed.), Transformational
Studies in European languages. Cambridge, MA: MIT Press, –.
Rizzi, Luigi () ‘Null objects in Italian and the theory of Pro’. Linguistic Inquiry : –.
() Relativized Minimality. Cambridge, MA: MIT Press.
282 References

Rizzi, Luigi () ‘The fine structure of the left periphery’, in Liliane Haegeman (ed.), Elements
of Grammar. Dordrecht: Kluwer, –.
Rodman, Robert () ‘On the stranding of prepositions in English’, in George Bedell (ed.)
Explorations in Syntactic Theory. UCLA Papers in Syntax . Los Angeles: Department of
Linguistics, UCLA, –.
Rodrigues, Cilene, Andrew I. Nevins, and Luis Vicente () ‘Cleaving the interactions
between sluicing and P-stranding’, in Danièle Torck and W. Leo Wetzels (eds.), Romance
Languages and Linguistic Theory . Amsterdam: John Benjamins, –.
Romero, Maribel () ‘Focus and reconstruction effects in wh-phrases’. Ph.D. dissertation,
University of Massachusetts at Amherst.
Rosen, Carol () ‘Guess what about?’, in A. Ford, J. Reighard and R. Singh (eds.), Papers from
the th Meeting of the North Eastern Linguistic Society (NELS). Montréal: Montréal Working
Papers in Linguistics, –.
Rosenbaum, Peter S. () The Grammar of English Predicate Complement Constructions.
Cambridge, MA: MIT Press.
Ross, John Robert () ‘Constraints on variables in syntax’. Ph.D. Dissertation, MIT.
() ‘Guess who?’, in Robert I. Binnick, Alice Davison, Georgia M. Green, and Jerry
L. Morgan (eds.), Papers from the th Regional Meeting of the Chicago Linguistic Society.
Chicago: CLS, –.
Rudin, Catherine () ‘On multiple questions and multiple wh fronting’. Natural Language
and Linguistic Theory : –.
Sabel, Joachim () ‘Malagasy as an optional multiple wh-fronting language’, in Cedric
Boeckx and Kleanthes Grohmann (eds.), Multiple Wh-Fronting. Amsterdam: John
Benjamins, –.
Sadakane, Kumi and Masatoshi Koizumi () ‘On the nature of the “dative” particle ni in
Japanese’. Linguistics : –.
Saito, Mamoru () ‘A derivational approach to the interpretation of scrambling chains’.
Lingua : –.
Sakai, Hiromu () Review of Kim () ‘Syntactic focus movement and ellipsis: A Mini-
malist approach’. Glot International : –.
Schütze, Carson () ‘On the nature of default case’. Syntax : –.
Schütze, Carson T. and Edward Gibson () ‘Argumenthood and English prepositional
phrase attachment’. Journal of Memory and Language : –.
Shimoyama, Junko () ‘On “sluicing” in Japanese’. Unpublished manuscript, University of
Massachusetts at Amherst.
Simpson, Andrew ()Wh-Movement and the Theory of Feature-Checking. Amsterdam: John
Benjamins.
Simpson, Andrew and Tanmoy Bhattacharya () ‘Obligatory overt wh-movement in a wh-
in-situ language’. Linguistic Inquiry : –.
Srivastav, Vaneeta () ‘Subjacency effects at LF: The case of Hindi WH’. Linguistic Inquiry
: –.
Steinberg, Danny D. and Leon Jakobovits (eds.) () Semantics: An Interdisciplinary Reader
in Philosophy, Linguistics and Psychology. Cambridge: Cambridge University Press.
References 283

Stjepanović, Sandra () ‘Extraction of adjuncts out of NPs’. Paper presented at the Compar-
ative Slavic Morphosyntax Workshop, June –, , Spencer, IN.
() ‘What do second position cliticization, scrambling, and multiple wh fronting have
in common?’. Ph.D. dissertation, University of Connecticut, Storrs.
() ‘Multiple wh-fronting in Serbo-Croatian matrix questions and the matrix sluicing
construction’, in Cedric Boeckx and Kleanthes K. Grohmann (eds.), Multiple Wh-Fronting.
Amsterdam: John Benjamins, –.
() ‘P-stranding under sluicing in a non-P-stranding language?’. Linguistic Inquiry
: –.
(in preparation) ‘Genitive of quantification in Slavic: A view from ellipsis’. Manuscript,
West Virginia University.
Sturm, Arie () ‘Over functionele projecties’ [On functional projections]. Nederlandse
Taalkunde : –.
Takahashi, Daiko () ‘Sluicing in Japanese’. Journal of East Asian Linguistics : –.
Takami, Ken-Ichi () Preposition Stranding: From Syntactic to Functional Analyses. Berlin:
Mouton de Gruyter.
Toosarvandani, Maziar () ‘Wh-movement and the syntax of sluicing’. Journal of Linguistics
: –.
Travis, Lisa () ‘VP-, D -movement languages: Malagasy in Wonderland’. Paper presented
at the th Meeting of the Austronesian Formal Linguistics Association, Zentrum fur Allge-
meine Sprachwissenschaft, Berlin.
Tsai, Wei-Tian Dylan () ‘On economizing the theory of A-bar dependencies’. Ph.D. disser-
tation, MIT.
Ueyama, Ayumi () ‘Two types of dependency’. Ph.D. dissertation, USC. Distributed by
GSIL publications, USC, Los Angeles. [Available at http://www.gges.org/hoji/research/hp-
Ayumi.cgi]
() ‘Two types of scrambling constructions in Japanese’, in Andrew Barss (ed.),
Anaphora: A Reference Guide. Malden, MA: Blackwell, –.
Urbina, J. O. de () ‘Operator feature percolation and clausal Pied Piping’, in Lisa Cheng and
Hamida Demirdash (eds.), MIT Working Papers in Linguistics : Papers on Wh-Movement.
Cambridge, MA: MITWPL, –.
Uriagereka, Juan () ‘Multiple spell-out’, in Samuel Epstein and Norbert Hornstein (eds.),
Working Minimalism. Cambridge, MA: MIT Press, –.
Wang, Chyan-An Arthur () ‘On sluicing in Mandarin Chinese’. Masters thesis, National
Tsing Hua University.
Wang, Chyan-An Arthur and Hsiao-Hung Iris Wu () ‘Sluicing and focus movement in
wh-in-situ languages’. Penn Working Papers in Linguistics : –. [Aviad Eilam, Tatjana
Scheffler, and Joshua Tauberer (eds.), Proceedings of the th Annual Penn Linguistics Collo-
quium]
Wei, Ting-Chi () ‘Predication and sluicing in Mandarin Chinese’. Ph.D. dissertation,
National Kaohsiung Normal University.
() ‘Some notes on sloppy identity in Mandarin sluicing’. Concentric: Studies in Lin-
guistics : –.
284 References

Wei, Ting-Chi () ‘Island repair effects of the Left Branch Condition in Mandarin Chinese’.
Journal of East Asian Linguistics : –.
Winkler, Susanne () Ellipsis and Focus in Generative Grammar. Berlin: Mouton de Gruyter.
Xu, Liejiong () ‘Free empty category’. Linguistic Inquiry : –.
Zanuttini, Rafaella and Paul Portner () ‘Exclamative clauses: At the syntax–semantics
interface’. Language : –.
Zlatić, Larisa () ‘Syntactico-semantic approach to binding: Evidence from Serbian’. Pre-
sented at First LINGUIST Electronic Conference on Geometric and Thematic Structure in
Binding.
Author Index
Adams, Perng Wang , , , , , Cornilescu, Alexandra 
,  Cooper, Robin 
Aelbrecht, Lobke  van Craenenbroeck, Jeroen –,  n,
Ai, Ruixi Ressy   n, , ,  n,  n, ,  n, 
Akar, Didar  n Culicover, Peter , 
Akmajian, Adrian , 
Alboiu, Gabriela ,  Dahl, Otto Christian 
Aldridge, Edith  Dayal, Vaneeta  n and n
Aoun, Joseph  n,  den Dikken, Marcel , 
Arslan, Z. Ceyda  n Dobrovie-Sorin, Carmen –, 
Aygen, Gülžat  n,  n,  n, Dougherty, Raymond C. 
 n
Enç, Murvet  n
Babby, Leonard H.  Erteschik-Shir, Nomi , 
Baker, C. , Evans, Gareth 
Baker, Mark  n
Barrs, Andrew  n Fanselow, Gisbert 
Benincà, Paolo  Fillmore, Charles J.  n
Bennis, Hans ,  n Fox, Danny , , , –, , , ,
Besler, Dilek  n , , , –, 
Bhattacharya, Tanmoy  n,  n, , Franks, Steven , , 
, ,  n,  Fukaya, Teruhiko  n,  n, ,  n,
Boeckx, Cedric ,  , , , , , , , , , ,
Bošković, Željko –, ,  
Burt, Kim  n Fukui Naoki , , ,  n

Cagri, Ilhan  n Gengel, Kirsten 


Caponigro, Ivano  n Gibson, Edward  n
Ćavar, Damir  Giannakidou, Anastasia 
Chao, Yuen-Ren  Göksel, Aslı  n
Cheng, Lisa –,  n,  n Grebenyova, Lydia 
Chierchia, Gennaro  Grohmann, Kleanthes 
Chomsky, Noam , , , ,  n, , , Groos, Anneke 
, , , , ,  n,  Grosu, Alexander , 
Chung, Sandra , ,  n, , , , Guéron, Jacqueline 
 n, –, , , , 
Cinque, Guglielmo  Halle, Morris , 
Cole, Peter  Hankamer Jorge , 
Comorovski, Ileana  n,  Harada, Kazuko –, 
286 Author Index

Harada, Naomi ,  Kuwabara, Kazuki , ,  n, –
Hartman, Jeremy 
Hasegawa, Hiroshi  Lahiri, Utpal  n
Heim, Irene – Lakoff, George 
Hermon, Gabriella  Lamontagne, Greg 
Hiemstra, Inge  Landman, Fred 
Hinskens, Frans Lappin, Shalom 
Hiraiwa, Ken , , , , , , , , Larson, Bradley 
, , , , , ,  Lasnik, Howard , , , –, –, ,
Hoekstra, Eric –, ,  , , , , ,  n, –, ,
Hoekstra, Jarich ,  ,  n,  n, 
Höhle, Tilman  Law, Paul –
Hoji, Hajime ,  n,  n, ,  n, Legate, Julie ,
, , , ,  n, , , , , Levin, Lori 
, , ,  Li, Audrey  n, , , 
Hornstein, Norbert  Li, Charles 
Hoyt, Frederick ,  n,  Lipták, Anikó 
Huang, C.-T. James  Lobeck, Ann , , , ,  n, 
Hudson, Richard  Longobardi, Giuseppe 

İnce, Atakan ,  n,  n MacLaughlin, Dawn 
Inoue, Kazuko  Mahajan, Anoop ,  n, 
Ishihara, Shinichiro , –, , , , Manetta, Emily  n,  n
, ,  Marantz, Alec , 
Massam, Diane 
Jackendoff, Ray ,  n,  May, Robert 
Johnson, Kyle ,  McCawley, James  n
McCloskey, James 
Kalin, Laura  McDaniel, Dana 
Keenan, Edward , ,  n Merchant, Jason , , , , , , –, ,
Kennedy, Christopher  , , , , , –, –, , ,
Kennelly, Sarah  n –, , , , ,  n, , ,
Kim, Jeong-Seok ,  n,  , , , , –, –, –, ,
Kimura, H. ,  n