This page intentionally left blank
SYMMETRY IN SYNTAX
While much has been written on asymmetric aspects of sentence ...
68 downloads
1001 Views
1MB Size
Report
This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!
Report copyright / DMCA form
This page intentionally left blank
SYMMETRY IN SYNTAX
While much has been written on asymmetric aspects of sentence structure, symmetric aspects have been largely ignored, or claimed to be non-existent. Does symmetry in syntax exist, and if it does, how do we account for it? Barbara Citko sets out to tackle these questions and offers a unified approach to a number of phenomena that have so far been studied only in isolation. Focusing on three core minimalist mechanisms, Merge, Move – and Labeling – she advances a new theory of these mechanisms, by showing that, under certain well-defined circumstances, Merge can create symmetric structures, Move can target either of two potentially moveable objects, and labels can be constructed symmetrically from the features of two objects. This book is aimed at researchers and graduate students interested in minimalist syntax, the structure of questions, relative clauses, coordination, double object constructions, and copular sentences. b arb ara c i t k o is Assistant Professor of Linguistics at the University of Washington in Seattle. Her research includes work on phrase structure, coordination, relative clauses, wh-questions and the syntax of Slavic languages.
CAMBRIDGE STUDIES IN LINGUISTICS General Editors: p. austin, j. bresnan, b. comrie, s. crain, w. dressler, c. j. ewen, r. lass, d. lightfoot, k. rice, i. roberts, s. romaine, n. v. smith
Symmetry in Syntax Merge, Move, and Labels
In this series 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129
n o m i e rt e s c hi k- s hi r: The dynamics of focus structure j o h n col e man: Phonological representations: their names, forms and powers c h ris t in a y. bethi n: Slavic prosody: language change and phonological theory b arb a ra d ancygi er: Conditionals and prediction c l a ir e l e f ebvre : Creole genesis and the acquisition of grammar: the case of Haitian creole h e in z g ie g e ri ch: Lexical strata in English k e re n r ic e : Morpheme order and semantic scope a p r il m cm ahon: Lexical phonology and the history of English m a t t h e w y. chen: Tone Sandhi: patterns across Chinese dialects g r e g o ry t . s tump: Inflectional morphology: a theory of paradigm structure j oan b y b e e : Phonology and language use l a u rie b a uer: Morphological productivity t h o m as e rns t: The syntax of adjuncts e l iz ab e t h c los s traugott and ri chard b. dasher : Regularity in semantic change m a ya h ic k mann: Children’s discourse: person, space and time across languages d ian e b l a kemore: Relevance and linguistic meaning: the semantics and pragmatics of discourse markers ian rob e rts and anna rous s ou: Syntactic change: a minimalist approach to grammaticalization d o n k a m in k ova: Alliteration and sound change in early English m ark c. b ake r: Lexical categories: verbs, nouns and adjectives carl o t a s . s mi th: Modes of discourse: the local structure of texts roch e l l e li ebe r: Morphology and lexical semantics h ol g e r d ies s el: The acquisition of complex sentences s h aron in k e las and che ryl zoll: Reduplication: doubling in morphology s u s an e d w ards : Fluent aphasia b a rb a ra d ancygi er and e ve s w e e ts er: Mental spaces in grammar: conditional constructions h e w b ae r m an, duns tan brow n and grevi lle g . c o r bett: The syntax–morphology interface: a study of syncretism m a rcu s t omali n: Linguistics and the formal sciences: the origins of generative grammar s a m u e l d . eps tei n and t. dani e l s e e ly: Derivations in minimalism p a u l d e l a cy: Markedness: reduction and preservation in phonology ye h u d a n . falk: Subjects and their properties p . h . m a t t he w s : Syntactic relations: a critical survey m ark c. b aker: The syntax of agreement and concord g il l ian catri ona ramchand: Verb meaning and the lexicon: a first phase syntax p ie t e r m u ys ken: Functional categories j u an u ria ge reka: Syntactic anchors: on semantic structuring d . rob e rt l add: Intonational phonology second edition l e on ard h . babby: The syntax of argument structure b . e l an d res her: The contrastive hierarchy in phonology d avid a d g er, dani el harbour and laure l j . wa tkins: Mirrors and microparameters: phrase structure beyond free word order n iin a n in g zhang: Coordination in syntax n e il s m it h : Acquiring phonology n in a t o p in tzi : Onsets: suprasegmental and prosodic behaviour ce d r ic b o e ckx, norbe rt horns tei n and j ai ro nuňes: Control as movement m ic h ae l is rael: The grammar of polarity: pragmatics, sensitivity, and the logic of scales m . rit a m anzi ni and leonardo m. s avoi a: Grammatical categories: variation in romance languages b arb a ra ci tko: Symmetry in syntax: Merge, move, and labels Earlier issues not listed are also available
SYMMETRY IN SYNTAX M E R G E , MOV E , AN D LA BELS
B A R B A R A CI T KO University of Washington, Seattle
cambridge university press Cambridge, New York, Melbourne, Madrid, Cape Town, Singapore, São Paulo, Delhi, Dubai, Tokyo, Mexico City Cambridge University Press The Edinburgh Building, Cambridge CB2 8RU, UK Published in the United States of America by Cambridge University Press, New York www.cambridge.org Information on this title: www.cambridge.org/9781107005556 © Barbara Citko 2011 This publication is in copyright. Subject to statutory exception and to the provisions of relevant collective licensing agreements, no reproduction of any part may take place without the written permission of Cambridge University Press. First published 2011 Printed in the United Kingdom at the University Press, Cambridge A catalogue record for this publication is available from the British Library
ISBN 978-1-107-00555-6 Hardback Cambridge University Press has no responsibility for the persistence or accuracy of URLs for external or third-party internet websites referred to in this publication, and does not guarantee that any content on such websites is, or will remain, accurate or appropriate.
Contents
Acknowledgements Abbreviations
page ix xi
1
Rationale 1.1 Introduction 1.2 Symmetry and asymmetry 1.3 Theoretical framework 1.4 The proposal 1.5 Disclaimer: other sources of symmetry? 1.6 Organization of the book
1 1 2 4 9 9 14
2
Asymmetry in syntax 2.1 Introduction 2.2 Antisymmetry Theory 2.3 Dynamic Antisymmetry Theory 2.4 Asymmetry Theory 2.5 Coordination 2.6 Double object constructions 2.7 Asymmetry in Move 2.8 Asymmetry in Labeling
17 17 17 20 23 25 29 35 39
3
Symmetry in Merge 3.1 Introduction 3.2 Symmetric Merge: Parallel Merge 3.3 Constraints on symmetric Merge 3.4 Feature checking in symmetric Merge structures 3.5 Linearization of symmetric Merge structures 3.6 Empirical support for symmetric Merge 3.6.1 Across-the-board wh-questions 3.6.2 Wh-questions with conjoined wh-pronouns
43 43 44 46 47 50 54 54 62 vii
viii
Contents 3.6.3 Right node raising 3.6.4 Gapping 3.6.5 Serial verbs 3.6.6 Free relatives 3.7 Conclusion
68 77 92 94 107
4
Symmetry in Move 4.1 Introduction 4.2 Symmetric and asymmetric passives 4.3 Polish double object constructions 4.4 Polish dative subjects 4.5 Interim summary and preview 4.6 Locality and symmetric passives 4.6.1 Word order variation 4.6.2 Movement through applicative phase edge 4.6.3 Wh-movement from applicatives 4.6.4 Quantifier raising from applicatives 4.7 Case and symmetric passives 4.7.1 Case absorption 4.7.2 Lexical case and passive movement 4.7.3 Dative intervention effects 4.7.4 Lack of multiple datives 4.7.5 Freezing and symmetric passives 4.8 Conclusion
109 109 110 115 122 125 127 127 130 135 140 144 144 146 148 151 154 161
5
Symmetric labels 5.1 Introduction 5.2 Arguments for (and against) eliminating labels 5.3 Symmetric labels 5.4 Serial verb constructions 5.5 Small clauses 5.6 Comparative correlatives 5.7 Wh-movement in correlative clauses 5.8 Conclusion
163 163 164 176 178 182 185 190 206
6
Conclusion
207
Notes to chapters References Index
212 248 268
Acknowledgements
This book is the culmination of many years of thinking, writing, rethinking, re-writing, revising and re-revising, which involved many people, places and jobs along the way. First and foremost, I would like to thank my friends, colleagues and students at the University of Washington, where this book was written. I would also like to thank the University of Washington’s Royalty Research Fund for the much needed (and appreciated) one-quarter teaching relief during the final stages of this project, and the serene Whiteley Center on San Juan Island for providing the necessary peace and quiet during many writing stages. Some of the ideas presented here were incubated and developed when I was a graduate student at Stony Brook University, a visiting student at MIT and a lecturer at the University of Utah, the University of Connecticut and Brandeis University, and I thank all of these departments for their support, hospitality and stimulating intellectual atmosphere. In particular, I would like to thank (in alphabetical order) the following people whose ideas inspired me, whose comments motivated me, and whose encouragement kept me going: Klaus Abels, Edith Aldridge, John Bailyn, Željko Bošković, Marcel den Dikken, Kat Dziwirek, Daniel Finer, Steven Franks, Martina Gračanin-Yüksek, Stephanie Harves, Julia Herschensohn, Sabine Iatridou, Ray Jackendoff, Brad Larson, Richard Larson, David Lightfoot, Terje Lohndal, Jairo Nunes, Asya Pereltsvaig, David Pesetsky, Dafina Raţiu, Henk van Riemsdijk, Catherine Rudin and Karen Zagona. Thank you all! This book would not have been possible without your feedback and support. I would also like to thank Andrew Winnard, Sarah Green and Elizabeth Davey at Cambridge University Press for their assistance throughout the entire process, two anonymous reviewers for raising many important points and forcing me to be more precise about some of the crucial claims, and Jill Lake for a wonderful job copyediting the manuscript. ix
x
Acknowledgements
Material related to this project was presented at the following conferences: FASL 18 at Cornell University, Ways of Structure Building in Vitoria-Gasteiz, GLOW 31 Workshop on Linearization in Newcastle, 83rd and 84th Annual LSA Meetings, NELS 39 at Cornell University and NELS 33 and 40 at MIT, and at colloquia at the University of Victoria, British Columbia, University of Calgary and the University of Washington. I thank the audiences at these events for useful feedback. Portions of the discussion of Parallel Merge and across-the-board whquestions in Chapter 3 draw on my dissertation and research findings published in Linguistic Inquiry, and the discussion of labels and comparative correlatives in Chapter 5 builds on a paper published in Lingua. Last but not least, I would like to thank my husband, Randy Collins, for more than I can express (in a second language), and Icarus (our Solomon Island eclectus parrot) for providing joy and distraction throughout the entire process. I dedicate this book to the memory of my parents, Krystyna and Stanisław Citko.
Abbreviations
acc appl asp cl cpr dat def dem epp exp fem foc fut fv gen hab imperf indef inf instr loc masc nact neg neut nml nom oa ob pass
accusative applicative aspect clitic comparative dative definite demonstrative Extended Projection Principle expletive feminine focus future final vowel genitive habitual imperfective (aspect) indefinite infinitive instrumental locative masculine non-active negative neuter nominalizer nominative object agreement object passive xi
xii
Abbreviations past perf pl poss pres prog refl rel sa se sg sp subj val
past (tense) perfective (aspect) plural possessive present progressive reflexive relative subject agreement se (reflexive marker) singular subject prefix subject value
1
Rationale
1.1
Introduction
While there has been a lot of research on asymmetry and antisymmetry in syntax, symmetry has been mostly ignored or claimed to be outright impossible (Kayne 1994, Di Sciullo 2002, 2005). This is somewhat surprising from a biolinguistic perspective, which seeks to integrate linguistics with the natural sciences, where symmetry is the normal state of affairs and asymmetry requires an explanation (as pointed out by Boeckx and Piattelli-Palmarini 2005, Brody 2006, Chomsky 2005, Jenkins 2000, among others). My main goal in this book is to remedy this gap by examining symmetric aspects of three fundamental syntactic mechanisms: the mechanism responsible for recursion, the mechanism responsible for displacement, and the mechanism responsible for determining the categories of syntactic objects. I look at these three mechanisms through the lens of Chomsky’s minimalist program, which takes the mechanism responsible for recursion to be External Merge (often referred to simply as Merge), the mechanism responsible for displacement to be Internal Merge (often referred to simply as Move) and the mechanism responsible for determining categories of both Merge and Move structures to be Labeling. The standard minimalist assumption is that the structures created by Merge are asymmetric (because only such structures can be linearized), that Move is asymmetric (because it ‘privileges’ one of two potentially movable elements) and that labels are asymmetric (because they contain features of only one element). In the course of the book I will challenge these three assumptions and argue that Merge can also create symmetric structures, that Move can sometimes treat two elements in a symmetric fashion, and that labels can sometimes contain features of two objects undergoing Merge. The rest of this introductory chapter serves three goals. First, it provides a general introduction to the concepts of symmetry, asymmetry and antisymmetry. It outlines what these concepts mean in general, as well as in 1
2
Rationale
more specific, linguistic terms. Second, it provides an overview of the theoretical framework assumed throughout the book, the minimalist program. The overview focuses on the workings of Merge, Move and Labeling, which are at the core of the claims I advance in the book. This chapter also explains why the empirical focus of the book is on symmetric aspects of these three mechanisms, as opposed to many other phenomena that the image of symmetry in syntax might conjure. And third, this introductory chapter provides an overview of the rest of the book.
1.2
Symmetry and asymmetry
The terms symmetry and asymmetry are used in two different ways in the literature. One is a fairly intuitive non-technical sense, and the other one is somewhat more technical and tends to vary from discipline to discipline. In its non-technical sense, the term symmetry is used to refer to the similarities between two parts of an object (or two objects), and the term asymmetry to the differences between them. In a linguistic context, the objects in question could be syntactic features, categories or transformations. Let us first look at a couple of simple cases. For example, we know that arguments differ from adjuncts in that they are bearers of theta roles. Thus we might speak of the symmetric behavior of different types of arguments (i.e. subjects and objects) with respect to theta theory, and the asymmetric behavior of arguments and adjuncts in the same respect. Another well-studied example involves cross-categorial symmetry, such as the symmetry between noun phrases and clauses, which has been studied quite extensively at least since Chomsky’s (1970) “Remarks on nominalization” (see Abney 1987, Douglas-Brown 1996 and Hiraiwa 2005, among others, for more recent ways to capture this symmetry). The data in (1a–b) illustrate the symmetric behavior of noun phrases and clauses with respect to theta role assignment. (1)
a. The RomansAgent destroyed the cityTheme b. the RomanAgent destruction of the cityTheme
And the Hungarian data in (2a–b) illustrate the symmetric behavior of subjects and possessors with respect to case marking; both are marked with the same (nominative) case. Furthermore, the possessee in (2b) agrees with the possessor in a way that parallels subject–verb agreement. (2)
a. Te ve-tt-el egy kalap-ot. 2SG.NOM buy-past-sg.indef indef hat-acc ‘You bought a hat.’
1.2 Symmetry and asymmetry
3
b. a te kalap-ja-i-d D 2SG.NOM hat-poss.pl-2.sg ‘your hats’ (Hiraiwa 2005:19–20, citing Szabolcsi 1994:186)
In a more technical (not necessarily linguistic) sense, the terms symmetry and asymmetry are used to describe geometric patterns, or relationships between two elements in a set. In geometric terms, an object is symmetric if it can remain unchanged when a transformation applies to it. Geometric figures under rotation transformation provide a straightforward illustration. A circle, for example, is symmetric under any rotation; if we rotate it by any degree, the result is still going to be a circle, as shown in (3a–c). A diamond, on the other hand, is only sometimes symmetric, as shown in (4a–c). If we rotate it by 45 degrees, the result is a square. However, if we rotate it by 90 degrees, the result is a diamond again. (3)
a.
b.
45° rotation
c. 90° rotation
(4)
a.
b.
45° rotation
c. 90° rotation
Mathematicians distinguish four types of symmetric transformations: reflection or mirror symmetry, rotation symmetry, translation symmetry and glide reflection symmetry (see Lee 2007 for an accessible overview). Rotation rotates an object (as we have just seen), translation shifts it (whilst preserving its orientation), reflection yields a mirror image of it, and glide reflection combines reflection and translation. As we will see shortly, the ones that apply most straightforwardly to linguistic patterns are translation and reflection symmetries, illustrated in (5a–b). (5)
a. translation symmetry
b. reflection or mirror symmetry
4
Rationale
In set theory, the terms symmetry and asymmetry are used to refer to binary relationships between elements in a set.1 This is by far the most common usage of the two terms in linguistics. A relationship between two elements in a set is symmetric if for every ordered pair <x,y> in the set, the pair is also in that set. A good illustration comes from the domain of kinship terms; the relationship ‘cousin of’ is an example of a symmetric relationship. If John is Bill’s cousin, Bill has to be John’s cousin as well. A relationship between two elements is asymmetric if it is never the case that for any pair <x,y> in the set, the pair is in the same set. The relation ‘is older than’ is asymmetric; if John is older than Bill, Bill cannot be older than John. A related concept is that of antisymmetry. A relationship between two elements in a set is antisymmetric if whenever both <x,y> and are members of the set, x must be the same as y. With this general background on symmetry (and asymmetry), we are almost ready to begin our examination of symmetry in syntax. First, however, let me briefly introduce the theoretical framework assumed in this book, the minimalist program. This is the topic of the next section. 1.3
Theoretical framework
The general framework of this book is the minimalist program pioneered by Chomsky (1995), in particular the version of it laid out in Chomsky (2000, 2001) and subsequent works, often referred to as Phase Theory.2 My goal in this section is not to provide a comprehensive overview of minimalism (or even a general introduction to it), but to give readers less familiar with it sufficient background to follow the rest of the book.3 The minimalist program is couched within the biolinguistic tradition, which takes the language faculty to be a biological organ, a product of evolutionary processes and pressures. The shape of the language faculty is determined by the following three factors, with the third factor gaining more prominence in recent years. (6)
(i) external data; (ii) genetic endowment (for language, the topic of UG); (iii) principles of structural architecture and developmental constraints that are not specific to the organ under investigation, and may be organism independent. (Chomsky 2008:133)
At the core of the minimalist program is the so-called Strong Minimalist Thesis (SMT), which states that “language is an optimal solution to interface conditions” (Chomsky 2008:135).4 The interface conditions are
1.3 Theoretical framework
5
those imposed by the sensorimotor (SM) and conceptual-intentional (C-I) systems. The SMT thus significantly changes the general architecture of the grammar. Readers well versed in Government and Binding theory (and its predecessors) will recognize the Y model of the grammar given in (7a) below, with four distinct levels of representation; D-structure, S-structure, Phonetic Form (PF) and Logical Form (LF). Operations could happen en route to any of these four levels. Likewise, conditions, principles and filters could apply at any level. The “new” minimalist architecture is given in (7b); there are only two relevant levels, the interface levels. Thus, all the syntactic conditions and principles have to be (re-)stated as interface conditions; there is no S-structure or D-structure levels to appeal to. (7)
a.
D-structure (X-bar theory, lexical insertion, Theta Criterion) overt transformations S-structure (expletive insertion, EPP) covert transformations PF
b.
LF (quantifier scope)
Numeration External Merge, Agree, Internal Merge
SM
C-I External Merge, Agree, Internal Merge
SM
C-I External Merge, Agree, Internal Merge
SM
C-I
Each derivation starts with a Numeration: a set of lexical items (or features, to be more accurate) to be manipulated in the course of the derivation. Once the Numeration is exhausted, the derivation is complete. Another crucial innovation in current minimalism is the idea that derivations proceed in chunks called phases and that transfer to the two interfaces can happen more than once per derivation. The terms Phase Theory or Multiple Spell-Out Theory reflect this aspect of the theory.5 The points of transfer to the interfaces are determined by phase heads, which are taken
6
Rationale
to be (transitive) v and C heads (perhaps also D heads).6 More specifically, every time one of these phase heads is merged, the complement of the lower phase head is transferred to the interfaces (and becomes inaccessible for further computation). For example, when C is merged, the complement of v is spelled out. This means that the v head itself, its specifiers and adjuncts (if any) remain accessible to the derivation. Otherwise, each derivation would stop with the first Transfer. The condition that ensures this is called the Phase Impenetrability Condition, one formulation of which is given in (8). (8)
The Phase Impenetrability Condition (PIC) (Chomsky 2004:108) a. PH = [α [ H β]] b. The domain of H is not accessible to operations, but only the edge of HP.
As mentioned above, the three mechanisms at the center of this book are External Merge, Internal Merge and Labeling. Let me thus conclude this overview with a brief discussion of how they work. The issue of what kinds of structures Merge can generate is arguably one of the most fundamental issues in syntactic theory. It becomes particularly pressing in the context of recent work by Hauser, Chomsky and Fitch (2002) (see also Fitch, Hauser and Chomsky 2005), who propose that recursive Merge (or some mechanism akin to it) is the only uniquely human property of language, and, as such, it is what separates human language from the communication systems of other species.7 Merge comes in two guises, External Merge and Internal Merge. Simply put, External Merge takes two disjoint syntactic objects and combines them to form one larger syntactic object, as shown in (9a). One of these objects could itself be a result of a previous Merge operation (which is what captures recursion), as shown in (9b). (9)
a. External Merge of α and β α
γ
β α
β
b. External Merge of γ and δ γ α
δ β
ε γ
δ α
β
1.3 Theoretical framework
7
Internal Merge, often referred to simply as Move, is an operation “responsible” for displacement in the grammar; it captures the intuition that syntactic objects can be pronounced and interpreted in different positions. Internal Merge is like External Merge in that it also takes two objects and combines them into one bigger object. What differentiates it from External Merge is that one of these two objects is a part of the other. (10a–b) below illustrate Internal Merge of α and β; (10a) represents it as a standard Copy and Merge (and Delete) operation, whereas (10b) represents it as literal Internal Merge; the moved element β, instead of being copied into a new position, is simply merged again in its new position.8 The issue of whether the choice between these two ways of conceptualizing movement is substantive or simply a matter of stylistic convention is not trivial but it is not an issue that is directly relevant for our purposes. (10)
a. Internal Merge of α and β α α
α
β
α
βi α
ti
b. Internal Merge of α and β α
α α
α
β α
β
The version of minimalism assumed here maintains the early minimalist assumption that uninterpretable features play a crucial role in syntactic computation. Uninterpretable features are the features that enter the derivation unvalued (marked in what follows as [uF]) and receive values in the course of the derivation via an operation called Agree. This is a major departure from early minimalism, where feature checking (now conceived of as feature valuation) required movement to a licensing position, typically a specifier of an appropriate functional projection. Now, they can get valued via Agree between a Probe (an item whose feature provides a value) and a Goal (an item whose feature is in need of a value), as shown in (11a–b).
8 (11)
Rationale a. ProbeF:val b. ProbeF:val
> >
GoaluF: GoaluF:val
For Agree to take place, the following conditions have to be met. First, the Goal has to be active, where being active means having an unvalued feature. Second, the Goal has to be in the c-command domain of the Probe. And third, there can be no closer potential Goal. Crucially, in a system with Agree, there is no direct relationship between the need to value unvalued features and movement. Features can be valued in situ and what drives movement is the (generalized) EPP feature (or property) of the Probe, which requires it to have an overt specifier. The last core concept that we will need is the concept of labels, whose existence has been implicit in the discussion of External Merge and Internal Merge, above. If External Merge were simply concatenation and Internal Merge displacement, we would expect their outputs to be (12a) and (12b), respectively. (12)
a. External Merge of α and β α
β
b. Internal Merge of α and β β δ
tβ
This, however, is not sufficient; the output of Merge also needs a label. The view of labels that I assume throughout this book is essentially that of Chomsky (1994), where features of one of the two objects undergoing Merge determine the label of the new complex object. The issue of whether Labeling is a separate mechanism (or part of Merge itself) is an interesting one, but it does not bear directly on the issue at stake here, which is the symmetric (or asymmetric) nature of labels. I will thus treat Labeling as distinct from Merge and Move (as argued recently by Hornstein 2009, for example), although nothing substantive hinges on this choice, and my conclusions hold irrespective of whether it is an independent mechanism or not. The idea that labels are necessary (or desirable) in a minimalist architecture is by no means uncontroversial. Chomsky (2004:109) states: “A still more attractive outcome is that L [language, B.C] requires no labels at all” (see also Collins 2002 and Seely 2006, among others, for arguments that labels might be dispensable in a minimalist system). In Chapter 5 of this book, however, I argue against this view, and provide concrete arguments
1.5 Disclaimer: other sources of symmetry?
9
in favor of the existence of labels. For now, let me proceed on what I consider the standard minimalist assumption, which is that the grammar needs labels, and that the result of External Merge is (13a), not (12a), and the result of Internal Merge is (13b), not (12b). (13)
a. External Merge of α and β
b. Internal Merge of α and β
γ α
γ
β
α
β δ
1.4
tβ
The proposal
Given that Merge, Move, and Labeling occupy a central spot in minimalist theorizing (for reasons outlined in the previous section), it seems natural to focus on them in our exploration into the locus and amount of symmetry in syntax. The three questions that I ask in the course of the book are: (14)
a. Does symmetric Merge exist? b. Does symmetric Move exist? c. Do symmetric labels exist?
The central claim of this book is that the answer to all three questions is “yes.” In the next five chapters, I will provide both theoretical and empirical arguments to support this claim, and, indirectly, against the claims that asymmetric relations are the core relations of the language faculty, articulated in various forms by various researchers (for example, as the Asymmetry Theory of Di Sciullo 2005 or the Antisymmetry Theory of Kayne 1994). 1.5
Disclaimer: other sources of symmetry?
My goal in this book is to argue for symmetry in Merge, Move, and Labeling. Arguably, these are not the first (or most obvious) mechanisms that come to mind when we think of symmetry in syntax. This raises a natural question of why focus on these three, as opposed to the perhaps more apparent symmetric patterns. Part of the reason is purely practical; the issue of symmetry (or the lack thereof) in syntax is vast, and I hope to make it more manageable by narrowing down the domain of inquiry to these three mechanisms. Furthermore, it is not clear that the symmetry we
10
Rationale
see elsewhere is true (or only apparent). Let me nevertheless conclude this chapter with a brief discussion of other syntactic phenomena that could be (or have been) classified as symmetric and explain why this book is not about them. Consider the schematic word order patterns given in (15a–b). If both exist, we have a case for the existence of mirror symmetry in the grammar; (15b) is a mirror image of (15a).9 (15)
a. ABCD b. DCBA
A fairly straightforward illustration of the two orders comes from the domain of adjective ordering, where both patterns are attested crosslinguistically (in addition to many others, not relevant here). For example, the ordering of French adjectives in (16a–b) is a mirror image of the ordering of their English counterparts in (17a–b). The factors that determine the ordering of the adjectives relative to each other are also not directly relevant here (for discussion and concrete proposals, see Bernstein 1993, Cinque 1994, Scott 2002, Sproat and Shih 1991, Svenonius 2008, among many others). (16)
(17)
a. une a b. une a
voiture car fusée rocket
italienne Italian américaine American
magnifique beautiful énorme huge
French
(Laenzlinger 2005:658)
a. a beautiful Italian car b. a huge American rocket
This is not the only possibility; in (18a), the ordering of postnominal adjectives matches the ordering of prenominal adjectives in English, and in (19a), some adjectives precede the noun and others follow it. (18)
(19)
a. une voiture rouge française a car red French b. a red French car
(Laenzlinger 2005:658)
a. un joli gros ballon rouge a pretty big ball red b. a pretty big red ball
(Cinque 1994:101)
The range of logical possibilities for a sequence consisting of two adjectives and a noun is given in (20a–f). Since there are three elements (two adjectives and a noun), there are 6 (3!) possible ways to order them. (20b) is a mirror image of (20a). The orderings in (20c) and (20d) have been
1.5 Disclaimer: other sources of symmetry?
11
shown by Bernstein (1993) and Cinque (1994), among many others, to be the result of head movement of the noun around Adj2 in (20c) and around both Adj1 and Adj2 in (20d), as shown in in (21b) and (21c) respectively. And the ordering in (20b) has been claimed to be the result of the so-called roll-up movement, schematized in (22a–c). First, the noun moves around Adj1 and, next, the constituent consisting of the noun and Adj1 moves around Adj2. (20)
a. b. c. d. e. f.
Adj2 N Adj2 N Adj1 Adj1
Adj1 Adj1 N Adj2 Adj2 N
N Adj2 Adj1 Adj1 N Adj2
(=17a–b) (=16a–b) (=19a)
(21)
a. Adj2 Adj1 N b. Adj2 Ni Adj1 ti c. Ni Adj2 ti Adj1 ti
(22)
a. Adj2 Adj1 N b. Adj2 Ni Adj1 ti c. [Ni Adj1]j Adj2 tj ti
head movement
roll-up movement
Given all the movement possibilities, the existence of such symmetric patterns does not seem to be evidence for symmetry inherent to phrase structure rules (or Merge itself). Rather, the illusion of symmetry is created by a sequence of movement operations, involving either head movement or roll-up movement. This brings us to movement itself, more specifically, to the question of what kind of movement would “count” as symmetric movement. We can consider this question from different angles, focusing either on the relationship between the basic and the derived order or the directionality of movement itself. Let us consider them in turn. There are two ways to think of a symmetric relationship between the basic and the derived order; the movement can either preserve the basic order or yield a mirror image of it, as shown in (23a) and (23b) respectively. The two could be thought of as examples of translation and reflection symmetries discussed above. (23)
a. ABCD b. ABCD
→ →
ABCD DCBA
translation symmetry reflection symmetry
If all elements move as individual units, the two illustrate crossing and nesting dependencies, schematized in (24a–b).
12 (24)
Rationale a.
b. A
D B
C B
C D
A tA
tA tB
tB tC
tC
tD
tD
The contrasts between the a and b examples in (25) and (26) below, due to Pesetsky (1982), suggest that in a language like English only nesting dependencies are possible. These examples establish this point for whmovement and tough movement, but the preference for nesting has since been generalized to other movement types. (25)
a. What subjecti do you know whoj PRO to talk to tj about ti? b. * Whoj do you know what subjecti PRO to talk to tj about ti?
(26)
a. What balalaikaj are these partitas easy OPi PRO to play ti on tj? b. * What partitasi is this balalaika easy OPj PRO to play ti on tj? (Pesetsky 1982:268–9)
Richards (2001), on the other hand, argues that only crossing dependencies are possible if movements target specifier positions of the same functional head. Crucially, neither Pesetsky nor Richards argues that either crossing or nesting is possible, all things being equal. Another possible way to conceive of symmetric movement involves directionality; example (27a) illustrates movement to the left and (27b) movement to the right. (27)
a.
b. B
B A
A tB
tB
Constructions that have been claimed to involve rightward movement include (but are not limited to) extraposition, Heavy NP Shift and right node raising, illustrated in (28a–c), respectively. (28)
a. We took a picture ti yesterday [PP of John]i. b. We gave ti to John [DP our most desirable king size suite]i. c. John wrote ti and Mary reviewed ti [DP a new article on right node raising]i.
1.5 Disclaimer: other sources of symmetry?
13
All of them, however, have since been reanalyzed in ways that either do not involve movement at all or involve leftward movement instead. Fox and Nissenbaum (1999), for example, analyze extraposition as countercyclic adjunction of the “extraposed element” to a covertly moved object, providing scope evidence in favor of such an analysis.10 Larson (1988; see also Kayne 1994) analyzes Heavy NP Shift as leftward movement of a larger constituent, consisting of the verb and the indirect object in (28b). In Chapter 3, we will see arguments against treating right node raising as rightward movement and in favor of analyzing it in a multidominant way (as also argued by Bachrach and Katzir 2009, Citko in press-b, Johnson 2007, McCawley 1982, De Vos and Vicente 2005, De Vries 2009, Wilder 1999b, 2008, among others). For now, it seems safe to conclude that the symmetry involving either leftward or rightward movement does not exist in the grammar. Yet another way to think of movement in a symmetric way is illustrated in (29a–b). (29a) represents raising, and (29b) lowering.11 (29)
a.
b. B A
A tB
tB B
It is typically assumed that only (29a) is possible, as (29b) violates wellestablished conditions on movement, such as the Extension Condition of Chomsky (1995), which requires all Merge and Move operations to target the root of the tree, or the more general No Tampering Condition of Chomsky (2008), which prohibits any modifications to structures that have been already built. Raising transformations are by far the norm in syntactic theory, but there are some cases of lowering transformations that come to mind, such as affix hopping. However, since affix hopping is better analyzed as a PF-merger operation (as argued or implied by Bošković and Lasnik 2003, Bobaljik 1995, Embick and Noyer 2001, Halle and Marantz 1993, among many others), it seems reasonable to conclude that the type of symmetry schematized in (29a–b) does not exist either.12 We might also wonder about more general sources of symmetry in the grammar: parameters whose different settings give rise to symmetric patterns. An obvious candidate for such a parameter is the Head Directionality Parameter. In a very intuitive sense, the existence of consistently head-initial and consistently head-final languages could be viewed as evidence for symmetry in syntax. However, the existence of this type of symmetry has also been questioned in the relevant literature. As pointed
14
Rationale
out by Kayne (1994), for example, this type of symmetry breaks down in many directions, and predicts the existence of many language types that we do not find. For example, we do not find mirror-image V2 effects (whose properties would follow from head-final and specifier-final setting of the Directionality Parameter), or wh-movement targeting the final position in the clause (which would follow from the same setting of the parameter). One could also imagine a symmetry parameter of sorts stating that some languages are inherently (more) symmetric than others. A hypothetical formulation of such a parameter is given in (30). If true, it could be seen as a macro-parameter par excellence, akin to Baker’s (1996) Polysynthesis Parameter, Hale’s (1983) Nonconfigurationality Parameter or Huang’s (2005) Analyticity Parameter. The existence of such macro-parameters, however, is not an uncontroversial matter (see Kayne 2005 and Baker 2008 for two opposing views on the matter). (30)
Symmetry Parameter A language is either symmetric or asymmetric.
I am not defending in this book anything even close to such a parameter, nor am I aware of anyone else trying to. Its simplicity is appealing but its consequences would be rather dire. First and foremost, it is not clear what it would mean for a language to be either symmetric or asymmetric. Perhaps symmetric languages have overwhelmingly symmetric structures, as implied by Hale’s (1983) Nonconfigurationality Parameter, which posits a flat structure (in which arguments c-command each other) for nonconfigurational languages like Warlpiri. Again, it has since been shown that there is nothing unusual about such languages, and their nonconfigurational properties follow from the interaction of perfectly configurational UG principles and parameters (see Legate 2001 and 2002 for discussion). More generally, we might wonder about the status of macro-parameters in minimalist syntax. Since the Symmetry Parameter cannot be stated as a property of individual lexical items (or a class of lexical items), it is not compatible with the view going back to Borer (1984) that the locus of crosslinguistic variation is in the lexicon, more specifically, in the features of functional heads.
1.6
Organization of the book
In this final section, I offer a brief outline of the rest of the book. Most of the novel data in the book come from Polish, a West Slavic language that provides a nice illustration of many of the phenomena I discuss. By focus-
1.6 Organization of the book
15
ing on Polish, I hope to make the book of interest not only to theoretically minded linguists interested in the workings of Merge, Move and Labeling, but also to more empirically minded linguists interested in the syntax of Slavic languages. I will, however, add a comparative perspective where it is useful and necessary. Chapter 2 “Asymmetry in syntax” takes a step back, and reviews both empirical and theoretical reasons behind the claims that syntax only allows asymmetric relations. It examines the evidence that Merge can only create structures that involve asymmetric c-command between nodes (since only such structures are linearizable, according to the Linear Correspondence Axiom of Kayne 1994), that Move is asymmetric (in that it always targets the higher of the two potentially movable elements), and that labels are asymmetric (in that they contain features of only one element). This chapter serves mostly as background for what is to come in the following three chapters, which show that Merge can also create symmetric structures, that Move can sometimes treat two constituents in a symmetric fashion, and that labels can include features of two elements. Chapter 3 “Symmetric Merge” provides both theoretical and empirical arguments in favor of a particular kind of symmetric structure, the multidominant structure of the kind given in (31), in which one node has two mothers.two mothers. (31)
K α
L β
γ
The existence of multidominant structures is by no means new in generative grammar (see Blevins 1990, Goodall 1987, McCawley 1988, Moltmann 1992a, Muadz 1991, for some pre-minimalist precursors and Citko 2000, 2003, 2005, Gračanin-Yüksek 2007, in press, Van Riemsdijk 2000, 2006a, 2006b and De Vries 2005a, 2005b, 2008, 2009 for more recent minimalist implementations). In my early work, I argued that adopting a symmetric structure of this kind can account for a number of otherwise mysterious properties of across-the-board wh-questions, questions such as the one given in (32), in which a single wh-element is extracted simultaneously from two (or more) conjuncts. (32)
What did John write and Bill review?
This chapter extends this analysis to a number of other constructions, thus providing new empirical evidence in favor of symmetric Merge structures. The constructions I discuss in this chapter are: questions with conjoined
16
Rationale
wh-pronouns, gapping, right node raising, standard and transparent free relatives, serial verbs, and they are illustrated in (33a–f). (33)
a. What and why did John eat? wh&wh questions b. John writes poems and Mary short stories. gapping c. John wrote and Mary reviewed an article on multidominance. right node raising d. John reads whatever Mary writes. standard free relatives e. John wrote what some might call a poem. transparent free relatives f. ò dà sε la nεnè ɔ`ɔ`. Dàgááré serial verbs 3sg past roast fem meat eat ‘He roasted the meat and ate it.’ (Hiraiwa and Bodomo 2008:243)
Chapter 4 “Symmetric Move” turns to the evidence for symmetric movement, i.e. a type of movement that can target two elements in a given structure with equally grammatical results. Passive movement in double object constructions provides a good testing ground; since there are two objects, there are two candidates for movement. In some languages, passive movement is a symmetric operation in that it can target either the direct or the indirect object, whereas in others it is an asymmetric operation in that it can target only one of them. This variation has been studied quite a bit and attributed either to various ways of circumventing a locality violation (such as the availability of the word order in which the direct object precedes the indirect one, either via base-generation or movement) or to case properties of the two objects (and various patterns of case absorption in passives). What I argue for in this chapter is that we need both types of account to reach empirical adequacy. Chapter 5 “Symmetric labels” gives arguments in favor of the existence of symmetric labels in the grammar. It also establishes a need for labels in the grammar. There are two ways a label can be symmetric – if both merged elements project as the label, or if neither of them does (as argued by Collins 2002, Seely 2006, among others). This chapter focuses on the former case – symmetric labels in which both constituents project – and argues that this is what is involved in the derivation of certain types of serial verb constructions, nominal copular constructions and comparative correlatives. By the end of Chapter 5 I hope to convince the reader that symmetry does exist in the domain of the three basic syntactic mechanisms: Merge, Move, and Labeling. Chapter 6 “Conclusion” offers a brief summary and some general thoughts on the nature and origin of symmetry (and asymmetry) in the grammar.
2
Asymmetry in syntax
2.1
Introduction
In this chapter, I take a step back and review the reasons behind the claim that syntactic structures, operations and relationships are fundamentally asymmetric. This claim has been articulated in various ways by various researchers: as the Antisymmetry Theory of Kayne (1994), the Dynamic Antisymmetry Theory of Moro (2000) or the Asymmetry Theory of Di Sciullo (2005). I focus on the asymmetric properties of Merge, Move and Labeling, as these are the mechanisms I take on in the chapters that follow. I proceed as follows. In Sections 2.2–2.4, I review Kayne’s (1994) Antisymmetry Theory, Moro’s (2000) Dynamic Antisymmetry Theory and Di Sciullo’s (2005) Asymmetry Theory (which all take asymmetric structures and mechanisms to be the norm). In Sections 2.5–2.6, I discuss two constructions that were analyzed as symmetric in the early days of generative grammar and have since been reanalyzed as asymmetric. The two are coordinate structures and double object constructions. In Section 2.7, I turn to the asymmetric properties of Move, focusing on locality considerations. And finally, in Section 2.8, I discuss the asymmetry of existing labeling algorithms. 2.2
Antisymmetry Theory
Kayne’s (1994) Linear Correspondence Axiom, stated formally in (1a) and more informally in (1b), only allows structures with asymmetric c-command relationships between non-terminal nodes. As a consequence, it derives precedence from asymmetric c-command; if one node asymmetrically c-commands another node in a tree, the terminal nodes dominated by the first node will precede the terminal nodes dominated by the second one.1 17
18 (1)
Asymmetry in syntax a. d(A) is a linear ordering of T. (Kayne 1994:6) b. Let X, Y be nonterminals and x, y terminals such that X dominates x and Y dominates y. Then, if X asymmetrically c-commands Y, x precedes y.
The LCA excludes structures in which non-terminal nodes are in a symmetric c-command relationship on the grounds that they violate one (or more) of the three requirements on linear ordering: transitivity, totality and antisymmetry, defined in (2a–c). R here stands for precedence but it can refer to any relationship between two elements in a set.2 (2)
a. it is transitive (xRy & yRz → xRz) b. it is total (for all distinct x, y, either xRy or yRx) c. it is antisymmetric (not (xRy & yRx))
To illustrate, let us consider a schematic tree representation given in (3a), with its A given in (3b) and its d(A) given in (3c). The resulting ordering is transitive, total and antisymmetric. Thus it satisfies all the requirements linear ordering has to satisfy. Note that because of the segment/category distinction, the lower E in (3a) does not c-command A. If it did, the terminals of the lower E would end up both preceding and following the terminals of A, in violation of antisymmetry. (3)
a.
E A a
E B
D
b
C c
b. A = { , , , } c. d(A) = { , , }
In X-bar theoretical terms, (3a) represents an asymmetric specifier-headcomplement structure. Examples of symmetric structures excluded by the LCA are given in (4a–c), with the problematic nodes indicated in bold.3 (4a) cannot be linearized because of the symmetric c-command between X and Z. This means that no order can be established between x and z, which violates the totality requirement on linear order. (4b) is excluded for similar reasons. Since X1 and X2 c-command each other, no ordering can be established between their respective terminals, x1 and x2. (4c) and (4d), in which YP and XP c-command each other, are excluded for the same reasons.
2.2 Antisymmetry Theory (4)
a.
19
XP YP
XP
Y
X
Z
y
x
z
b.
ZP Z z
XP X2 YP
X1 x1
Y
x2
y
c.
ZP Z z
d.
XP YP
XP
Y
Z
y
z
ZP XP
ZP
X
YP
x
Y
Z
WP
y
z
W
ZP
w
The ban on symmetric structures in the grammar can thus be attributed to the Linear Correspondence Axiom. A potential issue here concerns the compatibility of the LCA with the minimalist Bare Phrase Structure Theory of Chomsky (1994), in which Merge is the sole operation responsible for structure building. Merge creates structures of the kind given in (5a), rather than the “exploded” ones given in (5b), which the original formulation of the LCA relied upon.
20 (5)
Asymmetry in syntax a.
b.
x y
YP
x x
XP
z
X
Y’ X
ZP
Y x
Z
y
Z z
This is not an insurmountable problem. Chomsky (1995) takes the LCA to be compatible with core minimalist assumptions (and highly desirable given the Strong Minimalist Thesis), and suggests various ways to remedy this problem. One possibility is to reanalyze the symmetric portion of the tree as a morphological word, immune from the effects of the LCA. Alternatively, the complement of x in (5a) could itself be a more complex category, headed by a null projection. Kayne (2008), following Guimarães (2000), makes a similar suggestion and assumes that one of the two problematic elements can undergo self-Merge first, which yields a more complex object ({x} rather than x). This has essentially the same effect; it makes one of the problematic objects more complex. Yet another possibility would be for one of the problematic elements in (5a) to become invisible to the LCA in the course of the derivation, either through movement or ellipsis. This solution relies on the idea that the LCA, being a principle of linearization, only needs to apply at the level relevant for linearization, namely the level of Spell-Out. On this view, the grammar can generate symmetric structures as long as they become linearizable by the time of Spell-Out. Chomsky departs in this respect from Kayne (1994), for whom the LCA is an overarching principle of the grammar applying at all levels. This is also the basic insight behind Moro’s (2000) dynamic approach to antisymmetry, the details of which I discuss in the next section. 2.3
Dynamic Antisymmetry Theory
Moro’s (2000) Dynamic Antisymmetry Theory allows symmetric structures, which he dubs points of symmetry as long as these points of symmetry are destroyed during the derivation. Moro focuses on three kinds
2.3 Dynamic Antisymmetry
21
of symmetric structures: small clauses, multiple specifiers and clitic complementation structures, illustrated in (6a–c) respectively. In all three of them, the elements that stand in a symmetric c-command relationship are represented in bold. (6)
a. small clauses4 XP YP
ZP
b. multiple specifiers XP YP
XP XP
ZP
c. clitic adjunction XP X0
Y0
(Moro 2000:32)
An issue that is somewhat tangential for our purposes but that plays an important role in Moro’s theory concerns motivation for movement. For Moro, movement is driven by purely architectural considerations, i.e. by the need to break symmetry in the structure, not by the need to check uninterpretable features. This is a very strong, explicit, and innovative proposal, whose empirical consequences are quite clear. If true, it means that all moved elements have to involve a symmetric structure at the point of first Merge. Otherwise, there would be no need for them to move. The burden of proof is in the details; a cursory look at a couple of paradigmatic cases of movement – subject raising in (7a), wh-movement in (7b) and head movement in (7c) – shows no points of symmetry prior to movement.5 (7)
a.
TP DPi
T T
vP ti
v v
VP
22
Asymmetry in syntax b.
CP WHj
C TP
C T
vP tj
v DP
v VP
V V
c.
tj
vP DP
v VP
v Vi
v
ti
DP D
NP
Furthermore, if elimination of symmetry were the only motivation for movement (and the need to check (and delete) uninterpretable features were never an issue), we would expect a certain amount of optionality in the grammar, which we do not find. For example, we would expect either verb movement or wh-movement to be possible in (7b), since either would satisfy the need to “destroy the point of symmetry,” to use Moro’s terms. Part of Moro’s conceptual motivation for viewing movement as a symmetry-breaking phenomenon comes from the desire to rid the grammar of feature-driven movement. However, it is not clear how landing sites would be determined if formal features were not involved at all. As far as symmetry-breaking goes, movement to any position (as long as this position itself is not part of a symmetric configuration) should be fine. The two parts of the Dynamic Antisymmetry proposal, letting some amount of symmetry into the grammar and motivating movement by the need to break this symmetry, are in principle independent. The first part strikes me as generally plausible, and the discussion of symmetric Merge in Chapter 3 rests on the same idea. The second part faces three problems that I have just pointed out: it requires otherwise unmotivated new
2.4 Asymmetry Theory
23
movement configurations, it predicts optionality that we do not find, and it has nothing to say about potential landing sites. These are the reasons I maintain a more “conservative” approach to movement, in which it is driven by the need to eliminate uninterpretable features, even if such features are an imperfection in the system from the perspective of the Strong Minimalist Thesis. 2.4
Asymmetry Theory
Di Sciullo’s (2005) Asymmetry Theory extends many of the ideas discussed above to the morphological component. Even though our focus here is on symmetry (and somewhat indirectly, on asymmetry) in syntax, not morphology, and a complete discussion of Di Sciullo’s theory would take us too far off course, let me highlight its most relevant aspects. The main thesis is given in (8).6 (8)
Strict Asymmetry of Morphology
(Di Sciullo 2005:13)
Morphology combines and manipulates asymmetric relations only.
A very straightforward illustration comes from restrictions on morpheme ordering illustrated in (9a–c). (9a) shows that affixes and roots do not invert, (9b) shows that the order of affixes is fixed, and (9c) shows that the order within compounds is fixed as well. (9)
a. b. c.
write-er / # er-write comput-er-ize / # comput-ize-er elephant man / # man elephant
(Di Sciullo 2005:13–14)
To capture these asymmetries, Di Sciullo assumes that the most basic morphological object is an asymmetric minimal tree, which is a tree with exactly one head, one specifier and one complement. Different types of affixes can occupy different positions in such a morphological tree; some are specifiers and others heads. (10)
a.
b.
x dis-
α
x x
x
β
x − ed
β
Such minimal morphological trees can undergo three operations: M-Shift, M-Link and M-Flip. M-Shift combines two trees by substituting one into the complement of the other. M-Link establishes a featural relation
24
Asymmetry in syntax
between two objects in a single tree that stand in an asymmetric c-command relationship. And M-Flip takes one tree and creates a mirror image of it. Interestingly, M-Flip is a paradigmatic example of the reflection symmetry introduced in Chapter 1. (11)
M-Flip (T)
(Di Sciullo 2005:33)
Given a minimal tree such that the Spec of T has no PF features, M-Flip (T) is the tree obtained by creating a mirror image of T. (12)
a.
b.
x α
x
x α
x
FLIP
β
β
x
x
Di Sciullo gives three reasons why we need M-Flip. First, it can account for the fact that the linear order of affixes does not always reflect their scope. Words like unreadable illustrate this mismatch. (13)
a. un- > read > -able linear scope at PF b. un- > able- > read affixal scope at LF
(Di Sciullo 2005:128)
Second, M-Flip can capture the effects of head movement (without requiring actual movement). The result of Y to X head movement in (14a) is not (14b) but (14c). (14)
a.
b.
X α
c. X
α
X X
Y β
X X
δ
α
X Y
Yi X β
Y Y
X
Y Y
ti
β
Y δ
δ
X
Y
(Di Sciullo 2005:132)
And third, M-Flip is involved in the derivation of deverbal compounds such as spellchecker, in which the order of constituent parts mirrors their argument structure:
2.5 Coordination (15)
a.
25
b. α
α er
er β check χ
β check χ spell
δ
δ
spell
(Di Sciullo 2005:140)
The upshot of the discussion in this section (combined with the results of the previous two sections) is that the grammar either does not allow symmetric structures at all or allows them only to a very limited degree. From the perspective of the Antisymmetry Theory, the primary reason for this prohibition is linearization. In the next two sections, I add more empirical content to this claim and discuss two constructions (coordinate structures and double objects) that in the early days of generative grammar were analyzed as symmetric and have since been reanalyzed as asymmetric. The reason I choose these two has to do with the empirical content of the rest of the book. The discussion of symmetric Merge in Chapter 3 relies on the asymmetric approach to coordination, and the discussion of symmetric Move in Chapter 4 relies on the asymmetric structure for double object constructions. In general, I hope to show that reanalyzing double objects and coordinate structures in an asymmetric fashion does not warrant the conclusion that the grammar does not allow symmetric structures of any sort. 2.5
Coordination
Early accounts assigned coordinate structures a ternary branching structure with a symmetric c-command (i.e. mutual c-command) between the two conjuncts, as shown in (16). (16)
XP XP Conj XP
However, it has since been shown that the relationship between the two conjuncts has to be asymmetric instead (see Progovac 1998a and 1998b for a summary). First, the conjunction and forms a constituent with the second conjunct. This is shown, for example, by the fact that there is a
26
Asymmetry in syntax
natural pause (indicated in (17a) by a sentence boundary) after the first conjunct.7 (17)
a. John left. And he didn’t even say goodbye. b. * John left and. He didn’t even say goodbye.
(Ross 1967:90)
The fact that it is possible to extrapose the conjunction and the second conjunct, but not the first conjunct and the conjunction (or just the first conjunct) points towards the same conclusion: (18)
a. John bought a book ti yesterday, and a newspaperi. b. * John bought ti a newspaper yesterday a book andi. c. * John bought a book and ti yesterday, a newspaperi. (Munn 1993:15)
The contrast in (19) establishes a similar point; the pro-form etc can replace “and cookies” but not just “cookies”.8 (19)
a. I bought jam, bread, and cookies. b. I bought jam, bread etc. c. * I bought jam, bread, and etc.
(Zoerner 1995:17)
Second, there is evidence that the first conjunct c-commands the second one. In (20a) below, for example, the name contained in the second conjunct is c-commanded by the coindexed pronoun in the first one, in violation of Principle C. There is no c-command between John and he (or him) in (20b), since John is buried inside the first conjunct. (20)
a. * Hei and Johni’s dog went for a walk. b. Johni’s dog and hei/himi went for a walk.
(Munn 1993:16)
The contrast in (21) makes a similar point with respect to variable binding; bound variable interpretation is possible in (21a) because the quantified noun phrase every man c-commands his dog, but it is impossible in (21b) because the second conjunct does not c-command the first one. (21)
a. Every mani and hisi dog went to mow the meadow. b. * Hisi dog and everyi man went to mow the meadow. (Munn 1993:16)
There are many asymmetric structures that have been proposed for coordination which are consistent with the data above. They all capture the fact that the second conjunct does not c-command the first one. They differ, however, in whether the first conjunct c-commands the second one. The simplest (and most commonly assumed) structure is the one given in (22). This is the structure suggested by Larson (1990) and argued for more
2.5 Coordination
27
explicitly by Johannessen (1998) and Zoerner (1995), among many others.9 This is also the structure I adopt in this book.10 (22)
&P &
XP &
XP
(cf. Johannessen 1998:109)
This is by no means the only possibility. Munn (1993) and Progovac (1998a, 1998b) argue for a structure in which one conjunct and the conjunction are adjoined to the other conjunct. For Munn, the constituent consisting of the conjunction and the second conjunct is adjoined to the first conjunct, as shown in (23a). For Progovac, each conjunct is a complement of its own conjunction head, and the &P containing the first conjunct is adjoined to a null pronominal head, as shown in (23b).11 A yet different structure was proposed by Camacho (1997) and (2003). His structure, given in (23c) below, involves no conjunction phrase whatsoever. The conjunction occupies a variable position, marked as X in (23c). For example, with coordinate subjects, the conjunction is in T (I in Camacho’s terms) and with coordinate objects, it is in v (AgrO head in Camacho’s terms). (23)
a.
XP XP
BP (=Boolean Phrase) B
XP
b.
(cf. Munn 1993:13)
XP &P
XP XP
&P
XP
(cf. Progovac 1998b:5)
XP
&
c.
&
XP X
NP X
XP X
NP X
YP
(Camacho 1997:51)
28
Asymmetry in syntax
Progovac (1998a) and (1998b), however, raises some issues for the Johannessen/Zoerner structure given in (22) above. In particular, she questions the idea (shared by the structures in (22), (23a) and (23c)) that the first conjunct c-commands the second one. If it did, we would expect to find a contrast in grammaticality between (24a) and (24b). The fact that both are ungrammatical does not follow from the structure in which the first conjunct c-commands the second one. (24)
a. * He chased nobody and/or any dogs. b. * He chased any dogs and/or nobody.
(Progovac 1998a: 3)
Progovac also questions the Principle C data given in (20a–b) above and points out that the same contrast holds across clauses: (25)
* Hei finally arrived. John’si dog went for a walk.
(Progovac 1998a:3)
Furthermore, if c-command were the only factor responsible for the contrast in (20a–b), one would expect no contrast whatsoever between (26a) and (26b) below, as the c-command relationships are the same in both examples: John asymmetrically c-commands John’s wife in both examples, yet (26b) is markedly worse than (26a). (26)
a. Johni and John’si wife are certainly invited. b. ?* Johni certainly likes John’si wife.
(Progovac 1998a:3)
Another issue Progovac raises comes from languages with reflexive possessive pronouns. Polish, similarly to Serbo-Croatian (the language Progovac focuses on) has both possessive pronouns and possessive reflexives, illustrated in (27a–b) below. In such languages, a possessive pronoun is obligatorily disjoint from the subject. (27)
a. Jani lubi swóji dom. Jan likes refl house ‘Jan likes his (own) house.’ b. Jani lubi jegoj/*i dom. Jan likes his house ‘Jan likes his (someone else’s) house.’
Polish
The reflexive possessive, however, is impossible as the second conjunct in a coordinate structure, in spite of being c-commanded by a potential antecedent:12 (28)
* Jani
i
swojai żona przyszli.
Jan and refl wife arrived ‘Jan and his wife have arrived.’
2.6 Double object constructions
29
The ungrammaticality of (28), however, has an independent explanation, having to do with the fact that the Polish reflexive pronoun swój is a subject-oriented anaphor. In (29a) below, the indirect object cannot serve as the antecedent of swój even though it c-commands it. Furthermore, being a subject-oriented anaphor, swój itself cannot be inside a subject. This can also account for the ungrammaticality of (28) and (29b). In example (29b), c-command is also not a factor; the scrambled indirect object Janowi does c-command the subject anaphor swój. (29)
a.
Jani dał Tomkowij swojąi/*j książkę. Jan.nom gave Tom.dat refl book.acc ‘Jan gave Tom his book.’ b. * Janowii, swóji professor dał ti książkę. Jan.dat refl professor.nom gave book.acc ‘His professor gave Jan a book.’
The conclusion that the first conjunct c-commands the second one might seem somewhat weaker in view of Progovac’s data. However, the structure in which the first conjunct does not c-command the second one does not necessarily entail a symmetric structure. Thus the facts do not warrant abandoning the asymmetric approach to coordination. They also do not necessarily warrant abandoning the structure in which the first conjunct c-commands the second one. Rather, they call for an alternative explanation for why anaphor binding or negative polarity licensing across two conjuncts is impossible. One possibility would be to attribute it to the fact that the anaphor and its binder (the same goes for the negative polarity item and its licensor) have to have distinct thematic roles or distinct grammatical functions. In the next section, I turn to double object constructions, which were also initially analyzed as symmetric and have since been reanalyzed as asymmetric. This much is uncontroversial (and widely agreed upon). What is more controversial is the question of what kind of asymmetric structure double object constructions involve, especially in languages in which the order between the two objects is not fixed. 2.6
Double object constructions
Perhaps the first structure for double object constructions every syntax student encounters is a ternary branching one of the kind given in (30b), in which the two objects stand in a symmetric c-command relationship. This,
30
Asymmetry in syntax
for example, is the structure assumed by Oerhle (1976), recently “resurrected” by Culicover and Jackendoff (2005).13 (30)
a. Gina gave Sue a book. b.
VP V
NP
NP
gave Sue a book
However, as shown by Barss and Lasnik (1986), standard c-command diagnostics argue against such a symmetric structure and show that (at least in English) the indirect object c-commands the direct one. These diagnostics involve Principle A effects, variable binding, negative polarity item licensing, weak crossover and passivization. The contrast in (31a–b) shows that the indirect object can bind an anaphoric direct object but not vice versa, the contrast in (32a–b) shows that a quantified indirect object can bind a direct object variable but not vice versa, and the contrast in (33a–b) makes a similar point with respect to negative polarity item licensing; a negative indirect object can license a negative polarity direct object but a direct object cannot license an indirect object negpol. Examples (34a–b) show that a direct object cannot cross over the indirect object containing a coindexed pronoun.14 And finally, (35a–b) show that the indirect object is higher than the direct object, hence closer to the specifier of TP for the purposes of passivization. (31)
a. I showed Maryi herselfi . b. * I showed herselfi Maryi .
Principle A effects
(32)
a. I gave everyi worker hisi paycheck. b. * I sent itsi owner everyi paycheck.
variable binding
(33)
a. I showed no one anything. b. * I showed anyone nothing.
negpol licensing
(34)
a. % Which mani did you send ti hisi paycheck? b. * Whose payi did you send his mother ti?
(35)
a. Johni was given ti a book. b. * A booki was given John ti .
weak crossover
passivization (Larson 1988:336)
They all point towards an asymmetric structure in which the indirect object c-commands the direct one.
2.6 Double object constructions
31
(36) IO DO
There are many variants of such a structure, a sampling of which is given in (37a–c). They differ with respect to what kinds of elements mediate the relationship between the two objects; a verb in a VP shell structure (as argued by Larson 1988, for example), a prepositional element (as argued by Pesetsky 1995 and Harley 2002, among others), or a small clause head (as argued by Kayne 1984 and Beck and Johnson 2004). (37)
a.
VP V
IO V V
b.
DO tIO
(Larson 1988:353)
vP v
PP P
IO PHAVE
c.
DO
(Harley 2002:32)
VP V
HAVEP
IO
HAVE HAVE
DO
(cf. Beck and Johnson 2004:104)
The issue of what structure double object constructions involve becomes even more interesting in languages like Japanese or Polish, which allow either order of the two objects and mark them with different cases, as shown in (38a–b) for Polish and (39a–b) for Japanese. (38)
a. Jan dał Marii książkę. Jan.nom gave Maria.dat book.acc ‘Jan gave Maria a book.’ b. Jan dał Marii książkę. Jan.nom gave book.acc Maria.dat ‘Jan gave Maria a book.’
32 (39)
Asymmetry in syntax a. Taroo-ga Hanako-ni nimotu-o okutta. Taro-nom Hanako-dat package-acc sent ‘Taro sent Hanako a package.’ b. Taroo-ga nimotu-o Hanako-ni okutta. Taro-nom package-acc Hanako-dat sent ‘Taro gave a package to Hanako.’ (Miyagawa and Tsujioka 2004:5)
Such alternations raise the question of which order is basic and which one derived, or even more fundamentally, the question of whether the two orders are transformationally related at all. Both of these questions are far from being settled. Dornisch (1998), for example, takes the basic order in Polish to be DO IO (see also Bailyn 1995 and in press for similar claims about Russian). I argue in Chapter 4 that the basic order is IO DO instead. Dyakonova (2007, 2009) makes a similar case for Russian, and Dvořák (in press) and Kučerová (2007) for Czech. Hoji (1985) takes the IO DO order in Japanese to be basic, and the DO IO order to be derived via scrambling. Miyagawa (1995, 1997), Miyagawa and Tsujioka (2004), on the other hand, argue that the two orders are not transformationally derived, but are a reflex of two different argument structures, corresponding to the English double object structure and todative structure, respectively. What allows for such an analysis is the fact that the Japanese particle ni is ambiguous between a dative marker and a preposition. The structure I rely on in Chapter 4 (which deals with symmetric Move) assimilates double object constructions to applicative constructions. This is particularly apparent in languages in which the presence of an indirect object correlates with the presence of an applied morpheme on the verb. The examples in (40a–b) from Ndendeule, a Bantu language spoken in southern Tanzania, provide an illustration; (40a) is a transitive clause and (40b) a corresponding ditransitive one – it contains an extra argument mw-ana ‘child’, licensed by the applicative morpheme el.15 (40)
a. n-gɛhεni a-ki-hεmεl-a ngoβo. Ndendeule 1-guest 1-past-buy-fv 10cloth ‘The guest bought clothes.’ b. n-ghεni a-ki-n-hεmεl-εl-a mw-ana ngoβo 1-guest 1sa-past-1oa-buy-APPL-fv 1-child 10cloth ‘The guest bought the child clothes.’ (Ngonyani 1996:3)
If we allow applicative morphemes to be null, extending the applicative analysis to English double object constructions is pretty straightforward.
2.7 Double object constructions
33
Even though the verb’s morphology does not change in from (41a) to (41b), its valency does increase by one in (41b).16 (41)
a. John baked a cake. b. John baked Mary a cake.
Furthermore, applicatives do not have to be derived from transitive verbs; if an applied object is added to an intransitive verb, the result is a transitive construction. An example from Albanian is given in (42). (42)
I vrapova. him(dat.cl) ran.1sg ‘I ran for him.’
Albanian (Pylkkänen 2002:25)
Multiple applicatives are also possible. The following example from Kinyarwanda contains both a benefactive and a locative applicative. (43)
Umugóre a-rá-hé-er-a umagabo ímbwa íbiryo. woman she-pres-give-appl-asp man dog food ‘The woman is giving food to the dog for the man.’ (Kimenyi 1980:65)
One of the biggest issues in the research on applicatives concerns the syntactic behavior of the two objects. More specifically, the issue is whether both objects behave like true objects, or only one of them does (and if so, which one). Languages in which the two objects behave alike are often referred to as symmetric languages, and languages in which only one does are referred to as asymmetric languages. The properties that are typically used to diagnose the symmetric versus asymmetric behavior of the two objects involve their relative order (i.e. which one needs to be adjacent to the verb), agreement (i.e. which one determines verb agreement), passivization (i.e. which one can undergo passive movement).17 Another important development in the research on applicatives is the distinction between high and low applicatives, due to Pylkkänen (2002, 2008). A high applicative head establishes a relation between the indirect object and the event denoted by the VP. In structural terms, this means that the indirect object is the specifier of the high applicative head, and the VP (containing the direct object) is its complement, as shown in (44a) below. In a low applicative structure, the applicative head establishes a more direct relationship between the two objects; the indirect object is its specifier and the direct object is its complement, as shown in (44b). In both types, the relative relationship between the direct and the indirect object is the same in that the indirect object asymmetrically c-commands the direct one.
34 (44)
Asymmetry in syntax a. high applicatives
b. low applicatives
ApplHP
VP
IO
ApplH ApplH
IO
VP V
ApplLP
V
DO
ApplL ApplL
DO
This difference in structure is also reflected in the semantic interpretation of the two applicative heads. A high applicative head is a two-place predicate taking an event as one argument, and an individual as the other one, as shown in (45). It is compatible with a wide variety of semantic interpretations. (45)
λx. λe. APPL (e, x) collapsing APPLben, APPLinstr, APPLloc
high applicative (Pylkkänen 2002:21)
The semantics of a low applicative head is more complex; such a head takes two individuals and a function as its arguments. (46)
a. λx. λy. λf <e, <s,t>>. λe. f(e,x) & theme (e,x) & to-the-possession (x,y) b. λx. λy. λf<e, <s,t>>. λe. f(e,x) & theme (e,x) & from-the-possession (x,y) (Pylkkänen 2002:22)
Pylkkänen uses three independent diagnostics to determine whether a given applicative is high or low: compatibility with intransitive verbs, compatibility with stative verbs, and the ability for the object to be modified by a depictive. First, since low applicatives describe transfer of possession between two internal arguments, they are not possible with intransitive verbs (which lack one argument). Second, low applicatives are incompatible with verbs that describe states (rather than events) because such verbs are lexically incompatible with the idea of transfer. And third, low applicatives are incompatible with depictives modifying objects because they are of the incorrect semantic type.18 According to these three diagnostics, English lacks high applicatives. First, applied arguments are impossible with intransitive verbs, as shown by the ungrammaticality of (47a). Second, they are incompatible with stative predicates such as to hold a book, as shown in (47b).19 And third, they cannot be modified by depictives, as shown by the ungrammaticality of (47c). (47)
a. * I ran him. b. * I held Mary her book. c. * I gave Maryi the book hungryi.
2.7 Asymmetry in Move
35
High applicatives, on the other hand, are possible with both intransitive and stative verbs, and allow depictives predicated of the indirect object. These three properties are illustrated in (48a–c) for Albanian and in (49a–c) for Luganda. (48)
a. I vrapova. Albanian him(dat.cl) ran.1sg ‘I ran for him.’ b. Agimi i mban Drites çanten time. Agimi.nom dat.clholds Drita.dat bag my.acc ‘Agim holds my bag for Drita.’ (Pylkkänen 2002:25) c. Drita i poqi Agimiti të lodhuri. Drita.nom cl baked Agim.dat 3sg-masc tired ‘Drita baked for Agim tired.’ (Pylkkänen 2002:36)
(49)
a. Mukasa ya-tambu-le-dde Katonga. Luganda Mukasa 3sg.past-walk-appl-past Katonga ‘Mukasa walked for Katonga.’ b. Katonga ya-kwaant-i-dde Mukasa ensawo. Katonga 3sg.past-hold-appl-past Mukasa bag ‘Katonga held the bag for Mukasa.’ (Pylkkänen 2008:20) c. Mustafa ya-ko-le-dde Katongai nga mulwaddei. Mustafa 3sg.past-work-appl-past Katonga sick ‘Mustafa worked for Katonga while Katonga was sick.’ (Pylkkänen 2008:31)
The upshot of the discussion in this section is that irrespective of which structure for double object constructions we adopt, the relationship between the two objects has to be asymmetric. We saw both theoretical and empirical arguments in favor of such a structure. In the next section, I turn to the evidence typically used to argue for the asymmetric nature of movement. 2.7
Asymmetry in Move
The most convincing evidence in favor of the claim that movement is asymmetric comes from locality considerations. The data used to support this claim typically involve the configuration schematized in (50a–b), in which γ is the Probe (i.e. the element with some uninterpretable feature that needs to be checked) and α and β are two potential Goals (i.e. two elements that could in principle check this feature). Furthermore, both α and β have to be in the c-command domain of γ, and α has to c-command β. In such cases, the movement of α is possible, as shown in (50a), whereas the movement of β crossing α is not, as shown in (50b):20
36 (50)
Asymmetry in syntax a.
b. γ
γ
α
α
β
β
This is a familiar Relativized Minimality configuration, whose effects are found in the domain of A-bar movement, A-movement and head movement. Rizzi’s (1990) original formulation of Relativized Minimality is given in (51). Relativized Minimality requires all traces to be governed and disallows government by a head across an intervening head, by an element in an A-bar position across an intervening A-bar position, and by an element in an A position across an intervening A position. (51)
Relativized Minimality X α-governs Y only if there is no Z such that a. Z is a typical potential α-governor for Y; b. Z c-commands Y and Z does not c-command X.
(Rizzi 1990:7)
More recently, the effects of Relativized Minimality have been subsumed under various types of economy principles, such as the Minimal Link Condition, given in (52), or Attract Closest, given in (53). (52)
Minimal Link Condition (Chomsky 1995:296) α can raise to target K only if there is no legitimate operation Move β targeting K, where β is closer to K.
(53)
Attract Closest K attracts F if F is the closest feature that can enter into a checking relationship with a sublabel of K.
Chomsky’s (1973) Superiority Condition, given in (54) below, was an early principle that mandated rules to be asymmetric by privileging the structurally higher (or superior) element. It had the flavor of a general economy principle, thus predating current economy principles. (54)
Superiority Condition (Chomsky 1973:101) a. No rule can involve Χ, Υ in the structure ... X ... [ … Z ... Υ . . . ] . . . where the rule applies ambiguously to Z and Υ and Z is superior to Y. b. the category A is “superior” to the category B in the phrase marker if every major category dominating A dominates B as well but not conversely.
A well-known illustration of the asymmetric nature of A-bar movement comes from multiple wh-questions, in which (at least in languages like
2.7 Asymmetry in Move
37
English) movement of the higher wh-phrase is possible but movement of the lower one is not: (55)
a. Whoi ti bought what? b. * Whati did who buy ti ?
However, there are many well-known exceptions to the claim that whmovement is an asymmetric operation, coming both from English and other languages. For example, as pointed out by Barss (2000) and Pesetsky (2001), among others, superiority effects in multiple wh-questions get considerably weaker if the wh-phrases are discourse-linked. The result is symmetric wh-movement; movement of either the lower or the higher wh-phrase is possible: (56)
a. Which studenti ti read which article ? b. Which articlei did which student read ti?
Second, quite a few languages have been shown to allow violations of superiority even in non discourse-linked contexts. We see this both in languages that front only one wh-phrase overtly (such as German, shown in (57)) and in languages that front all wh-phrases overtly (such as SerboCroatian or Polish, shown in (58–59)). See Citko and Grohmann (2001) for further discussion. (57)
a. Weri hat ti was gekauft? who has what bought ‘Who bought what?’ b. Wasj hat wer tj gekauft? what has who bought
German
(58)
a. Koi je ti štaj kupio tj? who is what bought ‘Who bought what?’ b. Štaj je koi kupio tj? what is who bought
Serbo-Croatian
(59)
a. Ktoi coj ti kupił tj? who what bought ‘Who bought what?’ b. Co ktoi ti kupił tj? what who bought
Polish
However, these exceptions seem to be only apparent, as what look like superiority violations could be due to the fact that the movements involved here are of different types. Movement of discourse-linked whphrases is often analyzed as targeting a different position than movement of non discourse-linked ones (as argued by Boeckx and Grohmann 2004,
38
Asymmetry in syntax
for example), and movement of either non-initial or all wh-phrases in multiple wh-fronting languages is often analyzed as either targeting a different position or being driven by different types of features. Bošković (2002), for example, argues that the lack of superiority effects in Slavic languages can be attributed to the fact that the movement of wh-phrases is driven by Greed (rather than Attract); wh-phrases thus move to satisfy their own checking requirements, not because they are attracted by a C head. We see the same asymmetry in the domain of A-movement. A straightforward illustration comes from the process known as superraising, which involves movement of an argument (a subject, more specifically) from an A position to another A position across an intervening argument in yet another A position, as shown schematically in (60). (60)
* [TPsubjecti… [TPsubject … [TP ti … ] ] ]
A concrete example is given in (61) below; the subject of the embedded clause moves to the matrix subject position across the intermediate clause subject.21 (61)
* Johni seems that it is certain ti to fix the car.
A similar configuration, illustrated in (62), arises in raising to subject across an overt experiencer. Since the experiencer is closer to the matrix T, movement of the subject is predicted to be ungrammatical.22 (62)
* [TP subjecti … experiencer …
[TP ti … ] ]
Examples (63a–b) below illustrate such movement in Spanish and Italian.23 (63)
a. * Este taxistai parece a Maria [ti estar cansado]. Spanish this taxi driver.nom seems to Maria be tired. ‘It seems to Maria that this taxi driver is tired.’ (Torrego 1996:106) b. * Giannii sembra a Piero [ti non fare suo dovere]. Italian Gianni.nom seems to Piero not to.do his duty ‘Gianni seems to Piero not to do his duty.’ (Cuervo 2003a:203, citing Rizzi 1986)
A simple and straightforward illustration of asymmetry in head movement comes from subject–auxiliary inversion in examples involving two (or more) auxiliary verbs. In such cases, only the higher auxiliary can move: (64)
a. They could have left. b. Couldi they ti have left? c. * Havei they could ti left?
2.8 Asymmetry in Labeling 39 An early principle dealing with such cases is Travis’s Head Movement Constraint, given in (65) below. Informally, it says that a head may move to another head position only if there is no other head intervening. (65)
Head Movement Constraint (Travis 1984:131) An X0 may only move into the Y0 which properly governs it.
There are also exceptions to this generalization, involving the so-called long head movement, analyzed as precisely the kind of movement that the Head Movement Constraint deems to be impossible. These are the notorious cases of participle fronting in a subset of Slavic languages, discussed by Lema and Rivero (1989), Rivero (1991, 1994), among others, and illustrated in (66a–c). In all of them, the fronted participle crosses the auxiliary clitic, indicated in bold (examples from Rivero 1991). (66)
a. Pročeli e ti Petur knigata. read perf.3sg Peter book.def ‘Peter has read the book.’ b. Koupili jsem ti knihy. bought perf.1sg books ‘I have bought books.’ c. Čitaoi sam ti knjigu. read perf.1sg book ‘I have read a book.’
Bulgarian
Czech
Serbo-Croatian
However, there are many alternative accounts of this somewhat peculiar construction that do not rely on long head movement, and which, in fact, argue against long head movement. Bošković (2001), for example, analyzes the fronted participle plus clitic as a verb cluster in which the participle adjoins to the auxiliary, rather than moves across it. Ackema and Čamdžić (2003) analyze it as complex predicate formation at LF. I will not go into the details of these accounts; suffice it to say to that there are alternatives that do not rely on long head movement. To sum up, we saw in this section a fair amount of evidence for movement being asymmetric. In all three major types of movement, the element that is closest to the target is the one that moves. In the next section, I turn to the third type of asymmetry in syntax, involving labels in both Merge and Move structures. 2.8
Asymmetry in Labeling
The third mechanism considered in this book is the mechanism that determines labels in structures created by both Merge (External Merge) and
40
Asymmetry in syntax
Move (Internal Merge). In this section, I focus on the inherent asymmetry in this mechanism. As we saw in Chapter 1, the output of Merge (External and Internal alike) is (67a) in set notation and (67b) in tree notation. In both, γ represents a label of the syntactic object K. (67)
a. K = {γ,{α, β}} b. γ α
β
The answer to the question of what determines the label is constrained by the Inclusiveness Condition, which bans the introduction of new entities during the derivation. The label of an object consisting of α and β has to be constructed from the features of α and β. Chomsky (1995) discusses three possibilities; the label could be the intersection of α and β, the union of α and β, or one or the other of α and β. However, he allows only the last option because “the intersection of α and β will generally be irrelevant to output conditions, often null; and the union will be not only irrelevant but ‘contradictory’ if α and β differ in value for some feature,” which is often the case (Chomsky 1995:244). This is where the asymmetry of Labeling comes in; only one of the two elements determines the label. Furthermore, for any case of Merge, only one choice is possible. For example, when a preposition merges with a noun, only the preposition can project as the label. (68)
a. Merge to and school, project to
b. Merge to and school, project school
to to
school school
to
school
The reason behind this asymmetry seems fairly straightforward; it is the preposition (not the noun) that has its selectional feature satisfied by merging with the noun. There are many different implementations of this general idea. Adger (2003:91), for example, conceives of selectional features as uninterpretable categorical features that need to be satisfied by merging with objects of appropriate types. In both (68a) and (68b) above, the preposition has an uninterpretable nominal feature (i.e. [uD] feature), which is valued when it merges with a DP complement. The label is thus determined by the object that has its uninterpretable feature (its [uF] feature) satisfied via the Merge operation, as shown schematically in (69). αuF
(69) αuF
βF
2.8 Asymmetry in Labeling 41 This is also the conclusion reached by Pesetsky and Torrego (2006), who argue that both Internal and External Merge are feature-driven, governed by the so-called Vehicle Requirement on Merge, given in (70). (70)
Vehicle Requirement on Merge (Pesetsky and Torrego 2006:25) If α and β merge, some feature F of α must probe F on β.
Boeckx’s (2008) Probe-Label Correspondence Axiom, given in (71) below, also relies on the intuition that the Probe is the element that projects as the label. (71)
Probe-Label Correspondence Axiom (Boeckx 2008:96) The label of {α, β} is whichever α or β probes the other, where Probe = lexical item whose uF gets valued.
Interestingly, Chomsky (2008) departs from this fairly intuitive view of labels, and, instead of relying on feature checking, suggests the Labeling algorithm given in (72a–b). Notice that this algorithm also abandons the idea that there is a unified mechanism to determine labels in External and Internal Merge structures. The algorithm has two parts; (72a) deals with labels in External Merge, and (72b) with labels in Internal Merge structures. (72)
a. In {H, α}, H an LI, H is the label b. If α is internally merged to β, forming {α, β}, then the label of β is the label of {α, β} (Chomsky 2008:145)
H in (72a) stands for a head and LI for a lexical item. A lexical item is any element selected directly from the Numeration (as opposed to a complex object resulting from a previous Merge operation). From this perspective, both lexical and functional elements “count” as lexical items. However, there are cases, noted by Chomsky himself, that either do not follow from the algorithm or for which the algorithm makes incorrect predictions. It works nicely when a head merges with a non-head. However, it is problematic for cases involving two heads, or two non-heads, illustrated in (73a–b) respectively. In (73a), since we are merging two lexical items, either should be able to project, and in (73b), since we are merging two non-lexical items, neither should be able to project. The first Merge in every derivation is an example of (73a) and merging a subject as a specifier of vP or TP is an example of (73b). (73)
a.
b.
? X
Y
? XP
YP/Y
42
Asymmetry in syntax
The alternative views on Labeling outlined above, which link labels to selection or feature checking, avoid this problem. Even in the first Merge one element is selected by another one. Merging the subject in [Spec, vP] satisfies the selectional feature of v, and merging it in [Spec, TP] satisfies the EPP feature of T. To sum up briefly, while the different views on Labeling outlined in this section differ in specifics, they all take labels to be asymmetric. The asymmetry lies in the fact that the label is always determined by one of the two merged elements. We also saw in this chapter both theoretical and empirical reasons behind the standard minimalist view that Merge needs to create asymmetric structures (because only such structures are linearizable) and that Move is asymmetric (as evidenced by standard locality considerations).
3
Symmetry in Merge
3.1
Introduction
In the previous chapter, I surveyed the evidence typically used to support the claims that Merge, Move and Labeling are inherently asymmetric mechanisms. In the rest of the book, I turn to the evidence to support the claim that all three can also be symmetric under certain well-defined circumstances. This might come as a bit of a surprise; however, my goal here is not to renounce (or reanalyze as symmetric) all the phenomena that I showed to be asymmetric in Chapter 2, but, rather, to point out other sources of symmetry in the grammar. In the current chapter, I motivate the existence of a very specific kind of symmetric structure created by Merge: the multidominant structure given in (1).1 (1)
K α
L β
γ
The symmetry in (1) is quite obvious; α and β bear the same kind of relationship to γ. For now, the exact nature of this relationship is immaterial; it could be thought of as either mutual c-command (both α and β c-command γ and both are c-commanded by it) or sisterhood, which is an inherently symmetric relationship (both α and β are γ’s sisters).2 The symmetry becomes even more straightforward if we “flatten” (1), turning it into (2) below, and draw a line through the shared element γ. Now the parts on both sides of the line, K and L, are fully symmetric. In what follows, I refer to the two parallel parts as planes, borrowing Muadz’s (1991) term, and to the shared element as the pivot.3,4 (2)
L
K
Plane A
α
γ
Plane B
β pivot
43
44
Symmetry in Merge
The discussion in this chapter is structured as follows. In the next three sections (Sections 3.2–3.5), I introduce symmetric structures, focusing on how they are generated, linearized and have their features checked. In Section 3.6, I turn to the empirical evidence, surveying constructions that can be (or have been) analyzed as multidominant. The ones I focus on are: across-theboard wh-questions (Section 3.6.1), questions with conjoined wh-questions (Section 3.6.2), right node raising (Section 3.6.3), gapping (section 3.6.4), serial verbs (Section 3.6.5) and free relatives (Section 3.6.6). Note that not all of them are coordinate, which shows that multidominance is independent of coordination and which is consistent with the asymmetric approach to coordination that I advocated in Chapter 2. Section 3.7 is the conclusion. 3.2
Symmetric Merge: Parallel Merge5
Even though multidominant structures are not consistent with the standard views of phrase structure (as they violate the Single Root Condition), their existence goes back to the early days of generative grammar. What makes the mechanism introduced in this section different from many of its conceptual predecessors is that it can derive multidominance without introducing any extra assumptions into the grammar. Instead, it treats multidominance as a natural consequence of basic minimalist assumptions about the nature of structure building and movement operations, as it combines the properties of External Merge and Internal Merge. We saw in Chapter 1 that External Merge takes two disjoint objects as its input, whereas Internal Merge takes two objects, one of which is a part of the other, as its input. This third type, which in Citko (2000, 2003, 2005) I dubbed Parallel Merge, is like External Merge in that it takes two distinct objects as its input. However, it is also like Internal Merge in that it combines one with a subpart of the other. Parallel Merge structures are thus a result of a two-step process, illustrated in (3a–b). First, α merges with γ, and next, β merges with γ, as shown in (3b). As a result, γ becomes the pivot shared between α and β.6,7 (3)
a. Merge α and γ, project α α
γ
α α
γ
b. Merge β and γ, project β α α
β γ
α α
β β
γ
3.2 Symmetric Merge: Parallel Merge
45
Parallel Merge builds on the insights of earlier proposals that all rely on the same intuition, the intuition that one element is shared between two nodes, occupies two positions in a tree simultaneously, or exists in two dimensions at the same time. Let me discuss these proposals briefly before moving on to the specifics of the Parallel Merge mechanism. Many restrict multidominance to coordination, either by modifying existing phrase structure rules (or phrase structure building mechanisms) to generate multidominant structures, or by modifying existing tree axioms to allow multidominant representations. Williams (1978), for example, conceives of the conjuncts in a coordinate structure as so-called parallel factors and derives many of the constraints on ATB wh-movement from the parallelism constraint on such factors.8 He accomplishes this by allowing the following bracketed structures into the grammar: (4)
The structure [X1]C1 and [Xn]Cn
is a well-formed labeled bracketing if X1, …, Xn are.
(Williams 1978:32)
Goodall (1983, 1987) derives multidominance from a union of reduced phrase markers.9 If one phrase marker contains terminal nodes abc and the other one edc, they can be collapsed into one phrase marker, with c as a shared node, as shown in (5a–c). (5)
a.
b. a
c. e
b
c
a d
c
e b
d
c
A slightly different approach is taken by McCawley (1982), who relaxes standard tree axioms to allow both multidominant trees and trees with crossing nodes. A yet different approach is taken by Moltmann (1992a, 1992b), who, following Muadz (1991), captures multidominance by introducing the rule given in (6). The resulting system is multidimensional rather than multidominant, as the two conjuncts exist in two different dimensions. (6)
A → J where Bi is a legal expansion of A and J is a coordinator. (Moltmann 1992b: 265)
De Vries (2005a, 2005b, 2008) proposes a special kind of Merge, b-Merge (which stands for behindance Merge), which allows one conjunct in a coordinate structure to stand behind another one.
46
Symmetry in Merge
Multidominance is also not limited to minimalism (and its predecessors). Chen-Main (2006), for example, argues for the existence of multidominant structures in Tree Adjoining Grammar. In TAG, multidominant structures are a result of node contraction, a process introduced by Sarkar and Joshi (1996) to handle coordination. Node contraction is illustrated in (7a–c); it takes two nodes that are marked for contraction (marked in bold in (7a–b)) and collapses them into a single node, as shown in (7c). Node contraction is essentially equivalent to Goodall’s (1987) union of phrase markers given in (5a–c) above. (7)
a.
b.
XP1 X1
YP X1
ZP
XP2
XP1 X2
YP X2
ZP
YP
XP2 X1
X1
X2 ZP
X2
ZP
Irrespective of which of these mechanisms we adopt, there are a number of questions that the symmetry introduced by multidominance raises. The ones that strike me as most pressing in the context of current minimalist theory are listed in (8a–d). (8)
a. b. c. d.
What are the constraints on symmetric Merge? How are features valued (or checked) in symmetric Merge structures? How are symmetric Merge structures linearized? What empirical insights do symmetric Merge structures offer?
In Section 3.6, which constitutes the bulk of the discussion in this chapter, I address question (8d) by surveying constructions that can be (or have been) analyzed in a symmetric fashion. In Sections 3.3–3.5, I tackle the remaining questions listed in (8).
3.3
Constraints on symmetric Merge
The first question concerns the restrictions on shared elements, the pivots. The best-case scenario is that there are none. Let us see if this is feasible. What this means is that any element can be a pivot, irrespective of its phrase-theoretical status. In particular, pivots can be maximal (as shown in (9a–b)), minimal (as shown in (10a–b)), or intermediate (as shown in (11a–b)). The structures in (9–12a) are bare phrase structure theoretical representations, and the ones in (9–12b) are their more “traditional” X-bar theoretical counterparts. Furthermore, nothing prevents a single category from being simultaneously minimal and maximal. This is what happens in (12a–b); the shared
3.3 Constraints on symmetric Merge
47
element γ is both maximal (because it does not project any further) and minimal (because it itself is not a projection of γ). (9)
β
α
a. α
b. γ
β
γ
a. α
(11)
β
α
γ
b. γ
β
XP
XP
X XP X
Z
δ
X
b. γ
β
WP
Z
Y
β
α
XP Y
β
α
a.
X
b.
γ
(12)
XP
δ γ
α
a.
ZP Z
Y
γ
(10)
YP
YP Y
WP
ZP Z
XP X
Interestingly, there is one other option that Bare Phrase Structure Theory allows, which is excluded by X-bar theory. It is illustrated in (13a–b); the pivot γ is maximal with respect to one plane (the one dominated by β) but intermediate with respect to the other plane (the one dominated by γ). (13)
β
γ
a. α
b. γ
β γ
XP Y
δ
ZP X/XP
Z X
WP
There is nothing in the mechanism of Parallel Merge that would exclude such “mixed” cases. Thus, if they exist, they provide a good argument in favor of a bare approach to phrase structure, as standard X-bar theory does not allow a single element to be simultaneously maximal and nonmaximal. In fact, this possibility might allow us to capture case mismatches in free relatives and case proximity effects in right node raising constructions, which I will discuss in Sections 3.6.3 and 3.6.6, respectively.10 3.4
Feature checking in symmetric Merge structures
The second question in (8) involves feature valuation (or checking) in symmetric configurations. Agree, the operation responsible for providing values to unvalued features, is a one-to-one relationship between a Probe
48
Symmetry in Merge
and a Goal in its (c-command) domain. A paradigmatic example of Agree is a relationship between a finite T (the Probe) and a subject in [Spec, vP], schematized in (14), which results in T having its ϕ features valued and the subject having its case feature valued.11 (14)
a.
b.
TP
Tuϕ: __
vP
TP Tuϕ:val
DP uC:__ϕ:val v v
vP
DP uC:Nom ϕ:val v VP
v
VP
The configurations we encounter in symmetric Parallel Merge structures are different in that they involve one-to-many or many-to-one relationships between Probes and Goals. This raises the following question. If there are two clauses with a shared subject, how does this shared subject get its case features valued? The opposite issue arises if the shared element is a head that itself can value case features on the Goal, such as a v or a T head. These two scenarios are illustrated in (15a–b). (15a) involves two Probes (two v heads) and one Goal (a shared object), and (15b) one Probe (a shared T head) and two Goals (two subjects). (15)
a.
vP v uϕ:—
vP v uϕ:—
VP
V
b.
VP
V TP
Tuϕ:—
DPuC;— , ϕ:val TP
vP
DPuC;— ,ϕ:val v
vP v
DPuC;
— ϕ:val
VP
v
v VP
The issue is not unique to multidominant structures. Hiraiwa (2005), for example, explores the possibility that a single Probe can value features on two Goals simultaneously in a process he calls Multiple Agree.12 The cases he discusses are different from the ones under consideration here, as they do not involve multidominance at all. (16a) below provides a schematic representation of the data Hiraiwa was concerned with, and (16b–c) give illustrative examples. (16b) shows that in Icelandic, T can undergo Multiple Agree with both the associate of the expletive morgum studentum ‘many
3.4 Feature checking in symmetric Merge structures
49
students’ and the embedded subject tölvurnar ‘computers.’ As a result, the matrix verb shows plural agreement. If the intervening associate is singular, as in (16c), Multiple Agree is blocked, and only default agreement on the verb is possible. (16)
a. T uϕ
>
DP uCase,ϕ
>
DP uCase,ϕ
b. það finnst/finnast morgum stúdentum tölvurna ljótar exp find.sg/find.PL many student.dat.PL computers.dat.pl.nom ugly ‘Many students find the computers to be ugly.’ c. það finnst/*finnast einverjum stúdent tölvurnar ljótar. exp find.sg/*find.PL some student.dat.SG computers.dat.PL.NOM ugly ‘Some student finds the computers to be ugly.’ (Hiraiwa 2005:51)
The cases discussed by Hiraiwa involve a single Probe and multiple Goals, schematized in (17a), which suggests that the opposite should also be possible, and that Agree between two Probes and one Goal (schematized in (17b)) should also be allowed. (17)
a. Puϕ>GuCase,ϕ, GuCase,ϕ b. Puϕ, Puϕ>GuCase,ϕ
Multiple Goal Agree Multiple Probe Agree
Given Multiple Agree, the shared DP in (15a) can have its case features valued by two v heads simultaneously and can value ϕ-features of two v heads simultaneously as well, yielding (18a) below. And the shared T in (15b) can value case features on two subjects, as shown in (18b). The fact that the shared DP receives two case values (each from a different v) is not a problem if the two case values are identical. If they are distinct (for example, one is accusative and the other one nominative), the morphological component will determine whether the result is possible or not. If the lexical inventory of the language in question contains an appropriate syncretic form (nominative and accusative in this case), the result is going to be grammatical. (18)
a.
vP v uϕ:val
vP VP
V
v uϕ:val
V
VP
DPuC:val1, val2, ϕ:val
50
Symmetry in Merge b.
TP
TP
Tuϕ:__
vP
DPuC:Nom, ϕ:val v
vP DPuC;Nom, ϕ:val
v VP
v
v VP
A related question is which of the two DPs in (18a) values T’s unvalued ϕ-features. If the two subjects agree in ϕ-features, the issue does not arise. However, what happens if the two subjects disagree in features? Grosz (2009) addresses precisely this issue and argues that such configurations can give rise to what he calls cumulative agreement, which allows a shared functional head to end up plural if the two unshared elements it agrees with are singular. 3.5
Linearization of symmetric Merge structures
The next question in (8) concerns linearization of symmetric structures. As Williams (1978:42) puts it, “sentences are not spoken in ‘ATB format’–that is, with both conjuncts uttered simultaneously.” We saw in Chapter 2 that the Linear Correspondence Axiom of Kayne (1994) only allows structures of the kind given in (19a), in which asymmetric c-command maps onto linear precedence, as shown in (19b). The resulting ordering is transitive, total and antisymmetric. (19)
a.
E A a
E B
D
b
C c
b.
d(A) = {, , }
A schematic representation of a multidominant structure that is the focus of this chapter is given in (20). In this structure, C (and everything C dominates, which is c, D and d) is shared between A and B.
3.5 Linearization of symmetric Merge structures (20)
a.
51
B A a
B b
C c
D
d
b.
d(A) = {, , , , , , , }
The problem with the ordering in (20b) lies with the pairs in bold. Since a asymmetrically c-commands c and d, it should precede both of them. However, A (which dominates a) also asymmetrically c-commands b, and b in turn asymmetrically c-commands c and d. Thus, on the one hand, c and d should precede b (by virtue of A asymmetrically c-commanding b), and, on the other hand, b should precede c and d (by virtue of asymmetrically c-commanding them). The presence of both and in the set given in (20b) violates antisymmetry. The same goes for the pairs and . The issue of linearizing multidominant structures is more general, and, strictly speaking, it is not necessarily tied to the LCA. Irrespective of how we conceive of the mapping between hierarchical structure and linear order, there is the question of where (and how many times) to linearize the terminals of C in a structure like (20a). Logically speaking, they could be linearized in A, in B, or in both A and B, giving rise to the following possibilities. (21)
a. acdb b. abcd c. acdbcd
terminals of C linearized within A terminals of C linearized within B terminals of C linearized within both A and B
In Bare Phrase Structure Theory, a similar problem arises even for simple (i.e. not necessarily multidominant) structures created by Merge. We saw in Chapter 2 that the first Merge of two lexical items always creates a symmetric c-command configuration.13 But we also saw various ways to solve this problem. For example, either of the two objects in a symmetric configuration could undergo movement out of the symmetric configuration (which is the idea behind Moro’s Dynamic Antisymmetry Theory), or one of them could be null (which would make linearizing it a non-issue on the assumption that only overt elements need to be linearized), or one of them could have undergone self-Merge first, which would make it more complex for the purposes of linearization. The logic behind all these solutions is the same – to alter the structure in a way that makes it compatible with the LCA. We
52
Symmetry in Merge
could thus apply the same logic to multidominant structures. If C in (20a) above undergoes overt movement or is null to begin with, linearization is not a problem. This is the option I explored in Citko (2005); however, I now believe that it is not sufficient. Ultimately, it is an empirical question of what kinds of symmetric structures are linearizable. Does the grammar only allow multidominant structures in which the shared elements have moved (because only such structures are linearizable), or is there a way to linearize shared elements in situ? Two of the constructions I consider in this chapter (right node raising and wh-questions with conjoined wh-pronouns) suggest that there must be a way to linearize shared elements without movement. This shows that we need to explore non-movement solutions as well. There are many different ways that have been explored in the literature to handle linearization of multidominant structures. One is to rely on linearization algorithms devised specifically to handle such structures. Some examples of such linearization algorithms are given in (22a–d). (22)
a.
X1 and Xn
[X1 (and) … and Xn]
(Williams 1978:42)
b. Given an RPM [reduced phrase marker, B.C] containing distinct terminal strings, x1, x2, xn, for each element yi of xi, yi not an element of xi+1, there is an element yi+1 of xi+1, yi+1 not an element of xi, such that yi precedes yi+1. (Goodall 1987:23) c. If A is a 3D-node and A dominates B and C, then at PF A and B precede C. (Muadz 1991:59) d. Let X, Y, Z be variables ranging over terminal nodes or sequences of terminal nodes. (i) X P′ Y iff for every x in X′ and y in Y′ xPy.14 (ii) X Pi Y iff for the final element x in X′ and the initial element y in Y, xPiy. (Moltmann 1992a:76)
While such algorithms generally accomplish what they were designed to accomplish (namely, they linearize the structures they were designed to linearize), it is not clear how (if at all) they can be derived from more general syntactic principles. A different way to linearize multidominant structures comes from the work of Wilder (1999b, 2008), Gračanin-Yüksek (2007, in press) and Johnson (2007), who make multidominant structures linearizable by modifying the definition of c-command in a way that does not affect nonmultiply dominated structures. This is what allows them to linearize multiply dominated elements in situ. In general, they all rely on the intuition
3.5 Linearization of symmetric Merge structures
53
that the LCA can ignore some of the ordering statements (namely those that would otherwise lead to violations of antisymmetry). For example, in (20a) above, repeated as (23a), the instructions to linearize the multiply dominated node C inside A can be ignored. This excludes the orders given in (23b) and (23c) and leaves us with (23d) as the only possibility. (23)
a.
B A a
B b
C c
D
d
b. acdb c. acdbcd d. abcd
Wilder (1999b) accomplishes this by introducing the notion of full dominance, given in (24a–b). Full dominance is more restrictive than standard dominance. For example, in (23a), neither B nor A fully dominates the shared node C. (24)
a. X fully dominates α iff X dominates α and X does not share α. b. α is shared by X and Y iff (i) neither of X and Y dominates the other, and (ii) both X and Y dominate α. (Wilder 1999b:590–1)
Furthermore, both his definition of c-command and his definition of d(A), which is the image of A, rely on full dominance rather than standard dominance. (25)
a. X c-commands Y only if X does not fully dominate Y b. d(A) = the set of terminals fully dominated by A. (Wilder 1999b:590–1)
The original version of the LCA yields a set containing conflicting linearization specifications for (23a). The set is repeated in (26a). However, Wilder’s modified version of the LCA removes the pairs , from the set, yielding (26b). Now the terminals of C, not being fully dominated by A, are not included in the image of A. Thus, c-command by A does not cause the terminals of C to precede anything that A precedes. This means that the terminals of C are going to be linearized in their position in the
54
Symmetry in Merge
second “conjunct”. The resulting ordering in (26b) is total, transitive and antisymmetric. (26)
a. d(A) = {, , , , , , } b. d(A) = {, , , , }
I will not discuss the details of Wilder’s linearization system any further here. As we will see in Section 3.6.3, which deals with right node raising, it can derive one of the definitional characteristics of right node raising: the requirement that the shared element be rightmost in both conjuncts. The upshot of the discussion in this section is that there are ways to linearize multidominant structures with relatively minor modifications to the definition of c-command (which have no negative effect on standard non-multidominant structures). This removes perhaps the biggest theoretical obstacle to multidominant structures: the claim that they should be banned from the grammar because there is no way to linearize them. 3.6
Empirical support for symmetric Merge
3.6.1 Across-the-board wh-questions The insight that across-the-board (ATB) wh-questions involve a multidominant structure goes back at least to Williams (1978; see also Goodall 1983, 1987, Muadz 1991 and Moltmann 1992a for similar early proposals). For these researchers, however, multidominance was a property of coordination, not of ATB movement. The two conjuncts in a coordinate structure, irrespective of the presence or absence of ATB movement of any sort, were placed in two parallel dimensions or planes along the lines schematized in (27b).15 (27)
a. Jane and Alice saw Bill. b.
TP DP Jane
TP DP Alice T
T vP saw Bill
(cf. Goodall 1987:23)
The account I developed in Citko (2005), which is the account I expand on below, differs from these early accounts in that it divorced multidominance from coordination and thus maintained an asymmetric approach to coordination, for which we saw a lot of evidence in Chapter 2. To see
3.6 Empirical support for symmetric Merge
55
how it works, consider the derivation of the sentence given in (28a). Its Numeration, given in (28b), contains only one wh-pronoun, which is going to be shared between the two conjuncts. The end structure is given in (28c). (28)
a. What did John lose and Bill find? b. N = {what1, John1, lose1, Bill1, find1, v1, T1, C1, and1} c. CP C
&P &
TP1 T1 T
&
TP2 T2
vP1 v1
John v
VP1
vP2 Bill
v2
lose
VP2 find
what
The wh-pronoun is not the only shared element; v, T and C heads are also shared between the two conjuncts. The evidence that they are shared comes from the following considerations. The reason why T is shared is due to the fact that the two conjuncts in an ATB question have to agree in tense, as shown by the ungrammaticality of the examples in (29a–b), in which one conjunct is in the present and the other one in the past.16 (29)
a. * Whati did John lose ti and Bill will find ti? b. * Whati will Bill find ti and John lost ti?
By the same logic, the fact that voice mismatches are impossible in ATB questions implies that the v head has to be shared as well. In (30a–b), one conjunct is active and the other one passive, and the result is ungrammatical. (30)
a. * Whati did John lose ti and was found ti by John? b. * Whati was found ti by Bill and John lose ti?
There are a number of movements that take place in (28c); the subjects move to their respective [Spec,TP] positions, T moves to C, and the whphrase moves to [Spec,CP]. Furthermore, movement has to proceed
56
Symmetry in Merge
through phase edges, which means that the wh-phrase stops in [Spec,vP] (yet another shared position) on its way to [Spec,CP].17 (31)
CP C
whati C
&P
TP Johnj
& T &
TP
vP
T
Billj
ti
T
vP tj
vP v
v
vP VP
lose
tj
v v
VP find
ti
After all the movements have taken place, the structure can be linearized without any glitches, as all the shared elements either moved to non-shared positions or were null to begin with. There are a number of empirical advantages of such a symmetric approach to ATB wh-movement; (32a–d) lists the ones I focus on in the rest of this section. (32)
a. ATB wh-movement does not obey the Coordinate Structure Constraint; b. Multiple ATB wh-movement is not possible; c. ATB wh-movement is subject to matching effects; d. Covert ATB movement is not possible.
The exceptional status of ATB wh-questions with respect to the Coordinate Structure Constraint is illustrated in (33a–b). The ungrammatical (33a) involves movement out of only one conjunct, whereas the grammatical (33b) involves movement out of both conjuncts simultaneously. (33)
a. * Whati did John lose glasses and Bill find ti? b. Whati did John lose ti and Bill find ti ?
3.6 Empirical support for symmetric Merge
57
On the Parallel Merge account, this contrast follows from the fact that there is one wh-pronoun shared between two conjuncts. The second property, i.e. the lack of multiple fronting in ATB questions, is illustrated by the fact that there are no ATB questions involving a configuration schematized in (34), in which the number of fronted wh-elements matches the total number of gaps inside the two conjuncts. In (34), this number is two (there are two fronted wh-phrases and a total of two whtraces, one in each conjunct). (34)
[CP WHi WHj [TP …. ti…] and [TP …tj…] ]
The ungrammaticality of the English example in (35a), which involves precisely such a configuration, can be attributed to the fact that English complementizers do not allow multiple specifiers (or, to put it in a more theory-neutral fashion, to the fact that English does not allow multiple whfronting), which is what accounts for the ungrammaticality of (35b). (35)
a. * Whatj whati did John lose ti and Bill find tj? b. * Whoi whatj ti find tj ?
Interestingly, questions of this sort are also impossible in languages that allow multiple wh-fronting, as shown by the ungrammaticality of (36) in Polish, which is a multiple wh-fronting language. (36)
* Coi coj Jan zgubił ti a Piotr znalazł tj ? what what Jan lost and Piotr found ‘What did Jan lose and Piotr find?’
This does not imply that Polish ATB wh-questions can never contain more than one fronted wh-pronoun. Example (37a) is grammatical in spite of containing two fronted wh-pronouns. Note, however, that in this case both wh-pronouns are moved from both conjuncts. This contrasts with the ungrammatical example in (37b), in which one of the two moves only from one conjunct. (37)
a.
Coj komui Jan kupił tj ti a Piotr wysłał tj ti ? what whom Jan bought and Piotr sent ‘What did Jan buy for whom and Piotr send to whom?’ b. * Coj komui Jan kupił tj ti a Piotr zgubił tj? what whom Jan bought and Piotr lost ‘What did Jan buy for whom and Piotr lose?’
The third property listed in (32) involves case, in particular the requirement that the fronted wh-phrase satisfy the case requirements imposed on it within both conjuncts, as shown by Borsley (1983), Dyła (1984) and
58
Symmetry in Merge
Franks (1993, 1995). This is what accounts for the contrast in grammaticality between (38a) and (38b); (38a) is grammatical because the verbs inside the two conjuncts (lubić ‘like’ and kochać ‘love’) assign the same case on their complements. (38b), on the other hand, is ungrammatical because the verb in one conjunct (lubić ‘like’) assigns accusative case and the verb inside the other conjunct (ufać ‘trust’) dative. This results in ungrammaticality irrespective of whether a dative or an accusative wh-pronoun is used.18 (38)
a.
Kogo Jan lubi tACC a Piotr kocha tACC ? who.acc Jan like and Piotr love ‘Whom does Jan like and Piotr love?’ b. * Kogo/*komu Jan lubi tACC a Piotr ufa tDAT? who.acc/who.dat Jan like and Piotr trust ‘Whom does Jan like and Piotr trust?’
This case matching requirement follows from the fact that there is a single wh-pronoun shared between two clauses. Another property that follows nicely from the symmetric structure for ATB questions involves covert ATB movement. On the assumption that movement of the shared element is required for the structure to be linearizable, covert ATB wh-movement is predicted not to be possible. This is indeed what we find; the Japanese and Korean examples in (39a–b) with wh-pronouns in situ can only be interpreted as questions about two distinct individuals. They cannot be questions about a single individual. For such an interpretation to be possible, the wh-pronouns have to undergo overt ATB movement, just as in English-type languages.19 (39)
a. John-ga dono hito-o aisitei-te Mary-ga dono Japanese John-nom which person-acc love-and Mary-nom which hito-o nikundeiru-no? person-acc hate-Q ‘Which person does John love and which person does Mary hate?’ b. John-i enu salam-ul salangha-ko Mary-ka enu Korean John-nom which person-acc like-and Mary-nom which salam-ul miweha-ni? person-acc hate-Q ‘Which person does John like and which person does Mary hate’ (Cho and Zhou 2002:523)
Furthermore, as shown by Bošković and Franks (2000), ATB quantifier raising, which is covert in languages like English, does not exist. The lack of the interpretation in which the universally quantified noun phrases have wide scope over existentially quantified ones in (40a–b) provides evidence.
3.6 Empirical support for symmetric Merge (40)
59
a. Some delegate represented every candidate and nominated every candidate. some > every, * every > some b. Some boy hugged every girl and kissed every girl. some > every, * every > some (Bošković and Franks 2000:114)
The symmetric account outlined above is by no means the only conceivable account of ATB questions; prominent alternatives involve sideward movement (as proposed by Nunes 1999, 2001, 2004 and Hornstein and Nunes 2002) or null operator movement (as proposed by Franks 1993, 1995, Munn 1993, 2001, among others). Let me thus conclude this section by comparing the multidominant account advocated here with these two alternatives. Sideward movement involves movement across two disjoint trees, schematized in (41). (41)
K
L
… αi …
… ti …
For this derivation to converge, somewhere down the line K and L must be integrated into a single object. This is what happens in (42); L becomes the complement of X and K its specifier. (42)
XP X
K … αi … X
L … ti …
In ATB wh-questions, sideward movement would take place across two conjuncts, as shown in (43). (43)
CP C
whati C
&P TP1
John lose ti &
& TP2 Bill find ti
60
Symmetry in Merge
The Parallel Merge account and the sideward movement account are alike in that they both involve a single wh-phrase (either moving “from” one conjunct to the other or being merged in both conjuncts simultaneously).20 Both have to relax some fairly standard assumptions about phrase structure or movement; the Parallel Merge approach relaxes the Single Root Condition and the sideward movement approach relaxes the assumption that movement happens within a single syntactic object. Furthermore, both approaches have to allow a single element to enter into multiple relationships. A wh-phrase in an object ATB wh-question has to satisfy case requirements of two verbs, either by moving from the complement position of one verb to the complement position of the other verb or by being merged as a complement of both. Therein lies an important difference between the two approaches. We saw above that on the multidominant account, the process responsible for multiple case feature valuation can be thought of as a mirror image of Hiraiwa’s (2005) Multiple Agree, i.e. Agree between two distinct Probes and a single Goal. On the sideward movement account, the wh-phrase literally moves from one case position to another one. This seems to require a certain amount of look-ahead; the wh-phrase has to keep its case features active after Agree in the first conjunct. Furthermore, the values assigned to it in both conjuncts have to be identical. Typically, movement from one case position to another case position is deemed to be impossible; once an element has had its case feature valued, it is “frozen in place.” In other constructions that have been claimed to involve movement to a case position, the case that the moving element gets in its initial position can be “overridden” by the case it gets in its final position. This is typically what happens in headed relatives on the so-called head promotion account; the head moves from the relative clause-internal position to the relative clause-external head position. Both of these positions are case positions. The contrast from Polish in (44a–b) provides an illustration. In (44a), the relative clause-internal position is a position in which the head gets accusative case, and the relative clause external position is a nominative case position. And this is the case that surfaces on the head. Crucially, the head cannot maintain its “lower” accusative case, as is shown by the ungrammaticality of (44b).21 (44)
a. Książkai, którą wczoraj przeczytałam ti, jest na liście book.nom which.acc yesterday read tacc is on list bestsellerów. bestsellers ‘The book that I read yesterday is on the bestsellers list.’
3.6 Empirical support for symmetric Merge
61
b. * Książkęi, którą wczoraj przeczytałam ti , jest na liście book.acc which.acc yesterday read tacc is on list bestsellerów. bestsellers ‘The book that I read yesterday is on the bestsellers list.’
The null operator movement alternative, due to Munn (1993), assimilates ATB wh-questions to parasitic gaps, and assumes both involve movement of an empty operator (in addition to real wh-movement), as shown in (45a–b).22 (45)
a. Which papersj did Mary read tj and OPi John file ti ? b. Which paperj did Mary file tj OPi without reading ti?
The only difference between the two is the fact that ATB questions involve movement of a null operator inside the second conjunct (rather than an adjunct), as illustrated more explicitly in (46).23 (46)
CP which papersi
C
C did
TP TP
BP B
Mary read ti OPi B and
TP John file ti
While the empty operator movement account does not necessarily predict no differences whatsoever between parasitic gaps and ATB questions, it does establish a fundamental parallelism between the two.24 This is counter to the well-known observation, due to Postal (1993), that ATB whquestions are in general more permissive than parasitic gaps. For example, parasitic gaps are only possible with DP gaps, whereas ATB questions are not subject to this restriction.25 (47)
a. * How tiredi can one be ti without feeling pg? b. How heavyi do you think Bill is ti and Peter will become ti?
Hornstein and Nunes (2002) argue that an independently motivated parallelism requirement on coordinate structures is responsible for the more permissive nature of ATB questions.
62
Symmetry in Merge
An interesting question that arises for both the null operator account and the sideward movement account of ATB questions and parasitic gaps concerns a crosslinguistic correlation between the availability of ATB questions and the availability of parasitic gap constructions. There are languages that are reported to lack parasitic gap constructions; however, I am not aware of any reports of languages lacking ATB questions.26 While a typological study to test this correlation would be an interesting research endeavor, it goes beyond the scope of this chapter. To sum up briefly, we have seen in this section how the properties of ATB wh-questions follow from (and thus provide an argument for) a symmetric structure. This is my first empirical argument in favor of symmetry in Merge. In the next section, I turn to the second case study, wh-questions with conjoined wh-pronouns. 3.6.2 Wh-questions with conjoined wh-pronouns Another construction whose properties follow nicely from a symmetric structure is a wh-question with two (or more) conjoined wh-phrases, exemplified in (48a–b).27 (48)
a. What and where did Sally sing? b. What and why did you eat?
(Gračanin-Yüksek 2007:18) (Gračanin-Yüksek 2007:40)
I will henceforth refer to such questions as wh&wh questions. Descriptively speaking, they are a “mirror image” of ATB questions; in ATB questions, there are two clauses sharing a single wh-phrase, and in wh&wh questions, there are two wh-phrases “sharing” a single clause. Such questions exhibit a number of puzzling syntactic (as well as semantic) properties. The ones noted in the literature are listed in (49a–d). (49)
a. Wh&wh questions allow violations of the Law of Coordination of Likes; b. Arguments of different grammatical functions cannot be coordinated; c. Wh&wh questions are only possible with optionally transitive verbs; d. Wh&wh questions favor single pair readings.
Some of these properties, most notably (49b) and (49c) are subject to crosslinguistic variation. In this chapter, I focus solely on English wh&wh questions and refer the interested reader to Citko (in press-a), Citko and Gračanin-Yüksek (submitted), Gribanova (2009), Scott (2010) and
3.6 Empirical support for symmetric Merge
63
Tomaszewicz (2010) for a discussion of crosslinguistic variation, with the focus on Slavic languages (Croatian in Gračanin-Yüksek 2007, Polish in Citko in press-a and Tomaszewicz 2010 and Russian in Gribanova 2009 and Scott 2010). The description and account of English wh&wh questions I summarize below is that of Gračanin-Yüksek (2007), which is the first systematic description of such questions couched in minimalist terms. For a non-minimalist alternative, see Whitman (2002, 2004). The first property listed in (49) is illustrated by the contrast in (50a–b). It shows that two wh-pronouns of different categories, unlike their non-wh counterparts, can be coordinated without violating the Law of the Coordination of Likes. (50)
a. [What]DP and [where]AdvP did Sally sing? b. * Mary sang [an aria]DP and [at the Met]PP.
The second property involves restrictions on what kinds of wh-elements can be coordinated. The contrast in (51a–b) shows that in English, coordination of an argument wh-phrase with an adjunct wh-phrase is possible, but coordination of two argument wh-phrases (such as a subject with an object) is not. (51)
a. What and why did you read? b. * Who and what read?
The third property has to do with the kinds of verbs that are possible in wh&wh questions, to be more specific, with the transitivity restrictions on these verbs. In English, wh&wh questions are impossible with obligatorily transitive verbs (if one of the coordinated wh-phrases is an object), as observed by Whitman (2002). This is what accounts for the contrast in (52a–b). (52)
a. What and why did you eat? b. * What and why did you devour?
Intuitively, this contrast is related to the contrast in (53a–b) below, and the ungrammaticality of both (52b) and (53b) reduces to the fact that the verb devour is obligatorily transitive. (53)
a. What did you eat and why did you eat? b. *What did you devour and why did you devour?
64
Symmetry in Merge
The final property of wh&wh questions to consider involves interpretation, in particular the fact that they favor single pair readings. Thus, the question in (54) can be given the single pair answer in (55a) or (55b) but not the pair list one in (55c).28 (54)
What and where did Sally sing?
(55)
a. Sally sang Mozart and she sang at the Met. b. Sally sang Mozart at the Met. c. Sally sang an aria at the Met, a hymn at church and a lullaby at home.
√SP √SP *PL
In this respect, English wh&wh questions differ from standard (i.e. noncoordinated) multiple wh-questions, which exhibit the opposite pattern; they disallow single pair answers and allow pair list answers. The question in (56) allows the answer in (57a) but disallows the one in (57b). (56)
Where did Sally sing what?
(57)
a. Sally sang an aria at the Met, a hymn at church and a lullaby at home. √PL b. Sally sang an aria at the Met. *SP
This contrast between coordinated and non-coordinated multiple whquestions is particularly visible in examples which are pragmatically biased towards a single pair reading, such as the ones in (58a–b), modeled on Whitman’s (2004) examples, in which the event being questioned involves dying or being born: (58)
a. # How did John die when? b. # When was John born where?
Corresponding coordinated wh&wh questions, on the other hand, are fine, as they imply a single event of dying or being born: (59)
a. When and how did John die? b. When and where was John born?
Gračanin-Yüksek (2007) argues that English wh&wh questions involve a bi-clausal structure in which everything but the wh-phrases themselves is shared between two CPs, as shown in (60a). All the shared elements, indicated in bold, are in a symmetric relationship with other elements inside
3.6 Empirical support for symmetric Merge
65
the two conjuncts. In the course of the derivation, both wh-phrases move to the specifiers of their respective CPs, as shown in (60b).29 (60)
a.
b.
Gračanin-Yüksek (2007) further shows that all the properties of wh&wh questions listed in (49) follow from such a symmetric structure. The fact that wh&wh questions allow coordination of unlike elements is only apparent; what is really coordinated are two CPs (i.e. like categories), in perfect compliance with the Law of the Coordination of Likes. The fact that arguments with different grammatical functions (such as a subject and an object) cannot be coordinated is linked to the fact that English does not allow null subjects. The structure of the ungrammatical example in (61a) is given in (61b). Since the wh-phrases themselves are not shared, the second conjunct would have to a have a pro subject, which is not allowed in English.
66 (61)
Symmetry in Merge a. * Who and what ate? &P b. &’ CP1 whoi
and
CP2 whatj C2
C1 C
TP1
TP2 T1
ti T
pro
T2 vP2
vP1 v
VP1
VP2
ate
tj
Transitivity restrictions in wh&wh questions can also be attributed to the structure, in particular to the fact that each wh-pronoun is inside a different conjunct. Thus, in the ungrammatical example in (62a), the second conjunct contains just the adjunct wh-phrase, which leaves the selection requirements of the verb devour not met inside this conjunct. (62)
a. * What and why did you devour? &P b. &
CP1 whati did
C 1
CP2
and
whyj
TP1 T1
you T
TP2 T2
vP1 v
C 2
vP2
VP1 devour
ti
VP2 VP2
tj
Lastly, the fact that wh&wh questions favor single pair interpretations also follows from the structure, in particular from the fact that the two whphrases end up in two different conjuncts. As first pointed out by Kazenin (2001; see also Citko in press-a and Citko and Gračanin-Yüksek submitted),
3.6 Empirical support for symmetric Merge
67
this leads to a violation of the Coordinate Structure Constraint if the second wh-pronoun “tries” to adjoin to the first one. This in turn is necessary for LF absorption, which turns two unary operators into one binary operator (see Higginbotham and May 1981 and Kitahara 1993 for details on how LF absorption works). LF absorption is what gives rise to pair list readings. Crucially, it is subject to locality, and it is blocked if the two quantifiers are separated by an island boundary, as shown by Dayal (2002, 2006), Hagstrom (1998), among others. The lack of pair list readings in wh&wh questions thus reduces to a Coordinate Structure Condition violation. LF absorption is impossible in (63) because it would have to involve wh-movement of why from one conjunct to another. (63)
[&P [CPwhati did John eat ti] [& [CPwhyi did John eat ti]]]
Let me conclude this section with a brief comparison of the symmetric structure with an alternative, involving (backwards) ellipsis (proposed by Browne 1972 and Giannakidou and Merchant 1998).30 On such an account, wh&wh questions differ from sluicing only with respect to the directionality of ellipsis. (64)
a. [CP What did John eat] and [CP why did John eat]? b. [CP What did John eat] and [CP why did John eat]?
However, the contrast in (65a–b) would be problematic for such an account, as it shows that wh&wh questions are more restricted than their sluicing counterparts. This is somewhat surprising if they involve the same basic structure and differ only with respect to whether ellipsis takes place in the first or the second conjunct, as shown in (66a–b).31 (65)
a. * Who and why did John meet? b. Who did John meet and why?
(66)
a. [CP Who did John meet why] and [CP why did John meet who]? b. [CP Who did John meet why] and [CP why did John meet who]?
Gračanin-Yüksek (2007) gives two further arguments against assimilating wh&wh questions to backwards sluicing. One involves scope and the other one, which is the one I summarize below, a process called swiping. Swiping inverts wh-phrases and prepositions in sluicing constructions, as shown in (67a). It is also possible in backwards sluicing, as shown in (67b) but not in non-sluicing contexts, as shown in (67c). This predicts that swiping should be possible in wh&wh questions if they are derived by sluicing. The fact that it is not, as shown in (67d), shows that wh&wh questions cannot involve backwards sluicing.
68 (67)
Symmetry in Merge a. b.
I know that Mary danced with someone, but I don’t know who with. Although we don’t know yet who from, we know she received a package last Monday with instructions on bomb assembly. c. * I don’t know who with Mary danced. d. * Mary doesn’t know who with and why Bill danced. (Gračanin-Yüksek 2007:157)
3.6.3 Right node raising The next construction I consider in this chapter is right node raising, exemplified in (68). (68)
John likes __ and Bill dislikes __ TV shows about vampires.
The italicized string is understood to be shared between two conjuncts, which, given the nature of sharing argued for in this chapter, implies a Parallel Merge structure of the kind given in (69). This is the analysis of RNR argued for (or simply assumed by) Bachrach and Katzir (2009), Citko (in press-c), Johnson (2007), Larson (2007), McCawley (1982), De Vries (2005a, 2008, 2009), De Vos and Vicente (2005) and Wilder (1999a, 2008), among many others. The symmetry involved in this structure is quite obvious; the pivot TV shows about vampires stands in a symmetric relationship with other elements inside the two conjuncts. (69)
&P &
TP John
T &
T
TP T
vP Bill v likes
VP
T
vP v
VP dislikes
DP TV shows about vampires
The core properties of right node raising discussed in the literature are listed in (70a–g). And the question I focus on is how these properties favor the symmetric structure in (69) over alternatives involving movement. (70)
a. RNR does not obey islands; b. RNR can target non-constituents; c. RNR allows P-stranding in languages that disallow P-stranding;
3.6 Empirical support for symmetric Merge d. e. f. g. h.
69
RNR element is treated in situ by binding principles; RNR element is treated in situ by NPI licensing principles; RNR element has to be right-peripheral; RNR can have scope over both conjuncts; RNR shows case proximity effects.
Many of the properties listed in (70) show that the pivot behaves as if it has not moved. This is problematic for movement approaches to right node raising, in which the pivot, instead of being shared between two conjuncts as a result of symmetric Parallel Merge, undergoes ATB movement to some position above the coordination level, as argued for by Postal (1998), Ross (1967) and Sabbagh (2007), among others. The result of such movement is given in (71).32 (71) &P DPi
&P &
TP John
T &
T
TV shows about vampires TP T
vP Bill v likes
VP
T ti
vP v
VP dislikes
ti
As noted by most proponents of non-movement approaches to RNR, the first five properties in (70) pose significant problems for movement approaches. Perhaps the strongest argument comes from the insensitivity of RNR to island constraints. The contrast between (72a) and (72b) shows that RNR does not obey the Wh Island Constraint, and the contrast between (73a) and (73b) shows that it does not obey the Complex NP Island Constraint either. This is surprising in a movement account; in (72a) the pivot his great song about the death of Emmett Till would have to move out of the embedded wh-island headed by when, and in (73a) the pivot pictures of Fred would have to move out of the complex DP island headed by a man.33,34
70
Symmetry in Merge
(72)
a.
John wonders when Bob Dylan wrote _ and Mary wants to know when he recorded _ his great song about the death of Emmett Till. b. * Whati does John wonder when Bob Dylan wrote ti? (Abels 2004:5)
(73)
a.
Mary knows a man who buys __, and Bill knows a man who sells __ pictures of Fred. (Wexler and Culicover 1980:299) b. * Whati does Mary know a man who buys ti?
One could hypothesize that the reason why RNR does not show any island effects is somehow linked to the fact that movement is to the right (not left). This cannot be the reason, as RNR is also insensitive to constraints on rightward movement.35 As originally noted by Ross (1967), other cases of rightward movement (i.e. extraposition and Heavy NP Shift) are subject to the so-called Right Roof Constraint. One formulation of this constraint is given in (74a). More recent formulations, such as the one due to Sabbagh (2007), given in (74b), make reference to cyclic nodes (which are vPs and PPs for Sabbagh) or phase nodes instead: (74)
Right Roof Constraint a. An element cannot move rightward out of the clause in which it originates. b. Rightward movement may move and right-adjoin an element X to the cyclic node in which X is merged, but no further. (Sabbagh 2007:351)
The Right Roof Constraint is what accounts for the contrast below. In (75a), the moved element moves out of only one cyclic node, vP, whereas in (75b) it moves out of two cyclic nodes, PP and vP. (75)
a. John [vP edited a review __ for Sue] of Jamie’s article. b. *Josh [vP edited a review [PP of an article __] for Jamie], about verb movement. (Sabbagh 2007:350)
Interestingly, RNR does not obey the Right Roof Constraint; in (76) the pivot the dean’s office would have to move out of two cyclic nodes. (76)
Joss [vP walked suddenly [PP into__ ] ] and Maria [vP stormed quickly [PP out of __ ] ] the dean’s office. (Sabbagh 2007:351)
The second argument against a movement account comes from the fact that RNR does not need to target constituents, as shown in (77). On the symmetric approach advocated here, nothing prevents two elements from being shared independently, or a non-maximal projection from being shared between two conjuncts.36,37
3.6 Empirical support for symmetric Merge (77)
71
I borrowed __ and my sisters stole __a total of $3000 from the bank. (Abbott 1976:642)
The third argument against a movement account of RNR comes from preposition stranding, in particular from the fact that there are languages that do not allow preposition stranding in questions (or other movement constructions) but allow it in RNR. Examples (78a–c) show this for Irish, (79a–c) for Indonesian and (80a–c) for Polish, all of which are languages that disallow P-stranding in questions but allow it in RNR. (78)
a. * Bhí mé ag éisteacht le inné clár mór Irish was I listen.prog with yesterday program great fada ar an ráidió faoin toghachán. long on the radio about-the election ‘I was listening yesterday to a great long program on the radio about the election.’ b. Brian Mag Uidhir… ag glacadh le __ agus ag cabhru Brian Maguire take.prog with and help.prog le __ plandail a dtailte fein with planting their lands refl ‘Brian Maguire … accepting and helping with the planting of their own lands.’ (McCloskey 1986:184–5)
(79)
a. * Mana Merrie sudah pergi ke? Indonesian where Mary past go to ‘Where did Mary go to?’ b. * Allena pulang dari __dan Merrie pergi ke__ sekolah. Allena go.back from and Mary go to school ‘Allena went home from and Mary went to school.’ (Muadz 1991:273)
(80)
a. * Którym stołemi książka leży pod ti? Polish which table book lies under ‘Which table is the book under?’ b. Książka leży pod __ a lampa wisi nad __ naszym nowym. book lies under and lamp hangs over our new stołem table ‘The book is under and the lamp hangs over our new table.’
Preposition stranding can provide an argument against a movement account of RNR even in a language like English. As pointed out by Ross (1967), extraposition disallows preposition stranding, as shown in (81). Thus, if RNR were rightward movement, we would expect it to pattern with extraposition and disallow preposition stranding as well.
72 (81)
Symmetry in Merge * John read about ti yesterday [a new cure for cancer]i.
The fourth argument comes from the behavior of RNR with respect to binding, discussed by Levine (1985). Example (82a) shows that the pivot behaves as if it has not moved with respect to Principle C effects, and (82b–c) make the same point with respect to Principles A and B effects. (82)
a. * Shei disliked __ and I hated __ that picture of Maryi. (Levine 1985:496) b. I liked __ and Maryi disliked __ that picture of herselfi. c. * I liked __ and Maryi disliked __ a picture of heri.
Variable binding points towards the same conclusion. Not only can the pronoun contained in the pivot be bound by the elements contained inside the two conjuncts (every man and no one in (83a–b)) but it can be bound by both of them simultaneously. This is unexpected if the pivot his mother has moved out of the two conjuncts, as it would no longer be c-commanded by its binders. (83)
a. Every mani loves __ but no onej wants to marry __ hisi/j mother. b. Every mani loves __ but no onej marries __ hisi/j mother. (Jacobson 1999:167)
And the fifth argument against a movement account comes from negative polarity item licensing. Again, we would expect examples of the kind given in (84) to be ungrammatical if the constituent containing anyone moved out of the c-command domain of the negation licensing it. (84)
She couldn’t talk to __ and he wouldn’t talk to __ anyone with green hair. (Johnson 2007:17)
All the properties discussed so far receive a natural account on the symmetric structure given in (69) above. Since the pivot does not move, it is c-commanded by the material inside both conjuncts. This explains the behavior of the pivot with respect to binding and negative polarity licensing. The lack of movement also explains why RNR does not pattern with movement constructions with respect to island effects and preposition stranding. Some of the properties of RNR listed in (70) do not, strictly speaking, favor a symmetric approach over a movement approach, and some seem to allow only a movement approach. In the remainder of this section I tackle these. The first one involves the requirement that the pivot be at the right edge of both conjuncts. One formulation of this requirement, due to Sabbagh (2007), is given in (85) below.
3.6 Empirical support for symmetric Merge (85)
73
Right Edge Restriction (Sabbagh 2007:356) In the configuration: [[A….X…] Conj [B…X…]] X must be rightmost within A and B before either (i) X can be deleted from A; (ii) X can be rightward ATB-moved; or (iii) X can be multiply dominated by A and B.
The contrast between (86a) and (86b) provides an illustration; in (86a) the pivot (or the gap corresponding to it) is rightmost in both conjuncts, whereas in (86b) it is not. (86)
a. I [VP invited into my house __ ] and [VP congratulated all the winners] b. * I [VP gave __ a present] and [VP congratulated all the winners] (Wilder 1999b:587)
At first, the Right Edge Restriction does not seem to bear on the issue of whether RNR involves a symmetric multidominant structure or movement. However, upon further consideration it does seem to favor a multidominant approach. There is no other movement construction that requires the trace of the moved element to be conjunct-peripheral. ATB movement is not subject to a restriction of this sort. But some of the linearization algorithms discussed in Section 3.5 above do derive the Right Edge Restriction. Wilder’s linearization system is a good example. (87) below gives a partial representation of the ungrammatical example in (86b). Within the first conjunct, the verb gave is ordered before DP1 all the winners and DP2 a present. Within the second conjunct, DP2 is ordered after the conjunction and and the verb congratulated. Since VP1 precedes VP2, DP2 is ordered both before and after DP1. This is a violation of antisymmetry, which can be remedied by removing DP2 from the structure. (87)
&P & VP1 V gave
&
VP2
ApplP and V2 congratulated DP 1 Appl
all the winners Appl
DP2 a present
The last two properties listed in (70) seem to provide an argument against the symmetric account of RNR. First, Sabbagh (2007) shows
74
Symmetry in Merge
that the pivot can take scope over both conjuncts. In (88a), for example, the universal quantifier every Germanic language can have wide scope with respect to the existential quantifier someone inside each conjunct. Such wide scope is impossible in (88b), which involves a parallel example without right node raising. (88)
a. John knows someone who speaks, and Bill knows someone who wants to learn every Germanic language. (∃>∀,∀>∃) b. John knows someone who speaks every Germanic language. (∃>∀,*∀>∃) (Sabbagh 2007:367)
For Sabbagh (2007), this is an argument for a movement approach to RNR. Bachrach and Katzir (2009), however, provide an alternative explanation for the facts in (88a–b), relying on delayed Spell-Out of multiply dominated elements. More specifically, they derive wide scope of the pivot from the idea that Spell-Out applies only to completely dominated constituents. What this means is that in (88a) every Germanic language is not spelled out (and consequently, not interpreted as well) till the entire conjunction phrase is built. Case considerations also appear to raise some issues for a symmetric account of RNR. In Section 3.6.1, we saw that the shared wh-pronoun in ATB wh-questions has to satisfy the case requirements imposed on it within both conjuncts. Since the examples of RNR under consideration here also involve a DP shared between two conjuncts, we would expect the same kind of matching. What we find instead is a case proximity effect; mismatches in case are grammatical as long as the case of the pivot is the one that is assigned to it within the second conjunct. The relevant examples from Polish are given in (89a–b).38 The verb kupić ‘buy’ assigns accusative case, whereas the verb szukać ‘look-for’ assigns genitive case, as indicated by the subscripts in the glosses. RNR is grammatical as long as the pivot bears the case required by the verb in the second conjunct, which clearly shows that proximity is the crucial factor here, not case markedness or case hierarchy. (89)
a.
Maria kupiła __ a Jan szuka __ nowego Maria boughtacc and Jan looks.forgen new.gen ‘Maria bought and Jan is looking for a new car.’ b. * Maria kupiła __ a Jan szuka __ nowy Maria boughtacc and Jan looks. forgen new.acc ‘Maria bought and Jan is looking for a new car.’
samochodu. car.gen samochód. car.acc
This contrast between ATB questions and RNR constructions does not constitute a fatal blow to a symmetric structure, as there is an independent explanation for why RNR should differ from ATB questions in this respect.
3.6 Empirical support for symmetric Merge
75
RNR can target elements smaller than DPs, as shown by the examples of the kind given in (90), which involve two determiners sharing a noun. (90)
Jan sprzedał ten __ a Maria kupiła tamten __ samochód. Jan sold this.acc and Maria bought that.acc car.acc ‘Jan sold this and Maria bought that car.’
If the two determiners have different cases (as in the examples below), the shared noun agrees in case with the second determiner. (91)
a.
Maria kupi ten __ a Jan szuka tamtego __ samochodu. Maria buy this.acc and Jan looks.for that.gen car.gen ‘Maria will buy this (one) and Jan is looking for that car.’ b. * Maria kupi ten __ a Jan szuka tamtego __ samochód. Maria buys this.acc and Jan looks.for that.gen car.acc ‘Maria will buy this (one) and Jan is looking for that car.’
Right node raising involving numerals also shows that the shared noun gets the case assigned to it in the second conjunct. In both of the examples given in (92), only the noun samochody ‘cars’ is shared, the numerals are not. What is interesting about these examples is that the two numerals have different case properties; the numeral trzy ‘three’ agrees in case with the noun it modifies whereas the numeral pięć ‘five’ assigns genitive case to it (at least in structural case environments).39 This is a common pattern across Slavic languages. In such cases the case that surfaces on the noun is the case assigned to it by the numeral in the second conjunct. (92)
a. Maria kupiła pięć __ a Jan sprzedał trzy __ samochody. Maria bought five.acc and Jan sold three.acc cars.acc ‘Maria bought five and Jan sold three cars.’ b. Maria kupiła trzy __ a Jan sprzedał __ pięć samochodów. Maria bought three.acc and Jan sold five.acc cars.gen ‘Maria bought three and Jan sold five cars.’
This suggests that even in examples like (89a–b) a smaller constituent could be shared (an NP and not a DP). The structure of (89a) (repeated below as (93a)) is given in (93b). On the assumption that case is a property of DPs not NPs, the grammaticality of such examples follows from the structure. In (93b) Agree takes place between v and D; and since there are two D heads, there are two instances of Agree, which opens up the possibility that the two Agree operations could yield two different case values. (93)
a. Maria kupiła __ a Jan szuka __ nowego samochodu. Maria boughtacc and Jan looks.forgen new.gen car.gen ‘Maria bought and Jan is looking for a new car.’
76
Symmetry in Merge
b.
&P &
TP Maria
T &
TP T
vP Jan
T v
VP bought DuC:Acc
T DP
vP v
VP
looks for DuC:Gen
DP NP new car
The reason why the shared noun phrase agrees in case with the second determiner follows from the properties of concord, which is generally responsible for agreement between determiners and nouns. On the assumption that concord, being part of the morphological component, happens at Spell-Out, the noun will agree with the determiner with which it is spelled out. The idea that case matching is a consequence of D sharing (and case mismatches are a consequence of having two distinct D heads in the structure) can also explain why ATB wh-questions differ from RNR in that they require strict case matching. In (94a), in which one verb assigns accusative case and the other one genitive case, neither the accusative nor the genitive wh-pronoun can satisfy the case requirements of both verbs, as shown in (94a). The result is ungrammatical. Since sharing in ATB whquestions involves wh-phrases, which are always DPs, the structure with two distinct Ds is not available. The only possible structure is the one in (94b). (94)
a. *Coi/*czego Maria zgubiła ti-acc a Jan szukał ti-gen ? what.acc/what.gen Maria lost and Jan looked.for ‘What did Maria lose and Jan look for?’ VP b. VP V
V
DP D co ‘what’
3.6 Empirical support for symmetric Merge
77
ATB wh-questions involving complex wh-phrases, such as the one in (95a) below, provide further evidence that the entire DP has to be shared. If only the NP is shared, the result is ungrammatical, as shown in (95b). (95)
a.
Jakie artykułyi Jan czyta ti a Maria recenzjonuje ti ? what articles Jan reads and Maria reviews ‘What articles does Jan read and Maria review?’ b. * Jakie artykułyi Jan czyta ti a Maria recenzjonuje które ti? what articles Jan reads and Maria reviews which ‘What articles does Jan read and Maria review?’
The reason why D heads have to be shared in ATB wh-questions (but not in RNR constructions) could be linked to the fact that D is the locus of the wh-feature forcing movement to [Spec, CP]. To recap, we saw in this section that many properties of right node raising follow nicely from a symmetric structure in which the right node raised element is shared between the two conjuncts. We also saw that the properties that appear to be problematic for such a symmetric structure turn out to be compatible with it upon closer examination. For example, case proximity (rather than strict case matching) is compatible with a symmetric approach once we allow sharing of constituents that are smaller than DPs, which is something we saw independent evidence for. 3.6.4 Gapping The next construction that can be (and has been) analyzed as involving a symmetric structure is gapping, illustrated in (96). The second conjunct in a gapping construction typically contains two elements in addition to the gapped string (indicated with a strikethrough in all the examples in this section).40 Following the general consensus in the field, I refer to these two elements as remnants, and to the elements in the first conjunct they correspond to as their correlates. (96)
John likes movies and Mary likes books.
In this section I focus on Polish gapping, which resembles English gapping in many relevant respects. However, the morphology of Polish makes some of the properties of gapping I am about to discuss more transparent. Illustrative examples of Polish gapping are given in (97a–c); (97a) involves gapping of a main verb, (97b) of a modal auxiliary, and (97c) of a larger chunk involving a verb and its object (to the exclusion of a PP adjunct). (97)
a. Jan je ryża a Maria je fasolę. Jan eats rice and Maria eats beans ‘Jan eats rice and Maria beans.’
78
Symmetry in Merge b. Jan może jeść cukier a Maria może pić alcohol. Jan can eat sugar and Maria can drink alcohol ‘Jan can eat sugar and Maria drink alcohol.’ c. Jan je ryż na śniadanie a Maria je ryż na obiad. Jan eats rice for breakfast and Maria eats rice for dinner ‘Jan eats rice for breakfast and Maria for dinner.’
The general structure for gapping I defend in this section is given in (98) below (see also Kasai 2007). This is the structure underlying examples of the kind given in (97a). The basic idea is that the gapped portion is shared between the two conjuncts. In this example there are two gapped elements, the main verb and v. The two are shared individually, and more complex cases of gapping will involve more instances of sharing. The intuition that the verb is shared is pretty clear; the missing verb in the second conjunct is understood to be identical to the overt verb in the first conjunct. However, I argue below, as well as in Citko (submitted), that v also has to be shared. (98)
TP T
AspP Asp
&P &
vP
& v
DP v
DP VP
V
vP v VP
DP
DP
The symmetry in this structure involves the relationship of the shared v and V heads to their complements, the two VPs and the two DP objects. Such a symmetric structure combines the insights of two types of existing accounts; it follows Goodall (1987), Moltmann (1992a) and Muadz (1991) in that it analyzes gapping in a multidominant way, and it follows Johnson (1996/2003, 2000, 2009) in that it treats gapping as involving coordination of small conjuncts (vPs not TPs). While a complete discussion of Johnson’s arguments in favor of vP (as opposed to TP) coordination in gapping would lead us too far off course, let me briefly mention one argument here, involving scope. Johnson (2000) shows that small conjunct coordination can explain why gapped sentences differ from their non-gapped counterparts, in that they allow negation to have wide scope over both conjuncts
3.6 Empirical support for symmetric Merge
79
(see also Lin 2000, 2001, 2002, Lopez and Winkler 2003, Oerhle 1987, Repp 2008 and Siegel 1984 for a discussion of negation in gapping). This is illustrated in (99) and (100) below. (99a) is most naturally paraphrased as (99c), not (99b). Similarly, (100a) is paraphrased as (100c), not (100b). In other words, (99a) is interpreted as a disjunction of two negative statements in spite of the fact that it contains the conjunction and, whereas (100a) is interpreted as a conjunction of two negative statements in spite of the fact that it contains the disjunction or. This is surprising on the ellipsis account, on which (99a) is derived from (99b) and (100a) from (100b). (99)
a. Mrs. J can’t live in Boston and Mr. J can’t live in L.A. b. Mrs. J can’t live in Boston and Mr J can’t live in LA. c. Mrs J can’t live in Boston or Mr J can’t live in LA.
(100)
a. Mrs. J can’t live in Boston or Mr. J can’t live in LA. b. Mrs. J can’t live in Boston or Mr J can’t live in LA. c. Mrs J can’t live in Boston and Mr J can’t live in LA. (Lopez and Winkler 2003:237, citing Oerhle 1987)
However, if negation has scope over both conjuncts, the interpretation of these examples is not surprising at all. This is due to the equivalences given in (101a–b) below, known as De Morgan’s Laws. This logical equivalence can only hold if negation has wide scope over both conjuncts in a gapped structure, and thus provides a nice support for the structure in (98), in which negation would be merged above the level of coordination and, consequently, would have scope over both conjuncts. (101)
a. NOT (P OR Q) = (NOT P) AND (NOT Q) b. NOT (P AND Q) = NOT P OR NOT Q
Gapping is subject to a number of interesting restrictions, discussed quite extensively in the relevant literature (see Citko submitted, Coppock 2001, Culicover and Jackendoff 2005, Hankamer and Sag 1976, Jackendoff 1971, Johnson 2000, 2009, Kuno 1976, Lobeck 1995, Maling 1972, Repp 2008, Ross 1970 and Sag 1976, among many others). The ones I focus on here are listed in (102a–h), and illustrated below with Polish data. Some of them provide a direct argument for a symmetric account, others do so more indirectly, by virtue of being problematic for alternative accounts. (102)
a. b. c. d. e.
Gaps must be surrounded by lexical material; Remnants must be contrastive; Gapping is only possible in coordinate clauses; Gaps require linguistic antecedents; Gapping can target non-constituents;
80
Symmetry in Merge f. Gapping is sensitive to head directionality; g. Gapping obeys islands; h. Gapping requires matching in tense, voice and aspect.
First, a gap needs to be surrounded by lexical material. Example (103) is ungrammatical because there is nothing following the gap in the second conjunct. (103)
*Jan lubi ryż a Maria lubi. Jan likes rice and Maria likes
Second, the remnants have to be distinct. (104a) is ungrammatical because the subjects are identical, and (104b) is ungrammatical because the objects are. (104)
a. *Jan je ryż a Jan je fasolę. Jan eats rice and Jan eats beans ‘Jan eats rice and Jan eats beans.’ b. *Jan je ryż a Maria je ryż. Jan eats rice and Maria je rice ‘Jan eats rice and Maria eats rice.’
This follows from the requirement, proposed by Kuno (1976), that the remnants (and their correlates) have to represent new information. (105) below provides an illustration; the two subjects are distinct, but since they are coindexed, the subject inside the second (gapped) conjunct does not represent new information. (105)
*Johni eats peas and hei eats rice.
(Kuno 1976:309)
Polish provides a nice illustration of this contrastiveness requirement, as the conjunction used in gapping constructions is different from the conjunction used in run-of-the-mill coordinate structures. The two conjunctions are referred to in the literature as contrastive and consecutive conjunctions, respectively. In a typological study of coordinating strategies, Malchukov (2004) describes consecutive conjunctions as those that indicate a temporal or a logical sequence, and contrastive conjunctions as those that indicate contrast between two conjuncts. In English, both are expressed by means of the same conjunction and, as shown in (106a–b). (106a) expresses a sequence of events, whereas (106b) expresses contrast between two events. (106)
a. John went to the store and bought some bread. b. John went to the store and Mary went to the movies.
3.6 Empirical support for symmetric Merge
81
In Polish, on the other hand, consecutive and contrastive coordination is expressed by distinct lexical items, i and a, respectively. I will henceforth gloss the Polish contrastive conjunction as andc.41 Only the consecutive conjunction i is possible in (107a), which is a counterpart of (106a), and only the contrastive conjunction a is possible in (107b), which is a counterpart of (106b). (107)
a. Jan poszedł do sklepu i/ *a kupił chleb. Jan went to store and/andC bought bread ‘Jan went to the store and bought bread.’ b. Jan poszedł do sklepu a/ *i Maria pojechała do kina. Jan went to store andc/and Maria went to cinema ‘Jan went to the store and Maria to the movies.’
As shown by Frajzyngier (1986), there are other contexts that require the use of the contrastive conjunction in Polish. For example, it is also used in situations in which the second conjunct is not a logical consequence of the first one or contains an element of surprise, which is what accounts for the contrast between (108a) and (108b); drowning while skiing is surprising whereas drowning at the beach is somewhat less so. (108)
a. Poszedł na narty a utopił się. went to ski andc drowned refl ‘He went skiing and drowned.’ b. Poszedł na plażę i/ *a utopił się. went to beach and/andc drowned refl ‘He went to the beach and drowned.’ (Frajzyngier 1986:113–14)
What is interesting and relevant for our purposes is the fact that in gapping constructions, only the contrastive conjunction is possible: (109)
Maria je ryż a/*i Jan je fasolę. Maria eats rice andc/ and Jan eats beans ‘Maria eats rice and Jan beans.’
Furthermore, gapping is not unique in this respect; other types of ellipsis also require contrastive conjunctions. This is illustrated below for pseudogapping (see (110a)), right node raising (see (110b)) and a variant of gapping called determiner sharing (see (110c)).42 (110)
a. Jan musi czytać Chomskiego a/*i Ania będzie czytać Jan must read Chomsky andC /and Ania will read Lakoffa. Lakoff ‘Jan must read Chomsky and Ania will Lakoff.’
82
Symmetry in Merge b. Jan przeczytał ten artykuł a/*i Maria zrecenzjonowała ten Jan read this article andc / and Maria reviewed this artykuł. article ‘Jan read and Maria reviewed this article.’ c. Mało psów je Whiskas a/*i mało kotów je Alpo. few dogs eat Whiskas andc/and few cats eat Alpo ‘Few dogs eat Whiskas and cats Alpo.’
The third property of gapping listed in (102) concerns the ungrammaticality of gapping in non-coordinate contexts. All the ungrammatical examples in (111) below involve gaps in subordinate clauses: a purpose clause in (111a), a conditional clause in (111b) and a temporal clause in (111c).43 (111)
a. *Jan je ryż, żeby Maria zjadła fasolę. Jan eats rice in.order.to Maria ate beans ‘Jan eats rice in order for Maria to eat beans.’ b. *Jan zje ryż, jeżeli Maria zje fasolę. Jan eats rice if Maria eats beans ‘Jan will eat rice if Maria eats beans.’ c. *Jan zje ryż po tym, jak Maria zje fasolę. Jan eats rice after this how Maria eats beans ‘Jan will eat rice after Maria eats beans.’
The fourth property requires gaps to have linguistic antecedents. If the antecedent is only contextually implied, as in (112) below, modeled after Hankamer and Sag’s (1976) English example, gapping becomes ungrammatical.44 (112)
[Jan produces an orange, proceeds to peel it, and Maria produces an apple] *A Maria obierze jabłko. and Maria peels apple ‘And Maria will peel an apple.’
The next property has to do with the nature of the elements that can be gapped. As shown in (113a–b), gaps can be fairly complex, and they do not have to form a constituent.45 (113)
a. Jan chce spróbować zacząć pisać powieść a Maria Jan wants try.inf begin.inf write.inf novel andC Maria chce spróbować zacząć pisać sztukę. wants try.inf begin.inf write.inf play. ‘Jan wants to try to begin to write a novel and Maria a play.’ b. Jan je ryż na sniadanie a Maria je fasolę na śniadanie. Jan eats rice for breakfast andC Maria eats beans for breakfast ‘Jan eats rice for breakfast and Maria beans.’
3.6 Empirical support for symmetric Merge
83
The next property involves directionality of gapping. As has been known since the early days of generative research on gapping (see Maling 1972, Ross 1970 and Sjoblom 1980, among others), the direction of gapping (backward or forward) reflects the setting of the head directionality parameter in the language. The attested patterns, listed in (114), show that head-initial languages gap forwards, whereas head-final languages gap backwards.46 (114)
a. SVO + SO b. SOV + SO c. VSO + SO d. SO + SOV e. * SO + SVO f. * SO + VSO
(Maling 1972:101)
The examples below provide an illustration. The English and Polish examples in (115a–b) contain gaps inside the second conjunct, whereas the Japanese and Korean examples in (116a–b) contain gaps inside the first conjunct. (115)
(116)
a. John eats rice and Bill ___ beans. b. Jan je ryż a Maria ___ fasolę. Jan.nom eats rice.acc andC Maria.nom beans.acc ‘Jan eats rice and Maria beans.’ a. John-ga
Polish
Bill ___ sosite Mary-ga
Susan-nituite hanasita. Japanese John-nom Bill and Mary-nom Susan-about talked ‘Mary talked about Susan and John about Bill.’ (Abe and Hoshi 1999:193)
b. John-i Mary-eykey Kkoch-ul ___ kuliko Bill-i Korean John-nom Mary-dat flower-acc and Bill-nom Sue-eykey Chayk-ul cwuessta. Sue-dat book-acc gave ‘John gave Mary a flower and Bill Sue a book.’ (Repp 2008:6, citing J.-S. Kim 1997:178)
The generalization about directionality of gapping refers to the surface position of the verb. For example, German, whose verbs are initial in root clauses (due to the effects of V2) and final in embedded clauses, allows forward gapping in root clauses and backward gapping in embedded clauses: (117)
a. John gab Mary eine Blume und Bill ___ Sue ein Buch. German John gave Mary a flower and Bill Sue a book ‘John gave Mary a flower and Bill gave Sue a book.’
84
Symmetry in Merge b. Max sagt dass die Jungs drinnen ___ und die Madchen Max says that the boys inside and the girls drauβen spielen. outside play ‘Max says that the boys are playing inside and the girls, outside.’ (Repp 2008:6)
Another illustration comes from Polish, which is typically characterized as a free word order language. This is accurate as a descriptive characterization (but not as a theoretical claim about Polish); even though the neutral order is SVO, any other order of a subject, object and verb is grammatical, as shown in (118a–e). Different word orders differ in information status in ways that do not need to concern us here. (118)
a. Jan je ryż. Jan.nom eats rice.acc ‘Jan eats rice.’ b. Jan ryż je. Jan.nom rice.acc eats c. Ryż Jan je. rice.acc Jan.nom eats d. Ryż je Jan. rice.acc eats Jan.nom e. Je Jan ryż. eats Jan.nom rice.acc f. Je ryż Jan. eats rice.acc Jan.nom
SVO
SOV OSV OVS VSO VOS
This freedom of word order, in particular the variable position of the verb, creates more gapping possibilities. Depending on which order serves as the input, the result is either forward or backward gapping: (119)
(120)
a.
Jan je ryż a Maria je fasolę. Jan eats rice andc Maria eats beans ‘Jan eats rice and Maria beans.’ b. *Jan je ryż a Maria je fasolę. Jan eats rice andc Maria eats beans
a. b.
(121)
a.
Jan Jan Jan Jan
ryż rice ryż rice
je eats je eats
a andc a andc
Maria fasolę je. Maria beans eats Maria fasolę je. Maria beans eats
Je Jan ryż a je Maria fasolę. eat Jan rice andc eat Maria beans b. *Je Jan ryż a je Maria fasolę. eats Jan rice andc eats Maria beans
SVO + SO
* SO + SVO SOV + SO SO + SOV VSO + SO * SO + VSO
3.6 Empirical support for symmetric Merge
85
The next relevant property of gapping involves islands. As noted by Neijt (1979), gapping obeys island constraints. Representative English examples are given in (122a–b), and Polish ones (also ungrammatical) in (123a–b). (122)
a. *John discussed my question of which flowers they saw and Bill discussed my question (of) which animals. (Neijt 1979:135) b. *John wondered what to cook today and Peter wondered what to cook tomorrow. (Neijt 1979:138)
(123)
a. *Piotr lubi ludzi, którzy jedzą ryż a Maria lubi ludzi, Piotr likes people who eat rice andC Maria likes people którzy jedzą fasolę. who eat beans ‘Piotr likes people who eat rice and Maria likes people who eat beans.’ b. *Maria zastanawia się, co Piotr ugotuje dzisiaj a Ewa Maria wonders refl what Piotr cooks today andC Ewa zastanawia się, co Piotr ugotuje jutro. wonders refl what Piotr cooks tomorrow ‘Maria wonders what Piotr is cooking today and Ewa tomorrow.’
That gapping should show island effects is not immediately obvious, as on many accounts it involves deletion rather than movement.47 The last relevant property of gapping involves the relationship between the features of the gapped string and those of its licensor. The ungrammaticality of the examples in (124a–b) shows that the gapped verb has to match its antecedent in tense. The adverbials identify the tense of the gapped verb; in (124a) the antecedent verb is in the present tense and the gapped verb in the past tense, and in (124b) the antecedent verb is in the past and the gapped verb in the present. Both are ungrammatical. (124)
a. *Ewa czyta Gazetę Wyborczą dzisiaj a Maria przeczytała Ewa read.pres Gazeta Wyborcza today and Maria read.PAST Gazeta Wyborcza wczoraj Gazeta Wyborcza yesterday ‘Ewa is reading Gazeta Wyborcza today and Maria read it yesterday.’ b. *Ewa przeczytała Gazetę Wyborczą wczoraj aC Maria Ewa read.PAST Gazeta Wyborcza yesterday and Maria czyta Gazetę Wyborczą teraz. read.pres Gazeta Wyborcza now ‘Ewa read Gazeta Wyborcza yesterday and Maria is reading it now.’
The two verbs also have to match in aspectual features. I also use adverbials to establish whether the gapped verb is perfective or imperfective.
86
Symmetry in Merge
Imperfective verbs in Slavic languages are only compatible with atelic adverbials like godzinami ‘for hours’, whereas perfective verbs are only compatible with telic adverbials like w dwie godziny ‘in two hours’. This is what accounts for the contrasts in grammaticality in (125a–b) and (126a– b). (125b) is ungrammatical due to the presence of a telic adverbial with an imperfective verb, and (125b) is ungrammatical due to the presence of an atelic adverbial with a perfective verb. (125)
a.
Ewa czytała ten artykuł godzinami. Ewa read.IMPERF this article hours.instr ‘Ewa was reading this article for hours.’ b. *Ewa czytała ten artykuł w dwie godziny. Ewa read.IMPERF this article in two hours ‘Ewa read this article in two hours.’
(126)
a.
Ewa przeczytała ten artykuł w dwie godziny. Ewa read.PERF this article in two hours ‘Ewa read this article in two hours.’ b. *Ewa przeczytała ten artykuł godzinami. Ewa read.PERF this article hours.instr ‘Ewa was reading this article for hours.’
With this as background, we can test aspect matching in gapping constructions. In (127a) the gapped verb is perfective and its antecedent imperfective. In (127b), the opposite is the case; the gapped verb is imperfective and its antecedent perfective. Both are ungrammatical, which shows that gapping also requires aspect matching. (127)
a. *Ewa czytała ten artykuł godzinami a Maria Ewa read.IMPERF this article hours.INSTR and Maria przeczytała ten artykuł w dwa dni. read.PERF this article in two days ‘Ewa was reading this article for hours and Maria read it in two days.’ b. *Ewa przeczytała ten artykuł w dwie godziny a Maria Ewa read.PERF this article in two hours and Maria czytała ten artykuł godzinami. read.IMPERF this article hours.instr ‘Ewa read this article in two hours and Maria was reading it for hours.’
The two conjuncts also have to match in voice, as shown in (128a–b). In (128a), the gapped verb is passive and its antecedent active, and in (128b) the gapped verb is active and its antecedent passive. Again, both are ungrammatical.
3.6 Empirical support for symmetric Merge (128)
87
a. *Niektórzy studenci przeczytali Aspects a Syntactic some students read Aspects andC Syntactic Structures zostały przeczytane przez innych (studentów). Structures became read by others students ‘Some students read Aspects and Syntactic Structures were read by other students.’ b. *Aspects zostały przeczytane przez niektórych studentów a Aspects became read by some students andC inni (studenci) przeczytali Syntactic Structures other (students) read Syntactic Structures ‘Aspects were read by some students and others read Syntactic Structures.’
On the assumption that matching with respect to some feature implies the presence of one head responsible for that feature, matching of tense means there is only one T head, matching of aspect means that there is only one Aspect head, and matching of voice means that there is only one head responsible for voice.48 I take this head to be the v head, which is sometimes called a voice head (see Kratzer 1993 and Pylkkänen 2002, 2008, among many others). As an astute Polish-speaking reader might notice, gapping does not require matching with respect to all features. In particular, phi-features do not have to match. All the examples given in (129a–c) below are grammatical in spite of a mismatch in number, person or gender features between the gapped verb and its antecedent; in (129a) gender features do not match, in (129b) number features do not match, and in (129c) person features do not match. (129)
a. Maria zjadła ryż a Jan zjadł Maria eat.past.3sg.FEM rice andC Jan eat.past.3sg.MASC ‘Maria ate rice and Jan beans.’ b. Kowalscy gotują ryż a Maria gotuje Kowalskis cook.pres.3PL rice andC Maria cook.pres.3SG ‘The Kowalskis are making rice and Maria beans.’ c. Ja lubię ryż a ona lubi fasolę. I like.pres.1sg rice andC she like.pres.3sg beans ‘I like rice and she beans.’
fasolę. beans fasolę. beans
The grammaticality of mismatches in person, number and gender would receive a fairly natural explanation if agreement and tense features were mediated by distinct functional heads, AgrS and T respectively. This would allow T to be shared between two conjuncts without AgrS being shared as well. Such an explanation, however, is not available in the current version of the minimalist theory, which dispenses with agreement projections
88
Symmetry in Merge
altogether, and in which both tense and person, number and gender agreement are mediated by T. What seems to be more promising is appealing to the idea that only interpretable features have to match, as suggested by Repp (2008). The question I turn to now is how all the properties of gapping discussed above follow from a symmetric structure. The relevant structure has (at least) four elements that are shared between the two conjuncts. In cases of simple gaps, such as the one in (130a) below, the shared elements are: the main verb, v, Asp, and T heads. However, they are shared in different ways; T and Asp heads are shared by virtue of being merged above the coordination level. This not only maintains Johnson’s insight that gapping involves small conjunct coordination but it also accounts for tense and aspect matching in a very straightforward manner. V and v heads, on the other hand, are shared by virtue of Parallel Merge, which is what accounts for voice matching.49 (130)
a. Jan zjadł ryż a Maria zjadła fasolę. Jan ate rice andC Maria ate beans ‘Jan ate rice and Maria beans.’ b. TP T
AspP Asp
&P &
vP
& v
Jan v
Maria VP
ate
vP v VP
rice
beans
This is not the final structure. I further assume, following Johnson (2000, 2009), that each conjunct contains a discourse-related projection (which I dub FP), and that both the remnants and their antecedents move to the specifier of this projection.50 Consequently, there are a number of movements that take place in (130b); the subject of the first conjunct moves to [Spec,TP], the shared verb moves in a successive cyclic fashion to Asp, and both objects (as well as the second conjunct’s subject) move to the specifier of FP positions. The end result is given in (131).
3.6 Empirical support for symmetric Merge (131)
89
TP T
Janj T
AspP Asp atek
&P &
FP
& F
ricei tk
FP F
Mariaj
beansi
vP
F’
v
tj tk
vP
tk
v
tj
VP ti
VP ti
The symmetric account extends to cases involving more complex gaps, such as the one in (132a). In such cases, the lower segment of the twosegmented VP is shared, as shown in (132b). (132)
a. Jan je ryż na śniadanie a Maria je ryż Jan eats rice for breakfast andC Maria eats rice na obiad. for dinner ‘Jan eats rice for breakfast and Maria for dinner.’ b.
TP T
AspP Asp
&P &
vP
& v
Jan v
vP Maria
VP VP
eats rice
v VP
PP for breakfast
PP for dinner
90
Symmetry in Merge
Again, the subject of the first conjunct moves to the specifier of TP, the object to the specifier of FP inside the first conjunct, and both PPs to the lower specifier of FP inside both conjuncts. The end result is given in (133).51 (133)
TP T
Janj T
AspP Asp eats
&P &
FP
& F
ricei PPk
FP F
Mariaj F
for breakfast F
PPk vP
for dinner v
tj v
vP tj
VP VP
V
F
v VP
tk
tk
ti
The properties of gapping that we started this section with are repeated in (134a–h). (134)
a. b. c. d. e. f. g. h.
Gaps must be surrounded by lexical material; Remnants must be contrastive; Gaps cannot occur in subordinate clauses; Gaps require linguistic antecedents; Gapping can target non-constituents; Gapping is sensitive to head directionality; Gapping obeys islands; Gapping requires matching in tense, voice and aspect.
The fact that the gaps are surrounded by lexical material is only an illusion, as the gap is a combination of sharing and movement of the remnants to [Spec, FP]. The fact that the remnants must be contrastive is also related to their movement to [Spec, FP]. The ungrammaticality of gapping in noncoordinate clauses, however, cannot be linked to sharing being limited to
3.6 Empirical support for symmetric Merge
91
coordinate structures, as there are constructions that involve sharing but no coordination whatsoever, which I discuss in the next two sections. It could, however, follow from some form of a parallelism constraint on gapping. The fact that gapping can target non-constituents is to be expected on the symmetric account, since nothing prevents more than one element from being shared. If the shared elements are discontinuous, we get the effects of non-constituent gapping. Whether the directionality of gapping follows from a symmetric structure depends on independent assumptions about movement. If in a head-initial language, movement is to the left (thus creating the illusion of a gap in the second conjunct), whereas in a head-final language movement is to the right (thus creating the illusion of a gap in the first conjunct), as shown schematically in (135a–b), the correlation between directionality and the position of the gap follows naturally. This, however, is incompatible with the Linear Correspondence Axiom of Kayne (1994), which bans rightward movements (see Kasai 2007 for a possible solution). (135)
a.
b.
TP
TP
T
DPi T
T
DPi AspP
Asp
AspP &P
T
&P
Asp
& vP
& v
ti v
vP DP
VP V
& vP v
DP
v
ti VP
&
vP
VP DP
DP
v
DP VP DP
v V
The fact that gapping obeys islands follows from the fact that the remnants undergo movement to the specifier of focus-related projections contained inside both conjuncts. And the last property, the obligatory matching of tense, aspect and voice features, follows from the simple fact that there is only one Tense, Aspect, and voice head in this structure. These are all the properties of gapping I set out to account for in this section. All the constructions discussed so far as evidence for the existence of symmetric structures in the grammar were coordinate, which raises the question of what the relationship is between coordination and symmetric
92
Symmetry in Merge
structure sharing. Since coordination involves an asymmetric structure, there cannot be any direct relationship between coordination and structure sharing. The discussion in the next two sections makes this point even more transparent, as it introduces two symmetric constructions which are not coordinate at all. 3.6.5 Serial verbs The arguments in this section summarize Hiraiwa and Bodomo’s (2008) analysis of serial verb constructions in Dàgááré (a Gur language belonging to the Niger-Congo family). Generally speaking, a serial verb construction is a construction consisting of a sequence of verbs with no overt coordination or subordination marker mediating the relationship between them. An example is given in (136a); it contains two verbs sharing a single object. Hiraiwa and Bodomo take this sharing quite literally, and propose that the object undergoes Parallel Merge with the two verbs, as shown in (136b). As a result, the object is in a symmetric c-command relationship with the two verbs.52 (136)
a. ò dà sε la nεnè ɔ`ɔ`. 3sg past roast fem meat eat ‘He roasted the meat and ate it.’ b.
Dàgááré (Hiraiwa and Bodomo 2008:243)
vP v
FocP
Foc
Asp1+2P
Asp1P Asp1 roast
Asp2P
V1P Asp2 meat
V2P ate
If nothing else happens, such a symmetric structure is not linearizable. However, Hiraiwa and Bodomo (2008) also provide independent evidence in favor of short object shift (movement of the object to the specifier of AspP) and successive cyclic movement of V1 to Asp to v, shown in (137).
3.6 Empirical support for symmetric Merge (137)
93
vP roastj
FocP
Foc
Asp1+2P
meati
Asp1+2P
Asp1P tj
Asp2P
V1P Asp2 tj
ti
V 2P ate
As a result, the object remains “sandwiched” between the two verbs – however, without being multiply dominated by the two VPs. There is independent evidence that the verb (even in simple clauses) moves to v across the Focus head, which is occupied by the focus particle lá. Furthermore, this movement is obligatory, as shown by the ungrammaticality of (138b). (138)
a.
n´ ngmε´ lá Dàkóráá. 1sg hit foc Dàkóráá. ‘I hit Dakoraa.’ b. * n´ lá ngmε´ Dàkóráá. 1sg foc hit Dàkóráá. ‘I hit Dakoraa.’
(Hiraiwa and Bodomo 2008:249)
Hiraiwa and Bodomo (2008) also show that alternative structures for serial verb constructions (which have been proposed for other languages) will not work for Dàgááré. These include a ternary branching structure in (139a) proposed by Baker (1989), a VP shell structure in (139b) proposed by Collins (1997b), and a VP adjunction structure in (139c) proposed by Hale (1991). Since serial verb constructions are quite varied crosslinguistically (as discussed most recently by Muysken and Veenstra 2006 and Aikhenvald and Dixon 2007), it is not surprising that they might involve different structures in different languages.
94 (139)
Symmetry in Merge a.
b.
VP1
SUBJ
V1
NP
V1 OBJ V2 V2
c.
VP1 V1 V1
VP1 VP1
VP2
VP2
V1OBJ V2 pro
OBJ V2 V2 pro
VP3 V3 V3
(Hiraiwa and Bodomo 2008:243)
Hiraiwa and Bodomo’s strongest argument for a symmetric structure comes from the interaction of serialization and predicate clefting. They show that in Dàgááré, the object can be clefted with either verb. In (140a), it is clefted with the first verb, whereas in (140b) it is clefted with the second verb.53 (140)
a. nε´nè séε´ó lá ká ó se ɔ´ɔ´. meat roast.nml fem C 3sg roast eat ‘It is roasting meat that he roasted and ate.’ (clefting of V1 + object) b. nε´nè ɔ´ɔó lá ká ó sε´ ɔ´ɔ´. meat eat.nml fem C 3sg roast eat ‘It is eating meat that he roasted and ate.’ (clefting of V2 + object) (Hiraiwa and Bodomo 2008:246)
This pattern is problematic for the structures given in (139a–c) for fundamental structural reasons. In (139a), V1 and the shared object do not form a constituent. Neither do V2 and the shared object. In (139b), V1 and the object do not form a constituent, and in (139c), V2 and the object do not.54 These predicate clefting patterns are not a problem for the symmetric structure proposed by Hiraiwa and Bodomo (2008); clefting of V1 and the object would involve movement of a constituent marked as Asp1P in (137), whereas clefting of V2 and the object involves movement of the constituent marked as Asp2P. Since the object is shared between them, it can be piedpiped in either case. 3.6.6 Free relatives The last construction that lends itself to a symmetric multidominant account is a free relative construction. The insight that free relatives involve a multidominant structure goes back to Haider (1988), Citko (1998, 2000) and Van Riemsdijk (1998, 2000, 2006a). In this section, I survey and evalu-
3.6 Empirical support for symmetric Merge
95
ate the evidence that has been given to support such an analysis, starting with a brief description of what free relatives are and how they have been analyzed in the past. The account I argue for in this section is a variant of Van Riemsdijk’s (2006a) account. The term headless or free relative clause reflects the intuition that free relatives differ from their headed counterparts in that they lack overt heads. This difference is illustrated in (141a–b); the relative clause in (141a) is headed by the DP the woman, whereas the free relative in (141b) appears to either lack the head entirely or to be “headed” by the wh-phrase itself. (141)
a. We hired the woman who(m) Mary recommended. b. We hired whom Mary recommended.
headed relative free relative
Free relatives exhibit a number of interesting syntactic, semantic and morphological properties, a full consideration of which goes beyond the scope of this chapter. It is thus not surprising that they have inspired a healthy debate among linguists, starting in the late 1970s and early 1980s with the work of Bresnan and Grimshaw (1978) and Groos and Van Riemsdijk (1981). Perhaps the biggest source of (ongoing) debate concerns the structure of free relatives, in particular the position of the wh-phrase. There are two major views on this matter, referred to as the Comp Account and the Head Account, respectively. The two are illustrated in (142a–b). On the Comp Account, the head position is empty and the wh-phrase is in [Spec,CP], as argued by Caponigro (2003), Gračanin-Yüksek (2008), Groos and Van Riemsdijk (1981), Grosu (1994, 1996), and Grosu and Landman (1998), among many others. On the Head Account, the [Spec,CP] position is empty and the wh-phrase occupies the head position, as shown in (142b), as argued by Bresnan and Grimshaw (1978), Bury (2003), Citko (2000, 2002, 2008b), Donati (2006) and Larson (1987, 1998), among others.55,56 (142)
a. Mary eats [DP Ø [CPwhat(ever)i [TP Bill cooks ti ] ] ] b. Mary eats [DPwhat(ever)i [CP [TP Bill cooks ti ] ] ]
Comp Account Head Account
The parallelism between the ungrammatical wh-questions in (143a–d) and the ungrammatical free relatives in (144a–d) suggests that wh-pronouns undergo the same kind of movement in wh-questions and free relatives; both are subject to locality. This parallelism receives a straightforward explanation on the Comp Account, in which the landing site for whpronouns in both questions and free relatives is [Spec, CP]. (143)
a. * Whati did Mary made the claim that Bill cooked ti? COMPLEX NP ISLAND b. * Whati did Mary wonder why Bill cooked ti? WH-ISLAND
96
Symmetry in Merge c. * Whati did Mary eat because Bill cooked ti? ADJUNCT CONDITION d. * Whosei did Mary eat ti dish? LEFT BRANCH CONDITION
(144)
a. * Mary ate whati she made the claim that Bill cooked ti. COMPLEX NP ISLAND b. * Mary ate whati she wondered why Bill cooked ti. WH-ISLAND c. * Mary ate whati she could because Bill cooked ti? ADJUNCT CONDITION d. * Mary ate what Bill cooked ti dish. LEFT BRANCH CONDITION
The Head Account, on the other hand, provides a straightforward account of the so-called matching effect, which is the requirement that the wh-pronoun satisfy the category and case requirements of both the embedded and the main verb. In (145a), the category matching is satisfied because both the matrix verb like and the embedded verb cook subcategorize for a DP complement (and the free relative is headed by a DP). In (145b), by contrast, it is not, since the free relative is headed by a PP, which cannot satisfy the subcategorization requirements of the matrix verb like. (145)
a. Mary liked [DP [DP whatever]i Bill cooked ti ] ]. b. * John liked [DP [PP with whatever]i Bill flavored the dish ti] ].
Case matching is illustrated by the contrast in (146a–b) from Polish. In the grammatical (146a), both the matrix and the embedded verb assign accusative case (and thus an accusative wh-pronoun can satisfy the case requirements of both). In the ungrammatical (146b), the case assignment properties of the two verbs differ; the embedded verb szukać ‘to look for’ assigns genitive case, and the matrix one zgubić ‘to lose’ assigns accusative case. As a result, neither the genitive nor the accusative wh-pronoun can satisfy the case requirements of both verbs. (146)
a.
Maria recenzjonuje [co(kolwiek) Jan pisze tACC ] ACC Maria reviews what(ever).acc Jan writes ‘Maria reviews whatever Jan writes.’ b. * Maria zgubiła [co/ czego(kolwiek) Jan szuka tGEN ] ACC Maria lost what.acc/gen Jan look-for ‘Maria lost what Jan is looking for.’
There are many ways to divide free relatives into distinct types. The distinction that is most relevant for our purposes is that between standard free relatives (henceforth SFRs) and transparent free relatives (henceforth TFRs).57 The two are exemplified in (147a) and (147b), respectively. (147)
a. John read what Bill gave him. b. John read what appeared to be Bill’s new novel.
SFR TFR
3.6 Empirical support for symmetric Merge
97
The term transparent reflects the intuition that with respect to many diagnostics (which I will discuss shortly), part of the TFR behaves as if it were not there. For example, what appeared to be can omitted in (147b) without affecting the grammaticality of the sentence, as shown in (148a).58 No such omission is possible in SFRs, as is shown in (147b). (148)
a. John read Bill’s new novel. b. * John read.
This difference is related to the fact that in TFRs, the DP Bill’s new novel, rather than the wh-pronoun what, functions as the semantic head (also referred to in the literature as the nucleus). By contrast, in SFRs the whelement itself functions as the head. TFRs differ from SFRs in a number of other respects, discussed by Grosu (2003), Schelfhout, Coppen and Oostdijk (2004), Van Riemsdijk (1998, 2000, 2006a) and Wilder (1999a), among others. The table in (149) provides a quick overview. The examples I discuss below to illustrate these differences are from Van Riemsdijk (2006a:363–5) unless otherwise noted. (149)
Differences between standard and transparent free relatives
Property
Standard free relatives
Transparent free relatives
Possible wh-pronouns all wh-pronouns
only what
Compatibility with ever
ever optional
ever impossible
Agreement
determined by the wh-pronoun
determined by the semantic nucleus
Idiom chunks
impossible
possible
Extraction
impossible
possible
Interpretation
interpreted as definites
interpreted as indefinites
Law of the obey LCL Coordination of Likes
disobey LCL
The first difference has to do with the range of wh-pronouns allowed in free relatives. The wh-pronoun what is the only one that is possible in TFRs and it is used even if the TFR refers to a human. (150)
She invited what I took to be a policeman to stay overnight.
TFR
98
Symmetry in Merge
No such restriction is present in SFRs; all wh-pronouns are possible and the wh-pronoun determines the reference of the free relative; what cannot be used to refer to a human: (151)
* She invited what was carrying a baton and a gun to stay overnight. SFR
Furthermore, what in TFRs cannot be affixed with -ever. In this respect, TFRs also differ from SFRs, in which -ever is optional: (152)
a. What(ever) just appeared on the freeway is not a jet. b. There was what(*ever) appeared to be a jet on the freeway.
SFR TFR
Agreement also works differently in the two types of free relatives. In SFRs, what triggers singular agreement, as shown by the contrast in (153a– b), whereas in TFRs it can trigger either singular or plural agreement, as shown in (154a–b). This shows that in TFRs agreement is determined by the semantic nucleus and not by the wh-pronoun. (153)
a. What pleases me most adorns the living room wall. b. * What please me most adorn the living room wall.
SFR SFR
(154)
a. b.
TFR TFR
What seem to be several jets were landing on the freeway. What seems to be a jet was landing on the freeway.
Another difference between SFRs and TFRs involves their compatibility with idiom chunks. TFRs, unlike SFRs, allow one part of the idiom to be inside the matrix clause and the other one inside the relative clause: (155)
a.
Nicki lost what according to the dictionary are called hisi marbles. TFR b. * Nicki lost what hisi marbles cost in the store. SFR
The next difference between SFRs and TFRs involves movement. As shown by the ungrammaticality of both (156a) and (156b), SFRs are islands for extraction, paralleling in this respect headed relatives. TFRs, by contrast, allow movement, as shown in (156c). (156)
a. * Whoi did you order what represented the sole possession of ti? SFR b. * Whoi did you order a picture that represented the sole possession of ti? HR c. Whoi did you order what was advertised as a first rate picture of ti? TFR
The next set of diagnostics involves the interpretation of both SFRs and TFRs. SFRs can be interpreted as either definites or universals (see a brief discussion of this difference in note 57). As such, they are predicted to be
3.6 Empirical support for symmetric Merge
99
subject to the definiteness effect. This prediction is confirmed; the ungrammaticality of the SFR in (157a) is due to the same factors that rule out (157b). (157)
a. * There was what had incurred extensive engine failure on the freeway. SFR b. * There was the jet/every jet on the freeway.
TFRs, on the other hand, do not induce the definiteness effect and pattern with indefinites in this respect: (158)
a. There was what appeared to be a jet on the freeway. b. There was a jet on the freeway.
TFR
This shows that the definiteness of the TFR is determined by the semantic nucleus (a jet in (158a–b)). Replacing it with a definite DP leads to ungrammaticality: (159)
* There was what appeared to be the jet on the freeway.
And the last difference between SFRs and TFRs listed in (149) concerns their behavior with respect to the Law of the Coordination of Likes. If the category of the TFR were determined by the wh-pronoun what, we would expect (160a) to be grammatical as a coordination of two DPs, which is not what we find. However, a corresponding SFR in (160b) is grammatical, which shows that the wh-pronoun in SFRs does determine the categorical status of the entire relative. Furthermore, the grammaticality of (160c) shows that coordination in TFRs is between the first conjunct and the semantic nucleus of the second conjunct. (160)
a. * He dislikes [DPtrains] and [DPwhat I consider inconvenient]. TFR b. He dislikes [DPtrains] and [DPwhat the government proposes to replace them with]. SFR c. He dislikes [DPtrains] and what I consider an equally [DPinconvenient alternative]. TFR
All the diagnostics thus point towards the conclusion that TFRs and SFRs are structurally different. The question now is what kinds of structures could capture these differences, and, more importantly, whether they are symmetric in the relevant sense.59 The idea I would like to pursue in this section is that both SFRs and TFRs involve a Parallel Merge structure. They differ, however, with respect to what exactly is being shared. In SFRs, the wh-pronoun is shared between the matrix and the relative clause; in TFRs the nucleus is. The shared elements are indicated in bold in (161a–b).
100 (161)
Symmetry in Merge a. John read what(ever) Bill wrote. b. John read what he took to be Bill’s new novel.
SFR TFR
For SFRs, I assume, following Van Riemsdijk (2006a), that the wh-pronoun is shared between its theta position in the matrix clause and the specifier position of the relative CP. For example, in an object free relative, such as the one given in (162a) below, the shared position is the specifier of the relative CP and the complement of the matrix verb.60 (162)
a. John read what(ever) Mary wrote. b.
TP T
John j
vP
T
v
tj v
VP V read
CP C
DP whateveri C
TP T
Maryj
vP
T
v
ti
v
tj v
VP V wrote
ti
In a subject free relative, such as the one in (163a) below, the wh-pronoun is shared between the specifier of the relative CP and the specifier of the matrix vP.
3.6 Empirical support for symmetric Merge (163)
101
a. Whoever arrives first will win the race. b. TP T CP DP whoeveri
T will
vP v
C
C
v
TP
V win
T
ti T
VP DP the race
vP ti
v v
VP arrives first
These structures for free relatives have two roots (TP and CP), which violates the Single Root Condition. There is nothing unique about free relatives in this respect; all Parallel Merge structures discussed in this chapter violate it at some point in the derivation. We could relax the Single Root Condition to allow such structures, but a more fundamental issue with (162b) and (163b) concerns linearization. If the two roots do not undergo any further Merge operations, the structure is not linearizable. The solution to this problem in other cases of Parallel Merge was for the two “roots” to merge again into a single-rooted structure, which allowed linearization to proceed on standard assumptions. For example, in ATB wh-questions, the two TP conjuncts (which constituted the two independent roots) merged with the conjunction head (one becoming its complement and the other one its specifier). What I would like to suggest here is that something similar happens in free relatives, only that there is no coordinate head involved. In this respect, I depart from Van Riemsdijk (2006a), for whom the structures akin to (162b) and (163b) are the final ones. Two things thus have to happen in (162b) and (163b); the wh-pronouns have to move, and the TP and CP have to merge into a single-rooted structure. In subject free relatives, the shared wh-pronoun moves to the specifier of TP first, as shown in (164a). Next, the relative CP adjoins to the moved wh-pronoun, as shown in (164b).
102 (164)
Symmetry in Merge a.
TP DPi CP
whoever
T
vP
T will
C
ti C
v v
TP ti
VP DP
win
T
vP
T
the race v
ti v
VP arrives first
b.
TP T’
DP i CP
DP i whoever
vP
will C
ti C
v v
TP ti
VP win
T’ T
vP ti
DP the race
v v
VP arrives first
3.6 Empirical support for symmetric Merge
103
What this implies is that in object free relatives the shared wh-pronoun will also have to move to some higher (non-shared) position.61 I take this position to be the external specifier of vP. This further implies that (even in a language like English) the verb has to move to some position above vP but below T. I take this position to be the head of PredP (as proposed for independent reasons by Johnson 2000, for example). (165)
a.
TP Johnj
T T
PredP read k
vP
DPi
vP v
whatever t j tk
VP tk
CP ti
C C
TP T
Mary j
vP
T
vP
ti tj
v v
VP wrote
ti
104
Symmetry in Merge b.
TP T
Johnj T
PredP readk
vP DPi
DPi
vP CP
C
whatever
v
tj v TP
C
VP tk
ti
T
Maryj
vP
T ti
vP v
tj
VP
v wrote
ti
The final structure is a variant of the Head Account, as the wh-pronoun occupies the CP external head position.62 A natural question that arises here is whether the same analysis can be extended to TFRs. The structure in which the wh-pronoun is shared between the matrix and the relative clause in TFRs, as shown in (166b), would totally assimilate TFRs to SFRs, and thus leave all the differences between them unaccounted for.
3.6 Empirical support for symmetric Merge (166)
105
a. John read what he took to be Bill’s new novel. b. TP T
John j T
vP v
tj
CP
VP
v
V read
DPi
C
what C
TP
he took to be ti Bill’s new novel
Instead, it seems more reasonable to assume that in TFRs the nucleus, which in (166a) is Bill’s new novel, is shared between the matrix and the relative clause, as shown in (167). This is the basic insight of Van Riemsdijk’s (2006a) account. (167)
a.
TP T’
Johnj
vP
T
v
tj v
VP
CP
read what he took to be DP Bill’s new novel
106
Symmetry in Merge
One benefit of such a structure is that it can capture the transparency of TFRs. Intuitively speaking, the only material from the relative CP that the matrix clause can “see” is the nucleus, which is shared between the two. Many of the differences between SFRs and TFRs discussed in this section follow from this fact. For example, the fact that the nucleus determines agreement or definiteness of the relative clause follows from the fact that the nucleus is the complement of the matrix verb. This structure can also explain why the nucleus can contain part of an idiom or determine the category of the free relative for the purposes of coordination. However, if nothing else happens, this structure faces some of the same issues we saw above with the symmetric structure for SFRs. First, it also involves two roots, and second, the shared element has not moved out of the shared structure. The solution to this problem with TFRs parallels the solution I offered above for SFRs. The shared element (Bill’s new novel in this case) moves out of the shared structure and the relative CP adjoins to it. The end result is given in (168). (168)
TP Johnj
T
T
PredP readk
vP
DPi CP
v DPi
tj
Bill’s new novel what he took to be ti
v VP
v tk
ti
3.6 Conclusion
107
To sum up this section, we saw how both standard free relatives and transparent free relatives can be analyzed in a symmetric fashion. We also saw that such an analysis can account for many well-documented differences between these two types, as it allows for different elements to be shared between the matrix and the relative clause: wh-pronouns in standard free relatives and the semantic nuclei in transparent free relatives. 3.7
Conclusion
In this chapter, I provided both empirical and theoretical arguments in favor of a particular kind of symmetric structure created by Merge: a multidominant structure in which a single node stands in a symmetric c-command relationship with respect to two other nodes. I further argued that the mechanism responsible for creating such structures, Parallel Merge, does not require any extra assumptions, as it combines the properties of the two fundamental syntactic mechanisms, External Merge and Internal Merge. I also discussed the constraints on Parallel Merge structures and various ways that have been proposed in the literature to linearize them. On a more empirical level, I argued that a symmetric Parallel Merge approach can explain a number of otherwise mysterious properties of the constructions listed in (169a–g): (169)
a. b. c. d. e. f. g.
ATB wh-questions wh-questions with conjoined wh-pronouns right node raising gapping serial verbs standard free relatives transparent free relatives
In all of them, there is one (or more) element that is multiply dominated, thus standing in a symmetric relationship with respect to the rest of the structure. I referred to such elements as pivots or shared elements. In ATB wh-questions, there are three pivots: a wh-pronoun, a T and a v head. In questions with conjoined wh-pronouns, there are multiple pivots, as everything except the wh-phrases themselves is shared between the two conjuncts. In right node raising, the right node raised portion is the pivot, and in gapping constructions the gapped string is. In serial verb constructions
108
Symmetry in Merge
of the type discussed in this chapter, the object is shared between two verb phrases. Standard and transparent free relatives differ with respect to what is shared between the matrix and the relative clause; in standard free relatives the wh-phrase is shared, whereas in transparent free relatives, the semantic nucleus is.
4
Symmetry in Move
4.1
Introduction
In this chapter I turn to symmetry in movement, focusing on the configuration schematized in (1), in which γ is a Probe (i.e. an element endowed with an EPP feature or some other feature triggering movement) and α and β are two potential Goals (i.e. two elements of the same type that can check this feature). Symmetric movement is a type of movement that treats α and β alike; it allows (or disallows) movement of either one. (1) γuF αF βF
There are two (related) questions that such a configuration raises. First, under what circumstances is movement of either α or β possible? And second, what allows movement of β over α? Given Relativized Minimality (or any of its more current formulations, such as Attract Closest or Shortest Move), we would expect only the higher of the two, α in (1), to be able to move. Passive movement in double object constructions provides a very straightforward example of the configuration in (1); it involves one subject position (one Probe) and two objects (two potential Goals). Furthermore, it is wellknown that languages vary with respect to which object they can passivize. In some languages passivization is an asymmetric operation in that it can target only one of the two objects, whereas in others it is a symmetric operation in that it can target either object. Accounting for this crosslinguistic variation is my main goal in this chapter. I proceed as follows. In Section 4.2, I review the familiar data to illustrate the distinction between symmetric and asymmetric passives. In Sections 4.3 and 4.4., I provide the data from Polish that will inform the discussion in the rest of the chapter. After 109
110
Symmetry in Move
a brief summary (Section 4.5), I turn to previous accounts in Sections 4.6 and 4.7, pointing out the advantages and disadvantages of each of them. As noted in Anagnostopoulou (2003) and McGinnis (1998, 2004), they can be divided into two kinds: the ones that link the availability of symmetric passives to various ways of circumventing a locality violation (such as the availability of the word order in which the direct object precedes the indirect one, either via base-generation or movement) and the ones that link the availability of symmetric passives to case properties of the two objects (and various patterns of case absorption in passives). The data from Polish I discuss in this chapter suggest that both are partially correct, and that there are two ways in which the lower object can passivize; one involves movement of the lower object over the higher one, and the other one involves a case-related movement of the higher object.1 4.2
Symmetric and asymmetric passives
On the assumption that the indirect object c-commands the direct one, which I justified in Chapter 2, we would expect only the IO to be able to passivize in (2). (2)
TP
T[EPP] IO DO
This is not what we always find. Across languages, we find four distinct passivization patterns in double object constructions. The first type, schematized in (3a–b), is the expected type; it involves languages in which only indirect objects can passivize. (3)
a.
b
TP
TP
T[EPP]
T[EPP] IO
IO DO
This is what we find in American English and Danish, for example.2
DO
4.2 Symmetric and asymmetric passives (4)
a. Johni was given ti a book. b. A booki was given John ti .
(5)
a.
Hani blev tilbudt ti en stilling. he was offered a job ‘He was offered a job.’ b. * En stillingi blev tilbudt han ti. a job was offered him ‘A job was offered to him.’
111
Danish
(McGinnis 1998:73)
The second type involves languages in which only direct objects can passivize, as shown in (6a–b). (6)
a.
b.
TP
TP
T[EPP]
T [EPP] IO
IO DO
DO
This is a common pattern in languages that mark indirect objects with inherent case, such as German, Spanish or Polish.3 (7)
a.
Das Mädchen schenkte dem Jungen ein Buch. German the girl.nom gave the boy.dat a book.acc ‘The girl gave the boy a book.’ b. Ein Buchi wurde dem Jungen von dem Mädchen ti geschenkt. a book.nom was the boy.dat by the girl given ‘A book was given to the boy by the girl.’ c. * Der Jungei wurde ti von dem Mädchen ein Buch geschenkt. the boy.nom was by the girl a book given ‘The boy was given a book by the girl.’ (Woolford 1993:688)
(8)
a.
El diaro le entregó le primer premio the newspaper cl.dat awarded the first prize.acc a Carolina. Spanish Carolina.dat ‘The newspaper awarded the first prize to Carolina.’ b. El premier premioi le fue entregado ti a Carolina. the first prize.nom.cl.dat was awarded Carolina.dat ‘The first prize was awarded to Carolina.’ c. * Carolinai le fue entregado /a (a) el primer premio ti. Carolina cl was awarded.masc/fem the first prize ‘Carolina was awarded the first prize.’ (Cuervo 2003a:97)
112 (9)
Symmetry in Move a.
Jan wysłał Ewie paczkę. Polish John.nom sent Eve.dat package.acc ‘John sent Eve a package.’ b. Paczkai została wysłana (Ewie) ti przez Janka. package.nom was.3fem sent Eve.dat by John ‘The package was sent (to Eve) by John.’ c. * Ewai była wysłana ti paczkę przez Janka. Eve.nom was.3fem sent package.acc by John ‘Eve was sent a package.’ (Dziwirek 1994:84)
The third pattern involves languages that allow passivization of either object, as shown in (10a–b). (10)
a.
b.
TP
TP
T[EPP]
T[EPP]
IO
IO DO
DO
This is the symmetric pattern most relevant to the general theme of this book and this is reported in languages like British English, Norwegian or Icelandic. (11)
a. Johni was given ti a book. b. A booki was given John ti .
(12)
a. Joni ble gitt ti boken. John was given book.def ‘John was given the book.’ b. Bokeni ble gitt Jon ti. book.def was given John ‘The book was given John’
(13)
Norwegian
(Holmberg and Platzack 1995:215)
a. Jónii var gefin ti bókin. Icelandic John.dat was given book.def ‘John was given the book’ b. Bókini var gefin Jóni ti. book.def was given John.dat ‘The book was given John.’ (Holmberg and Platzack 1995:189)
And the fourth pattern, schematized in (14a–b), involves languages that disallow passivization of either object.
4.2 Symmetric and asymmetric passives (14)
a.
b.
TP
113
TP
T[EPP]
T[EPP]
IO
IO DO
DO
This is what we see in the following examples from Greek: (15)
a. ?* To vivlio charistike tis Marias apo ton Petro. Greek the book.nom award.nact the Maria.gen from the Petros ‘The book was awarded to Mary by Peter.’ b. ?* To grama tachidhromithike tu Petru apo tin Ilektra. the report.nom mailed.nact.3sg the Petros.gen from the Ilektra ‘The letter was mailed to Peter by Ilektra.’ (Anagnostopoulou 2003:22)
The contrast between languages (or dialects) with symmetric passives and those with asymmetric passives is widely discussed in the literature on Bantu languages. In this group, quite often though, the contrast is not between one language being symmetric and another one being asymmetric, but between one type of applicative construction being symmetric and another one being asymmetric. For example, Swahili locative applicatives are symmetric; either the applied or the direct object can be passivized: (16)
a. Ofisi-ni pa-li-l-i-w-a ch-akula. Swahili office-loc 16sa-past-eat-appl-pass-fv 7-food ‘In the office was eaten food.’ b. Ch-akula ki-li-l-i-w-a ofisi-ni. 7-food 7sa-past-eat-pass-fv 9office-loc ‘The food was eaten in the office.’ (Ngonyani 1996:39–40)
Benefactive and goal applicatives, on the other hand, are asymmetric; only the applied object can be passivized, as shown in (17a–b) for benefactives and in (18a–b) for goals. (17)
a.
M-toto a-li-nunul-i-w-a ki-tabu. 1-child 1sa-past-1oa-bring-appl-pass-fv 7-book ‘The child had a book bought for him.’ b. * Ki-tabu ki-li-nunul-i-w-a m-toto. 7-book 7sa-past-buy-appl-pass-fv 2-boy ‘The book was bought for the child.’
(18)
a.
Wa-vulana wa-li-sukum-i-w-a j-ongoo. 2-boy 2sa-past-push- appl-pass-fv 5-milipede ‘The boys had millipede pushed towards them.’
Swahili
114
Symmetry in Move b. * Jongoo a-li-sukum-i-w-a wa-vulana. 1milipede 1sa-past-push-appl-pass-fv 2-boy ‘The millipede was pushed towards the boys.’ (Ngonyani 1996:38–9)
Kinyarwanda also has two types of applicatives, which behave differently with respect to passivization. Goal applicatives are symmetric, whereas benefactive ones are asymmetric, as shown by the contrast between (19a–b) and (20a–b). (19)
a. Igitabo cy-a-haa-w-e umugore (n’umugabo). Kinyarwanda book sp-past-give-pass-asp woman (by-man) ‘The book was given to the woman by the man.’ b. Umugore y-a-haa-w-e igitabo (n’umugabo). woman sp-past-give-pass-asp book (by-man) ‘The woman was given the book by the man.’ (Kimenyi 1980:127)
(20)
a.
Ishuûri ry-oohere-j-w-é-ho igitabo n’úúmwáalímu. school it-send-asp-pass-asp-loc book by teacher ‘The school was sent the book by the teacher.’ b. * Igitabo cy-oohere-j-w-é-ho ishuûri n’úúmwáalímu. book it-send-asp-pass-asp-loc school by teacher ‘The book was sent to school by the teacher.’ (McGinnis 2001:93, citing Kimenyi 1980)
The passivization patterns discussed above bear on the issue of symmetry in movement in two ways. First, the existence of languages that allow passivization of either object provides evidence that the grammar has to allow symmetric movement. Second, the existence of two types of asymmetric pattern points toward a more general (crosslinguistic) symmetry, as these two asymmetric patterns “add up” to a symmetric pattern, so to speak. Let me conclude this section by noting that the issue of symmetric movement (or the lack thereof) is not limited to passive movement. As pointed out by Di Sciullo (2005), for example, locative inversion and predicate inversion could be seen as other examples of symmetric A-movement. If locative inversion involves the structure in (21a), the symmetry comes from the fact that either the DP or the PP can move to the specifier of TP, as shown in (21b–c), respectively.4 (21)
a. [TP __ T[EPP] [vP [DP John] … [PP down the hill] ] ] b. [TPJohni [T’ ti rolled down the hill] ] c. [TPDown the hilli [T’ rolled John ti ] ]
(Collins 1997a:10)
4.3 Polish double object constructions
115
Inversion in copular clauses leads to the same conclusion. On the assumption that specificational and predicational copular clauses are derived from the same underlying structure, given in (22a) below (as argued by Den Dikken 2006, Mikkelsen 2005, Moro 2000, among others), the fact that either the subject or the predicate can move to [Spec,TP] could be seen as another instance of symmetric movement.5 (22)
a. [TP __ [T’ T[EPP] [‘SC’ [DP1 John] [DP2 the cause of the riot] ] ] ] b. [TPJohni [T’ T[EPP] [‘SC’ti [DP2 the cause of the riot] ] ] ] c. [TPThe cause of the rioti [T’ T[EPP] [‘SC’ [DP1 John] ti] ] ]
(Moro 2000:41)
The issue extends to A-bar movement, where violations of superiority could be seen as yet another example of symmetric movement. We saw in Chapter 2 that English, for example, allows violations of superiority in discourse-linked wh-questions, such as the ones in (23a–c). (23)
a. [CP __ C [TP which student read which article] ] b. [CPWhich studenti [C’ C [TP ti read which article? ] ] ] c. [CPWhich articlei [C’ did [TP which student read ti? ] ] ]
However, the question of whether the movements that we see in locative inversion, copular constructions or multiple wh-questions are truly symmetric (in the sense of involving two elements marked with the same feature moving to the same position for the same reason) is far from being settled (see the discussion in Chapter 2 for some alternative proposals about superiority-violating wh-movement, for example). This is why I limit my attention in this chapter to passive movement in double object constructions, as it is clear that the movement of a direct or an indirect object to the subject position is the same kind of movement driven by the same kind of feature to the same kind of position. 4.3
Polish double object constructions
We saw above that Polish is an asymmetric language in that it only allows its direct objects to passivize. Further examples are given below; (24a–c) are the base examples in the active voice, (25a–c) are the grammatical direct object passives, and (26a–c) are the ungrammatical indirect object passives. (24)
a. Jan wysłał Piotrkowi present. John.nom sent Peter.dat present.acc ‘John sent Peter a present.’
116
Symmetry in Move b. Uszyję córce nową sukienkę. sew daughter.dat new.acc dress.acc ‘I’ll sew my daughter a new dress.’ c. Zamawiam sobie nowy płaszcz. order myself.dat new.acc coat.acc ‘I’m ordering myself a new coat.’
(Swan 2002:345)
(25)
a. Prezenti został wysłany Piotrkowi ti. present.nom became sent Piotr.dat ‘A present was sent to Piotr.’ b. Nowa sukienkai została uszyta córce ti. new dress.nom became sewn daughter.dat ‘A new dress was made for the daughter.’ c. Nowy płaszczi został mi zamówiony ti. new coat.nom became me.dat ordered ‘A new coat was ordered for me.’
(26)
a. * Piotri został wysłany ti present. Piotr.nom became sent present.acc b. * Córkai została uszyta ti nową sukienkę. daughter.nom became sewn new.acc dress.acc c. * Jai zostałam zamówiona ti nowy płaszcz. I.nom became ordered new.acc coat.acc
While passivization of direct objects is what we see with the vast majority of verbs, namely those that mark their indirect objects with dative case and their direct objects with accusative case, it is not the only possibility. First, as discussed by Dziwirek (1994, 2002), there is a small class of verbs that allows only indirect objects to passivize. They differ from the verbs considered above in that the indirect object is accusative (rather than dative) and the direct object is genitive (rather than accusative). These are verbs like uczyć ‘teach’ or pozbawić ‘deprive’.6 (27)
a. Ewa uczyła Janka matematyki. Eve.nom taught John.acc math.gen ‘Eve taught John math.’ (Dziwirek 2002:335) b. Ostatnie wiadomości pozbawiły Ewę nadziei. last news.nom deprived Eve.acc hope.gen ‘Recent news deprived Eve of hope.’ (Dziwirek 2002:341, fn 1)
They allow indirect objects to passivize but disallow direct ones to do so, which is the opposite of what happens with verbs like dać ‘give’ or wysłać ‘send’. (28)
a. ? Janeki będzie uczony ti matematyki przez Ewę. John.nom will.be taught math.gen by Eve ‘John will be taught math by Eve.’ (Dziwirek 2002:336)
4.3 Polish double object constructions b.
(29)
117
Ewai została pozbawiona ti nadziei przez ostatnie Eve became deprived hope.gen by last wiadomości. news ‘Eve was deprived of hope by recent news.’
a. * Matematykai będzie uczona Janka ti przez Ewę. math.nom will.be taught John.acc by Eve ‘Math will be taught to John by Eve.’ (Dziwirek 2002:336) b. * Nadziejai została pozbawiona Ewę ti przez-ostatnie hope.nom became deprived Eve.acc by last wiadomosci. news ‘Eve was deprived of hope by recent news.’
Furthermore, with uczyć ‘teach’, passivization of the direct object becomes possible if the indirect object object is absent: (30)
Matematykai była uczona ti przez filozofów. math.nom was taught by philosophers ‘Math was taught by philosophers.’
This is not surprising and is related to the fact that the verb uczyć ‘to teach’ is not an obligatorily ditransitive verb. It can also be transitive, with its sole internal argument being either a theme or a benefactive, as shown in (31a–b). The grammatical passive in (30) above is thus the passive of (31a). (31)
a. Filozofowie uczyli matematyki. philosophers.nom taught math.gen ‘Philosophers taught math.’ b. Filozofowie uczyli studentów. philosophers.nom taught students.acc ‘Philosophers taught students.’
Polish also has double object verbs that take dative indirect objects and genitive direct objects. Examples of such verbs include zabronić ‘forbid’, odmówić ‘refuse’ and udzielić ‘grant’. (32)
a. Strażnik zabronił wszystkim wejścia. guard.nom forbade all.dat entry.gen ‘The guard forbade entry to all.’ b. Rząd udzielił więźniom politycznym amnestii. government.nom granted prisoners.dat political.dat amnesty.gen ‘Political prisoners were granted amnesty by the government.’ c. Ambasada odmówiła nam wszystkim wiz. embassy.nom denied nam.dat all.dat visas.gen ‘The embassy denied us all visas.’
118
Symmetry in Move
With respect to passivization, these verbs behave the same way dative accusative verbs do; they allow direct objects to passivize and disallow indirect ones to do so. (33)
a. Wejściei zostało wszystkim ti zabronione. entry.nom became everyone.dat forbidden ‘Entry was forbidden to everyone.’ b. Amnestiai została udzielona więźniom politycznym ti. amnesty.nom became granted prisoners.dat political.dat ‘Amnesty was granted to political prisoners.’ c. Wizyi zostały nam wszystkim odmówione ti przez visas.nom became us.dat all.dat denied by ambasadę. embassy.acc ‘Visas were denied to us all by the embassy.’
(34)
a. ?* Wszyscyi zostali zabronieni ti wejścia. all became denied entry ‘All were denied entry.’ b. ?* Więźniowie politycznii zostali udzieleni ti amnestii prisoners.nom political.nom became granted amnesty.gen przez rząd. by government.acc ‘Political prisoners were granted amnesty by the government.’ c. ?* My wszyscyi zostaliśmy odmówieni ti wiz we.nom all.nom became denied visas.gen przez ambasadę. by embassy.acc ‘We all were denied visas by the embassy.’
The table in (35) summarizes the behavior of Polish double object verbs with respect to case marking and passivization. (35) Verb type
Polish double object verbs: summary Passive of DO
Verb IODAT DOACC ✓ Verb IOACC DOGEN * Verb IODAT DOGEN ✓
Passive of IO Examples * ✓ *
dać ‘give’ wysłać ‘send’ uczyć ‘teach’ pozbawić ‘deprive’ udzielić ‘grant’, odmówić ‘refuse’
The issue of whether the direct or the indirect object can passivize is in principle different from the issue of whether the higher or the lower object can passivize. How they are related depends on the underlying ordering of the two objects. In Chapter 2 we saw evidence that in English the indirect object c-commands the direct one. This is a somewhat more contentious
4.3 Polish double object constructions
119
issue in a language like Polish, which allows both orders of the two objects, as shown in (36a–b). (36)
a. Jan dał Marii książkę. Jan.nom gave Maria.dat book.acc ‘Jan gave Maria a book.’ b. Jan dał książkę Marii. Jan.nom gave book.acc Maria.dat ‘Jan gave Maria a book.’
IODAT > DOACC
DOACC > IODAT
The question is which of the two orders is basic and which one derived. Or even more fundamentally, are the two orders transformationally related at all? Both of the questions are far from being settled. Dornisch (1998), for example, takes the basic order in Polish to be DO IO (see also Bailyn 1995, in press, for similar claims about Russian). I argue below that the basic order is IO DO instead. Dyakonova (2007, 2009) makes a similar case for Russian, and Dvořák (in press) and Kučerová (2007) for Czech.7 This issue is not unique to Slavic languages. (37a–b) show that both orders are also possible in Japanese, a typologically very different language. (37)
a. Taroo-ga Hanako-ni nimotu-o okutta. Taro-nom Hanako-dat package-acc sent ‘Taro sent Hanako a package.’ b. Taoo-ga nimotu-o Hanako-ni okutta. Taro-nom package-acc Hanako-dat sent ‘Taro sent a package to Hanako.’ (Miyagawa and Tsujioka 2004:5)
Hoji (1985) takes the IO DO order in (37a) to be basic, and the DO IO order in (37b) to be derived via scrambling. Miyagawa (1995, 1997) and Miyagawa and Tsujioka (2004), on the other hand, argue that the two orders are not transformationally derived, but are a reflex of two different argument structures, corresponding to the English double object structure and to-dative structure, respectively. What allows for such an analysis is the fact that ni is ambiguous between a dative marker and a preposition in Japanese. We could extend this analysis to Polish and assume that the IO DO word order is a true double object construction and that the DO IO order corresponds to a PP ditransitive construction. However, if this were the case, we would expect the DO IO order in Polish to be subject to the same restrictions as English double object constructions. This is not what we find. One restriction, discussed by Green (1974) and Oehrle (1976), among others, concerns the infelicitous status of locations as indirect objects in double object constructions, illustrated in (38a–b):8
120 (38)
Symmetry in Move a. John sent a book to Paris. b. # John sent Paris a book.
In Polish, both orders are equally infelicitous with locative goals, and contrast with sentences containing locative PPs: (39)
a. # Jan wysłał książkę Paryżowi. Jan.nom sent book.acc Paris.dat b. # Jan wysłał Paryżowi książkę. Jan.nom sent Paris.dat book.acc c. Jan wysłał książkę do Paryża. Jan.nom sent book.acc to Paris ‘Jan sent a book to Paris.’
This suggests that we are dealing with a true double object construction in both cases, irrespective of the word order involved. Discourse considerations also suggest that for double object verbs like wysłać ‘send’ or dać ‘give’ the IO DO order is basic, and the DO IO order is derived. The IO DO word order is used as a response to a discourse-neutral question like What happened?, as shown in (40a–b).9 (40)
a. Co się stało? what refl happened ‘What happened?’ b. Jan wysłał Ewie paczkę. Jan.nom sent Ewe.dat package.acc ‘John sent Eve a package.’
The DO IO is also possible, but it would be an appropriate answer to a more specific question about the identity of the indirect object, as shown in (41a–b). (41)
a. Komu Jan wysłał paczkę? who.dat Jan.nom sent package.acc ‘Who did Jan send a package to?’ b. Jan wysłał paczkę Ewie. Jan.nom sent package.acc Ewe.dat ‘Jan sent Ewe a package.’
The relative height of the two objects in Polish can also be determined by applying the familiar Barss and Lasnik’s (1986) diagnostics. They confirm that the indirect object c-commands the direct one. The contrast in (42a–b) shows that the IO is higher than the DO with respect to pronominal binding; (43a–b) show that the IO is higher than the DO with respect to variable binding.10
4.3 Polish double object constructions (42)
Jan pokazał Mariii jeji szefową. Jan.nom showed Maria.dat her.acc boss.acc ‘Jan showed Maria her boss.’ b. * Jan pokazał jeji szefowej Marięi. Jan.nom showed her.dat boss.dat Maria.acc ‘Jan showed Maria to her boss.’
(43)
a.
121
a.
Jan dał każdemui pracownikowi jegoi czek. Jan.nom gave every.dat owner.dat his.acc check.acc ‘Jan gave every owner his check.’ b. * Jan dał jegoi właścicielowi każdy czek. Jan.nom gave his.dat owner.dat every.acc check.acc ‘Jan gave its owner every check.’
Not all of the Barss and Lasnik (1986) diagnostics are applicable to Polish. We saw above that passivization does not (necessarily) determine the relative height of the two objects, since the two objects are marked with different cases. The contrast in (44a–b) shows that only structural case is affected by passive morphology, not that direct objects are higher than indirect objects. (44)
a. * Jani został dany ti książkę. Jan.nom became given book.acc ‘Jan was given a book.’ b. Książkai została dana Janowi ti. book.nom became given Jan.dat ‘A book was given to Jan.’
Likewise, superiority is not a good diagnostic, since Polish in general does not show superiority effects. The grammaticality of both (45a) and (45b) thus does not tell us anything useful about the underlying order of the two objects either. (45)
a. Co komu Jan wysłał? what.acc who.dat Jan.nom sent ‘What did Jan send to whom?’ b. Komu co Jan wysłał? who.dat what.acc Jan.nom sent ‘To whom did Jan send what?’
Weak crossover effects, however, do point toward the IO DO being the base order, as this is the only order that yields the right crossover configuration in (46a–b). (46a) is correctly predicted to be grammatical, as the whmoved indirect object does not cross the coindexed direct object. (46b), on the other hand, in which the moved wh-phrase crosses a coindexed
122
Symmetry in Move
pronoun, is correctly predicted to be ungrammatical. If the base order were DO IO instead, we would expect the opposite results: (46a) to be ungrammatical as a violation of weak crossover and (46b) to be grammatical.11 (46)
a.
Któremu pracownikowii wysłałaś ti jegoi czek? which.dat employee.dat sent his.acc check.acc ‘Which employee did you send his check?’ b. * Czyji czek wysłałaś jegoi właścicielowi ti? whose check.acc sent his. dat owner.dat ‘Whose check did you send to its owner?’
4.4
Polish dative subjects
The generalization established in the previous section that Polish dative indirect objects do not passivize might seem falsified in view of examples such as the ones given in (47a–c), in which indirect objects undergo movement to what look like subject positions. (47)
a. Ewiei zostały dane ti kwiaty. Ewa.dat became given flowers .nom ‘Ewa was given flowers.’ b. Więźniom politycznymi została udzielona ti amnestia. prisoners.dat political.dat became granted amnesty.nom ‘The government granted political prisoners amnesty.’ c. Nami została odmówiona ti wiza. we.dat became denied visa.nom ‘We were denied a visa.’
Since Polish allows dative subjects, examples of which we will see shortly, it is not unreasonable to wonder whether the fronted datives in (47a–c) are subjects or not. To answer this question, I rely on standard subjecthood diagnostics established in the literature: subject–verb agreement, reflexive binding, subject-to-subject raising and adjunct control.12 These diagnostics were applied to Polish subjects (nominative and dative alike) by Dziwirek (1994), whose findings I rely on in this section. First, let us establish that Polish nominative subjects behave as expected with respect to all these diagnostics. The data in (48a–d) show that they do; they determine verb agreement (as shown in (48a)), they can bind subjectoriented reflexives (as shown in (48b)), they can undergo subject-to-subject raising (as shown in (48c)), and they can control adjuncts (as shown in (48d)).13
4.4 Polish dative subjects (48)
123
a. Jan napisał książkę. agreement Jan.nom write.past.3sg.masc book.acc ‘Jan wrote a book.’ b. Jani lubi swojei książki. reflexive binding Jan.nom likes refl books.acc ‘Jan likes his books.’ c. Jani wydaje się ti być dobrym kolegą. raising Jan.nom seems refl be good.instr friend.instr ‘Jan seems to be a good friend.’ d. Jani zobaczył Piotraj po pijanemui/*j. adjunct control Jan.nom saw Piotr.acc while drunk.instr ‘Jan saw Piotr (while Jan was drunk).’
With this as background, we can compare the behavior of fronted datives in double object constructions with the behavior of canonical dative subjects in the language. The ones I use for comparison here are experiencer subjects of the kind illustrated in (49a–b).14 (49)
a. Ewie było żal Janka. Ewa.dat be.past.3sg.neut sorry Jan.gen ‘Ewa was sorry for Jan.’ b. Ewie podoba się ta książka. Ewa.dat please.pres.3sg.neut refl this.nom book.nom ‘Ewa likes this book.’
Such dative subjects behave like nominative subjects with respect to all but one diagnostic. They do not trigger verb agreement; the verb forms used in (49a–b) above are the default third person singular neuter forms. Using agreeing forms instead results in ungrammaticality, as shown in (50a). However, they can bind reflexives (as shown in (50b)), they can undergo raising (as shown in (50c)), and they can control into adjunct clauses (as shown in (50d)). 15 (50)
a. * Ewie była żal Janka. Ewa.dat be.past.3sg.fem sorry Jan.gen ‘Ewa was sorry for Jan.’ b. Ewiei było żal siebiei. Ewa.dat was sorry refl ‘Ewa felt sorry for herself.’ c. Ewiei wydaje się ti być żal studentów. Ewa.dat seems refl be sorry students.gen ‘Ewa seems to feel sorry for the students.’ d. Jankowii żal jest Piotraj po pijanemui/*j. Jan.dat sorry is Piotr.gen while drunk ‘Jan feels sorry for Piotr when he (Jan) is drunk.’
124
Symmetry in Move
The issue of why dative subjects fail the agreement test, while interesting in itself, is not directly relevant for our purposes here. What is relevant is the fact that fronted datives in double object constructions do not behave like subjects with respect to any of the subjecthood diagnostics. First, they do not trigger verb agreement (as shown in (51a)); second, they cannot bind subject-oriented reflexives (as shown in (51b)); third, they do not allow subject-to-subject raising (as shown in (51c)); and fourth, they cannot control the adjunct po pijanemu ‘while drunk’ (as shown in (51d)). (51)
a. * Nami zostaliśmy wysłani ti paczka. we.dat become.past.1pl send.past.part.1pl package.nom ‘We got sent a package.’ b. * Ewiei została wysłana ti swojai/j paczka. Ewa.dat became sent refl package.nom przez Jankaj. by Jan ‘Ewa was sent her package by Jan.’ c. * Jankowii wydaje się ti zostać wysłana ti paczka. Jan.dat seems refl become.inf sent package.nom ‘Jan seems to have been sent a package.’ d. * Jankowii została wysłana ti paczka przez Piotra Jan.dat became sent package.nom by Piotr po pijanemui. while drunk ‘Jan was sent a package by Piotr while he [Jan] was drunk.’
To rule out the possibility that the subjecthood diagnostics used here work only for nonderived subjects, let us look at how passivized direct objects behave with respect to the same diagnostics. The examples in (52a–d) show that with respect to all the diagnostics, they behave like true subjects. Thus, they agree with both the auxiliary and the participle (as shown in (52a)); they can antecede subject-oriented reflexives (as shown in (52b)); they can undergo subject-to-subject raising (as shown in (52c)); and they can control the adjunct po pijanemu ‘while drunk’ (as shown in (52d)). (52)
a. Jani został przedstawiony ti Jan.NOM.3SG become.past.3SG.MASC introduce.part.3SG.MASC Marii. Maria.dat ‘Jan was introduced to Maria.’ b. Jani został przedstawiony ti swojemui szefowi. Jan.nom became introduced refl.dat boss.dat ‘Jan got introduced to his (own) boss.’
4.5 Interim summary and preview
125
c. Jani wydaje się ti być już przedstawiony ti wszystkim Jan.nom seems refl be already introduced all.dat pracownikom. employees.dat ‘Jan seems to have already been introduced to all his employees.’ d. Jani został przedstawiony ti wszystkim po pijanemui. Jan.nom became introduced all.dat while drunk ‘Jan was introduced to everyone while he (Jan) was drunk.’
The table in (53) summarizes the findings of this section. It shows that fronted datives in double object constructions are not subjects, as they differ from dative subjects (as well as passivized direct objects) with respect to standard subjecthood diagnostics. Thus their existence does not falsify the observation that Polish is an asymmetric passivization language, in which only one object can passivize (53)
Behavior of Polish subjects: summary Double object constructions
Verb Agreement Reflexive Binding Raising to Subject Adjunct Control
4.5
Nominative subjects
Dative subjects
Passivized direct objects
Fronted dative indirect objects
✓ ✓ ✓ ✓
* ✓ ✓ ✓
✓ ✓ ✓ ✓
* * * *
Interim summary and preview
To sum up the discussion so far, we saw four distinct passivization patterns in double object constructions. They distinguish not only one language (or dialect) from another, but also one type of a double object construction from another type. In asymmetric passives, only one of the two objects can undergo passive movement. In American English, Danish, Swahili benefactives and goals, and Kinyarwanda benefactives (and in the small set of Polish double object constructions), it is the indirect object (as shown schematically in (54a)), whereas in languages like German or Spanish (and in the vast majority of Polish double object constructions), it is the direct object (as shown in (54b)). In symmetric passives, either the direct or the indirect object can passivize, as shown in (54c). This is what we see in
126
Symmetry in Move
British English, Norwegian, Icelandic, Swahili locatives and Kinyarwarda goals, for example. (54)
a.
TP IOi T[EPP] ti DO
b.
TP DOi T[EPP] IO ti
c.
TP
TP DOi
IOi T[EPP]
T[EPP] ti
IO DO
ti
The main question I address in the rest of this chapter is what allows the lower object to move to the subject position, as movement of the higher object follows from standard locality considerations. Movement of the lower object is possible in symmetric passives as well as in one type of an asymmetric passive (i.e. the type illustrated in (54b)). Thus, the answer to this question will lead to a deeper understanding of the mechanism behind both symmetric and asymmetric passivization, and more generally, to a deeper understanding of the nature of symmetric movement in the grammar. Another question we might ask about asymmetric passives of type (54b) is what freezes the higher object so that it cannot passivize instead. I argue that the answers to these two questions are related. More specifically, I show that there are two ways for the lower object to move to the subject position.16 It can move over the higher object first, as shown in (55a), which makes it closer to T. If this movement is optional, the indirect object can passivize as well, as shown in (55b), resulting in a symmetric passive movement. This is what happens in Bantu languages like Kinyarwarda, as argued by McGinnis (2001), for example. I refer to this type of strategy as a locality-based strategy.17 Alternatively, the
4.6 Locality and symmetric passives
127
higher object can move to some position in which it is frozen for further movements of the same type (and, crucially, does not block movement of the lower object), as shown in (56a). This is what happens, I argue, in Polish double object constructions. Anagnostopoulou (2003) makes a similar claim about clitic doubling in Greek. Again, if the indirect object does not move first, it can move directly to the subject position, as shown in (56b). I refer to this type of strategy as a case-based strategy.18 (55) a.
b.
TP
TP IOi
DOi
T[EPP]
T ti IO
ti DO
ti
(56) a.
b.
TP DOi
TP IOi
T[EPP]
T[EPP] IOj tj
ti ti
DO
In this chapter, I motivate the need for these two strategies (following Doggett 2004, Anagnostopoulou 2003, and McGinnis 1998, among others). There are many approaches to the distinction between symmetric and asymmetric passives that rely purely on locality, and I start off by examining them. The first one capitalizes on the correlation between the availability of symmetric passives and the availability of the word order in which the direct object precedes the indirect one. And the second one links the availability of symmetric passives to the availability of high applicatives in the language, coupled with the assumption that high applicative heads are phase heads. Next, I turn to case-based approaches, which derive different passivization patterns from the differences in case-absorption mechanisms. 4.6 Locality and symmetric passives 4.6.1 Word order variation Many accounts link the availability of direct object passives to the availability of the word order in which the direct object precedes the indirect one. This could either be the result of movement (of the direct object over the
128
Symmetry in Move
indirect object) or base-generation of the DO IO order (in addition to the IO DO order), as shown in (57a–b). In both cases, the direct object ends up being closer to T and locality is respected when it passivizes. (57)
a.
b. T
T DOi DO
IO ti
IO
This is the approach favored by Anagnostopoulou (2003), Doggett (2004) and Ura (1996), among others.19 We saw above that there are dialects of British English that allow either object to passivize, as shown in (58a–b). (58)
a. Johni was given ti a book. b. A booki was given John ti.
There are also dialects of British English that allow “an alternative double object construction,” a double object construction in which the direct object precedes the indirect one. Representative examples are given in (59a–c). While it is more common with pronominal objects, (59b) shows that it is also possible with full DPs.20,21 (59)
a. I gave it him. (Gast 2007: 31) b. She gave a book the man. (Hughes and Trudgill 1979:21) c. I got the map from his secretary, and when I gave it him he spread it out on his desk. (Gast 2005:1)
If the alternative double object construction is the source of direct object passives, the fact that either object can passivize follows from the availability of the DO IO word order. However, the examples below show that the correlation between the DO IO word order and the ability to passivize DO breaks down across languages (see also Citko 2009, Doggett 2004, Yeong 2007, among others, for a discussion of this correlation). For example, Marantz (1993) notes that in Chaga benefactives only the IO DO order is possible, but either object can passivize:22 (60)
a.
N-a-i-lyi-i-a m-ka k-elya. foc-sp-pres-eat-appl-fv wife food ‘He is eating food for his wife.’ b. * N-a-i-lyi-i-a k-elya m-ka. foc-sp-pres-eat-appl-fv food wife ‘He is eating food for his wife.’
Chaga
(Marantz 1993:121)
4.6 Locality and symmetric passives (61)
129
a. M-ka n-a-i-lyi-i-o k-elya. wife foc-sp-prs-eat-appl-pass food ‘The wife is being affected by someone eating the food.’ b. K-elya k-i-lyi-o m-ka. food sp-pres-eat-appl-pass wife ‘The food is being eaten for the wife.’ (Marantz 1993:126)
In Swahili benefactives, on the other hand, both orders are possible but only a benefactive can passivize. (62)
(63)
a.
Halima alimpa Fatuma zawadi. Swahili Halima she-past-her-give Fatima gift ‘Halima gave Fatuma a gift.’ b. Halima alimpa zawadi Fatuma. Halima she-past-her-give gift Fatuma ‘Halima gave Fatuma a gift.’ a. Fatuma alipewa zawadi na Halima. Fatuma she-past-give-pass gift by Halima ‘Fatuma was given a gift by Halima.’ b. * Zawadi ilipewa Fatuma na Halima. gift it-past-give-pass Fatuma by Halima ‘A gift was given Fatuma by Halima.’ (Woolford 1993:686, citing Vitale 1981:130–1)
Languages like Polish also pose problems for the correlation between word order and passivizability of direct objects. We saw in Section 4.3 that they allow either order of the two objects but can only passivize one of them, the direct one (in most cases). Note that on the locality-based account under consideration here, passivization of direct objects has nothing to do with case; all that matters is the direct object’s proximity to T. In principle, it is also independent of the distinction between high and low applicatives and is thus consistent with accounts that assume only one structure for applicatives. Anagnostopoulou (2003) is an example of such an account; she assumes a single structure, given in (64) below, which corresponds to Pylkkänen’s high applicative structure in that a (verbal) applicative head takes a VP as its complement and an IO as its specifier. (64)
(Anagnostopoulou 2003:157)
vP v
IO
VP
vAPPL V
DO
Furthermore, she proposes the following parameter to distinguish languages with symmetric passives from languages with asymmetric pas-
130
Symmetry in Move
sives. The basic idea is the same; the direct object’s ability to move over the indirect object is what makes it accessible to passive movement.23 (65)
Specifier of vAPPL Parameter (Anagnostopoulou 2003:157) Symmetric movement languages license movement of DO to a specifier of vAPPL. In languages with asymmetric movement, movement of
DO may not proceed via vAPPL. To sum up, we saw in this section that the correlation between the availability of both IO DO and DO IO word orders and symmetric passivization works only for some languages. On the one hand, there are languages that allow both orders but lack symmetric passives, and on the other hand, there are languages that allow only one order but nevertheless allow symmetric passives. In the next section, I turn to a different type of a locality-based account, which correlates the availability of symmetric passives with the availability of a high applicative structure. It also relies on the movement of the lower object over the higher one, but allows it only in high applicatives. 4.6.2 Movement through applicative phase edge McGinnis in a series of papers attributes crosslinguistic variation in the availability of symmetric passives to independently motivated variation in the availability of high applicatives (McGinnis 2000, 2001, 2002, Gerdts and McGinnis 2003). In particular, she proposes that a high applicative head is a phase head, which means it has an EPP feature allowing the direct object to move above the indirect object (to its outer specifier).24 This is the position from which the direct object is closer to T for the purposes of passive movement, as shown in (66). (66)
TP vP
T SUBJ
vP ApplHP
v
ApplHP
DOi IO
ApplH ApplH[EPP] VP V
ti
4.6 Locality and symmetric passives
131
By contrast, a low applicative head is not a phase head. This means that the movement of the direct object to the specifier of ApplL, illustrated in (67a), is impossible. There are two reasons why. First, it lacks motivation (as a low applicative head does not have an EPP feature).25 Second, it involves movement from a complement position to a specifier position, which violates anti-locality by virtue of being too short, as argued by Doggett (2004), Jeong (2007), Lee (2004), among others. The direct object, being a complement of the applicative head, is already in its domain, thus no new checking relationship is established by moving it to the specifier of the same applicative head.26 Moving the direct object directly to [Spec,TP], schematized in (67b), would violate any version of locality, due to the presence of the intervening indirect object. (67)
a.
b.
TP
TP T
DOi T[EPP]
DOi vP
vP
T[EPP] VP
v
T
V
v ApplLP
ti
ApplLP
V
ApplLP IO
VP
ApplL
ApplL
ApplL
IO ApplL
ti
ti
Thus in a low applicative structure there is no way for the indirect object to move over the direct object. While intuitively very appealing, this analysis makes some interesting predictions, an examination of which is my goal in the remainder of this section (as well as in the two sections that follow). In this section, I focus on the crosslinguistic validity of the correlation between symmetric passives and high applicatives (see also the discussion in Jeong 2007), and in Sections 4.6.3 and 4.6.4 on the predictions it makes for other types of movement, such as wh-movement or quantifier raising. We saw above that a single language can have both high and low applicatives. In such a language, we would expect only high applicatives to be able to passivize. Swahili is a good example of a language in which
132
Symmetry in Move
different types of applicatives exbibit different passivization behavior, as shown in the table given in (68). Its benefactive, goal and malefactive applicatives are asymmetric, whereas its instrumental, motive and locative applicatives are symmetric. A natural question to ask in this context is whether there is any independent evidence for a split between these two groups of applicatives along the high versus low dimension. For this correlation to hold, benefactives, goals and malefactives would have to be shown to be low applicatives, since they only allow passivization of the higher (i.e. the applied) argument. While this might seem reasonable for goal applicatives (which, given their interpretation, arguably belong to the class of low applicatives), benefactive or malefactive applicatives generally behave like high applicatives (see Marantz 1993 for arguments that benefactive arguments are generally merged higher than locative or instrumental ones).27 (68)
Passivization patterns in Swahili (cf. Ngonyani 1996:47) Benefactive Goal Malefactive Instrumental Motive Locative
Passive of IO ✓ Passive of DO *
✓ *
✓ *
? ? ✓
? ✓
✓ ✓
Furthermore, there is more concrete evidence against the correlation between (the availability of) high applicatives and symmetric passivization. There exist languages that allow high applicatives (according to Pylkkänen’s original diagnostics) but lack symmetric passives. And there are also languages that allow only low applicatives but have symmetric passives. McGinnis (2001) discusses Chichewa, which has high applicatives but allows only indirect objects to passivize. Chichewa benefactives are high applicatives; the evidence comes from the fact that they do not necessarily involve transfer of possession (see (69a) below) and that they are compatible with static verbs (see (69b)). (69)
a.
b.
A-na-dy-er-a mkazi wake chakudya. Chichewa sp-past-eat-appl-fv woman his food. ‘He ate food for his wife.’ A-na-gwir-ir-a mkazi wake thumba. sp-past-hold-appl-fv woman his bag ‘He held the bag for his wife.’ (McGinnis 2001:134)
However, only applied arguments can passivize, which is what we would expect from a low applicative (not a high one).28
4.6 Locality and symmetric passives (70)
133
a.
Atsíkānai a-na-gúl-ír-idw-á ti mphâtso (ndí chítsîru). 2.girls 2s-pst-buy-appl-pass-fv 9.gift by 7.fool ‘The girls were bought a gift (by the fool).’ b. * Mphâtsoi i-na-gúl-ír-idw-á átsíkāna ti (ndí chítsîru). 9.gift 9s-pst-buy-appl-pass-fv 2.girls by 7.fool ‘A gift was bought for the girls (by the fool).’ (McGinnis 2001:134)
Slavic languages also allow high applicatives, as shown by the following examples from Russian (Polish behaves analogously in this respect). (71)
a. Ja budu vam pet’ i tantsevat’. Russian I.nom fut you.dat sing.inf and dance.inf ‘I will sing and dance for you.’ b. Ty sidiš mne na platje. you.nom sit.pres me.dat on dress.loc ‘You are sitting on my dress.’ c. Ona izlagala nam svoj plan uže izrjadno she.nom state.past us.dat self plan.acc already quite vypivšim tipsy.dat ‘She was telling us her plan while we were already quite tipsy.’ (Dyakonova 2007:19)
Yet, both Russian and Polish are asymmetric passivization languages, as shown in Section 4.3 for Polish and in (72a–b) below for Russian. (72)
a. * Petri byl dan/vyšit ti rubašku. Petr.nom was given/embroidered shirt.acc ‘Peter was given/ embroidered a shirt.’ b. Rubaškai byla vyšita Petru ti. shirt.nom was embroidered Petr.dat ‘Peter was embroidered a shirt.’
Russian (Soschen 2005:6)
(Soschen 2005:4)
Another potential issue concerns the dialects of (British) English that allow symmetric passives. There is no evidence that speakers of these dialects allow high applicatives. If they did, we would expect (73a–c) to be possible for them, contrary to fact (73)
a. * John ran Mary. (with the interpretation ‘John ran for Mary.’) b. * John held Mary the door. (with the interpretation ‘John held the door for Mary.’) c. *John gave Maryi a cookie hungryi. (with ‘hungry’ modifying ‘Mary’)
A somewhat more tangential issue for any account based purely on locality concerns applicatives that disallow passivization of either object. Postal (2004), following Fillmore (1965), points out two types of double object constructions in English: the ones that alternate with prepositional
134
Symmetry in Move
phrases headed by to, which he dubs Type A, and the ones that alternate with prepositional phrases headed by for, which he dubs Type B. In Type A direct object constructions, the indirect object is a goal (or a recipient), whereas in Type B constructions, it is a benefactive. (74)
a. Armand sold her the six Uzis. b. Amand sold the six Uzis to her.
(Type A)
(75)
a. Armand bought her the six Uzis. b. Armand bought the six Uzis for her.
(Type B)
Fillmore further points out that English benefactive double object constructions differ from recipient or goal ones in that they disallow passivization of either object: (76)
a. * Shei was bought ti the six Uzis by Armand. b. * The six Uzisi were bought her ti by Armand.
(77)
a. Shei was sold ti the six Uzis by Armand. b. The six Uzisi were sold her ti by Armand.
(Postal 2004:240)
More examples to illustrate ungrammatical passivization of either object in English benefactives are given below.29 (78)
a. * The visitorsi must have been found ti some food. b. * Some foodi must have been found the visitors ti.
(79)
a. * His parentsi were carved ti a statue. b. * A statuei was carved his parents ti.
(80)
a. * Maryi is being built ti a table by John. b. * A tablei is being built Mary ti by John.
(81)
a. * The guestsi have just been roasted ti a duck. b. * A ducki has just been roasted the guests ti.
According to Pylkkänen’s diagnostics, since English lacks high applicatives, such ungrammatical passives have to be low applicatives. This, however, seems at odds with Marantz’s (1993) observation that benefactive applied objects (even in languages like English) are higher than other types of applied objects. In fact, the structure he assigns to English benefactive applicatives (as opposed to locative or instrumental ones) corresponds to Pylkkänen’s high applicative structure. He motivates it by pointing out a number of differences between benefactive applicatives on the one hand and instrumental and locative ones on the other hand. One of these differences is the ability to form denominal verbs from instrumental and locative
4.6 Locality and symmetric passives
135
applicatives but not from benefactive ones, as shown by the contrast between the grammatical (82a–b) and the ungrammatical (82c):30 (82)
a. pound the nail with a hammer b. put the motion on the table c. do some favors for a friend
→ → →
hammer the nail table the motion * friend some favors (Marantz 1993:146)
This suggests that English benefactives are higher than locatives and instrumentals. However, they cannot be as high as, say, Kinyarwanda benefactives, because if they were, we would expect them to be compatible with unergative and stative predicates and allow symmetric passivization, contrary to fact. 4.6.3 Wh-movement from applicatives If the presence of the EPP feature on an applicative head is what allows movement of a direct object over an indirect one, we would expect all movements from applicatives to be subject to the same restrictions. In this section, I examine the behavior of applicatives with respect to wh-movement, focusing on the predictions the two applicative structures make with respect to superiority effects. In multiple questions involving high applicatives, a direct object whphrase moves over an indirect object wh-phrase to the outer specifier of the applicative head, as shown in (83a). From this position, it is closer to C, which predicts that we should find no superiority effects in high applicatives. Low applicatives, on the other hand, are predicted to show superiority effects and allow only indirect object wh-phrases to move. Wh-movement of the direct object, schematized in (83b), would violate locality. (83)
a. high applicative
b. low applicative
CP
CP C
WHi
C
WHi
C
C
ApplHP ti
VP
ApplHP
ApplLP
V WH
ApplH
WH
ApplL
VP
ApplH [EPP]
ApplL V
ti
ti
136
Symmetry in Move
Even though Polish has high applicatives, it is not a good language to test this prediction. Since Polish never shows superiority effects, the lack of superiority effects in high applicative constructions, illustrated in (84a–b), has nothing to do with the type (and properties) of its applicative heads. (84)
a. Coj komui Jan trzymał ti tj? what whom Jan held ‘Whad did Jan hold for whom?’ c. Komui coj Jan trzymał ti tj? whom what Jan held ‘For whom did Jan hold what?
Testing superiority effects in English double object constructions is also tricky, since for many speakers wh-movement of indirect objects is impossible even if the indirect object is the only wh-element. However, for speakers that do allow wh-extraction of indirect objects, the predicted contrast does appear to hold, as shown in (85a–b). (85)
a. ?? Whomi did John give ti what? b. * Whati did John give whom ti?
Abstracting away from independent factors (such as the lack of superiority in Polish and the ungrammaticality of indirect object extraction in English), the predictions the two types of applicatives make regarding multiple wh-movement are confirmed. In Citko (2009), I examined the predictions the two types of applicative structures make regarding wh-movement in questions containing a single wh-pronoun, which could be either a direct or an indirect object. To review briefly: we do not expect to find any restrictions on wh-movement in such cases; either the indirect or the direct object wh-pronoun should be able to move out of either a high or a low applicative structure.31 (86a–b) illustrate wh-movement of a direct object and (87a–b) of an indirect one. All four are predicted to be possible; since there is only one wh-phrase in each structure, no intervention effects arise.32
4.6 Locality and symmetric passives (86)
a. high applicative
b.
CP
low applicative CP
C
WHi
C
WHi
CQ
CQ
ApplHP
VP
ti
ApplHP
V
ApplLP
ApplH
IO
IO ApplH V
ApplL
ti
a. high applicative
b.
CP
ti
low applicative CP
C
WHi
ApplL
VP
[EPP]
(87)
137
C
WHi CQ
CQ ApplHP ti
VP ApplH
V VP
ApplH
ti
[EPP]
V
ApplLP
DO
ApplL ApplL
DO
However, it is well-documented in the literature on double object constructions that the movement schematized in (87b) is not possible for all, though it is for some, speakers of English, as shown by the following contrast.33,34 (88)
a. Whati shall we give John ti? b. * Whoi shall we give ti a present?
This immobility of indirect objects is not limited to wh-questions; we see the same pattern in other types of A-bar dependencies; Heavy NP Shift in (89), tough movement in (90), relativization in (91), topicalization in (92), and clefting in (93). (89)
Heavy NP Shift a. * John sent ti a letter every musician in the orchestrai of all. b. * Max gave ti a book about roses the tall man in the gardeni. c. * Mary promised ti to win some spectator in the grandstandsi. (Larson 1988:354)
138
Symmetry in Move d. * Anne gave ti a present the person she liked most of alli. (Hudson 1992:256) e. * Charlie baked ti that cake the girl who lives next doori. (Emonds and Whitney 2006:94)
(90)
Tough movement a. *? John is tough OPi to give ti a present. (Larson 1988:355) b. * Kids are always easy OPi to tell ti a story/buy presents. (Emonds and Whitney 2006:94)
(91)
Relativization a. * These are the girls OPi that the fool bought ti the gift. (Marantz 1993:133) b. % The girl whoi I gave ti flowers is Mary. c. % Nobody whoi I send ti an email message ever replies. (Hudson 1992:258)
(92) (93)
Topicalization * ? Johni, Mary said that she gave ti a present.
(Larson 1988:355)
Clefting * It is Bill OPi (that) John gave ti the book.
The data thus show that there is no correlation between the ability to passivize a given object and the ability to wh-move it. The table given in (94) shows that it is not a quirk of English; Swahili also shows no correlation between the ability to wh-move a given object and the ability to passivize it. (94)
Object properties in Swahili applicative constructions (adapted from Ngonyani 1996:47) Benef. Goal Malef. Instr. Motive Locative
Passivization
applied object direct object Wh-extraction applied object direct object
✓ * * ✓
✓ * * ✓
✓ * * ✓
? ?✓ ✓ ✓
? ✓ ✓ ✓
✓ ✓ ✓ ✓
Instrumental, motive and locative applicatives behave as expected; both direct and applied objects can undergo both passivization and whmovement. The examples below provide an illustration involving instrumental applicatives. Since Swahili is a wh-in-situ language, the examples in (95a–b) involve relativization (not wh-movement); in (95a) the applied
4.6 Locality and symmetric passives
139
object kisu ‘knife’ is relativized, and in (95b) the direct object muwa ‘sugar cane’ is. Either of them can be passivized, as shown in (96a–b). (95)
a. Ki-su amba-cho wa-li-kat-i-a mu-wa. 7-knife rel-7 2sa-past-cut-appl-fv 3-sugar cane ‘The knife with which they cut sugar cane.’ b. Mu-wa amba-o wa-li-kat-i-a ki-su. 3-sugar cane rel-3 2sa-past-cut-appl-fv 7-knife ‘The sugar cane which they cut with the knife.’ (Ngonyani 1996:44–5)
(96)
a. ?ma-we ya-li-vunj-i-w-a vy-ungu. 6-rock 6sa-past-break-appl-pass-fv 8-pot ‘The rocks were used to break pots with.’ b. ?ch-ungu ki-li-vunj-i-w-a ma-we. 7-pot 7-sa-past-break-appl-pass-fv 6-rock ‘The pot was broken with rocks.’ (Ngonyani 1996:38–9)
Benefactives, goals, and malefactives behave differently; only direct objects can be relativized but only applied objects can be passivized:35 (97)
a.
M-toto a-li-nunul-i-w-a ki-tabu. 1-child 1sa-past-1oa-bring-appl-pass-fv 7-book ‘The child had a book bought for him.’ b. * Ki-tabu ki-li-nunul-i-w-a m-toto. 7-book 7sa-past-buy-appl-pass-fv 1-boy ‘The book was bought for the child.’ (Ngonyani 1996: 38–9)
(98)
a. * M-toto amba-ye wa-li-nunul-i-a zawadi. 1-childrel-1 2sa-past-buy-appl-fv 9present ‘The child whom they bought a present for.’ b. Zawadi amba-zo wa-li-nunul-i-a wa-too. 10present rel-10 2sa-past-buy-appl-fv 2-child ‘The presents which they bought for the child.’ (Ngonyani 1996:44–5)
The lack of a correlation between the ability of a given object to undergo passive movement and its ability to undergo wh-movement is surprising if all movements have to proceed through phase edges. The phase theory does not allow a given head (be it a high or a low applicative head) to be a phase head with respect to some movements but not with respect to others. In the next section, I examine the predictions that the existence of two different applicative structures (coupled with the assumption that only one of them is a phase) makes for quantifier raising.
140
Symmetry in Move
4.6.4 Quantifier raising from applicatives The discussion in this section rests on two fairly standard assumptions about the nature of quantifier raising (QR). First, it assumes that QR is an instance of covert phrasal movement. Second, it assumes that the landing site for QR is the vP, which is the lowest constituent of the right semantic type.36 Since low applicative heads are not phase heads, raised quantifiers can move directly to vP. Logically speaking, there are two ways it can happen; either the IO moves first and the DO tucks in below it (as shown in (99a)) or the IO moves first and the DO stacks on top of it (as shown in (99b)). (99)
a.
b.
TP T
TP T
vP IOi
vP vP
DOj
vP DOj
IOi
vP v
SUBJ v
v
VP
ti
2 move
VP V
ApplLP ApplL ApplL
nd
v
SUBJ
V 1st move
vP
1st move
tj
ApplLP ti
ApplL ApplL
tj
2nd move
However, Bruening (2001) argues that QR only allows tucking in, which excludes (99b), and leaves us with (99a) as the only possibility. This is what allows him to account for the fact, noted by Aoun and Li (1989) and Larson (1990) (who attributes the observation to David Lebeaux, personal communication), that double object constructions in English, unlike their PP counterparts, have frozen scope. The examples in (100a–b) and (101a–b) provide an illustration; the PP ditransitives in (100a) and (101a) are scopally ambiguous; the universal quantifier inside the PP can have either wide or narrow scope with respect to the indefinite direct object. This is to be expected in a language like English, in which inverse scope is generally possible. What is interesting (and somewhat unexpected) is that their double object counterparts do not allow inverse scope. Thus, in (100b) every book does can have only narrow scope with respect to a different student, and in (101b), every problem can have only narrow scope with respect to one student. In other words, (100b) lacks the interpretation
4.6 Locality and symmetric passives
141
in which every book was given to a different student, and (101b) lacks the interpretation in which every problem was solved by a different student. (100)
a. The teacher gave a (different) book to every student. a > every, every > a b. The teacher gave a (# different) student every book. a > every, * every > a (Bruening 2001:235)
(101)
a. The teacher assigned one problem to every student. one > every, every > one b. The teacher assigned one student every problem. one > every, *every > one (Larson 1990:604)
Furthermore, it is not the case that the direct object has to get its scope in situ. The examples in (102a–b) show it can get wide scope with respect to the subject; in (102a) books can co-vary with teachers and in (102b) medals can co-vary with judges. (102)
a. A (different) teacher gave me every book. b. At least two judges awarded me every medal.
every > a every > at least two (Bruening 2001:243)
The trick is to allow the direct object to scope over the subject, while freezing its scope relative to the indirect object. To do so, Bruening (2001) proposes the following derivation for the example in (101b) above. Both objects undergo covert movement to the closest position in which they can be interpreted, which is the vP adjoined position. The indirect object moves first, as shown in (103a), and the direct object moves next, tucking in below the indirect object, as shown in (103b). (103)
a. [TP the teacherk [vP one studenti [vP tk v [VP assigned ti every problem] ] ] ] b. [TP the teacherk [vP one studenti [vP every problemj [vP tk v [VP assigned ti tj ] ] ] ] ]
The fact that indirect objects can have scope over subjects is not a problem, as the landing sites for the two QR-ed objects are above the base-generated position of the subject. In such cases, the subject is interpreted in its base position, thus it is going to have narrow scope with respect to both the direct and the indirect object. Bruening’s account of frozen scope in double object constructions makes an interesting crosslinguistic prediction. In particular, it predicts that a language with no superiority effects in wh-questions should allow both scope
142
Symmetry in Move
possibilities in double object constructions. This prediction was recently tested by Stepanov and Stateva (2009), who formulate it as in (104). (104)
A language shows Superiority effects with QR if and only if it shows Superiority effects with wh-movement. (Stepanov and Stateva 2009:177)
The presence of superiority effects in quantifier movement is diagnosed by scope; frozen scope is a sign of superiority effects. This correlation cannot hold in its strongest form without some extra caveats. While it holds in English double object constructions, it breaks down in simple transitive clauses. English shows superiority effects in multiple wh-questions involving subjects and objects (as shown in (105a–b)) but allows scope ambiguity with quantified subjects and objects (as shown in (106a–b)), which indicates that QR in English can violate superiority. (105)
a. Who visited whom? b. * Whom did who visit?
(106)
a. Some visitor visited every museum. b. Every visitor visited some museum.
some > every, every > some every > some, some > every
Polish, on the other hand, is a multiple wh-fronting language that allows violations of superiority in wh-questions, as shown in (107a–b). (107)
a. Komui coj nauczyciel zadał ti tj? who.dat what.acc teacher.nom assigned? ‘Whom did the teacher assign what?’ b. Coj komui nauczyciel zadał ti tj? what.acc who.dat teacher.nom assigned ‘What did the teacher assign to whom?’
This predicts that Polish double object constructions should be scopally ambiguous. This is not what we find; scope is fixed, as shown in (108a–b).37 (108)
a. Nauczyciel zadał każdemu studentowi jedno teacher.nom assigned every.dat student.dat one.acc zadanie. problem.acc ‘The teacher assigned every student one problem.’ every > one, * one > every b. Nauczyciel zadał jednemu studentowi każde zadanie. teacher.nom assigned one.dat student.dat every.acc problem.acc ‘The teacher assigned one student every problem.’ one > every, * every > one
Frozen scope in Polish double object constructions thus cannot be due to superiority, as Polish generally does not obey superiority.38
4.6 Locality and symmetric passives
143
The idea that high applicatives are phase heads also makes an interesting prediction about scope in high applicative constructions. Namely, it predicts that a direct object undergoing QR should have to move through the edge of a high applicative phrase, as shown in (109). Now the direct object is closer to v, so it will move first and the indirect object has to tuck in below it. (109)
TP T
vP DOi
vP IOj
vP v
SUBJ v
ApplHP ApplHP
ti 3rd move
ApplH
tj
ApplH
VP
EPP
2nd move
V 1st move
ti
This derivation predicts that high applicatives should only allow inverse scope unless we parametrize tucking in or assume the first step is optional, in which case either scope should be possible. However, this is not what we see in Polish. Both examples in (110a–b) involve high applicatives because of the use of the stative verb trzymać ‘to hold’. Scope, however, is also fixed in such examples, and reflects the order of the two objects, which is the opposite of what we would expect from the structure in (109).39 (110)
a. Maria trzymała każdemu gościowi jedne drzwi. Maria.nom held every.dat guest.dat one.acc door.acc ‘Maria was holding one door for every guest.’ every > one, * one > every b. Maria trzymała jednemu gościowi każde drzwi. Maria.nom held one.dat guest.dat every.acc door.acc ‘Maria was holding every door for one guest.’ * every > one, one > every
The conclusion that emerges from the discussion in this section (and the two sections preceding it) is that appealing to locality (which allows direct
144
Symmetry in Move
objects to passivize only if they move above indirect objects first) cannot be the only explanation for why in some languages passive movement is a symmetric operation and in others it is not. We saw problems with the correlation between symmetric passives and the availability of DO IO and IO DO word orders (Section 4.6.1) as well as with the correlation between symmetric passives and the availability of high applicatives (Sections 4.6.2–4.6.4). This suggests that while both types of locality-based accounts might be correct for some languages or dialects, they do not capture the full range of crosslinguistic variation. In the remainder of this chapter, I motivate the need for another strategy that languages can use to passivize their direct objects (which can also lead to symmetric passives). It involves case-driven movement of the indirect object, which freezes it for further case-driven movements, thus allowing the direct object to undergo passive movement, as shown in (111a). If the indirect object does not move first, it will be able to undergo passive movement, as shown in (111b). If both (111a) and (111b) are possible, the result is symmetric passive movement. (111)
a.
b. TP
TP
DOi
IOi T[EPP]
T[EPP] IOj tj
ti ti
DO
Since this movement strategy is linked to case, let me first introduce (and evaluate) existing case-based accounts of the distinction between symmetric and asymmetric passives. This is the focus of the next section. 4.7 Case and symmetric passives 4.7.1 Case absorption On the assumption that passive morphology absorbs structural case features (as argued by Baker, Johnson and Roberts 1989 and Jaeggli 1986, among others), passive movement could be related to the inability of direct objects to have their case valued. If we extend this idea to double object constructions, the difference between symmetric and asymmetric passives could be attributed to variation in case absorption (or more generally, case checking) mechanims. Broadly speaking, this is the view taken by Larson (1988), Legate (2001, 2002) McGinnis (1998) and Woolford (1993), among others.
4.7 Case and symmetric passives
145
One way to implement this view is to assume that there are two sources of case in applicative constructions, the applicative head and the v head. If both objects have structural case, the applicative head values the uninterpretable case feature of the direct object, whereas the v head values the uninterpretable case feature of the indirect object, as indicated by curved lines in (112a–b). Note that this account does not distriminate between high and low applicatives; in both types v undergoes Agree with the indirect object and the applicative head undergoes Agree with the direct object. (112)
a.
b. vP
vP v
v
ApplHP IO uC:Acc ApplH ApplH V
VP V
ApplLP IO uC:Acc ApplL
VP DO uC:Acc
ApplL
DOuC:Acc
The opposite pattern (with v valuing uCase on DO and Appl valuing uCase on IO) is excluded by locality conditions on Agree; IO is not in the c-command domain of Appl and IO intervenes for Agree between v and DO. There are thus two possibilities for case absorption in double object constructions; passive morphology can absorb either the case feature of the applicative head or the case feature of the v head. If v’s case feature is absorbed, the indirect object passivizes. If the applicative head’s case feature is absorbed, the direct object does. On this view, the crucial difference between languages with symmetric passives and those with asymmetric passives is that in languages with asymmetric passives only the first option is possible, i.e. only v’s case feature can be absorbed.40 A natural question that arises here concerns the existence of languages (or the lack thereof) with asymmetric passives in which only the case feature of the applicative head can be absorbed. This would yield a language in which both direct and indirect objects have structural case (valued by the applicative head and the v head, respectively) but only direct objects can passivize. I am not aware of the existence of such languages at this point. However, we know that the ability to passivize only direct objects is quite common in languages in which indirect objects are marked with inherent case (marked as iC in the diagrams that follow), such as German, Spanish or Polish, as shown by the following examples from Polish.
146 (113)
Symmetry in Move a.
Maria dała Janowi kwiaty. Maria.nom gave Jan.DAT flowers.ACC ‘Maria gave Jan flowers.’ b. Kwiatyi zostały dane Janowi ti. flowers.NOM became given Jan.DAT ‘Flowers were given to Jan.’ c. * Jani został dany ti kwiaty. Jan.NOM became given flowers.ACC ‘Jan was given flowers.’
If dative indirect objects enter the derivation with their case features valued, the v head will value the uC feature of the DO, as shown in (114a). If the ability of v to value case is absorbed by the passive morphology, the DO will get its case valued by T, as shown in (114b). (114)
a.
b.
TP T
TP vP
v
T
ApplHP IOiC:Dat
vP v
ApplH
V
V
ApplLP IO iC:Dat ApplL
VP
ApplH
VP
DOuC:Acc
ApplL
DOuC:Nom
As appealing at it may sound, such an explanation is not fully adequate, for two principled reasons. First, it predicts that dative arguments should not act as interveners for case-driven movements, and second, it relies on the underlying assumption that objects marked with non-structural cases should never be able to passivize. In the next two sections, I test these two predictions and show that (some) direct objects bearing non-structural case can passivize in Polish and that there is no correlation in grammaticality between passivization of a direct object across a dative indirect object and raising of a subject across a dative experiencer. 4.7.2 Lexical case and passive movement Many verbs in Polish assign lexical case to their objects.Some representative examples of such oblique objects are given below.41
4.7 Case and symmetric passives (115)
147
a. Ewa szuka mieszkania. Ewa.nom seek apartment.GEN ‘Ewa is looking for a new apartment.’ b. Ewa ufa Piotrowi. Ewa.nom trust Piotr.DAT ‘Ewa trusts Piotr.’ c. Ewa interesuje się składniom. Ewa.nom is.interested refl syntax.INSTR ‘Ewa is interested in syntax.’
What is interesting is that some of such oblique objects can passivize, as noted by Dziwirek (1994). (See also Swan 2002 and Citko 2009 for a discussion of such passives in Polish, and Fowler 1996 for a discussion of similar examples in Russian.)42 Examples (116) and (117) involve passivization of genitive objects and (118) of an instrumental object. (116)
a. Janek pilnował dziecka. John.nom watched child.GEN ‘John watched/minded the child.’ b. Dzieckoi było pilnowane ti przez Janka. child.NOM was watched by John ‘The child was being watched by John.’
(Dziwirek 1994:83)
(117)
a. Rząd zabronił demonstracji. government banned demonstrations.GEN ‘The government banned demonstrations.’ b. Demonstracjei zostały zabronione ti przez rząd. demonstrations.NOM became banned by government ‘The demonstrations were banned by the government.’
(118)
a. Janek kierował zakładem. John.nom manages factory.instr ‘John manages the factory.’ b. Ten zakładi był kierowany ti przez Janka. this. NOM factory.NOM was managed by John ‘This factory was managed by John.’ (Dziwirek 1994:83–4)
Dziwirek also notes that dative objects (even in simple transitive clauses) do not passivize (see, however, Zabrocki 1981 for some exceptions). (119)
a.
Janek pomagał Ewie. John.nom helped Eve.DAT ‘John helped Eve.’ b. * Ewai była pomagana ti przez Janka. Eve.NOM was helped by John ‘Eve was helped by John.’
148 (120)
Symmetry in Move a.
Janek powiedział Ewie o John.nom told Eve.DAT about ‘John told Eve about it.’ b. * Ewai była powiedziana ti (o tym) Eve.nom was told about ‘Eve was told about it by John.’
tym. it przez Janka. it by John (Dziwirek 1994:83–4)
The behavior of Polish oblique objects, in particular the fact that they can, under some circumstances, passivize, is problematic for the accounts of passives that rely on the idea that non-structural cases are not affected by passive morphology. The data discussed in this section suggest that it is not case but some other property of dative objects (direct and indirect alike) that makes them impervious to passive movement in Polish. Before examining the nature of this property in more detail, I turn to another prediction stemming from the idea that what allows direct objects in languages like Polish or German to passivize is the fact that dative indirect objects are case-marked in situ. Such an account makes predictions that extend beyond passive movement in double object constructions. In particular, it predicts that dative arguments in general should not block A-movement across them. The next section tests this prediction in constructions involving raising across dative experiencers.43 4.7.3 Dative intervention effects As often noted in the literature on double objects or dative intervention effects, subject-to-subject raising across a dative experiencer and passivization of a direct object across a dative indirect object involve the same general configuration; both target the specifier of TP and both cross an intervening dative argument. (121)
a. DPi T DPDAT ti b. DPi T DPDAT ti
subject raising passivization of direct object
A more specific representation of raising to subject is given in (122). It assumes that dative experiencers are introduced by applicative heads, and that they c-command (traces of) raised subjects.
4.7 Case and symmetric passives (122)
149
TP T
DPi T
vP v
Appl HP DPDAT
Appl H VP
ApplH
TP
V
T
ti T
vP
If dative case being inherent is what makes movement across dative indirect objects possible in double object constructions, we would expect raising across dative experiencer subjects to be possible as well. And conversely, if movement across a dative indirect object is impossible, we would expect movement across a dative experiencer to be impossible as well. We saw above data from three languages in which passivization across a dative direct object is possible; the prediction is that in these languages (if experiencers are also dative), raising across experiencers should also be possible. In general, we expect the same intervention effects (or the lack thereof) in both double object constructions and subject-to-subject raising; either movement in both cases is possible (as schematized in (123a)) or movement in both cases is impossible (as schematized in (123b)). We do not expect the patterns in (123c) and (123d), in which movement is possible in one case but not in the other: (123)
a.
[TP DPi [TP DPi b. * [TP DPi * [TP DPi c. [TP DPi * [TP DPi d. * [TP DPi [TP DPi
T … DPexp [TP ti … ] ] T … [ApplP DPind-obj … ti ] ] T … DPexp [TP ti … ] ] T … [ApplP DPind-obj … ti ] ] T … DPexp [TP ti … ] ] T … [ApplP DPind-obj … ti ] ] T … DPexp [TP ti … ] ] T … [ApplP DPind-obj … ti ] ]
subject raising passivization of direct object subject raising passivization of direct object subject raising passivization of direct object subject raising passivization of direct object
We saw above that passive movement across dative indirect objects is possible in languages like Polish or Spanish. In Polish, subject-to-subject raising
150
Symmetry in Move
across an intervening experiencer is also possible. The status of the dative experiencer does not make a difference; in (124a) it is a pronoun, in (124b) a full DP, and in (124c) a wh-trace. (124)
a. Jani wydaje mi się [ ti (być) najlepszym kandydatem]. Polish Jan seems me self be best candidate ‘Jan seems to me to be the best candidate.’ b. Jani wydaje się Marii [ti (być) najlepszym kandydatem]. Jan seems self Maria.dat be best candidate ‘Jan seems to Maria to be the best candidate.’ c. Komuj Jani wydaje się tj [ti być najlepszym kandydatem?] who.dat Jan.nom seems self be best candidate ‘To whom does Jan seem to be the best candidate?’
In Spanish, however, raising across the experiencer is blocked in all three cases, as pointed out by Torrego (1996). (125)
a. * Este taxistai me parece [ti estar cansado]. Spanish this taxi.driver.nom me.dat seems be tired ‘It seems to me that this taxi driver is tired.’ (Torrego 1996:106) b. * Este taxistai parece a Maria [ti estar cansado]. this taxi.driver.nom seems to Maria be tired ‘It seems to Maria that this taxi driver is tired.’ c. *A Mariaj, este taxistai tj parece [ti estar cansado]. to Maria this taxi.driver.nom seems be tired ‘To Maria, this taxi driver seems to be tired.’ (Kim 2005:650)
Icelandic is another language that allows us to test this correlation. Icelandic can passivize direct objects across dative indirect objects, and marks its experiencers with dative case. However, raising across experiencers is not possible (unless the wh-experiencer is a wh-trace).44 (126)
a. * Hestarnir virðast mér [ti vera seinir]. Icelandic the.horses.nom seem me.dat be slow ‘It seems to me that the horses are slow.’ b. Hverjumi hafa hsestarnirj virst ti [ tj vera seinir]? who.dat have the.horses.nom seemed be slow ‘To whom has it seemed that the horses are slow?’ (Holmberg and Hróarsdóttir 2003:998)
The table in (127) summarizes the raising patterns discussed in this section. It shows that there is no correlation between subject-to-subject raising across a dative experiencer and passive movement of a direct object across a dative indirect object.
4.7 Case and symmetric passives (127)
151
Movement across dative arguments: a summary Spanish Polish Icelandic
Raising across dative indirect objects ✓ Raising across dative experiencers *
✓ ✓
✓ * (unless Exp is wh-trace)
4.7.4 Lack of multiple datives We saw in Section 4.7.2 above that there is something unique about Polish datives which prevents them from undergoing passive movement. It cannot be the fact that dative case is non-structural, as other non-structural cases do sometimes passivize. Another unique fact about Polish datives, also noted by Dziwirek (1994), is their incompatibility with other datives, more specifically, the fact that Polish allows only one dative-marked nominal per clause. Before presenting the data to illustrate this restriction, let me introduce in (128a–f) other constructions involving dative case in Polish.45 (128a) involves a dative experiencer subject of a psych verb, (128b) a dative subject of a modal verb, (128c) a dative possessor, (128d) an ethical dative, (128e) an estimative dative, and (128f) an involuntary state construction. The first two terms are fairly self-explanatory; many psychological and modal verbs require dative subjects. A possessor raising construction is a construction in which the possessor is marked with dative case. An ethical dative construction is a construction in which the dative argument is directly affected by the action. In an estimative dative construction, the dative is the argument whose assessment of the event we are interested in. It typically occurs with a degree modifier. And in an involuntary state construction, the focus is more on the dative subject undergoing a certain state than being directly involved in a certain action. According to Dziwirek (1994:57), “the use of a dative-subject clause attributes a somewhat non-volitional quality to the activity/state, suggesting that it takes place independently, or in spite of the (human) subject involved.” (128)
a. Basi podobają się te kwiaty. Basia.DAT like refl these flowers.nom ‘Basia likes these flowers.’ b. Potrzeba nam więcej kwiatów. need we.DAT more flowers.gen ‘We need more flowers.’ c. Jan złamał Piotrowi rękę. Jan.nom broke Piotr.DAT arm.acc ‘Jan broke Piotr’s arm.’
psych verb
modal verb
possessor raising
152
Symmetry in Move d. Dziecko Basi zasnęło. ethical dative child.nom Basia.DAT fell.asleep ‘The child fell asleep on Basia.’ e. Jan jechał Basi za szybko. estimative dative Jan.nom drove Basia.DAT too fast ‘Jan drove too fast for Basia’s liking.’ f. Basi łatwo się pisze wiersze. involuntary state construction Basia.DAT easily refl write poems.acc ‘Writing poems comes easily to Basia.’
Now we can illustrate the ban against multiple datives alluded to above. Possessors in Polish can be either genitive or dative, as shown in (129a–b). Dative possessors are understood to be directly affected by the action. No such implication is present with genitive possessors, which therefore tend to be more common.46 (129)
a. Oczy Ewy zaszły eyes.nom Eve.GEN clouded ‘Eve’s eyes filled with tears.’ b. Oczy zaszły Ewie eyes.nom clouded Eve.DAT ‘Eve’s eyes filled with tears.’
łzami. tears.INSTR łzami. tears.INSTR (Dziwirek 1994:87)
Interestingly, dative possessors become impossible if there is another dative within the same clause. The examples in (130a) and (131a) are ungrammatical because they contain both a dative possessor and a dative experiencer. If the possessors are genitive instead, as in (130b) and (131b), the examples become grammatical: (130)
(131)
a. * Oczy Ewie podobają się eyes.nom Eve.DAT please refl ‘John likes Eve’s eyes.’ b. Oczy Ewy podobają się eyes.nom Eve.GEN please refl ‘John likes Eve’s eyes.’
Jankowi. John.DAT Jankowi. John.DAT
a. * Jan kupił Magdzie włosy Ewie John.nom bought Mary.DAT hair.acc Eve.DAT na perukę. for wig ‘John bought Mary Eve’s hair for a wig.’ b. Jan kupił Magdzie włosy Ewy John.nom bought Mary.DAT hair.acc Eve.GEN na perukę. for wig ‘John bought Mary Eve’s hair for a wig.’ (Dziwirek 1994:88)
4.7 Case and symmetric passives
153
Involuntary state constructions present a potential problem for the generalization that multiple datives are impossible in Polish. Example (132) is grammatical in spite of the fact that it contains two datives.47 (132)
Basi łatwo daje się Piotrowi prezenty. Basia.DAT easily give refl Piotr.DAT packages ‘It is easy for Basia to give presents to Piotr.’
The exceptional status of such examples, however, is only apparent, and can be attributed to the fact that involuntary state constructions are bi-clausal, as shown by Marušič and Žaucer (2006), for example. They argue, based on Slovenian data, that the matrix clause in an involuntary state construction contains a covert psych predicate and that the embedded clause is deficient in that it lacks a TP and CP layer, as shown in (133).48 (133)
[CP [ TP [AspP [vP DPDAT [ VP Ø [R-ModP [AspP [vP [VP ] ] ] ] ] ] ] ] ]
The details of their analysis are not directly relevant for our purposes; suffice it to note that there is an alternative explanation for why examples like (132) are grammatical; the two datives are not in the same clause. Before turning to the analysis of the ban against two datives in a single clause in Polish, let me briefly mention (and discount) two possible explanations. First, it cannot be due to a more general ban against two DPs marked with the same case in a clause; the grammatical example in (134a) contains two adjacent genitive arguments and the one in (134b) two instrumental DPs (one argument and one adjunct).49 (134)
a. Maria nie nauczyła Jana matematyki. Maria.nom not taught Jan.GEN math.GEN ‘Maria didn’t teach Jan math.’ b. Maria kierowała tym zakładem latami. Maria.nom directed this.INSTR factory.INSTR years.INSTR ‘Maria directed this factory for years.’
Second, the ban against double datives cannot be attributed to a more general ban against recursion of applicative structures. Such recursion is in principle possible; a low applicative phrase can be embedded under a high applicative phrase (with the main verb intervening between the two), as shown in (135a). Likewise, a high applicative phrase can be embedded under another high applicative phrase, as shown in (135b).50
154 (135)
Symmetry in Move a.
b.
ApplHP IO
ApplH ApplH
IO
VP V
ApplHP ApplH ApplH
ApplLP IO
IO
Appl ApplL
ApplHP ApplH ApplH
DO
VP V
DO
The analysis I develop in the next section suggests that both restrictions in Polish (i.e. the lack of dative passives and the ban on multiple datives) follow from a single mechanism of dative case licensing. 4.7.5 Freezing and symmetric passives The analysis I develop in this section to account for the behavior of Polish datives relies on a number of independently motivated assumptions.51 First, following Woolford (2006) most recently, I take dative case to be quirky case.52 Second, I take quirky case to be a combination of lexical and structural case. The lexical component of quirky case can be thought of as the presence of an interpretable case feature (represented as iC feature in the diagrams that follow), and its structural component can be thought of as the presence of the uninterpretable case feature (represented as uC feature). (136)
a. Quirky case = lexical case + structural case b. Quirky case = iC feature + uC feature
The standard minimalist assumption is that interpretable case features enter the derivation valued. I depart from this assumption slightly; nothing crucial, however, hinges on this departure. Even though interpretable case features might not need to checked (or valued) in the same way uninterpretable case features do, they do need to be licensed in specific positions by specific syntactic elements. For example, instrumental- or genitivemarked objects are only possible as complements of predicates that assign instrumental or genitive case. One way to capture this fact but still maintain the distinction between interpretable and uninterpretable features is to assume that both interpretable and uninterpretable features need to be valued but that they get valued in different ways. More specifically, interpretable case features get valued at first Merge, but uninterpretable case features get valued later, via Agree with a higher Probe, as argued by Bailyn
4.7 Case and symmetric passives
155
and Citko (1999) for instrumental versus so-called agreeing case on nonverbal predicates. On this view, dative indirect objects enter the derivation with both an iC feature and an uC feature. The iC feature is valued on Merge with an applicative head, either a low or a high one, as shown in (137a–b) respectively. (137)
a.
b.
VP
IO
Appl LP
V
Appl HP [ iC:Dat, uC: ]
ApplL
IO [ iC:Dat, uC:]
Appl L
Appl H
Appl H
VP V
DOuC:
DOuC
The crucial innovation lies in the valuing of the uC feature of the indirect object. I propose that it is valued by Agree with a new functional projection, which I dub a light applicative head (applP), generated above ApplP. I furthermore propose that this appl head has an EPP feature, which forces movement of the dative argument to its specifier.53 (138)
a.
b.
applP IOi [ iC:Dat, uC:Dat]
applP
appl
appl
IOi [iC:Dat, uC:Dat]
ApplLP
[EPP]
ti
appl
ApplHP
[EPP]
ApplL ApplL
appl
ti DOuC:
ApplH ApplH
VP V
DOuC:
This movement not only explains why dative objects (direct and indirect alike) cannot passivize but also captures the ban on multiple datives in Polish. To see how, let us look at some concrete examples, starting with low applicatives. The indirect object in (139) enters the derivation with an iC feature and an uC feature. Its iC feature gets valued on Merge (or enters the derivation valued, depending on one’s theoretical assumptions about the status and licensing of interpretable features). Its uC feature gets valued via Agree with a higher appl head. Since the appl head also has an EPP feature, the indirect object moves to its specifier. The subject moves to [Spec,TP] and the verb moves to v. The direct object has its uC feature valued via Agree with v.
156 (139)
Symmetry in Move a. Jan wysłał Piotrowi książkę. John.nom sent Peter.dat book.acc ‘John sent Peter a book.’ b. TP T
Jani
[uC:Nom]
T
vP
[EPP]
v
ti sent k
applP appl
Piotri
[iC:Dat, uC:Dat]
appl[EPP] tk V tk
VP Appl LP Appl L
ti ApplL
book [uC:Acc]
High applicatives, such as the one given in (140a), involve an analogous derivation, schematized in (140b). The only difference is that the dative indirect object starts out as a specifier of a high applicative head. (140)
a. Jan pobiegnie wam na zawodach. Jan.nom run.perf.3sg you.dat at competition ‘Jan will run for you at the competition.’ b. TP T
Jani
[uC:Nom]
T
vP
[EPP]
v
ti v runk
applP appl
you i [iC:Dat, uC:Dat]
appl[EPP] tk
ApplHP ti
Appl H Appl H tk
VP tk at a competition
4.7 Case and symmetric passives
157
Now we can turn to passive movement in both high and low applicatives. What allows the direct object to passivize in (141) is the fact that the indirect object has already had its uC feature valued (via movement to the specifier of the light applicative head). Thus it no longer intervenes for the purposes of A-movement. If the indirect object were to passivize instead, as shown in (142b), movement of the indirect object from the specifier of applP to the specifier of TP would violate economy; the indirect object would be moving from one case position to another one. Furthermore, the direct object would be left with its uC feature unvalued. (141)
a. Książkai została wysłana Piotrowi ti. book.nom became sent Piotr.dat ‘The book was sent to Piotr.’ b. TP book i
[uC:Nom]
T T[EPP] vP became v applP sentk Piotri appl [iC:Dat, uC:Dat]
appl[EPP] tk V tk
VP ApplLP ApplL
ti ApplL
(142)
a. * Piotri został wysłany ti książkę. Piotr.nom became sent book. acc ‘The book was sent to Piotr.’
ti
158
Symmetry in Move b.
TP T T[EPP] became v
vP VP applP
sentk Piotri
[iC:Dat, uC:Dat]
appl
appl[EPP] tk
VP V tk
ApplLP ApplL
ti ApplL
book uC:
We also saw above that the ban against passivization of indirect objects in Polish is not absolute. If the indirect object is accusative (not dative), the result is the opposite passivization pattern; the indirect object can passivize but the direct object cannot. This is what happens with the verb uczyć ‘to teach’, whose indirect object is accusative and direct object genitive. (143)
a.
Jan nauczył Piotra składni. Jan.nom taught Piotr.ACC syntax.GEN ‘Jan taught Piotr syntax.’ b. Piotri został nauczony ti składni. Piotr.nom became taught syntax.gen ‘Piotr was taught syntax.’ c. * Składniai została nauczona Piotra ti. syntax.nom became taught Piotr.acc ‘Syntax was taught to Piotr.’
In such cases, v values the uC feature of the IO as accusative, and the Appl head values the uC feature of the DO as genitive, as shown in (144a). Since passive morphology absorbs v’s ability to value case, T is the closest element that can value case of the IO, as shown in (144b). As a result, the IO is going to undergo passive movement.
4.7 Case and symmetric passives (144)
a.
159
TP T
Jan i
[uC:Nom]
T
vP
[EPP]
v
ti v
VP ApplLP
V taught
Piotr [uC:Acc]
ApplL ApplL
b.
syntax [uC:Gen]
TP T T[EPP] became v
vP VP V taught
ApplLP Piotr [uC:Nom]
ApplL ApplL
syntax [uC:Gen]
On this account, the fact that dative indirect objects cannot passivize and the fact that there can only be one dative argument per clause is tied to movement of dative arguments to the specifier of applP. In other words, the fact that dative indirect objects do not passivize is only indirectly linked to non-structural case. Such dative objects have already moved for caserelated reasons; thus they are frozen in place for further case-related movements. In other words, (145a) is ungrammatical for the same reasons (145b) and (145c) are. In (145b) the nominative subject John can have its case features valued inside the embedded clause, thus it does not need to move to the matrix subject position, and, likewise, in (145c) the wh-phrase what can have its wh-feature satisfied in the embedded clause, thus there is no reason for it to move to the matrix [Spec, CP].
160 (145)
Symmetry in Move a. * Jani został dany ti kwiaty przez Ewę. Jan.nom became given flowers.acc by Ewa ‘Jan was given flowers by Ewa.’ b. * Jani is likely that ti gave Ewa flowers. c. * Whati does Piotr wonder ti Jan gave Ewa ti?
The fact that Polish does not allow more than a single dative marked argument per clause is also related to the movement of datives to the specifier of applP. In particular, it follows from the assumption that the light applicative head has only one specifier position (and that there is only one applP projection per clause). The ungrammatical example in (146a) below contains two dative arguments. Since there is only one appl head, it can only value case features of one of them. This is going to be the one that raises to its specifier. If the benefactive argument raises, the possessor will be left with an unvalued case feature. If, instead, the possessor raises, the benefactive argument will be left with an unvalued case feature.54 (146)
a. * Jan kupił Magdzie włosy Ewie. John.nom bought Mary.DAT hair.acc Eve.DAT na perukę. for wig ‘John bought Mary Eve’s hair for a wig.’ (Dziwirek 1994:88) b. TP T
vP v
John [uC:Nom]
v
applP appl appl
ApplHP
[EPP]
Mary [iC:Dat,uC:]
ApplH ApplH
VP ApplLP
V
ApplL
Eve [iC:Dat, uC:]
ApplL
hair [uCase:Acc]
4.8 Conclusion
161
Thus the idea that there is a single light applicative phrase accounts for two otherwise rather peculiar properties of Polish datives: the fact that dative indirect objects cannot passivize and the fact that there can only be one dative argument per clause. 4.8
Conclusion
My main goal in this chapter was to provide evidence for the existence of symmetric movement in the grammar. I focused on passive movement in double object constructions, as it is a perfect example of a configuration in which there is a single landing site (the subject position) and two elements vying for this site (the two objects standing in an asymmetric c-command configuration). Movement is symmetric if either the direct or the indirect object can passivize with a grammatical result. On the assumption that the indirect object is merged above the direct object, we would expect only the indirect object, which is closer to T, to be able to passivize. This is what happens in some languages; however, there are others in which either both objects can passivize or only the lower of the two, the direct object, can. Since passivization of the higher object is the expected pattern, the bulk of the discussion in this chapter was devoted to the strategies languages use to allow the direct object, the lower of the two, to passivize. Combining the insights of previous accounts, I argued that there are two ways it can happen, leading to two types of symmetric movement. I referred to them as locality-based and case-based strategies. In a locality-based strategy, the direct object moves over the indirect one (before moving to the subject position). This is possible in languages in which both objects are marked with structural case and it does not correlate with the distinction between high and low applicatives. If the direct object does not move first, the indirect object will passivize. The result is symmetric movement. (147) a.
b.
TP
TP IOi
DOi
T[EPP]
T[EPP] ti IO
ti ti
DO
162
Symmetry in Move
However, I also argued in this chapter for the existence of a case-based strategy to passivize direct objects. It also involves movement; however, in this strategy movement involves the indirect object, which moves for case reasons to a non-subject position, as shown in (148a). This movement freezes the indirect object for further case-related movements and makes passivization of the direct object the only option. If the indirect object does not undergo this movement first, it will be able to passivize. Again, if both options are allowed, the result is symmetric movement. (148)
a.
b. TP
TP IOi
DOi T[EPP]
T[EPP]
IOj
ti
tj ti
DO
5
Symmetric labels
5.1
Introduction
In this chapter, I provide evidence for the existence of symmetric labels in the grammar. This is a departure from the standard minimalist view, which takes labels to be asymmetric. When two objects (such as α and β) merge, only one of them (either α or β) projects as the label of the new object, as shown in (1a–b).1 (1)
a. Merge α and β, project α
b. Merge α and β, project β
α α
β β
α
β
Logically speaking, there are two ways labels can be symmetric, illustrated in (2a) and (2b) respectively. In (2a), both α and β project, whereas in (2b) neither of them does.2 Crucially, in both cases α and β contribute equally to the identity of the label. (2)
a. Merge α and β, project both
b. Merge α and β, project neither
{α, β} α
β
α
β
My main focus in this chapter is on symmetric labels of the kind given in (2a).3 However, in order to argue for the existence of such labels, we first need to establish the existence of labels of any sort in the grammar. This is not a trivial pursuit, especially in light of recent claims to the contrary (see Collins 2002, Seely 2006, among others, for arguments to rid the grammar of labels altogether). I therefore devote the bulk of the discussion in the first part of this chapter to justifying the existence of labels. In the second part, I turn to the evidence for symmetric labels. The claim I defend is that labels can be symmetric, not that they have to be (or always are) symmetric. 163
164
Symmetric labels
This claim is similar in spirit to the claim about Merge in Chapter 3 (where I argued that Merge can also sometimes create symmetric structures) and the claim about Move in Chapter 4 (where I argued that Move can also sometimes treat two elements in a symmetric fashion). This chapter is structured as follows. In Section 5.2, I present (and counter) the proposals that eliminate labels from the grammar, focusing on Collins (2002), who gives us a detailed account of how a grammar without labels might work. As I review his proposal, I show that even though we can accomplish a lot without labels, there are some operations that crucially rely on their existence, both in Narrow Syntax and at the interfaces. In Section 5.3, I introduce symmetric labels, focusing on what theoretical conditions they have to meet. In Sections 5.4–5.7, I turn to the empirical support and present three case studies in symmetric Labeling: serial verbs (following Baker and Stewart 1999), small clauses (following Moro 2000, 2007 and Pereltsvaig 2008), and comparative correlatives (following Citko 2008b). I argue that all three involve symmetrically labeled constituents, consisting of two phrases of the same type (i.e. VPs, vPs, DPs or CPs). 5.2
Arguments for (and against) eliminating labels
Collins’s (2002) conceptual arguments for eliminating labels come from the fact that labels are vestiges of representations, and are thus undesirable in a highly derivational minimalist system. More concretely, he argues that labels violate the Inclusiveness Condition, given in (3) below. (3)
Inclusiveness Condition
(Chomsky 1995:228)
No new objects are added in the course of the computation apart from rearrangements of lexical properties (in particular, no indices, bar levels in the sense of X-bar theory).
This is a common objection to labels, found in different forms in the literature. However, this objection is only valid if a label is a new object, not present in the Numeration. The objection disappears in Bare Phrase Structure Theory, in which a label is simply one of the merged elements, or a set (or a subset) of its features. Both set-theoretical and tree-theoretical representations, given in (4a–b) below, reflect this “bare” view of labels. Labeling thus does not introduce any new objects into the derivation and is not a problem for the Inclusiveness Condition.
5.2 Arguments for (and against) eliminating labels (4)
165
a. {α, {α, β}} α
b. α
β
A slightly different conceptual argument in favor of (at least) partial elimination of labels comes from the insights of Distributed Morphology, in which the lexicon is assumed to contain label-less roots such as √construct, √destroy, rather than categorically labeled nouns, verbs or adjectives (as argued by Halle and Marantz 1993, 1994, Harley and Noyer 2003 and Marantz 1997, among others). The same root will end up as a noun if it merges with a determiner and a verb if it merges with a v or an Aspect head. Marantz (2001) takes it a step further, proposing that the heads that “turn” label-less roots into lexical categories (nouns, verbs, adjectives) are all phase heads.4 This view of label-less roots is also prominent in Borer (2005). Collins does not give many empirical reasons for eliminating labels; his focus is on showing how a grammar without labels would work. Thus, in order to understand his proposal, we need to understand what labels do (syntactically speaking). If they are not that useful to begin with, getting rid of them is not going to matter that much (and will certainly be desirable from a minimalist perspective). If, however, labels accomplish major syntactic tasks, dispensing with them will have dire consequences. We have to be somewhat cautious, however, in evaluating the need for labels (or the lack thereof) in the grammar. For example, we often appeal to labels in discussions of constituency tests. However, many of these tests do not distinguish between constituents with different labels (DPs versus VPs), but between maximal and non-maximal projections. For example, no reference to labels is necessary in order to account for the contrast between grammatical topicalizations of DPs and PPs in (5a–b) and the ungrammatical topicalizations of D-bars and P-bars in (6a–b).5 (5)
a. his best friendi, I met ti yesterday. b. Best friendi, John ran ti.
DP topicalization PP topicalization
(6)
a. * Best friendi, I met his ti. b. * Into the storei, John ran right ti.
D′ topicalization P′ topicalization
However, the existence of syntactic processes that apply to, or are triggered by, phrases of specific types should make us somewhat cautious of proposals that eliminate labels. To give just a couple of concrete examples,
166
Symmetric labels
only PPs allow their heads to be stranded via wh-movement, as shown in (7), and only &Ps require their specifiers and complements to be of the same category, as shown by the contrast in grammaticality between (8a) and (8b). (7)
Which articlei did you think [PP about ti ]?
(8)
a. [&P [DP John] [&′ and [DP his friend] ] ] b. * [&P [DP John] [&′ and [AdvP friendly] ] ]
Furthermore, processes like selection or movement crucially rely on (or make reference to) category labels. If the result of merging the determiner the and the noun book does not have a label, it is unclear how the verb read merging with the DP the book can have its selectional requirements satisfied. To address this issue, Collins restates selectional restrictions as “statements about derivational syntactic relations and economy conditions” (Collins 2002:44). The syntactic relations he appeals to are listed in (9a–d). (9)
a. b. c. d.
Theta (X, Y) EPP (X, Y) Agree (X, Y) Subcat (X, Y)
X assigns a theta role toY Y satisfies the EPP feature of X X matches Y, and Y values X X subcategorizes for Y
All the Xs and Ys in (9a–d) are lexical categories, not phrasal ones. Crucially, Collins does not argue for eliminating lexical labels (which could be thought of as features of lexical elements, represented by subscripts in (10a–b)). He does argue, however, that such lexical labels (or features) do not project up the tree. In other words, the result of merging the and book is (10a) not (10b). (10)
a.
b. theD
bookN
theD theD
bookN
In a system without labels, violations of selection are handled in two different ways, depending on the kind of violation. There are two kinds to consider; the selected element could be of a wrong type, or it could be absent altogether. The contrast between (11a) and (11b) illustrates the first type of violation. (11)
a. destroy cities b. * destroy to cities
5.2 Arguments for (and against) eliminating labels
167
In a system with labels, the ungrammaticality of (11b) is due to the fact that destroy subcategorizes for a DP not a PP. To account for its ungrammaticality in a system without labels, Collins extends the kind of minimality familiar from the research on movement and proposes the following condition on selection: (12)
If X selects Y (where Y is a lexical category), then * X Z Y where Z intervenes between X and Y, and Z is any lexical category (+/− V, +/− N). (Collins 2002:52)
If intervention reduces to c-command, (11b) above is ungrammatical because to intervenes between destroy and cities. However, the kind of minimality involved here is very different from the kind involved in movement, which is typically attributed to Relativized Minimality (or its conceptual successors). Relativized Minimality blocks relationships between elements of the same type, whereas the condition on selection in (12) allows any lexical item to act as an intervener.6 The second type of violation is illustrated in (13), which contains no element to meet the verb’s selectional requirements. To account for its ungrammaticality, Collins introduces the concept of a derivational locus. (13)
* John devoured.
Simply put, when two elements merge, a feature of one of them has to be satisfied. Collins refers to this feature as a probe (or selector), and the element bearing it as the locus of the derivation. Furthermore, he proposes the Locus Principle, given in (14) below, according to which no new elements can be introduced before the current selectors are satisfied. (14)
Locus Principle
(Collins 2002:46)
Let X be a lexical item that has one or more probe/selectors. Suppose X is chosen from the lexical array and introduced into the derivation. Then the probe/selectors of X must be satisfied before any new unsaturated lexical items are chosen from the lexical array. Let us call X the locus of the derivation.
Given the Locus Principle, the ungrammaticality of (13) above is due to the fact that a new unsaturated element (little v) is introduced into the derivation before the subcategorization feature of devour is satisfied. This results in the presence of two derivational loci, in violation of the Locus Principle.7
168
Symmetric labels
Another aspect of syntax where labels play a role concerns movement. There are two reasons why labels are useful here. First, they help determine how much needs to move. Second, they help determine what elements count as interveners for what kinds of movement. Let me examine these two issues in turn. In a wh-question involving a complex wh-phrase, such as the one given in (15a) below, the entire DP headed by the whdeterminer which needs to move, not just the lexical item bearing this wh-feature. In a system with labels, this can be captured quite straightforwardly by assuming that the wh-feature of the wh-determiner percolates up to the phrasal level or that the wh-determiner pied-pipes the entire phrase, as shown in (15b).8 With no labels, (15b) becomes (15c) and it is thus not quite clear what forces the entire DP to move (as opposed to just the wh-word which). (15)
a. Which booki are you reading ti? b. CP C’ C[Q,uwh]
…
DP[wh] which[wh]
book
c. C[Q,uwh]
…
which[wh]
book
Collins recognizes this issue and presents examples in which feature percolation does not work. In (16a) below, whose DP which man’s daughter has the structure in (16b), the wh-determiner does not head the DP which man’s daughter, which is the DP undergoing wh-movement. On standard assumptions about feature percolation, only the specifier of this DP (i.e. which man) is marked as interrogative. However, if the specifier which man agrees with the D head (as is typically assumed), the entire DP which man’s daughter becomes interrogative in the relevant sense.
5.2 Arguments for (and against) eliminating labels (16)
169
a. [Which man’s daughter]i did you see ti? b. DP DP[wh] D[wh] NP
D D
NP
which man ‘s daughter
A related issue for a grammar without labels concerns locality restrictions on movement. In the ungrammatical example in (17) below, why cannot move to the specifier of the matrix C because of the intervening wh-phrase which man’s daughter. With no labels, it is not clear how to capture such intervention effects, since the wh-determiner which in its premovement position does not c-command the adjunct why (the entire DP which man’s daughter, on the other hand, does). (17)
* Whyi do you wonder which man’s daughter stole the money ti?
Collins argues that the ungrammaticality of (17) is linked to the fact that the embedded interrogative complementizer intervenes between the matrix interrogative complementizer and the adjunct why, as shown in (18). Such a formulation of the Minimal Link Condition does not require reference to phrasal labels. (18)
[ C[+wh] you wonder [ [which man’s daughter] C[+wh] [stole the money why ] ] ] (Collins 2002:57)
The upshot of the discussion above is that even though some work traditionally attributed to labels can be relegated to other mechanisms in the grammar, there are aspects of selection and movement that remain problematic in a system with no labels. In what follows, I examine two more phenomena that require the presence of category labels. One involves operations on labels, and the other one the differences between arguments and adjuncts that can be reduced to different types of labels. If there are operations that make reference to category labels (which cannot be restated as operations on lexical items), we have a solid argument in favor of the existence of labels. One argument that such operations exist comes from the work of Hornstein and Uriagereka (2002) on reprojection. Descriptively speaking, reprojection is a mechanism that allows a category to change its label in the course of the derivation. What
170
Symmetric labels
this means is that when α and β merge and α projects, the result can later reproject as β. In X-bar theoretical terms, reprojection can be represented as follows: 9 (19)
a.
b.
YP Y
XPi Y
XPi Xi
→ REPROJECTION →
YP Y
ti
ti
(Hornstein and Uriagereka 2002:107)
In more current, bare phrase structure theoretical terms, reprojection is represented as in (20a–b). When x (internally) merges with y, y (the Probe) becomes the label. Later, however, x (the Goal) can reproject, labeling the entire structure as x. (20)
y
a. xi
b. y
y
→ REPROJECTION →
x xi
y
ti
y
ti
Needless to say, if reprojection exists, labels have to exist as well. Hornstein and Uriagereka’s (2002) main empirical argument in favor of reprojection comes from the properties of binary quantifiers such as most. Reprojection provides a very intuitive account of the fact that the restriction and scope of such quantifiers behave like its two arguments (which parallels the subject and the object being two arguments of the verb).10 Example (21) provides a concrete illustration. Before reprojection, its structure is (22a). After reprojection, the restriction and the scope of the quantifier most literally become its two arguments; the NP people is the complement of most and the IP love children is its (rightward) specifier, as shown in (22b). (21)
Most people love children.
(22)
a.
I
QPi Q
NP
most people
b
IP
ti
Qi
→ REPROJECTION → Q
VP
I
V love children
QP IP NP I
most people
VP ti
V love children
(Hornstein and Uriagereka 2002:109)
5.2 Arguments for (and against) eliminating labels
171
Strictly speaking, Hornstein and Uriagereka do not conceive of reprojection as a new (or a non-standard) syntactic operation. Rather, they assume that at different stages in the derivation, different elements can count as labels. This, however, does not affect the main point of this section; either way of thinking of reprojection requires reference to (and thus the existence of) labels, which is all I set out to establish here.11 A slightly different argument for the existence of labels comes from cases of Merge in which either of the two merged elements can project with a grammatical result. A case in point comes from Rooryck’s (2000) analysis of qui/que alternation in French. What is relevant for our purposes is not the factors that determine when qui is used and when que is used, but the fact that both qui and que can function either as interrogative pronouns or as complementizers. Example (23a) illustrates the use of qui as an interrogative pronoun, and (23b) its use as a complementizer; (24a–b) do the same for que. (23)
a. Quii ti est arrivé? who is arrived ‘Who arrived?’ b. Le paqueti qui ti est venu. the package that is arrived ‘the package that came’
(24)
a. Qu’ as-tu vu? what have-you seen ‘What did you see?’ b. Le paqueti que j’ ai vu ti. the package that I have seen ‘the package that I have seen’
(Rooryck 2000)
Rooryck’s (2000) account of this alternation relies crucially on the existence of labels. First, he proposes that morphologically qui is a combination of the complementizer que and the subject pronoun clitic il. What this means is that both the complementizer qui and the wh-pronoun qui involve Merge of the complementizer qu(e) and the subject clitic il. The difference between them lies in which of the two (the complementizer or the clitic) projects. If C projects, the result is a complementizer, and if D projects, the result is a wh-pronoun. Such treatment of the complementizer qui departs from many standard analyses, which treat qui as an agreeing complementizer (see Rizzi 1990, for example).12
172 (25)
Symmetric labels a. Merge C and D, project C
b.
Merge C and D, project D
C
D
C
D
qu–
(l)i
C
D
qu–
(l)i
On this account, the difference between qui and que reduces to the nature of the clitic. Que is also a result of Merge of a C and a D head, but this time the D head is the accusative clitic le ‘him/it’. As expected given the discussion of qui above, if C projects, the result is a complementizer and if D projects, the result is an interrogative pronoun, as shown in (26a–b). (26)
a. Merge C and D, project C
b.
Merge C and D, project D
C C qu–
D D
C
D
(l)e
qu–
(l)e
A somewhat different argument for the existence of labels comes from the proposals that distinguish syntactic objects by reference to their labels. By different labels, I do not mean a DP versus a VP, but a general “architecture” of a label. On this view, some elements have simple labels (in the sense illustrated in (27a) or (27b) below), whereas others have more complex ones, consisting of a set or an ordered pair, as shown in (27c) and (27d) respectively.13 (27)
α
a.
α
β
b.
β
α
c.
β
{α,β} α
β
d.
α
β
Chomsky (1995) proposes that such complex labels are necessary in order to distinguish arguments from adjuncts. X-bar theory was able to capture this distinction in a very straightforward manner; adjuncts were defined as elements whose mothers and sisters were identical. In (28a) and (28b), YP is an adjunct (of XP and X′, respectively), and in (28c), Y is an adjunct of X.
5.2 Arguments for (and against) eliminating labels (28)
a.
b.
XP YP
X
ZP
XP X
c.
XP
XP X
ZP
X
YP
WP
X
WP
WP
X
173
Y X
In Bare Phrase Structure Theory, however, adjuncts become indistinguishable from specifiers. Consider the bare phrase structure counterparts of (28a–c), given in (29a–c) below. In (29a) and (29b), it is impossible to tell whether y is a specifier or an adjunct of x, and in (29c) it is impossible to tell whether y is adjoined to x or is its leftward complement.14 (29)
a.
b.
x
z
x
y x
c.
x x
w
x z
y
x
x x
x x
w
y x
To solve this problem (and thus capture the differences between arguments and adjuncts in structural terms), Chomsky (1995) proposes that adjunction structures have different labels from complementation structures. When an adjunct is introduced into the structure, the label of the resulting structure is an ordered pair, represented in set notation in (30a) and in tree notation in (30b).15 (30)
a. L = { < H(K), H(K)>, {α, K}} b. α
(Chomsky 1995:248)
K
This reflects the intuition behind what is traditionally referred to as Chomsky-adjunction; the process in which one node is essentially “doubled” by the structure-building operation. Given this notation, the three types of adjunction structures under consideration here (i.e. adjunction to a maximal projection, adjunction to an intermediate projection, and adjunction to a head) can be represented as in (31a–c). (31)
a. <x,x> y
b. z
x x
w
c.
x
<x,x>
x x
x
z
<x,y> y
x
w
y x
w
174
Symmetric labels
Chomsky (2004), however, departs from this view of adjunction and proposes instead that the label of an adjunction structure is an ordered pair consisting of the adjunct and the element being adjoined to, rather than two segments of the latter. This is illustrated in (32a–c).). (32)
a.
b.
<x,y> y
z
x x
c.
x
w
x
z
<x,y>
<x,y>
x
y
x
x
w
w
y x
The earlier version (i.e. the one given in (31a–c)) captures more closely the spirit of Chomsky-adjunction (namely, the idea that adjunction splits a single node into two). An interesting question that arises for both views is whether the difference between complements and adjuncts concerns only labels, or whether there is also something fundamentally different about the way adjuncts and arguments are introduced into the structure. Chomsky (2004) takes the latter view, and assumes that adjuncts and arguments are introduced into the structure via a different kind of Merge. The two types of Merge he proposes are Set Merge and Pair Merge. Set Merge is the familiar kind of Merge; it takes two objects, combines them into one, and projects one of them as the label. Pair Merge takes two objects, combines them into one, and projects both of them as the label (in the form of an ordered pair). (33)
a. Set Merge
b. Pair Merge
α
α
β
α
β
A slightly different way of looking at the distinction between arguments and adjuncts, which also relies on the existence of labels in the grammar, comes from the work of Hornstein and Nunes (2008). It combines aspects of the two ideas we just looked at: the idea that there are some label-less constituents in the syntax, and the idea that the labels of adjuncts differ from the labels of complements or specifiers. More specifically, they argue that adjunction structures have no labels, as shown in (34), where y is an adjunct of x. (34)
y
x x
w
5.2 Arguments for (and against) eliminating labels
175
What is interesting about their proposal (and what distinguishes it from Chomsky’s) is the idea that adjuncts are introduced into the structure via a simpler mechanism than arguments. Adjunction is a simple concatenation operation (Merge without a label). This, they argue, can capture a number of properties of adjunction, the most fundamental one being the fact that adding an adjunct does not change the status of the element it is adjoined to. Since in (34) neither x nor y projects, the X-bar theoretic status of neither of them changes. Choosing between these three slightly different views of adjunction is not crucial for our purposes in this chapter; what is crucial is the conclusion that emerges from the discussion above, which is that we need labels in order to distinguish arguments from adjuncts in structural terms. This conclusion holds irrespective of whether adjuncts have no labels at all or have an ordered pair as the label. Let me conclude this section by noting that even though, on Chomsky’s view, adjuncts have complex labels (consisting of an ordered pair), they do not have symmetric labels. Since the two elements contributing to the label form an ordered pair, one is in a sense subordinate to the other. This can capture the well-documented asymmetries between adjuncts and nonadjuncts, such as the fact that extraction from adjuncts (unlike from complements) is banned, as shown by the following contrast. (35)
a. * Whati did they hire Bill because they needed more help with ti? b. Whati did they tell Bill that he needed to finish soon ti ?
One final conceptual argument in favor of the existence of labels comes from Hornstein (2009), who suggests that “labeling, understood as it is within Bare Phrase Structure, supplies the necessary ingredient to get one from a flat beads-on-a-string system to a hierarchical nesting system” (Hornstein 2009:55). Paraphrasing slightly, labels make syntactic objects visible for further concatenation processes and are thus necessary in order to build hierarchical structure. Merge for Hornstein thus consists of two distinct operations: Concatenate and Label. He further hypothesizes that it is Labeling (rather than concatenation) that is unique to human language (contra Hauser, Chomsky and Fitch 2002, for whom Merge itself is unique to the human language faculty).16
176 5.3
Symmetric labels Symmetric labels
The discussion in the previous section shows that the output of Merge cannot be a set consisting of two input elements, and that it also needs to include a label, marked as γ in (36). (36)
K = {γ, {α, β}}
As we saw in Chapter 2, Chomsky (1995) considers three ways to determine what γ is; it could be the intersection of α and β, the union of α and β, or one or the other of α and β. He excludes the first two on the following grounds: “the intersection of α and β will generally be irrelevant to output conditions, often null; and the union will be not only irrelevant but ‘contradictory’ if α, β differ in value for some feature,” which is often the case (Chomsky 1995:244). A hypothetical case, adapted from Pereltsvaig (2008), provides an illustration. It involves the Merge of α and β, where α and β have two features each (F and G), with conflicting specifications of these two features. (37)
a. α = {−F, +G} b. β = {+F, −G}
Without violating inclusiveness, the label of the new object (marked as γ in (38a–f)) could be any of the following combinations of these two features.17 (38)
a. b. c. d. e. f.
γ = {−F, +G} γ = {+F, −G} γ=0 γ = {+F, −G, −F, +G} γ = {+F, +G} γ ={-F, −G}
(α) (β) (intersection of α and β) (union of α and β)
In set theoretical terms, the set in (38a) is the complement of β (the members of α that are not in β) and the one in (38b) is the complement of α (the members of β that are not in α). These two sets are the result of either α or β projecting. (38c) is the intersection of α and β (which is empty in this case) and (39d) is the union of these two sets (which contains contradictory features). The sets given in (38e) and (38f) include one element from α and one from β. While these last two options do not violate inclusiveness (and do not result in contradiction), I will not explore them any further here. The question now is what F and G in (37) and (38) stand for. This is the fundamental question of what features labels contain. It is clear that
5.3 Symmetric labels
177
at the very least a label has to include information about categorial status of the constituent it labels. However, does it need to include anything else? In Bare Phrase Structure Theory, the simplest assumption is that one of the objects undergoing Merge simply becomes its label. This implies that a label includes all the features of this object. For example, a label of a noun will not only include features like person, number, gender (or class), but also purely semantic or purely phonological features. It is not clear, however, why labels would need to contain features that are never accessed in a syntactic derivation: features like +adult (a purely semantic feature) or +labial (a purely phonological feature). A more reasonable assumption is that labels need to contain only syntactically relevant features. These are categorial features and a subset of semantic features; features such as person, number and gender on nouns, and tense, aspect, voice, mood on verbs. Such features could be thought of as being parasitic on categorial features, as, for example, only nouns and pronouns are inherently marked for person, number, gender or class. Let us look at one concrete example, in which the verb see merges with the pronoun her. On the assumption that only categorial features and their dependents project when the two of them merge, the union contains contradictory features (both +N and −N, as well both +V and −V) and the intersection is null, as shown in (39a–b). (39)
a.
{+V, –N, –V, +N} see{+V, –N}
her{–V, +N}
b.
Ø
see{+V, –N}
her{–V, +N}
However, while it is true that the intersection of α and β is often null and that the union is often contradictory, it is by no means a logical necessity. If the two elements undergoing Merge have the same features, their union is not going to be contradictory and their intersection is not going to be null. This opens up a new theoretical possibility, explored by Baker and Stewart (1999), which is that symmetric labels are possible if the two elements undergoing Merge do not conflict in categorial features.18 If α and β with the same specifications of the two features F and G merge, the resulting label could be the union of the features of α and β (as shown in (40a)) or their intersection (as shown in (40b)). (40)
a.
γ{–F, +G, –F, +G} α{–F, +G}
β{–F, +G}
b.
γ{–F, +G} α{–F, +G}
β{–F, +G}
178
Symmetric labels
Both labels are symmetric since the features of α and β contribute in equal amounts to the identity of the label. In the rest of this chapter, I examine the empirical consequences of this theoretical possibility. I focus on the option illustrated in (40a), in which the label is the union of its constituents. Such symmetric labels are allowed irrespective of the phrase-theoretical status of the elements involved, as shown in (41a–c). By contrast, the labels in (42a–c) are excluded due to the lack of identity between Y and X. {X, X}
{XP, XP} XP
(41)
XP
a.
X
b.
a.
YP
XP
X
c.
*{Y, X} X
Y’
b.
X
X
*{Y, X}
*{XP, YP}
(42)
{X, X}
X
Y
c.
More specifically, I motivate the existence of the following symmetrically labeled constituents: {vP, vP}
{V, V}
(43)
a.
V
V
b.
vP
vP
{DP, DP}
{VoiP, VoiP}
c.
VoiP
VoiP
d.
DP
DP
{CP, CP}
e.
CP
CP
The ones in (43a–c) were proposed by Baker and Stewart (1999) to account for the properties of three distinct types of serial verb constructions.19 A variant of (43d) was proposed by Moro (2000) for (a subset of) Italian small clauses and by Pereltsvaig (2008) for a subset of Russian small clauses. And (43e) is the structure I suggested in Citko (2008b) for so-called comparative correlatives, which is the structure (and analysis) I expand on below. A natural question that arises for all the symmetric options under consideration here is what determines when symmetric labels are possible. If categorial identity were the determining factor, we would expect a symmetric label whenever two elements of the same category merge. However, this is not what we find. Sometimes the result is a symmetric label, and other times an asymmetric one. Since the answer to this question depends on the specifics of the symmetric construction involved, I address it over the next three sections, where I discuss symmetrically labeled constructions in more detail. 5.4
Serial verb constructions
Baker and Stewart (1999) propose a doubly headed structure for three types of serial verb constructions in Edo. In our terms, a doubly headed constituent is simply a symmetrically labeled one, in which each of the
5.4 Serial verb constructions
179
two heads contributes in equal amounts to the identity of the label.20 Baker and Stewart (1999) dub the three types, illustrated in (44a–c) below, covert coordination, a consequential serial verb construction and a resultative serial verb construction. Covert coordination involves two transitive verbs (each with its own object), a consequential serial verb construction involves two transitive verbs sharing an object (where sharing does not mean literal structure sharing or multidominance) and a resultative serial verb construction involves an unaccusative verb as the second verb in a series. (44)
a. Òzó ghá tòbórè lé evbàré rri órè. Ozo fut by.self cook food eat it ‘Ozo will cook the food by himself and eat it.’ b. Òzó ghá tòbórè lé èvbàré ré. Ozo fut by.self cook food eat ‘Ozo will by himself cook the food and eat it.’ c. Òzó suá Úyi dé. Ozo push Uyi fall ‘Ozo Pushed Uyi, causing him to fall.’
(Baker and Stewart 1999:10)
(Baker and Stewart 1999:13)
(Baker and Stewart 1999:17)
Furthermore, they argue that the three types differ with respect to which constituent is doubly headed (or symmetrically labeled in our terms). Thus, covert coordination involves a symmetrically labeled VoiceP (as shown in (45a)), a consequential serial verb construction involves a symmetrically labeled vP (as shown in (45b)), and a resultative verb construction involves a symmetrically labeled V (as shown in (45c)). (45)
a. covert coordination TP T
DP Ozoi
T
{VoiceP, VoiceP}
will VoiceP DP ti
VoiceP
Voice
DP
Voice vP v
ti
Voice vP
Voice v
VP
cook NP
V
food
e
VP
eat NP it
V e
(cf. Baker and Stewart 1999:11)
180
Symmetric labels b. consequential serial verb construction TP T
DP Ozoi T
VoiceP
will DP Voice ti Voice {vP, vP} vP
vP v
v
VP
cook DP
V eat DP
V
prok
e
e
foodk
VP
(cf. Baker and Stewart 1999:13) c. resultative serial verb construction TP T
DP Ozoi T
VoiceP
will DP ti
Voice Voice vP v CAUSE
VP DP {V, V} Uyi
V V push fall
(cf. Baker and Stewart 1999:18)
Their evidence in favor of analyzing covert coordination as a doubly headed VoiceP comes from the distribution of the reflexive element tòbórè. As shown in (46a–b) below, this reflexive does not have to be stringadjacent to the element it modifies. Baker and Stewart (1999) assimilate (46b) to quantifier stranding (or floating); the subject moves to [Spec,TP] and the reflexive stays in its base-generated position.
5.4 Serial verb constructions (46)
181
a. Òzó tòbórè ghá gié!gié lé èvbàré. Ozo by.self fut quickly cook food ‘Ozo by himself will quickly cook the food.’ b. Òzó ghá (gié!gié) tòbórè lé èvbàré. Ozo fut quickly by.self cook food ‘Ozo will quickly cook the food by himself.’ (Baker and Stewart 1999:10)
The fact that the reflexive can appear inside either verbal projection in a covert coordination suggests that there are two VoicePs. Since for Baker and Stewart subjects are specifiers of VoiceP, this yields two subject positions. (47)
a. Òzó ghá tòbórè lé èvbàré rrí órè. Ozo fut by.self cook food eat it ‘Ozo will cook the food by himself and eat it.’ b. Òzó lé ízè tòbórè rrì órè. Ozo cook rice by.self ate it ‘Ozo cooked rice and ate it by himself.’ (Baker and Stewart 1999:10)
The placement of the same reflexive element also provides evidence that consequential serial verb constructions involve symmetry at a lower level (vP as opposed to VoiceP). In serial verb constructions of this type, tòbórè can only occur inside the first verbal projection, which shows that the second verbal projection is smaller than VoiceP. (48)
a.
Òzó ghá tòbórè lé èvbàré ré. Ozo fut by.self cook food eat ‘Ozo will by himself cook the food and eat it.’ b. *Òzó lé èvbàré tòbórè ré. Ozo cooked food by.self ate ‘Ozo cooked some food and by himself ate it.’ (Baker and Stewart 1999:13)
And finally, the evidence that resultative serial verbs involve a doubly headed VP comes from the fact that only unaccusatives are possible as the second verb in this type of a serial verb construction. (49)
a. *Òzó suá Úyì só. Ozo push Uyi shout ‘Ozo pushed Uyi, causing him to shout.’ b. *Ékítà khú áhiánmwèn tín. dog chase bird fly ‘The dog chased a bird, causing it to fly.’ (Baker and Stewart 1999:17)
182
Symmetric labels
Baker and Stewart (1999) further argue that doubly headed (symmetrically labeled) phrases are possible if neither of the two heads in such a structure is attracted by some higher head, according to the condition stated in (50), which I will refer to as a ban on ambiguous Attract. (50)
X attracts a head Y iff Y can check a feature of X, and for all Z such that Z is not equal to Y and Z can check this feature of X, Y asymmetrically c-commands Z. (Baker and Stewart 1999:7)
To see how it works, consider the two schematic representations given in (51a–b), in which the subscripts P and G stand for Probe and Goal respectively. In a doubly headed structure, the two heads (such as YG and ZG in (51b)) are in a symmetric c-command relationship. According to (50), the Probe XP is not able to unambiguously attract either of them. Interestingly, the intuition behind (50) is the opposite of the intuition behind equidistance (which would allow either Y or Z to move in such a configuration). According to (50), however, X can attract Y in (51a) but not in (51b), due to the fact that Y and Z stand in a mutual c-command relationship. (51)
a.
b. XP
XP YG
YG
ZG
ZG
The ban on ambiguous Attract has the following implications, discussed by Baker and Stewart (1999); in covert coordinations, T cannot attract the head of VoiceP, in resultative serial verb constructions, the Voice head cannot attract the head of vP, and in resultative serial verb constructions, the v head cannot attract either of the two Vs. 5.5
Small clauses
Another argument for allowing symmetrically labeled constituents into the grammar comes from the domain of small clauses. I am not going to be able to do justice here to all the arguments that have been levied in favor of (and against) the idea that small clauses involve a symmetric structure.21 My point is merely to point out, following Moro (2000, 2007) and Pereltsvaig (2008), that certain types of small clauses might involve symmetric labels. Moro (2000, 2007) argues that Italian has two types of small clauses, which he dubs bare small clauses and rich small clauses, respectively. A bare
5.5 Small clauses
183
small clause is a small clause consisting of two DPs (one being the subject and the other one the predicate), as shown in (52a), whereas a rich small clause also includes a functional projection mediating the relationship between the subject and the predicate, as shown in (52b). For our purposes, the precise nature of the functional projection heading the small clause is immaterial; it could be a Pred head (as argued by Bailyn 2001, Bailyn and Citko 1999, Bowers 1993), a π head (as argued by Adger and Ramchand 2003 and Citko 2006, 2008a), or a Relator head (as argued by Den Dikken 2006). (52)
a.
b. DP
FP
DP DP F
F DP
For Moro, bare small clauses are headless and thus lack labels. This makes them another example of a symmetrically labeled constituent, as both parts of the small clause contribute in equal amounts to the identity of the label; namely, both contribute nothing. However, the arguments Moro gives in support of such a structure are in principle compatible with small clauses being symmetrically labeled, as shown in (53). (53)
a.
{DP, DP} DP
DP
The bulk of Moro’s arguments come from Italian. Italian is a canonical pro-drop language. This means, among other things, that in passives and unaccusatives either movement of an argument or insertion of pro can satisfy the EPP feature of T, as shown in (54a–b) and (55a–b) respectively. (54)
a. Giannii arriva ti. Gianni arrives ‘Gianni arrives.’ b. pro arriva Gianni. pro arrives Gianni ‘Gianni arrives’
(55)
a. Giannii è stato ucciso ti. Gianni was killed ‘Gianni was killed.’ b. pro è stato ucciso Gianni. pro was killed Gianni ‘Gianni was killed.’
(Moro 2007:1)
184
Symmetric labels
This is not, however, what happens in copular constructions. The EPP feature can only be satisfied via movement of the subject or the predicate DP, as shown in (56a–c). (56)
a.
Molte foto del muroi sono [ti la causa della rivolta] many pictures of the wall are the cause of the riot ‘Many pictures of the wall are the cause of the riot.’ b. La causa della rivolta sono [molte foto del muro ti] the cause of the riot are many pictures of the wall c. * pro sono [molte foto del muro la causa della rivolta] pro are many pictures of the wall the cause of the riot (Moro 2007:1–2)
Moro attributes the ungrammaticality of (56c) to the fact that the bracketed constituent lacks a head, which is what makes it unstable, to use his terminology, and requires one of the DPs to move out of the small clause. To rule out such unstable structures, he proposes the following condition: (57)
For probe P and goal G (within a phase Φ): a. * … P… G G … (Ambiguous Goal Violation) b. * … G … P … G … (Discontinuous Goal Violation) (cf. Moro 2007:8)
The reason why nominal small clauses violate (57a) is the fact that there are two Goals, two DPs constituting the small clause and a single Probe: the copula verb, which Moro takes to occupy a higher head position.22 This makes the structure unstable and forces one of the two DPs to move.23 The idea that nominal small clauses have symmetric labels is also defended by Pereltsvaig (2008), who follows Baker and Stewart (1999) and assumes that this is possible if there is no feature conflict between the two elements contributing to the label. Pereltsvaig’s empirical focus is on Russian copular sentences in which both the subject and the predicate are DPs marked with nominative case, such as the ones in (58a–b).24 (58)
a. Vysotskij byl Gamlet. Vysotsky.NOM was Hamlet.NOM ‘Vysotsky was Hamlet.’ b. Gamlet byl Vysotskij. Hamlet.NOM was Vysotsky.NOM ‘Vysotsky was Hamlet.’ (Pereltsvaig 2008:52)
The properties of Russian copular constructions also show that the feature match does not have to be complete for a symmetric label to be possible.
5.6 Comparative correlatives
185
A full match would only be the case if two identical DPs were merged; in (59a–b) the two DPs differ (at least) with respect to their phonological features; one is Vysotskij and the other one Gamlet. It is also not the case that all their syntactic features have to match. This is made obvious by the examples in (59a–b), in which the gender features of the two DPs do not match; (59a) involves the noun sekretar ‘secretary’, which lacks a distinct feminine form, and (59b) involves the noun nerjaxa ‘sloven’, which only has a feminine form. (59)
a. Valentina Ivanova – sekretar’ gorkoma. Valentina Ivanova.FEM secretary.MASC city.commitee i vernyj tovarišč po partii. and loyal.MASC comrade.MASC at party ‘Valentina Ivanova is the secretary of the city committee and a loyal party comrade.’ (Pereltsvaig 2008:54) b. Etot mal’cik – takaja bol’šaja nerjaxa. this.MASC boy.MASC such.FEM big.FEM sloven. FEM ‘This boy is such a big sloven.’ (Pereltsvaig 2008:54, citing Comrie and Stone 1978:77)
In order to solve this problem (and maintain the intuition that all syntactically relevant features have to match), Pereltsvaig introduces a distinction between grammatical gender features of nouns and phi-features of DPs “housing” these nouns. Furthermore, she allows the two not to match. What this means is that in the examples in (59a–b) above, the phi-features of the two DPs can match even though the gender features of the nouns themselves do not. It is not clear, however, how this would extend to mismatches in other features or in other types of symmetrically labeled objects. Furthermore, since we have to allow for some feature mismatches anyway (namely, mismatches in phonological and semantic features), I do not adopt Pereltsvaig’s solution and rather, maintain the assumption that categorial feature identity is all that is required for symmetric labels to be possible.
5.6
Comparative correlatives
This discussion expands on the analysis in Citko (2008b), where I proposed that symmetric labels of the kind given in (60a) are involved in (at least a subset of) the so-called comparative correlatives, an example of which is given in (60b).25 The subscripts in all the diagrams in this section are purely for expository purposes and have no theoretical significance whatsoever.
186 (60)
Symmetric labels a.
{CP1,CP2} CP1
CP2
b. The more I read, the less I know.
The label in (60a) is different from a label of a CP adjunction structure, given in (61) below, which is an ordered pair of two segments of the matrix CP: (61)
CP1
CP2
The crucial argument in favor of a symmetrically labeled structure for comparative correlatives comes from the differences between standard and comparative correlatives with respect to wh-movement. In Citko (2008b), I contrasted the behavior of comparative correlatives and standard correlatives with respect to wh-movement in three languages that allow both: Polish, Hindi, and Hungarian. Here, I briefly summarize the results of that paper, extend the analysis to comparative correlatives in other languages (where we see different extraction patterns), and point out other (non movement-related) differences between comparative and standard correlatives on the one hand, and comparative correlatives and other types of adjunct clauses on the other. Let me start with a brief description of what correlatives are and how they have been previously analyzed. Correlativization is a relativization strategy in which the relative clause is adjoined to the matrix clause (rather than to the relativized constituent). Here I limit my attention to cases in which the relative clause is left-adjoined to the matrix clause, as these are widely believed to constitute prototypical examples of correlatives. A schematic representation is given in (62). The relative clause contains a relative pronoun or a wh-pronoun, and the matrix clause contains a corresponding nominal (typically a demonstrative pronoun), which is typically referred to as a correlative pronoun, a correlative phrase or simply a correlate. (62)
[ [rel clause … WH/REL … ] [matrix clause…DEM …] ]
There are two types of correlatives, which I henceforth refer to as standard and comparative correlatives. I illustrate this distinction with data from Polish (see Izvorski 1997 and Rudin 2006, among others, for examples from other Slavic languages). Standard correlatives are illustrated in (63a–c) and (64a–c). In all of them, the relative clause is in italics, and the correlative
5.6 Comparative correlatives
187
pronoun is in bold. The examples in (63a–c) are all proverbs; however, the ones in (64a–c) are not, which shows that correlativization in Polish is not an archaic phenomenon, limited to proverbs or idiomatic statements. (63)
a. Kto rano wstaje, temu Pan Bóg daje. who morning raises dem Mr God gives ‘The early bird gets the worm.’ b. Jak sobie pościelisz, tak się wyśpisz. how refl make.bed this.way refl sleep ‘You made your bed, now lie in it.’ c. Gdzie kucharek sześć, tam nie ma co jeść. where cooks six there not have what eat ‘Too many cooks spoil the broth.’
(64)
a. Kto pierwszy przybiegnie, ten dostanie nagrodę. who first runs this gets prize ‘The one who arrives first will get the prize.’ b. Jak mi każesz, tak to zrobię. how me order this.way this do ‘I will do it the way you tell me to.’ c. Gdzie chcesz, tam pojedź na wakacje. where want there go for holidays ‘Go where you want for vacation.’
Standard correlatives are typically analyzed as involving an adjunction structure of the kind given in (65) (see Bhatt 2003, Dayal 1996 and Srivastav 1991, among others, for arguments in favor of such a structure).26 (65)
IP IP
CPi DP
… Dem-XPi V-I
(cf. Bhatt 2003:495)
Comparative correlatives differ from standard ones in that the two clauses establish a comparison between two degrees, amounts or individuals. However, they resemble standard correlatives in many other respects. For example, in comparative correlatives the first clause also contains a relative or an interrogative pronoun, and the matrix clause a corresponding demonstrative pronoun, as shown in (66). (66)
Im
więcej się uśmiechasz, tym lepiej się czujesz. more refl smile, DEM.INSTR better refl feel ‘The more you smile, the better you feel.’ REL.INSTR
188
Symmetric labels
The elements introducing the two clauses in Polish comparative correlatives are specific to this construction. Modern Polish does not distinguish between relative and interrogative pronouns, but the im form that we see in comparative correlatives looks like a remnant of the Old Slavic j-series relative pronouns. Polish also allows comparative correlatives in which the first clause contains a regular wh-pronoun and the second one a corresponding demonstrative pronoun, as shown in (67). The wh-pronoun used in such cases is also instrumental. (67)
Czym więcej się uśmiechasz, tym lepiej się czujesz. what.INSTR more refl smile, dem.INSTR better refl feel ‘The more you smile, the better you feel.’
Furthermore, comparative correlatives are more common than standard ones from a crosslinguistic perspective. Some languages (such as Polish, Hungarian or Hindi) allow both types. However, there are many others that only allow comparative correlatives. (68a–c) below provide examples from English, Greek and Dutch (see Taylor 2006 for a larger sample). (68)
a. The more you smile, the happier you get. b. Oso pio poli zahari tros toso pio poles Greek as.much.as more sugar eat.2sg.pres that.much more thermides pernis. calories get.2sg.pres ‘The more sugar you eat, the more calories you get.’ (Kapetangianni and Taylor 2007:7) c. Hoe meer je leest, hoe minder je begrijpt. Dutch how more you read how less you understand ‘The more you read, the less you understand.’ (Den Dikken 2005:501)
Comparative correlatives are sometimes referred to in the literature as comparative conditionals, to reflect their conditional-like interpretation. This reflects the intuition that (69a) below can be quite naturally paraphrased as (69b).27 (69)
a. The more you read, the more you understand. b. If you read more, you will understand more.
Comparative correlatives are often assimilated to standard correlatives and assigned a similar kind of adjunction structure. This is the view taken by Beck (1997), Den Dikken (2005, 2009), Huber (2007), McCawley (1988, 1998), Taylor (2006), among many others, and illustrated in (70–72) with concrete examples.
5.6 Comparative correlatives (70)
189
a. Je schneller Hans rennt, umso schneller wird er müde. the faster Hans runs the faster gets he tired ‘Hans will get tired faster, the faster he runs.’ b. CP2 CP2
CP1 C
DegP Deg C
je
TP
schneller
C
DegPi Deg C
umso
Hans rennt
TP
schneller wird er müde
(Beck 1997:234) (71)
a. The more you read, the more you understand. b. CP2
CP1
CP2 C
DegPi QP
Deg C
TP
OP Q Deg AP the more
C
DegPi QP
Deg C
TP
you read ti OP Q Deg AP you understand ti the more
(cf. Den Dikken 2005:510) (72)
a. The more pizza you eat, the more you need to diet. b. CP2 CP1
CP2 C
C’ C the
FP F
DPi more pizza
C the
F
FP F
DegPi IP
you eat ti
more Ø F
IP
you need to diet ti
(Kapetangianni and Taylor 2007:13)
All three involve adjunction of one CP to another one, and in all three of them, there is A-bar movement internal to both CPs. The similarities
190
Symmetric labels
end here, however. Even though the focus of this chapter is on movement from correlative clauses (not on their internal structure), let me briefly comment on the most significant differences between the structures in (70–72) involving constituency, the status of the correlative element the and the type of movement that takes place inside each clause. For Taylor (2006), for example, the is a complementizer that does not form a constituent with the compared DP more pizza. For Den Dikken (2005), the is the head of a Degree Phrase forming a constituent with the compared element. For Beck (1997), the is a specifier of DegP.28 The three accounts also differ with respect to what kind of movement takes place. In Taylor’s structure, for example, the compared element moves to the specifier of a functional projection below CP. In Den Dikken’s structure, the DegP (containing the plus the compared phrase) moves to the specifier of CP. What matters for us is that in all three structures, the first clause is adjoined to the second one. Thus they all predict that comparative correlatives should exhibit the matrix/adjunct clause asymmetry with respect to wh-movement. Furthermore, since both standard and comparative correlatives involve the same type of adjunction structure, they should also exhibit the same behavior with respect to wh-movement. In the next two sections, I test these two predictions, starting with a discussion of whmovement from standard correlatives.
5.7
Wh-movement in correlative clauses
The adjunction structure makes correct predictions about wh-movement from standard correlatives; wh-movement is possible from the matrix clause but not from the adjunct clause. This is illustrated in (73a–c) for Polish, (74a–c) for Hindi, and (75a–c) for Hungarian. The “a” examples show extraction from the adjunct clause, the “b” examples show extraction from the matrix clause and the “c” examples show ATB-style extraction from both clauses simultaneously.29 (73)
a.
* Kogoi Jan myśli, że [kto lubi ti], [ten powinien przeczytać who Jan thinks that who likes dem should read Annę Kareninę]? Anna Karenina ‘Who does Jan think that the one that likes (him) should read Anna Karenina?’
5.7 Wh-movement in correlative clauses b.
c.
? Coi Jan myśli, że [kto lubi Tołstoja], [ten powinien what Jan thinks that who likes Tolstoy dem should przeczytać ti]? read ‘What does Jan think that the one who likes Tolstoy should read?’ * Coi Maria myśli, że [kto lubi ti], ten powinien przeczytać ti]? what Maria thinks that who likes dem should read ‘What does Maria think that the one who likes should read?’
(74)
hai ki [jis-ko ti a. ???/* Kaun-sii kitaabi Mary soch-tii which book Mary think-hab.f be.prs that rel-dat pasand] hai [us-ko Anna Karenina paRh-nii chaahiye] like is he-dat Anna Karenina read-inf should ‘Which book does Mary think that the one who likes (it) should read Anna Karenina?’ b. ? Kaun-sii kitaabi Mary soch-tii hai ki [jis-ko which book Mary think-hab.f is that rel-dat War and Peace pasand hai] [us-ko ti paRh-nii chaahiye] War and Peace like is he-dat read-inf should ‘Which book does Mary think that the one who likes War and Peace should read?’ c. * Kaun-sii kitaabi Mary soch-tii hai ki [jis-ko ti pasand which book Mary think-hab.f is that rel-dat like hai] [us-ko ti paRh-nii chaahiye] is he-dat read-inf should ‘Which book does Mary think that the one who likes (it) should read (it)?’
(75)
a.
b.
c.
191
* Kiketi hallott Mari, [hogy aki szeret ti], [annak el who heard Mari that rel-who likes, that-dat pv kell olvasni az Anna Kareninát]? need read-inf the Anna Karenina-acc ‘Who did Mary hear that the one who likes (him) should read Anna Karenina?’ ? Miketi hallott Mari, hogy [aki szereti Tolstojt], what heard Mari that rel-who likes Tolstoy-acc [annak el kell olvasnia ti]? that-datpv need read-inf-3sg ‘What did Mary hear that the one who likes Tolstoy needs to read?’ * Miketi hallott Mari, hogy [aki szeret ti] [annak el kell what heard Mari that rel.who likes that-datpv need olvasni ti]? read-inf ‘What did Mary hear that the one who likes should read?’ (Citko 2008b:919)
Note that all these examples involve long distance extraction; the entire correlative construction is embedded under a verb like hear or think. This
192
Symmetric labels
could be a confounding factor in a language like Polish, in which long distance wh-movement from non-subjunctive clauses is somewhat degraded for many speakers (as pointed out by Willim 1989, for example). However, if the depth of embedding were solely to blame, we would not expect any contrast between the ‘a’ and ‘b’ examples in (73–75) above. Furthermore, we would expect an improvement if the “extra” level of embedding is removed. This is not, however, what we find. The same matrix/adjunct contrast holds in (76a–b), which do not involve long distance movement.30 (76)
a. ? Coi kto lubi Tołstoja, ten czyta ti? what.acc who likes Tolstoy dem reads ‘What does the one that likes Tolstoy read?’ b. * Kogoi kto lubi ti , ten czyta Annę Kareninę? who.acc who.nom likes dem reads Anna Karenina Lit. ‘Who does the one that likes reads Anna Karenina?’
Having established that standard correlatives show the expected adjunct/ matrix clause asymmetry with respect to wh-movement, I turn now to comparative correlatives. The class of comparative correlatives I consider first differs from the standard correlatives considered above in that they do not show the expected asymmetry; either both clauses allow extraction or both disallow it. In Polish and Hindi, wh-movement is possible from either clause, as shown in (77–78).31 (77)
a. Kogoi myślisz, że [im lepiej poznasz ti ], [tym bardziej go whom think.2sg that the better know.2sg the more him polubisz]? like.2sg ‘Whom do you think that the more you get to know, the more you will like him?’ b. Kogoi myślisz, że [im lepiej go poznasz], [tym bardziej whom think.2sg that the better him know.2sg the better polubisz ti]? like.2sg ‘Who do you think that the more you get to know him, the more you will like?’ c. Kogoi myslisz, że [im lepiej poznasz ti], [tym bardziej who think.2sg that the better know.2sg the more polubisz ti]? like.2sg ‘Who do you think that the more you get to know, the more you will like?’
(78)
a. ? Kis-koi Mary soch-tii hai ki [tum ti jitnaa ? who-acc Mary think-hab.fem be.prs that you how.much
5.7 Wh-movement in correlative clauses
193
? zyaadaa jaanoge] [tum us-ko utnaa-hii zyaadaa pasand ? more know.fut you he-acc that.much-only more like ? karoge]? ? do.fut ?‘Who does Mary think that the more you get to know, the more you like him?’ b. ? Kis-koi Mary soch-tii hai ki [tum us-ko ? who-acc Mary think-hab.fem be.prs that you he-acc ? jitnaa zyaadaa jaanoge] [tum ti utnaa-hii zyaadaa ? how.much more know.fut you that.much-only more ? pasand karoge]? ? like do.fut ? ‘Whom does Mary think that the more you get to know him, the more you will like (him)?’ c. ? Kis-koi Mary soch-tii hai ki [tum ti jitnaa ? who-acc Mary think-hab.fem be.prs that you how.much ? zyaadaa jaanoge] [tum ti utnaa-hii zyaadaa pasand karoge] ? more know.fut you that.much-only more like do.fut ? ‘Who does Mary think that the more you get to know (him), the more you will like (him)?’ (Citko 2008b:920–1)
The Hungarian examples in (79a–c) show that extraction is impossible from either clause. However, the pattern changes if the verb inside the clause containing the extraction site is marked with an indefinite object agreement marker. In such cases, extraction from either clause becomes grammatical, as shown in (80a–b). (79)
a.
* Kiketi hallott Mari, hogy [minél jobban ismerted ti], who.pl.acc heard Mari that what.at better knew.2sg [annál jobban szeretted öket]? that.at better liked-2sg they-acc ‘Who did Mari hear that the better you knew (them), the better you liked (them)?’ b. ??? Kike ti hallott Mari, hogy [minél jobban ismerted who.pl.acc heard Mari that what.at better knew.2sg öket], [annál jobban szeretted ti]? them that.at better liked.2sg ‘Who did Mari hear that the better you knew (them), the better you liked (them)?’ c. * Kiketi hallott Mari, [hogy minél jobban ismerted ti], who.pl.acc heard Mari that what.at better knew.2sg [annál jobban szeretted ti] ? that.at better liked.2sg ‘Who did Mari hear that the better you knew (them), the better you liked (them)?’ (Citko 2008b:921)
194 (80)
Symmetric labels a.
? Kiketi hallott Mari, hogy [minél jobban ismerted öket], who.pl.acc heard Mari that what.at better knew.2sg them, [annál jobban szerett-él ti]? that.at better liked.2sg.indef.ob ‘Who did Mari hear that the better you knew (them), the better you liked (them)?’ b. (?) Kiketi hallott Mari, [hogy minél jobban ismerted ti], who.pl.acc heard Mari that what.at better knew.2sg [annál jobban szerett-él ti] ? that.at better liked.2sg.indef.ob ‘Who did Mari hear that the better you knew (them), the better you liked (them)?’ (Aniko Csirmaz, p.c.)
Since comparative correlatives have a wider crosslinguistic distribution than standard correlatives, we can also examine their behavior with respect to wh-movement in languages that do not allow standard correlatives. Let us start with English. Culicover and Jackendoff (1999) observe that extraction is possible from either the first or the second clause, and Den Dikken (2005) adds that ATB movement is possible as well.32 (81)
a. a person whoi [the more you meet ti], [the more you hate him] b. a person whoi [the more you meet him], [the more you hate ti] c. a person whoi [the more you meet ti], [the more you hate ti] (Den Dikken 2005:504)
Furthermore, (82a–b) show that the grammaticality of (81a–b) has nothing to do with the fact that one of the clauses contains a pronoun, which could be construed as a resumptive pronoun. In (82a–b) below, there is no pronoun inside the non-extracted clause, and the result is equally grammatical: (82)
a. This is the sort of problem whichi the sooner you solve ti, the more easily you’ll satisfy the folks up at corporate headquarters. b. The folks up at the corporate headquarters are the sort of people whoi the sooner you solve the problem, the more easily you’ll satisfy ti. (Culicover and Jackendoff 1999:564)
Culicover and Jackendoff (1999) assimilate wh-movement from comparative correlatives to movement involved in parasitic gap constructions. However, parasitic gap constructions do show the matrix/adjunct asymmetry; a gap inside the adjunct clause is only possible if there is movement from the matrix clause, as shown in (83a–b). (83)
a. Whati did John file ti without reading pg? b. * Whoi did John file the letter without listening to ti?
5.7 Wh-movement in correlative clauses
195
Thus if we were dealing with a parasitic gap phenomenon in comparative correlatives, we would expect movement from the matrix clause (as well as movement from both clauses simultaneously) to be possible, and movement from the adjunct clause to be impossible. This is not what we find; movement from the adjunct clause alone is possible. Culicover and Jackendoff (1999) recognize this issue and suggest that comparative correlatives are more like the parasitic gaps of the kind given in (84a–b) below, in which either gap is independently possible. To abstract away from the issue of which gap is real (and which one parasitic), they mark both as t. (84)
a. Whoi did you tell ti that you would pay a call on ti? b. Whoi did you give pictures of ti to friends of ti? (Culicover and Jackendoff 1999:564)
Such an explanation, however, does not extend to languages like Polish, in which such parasitic gaps are impossible and which behave similarly to English with respect to movement from comparative correlatives.33 (85)
a. * Komui powiedziałeś ti, że odwiedzisz ti? who.dat say.past.2sg that visit.2sg ‘Who did you tell that you will visit?’ b. * Kogoi dałeś zdjęcia ti znajomym ti ? who.gen give.past.2sg pictures friends.dat ‘Who did you give pictures of to friends of ?’
Symmetric behavior of comparative correlatives is not limited to the languages considered so far. Kapetangianni and Taylor (2007) make a similar point based on Greek data.34 (86)
a. Pia thematai nomizis oti oso pio poli diavasis ti which materials think.2sg that as.much.as more read.2sg toso pio kala tha grapsis stis eksetasis? that.much more well fut write.2sg to.the exams ‘Which materials do you think that the more you read, the better you will do on the exam?’ b. Pioni nomizis oti oso pio poli argis na paris who think.2sg that as.much.as more take.time.2sg subj get to didaktoriko su toso pio poli apogoitevis ti? the PhD your that.much more disappoint.2sg ‘Who do you think the longer you spend in your PhD, the more you will disappoint?’ (Kapetangianni and Taylor 2007:16)
Abeillé, Borsley and Espinal (2006) note a symmetric extraction pattern in French comparative correlatives, illustrated in (87a–c). (87)
a. * C’est un livre dont plus tu te souviens ti, plus it.is a book of.which more you you remember more tu l’apprécies. you it.like
196
Symmetric labels ‘It is a book which the more you remember the more you like.’ b. * C’est un livre dont plus tu le lis, plus tu te it.is a book of.which more you it read more you you souviens ti. remember ‘It is a book which the more you read the more you remember.’ c. C’est un livre OPi que plus tu lis ti, plus tu apprécies ti. it is a book that more you read more you like ‘It is a book that the more you read, the more you like.’ (Abeillé, Borsley and Espinal 2006:12)
They also discuss Spanish, noting that it has two types of comparative correlatives: one symmetric and one asymmetric. The examples given in (88a–c) illustrate the symmetric ones, whereas the examples in (89) are their asymmetric counterparts. (88)
a. * Dime a quién más conoces ti y menos te tell.me to whom more you.know and less you fías de él. trust of him ‘Tell me whom the more you know the less you trust him.’ b. * Dime de quiéni más lo conoces y menos te tell-me of whom more him you.know and less you fías ti. trust c. Este es un tipo de aceite del que más uno compra ti this.is a type of oil of which more one buys y más utilizati i en las ensaladas. and more uses in the salads ‘This is a type of oil which the more one buys, the more one uses in the salads.’ (Abeillé, Borsley and Espinal 2006:18)
(89)
a.
Dime de quiéni cuanto más lo conoces tell.me of whom how.much more him you.know menos te fías ti. less you trust ‘Tell me whom the more you know him, the less you trust.’ b. * Dime a quiéni cuanto más conoces ti menos tell.me to whom how.much more you.know less te fías de él. you trust of him ‘Tell me whom the more you know, the less you trust him.’ (Abeillé, Borsley and Espinal 2006:17)
At first glance, Dutch and German also behave symmetrically, as shown in (90a–c) for Dutch and in (91a–c) for German.35 (90)
a.
* Een man die [hoe vaker je ec ziet], [hoe/ des a man who how often-cpr you see how/the-gen
5.7 Wh-movement in correlative clauses
197
te meer je ’m gaat haten] te more you him go hate ‘a man who the more often you see (him), the more you get to hate (him)’ b. ?* Een man die [ hoe vaker je ’m ziet], [hoe/ des te a man who how often-cpr you him see how/the-gen te meer je ec gaat haten] more you go hate ‘a man who the more often you see (him), the more you get to hate (him)’ c. ?* Een man die [ hoe vaker je ec ziet], [hoe/ des te a man who how often-cpr you see how/the- gen te meer je ec gaat haten] more you go hate ‘a man who the more often you see (him), the more you get to hate (him)’ (91)
a.
* Ein Sänger, den, [ je öfter du ec hörst], [desto a singer who q often-cpr you hear the-gen-to mehr du ihn treffen möchtest] more you him meet would-like ‘a singer who the more often you hear (him), the more you would like to meet (him)’ b. ?* Ein Sänger, den, [ je öfter du ihn hörst], [desto a singer who q often-cpr you him hear the-gen-to mehr du ec treffen möchtest] more you meet would-like ‘a singer who the more often you hear (him), the more you would like to meet (him)’ c. ?* Ein Sänger, den, [ je öfter du ec hörst], [desto a singer who q often-cpr you hear the-gen-to mehr du ec treffen möchtest] more you meet would-like ‘a singer who the more often you hear (him), the more you would like to meet (him)’ (Den Dikken 2005:505)
Den Dikken (2005), however, argues against a symmetric treatment of Dutch and German comparative correlatives and points out that symmetric movement is only possible under very limited conditions, namely when the comparative element undergoes fronting in the main clause. If the comparative phrase inside the second clause is not fronted, an asymmetric pattern emerges instead.36 (92)
a.
* Een man die [ hoe vaker je ec ziet], [je ’m des te a man who how often-cpr you see you him the-gen te meer gaat haten] more go hate
198
Symmetric labels b. ?(?) Een man die [ hoe vaker je ’m ziet], [je ec des te meer a man who how often-cpr you him see you the-gen te more gaat haten] go hate c. % Een man die [ hoe vaker je ec ziet], [je ec des te meer a man who how often-cpr you see you the-gen te more gaat haten] go hate all: ‘a man who the more often you see (him), the more you get to hate (him)’
(93)
a.
* Ein Sänger, den, [ je öfter du ec hörst], [du ihn desto a singer who q often-cpr you hear you him the-gen-to mehr treffen möchtest] more meet would-like b. ?(?) Ein Sänger, den, [je öfter du ihn hörst], [du ec a singer who q often-cpr you him hear you desto mehr treffen möchtest] the- gen-to more meet would-like c. % Ein Sänger den, [ je öfter du ihn hörst], [du ec a singer who q often-cpr you him hear you desto mehr treffen möchtest] the-gen-to more meet would-like All: ‘a singer who the more often you hear (him), the more you would like to meet (him)’ (Den Dikken 2005:506)
In Table (94) summarizes the extraction patterns discussed so far.37 There are four patterns we have seen evidence of, three of which are symmetric. (94) Wh-movement patterns in comparative correlatives Type
Language
Extraction from CP1
Extraction from CP2
ATB extraction
I
Polish Hindi English Greek Hungarian
✓ ✓ ✓ ✓ ✓
✓ ✓ ✓ ✓ ✓
✓ ✓ ✓
II
Hungarian * German * Dutch *
* * *
* * *
III
German Dutch Spanish
* * *
✓ ✓ ✓
* *
IV
French Spanish
* *
* *
✓ ✓
5.7 Wh-movement in correlative clauses
199
Logically speaking, these are not the only possibilities. The four given in (95) below complete the picture. (95) Type
Extraction from CP1
Extraction from CP2
ATB extraction
V VI VII VIII
✓ * ✓ ✓
✓ ✓ * *
* * ✓ *
Let me discuss them in turn, starting with the ones listed in (95). We have not seen any evidence of these patterns, which raises the question of whether they are possible at all. If wh-movement from either CP1 or CP2 is possible independently, there are no syntactic reasons why movement from both CPs simultaneously should be impossible. This is what excludes Type V. Type VI, in which only movement from CP2 (the matrix clause) is allowed, might be what we find in a language that has asymmetric correlatives but no parasitic gaps (which would explain why movement from both clauses is impossible). In the last two types (VII and VIII), extraction from the adjunct clause (CP1) is possible but extraction from the matrix clause (CP2) is not. Needless to say, this would be unexpected on any assumptions concerning the differences between arguments and adjuncts. The upshot of the discussion in this section is that there exist comparative correlatives that allow symmetric wh-extraction (grammatical extraction from either the matrix or the relative CP). These are Types I, II, and IV of the table (94), which are the ones I focus on in the rest of this chapter. More specifically, I show how symmetric extraction follows from a symmetrically labeled structure, thus providing another argument in favor of the presence of symmetric labels in the syntax. However, crosslinguistic variation with respect to wh-movement suggests that there are two types of correlative clauses. In both of them, two CPs undergo Merge. The difference between them lies in whether the features of one of these two CPs project as the label, which yields a standard case of adjunction (even if the label is a two-segmented object, as is typically assumed), or both of them do (which yields a structure in which the two CPs behave in the same way), as shown in (96a) and (96b) respectively. The label in (96b) is symmetric because it contains the features of both constituent CPs.
200 (96)
Symmetric labels a. asymmetric correlatives
b.
symmetric correlatives
α
{α, β} β
α
β
Crucially, the distinction between symmetric and asymmetric correlatives does not correlate with the distinction between standard and comparative correlatives that we started this section with. We saw above that not all comparative correlatives allow symmetric wh-movement, which raises the question of whether the opposite is the case. Namely, do all standard correlatives show asymmetric behavior with respect to extraction? For all the languages considered in this chapter, the answer appears to be yes. The reason why there are no symmetric standard correlatives could also be linked to their structure. If the relative CP adjoins to a TP (rather than a CP), as shown in (97a), a symmetric label of the kind given in (97b) is impossible because the two constituents contributing to the label (TP and CP) are not of the same category.38 (97)
a.
CP
TP
b.
{CP, TP} CP
TP
Another question that arises here, brought to my attention by one of the reviewers, is whether the relative clause in symmetric correlatives is related to the main clause paratactically or hypotactically, and, more generally, whether the difference between parataxis and hypotaxis could be reduced to the difference in labeling along the symmetric/asymmetric dimension. The crosslinguistic data presented in this section are consistent with the assumption that correlatives form a continuum on the parataxis/hypotaxis grammaticalization cline. While neither of the two types considered here is fully paratactic, the symmetric ones are closer to the paratactic end of the spectrum than the asymmetric ones. Given that the other two symmetrically labeled constituents considered in this chapter, serial verbs and nominal small clauses, are not paratactic in any straightforward sense of the term, the answer to the general question of whether symmetric Labeling is a structural way to represent parataxis seems to be no. The fact that some languages allow symmetric extraction does not mean that there are no asymmetries between the two clauses whatsoever. As pointed out by Culicover and Jackendoff (1999), the second clause in comparative correlatives behaves like a matrix clause with respect to diagnostics
5.7 Wh-movement in correlative clauses
201
such as subjunctive mood selection or tag question formation. The relevant examples are given below. (98)
a. I demand that John pay more, the more he eats. b. * I demand that John pays more, the more he eats.
(99)
a. You get angrier, the more we eat, don’t you? b. * You get angrier, the more we eat, don’t we? (Culicover and Jackendoff 1999:549)
The fact that comparative correlatives behave symmetrically with respect to some diagnostics and asymmetrically with respect to others is somewhat of an issue for any account. There are two possible ways to address this issue. One is to assume that comparative correlatives behave symmetrically with respect to Narrow Syntax processes (such as wh-movement) but asymmetrically with respect to interface processes such as interpretation or morphological agreement. Another one is to assume the existence of a process that converts symmetrically labeled structures to more standard asymmetric ones.39 With this as background, we can now turn to the account of different extraction patterns in comparative correlatives. The most symmetric ones, so to speak, are Type I correlatives. This is what we find in languages like Polish, Hindi and English, in which wh-movement from either clause is possible, as is movement from both clauses simultaneously. Culicover and Jackendoff (1999) attribute symmetric extraction from comparative correlatives to their quasi-coordinate syntax, arguing that comparative correlatives provide an example of a syntax–semantics mismatch: a construction with coordinate syntax but conditional semantics. However, the extraction patterns that we see in comparative correlatives are different from the extraction patterns that we see in standard coordinate structures. If comparative correlatives were coordinate, we would expect movement from one clause to be impossible (as a violation of the Coordinate Structure Constraint), and movement from both clauses to be possible (as an instance of ATB movement). This is what we find in run-of-the-mill coordinate structures:40 (100)
a. * Whati did John write ti and Mary review an article? b. * Whati did John write a book and Mary review ti? c. Whati did John write ti and Mary review ti?
Interestingly, this is what we see in Type III comparative correlatives, exemplified by Spanish and French, which suggests that in these languages
202
Symmetric labels
comparative correlatives might really be quasi-coordinate. The fact that they allow an overt conjunction between the matrix and the relative clause, as shown in (101) below for French and in (88a–c) above for Spanish, adds plausibility to this view. (101)
Plus je lis et plus je comprends. more I read and more I understand ‘The more I read, the more I understand.’ (Abeillé and Borsley 2008:1153)
This, however, cannot be the right analysis for comparative correlatives of Type I, as they do not allow overt conjunctions, as shown in (102a–b) for Polish and English. (102)
a. * Im więcej czytam i tym więcej rozumiem. rel.instr more read.1sg and dem.instr more understand.1sg ‘The more I read, the more I understand.’ b. * The more I read and the more I understand.
Type II correlatives, in which extraction is impossible from either clause (as well as from both clauses simultaneously), also cannot be treated as coordinate. If comparative correlatives belonging of this type were coordinate, we would expect ATB-style movement from both clauses simultaneously to be possible. We can, however, explain the behavior of Type I comparative correlatives by resorting to the by now familiar idea that such symmetric correlatives have symmetric labels. More specifically, their labels contain the features of both the relative and the matrix CP, as shown below: (103)
Merge CP1 and CP2 {CP1, CP2} CP 1
CP2
Independent support in favor of such a structure comes from the fact, discussed by Borsley (2004), that both CPs are obligatory, as shown in (104a–b) for English and in (105a–b) for Polish. (104)
a. The more I read, the less I know. b. * The less I know.
(105)
a.
Im więcej czytam, tym mniej wiem. rel.instr more I.read this.instr less I.know ‘The more I read, the less I know.’
5.7 Wh-movement in correlative clauses
203
b. * Tym mniej wiem. this.instr less I.know ‘The less I know.’
If the first clause were an adjunct, we would expect it to be optional. This is what we see in standard correlatives. The Polish examples in (106a–c) show that the bracketed CP can be omitted without affecting the grammaticality of the sentence. (106)
a. (Kto pierwszy przybiegnie), ten dostanie nagrodę. who first runs this gets prize ‘The one who arrives first will get the prize.’ b. (Jak mi pokazałaś), tak to zrobię. how me showed this.way this do ‘I will do it the way you showed me.’ c. (Gdzie chcesz), tam pojedź na wakacje. where want there go for holidays ‘Go where you want for vacation.’
Furthermore, as also pointed out by Borsley (2004), the first CP is incompatible with other types of matrix clauses, which points towards a tighter relationship between the two CPs in comparative correlatives. (107)
a. *The more books I read, I know the answer. b. *Im wiecej książek czytam, znam odpowiedź. rel.instr more books read.1sg know.1sg answer (Borsley 2004:70)
The crucial question now is how symmetric labels make symmetric whmovement possible. To answer this question, I adopt Rackowski and Richards’ (2005) proposal that the constituent from which wh-movement takes place has to (independently) Agree with the head triggering movement. It is this Agree relationship that makes it transparent for extraction, which is how Rackowski and Richards derive the effects of the Adjunct Condition.41 Let us first see how their proposal works in simple cases of extraction from adjunct versus complement clauses, before extending it to correlative clauses. Example (108a) involves grammatical wh-movement from a complement clause. The crucial point in its derivation is schematized in (108b). The matrix v undergoes Agree, indicated by a dotted line, with the closest potentially movable phase containing the wh-phrase, which in this case is the embedded CP. The CP and the wh-phrase in its specifier are equidistant from the higher Probe (the matrix v), which is what allows the wh-phrase in the specifier of CP to legitimately move from the embedded CP. If the entire CP moves instead, the result is clausal pied-piping of the kind found in Basque.42
204 (108)
Symmetric labels a. [CP Whoi do you [vP ti think [CP ti that we should [vP ti hire ti ] ] ] ] ? b. CP C C v Agree (v, CP)
CP C
whoi C
TP … ti …
The derivation of the ungrammatical examples involving extraction from adjunct clauses (such as the one in (109a) below) differs in that the wh-pronoun is not in the specifier of the CP which the matrix v undergoes Agree with. Agree in this case is between v and the embedded CP (marked as CP1 in (109b)). However, the wh-pronoun is “buried” inside the adjunct CP, marked as CP2. Since v undergoes Agree with CP1 (not CP2), movement from CP2 is blocked. (109)
a. * [CP Whoi do you [vP ti think [CP that [CP ti after we [vP hire ti] ] we will [vP
have to move to a new office] ] ] ]? b. CP C C v CP1 C
Agree (v, CP1)
CP2 whoi C
TP
C TP … ti …
5.7 Wh-movement in correlative clauses
205
Rackowski and Richards’ Agree-based version of the Adjunct Condition outlined above can be extended quite straightforwardly to correlative clauses. We saw that (at least some) comparative correlatives differ from standard comparatives in that they allow extraction from either CP. I also argued that such correlatives have symmetric labels, which consist of the features of both the matrix and the relative CP. This means that in some correlatives, v is going to Agree with a symmetrically labeled CP. In both symmetric and asymmetric correlatives, the matrix v will undergo Agree with the closest phase containing a wh-phrase, the CP phase. Extraction from asymmetric correlatives (which includes all standard correlatives and a subset of comparative ones) proceeds similarly to extraction from adjunct clauses illustrated in (109b) above. The matrix v undergoes Agree with the embedded CP (marked as CP2 in (110a–b)), which makes this CP transparent for extraction; WH2 can undergo further movement, but WH1 cannot (due the lack of Agree between v and CP1).43, 44 (110)
a. movement from CP1
b.
CP
movement from CP2 CP
C
C
C
C vP
v
vP
VP
v
V CP1 WH1 C
V
CP2
CP1
C1 WH2 C2 TP C … t1 …
VP
TP … t2 …
WH1 C
CP2 C1 WH2 C2 TP … t1 …
C
TP … t2 …
By contrast, in symmetric correlatives, the matrix v undergoes Agree with a symmetrically labeled constituent, consisting of both CP1 and CP2. This means that movement of either WH1 or WH2 should be possible, as shown in (111a–b) respectively.45 And this is indeed what we find.
206 (111)
Symmetric labels a. movement from CP1
b. movement from CP2
CP
CP C
C
C
C vP
v
vP
VP
v
V {CP1,CP2} CP1 WH1 C
V {CP1,CP2} CP2
TP C
CP2
CP1
C1 WH2 C2
… t1 …
5.8
VP
TP … t2 …
WH1 C
C1 WH2 C2 TP … t1 …
C
TP … t2 …
Conclusion
To conclude briefly, I established two points in this chapter. First, I showed that labels are necessary, as many syntactic principles and operations that make reference to labels could not be easily stated in a system without labels. Second, I argued that labels can be symmetric, thus adding to the evidence in favor of the symmetric nature of core syntactic operations presented in Chapters 3 and 4. More specifically, I argued that a label of a complex syntactic object is symmetric if it includes features of both of its constituent parts. This departs from the standard minimalist view, which is that a label is asymmetric in that it is determined by only one part. To support the existence of such symmetric labels in the grammar, I showed that they can account for a number of properties of certain types of serial verb constructions, nominal small clauses and comparative correlatives.
6
Conclusion
My goal in this book was to make a case for symmetry in syntax, and to show that syntactic structures and processes do not necessarily have to be asymmetric or antisymmetric. In the process, I hope to have added to our understanding of the structure and properties of the following constructions: across-the-board wh-questions, right node raising, gapping, free relatives (standard and comparative alike), serial verbs, copular clauses, double object constructions, standard correlatives and comparative correlatives. On a more theoretical level, I hope to have contributed to our understanding of the three basic mechanisms in minimalist syntax (i.e. Merge, Move and Labeling) and, more generally, to our understanding of the combinatorial properties underlying the human language faculty. The three questions that I started with are given in (1). (1)
a. Does symmetric Merge exist? b. Does symmetric Move exist? c. Do symmetric labels exist?
In the course of this book I argued that the answer to all three questions is yes. I provided both theoretical and empirical arguments to support this claim. After a brief introduction to the concepts of symmetry, asymmetry and antisymmetry in Chapter 1, in Chapter 2 I reviewed both theoretical and empirical reasons not to expect any symmetry in the grammar. We expect Merge to be asymmetric because of the Linear Correspondence Axiom (which can only linearize asymmetric structures); we expect Move to be asymmetric because of principles like Attract Closest or the Minimal Link Condition (which privilege a higher element with respect to movement); and we expect labels to be asymmetric because of the interface conditions (which, as pointed out by Boeckx 2008, require labels to be unambiguous and non-contradictory). In Chapters 3, 4 and 5, I challenged these views and provided arguments in favor of the existence of: symmetric Merge structures, in which a single element undergoes Parallel Merge with two other elements (as shown in (2)); symmetric Move operations, in 207
208
Conclusion
which either of the two potentially movable elements can move (as shown in (3a–b)); and symmetric labels, which are determined by both merged elements (as shown in (4)). (2)
Symmetric Merge L
K α
(3)
γ
β
Symmetric Move a.
b. γuF
γuF
(4)
αF
βF
αF
βF
Symmetric labels {α, β} α
β
The main empirical argument for symmetric Merge came from the fact that it can account for a number of otherwise quite mysterious properties of a number of distinct constructions. The ones I focused on in Chapter 3 were: across-the-board wh-questions, wh-questions with coordinated whpronouns, free relatives, serial verb constructions, gapping and right node raising. I showed that in all of them, there is at least one element that is involved in Parallel Merge structure. A more general question that the discussion in this chapter raised was whether we can establish a general diagnostic of a Parallel Merge structure. A possibility I considered briefly is that coordination is a hallmark of symmetric Parallel Merge. This, however, cannot be true, for the simple reason that not all Parallel Merge structures discussed in this chapter were coordinate. I also argued that free relatives (standard and transparent alike) involve Parallel Merge, as do serial verb constructions. Neither of them is coordinate in any straightforward sense. Even though certain types of serial verb constructions might appear to be coordinate (with a null conjunction mediating the relationship between the two verbs in a sequence), there are solid arguments against treating them as such. Likewise, the relationship between the matrix and the relative clause in a free relative clause is that of subordination not coordination. Thus, the correlation between Parallel Merge and coordination breaks down in both directions. This
Conclusion
209
suggests that interpretation (a sense that the shared element is interpreted in two places simultaneously without being moved from one to the other) and morphosyntactic properties (a sense that a single element has to simultaneously satisfy the requirements imposed on it from two distinct positions in the structure) might be more reliable as diagnostics of a Parallel Merge structure. The main empirical evidence in favor of symmetric Move came from the domain of passive movement in double object constructions. In Chapter 4, we saw that some languages allow symmetric passivization, i.e. passivization of either object in a double object construction. We also saw evidence of crosslinguistic symmetry of sorts; in some languages only the higher object of the two can passivize, whereas in others only the lower one can. I argued that we need both locality- and case-based approaches to account for this crosslinguistic variation. However, we also saw reasons not to reduce the pattern in which the indirect object is marked with dative case and only the direct object can passivize to the non-structural nature of dative case. The crucial evidence came from Polish, in which some objects marked with non-structural cases do passivize but dative objects do not. To account for this asymmetric pattern, I proposed that there is a structural component to dative case, manifested through the presence of an uninterpretable case feature, in addition to the lexical component, which is manifested through the presence of an interpretable case feature. The uninterpretable case feature is checked/valued via movement to the specifier of a light applicative head. This is a case-checking position, which makes the element occupying it frozen for further case-related movements. A natural question to ask here is whether symmetric passivization considered in this chapter is the only place where symmetric movement arises in the grammar, or more generally, whether this is the only way movement could be symmetric. It would be rather surprising if the answer were “yes.” We could easily imagine analogous symmetry within the domain of A-bar movement, involving wh-movement, scope, or focus-driven movements, some of which I alluded to in Chapter 4. The last domain of symmetry I examined in this book involved labels, in particular the possibility that a label of a complex object can sometimes be constructed “symmetrically” from the labels of both of its parts. The result is a symmetric label of the kind given in (4). The main empirical argument in favor of such symmetric labels came from the domain of comparative correlatives, which (at least in some languages) allow symmetric wh-movement; movement is allowed from either the matrix or the relative clause (or
210
Conclusion
both clauses simultaneously), or it is disallowed from either clause. The two other candidates for symmetrically labeled constituents that I discussed in Chapter 5 involved serial verb constructions and nominal small clauses. Again, I think it would be very surprising to find that symmetric labels are restricted to these three constructions. However, my goal in this chapter was not to give a complete and definitive theory of symmetrically labeled syntactic objects, but rather to present a couple of case studies and to show that the existence of such objects is both theoretically feasible and empirically attested. In short, the general claim I argued for in this book is that symmetry exists within the domain of the three fundamental syntactic operations: Merge, Move, and Labeling. This claim has many interesting implications beyond the ones discussed in the book, and it may raise more questions than it answers, but I hope it will open new research avenues and inspire others to investigate other sources of symmetry within syntax. A general question to ask in the context of the discussion in this book is whether the existence of symmetric structures, symmetric operations and symmetric labels simplifies syntax or complicates it. One might be tempted to think that it is a complication, because it introduces new mechanisms into the grammar, such as a new type of Merge (Parallel Merge) or new types of labels (complex labels involving unordered sets or ordered pairs). I believe something quite opposite is the case. The grammar with Parallel Merge and symmetric labels is simpler than the grammar without them, as it does not require any specific constraints to exclude them. Such a grammar is thus more plausible from the perspective of the Strong Minimalist Thesis, which strives for an optimal system with minimal constraints beyond the ones imposed by the interfaces. This brings up another issue: the issue of constraints on symmetry. If there is nothing that excludes symmetry on principled grounds, why do we not find it more often? I showed in this book that symmetry is in fact more pervasive than it appears on the surface, as we see its effects in Merge, Move and Labeling. It is nevertheless true that the effects of symmetry are often obscured by independent factors. For example, most constructions involving Parallel Merge considered here involve movement of the shared element out of the Parallel Merge structure. A related question concerns the origin of symmetry and asymmetry in the grammar. We know that the output of syntactic operations is subject to the conditions imposed by the interfaces. The interface conditions are the familiar conditions governing interpretation and pronunciation, which are typically assumed to require (and rely on) asymmetric relationships.
Conclusion
211
For example, functional application is an asymmetric operation and linear ordering is a direct reflection of asymmetric relations between nodes in a tree. However, I believe it would be wrong to conclude from these interface conditions that the operations of Narrow Syntax must be strictly asymmetric. What seems much more plausible is something akin to the conclusion reached by Boeckx (2009), which is that conditions requiring asymmetry are imposed by the interfaces not by the narrow computational system itself. My conclusion, however, is slightly different. For Boeckx, “narrow syntax should be formulated in a symmetric fashion and all the sources of asymmetries should be relegated to the point of cyclic transfer” (Boeckx 2009:41). What I argued for here is not that that the operations of Narrow Syntax have to be symmetric but that they do not have to be asymmetric. This strikes me as plausible from the perspective of the Strong Minimalist Thesis, as it does not require any extra conditions to rule out asymmetric structures or operations. It simply adds a new perspective to our syntactic theorizing, by allowing both symmetric and asymmetric operations.
Notes to chapters 000–000
1
Rationale
1 The definitions and examples in this section are based on Partee, ter Meulen and Wall (1993). 2 For expository purposes, I will refer to the minimalist theory before the publication of “Derivation by phase” as early minimalism. 3 Readers familiar with minimalism will feel comfortable skipping to Section 1.4. Readers less familiar with minimalism might benefit from a more general and comprehensive introduction, such as Adger (2003) or Boeckx (2008). For readers familiar with the Government and Binding framework, Hornstein, Nunes and Grohmann (2005) offers a good transition to minimalism. 4 We should keep in mind that the idea of optimality in the grammar is by no means new (or unique) to the minimalist program. Consider, for example, Culicover and Jackendoff’s (2005) Simpler Syntax Hypothesis, given in (i) below: (i) The most explanatory syntactic theory is one that imputes the minimum structure necessary to mediate between phonology and meaning. (Culicover and Jackendoff 2005:5) There is, however, an important difference between the Strong Minimalist Thesis and the Simpler Syntax Hypothesis. The latter might be viewed as an instance of what Martin and Uriagereka (2000) call methodological minimalism, which amounts to Occam’s Razor and is certainly not limited to minimalist approaches to language. The more relevant kind of minimalism is what they refer to as ontological minimalism, which emphasizes optimality in language design itself, not just in the theory of language. 5 The term Multiple Spell-Out is somewhat unfortunate, as Spell-Out, technically speaking, designates only the point of transfer to the interface with the sensorimotor system. A more accurate term would be Multiple Transfer. This is not a substantive issue, as transfer to both interfaces is typically assumed to happen at the same time, which means that Spell-Out and Transfer denote the same points in the derivation. However, this assumption has been recently challenged (see Marušič 2009 for arguments that Transfer to the two interfaces can happen at different points in the derivation). 212
Notes to pages 6–12
213
6 The issue of what phase heads are is far from being settled in syntactic theory, and the list of potential phase heads seems to be growing. For example, for Chomsky only transitive (and unergative) v heads are phase heads. Legate (2003), however, argues that unaccusative and passive v heads are phase heads as well. 7 Studies show that other species are capable of acquiring recursive structures of a less complex kind than the kinds involved in human language. One possibility is that other animals can acquire strings generated by finite state grammars but not those generated by phrase structure grammars. See Saffran et al. (2008) and Newport et al. (2004), among others, for a discussion of the differences between the kinds of grammars tamarin monkeys and human infants can handle. 8 This is a somewhat idealized view of how Internal Merge works. There are good arguments (see Abels 2003, Grohmann 2003, among others) that in reality the kind of movement schematized here, i.e. movement from the complement to the specifier of the same head, is not allowed. Since β, being α’s complement, is already in a very local relationship to α in (10), moving it to the specifier of α is redundant, and thus banned by economy. 9 Baker’s (1985) Mirror Principle, given in (i), could also be viewed as an example of a symmetric mapping between the order of morphological affixes and the order of functional heads: (i) Morphological derivations must directly reflect syntactic derivations (and vice versa). (Baker 1985:375) 10 Fox and Nissenbaum’s (1999) proposal does not totally eliminate the need for rightward movement. The object undergoes covert Quantifier Raising to the right and the “extraposed” PP adjoins to it countercyclically. This is only possible in a “single cycle” architecture where overt and covert movements can interleave. 11 Perhaps an even stronger case for symmetry would come from movement in which the same element can undergo either a raising or a lowering operation. 12 Another potential example of lowering in the syntax, due to Bošković and Takahashi (1998), comes from Japanese long distance scrambling. For them, the scrambled element such as sono hon-o ‘that book’ in (i) enters the derivation in a position in which it is pronounced (the IP adjoined position) and undergoes lowering at LF to its theta position, as shown in (ii). This is what accounts for radical reconstruction properties of scrambled elements; the fact that they are interpreted in their “unscrambled” positions. It is not clear, however, how their proposal translates into the Multiple Spell-Out architecture assumed here. (i) [IP Sono hon-o [IP John-ga [CP[IP Mary-ga [VP[V katta] ] ] to] omotteiru] ] that book-acc John-nom Mary-nom bought that thinks ‘John thinks that Mary bought that book.’ (ii) [IP John-ga [CP[IP Mary-ga [VP sono hon-o [V katta] ] ] to] omotteiru] (Bošković and Takahashi 1998:350)
214 2
Notes to pages 17–21
Asymmetry in syntax
1 The symbol T in (1a) stands for a set of all terminal nodes, A is a set of ordered pairs <X,Y> such that the first member asymmetrically c-commands the second one, and d(A) (the image under d of A) is the set of terminals that A dominates. The definitions of c-command and asymmetric c-command the LCA relies on are given in (i) and (ii) below. (i) X c-commands Y iff X and Y are categories and X excludes Y, and every category that dominates X dominates Y. (Kayne 1994:16) (ii) X asymmetrically c-commands Y, if X c-commands Y and Y does not c-command X. 2 A relationship between two elements is transitive if for all pairs <x,y> and in the set, the pair <x,z> is also present in the set. For example, “be a sibling of” is such a relation. If John is Bill’s sibling and Bill is Peter’s sibling, John is Peter’s sibling as well. A relationship between two elements is total if every two distinct elements in the set are ordered by this relationship. As noted by Di Sciullo (2005), Kayne’s definition of antisymmetry departs slightly from the definition of antisymmetry given in the previous chapter, where we defined the relationship between two elements in a set as antisymmetric if whenever both <x,y> and are members of the same set, then x must be the same as y. Furthermore, we took the relationship between two elements to be asymmetric if it is never the case that for any pair <x,y> in the set, the pair is in the same set, which is what (2c) says. In the discussion in this chapter, I will assume the definition of antisymmetry given in (2c). 3 The LCA thus derives the basic properties of X-bar theory; (4a) is a head taking another head as its complement, (4b) is a phrase with two heads, and (4c) is a phrase with no head. However, at the same time, the LCA is more restrictive than X-bar theory; it only allows one adjunct per phrase ((4d) is a multiple adjunction structure), and treats specifiers as adjuncts. 4 An issue that I come back to in Chapter 5 involves the label of the small clause in (6a). According to Moro, a small clause is a construction consisting of a subject and a predicate whose label is nevertheless different from both the subject and the predicate. This leads him to conclude that small clauses have no labels at all. I depart from him in Chapter 5, where I argue that (some) small clauses have symmetric labels. 5 There is a variant of (7c) which does indeed involve a symmetric configuration. It involves a non-branching object, as shown in (i) below. However, in such cases, movement of either the verb or the object would resolve symmetry. (i)
VP V
D
6 Di Sciullo also argues that morphology is “more asymmetric” than syntax. In particular, she assumes that syntax allows symmetry to the extent allowed by Moro’s Dynamic Antisymmetry Theory (see the discussion in the previous
Notes to pages 26–7
215
section), but morphology never does. One example comes from scope considerations. The ambiguity of (i) shows that scope is symmetric at a clausal level: (i) Every student solved some problem.
every > some, some > every
No analogous scope ambiguity exists at a word level; (iia) below, involving a modal and negative prefix, is not ambiguous: (ii) a. This situation is unbearable. neg> modal, * modal> neg b. It is not possible to bear this situation. c. # It is possible not to bear this situation. (Di Sciullo 2005:15) 7 The structural diagnostics discussed in this section come from Progovac’s (1998a) and (1998b) GLOT overview articles on coordination. 8 Of course, one could argue that this is simply because et cetera means ‘and the rest’. However, even speakers who do not know the origin or the meaning of etc have very clear intuitions about this contrast. 9 One insight that appears to be lost on the asymmetric structure involves the Law of the Coordination of Likes, which requires the two conjuncts to be of the same category. This intuition, however, is only apparent. The fact that the two conjuncts have to be identical also did not follow from the symmetric structure generated by the phrase structure rule XP Å XP conj XP, as it is not generally the case that the elements on the right hand side of the phrase structure rule have to be of the same type as the elements on the left hand side (just as it is not the case that the specifier and the complement of the same head have to be of the same type). Thus in either the symmetric or the asymmetric approach, this identity requirement has to be stated as a property specific to coordination. One possibility is to derive it from the semantics of coordination. This is what Munn (1993), for example, does. Alternatively, we could attribute it to the selectional requirements, or the lack thereof, of the conjunction head. One way to do it is to assume that the conjunction head has no selectional properties of its own. However, once it merges with one conjunct (which will end up being its complement), it “picks up” the requirement that the conjunct it merges with next (its specifier) cannot conflict in features with the first conjunct. 10 Johannessen (1998) refers to &P as a CoP (Conjunction Phrase). 11 This pronominal head can sometimes be overt, as in the following example: (i) (they), Tom, Mary and Jim.
(Progovac 1998b:5)
Likewise, the first conjunction can be overt, in so-called correlative coordination. In some languages (like English) correlative coordination involves the quantifier both, while in others, exemplified in (iiia–c), the repetition of the conjunction. (ii) Both John and Bill like movies. (iii) a. Jean connait (et) Paul et Michel. Jean knows and Paul and Michel ‘Jean knows (both) Paul and Michel.’
French
216
Notes to pages 28–33 b. Sono arrivati (e) Anna e Roberto. are arrived (and) Anna and Roberto ‘(Both) Anna and Roberto have arrived.’ c. (I) Marija i Milan donose vino. and Mary and Milan bring wine ‘Both Mary and Milan will bring wine.’
Italian
Serbo-Croatian (Progovac 1999:146)
12 Example (28) is modeled upon Progovac’s Serbo-Croatian one. The pattern goes beyond Slavic languages though. The fact that the anaphoric second conjunct (or an anaphor inside the second conjunct) cannot be licensed by the first conjunct is also shown by the following data from Dutch: (i) *Joopi en (een foto van) *Joopi and (a picture of) *‘Joopi and (a picture of) himselfi’
zichzelfi SE-SELFi (De Vries 2005b:92)
13 I will not in this chapter discuss the relationship between double object constructions, such as the one in (30) in the text, and PP dative constructions, such as (i) below: (i) Gina gave a book to Sue. Larson (1988), for example, argues that double objects are derived from PP datives, which allows him to maintain a strict approach to theta role assignments (cf. Baker’s Uniformity of Theta Role Assignment Hypothesis). Others, such as Harley (2002), Ormazabal and Romero (2007) and Pesetsky (1995), among many others, argue against such a derivational account, based on the differences between double objects and prepositional datives of the kind discussed by Green (1974) and Oerhle (1976). 14 There is some variation in judgments regarding wh-movement in double object constructions. Many speakers of American English find wh-extraction of indirect objects ungrammatical. I come back to this variation in Chapter 4. 15 The extra objects, which are often referred to as applied objects, can bear a variety of thematic roles. In (40b), for example, the applied object is benefactive. Other possible thematic roles are goals, recipients, malefactives, instrumentals, locatives and reasons. 16 English double objects, however, differ from more ‘paradigmatic’ cases of applicative constructions in that not all applied objects are optional. The verb give differs from bake in that its objects are obligatory without special context. (i) a. John gave Mary a cake. b. *John gave a cake. c. *John gave Mary. With verbs like teach on the other hand, either object can be omitted: (ii) a. Mary taught students calculus. b. Mary taught students. c. Mary taught calculus.
Notes to pages 33–8
217
Verbs like feed, on the other hand, require the indirect object to be present: (iii) a. John fed cows hay. b. John fed cows. c. ?? John fed some hay. d. *John fed (last week).
(Pesetsky 1995:158)
Such differences do not bear on the issues discussed in this chapter. 17 These diagnostics do not have to go together. Alsina and Mchombo (1993) discuss cases in which they diverge. In Chichewa, for example, benefactive and instrumental applicatives behave alike with respect to passivization; both allow only applied arguments to undergo passive movement. However, they behave differently with respect to other diagnostics: word order and object marking. In instrumental applicatives, both IO DO and DO IO word orders are possible, whereas in beneficiary applicatives, the IO has to be adjacent to the verb. Furthermore, in instrumental applicatives, either the IO or the DO can determine verb-object agreement, whereas in beneficiary applicatives, only the IO can. 18 More concretely, depictives (which Pylkkänen takes to be embedded inside a Dep(ictive) Phrase can only modify categories of type <e,st>. The low applicative is of a more complex type, which explains why it cannot be modified by a Depictive Phrase. 19 The intended interpretation of (47a) is ‘I ran for him’ and of (47b) ‘I held her book for Mary,’ with him and Mary being benefactors. 20 One might wonder whether equidistance might change this conclusion (and make the movement of either α or β possible). Chomsky’s (1995) definition of equidistance is given in (i), and the series of definitions leading up to the definition of minimal domain are given in (ii). (i) β is closer to K than α unless β is in the same minimal domain as (a) τ [τ being the target of raising, B.C] or (b) α. (Chomsky 1995:356) (ii) a. Max (α) is the smallest maximal projection including α. b. The domain δ(CH) of CH is the set of categories included in Max (α) that are distinct from and do not contain α or t. c. The minimal domain Min(δ(CH)) of CH is the smallest subset K of δ(CH) such that for any γ ε δ(CH), some β ε K reflexively dominates γ. (Chomsky 1995: 299) Chomsky (2004), however, dispenses with the concept of equidistance and assumes a strict locality-based approach to movement. This is also the approach I adopt throughout this book. 21 What distinguishes superraising from the cases of A-bar movement considered above is the fact that movement of the higher subject instead does not result in a grammatical string either: (i) *Iti seems that ti is certain John to fix the car.
218
Notes to pages 38–43
22 In this respect, English seems to be the odd language out, since movement of an embedded subject across an intervening experiencer is possible: (i) Maryi seems to John ti to like syntax. This is not a problem for the generalization established here, since the experiencer in English is a PP not a DP; see, however, Kitahara (1997) and Boeckx (1999, 2008) for some complications stemming from this view. 23 The nature of the experiencer makes a difference. For example, in Italian extraction is impossible across a full DP experiencer, but becomes possible if the experiencer is a pronoun: (i) Vickii mi sembra [ti aver dormito bene]. Vicki.nom me.dat seems have slept well ‘Vicki seem to me to have slept well.’ (Cuervo 2003b:127) Icelandic exhibits a yet different pattern; extraction is impossible across both a DP and clitic experiencer, but becomes possible across a copy of an A-bar moved experiencer, as shown below. (ii) *Hestarnir virðast mér [ti vera seinir]. the horses.nom seem me.dat be slow ‘It seems to me that the horses are slow.’ (iii) Hverjumi hafa hsestarnirj virst ti [tj vera seinir]? who.dat have the horses.nom seemed be slow ‘To whom does it seem that the horses are slow?’ (Holmberg and Hróarsdóttir 2003:998) 3
Symmetry in Merge
1 The ideas presented here are consistent with Merge being an asymmetric operation, but leading to a symmetric structure through a series of asymmetric steps. 2 One could ask here whether there is anything special about the symmetry involved in multidominant structures. The run-of-the-mill structures created by Merge, such as the one given in (i) below, also are symmetric; α and γ stand in a symmetric c-command relationship. Multidominance, however, adds an extra layer of symmetry, coming from the presence of two distinct roots. (i)
K α
γ
3 The term plane implies two distinct dimensions and some researchers do indeed use the two terms interchangeably. For the purposes of this chapter, I will keep the two distinct and focus on multidominant structures, and refer the interested reader to Carnie (2008) for a discussion of the differences between multidominant, multidimensional and multiplanar structures. 4 These are not the only terms that have been used in the literature to describe multidominant trees. Botanical metaphors abound. Svenonius (2005), for
Notes to pages 44–5
219
example, refers to multidominant trees as banyan trees. Van Riemsdijk (2006a, 2006b) analogizes the process of forming multidominant structures to the process of grafting one tree onto another. He refers to the shared element as the callus, one tree as the stock and the other one as the graft (or scion). Furthermore, for him the two planes are not equal in status; one is in a sense “parasitic” on the other. 5 Some of the ideas discussed in this chapter draw on previous work of mine. The discussion of the Parallel Merge mechanism itself and its application to ATB wh-questions builds on Citko (2003) and (2005). The discussion of free relatives builds on my dissertation (Citko 2000). Citko (in press-b) provides a general overview of multidominance in minimalist syntax, which summarizes some of the ideas discussed in this chapter. 6 One might wonder whether the second step (3b) violates the Extension Condition (which Chomsky more recently subsumes under the No Tampering Condition), which requires operations to target the root of the tree. This might be a valid objection, but I do not believe it is a deadly blow to multidominance, given the existence of other movements that violate the Extension Condition, such as head movement or covert movement. 7 In recent minimalist literature, movement is often conceived as resulting in a multidominant structure, which is a fairly standard way to interpret Chomsky’s (2004) idea that Move reduces to Internal Merge. On this view, a wh-moved subject would be multiply dominated (by CP and TP). Here, however, I maintain a distinction between multidominance resulting from Parallel Merge (which, itself, could be a sequence of two applications of External Merge) and multidominance resulting from movement. In this respect, I follow GračaninYüksek (2007), who also assumes that there are two types of multidominance, one resulting from Parallel Merge (or a process akin to it) and the other one from movement. She dubs them vertical and horizontal sharing, respectively. Her definitions of the two are given below. (i) X and Y horizontally share α if X ≠ Y and: X does not dominate Y and Y does not dominate X; and X reflexively dominates X′ and Y reflexively dominates Y′, and X′ and Y’ immediately share α. (ii) X and Y vertically share α if X ≠ Y and: X dominates Y or Y dominates X, and X and Y immediately share α. (Gračanin-Yüksek 2007:10-11) 8 Following Peters and Ritchie (1973), Williams (1978:32) takes a factor to be “any substring of a well-formed labeled bracketing that does not begin with a right bracket or end with a left bracket.” The parallelism constraint requires all conjuncts to be split by a factor line if one conjunct is split. It also requires left conjunct brackets to belong to the same factor. 9 The concept of a reduced phrase marker (RPM) is due to Lasnik and Kuppin (1977) and is discussed by Lasnik (2000), where the examples and definitions given below come from. A phrase marker (PM) is a set consisting of all the lines
220
Notes to pages 47–56
in the derivation of a given string. For example, a PM for the sentence He left is the one in (i). (i) PM = {S, he left, he VP, he V, NP left, NP VP, NP V} A monostring in turn consists of exactly one nonterminal symbol and any number of terminal symbols. The monostrings are in bold in (i). Reduced phrase markers (RPMs) are sets that contain only terminal strings and monostrings. The RPM for He left is given in (ii). (ii) RPM = {S, he left, he VP, he V, NP left}
10
11 12
13
14 15 16
17
The theory of phrase structure based on reduced phrase markers is more restrictive than the theory based on standard phrase markers; every reduced phrase marker is a phrase marker but not every phrase marker is a reduced phrase marker. One potential question about such “mixed” structures is linearization. As we will see shortly, one of the most straightforward ways to linearize a multidominant structure is to move the pivot (and spell it out in a post-movement non-symmetric position). This is impossible if the pivot is a “mixed” category of the kind given in (13a–b). However, such linearization considerations would also exclude structures in which the pivot is intermediate in both planes, such as the one given in (11a–b) above. This is not a fatal blow to the feasibility of such structures, since (as we will see shortly), there are ways to linearize pivots in situ. The abbreviation val in (14a–b) stands for any combination of person, number and gender features. The existence of the Multiple Agree mechanism is not uncontroversial. See, for example, Haegeman and Lohndal (2010) for arguments against Multiple Agree and a “binary” Agree reanalysis of West Flemish negative concord facts. There are views of Merge that avoid this problem. For example, De Vries (2005a) treats Merge as an inherently asymmetric operation that creates ordered pairs . P’ and Pi’ are precedence relations among strings of terminal elements. The structure in (27b) is an updated version of Goodall’s structure, reflecting current theoretical assumptions about phrase structure. This is only true on the fairly natural assumption that identity (or matching) with respect to some feature implies the presence of one head responsible for that feature. Thus, identity in tense between two conjuncts implies the presence of one T head. This is not the only logical possibility. The wh-phrase could move to the specifier of vP within only one conjunct. If it is the first conjunct, such a derivation might offer some insight into the reconstruction asymmetries in ATB questions discussed by Munn (1993). It would predict that the wh-phrase can be reconstructed either to the specifier of vP inside the first conjunct or to the base-generated position shared between both conjuncts, but never to a position within only the second conjunct. This might shed light on the first versus
Notes to pages 58–61
221
second conjunct asymmetry with respect to diagnostics such as Principle C and Principle A effects, illustrated in (i–ii). (i) a. *Which picture of Johni did hei like and Mary dislike? b. Which picture of Johni did Mary like and hei dislike? (ii) a. Which pictures of himselfi did Johni buy and Mary paint? b. *Which pictures of herselfj did John buy and Maryj paint? (Munn 1993:52) However, one of the anonymous reviewers brings to my attention examples involving an extra level of embedding, which could decide whether the issue with (i–ii) is reconstruction into the first conjunct, or some kind of proximity effect. The fact that they are possible (on the coindexed reading) suggest that the latter is more likely to be the case. (iii) a. Which picture of Johni do you think hei liked and Mary disliked? b. Which picture of Johni do you think Mary liked and hei disliked? 18 There is a principled exception to this generalization involving case-syncretic forms. I refer the interested reader to Citko (2005) for more discussion of case matching and case syncretism in ATB wh-questions, and to Citko (2000) and Section 3.3.6 of this chapter for a more comprehensive discussion of case matching (and case syncretism effects) in free relatives. 19 Munn (1999) discusses factors that make a multiple individual interpretation possible in English ATB wh-questions (see also Gawron and Kehler 2003 and Citko 2007). In Citko (2007), I discuss factors that influence the availability of a single individual, versus a multiple individual, interpretation in Polish ATB questions, such as the choice of a conjunction (a consecutive or a contrastive one) or the choice of a coordination strategy (simple coordination versus socalled correlative coordination). 20 They become even more similar if movement is treated as Internal Merge/ Remerge rather than Copy and Merge. Then sideward movement literally becomes Parallel Merge. 21 One could simply conclude from this contrast that headed relatives do not involve a head promotion account. This, I believe, would be a rash conclusion in view of the overwhelming evidence for promotion in relative clauses (see Vergnaud 1974, Kayne 1994, Bhatt 2002, among many others). 22 Logically speaking, another direction of assimilation is possible as well. Instead of assimilating ATB questions to parasitic gaps, one could assimilate parasitic gaps to ATB questions. This is the line of thought pursued by Williams (1990). Alternatively, we could treat both ATB wh-questions and parasitic gaps as involving sideward movement. This is what, for example, Hornstein and Nunes (2002) do. 23 The structure in (46) assumes Munn’s (1993) structure for coordination, in which the conjunction phrase (Boolean Phrase in his terminology) is adjoined to the first conjunct. The empty operator movement analysis, however, is compatible with the accounts that treat the first conjunct as a specifier and
222
24
25
26 27
28
Notes to pages 61–9
the second one as a complement of a conjunction head. This is the account I favored in Chapter 2, and have adopted since. See Munn (2001) for an account of the differences between ATB questions and parasitic gaps on the null operator movement account, and Hornstein and Nunes (2002) for an account of these differences on the sideward movement account. This is contra Munn’s (1993) observation that ATB questions actually parallel parasitic gaps in this respect. He considers example (47b) as an exception to this generalization, rather than an illustration of the general pattern. Even wh in situ languages can form ATB wh-questions, via scrambling, topicalization, or some other (non-wh-related) movement of the wh-pronoun. Not all English speakers find such questions grammatical. I will not attempt to explain this variation here. Instead, I will focus on describing the grammar of the speakers that do allow them. A simple explanation would be that there is some kind of variation with respect to the Law of Coordination of Likes. For some speakers, i.e. the ones that allow wh&wh questions, the fact that the coordinated elements are both wh-phrases is sufficient, for others categorical identity is also necessary. The judgments are more subtle as to whether wh&wh questions imply two events of singing (or just a single one). Gračanin-Yüksek (2007) refers to the two readings as it-reading and at-all reading, respectively, to reflect their paraphrases given below: (i) What did Sally sing and where did she sing it? (ii) What did Sally sing and where did she sing at all?
29 For clarity’s sake, this derivation ignores movement to the edge of vP. 30 I will not discuss monoclausal accounts here, such as the ones proposed by Zhang (2007), Gribanova (2009) and Citko and Gračanin-Yüksek (submitted), as English only allows bi-clausal wh&wh questions. 31 Kazenin (2001) makes a similar point based on data from Russian. What is interesting about Russian is that in some cases, sluicing is more restricted than wh&wh questions. 32 Yet another way to analyze right node raising, argued for by An (2007), Ha (2006) and Hartmann (2000), involves backward deletion, shown in (i). I will not discuss it any further here and focus on a comparison between movement and multidominant accounts instead. (i) John likes TV shows about vampires and Mary dislikes TV shows about vampires. 33 A potential counterexample might come from the fact that RNR obeys the Coordinate Structure Constraint, as shown in (i). This, however, might have to do with an independent parallelism constraint on RNR rather than the movement itself. (i) *? Josh was looking for the dean’s office, Maria was waiting in __, and reporters were trying to find __, Joss’s office. (Sabbagh 2007:376)
Notes to pages 69–75
223
34 A similar argument, due to McCawley (1982), comes from the fact that VP deletion includes the pivot. In (i-ii), for example, the missing VP is interpreted as “admire Adolf Hitler” not just “admire,” which is what we might expect if the pivot “Adolf Hitler” moved out of this VP. This argument, however, is somewhat weakened by the Copy Theory of Movement (since the VP would contain the copy of the moved pivot). (i) Tom admires, is sure that everyone else admires, Adolf Hilter, but of course you and I don’t [VP Ø] (ii) Tom admires, is sure that everyone else admires, Adolf Hilter, which of course hardly anyone does [VP Ø]. (McCawley 1982:100) 35 The issue of why rightward movement should be different in this respect from leftward movement remains a mystery though. 36 One could imagine a scenario in which the two elements undergo independent rightward movement (rather than moving as a unit). But it is not clear what would force movement of both. 37 The fact that right node raising can target non-constituents is even more apparent in OV languages. Thank you to an anonymous reviewer for bringing this fact to my attention, as well as for providing the following example from German: (i) weil Hans vor __ und Peter nach __ dem Krieg starb. because Hans before and Peter after the war died 38 Kluck (2009) makes a similar observation based on Dutch data and Larson (2007) based on German data. 39 Polish numerals belong to three different classes, as shown by Rutkowski and Maliszewska (2007). The Table in (i) gives examples of numerals belonging to each class, and lists their properties. (i) Types of numerals in Polish (based on Rutkowski and Maliszewska 2007) Type Examples
Properties
A-numeral jeden ‘one’ dwa ‘two’ trzy ‘three’ cztery ‘four’ agree with nouns they quantify
N-numeral tysia˛c ‘thousand’ million ‘million’ milliard ‘billion’
Q-numeral pie˛ ‘five’ szes´c ‘six’ siedem ‘seven’ etc.
always assign genitive case to nouns they modify
agree with head nouns in inherent case contexts; assign genitive case in structural case contexts
224
Notes to pages 77–82
40 This is related to the requirement that typically there are two remnants. However, the grammaticality of (i) below, which contains three remnants, shows that it is not an absolute requirement. (i) Peter talked to his boss on Tuesday and Betsy to her supervisor on Wednesday. (Sjoblom 1980:97) 41 This contrastive conjunction is also different from the so-called adversative conjunction but (ale in Polish), which, according to Malchukov (2004), is a conjunction that denies the presupposition of the first conjunct. An example is given in (i). (i) John went to the store but nobody saw him there. 42 Interestingly, in ATB wh-questions either conjunction is possible: (i) Który artykuł Maria napisała a/i Ewa zrecenzjonowała? which article Maria wrote andC /and Ewa reviewed ‘Which article did Maria write and Ewa review? The choice of a conjunction, however, affects the interpretation. For example, (ii) below is most naturally interpreted as a question about a single quantity of books with the consecutive conjunction i. With the contrastive conjunction, it can be a question about two distinct quantities of books. (ii) Ile książek Maria napisała a/i Ewa przeczytała? how.many books Maria wrote andC /and Ewa read ‘How many books did Maria write and Ewa read?’ See Citko (2007) for further discussion of factors that influence the interpretation of ATB questions. 43 Comparative clauses are an exception: (i) Jan zjadł więcej pierogów niż Maria zjadła pączków. Jan ate more pierogis than Maria ate donuts ‘Jan ate more pierogis than Maria donuts.’ Jackendoff (1971) notes that the deletion that we encounter in comparatives is much freer than the deletion that we see in gapping, which suggests that the two might be different processes. See Moltmann (1992a), however, for a multidominant analysis of comparatives. 44 With respect to the last three diagnostics (contrastiveness, obligatory linguistic antecedents and ungrammaticality in non-coordinate structures), gapping differs from other types of ellipsis, such as VP deletion or pseudogapping. I will not discuss these any further here (see Citko submitted, Johnson 2009 and Coppock 2001 for a summary of these differences). 45 This example is modeled after Ross’s (1970) English example: (i) I want to try to begin to write a novel and Mary a play. (Ross 1970:250) Further examples of non-constituent gapping from English are given in (ii–v):
Notes to pages 83–7
225
(ii) Arizona elected Goldwater Senator and Pennsylvania Schweiker. (iii) Max writes poetry in the bathroom, and Schwarz radical pamphlets. (iv) Maytag will give a brand new dryer to the winner of the Mrs. Albania contest, and General Electric four hundred light bulbs. (v) Jack begged Elsie to get married and Wilfred Phoebe. (Jackendoff 1971:24) 46 One could analyze gapping in verb-final languages as an instance of right node raising. Johnson (2006), however, shows four differences between right node raising and gapping, which argue against assimilating the two. First, gapping allows mismatches in inflectional features whereas right node raising does not, as shown in (ia–b). Second, right node raising (but not gapping) can target elements below word level, as shown in (iia–b). Third, right node raising needs to target continuous strings, whereas we saw examples of gapping that do not need to. And fourth, gapping cannot strand prepositions but right node raising can, as shown in (iiia–b). (i) a. b. (ii) a. b. (iii) a.
*He likes beans and you like rice. *He always complains and you sometimes complain. *Carly is overpaid and Will underpaid. *Carly is overpaid and Will underpaid. *John is confident of a successful outing and Peter is dependent on a successful outing. b. *John is confident of a successful outing and Peter is dependent on a successful outing. (Johnson 2006:410)
47 This is different from the conclusion reached by Culicover and Jackendoff (2005), who give the following examples as evidence that gapping cannot involve movement. In (i) the DP cats would have to move from the subject position of the embedded wh-clause, which would violate the Wh Island Constraint, and in (ii) the possessor Kissinger’s would have to move in a way that violates the Left Branch Condition. (i) Robin knows a lot of reasons why dogs are good pets and Lesliek catsj tk knows a lot of reasons why tj are good pets. (ii) I enjoyed reading both Kissinger’s criticism of Hitchens in the New York Times and Hitchens’k of Kissingerj tk criticism tj in the New York Times. (Culicover and Jackendoff 2005:275) 48 The generalization about matching I establish here for Polish differs from McShane’s (2005), who claims that in Russian gapping constructions mismatches are possible. The example she provides to illustrate this claim is given in (i). Note that the missing gap is in a subordinate clause, which suggests that it is not a gapping construction. (i) Podberëzovikov pogljadel na pomoščnicu, kak redaktor gljadit/ Podberezovikov looked.3sg.masc at assistant like editor looks.3sg/look.3sg. pogljadel by na opečatku cond at typo masc
226
Notes to pages 88–95
‘Podberezovikov looked at his assistant like an editor looking at a typo.’ (McShane 2005:137) 49 See Citko (submitted) for a detailed discussion of why voice matching requires v sharing. 50 The idea that the remnants in a gapping construction undergo movement goes back at least to Pesetsky (1982). Pesetsky’s (1982) argument in favor of such movement came from the fact that it can explain why gapping shows subjectobject asymmetries characteristic of movement constructions. The relevant contrast is given in (i–ii). The degraded status of (i) is presumably due to the violation of the Wh Island Constraint; what matters is the relevant contrast between these two examples. (i) ?This man knows why you eat spaghetti and that man knows why you eat macaroni. (Pesetsky 1982:644, citing Kuno 1976) (ii) *This man knows why spaghetti makes you sick, and that doctor knows why salmon makes you sick. (Pesetsky 1982:645) 51 Alternatively, the remnants and their correspondents could move out of their respective conjuncts and the remaining vPs would move to [Spec, AspP] (in an ATB fashion). This would be closer in spirit to Johnson’s account. However, I refer the interested reader to Citko (submitted) for a detailed discussion of reasons to prefer an account that does not rely on remnant movement. 52 There are two symmetric aspects to the structure Hiraiwa and Bodomo (2008) propose for serial verb constructions. One is the symmetric object sharing, and the other one is the symmetric nature of the phrase marked as Asp1+2P. They assume that it involves one phrase headed by two heads, following Baker and Stewart (1999), who argue that such double-headed structures are allowed as long as the two heads are of the same type. This could be viewed as an example of a structure with symmetric labels, the possibility I discuss in Chapter 5. 53 These two are not the only possibilities. The patterns in (i)–(iv) are also possible in Dàgááré (see Hiraiwa and Bodomo 2008 for data and analysis). (i) (ii) (iii) (iv)
clefting of V1 clefting of V2 clefting of V1 + V2 clefting of V1 + V2 + object
54 This predicate clefting pattern, namely the possibility of clefting the object with either verb, is not a universal property of serial verb constructions. Consequently, the analysis Hiraiwa and Bodomo propose for Dàgááré cannot be a universal structure for serial verb constructions. 55 There are actually two variants of the Head Account. On one, due to Bresnan and Grimshaw (1978), the wh-phrase is merged directly in the head position and is related to the gap via binding. On the other, the wh-pronoun moves from the relative clause-internal position to the head position (as suggested by Bury 2003, Citko 2008b and Larson 1998, among others).
Notes to pages 95–104
227
56 It is logically possible that there is no single right structure for free relatives, and that in some languages they have the Head structure and in others the Comp structure. 57 We could also divide free relatives into types based on whether they are headed by simple wh-pronouns or by complex wh-phrases: (i) a. I will eat what John cooks. b. I will eat whatever dish John cooks. The issue of whether free relatives headed by complex wh-phrases are true free relatives (or perhaps headed relatives in disguise, as argued recently by Donati 2006) is somewhat controversial. In Citko (2008b), I argue against Donati’s claims, and show that the two types pattern alike with respect to many diagnostics that distinguish headed from free relatives, such as complementizer selection, matching effects and extraposition in German. Another important distinction (but largely irrelevant for our purposes) concerns the interpretation of free relatives; more specifically, the fact that they can be interpreted as either definites or universals. Larson (1987) points out that two interpretations can be distinguished by the fact that the free relative in (iia) is most naturally paraphrased by a definite noun phrase, as shown in (iib), and the free relative in (iiia) is most naturally paraphrased by a definite noun phrase, as shown in (iiib). (ii) a. b. (iii) a. b.
58
59
60
61 62
I will eat what John cooks. I will eat the thing John cooks. I read whatever John writes. I read everything John writes.
This interpretive distinction was originally claimed to correlate with the presence (or absence) of -ever, but has since been shown to be independent of -ever. The interpretation changes though. TRFs contribute intensionality to the interpretation of the clause, and can be paraphrased by intensional modifiers such as allegedly, formerly, supposedly etc. This is by no means the only possible conclusion. Grosu (2003), for example, assumes that SFRs and TFRs involve the same structure (and derivation), and attributes the differences between them to independently motivated semantic considerations. In Citko (2000), I proposed a slightly different symmetric structure for free relatives, in which the wh-pronoun was shared between the matrix and the relative clause in its theta position. Similar considerations apply to adjunct free relatives. As pointed out by one of the anonymous reviewers, the resulting structure, being a variant of the Head Account, faces the same issues that the original Head Account faced. In particular, it fails to account for the relative clause extraposition data in (i–ii) advanced by Groos and Van Riemsdijk (1981) in support of the Comp structure. They attribute the ungrammaticality of (i) and
228
Notes to pages 110–14
the grammaticality of (ii) to the fact that the wh-pronoun is in [Spec, CP] and only CPs can extrapose in German. (i) * Hans hat was gelesen sein Vater * Hans has what read his father ‘Hans read what his father sent him.’ (ii) Hans hat gelesen was sein Vater Hans has read what his father
ihm geschickt hat. him sent has ihm geschickt hat. him sent has
This contrast is also consistent with the possibility that free relatives in some languages involve the Comp structure, and in other languages the Head structure. Furthermore, in (165b) there is a copy of the wh-pronoun in both the Head position and the Comp position. Crosslinguistic variation can thus be captured by allowing some languages to spell out the higher copy (i.e. the one in the Head position) and other languages to spell out the lower copy (i.e. the one in the Comp position). 4
Symmetry in Move 1 In this respect, my conclusions parallel McGinnis (2004), who also argues that long A-movement can arise in two ways, “by an EPP driven derivation and by a Case-driven derivation.” My account, however, differs in two respects, which I elaborate on as I discuss McGinnis’s account (Section 4.6.2) and introduce my alternative (Section 4.7.5). 2 There is variation with respect to the acceptability of (4b). For some speakers of American English, passivization of direct objects is also possible (as reported by Emonds 1972, Fillmore 1965, Postal 2004, among many others). Further examples are given below. (ii) a. Hei was given ti several books by the teacher. b. Several booksi were given him ti by the teacher. (Chomsky 1975: 493) (iii) a. Lagniappe: a small gifti given a customer ti by a merchant at the time of purchase. (Webster’s 3rd New International Dictionary) (Postal 2004:242, crediting Chris Potts) b. For his allowance, a regular allowancei was given him ti by the King of Babylon. (Jeremiah 52:34) Furthermore, there are factors that improve passivization of direct objects even for speakers that generally disallow it. For example, if the indirect object is a weak pronoun, passivization of the direct object becomes acceptable. (i) A letter was given ’im /*HIM by Mary.
(Larson 1988:364)
3 These patterns are not the only patterns possible in these languages. I discuss others as they become relevant. 4 Collins (1997a) suggests a parallel derivation for quotative inversion structures, illustrated in (i–ii) below. (i) “I am so happy,” Mary thought. (ii) “I am so happy,” thought Mary.
Notes to pages 115–20
229
5 The issue of symmetric movement in copular sentences is independent of the issue of whether copular sentences involve a symmetric structure, in which the subject and the predicate c-command each other, as well as the issue of what the label of this structure is. In Citko (2008a) I argue against a symmetric structure for small clauses (i.e. a structure in which the subject and the predicate c-command each other). However, in Chapter 5 I argue in favor of symmetric labels for a subset of small clauses. 6 Dziwirek (2002) also lists nabawić ‘cause’ as belonging to this class. It differs from both kinds of double object verbs considered in this section in that it disallows either object to passivize. (i)
Janek nabawił Ewę kłopotu. John.nom caused Eve.acc trouble.gen ‘John caused Eve trouble.’ (Dziwirek 2002:341, fn 1) (ii) * Ewai została nabawiona ti kłopotu przez Janka. * Eve.nom became caused trouble.gen by John * ‘Eve was caused trouble by John.’ (iii) * Kłopoti został nabawiony Ewę ti przez Janka. problem.nom became caused Eve.acc by John ‘A problem was caused for Eve by John.’ 7 The answer to the question of which order is basic and which one derived can also depend on the verb involved. Verbs can differ in this respect; for some the neutral order is one in which the indirect object precedes the direct one, whereas for others it is the other way round. The first class, however, strikes me as the larger of the two (in Polish) and this is the class I focus on here. 8 Not all diagnostics are directly applicable to Polish. Example (ia), unlike (ib), implies that the students learned some French. In Polish, aspect, not word order, is what conveys this implication. Example (iia) does not imply that the students learned French (in spite of the IO DO order) but (iib), with the same word order but perfective aspect instead, does. (i) a. John taught the students French. b. John taught French to the students. (ii) a. Jan uczył studentów Jan.nom taught.imperf students.acc b. Jan nauczył studentów Jan.nom taught.perf students.acc
francuskiego. French.gen francuskiego. French.gen
9 There are some double object verbs, such as powierzyć ‘entrust’, for which DO IO word order is more neutral (or even preferred): (i) ??Powierzyliśmy sprawę dobrej kancelarii. ??entrust.past.1pl case.acc good.dat law.firm.dat ??‘We entrusted this case to a good law firm.’ (ii) ??Powierzyliśmy dobrej kancelarii sprawę. ??entrust.past.1pl good.dat law.firm.dat case.acc ??‘We entrusted this case to a good law firm.’
230
Notes to pages 120–2
See Dvořák (in press) for a discussion of such verbs in Czech and Bailyn (in press) for a discussion of factors that influence the choice between two orders in Russian. My main focus in this chapter, however, is on Polish verbs for which IO DO order is the unmarked one. 10 With the opposite order of the two objects, opposite results obtain, as shown in (i–ii). This, however, cannot be used as an argument that the DO is higher than the IO, as it is also consistent with the DO IO order being derived from the IO DO order via movement (or both orders being base-generated). (i) a.
Jan pokazał Marięi jeji szefowej. Jan.nom showed Maria.acc her.dat boss.dat ‘Jan showed Maria to her boss.’ b. *Jan pokazał jeji szefową Mariii. Jan.nom showed her.acc boss.acc Maria.dat ‘Jan showed her boss to Maria.’ (ii) a. Jan dał każdyi czek jegoi właścicielowi. Jan.nom gave every.acc check.acc his.dat owner.dat ‘Jan gave every check to its owner.’ b. *Jan dał jegoi czek każdemu pracownikowii. Jan.nom gave his.acc check.acc every.dat employee.dat ‘Jan gave his check to every employee.’ 11 Willim (1989) and Wiland (2009) also show that in Polish movement of the direct object can induce a crossover configuration. Their examples are given below. (i) *Czyjąi książkę mui pokazałeś ti? whose book him showed ‘Whose book did you show to him?’ (Willim 1989:122) (ii) ??Synai Kowalskich policja odesłał jegoi mamie ti. son.acc Kowalski.gen police.nom sent.back his Mom.dat ‘The Kowalski’s son, the police sent back to his parents.’ (Wiland 2009:2) 12 Moore and Perlmutter (2000) apply some of these diagnostics to Russian. They contrast the behavior of two types of dative subjects (inversion nominals and subjects of infinitives) and argue that only the latter (subjects of infinitives) are true subjects. The two are exemplified in (i–ii) below. (i) Borisu ne istratit’ tak mnogo deneg na sebja. Boris.dat neg spend.inf so much money on himself. ‘It’s not (in the cards) for Boris to spend so much money on himself.’ (Moore and Perlmutter 2000:377) (ii) Borisu bylo veselo. Boris.dat was.neut.sg merry.neut.sg ‘Boris was having fun.’ (Moore and Perlmutter 2000:374) 13 Conjunction reduction is another standard subjecthood diagnostic. It is not a very reliable one for pro-drop languages like Polish.
Notes to pages 123–8
231
14 Other examples include so-called involuntary state constructions (to borrow Rivero’s 2009 term), illustrated in (i). The term reflects the intuition that the subject is in a certain state (or experiences a certain state) rather than being an Agent of a certain action. Dziwirek (1994) also shows that they behave like subjects with respect to most of the subjecthood diagnostics under consideration here. (i) Dobrze mi się dzisiaj spało. well me.dat refl today sleep.past.neut.3sg ‘I slept well today.’ (Dziwirek 1994:57) 15 In the Relational Grammar framework, the fact that dative subjects behave like subjects with respect to some diagnostics but not with respect to others is captured by inversion (more specifically, 1–3 arc demotion). Space considerations prevent me from doing justice to this analysis here; I refer the interested reader to Dziwirek (1994) for an overview of the framework, and arguments for analyzing dative subjects as intial 1s and final 3s. 16 This is reminiscent of McGinnis’s (2004) account, in which long A-movement can also arise in two ways, “by an EPP driven derivation and by a Case-driven derivation.” I differ from McGinnis in the following two respects. For me the EPP driven derivation is not limited to high applicatives and the case-driven derivation also involves movement. I will elaborate on these differences as I discuss McGinnis’s account (Section 4.6.2) and introduce my alternative (Section 4.7.5). 17 Doggett (2004) refers to this type of movement as hurdling and McGinnis (1998) as leapfrogging. 18 The reason why I refer to the second strategy as a case-based strategy has to do with the fact that the movement of the indirect object is related to case. This might not be quite obvious yet, but, hopefully will become apparent by the end of the chapter. 19 McGinnis’s (2001) approach also belongs to this category. Since it makes more specific claims about the nature of the movement schematized in (57a), I defer a discussion of it till the next section. 20 Siewierska and Hollmann (2007) note the presence of the give it me variant in the Northern varieties of British English. Haddican (2010) focuses on this alternation in the greater Manchester area. 21 An anonymous reviewer brings to my attention examples involving idioms such as take advantage of or keep tabs on, which pattern with symmetric double object constructions in that they allow two kinds of passives, as shown below. (i) a. b. (ii) a. b.
Johni was taken unfair advantage of ti. Unfair advantagei was taken ti of John. Johni was kept tabs on ti by the FBI. Tabsi were kept ti on John by the FBI.
As pointed out by the reviewer, the accounts of symmetric passives discussed in this chapter will not extend to such cases. It is not clear that they should, as we are dealing with idioms here, which suggests that something else might
232
Notes to pages 128–32
be going on. For example, passivization of the prepositional object could be a consequence of reanalysis. 22 Doggett (2004) also points out that Norwegian and Swedish are counterexamples to the generalization. They have symmetric passives but allow only one ordering of the two objects (IO DO order). 23 Anagnostopoulou explicitly argues against case-based accounts, based on the following data from Greek. What is interesting about Greek is that the indirect object has inherent genitive case but it nevertheless blocks movement of the direct object across it, as shown by the ungrammaticality of the examples in (i–ii). (i) ?*To vivlioi charistike tis Marias ti apo ton Petro. ?*the book.nom award.nact the Maria.gen from the Petros ?*‘The book was awarded to Mary by Peter.’ (ii) ?*To gramai tachidhromithike tu Petru ti apo tin Ilektra. ?*the report.nom mailed.nact.3sg the Petros.gen from the Ilektra ?*‘The letter was mailed to Peter by Ilektra.’ (Anagnostopoulou 2003:22) Anagnostopoulou’s locality-based account capitalizes on the fact that passivization improves if the goal is not expressed overtly (or if it is clitic doubled). Clitic doubling, she argues, has the effect of removing the formal features of the intervening indirect object. The structure is similar to the structure proposed here for the movement of Polish dative IOs. (iii) To vivlio tis charistike (tis Marias) the book.nom cl.gen award.nact the Maria.gen ‘The book was awarded to Mary.’ (Anagnostopoulou 2003:22) 24 The idea that symmetric and asymmetric applicatives involve different structures is also behind Baker’s (1988) analysis of the differences between symmetric and asymmetric applicatives. For him, in symmetric applicatives, the verb theta-marks both objects (which explains why they behave alike with respect to diagnostics such as adjacency to the verb, verbal agreement and passive movement). In asymmetric applicatives, the applied object gets its theta role from a null preposition, which incorporates into the verb. As a result of this incorporation, the applied object acquires what Baker refers to as “distinctive object properties.” One of these “distinctive” object properties is the ability to passivize. 25 A head can have an EPP feature either by virtue of being a phase head itself or by virtue of inheriting it from a phase head. For example, C and v (and now a high applicative head) have EPP features because they are phase heads, but T inherits its EPP feature from C. The low applicative head cannot inherit an EPP feature since it is not directly dominated by a phase head. 26 For both theoretical and empirical arguments in favor of anti-locality, see Abels (2003) and Grohmann (2003). 27 There is a more general question concerning the semantic interpretation of the two types of applicatives. In Pylkännen’s terms, transfer of possession is a crucial aspect of the meaning of the low applicative head, as shown in (i–ii).
Notes to pages 132–7
233
(i) λx. λy. λf<e, <s,t>> . λe. f(e,x) & theme (e,x) & to-the-possession (x,y) (ii) λx. λy. λf<e, <s,t>> . λe. f(e,x) & theme (e,x) & from-the-possession (x,y) Given this difference in meaning, we would expect all applicatives involving pure transfer of possession (with no implication that the applied object is benefiting from this transfer) to be analyzed as low applicatives. In more concrete terms, one wonders if this would not imply that the verb give should universally involve a low applicative structure, and therefore universally disallow symmetric passivization. However, many of the examples discussed above to illustrate crosslinguistic variation in the availability of symmetric passives involved a verb give or its close counterparts. To solve this issue, we would need to assume that the meaning of the verb give can vary from language to language. In a language with high applicatives, its indirect object would be interpreted as a benefactive, whereas in a language with low applicatives, it would be interpreted as a goal or a recipient. 28 McGinnis (2002) accounts for this contrast by assuming that the DO tucks in under the IO (rather than moves to the outer specifier of ApplH) when it moves to check the EPP feature of ApplH . As a result of such tucking in, the IO is still going to be closer to T for the purposes of passivization: (i) [TP T[EPP] [vP v [ApplHP IO [ApplH’ DOi ApplH[EPP] [VP V ti] ] ] ] ] This adds an extra parameter of variation (tucking in versus stacking), thus obscuring the correlation between high applicatives and symmetric passives. 29 The (a) examples in (78–81) are from Emonds (1976:79). I added the (b) examples to complete the paradigm. 30 A related difference, also discussed by Marantz (1993), concerns the inability of benefactives to form adjectival passives: (i) *children-baked cookies (ii) *boss-given flowers Again, in this respect benefactives differ from locatives and instruments: (iii) home-made cookies (iv) hand-made cookies 31 McGinnis (2001) notes that the lower object in a low applicative structure is not frozen with respect to wh-movement or quantifier raising. She does not discuss multiple wh-questions (and the restrictions on wh-movement of the higher object). 32 In a high applicative structure, the direct object moves through the specifier of the applicative head, which also makes an interesting prediction concerning reconstruction effects. In particular, it predicts that a direct object should be able to reconstruct into this intermediate position, in which it should have wide scope with respect to the indirect object (which remains in its base-generated position) but narrow scope with respect to the subject. 33 Den Dikken (1995) further notes that long distance movement of indirect objects is markedly worse than short (intra-clausal) distance movement:
234
Notes to pages 137–42 (i) % Whoi did John give ti a book? (ii) * Whoi did Bill think that John gave ti a book? (Den Dikken 1995:183)
He also notes that empty operator movement involving indirect objects is worse than overt wh-movement. His account capitalizes on this observation, a move that leads him to reanalyze long distance wh-movement (and short distance movement for speakers for whom the contrast in (88a–b) holds) as empty operator movement. 34 Hudson (1992) also observes an interesting contrast between extraction of indirect objects and subextraction from indirect objects, given in (i–ii). At present, I do not have an explanation for this contrast. (i) Which booki shall we give the author of ti a prize? (Hudson 1992:258) (ii) % Which authori shall we give ti a prize? 35 Extraction of applied argument becomes possible if the applied argument is clitic doubled, as shown in (i). (i) M-toto amba-ye wa-li-m-nunul-i-a zawadi. 1-child rel-1 2sa-past-buy-1OA-appl-fv 9present ‘The child whom they bought a present for.’ (Ngonyani 1996:46) Anagnostopoulou (2003) discusses a similar effect of clitic doubling in Greek. 36 The standard GB-style assumption that quantifier raising is covert phrasal movement is not trivial on current minimalist assumptions, where, generally speaking, the effects of covert movement are derived from the Agree relationship between a Probe and a Goal in its domain. For example, direct objects do not have to undergo covert movement to [Spec, vP] to get their case features checked (or valued). Case valuation can happen in situ, via Agree. The issue of whether quantifier scope can also be handled via Agree is a very interesting one, but one that goes beyond the scope of this chapter. I therefore assume a fairly uncontroversial approach to QR, on which quantified noun phrases undergo covert phrasal movement to their scope positions. 37 In order to get inverse scope, the lower object has to scramble over the higher one: (i) Nauczyciel zadał jedno zadaniei każdemu teacher.nom assigned one.acc problem.acc every.dat ‘The teacher assigned one problem to every student.’ (ii) Nauczyciel zadał każde zadaniei jednemu teacher.nom assigned every.acc problem.acc one.dat ‘The teacher assigned every problem to one student.’
studentowi ti. student.dat studentowi ti. student.dat
38 Frozen scope is a more general fact about Polish, extending beyond double object constructions. This is not surprising; in free word order languages (such as Polish) scope generally reflects the linear ordering of the two quantifiers. Ionin (2002) makes a similar observation about Russian. However, she also notes that inverse scope becomes possible if a universal quantifier is above a stressed indefinite, as shown in (i) (see also Stepanov and Stateva 2009 for a discussion of such examples).
Notes to pages 143–7
235
(i) Každaja koška ukusila odnu sobaku. Russian every.nom cat.nom bit one.acc dog.acc ‘Every cat bit one dog.’ one > every, every > one Polish behaves similarly in this respect. However, I do not believe that the availability of inverse scope under such restricted circumstances falsifies the generalization that scope is generally fixed in Polish or Russian. This is the reason why I use examples with neutral intonation to test the correlation between superiority and wh-movement. I depart in this respect from Stepanov and Stateva (2009), who use focused examples to test this correlation. They show that inverse scope obtains in Russian double object constructions if the two quantifiers are contrastively focused. Since the two wh-phrases in multiple wh-questions do not have to be contrastively focused for violations of superiority to be possible, using examples with non-focused quantifiers makes the data more parallel. 39 Again, if the order of the two objects is reversed (via short scrambling, for example), the opposite scope becomes possible: (i) Maria trzymała jedne drzwii każdemu Maria.nom held one.acc door.acc every.dat ‘Maria was holding one door for every guest.’ (ii) Maria trzymała każde drzwi jednemu Maria.nom held every.acc door.acc one.dat ‘Maria was holding every door for one guest.’
gościowi ti. guest.dat one > every gościowii ti. guest.dat one > every
40 This is somewhat of an idealization; we have seen above that in a single language, some types of applicatives can be symmetric and others asymmetric. 41 The genitive case in these examples is lexically selected by the verbs, and it differs from the so-called genitive of negation, a structural variant of the accusative case that appears on negated objects, as shown by the following contrast: (i) Jan czyta książkę. Jan.nom reads book.acc ‘Jan is reading a book.’ (ii) Jan nie czyta książki/*książkę. Jan.nom not reads book.gen/*acc ‘Jan is not reading a book.’ 42 Interestingly, not all lexically case-marked objects passivize. The examples in (i–ii) involve genitives that do not passivize, and the ones in (iii–iv) instrumentals that do not passivize. At present, the factors that distinguish the two types of passives remain somewhat unclear. The distinction between adjectival and verbal passives cannot be the crucial factor, since these passives allow by phrases, which is a property typically associated with verbal passives. (The active sentences, i.e. the ‘a’ sentences, are from Jagodzinski 2007. I have added the translations and the ungrammatical passives – the ‘b’ sentences.) (i) a.
Każdy pragnie czułości. everyone.nom desires tenderness.gen ‘Everyone desires tenderness.’
236
Notes to page 148 b. * Czułośći jest pragniona ti przez każdego. tenderness.nom is desired by everyone ‘Tenderness is desired by everyone.’ (ii) a. Staszek dostał grypy. Stanislaus.nom caught influenza.gen ‘Stanislaus caught a flu.’ b. * Grypai była dostana ti przez Staszka. flu.nom was caught by Stanislaus ‘Flu was caught by Stanislaus.’ (iii) a. Magda gardzi niebezpieczeństwami. Magda.nom despises perils.instr ‘Magda despises perils.’ b. * Niebezpieczeństwai są gardzone ti przez Magdę. perils.nom are despised by Magda ‘Perils are despised by Magda.’ (iv) a. Anna bawi się lalkami. Anna.nom plays refl dolls.instr ‘Anna plays with her dolls.’ b. * Lalkii zostały bawione ti się przez Annę. dolls were played refl by Anna ‘The dolls were played with by Anna.’ (Jagodziński 2007)
43 The behavior of English experiencers is not directly relevant here, as experiencers (in raising constructions) and indirect objects are of different categories (PPs headed by the preposition to versus DPs marked with accusative case). The lack of intervention effects of any sort, illustrated in (i–iii), might thus be attributed to the fact that the experiencer does not c-command the embedded subject (since it is buried within the PP). (i) Johni seems to Mary [ti to be the best candidate]. (ii) Johni seems to her [ti to be the best candidate]. (iii) To whomj does Johni seem tj [ti to be the best candidate]. However, standard c-command diagnostics suggest that the experiencer does indeed c-command the embedded subject; otherwise it would be impossible to explain the data in (iv–v) showing a c-command relationship between the experiencer and the embedded subject; (iv) is a Principle C violation, and the anaphor in (v) has to be in the c-command domain of the experiencer (at some point in the derivation). (iv) *Theyj seem to himi tj to like Johni. (v) Pictures of himselfi seem to Johni t to be ugly. A lot of research has gone into explaining why English experiencers behave as interveners with respect to some diagnostics but not with respect to others. Kitahara (1997), for example, accounts for this difference by assuming that the DP experiencer c-commands into the embedded clause only at LF, as a result of covert adjunction to PP.
Notes to pages 150–3
237
44 McGinnis (2001) attributes the Icelandic pattern to the fact that the dative experiencer is a closer candidate for movement because it is in a low applicative structure instead. 45 I adopt Lee-Schoenfeld’s (2006) terminology, who focuses on German (not Polish) to refer to different types of dative constructions, abstracting away from the issue of how to analyze them all. For a detailed description of Polish datives, I refer the interested reader to Dąbrowska (1997). 46 There are two lines of thought on the structure (and derivation) of dative possessors. One is a possessor raising analysis, on which the possessor raises from its DP-internal position to some higher position within the verbal domain, such as the specifier of VP or the specifier of an applicative head, as shown in (i). This, for example, is the analysis argued for by Landau (1999) for Hebrew and Lee-Schoenfeld (2006) for German. Pylkkänen (2008), on the other hand, argues against such a possessor raising analysis, and treats “raised” possessors as specifiers of low applicative heads, as shown in (ii). The choice between the two types of account is not crucial for our purposes here.
′
′
′ ′ ′
47 My judgment here differs from Dziwirek’s (1994), who deems the following examples to be ungrammatical: (i) *Szybko wysłało mi się Ewie paczkę. quickly send. me.DAT refl Eve.DAT package.acc ‘I sent Eve the package quickly.’ (ii) *Jankowi z trudem wytłumaczyło się Ewie ten John.DAT with difficulty explain.perf refl Eve.DAT this problem. problem.acc ‘John explained this problem to Eve with difficulty.’ (Dziwirek 1994:64) These are only slightly marginal, and their marginality has to do with the episodic nature of the predicates involved. If the examples in (i–ii) are made more generic, by replacing perfective verbs with imperfective ones or by making the object plural, the results become fully grammatical:
238
Notes to pages 153–4 (iii) Łatwo wysyła mi się Ewie paczki. easily send.imperf me.DAT refl Eve.DAT packages.acc ‘It is easy for me to send Eve packages.’ (iv) Jankowi z trudem tłumaczyło się Ewie ten problem. John.DAT with difficulty explain.imperf refl Eve.DAT this problem.acc ‘John explained this problem to Eve with difficulty.’
48 Rivero (2009), however, argues against a bi-clausal analysis, and proposes instead the following structure for the involuntary state construction: (i) [ApplP NPDAT [Appl’ Appl [TP Tense [AspP IMPOP [vP v VP] ] ] ] ] 49 There is a fair amount of crosslinguistic variation with respect to this restriction. Abraham (2006) notes that German disallows sequences of two datives or two genitives. However, he also notes that ethical datives constitute an exception to this generalization, as shown in (i). (i) Sie hilft mir ihr. she helps medat:ethical herdat:benefactive ‘I can’t believe it, she helps her!’
German (Abraham 2006:11)
Japanese allows two datives, as shown in (ii). However, this could be due to the fact that the dative particle ni in Japanese is ambiguous between a preposition and a dative marker. (ii) Taroo-ga Hanako-ni Tokyo-ni nimotu-o okutta. Japanese Taro.nom Hanako.DAT Tokyo-to package.acc sent ‘Taro sent Hanako a package to Tokyo.’ (Miyagawa and Tsujioka 2004:9) Thus, it could be the prepositional status of one of the alleged dative markers that is responsible for the grammaticality of (i) in Japanese. This, however, is not going to generalize to other languages that allow multiple datives, in which there is no evidence whatsoever for the prepositional status of the dative marker. Croatian, for example, is such a language. (iii) Vid je Danu poslao knjigu Hani. Croatian Vid aux Dan.dat sent book.acc Hana.dat ‘Vid sent Dan the book to Hana (to Hana’s place). (Gračanin-Yüksek 2006) The issue of why some languages allow multiple datives and others do not is a very interesting one, but it is not an issue that I will be able to solve here. 50 The only recursion structure that is banned involves a low applicative phrase embedded under another low applicative phrase. This is because a low applicative head relates two individuals, which in structural terms means that it requires an individual denoting expression in both its specifier and its complement. It thus would not allow another applicative phrase in its complement position. 51 The analysis presented in this section is a revision and expansion of the analysis first presented in Citko (2009).
Notes to pages 154–67
239
52 I take the terms “quirky” and “inherent” case to refer to the same phenomenon. Woolford (2006) argues for a three-way division in the case system, between inherent, lexical and structural cases. Both inherent and lexical cases are thetarelated and thus not structural. Lexical cases are restricted to internal arguments, whereas inherent case is the case of goals and external arguments. 53 The idea that dative indirect object undergoes movement is also behind the dative shift accounts of double object constructions. For Larson (1988), the indirect object undergoes “passive-like” movement to the specifier of VP. For Baker (1997), the indirect object moves to the specifier of an Aspect head. What distinguishes the present proposal is that the dative indirect object does not cross the accusative direct object. 54 The structure in (146b) follows Pylkkänen’s analysis of possessor raising structures as low applicatives. 5
Symmetric labels
1 The question that arises here is whether the asymmetry illustrated in (1a–b) shows that labels are asymmetric or that Merge itself is asymmetric. The evidence for symmetric labels I present in this chapter is different in nature from the evidence for symmetric Merge I presented in Chapter 3, which suggests that the existence of symmetric Merge is in principle distinct from the existence of symmetric labels. See, however, Boeckx (2008) for claims that Merge with no labels is a symmetric operation and that labels are necessary to make symmetric Merge structures interpretable. 2 Logically speaking, there exists a possibility of neither α nor β projecting and the resulting object nevertheless having a label (some element that is neither α nor β). Such “third-party” labels, however, would violate the Inclusiveness Condition, which bans the introduction of new elements in the course of the derivation. 3 There is another variant of (2a). The label, instead of being a set consisting of α and β, could be an ordered pair consisting of α and β, as shown in (i). I examine both types of labels in this chapter, as both are symmetric in the requisite sense. (i) Merge α and β, project both α
β
4 This requires the assumption that all v heads (not just transitive and unergative ones) are phase heads. See Legate (2003) for arguments in favor of this claim. Marantz (to appear) also addresses this point. 5 As noted by Collins (2002), in a system without labels, the notion of a maximal projection can be straightforwardly replaced with the notion of a saturated constituent. However, the calculation of which elements are saturated requires a lot of backtracking and thus an increase in computational power. 6 The minimality condition in (12) needs to be extended to functional categories
240
7
8
9
10
Notes to pages 167–71
in order to capture selection by functional heads as well as intervention effects caused by functional categories. It is, however, not clear whether functional categories participate in the same kinds of selection relationships as lexical categories, or form extended projections in the sense of Grimshaw (1991/2005) instead. Collins (2002) also discusses the relationship between labels and derivational loci. Superficially, the two look very much alike. However, there are two nontrivial differences between them. One concerns the fact that there can only be one locus at a given time, but there are multiple labels. Second, the locus changes throughout the derivation but the label stays the same, at least on standard minimalist assumptions about labels. However, there are reasons to believe that under certain (very restricted) circumstances labels can also change throughout the derivation. I discuss one such mechanism shortly, the reprojection mechanism of Hornstein and Uriagereka (2002). I do not mean to trivialize the issues surrounding pied-piping here. I simply assume its existence, following what strikes me as a general consensus in the field. For a recent discussion of the status of pied-piping in the grammar, see Cable (2007) and Heck (2008, 2009). There are two potential issues that reprojection might raise, which Hornstein and Uriagereka recognize. First, after reprojection the movement chain is nonuniform in that its head is an intermediate projection and its tail a maximal one. Second, the checking relationships are altered by reprojection; XP is the specifier of Y before reprojection, and YP becomes the specifier of X after reprojection. According to Hornstein and Uriagereka, this is not a problem, since reprojection is possible only at LF, after all the chain- and feature-checking configurations have been evaluated. More concretely, they assume that “chains […, B.C] are read by the interfaces prior to their becoming either nonuniform or improper” (Hornstein and Uriagereka 2002:108). Hornstein and Uriagereka (2002) argue that reprojection applies only to asymmetric quantifiers. The difference between symmetric and asymmetric quantifiers lies in whether the following equivalence holds. (i) Det X are Y Ù Det Y are X According to this diagnostic, the quantifier/determiner some is symmetric because (iia) and (iib) are truth-conditionally equivalent. The quantifier most, on the other hand, is not symmetric, since (iiia) and (iiib) are not equivalent in the same sense. (ii) a. b. (iii) a. b.
Some Basques are Spaniards. Some Spaniards are Basques. Most Basques are Spaniards. Most Spaniards are Basques. (Hornstein and Uriagereka 2002:111)
11 Crucially, reprojection can only apply at LF, after Spell-Out. If it happened before Spell-Out, the structure in (22b) above would be incorrectly linearized by the LCA as Love children most people, since the IP love children asym-
Notes to pages 171–2
241
metrically c-commands the quantifier most and the NP people. Furthermore, Hornstein and Uriagereka (2002) show that treating reprojection as a covert operation can account for the so-called quantifier-induced intervention effects illustrated below. (i) *Nobody gave most children a red cent. (ii) Whati did nobody give most children ti? (Hornstein and Uriagereka 2002:110) In (i) the quantifier most children intervenes between the negative polarity item a red cent and its licensor nobody. Interestingly, as shown in (ii), the same element (namely most children) does not intervene for the relationship between what and its trace. This contrast follows nicely from the assumption that whatever configuration is responsible for the ungrammaticality of (i) arises only at LF, after reprojection takes place. Hornstein and Uriagereka do not go into the specifics of how reprojection works in this case; presumably (i) ends up with the structure in (iii), assuming that QR targets VP (or vP on current assumptions). The negpol a red cent is buried inside the specifier of Q, which suggests that whatever factors are responsible for specifiers being islands are also at play here.
′
′ ′
′
12 Taraldsen (2002) also points out some problems with the idea that qui in French is an agreeing complementizer. First, qui lacks the properties typically associated with agreeing complementizers. For example, it lacks the person/ number/gender features that we find in Germanic agreeing complementizers. Second, it only “agrees” with the subject, not any phrase in its specifier. 13 In Citko (2008b), I argued that all these labels are in fact possible. My goal here is more modest – to provide evidence in favor of one of the two symmetric options, the one given in (27c). For now, suffice it to say that capturing the variation illustrated in (27a–d) would be impossible in a grammar with no labels.
242
Notes to pages 172–7
14 The issue arises with even more vengeance in a system without labels, as is recognized by Collins (2002). The only type of adjunction that might be amenable to a bare phrase structure analysis would be adjunction to categories that are simultaneously non-minimal and non-maximal. A schematic representation of such a case is given in (i), y’s mother and sister are the same in terms of their phrase-theoretical status (determined relationally); both are non-minimal and non-maximal. This, however, is not going to work for all cases of adjunction.
15 An issue that I will not address here is whether distinguishing adjuncts from non-adjuncts solely in terms of labels is sufficient to capture all the properties of adjuncts, such as ordering or co-occurrence restrictions of the kind discussed by Alexiadou (1997), Cinque (1999) and Ernst (2002), among many others. 16 See, however, Jackendoff and Pinker (2005) for a dissenting view. 17 The feature structures in (38a–f) depart slightly from Pereltsvaig’s structures, as hers also include a third optional feature (+/−H). Here I stick to two (conflicting) features. 18 Baker and Stewart (1999) (see also Pereltsvaig 2008) argue that “doubly-headed phrases emerge as a natural possibility in the Minimalist Program when (and only when) the two phrases combined by Merge are identical or nearly identical in syntactic features” (Baker and Stewart 1999:5). An anonymous reviewer also brings to my attention some pre-minimalist precursors of the symmetric Labeling argued for here. For example, Haider’s (1988) matching projections could be thought of as doubly headed (or symmetrically labeled) constituents. His definition of a matching projection is given in (i). (i) A matching projection is a projection superimposed on an existing projection such that the nodes of the primary projection serve as secondary nodes of the superimposed projection. (Haider 1988:112) Matching projections are subject to a similar identity requirement; a given head can serve as a secondary head only if its features are consistent with the features of the primary head. In (ii), for example, z and x cannot have opposite values.
′ ′
(Haider 1988:113)
Notes to pages 178–87
19
20 21
22
23 24
25
243
The idea of co-projection is also found in the discussion of extended projections. Bayer, Schmid and Bader (2005), for example, allow features of functional heads to co-project with features of lexical heads (as is also suggested in Citko (2008b)). Baker and Stewart (1999) distinguish between vPs and VoicePs (and assume that the subjects are base-generated as specifiers of VoiceP (rather than vP), whereas most researchers conflate the two. In their analysis, a voice head selects an external argument, and a v head encodes cause (and is thus absent in unaccusatives). Note that the symmetry we see in Edo serial verb constructions is different from the symmetry we saw in Dàgááré serial verbs, discussed in Chapter 3. In principle, there are two independent ways a small clause can be symmetric: by virtue of involving two elements (i.e. the subject and the predicate) that c-command each other, or by virtue of being symmetrically labeled. Here, I focus on the latter option, and refer the reader to Citko (2008a) for a discussion of the former. It is not clear how Moro’s condition stated in (57) can be reconciled with Baker and Stewart’s (1999) idea that doubly headed constituents are possible only if some higher head does not attract either of the two heads. Conceptually, the two are alike in that they place certain restrictions on doubly headed (or nonheaded) structures. However, the restrictions they place are opposites of each other; for Baker and Stewart, neither Goal can move in a configuration like (57a), whereas for Moro, one of the two Goals has to move. I leave the issue of how to reconcile the two for future research. (57b) rules out cases if which the Goal moves to a position in which it c-commands the Probe. Such an analysis is also in principle compatible with a functional head mediating the relationship between the two DPs, as long as this functional head itself does not contribute any features to the identity of the label. In Citko (2008a), I took this functional projection to be defective in the requisite sense. My goal in Citko 2008b was not to argue for symmetric labels but to look at all the logically possible labeling possibilities and argue that all of them are in fact attested. The labeling options I argued for in Internal Merge structures are listed in (i), and the ones I argued for in Internal Merge (Move) structures are listed in (ii). I refer the interested reader to that paper for details. (i) a. b. c. (ii) a. b. c.
Project α or β (standard option) labels in External Merge structures Project Both (comparative correlatives, extended projections) Project None (last instance of External Merge) Project Probe (standard option) labels in Internal Merge structures Project Goal (free relatives) Project Both Probe and Goal (head movement)
26 Bhatt (2003) argues that the correlative CP, instead of being generated in an IP adjoined position (and binding the demonstrative), starts out adjoined to the demonstrative and moves to the IP adjoined position. Nothing in what follows hinges on this assumption.
244
Notes to pages 188–92
27 Brasoveanu (2008), based on Romanian data, takes issue with this intuition, and argues that not all comparative correlatives are conditional. An example of such a non-conditional comparative correlative is given in (i) below. He refers to such correlatives as equative correlatives, as they equate two degrees or amounts. In (i), for example, what is equated is the amount by which the brother is taller than the sister and the amount by which the father is taller than the mother. It is not clear to me, however, whether these are comparative correlatives, rather than standard correlatives in which what is relativized is a degree (rather than an individual). (i) Cu cît e mai înalt fratele decît sora, Romanian with how much is more tall brother.the than sister.the, (tot) cu atît e mai înalt tatăl decît mama. (also) with that much is more tall father.the than mother.the ‘The brother is taller than the sister by a certain amount and the father is taller than the mother by the same amount.’ Its Polish counterpart (somewhat simplified in insignificant ways) is given in (ii). The reason why this is a standard correlative is that it does not involve the correlative pronoun strategy im tym; instead, it involves the same combination of relative and demonstrative pronouns that are used in standard correlatives, as shown by the parallelism between (ii) and (iii). (ii) O ile Jan jest wyższy od Marii, o tyle Polish by how.much Jan is taller than Maria by that.much Ewa jest niższa od Zosii. Ewa is shorter than Zosia ‘Jan is taller than Maria by the same amount that Ewa is shorter than Zosia.’ (iii) Ile Jan waży, tyle Maria waży. how.much Jan weighs that.much Maria weighs ‘Jan weighs as much as Maria weighs.’ 28 Beck (1997) focuses on German but this seems to be a plausible way to extend her analysis to English. 29 In Citko (2008b), I did not discuss ATB-style extraction from standard correlatives. I include it here for the sake of completeness. 30 (76a) requires a pause after the wh-pronoun co ‘what’. Otherwise the result is a garden path effect, due to a sequence of two wh-pronouns, which is what we find in run-of-the-mill multiple wh-questions: (i) Co kto lubi? what who likes ‘Who likes whom?’ 31 One could argue that the grammaticality of the examples in (77a–b) is due to the fact that one of the clauses contains a pronoun which is coreferential with the gap inside the other clause. This could suggest that a resumptive pronoun
Notes to pages 194–200
245
strategy is at play. However, we find the same pattern if neither clause contains a (potentially resumptive) pronoun. This is shown in (i–ii) for Polish. Culicover and Jackendoff (1999) make a similar point based on English data. (i) ? Komui myślisz, że im więcej zarobisz, tym bardziej who.dat think.2sg that the more earn.2sg the more zaimponujesz ti? impress.2sg ‘Who do you think that the more you earn the more you will impress?’ (ii) Komui myślisz, że im więcej zaimponujesz ti, tym who.dat think.2sg that the more impress.2sg the więcej zarobisz? more earn.2sg ‘Who do you think that the more you impress the more you will earn?’
32
33
34
35 36
37
38
If anything, extraction from the first clause (the adjunct clause) is slightly better, which is the opposite of what we would expect. I depart from Culicover and Jackendoff (1999) (see also Borsley 2004 and Abeillé and Borsley 2008), who suggest that such unique and idiosyncratic properties of comparative correlatives are a problem for “mainstream generative” accounts of these constructions. I hope to convince the reader of something quite the opposite; it is precisely the “mainstream generative” assumptions (more specifically, the minimalist view of phrase structure and labeling) that allow us to capture the properties of the comparative correlatives under consideration here. An independent reason why the example in (85a) is ungrammatical involves case matching. The verb powiedzieć‘tell’ requires a dative complement, whereas the verb odwiedzić ‘visit’ requires an accusative one. A dative pronoun cannot satisfy the case requirements of both verbs (and neither can a genitive one). However, (85b) involves no case mismatch; both gaps involve adnominal genitive positions. Kapetangianni and Taylor (2007) also observe that focus movement from comparative correlatives yields different results. In particular, it is ungrammatical from either clause. They attribute it to the fact that focused elements and compared phrases in comparative correlatives move to the same position (the specifier of a Focus Phrase). The element glossed as te is a Dutch syncategorematic element. On Den Dikken’s assumptions, the fronted comparative creates a wh-island. The comparative fronting in the first clause is obligatory, which explains why movement from the first clause is never possible. Some languages are included in more than one group; this means that they allow more than one extraction pattern. Blank spaces mean that the sources cited did not include the relevant data. There is, however, one potential problem with the structure in (97a) for standard correlatives in Slavic languages. Izvorski (1997) shows that Slavic correlative pronouns (or phrases) undergo fronting to a clause-initial position, which
246
Notes to pages 201–5
she takes to be the specifier of CP. The relevant data from Polish (Izvorski’s own examples are from Russian and Bulgarian) are given in (i–ii). The contrast between these two examples shows that the correlate cannot stay in situ. (i)
Kogoi lubimy ti, tegoi zaprosimy ti. who.acc like.1pl this.acc invite.1pl ‘We will invite who we like.’ (ii) * Kogoi lubimy ti, zaprosimy tego. who.acc like.1pl invite.1pl this.acc ‘We will invite who we like.’ Furthermore, the ungrammaticality of (iii) below shows that the ungrammaticality in (ii) cannot be due to some kind of a parallelism requirement, which would require the wh-pronoun inside the relative clause and the correlative pronoun inside the matrix clause to occupy analogous positions. In (iii) both remain in situ, but the result is nevertheless ungrammatical: (iii) *Lubimy kogo, zaprosimy tego. like who.acc invite this.acc ‘We will invite who we like.’
39
40
41
42
43
This, however, does not necessarily imply that both the relative and the matrix clause have to be CPs; it merely shows that the matrix clause in Slavic correlatives has to be larger than a TP in order to accommodate the fronted whpronoun. Chomsky (2004), for example, suggests that the process in question is SIMPL, which transforms a symmetric Pair Merge structure into an asymmetric Set Merge one. Admittedly, the details of this process remain somewhat mysterious. A reader will think of exceptions to the Coordinate Structure Constraint. The grammatical cases of extraction from comparative correlatives, however, do not have a similar flavor. I depart from them in that I do not assume that movement proceeds in a successive cyclic fashion from the specifier of the embedded vP directly to the specifier of the matrix vP, skipping the specifier of the embedded CP. This may be motivated for Austronesian languages such as Tagalog (which is the main focus of their paper) but seems problematic for Irish or dialects of German, which provide direct evidence in the form of complementizer agreement that movement has to proceed through the specifier of CP. As mentioned in note 41, in Rackowski and Richards’ original account, the wh-phrase moves from the specifier of the embedded vP directly to the specifier of the matrix vP. Relying on the Principle of Minimal Compliance, they allow the matrix v to skip the CP layer and undergo Agree with the next closest phase, the embedded vP. Now the specifier of vP and the entire vP are also equidistant from any higher Probe, which is what allows the wh-phrase to move. This structure departs slightly from the structure I proposed in Citko (2008b),
Notes to page 205
247
where I took the label of adjunction structures to be a simple object (not an ordered pair). 44 Nothing changes if the relative clause (marked as CP1 in (110a–b) is adjoined to TP instead. Such a derivation would be identical to the one schematized in (109b). 45 There is another symmetric pattern to consider, involving languages in which wh-movement is impossible out of either clause in comparative correlatives. This is Type II in (94), which is what we see (under certain circumstances) in Hungarian, German and Dutch. I can only offer a suggestion at this point that in such languages Agree is only possible with “simple” objects. In other words, there is something that blocks Agree with a symmetrically labeled constituent. In the absence of such an Agree relationship, movement out of either CP is blocked.
References
Abe, Jun and Hoshi, Hiroto 1999. “Directionality of movement in ellipsis resolution in English and Japanese,” in Lappin and Benmamoun (eds.), pp. 193– 226. Abbott, Barbara 1976. “Right node raising as a test for constituenthood,” Linguistic Inquiry 7:639–42. Abeillé, Anne and Borsley, Robert D. 2008. “Comparative correlatives and parameters,” Lingua 118:1139–57. Abeillé, Anne, Borsley, Robert D. and Espinal, Maria-Teresa 2006. “The syntax of comparative correlatives in French and Spanish,” in Müller (ed.), pp. 6–26. Abels, Klaus 2003. “Successive cyclicity, anti-locality and adposition stranding,” PhD thesis, University of Connecticut, Storrs. 2004. “Right node raising: ellipsis or across the board movement?”, in Moulton and Wolf (eds.), pp. 45–59. Abney, Steven 1987. “The English noun phrase in its sentential aspect,” PhD thesis, MIT, Cambridge, MA. Abraham, Werner 2006. “Introduction. Datives: structural vs. inherent – abstract vs. morphological – autonomous vs. combinatory – universally vs. languagespecifically configured?”, in Hole, Abraham and Meinunger (eds.), pp. 3–46. Ackema, Peter and Čamdžić, Amela 2003. “LF complex predicate formation: the case of participle fronting in Serbo-Croatian,” UCL Working Papers in Linguistics 15:131–75. Adger, David 2003. Core Syntax. Oxford: Oxford University Press. Adger, David and Ramchand, Gillian 2003. “Predication and equation,” Linguistic Inquiry 34:325–60. Adger, David, de Cat, Cécile and Tsoulas, George (eds.) 2004. Peripheries: Syntactic Edges and Their Effects. Dordrecht: Kluwer Academic Publishers. Aikhenvald, Alexandra Y. and Dixon, Robert M. W. 2007. Serial Verb Constructions – A Cross-Linguistic Typology. Oxford: Oxford University Press. Akatsuka, Noriko and Strauss, Susan (eds.) 2002. Japanese/Korean Linguistics 10. Stanford: CSLI Publications. Alderete, John, Han, Chung-hye and Kochetov, Alexei (eds.) 2005. Proceedings of the 24th West Coast Conference on Formal Linguistics. Somerville, MA: Cascadilla Press. 248
References
249
Alexiadou, Artemis 1997. Adverb Placement: A Case Study in Antisymmetric Syntax. Amsterdam/Philadelphia: John Benjamins Publishing Company. Alsina, Alex and Mchombo, Sam A. 1993. “Object asymmetries and the Chichewa applicative construction,” in Mchombo (ed.), pp. 17–45. An, Duk-Ho 2007. “Syntax at the PF interface: prosodic mapping, linear order, and deletion,” PhD thesis, University of Connecticut, Storrs. Anagnostopoulou, Elena 2003. The Syntax of Ditransitives: Evidence from Clitics. Berlin: Mouton de Gruyter. Anderson, Stephen R. and Kiparsky, Paul (eds.) 1973. A Festschrift for Morris Halle. New York: Holt, Rinehart and Winston. Andronis, Mary, Ball, Christopher, Elston, Heidi and Neuvel, Sylvain (eds.) 2001. Chicago Linguistic Society: The Main Session 37–1. Chicago: Chicago University Press. Aoun, Joseph and Li, Yen-hui Audrey 1989. “Constituency and scope,” Linguistic Inquiry 20:141–72. Arnaudova, Olga, Browne, Wayles, Rivero, Maria Luisa and Stojanovic, Danijela (eds.) 2004. Formal Approaches to Slavic Linguistics # 12: The Ottawa Meeting. Ann Arbor, MI: Michigan Slavic Publications. Bach, Emmon, Jelinek, Eloise, Kratzer, Angelika and Partee, Barbara H. (eds.) 1996. Quantification in Natural Languages, Studies in Linguistics and Philosophy 54. Dordrecht: Kluwer Academic Publishers. Bachrach, Asaf and Katzir, Roni 2009. “Right-node raising and delayed Spellout,” in Grohmann (ed.), pp. 283–316. Bailyn, John 1995. “A configurational approach to Russian ‘free’ word order,” PhD thesis, Cornell University. 2001. “The syntax of Slavic predicate case,” in Jäger et al. (eds.), pp. 1–23. in press. “What”s inside VP? New (and old) evidence from Russian,” Formal Approaches to Slavic Linguistics # 18: The Cornell Meeting. Bailyn, John and Citko, Barbara 1999. “Case and agreement in Slavic predicates,” in Dziwirek, Coats and Vakareliyska (eds.), pp. 17–39. Baker, Mark 1985. “The mirror principle and morphosyntactic explanation,” Linguistic Inquiry 16:373–415. 1988. “Theta theory and the syntax of applicative constructions in Chichewa,” Natural Language and Linguistic Theory 6:353–89. 1989. “Object sharing and projection in serial verb constructions,” Linguistic Inquiry 20:513–53. 1996. The Polysynthesis Parameter. Oxford: Oxford University Press. 1997. “Thematic roles and syntactic structure,” in Haegeman (ed.), pp. 73–137. 2008. “The macroparameter in a microparametric world,” in Biberauer (ed.), pp. 351–73. Baker, Mark C. and Stewart, Osamuyimen T. 1999. “On double-headedness and the anatomy of the clause,” manuscript, Rutgers University. Baker, Mark C., Johnson, Kyle and Roberts, Ian 1989. “Passive arguments raised,” Linguistic Inquiry 20:219–51.
250
References
Baltin, Mark and Postal, Paul M. 1996. “More on the reanalysis hypothesis,” Linguistic Inquiry 27:127–45. Barss, Andrew 2000. “Minimalism and asymmetric wh-interpretation,” in Martin, Michaels and Uriagereka (eds.), pp. 31–52. Barss, Andrew and Lasnik, Howard 1986. “A note on anaphora and double objects,” Linguistic Inquiry 17:347–54. Bayer, Josef, Schmid, Tanja and Bader, Markus 2005. “Clause union and clausal position,” in Den Dikken and Tortora (eds.) pp. 79–113. Beck, Sigrid 1997. “On the semantics of comparative conditionals,” Linguistics and Philosophy 20:229–71. Beck, Sigrid and Johnson, Kyle 2004. “Double objects again,” Linguistic Inquiry 35:97–123. Belletti, Adriana (ed.) 2004. Structures and Beyond. Oxford: Oxford University Press. Belletti, Adriana, Brandi, Luciana and Rizzi, Luigi (eds.) 1981. Theory of Markedness in Generative Grammar. Pisa: Scuola Normale Superiore di Pisa. Bernstein, Judy 1993. “Topics in the syntax of nominal structure across Romance,” PhD thesis, City University of New York, New York. Bhatt, Rajesh 2002. “The raising analysis of relative clauses: evidence from adjectival modification,” Natural Language Semantics 10:43–90. 2003 “The locality in correlativization,” Natural Language and Linguistic Theory 21:485–541. Biberauer, Theresa (ed.) 2008. The Limits of Syntactic Variation. Amsterdam/ Philadelphia: John Benjamins Publishing Company. Biberauer, Theresa and Roberts, Ian (eds.) In press. Principles of Linearization. Berlin: Mouton de Gruyter. Bierwisch, Manfred and Heidolph, Karl Erich (eds.) 1970. Progress in Linguistics. The Hague: Mouton. Billerey, Roger and Lillehaugen, Brook Danielle (eds.) 2000. Proceedings of the 19th West Coast Conference on Formal Linguistics. Somerville, MA: Cascadilla Press. Bird, Sonya, Carnie, Andrew, Haugen, Jason D. and Norquest, Peter (eds.) 1999. Proceedings of the 18th West Coast Conference on Formal Linguistics. Somerville, MA: Cascadilla Press. Blaho, Sylvia, Vicente, Luis and de Vos, Mark (eds.) 2004. Proceedings of CONSOLE XII. Leiden University. Blevins, James 1990. “Syntactic complexity: evidence for discontinuity and multidomination,” PhD thesis, University of Massachusetts, Amherst. Bobaljik, Jonathan 1995. “Morphosyntax: the syntax of verbal inflection,” PhD thesis, MIT. Boeckx, Cedric 1999. “Conflicting c-command requirements,” Studia Linguistica 53:227–50. 2006. Minimalist Essays. Amsterdam/Philadelphia: John Benjamins Publishing Company. 2008. Bare Syntax. Oxford: Oxford University Press. 2009. “The locus of asymmetry in UG,” Catalan Journal of Linguistics 8:41–53.
References
251
(ed.) In press. Oxford Handbook of Minimalist Syntax. Oxford: Oxford University Press. Boeckx, Cedric and Grohmann, Kleanthes 2004. “Sub-move: Towards a unified account of scrambling and D-Linking,” in Adger, de Cat and Tsoulas (eds.), pp. 341–57. Boeckx, Cedric and Piattelli-Palmarini, Massimo 2005. “Language as a natural object, linguistics as a natural science,” The Linguistic Review 22:447–466. Borer, Hagit 1984. Parametric Syntax: Case Studies in Semitic and Romance Languages. Dordrecht: Foris Publications. (ed.) 1986. Syntax and Semantics 19: The Syntax of Pronominal Clitics. New York: Academic Press. 2005. Structuring Sense: Volumes I and II. Oxford: Oxford University Press. Borsley, Robert D. 1983. “A note on the Generalized Left Branch Condition,” Linguistic Inquiry 14:169–74. 2004. “On the periphery: comparative correlatives in Polish and English,” in Arnaudova, Browne, Rivero and Stojanovic (eds.) pp. 59–90. Bošković, Željko 2001. On the Nature of the Syntax–Phonology Interface. Amsterdam: North-Holland. 2002. “On multiple wh-fronting,” Linguistic Inquiry 33:351–84. Bošković, Željko and Franks, Steven 2000. “Across-the-board movement and LF,” Syntax 3:107–28. Bošković, Željko and Lasnik, Howard 2003. “On the distribution of null complementizers,” Linguistic Inquiry 34:527–46. Bošković, Željko and Takahashi, Daiko 1998. “Scrambling and last resort,” Linguistic Inquiry 29:347–66. Bowers, John. 1993. “The syntax of predication,” Linguistic Inquiry 24:591–656. Brasoveanu, Adrian 2008. “Comparative and equative correlatives as anaphora to differentials,” Proceedings of SALT XVIII, downloadable at http://hdl. handle.net/1813/13030. Bresnan, Joan, and Grimshaw, Jane 1978. “The syntax of free relatives in English,” Linguistic Inquiry 9:331–91. Brody, Michael 2006. “Syntax and symmetry,” manuscript, downloadable at http://ling.auf.net/lingBuzz/000260. Browne, Wayles 1972. “Conjoined questions and the limitation on English surface structure,” Linguistic Inquiry 3:223–6. Bruening, Benjamin 2001. “QR obeys superiority: frozen scope and ACD,” Linguistic Inquiry 32:233–73. Bunting, Jacqueline, Desai, Sapna, Peachey, Robert, Straughn, Christopher and Tomková, Zuzana (eds.) 2006. Proceedings from the 42nd Annual Meeting of the Chicago Linguistic Society – Main Session. Chicago: Chicago Linguistic Society. Burelle, Sophie and Somesfalean, Stanca (eds.) 2003. Proceedings of the 2003 Annual Conference of the Canadian Linguistic Association. Montréal: Université de Québec.
252
References
Bury, Dirk 2003 “Phrase structure and derived heads,” PhD thesis, University College London. Cable, Seth 2007. “The grammar of Q: Q-particles and the nature of wh-fronting, as revealed by the wh-questions of Tlingit,” PhD thesis, MIT. Camacho, José 1997. “The syntax of NP conjunction,” PhD thesis, University of Southern California. 2003. The Structure of Coordination. Conjunction and Agreement Phenomena in Spanish and Other Languages. Dordrecht: Kluwer Academic Publishers. Caponigro, Ivano 2003. “Free not to ask: On the semantics of free relatives and wh-words cross-linguistically,” PhD thesis, University of California, Los Angeles. Cardinaletti, Anna, Cinque, Guglielmo and Giusti, Giuliana (eds.) 1988. Constituent Structure: Papers from the 1987 GLOW Conference. Dordrecht, Foris. Carnie, Andrew 2008. Constituent Structure. Oxford: Oxford University Press. Carter, Julie, Dechaine, Rose-Marie et al. (eds.) 1986. Proceedings of the 20th Northeastern Linguistic Society. Amherst, MA: GLSA Publications. Chang , Charles B. and Haynie, Hannah J. et al. (eds.) 2008. Proceedings of the 26th West Coast Conference on Formal Linguistics. Somerville, MA: Cascadilla Proceedings Project. Chen-Main, Joan 2006. “On the generation and linearization of multi-dominance structures,” PhD thesis, Johns Hopkins University. Cheng, Lisa and Corver, Norbert (eds.) 2006. Wh-Movement. Moving on. Cambridge, MA: MIT Press. Cheng, Lisa and Sybesma, Rint (eds.) 2003. The Second GLOT International Stateof-the-Article Book. Berlin: Mouton de Gruyter. Cho, Sungeun and Zhou, Xuan 2002. “The interpretations of wh-elements in conjoined wh-questions,” in Akatsuka and Strauss (eds.), pp. 522–31. Chomsky, Noam 1970. “Remarks on nominalization,” in Jacobs and Rosenbaum (eds.), pp. 184–221. 1973. “Conditions on transformations,” in Anderson and Kiparsky (eds.), pp. 232–83 1975. Reflections on Language. New York: Pantheon Books. 1994. “Bare phrase structure,” in Webelhuth (ed.), pp. 383–439. 1995. The Minimalist Program. Cambridge, MA: MIT Press. 2000. “Minimalist inquiries: the framework,” in Martin, Michaels and Uriagereka (eds.), pp. 89–155. 2001. “Derivation by phase,” in Hale and Keyser (eds.), pp. 1–52. 2004. “Beyond explanatory adequacy,” in Belletti (ed.), pp. 104–31. 2005. “Three factors in language design,” Linguistic Inquiry 36:1–22. 2008. “On phases,” in Freidin, Otero and Zubizaretta (eds.), pp. 133–66. Chukerman, Amy, Marks, Mitchell and Richardson, John F. (eds.) 1983. Papers from the 19th Regional Meeting of the Chicago Linguistic Society. Chicago: Chicago Linguistic Society. Cinque, Gugliemo 1994. “On the evidence for partial N movement in the Romance DP,” in Cinque, Koster, Pollock, Rizzi and Zanuttini (eds.), pp. 85–110.
References
253
1999. Adverbs and Functional Heads: A Cross-Linguistic Perspective. Oxford: Oxford University Press. (ed.) 2002. Functional Structure in DP and IP: the Cartography of Syntactic Structures. Oxford: Oxford University Press. Cinque, Guglielmo and Kayne, Richard (eds.) 2005. The Oxford Handbook of Comparative Syntax. Oxford: Oxford University Press. Cinque, Gugliemo, Koster, Jan, Pollock, Jean-Yves, Rizzi, Luigi and Zanuttini, Rafaela (eds.) 1994. Paths Towards Universal Grammar. Georgetown: Georgetown University Press. Citko, Barbara 1998. “An ATB account of free relatives and correlatives,” in Gruber, Higgins, Olson and Wysocki (eds.), pp. 69–81. 2000. “Parallel merge and the syntax of free relatives,” PhD thesis, Stony Brook University. 2002. “(Anti)reconstruction effects in free relatives: a new argument against the comp account,” Linguistic Inquiry 33:507–11. 2003. “ATB wh-questions and the nature of Merge,” in Kadowaki and Kawahara (eds.), pp. 87–102. 2005. “On the nature of Merge: External Merge, Internal Merge, and Parallel Merge,” Linguistic Inquiry 36:475–97. 2006. “Copula sentences reconsidered,” in Lavine, Franks, Tasseva-Kurktchieva and Filip (eds.), pp. 83–93. 2007. “Scope disambiguation strategies in ATB wh-questions,” Paper presented at the 33rd Annual Meeting of the Berkeley Linguistics Society. 2008a. “Small clauses reconsidered: not so small and not all alike,” Lingua 118:261–95. 2008b. “Missing labels,” Lingua 118:907–44. 2009. “A ‘new’ look at symmetric and asymmetric passives,” Proceedings of the 39th North East Linguistics Society. In press-a. “The puzzles of wh-questions with coordinated pronouns,” in Biberauer and Roberts (eds.). In press-b. “Multidominance,” in Boeckx (ed.). Submitted. “A multidominant solution to the Merchant/Johnson paradox,” In Uribe-Etxebarria and Valmala (eds.). Citko, Barbara and Gračanin-Yüksek, Martina (submitted). “Towards a new typology of wh-questions with coordinated wh-pronouns.” Citko, Barbara and Grohmann, Kleanthes 2001. “On the non-uniqueness of multiple wh-fronting: German = Bulgarian,” in Franks, Holloway-King and Yadroff (eds.), pp. 117–37. Collins, Chris 1997a. Local Economy. Cambridge, MA: MIT Press. 1997b. “Argument sharing in serial verb constructions,” Linguistic Inquiry 28:461–97. 2002. “Eliminating labels,” in Epstein and Seely (eds.), pp. 42–64. Compton, Richard, Goledzinowska, Magdalena and Savchenko, Ulyana (eds.) 2006. Formal Approaches to Slavic Linguistics # 15: The Toronto Meeting. Ann Arbor, MI: Michigan Slavic Publications.
254
References
Comrie, Bernard and Stone, Gerald 1978. The Russian Language since the Revolution. Oxford: Clarendon Press. Coppock, Elizabeth 2001. “Gapping: in defense of deletion,” in Andronis, Ball, Elston and Neuvel (eds.), pp. 133–48. Cuervo, Maria Cristina 2003a. “Datives at large,” PhD thesis, MIT. 2003b. “A control vs. raising theory of dative experiencers,” in Pérez-Leroux and Roberge (eds.), pp. 111–30. Culicover, Peter and Jackendoff, Ray 1999. “The view from the periphery,” Linguistic Inquiry 30:543–71. 2005. Simpler Syntax. Oxford: Oxford University Press. Culicover, Peter and Postal, Paul (eds.) 2001. Parasitic Gaps. Cambridge, MA: MIT Press. Dąbrowska, Ewa 1997. Cognitive Semantics and the Polish Dative. Berlin: Mouton de Gruyter. Dayal, Veneeta 1996. “Quantification in correlatives,” in Bach, Jelinek, Kratzer and Partee (eds.), pp. 179–205. 2002. “Single-pair versus multiple-pair answers: wh-in-situ and scope,” Linguistic Inquiry 33:512–20. 2006. “Multiple wh-questions,” in Everaert and van Riemsdijk (eds.), pp. 275– 326. De Vos, Mark and Vicente, Luis 2005. “Coordination under Right Node Raising,” in Alderete, Han and Kochetov (eds.), pp. 97–104. DeLancey, Scott and Tomlin, Russell (eds.) 1986. Papers from the First Pacific Linguistic Conference. Eugene, OR: Dept of Linguistics. Di Sciullo, Anna Maria 2002. Asymmetry in Grammar. Amsterdam/Philadelphia: John Benjamins Publishing Company. 2005. Asymmetry in Morphology. Cambridge, MA: MIT Press. Dikken, Marcel den 1995. Particles: On the Syntax of Verb-Particle, Triadic and Causative Constructions. Oxford: Oxford University Press. 2005. “Comparative correlatives comparatively,” Linguistic Inquiry 36:497–533. 2006. Relators and Linkers. Cambridge, MA: MIT Press. 2009. “Comparative correlatives and successive cyclicity,” in Lipták (ed.), pp. 263–306. Dikken, Marcel den and Tortora, Christina (eds.) 2005. The Function of Function Words and Functional Categories. Amsterdam/Philadelphia: John Benjamins Publishing Company. Dimitriadis, Alexis, Siegel, Laura, Surek-Clark, Clarissa and Williams, Alexander (eds.) 1997. Proceedings of the 21st Annual Penn Linguistics Colloquium. University of Pennsylvania Working Papers. Doggett, Teal Bissell 2004. “All things being unequal: locality in movement,” PhD thesis, MIT. Donati, Valentina 2006. “On wh-head movement,” in Cheng and Corver (eds.), pp. 21–46. Dornisch, Ewa 1998. “Multiple wh-questions in Polish. The interactions between wh-phrases and clitics,” PhD thesis, Cornell University.
References
255
Douglas-Brown, Denise 1996. “In search of syntactic symmetry: on the parallels between clausal and nominal hierarchical structure,” PhD thesis, Univesity of Durham, UK. Dvořák, Věra. In press. “Ditransitive verbs in Czech,” Formal Approaches to Slavic Linguistics # 18: The Cornell Meeting. Dyakonova, Marina 2007. “Russian double object constructions,” ACLC Working Papers 2(1):3–30. 2009. “A phase-based approach to Russian word order,” PhD thesis, Netherlands Graduate School of Linguistics. Dyła, Stefan 1984. “Across-the-board dependencies and case in Polish,” Linguistic Inquiry 15:701–05. Dziwirek, Katarzyna 1994. Polish Subjects. New York/London: Garland Publishing. 2002. “A different kind of non-canonical case marking: the Slavic verb ‘to teach’,” Slavic and East European Journal 46:319–47. Dziwirek, Katarzyna, Coats, Herb and Vakareliyska, Cynthia (eds.) 1999. Formal Approaches to Slavic Linguistics #7: The Seattle Meeting. Ann Arbor: MI: Michigan Slavic Publications. Embick, David and Noyer, Rolf 2001. “Movement operations after syntax,” Linguistic Inquiry 32:555–95. Emonds, Joseph 1972. “Evidence that indirect object movement is a structure-preserving rule,” Foundations of Language 8:546–61. 1976. A Transformational Approach to Syntax. New York: Academic Press. Emonds, Joseph and Whitney, Rosemary 2006. “Double object constructions,” in Everaert and van Riemsdijk (eds.), pp.73–145. Epstein, Samuel and Seely, Daniel (eds.) 2002. Derivation and Explanation in the Minimalist Program. Oxford: Blackwell. Ernst, Thomas 2002. The Syntax of Adjuncts, Cambridge: Cambridge University Press. Everaert, Martin and van Riemsdijk, Henk (eds.) 2006. The Blackwell Companion to Syntax: Vols I–V. Malden/Oxford: Blackwell Publishing. Fillmore, Charles 1965. Indirect Object Constructions in English and the Ordering of Transformations. The Hague: Mouton. Fitch, W. Tecumseh, Hauser, Marc, and Chomsky, Noam 2005. “The evolution of the language faculty: clarifications and implications,” Cognition 97(2):179– 210. Fowler, George 1996. “Oblique passivization in Russian,” Slavic and East European Journal 40:519–45. Fox, Danny and Nissenbaum, Jon 1999. “Extraposition and scope: a case for overt QR,” in Bird, Carnie, Haugen, and Norquest (eds.), pp.132–44. Frajzyngier, Zygmunt 1986. “Propositional characterization of categories,” in DeLancey and Tomlin (eds.), pp. 108–19. Franks, Steven 1993. “On parallelism in across-the-board dependencies,” Linguistic Inquiry 4:509–29. 1995. Parameters of Slavic Morphosyntax. New York: Oxford University Press.
256
References
Franks, Steven, Holloway-King, Tracy and Yadroff, Michael (eds.) 2001. Formal Approaches to Slavic Linguistics: The Bloomington Meeting 2000. Ann Arbor MI: Michigan Slavic Publications. Frascarelli, Mara (ed.) 2006. Phases of Interpretation. Berlin: Mouton de Gruyter. Freidin, Robert (ed.) 1996. Current Issues in Comparative Grammar. Dordrecht: Kluwer, Academic Publishers. Freidin, Robert, Otero, Carlos and Zubizaretta, Maria-Luisa (eds.) 2008. Foundational Issues in Linguistic Theory. Cambridge, MA: MIT Press. Gast, Volker 2005. “I gave it him – On the motivation of the ‘alternative double object construction’ in English,” Paper presented at the Workshop on Ditransitive Constructions, Padang, July 21, 2005. 2007. “I gave it him – On the motivation of the ‘alternative double object construction’ in varieties of British English,” Functions of Language 14, 1:31–56. Gawron, Mark and Kehler, Andrew 2003. “Respective answers to coordinated questions,” Proceedings of the Thirteenth Conference on Semantics and Linguistic Theory. Georgopoulos, Carol and Ishihara, Roberta (eds.) 1991. Interdisciplinary Approaches to Language. Essays in Honor of S. Y. Kuroda. Dordrecht: Kluwer Academic Publishers. Gerdts, Donna and McGinnis, Martha 2003. “A phase theoretic analysis of Kinyarwanda multiple applicatives,” in Burelle and Somesfalean (eds.), pp. 154–65. Giannakidou, Anastasia, and Merchant, Jason 1998. “Reverse sluicing in English and Greek,” Linguistic Review 15:233–56. Goodall, Grant 1983. “A three-dimensional analysis of coordination,” in Chukerman, Marks and Richardson (eds.), pp. 146–54. 1987. Parallel Structures in Syntax. Cambridge: Cambridge University Press. Gračanin-Yüksek, Martina 2006. “Double object construction in Croatian: arguments against Appl0,” in Compton, Goledzinowska and Savchenko (eds.), pp. 94–112. 2007. “About sharing,” PhD thesis, MIT. 2008. “Free relatives in Croatian: arguments for the COMP Account,” Linguistic Inquiry 39:275–94. In press. “Linearizing multidominance structures,” in Biberauer and Roberts (eds.) Green, Georgia M. 1974. Semantics and Syntactic Regularity. Bloomington, IN: Indiana University Press. Gribanova, Vera 2009. “Structural adjacency and the typology of interrogative interpretations,” Linguistic Inquiry 40:133–54. Grimshaw, Jane 1991. “Extended projections,” manuscript, Brandeis University. [Published as Grimshaw 2005. Words and Structure. CSLI Lecture Notes Number 151, Stanford, CA: CSLI Publications]. Grohmann, Kleanthes 2003. Prolific Domains: On the Anti-locality of Movement Dependencies. Amsterdam/Philadelphia: John Benjamins Publishing Company.
References
257
(ed.) 2009. InterPhases: Phase-Theoretic Investigations of Linguistic Interfaces. Oxford: Oxford University Press. Groos, Anneke and van Riemsdijk, Henk 1981. “Matching effects in free relatives: a parameter of core grammar,” in Belletti, Brandi and Rizzi (eds.), pp. 171– 216. Grosu, Alexander 1994. Three Studies in Locality and Case, London: Routledge. 1996. “The proper analysis of ‘Missing-P’ free relative constructions,” Linguistic Inquiry 27:257–93. 2003. “A unified theory of ‘standard’ and ‘transparent’ free relatives,” Natural Language and Linguistic Theory 21:247–331. Grosu, Alexander and Landman, Fred 1998. “Strange relatives of the third kind,” Natural Language Semantics 6:125–70. Grosz, Patrick 2009. “Movement and agreement in right-node raising constructions,” unpublished manuscript, MIT. Gruber, Catherine, Higgins, Derrick, Olson, Kenneth S. and Wysocki, Tamra (eds.) 1998. Chicago Linguistic Society 34: The Main Session. Chicago, IL: University of Chicago Press. Guéron, Jacqueline and Lecarme, Jacqueline (eds.) 2004. The Syntax of Time. Cambridge, MA: MIT Press. Guimarães, Maximiliano 2000. “In defense of vacuous projections in bare phrase structure,” in Guimarães, Meroni, Rodrigues and San Martin (eds.) Guimarães, Maximiliano, Meroni, Luisa, Rodrigues, Cilene and San Martin, Itziar (eds.) 2000. University of Maryland Working Papers in Linguistics 9. College Park: University of Maryland Department of Linguistics. Ha, Seungwan 2006. “Multiple dominance CAN’T, but ellipsis can account for right node raising,” in Bunting, Desai, Peachey, Straughn, and Tomková (eds.), pp. 17–31. Haddican, William 2010. “Theme-goal ditransitives and theme passivisation in British English dialects,” Lingua 120:2424–34. Haegeman, Liliane (ed.) 1997. Elements of Grammar. Dordrecht: Kluwer Academic Publishers. Haegeman, Liliane and Lohndal, Terje 2010. “Negative concord and (multiple) agree: a case study of west Flemish,” Linguistic Inquiry 41:181–211. Hagstrom, Paul 1998. “Decomposing questions,” PhD thesis, MIT. Haider, Hubert 1988. “Matching projections,” in Cardinaletti, Cinque and Giusti (eds), pp. 101–23. Hale, Ken 1983. “Warlpiri and the grammar of non-configurational languages,” Natural Language and Linguistic Theory 1:5–47. 1991. “Misumalpan verb sequencing constructions,” in Lefebvre (ed.), pp. 1–35. Hale, Kenneth and Keyser, Samuel Jay (eds.) 2001. The View from Building 20. Cambridge: MA: MIT Press. Halle, Morris and Marantz, Alec 1993. “Distributed morphology and the pieces of inflection,” in Hale and Keyser (eds.), pp. 111–76. 1994. “Some key features of distributed morphology,” MIT Working Papers in Linguistics 21:275–88.
258
References
Hankamer, Jorge and Sag, Ivan 1976. “Deep and surface anaphora,” Linguistic Inquiry 7:391–428. Hannay, Mike and Steen, Gerard J. (eds.) 2007. Structural-Functional Studies in English Grammar. Amsterdam/Philadelphia: John Benjamins Publishing Company. Harley, Heidi 2002. “Possession and the double object construction,” Linguistic Variation Yearbook 2:29–68. Harley, Heidi and Noyer, Rolf 2003. “Distributed morphology,” in Cheng and Sybesma (eds.), pp. 463–96. Hartmann, Katharina 2000. Right Node Raising and Gapping: Interface Conditions on Prosodic Deletion. Amsterdam/Philadelphia: John Benjamins Publishing Company. Hauser, Mark, Chomsky, Noam and Fitch, Tecumseh 2002. “The faculty of language: what is it, who has it, and how did it evolve?” Science 298:1569–79. Heck, Fabian 2008. On Pied-piping: Wh-movement and Beyond. Berlin: Mouton de Gruyter. 2009. “On certain properties of pied-piping,” Linguistic Inquiry 40:75–111. Higginbotham, James and May, Robert 1981. “Questions, quantifiers, and crossing,” The Linguistic Review 1:41–79. Hiraiwa, Ken 2005. “Dimensions of symmetry in syntax: agreement and clausal architecture,” PhD thesis, MIT. Hiraiwa, Ken and Bodomo, Adams 2008. “Object-sharing as symmetric sharing: evidence from Dagaare,” in Chang and Haynie (eds.), pp. 243–51. Hoji, Hajime 1985. “Logical Form constraints and configurational structures in Japanese,” PhD thesis, University of Washington. Hole, Daniel P., Abraham, Werner and Meinunger, André (eds.) 2006. Datives and Other Cases. Amsterdam/Philadelphia: John Benjamins Publishing Company. Holmberg, Anders and Hróarsdóttir, Thorbjörg 2003. “Agreement and movement in Icelandic raising constructions,” Lingua 113:997–1019. Holmberg, Anders and Platzack, Christer 1995. The Role of Inflection in Scandinavian Syntax. Oxford: Oxford University Press. Hornstein, Norbert 2009. A Theory of Syntax: Minimal Operations and Universal Grammar. Cambridge: Cambridge University Press. Hornstein, Norbert and Nunes, Jairo 2002. “On asymmetries between parasitic gaps and across-the-board constructions,” Syntax 5:26–54. 2008. “Adjunction, labeling, and bare phrase structure,” Biolinguistics 2.1:57– 86. Hornstein, Norbert and Uriagereka, Juan 2002. “Reprojections,” in Epstein and Seely (eds.), pp. 106–32. Hornstein, Norbert, Nunes, Jairo and Grohmann, Kleanthes 2005. Understanding Minimalism. Cambridge: Cambridge University Press. Huang, James C.-T. 2005. “Syntactic analyticity and the other end of the parameter,” Lecture notes from the LSA Summer Institute. Huber, Stefan 2007. “Comparative correlatives – the case of German,” Proceedings of the 33th Annual Meeting of the Berkeley Linguistics Society.
References
259
Hudson, Richard 1992. “So-called ‘double objects’ and grammatical relations,” Language 68:251–76. Hughes, Arthur and Trudgill, Peter 1979. English Accents and Dialects: An Introduction to the Social and Regional Varieties in the British Isles. London: Arnold. Ionin, Tania 2002. “The one girl who was kissed by every boy: scope, scrambling and discourse function in Russian,” in van Koppen, Sio and de Vos (eds.), pp. 79–94. Izvorski, Roumyana 1997. “The syntax and semantics of correlative proforms,” in Kusumoto (ed.), pp. 133–47. Jackendoff, Ray 1971. “Gapping and related rules,” Linguistic Inquiry 2:21–35. Jackendoff, Ray and Pinker, Steven 2005. “The nature of the language faculty and its implications for evolution of language (Reply to Fitch, Hauser, and Chomsky),” Cognition 97:211–25. Jacobs, Roderick A. and Rosenbaum, Peter S. (eds.) 1970. Readings in English Transformational Grammar. Waltham, MA: Blaisdell Publishing. Jacobson, Pauline 1999. “Towards a variable-free semantics,” Linguistics and Philosophy 22:117–84. Jäger, Gerhard, Strigin, Anatoli, Wilder, Chris, Zhang, Niina (eds.) 2001, ZAS Papers in Linguistics 22. Berlin: ZAS. Jaeggli, Osvaldo 1986. “Passive,” Linguistic Inquiry 17:593–9. Jagodziński, Grzegorz 2007. “More on syntactic typology,” manuscript, downloadable at www.aries.com.pl/grzegorzj/thp/morfosynt.html. Jenkins, Lyle 2000. Biolinguistics: Exploring the Biology of Language. Cambridge: Cambridge University Press. Jensen, John T. and van Herk, Gerard (eds.) 2002. Proceedings of the 2001 CLA Annual Conference. Department of Linguistics, University of Ottawa: Cahiers Linguistiques d’Ottawa. Jeong, Youngmi 2007. Applicatives: Structure and Interpretation from a Minimalist Perspective. Amsterdam/Philadelphia: John Benjamins Publishing Company. Johannessen, Janne Bondi 1998. Coordination. Oxford: Oxford University Press. Johnson, Kyle. 1996/2003. “In search of the middle field,” manuscript, University of Massachusetts, Amherst. 2000. “Few dogs eat Whiskas or cats Alpo,” in Kusumoto and Villalta (eds.), pp. 59–82. 2006. “Gapping,” in Everaert and van Riemsdijk (eds.), pp. 407–35. 2007. “LCA+Alignment=RNR,” Handout of a talk presented at the Workshop on Coordination, Subordination and Ellipsis, Tubingen, June 2007. (ed.) 2008. Topics in Ellipsis. Cambridge: Cambridge University Press. 2009. “Gapping is not (VP) ellipsis,” Linguistic Inquiry 40:289–328. Kadowaki, Makoto and Kawahara, Shigeto (eds.) 2003. Proceedings of the 33rd Annual Meeting of the North East Linguistics Society. Amherst, MA: GLSA Publications. Kapetangianni, Konstantia and Taylor, Heather 2007. “Comparative correlatives
260
References
in Greek: the syntax of oso,” Paper presented at the Workshop on Greek Syntax and Semantics. Kasai, Hironobu 2007. “Multiple dominance in syntax,” PhD thesis, Harvard University. Kayne, Richard 1984. Connectedness and Binary Branching. Dordrecht: Foris. 1994. The Antisymmetry of Syntax, Cambridge, MA: MIT Press. 2005. “Some notes on comparative syntax,” in Cinque and Kayne (eds.), pp. 3–69. 2008. “Antisymmetry and the lexicon,” manuscript, downloadable at http://ling. auf.net/lingBuzz/000598. Kazenin, Konstantin 2001. “On coordination of wh-phrases in Russian,” Paper presented at the Fourth European Conference on Formal Description of Slavic Languages. Kiefer, Ferenc, Anderson, Stephen R. and Kiss, Katalin E. (eds.) 1994. The Syntactic Structure of Hungarian. Syntax and Semantics 27. New York: Academic Press. Kim, Jeong-Seuk 1997. “Syntactic focus movement and ellipsis: a minimalist approach,” PhD thesis, Univesity of Connecticut. Kim, Min-Joo and Strauss, Uri (eds.) 2000. Proceedings of the 31st Annual Meeting of the North East Linguistics Society. Amherst, MA: GLSA Publications. Kim, Sun-Woong 2005. “On the dative intervention effect in raising to subject,” Studies in Generative Grammar 4:649–60. Kimenyi, Alexandre 1980. A Relational Grammar of Kinyarwanda. Berkeley: University of California Press. Kitahara, Hisatsugu 1993. “Deducing superiority effects from the shortest chain requirement,” Harvard Working Papers in Linguistics 3:109–120. 1997. Elementary Operations and Optimal Derivations. Cambridge, MA: MIT Press. Kluck, Marlies 2009. “Good neighbors or far friends. Matching and proximity effects in Dutch Right Node Raising,” Groninger Arbeiten zur Germanistischen Linguistik (GAGL) 48. Koppen, Mario van, Sio, Joanna, de Vos, Mark (eds.) 2002. Proceedings of ConSOLE X. University of Leiden. Kratzer, Angelika 1993. “The event argument and the semantics of voice,” manuscript, University of Massachusetts, Amherst. Kučerová, Ivona 2007. “The syntax of givenness,” PhD thesis, MIT. Kuno, Susumo 1976. “Gapping: a functional analysis,” Linguistic Inquiry 7:300– 18. Kusumoto, Kiyomi (ed.) 1997. Proceedings of the 26th Annual Meeting of the North East Lingustics Society. Amherst, MA: GLSA Publications. Kusumoto, Kiyomi and Villalta, Elisabeth (eds.) 2000. UMOP 23: Issues in Semantics and its Interface. Amherst, MA: GLSA Publications. Laenzlinger, Christopher 2005. “French adjective ordering: perspectives on DP internal movement types,” Lingua 115:645–89. Landau, Idan 1999. “Possessor raising and the structure of VP,” Lingua 107:1–37.
References
261
Lappin, Shalom and Benmamoun, Elabbas (eds.) 1999. Fragments: Studies in Ellipsis and Gapping. Oxford: Oxford University Press. Larson, Bradley 2007. “In- or ex-situ: a diagnosis of right node raising,” undergraduate honors thesis, University of Washington. Larson, Richard 1987. “Missing prepositions and the analysis of English free relative clauses,” Linguistic Inquiry 19:239–66. 1988. “On the double object construction,” Linguistic Inquiry 19:335–91. 1990. “Double objects revisited: reply to Jackendoff,” Linguistic Inquiry 21:589– 632. 1998. “Free relative clauses and missing P’s: reply to Grosu,” manuscript, Stony Brook University. Lasnik, Howard 2000. Syntactic Structures Revisited: Contemporary Lectures on Classic Transformational Theory. Cambridge: MA: MIT Press. Lasnik, Howard and Kuppin, Joseph J. 1977. “A restrictive theory of transformational grammar,” Theoretical Linguistics 4:173–96. Lavine, James E., Franks, Steven, Tasseva-Kurktchieva, Mila, Filip, Hana (eds.) 2006. Formal Approaches to Slavic Linguistics 14: The Princeton Meeting 2005. Ann Arbor, MI: Michigan Slavic Publications. Lee, Ju-Eun 2004. “Ditransitive structures and the (anti-)locality principle,” PhD thesis, Harvard University. Lee, Xah 2007. “Introduction to symmetry,” manuscript, downloadable at http:// xahlee.org/Wallpaper_dir/symmetry.html. Lee-Shoenfeld, Vera 2006. “German possessor datives: raised and affected,” The Journal of Comparative Germanic Linguistics 9:101–142. Lefebvre, Claire (ed.) 1991. Serial Verbs: Grammatical, Comparative, and Cognitive Approaches. Amsterdam/Philadelphia: John Benjamins Publishing Company. Legate, Julie 2001. “The configuration nature of a non-configurational language,” Linguistic Variation Yearbook 1:63–99. 2002. “Warlpiri: theoretical implications,” PhD thesis, MIT. 2003. “Some interface properties of the phase,” Linguistic Inquiry 34:506–16. Lema, José and Rivero, Maria-Luisa 1989. “Long Head Movement: ECP vs. HMC,” in Carter, Dechaine et al. (eds.), pp. 333–47. Levine, Robert D. 1985. “Right node (non-)raising,” Linguistic Inquiry 16:492–97. Lin, Vivian 2000. “Determiner sharing,” in Billerey and Lillehaugen (eds.), pp. 274–87. 2001. “A way to undo A-movement,” in Megerdoomian and Bar-el (eds.), pp. 358–71. Somerville, MA: Cascadilla Press. 2002. “Coordination and sharing at the interfaces,” PhD thesis, MIT. Lipták, Aniko (ed.) 2009. Correlatives Crosslinguistically. Amsterdam/ Philadelphia: John Benjamins Publishing Company. Lobeck, Anne 1995. Ellipsis: Functional Heads, Licensing, and Identification. Oxford: Oxford University Press. López, Luis and Winkler, Susanne 2003. “Variation at the syntax–semantics interface,” in Schwabe and Winkler (eds.), pp. 227–48.
262
References
Malchukov, Andrej 2004. “Towards a semantic typology of adversative and contrast marking,” Journal of Semantics 21:177–98. Maling, Joan 1972. “On gapping and the order of constituents,” Linguistic Inquiry 3:101–08. Marantz, Alec 1993. “Implications of asymmetries in double object constructions,” in Mchombo (ed.), pp. 113–50. 1997. “No escape from syntax: Don’t try morphological analysis in the privacy of your own lexicon,” in Dimitriadis, Siegel, Surek-Clark and Williams (eds.), pp. 201–25. 2001. “Words.” Paper presented at WCCFL 20, USC, February 2001. To appear. “Phases and words”. Martin, Roger and Uriagereka, Juan 2000. “Introduction: Some possible foundations of the minimalist program,” in Martin, Michaels and Uriagereka (eds.), pp. 1–29. Martin, Roger, Michaels, David and Uriagereka, Juan (eds.) 2000. Step by Step: Essays in Honor of Howard Lasnik. Cambridge, MA: MIT Press. Marušič, Franc 2009. “Non-simultaneous Spell-Out in the clausal and nominal domain,” in Grohmann (ed.) pp. 151–81. Marušič, Franc and Žaucer, Rok 2006. “On the intensional feel-like construction in Slovenian,” Natural Language and Linguistic Theory 24:1093–1159. McCawley, James 1982. “Parentheticals and discontinuous constituent structure,” Linguistic Inquiry 13:91–106. 1988. “The comparative conditional constructions in English, German and Chinese,” Berkeley Linguistics Society 14:176–87. 1998. The Syntactic Phenomena of English (2nd edn). Chicago: Chicago University Press. McCloskey, James 1986. “Right node raising and preposition stranding,” Linguistic Inquiry 17:183–86. McGinnis, Martha 1998. “Locality in A-movement,” PhD thesis, MIT. 2000. “Phases and the syntax of applicatives,” in Kim and Strauss (eds.), pp. 333–49. 2001 “Variation in the phase structure of applicatives,” Linguistic Variation Yearbook 1:105–46. 2002. “Object asymmetries in a phase theory of syntax,” in Jensen and van Herk (eds.), pp. 245–56. 2004. “Lethal ambiguity,” Linguistic Inquiry 35:47–95. 2005. “UTAH at merge: evidence from multiple applicatives,” MIT Working Papers in Linguistics 49:183–200. 2008. “Applicatives,” Language and Linguistics Compass 2:1225–45. Mchombo, Sam A. (ed.) 1993. Theoretical Aspects of Bantu Grammar 1. Stanford, CA: CSLI Publications. McNally, Louise and Kennedy, Chris (eds.) 2008. Adjectives and Adverbs: Syntax, Semantics, and Discourse. Oxford: Oxford University Press. McShane, Marjorie 2005. A Theory of Ellipsis. Oxford: Oxford University Press. Megerdoomian, Karine and Bar-el, Leora Anne (eds.) 2001. Proceedings of
References
263
the 20th West Coast Conference on Formal Linguistics. Somerville, MA: Cascadilla Press. Mikkelsen, Line 2005. Copular Clauses: Specification, Predication and Equation. Amsterdam/Philadelphia: John Benjamins Publishing Company. Miyagawa, Shigeru 1995. “Scrambling as an obligatory movement, Proceedings of the Nanzan Conference on Japanese Linguistics and Language Teaching, Nanzan University, Nagoya, Japan. 1997. “Against optional scrambling,” Linguistic Inquiry 28:1–26. Miyagawa, Shigeru and Tsujioka, Takae 2004. “Argument structure and ditransitive verbs in Japanese,” Journal of East Asian Linguistics 13:1–38. Moltmann, Friederike 1992a. “Coordination and comparatives,” PhD thesis, MIT. 1992b. “On the interpretation of three-dimensional syntactic trees,” in Barker, C. and Dowty, D. (eds.) Columbus, OH: Ohio State Univesity, Department of Linguistics, Proceedings of SALT II. Moore, John, and Perlmutter, David 2000. “What does it take to be a dative subject?,” Natural Language and Linguistic Theory 18:373–416. Moro, Andrea 2000. Dynamic Antisymmetry. Cambridge, MA: MIT Press. 2007. “Some notes on unstable structures,” manuscript, Università San Raffaele di Milano. Moulton, Keir and Wolf, Matthew (eds.) 2004. Proceedings of the 34th Annual Meeting of the North East Linguistics Society. Amherst, MA: GLSA. Muadz, Husni 1991. “Coordinate structures: a planar representation,” PhD thesis, University of Arizona. Müller, Stefan (ed.) 2006. The Proceedings of the 13th International Conference on Head-Driven Phrase Structure Grammar. Stanford: CSLI Publications. Munn, Alan 1993. “Topics in the syntax and semantics of coordinate structures,” PhD thesis, University of Maryland, College Park. 1999. “On the identity requirement of ATB movement,” Natural Language Semantics 7:421–25. 2001. “Explaining parasitic gap restrictions,” in Culicover and Postal (eds.), pp. 369–92. Muysken, Pieter, and Veenstra, Tonjes 2006. “Serial verbs,” in Everaert and van Riemsdijk (eds.), pp. 234–70. Neijt, Anneke 1979. Gapping: A Contribution to Sentence Grammar. Dordrecht: Foris Publications. Newport, Elissa, Hauser, Marc, Spaepen, Geertrui and Aslin, Richard 2004. “Learning at a distance: II. Statistical learning of non-adjacent dependencies in a non-human primate,” Cognitive Psychology 49:85–117. Ngonyani, Deogratias S. 1996. “The morphosyntax of applicatives,” PhD thesis, University of California, Los Angeles. Nunes, Jairo 1999. “Linearization of chains and phonetic realization of chain links in the minimalist program,” PhD thesis, University of Maryland, College Park. 2001. “Sideward movement,” Linguistic Inquiry 32:303–44.
264
References
2004. Linearization of Chains and Sideward Movement. Cambridge, MA: MIT Press. Oehrle, Richard T. 1976. “The grammatical status of the English dative alternation,” PhD thesis, MIT. 1987. “Boolean properties in the analysis of gapping,” Syntax and Semantics 20:201–40. Ormazabal, Javier and Juan Romero. 2007. “The derivation of dative alternation,” Paper presented at the Workshop on Argument Structure, University of the Basque Country, Vitoria-Gasteiz. Otsu, Yukio. (ed.) 2006. Proceedings of the 7th Annual Tokyo Conference on Psycholinguistics. Tokyo: Hituzi Syobo Publishing Company. Partee, Barbara H., ter Meulen, Alice & Wall, Robert E. 1993. Mathematical Methods in Linguistics, Dordrecht/Boston/London: Kluwer Academic Publishers. Pereltsvaig, Asya 2008. Copular Sentences in Russian: A Theory of Intra-Clausal Relations. New York: Springer. Pérez-Leroux, Ana Teresa and Roberge, Yves (eds.) 2003. Romance Linguistics: Theory and Acquisition. Amsterdam/Philadelphia: John Benjamins. Pesetsky, David 1982. “Paths and categories,” PhD thesis, MIT. 1995. Zero Syntax. Cambridge, MA: MIT Press. 2001. Phrasal Movement and its Kin. Cambridge: MA: MIT Press. Pesetsky, David and Torrego, Esther 2006. “Probes, goals and syntactic categories,” in Otsu (eds.), pp. 25–60. Peters, Stanley and Richie, Robert 1973. “On the generative power of transformational grammars,” Information Sciences 6: 49–83. Postal, Paul 1993. “Parasitic gaps, an across-the-board phenomenon,” Linguistic Inquiry 24:735–54. 1998. Three Investigations of Extraction. Cambridge: MA: MIT Press. 2004. Skeptical Linguistic Essays. Oxford: Oxford University Press. Progovac, Ljiljana 1998a. “Structure for coordination (part I)” GLOT International 3.7:3–6. 1998b. “Structure for coordination (part II),” GLOT International 3.8:3–9. 1999. “Events and economy of coordination,” Syntax 2:141–59. Pylkkänen, Liina 2002. “Introducing arguments,” PhD thesis, MIT. 2008. Introducing Arguments. Cambridge, MA: MIT Press. Rackowski, Andrea and Richards, Norvin 2005. “Phase edge and extraction: a Tagalog case study,” Linguistic Inquiry 36:565–99. Repp, Sophie 2008. Negation in Gapping, Oxford: Oxford University Press. Richards, Marc 2004. “Object shift and scrambling in North and West Germanic. A case study in symmetrical syntax,” PhD thesis, University of Cambridge. Richards, Norvin 2001. Movement in Language: Interactions and Architectures. Oxford: Oxford University Press. Riemsdijk, Henk van 1998. “Trees and scions – science and trees,” manuscript, Fest-Web-Page for Noam Chomsky.
References
265
2000. “Free relatives inside out: transparent free relatives as grafts,” Proceedings of the 8th Annual Conference of the Polish Association for the Study of English. 2006a. “Free relatives,” in Everaert and van Riemsdijk (eds.), pp. 338–82. 2006b. “Grafts follow from Merge,” in Frascarelli (ed.), pp. 17–44. Rivero, Maria Luisa 1991. “Long head movement and negation: Serbo-Croatian vs. Slovak and Czech,” The Linguistic Review 8:319–51. 1994. “Clause structure and V-movement in the languages of the Balkans,” Natural Language and Linguistic Theory 12:63–120. 2009. “Intensionality, high applicatives, and aspect: Involuntary state constructions in Bulgarian and Slovenian,” Natural Language and Linguistic Theory 27:151–96. Rizzi, Luigi 1986. “On chain formation,” in Borer (ed.), pp. 65–96. 1990. Relativized Minimality. Cambridge, MA: MIT Press. Rooryck, Johan 2000. Configurations of Sentential Complementation: Perspectives from Romance Languages. London: Routledge. Ross, John R. 1967. “Constraints on variables in syntax,” PhD thesis, MIT. 1970. “Gapping and the order of constituents,” in Bierwisch and Heidolph (eds.), pp. 249–59. Rudin, Catherine 2006. “Multiple wh relatives in Slavic,” in Compton, Goledzinowska and Savchenko (eds.), pp. 299–323. Rutkowski, Paweł and Maliszewska, Hanna 2007. “On prepositional phrases inside numeral phrases in Polish,” Lingua 117:784–813. Sabbagh, Joseph 2007. “Ordering and linearizing rightward movement,” Natural Language and Linguistic Theory 25:349–401. Saffran, Jenny, Hauser, Marc, Seibel, Rebecca, Kapfhamer, Joshua, Tsao, Fritz and Cushman, Fiery 2008. “Grammatical pattern learning by human infants and cotton-top tamarin monkeys,” Cognition 107:479–500. Sag, Ivan A. 1976. “Deletion and logical form,” PhD thesis, MIT. Sampson, Geoffrey 1975. “The single mother condition,” Journal of Linguistics 11:1–11. Sarkar, Anoop and Joshi, Aravind 1996. “Coordination in Tree Adjoining Grammars: formalization and implementation,” in Tsujii (ed.), pp. 610–15. Schelfhout, Carla, Coppen, Peter-Arno and Oostdijk, Nelleke 2004. “Transparent free relatives,” in Blaho, Vicente and de Vos (eds.), pp. 81–90. Schwabe, Kerstin and Winkler, Susanne (eds.) 2003. The Interfaces: Deriving and Interpreting Omitted Structures. Amsterdam/Philadelphia: John Benjamins Publishing Company. Scott, Gary-John 2002. “Stacked adjectival modification and the structure of nominal phrases,” in Cinque (ed.), pp. 91–116. Scott, Tanya 2010. “Spurious coordination in Russian multiple WH,” Paper presented at the 2010 LSA Annual Meeting, Baltimore. Seely, T. Daniel 2006. “Merge, derivational c-command, and subcategorization in a label- free syntax,” in Boeckx (ed.), pp. 182–217. Shahin, Kimary N., Blake, Susan and Kim, Eun-Sook (eds.) 1999. Proceedings of
266
References
the Seventeenth West Coast Conference on Formal Linguistics. Stanford: CSLI Publications. Siegel, Muffy 1984. “Gapping and interpretation,” Linguistic Inquiry 15:523–30. Siewierska, Anna and Hollmann, W. 2007. “Ditransitive clauses in English with special reference to Lancashire dialect,” in Hannay and Steen (eds.), pp. 83–102. Sjoblom, Todd 1980. “Coordination,” PhD thesis, MIT. Soschen, Alona 2005. “Derivation by phase: Russian applicatives,” Proceedings of the 2005 Canadian Linguistic Association Annual Conference. Sportiche, Dominique 1997. “Reconstruction and constituent structure,” manuscript, University of California, Los Angeles. Sproat, Richard and Shih, Chilin 1991. “The cross-linguistic distribution of adjective ordering restrictions,” in Georgopolous and Ishihara (eds.), pp. 565–93. Srivastav, Veneeta 1991. “The syntax and semantics of correlatives,” Natural Language and Linguistic Theory 9:637–86. Stepanov, Arthur and Stateva, Penka 2009. “When QR disobeys superiority,” Linguistic Inquiry 40:176–85. Svenonius, Peter 2005. “Extending the Extension Condition to discontinuous idioms,” Linguistic Variation Yearbook 5:227–63. (ed.) 2002. Subjects, Expletives, and the EPP. Oxford: Oxford University Press. 2008 “The position of adjectives and other phrasal modifiers in the decomposition of DP,” in McNally and Kennedy (eds.), pp. 16–42. Swan, Oscar 2002. A Grammar of Contemporary Polish. Bloomington, IN: Slavica. Szabolcsi, Anna 1994. “The noun phrase,” in Kiefer, Anderson and Kiss (eds.), pp. 179–275. Taraldsen, Knut Tarald 2002. “The que/qui alternation and the distribution of expletives,” in Svenonius (ed.), pp. 29–42. Taylor, Heather 2006. “Out on good syntactic behavior,” generals paper. University of Maryland, College Park. Tomaszewicz, Barbara 2010. “Against spurious coordination in multiple wh-questions,” Paper presented at WCCFL 20, University of Southern California. Torrego, Esther 1996. “Experiencers and raising verbs,” in Freidin (ed.), pp. 101–20. Travis, Lisa 1984. “Parameters and effects of word order variation,” PhD thesis, MIT. Tsujii, J. (ed.) 1996. Proceedings of the 16th Conference on Computational Linguistics, vol. 2. Morristown, NJ: Association for Computational Linguistics. Ura, Hiroyuki 1996. “Multiple feature checking: a theory of grammatical function splitting,” PhD thesis, MIT. Vergnaud, Jean-Roger 1974. “French relative clauses,” PhD thesis, MIT. Vitale, Anthony 1981. Swahili Syntax. Dordrecht: Foris. Vries, Mark de 2005a. “Merge: properties and boundary conditions,” Linguistics in the Netherlands 22:219–30. 2005b. “Coordination and syntactic hierarchy,” Studia Linguistica 59:83–105.
References
267
2008. “Asymmetric merge and parataxis,” Canadian Journal of Linguistics 53:355–85. 2009. “On multidominance and linearization,” Biolinguistics 3:344–403. Webelhuth, Gert (ed.) 1994. Government and Binding Theory and the Minimalist Program. Malden/Oxford: Blackwell Publishers. Wexler, Kenneth and Culicover, Peter W 1980. Formal Principles of Language Acquisition. Cambridge, MA: MIT Press. Whitman, Neal 2002. “Category neutrality: A type-logical investigation,” PhD thesis, Ohio State University. 2004. “Semantics and pragmatics of English verbal dependent coordination,” Language 80:403–34. Wiland, Bartosz 2009. “Paths to remnant movement: a single solution to three problems in the Polish OVS syntax,” Paper presented at NELS 40, MIT. Wilder, Chris 1999a. “Transparent free relatives,” in Shahin, Blake, and Kim (eds.), pp. 685–99. 1999b “Right node raising and the LCA,” in Bird, Carnie, Haugen and Norquest (eds.), pp. 586–98. 2008. “Shared constituents and linearization,” in Johnson (ed.), pp. 229–58. Williams, Edwin 1978. “Across-the-board rule application,” Linguistic Inquiry 9:31–43. 1990. “The ATB theory of parasitic gaps,” The Linguistic Review 6:265–79. Willim, Ewa 1989. On Word Order: A Government and Binding Study of English and Polish. Kraków: Zeszyty Naukowe Uniwersytetu Jagiellońskiego. Woolford, Ellen 1993. “Symmetric and asymmetric passives,” Natural Language and Linguistic Theory 11:679–728. 2006. “Lexical case, inherent case, and argument structure,” Linguistic Inquiry 37:111–30. Zabrocki, Tadeusz 1981. Lexical Rules of Semantic Interpretation. Control and NP Movement in English and Polish. Poznań: Uniwersytet im. Adama Mickiewicza. Zhang, Niina 2007. “Derivations of two paired dependency constructions,” Lingua 117:2134–58. Zoerner, Edward 1995. “Coordination: the syntax of &P,” PhD thesis, University of California, Irvine.
Index
A-bar movement 36–8, 115, 137, 189, 209 A-movement 38, 218n23 Abbott, Barbara 71 Abeillé, Anne 195–6, 202, 245n32 Abels, Klaus 70, 232n26 Abraham, Werner 238n49 Ackema, Peter 39 across-the-board (ATB) wh-questions 54–62 Japanese 58 Korean 58 Polish 57–8, 221n19 Adger, David 40, 183 adjectival order 10–11 Adjunct Condition 96, 203, 205 adjuncts 2, 172–5 affix hopping 12 Agree 7–8, 47–8 cumulative agreement 50 Multiple Agree 48–9, 60, 220n12 Aikhenvald, Alexandra Y. 93 Albanian: double object constructions 33, 35 Alexiadou, Artemis 242n15 Alsina, Alex 217n17 American English passivization 110–11 wh-movement 216n14 An, Duk-Ho 222n32 Anagnostopoulou, Elena 110, 113, 127–30, 232n23 Analyticity Parameter 14 antisymmetry 4 Antisymmetry Theory 17–20 Dynamic Antisymmetry Theory 20–3, 51 Aoun, Joseph 140 applicative constructions 32–5, 217nn17–18 see also locality and symmetric passives arguments 2, 172 Aslin, Richard 213n7 asymmetric languages 33 asymmetry 2–3, 4, 15, 17–42 Antisymmetry Theory 17–20
268
Asymmetry Theory 23–5 coordination 25–9 double object constructions 29–35 Dynamic Antisymmetry Theory 20–3, 51 in Labeling 1, 39–42 in Merge 1 in Move 1, 35–9 Asymmetry Theory 23–5 ATB: see across-the-board (ATB) wh-questions Attract Closest 36, 109 b-Merge (behindance Merge) 45 Bachrach, Asaf 68, 74 Bader, Markus 243n18 Bailyn, John 32, 119, 155, 183, 230n9 Baker, Mark 14, 93, 144, 164, 177–82, 184, 213n9, 216n13, 226n52, 232n24, 242n18, 243n19, 243n22 ban on ambiguous Attract 182 Bantu languages passivization 113 see also Ndendeule; Swahili Bare Phrase Structure Theory 19–20, 46–7, 51, 164, 173, 175, 177 Barss, Andrew 30, 37, 120–1 Basque: pied-piping 203 Bayer, Josef 243n18 Beck, Sigrid 31, 188, 189, 190, 244n28 Bernstein, Judy 11 Bhatt, Rajesh 187, 243n26 binding 72 Bodomo, Adams 17, 92–4 Boeckx, Cedric 37, 41, 207, 211, 218n22, 239n1 Boolean Phrase 221n23 Borer, Hagit 14, 165 Borsley, Robert D. 57, 195–6, 202–3, 243n32, 245n32 Boskovic, Zeljko 38–9, 58–9, 213n12 Bowers, John 183 Brasoveanu, Adrian 244n27
Index Bresnan, Joan 95, 226n55 British English: passivization 112, 126, 128, 133 Browne, Wayles 67 Bruening, Benjamin 140–2 Bulgarian: participle fronting 39 Bury, Dirk 95, 226n55 Cable, Seth 240n8 Camacho, José 27 Camdzic, Amela 39 Caponigro, Ivano 95 Carnie, Andrew 218n3 case and symmetric passives case absorption 144–6 dative intervention effects 148–51 freezing and symmetric passives 154–61 lack of multiple datives 151–4 lexical case and passive movement 146–8, 239n52 categories of syntactic objects: see Labeling Chaga: passivization 128 Chen-Main, Joan 46 Chichewa: applicatives 132, 217n17 Cho, Sungeun 58 Chomsky-adjunction 173 Chomsky, Noam 2, 4, 6, 8, 13, 19–20, 36, 40–1, 164, 172–6, 217n20, 219nn6–7, 246n39 C-I (conceptual-intentional) system 5 Cinque, Gugliemo 10, 11, 242n15 Citko, Barbara 37, 52, 54, 62–3, 66–8, 94–5, 128, 136, 147, 155, 164, 178, 183, 185–6, 191–3, 219n5, 221nn18–19, 222n30, 224n42, 224n44, 226n51, 226n55, 227n57, 227n60, 229n5, 238n51, 241n13, 243n18, 243n21, 243nn24–25, 244n29, 246–7n43 clauses free relatives 94–107 theta role assignment 2 clefting 138 clitic complementation structures 21 Collins, Chris 93, 114, 163–9, 239n5, 240n7, 242n14 Comp Account 95, 227–8n62 comparative conditionals: see comparative correlatives comparative correlatives 178, 185–90, 192–206 Dutch 188, 196–8, 247n45 English 188, 194–5, 198, 201–3 French 195–6, 198, 201–2 German 196–8, 246n41, 247n45 Greek 188, 195, 198 Hindi 198, 201
269
Hungarian 198, 247n45 Irish 246n41 Polish 186–8, 195, 198, 201–3, 244n27, 245n31, 246n38 Romanian 244n27 Spanish 196, 198, 201–2 Tagalog 246n41 Comrie, Bernard 185 concatenation 175 conceptual-intentional (C-I) system 5 Conjunction Phrase 215n10 conjunctions consecutive 80–1 contrastive 80–2 consequential serial verb construction 179, 180 Coordinate Structure Constraint 56, 67, 201 coordination 25–9, 221n19 correlative coordination 215–16n11, 221n19 covert coordination 179 node contraction 46 Coppen, Peter-Arno 97 Coppock, Elizabeth 224n44 correlates 77, 186 correlative coordination 215–16n11, 221n19 phrases 186 pronouns 186 correlatives 186, 188 comparative correlatives 178, 185–90, 192–206 equative correlatives 244n27 standard correlatives 186–7, 190–2, 200 wh-movement in correlative clauses 190–206 covert coordination 179 Croatian datives 238n49 wh-questions 63 cross-categorial symmetry 2 crossing 12 Csirmaz, Aniko 194 Cuervo, Maria Cristina 38, 111, 218n23 Culicover, Peter 30, 70, 194–5, 200–1, 212n4, 225n47, 245nn31–32 cumulative agreement 50 Czech double object constructions 32, 119, 230n9 participle fronting 39 D-structure 5 Dabrowska, Ewa 237n45 Dàgááré: serial verb constructions 92–4
270
Index
Danish: passivization 110–11 datives Croatian 238n49 German 237n46, 238n49 Hebrew 237n46 intervention effects 148–51 Japanese 238n49 lack of multiple datives 151–4 Polish 122–5, 147–8, 150–61 Russian 230n12 Slovenian 153 Dayal, Veneeta 67, 187 De Morgan’s Laws 79 De Vos, Mark 68 De Vries, Mark 45, 68, 216n12, 220n13 Den Dikken, Marcel 115, 183, 188–90, 194, 197–8, 233–4n33, 245n36 depictives 34–5, 217n18 derivational locus 167 Di Sciullo, Anna Maria 17, 23–5, 114, 214n2, 214–15n6 directionality 12 displacement: see Move Distributed Morphology 165 Dixon, Robert M. W. 93 Doggett, Teal Bissell 127–8, 131, 231n17, 232n22 dominance, full 53 Donati, Valentina 95, 227n57 Dornisch, Ewa 32, 119 double object constructions 29–35 Albanian 33, 35 Czech 32, 119, 230n9 English 30, 34, 134–7, 140, 142 Japanese 31–2, 119 Kinyarwanda 33 Luganda 35 Ndendeule 32 Polish 31–2, 115–22, 127, 142–3 Russian 32, 119, 230n9, 234n38 Dutch comparative correlatives 188, 196–8, 247n45 reflexive possessive pronouns 216n12 right node raising 223n38 Dvořák, Vera 32, 119, 230n9 Dyakonova, Marina 32, 119, 133 Dyła, Stefan 57 Dynamic Antisymmetry Theory 20–3, 51 Dziwirek, Katarzyna 112, 116–17, 122, 147–8, 151–2, 160, 229n6, 231nn14–15, 237n47 Edo: serial verb constructions 178–82, 243n20 Emonds, Joseph 134, 138, 228n2
English adjectives 10 comparative correlatives 188, 194–5, 198, 201–3 double object constructions 30, 34, 134–7, 140, 142 gapping 77, 80, 82–3, 85 passivization 110–12, 125–6, 128, 133 right node raising 71–2 wh-questions 37, 57–8 wh&wh questions 62–8 equative correlatives 244n27 Ernst, Thomas 242n15 Espinal, Maria-Teresa 195–6 Extension Condition 12, 219n6 External Merge 6, 44 see also Merge: asymmetry; Merge: symmetry feature checking/feature valuation 7–8, 47–50 Fillmore, Charles 133–4, 228n2 Fitch, Tecumseh 6, 175 Fowler, George 147 Fox, Danny 13, 213n10 Frajzyngier, Zygmunt 81 Franks, Steven 58–9 free relatives 94–107 Comp Account 95, 227–8n62 Head Account 95–6, 104, 227–8n62 standard free relatives (SFRs) 96–100, 106 transparent free relatives (TFRs) 96–9, 104–6 French adjectives 10 comparative correlatives 195–6, 198, 201–2 correlative coordination 215n11 qui/que alternation 171–2 full dominance 53 gapping 77–92 directionality 83–4 English 77, 80, 82–3, 85 German 83–4 islands 85 Japanese 83 Korean 83 lexical material 80 linguistic antecedents 82 matching 85–8 negation 79 non-constituents 82 non-coordinate contexts 82 Polish 77–8, 80–90
Index remnants 77, 80–2 restrictions 79–92 and right node raising 225n46 Russian 225–6n48 Gast, Volker 128 geometric patterns 3 Gerdts, Donna 130 German comparative correlatives 196–8, 246n41, 247n45 datives 237n46, 238n49 gapping 83–4 passivization 111, 145 right node raising 223nn37–38 wh-phrases 37 Giannakidou, Anastasia 67 glide reflection symmetry 3 Goal 7–8 Goodall, Grant 45–6, 52, 54, 78 Government and Binding theory 5 Gracanin-Yüksek, Martina 52, 62–8, 95, 219n7, 222n28, 222n30, 238n49 Greek comparative correlatives 188, 195, 198 passivization 113, 232n23 Green, Georgia M. 119, 216n13 Gribanova, Vera 62–3, 222n30 Grimshaw, Jane 95, 226n55, 240n6 Grohmann, Kleanthes 37, 232n26 Groos, Anneke 95, 227–8n62 Grosu, Alexander 95, 97, 227n59 Grosz, Patrick 50 Guimarães, Maximiliano 20 Ha, Seungwan 222n32 Haddican, William 231n19 Haegeman, Liliane 220n12 Hagstrom, Paul 67 Haider, Hubert 94, 242n18 Hale, Ken 14, 93 Halle, Morris 165 Hankamer, Jorge 82 Harley, Heidi 31, 165, 216n13 Hartmann, Katharina 222n32 Hauser, Mark 6, 175, 213n7 Head Account 95–6, 104, 227–8n62 Head Directionality Parameter 13–14 head movement 38–9 Head Movement Constraint 39 head promotion account 60–1 headless clauses see free relatives Hebrew: datives 237n46 Heck, Fabian 240n8 Higginbotham, James 67 Hindi comparative correlatives 198, 201
271
correlatives 186, 188 wh-movement 190–3 Hiraiwa, Ken 3, 16, 48–9, 60, 92–4 Hoji, Hajime 32, 119 Hollmann, W. 231n19 Holmberg, Anders 112, 150, 218n23 horizontal sharing 219n7 Hornstein, Norbert 61, 169–71, 174–5, 221n22, 222n24, 240n7, 240n9–10, 241n11 Hróarsdóttir, Thorbjörg 150, 218n23 Huang, James C.-T. 14 Huber, Stefan 188 Hudson, Richard 138, 234n34 Hughes, Arthur 128 Hungarian comparative correlatives 198, 247n45 correlatives 186, 188 subjects and possessors 2–3 wh-movement 190–4 hurdling 231n17 Icelandic A-movement 38, 218n23 Multiple Agree 48–9 passivization 112, 150–1 Inclusiveness Condition 40, 164 Indonesian: preposition stranding 71 inherent case: see quirky case Internal Merge: see Move involuntary state constructions 151, 153, 231n14 Ionin, Tania 234n38 Irish comparative correlatives 246n41 preposition stranding 71 Island Constraints 69–70, 85 Italian A-movement 218n23 correlative coordination 216n11 small clauses 178, 182–3 Izvorski, Roumyana 186, 245–6n38 Jackendoff, Ray 30, 194–5, 200–1, 212n4, 224n43, 225n45, 225n47, 242n16, 245nn31–32 Jacobson, Pauline 72 Jaeggli, Osvaldo 144 Jagodzinski, Grzegorz 235–6n42 Japanese ATB wh-questions 58 datives 238n49 double object constructions 31–2, 119 gapping 83 scrambling 213n12 Jeong, Youngmi 128, 131
272
Index
Johannessen, Janne Bondi 27, 215n9 Johnson, Kyle 31, 52, 68, 72, 78–9, 88, 103, 144, 224n44, 225n46 Joshi, Aravind 46 Kapetangianni, Konstantia 188–9, 195, 245n34 Kasai, Hironobu 78, 91 Katzir, Roni 68, 74 Kayne, Richard 13–14, 17–20, 31, 50, 91, 214n1 Kazenin, Konstantin 66, 222n31 Kim, Jeong-Seuk 83 Kim, Sun-Woong 150 Kimenyi, Alexandre 33, 114 Kinyarwanda double object constructions 33 passivization 114, 126, 135 Kitahara, Hisatsugu 67, 218n22, 236n43 Kluck, Marlies 223n38 Korean ATB wh-questions 58 gapping 83 Kucerová, Ivona 32, 119 Kuno, Susumo 80, 226n50 Kuppin, Joseph J. 219n9 Labeling 1, 8–9 Labeling: asymmetry 1, 39–42 Labeling: symmetry 16, 163–206 arguments for eliminating labels 164–75 comparative correlatives 178, 185–90, 192–206 serial verb constructions 178–82 small clauses 178, 182–5 symmetric labels 176–8 wh-movement in correlative clauses 190–206 Laenzlinger, Christopher 10 Landau, Idan 237n46 Landman, Fred 95 language faculty 4 Larson, Bradley 68, 223n38 Larson, Richard 13, 26, 30–1, 95, 137–8, 140, 144, 216n13, 226n55, 227n57, 228n2, 239n53 Lasnik, Howard 30, 120–1, 219n9 Law of Coordination of Likes 62–3, 99, 215n9, 222n27 LCA (Linear Correspondence Axiom) 17–18, 91 leapfrogging 231n17 Lebeaux, David 140 Lee, Ju-Eun 131 Lee-Schoenfeld, Vera 237nn45–46
Legate, Julie 144, 213n6, 239n4 Levine, Robert D. 72 lexical case and passive movement 146–8, 239n52 LF (Logical Form) 5, 67 LF absorption 67 Li, Yen-hui Audrey 140 Lin, Vivian 79 Linear Correspondence Axiom (LCA) 17–18, 91 linearization 50–4, 73 locality and symmetric passives 126 movement through applicative phase edge 130–5 quantifier raising from applicatives 140–4 wh-movement from applicatives 135–9 word order variation 127–30 Locus Principle 167 Logical Form (LF) 5, 67 Lohndal, Terje 220n12 long head movement 39 López, Luis 79 lowering transformations 12 Luganda: double object constructions 35 Malchukov, Andrej 80, 224n41 Maling, Joan 83 Maliszewska, Hanna 223n39 Marantz, Alec 128–9, 132, 134–5, 138, 165, 233n30, 239n4 Martin, Roger 212n4 Marusic, Franc 153 matching free relatives 96 gapping 85–8 May, Robert 67 McCawley, James 45, 68, 188, 223n34 McCloskey, James 71 McGinnis, Martha 110–11, 114, 126–7, 130, 132–3, 144, 228n1, 231n16–17, 231n19, 233n28, 233n31 Mchombo, Sam A. 217n17 McShane, Marjorie 225–6n48 Merchant, Jason 67 Merge 44 b-Merge 45 External Merge 6, 44 Internal Merge (see Move) see also Parallel Merge Merge: asymmetry 1 Merge: symmetry 15, 43–108 across-the-board wh-questions 54–62 constraints 46–7 feature checking 47–50
Index free relatives 94–107 gapping 77–92 linearization of symmetric structures 50–4, 73 Parallel Merge 44–6, 60, 68, 88, 99, 101, 219n7 right node raising 68–77 serial verbs 92–4 wh&wh questions 62–8 methodological minimalism 212n4 Mikkelsen, Line 115 Minimal Link Condition 36, 169 minimalist program 1, 4–9, 212n2, 242n18 Minimum Compliance, Principle of 246n42 mirror symmetry 3, 10 Miyagawa, Shigeru 32, 119, 238n49 Moltmann, Friederike 45, 52, 54, 78, 224n43 Moore, John 230n12 Moro, Andrea 17, 20–3, 115, 164, 178, 182–4, 214n4, 243n22 morphology 23–4, 165, 214–15n6 Move (Internal Merge) 7–8, 44 Move: asymmetry 1, 35–9 Move: symmetry 16, 109–62 case and symmetric passives 144–61 locality and symmetric passives 126–44 Polish dative subjects 122–5 Polish double object constructions 115–22 symmetric and asymmetric passives 109–15, 125–7 Muadz, Husni 43, 45, 52, 54, 71, 78 multidominance 43–6, 218–19nn3–4, 219n7 Multiple Agree 48–9, 60, 220n12 multiple specifiers 21 Multiple Spell-Out Theory: see Phase Theory Munn, Alan 26–7, 61, 215n9, 220–1n17, 221n19, 221n23, 222nn24–25 Muysken, Pieter 93 Ndendeule: double object constructions 32 negative polarity licensing 30, 72 nesting 12 Newport, Elissa 213n7 Ngonyani, Deogratias S. 32, 113–14, 132, 138–9, 234n35 Nissenbaum, Jon 13, 213n10 No Tampering Condition 12, 219n6 node contraction 46 Nonconfigurationality Parameter 14 Norwegian: passivization 112, 232n22 noun phrases Heavy NP Shift 137 theta role assignment 2
273
Noyer, Rolf 165 null operator movement 61–2 Numeration 5 Nunes, Jairo 61, 174–5, 221n22, 222n24 Oehrle, Richard T. 30, 79, 119, 216n13 ontological minimalism 212n4 Oostdijk, Nelleke 97 optimality 4 Ormazabal, Javier 216n13 parallel factors 45 Parallel Merge 44–6, 60, 68, 88, 99, 101, 219n7 parasitic gaps 61–2 participle fronting 39 passivization 30 American English 110–11 Bantu languages 113 British English 112, 126, 128, 133 case and symmetric passives 144–61 Chaga 128 Danish 110–11 English 110–12, 125–6, 128, 133 German 111, 145 Greek 113, 232n23 Icelandic 112, 150–1 Kinyarwanda 114, 126, 135 locality and symmetric passives 126–44 Norwegian 122, 232n22 Polish 112, 129, 133, 136, 142–3, 145–51 Russian 133, 147 Spanish 111, 145, 150–1 Swahili 113–14, 129, 131–2, 138–9 Swedish 232n22 symmetric and asymmetric 109, 110–15, 125–7 Pereltsvaig, Asya 164, 176, 178, 182, 184–5, 242n18 Perlmutter, David 230n12 Pesetsky, David 12, 31, 37, 41, 216n13, 217n16, 226n50 Peters, Stanley 219n8 Phase Impenetrability Condition 6 Phase Theory 4–6, 212n5 Phonetic Form (PF) 5 phrase markers: see reduced phrase markers (RPM) pied-piping 168, 203 Pinker, Steven 242n16 pivots 43, 46–7 planes 43, 218n3 Platzack, Christer 112 points of symmetry 20
274
Index
Polish 14–15 ATB wh-questions 57–8, 221n19 comparative correlatives 186–8, 195, 198, 201–3, 244n27, 245n31, 246n38 datives 122–5, 147–8, 150–61 double object constructions 31–2, 115–22, 127, 142–3 double object verbs 118 free relatives 96 gapping 77–8, 80–90 head promotion account 60–1 imperfective verbs 86 lexical case 146–7 numerals 75, 223n39 passivization 112, 129, 133, 136, 142–3, 145–51 possessive pronouns 28 preposition stranding 71 reflexive possessive pronouns 28–9 right node raising 71, 74–7 standard correlatives 186–7 wh-movement 190–2 wh-phrases 37 wh-questions 63 word order 84 Polysynthesis Parameter 14 possessive pronouns 28–9 Postal, Paul 61, 69, 133–4, 228n2 preposition stranding 71–2 Principle A effects 30 Principle of Minimum Compliance 246n42 pro-drop languages 183 Probe 7–8 Probe-Label Correspondence Axiom 41 probes (selectors) 167 Progovac, Ljiljana 25, 27–8, 215n7, 215n11, 216n11 Pylkkänen, Liina 33–4, 129, 132, 134, 217n18, 232n27, 237n46, 239n54 quantifier raising (QR) 140–4 quirky case 154, 239n52 Rackowski, Andrea 203, 205, 246nn41–2 raising transformations 12 Ramchand, Gillian 183 recursion: see Merge reduced phrase markers (RPM) 45, 219– 20n9 reflection symmetry 3, 11 reflexive possessive pronouns 28 Relational Grammar framework 231n15 relativization 138 Relativized Minimality 36, 109, 167 remnants 77, 80–2 Repp, Sophie 79, 83–4, 88
reprojection 169–71 resultative serial verb construction 179–80 Richards, Norvin 12, 203, 205, 246nn41–42 Riemsdijk, Henk van: see Van Riemsdijk, Henk Right Edge Restriction 72–3 right node raising (RNR) 68–77 Dutch 223n38 English 71–2 and gapping 225n46 German 223n37–38 Polish 71, 74–7 Right Roof Constraint 70 Ritchie, Robert 219n8 Rivero, Maria Luisa 231n14, 238n48 Rizzi, Luigi 36–8, 171 RNR: see right node raising Roberts, Ian 144 roll-up movement 11 Romanian: comparative correlatives 244n27 Romero, Juan 216n13 Rooryck, Johan 171 Ross, John R. 26, 69–71, 83, 224n45 rotation symmetry 3 RPM (reduced phrase markers) 45, 219– 20n9 Rudin, Catherine 186 Russian copular constructions 184–5 dative subjects 230n12 double object constructions 32, 119, 230n9, 234n38 gapping 225–6n48 passivization 133, 147 small clauses 178 wh-questions 63 Rutkowski, Paweł 223n39 S-structure 5 Sabbagh, Joseph 69–70, 72–4, 222n33 Saffran, Jenny et al. 213n7 Sag, Ivan 82 Sarkar, Anoop 46 Schelfhout, Carla 97 Schmid, Tanja 243n18 Scott, Tanya 62–3 Seely, T. Daniel 163 selectors 167 sensorimotor (SM) system 5 Serbo-Croatian correlative coordination 216n11 participle fronting 39 wh-phrases 37 serial verb constructions Labeling 178–82 Merge 92–4
Index set theory 4 SFRs (standard free relatives) 96–100, 106 Shortest Move 109 sideward movement 59–60, 62 Siegel, Muffy 79 Siewierska, Anna 231n19 Simpler Syntax Hypothesis 212n4 Single Root Condition 60, 101 Sjoblom, Todd 83, 224n40 Slovenian: datives 153 sluicing 67 SM (sensorimotor) system 5 small clauses 21, 178, 182–5, 214n4 SMT (Strong Minimalist Thesis) 4–5 Soschen, Alona 133 Spaepen, Geertrui 213n7 Spanish A-movement 38 comparative correlatives 196, 198, 201–2 passivization 111, 145, 150–1 Spell-Out 74, 212n5 Srivastav, Veneeta 187 standard correlatives 186–7, 190–2, 200 standard free relatives (SFRs) 96–100, 106 Stateva, Penka 142, 234n38 Stepanov, Arthur 142, 234n38 Stewart, Osamuyimen T. 164, 177–82, 184, 226n52, 242n18, 243n19, 243n22 Stone, Gerald 185 Strong Minimalist Thesis (SMT) 4–5 Superiority Condition 36–7 superraising 38 Svenonius, Peter 218–19n4 Swahili: passivization 113–14, 129, 131–2, 138–9 Swan, Oscar 116, 147 Swedish: passivization 232n22 swiping 67–8 symmetric languages 33 symmetric passives: see case and symmetric passives symmetry 1, 2–4 other sources 9–14 see also Labeling: symmetry; Merge: symmetry; Move: symmetry Symmetry Parameter 14 Szabolcsi, Anna 3 TAG (Tree Adjoining Grammar) 46 Tagalog: comparative correlatives 246n41 Takahashi, Daiko 213n12 Taraldsen, Knut Tarald 241n12 Taylor, Heather 188–90, 195, 245n34 TFRs (transparent free relatives) 96–9, 104–6 theoretical framework 4–9
275
theta roles 2 Tomaszewicz, Barbara 63 topicalization 138 Torrego, Esther 41 tough movement 12, 138 transformations 3 translation symmetry 3, 11 transparent free relatives (TFRs) 96–9, 104–6 Travis, Lisa 39 Tree Adjoining Grammar (TAG) 46 Trudgill, Peter 128 Tsujioka, Takae 32, 119, 238n49 Ura, Hiroyuki 128 Uriagereka, Juan 169–71, 212n4, 240n7, 240nn9–10, 241n11 Van Riemsdijk, Henk 94–5, 97, 100–1, 105, 219n4, 227–8n62 variable binding 30, 72 Veenstra, Tonjes 93 Vehicle Requirement on Merge 41 verbs Polish double object verbs 118 Polish imperfective verbs 86 serial verb constructions 92–4, 178–82 vertical sharing 219n7 Vicente, Luis 68 Vitale, Anthony 129 Vries, Mark de: see De Vries, Mark Warlpiri 14 weak crossover 30 West Flemish: Agree 220n12 Wexler, Kenneth 70 wh-movement 12 American English 216n14 from applicatives 135–9 in correlative clauses 190–206 Hindi 190–3 Hungarian 190–4 Polish 190–2 wh-phrases 37–8 wh-questions 36–7 across-the-board (ATB) 54–62 Croatian 63 English 37, 57–8, 62–8 Japanese 58 Korean 58 Polish 57–8, 63, 221n19 Russian 63 wh&wh questions 62–8 Whitman, Neal 63–4 Whitney, Rosemary 138 Wiland, Bartosz 230n11
276
Index
Wilder, Chris 52–3, 68, 73, 97 Williams, Edwin 45, 50, 52, 54, 219n8, 221n22 Willim, Ewa 192, 230n11 Winkler, Susanne 79 Woolford, Ellen 111, 129, 144, 154, 239n52 word order 84, 127–30
X-bar theory 46–7, 172 Zabrocki, Tadeusz 147 Zaucer, Rok 153 Zhang, Niina 222n30 Zhou, Xuan 258 Zoerner, Edward 26, 27