The Interfaces
Linguistik Aktuell/Linguistics Today Linguistik Aktuell/Linguistics Today (LA) provides a platform for...
140 downloads
339 Views
2MB 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
The Interfaces
Linguistik Aktuell/Linguistics Today Linguistik Aktuell/Linguistics Today (LA) provides a platform for original monograph studies into synchronic and diachronic linguistics. Studies in LA confront empirical and theoretical problems as these are currently discussed in syntax, semantics, morphology, phonology, and systematic pragmatics with the aim to establish robust empirical generalizations within a universalistic perspective.
Series Editor Werner Abraham University of Vienna
Advisory Editorial Board Guglielmo Cinque (University of Venice) Günther Grewendorf (J.W. Goethe-University, Frankfurt) Liliane Haegeman (University of Lille, France) Hubert Haider (University of Salzburg) Christer Platzack (University of Lund) Ian Roberts (Cambridge University) Ken Safir (Rutgers University, New Brunswick NJ) Lisa deMena Travis (McGill University) Sten Vikner (University of Aarhus) C. Jan-Wouter Zwart (University of Groningen)
Volume 61 The Interfaces: Deriving and interpreting omitted structures Edited by Kerstin Schwabe and Susanne Winkler
The Interfaces Deriving and interpreting omitted structures Edited by
Kerstin Schwabe ZAS, Berlin
Susanne Winkler University of Tübingen
John Benjamins Publishing Company Amsterdam/Philadelphia
8
TM
The paper used in this publication meets the minimum requirements of American National Standard for Information Sciences – Permanence of Paper for Printed Library Materials, ansi z39.48-1984.
Library of Congress Cataloging-in-Publication Data The interfaces : deriving and interpreting omitted structures / edited by Kerstin Schwabe, Susanne Winkler. p. cm. (Linguistik Aktuell/Linguistics Today, issn 0166–0829 ; v. 61) Includes bibliographical references and indexes. 1. Grammar, Comparative and general--Ellipsis. 2. Grammar, Comparative and general--Syntax. 3. Semantics. I. Schwabe, Kerstin. II. Winkler, Susanne, 1960- III. Linguistik aktuell ; Bd. 61. P291.3 I58 2003 415-dc21 isbn 9027227845 (Eur.) / 1588113302 (US) (Hb; alk. paper)
2002034217
© 2003 – John Benjamins B.V. No part of this book may be reproduced in any form, by print, photoprint, microfilm, or any other means, without written permission from the publisher. John Benjamins Publishing Co. · P.O. Box 36224 · 1020 me Amsterdam · The Netherlands John Benjamins North America · P.O. Box 27519 · Philadelphia pa 19118-0519 · usa
Table of contents
Exploring the interfaces from the perspective of omitted structures Susanne Winkler and Kerstin Schwabe
1
I. Towards the exploration of PF-deletion accounts Ellipsis and syntactic representation Christopher Kennedy
29
Subject-auxiliary inversion in comparatives and PF output constraints Jason Merchant
55
Antecedent-containment and ellipsis Chris Wilder
79
Background matching in right node raising constructions Katharina Hartmann
121
II. From the computational system to the syntax-semantics interface Merge copy Caterina Donati
155
Phrase structure paradoxes, movement and ellipsis Winfried Lechner
177
Unpronounced heads in relative clauses Uli Sauerland
205
Variation at the syntax-semantics interface: Evidence from gapping Luis López and Susanne Winkler
227
Table of contents
III. The semantic component and its connection to focus and discourse structure Ellipsis and the structure of discourse Daniel Hardt
251
Correlate restriction and definiteness effect in ellipsis Maribel Romero
263
F-marking and specificity in sluicing constructions Kerstin Schwabe
301
The semantics of Japanese null pronouns and its cross-linguistic implications 321 Satoshi Tomioka Omission impossible? Topic and focus in focal ellipsis Petra Gretsch
341
References
367
Name index
389
Subject index
395
Exploring the interfaces from the perspective of omitted structures* Susanne Winkler and Kerstin Schwabe
.
Introduction
The investigation of the structure and the precise nature of the interfaces which connect the sound and the meaning components to the core computational system has been located at the center of the research in contemporary generative linguistics. Traditionally, the interfaces comprise the syntax-semantics and the syntaxphonology interfaces. In recent years, however, the scope of investigation has been considerably expanded by asking how the syntax-semantics and the syntaxphonology interfaces connect with the discourse and the information structural component. The prevailing questions are: What exactly counts as evidence, and what counts as an argument for a particular set-up of the interfaces? This is where the enterprise of this book begins: the investigation of elliptical phenomena provides crucial data for the theories that are concerned with the architecture of the interfaces, because ellipsis lacks form and still has meaning. The theories of ellipsis in this volume are concerned with questions of the representation and interpretation of ellipsis and the functions of information structure. The most crucial question is: How do the syntax-phonology, the syntax-semantics, and the information structural component interact in the context of ellipsis? Although the research in this area has grown to incorporate a greater variety of data, it has become increasingly clear that many questions about the architecture of grammar cannot be answered by following a strict syntactic, semantic, phonological, or discourse account. The theoretical complexity of the study of interfaces and the rapid changes in leading theories like the Principles and Parameters Framework (Chomsky 1981) and the Minimalist Program (Chomsky 1995) require that syntacticians, semanticists, and phonologists collaborate in the exploration of the interfaces. Collaboration is needed in the investigations of the whole range of available data, including evidence from focus and intonational data. The development of a comprehensive theory of meaning which also incorporates insights from the study of information
Susanne Winkler and Kerstin Schwabe
structure is essential for a comprehensive theory of ellipsis. The chapters in this volume bring together the expertise of semanticists and syntacticians with longstanding interests in the syntax-semantics-phonology interfaces and particularly their interrelatedness with focus and information structure. In the classic view set out by Chomsky (see Chomsky 1965, 1995, 2000, 2001), the syntactic component of the grammar (also called the computational system) accounts for the matching of sound and meaning (see Jackendoff 1998; Platzack 2000; López in preparation, for opposite views). Under this conception, one of the most central issues has been to find an explanation of how it is possible for speakers and listeners to interpret ellipsis in the absence of form, in some sense or other, such as in the case of VP ellipses in English, as in (1), or in the case of gapping in German, as exemplified by (2). (1) a. John voted for himself and his lawyer did too. b. John voted for himself, because nobody else did. (2) Die Hella hat mir geschrieben, dass sie sich langweilt ohne mich; the Hella has me written that she herself is-bored without me; ich mich auch. I myself too ‘Hella has written that she is bored without me; me too.’ In this respect the mere occurrence of ellipsis is a challenge to our understanding of the architecture of grammar (in current syntactic terms as in (3)), where the most powerful constraint (the former Projection Principle) is that grammatical information cannot simply be lost on its way from the computational system to the interfaces, in (3) referred to as phonological form (PF) and logical form (LF). (3)
Part of the answer to the question of how words or phrases can be understood without being pronounced (or otherwise referred to) already lies in the way that we assume elliptical material is represented and where. Within the current theory of grammar, three prevailing research paradigms can be isolated: the phonological deletion theory, the so-called syntax-first approach, and the semantic theory. The phonological deletion theory assumes that elliptical material is fully syntactically represented, but deleted at a certain point in the derivation of the sentence. The syntax-first approach investigates the computational system proper and aims at an answer to the question of whether ellipsis can be reduced to movement or other independently existing syntactic processes. Although the starting point of any syntax-
Exploring the interfaces from the perspective of omitted structures
first methodology is, as the name says, essentially syntactic, the goal is to find out exactly which parts of the analysis of omitted structures must be handed over to the interfaces. The semantic theory assumes that semantic mechanisms governing the recovery of meanings from discourse can be put to work in the analysis of elliptical constructions (e.g., the proform theory). Closely connected to the question of how ellipsis is represented is the question of how information structure and focus contribute to the interpretation of ellipsis. Assuming the grammatical model in (3), it is difficult to see how and exactly where the information structural component and the theory of focus should come into play. There are basically four different approaches that address the integration of the information structural component in the theory of grammar: The first approach is essentially syntactic and assumes that focus movement, like wh-movement, is interpreted at the syntax-semantics interface (see Platzack 2000; Uriagereka 1999). The second is an essentially semantic approach and finds its instantiation either in Rooth’s (1992b) theory of Alternative Semantics or in Schwarzschild’s (1999) theory of Givenness, both of which operate at the semantic component of grammar. The third approach assumes that the semantic component interfaces with information structure (e.g., Vergnaud & Zubizarreta 1992; Zubizarreta 1998; see Winkler & Göbbel 2000). Vallduví (1992), finally, assumes that the information structural component is an interface separate from PF and LF, possibly also located outside the grammatical model proper (see also Vallduví & Vilkuna 1998). The contributions in this volume fall into three groups. The groups are arranged according to their theoretical point of view and according to the interface problem that they investigate and solve. The papers in the first group (Kennedy, Hartmann, Merchant, and Wilder) pursue a PF-deletion account. Kennedy’s paper starts out by directly addressing the question of how much syntax, how much phonology and semantics are involved in ellipsis. On the basis of the interaction of ellipsis with several different grammatical constraints he proposes that elided constituents have syntactic structure that is deleted at PF. At PF, deletion is understood as an instruction to bypass morpho-phonological instantiation in the sense of Wasow (1972). The papers by Merchant, Wilder, and Hartmann discuss different types of elliptical constructions (comparative deletion, antecedent-contained deletion and right node raising) and show that their specific behavior can only be adequately accounted for under a PF-deletion account. The second group (Donati, Lechner, Sauerland, López & Winkler) explores the possibility that certain kinds of ellipsis should be analyzed using the well-known machinery of syntactic displacement. Donati investigates the idea that movement and ellipsis may turn out to be the same operation. Lechner addresses a similar problem. He suggests that VP-fronting and VP-deletion are essentially the same, and that their apparent differences are derivable from principles of economy.
Susanne Winkler and Kerstin Schwabe
Sauerland assumes that one type of relative clauses, namely the matching relative clause, involves ellipsis and not movement. López and Winkler take up Johnson’s (1997a) suggestion that gapping does not involve deletion, but rather Across-TheBoard (ATB) movement of the verb and use it as a platform to investigate possible sources of linguistic variation. The third group of papers sets out to explore the semantic component and its connection to focus and discourse structure (Hardt, Romero, Schwabe, Tomioka, and Gretsch). In contrast to the PF-deletion accounts, Hardt rejects the claim that unpronounced material has a syntactic representation, and he suggests that a semantic matching condition that has access to discourse structure governs the recovery of the meaning of the elliptical site. Romero and Schwabe investigate constructions such as sluicing and reduced conditionals and claim that semantic restrictions on their wellformedness can be explained in terms of the semantics of questions and focus. Tomioka studies the nature of phonologically silent pronouns in Japanese. He proposes that they are simply property anaphora and that their different uses correspond to the same semantic operations that affect full-fledged NPs. Gretsch investigates the gaps in operatorless questions such as The keys are _? and proposes that what is missing in these sentences is the focus of the question. The major emphasis of each contributor has been placed on developing a theory of how syntactic conditions on chains and copies, semantic identity conditions, and discourse and focus conditions interact in the deletion process. The syntactic accounts are couched in the generative framework of Chomsky (1995, 2000), and the semantic accounts are set in Discourse Representation Theory and Alternative Semantics. The range of evidence involves new data on ellipsis from English, German, Japanese, Italian and French, as well as new intonational data.
. Burning issues in the study of ellipsis The goal of this volume is to present an overview of the current state of the art in the research of ellipsis and omitted elements. Certainly, this cannot be done without also taking into account the changes in the research paradigms of the leading theories over the last three decades. The leading idea of this section is that we are witnessing a paradigm change from the so-called syntactic deletion theories, integrated in the early generative theory of transformational grammar, to the licensing theories of ellipsis within the Principles and Parameters Framework, and back again to phonological deletion theories hosted within the Minimalist Framework, also basically a generative theory. In the second part of this section, we take up the question of what counts as an argument for a particular set-up of the interfaces. There we shall take a look at one
Exploring the interfaces from the perspective of omitted structures
single piece of evidence from gapping that undermines the hypothesis of a simple version of the PF-deletion account.
. A historical perspective A central issue in the research on elliptical constructions has been to find out what sorts of representations are involved in the resolution and licensing of elided material. The three central questions are given in (4): (4) i. Does ellipsis have internal structure? ii. How is ellipsis interpreted? iii. What role does information structure play in the structural representation and interpretation of ellipsis? Two lines of thought, differing in their assumptions about the role of syntax in the ellipsis, have been predominant: deletion approaches vs. non-deletion approaches. The deletion approaches comprise the syntactic deletion vs. the PF-deletion approach, in essence, a historical distinction which reflects the paradigm change within the generative theory. The non-deletion approaches comprise interpretive accounts and syntactic movement accounts. The different approaches are listed schematically in (5):1 Taxonomy of approaches to ellipsis
(5) deletion
no deletion
syntactic deletion
PF-deletion
interpretive
e.g., Ross 1967;1969 Postal 1970, 1972 Sag 1976a, 1976b Hankamer & Sag 1976 Hankamer 1979
e.g., Chomsky & Lasnik 1993 Tancredi 1992 Klein 1993 Lasnik 1999b Hartmann 2000, itv Kennedy & Merchant 2000a Merchant 2001, itv Kennedy itv Wilder 1997d, itv Sauerland 1998a, itv
Neijt 1979 Proform Theory: e.g. Johnson 1997a e.g., Hardt 1993, itv Dalrymple et al. 1991 Lechner 1999, itv Donati itv Reconstruction: López & Winkler itv Williams 1977a Sauerland itv Fiengo & May 1994 Lappin 1996 Proform and Reconstr. Chao 1988 Lobeck 1995, 1999 Chung et al.1995 López & Winkler 2000 Semantic Focus Licensing Rooth 1992a Schwarzschild 1999 Romero 1998, itv Schwabe 1994, 2000, itv Tomioka 1998, itv
syntactic movement
Susanne Winkler and Kerstin Schwabe
The earliest generative approaches to anaphora and ellipses were syntactic deletion approaches in the pre-Aspects model proposed by, e.g., Ross (1967, 1969) and Postal (1970, 1972). The proponents of this approach, which is also called the strict transformational position, claim that “all anaphoric processes are transformations that involve deletion” (Hankamer & Sag 1976: 394). This deletion process is assumed to occur in the derivation of S-structure and not, as is proposed by the more recent deletion approaches, at the level of PF. The strict transformational position is opposed by the proponents of the strict interpretive position, whose strongest arguments against the transformational position include: the Bach-Peter Paradox (Bach 1970); Bresnan’s (1970) arguments showing that certain rules (such as there-insertion, dative movement, particle movement, and the nuclear stress rule) are sensitive to properties that differentiate full NPs from pronouns; Dougherty’s (1969) observations that a transformational account of pronominalization whose antecedents contain quantifiers is problematic; and also Wasow’s (1979) observation that a transformational approach to pronominalization derives certain ungrammatical sentences which violate Binding Principle C. The core idea of the strict interpretive position (cf. Wasow 1972, 1979; Shopen 1972) is that “all anaphors (pronominal or null) are present in the underlying representation, and that no anaphors are derived transformationally” (Hankamer, & Sag 1976: 394). Wasow (1979) aims at an unified approach of all anaphora types, including the so-called null-anaphora type. His basic claim is “that null-anaphora underlie the same restrictions as other anaphors with the difference that null-anaphors are generated with no phonetic realization” (p. 105). Trying to avoid the criticism advanced against Akmajian’s (1968) non-expansion hypothesis, which claims that “null anaphora ∆ have no internal structure” (cf. Ross 1969), Wasow proposes the empty structure hypothesis, which assumes that “null anaphors have all the structure of their antecedents, lacking only phonetic material” (p. 109). If we consider Wasow’s proposal from today’s perspective, we can conclude that it classifies as a direct historical precursor of the modern PF-deletion proposals. More generally, modern discussion of omitted structures recapitulates to a certain extend the discussion of the very early approaches to ellipsis. The strict transformational and the strict interpretive positions were superseded by the mixed approaches, most notably Hankamer and Sag’s (1976) deep vs. surface anaphora proposal and Williams’ (1977a) discourse vs. sentence grammar proposal. Hankamer and Sag’s proposal is based on the insight that anaphoric processes fall into two classes, the deep anaphora cases and the surface anaphora cases. Only the latter, which are syntactically controlled, are claimed to be derived by a deletion process, whereas the former, which are pragmatically controlled, are basegenerated elements subject to an interpretive process. Pronominal anaphora, sentential it, and null complement anaphora are categorized as deep anaphora, whereas VP-ellipsis, sluicing, gapping, stripping, and so-anaphora are categorized as surface
Exploring the interfaces from the perspective of omitted structures
anaphora. That is, most types of ellipses (except the null complement anaphora) are assumed to be derived by a deletion process. In addition to the fact that there are various well-known counterexamples to the classification proposal of the mixed approaches (see Williams 1977b; Chao 1988; Hardt 1993), one feature of this proposal has come under particularly heavy attack, namely, the fact that the different types of anaphora do not receive a uniform treatment. This is the point at which Williams’ (1977a) discourse vs. sentence grammar proposal steps in. Instead of assuming different derivational processes for different types of anaphora, he adopts Wasow’s (1972, 1979) empty structure hypothesis, which applies to all types of anaphora, and proposes that they are subject to different rule types which operate on two distinct subsystems of the grammar: Those rules like Comparative Deletion and Gapping that are sentence-bound and obey Ross’s constraints are called Sentence Grammar rules. These rules define the form and meaning of sentences. Those rules which, like VP Deletion, apply across sentences in a discourse and which do not obey Ross’s constraints we will assign to a component called Discourse Grammar. (Williams 1977a: 102)
Williams explicitly opposes the deletion approach and argues in favor of the interpretive hypothesis, which assumes that an empty element in the surface representation receives its interpretation at the newly introduced level of LF by copying the lexical material of the antecedent into the position of the empty element. The idea is that sentence grammar (SG) comprises the processes at DS and SS, including the rules deriving LF representations. Discourse grammar (DG) rules, on the other hand, apply exclusively to LF representations provided by sentence grammar. Thus, gapping and comparative ellipsis count as phenomena of SG and are subject to a rule of semantic interpretation whereas VP-ellipsis counts as a rule of DG that establishes an anaphoric relation between a string minimally including an empty V and the V in the preceding conjunct and copies the material of the antecedent V into the empty V at LF. The level of LF in this model forms an interface representation between SG and DG, at which Williams’ representation of ellipsis is situated. Many later works on ellipsis concentrate on the study of either the SG component (Zagona 1982; Chao 1988; Lobeck 1995; Chung et al. 1995) or the DG component (Klein 1987; Hardt 1993; Dalrymple et al. 1991) of specific elliptical constructions. Some of the contributions in this book present their own proposals (López & Winkler, Romero, Schwabe, Hardt, Tomioka), which characterize VP-ellipsis, IP ellipsis, NP ellipsis and gapping as being located at the interface between SG and DG, like Williams’ original proposal. Translated into a modern framework, these proposals are situated at the syntax-semantics interface.
Susanne Winkler and Kerstin Schwabe
Since Williams (1977a), a third component has come into play, whose relation to both SG and DG is subject to much discussion. This component is PF, and the arguments of the studies which propose that the deletion process involved in ellipsis is situated at this level are as strong as those for the original syntactic deletion accounts. Moreover, the PF-deletion process assumed in Chomsky and Lasnik (1993), Chomsky (1995), Tancredi (1992), and Klein (1993), among others, has the advantage that it leaves the syntactic representation of the sentence completely intact for the interpretation at LF. PF-deletion accounts rest on the assumption that ellipsis has abstract features of phonetic and phonological representation, but that grammar contains a means of blocking its pronounciation in the surface form (here understood as the actual realization). Although this approach seems close to being unrefutable (since the assumption that every type of ellipsis involves the absence of phonological form is trivially true), there are still some questions unsolved: What exactly triggers the (optional) deletion process? Are the triggers syntactic or pragmatic wellformedness conditions? If the trigger is pragmatic, does that mean that PF has access to the information structural component? Furthermore, when does PF-deletion actually apply? Before or after verb movement, quantifier movement, focus phrase movement? What is the exact order between movement and PF-deletion? Many of these questions are investigated in the context of this book. Before we turn to the individual proposals in this book, let us briefly address the question of what counts as an argument for a particular arrangement of the interfaces. A paradigmatic case is presented in the next Section 2.2.
. Evidence for the architecture of the interfaces from ellipsis In this section, we present a puzzling case from gapping and focus that, to our knowledge, has not been observed so far. Let us take up the three central questions from above, repeated in (6), and look at the relevant gapping example in (7) (capitalization signals focus): (6) The three central questions are: i. Does gapping have internal structure? ii. How is gapping interpreted in the semantic component? iii. What role does information structure play in the structural representation and interpretation of gapping? (7) She wrote long [. . . ] letters, which she sent to her SISTER and SHE to my MOTHER. (Ruth Prawer Ihabrala (1975) Heat and Dust) The first question – does gapping have internal structure – can be straightforwardly answered in the affirmative. Two arguments underlie this assumption: one based
Exploring the interfaces from the perspective of omitted structures
on the Parallel Structure Hypothesis, the other based on the traditional Island Constraints (Ross 1967). The first argument is based on the assumption that if the antecedent of a gap in the first conjunct contains a wh-trace, the gap itself contains a wh-trace, as is the case in (8) (missing material appears in strike-through): (8) She wrote long lettersi , whichi [shek [sent ti ] to her SISTERj ] and [SHEj [sent ti ] to my MOTHER]. The second argument is based on the observation that gapping is subject to island constraints and must therefore contain syntactically represented material, just like wh-constituents. The ungrammaticality in (9b) results from the violation of the Complex NP Constraint. (9) a. SHE asked which LETTERS we wrote and HE which BOOKS. b. *SHE discussed my question which LETTERS we wrote and HE which BOOKS. It is not possible for a wh-phrase to be moved out of a complex NP in (9b). Island Constraint violations are generally assumed to constitute a strong argument for movement, and if movement out of an elliptical construction is barred, it is an argument for the assumption that the elided material still has a syntactic effect. Now let us turn to questions (6ii-iii) above: How is ellipsis interpreted, and what role do focus and information structure play. Consider (10), a gapping example derived by a corpus-based search of the Penn Treebank (cf. Marcus et al. 1993): (10) It’s probably true . . . that the system is so hierarchical that only the assistant manager can talk to the manager and the manager to the general manager. . . (Penn Treebank, #. . . wsj_0037.mrg 805) A typical pitch extraction contour is given in Figure 1. The contour shows that there are four focus accents in the utterance: a falling accent on ASSISTANT and on MANAGER in the first conjunct, and a falling accent on MANAGER and on GENERAL manager in the second conjunct. The Beckman and Elam (1997) notation is given in (11) below (the reflex of pitch accents is signaled by capitalization): (11)
H* LH* L- L% only the ASSISTANT manager can talk to the MANAGER H* LH* LL% and the MANAGER to the GENERAL manager
Susanne Winkler and Kerstin Schwabe
Figure 1.
The result of either a syntactic deletion account (e.g., Ross 1970) or a PF-deletion account (e.g., Hartmann 2000, and in this volume) would suggest for (12) the simplified structure in (13): (12) [[S1/TP. . . the ASSISTANT manager can talk to the MANAGER] and [S2/TP the MANAGER can talk to the GENERAL manager]]. (13)
S S1
and
S2
In (13), two clauses are coordinated. It is generally assumed that in addition to a parallel syntactic structure gapping also requires a parallel focus structure, as expressed in the contrastive focus principle in (14): (14) Contrastive Focus Principle: In gapping, the deleted elements must belong to the background. The remnants must occur in a contrastive relation to their correlates. (adopted from López & Winkler, in this volume)
Exploring the interfaces from the perspective of omitted structures
Principle (14) applied to the manager example in (10) provides the following picture: There are parallel foci, and the remnants (MANAGER, GENERAL manager) occur in a contrastive relationship to their correspondents in the first conjunct (ASSISTANT manager, MANAGER). Within the PF-deletion account it is possible to argue that the given part of a sentence is phonologically reduced, or p-reduced, as proposed by Klein (1993): (15) Exactly those lexical elements that constitute a maintained topic can be p-reduced. (Klein 1993: 791; our translation) The term maintained topic refers to the background of the sentence and is generated by replacing the focused constituents with a variable and building up the resulting structure at LF. The maintained topic in example (10) is the string can talk, as seen in (16). (16) It’s probably true . . . that the system is so hierarchical that only [x can talk to y and y can talk to z]. . . A straightforward test that shows that the modal verb can and the matrix verb talk are actually maintained topics in Klein’s sense is shown when (12) is used as an answer to a multiple wh-question, as in (17): (17) A: Who can talk to whom? B: The ASSISTANT manager can talk to the MANAGER and the MANAGER can talk to the GENERAL manager. The wh-phrases in (17) specify the sets of alternatives from which the relevant set must be chosen; in the case of (17B), the answer specifies two pairs namely the pair assistant manager and manager and the pair manager and general manager. Up to this point, we have intentionally ignored the function of only in (10). Let us look at it more carefully, as represented in (18). (18) It’s probably true . . . that the system is so hierarchical that only the ASSISTANT manager can talk to the MANAGER and the MANAGER to the GENERAL manager. . . It is generally assumed that the focus sensitive particle only is associated with the focus (cf. semantic works ranging from Jacobs 1983, 1986, over Rooth 1985, 1992b, and Tancredi 1990, to von Stechow 1991, and syntactic works ranging from Anderson 1972, over McCawley 1996 and Kayne 1998, to Büring & Hartmann 2001). That is, only is a focus particle that takes as its associated focus a prosodically marked element that is implicitly contrasted with alternatives, as for example in (19):
Susanne Winkler and Kerstin Schwabe
(19) a. Only John came. b. John only introduced Bill to Sue.
(but not BILL) (but not his MOTHER)
Under the assumption that p-reduction of the maintained topic at PF in (18) depends upon the focused remnants in the second conjunct, then it could be assumed that the complete syntactic representation is handed over to the LF-component. A simplified structure of example (18) is provided in (20). Note that the representation follows Rooth’s (1985: 28, Chapter 3) original assumption that only is part of the NP (DP) constituent: S
(20) S DP
and M
FP
DP
only
the ASSISTANT manager
VP
DP FP
can
talk to the MANager
S
only
M
VP
DP the MANager
can
talk to the GENeral manager
In (20), p-reduction, according to Klein, would affect the focus particle only, the modal verb can and the main verb talk in the second conjunct. The question that we have not addressed so far is: What is a possible paraphrase of the manager example as represented in (20)? A natural paraphrase is given in (21): (21) P1: As for the MANAGER, only the ASSISTANT manager can talk to him; and as for the GENERAL manager, only the MANAGER can talk to him. If the paraphrase in (21) models our interpretation, then this sentence contains a pair of contrastive foci, as well as two foci that are associated with only. However, for most informants example (18) allows only the paraphrase, which is given in (22): (22) P2: There are only two pairs (x, y) with the property that x can talk to y – namely, the pair (assistant manager, manager) and the pair (manager, general manager) and no other pair. We can see immediately that the paraphrases P1 and P2 actually differ when we test them with an additional focus pair, such as the pair (assistant manager, executive director). This pair is excluded by the paraphrase P2, but not by P1. The relevant observation is the following: the reading P2 cannot be explained under a straightforward PF-deletion account, as it is represented in (20).2 However,
Exploring the interfaces from the perspective of omitted structures
it can be explained under the so-called syntax-first approach that assumes that a contrastively focused phrase must be moved to an A-bar position in overt syntax and is then interpreted at the syntax-semantics interface. A very brief sketch of such an account of (18) is given below. The idea is the following: Independent evidence suggests that gapping involves the coordination of vPs and not of sentences (see Johnson 1997a as well as López & Winkler, in this volume). Thus, gapping is derived by two independent movement operations: focus movement in a Spec,v-position with subsequent ATB-movement of the vacated vP. The focus particle only in (18) does not take a DP, but functions as a quantifying sentential adverb in the sense of Jacobs (1983) and Büring and Hartmann (2001) and is adjoined to vP in (23). From this position, only has scope over the coordination and therefore over the pairs (assistant manager, manager) and (manager, general manager). (23)
TP
DP
T’ T can
XP FP
only
XP vPx talk
vP and
vP
DP
vP DP
vP
the ASSISTANT manager1 PP to the MANager2
vP
the MANager3
vPx t1
PP t2
to the GENeral manager4
vPx t3
t4
Note that the structure in (23) is not the surface structure of (18). The subject the assistant manager and the focus particle only must somehow form a phrase and move to the canonical subject position in English. The above exemplary discussion of a single example is not intended to show that there is no PF-deletion. PF-deletion is needed for many different deletion operations. The question, however, is how much of the explanatory force do we
Susanne Winkler and Kerstin Schwabe
actually need to load onto the PF-component in the discussion of elliptical phenomena, and how do the deletion operations interact with the information structural component and focus? These are two of the major questions addressed by the contributors to this volume.
. The chapters . Towards the exploration of PF-deletion accounts Since Chomsky and Lasnik (1993) as well as Tancredi (1992) introduced the phonological reduction hypothesis as an explanation for ellipsis, various different approaches to how this hypothesis could be implemented have emerged. One branch of research is inspired by Lasnik (1995b, 1999b) and investigates various syntactic principles that operate at PF (Kennedy, Merchant, Wilder, in this volume), another branch of research, inspired by Klein (1993), assumes that pragmatic wellformedness conditions apply at PF and trigger deletion processes (Hartmann, in this volume).
Chris Kennedy: Ellipsis and syntactic representation On the traditional view outlined in (3), the syntactic representation feeds the two performance systems, the sensori-motor system and the system of thought. That is, the pairing of sound and meaning is essentially driven by syntactic representation. Kennedy, as part of a general research program, puts this view on test using the case of ellipsis. He addresses a question that has been of interest to researchers on ellipsis phenomena since the very early days of generative grammar: Do constituents targeted by various types of ellipsis operations have syntactic structure at some level (or levels) of representation, or can the various properties of ellipsis constructions be accounted for purely in terms of recovery of meaning, without assuming an intermediate level of syntactic representation? Kennedy argues that while semantic (and discourse) factors clearly play an important role in licensing ellipsis, it is nevertheless the case that elided constituents must also have syntactic representation. To support this claim, he focuses on the interaction of ellipsis (VP-ellipsis, attributive comparative deletion, pseudogapping) and several different grammatical phenomena and constraints, including parasitic gaps, binding theory, and island constraints, which he shows provide clear evidence that ellipsis constructions are sensitive to configurational constraints on syntactic representations. However, Kennedy points to a phenomenon that seems to challenge this result: attributive comparative deletion seems to be insensitive to Ross’ Left Branch Constraint (LBC), as is shown by the ungrammatical *Jones produced as successful a film as Smith produced a play. The solution to this problem lies in the assumption that the LBC holds at PF and not at LF (as argued in Kennedy & Merchant 2000). More
Exploring the interfaces from the perspective of omitted structures
precisely, if the LBC is hypothesized to apply to morpho-phonological properties of lexical items at PF, then the constraint is inactive if ellipsis has taken place. The prediction on ellipsis and syntactic representation involves two cases: If ellipsis involves deletion of syntactic structure, then elided constituents should be sensitive to syntactic constraints in general. However, since ellipsis (by definition) does not require pronunciation of the “missing” structure, elided constituents should be insensitive to syntactic constraints that derive from morphophonological properties of the lexicon. The second part of this prediction is based on the understanding that ellipsis always requires deletion of syntactic structure (where deletion can be formalized either as actual elimination of syntactic structure, or as elimination of morpho-phonological information only). The above prediction requires as the next step in the research program the identification of further ellipsis constructions that have properties of structures that could not appear overtly. In other words, the task is to find in ellipsis constructions evidence for the presence of objects that do not appear in other (wellformed) constructions of the language (such as the F0 [+wh] head in English). Initial support for this conclusion comes from Merchant’s (2001) work on sluicing and Kennedy and Lidz’s (2001) work on sloppy vs. strict readings of comparative stripping constructions. Merchant’s paper on subject-auxiliary inversion in comparatives in this volume constitutes further support for this prediction.
Jason Merchant: Subject-auxiliary inversion in comparatives and PF output constraints The paper “Subject-auxiliary inversion in comparatives and PF output constraints” contributes to the discussion of ellipsis by tackling a very important and long overlooked phenomenon: the required cooccurrence of subject-auxiliary inversion (SAI) with VP-ellipsis in comparative constructions. Merchant proposes an explanation of this phenomenon within the PF-deletion account of VP-ellipsis that attributes the requirement of VP-ellipsis under SAI to an interaction of properties of head movement involved in SAI with the licensing of traces of wh-movement involved in comparative clause formation. More precisely, it is assumed that A’movement out of VP with the destination of SpecCP proceeds via adjunction to that VP, as is also argued in López and Winkler (in this volume). Moreover, A’traces – including the intermediate trace adjoined to VP – are subject to the Empty Category Principle (ECP), reformulated as a condition operating at PF. Crucial to the explanation of why SAI in comparatives requires VP-ellipsis is the assumption that in comparatives the intermediate trace constitutes a violation of the ECP just in those cases where I-to-C movement has occurred. The logic is that a violation due to an illformed/ungoverned trace at PF can be remedied by deleting the VP and with it the offending trace.
Susanne Winkler and Kerstin Schwabe
Chris Wilder: Antecedent-containment and ellipsis Wilder reexamines antecedent-contained-deletion (ACD) constructions in the light of new empirical evidence from sentences involving ‘wide scope’ VP-ellipsis, as, for example, in John said that more trees had died than Mary did, which has the wide scope reading John said that more trees had died than Mary said had died. Wilder’s empirical interest is to establish a valid paradigm of ACD constructions. His primary theoretical interest is to find out whether the wide scope ACDs provide evidence for either of the hitherto assumed LF movement processes, namely, A’movement (Quantifier Raising, May 1985), or A-movement of objects in English (Hornstein 1994). In fact, the wide scope VP-ellipsis cases provide strong support for the following three claims: (i) ACDs involve A’-movement, (ii) an independent PF principle is at work (“A VP-ellipsis site E may not be linearly contained in its antecedent site”), and (iii) not all cases of alleged ACDs are indeed ACDs, but are rather the results of pseudogapping (It’s harder to persuade the Americans of this than it is_, the English) or backward deletion (Go from the last town north to the first town south of that mosquito-infested river) or both (John gave whoever he could _two dollars). Katharina Hartmann: Background matching in right node raising constructions Hartmann’s paper also sets out to explore the question of whether ellipsis is the result of syntactic processes or whether it is the result of phonological deletion (cf. Klein 1993; Chomsky 1995; Lasnik 1995b) and arrives at an essentially different PF-deletion account. While Kennedy, Merchant, and Wilder formulate syntactic principles that hold at PF, Hartmann formulates a pragmatic wellformedness condition (for background matching on the basis of Givenness) that functions as a licensing condition on ellipsis at PF. The paper investigates the right node raising (RNR) construction in German. On the basis of a detailed analysis of the interaction of grammatical processes that license elliptical constructions, Hartmann argues against the traditional movement accounts (cf. Postal 1974) and for an in situ theory that claims that RNR is derived by phonological deletion without modification of the syntactic structure. However, although she proposes that deletion takes place at PF, the principles that license deletion are not phonological but pragmatic. The relevant condition states that an utterance is pragmatically licensed if it has a background match, where background match is defined as Givenness in the sense of Schwarzschild (1999). More precisely, Hartmann shows that the interaction of information structure, a parallel syntactic configuration and a specific intonational pattern licenses phonetic deletion in RNR. This is the point at which she departs from the model in (3) and from the straightforward Lasnik-type PF-approach to ellipsis.
Exploring the interfaces from the perspective of omitted structures
In a nutshell, while the Kennedy, Merchant and Wilder approaches hypothesize that essentially syntactic principles and reflexes trigger PF-deletion, Hartmann suggests that essentially pragmatic principles cause PF-deletion. In terms of our original model of grammar in (3), Hartmann proposes that the PF component also interfaces with a pragmatic component. Although not explicitly stated, the model she proposes for RNR could be hypothesized to have the form in (24) where PF has direct access to the pragmatic component. (24)
Lexicon (Spell-Out) PF
LF
Discourse/Pragmatics
Hartmann’s paper, then, argues that RNR does not involve syntactic movement, but is best analyzed in terms of phonological reduction, where PF-reduction is licensed by the specific focus structure of RNR which is defined in terms of a specific pragmatic filter.
. From the computational system to the syntax-semantics interface All four papers in this group investigate the question of whether a special process of ellipsis must be added to the system of grammar or whether the existing means are sufficient to account for the different kinds of ellipses. More precisely, the first three papers explore the differences between movement and ellipsis, and the fourth paper proceeds from the hypothesis that gapping, in fact, is a special form of movement. This hypothesis forms the basis of an investigation of crosslinguistic syntactic variation, explaining it as variation at the syntax-semantics interface.
Caterina Donati: Merge copy Donati’s paper explores the null-hypothesis that ellipsis can be reduced to movement. It thereby challenges the standard view on ellipsis that perceives of ellipsis as an interface phenomenon. In essence, the paper proposes that ellipsis is not an interface phenomenon; rather, it is a purely syntactic phenomena. As all syntactic phenomena, ellipsis gets interpreted at the interfaces, but it is not in itself an interface process. The paper starts out by analyzing the fundamental similarities and differences between well-established movement operations and ellipsis. Donati points out that the copy theory of movement already makes available a deletion operation which is defined independently of standard ellipsis cases, such as in wh-movement where
Susanne Winkler and Kerstin Schwabe
the lower copy is deleted at PF. With respect to the phonological reduction of elements, which are interpreted without being pronounced, ellipsis doesn’t seem to be anything special. Grammar is already equipped with a general copy-deletion mechanism. Donati hypothesizes that in ellipsis contexts, the lower of two identical constituents (i.e., copies) is deleted at PF and this is simply due to a uniform deletion operation applying on copies. The paper further investigates the possibility of reducing ellipsis to movement altogether. The discussion, however, shows that the process merge copy has two instantiations: movement (e.g., wh-movement) and reduplication (e.g., ellipsis). In movement contexts, the copies are links of the same chain, where the moved copy is the goal of an agreement relation (feature checking); in the ellipsis contexts, in contrast, the copies are members of different chains and no agreement relation is involved (no feature checking). Reduplication is understood as a free operation, applying freely in the derivation like any other instance of merge.
Winfried Lechner: Phrase structure paradoxes, movement and ellipsis Lechner’s paper studies VP-fronting and pseudogapping, two constructions which superficially seem to differ only in the mechanism (movement vs. ellipsis) which removes the missing VP. At first sight, both constructions look very similar. They have in common a phonologically silent VP followed by a remnant: VP-fronting, as in (26) (Pesetsky 1995; Phillips 1996, to appear), and pseudogapping, as in (27): (26) John intended to give the books to the children, and [VP give the books to themi ] he did on each otheri ’s birthdays. (27) John gave the books to the children on Monday, and Mary did on Friday. Parting from earlier analyses (Phillip’s 1996 account of the type of phrase structure paradoxes discussed by Pesetsky 1995), Lechner argues for a movement analysis of PS-paradoxes, which postulates extraction of the remnant PP prior to topicalization. Thus, VP-fronting as in (26) receives an analysis analogous to the pseudogapping example in (27) (cf. Jayaseelan 1990; Johnson 1997a; Lasnik 1995b), the only difference being that the VP in VP-fronting is removed from its base position by movement and not by ellipsis. Lechner investigates various interpretive differences between VP-fronting and pseudogapping which manifest themselves in the complex scope and binding properties of remnants: On the one hand, he demonstrates that certain disparities between traces and ellipsis copies can be derived from the assumption that (VP) adjuncts may be merged in any position in which they are interpretable by the principles of type driven interpretation; On the other hand, he shows that the actual point in the derivation at which adjuncts are inserted is (indirectly) determined by economy conditions (Fewest Step). This elegant analysis of the contrasts between
Exploring the interfaces from the perspective of omitted structures
VP-fronting and pseudogapping entails two consequences: First, it generates an argument for the view that economy not only restricts possible movement relations, but also regulates the second structure building operation, merge. Second, it follows as a corollary that the metric which evaluates derivations cannot operate on a strictly local basis, but needs to be able to compare larger units of information (such as phases). He attributes these disparities to the interaction between economy conditions and the assumption that adjuncts may be merged with the root at different stages of the derivation.
Uli Sauerland: Unpronounced heads in relative clauses Sauerland’s paper investigates English relative clauses, such as in The pandas that we saw t at Ueno were cute, and tries to identify and explain the semantic relation between the relative clause internal trace position t, the position of the complementizer, and the head of the relative clause, here pandas. There are basically two hypotheses (cf. Carlson 1977a) that describe this relation in syntax: the matching hypothesis and the raising hypothesis. The matching hypothesis assumes that there is no direct transformational relationship between the head NP and the relative clause internal trace position. The raising hypothesis assumes that there is a transformational connection. Sauerland’s contribution to this discussion is threefold: First, he uses the diagnostics for wh-movement (reconstruction) that have been developed in recent years (Chomsky 1995; Fox 1998, 1999; Heycock 1995; Huang 1993; Lebeaux 1991, 1998; Sauerland 1998a, and others) and applies them to relative clauses (RCs). He thereby shows that indeed both types of RCs exist: a matching RC is shown in (28a) and a raising RC in (28b). The relevant factor to distinguish the two are reconstruction effects. An RC is a matching RC if the head NP must not be reconstructed. An RC is a raising RC if the RC must be reconstructed. (28) a. The relative of Johni that hei likes lives far away. b. The relative of hisi that everybodyi likes lives far away. Second, he concentrates on matching RCs and argues that matching RCs, like raising RCs, have complex heads. Third, Sauerland proposes that the relation of the silent internal head and the external head in matching RCs is best described as a process, called relative deletion, defined as follows: In matching RCs, the internal head must not be pronounced. Furthermore, the external head must be the antecedent of the internal head.3 Sauerland analyzes relative deletion as an obligatory ellipsis process, and shows that the standard assumption that vehicle change is possible in ellipsis, but not in movement chains, accounts for the special properties of reconstruction in relative
Susanne Winkler and Kerstin Schwabe
and comparative clauses. Thus, the novel claim of this paper is that the material in the trace position is related to the head not by movement but by ellipsis. The goal of the fourth paper in this group is an explanation of crosslinguistic variation in terms of syntax-semantics interface.
Luis López and Susanne Winkler: Variation at the syntax-semantics interface: Evidence from gapping López and Winkler investigate gapping in English and propose that it is a focus construction which is the result of two movement operations: first, remnant movement to a stacked A-bar position in the coordinated vP, and second, subsequent ATB-movement of the verb (cf. Johnson 1997a). The first type of movement is a version of contrastive focus movement; the second movement corresponds to the information structural function of defocusing López and Winkler use this gapping analysis as a platform to investigate the sources of cross-linguistic syntactic variation within Chomsky’s (1995, 2000) model of grammar. Within the Minimalist Program, the computational system should be conceived of as an autonomous system of composition and manipulation; its properties are therefore fixed and cannot be the locus of variation. The paper identifies two possible sources of variation: The first is the lexicon, where features of functional categories in particular have been identified as triggering movement (cf. Borer 1984). The second is the syntax-semantics interface, where it is hypothesized that rules of interpretation can be subject to variation. More specifically, López and Winkler have argued that the differences concerning wh- and focus movement between Western Bade and English cannot be traced to a property of a functional category, but rather to rules of interpretation. The same holds true if the variation occurs within a language, as in Chinese, where foci and wh-phrases surface in different positions. The first provisional hypothesis is that the parameter that distinguishes Western Bade from English involves interpretive rules (INT): While Spec,v receives INT in Western Bade, Spec,C receives INT in English, where INT is an interpretive rule that assigns wh-phrases and foci an interpretation. However, evidence from gapping in English shows that Spec,v can be a place where foci and wh-phrases can be interpreted even in English. In contrast to the sentential coordination plus deletion accounts (Ross 1970, among others), gapping is analyzed as vP coordination plus ATB-movement. The close investigation of the movement processes involved provides evidence for the claim that the focused remnants must occur in Spec,v in English. The authors conclude that the difference between Western Bade and English cannot be a fixed parameterized rule. Rather, they propose that interpretive rules are universal but ranked according to markedness. That is, the syntax-semantics
Exploring the interfaces from the perspective of omitted structures
interface is free to assign INT to the edges of phases. The exact choice of where INT is assigned is the source of parametric variation.
. The semantic component and its connection to focus and discourse structure The papers of this group focus on the semantics of the silent elements in the ellipsis site. Hardt, Romero, Schwabe, and Tomioka explore anaphoric elliptical material. Gretsch, in contrast concentrates on elliptical expressions that have an interrogative function in that they occur in a cataphoric relationship to their term answers.
Daniel Hardt: Ellipsis and the structure of discourse Hardt’s paper explores the relation that holds between an ellipsis site (here VPellipsis) and its antecedent, and how the ellipsis site is interpreted in the absence of form. Many authors have suggested that ellipsis and other processes of reduction require a certain similarity or parallelism between the ellipsis clause and some antecedent clause in discourse (Dalrymple et al. 1991; Rooth 1992b). In this paper, Hardt argues that this matching condition on ellipsis must be applied in accordance with discourse structure, where discourse structure is assumed to be a postLF-component that can be freely accessed by semantics. This results in a stronger condition on ellipsis: The matching clauses must also participate in a discourse relation. The effect of this condition can be clearly observed in examples involving multiple ellipsis, where discourse structure plays a key role in determining possible readings. The paper begins by considering two types of approaches to the matching condition: the semantic matching approach, exemplified by Rooth (1992a), and the path identity approach (originally due to Fiengo & May 1994 and modified by Asher et al. 2001). Hardt’s approach is that semantic matching must be applied according to discourse structure. He shows that in the case of multiple potential antecedents, discourse relations determine the correct choice. With respect to sloppy identity, discourse relations limit the possible readings. Maribel Romero: Correlate restriction and definiteness effects in ellipsis This paper investigates two ellipsis constructions – reduced conditionals in German and sluicing – and shows that both constructions are subject to a restriction on possible antecedent phrases for the remnants of ellipsis and to a definiteness effect that makes nondefinite phrases behave semantically as definites in ellipsis sites. Interestingly, these two properties are found only in German reduced conditionals, as in (29) and in sluicing, as in (30):
Susanne Winkler and Kerstin Schwabe
(29) a.
Wenn ich jemanden besuche, dann immer den Peter. if I somebody visit then always the Peter ‘If/whenever I visit somebody, then Peter/it’s Peter’ b. *Wenn ich den Karl besuche, dann immer den Peter.
(30) a. I know that four students came to the party, but I don’t know who. b. *I know that four students came to the party but I don’t know how many. Both restrictions are shown to follow from the semantics of focus. Romero brings together Rooth’s theory of focus with Schwarzschild’s Avoid Focus Constraint. Rooth distinguishes two felicity conditions: (i) the subset condition, which is relevant for reduced conditionals, and (ii) the membership condition, which holds for sluicing. The subset condition typically occurs in question-answer pairs (cf. Rooth 1992b) and licenses free focus (non-operator focus, according to von Stechow 1991), which typically occurs in question-answer pairs. According to Rooth (1992b), the denotation of the question (in Hamblin’s 1973 formalism) must be a subset of the semantic value of the answer. The membership condition licenses contrastive focus, that is, contrastive focus is licensed if there is an alternative to the focused element which is different from the element itself. Schwarzschild’s Avoid Focus Constraint (economy constraint) is needed to fulfill the membership condition. Romero proposes two generalizations. The first one is a generalization on proper remnants and involves two parts: (i) A remnant with free focus introduces an exhaustivity implicature; (ii) A remnant with contrastive focus must contrast with its correlate phrase. The second one is a descriptive generalization operating on the antecedent of the wh-remnant: An NP is a good correlate for a sluiced focused wh-phrase if its clause expresses (the answer to) a question that contrasts with the question expressed by the sluiced interrogative. Romero claims that the definiteness effect in sluicing follows from the interaction of the semantics of questions and focus. The key intuition is that the wh-phrase who in a full interrogative like I just heard that a student in your class likes most professors in this school. I need to find out who likes most professors here/in this school may range over a wider set of people than the sluiced WHO in I just heard that a student in your class likes most professors in this school. I need to find out WHO _, where focused WHO can only stand for which student in your class. Basically, focus requires that the question implied by the A-constituent and the question expressed by the E-constituent match semantically and that they only differ in the denotation of the focused material. The first question asks about the existence of a particular type of student and the second question asks about the identity of that same type of student. This is required by the subset condition plus Avoid Focus. Since the student is the same in
Exploring the interfaces from the perspective of omitted structures
both questions, the (unique) majority of professors that he likes is also the same in both questions. Hence the definiteness effect. This is the point at which Romero’s and Schwabe’s approach to sluicing meet.
Kerstin Schwabe: Focus marking and specificity in sluicing constructions This paper puts forward a novel and integrated analysis of two factors that play an important role in the semantic licensing of sluicing constructions, namely, the focus restriction on the antecedent and the specificity restriction on the antecedent. The basic claim is that the correlate of the wh-phrase must allow a specific interpretation and that the restriction as well as the nucleus scope of the wh-phrase must be given within the antecedent clause. The focus analysis is based on Schwarzschild (1999) and Merchant (2001) and shows that in cases where the antecedent clause of the sluicing clause is not a judgement, a process of accommodation must be assumed. With respect to the specificity restriction on the antecedent, Schwabe proposes a solution within the theory of von Heusinger (1997, 2000). She analyzes specific indefinite expressions as denoting a discourse referent that is anchored to an accessible discourse participant and shows that the specificity restriction of the indefinite in sluicing depends on the semantics of wh-phrases. She shows that the semantics of the wh-phrases presupposes (i) that the entity denoted by the indefinite in the first conjunct must not be anchored for a discourse participant a, and (ii) that it must be anchored for some other discourse participant b. The first presupposition prevents the entity from being in the background and/or from being expressed by a definite expression. The second presupposition prohibits the entity from occurring in contexts that do not allow a specific interpretation such as thematic embedded clauses and the scope of certain quantificational or modal phrases. Finally, the paper shows in various applications that the focus restriction and the indefinite restriction are two mutually dependent conditions. Satoshi Tomioka: The semantics of Japanese null pronouns and its crosslinguistic implications Tomioka investigates the nature of phonologically silent pronouns in Japanese and shows that they can receive a wide variety of semantic interpretations. One of his main claims is that different interpretations of pronouns do not come from semantic ambiguities of pronouns themselves. Four out of the six interpretations of null pronouns in Japanese (namely, E-type, (definite) laziness pronoun, indefinite pronoun, and property anaphora) are due to the inherent semantic flexibility of full-fledged bare NPs in Japanese. Bare NPs in Japanese appear freely in argument position, and their in/definiteness is achieved secondarily by means of semantic operations. Tomioka proposes that the four aforementioned uses of Japanese null
Susanne Winkler and Kerstin Schwabe
pronouns are simply property anaphora (type <e,t>), and that their differences are the results of the same semantic operations that affect full-fledged NPs. Additional crosslinguistic evidence for Tomioka’s analysis comes from Chinese, Korean and Modern Greek. Chinese and Korean, which allow Bare NPs and employ the same interpretive strategies as Japanese, show exactly the same range of semantic variability of null pronouns as Japanese. Modern Greek allows null pronouns only when the antecedents are ‘weak’ quantifiers. Such selectiveness is easy to analyze within Tomioka’s proposal. The existential quantification of those weak quantifiers does not come from determiner quantification but rather is due to a covert semantic operation. Then, Greek null pronouns with weak antecedents are analyzed as property anaphora, just like the Japanese counterpart. Given the view that there is crosslinguistic variation in the availability of the semantic tools which play an important role in the interpretation of null pronouns, an entirely new question arises: Does semantics have something to do with what has been known as the pro-drop parameter? Tomioka’s answer to this question is ‘Yes’. His argument is based on two crosslinguistic generalizations. The first generalization is about discourse pro-drop languages, which permit null pronouns when a certain contextual familiarity condition is met: Discourse pro-drop languages allow robust bare NP arguments. The second generalization is that phonologically null NP anaphora, also known as N-bar-deletion/NP ellipsis, seems available in most, if not all, languages. Tomioka points out that phonologically null arguments and N-bar-deletion/NP ellipsis would have the identical outcome in languages with NP arguments. The hypothesis forwarded in this paper is that they are indeed the same: Null pronouns in discourse pro-drop languages are simply the result of N-bar-deletion/NP ellipsis without determiner stranding. In Chierchia’s (1998) theory, the availability of bare NP arguments is determined by the nominal mapping parameter, which dictates what basic semantic types each language assigns to nouns. Discourse pro-drop crucially relies on the availability of bare NPs in argument positions, which is determined by the semantic parameter of nouns. Thus, semantics has a great deal of influence on the pro-drop parameter. The paper concludes by pointing out two questions for further research: First, why is it that although all discourse pro-drop languages seem to allow bare NP arguments, the reverse is not true? Not all bare NP languages are discourse prodrop. Second, what is the nature of this null NP anaphora? Is it a pro-form or an instance of ellipsis/deletion? These open questions need to be explored further.
Petra Gretsch: Topic and focus in focal ellipsis Gretsch’s paper constitute a novel exploration of what can possibly count as an instance of ellipsis. She investigates operatorless questions such as The keys are _?, called the focal ellipsis (FE) construction and proposes that focal constituents actually can be omitted, which results in an erotetical utterance. She suggests that the
Exploring the interfaces from the perspective of omitted structures
gap in the FE construction constitutes the focus of the question which challenges the common view on the information structure of elliptical constructions where the gap corresponds to the given constituents. Using German, Korean and Chinese data, Gretsch argues that there are different types of FEs (contrastive and presentational FEs) and proposes that the appearance of contrastive FE is universal, but that the occurrence of presentational FE is restricted to languages which allow for morphologically unmarked topics. The above abstracts show that the papers of this volume connect in at least four important respects: First, all the papers investigate the fine structure of the syntaxsemantics-phonology interfaces. Second, they look at different types of ellipses or deletion processes with the goal of finding an answer to the theoretical questions posed in (4) above. Third, the papers build on current investigations in the theory of grammar and information structure and mirror an important shift in the theoretical approaches to ellipsis. The shift from syntactic deletion to PF-deletion is counterbalanced by many researchers who investigate the interface between syntax and semantics in the search for an answer to many of the intricate questions of ellipsis. And last but not least, the papers reflect the cooperation of a group of researchers whose recent investigations into the interfaces from the perspective of omitted structures have yielded important results.
Notes * We would like to thank the participants of the joint workshop of the ZAS (Berlin) and the University of Tübingen on Ellipsis and Information Structure, which took place in Berlin in November 1999 and the SFB441-colloquium participants at the University of Tübingen in the summer of 2001 for stimulating discussions of the ideas and issues discussed here. Particular thanks go to Werner Abraham, Ralph Albrecht, Mechthild Bernhard, Kirsten Brock, Paul David Doherty, Bernie Drubig, Remus Gergel, Daniel Hardt, Jutta Hartmann, Katharina Hartmann, Winnie Lechner, Luis López, and Uli Sauerland for comments on an early draft. Remaining errors of interpretation and analysis remain ours solely. . The list of authors referred to in (5) is exemplary and non-exhaustive. Since ellipsis is an interface phenomenon and a nonuniform phenomenon, there are cross-classifications to be expected that are worked out in the individual contributions to this volume. The abbreviation itv stands for “in this volume”. . A further problem for the PF-deletion account stems from the observation that only in the second conjunct can only be understood if can talk is deleted. A similar observation has been made by Johnson 2000 with respect to VP-adverbs. (i) It’s probably true . . . that the system is so hierarchical that only the ASSISTANT manager can talk to the MANAGER – a. and the manager can talk to the general manager (as everybody else does). b. and the manager can talk to whoever he wants.
Susanne Winkler and Kerstin Schwabe . Relative deletion is modeled after comparative deletion (Bresnan 1973, 1975; Lechner 1999), which also involves the deletion of “an internal copy of the head that doesn’t seem to be related by movement to the external head”.
I. Towards the exploration of PF-deletion accounts
Ellipsis and syntactic representation Christopher Kennedy This paper addresses a question that has been of interest to researchers on ellipsis since the very early days of work in generative grammar: do constituents targeted by various types of ellipsis operations have syntactic structure at some level (or levels) of representation, or can the various properties of ellipsis constructions be accounted for purely in terms of recovery of meanings, without positing syntactic representation at the ellipsis site? Focusing on the interaction of ellipsis and several different grammatical phenomena and constraints, including parasitic gaps, binding theory, and extraction islands, I will present evidence that ellipsis constructions are sensitive to configurational constraints on syntactic representations, but not to constraints that are based on morphophonological properties of lexical items, thus supporting a view of ellipsis as deletion of syntactic material.*
.
The representation of nothing
Since at least Hankamer and Sag 1976a, a central question in research on ellipsis has been what sorts of representations are involved in the resolution and licensing of unpronounced linguistic information? Two lines of thought have predominated, which differ in their assumptions about the role of syntax in ellipsis. The first approach, which has a long tradition in generative grammar, postulates that elided material has syntactic structure at some level of representation, but the grammar contains a means of blocking its pronunciation in the surface form. The second approach rejects the claim that unpronounced material has syntactic representation, hypothesizing instead that general mechanisms governing the recovery of meanings from context can be put to work to resolve ellipsis. The purpose of this paper is to provide arguments in favor of a version of the first approach, and to show that an analysis in which ellipsis involves only the recovery of meanings, without reference to syntax, fails to provide an empirically adequate account of the facts. Before making an argument in favor of this position, however, I will present an overview of the positive and negative aspects of several approaches that are representative of these different answers to the question of representation.
Christopher Kennedy
. Syntactic analyses of ellipsis The hypothesis that ellipsis involves syntactic representation can be implemented in two ways: deletion of syntactic material from the representation that is the input to the phonological component or recovery of syntactic structure at some level of logical representation. The former approach goes back to the very early days of generative grammar, and has been revitalized in recent work in the Principles and Parameters framework (see e.g. Hankamer 1979; Sag 1976a; Tancredi 1992; Wilder 1995b; Merchant 2001; Kennedy & Merchant 2000a); the latter approach has also appeared in different forms over the past twenty-five years (see e.g. Wasow 1972; Williams 1977a; Haïk 1987; Kitagawa 1991; Fiengo & May 1994). The crucial assumption that both sorts of syntactic analyses share is that elided material has syntactic structure at some level of representation. A central result of such approaches, therefore, is that they can account for syntactic effects within the ellipsis site. For example, the fact that island effects appear under ellipsis, as shown by the contrast between the (a) and (b) examples in (1) and (2), receives a straightforward explanation: if ellipsis involves deletion, then the (b) sentences are derived from the representations in (1c) and (2c) (where struck-through text indicates material deleted from the pronounced form), which involve extraction out of an island (indicated by a subscript I). (1) a. Sterling criticized every decision that Lou did. b. *Sterling criticized every decision that Doug was upset because Lou did. c. *Sterling criticized every decision [whi that Doug was upset [I because Lou did [ VP criticized t i]]]. (2) a. Dogs, I understand, but cats, I don’t. b. *Dogs, I understand, but cats, I don’t know a single person who does. c. *Catsi I don’t know [I a single person who does [VP understand t i]]. Note that in the absence of an A’-dependency into the ellipsis site, the relation between an elided VP and its antecedent is not sensitive to island constraints, as originally observed by Ross (1967). The appearance of Binding Theory effects in the ellipsis site is another property that is expected in a syntactic analysis. For example, the fact that (3a) strongly disfavors a ‘strict’ interpretation, in which Sterling also blames Doug for the band’s collapse, follows from the fact that it is derived from (3b): the strict reading would violate Condition A, which requires a reflexive pronoun to find its antecedent locally. Likewise, the disjoint reference effect in (4a) is a direct consequence of Condition B of the Binding Theory, which rules out coreference between a pronoun and a co-argument.
Ellipsis and syntactic representation
(3) a. Doug blamed himself for the band’s collapse, and Sterling did too. b. Doug blamed himself for the band’s collapse, and Sterling did [VP blame himself] too. (4) a. *Kim takes care of himi because hei won’t. b. Kim takes care of himi because hei won’t [VP take care of him i]. Finally, a syntactic approach to ellipsis, augmented with sufficiently strict requirements on the type of identity relation that licenses deletion, provides the basis of an account of the unacceptability of examples involving syntactic nonidentity. (5) a. ??Only 43 percent of registered voters did. b. ??A lot of this material can be presented in a fairly informal and accessible fashion, and often I do. There appear to be a number of serious problems for a syntactic account of ellipsis, however, the most important of which is the fact that there are contexts in which syntactic effects within the ellipsis site seem to disappear. One such context involves comparatives constructed out of attributive adjective phrases, such as those in (6). (6) a. *The Cubs start a more talented infield than the Sox start an outfield. b. *Jones produced as successful a film as Smith produced a play. Kennedy & Merchant (2000) demonstrate that the unacceptability of the examples in (6) is due to the Left Branch Constraint (LBC), which blocks movement of left branch attributive modifiers (see also Pinkham 1982). Assuming that comparatives are derived through an operation of A’-movement that targets the compared constituent in the than-clause (Ross 1967; Chomsky 1977b), the syntactic structures assigned to the examples in (6) are those in (7), which, just like the questions in (8), violate the LBC. (7) a. *The Cubs start a more talented infield than [whi the Sox start an ti outfield] b. *Jones produced as successful a film as [whi Smith produced a ti play]. (8) a. *How talentedi do the Sox start an ti outfield? b. *How successfuli did Smith produce a ti play? The problem for a syntactic theory of ellipsis comes from examples like (9a) and (9b), which appear to indicate that island effects disappear under ellipsis, contrary to what we saw above in (lb) and (2b). If ellipsis involves deletion, then (9a) and (9b) should be derived from (l0a) and (10b), respectively, which should be just as ill-formed as (7) and (8) above. (9) a. The Cubs start a more talented infield than the Sox (do). b. Jones produced as successful a film as Smith (did).
Christopher Kennedy
(10) a.
The Cubs start a more talented infield than [whi the Sox (do) [VP start [DP a t i infield]]]. b. Jones produced as successful a film as [whi Smith (did) [VP produced [DP a t i film]]].
A second problem for syntactic analyses of ellipsis is that there are also contexts in which binding effects seem to disappear. (11a), for example, can clearly have a strict reading, despite the fact that it should be derived from (11b), which does not support such a reading. (11) a.
Doug blamed himself for the band’s collapse because everyone else did. b. Doug blamed himself for the band’s collapse because everyone else did [VP blame himself]
Similarly, (12a) fails to show the disjoint reference effect that we would expect to see if it were derived from the representation in (12b), which does not permit coreference between the pronominal arguments in the second clause. Most Americans expected himi to be acquitted, and obviously hei did too. b. Most Americans expected himi to be acquitted, and obviously hei did [VP expect himi to be acquitted].
(12) a.
Finally, although it is the case that syntactic non-identity typically results in judgments of unacceptability, it is a fact about English that many such examples are attested. In particular, both of the examples presented above in (5) are naturally occurring sentences: (13) a.
In yesterday’s elections, only 43 percent of registered voters did. (heard on National Public Radio by CK in November 1996) b. A lot of this material can be presented in a fairly informal and accessible fashion, and often I do. (Chomsky 1982: 41; cited in Dalrymple et al. 1991)
. Semantic analyses of ellipsis A second approach to ellipsis claims that elided constituents have no syntactic representation at all, but rather can be fully explained in terms of a more general theory of information retrieval (see e.g. Dalrymple et al. 1991; Hardt 1992, 1999; Jacobson 1992; Hendriks & de Hoop 2001). In the higher-order unification approach advocated by Dalrymple et al. (1991), for example, a structure like (14a) is assigned a semantic representation of the sort in (14b), where P is a free variable over prop-
Ellipsis and syntactic representation
erties that needs to be resolved. The problem of ellipsis is the problem of solving the value of P, which is done by abstracting over parallel elements in some previous clause to generate a property-denoting expression, as shown in (14c), and substituting this expression for P. (14) a. b. c. d.
Sterling quit the band because Lou did. quit(Sterling, the band) BECAUSE P(Lou) P = λx.quit(x, theband) quit(Sterling, the band) BECAUSE λx.quit(x, theband)(Lou)
One obvious positive aspect of this type of approach is that it doesn’t run into the problems associated with purely syntactic accounts: because ellipsis does not involve syntactic representation, we should not expect to find syntactic effects inside the ellipsis site. However, this advantage is also its disadvantage: as noted above, there are a number of contexts in which we do find clear evidence of syntactic effects within the ellipsis site. In order to account for facts like those discussed in Section 1.1, we would need to significantly weaken assumptions about the nature of the syntax-semantics interface. For example, we could adopt Haïk’s (1987) position that the ellipsis site itself can serve as the “gap” for a syntactic operator, as illustrated in the following examples. (15) a. *Sterling criticized every decision [CP whi that Doug was upset [AdvP because Lou did ti ]]. b. *Dogs, I understand. Catsi I don’t know [DP a single person who does ti ]. Since this approach requires us to abandon the assumption that semantic type of a bound variable corresponds to syntactic category of a corresponding gap (the bound variable in the interpretation of the relative clause in (15a) has the type of an individual, but the syntactic category of the gap in the syntactic representation is VP), it should be adopted only if a less costly approach fails to materialize.
. A “mixed” analysis In recent work, Andy Kehler (1995, 2000) attempts to account for the apparently paradoxical sensitivity of ellipsis to syntactic constraints by developing a “mixed” syntactic/semantic analysis, in which whether an elided constituent has internal syntactic structure or not depends on the discourse context in which it appears. In particular, Kehler argues that the requirement for syntactic representation in ellipsis depends on the type of “coherence relation” an elided VP participates in (see Kehler 1995; Hobbs 1979). Coherence relations determine coherence between
Christopher Kennedy
sentences in a discourse, which in turn affects acceptability. The two types of coherence relations that are relevant to ellipsis are cause-effect relations and resemblance (parallelism and contrast) relations, the action of which is illustrated by the following coherent (acceptable) and incoherent (unacceptable) discourses. (16) cause-effect relations (coherent) a. Smith is a politician. He’s bound to be dishonest. b. Smith is dishonest because he’s a politician. c. Jones is a politician, but she’s honest. (17) cause-effect relations (incoherent) a. ??Smith is a politician. He’s bound to wear long underwear. b. ??Smith is dishonest because he wears long underwear. c. ??Jones is a politician, but she has a nice daughter. (18) resemblance relations (coherent) a. Smith likes to play golf. Jones enjoys surfing the net. b. Smith stepped up to the podium. His critics stepped into their offices. c. Jones seems unbeatable, while her opponent seems beaten. (19) resemblance relations (incoherent) a. ??Smith likes to play golf. Jones ate lunch. b. ??Smith stepped up to the podium. Jones bought a new Mercedes. c. ??Jones seems unbeatable. The Senate is being painted. Kehler suggests that an elided expression that contained in a sentence that is part of a cause-effect relation does not require syntactic representation, but an elided expression that is part of a resemblance relation must have syntactic representation. The reasoning underlying this proposal is that resemblance relations are, to a large extent, at least, identified on the basis of syntactic structure, while causeeffect relations care only about propositional content (see Kehler 2000: 540–543). The prediction of this analysis, then, is that syntactic effects should show up only in the context of resemblance relations. This seems to be exactly right for some of the cases that are problematic for a syntactic approach, such as examples involving syntactic non-identity (20)–(21) and the absence of Condition A effects (22)–(23) (but see also Hestvik 1995). (20) a. ??This problem was looked into by Kim, and Lee did too. (resemblance) b. ?This problem was looked into by Kim, even though Lee already had. (cause-effect) (21) a.
??The letter from the Dean provoked a response from the Chair, and the
Provost did too. (resemblance)
Ellipsis and syntactic representation
b. ?The Dean’s actions provoked a response from the Chair, despite the fact that the Provost already had. (cause-effect) (22) a.
Doug2 blamed himselfi for the band’s collapse, and everyone else did too. (resemblance; strict reading difficult) b. Doug blamed himself for the band’s collapse, because everyone else did. (cause-effect; strict reading possible)
(23) a.
John wouldn’t introduce himself to everyone, but Mary did. (resemblance; no strict reading) b. Since John wouldn’t introduce himself to everyone, Mary did. (causeeffect; strict reading possible)
However, Kehler’s proposal also suffers from the same problems that afflict a purely semantic account of ellipsis: there are contexts in which an elided VP clearly occurs in a constituent that participates in a cause-effect relation, yet also appears to be sensitive to syntactic constraints. For example, both (24a) and (25a) manifest cause-effect relations, yet the elided VPs inside the relative clauses are sensitive to the Adjunct Island Constraint (islands enclosed in brackets). (24b) and (25b) demonstrate that ellipsis is possible in these examples as long as the elided VP is not in an island. (24) a. *Sterling criticized every decision that Doug was upset [because Lou did]. b. Sterling criticized every decision that Lou did. (25) a. *Max refused to buy the shirt that I picked out even though it was less expensive than the one that the salesperson complimented him [after he did]. b. Max refused to buy the shirt that I picked out even though it was less expensive than the one that he did. Similarly, (26a) and (26b) show that Condition B effects arise in cause-effect environments. (26) a. *Kim takes care of himi because hei won’t. b. *His closest allies supported himi throughout this ordeal, even though hei probably wouldn’t have. In addition to these empirical problems, there is a third, more general problem with a mixed approach such as Kehler’s. If a purely semantic analysis is available in some examples, then it ought to be in principle available in all examples, even if a syntactic analysis is preferred. In other words, an approach that allows for the possibility of semantic recovery of VP meanings without concomitant syntactic representation of an elided VP predicts that examples like (27a) and (28a) should
Christopher Kennedy
be no worse than (27b) and (28b), which violate coherence relations but do not violate any syntactic constraints. (27) a. ??This problem was looked into by Kim, and Lee did too. b. This problem was looked into by Kim, and Lee looked into it too. (28) a.
??The letter from the Dean provoked a response from the Chair, and the
Provost did too. (resemblance) b. The letter from the Dean provoked a response from the Chair, and the Provost responded to it, too. This is not an accurate characterization of the facts, however. While (27b) and (28b) are less fluent than completely parallel structures, they are quite clearly more acceptable than (27a) and (28a).
. Ellipsis and syntactic representation The discussion in the previous section does not provide conclusive arguments for one answer to the representation question in ellipsis over another, but it lays out some of the empirical ground that must be covered by any account. At the same time, this discussion allows us to formulate very precisely a central prediction of analyses that posit syntactic representation in the ellipsis site. Taking the deletion analysis as the representative of such an approach (for simplicity; most of what I will say here holds of a copying analysis as well), we can formulate the predictions in (29). (29) Ellipsis and syntactic representation If ellipsis involves deletion of syntactic structure, then: a. Elided constituents should be sensitive to syntactic constraints in general. b. However, since ellipsis does not require pronunciation of the omitted structure, elided constituents should be insensitive to syntactic constraints that derive from morphophonological properties of lexical items. In other words, the predictions of a syntactic analysis of ellipsis are more subtle than they appear at first. For the most part, we should see the same sort of syntactic behavior in overt and elided XPs; in particular, both elided and overt XPs should be subject to the same set of configurational constraints. However, the syntactic analysis does not predict that an ellipsis construction should have exactly the same syntactic properties as a corresponding overt form. Crucially, because ellipsis bypasses pronunciation, any constraints that make reference to the interface be-
Ellipsis and syntactic representation
tween the syntax and the phonological component should be vacuously satisfied, and therefore effectively “turned off ”, in the case of ellipsis. In the following two sections, I will show that at least one form of ellipsis – VPdeletion in English – behaves exactly as the predictions in (29) lead us to expect. For the purpose of this paper, I will assume a multistratal framework in which the syntactic component generates a pair of representations, one that interfaces with the phonological component (PF), and one that interfaces with the semantic component (LF) (as in Chomsky 1995, etc.). I will show that VP-ellipsis constructions are sensitive to configurational constraints on (LF) representations, but not to morphophonological constraints governing the interpretation of PF representations, in line with the predictions in (29). I should note that I am adopting a multistratal framework for convenience. The predictions in (29) hold of any theory that includes both configurational and morphophonologically defined “interpretive” constraints on syntactic well-formedness: ellipsis constructions should be sensitive to the former but not the latter.
. Ellipsis and parasitic gaps . “Non parasitic” gaps As originally observed by Kim and Lyle (1996), apparent parasitic gap chains do not show island effects when the expected position of the parasitic gap is contained in a deleted VP (see also Lappin 1992). (30) Wh-islands a. *Which articlei did you summarize ti after Jim asked [who had read PGi ]? b. Which articlei did you summarize ti after Jim asked [who would be willing to]? (31) Adjuncts a. *Which moviei did you see ti because Polly was so excited [after going to PGi ]? b. Which moviei did you see ti because Polly was so excited [after she did]? (32) Complex NPs a. *Mayor Daleyi , whom everyone met ti except [the people who didn’t know that there would be an opportunity to see PGi ], discussed Chicago politics.
Christopher Kennedy
b. Mayor Daleyi , whom everyone met ti except [the people who didn’t know that there would be an opportunity to], discussed Chicago politics. (33) Coordinate structures a. *Which booksi did you read ti after learning that Erik had bought PGi and enjoyed themi ? b. Which booksi did you read ti after learning that Erik had and enjoyed themi ? Kennedy (1997a) argues that these facts have a fairly straightforward and simple explanation: the putative parasitic gap chains in (30b)–(33b) do not show island effects because, contrary to what might be initially assumed, these sentences do not actually contain parasitic gaps. Instead, the “gap” in the deleted VP is actually a pronoun. In other words, the structural descriptions of (30b)–(33b) are as in (34a)–(34d), where the struck-through text corresponds to the deleted VP. (34) a.
Which articlei did you summarize ti after Jim asked who would be willing to [VP summarize iti]. b. Which moviei did you see ti because Polly was so excited after she did [VP saw iti]? c. Mayor Daleyi , who everyone met ti except the people who didn’t know that there would be an opportunity to [VP meet himi], discussed Chicago politics. d. Which booksi did you read ti after learning that Erik did [VP read themi ] and enjoyed themi ?
The reason that VP-deletion is possible here, despite the apparent non-identity between deleted and antecedent VPs, is that these examples are just instances of what Fiengo and May (1994) call “vehicle change”: the observation that in many contexts, pronouns and other expressions (in particular, A’-traces) “count as” identical for the purpose of licensing deletion.1 One of the arguments that Kennedy presents in favor of the claim that the elided VPs in (30b)–(33b) contain pronouns, not parasitic gaps, comes from crossover effects. As is well known, parasitic gaps show strong crossover effects (i.e., they are subject to Condition C; see Postal 1993; Cinque 1990). This is illustrated by the contrast in (35). (35) a. *Whoi were they investigating ti before hei knew they suspected PGi ? b. Whoi were they investigating ti before you knew they suspected PGi ? Sentences like (30b)–(33b), however, do not show strong crossover effects:
Ellipsis and syntactic representation
(36) a. Whoi did Maureen vote for because hei asked her to? b. Which studentsi did Otis report without themi knowing he had? c. Whoi did you call before shei asked you to? If the elided VPs in these examples contain pronouns, rather than parasitic gaps, then the facts in (36) follow: these sentences have structures that are completely parallel to examples like the ones in (37), which also do not show crossover effects. (37) a. Whoi did Maureen vote for ti because hei asked her to vote for himi ? b. Which studentsi did Otis report ti without themi knowing he had reported themi ? c. Whoi did you call ti before shei asked you to call heri ? If, however, the elided VPs in these examples contained parasitic gaps, then they should be as ungrammatical as corresponding examples with overt parasitic gaps, such as those in (38). (38) a.
Whoi did Maureen vote for ti because *hei /Charles asked her to support PGi ? b. Which studentsi did Otis report ti without *themi /you knowing he suspected PGi ? c. Whoi did you call ti before *shei /Marcus asked you to visit PGi ?
While the sentences under consideration do not show crossover (Condition C) effects, they do show Condition B (39) a. *Whoi did Maureen recommend because hei /Louis wouldn’t? b. *Which studentsi did Otis report even after telling themi /you to? c. *Whoi did you try to serve before seeing that shei /I already had? Again, this follows if the “gaps” in the ellipsis sites are pronouns: (40) a. *Whoi did Maureen recommend ti because hei wouldn’t recommend himi ? b. *Which studentsi did Otis report ti even after telling themi to report themi ? c. *Whoi did you try to serve t i before seeing that shei already had served heri ? The conclusion to draw from these facts is that the gaps in the elided VPs in these examples are pronouns. It follows that (30b)–(33b) can be assigned the syntactic representations in (34a)–(34d), explaining the absence of island effects. More generally, if the principles of the Binding Theory (in particular, Condition B) apply to syntactic representations, as is standardly assumed, then it must be the case that elided VPs are syntactically represented.
Christopher Kennedy
Before moving on, we should first consider an alternative view of the Binding Theory, which views constraints on coreference as constraints on the morphosyntactic expression of particular types of meanings, rather than as structure-based constraints on the possible interpretations. In such a model, the facts discussed here would not necessarily provide evidence for syntactic representation in ellipsis, as pointed out in Hardt 1999. An example of this type of approach to the Binding Theory is developed in Reinhart and Reuland 1993, in which Condition B is stated as in (41). (41) A reflexive predicate is reflexive-marked. In essence, this constraint requires any predicate (at least) two of which’s arguments are co-valued to be morphologically (or lexically) marked as reflexive. (41) thus correctly predicts that an example like (42a) is ungrammatical on the reading indicated by the coindexing even if there is no internal structure to the VP (42) a. *Whoi did Otis nominate ti because shei couldn’t? b. wh x[Otis nominated x BECAUSE ¬ ♦ ◊ [x nominate x]] The predicate is reflexive (in Reinhart and Reuland’s sense), as indicated by the logical representation in (42b), but it is clearly not reflexive-marked. More precisely, if there is no structure to the elided VP, it could never be reflexive-marked.2 The problem with this analysis is that it is too strong. In particular, without some weakening, it rules out reflexive interpretations across the board: the second conjunct in a simple example like (43a) should also violate (41), because the reflexiv e predicate in the second conjunct is not reflexive-marked. If the elided VP has internal structure, however, as indicated in (43b), then it is reflexive-marked, even if this is not apparent in the surface string. (43) Otis served himself, and Alex did too. (44) Otis served himself, and Alex did [VP serve himself] too.
. “Missing” parasitic gaps The analysis of the facts in the previous section builds on the idea that the deleted VP can be given the “non-parasitic” gap analysis in (45a). However, as pointed out by Shimada (1999) and Postal (2001), the alternative “missing” parasitic gap structure in (45b) is a possible analysis of the elided VP, but it is not chosen since it would result in an ill-formed structure. (45) a. [VP . . . pro . . . ] b. [VP . . . PG . . . ]
the non-parasitic gap structure the missing parasitic gap structure
Ellipsis and syntactic representation
If we could find contexts in which (45b) had to be the actual analysis, and if such contexts behaved syntactically just like p-gap constructions (obeyed islands, showed strong crossover effects, etc.), then we would have even more evidence that elided VPs have syntactic structure. The only difference between (45a) and (45b) is a structural/syntactic one (semantically, both pro and PG are interpreted as bound variables), therefore if we can show that some instances of VP-deletion must have the former structure and some the latter, we will have provided evidence for syntactic representation in the ellipsis site. Postal (2001) makes essentially this point in the conclusion of his paper; here I am simply fleshing out the argument in more detail. We begin with Engdahl’s (1985: 41, fn. 19) observation that VP-deletion improves some parasitic gaps outside the deletion site: (46) a. *Otis is a person whoi I admire ti because close friends of PGi became famous. b. Otis is a person whoi I admire ti because close friends of PGi seem to. (47) a. *Which filmi did you see ti because a critic of PGi was excited? b. Which filmi did you see ti because a critic of PGi advised you to? According to Postal (2001), these contrasts provide evidence for a structural analysis of the deleted VP as in (45b). Postal observes that subject-internal parasitic gaps such as those in the preceding examples require the presence of a local Adependency; since no such dependency exists in the (a) sentences, the subjectinternal gap is not licensed. The gap may be licensed by another parasitic gap chain, however: (48) a.
Otis is a person whoi I admire ti because close friends of PGi seem to respect PGi . b. Which filmi , did you see ti because a critic of PGi had recommended PGi ?
The fact that (46b) and (47b) are grammatical, then, means that the deleted VPs must contain parasitic gaps. That is, they must have the structure in (45b), as illustrated in (49). Otis is a person whoi I admire ti because close friends of PGi seem to [VP admire PG i]. b. Which filmi , did you see ti because a critic of PGi advised you to [VP see PG i]?
(49) a.
Postal’s observation holds of other “dependent” parasitic gaps as well. In the following examples, the dependent parasitic gap appears a clause that is adjoined to another adjunct clause. As shown by (50b), a parasitic gap in the second adjunct requires the presence of a parasitic gap (or some other A’-dependency) in the first
Christopher Kennedy
adjunct it cannot be directly licensed by the wh-chain in the main clause, as this would violate the Adjunct Island Constraint. (50) a.
Whoi did you call t i , [before learning that I already had gotten in touch with PGi , [after seeing PGi , in the street]]? b. *Whoi did you call t i , [before learning that I already had gotten in touch with himi , [after seeing PGi , in the street]]?
The fact that an example like (51a) is well-formed, then, means that the elided VP in the first adjunct must have an analysis in which it contains a parasitic gap, as in (45b). Whoi did you call t i , [before learning that I already had [after seeing PGi in the street]]? b. Whoi did you call t i before learning that I already had [VP called PGi ] after seeing PGi in the street?
(51) a.
Postal (2001) strengthens the conclusion that both (45a) and (45b) are possible structures for elided VPs by showing that examples involving dependent parasitic gaps, unlike the “non-parasitic” gap structures discussed in Section 3.1, are sensitive to islands. This is illustrated by the following examples, which are parallel to the examples in (30)–(33) except that VP-deletion does not save the island violations. (52) Wh-islands a. *Which bandi did you hire ti only after people told you why you should book PGi while insisting they adored PGi ? b. *Which bandi did you hire ti only after people told you why you should while insisting they adored PGi ? (53) Adjuncts a. *Which filmi did you refuse to see ti because Roger was so revolted while he watched PGi after renting PGi ? b. *Which filmi did you refuse to see ti because Roger was so revolted when he did after renting PGi ? (54) Complex NPs a. *Mayor Daley, whomi everyone met ti after grabbing the person who had arranged the opportunity to see PGi while pointing at PGi , discussed Chicago politics. b. *Mayor Daley, whomi everyone met ti after grabbing the person who had arranged the opportunity to while pointing at PGi , discussed Chicago politics.
Ellipsis and syntactic representation
As Postal points out, the unacceptability of the (b) sentences is due to the fact that both of the possible syntactic representations of the elided VP – the one corresponding to (45a) and the one corresponding to (45b) – result in ill-formed structures: the missing parasitic gap analysis (45b) triggers an island violation in the first adjunct, and the non-parasitic gap analysis (45a) fails to license the dependent parasitic gap in the second adjunct. This is illustrated by the examples in (55), which show the two possible syntactic representations of (52)–(54). (55) a. *Which bandi did you hire ti [only after people told you why you should [VP hire themi /PGi ] [while insisting they adored PGi ]]? b. *Which filmi did you refuse to see ti [because Roger was so revolted when he did [VP see iti/PGi ] [after renting PGi ]]? c. *Mayor Daley, whomi everyone met ti [after grabbing the person who had arranged the opportunity to [VP meet himi /PGi ] [while pointing at PGi ]], discussed Chicago politics. Finally, Postal shows that missing parasitic gaps also show strong crossover effects: (56) a. *Which candidatei did the linguists support ti because close friends of PGi claimed hei could also get the philosophers to back PGi ? b. *Which candidatei did the linguists support ti because close friends of PGi claimed hei could also get the philosophers to? c. Which candidatei did the linguists support ti because Maureen claimed hei , could also get the philosophers to? (57) a. *Whoi did you call after learning that hei was expecting grey aliens to abduct PGi after locating PGi ? b. *Whoi did you call after learning that hei was expecting grey aliens to after locating PGi ? c. Whoi did you call after learning that hei was expecting grey aliens to? Only the missing parasitic gap analysis of the elided VP licenses the dependent gap, therefore the Us of (56b) and (57b) must be as in (58), triggering a SCO effect. (58) a. *Which candidate did the linguists support ti , because close friends of PGi claimed hei could also get the philosophers to [VP support PGi]. b. *Whoi did you call after learning that hei expected you to [VP call PGi] after locating PGi ?
. Summary To summarize, the interaction of VP-deletion and parasitic gaps provides strong support for a syntactic analysis of ellipsis. First, the facts discussed here clearly
Christopher Kennedy
demonstrate that the elided constituent is sensitive to (at least) Condition B effects, strong crossover (Condition C effects), and various island constraints (including wh-islands, Complex NP Islands, Adjunct Islands and the Coordinate Structures Constraint).3 Assuming that these constraints are constraints on syntactic representations, it must the be the case that elided VPs have syntactic structure. Second, the fact that an elided VP in these contexts can be shown to require a “nonparasitic” gap analysis in some contexts and a “missing” parasitic gap analysis in others – a distinction that is purely syntactic in nature – further strengthens the conclusion that ellipsis involves syntactic representation. In the next section, I will examine a phenomenon that at first glance appears to challenge this conclusion, since it seems to indicate that elided constituents are insensitive to the Left Branch Constraint. As we will see, however, the facts actually substantiate the second part of the predictions of a syntactic analysis outlined in (29): ellipsis constructions are insensitive to syntactic constraints that derive from the morphophonological properties of lexical items.
. Ellipsis and left branch extractions . Attributive comparative deletion As shown by the sentences in (59), comparative deletion constructions that target just an attributive adjective (henceforth “attributive CD” constructions) are ungrammatical (Pilch 1965; Pinkham 1982; Kennedy & Merchant 2000). (59) a. *The Cubs start a more talented infield than the Sox start an outfield. b. *Jones produced as successful a film as Smith produced a play. This fact is unsurprising given Ross’ (1967) observation that comparative deletion is subject to the full range of syntactic island constraints. Assuming for concreteness that comparatives involve null or deleted wh-phrases which originate in the position of the gap (as in e.g. Chomsky 1977b), the examples in (59) have the structures in (60a) and (60b). These are completely parallel to the questions in (61), which violate the Left Branch Constraint (LBC). (60) a. *The Cubs start a more talented infield than [whi the Sox start [DP an ti outfield]]. b. *Jones produced as successful a film as [whi Smith produced [DP a ti play]]. (61) a. *How talentedi do the Sox start [DP an ti outfield]? b. *How successfuli did Smith produce [DP a ti play]? What is surprising is that ellipsis in the comparative clause seems to eliminate LBC violations (Pinkham 1982; Kennedy & Merchant 2000). This is illustrated
Ellipsis and syntactic representation
by the examples in (62), which are just like the sentences in (59) except that a constituent containing the gap has been elided. (62) a. The Cubs start a more talented infield than the Sox (do). b. Jones produced as successful a film as Smith (did). Given the main conclusion of the previous section – that elided constituents have syntactic representation and are subject to configurational constraints – the wellformedness of the examples in (62) is extremely puzzling. On this view, these sentences have the structures shown in (63), which should violate the LBC just as much as those in (60), assuming that the LBC is stated in terms of configurational relations (e.g., in terms of the ECP, as proposed in Corver 1990). (63) a.
The Cubs start a more talented infield than [whi the Sox (do) [VP start [DP an t i infield]]]. b. Jones produced as successful a film as [whi Smith (did) [VP produced [DP a ti film]]].
The solution to this puzzle developed and substantiated in Kennedy & Merchant 2000 (K&M) builds on an analysis of LBC effects in which extraction of attributive modifiers is ruled out not by a configurational constraint governing empty categories or operator/variable dependencies, but rather by a constraint on the morphophonological instantiation of syntactic feature combinations. If this version of the LBC is correct, then these constructions actually instantiate the prediction of the syntactic theory of ellipsis stated in (29b): since ellipsis does not require pronunciation of the omitted structure, elided constituents should be insensitive to syntactic constraints that derive from morphophonological properties of lexical items. K&M’s arguments for a PF characterization of the LBC are summarized in the next section.4
. A PF analysis of the Left Branch Constraint We begin with the syntax of attributive modifiers. Following Svenonius (1992), K&M assume that the base position of an attributive modifier (categorically a DegP) is as a right adjunct to NP Inverted DegPs, such as those in (64)–(65), indicate that some attributive modifiers must move to the left of the determiner (Bolinger 1972; Bresnan 1973; Woisetschlager 1981; Baker 1989; Corver 1990; Hendrik 1990). (64) a. [How interesting a play] did Brio write? b. I ate [too big a piece]. c. If I ever see [that disgusting a movie] again, I’ll ask for my money back.
Christopher Kennedy
d. Bob didn’t write [as detailed a proposal] as Sheila did. e. He took [so big a piece] that he couldn’t finish it. (65) a. *[A how interesting a play] did Brio write? b. *I ate [a too big piece]. c. *If I ever see [a that disgusting movie] again, I’ll ask for my money back. d. *Bob didn’t write [an as detailed proposal] as Sheila did. e. *He took [a so big piece] that he couldn’t finish it. K&M propose that the position an inverted DegP – as well as an intermediate landing site for the wh-operator in comparatives – is the specifier of a functional phrase within the nominal projection but above DP: “FP” (cf. Bennis et al. 1998; see Corver 1990 for arguments that the landing site of inversion is not SpecDP). The basic structure is illustrated in (66). (66)
FP DegP how interestingi
F F
DP NP
D
a DegP ti
NP N play
K&M provide three pieces of independent evidence for this syntactic analysis. First, the head of FP can sometimes be morphologically realized as of, as in the following examples (cf. constructions like a bear of a guy discussed in Bennis et al. 1998). (67) How long of a novel did Brio write? a. I ate [too big of a piece]. b. If I ever see [that disgusting of a movie] again, I’ll ask for my money back. c. Bob didn’t write [as detailed of a proposal] as Sheila did. d. He took [so big of a piece] that he couldn’t finish it.
Ellipsis and syntactic representation
Second, an attributive modifier can be “caught” by pseudogapping. This is illustrated by (68a)–(70a), which are ambiguous between the (b) and the (c) interpretations. (68) a. I have written a successful play, but you have a novel. b. I have written a successful play, but you have written a novel. c. I have written a successful play, but you have written a successful novel. (69) a. The Cubs need a left-handed hitter more than they do a pitcher. b. The Cubs need a left-handed hitter more than they need a pitcher. c. The Cubs need a left-handed hitter more than they need a left-handed pitcher. (70) a. I buy expensive shoes because I don’t suits. b. I buy expensive shoes because I don’t buy suits. c. I buy expensive shoes because I don’t buy expensive suits. Given the assumption that pseudogapping involves movement of the “remnant” phrase out of VP, followed by VP-deletion (Kuno 1981; Jayaseelan 1990; Lasnik 1995b; Johnson 1997b; see Sag 1976a; Levin 1986; Miller 1992 for qualifications), the possibility of the (c) interpretations follows directly from the structure in (66): these readings correspond to structures in which the attributive modifier raises to SpecFP, and DP moves out of (the deleted) VP. This is illustrated for (69a) in (71), where the deleted VP is enclosed in a box. (71)
VP VP V need
DPj D
FP
left-handedi
a
F’
DegP F
DP
NP DegP
NP
ti
pitcher
tj
Crucially, only a syntactic analysis of the sort given in (66) (together with the assumption that pseudogapping involves VP-deletion plus extraction of the remnant XP) can account for the (c) readings in (68)–(70). An alternative explanation of the facts would be one that posited some kind of “attributive modifier ellipsis”, in which the structure assigned to e.g. (69a) would be (72).
Christopher Kennedy
(72) The Cubs need a left-handed hitter more than they do [VP need t i] [DP a [NP [DegP left handed] [NP pitcher]]]i The problem with this sort of analysis is that it also predicts that the (b) sentences in (68)–(70) should have (c) readings, which is false. That is, such an analysis cannot capture the fact that ellipsis of the attributive modifier in these examples is “parasitic” on pseudogapping, whereas this follows directly from the structural analysis proposed in K&M.5 Only an analysis that assumes movement of DegP to SpecFP plus deletion (i.e., syntax on the PF side of the derivation) allows for the construction of a syntactic constituent of the type needed to get the (c) reading only when the verb (phrase) is also deleted. The third argument presented by K&M for the FP-structure in (66) comes from particular uses of the verb make. This verb has an “evaluative” use that requires an attributive modifier, as shown in (73). (73) a. Peaches make delicious tarts. b. #Peaches make tarts. This restriction appears to be relaxed in pseudogapping contexts, however: (74) a. Peaches make delicious pies more often than they do tarts. b. #Peaches make delicious pies more often than they make tarts. The acceptability of (74a) follows directly given the FP structure in (66) and the analysis of the pseudogapping sentences discussed above: (74a) can be assigned the structure in (75), which satisfies the requirement that the complement of make have an attributive modifier.6 (75) Peaches make delicious pies more often than they do [VP make [FP [DegP delicious]j F0 tj ]] [DP ti tarts]j . Taken together, these facts provide strong support for the hypothesis that attributive modifiers can, and sometimes must (see (65) above), raise from their base position at the NP level to the specifier of a functional head within the nominal projection but above the determiner. Since this movement is clearly licensed, it must be the case that the locus of LBC violations is movement out of SpecFP, not movement from the base position of the attributive modifier. That is, the problem must be at the FP level, not the DP level. This conclusion, together with the fact that ellipsis eliminates LBC effects in attributive comparative deletion, lead K&M to propose a formulation of the LBC in terms of the morphophonological expression of syntactic feature combinations. Specifically, K&M claim that extraction of left branch modifiers in English (and related languages) is regulated by the principle of Full Interpretation (Chomsky 1981, 1986a, 1995), which requires that every element in a particular interface represen-
Ellipsis and syntactic representation
tation have an interpretation at that interface. In the case of the syntax-phonology interface, this means that all terminal nodes structured bundles of syntactic features must have a phonological value. Following Halle and Marantz 1993, K&M assume that a syntactic object “has a phonological value” if and only if it can be paired with a corresponding morphophonological matrix from the lexicon. This leaves open the possibility that the syntactic component can derive representations that are well formed in all respects except that they contain objects without morphophonological instantiations. According to K&M, this is exactly what happens in LBC contexts. Like the inverted DegPs, an attributive wh-operator must move through SpecFP (cf. how tall a man vs. *a how tall man), with the result that the head of FP is assigned a [+wh] feature by spec-head agreement. K&M’s proposal is that left branch effects in English arise because the lexicon lacks a F0[+wh] head. On this view, examples like those in (76)–(77) are ungrammatical not because of a constraint on movement per se, but because the syntactic representations are unpronounceable: they violate Full Interpretation at the PF interface, because F0[+wh] has no morphophonological instantiation. (76) *The Cubs start a more talented infield than [whi the Sox start [FP ti F0[+wh] [DP an ti outfield]]]. (77) *How talentedi do the Sox start [FP ti F0[+wh] [DP an ti outfield]]? In order to generate a well-formed structure, the [+wh] feature on F0 must be eliminated. This can happen in two ways. The first option is to pied-pipe the entire FP with the wh-operator, in which case the [+wh] feature on F0 can be checked in the normal way. This is the strategy taken in questions such as (78), but it this option is unavailable in null operator constructions (see Grosu 1994).7 The second option is to delete a constituent containing the offending F0[+wh] object. This is what happens in the well-formed examples of attributive comparative deletion, in which a constituent containing the gap is deleted. (78) illustrates the case where a VP has been deleted. (78) The Cubs start a more talented infield than [whi the Sox (do) [VP start [FP t i F0 [+wh] [DP an t i infield]]]]. We now have an answer to our puzzle: if the LBC is a constraint on the morphophonological instantiation of syntactic representations, and if ellipsis involves deletion of syntactic representations (or, alternatively, an instruction to “bypass” morphophonological instantiation, à la Wasow 1972), then the fact that ellipsis of a constituent containing the gap in attributive comparatives bypasses the LBC follows.
Christopher Kennedy
The formulation of the LBC presented here makes a number of predictions. First, if pseudogapping works as claimed above (see the discussion of (68)–(70)), then like other examples of ellipsis in which FP is included in the deleted VP, it should also license left branch extractions in attributive CD. As K&M point out, this is correct: (79) a. b. c. d.
The Sox start a more talented infield than they do an outfield. Jones produced as successful a film as she did a play. Abby wrote a more interesting novel than she did a play. Erik drives a more expensive car than he does a motorcycle.
Second, if the LBC is a reflex of lexical inventory, reflecting whether or not a language has a morphophonological instantiation of a F0[+wh] head, rather than a structural condition, then it should show a wide degree of cross-linguistic variation. This is true, as pointed out in Ross (1967) and Grosu (1974, 1994). Finally, this analysis predicts the following pattern cross-linguistically (all other things being equal): if a language obeys the left branch constraint, then attributive CD should be acceptable only if deletion also applies; if, however, a language does not obey the left branch constraint, then attributive CD should be acceptable without deletion. This is also correct (see Kennedy & Merchant 2000).
. Summary The facts of attributive comparative deletion show that the elided constituent is insensitive to the Left Branch Constraint. At first glance, this fact appears to challenge the assumption that elided constituents are subject to syntactic constraints. However, if Kennedy and Merchant’s (2000) characterization of the LBC in terms of the principle of Full Interpretation at the PF interface is correct, then these facts represent exactly the type of data we expect to see if ellipsis involves deletion of syntactic structure (where deletion can be formalized either as actual elimination of structure, or as elimination of morphophonological information only). Since deletion bypasses the need for PF interpretation of syntactic representations, Full Interpretation (and similar constraints) is satisfied vacuously by an elided XP.
. Final thoughts The empirical data considered in this paper clearly show that elided constituents must have syntactic representation, but also that it is not the case that elided constituents are predicted to be subject to exactly the same set of constraints and principles as their unelided counterparts. Specifically, elided constituents are predicted
Ellipsis and syntactic representation
to be insensitive to just those constraints that make reference to morphophonological properties of syntactic objects, as specified in (80). (80) Ellipsis and syntactic representation If ellipsis involves deletion of syntactic structure, then: a. Elided constituents should be sensitive to syntactic constraints in general. b. However, since ellipsis does not require pronunciation of the omitted structure, elided constituents should be insensitive to syntactic constraints that derive from morphophonological properties of lexical items. The next step in the research program, then, is to explore further the prediction in (80b). If ellipsis really does work this way, then we should be able to identify a set of (potentially otherwise unrelated) ellipsis constructions that have properties of structures that could not appear overtly. In other words, we should find in ellipsis constructions evidence for the presence of objects that do not appear in other (well-formed) constructions of the language (such as the F0[+wh] head in Englis h). Initial support for this conclusion comes from Merchant’s (2001) work on sluicing and Kennedy and Lidz’s (2001) work on strict/sloppy readings of comparativ e stripping constructions, but there remains much to be done in exploring this prediction in detail. The final question we are faced with is whether a syntactic analysis as I have laid it out here has anything to say about the problematic data discussed in Section 1.1. Of particular interest are cases of active/passive mismatch and syntactic category mismatch, where native speaker intuitions often collide with observations about naturally occurring data. Answering this question will have to be the topic of another paper, but my guess is that Kehler’s (2000) observations about the role of coherence relations in ellipsis, together with recent work on the role of parallelis m and focus structure (in particular the work of Rooth 1992a and Fox 1999a) may ultimately provide a basis for a pragmatic, rather than a syntactic or semantic, explanation of the facts. On this view, the coherence-based principles identified by Kehler would not govern the syntax of ellipsis per se, but rather would govern the felicity of particular uses of ellipsis, as in, for example, the theory of preposing and information structure defended in Ward 1988 and Ward and Birner 1998, whereby the latter crucially determines the felicity of the former, but not its syntactic wellformedness.
Christopher Kennedy
Notes * I am grateful to Susanne Winkler and Caterina Donati for their thoughtful comments on an earlier draft of this paper, and to the participants in the Workshop on Adding and Omitting at the 21st Meeting of the DGfS in Konstanz for very helpful discussion of the material presented here. . See Merchant (2001) for a semantic licensing condition on ellipsis, based on Schwarzschild’s (1999) theory of focus and deaccenting, that derives vehicle change. . This is presumably exactly the right analysis of (i), in which the presence of the VPanaphor it indicates an absence of internal structure. (i) *Whoi did Otis nominate ti because shei couldn’t do it? If this predicate is reflexive-marked, then the sentence is perfectly acceptable, as expected: (ii) Whoi did Otis nominate ti because shei couldn’t do it herselfi ? . It should also be noted that many of the examples discussed above involve cause-effect coherence relations (e.g. the because-adjuncts), and are therefore contexts in which Kehler (1995, 2000) predicts syntactic effects to disappear. . It should be noted that K&M focus specifically on the case of extraction of left-branch attributive modifiers, not e.g. extractions such as (i), which violate the head-movement constraint, assuming that how is the head of a Degree Phrase (DegP) with the AP headed by strong as its complement (see Abney 1987; Corver 1990; Grimshaw 1991; Kennedy 1999). (i) *Howi is Sammy ti strong? As originally observed by Grosu (1974) (see also Corver 1990), the various phenomena that Ross subsumed under the Left Branch Constraint are more properly explained in terms of distinct constraints. . This discussion is stated in terms of a deletion analysis of ellipsis, but the same arguments hold of semantic analyses. A purely interpretive analysis such as the one developed in Dalrymple et al. 1991 runs into exactly the same problems of overgeneration outlined above, since it must posit an operation that recovers “missing” attributive modifier meanings. If such an operation is possible in the (a) sentences in (68)–(70), then it should also be possible in the (b) sentences, contrary to fact. . I set aside here the question of what is responsible for this constraint. One possibility, given the well-formedness of sentences like These peaches would make a hell of a pie is that evaluative make subcategorizes for an FP. . Caterina Donati points out that we also don’t see this kind of pied-piping in comparatives in which the operator is overt, however (in Italian, Bulgarian, and Romanian), so this may not be a complete answer. Movement of the entire FP to SpecCP may be exactly what is going on in examples of attributive CD in which only an argument is missing, such as (i), however.
Ellipsis and syntactic representation
(i) Jones produced as successful a film as [[FP whi F0 [+wh] [DP a t i film]]j Smith produced tj ] This sort of analysis cannot be ruled out, if movement and deletion in SpecCP is part of the grammar of comparatives (see Kennedy 2002 for arguments to this effect). Moreover, given that English does not allow deletion of argument DPs except in movement constructions (assuming a ‘copy and delete’ theory of movement), this may be the only plausible analysis for such comparatives.
Subject-auxiliary inversion in comparatives and PF output constraints Jason Merchant This paper establishes the novel generalization that Subject-auxiliary inversion (SAI) in comparative clauses requires the co-presence of VP-ellipsis, and argues that this peculiar fact follows from a disjunctive formulation of an ECP that applies at PF. The analysis relies crucially on the presence of an intermediate trace of the A -moved comparative operator at the edge of VP, which is subject to the ECP at PF, and which interacts with the head movement involved in SAI. This trace is unlicensed in structures with I-to-C movement, but VP-ellipsis repairs the violation, providing further evidence that ellipsis can repair otherwise deviant structures.*
Introduction The object of this paper is to document and assay an explanation of a novel generalization concerning the co-occurrence of subject-auxiliary inversion (SAI; analyzed here as I-to-C movement) and VP-ellipsis in comparative clauses. Although VPellipsis is usually optional, in comparative clauses in which I-to-C movement has occurred, it is obligatory. This generalization is stated in (1). (1) Comparative SAI and VP-ellipsis generalization I-to-C movement in comparative clauses can occur only if VP-ellipsis has deleted the VP complement to I0 . While this peculiar fact might be easy to encode in grammar formalisms that make direct reference to constructions (such as Construction Grammar or some recent versions of Head-Driven Phrase Structure Grammar; see Goldberg 1995; Sag 1997; Kay & Fillmore 1999), presumably simply by stating the co-occurrence restriction on the type-hierarchies directly (e.g., {SAI} ∩ {comparative wh-movement} ∩ {non-VP-ellipsis} = Ø), it poses a much more severe challenge for theories of grammar that attempt to reduce such apparently construction-specific phenomena to general principles that operate across the grammar without restriction. In this
Jason Merchant
paper, I attempt to show that such a general account is possible, and that the seemingly parochial generalization in (1) does not provide evidence against grammatical theories that eschew reference to constructions (such as current work in Minimalism and Optimality Theory; see Chomsky 1995; Grimshaw 1997; Barbosa et al. 1998). In particular, I argue that the peculiar generalization in (1) finds its explanation in the interaction of the properties of head-movement involved in SAI with the licensing of traces of wh-movement, where the licensing in question must be formulated as a kind of ECP applying at PF. This analysis lends new support both to the claim that wh-movement out of a VP proceeds via adjunction to that VP, and to the idea that certain types of constraints are operative as static, representational output constraints at the PF-interface.
.
Establishing the generalization
I begin by laying out the data that gives rise to the generalization in (1)1 , and continue with a brief comparison to the more well-known case of SAI in matrix wh-questions.
. Comparatives In addition to the usual embedded word order seen in English comparative clauses (CPs selected by than or as), under certain circumstances the highest auxiliary can raise to C:2 (2) a.
Abby knows more languages than does her father. (cf. Abby knows more languages than her father does.) b. Abby can play more instruments than can her father. (cf. Abby can play more instruments than her father can.) c. Abby is taller than is her father. (cf. Abby is taller than her father is.)
When I raises to C, however, the VP complement to I must be elided, as a comparison of (2a, b) with (3a, b) shows; when SAI does not take place, VP-ellipsis is not required, however, as shown in (4). (3) a. *Abby knows more languages than does her father know. b. *Abby can play more instruments than can her father play. (4) a. Abby knows more languages than her father knows. b. Abby can play more instruments than her father can play.
Subject-auxiliary inversion in comparatives and PF output constraints
This can be seen clearly also in the contrastive data with stacked auxiliaries in (5)– (7) (a nominal amount comparative, an attributive adjectival comparative, and an adverbial comparative respectively): (5) a.
Abby has been awarded more accolades than has her father (*been awarded). b. Abby has been awarded more accolades than her father has been awarded.
(6) a.
Abby has been awarded a more prestigious accolade than has her father (*been awarded). b. Abby has been awarded a more prestigious accolade than her father has (been awarded).
(7) a.
Abby has been playing piano longer than has her father (*been playing piano). b. Abby has been playing piano longer than her father has been playing piano.
It is not the case that SAI in comparatives is possible just in case some kind or other of VP-ellipsis occurs in the comparative clause – the target of VP deletion must be the highest VP, the complement to I. In (8a), for example, the lower VP headed by awarded has been deleted, leaving the higher VP headed by been. In each case, the result is ungrammatical. In the absence of SAI, however, VP-ellipsis is free to target the lower VP; this is shown in the control cases in (9). (8) a. *Abby has been awarded more accolades than has her father been. b. *Abby has been awarded a more prestigious accolade than has her father been. c. *Abby has been playing piano longer than has her father been. (9) a. Abby has been awarded more accolades than her father has been. b. Abby has been awarded a more prestigious accolade than her father has been. c. Abby has been playing piano longer than her father has been. These data are supported by the results of two corpus searches. The first consisted of a search for SAI in comparative clauses using the AltaVista search engine, and was concluded on 29 October 1999, with the results given in Table 1. The search strings were than does, than do, and than did (other auxiliaries returned too high a rate of false hits to be useful). The number of hits (web pages that contained at least one occurrence of the search string) for each search string is given in column B. The AltaVista search engine, however, only makes available to the user a maximum of 200 web pages, reducing the actual available sample to 200 in each case (the crite-
Jason Merchant
Table 1. AltaVista search for SAI with VP-ellipsis in comparatives A
B
C
D
E
F
Search string
Number Sample Sample Number of Number of of hits available inspected than + SAI (N) N containing VP-ellipsis (V) than does 45,581 200 20 26 26 than do 68,627 200 20 22 22 than did 46,086 200 20 21 21
G V as a percentage of N 100 % 100 % 100 %
ria for the selection of the 200 is proprietary information of the AltaVista search engine). A random sample of ten percent of the available pages was examined for the search string. The number of occurrences of the search string that instantiated than with SAI (N) is given in column E (in some cases, a single webpage contained multiple instances of the search string). In each case, the context of the string was examined to determine whether or not VP-ellipsis was co-present, given as V in column F. The resulting percentage is reported in column G. No counterexamples to the generalization in (1) were found. The second corpus search was a search for the same strings in J. Hoeksema’s 16 million word database of texts. The number of hits were as follows: 24 occurrences of than does, 26 occurrences of than do, and 28 occurrences of than did. And again the results were identical: in all cases, SAI in a comparative clause was accompanied by VP-ellipsis.
. Matrix wh-questions The connection between SAI and VP-ellipsis seen in comparatives contrasts with the state of affairs found in the most common environment for SAI in English: matrix wh-interrogatives, where SAI is obligatory. Here, the restriction noted above for comparatives does not hold – no VP-ellipsis is necessary: (10) a. b. c. d. e.
How many languages does Abby know? How many instruments can Abby play? How many accolades has Abby been awarded? How prestigious an accolade has Abby been awarded? How long has Abby been playing the piano?
There are several other environments in which SAI occurs in standard English, but comparison with such cases is less illuminating, for reasons we will see directly. Consideration of such cases will thus be postponed until below, and I will concentrate first on the asymmetry between the constraints on SAI in comparatives vs. matrix wh-questions.
Subject-auxiliary inversion in comparatives and PF output constraints
. Intermediate wh-traces and the Empty Category Principle at PF We are now in a position to confront the analytical challenge presented by the data above. First, why is VP-ellipsis necessary in comparatives with SAI, and second, why does SAI in matrix wh-questions not have a similar effect? Let us begin by answering the first question. In essence, it seems that the VP deletion is saving an otherwise illicit structure; VP-ellipsis repairs some kind of defect brought about by SAI. I would like to suggest that this defect is the ill-formedness of the intermediate trace of wh-movement that occurs in the comparative clause. A number of lines of evidence indicate that wh-extraction out of a VP proceeds via adjunction to that VP (see Chomsky 1986a; Fox 1999b; López & Winkler, in this volume; cf. also Koster 1978). If this is so, a regular comparative clause such as that in (4b), repeated here in (11a), will have the representation in (11b). (11) a. Abby can play more instruments than her father can play. b. . . . than [CP Op1 her father can [VP t1 [VP tSU play t1 ]]] In this structure, the comparative operator (here represented simply as Op; see Lechner 1999; Kennedy 2002 for refinements) has been extracted from the object position of play, marked by t1 , and has moved to specCP by first adjoining to the VP, marked by t1 . The intermediate trace t1 will be subject to general well-formedness conditions on traces, in particular, it will be subject to the Empty Category Principle (ECP). I modify here the disjunctive approach to the ECP advocated in Chomsky (1981), (1986a) and much other work. Crucially, however, this condition applies at PF, and is independent of any condition that may apply at LF as well, about which I will have nothing to say here. Conceptually, this approach is reminiscent especially of the proposal in Aoun et al. (1987) (also Rizzi (1990: 39) for proper head-government, and cf. Jaeggli (1982)), who argue that part of the ECP applies at PF, though the details differ substantially. (12) The Empty Category Principle at PF (ECPPF ): At PF, a trace of A -movement must either be i. PF-head-governed, or ii. PF-antecedent-governed Both disjuncts contain versions of familiar notions, here applied to PF representations. These PF versions of head-government and antecedent-government are defined below; the (i) clauses follow the definitions in Rizzi (1990) in essentials – the innovations are the additional (PF) requirements imposed by the (ii) clauses. I begin with the definition of PF-head-government, given in (13).
Jason Merchant
(13) α PF-head-governs β iff i. a. α is a head, and b. α c-commands β, and c. α respects Relativized Minimality wrt β, and ii. α is PF-active (14) A link αi in a chain < α1 , . . . , αn > is PF-active iff αi is the link at which lexical insertion occurs The idea in the definition in (14) is that the PF interface privileges one copy – αi – of a moved element for lexical insertion (in the sense of Halle & Marantz 1993) over other links in the chain formed by movement (the notion ‘chain’ being only a derivative one, presumably reducible to a consequence of the definition of move); the features of αi are said to be active at PF (note that this does not entail that this element will be pronounced, given the existence of null lexical items such as null operators, null I0 , null D0 , etc.). How this computation is effected is not relevant here (but see Richards (2001) for one theory of this selection); in general, αi in (14) will be the head of the chain in the cases of present interest.3 The definition of PF-antecedent-government is given in (15). (15) α PF-antecedent-governs β iff i. a. α and β are co-indexed, and b. α c-commands β, and c. α respects Relativized Minimality wrt β, and ii. α is PF-visible (16) An expression α is PF-visible iff α has phonetic exponence I understand having phonetic exponence as meaning that the element in questioned is pronounced (cf. the equivalent notion of having a phonetic matrix). In the present cases of interest, wh-operators, the notion corresponds exactly to the traditional distinction between overt and empty (or null) operators; only overt operators can satisfy clause (ii) or the ECPPF .4 In an example such as (11) above, then, these definitions will have the following consequences. In (11b), the lower trace of wh-movement t1 is PF-headgoverned by plays: plays is a head, c-commands t1 , no other head intervenes, and at PF, it forms a single-membered chain, and lexical insertion targets it. The intermediate trace of movement, t1 , is PF-head-governed by the modal can in I0 : can is a head, c-commands t1 , no other head intervenes, and lexical insertion targets it at PF. Therefore, both traces in (11b) satisfy the ECPPF , by virtue of satisfying clause (i) of (12).5 Consider now what happens when SAI occurs, as in (3b), repeated here in (17a).
Subject-auxiliary inversion in comparatives and PF output constraints
(17) a. *Abby can play more instruments than can her father play. b. . . . than [CP Op1 can [IP her father tcan [VP t1 [VP tSU play t1 ]]]] This example has the structure in (17b), parallel to (11b) above except for the presence of I-to-C movement. I-to-C movement in the comparative clause moves the modal can out of the IP, forming the chain < can, tcan > (again, presumably independent principles determine that the higher copy is pronounced in this structure, and that the lower copy remains unpronounced, signaled here with the traditional t). While the lower trace t1 satisfies the ECPPF as above, the I-to-C movement changes the licensing status of the intermediate trace t1 : the lower copy of can, tcan in I0 , does not PF-head-govern t1 because it is not PF-active (the higher copy is, in C). This means that the ECPPF in (12) cannot be satisfied via clause (i) as it was in (11). The other possibility for satisfying the ECPPF is via clause (ii) of (12), which requires PF-antecedent-government. While Op in specCP does antecedent-govern t1 , it does not satisfy the crucial PF-requirement in (15ii): Op is not PF-visible, that is, pronounced (perhaps for substantive reasons, as explored in Lechner 1999; Kennedy 2002). Thus the intermediate trace of Op violates the ECPPF and (17a) is correctly ruled out. We are now in a position to account for the otherwise puzzling generalization that SAI is possible in comparatives just in case VP-ellipsis has targeted the VP sister to I0 , as established in Section 1.1. I assume that ellipsis is deletion of a constituent at PF (see Merchant 2001 for extensive discussion and references); for VP-ellipsis, this means that a VP is deleted in the derivation from Spell-Out to the PF representation. For a non-comparative example such as (18a), this deletion is represented as in (18b), where strikethrough indicates deleted material, inactive at PF. (18) a. Abby can play the piano, but her father can’t. b. Abby can play the piano, but her father can’t [VP t SU play the piano]. Consider now what effect such deletion will have in comparatives in which SAI has occurred, such as (2b), repeated here. (19) Abby can play more instruments than can her father. In such an example, wh-movement leaves an intermediate trace adjoined to the highest VP, as we have seen. But in addition, VP-ellipsis has applied, deleting the highest VP segment. The resulting PF structure is given in (20). (20) . . . than [CP Op1 can [IP her father tcan [VP t 1 [VP t SU play t 1 ]] ]] Because VP-ellipsis has deleted the intermediate trace t1 contained in the deleted VP, t1 will not violate the ECPPF ; deletion removes the offending trace from the PFobject considered by the ECPPF , since the ECPPF is a well-formedness condition
Jason Merchant
applying at the PF interface. The logic here is very similar, then, to the logic applied to other PF violations voided by ellipsis as discussed in Lasnik (1995b, 1999), Kennedy and Merchant (2000), and Merchant (2001). This explanation extends as well to the data in (5)–(8) above; as we saw there, VP-ellipsis must target the highest VP. The deletion of a lower VP does not suffice to rescue the structure.6 Under the proposed analysis, this is the expected result, assuming as before that the wh-movement targets (at least) the highest VP.7 In (8a), repeated here, for example, the deletion of the VP headed by awarded does not remove the intermediate trace t1 adjoined to the VP headed by been from the PF structure submitted to the PF well-formedness requirements – as such, t1 remains at PF and violates the ECPPF . (21) a. *Abby has been awarded more accolades than has her father been. b. . . . than [CP Op1 has [IP her father thas [VP t1 [VP been [VP awarded t SU t 1 ]]]]] Let us now reconsider the case of SAI in matrix wh-questions, where VPellipsis is not required, such as in (22), repeated from (10b) above. (22) How many instruments can Abby play? This has the structure given in (23). (23) [CP How many instruments1 can [IP Abby tcan [VP t1 [VP tSU play t1 ]]]]? As in (17) above, t1 does not satisfy clause (i) of the ECPPF . However, unlike in (17), t1 here satisfies clause (ii): the operator How many instruments is PF-visible, and PF-antecedent-governs t1 . Because of this, there is no similar requirement for VP-ellipsis in matrix wh-questions.8 At this point, it is important to examine the generality of the analysis presented thus far. While it is successful in handling for an otherwise puzzling and until now unaccounted-for generalization, one may wonder whether it has undesired consequences in domains other than the two examined so far. In particular, it makes the claim that there will be no instance of wh-movement out of VP co-occurring with I-to-C movement in which the operator is not PF-visible (in traditional terms, which involves an empty operator). This claim is fairly simple to check, since the environments in which I-to-C movement in English occur are quite limited, which we can consider in two groups. The first group consists of yes-no questions (24), literary counterfactuals and concessives (25), non-wh-exclamatives (26), and imperatives and hortatives (27). (24) a. Can she play the piano? b. Does he know many languages?
Subject-auxiliary inversion in comparatives and PF output constraints
(25) a. Had he been on time, we might have made it. b. Be it ever so humble, there’s no place like home. (26) a. Man, can she play the piano! b. Am I ever glad to see you! (27) a. Don’t everybody get up at once! b. May he be a joy to you forever! These various constructions have in common that there is no overt operator in specCP, and indeed no reason to assume that there has been wh-movement out of the VP at all (though it is sometimes argued that there is some kind of null operator present in specCP for various reasons, this operator is not usually assumed to originate within the clause; see McCloskey (1991: 295), Potsdam (1996), and Bennis (1998)). These constructions are thus irrelevant to our present concerns. The second group consists of affective inversion structures (28), and no sooner(29a), little- (29b), so- (30), and as-constructions (31). (28) Rarely have I ever been so surprised. (29) a. No sooner had he arrived than it started to rain. b. Little did he suspect that he had already been betrayed. (30) Abby can swim a mile, and so can Ben. (31) Abby got the Nobel Prize, as did her father. In these cases, some element does precede the moved auxiliary in C0 . For affective inversion structures (see Liberman 1974), it has been argued in Haegeman and Zanuttini (1991) that the extracted operator (rarely in (28)) is in specNegP, but the exact location of this specifier and its head (into which I0 moves) does not matter: in either case, since the operator is overt, the ECPPF will be satisfied by clause (ii) of (12). Similar remarks apply to (29) and (30), though it is even less clear that any extraction has occurred (though VP-ellipsis may be highly preferred with so, it apparently does not have the same categorical status that the comparative examples above do; see Chomsky and Lasnik (1993) and Quirk et al. (1985: 882) for examples of so with SAI without VP-ellipsis9 ). This leaves the case of as. This case bears the strongest resemblance to the comparative data considered above, since VP-ellipsis is required as well (compare (31) with *Abby got the Nobel Prize, as did her father get the Nobel Prize). This state of affairs can be accounted for if the as-construction involves some kind of extraction of an operator to specCP, an assumption strongly supported by island -sensitivity and other evidence (see Postal 1998 and Potts 2002). In particular, it supports the idea that as is not itself the extractee (since in that case, it would be an overt operator, and should not require VP-ellipsis with SAI), but rather that as
Jason Merchant
is a head that co-occurs with a null operator (as with the equative as in Abby speaks as many languages as does her father (*speak) etc.).10 The intermediate trace of this null operator, then, will be subject to the ECPPF as above, with the expected results. Thus it appears that the present analysis achieves sufficient generality, applying equally well to other cases in which SAI occurs. Similarly, the vast majority of constructions in which wh-movement takes place (including relative clauses, clefts, pseudoclefts, ‘topicalizations’, free relatives, and embedded wh-questions) will not be relevant to distinguishing the effects of the ECPPF from those of the traditional ECP applying only at LF, since in none of these cases do we find concomitant I-to-C movement.11
. Further considerations At this point, we have seen that the ECPPF can successfully account for the generalization given in (1), and have been led to an interesting view of the kinds of constraints that can operate at the PF interface, an issue that I return to in the Conclusion below. Before concluding however, three further points are worth briefly exploring: comparatives with overt operators, pseudogapping, and V-to-I movement, which will be taken up in the following three sections in turn.
. Comparatives with PF-visible operators The first point is an intriguing prediction of sorts that the analysis makes, which as it turns out is unfortunately difficult to test for independent reasons. We have seen that in standard English comparatives, in which the operator in specCP is nonovert, SAI is impossible without concomitant VP-ellipsis. But the account given leads us to expect that if the operator were overt, SAI should be possible even without VP-ellipsis. There are varieties of American English that permit an overt operator, namely the invariant what, in some comparatives (see Hankamer 1973b). Although it is somewhat difficult to find speakers for whom judgments involving such comparatives are absolutely secure, due to prescriptive prohibitions against the overt operator, it appears that the kind of example we would expect to find are nevertheless ungrammatical: (32) a. *Abby can sing more songs than what can her dad sing. b. *Abby knows more languages than what does her father know. Despite first, worrisome appearances, these data in fact shed no light on the account given above, because SAI in comparatives is unacceptable in these varieties even with VP-ellipsis:
Subject-auxiliary inversion in comparatives and PF output constraints
(33) a. *Abby can sing more songs than what can her dad. b. *Abby knows more languages than what does her father. Thus it is clear that the source of the judgments in (32) lies elsewhere, and does not have to do with the ECPPF . One possibility is that the examples in both (32) and (33) are judged unacceptable because they involve a fatal register clash: while use of the overt operator what in comparatives is colloquial and belongs to a very low, informal spoken register, the use of SAI in comparatives belongs to an extremely high, formal register of English. In other words, no single register’s grammar can produce these two in a single sentence (barring intentional register mixing).12 Another possibility, suggested to me by E. Potsdam, is that the syntax of the what-clauses in (32) and (33) is essentially that of (free) relatives, an analysis made plausible by other evidence as well (animacy restrictions, the unrestricted interpretations found in attributive comparatives, etc.); see den Besten (1978) for extensive discussion. If this is the case, then we do not expect to find SAI in any case, since SAI is excluded in (free) relatives in general.
. SAI, VP-ellipsis, and pseudogapping The second point worth exploring is the interaction of the present cases with pseudogapping. Pseudogapping as in (34a) is most widely analyzed as consisting of adjunction of a remnant XP to VP followed by VP-ellipsis (see Kuno (1981), Jayaseelan (1990), Kennedy and Merchant (2000), Johnson (2001); see Lasnik (1995b) for a related but differing viewpoint), represented in (34b) as left-adjunction for convenience. (34) a. I eat meat, but I don’t seafood. b. . . . I don’t [VP seafood [VP t SU eat t OB ] ] Pseudogapping is compatible with SAI, as the examples in (35) show (from Sag (1976a) and Kempson et al. (1999: 282 fn 43), respectively), though as Levin (1986) points out, such examples are somewhat marked. (35) a. A: Hey! I’ve never seen you on campus before. B: Nor have I you! b. John didn’t give a nickel to Mary, nor did I a dime to Sue. Pseudogapping also occurs in comparative clauses, as in (36); in fact, the best examples of pseudogapping are precisely in comparatives, as Levin (1986) observes. (36) Abby plays the flute better than her father does the trumpet. Given Levin’s observation, it is particularly surprising that pseudogapping with SAI in comparatives is impossible, as (37) shows.13
Jason Merchant
(37) *Abby plays the flute better than does her father the trumpet. We can make sense of this fact if wh-extraction out of a VP must target the highest VP-segment. In this case, an example like (37) will have the structure in (38). In (38), VP-ellipsis has deleted the VP segment below the adjoined remnant the trumpet, leaving the intermediate trace t1 intact at PF, where it violates the ECPPF . (38) . . . than Op1 does her father tdoes [VP t1 [VP the trumpet [VP t SU play t OB t 1 ] ]] Crucially, a derivation in which wh-extraction targets VP first, followed by remnant adjunction, followed by VP-ellipsis must be ruled out, as this would lead to the PF representation in (39), where t1 can be deleted along with the intermediate VP segment (see Sag 1976c; Merchant 2000a for discussion of VP segments as targets of deletion). (39) . . . than Op1 does her father tdoes [VP the trumpet [VP t 1 [VP t SU play t OB t 1 ]] ] There is in fact independent indication that such a restriction on the ordering of multiple A -movement is in force (see Fox 1999b), which presumably should be derived (see Chomsky 2001’s notion of ‘phases’ and extractions from them). If (38) represents the only licit derivation, then, the absence of pseudogapping with SAI in comparatives also falls out from the analysis proposed above. Crucially, this restriction holds only for multiple (A -) extractions out of VP, utilizing adjunction to VP as a landing site, and does not hold of base-generated VP-adjuncts such as adverbials. As Potts (2002) points out, the restriction regarding pseudogapping does not extend to what are presumably VP adjuncts: (40) Klaus would be happier in the north than would Chuck in the south. (Potts 2002: 639(39a)) Fully exploring the contrast between examples like (40) and that in (37) would take us beyond the scope of the present inquiry, but several lines of analysis are open to us. First, one might wonder whether in these case the attested adverbials are in fact adjoined to VP, and not to some higher projection (I , IP); determining the exact adjunction site of right-adjoined adverbials is notoriously difficult, and is complicated by the fact that they can undergo rightward movement as well (in this regard it is important to note that the remnant of pseudogapping must precede such adverbials: cf. I’d build a house in the north before I would a cabin in the south vs. *I’d build a house in the north before I would in the south a cabin; see Sobin (1981) for contrasting data from gapping, for instance). A second possibility, assuming in the south in (40) is in fact right-adjoined to VP, would be to explore whether the
Subject-auxiliary inversion in comparatives and PF output constraints
directionality of adjunction might make a difference: if it could be shown that the remnant of pseudogapping, presumably like wh-movement, left-adjoins to VP, one might pursue an account of the differences between such remnants and adjuncts based on this fact. Third, one might claim simply that adverbial elements don’t factor in the computation at PF in a relevant way, as proposed for other reasons by Bobaljik (1995); here, we might assume that VP-ellipsis can target the highest VP segment, eliminating the illicit trace in the way discussed, but not necessarily delete a right-adjoined adverbial (following in essence Sag’s 1976c proposal). Finally, one might claim, if the attested restriction on extraction ordering is to be derived from phase-edge effects following Chomsky (2001), that merger of adjuncts at the VP doesn’t affect the computation of the edge, allowing a moved element below the merged adjunct to still qualify as at the phase-edge (allowing further extraction), whereas multiple movements to the VP edge (as in the pseudogapping cases) does affect this computation. Essentially, we would claim that only one moved element, the highest one, counts as being at the phase-edge (this must be related in some manner to the differing features driving the movements involved, since multiple wh-extraction is permitted, of course; I leave open the many questions that arise). If this latter line of analysis is on the right track, the following picture emerges: when both the comparative operator and a pseudogapped remnant are moved, the resulting VP structure is that in (41a), where underlining indicates elements at the edge of the VP-phase – because the wh-operator must be extracted further, to specCP, it must be on the edge of the phase, as discussed above. But Merge targeting a VP that has a comparative operator adjoined to it by previous movement, adjoining a VP adjunct (represented as left-adjoined in (41b) for ease of comparison with (41a)), does not alter the status of the moved XP at the edge, resulting in either an expanded edge, as indicated with underlining in (41b), or perhaps simply not affecting the edge computation at all.14 . . . than Op1 does her father tdoes [VP t1 [VP the trumpet [VP t SU play t OB t 1 ] ]] b. . . . than Op1 would Chuck twould [VP in the south [VP t 1 [VP t SU be t 1 ]] ]
(41) a.
If this analysis is on the right track, the contrast between base-generated adjuncts and remnants of pseudogapping for blocking further wh-movement of the comparative operator falls out from the independent computation of phase-edges. Because the intermediate landing site of the comparative operator in (41b) is below the adjunct, but still on the edge, the operator can be extracted to specCP, and the VP to which t1 is adjoined can be deleted as indicated, saving the structure from an ECPPF violation as in (20) above. It is thus clear that there are various analytical options available to us to account for the contrast between (37) and (40), though deciding between them
Jason Merchant
should ideally be done on the basis of evidence unrelated to the data from comparatives; since doing so is not material to my present purposes, such an expanded investigation can safely await further research. Pseudogapping raises another interesting issue, however. In such examples, the object of comparison can be an attributive adjective originating in the remnant of pseudogapping, as in (42a). Following Kennedy and Merchant (2000), I assume that such examples involve movement of the comparative operator (here, a DegP) out of the DP a concerto, followed by movement of a concerto to VP, followed by VPellipsis (here obligatory, for reasons discussed in detail in Kennedy and Merchant 2000); this derivation results in the structure in (42b). (42) a. Abby played a longer sonata than her father did a concerto. b. . . . than Op2 her father did [VP t2 [VP [a t2 concerto]1 [VP play [FP t 2 [F F0 t 1 ]]] ]] Given the analysis developed so far, we expect that in such cases, SAI will be impossible, since t2 in (42b) will violate the ECPPF . This prediction is correct: (43) *Abby played a longer sonata than did her father a concerto. Further, it is interesting to note that the remnant of pseudogapping can itself be the object of comparison, in nominal amount comparatives such as (44) (here, VP-ellipsis is only optional, as noted in Kennedy and Merchant (2000: 101 fn 10)). (44) Abby can play more sonatas than her father can concertos. In these cases, as with attributive comparatives, SAI and pseudogapping are incompatible: (45) *Abby can play more sonatas than can her father concertos. This follows under the present account if such nominal subdeletions involve movement of an empty operator to specCP, as has sometimes been argued for (see Chomsky 1977b; Kayne 1981: 98–99; and Izvorksi 1995, but see Bresnan 1975; Grimshaw 1987; Corver 1993; Kennedy 2002 for complications), an analysis suggested especially by the fact that nominal subdeletions are sometimes sensitive to the availability in a given language of preposition-stranding (see Milner 1978; Kayne 1981).15 Finally, note that SAI is not found in predicative adjectival subdeletion structures either: (46) *Abby is taller than is her father short. This fact is again expected under the present analysis if these structures involve movement of an empty operator from a VP-adjoined position (as mooted by
Subject-auxiliary inversion in comparatives and PF output constraints
Izvorski 1995 in particular) or a VP-internal one (a possibility supported by the fact that attributive adjectival subdeletion is impossible in general, reducible to a Left Branch effect if the operator moves from the DegP inside the DP).
. V-to-I movement The last point worth examining here is the interaction of V-to-I movement with the ECPPF , with a particular view to seeing what light the present analysis can shed on the nature of this movement, especially vis-à-vis I-to-C movement. The question that arises is whether V-to-I movement has the same effect on the licensing of wh-traces as I-to-C movement has. The answer is no, as we will see immediately. This fact raises a substantive issue: is this non-parallel behavior of head movement to be traced somehow to the differing nature of V-to-I vs. I-to-C movement, or does it derive from independent properties of the resulting structures? This in turn raises an analytical issue: is it possible to reformulate the definitions given above to retain the advantages of the account constructed thus far without faltering on V-to-I movement, and if so, what is the best way to do this? The evidence bearing on these questions is at times quite subtle, and to a great extent dependent on considerations of a more global nature than I have been concerned with here. Before tackling these questions, let me lay out what the relevant data will be. V-to-I movement occurs in the most widely spoken varieties of English only with inflected forms of the verb be and the auxiliary have (modals being basegenerated in I0 , and other auxiliaries being quite marginal). There are two kinds of cases that will be of interest here: first, where the operator involved originates as the complement to be, as in (47), and second, where the operator extracts from a series of VPs selected by a verb that raises into I0 , as in (48). (47) a. I didn’t know how fast she was. b. He’s not the man he was. (48) a. They asked what instruments she had been playing. b. Abby has been playing the same instruments her father has been. c. The Nobel is the only prize (which) Abby hasn’t been awarded. For the first set of cases, the relevant portion of the structure is given in (49). (49) a. . . . [CP how fast1 [IP she was [VP t1 [VP twas t1 ]]]] b. . . . [CP Op2 [IP he was [VP t2 [VP twas t2 ]]]] The problem that emerges here is how the lower traces (t1 and t2 , respectively) satisfy the ECPPF . (The higher, VP-adjoined traces will be licensed by clause (i) of the ECPPF as we have seen.) In (49a), t1 does not satisfy clause (i) of the ECPPF as formulated in (12) because its head-governor, the trace of was, is not PF-active as re-
Jason Merchant
quired for PF-head-government in (13); was in I0 itself is prevented from being the head-governor by Relativized Minimality (necessarily so, since if all moved heads could govern through their traces, the I-to-C movement effects would remain a mystery). This leaves the possibility that the ECPPF is satisfied via PF-antecedentgovernment. But this possibility also fails: the closest antecedent-governor for t1 is t1 , by Relativized Minimality, and this trace is not PF-visible as required for PF-antecedent-government. So t1 in (49a) fails to be either PF-head-governed or PF-antecedent-governed, and the structure should violate the ECPPF , a conclusion in conflict with its status as perfectly grammatical. Precisely the same reasoning applies to (49b). The examples in (48) present a different prima facie difficulty, if we assume that wh-movement proceeds cyclically by adjunction to every VP in stacked VP structures. If this is so, these examples will have the structures in (50), where only the traces of interest are given. . . . [CP what instruments1 [IP she had [VP t1 [VP thad [VP t1 [VP been playing t1 ]]]]] b. . . . [CP Op2 [IP her father has [VP t2 [VP thas [VP t2 [VP been [VP playing t 2 ] ]]]]]] c. . . . [CP (which3 )[IP she hasn’t [VP t3 [VP thas [VP t3 [VP been [VP awarded t3 ]]]]]]]
(50) a.
The difficulty comes from the trace which is adjoined to the VP complement of the base position of the raised verb: t1 , t2 , and t3 in (50a), (b), and (c), respectively. Here we face a similar situation to the one just described for complements of moved be: t1 in (50a) is neither PF-antecedent-governed (since the closest antecedentgovernor is t1 , which is not PF-visible) nor PF-head-governed (since thad is not PF-active). Of course, this kind of example only poses a difficulty if wh-extraction proceeds via adjunction to every VP: if, on the other hand, extraction proceeds only via adjunction to the highest VP in such stacked VP structures, nothing more need be said, since a trace adjoined to this VP is PF-head-governed by the raised verb in I0 . Although it has not been settled in the literature whether such cyclic adjunction is necessary, I will proceed on the assumption that under the system generally used here, extraction will result in such intermediate adjoined traces. I take it that the difficulty in these two kinds of examples can be traced back to the same source and that, since the difficulty cuts across the distinction between overt and null operators, this source is to be located in the nature of PF-headgovernment. In some fashion, and counter the naive expectation raised by the case of I-to-C movement, the structures in (49) and (50) manage to satisfy the ECPPF despite the attested verb movement: for some reason, V-to-I movement does not have the effect of disrupting PF-head-government of traces that would have been
Subject-auxiliary inversion in comparatives and PF output constraints
PF-head-governed from the original position of the moved V. The question that must be answered, then, is why this should be so. There are two possible answers to this question. The first takes as its starting point the idea that the traces of the moved verbs are PF-head-governors despite that fact that they are not, under our current definitions, PF-active; in other words, why is it that a trace of a verb (tV ) raised to I0 qualifies as a PF-head-governor, but the trace of I (tI ) raised to C0 does not? Posing the question this way leads us to search for a principled distinction between the kinds of movement involved, considering especially properties of the moved elements and the positions they are moved from. Indeed, it is not difficult to find discussions of the quite different nature of the inherent governing properties of V vs. I: much of the literature on the nature of (especially proper) head-government has stressed exactly the defective nature of I0 , in contrast to the more durable ability of V0 to properly head-govern, accounting especially for subject-object extraction asymmetries in these terms. Following this line of thought, we could revise the definition of PF-active in (14) to be sensitive to these differences: in terms of the theory of Chomsky (1986a), for example, where V L-marks its complement but I does not (Chomsky 1986a: 79), we can redefine PF-active as follows: (51) A link αi in a chain < α1 , . . . , αn > is PF-active iff αi is the link at which lexical insertion or L-marking occurs. Under this definition, the lowest position in a chain of verb movement is always a PF-head-governor, since it is from this position (αn ) that L-marking proceeds (i.e., a verb V L-marks its complement when V is in the position αn , not in any other, derived, position). Viewed from the perspective of positions in the tree, this kind of solution claims essentially that the position I is defective with respect to PF-head-government, and must be targeted by lexical insertion in order to qualify as a PF-head-governor, whereas V always qualifies as such, regardless of whether it is the position targeted by lexical insertion or not. This approach bears most similarity to the approach to defining head governors advocated in Chung (1991), who restricts head governors to those heads that either belong to a given set (among which V), or else are phonetically overt (Chung 1991: 120). Though the class picked out there is slightly different from the one required here, the similarity is otherwise close: lexical heads are ‘prototypical head governors’ (Chung 1991: 107) whose head-government properties are invariant, while more defective functional heads like D must be phonetically overt to qualify as head governors (see also Chung (1998: 276–322) for extensive recent discussion). The approach also bears a similarity to the definition of ‘PF-visible’ advocated in Aoun et al. (1987), who propose that a head H will qualify as a PF-head-governor either if H is phonetically overt or if H has acquired an ‘index through the ap-
Jason Merchant
plication of grammatical processes like a syntactic movement rule. . . ’ (Aoun et al. 1987: 539). They leave it somewhat open what may count as a qualifying ‘grammatical process’; if V-to-I movement, but not I-to-C movement, were posited to pass on an index in their system, the distinction necessary here could be made as well (a distinction that may well be derivable from independent factors, if V-to-I movement occurs in the syntax before Spell-Out, but I-to-C at PF).16 Such an analysis, though perfectly adequate and consistent with previous theoretical work, suffers primarily from the need to posit a disjunction in the definition of PF-active; as Rizzi (1990: 76–77) points out, disjunctive statements sometimes point to a deficient understanding of the underlying properties at issue, though they may be productive and useful research vehicles nonetheless. This metatheoretical concern is essentially the only objection one could bring to the redefinition of PF-active in terms sensitive either to lexical insertion or L-marking (or its equivalent) in (51). The second possible answer to the question of why V-to-I and I-to-C movement have the variable effects seen in English explores the possibility of imposing a PF-adjacency requirement on PF-head-government, as Aoun et al. (1987: 573–575) do (cf. also Kayne 1981: 98). The simplest way to do this is to alter the definition of PF-head-government in (13) above by replacing clause (i.c) (the Relativized Minimality requirement) by a clause requiring that α be PF-adjacent to β. Two elements can then be defined as PF-adjacent iff no PF-visible head or specifier separates them.17 This additional requirement draws the desired distinctions, as a brief consideration of the relevant cases shows. In the case of I-to-C movement, the subject will intervene between I in C and the VP-adjoined wh-trace, blocking PF-adjacency, with the results of the analysis unaffected. But in the case of V-to-I movement, the raised V remains PF-adjacent to the trace in question. In the first set of cases, where the lowest trace of wh-movement is the complement position of the raised verb, the redefinition has the following effect. In (49a), for example, (the relevant part of whose structure is . . . was [VP t1 [VP tSU twas t1 ]])t1 and t1 are both PF-adjacent to the raised verb was in I0 : t1 because nothing intervenes between it and I0 , and t1 because neither the trace of the raised subject in specVP nor the trace of was itself counts as an intervener, since they are not PF-visible. In the second set of cases, the problematic trace is adjoined to the VP complement to the origin site of the raised verb in I0 . Consider for example (50a), the relevant part of whose structure is the following: . . . had [VP t1 [VP thad [VP t1 [. . . In this structure, t1 is PF-head-governed as before, since had in I0 is PF-adjacent to it. The intermediate trace that concerned us before, t1 , is now PF-head-governed by had in I0 as well, since t1 is PF-adjacent to had – the intervening trace thad does not block
Subject-auxiliary inversion in comparatives and PF output constraints
PF-adjacency because it is not PF-visible (and likewise for the possible trace of the subject in specVP, if subjects must move through all intervening VP-specifiers). Again, we see that it is possible to construct an adequate account of the facts by appropriately redefining PF-head-government, in this case to include a PFadjacency condition. And once again, primarily metatheoretical concerns might trouble us on this point: adjacency conditions are usually proposed with a modicum of embarrassment, since in general it seems that adjacency is something we would like to derive from other requirements, and not stipulate in the syntax. But note that it is not debated that adjacency plays an important role in grammar, most prominently in numerous phonological effects (see Walker 1998 for discussion). Any definition making use of adjacency, then, might be expected to be active on the PF side of the grammar, precisely the case here. Thus while Stowell’s (1981) analysis of adjacency effects on Case-assignment are suspect if Case is not merely a PF phenomenon (cf. also McCloskey’s 1991 analysis of Case-assignment under government from I0 in Irish), his analysis of the conditions on that-deletion continue to play a role in current theorizing (Fox 2000a), since it seems that this deletion, like many others, occurs on the PF side of the derivation.18 In sum, the facts of V-to-I movement are compatible with the present analysis under two different and equally viable minor redefinitions of PF-headgovernment. The first builds upon a posited difference between V-to-I and I-toC movement related to the lexical nature of the base position V, while the second capitalizes on the fact that subjects in English appear between I and C and disrupt PF-adjacency. In either case, we have a workable adaptation of the above analysis of the fact that I0 moving into C0 causes a disruption of PF-head-government in the manner we have seen, while verbs moving into I0 have the ability to satisfy the ECPPF via, or in spite of, their traces.
Conclusion The analysis presented here rests on three elements: first, that wh-movement of the comparative operator proceeds via adjunction to the VP sister of I0 ; second, that the resulting intermediate trace is subject to an ECP operating at the PF interface; and third, that I-to-C movement destroys the possibility that this ECP can be satisfied by head-government by I0 . These three elements have the result that in comparatives, where the extracted operator is nonovert, structures in which Ito-C movement has applied will always be accompanied by VP-ellipsis. The result is an internally coherent analysis of the surprising and subtle generalization that opened this paper, an analysis which furthermore is compatible with all other
Jason Merchant
instances of subject-auxiliary inversion in English, and which does not rely on construction-specific statements or restrictions. The theoretical import of the present analysis is not merely that there must be a licensing condition on traces that applies at PF, and that this condition must have the disjunctive character of Chomsky’s (1981, 1986a) ECP (not the conjunctive character of Rizzi 1990, for example), but simpliciter that there must be such a licensing condition at all. This is a conclusion some recent work has attempted to avoid, instead building the traditional representational well-formedness conditions such as the ECP into the well-formedness conditions on the movement operation itself. If the analysis presented here is correct however, such efforts cannot suffice to account for the full range of data a static ECPPF can: presumably no derivational version of the ECP can be devised that would have the ability to look ahead to the output of PF deletion, and license the intermediate traces in question just in case such a later PF operation were to apply. Since it is exactly VP deletion that rescues otherwise ill-formed structures, and since this deletion happens in the mapping from Spell-Out to PF, no derivational constraint applying (stepwise or otherwise) before Spell-Out will be able to take this deletion into account. A static well-formedness condition such as the ECPPF pursued here, on the other hand, will apply only after all deletion operations have, at the PF interface itself, and will be sensitive to exactly this kind of information, with the results we have seen. Indeed, if Chomsky (1995) and others are correct in analyzing I-to-C movement itself as a kind of movement that takes place on the PF side of the derivation (though see Zwart 2001 for important qualifications), then we have no recourse but to look to conditions on that side of the derivation to account for the generalization established here, since it is only in structures in which SAI has occurred that such VP-ellipsis is required. Putting these considerations together, we find all the more reason to posit well-formedness constraints such as the ECPPF that operate exclusively at the PF interface.
Notes * I am grateful to Chris Kennedy and Gregory Ward for discussions that stimulated this work and especially for bringing some of the initial data to my attention, though they should not be held guilty in any way for my interpretation of those data or for the present analysis. Thanks also to Anastasia Giannakidou, Jack Hoeksema, Sylvain Neuvel, Eric Potsdam, Chris Potts, Kerstin Schwabe, Susanne Winkler, and Jan-Wouter Zwart for comments and to Jack Hoeksema for making his corpus available to me for searching. . Potts (2000) gives related data from as-parentheticals.
Subject-auxiliary inversion in comparatives and PF output constraints . What triggers this movement, and what determines the circumstances under which it can occur, are not germane to the issue that interests me here; see Pollard and Sag (1994: Ch. 1), Pesetsky and Torrego (2001) for two recent approaches and references. . The notion of PF-active bears some similarity to Aoun et al.’s (1987) notion of ‘PF-visible’, though their definition picks out a slightly different class of elements, those that either are pronounced or bear certain kinds of indices (see Aoun et al. (1987: 539, 546)). Cf. also Chung’s (1991: 120) definition of ‘strong’ head governors (those that either are pronounced or are an element of {V, A, Infl}). . The notions of PF-visibility and PF-activeness clearly have much in common; whether they could be collapsed entirely, at least for case of heads, depends on details of the analysis of Affix-hopping and complementizer ‘deletion’ phenomena that go beyond the bounds of the present paper. . Similarly, presumably, when I0 has no pronounced content, as in (4a), where no dosupport is necessary; exactly how lexical insertion spells out the information in I0 on the verb in such cases is a topic of much debate (see e.g. Bobaljik 1995). The present cases are incompatible with literal I-to-V lowering analyses – what is clear is that in these cases I0 must not move, so it is the features of I0 , though not necessarily pronounced in I0 , that are active at the PF interface, and operate in the desired manner for purposes of the ECPPF . . This fact shows that simply stating a type-hierarchy co-occurrence restriction like [{SAI} ∩ {comparative wh-movement} ∩ {non-VP-ellipsis} = Ø], as considered at the beginning of this paper for Construction Grammar or some versions of HPSG, is too facile: somehow, the VP-ellipsis must also be restricted to occurring in a certain position. I am unsure whether such additional requirements pose a difficulty for construction-sensitive theories as currently formulated. . Whether every VP in stacked VP structures must be targeted by wh-movement cannot be decided on the basis of the SAI facts alone: putative intermediate traces adjoined to lower VPs would satisfy the ECPPF in any case. See Section 3.3 below for more discussion. . Likewise for the trace of the A-moved subject, if the ECPPF is generalized to A-traces as well, a possible extension outside the scope of this paper; because the subject is overt, it will satisfy clause (ii) of the ECPPF . In general, I do not consider subjects here for this reason; wh-moved subjects of comparatives (such as More people can sing than can play the piano) will not be relevant in any case, since in these cases, no SAI is found (see Aoun et al. (1987: 567–568); Rizzi (1990: 40–41); and Grimshaw (1997) for various explanations of this fact). . As E. Potsdam reminds me, for many speakers, VP-ellipsis is obligatory with so. For such speakers, we must assume that so is like as in selecting a CP whose specifier is filled by an empty operator, presumably an extracted VP. For speakers who share the judgments of Chomsky, Lasnik, and Quirk et al., so is presumably in specCP. Another possibility, suggested to me by S. Winkler, is that the construction with so simply doesn’t involve ellipsis at all, so being itself the (moved) predicate, as in López (1995), and hence irrelevant here. . This is supported as well by that fact that as can be used as a complementizer in some dialects, equivalent to that in standard English (occurring in selected complement clauses
Jason Merchant
and in headed relative clauses, for example). The data in (i) and (ii) illustrate these uses of as respectively, taken from the lower class British speech of the characters of Dorothy Sayers: (i) a. b.
She was quite ready to believe as she done wrong. [D. Sayers, Unnatural Death, 1927, p. 109] I don’t know as I ought to tell you. [loc. cit.]
(ii) This is Mr. Murbles, as put in that unfortunit advertisement, as I truly believes was the beginnin’ of it all. [op. cit. p. 104] . We are of course left with the intriguing residual question of why the observed difference should exist between the ability of overt (PF-visible) and empty (not PF-visible) operators to license intermediate traces. Clearly this fact should find its place in a general theory of how these kinds of operators differ, presumably relating to more general considerations of recoverability and identification; see in particular Grosu (1994) for extensive discussion. . A parallel to this situation might be found with pied-piping of prepositions and the use of the archaic form whom: many speakers reject examples with pied-piping of who, since the latter, bare, form belongs to a more colloquial register that prohibits pied-piping in favor of preposition stranding; cf. ®*With who did he speak? (where ®* indicates a register clash) vs. the prescriptively correct With whom did he speak?. See Lasnik and Sobin (2000) for discussion. . Levin also observes that the level of acceptability of examples of pseudogapping deteriorates with unlike subjects (as with SAI); this fact alone may account for some of the unacceptability of the examples in this section. I am operating on the assumption that this fact itself is in need of an explanation, and in any case, Levin’s observations are more of a general nature, and not meant as an absolute measure – she herself notes several examples where these conditions are not met, but pseudogapping remains possible. The pseudogapping examples in this section use unlike subjects because SAI in comparatives prohibits like subjects in any case: Abby1 is taller now than was {her father / *she1 } last year vs. Abby1 is taller now than {her father / she1 } was last year (see also Potts (2002) for discussion with respect to as as well; Potts points out that inversion creates an anti-pronominal context for unstressed pronouns, following the observation made in Quirk et al. (1985: 1382). . Such a conclusion is needed independently to allow wh-extraction in the presence of VP-adjuncts in any case (or VP-internal subjects, for that matter). . I will have to leave it open how these facts could be reconciled with the account of subdeletion proposed in Kennedy (2002). . In fact, Aoun et al. are somewhat unclear on whether head movement leaves an indexed trace: on p. 539 they assume that ‘rules moving . . . nonphrasal categories do not leave . . . indices’, while on p. 546, they claim that ‘the trace of a moved verb . . . is indexed’, in order to account for V2 effects in Dutch; they further point out that the literature on the question is equivocal, citing Torrego (1984) and Bouchard (1985) for conflicting claims regarding the ability of moved verbs to count as governors for the purposes of the ECP. . For purposes of adjacency, adjoined elements such as adverbs do not count, either for the reasons posited by Bobaljik (1995), or, at least for the cases in (48), simply because the option of adjoining the adverbs below the wh-trace has been taken. In this regard, it is
Subject-auxiliary inversion in comparatives and PF output constraints
also interesting to note that negation does not count, as (48c) above indicates (while in the computation relevant for do-support, negation destroys adjacency; I conclude from this fact that ‘affix-hopping’ is not dependent on PF-head-government as defined here). . A more substantive objection (besides the general problems inherent in Bobaljik’s stipulation, also required here, that adjuncts do not count for purposes of adjacency) might be raised if one wished to extend the analysis directly to languages other than English. In particular, a question arises concerning the application of the analysis to languages that have generalized V-to-I raising as well as null operators, such as Irish, Greek, French, Spanish, Hebrew, Arabic, colloquial Russian, and many others; in these languages, the facts are much the same as for their more limited English counterparts: V-to-I raising does not interfere with wh-movement of null operator complements of V. For example, the Irish relative clause in (i) shows both types of movements (McCloskey 1979: 6 (5)): (i) an scríbhneoir [CP Op1 aL [IP mholannV [VP t1 [VP na mic léinn tV t1 ]]]]. the writer Ctrace praise the students ‘the writer whom the students praise’ In such a case, the adjacency requirement as formulated in the text will not account for the satisfaction of the ECPPF , since the subject na mic léinn intervenes between the raised verb mholann and t1 . We can either take this fact to indicate that some cross-linguistic variability must be countenanced in this domain (cf. Chung 1991’s Chamorro-particular definition of ‘strong’ governors), or simply that the approach using the redefinition of PF-active in (51) is to be preferred to the PF-adjacency approach under consideration. Since the goal of this paper is restricted to analyzing the English facts, I will leave the choice between these options open. (Note finally that the I-to-C movement in V2 languages such as Dutch and German won’t be relevant, since V2 and wh-movement only co-occur in matrix clauses with overt operators, given Reis (2000)’s convincing re-analysis of the purported cases of embedded V2 as a special kind of parenthetical.)
Antecedent-containment and ellipsis Chris Wilder This paper examines a class of Antecedent-Contained Deletions involving ‘wide scope’ VP-ellipsis (e.g. John said that more trees had died than Mary did), in which the relative or comparative clause containing the elided VP is itself contained in a finite complement clause CP* inside the VP that antecedes the ellipsis. Under the Quantifier Raising account of ACD resolution, such examples involve ‘long-distance’ covert raising of the constituent containing the ellipsis (more trees than Mary did) out of CP* to a position higher than the antecedent VP (said that . . . had died). The question is raised as to why wide scope ACDs are possible only when the relative/comparative clause is extraposed (right-peripheral) within the complement CP* in surface order. It is suggested that to accommodate this fact, the account of ACDs needs to refer to a condition prohibiting ‘PF-containment’ of the ellipsis site by its antecedent, in addition to the condition prohibiting antecedent-containment at LF. It is further argued that examples apparently involving a VP-ellipsis site properly contained in its antecedent string in surface order (in violation of the ‘PF-containment’ condition), have been misanalysed – such examples result from the interaction of two independent ellipsis rules (‘pseudogapping’ and ‘backward deletion’) yielding discrete adjacent ellipsis sites, neither of which is antecedent-contained.
.
VP ellipsis and antecedent-containment
The primary theoretical interest of the English Antecedent-Contained Deletion (ACD) phenomenon lies in the evidence it provides for the assumption of LF movement (Quantifier Raising, cf. May 1985; Kennedy 1997; and others). This paper takes the QR account of ACD resolution as given, and examines a class of examples involving ‘wide scope’ VP-ellipsis, as in John said that more trees had died than Mary did, that has largely been overlooked. Such cases challenge the widespread view that QR is finite-clause bounded (see Wilder 1997a). They also provide evidence that the account of ACDs needs to refer to a condition prohibiting ‘PF-containment’ (of the ellipsis site by its antecedent), in addition to the usual
Chris Wilder
condition prohibiting ‘LF-containment’ (similar to a claim advanced, for different reasons, in Baltin 1987). It is the latter issue that forms the focus of this paper. ACDs form a subcase of the VP ellipsis phenomenon illustrated in (1)–(3): (1) a. John [VP2 met Mary], and Bill did [VP1 e ], too. b. Although John [VP2 met Mary], Bill didn’t [VP1 e ]. c. Although John did [VP1 e ], Bill didn’t [VP2 meet Mary]. (2) John [VP2 met everyone that Mary did [VP1 e ]]. (3) John [VP2 said that Mary did [VP1 e ]]. A VP ellipsis site (VPE), such as VP1 in these examples, is ‘anaphoric’, depending on an antecedent VP for its interpretation. Its antecedent can be in the same sentence (arbitrarily distant), or a VP in the external discourse (Sag 1976a; Williams 1977a). Embedded in a suitable discourse, all of (1)–(3) are well-formed, with VP1 taking a discourse antecedent. In (1), VP1 may take a VP in the same structure (i.e. VP2) as its antecedent, with no issue of antecedent containment arising. The ACD configuration is represented by (2)–(3), where VP1 is contained within VP2, the issue being whether VP2 can be understood as the antecedent of VP1. The puzzle is that while (3) is ill-formed (such a reading is simply impossible), (2) is not. Various accounts of this contrast between (2) and (3) (May 1985; Baltin 1987; Hornstein 1994; and others) assume that (4) holds at some level: (4) A VPE may not be contained in its antecedent. While (3) violates (4), the fact that (2) is well-formed (and has the interpretation it has) is accounted for by assuming that the example has (at the relevant level) a different syntactic representation, which avoids violating (4). Assumption (4) is generally assumed to hold at LF (May 1985, Hornstein 1994, etc.). One exception is Baltin (1987), who claims that (4) holds at S-structure. In the following, I make the claim (somewhat related to Baltin’s) that reference must be made to two independent conditions, i.e. (4), holding at LF, and (5), holding at PF: (5) A VPE may not be contained in its antecedent string at PF. While a VPE may in surface order either follow or precede the string of terminals corresponding to its LF-antecedent (cf. (1b) vs. (1c)), (5) states that an ellipsis site E may not be properly contained within that string: i.e. for any two terminals (α, β) belonging to the string spelling out the antecedent of E, if α precedes E, then E must not precede β. This condition is independent of (4). (5) is motivated by some new evidence showing that the distribution of VPE is governed by generalizations over surface orders that cannot be reduced to (4) holding at LF (Sections 3–5); and supported by a reinterpretation of some apparent counterexamples (Section 6).
Antecedent-containment and ellipsis
Assumptions differ on how VPEs and their interpretation are analyzed. In the ‘LF-copying’ proposal (Williams 1977a; May 1985; Hornstein 1994), a VPE is generated as a VP node dominating a phonologically null terminal, [VP e ]; in the covert syntax (following S-structure and prior to LF), the content of the antecedent VP is ‘copied in’, i.e. a copy of the antecedent VP is substituted for [VP e]. In the ‘deletion’ proposal (assumed e.g. in Chomsky 1995), the gap arises through phonological deletion, i.e. material generated as a normal VP gets deleted, in the PF-wing of the grammar, under identity with material belonging to the antecedent (to ensure the correspondence between deletion in PF and interpretive dependence at LF, material so deleted has to have been marked in some way in the syntax already). The choice between the two approaches is essentially orthogonal to the issues surrounding ACDs. While I assume a PF-deletion approach to be correct, nothing hangs on this; in reviewing previous accounts, discussion is framed in terms of the LF-copying approach where appropriate.1 Two further assumptions underlying the discussion of ACDs are: (6) a. The target of VPE is a single constituent. b. That constituent is VP. One problem that threatens the whole discussion is the possibility that one or both of these assumptions is false. Consider how (6b) could be false. If the deletion were only to target the verb (V1 in (7)), then the elided element would not be contained in its antecedent (which would be V2). The trace of the relative operator would then be a ‘remnant’ not contained within the deletion site. Deletion of a verb alone, leaving its object as remnant, is independently attested in the pseudogapping construction (8). (In these examples and others below, ‘deleted’ words appear in italics, their antecedents in bold): (7) John [V2 met] everyone [ Op that Mary did [V1 meet] t]. (8) a. ?John met Bill and Mary did, Paul. b. John [V2 met] Bill and Mary did [V1 meet ] Paul. Hence, (2)/(7) are potentially irrelevant to the issue of antecedent-containment. Furthermore, the pseudogapping construction shows that deletion sites can be discontinuous (in surface order at least), cf. (9): (9) a. ?John sent the book to Bill and Mary did, the papers. b. John [V2 sent] the book [PP2 to Bill] and Mary did [V1 send] the papers [PP2 to Bill]. Unless it can be shown that such discontinuous deletions form a constituent (at some level), the possibility must be countenanced that apparent VP-deletions are
Chris Wilder
actually composed of adjacent deletions of smaller constituents. So (1a) (=10a) may involve adjacent V-deletion and NP-deletion sites, as in (10b): (10) a. John met Mary, and Bill did, too. b. John [V2 met ] [NP1 Mary ] and Bill did [V1 meet ] [NP1 Mary ], too. Hence, more complex examples of apparent ACDs that avoid the objection posed by (8) may turn out to consist of several deletion dependencies, none antecedentcontained. Instead of involving an ACD as in (11b), (11a) may be analysed, by analogy with (10b), as involving two adjacent, independent deletions, neither antecedent-contained (11c): (11) a. John met Mary everywhere that Bill did. b. John [VP2 met Mary everywhere that Bill did [VP1 meet Mary t]]. c. John [V2 met ] [NP1 Mary ] everywhere that Bill did [V1 meet ] [NP1 Mary ] t. This issue is taken up in Section 6.1. Section 2 reviews some previous accounts of the ACD construction, highlighting crucial data which these sought to account for. In Section 3, these accounts are confronted with new data which motivate the proposal to be defended here.
. Three approaches to ACD resolution . QR (May 1985) Consider (3), repeated as (12a), under the LF-copy theory, assuming the structure given as input to the copy-operation. There is a syntactic reason for the fact that VP1 cannot take VP2 as its antecedent. The output of copying results in a structure (12b) which contains a second instance of VP1, requiring to be replaced by a copy of VP2. The derivation would thus slip into infinite regress, a finite well-formed LF cannot be derived:2 (12) a. John [VP2 said that Mary did [VP1 e ]]. b. John [VP2 said that Mary did [VP2 say that Mary did [VP1 e]]]. c. etc. In (13a) (=2), copying VP2 into VP1 leads to the same problem of infinite regress: (13) a. John [VP2 met everyone that Mary did [VP1 e]]. b. John [VP2 met everyone that Mary did [VP2 meet everyone that Mary did [VP1 e]]]. c. etc.
Antecedent-containment and ellipsis
However, VP2 can antecede VP1 in this case. May (1985) concludes that the structure given is not the correct one. Since antecedent-containment obtains at S-structure, “S-structure cannot be the level for the copying operation”. Instead, copying applies in the LF-component, after a covert operation which alters the Sstructure in such a way that the elided VP is no longer contained in its antecedent. This operation is Quantifier Raising, which raises the object NP (including the relative clause containing the VPE) out of the matrix VP, adjoining it to IP (14b). The VP that is subsequently copied consists of the verb and the trace left by QR; since it no longer contains a VPE, copying avoids regress (14c). (14) a. SS John met everyone that Mary did [VP1 e]. b. QR [IP [NP everyone that Mary did [VP1 e]] [IP John met t]]. c. Copy [IP [NP everyone [CP Op that Mary did [VP1 meet t]]] [IP John past [VP2 meet t]]]. This analysis yields an LF-representation (14c) which satisfies other wellformedness conditions (e.g. the relative operator c-commands a trace/variable which it can bind); and which captures relevant aspects of the meaning of the sentence (15).3 (15) ∀x ( ( person(x) ∧ Mary met x ) → John met x ) This analysis of ACDs supplies a major piece of evidence for the existence of a Quantifier Raising operation in covert syntax. One argument that May used to support the QR approach was that only quantified expressions license ACDs in relative clauses modifying them. Thus (16), with a name modified by a nonrestrictive relative, is deviant under the antecedent-contained reading; this is predicted, since a name, not being a quantified expression, does not undergo QR (May 1985: 12): (16) *?John accused Mary, who Peter did [VP e]. The object NP remains in VP at LF, hence the VPE cannot take the VP headed by accused as its antecedent without inducing infinite regress. This argument is put in doubt by the fact that (16) improves significantly if too is added, cf. (17) (Wyngaerd & Zwart 1991). The pattern can be reproduced for VPE in coordinate contexts, cf. (18), suggesting that the cause of the deviance of (17) was incorrectly diagnosed by May, and that it is due to some completely independent factor (cf. also Hornstein 1994): (17) ?John accused Mary, who Peter did [VP e], too. (18) a. *?John accused Mary, and Peter did. b. John accused Mary, and Peter did, too.
Chris Wilder
The well-formedness of (17) does not, however, necessarily threaten the QR approach to ACDs in restrictive relatives. Lasnik (1995b) notes some differences between ACDs in nonrestrictive relatives and those in restrictive relatives, including the contrast (19): (19) a. John stood near everyone Mary did [VP e]. b. *John stood near Sue, who Mary did [VP e] (too). Lasnik suggests that non-restrictive cases involve a pseudogapping derivation of the type discussed in Section 1. Then, (17) is to be analyzed as in (20), and (19b) as in (21): (20) John [V ◦ accused] Mary, whoj Peter did [V ◦ accuse]t j too. (21) *John [V ◦ stood] [P◦ near] Sue, whoj Peter did [V ◦ stand][P◦ near]t j too. The deviance of (21) can then be related to the impossibility of deletion of the V+P combination (stand near) in the coordinate structure (22a):4 (22) a. *John stood near Sue and Peter did, Mary. b. John stood near Sue and Peter did, near Mary. If the pseudogapping approach generalizes to all non-restrictive examples, then these are not ACDs at all, and the well-formedness of (17) is irrelevant. At the same time, the well-formedness of (19a), contrasting with (19b), indicates that a pseudogapping analysis cannot be generalized to all ACDs in restrictive relatives. The contrast can be handled by assuming that ellipsis of the whole VP including the stranded preposition (near) is possible in (19a) (QR can apply to eliminate the antecedent containment configuration prior to LF) but not in (19b) (QR cannot apply).
. Extraposition (Baltin 1987) Baltin (1987) constructs an argument against the QR (LF-movement) account of the resolution of ACDs from the observation that the account overgenerates: readings are predicted to be possible that do not in fact exist. One such example is (23a). The VPE in the relative clause may only take the embedded VP as its antecedent (23b), although the QR account predicts that the higher VP headed by thought should be a possible antecedent as well (i.e. (23a) should be ambiguous between (23b) and (23c)): (23) a. Who thought that Fred read how many of the books that Bill did? b. = Who thought that Fred read how many of the books that Bill read?
Antecedent-containment and ellipsis
c. = Who thought that Fred read how many of the books that Bill thought that Fred read ? As well as applying to quantificational expressions such as everyone, QR is assumed to apply in multiple interrogatives to wh-phrases located in situ at S-Structure, such as phrase headed by how many in (23). This wh-phrase raises covertly to a position from which it takes scope as a wh-operator (the specifier of the root CP), where it adjoins to the wh-phrase who in the standard analysis. It is crucial to the QR account of ACD-resolution that the whole quantifier phrase, including the relative clause containing the VPE, is (or can be) raised by QR. If the relative clause is piedpiped, the LF of (23a) will look like (24a), with the VPE contained in the highest SPEC (24b). Since neither VP1 nor VP2 contains the VPE in (24), both should be available as possible antecedents. But factually, only VP2 can antecede it (cf. (23)). (24) a. [ SPEC C◦ [IP t j [VP1 thought [ that Fred [VP2 read t k ]]]]] b. SPEC = [ whoj [ how many of the books that Bill did [VP e]]k ] Let us call a reading where a VPE takes only the ‘first VP up’ as its antecedent (e.g. (23b)), a narrow reading; and one where a higher VP is intended as antecedent (e.g. (23c)) a wide reading. There is no independent reason why the VPE in (23) should not take VP1 instead of VP2 as its antecedent. In (25a), where the relevant wh-phrase has raised overtly, both narrow and wide scope readings are possible: (25) a. How many of the books that Bill did, did Mary think that Fred read ? b. = How many of the books that Bill read, did Mary think that Fred read ? c. = How many of the books that Bill thought that Fred read, did Mary think that Fred read ? Thus, there is a mismatch between the scope taken by a scope-bearing element E and the scope of a VPE contained within the phrase headed by E. While QR at LF may assign the quantifier wide scope, the VPE may only get a narrow scope reading. Baltin calls this restriction on ACDs the Boundedness restriction. In his analysis of ACDs, Baltin propsoses a surface filter (26): (26) There is no well-formed sentence in which a VPE is contained in its antecedent in the S-structure representation. Given (26), QR in the LF-component does not suffice to resolve antecedentcontainment. Even if QR results in a VPE not being antecedent-contained at LF, this would happen ‘too late’ in the derivation to save the sentence, which will be marked ungrammatical (violation of (26)) at S-structure.5 Baltin proposes that the operation is responsible for resolving antecedentcontainment is not QR but Relative Clause Extraposition (RCE). RCE is assumed to apply in the overt syntax, extracting a relative clause from the NP it modifies
Chris Wilder
and right-adjoining it to the VP (or some higher projection). Supposing that RCE has applied – string-vacuously – to (2), the S-structure representation looks like (27a), and (26) is satisfied. QR may subsequently apply to yield (27b), on the basis of which VP1 may take VP2 as its antecedent without incurring regress (27c): (27) a. John [VP3 [VP2 met everyone] [that Mary did [VP1 e]] b. [everyonej [John [VP3 [VP2 met t j ][that Mary did [VP1 e]] c. [everyonej [John [VP3 [VP2 met t j ] [that Mary did [VP1 meet t j ]] The hypothesis that RCE resolves containment is not only compatible with (26), but also provides a means for accounting for the Boundedness Restriction on ACDresolution. Extraposition is subject to a limitation known as the ‘Right Roof Constraint’ which prohibits it from moving a constituent out of the minimal sentence (CP) containing it. The Right Roof Constraint is usually illustrated by pointing to word order facts such as (28): (28) a. Bill will say that John kissed everyone that he met tomorrow. b. *Bill will say that John kissed everyone tomorrow that he met. The relative clause, which modifies the object of the embedded clause, may not appear to the right of the adverbial modifying the matrix VP, a linear position corresponding to a hierarchical position higher than the matrix adverbial, hence outside the embedded clause. Given this interpretation of RCE and the Right Roof constraint, the impossibility of wide scope interpretation in (23) can be explained. RCE can only raise the relative clause prior to S-structure to a position within the clause containing the object how many of the books, so that the VPE inside it, while no longer contained in VP2 (headed by read), fails to escape the higher VP3 headed by thought, as indicated in (29). (29) Who [VP3 thought [CP that Fred [VP2 read how many of the books [that Bill did [VP1 e]] CP ] VP3 ]
VP2 ]
Assuming that no post-S-structure operation affects the relative hierarchy of the VPs in (29), there is exactly one VP which does not contain VP1 at LF and which may function as its antecedent, namely VP2. (The phrase how many of the books may undergo covert movement, which ensures that at LF, VP1 contains a trace that can be bound as a variable by that phrase). In this way, Baltin is able to derive the Boundedness Restriction on ACDs.
Antecedent-containment and ellipsis
. Larson and May (1990) In their reply to Baltin (1987), Larson & May (1990) give counterarguments to the assumption (26), and to the proposal that all relative clauses with antecedentcontained VPEs undergo (possibly string-vacuous) extraposition. Direct counterevidence to (26) comes in the form of examples involving a VPE both preceded and followed by terminals belonging to its antecedent at S-structure. In (30a), the VPE is contained in a free relative that functions as the goal object in the double object construction. Its antecedent includes the preceding verb and the following NP (the theme argument of gave).6 ACDs involving a VPE contained within the accusative subject in ECM constructions are also held to be acceptable (in disagreement with Baltin – cf. also Hornstein 1994: 459): (30) a. John gave [ whoever he could _ ] two dollars. b. ?John believes [ everyone I do _ ] to be a genius. Larson and May contend that QR alone makes the right distinctions among possible and impossible ACDs, i.e. that ACDs are found in exactly those cases in which QR raises a constituent containing the VPE out of its antecedent VP before LF. The accusative subject in the infinitive complement (30b), for example, contrasts with the nominative subject of its finite counterpart, in that the latter may not host an ACD taking the VP headed by believe as its antecedent: (31) *John believes [ everyone I do _ ] is a genius. The claim that an accusative subject can be QR-ed to a position above the higher verb, while a nominative subject is unable to leave its own clause correlates with the possibilities for scopal interactions between quantifiers. The accusative subject in (32a) can take scope over the subject of the higher clause, but the inverse scope reading is not possible in (32b): (32) a. Some student believes every teacher to be intelligent. b. Some student believes that every teacher is intelligent.
∀∃ *∀∃
. Covert A-movement (Hornstein 1994) In the context of the Minimalist framework, Hornstein (1994) proposes an alternative to the QR account of ACD resolution. Following Larson and May (1990), Hornstein rejects (26), assuming that the sole filter on antecedent containment holds of LF-representations, and concluding that ACDs provide evidence for phrasal movement in the covert syntax. Instead of QR (movement to an A’position), Hornstein argues that it is A-movement of objects at LF that resolves antecedent-containment in ACDs.
Chris Wilder
In Chomsky’s (1993) proposal, direct objects in English (in VP at S-structure/Spell-Out) raise at LF to the specifier of a functional agreement projection (AGRo) to check Case and agreement features. Hornstein seeks to establish that covert A-movement provides the basis for an account of ACD resolution without recourse to QR. The simplified derivation for (2), assuming an LF-copy approach to VPE, is given in (33). The direct object containing the VPE is in VP at Spell-Out (33a). Covert A-movement places it in the specifier of AGRo (33b). LF-copying may now substitute the main VP for the VPE with no regress arising.7 (33) a. John [VP met everyone that Mary did [VP e]]. b. [IP John past[AGRoP [DP everyone that Mary did [VP e]] AGRo [VP t meet t]]]. c. [IP John past [AGRoP [DP everyone that Mary did [VP t meet t]] AGRo [VP t meet t]]]. Hornstein cites conceptual and empirical reasons for rejecting the QR-based solution to the ACD issue. Firstly, given the Minimalist hypothesis that all movement processes are driven by purely formal (morphological) requirements, it is not clear if QR even exists. Aside from wh-movement, there is no apparent morphological requirement to which QR may be the response. Secondly, even if QR is adopted as the means for handling scope facts, this does not suffice to ensure that QR also provides for ACD-resolution. The latter depends on the further assumption that relative clauses are pied-piped under QR at LF; while relative scope facts could be accounted for even if QR affects the quantifier alone without pied-piping further material. That covert NP-movement, on the other hand, must pied-pipe the whole DP, Hornstein takes as given.8 Beyond this, Hornstein cites worries about the coextension of QR and ACD resolution (ACDs in appositive relatives; Baltin’s Boundedness Restriction). The A-movement solution is claimed to capture all facts discussed so far, hence to be superior to previous accounts. Correct predictions are generated concering cases where QR targets and ACDs diverge. While names do not undergo QR, they do undergo LF-A-movement, like any other DP; hence, the occurrence of ACDs in appositive relatives is expected. A-movement also differs from QR in that it is clause-bound; this fact can be used to explain the Boundedness Restriction observed by Baltin (cf. (23)), without recourse to extraposition. Cases like (30), counterexamples to Baltin’s extraposition account, also fall out correctly; VPEs contained within their antecedents at S-structure are predicted to be permitted precisely where covert A-movement undoes the containment configuration.9 The approach also correctly distinguishes accusative subjects from nominative subjects (30b) vs. (32).
Antecedent-containment and ellipsis
. Kennedy (1997a) The A-movement proposal faces a number of serious problems problems that arise from it being too restrictive in its choice of target, discussed in detail in Kennedy’s (1997a) reply to Hornstein. An ACD configuration may arise not only when a VPE is contained in a relative modifier of a direct object, but also when contained in a PP-complement or an adverbial, categories which are not normally considered to be targets for A-movement: (34) a. John [talked [to everyone who Mary did _ ]]. b. John [recited his lines [in the same way that Bill did _ ]]. Hornstein suggests that the former case be handled in terms of A-movement of the DP out of PP into a higher agreement projection. In other words, such examples are used to motivate an extension of the LF-A-movement hypothesis to DPs in selected PPs. The A-movement analysis predicts that there are no genuine ACDs contained in adjuncts: A-movement resolves antecedent-containment, and no DP can be Amovement out of an adjunct phrase. Hornstein claims that examples such as (34b) involving adverbials at the right edge of VP do not constitute ACDs at all, even at Sstructure. The adverbial in (34b) is assumed to be right-adjoined to (hence higher than) the antecedent VP; and the trace of the relatvized adjunct outside the VPE. However, as Kennedy points out, there are cases of ‘wide scope’ adjunct ACDs in which the adjunct is clearly contained in the antecedent VP. In (35a), the antecedent VP (headed by likes) contains the locative adjunct PP which modifies the embedded VP (perform). While ungenerable under the A-movement analysis, such cases are relatively unproblematic for the QR analysis. Assuming that QR is able to move the quantified NP out of the embedded infinitive (35b), the VPE can take the matrix VP as its antecedent (35c): (35) a. John likes to perform in the same clubs as Bill does _. b. [IP [the same clubs as Bill does _ ] [IP John likes to perform in t ]]. QR c. [IP [the same clubs as Bill does [VP1 like to perform in t ]] Copy John [VP2 likes to perform in t ]]. Another telling class of examples which Kennedy draws attention to involves NP-contained ACDs, as in (36): (36) Bill read [NP a report on [QNP every suspect [RC that John did _ ]]]. This type of example provides a double problem for the A-movement account. Firstly, while covert A-movement of the direct object of read (37a) resolves the antecedent-containment configuration, it generates the ‘wrong LF’. Copying the
Chris Wilder
matrix VP into the VPE yields an LF (37b) that would give the interpretation (37c). Clearly, (36) does not have this (pragmatically absurd) reading: (37) a. Bill [ [a report on every suspect that John did _ ] AGRo [read t OB ]. b. Bill [a report on every suspect that John did [ read t OB ]] AGRo [read t OB ]. # c. Bill read a report on every suspect that John read. Secondly, the covert movement required to generate the ‘correct’ LF cannot be Amovement. The reading that (36) actually has is paraphrased in (38c), and can be derived by assuming that the quantified NP headed by every is raised out of the object NP, leaving the remainder of the higher NP (the direct object of read) inside the matrix VP (38a, b). (38) a.
[IP [every suspect that John did _ ] [IP Bill read [OB a report on t ]]]. QR b. [IP [every suspect that John did [VP1 read [ a report on t ]]] Copy Bill [VP2 read [ a report on t ]]]. c. Bill read a report on every suspect that John read a report on.
The antecedent required by the VPE in the relative clause is not a VP of the type [VP V t NP ] generated by covert A-movement, but a VP of the type [VP V [NP Det N .. t QNP ]]. While the correct antecedent cannot be provided by A-movement, it is reasonable to assume that covert QR (of a QNP out of a weak indefinite object NP) can do the job. Cases like (35a) and (36) clearly indicate that ACD-resolution does not shadow covert A-movement. Another class of examples illustrating the same point are wide scope ACDs involving QR out of finite complement clauses, to which we now turn.
. Wide scope ACDs . A new problem As noted in the previous sections, a VPE in a relative clause modifying the subject of a finite complement may not take a wide scope reading (39a) (cf. (32) above). However, as Lasnik (1995a) observes, citing Tiedeman (1995), wide scope VP-ellipsis is possible in a relative modifying the subject of a finite complement clause – if the relative is extraposed (39b):10 (39) a. *John believes [ everyone I do _ ] is a genius. b. ?John believes everyone is a genius [ that I do _ ].
Antecedent-containment and ellipsis
Similar pairs involving comparatives can also be constructed: (40) a. *John claimed that more people than you did _ were stupid. b. John claimed that more people were stupid than you did _. The wide scope interpretation of the VPE in (39b) and (40b) is quite clear: the matrix VP is reconstructed into the VPE, so that they mean the same as the sentences (41) and (42) (with or without extraposition), respectively:11 (41) a. John believes everyone is a genius that I believe is a genius. b. John believes everyone that I believe is a genius, is a genius. (42) a.
John claimed that more people were stupid than you claimed were stupid. b. John claimed that more people than you claimed were stupid, were stupid.
With VPE, extraposition of the relative or comparative clause has a dramatic effect on acceptability, which none of the analyses discussed in the previous section is in a position to explain. The problem posed by the contrast is different in each case.
.. The QR account For the QR account (May 1985 etc.), the question raised is: why should extraposition of a modifier affect the ability of the head NP to undergo QR? In order for the QR account to be able to explain the wide scope reading of the VPE in (39b) and (40b), it must be assumed that the nominative can undergo QR, pied-piping its modifier, to a position above the matrix VP.12 In order to account for the ungrammaticality of (39a) and (40a) under the wide scope reading, QR must be unable to raise the nominative to a position above the matrix VP. Either QR is prevented from applying, or an independent explanation must be found. Larson and May (1990: 108) explicitly claim that the reason for the ill-formedness of (39a) is that the subject of a finite complement clause may not undergo QR into the matrix sentence (“quantified subjects are clause-bound in finite sentences”). Suppose ECP is the constraining principle. The application of ECP should be blind to whether or not a raised QNP contains a relative clause. How extraposition of a modifier could neutralize the ECP in this case, is completely mysterious.13 .. The extraposition account For Baltin (1987), the problem is a different one. Given the S-structure constraint (26) of Section 2.2 above, it appears at first glance that the paradigm (39)–(40) actually supports Baltin’s approach. (39a) and (40a) clearly involve antecedent containment at S-structure, regardless of what happens at LF. It seems as if extraposition in (39b) and (40b) has the effect of removing the VPE from the contain-
Chris Wilder
ment configuration, exactly what Baltin claims. However, if extraposition obeys the Right Roof constraint, the relative clause in these examples can be adjoined no higher than the CP of the complement, where it will still be contained within the VPE-antecedent (the matrix VP), in violation of the S-structure contraint (26). In other words, the S-structure (43a), which obeys Right Roof, violates (26); while (43b) satisfies (26) but violates Right Roof: (43) a.
John [VP2 believes [CP that everyone is a genius [that I do [VP1 e]] CP ] VP2 ]. b. John [VP2 believes [CP that everyone is a genius CP ] VP2 ] [that I do [VP1 e]].
The Right Roof Constraint plays a crucial role in deriving the Boundedness restriction discussed in connection with example (23) above. (39b) and (40b) are clear counterexamples to the Boundedness Restriction. Baltin assumed that the Boundedness restriction holds without exception. If it were the case that the Right Roof constraint were not operative in these examples, then (39b) could be assigned the S-structure (43b). Thus, the idea may seem worth exploring, that these sentences counterexemplify the Boundedness restriction for the very reason that extraposition is not constrained by the Right Roof Constraint. However, it is doubtful whether the Right Roof effect is cancelled (i.e. the ‘roof ’ is extended) in these constructions: the word order test clearly indicates that the restriction is still in place – no matrix adverbial may separate the extraposed relative from the rest of the complement clause: (44) a. *John said that everyone is a genius yesterday that I did. b. *John claimed that more people were stupid when I asked him, than you did. Without taking any stand on the nature of the extraposition process, I proceed on the assumption that the extraposed clauses in (39b) and (40b) are located within the complement clause at S-structure, as in (45), so that the ellipsis sites are antecedent-contained at that level. (45) a. John [VP believes [CP that everyone is a genius that I do [VP1 e ]]]. b. John [VP claimed [CP that more people were stupid than I did [VP1 e ]]].
.. The A-movement account For the A-movement account, the problem posed by wide scope ACDs is as simple as it is unresolvable. Hornstein’s goal of establishing the ACD phenomenon as independent evidence for covert A-movement in English rests on showing that
Antecedent-containment and ellipsis
the resolution (and interpretation) of ACDs tracks covert A-movement (which is refuted by Kennedy’s evidence in any case, see Section 2.5). In addition to Baltin’s boundedness restriction, that account depends crucially on the accusativenominative contrast (46a) vs. (46b): (46) a. John believes everyone I do _ to be intelligent. b. *John believes everyone I do _ is intelligent. c. John believes everyone is intelligent that I do _. Adding (46c) (=39b) to the paradigm ruins the picture. (46c) should be as ungrammatical as (46b). It is hardly plausible to assume LF A-movement of nominatives into a higher clause; even less so, if that operation depends on the presence of a relative clause that is extraposed.14
.. Further properties of wide scope ACDs What is the extent of the problem? Wide scope ACDs are not freely constructible, one has to hunt them. As Hornstein notes (fn. 9), there are unclear restrictions on matrix predicates and tenses (see also Fiengo & May 1994: 255–256, fn.17): (47) a. b. c. d.
??John has heard
that everyone is a genius [ that I have _ ].
??John said that Bill was reporting everything [ Sam did _ ].
*John doubted that the yacht was as long as she did. *John insists that my yacht is flashier than you do.
This may help to explain why they have largely escaped notice so far (in addition, Baltin’s examples illustrating the Boundedness restriction (23) might also have served to distract attention from them). Nevertheless, there are enough good examples to establish that this is indeed a productive phenomenon which cannot be simply dismissed. The availability of a narrow scope reading can make a wide scope reading barely, if at all, accessible. For (48a), the reading (48b) obscures (48c), for example: (48) a. John thought that Mary bought more books than Bill did. b. John thought that Mary bought more books than Bill did buy. c. John thought that Mary bought more books than Bill did think that Mary bought. This ‘local antecedent’ effect can be controlled for. In (49), where the embedded subject in the antecedent does not match the subject of the VPE in animacy, but the matrix subject does, the wide reading is at least as accessible as the narrow reading: (49) a. John thought that the fire destroyed more books than Bill did. b. John thought that the fire destroyed more books than Bill did destroy.
Chris Wilder
c.
John thought that the fire destroyed more books than Bill did think that it destroyed.
Prosodic factors can be important. A wide scope reading can be facilitated by parallel accent (indicated by capitals) on the subject of the VPE and the subject of antecedent VP, with the subject of the complement clause in the antecedent deaccented. In fact, wide and narrow readings for the VPE can be disambiguated prosodically in this way (cf. also discussion of deaccenting and VPE in Tancredi 1992): (50) a.
JOHN thought that the fire destroyed more books than BILL did. (wide only) b. John thought that the FIRE destroyed more books than BILL did. (narrow only)
Prosodic manipulation cannot save the wide scope reading in Baltin’s example (51) (= (23) in Section 2.2), however: it can only exclude the narrow scope reading as well: (51) *WHO thought that Bill read how many of the books that MARY did? This strongly suggests that long distance association of wh-in situ (i.e. the association of how many with who in (51)), even if it involves covert movement, does not involve pied-piping of the relative clause (as Hornstein suggests). Two further properties of wide scope ACDs are relevant to the proposal to be made in Section 3.2. Firstly, they are only found in right-peripheral position (cf. Hornstein 1994: fn. 9):15 (52) a. John said that there were more people than Mary did _. b. John said that more people were there than Mary did _. c. *John said that more people than Mary did _were there. The function of the head of the relative or comparative containing the ACD (subject, object, predicate, etc.) does not seem to matter, as long as the VPE is in final position; i.e. if the wide scope reading is possible at all, then only if the VPE is in final position. Extraposition does not seem necessary (53a), (54a) – although it may operate string-vacuously, as in Baltin’s proposal – unless the relative or comparative is not in final position (53b), (54b): (53) a. John thinks that this will cause more harm than Mary does. b. *John thinks that this will cause more harm than Mary does to the proposal. c. John thinks that this will cause more harm to the proposal than Mary does.
Antecedent-containment and ellipsis
(54) a. John thinks the situation has got worse than Mary does. b. *John thinks the situation has got worse than Mary does through this. c. John thinks the situation has got worse through this than Mary does. Secondly, wide scope ACDs are found in relative clauses and comparative clauses, but they are not licensed in adjunct clauses (in contrast to relative clauses contained within adjuncts, cf. (35) above). A sentence like (55) is ambiguous with respect to which verbal projection the temporal adjunct modifies: (55) John said that Mary would arrive when Peter did. Adding the wide – narrow scope dimension for the VPE (and ignoring the possibility for an external antecedent), we expect the sentence to have four possible readings (56). In fact, it only has three. (56) a. [[ John said that Mary would arrive ] when Peter did arrive ]. b. [[ John said that Mary would arrive ] when Peter did say that Mary would arrive ]. c. John said that [[ Mary would arrive ] when Peter did arrive ]. d. *John said that [[ Mary would arrive ] when Peter did say that Mary would arrive ]. If the temporal clause modifies said, then the VPE may take either the embedded (56a) or the matrix VP (56b) as antecedent. If the temporal clause is understood as modifying would arrive, then the VPE may only take the embedded VP as antecedent (56c). The intended reading for (56d) can be paraphrased as (57): (57) John said at time t1, that Mary would arrive at time t2, t2 being such that Peter said (at time t3) that Mary would arrive at t2. This reading is utterly impossible for (55) – although it represents a possible interpretation of (58), the ‘undeleted’ counterpart to (55): (58) John said that Mary would arrive when Peter said that she would arrive.
. A new proposal The evidence from Section 3.1 speaks strongly against the proposal of Hornstein (1994): LF A-movement is too restrictive to be able to account for limitations on wide scope readings of ACDs. It is consistent with an A’-movement (i.e. QR) account, assuming that restrictions on QR are relaxed to permit LF-extraction of certain phrases from finite complement clauses. But that account is then too liberal to account for the distribution of well-formed and ill-formed ACDs – in particular, the contrasts involving extraposition. These in turn motivate a proposal in the
Chris Wilder
spirit of Baltin’s. There is not only an LF condition barring antecedent containment, which is voided by QR in certain cases; there is also an independent surface condition, which is voided by extraposition in certain cases. That is, in addition to the condition (59) (= (4) of Section 1 above) holding at LF, there is a separate principle (60) (= 5), holding at PF: (59) A VP-deletion site may not be contained in its antecedent constituent at LF. (60) A VP-deletion site may not be contained in its antecedent string at PF. In view of the considerations on the status of extraposition mentioned in Section 3.1.2, the surface condition cannot be stated in terms of hierarchical containment (as Baltin’s condition is). Extraposition only serves to ensure that the VPE is not contained within the string corresponding to its antecedent; it does not remove the VPE from the hierarchical constituent of its antecedent. Instead, the condition is formulated in terms of ‘string-containment’. What (60) excludes is any PF-string in which two terminals α, β, belonging to the antecedent VP, are linearized with respect to the VPE such that α precedes the ellipsis site and β follows the ellipsis site: (61) *α < E < β, where α, β are elements of the antecedent of E. If syntactic objects (including S-structure/Spell-Out trees) do not contain precedence information, then (60) cannot hold of S-structure. The locution “at PF” is however intended to be vague, referring to some stage within the derivation from syntactic component to the PF interface, in the sense discussed in Chomsky (1995). The earliest stage at which (60) can be assumed to hold is the level at which representations are linearized. The problem with assuming that (60) holds at the interface is of course that it refers to a phonetically empty element, by hypothesis not ‘visible’ at the interface.16 Each of (59) and (60) is necessary independently of the other. The impossibility of a wide scope reading for a VPE in an embedded adjunct clause provides evidence for the necessity of the ban on structural containment (59) – since the adjunct appears string finally, the ban on string containment (60) would be insufficient. The adjunct – relative clause asymmetry (a VPE in a final embedded relative clause can receive wide scope interpretation) supports the assumption of the QR-escape hatch (assuming that an extraposed relative clause is ‘reconstructed’, so that it can be pied-piped by LF-movement; cf. Wilder 1997a). The extraposition / intraposition asymmetry found with embedded relative clauses then provides the evidence for the existence of the additional surface filter (60). This filter must be formulated in terms of linear containment; hierarchical containment would be too strong (unless extraposition is rightward movement and the Right-Roof Constraint is lifted in the relevant cases).
Antecedent-containment and ellipsis
Assuming that adjunct clauses do not undergo QR, wide scope VP-deletion in a final embedded adjunct (56d) violates (59), but not (60). A wide scope VPE in an embedded intraposed relative (62) violates the surface filter (60), but not (59) (assuming QR). A wide scope in an embedded final relative (63) violates neither. (62) a. John believes everyone [ that I do _ ] is a genius. b. PF *John believes everyone that I do (believe (x) is a genius) is a genius c. LF [ everyone that I do (believe (x) is a genius) ] [John believes x is a genius ] (63) a. John believes everyone is a genius [ that I do _ ]. b. PF John believes everyone is a genius that I do (believe (everyone) is a genius) c. LF [ everyone that I do (believe (x) is a genius) ] [ John believes x is a genius ] Additional evidence for the analysis of (62) in terms of LF-convergence but PFcrash comes from typical native speaker reaction to such examples – they are held to be “interpretable, but the word order is wrong”. This reaction contrasts strongly with the reaction to embedded adjunct examples where a narrow scope reading is independently excluded: the wide scope reading is felt to be simply impossible (“the example is uninterpretable”). Their status is comparable to that of cases like (64), under the ACD reading: (64) *John said that Mary did. There is one issue not yet addressed: what to do with ‘genuine’ string contained ACDs like (65), which – apparently – violate (60)? (65) a. John gave whoever he could _ two dollars. b. ?John finds everyone I do _ easy to work with. c. ?John believes everyone I do _ to be a genius. These cases are examined in Section 6 below, where it is proposed that they do not involve antecedent-containment at all, hence do not pose any problem for (60).
. Hierarchical containment and QR In this section, two questions concerning the LF-side of the account are addressed: (i) what type of constituents undergo QR out of finite complement clauses (i.e. license wide scope readings for VPEs)? (ii) how far do such constituents move (what is the landing site of ‘long QR’)?
Chris Wilder
. What undergoes QR? Taking wide-scope VP-deletion as an indicator for QR, it turns out that QR can affect argument DPs of various types which have been claimed by other authors to be QR targets. Additionally, compared predicates must be assumed to be able to undergo QR. The VPE in each case is contained within a relative clause modifying the head noun, or, in the case of compared noun phrases and adjectival predicates, within the comparative clause selected by the degree word (e.g. the than-clause selected by more, the as-clause selected by equative as or by the same, etc.). The possibility for argument noun phrases that are modified by relative clauses to host wide scope VPEs also depends on the nature of the determiner. Wide scope VPEs occur most felicitously in relatives modifying DPs headed by strong determiners (every, each, most, etc – cf. (66)); definites, especially in combinations such as the same N, or the very N (67); and partitive DPs headed by weak determiners (68) (‘strong’ and ‘weak’ in the sense of Milsark (1977), Diesing (1992)). In each case, the extraposition-intraposition contrast is evident, even where the extraposed variant itself is only marginally acceptable: (66) a. b. c. d.
John thought that most people were there that Mary did. *John thought that most people that Mary did were there. John thought that each proposal should be accepted that Mary did. *John thought that each proposal that Mary did should be accepted.
(67) a. John said that the (very) same problem would arise that/as Mary did. b. *John said that the (very) same problem that/as Mary did would arise. (68) a. John said that many/none of the problems would arise that Mary did. b. *John said that many/none of the problems that Mary did would arise. All these DP types have been claimed by May (1985) and/or Diesing (1992) to undergo QR. Examples involving relatives whose DP is headed by a weak determiner that is not a partitive, are marginal, bordering on the unacceptable: (69) ???John said that a / one / many / few / no problem(s) would arise that Mary did. In her discussion of indefinites, Diesing (1992: Ch. 3) distinguishes a presuppositional and a non-presuppositional reading of indefinite DPs headed by weak determiners, claiming that presuppositional indefinites undergo QR, while nonpresuppositional indefinites do not. Partitive indefinites are invariably presuppositional. Using ACDs as a diagnostic, the contrast between the two readings comes out clearly in simple examples (Diesing 1992: 71):
Antecedent-containment and ellipsis
(70) a. ??I read few books that you did. b. I read few of the books that you did. Under certain circumstances, (70a) can be rendered acceptable – namely when the context facilitates a presuppositional reading of the DP (in which there a salient set of books over which the determiner ranges, as in the sole possible reading of (70b)). As Diesing notes, (70a) is “unquestionably ungrammatical in the case of the cardinal (nonpartitive) reading”.17 Plausibly, the difficulty with (69) can be attributed to the difficulty in accessing the presuppositional reading for the DP that would accompany the QR needed to resolve the VPE, just as in (70a). Additional support for this speculation can be derived from considering the weak determiner some. As Milsark (1977) has pointed out, the non-presuppositional (cardinal) reading of DPs headed by some typically involve an unstressed variant sm; while stressing the determiner triggers a presuppositional reading. A wide scope VPE associated with sm is impossible, its counterpart with stressed some is relatively good: (71) a. *John said that sm problems would arise that Mary did. b. ?John said that some problems would arise that Mary did. Already on the basis of simple examples like (72a), Diesing is led to suggest that comparatives undergo QR. Cases involving compared DPs seem to yield the best examples of wide scope ACDs (72b): (72) a. I read more books than you did. b. John said that more problems would arise than Mary did. Compared embedded predicates also license wide scope ACDs: (73) John said that his yacht was longer than Bill did. This indicates that it is the comparative morpheme, as a functional head governing the predicative AP analogous to the way a determiner governs NP, that licenses QR of its phrase, pied-piping the predicate along with the than-clause. It is even possible for a compared adverbial modifying a verb in a finite complement to license a wide scope VPE in its than-clause: (74) John said that it would rain more often than Mary did. (74) has a reading, albeit marginal, in which more often. . . modifies the embedded verb rain, while the than-clause contains a VPE taking the matrix VP said that. . . as its antecedent; which contrasts with the impossibility for a non-compared embedded temporal adverbial to contain a wide scope VPE (cf. (56) above): (75) John said that it would rain when Mary did.
Chris Wilder
The contrast (74) vs. (75) provides further support for the hypothesis that the comparative morpheme licenses QR of its phrase. One problem for the QR approach noted above concerns that apparent possibility for appositive relatives modifying names to contain ACDs, as in (76a) – the problem being that names are not QR targets. However, appositive relatives in complement clauses do not support wide scope ACDs. (76b), which would have the reading (76c) if the DP headed by the name Peter could undergo QR, is deviant: (76) a. John met Peter, who Mary did, too. b. *John said it would please Peter, who Mary did, too. c. John said it would please Peter, who Mary said it would please, too. In Section 2.1 above, it was suggested (following Lasnik 1995b) that (76a) involves pseudogapping, hence does not bear on the issue of antecedent containment and QR at all. That solution is further supported by the fact that (77b) – in the reading (77a) – matches the ill-formedness of (76b): (77) a. John said it would please Peter, and Mary said it would please Bill. b. *John said it would please Peter, and Mary did, Bill.
. What is the landing site of long-distance QR? The existence of wide scope ACDs motivates the claim that QR can raise constituents out of finite complement clauses into the next clause up (see Wilder 1997a for corroborating evidence). If this is so, then it is expected that the raised constituent will enter into scopal interactions with other quantified expressions originating in the higher clause. This expectation is not met, however. It is already clear for simple cases (not involving ACDs) that a quantified expression originating in a finite complement clause does not interact scopally with any quantified argument (subject, direct or indirect object) in the matrix clause. The embedded subject in (78) cannot take scope over an indefinite in the matrix: (78) a. Someone believes that everyone is a genius. b. John told someone that everyone is a spy.
(= ∀∃) (= ∀∃)
Addition of an extraposed relative clause modifiying the embedded subject, and containing a wide scope VPE does not alter this fact – the phrase headed by everyone cannot take scope over someone in (79): (79) a. Someone believes that everyone is a genius that Mary does. b. John told someone that everyone is a spy that Mary did.
(= ∀∃) (= ∀∃)
Antecedent-containment and ellipsis
In order to escape antecedent containment at LF, some phrase containing the wide-scope VPE (by hypothesis, the DP headed by everyone in (79)) must minimally gain scope over the constituent (V’) containing the matrix verb and its complement (the finite clause). At the same time, it must be ensured that rules of interpretation do not permit everyone to take scope over someone. In the theory of May (1985), quantified expressions may only move to positions adjoined to VP or IP (leaving aside movement to Spec,CP). Supposing that the phrase headed by everyone is adjoined to VP, and that the matrix subject someone is higher up the tree (e.g. in Spec,IP or adjoined to IP), an LF like (80a) is derived: (80) a.
someoney [everyonex that Mary believes [ x is a genius ]] [ y believes [ x is a genius ]] b. ∃y ∀x ( believe(m) (genius(x)) ) → ( believe(y) (genius(x)) ) c. There is a y such that for all x, if Mary believes x is a genius, then y believes x is a genius.
From (80a), the formula (80b) would be derivable, which correctly represents the interpretation of (79a). The problem posed by (80a) is that in current theories of scopal interactions, the matrix subject someone would be able to get into the scope of the raised DP, either by a rule of syntax such as quantifier lowering, or a rule of interpretation. In May’s theory, for example, once an expression is adjoined to some VP, rules of interpretation enable it to interact scopally with any other expression adjoined either to that VP or to the IP immediately dominating. The lowest adjunction site consistent with resolving antecedent containment in (79) would be the VP headed by believes (or told). If the phrase headed by everyone has adjoined to that VP, then nothing would prevent it from being able to take scope over someone. It does not seem possible however, to maintain that the landing site of QR is some lower position from which such scopal interactions would not be possible. If QR in (79a) were restricted to the finite complement clause, an LF would be generated in which the phrase headed by everyone would be in the scope of the matrix verb believe. The reading expected would be (81): (81) a. ∃y believe(y) ( ∀x ( believe(m) (genius(x)) ) → (genius(x)) ) b. There is a y who believes that for all x, if Mary believes x is a genius, then x is a genius. Quite apart from the fact that this assumption about the landing site of QR would lead to an antecedent-containment configuration at LF, the reading (81) – which is available for the ‘undeleted’ examples (82), is not available for (79a):
Chris Wilder
(82) a.
Someone believes that everyone that Mary believes is a genius, is a genius. b. Someone believes that everyone is a genius that Mary believes is a genius.
The facts about wide scope ACDs suggest instead that the account of scopal interactions may be in need of revision. There is one further constraint on wide scope ACDs that may be relevant to the issue of absent scopal interactions. The antecedent to the VPE in a wide scope ACD is restricted to the first VP dominating the finite clause containing the quantifier phrase. While (83a,b,c) are all possible, (84) is not: (83) a. John thinks that more trees died than Mary does _. b. John thinks that more trees seem to have died than Mary does _. c. John thinks that it seems that more trees have died than Mary thinks it does _. (84) *John thinks that it seems that more trees died than Mary does _ In each of (83a,b,c), the constituent (more trees. . . ) that undergoes QR is located in the clause directly below the VP-antecedent to the VPE in the comparative clause: (85) a. John [ thinks [ that more trees died than Mary does think died ]]. b. John [ thinks [ that more trees seem to have died than Mary does think seem to have died ]]. c. John thinks that it [ seems [ that more trees have died than Mary thinks it does seem have died ]]. In (84), the constituent (more trees. . . ) that undergoes QR is located in the second clause below the VP-antecedent to the VPE in the comparative clause: (86) *John [ thinks [ that it seems [ that more trees died than Mary does think that it seems died ]]]. This paradigm suggest that long QR is strictly local, in the sense that it may not raise a constituent further than the VP immediately dominating the first finite clause containing it. If this is correct, then the other facts discussed in this section would fall out, if the theory of scopal interaction is designed in such a way that a phrase raised from a finite complement and adjoined to VP cannot interact scopally with the arguments of the verb heading that VP.
Antecedent-containment and ellipsis
. German comparatives Cross-linguistic support for the account of wide-scope VP-ellipsis sketched here is not easy to come by, since many languages do not show VP-ellipsis in the English pattern at all, or only to a much more restricted extent. However, a similar contrast to the one discussed in Section 3 is found in an elliptical comparative construction that is not confined to English. Consider first the English examples (87)–(88). (87) Peter thought about it more often than Mary. (88) a. Peter thought that more trees had died than Mary. b. *Peter thought that more trees than Mary had died. Assuming that the complement to than in (87), superficially a bare DP, is underlyingly clausal, this clause undergoes ellipsis which deletes all daughters of I’, leaving the subject as the sole remnant. (87) thus has an abstract representation like (89): (89) Peter thought about it more often than Mary thought about it x-often Assuming further that I’-ellipsis, like VP-ellipsis, is subject to the PF-constraint (60), it is clear that (60) is satisfied by (89). Where the elliptical than-clause is associated with an embedded nominative, wide-scope interpretation is possible only when the comparative clause is extraposed (89a). (89b) only has the nonsensical interpretation ‘more trees than just Mary died’, i.e. that Mary is a dead tree. The contrast can be accounted for if the examples (88) have representations as in (90). (90) a.
Peter thought [that [more trees _] had died than Mary thought that x-many trees had died]. b. *Peter thought [that [ more trees than Mary thought x-many trees had died ] had died ].
(90b) violates the PF-constraint (60), while (90a) does not. (90a) also avoids antecedent-containment at LF, assuming that the extraposed elliptical than-clause is reconstructed into the comparative DP, yielding a constituent more trees than Mary thought that x-many trees died which then undergoes QR to a position above the matrix verb thought. Given that the same QR option is available for (90b), the ungrammaticality of this example can be taken as support for the hypothesis that I’-ellipsis is subject to (60). Unlike VP-ellipsis, I’-ellipsis in comparative clauses is found in other languages, for example, German. The contrast in (88) can be reproduced for German, as illustrated in (91):18
Chris Wilder
(91) a. *weil Peter meinte, daß es öfter als ich geregnet hat. since Peter said that it more-often than I rained has b. weil Peter meinte, daß es öfter geregnet hat als ich. since Peter said that it more-often rained has than I ‘since Peter said that it has rained more often than I (did)’ The German comparative morpheme selects the complementizer als which heads the comparative clause (the precise status of than/als is not crucial to the argument). This clause may undergo I’-ellipsis, with the result that only the subject DP is overtly expressed. In simple clauses, in contrast to (91), extraposition of the als-clause (subject to the Right Roof constraint) is possible but not obligatory, regardless of whether it is elliptical or not, as shown in (92)–(93): (92) a.
weil er öfter als ich darüber nachgedacht hat. since he more-often than I there-over thought has ‘since he has thought about this more often than I (have)’ b. weil er öfter darüber nachgedacht hat als ich. since he more-often there-over thought has than I ‘since he has thought about this more often than I (have)’
(93) a.
weil er öfter, als ich darüber nachgedacht habe, darüber nachgedacht hat. since he more-often than I there-over thought have there-over thought has ‘since he has thought about this more often than I have thought about it’ b. weil er öfter darüber nachgedacht hat, als ich darüber nachgedacht habe. since he more-often there-over thought has than I there-over thought have ‘since he has thought about this more often than I have thought about it’
The examples (92) have the analysis (94). Neither (94a) nor (94b) violate (60): (94) a.
weil er [öfter als ich x-oft darüber nachgedacht habe] darüber nachgedacht hat b. weil er [öfter _] darüber nachgedacht hat [als ich x-oft darüber nachgedacht habe]
By the same reasoning, the examples (91) are represented as (95): (95) a. *weil Peter meinte, daß es [öfter als ich meinte daß es x-oft geregnet hat] geregnet hat
Antecedent-containment and ellipsis
b. weil Peter meinte, daß es [öfter _] geregnet hat [als ich meinte daß es x-oft geregnet hat] (95b) does not violate (60), but (95a) does. Assuming that in (95b), the extraposed clause reconstructs into the phrase headed by öfter, with the latter undergoing QR to a position above the matrix verb meinte, antecedent-containment is avoided at LF. Given that the second step, i.e. QR, is equally available to (95a), the ungrammaticality of (95a) provides support for the claim that (60) applies also in German.19
. VP-ellipsis and string-containment The account of wide scope ellipsis contrasts in the preceding sections depended crucially on the PF-constraint (60). Now we return to the examples (96) (=65), put aside in Section 3.2. (96) a. John gave whoever he could _ two dollars. b. ?John finds everyone I do _ easy to work with. c. ?John believes everyone I do _ to be a genius. The question is, given that (60) excludes examples with an ellipsis site (properly) contained within the string corresponding to its antecedent in surface order, how do we account for such well-formed string-contained ACDs? The idea pursued here is that examples of this type are not ACDs at all. Rather, they involve two adjacent deletions, neither antecedent-contained in surface order, one due to pseudogapping, the other due to the independent rule of backward deletion (BWD). (96) are therefore related to examples like (97), where only part of the VP in the relative clause undergoes deletion under identity with material belonging to the matrix VP: (97) a. ?John gave [ whoever Mary did _ two dollars ] three dollars. b. ?John finds [ everyone I do _ hard to work with ] easy to work with. c. ?John believes [ everyone I do _ to be a fool ] to be a genius. The claim is that (96) involves pseudogapping as in (97), with additional backward deletion of the right-remnant of the embedded VP, under identity with corresponding material of the matrix VP, as in (98). Neither deletion site is properly contained within the string corresponding to its antecedent. (98) a. John gave [ whoever he could _ two dollars ] two dollars. b. John finds [ everyone I do _ easy to work with ] easy to work with. c. John believes [ everyone I do _ to be a genius ] to be a genius.
Chris Wilder
Put another way, the claim is that a derivation of (96a) involving VP-ellipsis (99a) is ill-formed, since it violates (60); while a derivation of the same string involving partial VP-ellipsis and BWD (99b) is well-formed, since it violates neither (60), nor any other constraint. (Here and below, BWD-dependencies are indicated by crossing-through of items deleted under identity with underlined items. As before VP-ellipsis sites, including partial VPE, are given in italics, their antecedents in bold.) (99) a. *John gave [ whoever he could give t two dollars ] two dollars. b. John gave [ whoever he could give t two dollars ] two dollars. If this is correct, then the examples in (96) have been mis-identified as ACDs arising from VP-ellipsis. What they actually indicate is that independent deletion processes may interact to give rise to adjacent ellipsis sites. Hence, such examples have no direct bearing on antecedent-containment per se; instead, they relate to the question of how apparent VP-deletions are to be analyzed in the first place (cf. Section 1). To make this idea plausible, it is necessary to show firstly, that in the good examples, the component deletions meet independently motivated constraints holding of each deletion process individually; and secondly, that any violation of a constraint leads to ungrammaticality, i.e. a bad derivation. The task is complicated by the fact that for a given example, there may be more than one derivation to consider (as in (99)). In the next section, the analysis of pseudogapping is discussed, and constraints relevant to the constructions at hand are identified. In Section 6.2, constraints on BWD are introduced. Interactions between them are discussed in Section 6.3.
. Partial VPE (pseudogapping) Pseudogapping can be regarded as ‘partial VPE’ in that it involves deletion of a VP that strands a subconstituent of that VP (complement or adjunct) – call it the ‘remnant’ – to the right of the finite auxiliary. (The class of possible remnants under pseudogapping is remarkably similar to the class of possible remnants under ordinary gapping, identified by the Major Constituent Condition of Hankamer (1973a).) (100) a. ?John read the book and Mary did, the newspaper. b. John read the book on the plane and Mary did, on the train. The following discussion focusses on complement remnants, since these are the cases we are interested in (argument (96a), small clause AP predicate (96b), complement infinitive (96c)).
Antecedent-containment and ellipsis
The claim that pseudogapping is essentially the same phenomenon as VPE is supported by the fact that both ellipsis types involve obligatory stranding of a finite auxiliary, and both show the same distribution, occurring in coordinate and subordinate contexts (pseudogapping thus contrasts with ordinary gapping, which involves obligatory ellipsis of the finite auxiliary and is limited to coordinate contexts). Beyond this, attempts to establish generalizations on pseudogapping are hampered by the fact that simple examples are generally felt to substandard. Acceptability varies across speakers and also across construction types – for instance, pseudogapping leaving a direct object remnant tends to be less awkward in comparatives ((101a) due to Sten Vikner) than in coordination (cf. (100a)): (101) a. It is harder to persuade the Americans of this than it is, the English. b. John started his paper at the same time as Mary did, her book. As noted above, pseudogapping can involve a discontinuous deletion site (cf. (9) above, and (102a)). At first sight, this makes pseudogapping look rather different from VP-ellipsis, which targets a constituent of a specific type. However, a unification of VPE and pseudogapping would be possible if it could be shown that the remnant (apparently stranded in VP) has actually moved out of VP prior to application of VP ellipsis, so that what is deleted in pseudogapping is a VP containing a trace of a remnant category (102b) (Jayaseelan 1990). (102) a. ?John sent the book to Bill and Mary did, the papers. b. . . . and Mary did [VP send tOB to Bill ] [the papers] The question arises then as to what is the nature of the operation that removes remnants from VP prior to ellipsis. This line is pursued by Lasnik (1995b), who proposes that the movement operation involved is raising to AGRo, i.e. overt ‘object shift’.20 Example (102) is analysed approximately as in (103): (103) . . . and Mary did [AgrP the newspaper [ Agr [VP send tOB to Bill ] Lasnik’s approach is supported by a generalization about possible pseudogapping remnants. Direct objects in simple and complex VPs (including ECM subjects) make good remnants. Bad remnants include complements of various types that are not direct objects of the verb heading the elided VP: AP predicates (104a); and ‘second complement’ XPs in the frame [ V NP XP ], such as dative and directional PPs (104b-c) and the second (theme) object in the double object construction (104d). Predicate APs of transitive small clause constructions (104e), ECMinfinitives (104f), and embedded direct objects of ECM-constructions (104g) likewise make bad pseudogapping remnants (though with heavy stress on the remnant, (104g) is less than fully deviant):
Chris Wilder
(104) a. *This tastes nice and that does _ horrible. b. *John gave a lot of money to Bill and Mary will _ to Susan. c. *John put some milk in the fridge and Bill did _ on the balcony. d. *John gave Mary a lot of money and Bill did _ a lot of advice. e. *John considers him smart and Bill does _ stupid. f. *John believes him to be smart but Bill does _ to be stupid. ?? g. He believes them to like beans, and she does, spaghetti.
(tastes) (give a lot of money) (put some milk) (give Mary) (consider him) (believe him) (believe them to like)
The paradigm does not support a simple positional generalization (the first postverbal complement always makes a good remnant, except in (104a)), nor a categorial generalization (NPs make good remnants, except in (104d) and (104g)). The correct generalization is that only the highest NP makes a good remnant – exactly the class of complements held in the Minimalist model to undergo Case-checking movement to AGRo. Lasnik’s generalization is problematic in various respects. For example, finite CP-complements and PP complements often make good remnants (if the sole complement of V), as in (105), although these categories are generally held not to participate in Case-checking: (105) a.
John claimed that Bill was sacked, and Mary did, that he was promoted. b. First, John spoke to Mary and then I did, to Sue.
More importantly for present purposes, the pseudogapping data presented in Lasnik’s paper almost uniformly involve coordinations. It turns out that categories that yield bad pseudogapping remnants in simple coordinations yield remnants that are ‘good’ (though awkward) in relative clauses and comparatives (cf. the relative clauses involving relevant overt pseudogapping remnants presented in (97)). I do not understand this asymmetry; but it is crucial to the present argument, since the categories in (96) which we argue correspond to pseudogapping remnants all constitute ‘bad’ remnants according to Lasnik’s approach (second NP in a double object construction (96a), small clause AP predicate (96b), ECM infinitive (96c)). Further data showing comparatives and relative clauses with small clause APs, directional and dative PPs, second objects, and ECM-infinitives as remnants, are given in (106)-(110). Even an embedded object in an ECM infinitive is marginally (and with heavy stress) acceptable as a remnant (111):21
Antecedent-containment and ellipsis
(106) a. ?John considers more students smart than Bill does, stupid. b. ?John considers everyone smart who Bill does, stupid. (107) a. ?John put more into the fridge than Bill did, onto the balcony. b. ?John put everything into the fridge that Bill had, onto the balcony. (108) a. ?John showed more sights to Susan than Bill did, to Mary. b. ?John showed every sight to Susan that Bill did, to Mary. (109) a. ?John gave more people two dollars than Bill did, three dollars. b. ?John gave everyone two dollars who Bill did, three dollars. (110) a. ?John believes more students to be geniuses than I do, to be fools. b. ?John believes everyone to be a genius that I do, to be a fool. (111) a. ??John claimed more people to have eaten beans than she did, spaghetti. b. ??John claimed everyone to have eaten beans that she did, spaghetti. Whatever the reason behind the asymmetry, the relative well-formedness of cases like (97) and (106)–(111) means that the current proposal – that examples of type (96) involve pseudogapping – can be upheld. There is a definite contrast between examples (110) and (111), with ECM infinitives, and corresponding cases with finite complements (112). It appears generally true that pseudogapping may not remove any subpart of a finite clause remnant, whether in coordination or in comparatives.22 (112) a. *John believes more students to be geniuses than I do, are fools. (believe (that). . . ) b. *He believes that they like beans, and she does, spaghetti. (*believe that they like. . . ) A constraint which plays a specific role in Section 6.3 is that pseudogapping of a verb which is left-adjacent to a finite clause lacking a complementizer is impossible. This holds even in contexts where ‘complementizer deletion’ is normally possible:23 (113) a. *Mary believes (that) Peter is a fool and Bill does, he is a genius. b. Mary believes Peter is a fool and Bill does, that he is a genius. The ban on deletion of a complementizer following a gapped verb is independent of whether the complementizer of the complement of the antecedent verb is deleted (113b). A subcase of this generalization is illustrated in (114a). (114) shows that a verb whose complement contains a subject trace cannot undergo pseudogapping: (114) a. *someone who he believes is a fool and (who) she does, is a genius. b. *someone who he believes is a fool and (who) she does, that is a genius.
Chris Wilder
This fact follows from two conflicting requirements – the presence of the complementizer required by the constraint on pseudogapping induces a violation of (whatever underlies) the that-trace filter. Where the latter is inoperative (object extraction), the complementizer may be overt, and pseudogapping is possible: (115) someone who he believes is a fool and (who) she does, that we should hire. Recall from Section 1 the question raised by the existence of pseudogapping, whether ordinary ACD sentences really involve antecedent-containment. Given that discontinuous gaps can arise, it may be that apparent VP-gaps are actually a sequence of adjacent, smaller gaps. Then the antecedent for each gap would be a correspondingly smaller constituent, possibly never containing its dependent gap (in surface structure). An argument that genuine antecedent-containment arises at S-structure comes from considering wide scope ACDs. The fact that CP cannot be ‘cut into’ can be used to deny the possibility of analyzing the deleted string corresponding to CP as a collection of individual deletion sites. If the deletion site in (116a) is dismantled into its smallest independently motivated parts, the deleted CP is still a single deletion site (116b); the analysis (116c), for instance, not meeting the Major Constituent Condition: (116) a. b. c. d.
John thinks everyone is intelligent that Mary does __. . . . that Mary does [think] [ t is intelligent] *. . . that Mary does [think] t [is ] [ intelligent] John [V2 thinks] [CP2 everyone is intelligent that Mary does [V1 think] [CP1 t is intelligent CP1 ]CP2 ]
Assuming the Right Roof Constraint to be valid, then that deletion site (CP) must be hierarchically contained in its antecedent (at S-structure) (116d).
. Backward deletion The backward deletion (BWD) operation assumed here covers what is normally termed right node raising, most commonly found in coordinate structures (117a). Since we are dealing with noncoordinate structures, it is important to realize that the process also applies in noncoordinate structures, as in (117b) (Williams 1990) and (117c): (117) a. John talked to __ ) and really got to like my new friend from Kansas). b. Anyone who talks to __ ) really gets to like my new friend from Kansas). c. . . . go from the last town north __ ) to the first town south of that mosquito-infested river).
Antecedent-containment and ellipsis
The conditions on BWD that are relevant below are:24 (118) a. The deletion site must be at right edge of its domain. b. The antecedent must be at right edge of its domain. There is no satisfactory account of what counts as a domain for BWD, that covers both coordinate and noncoordinate types.25 Considering BWD in coordination, the domains mentioned in (118) must be conjuncts; moreover, these domains must be adjacent, with the conjunct containing the deletion site immediately preceding the conjunct containing its ‘antecedent’ (hence the backward directionality). Assuming a view of phrase structure like that of Kayne (1994), including a Larsonian structure for complex VPs, one can generalize from the coordinate to the noncoordinate cases by stating that the domain for the deletion site is a specifier, and the domain for its antecedent is the X’-sister of that specifier (119a). Under the analysis of conjoined structures in which the conjunction is a head taking conjuncts as its complement and specifier, coordinate BWD configurations realize the subcase (119b). The non-coordinate cases also fit the scheme. (117b) involves the specifier of IP as the deletion domain, and I’ as the antecedent domain (119c); (117c) might involve two PPs embedded in a Larsonian VP (119d): (119) a. . . . [XP [SPEC . . . YP ] [X’ . . . YP ]] . . . b. . . . [&P [SPEC=CP . . . YP ] [&’ and [CP . . . YP ] ] . . . c. [IP [SPEC=DP . . . YP ] [&I’ . . . YP ] ] d. . . . [VP [SPEC=PP . . . YP ] [V’ . . . [PP . . . YP ] ] ] The cases we are interested in involve deletion in a relative clause modifiying a DP that is nonfinal in VP (120a). Under Larsonian assumptions, that relative clause forms the right edge of a complex DP in a specifier of an internal V’-constituent, as in (120b): (120) a. John gave [ whoever he could _ two dollars ] two dollars. b. . . . [VP [SPEC=DP . . . [CP . . . YP ] ] [V’ . . . YP ] ] . . . In the following, whether this characterisation is correct or not, I assume that the right edge of the BWD domain must coincide with the right edge of the DP containing the relative clause.26
. VPE/BWD interactions For the account of ‘string-contained’ ACDs (96) as adjacent VPE and BWD deletions (98) to go through, it needs to be shown that independent word order rules interact with partial VPE and BWD to give the right results for the acceptable ‘con-
Chris Wilder
tainment’ examples. Beyond that, it is desirable to show that an independently motivated failure of pseudogapping can be used to explain the ungrammaticality of a ‘string-contained’ ACD ; and to show that the violation of a BWD-constraint leads to ill-formedness of a ‘string-contained’ ACD. The latter two points could prove impossible to establish (the existence of a ‘bad’ derivation can be obscured by a ‘good’ derivation of the same string); nonetheless, as it turns out, all three goals can be secured:27 I.
For each ‘good’ string-contained ACD, there is at least one derivation involving (partial) VPE, BWD, applying on independently well-formed surface word order. II. If VPE fails, then the derivation fails. III. If BWD fails, then the derivation fails.
– ad (I): All three examples (96a–c) have a derivation involving partial VPE and BWD and meeting requirements of both: (121) a. John gave [ whoever he could give t two dollars ] two dollars. b. John finds [ everyone I do find t easy to work with ] easy to work with. c. John believes [ everyone I do believe t to be a genius ] to be a genius. Partial VPE is independently established by the (relatively) acceptable examples (97) without BWD. BWD itself meets the requirements (118): the deletion target is right-peripheral in its domain (the DP containing the relative clause), and the antecedent to the deletion is right-peripheral in its domain – the matrix VP (strictly speaking, the V’ sister of the DP containing the relative clause).
– ad (II): For the original example (122) (=62), whose ungrammaticality was attributed to the PF condition (60) under the VPE-derivation (123a), it is now necessary also to exclude an alternative derivation, involving partial VPE interacting with BWD (123b): (122) *John believes everyone I do, is a genius. (123) a. *John believes [ everyone I do believe t is a genius ] is a genius ] b. *John believes [ everyone I do believe is a genius ] is a genius In (123b), partial BWD targeting the string is a genius at the right edge of the relative clause, under identity with the same string at the right edge of the complement to the matrix believe should be possible. In this derivation, however, partial VPE fails, for reasons discussed above: a verb whose complement clause contains a subject trace cannot undergo pseudogapping (114). Inserting a complementizer in the antecedent does nothing to save
Antecedent-containment and ellipsis
the example – the complementizer governed by believe in the relative clause cannot be included in a partial VPE, since VPE cannot delete ‘into’ a finite clause:28 (124) *John believes [ that [ everyone I do believe [ that t is a genius ]] is a genius] Nor may the complementizer in the relative clause be left undeleted (to save pseudogapping) on pain of a that-trace violation: (125) a. *John believes that everyone I do, that, is a genius. b. *John believes that [ everyone I do believe that t is a genius ] is a genius. Hence, it can be concluded that an independently motivated constraint on partial VPE does real work in correctly excluding ‘string-contained’ ACDs.
– ad (III): To show that failure of BWD can lead to ungrammaticality of ‘stringcontained’ ACDs, it is sufficient to consider examples in which the part of the gap that is due to BWD (i.e. the part whose antecedent follows the gap) is itself not right-peripheral in the relative clause. While the judgements are subtle, the contrasts are clear enough. Consider the paradigm (126): (126) a. b. c. d.
?John believes [ everyone I do _ to like spaghetti ], to like beans. ?John believes [ everyone I do_ ] to like beans.
*John believes [ everyone I do _ spaghetti ], to like beans. ??John believes to like beans [ everyone I do _ spaghetti ].
(126a) involves partial VPE of believe in an ECM construction (cf. (97c) above). The same partial VPE is involved in (126b); the remainder of the gap (to like beans) is BWD-site, whose antecedent follows the gap, as in (126b ). The competing derivation (126b ) which treats the whole gap as a VPE-site, violates stringcontainment (60): (126) b . John believes everyone I do believe t to like beans, to like beans. b *John believes everyone I do believe t to like beans, to like beans. (*PF-containment) (126c) has no good derivation. Treating the deletion of believes to like as partial VPE as in (126c ) – marginally possible, cf. (111) above – falls foul of the PF-condition (60). The alternative (126c) treats the gap as two deletions: pseudogapping of believe, as in (126a), and BWD of to like. The latter – crucially – violates the right-edge condition on BWD-sites (118a): the deletion is not right-peripheral in its domain (=the relative clause), since it is followed by non-deleted spaghetti:
Chris Wilder
(126) c . *John believes everyone I do believe to like spaghetti, to like beans. (*PF-containment) c . *John believes everyone I do believe to like spaghetti, to like beans. (*BWD) It is possible for the deletion site to be displaced to the right, by Heavy NP Shift of the accusative subject containing the relative clause, as in (126d). Crucially, now, the gap in the relative clause is no longer followed by part of its antecedent in surface order. Hence, there is a derivation involving partial VPE of the whole gap which does not violate (60); and BWD is not necessary to derive any part of the gap: (126) d . John believes to like beans everyone I do believe to like spaghetti. The example is awkard (requiring heavy stress on the remnant spaghetti, recalling (104g) and (111)), but does not have the sharp deviance of (126c). Analogous paradigms are given in (127) and (128). The offending BWD-site in (127c) is in the fridge, the PP-complement of put; the element blocking rightperipherality of that BWD-site is the temporal adverb at night, distinct from the matrix during the day – hence necessarily part of the relative clause. In (128c), BWD of three dollars is blocked by embedded yesterday – again, distinct from the matrix today, hence necessarily part of the relative clause. Reordering the gap to the right of the antecedent string in (127d) and (128d) again permits the gap to escape the conspiracy of the PF-constraint (60) and the BWD peripherality requirement (118a):29 (127) a.
John puts [ everything I do _ on the balcony at night ] in the fridge during the day. b. John puts [ everything I do _ ] in the fridge during the day. c. *John puts [ everything I do _ at night ] in the fridge during the day. d. John puts in the fridge during the day [ everything I do _ at night ]. ?John
gave [everyone that I did _ two dollars yesterday ] three dollars today. b. ?John gave [everyone that I did _ ] three dollars today. c. *John gave [everyone that I did _ yesterday ] three dollars today. d. John gave three dollars today [ to everyone that I did _ yesterday ].
(128) a.
From these paradigms, it can be concluded that constraints on BWD are independently needed to derive acceptability patterns among ‘string- contained’ ACDs. In sum, the contention that such examples do not constitute genuine ACDs finds subtle, hence strong, empirical support.
Antecedent-containment and ellipsis
Acknowledgements An earlier version of this material was presented in Berlin (July and November 1995) and Potsdam (November 1995) and circulated as Wilder (1995a). For helpful comments, I thank the following: M. Bierwisch, D. Büring, M. Den Dikken, G. Fanselow, H.-M. Gärtner, G. Jäger, P. Ludlow, A. Mahajan, R. May, T. Roeper, M. Starke, M. Steinbach, A. Strigin, R. Vogel, S. Vikner, J.-W. Zwart. The paper was indirectly inspired by Howard Lasnik’s lectures on Ellipsis and QR in Berlin in June 1995.
Notes . A third proposal, that a VPE is some kind of pronominal element with no internal structure at any syntactic level (cf. Dalrymple et al. 1991), is left out of consideration here. Under a derivational approach to chains, the fact that a VPE may contain a trace of wh-movement, cf. (i) and ACDs like (2), is most naturally accommodated under a deletion approach, which attributes internal syntactic structure to a VPE throughout the syntactic derivation. The copy approach must allow for the presence of a wh-phrase in Spec,CP at S-structure that binds no trace at that point in the derivation (the trace is only provided by subsequent copying of VP2 into VP1). (i) I know who Mary [VP2 met t], but I don’t know who John did [VP1 e]. . Under a deletion approach, the ‘infinite regress’ account can be recast in terms of ‘nonderivable structures’. If VP1 and VP2 must be structurally identical, i.e. syntactic copies of each other according to some measure of identity, for a VPE-dependency to arise, then no finite representation corresponding to (12a) can be generated. VP1 cannot be both structurally identical to and properly contained in VP1. . This account involves assumptions about identity (spelled out in Fiengo & May 1994) that are far from innocuous: the trace left by QR in the antecedent VP counts as identical to the trace bound by the relative clause operator in the VPE. In cases like (i), a focussed constituent (Bill) of the antecedent VP counts as identical to the trace of the wh-phrase in the VPE, cf. (ii). I sidestep this issue here; see Merchant (2001) for discussion. (i) I know that Mary called Bill a fool but I don’t know who John did [VP1 e]. (ii) I know that Mary [VP2 called [Bill]F a fool], but I don’t know whoj John did [VP1 call tj a fool]. . Although near can be stranded under wh-movement (i), the combination with pseudogapping in a relative clause (ii) is not good. Pied-piping of the preposition (iii) is much better: (i) ?John stood near Sue, who Mary stood near, too. (ii) *John stood near Sue, who Mary did near too. (iii) ?John stood near Sue, near whom Mary did too.
Chris Wilder
Suppose that pseudogapping involves VP-ellipsis in combination with raising of the ‘remnant’ out of VP. If the remnant is moved to an A’-position, the deviance of (ii) might be explained as a ‘freezing’ effect (an phrase moved to an A’-position is ‘frozen’, so that nothing may be subsequently extracted from it): (iv)*. . . whoj Mary did [VP stand t PP ][PP near t j ]. . Baltin does not comment on whether in addition to (26), there is a filter banning antecedent containment at LF. Nor does he make explicit whether he endorses QR, or LFcopying as the mechanism behind VPE-interpretation. . An analysis of (30a) in which both complements have shifted rightwards, vacating VP, as in (i), is precluded by independent considerations: (i) John [VP gave t j t k ] [whoever he could _ ]j [two dollars]k . While RCE may apply to a goal NP, splitting off the relative clause from the NP-head as in (ii), free relatives are never split by RCE (iii). In argument positions, free relatives behave like NPs; in particular, in the goal argument position in the double object construction, neither undergoes rightward displacement of the Heavy NP Shift type (iv, v). (ii) John gave [ everyone _ ] two dollars [who he met at the party]. (iii)*John gave [ whoever_ ] two dollars [he met at the party]. (iv)*John gave _ two dollars [whoever he met at the party]. (v) *John gave _ two dollars [the boy next door]. Hence, under standard assumptions, the VPE in (30a) is contained within its antecedent VP at S-structure. . Given the VP-internal subject hypothesis, the VP that is copied in (33c) contains an Atrace left by the main clause subject John, which functions in the relative clause as the Atrace of the subject Mary. It must therefore be assumed that NP-traces count as identical for purposes of VPE. This assumption is necessary even for the simplest cases: cf. (i), where the trace left by movement of John in the first conjunct must count as identical with the NP-trace forming a chain with Mary in the second conjunct: (i) John was [kissed t ] and Mary was _ too. Other technical assumptions are left aside here. . Chomsky (1995: Ch. 4) subsequently developed an account of covert A-movement (object raising) that does not involve pied-piping of the whole DP. . Example (30a), with the VPE contained within the goal NP object of give in a double object construction, can plausibly be handled via A-movement. . (39b) is slightly degraded in Lasnik’s judgement. I find the example perfect. . There is no narrow scope reading for the VPE in these examples: be+XP (is a genius/were stupid) cannot reconstruct under do/did. . It must be assumed in addition that the extraposed relative reconstructs into the subject prior to QR. See Section 3.2.
Antecedent-containment and ellipsis . It has been proposed that QR may extend the domain of extraposition, such that the extraposed clause may cross a sentence boundary rightwards, in violation of the Right Roof Constraint, just in case the head of the extraposed clause itself raises the same distance leftwards – cf. Guéron and May (1984). But nowhere has it been proposed that extraposition may violate the right-roof constraint in order to extend the domain of QR. . In fact, Hornstein is aware of the problem. In a footnote, he addresses the issue of widescope ACDs and seeks to deny the facts (Hornstein 1994: 462, fn. 9 – my emphasis): May (p.c.) notes cases that are problematic for the above claims: (vi) Max thinks the yacht is longer than Oscar does (vi), he says, has the interpretation ‘. . . longer than Oscar thinks it is’ [. . . ] These cases all strike me as unacceptable. However, it is possible that more is going on here than present analysis accounts for [. . . ] if matrix ACD readings are regularly possible from null VPs in embedded clauses, then this would tell against the minimalist approach sketched here. I will proceed as if this is not the case. . Fiengo and May (1994: 254, fn.16) mention a relevant contrast between (i) and (ii), which they attribute to Pauline Jacobson. According to Jacobson, (i) permits a wide scope reading but (ii) does not: (i) His father said he read every book that his teacher did _. (= read, or said he read) (ii) John said Mary put every book that Paul also did _ on the shelf. (= put on the shelf; = said he put on the shelf ) Fiengo and May seek to deny the contrast; I agree with Jacobson. . This discussion presupposes that (60) belongs to UG. One may speculate that the facts (60) is intended to capture may in fact not be linguistic at all, but rather reflect an extralinguistic condition, perhaps relating to the parsing mechanism. . Thanks to A. Mahajan and H-M. Gärtner for directing me to this reference. It is important to realize that the non-presuppositional reading is not to be equated with the narrow scope reading in sentences with an additional quantifier. Rather, only in its presuppositional reading does a weak quantifier undergo QR, in which case it may take narrow or wide scope with respect to the second quantifier. . Some speakers find (91b) degraded, but the contrast with (91a) seems to be robust. . This account of the contrast in (88) (English), and between (94) and (95) (German) depends crucially on the assumption that the complements of than and als in the comparative is always clausal, with the bare DP in these cases being left behind by ellipsis. If there is the possibility for than/als to govern a bare DP, then there is the possibility for an abstract representation of (88) or (95) without ellipsis, and the constrast goes unexplained. This assumption has further consequences regarding the notion of identity required for ellipsis. In (i), the DP Peter, in the complement of than, is interpreted as the subject of a VP – more correctly, an I’-constituent – whose antecedent is nonfinite. Given that the DP Peter is overt, it must be in a Case position (in terms of Chomsky & Lasnik 1993, not a position in which ‘null Case’ is licensed). Hence this DP cannot be the subject of a non-finite IP (ii). It must
Chris Wilder
be concluded that a finite I’ can be deleted under identity with a non-finite I’ (iii). The same goes for the German equivalent (iv): (i) To eat more apples than Peter is not advisable. (ii) *[PRO to eat more apples than Peter to eat x-many apples], is not advisable. (iii) [PRO to eat more apples than Peter eats x-many apples], is not advisable. (iv) Mehr Äpfel zu essen als Peter, ist nicht ratsam. more apples to eat than Peter is not advisable. . Object Case-checking movement is argued by Lasnik to apply pre-S-structure in English. This entails that the verb (when overt) also raises pre-S-structure to a position higher than the SpecAgro. In VP-ellipsis, Lasnik suggests that the verb does not raise. The fact that the verb does not raise when the VP is deleted can be accounted for, by assuming that overt V-raising is triggered by a ‘strong feature’ of V which needs to be eliminated before PF: VPdeletion eliminates the strong feature of V, removing the need for V-raising. The embedded object in the ECM construction (104g) is unable to be a pseudogapping remnant since it checks Case within the VP affected by ellipsis; a similar account can be given for the second object in the double object construction (104d). . Not all of Lasnik’s ‘bad remnants’ improve in comparatives; the V+AP case (104a) remains quite deviant: (i) *?John feels happy more often than he does _ sad. . Notice that two independent (partial) VPE operations can coexist in a complex structure, even a verb-complement construction such as (ii): (i) John will leave, if you come, and Mary will, if I do. (ii) John believes that Mary likes beans, and Bill does, that Sue does, spaghetti. Example (ii) does not counterexemplify the text claim. The deletion of the lower finite verb begins a new (partial) VPE domain. . This fact is probably related to Stowell’s (1981) observation that a deleted complementizer has to be adjacent to its governing verb: cf. Mary believes sincerely *(that) Peter is a fool. . See Wilder (1995c, 1997b) for fuller discussion of BWD. . Williams (1990) suggests that noncoordinate RNR contexts are really conjoined structures – subject conjoined with predicate, etc. . In (120a), the antecedent string exhausts the overt terminals of the antecedent domain (=V’). The converse situation – where the deletion site exhausts the deletion domain (=Spec), must probably be excluded. Kayne (1994), Wilder (1995c) assume that a deletion site may not c-command its antecedent. . If (I)–(III) are not met, we are faced with a problem of undergeneration, avoidance of which is a sine qua non. Independently of this, questions of overgeneration also arise. One such concerns the fact that BWD yields unacceptable examples when applying independently of pseudogapping, cf. (i) (R. May, p.c.): (i) *?John gave whoever he could give, two dollars. (ii) John gave [ whoever he could give t two dollars ] two dollars.
Antecedent-containment and ellipsis
One factor contributing to the deviant status of (i) may be the identity of the verbs, which makes pseudogapping possible in the first place. If non-identical verbs are chosen, the example seems to improve somewhat: (iii)??John sent whoever he couldn’t give, two dollars. However, the real issue here is the mystery of what governs the applicability of noncoordinate BWD (cf. Williams 1990). . In fact, in the version of (124) that avoids the PF constraint (60) by extraposition, it makes no difference whether or not the antecedent contains overt that, although the appearance of that in the relative clause would cause a that-trace violation: (i) John believes that everyone is intelligent that I do _. (ii) *John believes that everyone is intelligent that I believe that is intelligent. From this example alone it is impossible to decide whether (a) the overt that is reconstructed into the VPE site, with the that-trace filter only applicable to overt that; or (b) the zero complementizer is treated as nondistinct from the overt complementizer for purposes of VPE-identity. Other examples indicate the former. Overt verbs of manner-of-speaking such as mutter do not license the zero complementizer: (iii)*John muttered everyone is intelligent. However, wide scope VPE, involving reconstruction of mutter together with its complement CP containing a subject trace, is possible: (iv) John muttered that everyone was intelligent that I did. (v) *John muttered that everyone was intelligent that I did mutter that was intelligent. (vi) John [ muttered [ that [ everyone _ ] was intelligent [ that I did mutter that t was intelligent ] ]] (vi) shows that the that-trace effect can be voided in VPE contexts for the verbs in question. The that-trace effect appears to be sensitive solely to the PF-status (overtly realized or not) of the complementizer. . The bare DP goal object of give does not undergo Heavy NP Shift; hence the effect of reordering the impossible deletion site of (128c) is shown using the variant with to in (128d).
Background matching in right node raising constructions Katharina Hartmann The first part of this paper argues that right node raising in German is not derived by movement, but by phonetic deletion of the right node raising target in the first conjunct. Several syntactic and semantic arguments are adduced to corroborate this claim. The second part of the paper elaborates the demands on a theory of phonetic deletion of right node raising constructions. The main condition for phonetic deletion is a parallel focus structure of the two conjuncts, which requires syntactic parallelism, as well as a symmetric distribution of pitch accents.*
.
Introduction
This paper analyzes German right node raising constructions. Right node raising (RNR) is one kind of elliptical coordination. Typically, the conjuncts share a string of elements which appears at the end of the coordinated construction, vide (1) and (2). The aim of this paper is to develop a theory of RNR in German which is based on the assumption that the first conjunct is elliptical at its right periphery: ellipsis in RNR is due to a process of phonetic deletion which is triggered if the right peripheral elements of the two conjuncts are phonetically identical. The central claim is that ellipsis in RNR, which is optional, is an epiphenomenon of the focus structure and the accentuation pattern found in RNR. Both will be thoroughly analyzed. This paper has two parts. In the first part (Section 2), I present strong evidence against the traditional assumption that RNR is derived by movement of the RNR target, i.e. the string of elements shared by the two conjuncts, to the right periphery of the coordinated structure. Instead I defend the claim that RNR involves deletion of the phonetic matrix of a right peripheral element (or string of elements) in the first conjunct. This In Situ Theory of RNR is corroborated by the fact that the target of RNR does not behave as if rightward moved. That is, it does not have to be a constituent, but it can be a DP, differing from extraposition which is only possible
Katharina Hartmann
with PPs and clauses. The alleged movement neither respects the Right Roof Constraint, which is known to restrict extraposition, nor is it sensitive to other locality constraints. Furthermore it is shown that the target behaves as if not moved with respect to scope. In the second part (Sections 3 to 6), I develop the In Situ Theory of RNR. In a nutshell, it is claimed that a RNR construction is subject to the principle of Pragmatic Licensing of Schwarzschild (1999) which establishes the pragmatic wellformedness of an utterance in relation to the linguistic context. A RNR construction can fulfil Pragmatic Licensing only by the assumption that each conjunct serves as a linguistic context for the other conjunct. The typical RNR format which allows PF-deletion then turns out to naturally follow from the accentuation pattern a conjunct is necessarily associated with in order to qualify as a linguistic context for the other conjunct.
. An in situ analysis of right node raising The term right node raising describes a coordination construction whose characteristic property is that the conjuncts share a string of elements which appears only at the end of the second conjunct.1 An example of right node raising (RNR) in German, which is taken from the newspaper Frankfurter Rundschau (26.11.1998), is given in (1). The shared elements are underlined. (1) Vicky Leandros besang und Petra Roth beschwor die Schönheit des V.L. sang.of and P.R. evoked the beauty of the Frankfurter Weihnachtsmarkts. Frankfurt Christmas.fair ‘Vicky Leandros sang of and Petra Roth evoked the beauty of the Frankfurt Christmas fair.’ RNR constructions also exist in English, an example is presented in (2). This paper, however, focuses mainly on German RNR. (2) Timothy wrote and Paula read a funny story. The term right node raising (due to Postal 1974) implicitly carries the following instructions for use: Take some element (i.e. the string-identical right peripheral element from each conjunct of a coordinated structure) and right-adjoin it to a node which dominates the conjunction, as illustrated in (3). Such an analysis describes right node raising (RNR) as an instance of across-the-board (ATB) extraposition; I will refer to it as the Movement Theory. (3) [XP [XP . . . tYP ] and [XP . . . tYP ]] YP
Background matching in right node raising constructions
An alternative analysis to the Movement Theory is what I call the In Situ Theory. This theory derives RNR not by movement, but by phonetic deletion of a rightperipheral element of the first conjunct. The phonetically deleted constituent is orthographically represented by crossed out letters. (4) [XP [XP . . . YP] and [XP . . . YP]] In what follows, I will provide robust arguments which are incompatible with the Movement Theory and at the same time clearly corroborate the In Situ theory. Concretely, I show that the target of RNR (i.e. YP in (3) and (4))2 does not behave as if moved with respect to the following six points: (i) RNR can involve nonconstituents; (ii) RNR can strand prepositions in languages which do normally not allow preposition stranding; (iii) the target can be a DP which, however, does not undergo extraposition in general; (iv) RNR does not respect island constraints; (v) the targets in the two conjuncts may be DPs which represent different referents – a fact which hardly follows from ATB-extraposition; and (vi) the target behaves as if not moved with respect to binding, negative polarity, and pronoun binding. It will be demonstrated that the first conjunct, although being elliptic, preserves a complete syntactic structure, a conclusion also drawn in Kennedy’s and Merchant’s contributions to this volume.
. Constituency A first indication against the Movement Theory is that RNR, if it were analyzed according to the Movement Theory, would not have to respect constituency. In contrast to Hankamer (1979), Postal (1974), Bresnan (1974) and Reinhart (1991: 370) who says: “In the standard cases of ellipsis, which may be labeled constituent ellipsis, the elliptic, or ‘missing’ material corresponds to a constituent”, the target of RNR does not have to be a constituent (as noted for instance in Klein 1981). As the data below illustrate, the RNR target may involve more than one syntactic constituent.3 This claim is supported by the fact that none of the target strings can be subject to any movement transformation in non-coordinated contexts (cf. (5b) and (6b)). Peter verspricht seiner Mutter, in die Kirche zu gehen, und mother in the church to go and P. promises his Maria verspricht ihrer Mutter, in die Kirche zu gehen. M. promises her mother in the church to go ‘Peter promises his mother to go to church, and Maria promises her mother to go to church.’ b. *[Mutter in die Kirche zu gehen] verspricht Peter seiner.
(5) a.
Katharina Hartmann
(6) a.
Ramona hat ein Lied gesungen über die Revolution, und R. has a song sung about the revolution and Romana hat eine Hymne gesungen über die Revolution. R. has a hymne sung about the revolution ‘Ramona sang a song, and Romana sang a hymne about the revolution.’ b. *[Gesungen über die Revolution] hat Ramona ein Lied.
In (5) the target consists of the NP of the indirect object DP plus an infinitival complement clause; the target in (6) contains the participle and the extraposed PP. In none of the examples does the target consist of one constituent only. This is evident in (5) where the NP is severed from its possessive pronoun belonging to the remnant (i.e. the conjunct except for the target). It also holds for (6) assuming that extraposition is movement to some adjoined position to the right of VP (cf. Büring & Hartmann 1997). As syntactic movement is sensitive to constituency, the Movement Theory of RNR cannot account for this type of data. By no means is NP plus an infinitival clause a constituent, and neither is a participle together with an extraposed PP.4
. Preposition stranding A second argument in favor of the In Situ Theory concerns the remnant of RNR. McCloskey (1986) argues that languages which do not allow preposition stranding in general nevertheless seem to permit it in RNR. This effect can be observed in German. In (7), a preposition can be left behind after RNR, although leftward movement in these languages does not allow a preposition to strand. (The ‘stranded’ prepositions are printed in boldface.) Die Katze Halma sitzt auf dem Ofen, und der Kater Mikado the cat H. sits on the oven and the tomcat M. sitzt neben dem Ofen. sits beside the oven ‘The cat Halma sits on the oven, and the tomcat Mikado sits beside the oven.’ b. *[Dem Ofen]1 sitzt die Katze Halma auf t1 . c. [Auf dem Ofen]1 sitzt die Katze Halma t1 .
(7) a.
It has been argued, however, that rightward movement and leftward movement are limited by quite different constraints. Thus, preposition stranding – although excluded with leftward movement in German – could still be a property of extraposition. In order to make the argument water-proof, it has to be shown that prepositions are not allowed to strand after extraposition in German either. DPs
Background matching in right node raising constructions
optionally extrapose if they are prosodically heavy. Consider (8) where the passive subject appears after the participle which occupies the final position, given that German is a SOV language. (8) Auf dem Empfang wurde uns t1 serviert [DP ein sehr reichhaltiges usDAT served a very plentiful at the reception was Angebot an erlesenen Speisen und Getränken]1 . collection of selected dishes and drinks ‘We were served a very plentiful collection of selected dishes and drinks at the reception.’ The same heavy DP, however, can not extrapose if it originates in a PP thereby stranding the preposition, cf. the ungrammatical (9a). Extraposition of the whole PP is obligatory in this case (cf. (9b)). (9) a. *Vor dem Empfang wurden wir [PP über t1 ] informiert [DP das we about informed the before the reception were sehr reichhaltige Angebot an erlesenen Speisen und very plentiful collection of selected dishes and Getränken]1 . drinks ‘Before the reception we were informed about the very plentiful collection of selected dishes and drinks.’ b. Vor dem Empfang wurden wir t1 informiert [PP über [DP das sehr reichhaltige Angebot an erlesenen Speisen und Getränken]]1 . To sum up, extraposition in German does not allow to strand prepositions. If RNR would be an instance of ATB-extraposition, as advocated by the Movement Theory, the grammaticality of (7a) would not be accounted for.
. Extraposition Apart from heavy DPs, extraposition cannot involve DPs, differing from RNR again. As illustrated below, a simple DP can be the target of RNR (10a), but not of extraposition (10b) which is possible only with PPs and CPs ((10c), cf. Büring and Hartmann 1997). (10) a.
Hans reizt mit [DP vier Buben], und Peter reizt ohne [DP vier H. bids with four jacks and P. bids without four Buben]. jacks ‘Hans bids with, and Peter without four jacks.’
Katharina Hartmann
b. *Hans hat t1 gezogen [DP vier Buben]1. has drawn four jacks H. ‘Hans has drawn four jacks.’ c. Hans hat t1 geglaubt, [CP dass er mit diesem Blatt ein has thought that he with this hand an H. Null-Ouvert gewinnen könnte]1 . open.null win could ‘Hans thought that he could win an open null with this hand.’ A further comparison between RNR and extraposition again stresses more differences rather than common properties of these two constructions. As is well-known, extraposition obeys the Right Roof Constraint or Constraint on Upward Boundedness (Ross 1967: 174ff.) which states that extraposition is a clause-bound transformation, i.e. it cannot leave a CP by successive cyclic movement through SpecC. The Right Roof Constraint can also be observed in German (cf. Müller 1995). While the PP in (11a) is locally extraposed, it crosses a CP in (11b), thereby provoking the contrast in grammaticality between (11a) and (11b). [CP Eric versprach [CP dass er [DP das Buch t1 ] lesen will that he the book E. promised read want [PP über die Geschichte von Hessen]1 ][CP nachdem es ihm after it him about the history of Hesse empfohlen wurde]]. recommended was ‘Eric promised that he will read the book on the history of Hesse after it was recommended to him.’ b. *[CP Eric versprach [CP dass er [DP das Buch t1 ] lesen will [CP nachdem es ihm empfohlen wurde]] [PP über die Geschichte von Hessen]1 ].
(11) a.
The Right Roof Constraint does not hold in RNR constructions. As example (12) illustrates, the RNR target geflogen ist (‘has flown’), if it had indeed moved, would have to have done so across several CP boundaries. (12) [CP [CP Hans erzählte uns [CP dass Christian nach Chile us that C. to Chile H. told geflogen ist und [CP Klaus erzählte uns [CP dass Claudia nach us that C. to flown is and K. told Ecuador geflogen ist]]]. Ecuador flown is ‘Hans told us that Christian flew to Chile, and Klaus told us that Claudia flew to Ecuador.’
Background matching in right node raising constructions
Sections 2.2 and 2.3 both showed that RNR behaves quite differently not only from leftward movement but also from rightward movement. This behavior of RNR represents clear evidence against the Movement Theory.
. Islands A further prediction of the Movement Theory is that RNR should be sensitive to locality constraints. As (13) illustrates, however, RNR freely ignores all kinds of potential island violations. In (13a), the target is contained within a complex DP, in (13b) it is embedded in a wh-island. In both cases RNR does not lead to a violation of the complex NP constraint and the wh-island constraint, respectively. Ulf verbreitete [DP die Behauptung [CP dass Petra schwanger sei the rumor that P. U. spread pregnant is und Martin äußerte [DP die Vermutung [CP dass Claudia expressed the supposition that C. and M. schwanger sei]]. pregnant is ‘Ulf spread the rumor that Petra is pregnant, and Martin expressed the supposition that Claudia is pregnant.’ Ulf bleibt], und Hans fragt [CP wo b. Peter fragt [CP wo asks where U. is asks where and H. P. Klaus bleibt]. K. is ‘Peter asks where Ulf is, and Hans asks where Klaus is.’
(13) a.
In the following sections, I turn to discuss interpretational facts which suggest exactly the same conclusion: RNR is not derivable by movement.
. Quantified expressions Another prediction of the Movement Theory is that the traces purportedly created should exhibit interpretational identity. That is, a DP raised across-the-board by RNR should only allow the same referent in both conjuncts. However, if the target is a quantified expression, such a condition does not obtain, as verified by the following example. (14) Hans kauft einen Roman, und Maria liest einen Roman. H. buys a novel and M. reads a novel ‘Hans buys, and Maria reads a novel.’
Katharina Hartmann
The predominant interpretation of the indefinite in (14) is referential diversity of the existential expression in the target of the first and second conjunct. In other words, the book bought by Hans is not the same book as the book read by Mary under the predominant reading. This contrasts with leftward ATB-movement. The only meaning attributable to the wh-construction in (15) is the following: for which novel is it true that Hans bought it and Maria read it. (15) [Welchen Roman]1 hat Hans t1 gekauft und Maria t1 gelesen? novel has H. bought and M. read which ‘Which novel did Hans buy and Maria read?’ The In Situ Theory of RNR predicts the right interpretation of (14): if RNR is derived by phonetic deletion, the referential properties of the object DP in the first conjunct are independent of the referential properties of the object DP in the second conjunct. A similar point has been made on several occasions in the literature. As e.g. Klein (1993: 773) notes, the interpretation of the indefinite jemand in (16a) allows a reading where the arriving person and the leaving person are not identical. This RNR example contrasts with constituent coordination in (16b), where the subjects of the two conjuncts have an obligatorily identical reference. ging Um vier Uhr kam jemand, und um fünf Uhr at four o’clock came someone and at five o’clock left jemand. someone ‘At four o’clock somebody came, and at five o’clock somebody left.’ b. Jemand [C’ kam um vier Uhr] und [C’ ging um fünf Uhr].
(16) a.
The difference between (16a) and (16b) can be reduced to a phrase structure variation. As RNR takes place at a post-syntactic level, both conjuncts are syntactically full CPs containing an indefinite subject each. (16b), on the other hand, is an instance of constituent coordination where the indefinite subject is moved ATB to SpecC. It is therefore interpreted as referentially identical in the two conjuncts.
. Scope The last argument which I take to bear against the Movement Theory concerns the scope of the target. The present section examines whether RNR alters scope relations between an element of the remnant, i.e. the conjunct minus the target, and (an element contained within) the target with respect to Binding Theory, pronouns bound by a quantified DP, and negative polarity. Assuming the Movement Theory, the RNR target right-adjoins to a sentential projection – probably the root
Background matching in right node raising constructions
node – thereby raising out of the scope of the remnants. This is schematically illustrated in (17): (17) [CP [CP [CP . . . α. . . t1 ] & [CP . . . α. . . t1 ]] [. . . β. . . ]1 ] As (17) shows, the element β, which is c-commanded by α prior to RNR, will be outside of α’s scope after RNR. Under the In Situ Theory, on the other hand, the target remains unmoved and the scope relations are identical before and after RNR. (18) [CP [CP . . . α. . . [. . . β. . . ]] & [CP . . . α. . . [. . . β. . . ]]] Thus, scope can be advanced as a further argument when investigating the nature of right node raising. Several authors (Levine 1984 and Phillips 1996) observe that in English, a RNR target behaves as if it had not moved with respect to Binding Theory, bound pronouns, and negative polarity. Similar ‘still-there’ effects hold in German, too. Assuming the Movement Theory, the target in (19) (i.e. the complement clause containing the R-expression Eva) would end up in a position where it would be out of the c-command domain of the subject pronoun. An interpretation with the pronominal subject from the main clause and the subject from the embedded clause being coindexed should be available, contrary to fact.5 (19) *Siei vermutet, dass Evai schwanger ist, und ich bestätige die she supposes that Eva pregnant is and I confirm the Vermutung, dass Evai schwanger ist. supposition that Eva pregnant is ‘She supposes, and I confirm the supposition that Eva is pregnant.’ Similarly, movement does not seem to be involved in (20) either. The target in these examples contains a negative polarity item (jemals ‘ever’, and auch nur ‘not even’) which is licensed by the negative verb bezweifeln (‘to doubt’) in (20a) and the negative DP keiner meiner Studenten (‘none of my students’) in (20b). Negative polarity items must be c-commanded by a negative element. If the targets in (20) move, they raise out of the scope of the negative elements which should yield an ungrammatical result. However, the examples are fine. (20) a.
Ich bezweifle, dass Hans jemals in Indien war, und Ute glaubt that H. ever in India was and U. believes I doubt auch nicht, dass Hans jemals in Indien war. also not that H. ever in India was ‘I doubt that Hans ever went to India, and Ute doesn’t believe it either.’
Katharina Hartmann
b. Keiner meiner Studenten liest auch nur...., und keiner aus none my students reads even only and nobody from Karls Seminar versteht auch nur einen einzigen Aufsatz. K.’s class understands even only one single article ‘None of my students reads and no-one from Karl’s seminar understands not even a single article.’ Finally, a quantifier may bind a pronoun in a target, which again is only possible if the target is not attached higher in the tree than the quantifier, given that the quantifier must c-command the pronoun in order to bind it. (21) Viele Leutei denken, dass siei gesünder leben sollten, und nur that they healthier live should and only many people think wenigei bezweifeln, dass siei gesünder leben sollten. few doubt that they healthier live should ‘Many people think and only few people doubt that they should live more healthily.’ All these examples show that the targets behave as if they had not moved, which at first glance is problematic for the Movement Theory. The data, however, are not really decisive. D-structure interpretation is not in general incompatible with a movement account, as the interpretational effects can be obtained assuming that RNR – being A’-movement – is reconstructed at LF. The (un)grammaticality of the examples discussed in this subsection is therefore independent of the question of whether or not movement has taken place. Given the possibility of reconstruction, the data can be explained under both accounts of RNR, the Movement Theory and the In Situ Theory. In order to actually choose between these two accounts, one has to look at sentences where the alleged movement should have given an ungrammatical result. Such examples involve strong crossover. Following Postal (1971), Wasow (1979), Chomsky (1977a), Chomsky (1981) and Chomsky (1986b) a wh-trace must be A-free in the domain of its operator, i.e. it may not be coindexed with a constituent in A-position which intervenes between the trace and its antecedent. Hence, the ungrammaticality of (22) is due to a violation of the strong crossover condition. (Movement indices are numerical, coindexing is expressed by alphabetical indices.) (22) *Who1 did [IP hei say [CP t1 that Mary met t1/i ]]? The RNR example in (23), which is analyzed according to the Movement Theory, exactly exhibits the problematic crossover structure: the subjects of the conjuncts hierarchically intervene between the wh-traces and their operators, therefore, the traces are not A-free. (23) should also represent a strong crossover violation, but the example is grammatical.
Background matching in right node raising constructions
(23) Mariai behauptet, Carlo liebe t1 , und Utaj behauptet, Roberto M. claims C. R. loves and U. claims hasse t1 sie1/i/j . hates her ‘Maria claims that Carlo loves her, and Uta claims that Roberto hates her.’ The example can be naturally explained given the In Situ Theory. If the pronouns are not moved to begin with, a potential crossover violation does not arise. To sum up Section 2, I provided ample evidence that the target of RNR remains in situ. RNR does not show any of the typical properties of movement. It is neither conditioned by the relevant constraints on leftward movement nor can it be compared to rightward movement. Therefore the Movement Theory has proven to be inferior to the In Situ Theory of RNR. In the following sections, I develop an account of RNR which is in accordance with the In Situ Theory.
. Contrast A first property of the focus structure of RNR is that the focused part in one conjunct must semantically contrast with the focused counterpart in the other conjunct. If the focused parts do not contrast, RNR is not possible. As illustrated by the following examples, contrast between the focused parts is impossible if these parts are either identical, or too similar to each other. The relevant focused parts are printed in bold face, the pitch accent, which is the prosodic realization of semantic contrast, is indicated by capital letters of the syllable associated with this accent. (24) a. *Peter P. b. *Jonas J. Brief. letter
verkaufte Utas Auto, und sein Bruder kaufte Utas Auto. sold U.’s car and his brother bought U.’s car SCHICKte einen Brief und Claus verSCHICKte einen a letter and C. PRT-sent a sent
In (24a) the focused constituents are identical and therefore do not contrast at all. In (24b), contrast between the verbs schickte (‘sent’) and verschickte (‘PRT-sent’) is semantically not strong enough. If the examples are slightly changed such that the constituents contrast, they become grammatical immediately.6 (25) a.
Peter verkaufte Utas Auto, und sein Bruder kaufte CLAUdias P. sold U.’s car and his brother bought C.’s Auto. car ‘Peter sold Uta’s, and his brother bought Claudia’s car.’
Katharina Hartmann
b. Jonas SCHICKte einen Brief, und Claus erHIELT einen Brief. Jonas sent a letter and Claus received a letter ‘Jonas sent, and Claus received a letter.’ In the following examples, the focus constituents are not identical, but still, they cannot contrast. The reason for the ungrammaticality of these examples is that the focus constituents in the two conjuncts belong to different semantic fields and therefore are not contrastable. The indefinite article in (26a) does not contrast with a proper name. In (26b), the adjectives verschiedene (‘different’) and juristische (‘law’) are equally too different to stay in a relation of contrast. (26) a. *Olaf wollte immer schon EInen Eierkocher, aber egg.boiler but O. wanted always already a gab ich ihm MARthas Eierkocher. gave I him M.’s egg.boiler b. *Peter las verSCHIEdene Zeitschriften, und Martin journals and Martin P. read different juRIstische Zeitschriften. law journals
schließlich finally
las read
Again, slight changes in the examples in (26), such that the focus constituents are potentially contrastive, make the RNR constructions grammatical. As (27a) illustrates, two possessive pronouns are potentially contrastive and can therefore be focused in RNR. (The example in (27a) is a slightly modified translation of an English example from Swingle (1993).) Example (27b) shows that two different types of journals also represent possible contrastive pairs. Olaf wollte immer schon MEInen Eierkocher, aber egg.boiler but Olaf wanted always already my schließlich gab ich ihm DEInen Eierkocher. finally gave I him your egg.boiler ‘Olaf always wanted to have my egg boiler, but finally I gave him you egg boiler.’ b. Peter las bioLOgische Zeitschriften und Martin las journals and Martin read P. read biological juRIstische Zeitschriften. law journals ‘Peter read biological, and Martin read law journals.’
(27) a.
To summarize, the focused parts in one conjunct and the focused counterparts of the other conjunct of a RNR construction must stand in a relation of semantic contrast.7 If contrast is not possible either because the focused parts and counterparts
Background matching in right node raising constructions
are too similar to each other or because they are not of the same semantic kind, RNR is ungrammatical.
. Theoretical background The In Situ Theory claims that RNR is derived by phonological deletion without any modification of the syntactic structure. However, although deletion takes place at PF, the principles which license it are not exclusively phonological. The conjuncts of a RNR construction have to appear in a special format in order to allow phonetic deletion and this format is constrained by pragmatic principles.8 Ideally, these principles are not specific to RNR (and other elliptical coordinations) but generally determine the pragmatic wellformedness of utterances. Such a theory of pragmatic wellformedness is Schwarzschild’s theory of givenness which I adopt in this article. Following Schwarzschild (1999), any utterance must be anchored in the discourse which is achieved by the principle of Pragmatic Licensing. (28) Pragmatic Licensing An utterance U is pragmatically licensed if U has a background match, i.e. if it is given. The following section spells out the ingredients of (28).
. Focus assignment and focus percolation Schwarzschild’s system is based on Selkirk’s rules of focus assignment.9 Selkirk (1995) assumes that pitch accents are freely assigned. An accent on a word licenses a focus (F-) feature on this word10 – this is the content of the Basic Focus Rule in (29a). F-marking is sensitive to syntactic structure in the following way: F-features may project and thereby F-mark larger constituents. Focus projection proceeds as indicated in (29b): an F-feature may project from a head to its maximal projection. If this is an internal argument, the F-feature may further project to the selecting head. (29) F-Assignment Rules (Selkirk 1995: 555) a. Basic Focus Rule An accented word is F-marked. b. Focus Projection (i) F-marking of the head of a phrase licenses the F-marking of the phrase. (ii) F-marking of an internal argument of a head licenses the Fmarking of the head.
Katharina Hartmann
Focus projection is illustrated in example (30). The nominal head of the argument of the preposition SOfa is accented (the accented syllable is printed in capital letters) and, by the Basic Focus Rule, receives an F-feature. This F-feature projects according to the rules of Focus Projection: it projects from N0 to NP, from NP to the selecting D0 , from D0 to DP, and further up the sentential projection line to I’. (30) Where are the cats? [IP They [I’ areF [PP onF [DP theF SOfaF ]F ]F ]F ]. Having established how F-features are assigned, I turn to the interpretation of Ffeatures in the next subsection.
. The interpretation of F-features The theory of Schwarzschild (1999) offers an account how to interpret Selkirk’s F-features: they are interpreted not at all. What is interpreted instead are the nonF-marked constituents. According to Schwarzschild, a non-F-marked constituent must have a salient antecedent in the linguistic context of the clause the constituent occurs in, or it must be derived by this context. The technical term for a non-Fmarked constituent c to be present in the linguistic context is that c is given. This is expressed in the Rule of Interpretation: (31) Rule of Interpretation Non-F-marked constituents are given.
(Schwarzschild 1999)
For illustration, consider example (32). The subject carries a pitch accent and is F-marked. The predicate, on the other hand, is not F-marked and, in accordance with (31) must be interpreted as given. Due to the fact that it is mentioned in the linguistic context provided by the wh-question, it is given indeed. (32) Q32 : Who finished the beer? [IP PEterF finished the beer]. As the subject Peter is a specifier, the F-feature cannot project to the root node IP which is therefore not F-marked. But is IP given? Although IP is not F-marked itself, it contains an F-marked constituent, namely the subject. The definition of givenness obviously has to make reference to such cases. Schwarzschild assumes that an F-marked constituent equally to or contained in a constituent c is replaced by a variable and that this variable is existentially quantified over. He calls this the focus closure of c. The focus closure of the IP in (32) is given in (33). The focused subject Peter is replaced by a variable, and the result is existentially closed. According to the Rule of Interpretation in (31) and the (non-final) definition of given
Background matching in right node raising constructions
in (34), FC(IP), i.e. the focus closure of IP, is entailed by Q32 , the wh-question preceding IP in (32). Therefore, this IP is given. (33) FC(IP) = ∃x.finished’(the’ beer’)(x) (34) given (non-final version) An utterance U counts as given iff it has a salient antecedent A and A entails the Focus-Closure of U. As entailment relations are relations between propositions, non-propositional expressions are existentially type-shifted to propositions by adding variables to them until the expressions are propositional. The result is existentially closed. This extension is integrated into the definition of given, my final version of which is presented in (35). (35) given (Schwarzschild 1999: 149) An utterance U counts as given iff it has a salient antecedent A and modulo ∃-type shifting, A entails the Existential-Focus-Closure of U. The theory of givenness as it stands is illustrated by the example in (36). (36) Q36 : Where are our cats? They are [PP onF [DP theF SOfaF ]F ]F . a. [DP theF SOfaF ]F given because Q entails the existential closure of the focus closure of DP, i.e. EC(FC(DP)), i.e. ∃Y∃x.Y(x) b. [PP onF [DP theF SOfaF ]F ]F given because Q entails EC(FC(PP)), i.e. ∃Y∃x.Y(x) c. [I’ are [PP onF [theF SOfaF ]F ]F ] given because Q entails EC(FC(I’)), i.e. ∃y∃x.are’(y)(x) d. [IP They are [onF [theF SOfaF ]F ]F ] given because Q entails FC(IP), i.e. ∃x.are’(x)(they’) Let us walk through this example step by step. The nominal head of the prepositional complement SOfa has a pitch accent and, by the Basic Focus Rule in (29a), is assigned an F-feature which projects to the PP F-marking NP, D0 , DP and P0 on its way. We start by considering the DP complement of the preposition ((36a)). The DP is F-marked by focus projection. Hence, in order to determine if this DP is given, it is replaced by an existentially bound variable and the result (∃x.x) is existentially type-shifted to a proposition along the lines given above. The result of this, i.e. ∃Y∃x.Y(x), is entailed by the wh-question Q36 . Therefore, DP is given. The PP is equally given (cf. (36b)). As the PP is F-marked, the existentially type-shifted focus closure of PP (again ∃Y∃x.Y(x)) is entailed by Q36 . Notice that at this point the F-features embedded in the PP are disregarded. Let’s consider I’. After replac-
Katharina Hartmann
ing the F-marked PP by a variable and existentially quantifying over it, the result denotes a relation of being somewhere. The existentially type-shifted expression is entailed by Q36 , since it is entailed that somebody is somewhere, hence I’ is given. Finally, IP is also entailed by Q36 since the question presupposes that there must be a place where the cats are, therefore IP is also given. According to the principle of Pragmatic Licensing (28), the utterance in (36) is pragmatically well-formed. F-marking as in (36) is the only possibility given the indicated accent pattern. I sketch some ungrammatical alternatives which are all incompatible with the principle of Pragmatic Licensing. First imagine that the F-feature assigned by the Basic Focus Rule to the accented SOfa does not project at all ((37a)), or not up to the PP ((37b)). The utterance would not be entailed by the question Q36 (Where are our cats?) any longer and consequently not be pragmatically well-formed. (37) a.
[IP They are on the SOfaF ]. not given because Q does not entail ∃x.are’(on’ the’ x)(they’) b. [IP They are on [DP theF SOfaF ]F ]. not given because Q does not entail ∃x.are’(on’ x)(they’)
While focus does not project at all in (37a) or not far enough as in (37b), it projects beyond the PP in (38). The theory of givenness as it stands does not restrict focus from always projecting up to the highest node of the projection line of the focused constituent, which is an intuitively unwarranted theoretical stance. In order to remedy this problem, Schwarzschild (1999: 156) proposes the economy constraint AvoidF which says that F-projection is allowed only as far as necessary for the Rule of Interpretation. In view of the constraint AvoidF, (38a) and (38b) are not well-formed because, although their focus closure is given, the representations contain more F-features than the equally given (36). They [areF [onF [theF SOfaF ]F ]F ]F . given because Q’ entails ∃Y.Y(they’) but ruled out by AvoidF b. [They [areF [onF [theF SOfaF ]F ]F ]F ]F . given because Q’ entails ∃Y.Y but ruled out by AvoidF
(38) a.
To summarize, the last two subsections discussed some portions of the complicated interaction between prosodic structure and pragmatic well-formedness. I used the theories of focus assignment and focus interpretation presented in Selkirk (1995) and Schwarzschild (1999) to illustrate the connection between the two systems intonation on the one hand, and pragmatics on the other: The (freely assigned) pitch accents introduce F-features into the syntactic representation which potentially project. These F-features were shown to play an important role when matching an utterance against the linguistic background: non-F-marked constituents have
Background matching in right node raising constructions
to be given in Schwarzschild’s sense, i.e. they need an appropriate antecedent in the linguistic context. If the background match turns out to be successful, in other words, if a matched utterance is given, the prosodic structure associated with this utterance is well-formed. Otherwise, the pitch accents must be differently assigned. The antecedent against which utterances were matched in this subsection was always provided by a wh-question. To give wh-questions as contexts is a very common strategy to determine the focus structure of a subsequent utterance. However, it is not the only possible strategy. In the next section, I will discuss a variety of examples where an appropriate antecedent for the determination of givenness of an utterance is either provided by some non-interrogative element in the context, or by a part of the utterance itself. It is this last strategy which will become relevant when finally coming back to RNR in Section 6.
. Context extensions Appropriate antecedents for the determination of givenness of an utterance U do neither have to immediately precede U nor do they necessarily have to be external to U. Constituents within U can also function as antecedents for other constituents contained in U. In this section, I discuss three cases which document the availability of such alternative antecedents. Consider first the example in (39) which is taken from Schwarzschild (1999). (39) John drove Mary’s red convertible. What did he drive before that? a. He drove her BLUE MerCEdes. b. He drove her BLUE convertible. Both, (39a) and (39b) are felicitous continuations of the preceding context. The difference between them is that in (39a), the head noun contained in the object DP as well as the modifying adjective carry accents, whereas in the latter it is only the adjective. The last-mentioned case involves deaccenting of the NP convertible. Very informally, the term deaccenting describes a state of affairs where a constituent which would be accented under regular circumstances is disposed of its accent for some reason (cf. Ladd 1978). Deaccenting of the NP convertible in (39b) is unexpected regarding the context provided by the wh-question. The wh-word corresponds to a DP which requires the nominal head to be accented.11 However, deaccenting follows naturally if the sentence preceding the question is also taken into consideration in order to evaluate givenness, because convertible is mentioned there. Hence, FC(39b), i.e. ∃x[drove(her x convertible)(he)] is actually given by the extended context preceding (39b).
Katharina Hartmann
A similar example is provided in Selkirk (1995). Taken the embedded whquestion (Guess) what Judy did? in isolation, the only possible intonational contour for an answer would be (40b), not (40a). Note that F-marking the accented nominal head book as in (40a) does not license an F-mark on the more deeply embedded PP which therefore must be interpreted as given. While the bare wh-question (Q) is not an appropriate antecedent for the object DP in (40a), the preceding temporal adjunct clause (T) is: it licenses deaccenting of the given PP about bats as illustrated in (41). (40) After all this talking about bats, guess what Judith did? a. Judith bought a BOOK about bats. b. *Judith bought a book about BATS. (41) [IP Judith [VP boughtF [DP aF BOOKF [PP about bats]]F ]F ]. a. EC(FC(PP)) = ∃P.P(about-bats’) b. EC(FC(DP)) = ∃P∃Y∃x.Y(x) & about-bats’(x) & P(x) c. EC(FC(VP)) = ∃x∃P.P(x) d. FC(IP) = ∃P.P(Judith’)
given by T given by T given by Q given by Q
While T is responsible for the evaluation of givenness of the PP, the context provided by Q becomes relevant at the VP level. The example in (41) illustrates how different parts of a linguistic context determine the accentuation pattern of an utterance. Antecedents which determine the intonational contour of an utterance or parts of it do not necessarily have to precede this utterance. Entailment relations are also possible between elements of one and the same utterance.12 This is shown in example (42) which is originally due to Rooth (1992b). (42) Q42 : Who was beating whom? a. [DP1 AF BELgianF FANF ]F was beating [DP2 anF ENGlishF HOOliganF ]F . b. [DP1 A BELgianF fan] was beating [DP2 an ENGlishF fan]. c. *[DP1 AF BELgianF FANF ]F was beating [DP2 anF ENGlishF FANF ]F . (42a) and (42b) are both natural answers to the question Q42 . In (42a), the subject and object DPs are focused (via the multiple wh-question). The metrically most prominent syllable of the NP-heads are associated with the pitch accents and Fmarked. The F-features project across the D0 –heads to the DPs. The prenominal adjectives are excluded from being F-marked by projection and receive an F-mark only by secondary stress. At first glance things are not as straightforward in (42b), which differs from (42a) in that the subject and object NP-arguments are identical. In this case, only the adjectives carry a pitch accent. Notice that the F-marks on these modifiers cannot project. An accentuation pattern for (42b) according to the one in (42a) yields an intonationally marked sentence, cf. (42c). The interest-
Background matching in right node raising constructions
ing question is: Is the F-marking in (42b) licensed? Obviously, Q42 alone is not a suitable context. But in contrast to the examples (39) and (40) above, the context is not further specified, i.e. the wh-question is the only available context here. One therefore has to allow an extension of the notion of context to the utterance itself. Thus, parts of the utterance constitute “backgrounds” for other parts of the same utterance. This is illustrated for (42b) in (43).13 (43) a.
[DP1 a BELgianF fan] given because EC(DP2 )entails EC(FC([DP1 a BELgianF fan])), i.e. ∃P∃x.fan’(x) & English’(x) & P(x) entails ∃P∃Q∃x.fan’(x) & Q(x) & P(x) b. [DP2 an ENGlishF fan] given because EC(DP1 )entails EC(FC([DP2 an ENGlishF fan])), i.e. ∃P∃x.fan’(x) & Belgian’(x) & P(x) entails ∃P∃Q∃x.fan’(x) & Q(x) & P(x)
The accentuation patterns of the DPs in (42c) are excluded. They violate the constraint AvoidF because the DPs contain more F-features than the DPs in (42b) given the same interpretation. This is illustrated for DP1 in (44). (44) [DP1 aF BELgianF FANF ]F given because ∃P∃x.fan’(x) & English’(x) & P(x) entails ∃P∃Q∃x∃y[y(x) & Q(x) & P(x)], but ruled out by AvoidF Having established that linguistic contexts which license the accentuation pattern of an utterance U are not restricted to wh-questions preceding U, but may either consist of bigger chunks of linguistic material, or even of parts of U itself, we are now ready to turn back to right node raising. As it will turn out, sentence internal contexts will be essential in order to license the accentuation pattern of RNR, which is taken to be responsible for ellipsis in these constructions.
. Background matching in right node raising Section 2 adduced evidence that the derivation of RNR constructions does not involve syntactic movement. It was proposed instead that RNR is derived by a process of phonetic deletion of the target in the first conjunct. The aim of this section is to work out this proposal. The process of phonetic deletion in RNR is constrained by pragmatic as well as by prosodic conditions, in particular by the characteristic intonational contour of RNR constructions. I analyze the typical accent pattern of RNR and its consequences for phonetic deletion in this section. I do not discuss other essential factors as the tones which are associated with the words immediately preceding the targets, the intonational break which occurs after the first conjunct, and the tonal reset at the beginning of the second, but see Hartmann (2000) and espe-
Katharina Hartmann
cially Féry and Hartmann (2001), who claim that the strong L*H tone obligatorily associated with the element immediately preceding the target phonetically licenses PF-deletion. Concretely, it will be argued in this section that the accent pattern typically associated with RNR triggers a distribution of F-features which makes RNR constructions fulfill the principle of Pragmatic Licensing introduced in Section 2 and repeated in (45). (45) Pragmatic Licensing An utterance U is pragmatically licensed if U has a background match, i.e. if it is given. String identity at the right periphery of the conjuncts as an obligatory precondition for RNR to take place will follow from F-feature interpretation as outlined in the above paragraphs: As we have shown, entailment relations which determine whether a constituent c is given may exist not only between c and an expression external to the utterance containing c but also between c and another expression contained within the same utterance. Such an extension of the notion of discourse antecedent to sentence internal expressions also accounts for the distribution of F-features in RNR. After F-assignment and F-projection, one conjunct of a RNR construction may serve as a discourse antecedent for the other conjunct for establishing the entailment relations which determine givenness. Due to this fact, an overlap of phonetically identical material in the conjuncts becomes unavoidable. Phonetic deletion in RNR, which is optional, is then characterized as an economizing strategy. It is triggered by the tonal facts, and, indirectly, by the distribution of F-features.14 This section is structured as follows. Subsection 6.1 shows that RNR may involve deaccenting. The proposal that the conjuncts may function as discourse antecedents for each other will account for deaccenting in such cases. Subsection 6.2 considers the question whether targets can be identified with distinctive parts of informational structure, more concretely, with the background. In contrary to what has been proposed in the literature it is shown that no such correlation can be maintained. Due to the claim that the distribution of F-features and therefore the accent pattern is responsible for subsequent phonetic deletion, I consider the pre-deletion structures of RNR throughout this section.
. Accenting and deaccenting The intonational contour of RNR constructions is determined by the pitch accents associated with the contrasting elements in the two conjuncts. These accents may trigger deaccenting of other constituents. Deaccenting is illustrated in (46).
Background matching in right node raising constructions
In both conjuncts, the pitch accent is associated with the contrasting postpositions hiNUNter and hiNAUF which immediately precede the target. (46) Q46 : Where did they go? Ich glaube, dass Hans [PP die Straße hiNUNter]F gegangen ist und the street down I believe that H. walked is and Klaus [PP die Straße hiNAUF]F gegangen ist. walked is the street up K. ‘I believe that Hans walked down the street, and Klaus walked up the street.’ We may note that if a PP contains a pitch accent in a non-coordinated sentence, the accent is associated with the nominal head of the pre- or postposition’s complement. An accent pattern analogous to the RNR example in (46) is excluded in simple sentences. (47) Q47 : Where did he go?15 a. Hans ist [PP die STRAße hinunter] gegangen. b. *Hans ist [PP die Straße hiNUNter] gegangen. That is, the distribution of pitch accents within PP may vary depending on whether the PP is contained in a single clause or in a RNR example where it precedes the target: in single clauses, the nominal head carries the main stress whereas in RNR it is the postpositional head. This surprising fact follows naturally from the theory of givenness as outlined in Section 4. Consider F-marking of the respective PPs and the projection of the F-feature. (48) a. [PP [DP dieF STRAßeF ]F hinunterF ]F b. [PP [DP die Straße] hiNUNterF ]F
single clause RNR
In (48a), the nominal head is accented and F-marked by the Basic Focus Rule. The F-feature projects across NP, D0 and DP to P0 and finally PP F-marking all lexical elements on its way. Thus, (48a) is given because ∃X∃Y.X(Y), the existential focus closure of (48a), is trivially given by Q47 . In (48b), on the other hand, the F-feature projects from the accented postposition to its maximal projection such that the DP does not receive an F-feature by projection. The existential-Fclosure of (48b) is ∃X∃Y.X(Y(die Straße)), which is not given by Q47 . There is, however, a context which is appropriate for matching the PP in (48b) – call it PP1 – against, namely the PP contained in the other conjunct of the RNR construction – PP2 – [PP2 die Straße hiNAUF]. This PP2 entails EC(FC([PP1 die Straße hiNUNter])), which is ∃X∃Y.X(Y(die Straße)). Therefore, EC(FC(PP1 )) is given by PP2 in the subsequent conjunct. Notice that the entailment relations are mu-
Katharina Hartmann
tual in RNR. That is, PP1 entails the existential-F-closure of PP2 . The licensing mechanisms are summarized in (49). (49) a.
[PP1 die Straße hiNUNter] is given because [PP2 die Straße hiNAUF] entails EC(FC(PP1 )), i.e. ∃X∃Y.X(Y(die Straße)) b. [PP2 die Straße hiNAUF] is given because [PP1 die Straße hiNUNter] entails EC(FC(PP2 )), i.e. ∃X∃Y.X(Y(die Straße))
To summarize, accenting the prepositions and deaccenting the DP-complement in the RNR example (46) leads to a distribution of F-features such that the linguistic context provided by Q46 does not entail the existential F-closure neither of the PPs, nor of course of the conjuncts containing them. However, the internal structure of a RNR construction is such that the existential F-closure of a deaccented constituent is entailed by the respective parallel constituent of the other conjunct. The following examples further corroborate the claim that the accent pattern typically found with RNR involves deaccenting. In (50), the pitch accent is again associated with two prepositions, neben (‘beside’) and auf (‘on’), and not with the nominal head of their complements, as required in uncoordinated sentences if PP is focused. In (50), the verbal stems are deaccented and the pitch accents are realized on the verbal particles. (50) Q50 : Where are our two cats? [CP1 Halma liegt [PP NEbenF dem Sofa]] und [CP2 Mikado liegt lies beside the sofa and M. lies H. [PP AUFF dem Sofa]]. on the sofa ‘Halma lies beside the sofa and Mikado lies on the sofa.’ (51) Q51 : How do you think they estimate the dollar? Ich glaube, dass [IP1 Anna den Dollar ÜBERschätzt] und I think that A. the dollar overestimates and [IP2 Ute den Dollar UNterschätzt]. U. the dollar underestimates ‘I think that Anna overestimates the dollar and Ute underestimates the dollar.’ In none of the two examples (50) and (51) does the question context alone license their intonational contours. The sentence internal context is necessary in order to license the distribution of F-features resulting from F-marking the accented words and F-projection.
Background matching in right node raising constructions
. The target of RNR Hartmann (2000) and Féry and Hartmann (2001) observe the following typical characteristics of the intonational contour of German RNR constructions: Firstly, the words immediately preceding the targets in the two conjuncts are associated with the pitch accents. The accent is a L*H %H tone in the first conjunct, and a H*L %L in the second.16 Secondly, the target, which is phonetically deleted in the first conjunct, very frequently has a flat intonational contour in the second conjunct where it is deaccented. An assumption which immediately suggests itself is that the target corresponds to the informational background. This hypothesis seems to be corroborated by the following example. (52) Whom did they buy a beer? Ute hat HANSF ein Bier gekauft, und Klaus hat PEterF ein Bier a beer a beer bought and K. has P. U. has H. gekauft. bought ‘Ute bought Hans, and Klaus bought Peter a beer.’ It is a common strategy to separate the focus from the background in a sentence S by identifying the focus of S with the wh-phrase of a preceding wh-question. This strategy is called question-answer-congruence. According to this, the target in (52) indeed corresponds to the background. Given that backgrounds are in general not associated with pitch accents, it naturally follows that the target of RNR has a flat intonational contour. However, it will turn out immediately that (52) only represents a special case of a much broader picture. As the data below prove, the identification of the target with the informational background cannot be maintained. Targets may constitute a part of the focus ((53)), or they may even be the focus themselves ((54)). In (53), the focus constituents are DP1 ein großes Stück Kuchen in the first conjunct, and DP2 ein kleines Stück Kuchen in the second. The NP Stück Kuchen is the target. (53) Q53 : What did they buy for Hans and Peter? [CP1 Maria kaufte Hans [DP1 ein GROßes Stück Kuchen]], a big M. bought H. piece.of cake und [CP2 Klaus kaufte Peter [DP2 ein KLEInes Stück Kuchen]]. and K. bought P. a small piece.of cake ‘Maria bought a big piece of cake for Hans, and Klaus bought a small piece of cake for Peter.’ In (54), a RNR construction is an answer to a multiple wh-question. The target is one of the focus constituents, it is the direct object, which is questioned. Being a
Katharina Hartmann
focus constituent, the target contains a pitch accent which is associated with the focus exponent, i.e. the nominal head of the PP nach Berlin.17 (54) Q54 : What did Paul and Erika promise to whom? Paul versprach LIsa [nach Berlin zu kommen], und Erika versprach P. promised L. to Berlin to come and E. promised HANS, [nach BerLIN zu kommen]. H. to Berlin to come ‘Paul promised Lisa to come to Berlin and Erika promised Hans to come to Berlin.’ Thus, these examples show that targets can be a focus constituent or be contained in a focus constituent. Therefore, the hypothesis that targets correspond to backgrounds has to be dismissed. A theory which captures the data correctly is the theory of givenness as developed in Sections 4 and 5. I demonstrate below that the distribution of pitch accents found (53) and (54) yields the only accent pattern which satisfies the principle of Pragmatic Licensing introduced in Section 4. Furthermore, I claim that there is no correlation whatsoever between targets and informational status, hence, the licensing conditions of phonetic deletion have to be sought elsewhere. Consider (53) again. Remember that the Basic Focus Rule F-marks words which carry a pitch accent, i.e. the adjectives GROßes and KLEInes in the respective conjuncts. Given that the adjectives are modifiers and therefore not selected, the F-features do not project. Thus, we observe a discrepancy between the F-marking which follows from Selkirk’s rules of F-marking and question-answercongruence, which predicts a correlation between wh-phrases of questions and focus constituents in the corresponding answer: While Selkirk’s rules yield the adjectives GROßes and KLEInes as the only F-marked constituents, the focus as expected from the question is bigger, namely the whole direct object DPs. (55) a. focus constituents in (53) according to Selkirk 1995: GROßes, KLEInes b. focus constituents in (53) according to question-answer-congruence: [DP ein GROßes Stück Kuchen], [DP ein KLEInes Stück Kuchen] The two approaches can be reconciled in the givenness approach outlined in Section 3. Selkirk’s mechanism of F-assignment is the basis for establishing the entailment relations between an expression and an appropriate discourse antecedent. Such entailment relations were claimed to be necessary to determine if an expression is given. Question-answer-congruence, on the other hand, is absorbed into the theory of givenness. The focus of a sentence, i.e. the constituent which is marked with the highest F-feature is subject to the same requirement as any other constituent in an utterance: it must be given in order to fulfill the principle of Prag-
Background matching in right node raising constructions
matic Licensing. According to the modifications elaborated in Section 4, which extend the definition of discourse antecedent of a constituent c to linguistic contexts which are part of the utterance containing c, (53) fulfills the principle of Pragmatic Licensing. This is illustrated in (56). [DP1 ein GROßesF Stück Kuchen] is given because DP2 entails EC(FC(DP1 )), i.e. ∃x∃Y∃P.Stück-Kuchen(x) & Y(x) & P(x) [DP2 ein KLEInesF Stück Kuchen] is given because DP1 entails EC(FC(DP2 )), i.e. ∃x∃Y∃P.Stück-Kuchen(x) & Y(x) & P(x) b. [CP1 Maria kaufte Hans [DP1 ein GROßesF Stück Kuchen]] is given because DP2 and Q together entail FC(CP1 )), i.e. ∃x∃Y.Stück-Kuchen(x) & Y(x) & kaufte(x)(Hans)(Maria)
(56) a.
Thus, the F-closure of the first conjunct is entailed by the context provided by the question in conjunction with the context provided by the target of the second conjunct. Note that Q53 alone does not entail the F-closure of CP1 because the question does not say anything about cakes. The F-closure of CP2 is entailed similarly, i.e. again by Q53 and the target of the first conjunct prior to phonetic deletion. The accent pattern in (53) is the only well-formed pattern for this sentence. Two examples of ungrammatical accent patterns are given in (57). (57a) does not contain any pitch accent at all. This example is excluded because neither the existential F-closure DP1 , ∃x∃P.Stück-Kuchen(x) & groß(x) & P(x), nor the existential F-closure of DP2 are entailed. Similarly, if the indirect objects are pitch accented (cf. (57b)), the existential F-closures of DP1 and DP2 are not entailed. (Moreover, the F-feature on the indirect object violates AvoidF: as the indirect objects are mentioned in Q53 , the F-feature is not required.) (57) a. *[CP1 Maria kaufte Hans [DP1 ein großes Stück Kuchen]] und [CP2 Klaus kaufte Peter [DP2 ein kleines Stück Kuchen]]. b. *[CP1 Maria kaufte HANSF [DP1 ein großes Stück Kuchen]] und [CP2 Klaus kaufte PEterF [DP2 ein kleines Stück Kuchen]]. That targets can also contain pitch accents is shown in (53), provided with Ffeatures in (58) below. This is a multiple focus construction where one focus constituent comprises the target. (58) Q58 : What did Paul and Erika promise to whom? Paul versprach LIsaF [nach Berlin zu kommen], und Erika versprach P. promised L. to Berlin to come and E. promised HANSF, [[PP nachF BerLINF ]F zuF kommenF ]F . to Berlin to come H. ‘Paul promised Lisa to come to Berlin and Erika promised Hans to come to Berlin.’
Katharina Hartmann
It is obvious that the wh-question entails the F-closure of the two conjuncts. Nevertheless, two rather speculative remarks are in order here: Firstly, the theory of givenness with the extension proposed in this paper predicts the pitch accent on BerLIN to be superfluous. The reason is that (the existential F-closure of) a target is always entailed by the presence of the other conjunct. F-marking, and consequently also pitch accents within a target should therefore be ruled out by the economy constraint AvoidF. This shows the necessity of an additional requirement which states that a constituent which correspond to a wh-phrase in a preceding question must contain a pitch accent, i.e. it may not be fully deaccented. Secondly, while phonetic deletion is optional if the target in the second conjunct is deaccented (the target in the first conjunct can be deaccented, too), it seems obligatory if the target contains an accent. I cannot offer an explanation for this fact.
. Conclusion The aim of this paper was to investigate the distribution of accents and the consequences this distribution has on the focus structure of RNR constructions as well as on the possibility to phonetically delete the target in the first conjunct. I started by providing ample evidence for the claim that RNR cannot be syntactically derived. In a nutshell, I showed that the target of RNR does not have to be a syntactic constituent, that RNR can strand prepositions, that it can violate islands and that the target can be a DP, a constituent which usually does not extrapose in noncoordinated clauses. Furthermore, the target behaves as if not moved with respect to several scope phenomena. I went on to show that the elements immediately preceding the targets in the first and second conjunct must contrast. They carry a pitch accent and must be semantically licensed such that contrast is possible. If these elements are either too similar, or too different, they cannot contrast, and RNR is ruled out. The 4th and 5th section introduced the theoretical background of the In Situ Theory of RNR. Section 4 presented a summary of Schwarzschild’s theory of givenness, and Section 5 extended the notion of discourse antecedents which are necessary in order to establish entailment relations to linguistic material which is contained in the utterance also containing the constituent to be matched. Finally, Section 6 analyzed the focus structure of RNR. I showed that the targets of a RNR construction mutually serve as discourse antecedents for the other conjunct, a fact which licenses deaccenting in RNR. Furthermore, I dismissed the hypothesis that the target corresponds to the informational background proving that targets can constitute a part of the focus, or be the focus themselves. The data were successfully reconciled under the givenness approach. Thus, the distribution of accents as
Background matching in right node raising constructions
well as the phonetic identity of the targets were shown to be obligatory conditions to fulfill the Principle of Pragmatic Licensing.
Notes * For comments and suggestions on various stages of this article, I thank Kerstin Schwabe and Susanne Winkler, the editors of this volume, as well as Chris Kennedy. . The shared element can also consist of just the verb, cf. (i). In such a case, RNR appears to be a reflection of gapping. In gapping constructions, the finite verb (and optionally additional constituents) are omitted in the second conjunct, cf. (ii). (i) (Ich (I (ii) (Ich (I
möchte,) want), möchte), want),
dass that dass that
Hans Hans Hans Hans
die the die the
Nüsse nuts Nüsse nuts
und Peter die Mandeln knackt. and Peter the almonds cracks knackt und Peter die Mandeln. cracks and Peter the almonds
Notice, however, that the similarity between (i) and (ii) is a mere coincidence since RNR and Gapping differ in many respects (for a comparison of the different syntactic and prosodic properties of the two constructions, cf. Hartmann (2000) and Féry & Hartmann (2001)). . From now on, I refer to the elided part in the first conjunct and its overt counterpart in the second conjunct as the “target” of RNR. . That RNR targets do not have to be constituents in German is shown in Wesche (1992) and Wilder (1997b). The same claim can be made for English as shown in Gleitman (1965), Abbott (1976), Grosu (1976), Williams (1978), van Oirsouw (1983), van Oirsouw (1987) and Larson (1990). . RNR can also involve only words, and even smaller units like morphemes, cf. (i) (from Booij 1985, cf. also Wiese 1993) and (ii). These examples illustrate that RNR may also apply below the X0 –level. Assuming the Principle of Lexical Integrity (DiSciullo & Williams 1987; Lapointe 1980), syntax cannot refer to parts of words, therefore (i) and (ii) cannot be derived by the Movement Theory either. (i) Frühlingsblumen und Herbstblumen springtime.flowers and autumn.flowers ‘springtime and autumn flowers’ (ii) beladen und entladen, überschätzen und unterschätzen load and unload, overestimate and underestimate ‘load and unload, overestimate and underestimate’ . Chris Kennedy (in his review of this paper) points out that the ungrammaticality of (19) is unexpected given that names contained in an ellipsis may undergo vehicle change Fiengo and May (1994). Vehicle change from names to pronouns is illustrated in the following example involving VP-ellipsis (from Fiengo & May 1994: 220). Notice that, in (i), the Rexpression contained in the ellipsis (John) can corefer with the pronominal subject of the second conjunct’s matrix clause (he). Such coreference, however, is excluded by the binding
Katharina Hartmann
theory, because the pronoun c-commands the R-expression after reconstructing the ellipsis (cf. (ii)). In order to derive the grammaticality of (i), the theory of vehicle change allows the R-expression to appear as a pronoun in the ellipsis, as shown in (iii). (i) Mary loves Johni , and hei thinks that Sally does, too. (ii) *Mary loves Johni , and hei thinks that Sally , too. (iii) Mary loves Johni , and hei thinks that Sally , too. Now, if vehicle change applied to (19), the R-expression in the ellipsis (Eva) could be changed to a pronoun, and no Principle C violation should occur. However, vehicle change does not seem to take place in RNR – probably a manifestation of its purely phonological nature. Consider (iv) to (vi). In none of these examples is coreference as indicated possible. (iv)*Siei kämmt und er frisiert Monikasi Haar. she combes and he does M’s hair liebe den Tratsch über [Prince Charles]i . (v) *Eri verabscheut, ich aber he detests I instead love the gossip about Prince Charles (vi)*Eri behauptete, Petra WÜrde, aber ich wusste, sie HATte Karli bereits he claimed P. would but I knew she had K. already vergessen. forgotten Observe that vehicle change should allow a pronoun in the reconstructed ellipsis site and make the coreferent reading available – contrary to fact. . It is not trivial to define what it means for a semantic contrast to be “strong enough”. The problematic nature of this notion is best illuminated by the following two remarks, both brought up to my attention by Susanne Winkler (p.c.). Firstly, example (24b) slightly improves if stress on the verb verSCHICKte, which is associated with the prosodically most prominent syllable of the word, shifts to the prefix. The reason for this is that by shifting the stress to the prefix one tries to emphasize the lexical semantic differences between the verbs SCHICKen and VERschicken. These two verbs, however, are semantically almost identical. Still, the improvement is observable. Secondly, the adverbial particle auch (‘as well’) obviously adds enough information to a verb as to make it distinct enough from its counterpart without a particle. This is illustrated in example (i) which I owe to Susanne Winkler. (i) Peter GLAUBT, dass die BSE-Krise zum Umdenken führen wird, und that the BSE-crisis to.the rethinking lead will and Peter thinks Maria glaubt AUCH, dass die BSE-Krise zum Umdenken führen wird. Maria thinks as.well that the BSE-crisis to rethinking lead will ‘Peter thinks, and Maria thinks as well that the BSE-crisis will lead people to a process of rethinking.’ Although the difference between the verbs in (i) consists only of the additional adverbial particle in the second conjunct, this difference seems to be strong enough to make the verbs semantically contrastable. . It may also happen that contrast is semantically possible but phonologically excluded. A case in point are clitics, which cannot carry an accent. Therefore, clitics are impossible RNR
Background matching in right node raising constructions
targets. Note that the ungrammaticality of (i) is not due to semantic reasons. Given that full personal pronouns may be stressed, they are consequently able to contrast (cf. (ii)). (i) *Peter hat’n und Rosa hat’se getroffen. P. has’himCL and R. has’herCL met (ii) Peter hat IHN, und Rosa hat SIE getroffen. P. has him and Rosa has her met ‘Peter met him, and Rosa met her.’ . Phonological constraints play a role as well. Thus, as Swingle (1993) observes, the size of a target is determined by prosody, i.e. it must correspond to an intonational phrase. In this article, I will not dwell on this issue. . I am basing my account on the theory of Schwarzschild (1999) as developed in Sections 1–5 of his paper. Later on, in Section 6, Schwarzschild proposes to dismiss Selkirk’s focus projection rules and proposes an Optimality Theory account instead. As the data I am analyzing here may well be couched within the combined Selkirk-Schwarzschild mechanism, I stick with this version of the theory. . The idea of such F-features goes back to Jackendoff (1972). The assumption that they are inserted into a syntactic representation suggests that their function is primarily syntactic. However, F-features differ from purely syntactic features as used in e.g. the Minimalist Program (Chomsky 1995) in that they do not take part in purely syntactic processes (they do not trigger movement, for instance). Rather, the task of F-features is to mediate between syntax, phonology and semantics. Thus, while their assignment depends on a phonetic element, the pitch accent, their interpretation is driven by semantic and pragmatic conditions such as focus-background structuring. . This can be easily verified: (i) Qi : What did Mary drive? [IP She drove [DP a conVERtible]F ]. given because Qi entails FC(IP), i.e. ∃x.drove’(x)(she’) If the head noun of the object would not be accented, the IP would not contain any F-mark at all, and Qi would not entail the FC(IP), i.e. ∃x.convertible (x) & drove (she )(x). . Deaccenting can also be licensed cataphorically. This can be observed in quite a variety of different cases. Ray Jackendoff (p.c.) for instance considers score keeping in a baseball game. He notices that the natural way to announce the score is by accenting the number of scored points, as shown in (i). (i) The Red Sox FOUR, the Yankees THREE. However, if the game is at a draw, the score numbers are deaccented, and the teams get an accent. Notice that only after expressing the score of the second team (the Yankees) does one know that the game is at a level. This, however, is an obligatory information when deaccenting the score of the first team (the Red Sox). (ii) The RED Sox four, the YANkees four.
Katharina Hartmann
Another example where the information which licenses deaccenting of an element follows this element are enumerations. Consider (iii): (iii) I talked to the following linguists at yesterday’s dinner party: Jaklin KORNfilt, Roland PFAU, Ray JACKendoff, Daniel BÜring, Ede ZIMmermann, UTpal Lahiri and ADIti Lahiri. While the main stress on a name is on the surname, if the surname of two persons is identical, it is deaccented and the first name is accented instead. Again, when the enumeration in (iii) comes to ‘Utpal Lahiri’, the necessary information. i.e. the identity of the surname in ‘Aditi Lahiri’, is not available yet, still, deaccenting as in (iii) is possible. . Such a condition is also implicit in Merchant’s (2001) definition of “e-Givenness”, which is a variation of Schwarzschild’s definition of Givenness: An expression E counts as e-given iff E has a salient antecedent A and, modulo ∃-type shifting, (i) A entails the F-closure of E, and (ii) E entails the F-closure of A. . C. Kennedy (reviewing this paper) touches on a delicate point of the analysis proposed here. He correctly observes that RNR is limited to coordinated structures and not possible in subordinations – a fact which is couched in terms of the Movement Theory, but does not follow from my analysis without further stipulations. I cannot present a satisfying solution to this problem of the In Situ Theory. But note that RNR is the only type of coordination which contains an ellipsis in the first conjunct. Therefore, resolution of the ellipsis has to take place backwards, i.e. the ellipsis site precedes its overt correlate. Apart from the striking intonational contour, the identification of the ellipsis site in a RNR construction requires syntactic parallelism. Coordination, but not subordination facilitates the backward identification of the ellipsis, which I think might be due to processing factors. . The contexts provided by the questions Q46 and Q47 do not necessarily constitute the same context. While Q46 may be understood as a hidden multiple wh-question (Who did go where?), Q47 is a simple wh-question. This could be a problem since the discussion of these sentences depends upon the assumption that the contexts are identical (I am grateful to Susanne Winkler for pointing this out to me). But even if we use the same multiple whquestion as a context for (46) and (47), the facts concerning the accentuation pattern inside PP remain the same: (i) Q: Who did go where? (ii) Ich glaube, dass HANS [PP die Straße hiNUNter]F gegangen ist und KLAUS [PP die Straße hiNAUF]F gegangen ist. (iii) HANS ist [PP die STRAße hinunter] gegangen. . There are some few exceptions to this generalization. Thus, elements which cannot bear stress can intervene between the stressed words preceding the target and the target itself. Such elements are unstressable pronouns and clitics. The examples in (i) and (ii) are from Truckenbrodt (1995). (i) Peter SOLLlte es Hans erzählen, und Maria WOLLte es Hans erzählen. tell and M. wanted it H. tell P. should it H. ‘Peter should tell it Hans and Maria wanted to tell it Hans.’
Background matching in right node raising constructions
(ii) Peter SETZte sich auf den Tisch, und Maria STELLte sich auf den Tisch. P. sat refl on the table and M. stood refl on the table ‘Peter sat and Maria stood on the table.’ (iii) Halma liegt AUF’m Sofa, und Mikado liegt UNter’m Sofa. H. lies on’det sofa and M. lies below’det sofa ‘Halma lies on the sofa and Mikado lies below the sofa.’ A precondition for the pronouns and clitics to intervene is that they are phonetically identical in both conjuncts. . A focus accent may also be realized in the target of the example (53) which is part of the focus. Here, the NP KUCHEN may carry an accent. Speaker seem to freely chose one or the other option. For further discussion, cf. Féry and Hartmann (2001).
II. From the computational system to the syntax-semantics interface
Merge copy Caterina Donati The paper explores the consequences of an approach underlining the similarities between movement and ellipsis: both are syntactic operations creating a copy which gets deleted at PF. Starting from the observation that a complete identification of the two operations is not possible given that the two contrast in a number of syntactic and semantic properties, the paper proposes to differentiate movement and ellipsis on the basis of the mechanism creating the copy in the two cases: a complex and costly operation involving feature agreement and merge copy in the case of movement; a simple and free operation of merge copy in the case of ellipsis. The second part of the paper explores some empirical and conceptual consequences of this approach.
Introduction The two major lines of analysis pursued in the literature both basically define ellipsis as an interface phenomenon: either at PF, as a deletion procedure, (since Sag 1976a), or at LF, as a reconstruction operation (since Williams 1977a).1 There is a priori a third possibility, claiming that these approaches are indeed both right, in that a PF process and an LF process are needed to account for ellipsis; but that they are also fundamentally wrong, in that the basic mechanism defining ellipsis contexts is neither phonological nor semantic, but purely syntactic. As all syntactic phenomena, by virtue of the very architecture of grammar, it gets interpreted at both interfaces, but it is not in itself an interface process. This work is a tentative exploration of this logical possibility, which appears to give interesting results, and to provide new understanding of ellipsis phenomena. Starting from the observation of an obvious though superficial similarity between movement and ellipsis, the first section proposes to derive the gap involved in both contexts from the same copy deletion rule applying at PF. However this reduction of the gap to the same phonological rule cannot mean a complete reduction of ellipsis to movement, since the two contexts contrast in a number of syntactic and semantic properties. Section 2, starting from a brief presentation of
Caterina Donati
Chomsky’s theory of movement as a complex operation involving feature agree + merge copy, proposes to identify ellipsis as an instance of merge copy not involving feature checking. Sections 3 and 4 discuss some consequences of this approach potentially deriving both the interpretive (Section 3) and the locality (Section 4) differences between ellipsis and movement. Finally Section 5 discusses some evidence that ellipsis is indeed a syntactic phenomenon.
.
Delete copy in movement and ellipsis
The simplest and most neutral definition of an ellipsis context is that of a string containing a constituent which gets interpreted as identical to some other constituent, without being pronounced. The same holds at first approximation for movement contexts: both in (1a) – a case of VP-ellipsis – and in (1b) – a simple Wh-structure – a constituent fulfilling the selectional requirements of its local environment and interpreted under identity with some other constituent (the root VP in (1a), the Wh-element in (1b)) is left unpronounced. (1) a. John will move tomorrow, and Mary will [e] the day after. b. When will John leave [e] ? There is thus a strong similarity at least at first sight between movement and ellipsis, due to the fact that in both cases a gap is involved which is dependent in its interpretation from some local constituent. The first step of this paper will be to verify how far this superficial similarity can be pushed forward, given the current theory of movement on the one hand and the actual syntactic properties of the two environments on the other hand.
. The copy theory In the minimalist perspective inaugurated by Chomsky (1995), reference to “movement” is really intended in a metaphorical way to indicate a specific instance of the primitive operation merge in which the item inserted in the structure is not extracted directly from the numeration, but rather copied from an element already merged .To clarify, a simple Wh-question like (2) is derived as follows.2 (2) I wonder what Mary eats a. I wonder Mary eats what. Merge copy (move) → b. I wonder what Mary eats what. Delete copy → c. I wonder what Mary eats what
Merge copy
Merge inserts in the structure a copy of the Wh-element, and a PF interface rule of copy deletion erases the lower copy. The main motivation for taking move to be merge copy, in addition to the theory-internal necessity of getting rid of S-structure, comes from reconstruction effects we shall not discuss here, cf. Chomsky (1995: 202–210). All in all, given the copy theory of movement, what directly concerns us here is that grammar already makes available a deletion operation which is defined independently from standard ellipsis cases. This means that as far as its phonological aspect is concerned, namely the fact that it involves elements which are interpreted without being pronounced, ellipsis does not seem to have anything specific. It is thus perfectly legitimate to reduce this phonological aspect of ellipsis to an application of the general copy deletion mechanism. In ellipsis contexts, just like in movement ones, the lower of two identical constituents (i.e. copies) gets deleted at PF and this is simply due to a uniform deletion operation applying on copies. Ellipsis contexts, however, do exhibit some specificity if compared with the contexts traditionally analyzed as movement outputs. If this specificity cannot be explained in terms of the PF process involved, it is to be searched in some other component.
. Specificity of ellipsis contexts Rather standardly, we may identify movement structures by three criteria: (i) they involve a gap and a dependency; (ii) the dependency is virtually unbounded; (iii) it is sensitive to islands configurations. Out of these three standard properties, ellipsis contexts display uniformly only the first, and in a rather peculiar way. The first thing to notice is that ellipsis contexts exhibit a wide variation as far as locality restrictions are concerned. VP-ellipsis and gapping, to take some well known cases, exhibit a dramatically different behavior.3 Consider first socalled VP-ellipsis: as the contrasts in (3) to (5) show, this typology of ellipsis is not sensitive to main islands configurations Wh-movement is subject to. (3) CNPC a. The man who doesn’t eat cookies hates [the man who does [e]]. b. *I wonder what is [the man who eats [e]]. (4) CSC a. John didn’t immediately open the door. First he shut the window, and then he did [e]. b. *I wonder what John first shut the door and then opened [e].
Caterina Donati
(5) Wh-island a. Paul will eat [whatever Mary will [e]]. b. *When will Paul eat tomorrow [whatever Mary cooked [e]] ? VP-ellipsis thus appears to be much freer and more unconstrained than movement, lacking the third property listed above. The reverse appears to hold for gapping, which not only displays a neat sensitivity to islands, but appears to be systematically clause-bounded (hence lacking the unboundedness property (ii)): while it seems to behave coherently with movement contexts in (6), it sharply contrasts with them in (7). (6) CNPC a. *The man who doesn’t eat cookies hates [the man who candies [e]]. CSC b. *John loves Mary but Paul prefers to work and [Peter Jane [e]]. Wh-island c. *Paul will eat [whatever Mary [e]]. (7) Clause-boundedness a. *Mary loves John and [I have been told that Julia Paul [e]]. b. Who have you been told [that Paul loves [e]] ? Considering a little closer the first property standardly associated with movement, that of involving a dependency, another clear specificity of ellipsis contexts arises: while it is true that both in ellipsis and in movement contexts the content of the gap is dependent from the content of an overt antecedent, the way this dependency is interpreted varies a lot. In movement, only one of the two copies (the “gap”) gets interpreted at LF, and thus their dependency is to be intended as a complete identification; in ellipsis contexts both copies get a full interpretation, although identical in content. In (7a), to clarify, the copied inflected verb loves gets interpreted both in its lower-position and in its higher position: the two copies behave as two independent, although identical, predicates. In (7b), on the other hand, the copied Wh-element who is interpreted only once, and it gets both its case and its θ-role in its lower position. The two copies behave as two “occurrences”4 of the same argument. I shall return on these two important differences of ellipsis akin to locality and interpretation, providing a tentative explanation for both . Summarizing so far, we are faced with an empirical dilemma: on the one hand there are some empirical similarities and a theoretical possibility which encourage an identification of the phonological operation involved in movement and in ellipsis. On the other hand ellipsis contexts exhibit strong specificities akin to locality and interpretation. This means that ellipsis cannot be simply reduced to move-
Merge copy
ment. If the phonological operation of deletion is common to both operations, we need to look somewhere else for defining ellipsis contexts as a natural class. One possibility is to invoke here the properly syntactic operation which provides its input to the copy deletion, generating the copy itself.
. Merge copy in movement and ellipsis What is a syntactic operation? One basic aim of the minimalist program is to reduce to the minimum the range of possible answers to this question: the restrictiveness of available operations is indeed one of the crucial assumptions of the program. In Chomsky (1995, 2000, 2001), the rules, or operations, of the grammar, are nothing but two: Merge, i.e. the union in a structure of items extracted from the numeration (and thus ultimately from the lexicon) or of structured phrases; and move, which is a complex operation involving merging of a copy which gets deleted at PF. In such a restrictive framework as minimalism the specificity of ellipsis contexts with respect to movement ones may be reduced to the syntactic mechanism generating the copy which undergoes deletion.
. Decomposing movement Following the approach detailed in Chomsky (2000, 2001), movement is not to be intended as a primitive operation, but as a complex derivation to be decomposed in at least three steps. a.
Agree: Agree is an operation by which a probe (i.e. a set of uninterpretable formal features which need to be deleted) searches for a goal, i.e. a set of matching features. If probe and goal are in a sufficiently local configuration, the uninterpretable features on the probe are deleted long distance, without triggering any further operation. Agree yields instead a “movement” derivation when a particular type of uninterpretable features is involved, called EPPfeatures: features which need for some reason to be checked in a [Specifier, Head] configuration. In that case, Agree identifies the goal, and the derivation proceeds to: b. Merge copy: a copy of the goal is merged in the Spec position of the probe; c. the lower copy is deleted. Notice that under this approach feature checking and merge copy are not identified: feature checking can occur without merging a copy of the goal, the latter operation being triggered only in certain derivations given certain features.
Caterina Donati
. Reduplication The “decomposed” approach to movement briefly summarized above leaves open one possibility for syntax, which appears to be worth exploring: namely that not only feature checking can be done without merge copy, but that also merge copy is an operation that can apply without any feature checking. Suppose merge copy is an operation that can apply freely in syntax. When no agree relation is involved in the derivation, the output of this operation will simply be the copy of a constituent already merged in the structure, which undergoes deletion at PF. This is a good definition of what we find in ellipsis contexts. We shall call this simple operation of free insertion of a copy “reduplication” to distinguish it from the more complex operation “movement” (agree + merge copy). Summarizing, syntactic theory as is construed in the minimalist program makes available the natural possibility of assuming two distinct operations of merge copy: movement, where the copy is the goal of an agree-checking mechanism, and reduplication, where the copy is not the goal of any feature dependency. Technically, taking a given numeration, and having construed K merging α and β, it is possible to apply merge again, merging K with a copy of β. (8)
(9)
K
K
At this point, PF identifies the lower β as a copy and thus deletes it. Remember we do not want ellipsis to be reduced to movement, given the syntactic differences between the two contexts discussed in §1.2. We need in other words to derive the specificities of ellipsis from the specificity of reduplication with respect to movement. This is what we shall try to do in the next two sections.
. Interpretation in movement and ellipsis In both cases, reduplication and movement, there is a dependency in interpretation between the two identical constituents merged in syntax. However, we have seen that an important difference holds. In movement, only one of the two copies receives an interpretation at LF. In reduplication, both copies, although identical in content, are interpreted at LF. Technically, this important difference can be stated through the notion of chain: the copies created by movement are links of the same chain, which is the one and only legitimate object at the LF interface; the copies created by reduplication, instead, are interpreted as members of different chains.
Merge copy
. Definition of chain As for the definition of chain, various possibilities are available. It is clear that the simpliest definition, as a sequence of identical items, i.e. of copies, would not do: given this definition, merge copy would amount to a “create chain” operation, and we would be left with no way of distinguishing movement and reduplication. One alternative is to assume that whether two copies form a chain or not is decided at LF, in other words that chains are solely a matter of interpretation at LF. At LF two identical items a priori can either be interpreted as two segments of a non trivial chain or can be interpreted as two trivial chains. At least in typical cases, however, only one of these two options is actually available, because the other one leads the derivation to crash, violating selectional restrictions, the Θ-criterion and standard constraints on interpretation. This seems an interesting possibility, but possibly falling beside the point: first, it is a fact that movement systematically creates chains, while reduplication never does so. It would be necessary to demonstrate that legitimate LF objects can be uniquely identified, with copies belonging to the same chain properly distinguished from those belonging to different chains (see Chomsky 1995). Which is not obvious. And second, there is already in the syntax an important difference between movement copies and reduplication copies. A natural (and “minimalist”) move would be to derive the interpretive difference observable at LF from the different syntactic operation generating the copies in the two cases. In movement, the constituent a copy of which is merged is the goal of an agreement relation, while no agreement relation is involved in reduplication. Technically, this difference can be built in the definition of chain itself, reducing chain formation to agree, rather than to merge copy: it is the agreement relation which creates a chain among copies, identifying them as the goal of a given uninterpretable probe. When no such agree relation is established, no chain relation holds between the merged copies. (10) Chain A pair < α, β > where α=β is a chain when α is the goal of an agree relation. Another piece of evidence that chain formation is to be linked to agree in the sense of Chomsky (2000) rather than to movement, comes from Cecchetto’s (1999) treatment of so-called optional movement. If Cecchetto is right, a chain can be created in one of two ways: through a standard movement derivation (decomposed as sketched in the preceding paragraph), or through a simple agree operation identifying two copies extracted from the numeration (and thus not created in the syntax) as the goal of a given probe.
Caterina Donati
. Parallelism effects As is well known, the recovery of the deleted constituent in ellipsis contexts gives rise to “parallelism effects” with respect to its antecedent: although not interpreted through a chain in the technical sense defined above, the two copies generated by reduplication still display a dependency constraining their interpretation. A rather standard illustration of these restrictions on interpretation is given in (11). (11) The children are ready to eat and so are the chickens. The VP in the first conjunct is ambiguous between two comically opposed readings. If the deleted VP-copy in the second conjunct received its interpretation independently from the one of its antecedent, we would expect it to be ambiguous as well. But this is never the case: the deleted copy inherits its interpretation from the overt copy. As a result, the sentence in (11) is only two-ways ambiguous, and not four-ways ambiguous. I shall not explore in details any of these facts. It will suffice here to indicate that constraints of this kind are indeed predicted by the reduplication approach pursued here: the deleted constituent and its antecedent are not two lexical items taken from the numeration, but two copies generated in the syntax. Presumably these constituents reach the interface with a special copy marker which constrains their interpretation.
. A sample derivation We have now all the tools for interpreting reduplication. Let’s now turn to a sample derivation of a VP-reduplication. Suppose we start from a numeration like (12) (abstracting from details of the inflectional and complementizer systems). (12) Num = {Paul, I, Bill, I, and, eat, candies} Proceeding bottom up, the derivation first construes an IP constituent like (13): (13)5
IP DP Bill
I’ I
VP [e]
V’ eat candies
Then the conjunction is merged cyclically ending up with (14).
Merge copy
(14)
IP and
IP I
DP Bill I will
VP V
[e]
eat candies
Suppose then that I◦ is merged directly, yielding (15). *I
(15) I
IP and
IP I
DP Bill I will
VP [e]
V eat candies
The derivation will crash because I◦ does not select for an IP. No other item in the numeration (12) yields an interpretable derivation, since I◦ selects for a VP, which is not present in the numeration at this stage of the derivation. The only possibility id thus to reduplicate a constituent satisfying I◦ ’s selection, merging a copy of the VP, and yielding (16).
Caterina Donati
I
(16)
VP
I
IP
VP V
[e]
and
eat candies
IP I
NP Bill I eat
VP V
[e]
eat candies
Then the derivation proceeds, merging the last LI of the numeration. At Spell Out the lower copy gets deleted. (17)
IP I
DP Paul I will
VP VP
[e]
IP V
eat candies
and
IP I
NP Bill I will
VP [e]
V eat candies
For ease of description, the derivation stemming from the numeration (12) has been presented in a way that suggests that selectional restrictions must operate as early as lexical elements enter the derivation. This way of describing things may also suggest that reduplication is indeed triggered by selectional requirements: this
Merge copy
is not what I have in mind. The basic idea of reduplication is that of a free operation, applying freely in the derivation as any other instance of merge. As usually assumed, any configuration violating selectional requirements will crash. This conclusion remains even if selection is checked as late as at LF, when syntax provides its final output at the interface.
. Locality in movement and ellipsis As we noticed above (§1.2), the other major difference between movement and ellipsis has to do with the locality constraints on the dependency involved. We saw in particular two differences: first, different instances of ellipsis such as VP-deletion and gapping appear to obey different locality restrictions. The relevant facts are repeated in (18) through (20): VP-deletion (a-cases) systematically violates islands conditions while gapping exhibits all the typical islands effects (b-cases); (18) CNPC a. The man who doesn’t eat cookies hates [the man who does [e]]. b. *The man who doesn’t eat cookies hates [the man who candies [e]]. (19) CSC a. John didn’t immediately open the door. First he shut the window, and then he did [e]. b. *John loves Mary but Paul prefers to work and [Peter Jane [e]]. (20) Wh-island a. Paul will eat [whatever Mary will [e]]. b. *Paul will eat [whatever Mary [e]]. Second, ellipsis phenomena as a whole seem to obey a sort of locality which is not reducible to the well known conditions movement is restricted by: even gapping, which is closer to movement in that it obeys islands (b-cases above), strongly diverges, being clause-bounded, as illustrated in (7) above repeated here as (21). (21) Clause-boundedness a. *Mary loves John and [I have been told that Julia Paul [e]]. b. Who have you been told [that Paul loves [e]] ?
. Williams (1977a) As for the first point, i.e. the variation internal to ellipsis phenomena with respect to locality restrictions, it is a well known fact: Williams (1977a) proposed on the basis of these differences to ascribe the various subcases of ellipsis to different mod-
Caterina Donati
ules of the grammar: on one hand gapping, which obeys island constraints and is thus more similar to movement, and thus belongs to “sentence grammar”; on the other hand VP-ellipsis, virtually unconstrained and thus external to the core grammar module and reducible to “discourse grammar”. This splitting of ellipsis into two natural classes is not compatible with the uniformist approach pursued here, where all ellipsis phenomena are reduced to the (core) syntactic operation of reduplication. On an empirical side, moreover, Williams’ proposal is not resolutive: while it may provide an explanation for the first order of facts, illustrated in (18)–(20), it wrongly predicts that gapping, being a “sentential grammar” rule, obeys the same constraints movement does: in other words, it wrongly rules out the contrast in (21).
. Minimalist locality Remember that under the minimalist approach movement can be defined as an instance of the operation merge copy triggered by the necessity for a feature (a probe) to be checked by another feature (a goal). Crucially, the locality constraints movement undergoes are not defined on the operation merge copy, but on the mechanism of feature checking (or Agree). The only necessary condition constraining movement in this framework can be stated as the Minimal Link Condition in (22). (22) MLC 6 K attracts α if and only if there is not β, β closer than α, such that K attracts β. In this perspective, it is perfectly conceivable that other instances of merge copy which are not driven by any feature checking requirement, as is ellipsis according to the analysis here proposed (i.e. reduplication), do not obey at all to MLC, which is a condition on features relations. On the other hand, we do not want to exclude any locality-like restrictions for reduplication, since ellipsis does seem to be constrained in this direction. Merge copy involves a searching procedure, which is likely to be constrained by some economy of computation considerations involving reduction of the search space.7 The restrictions the copy dependency obeys will thus be a function of the context and of the nature of the copies involved.
. C-command There is only one rigid and invariable aspect which restricts merge copy in each of its applications, no matter whether “movement” or “reduplication”. From the cyclic nature of the computation procedure, which imposes to merge to always
Merge copy
extend its target, it follows that the two copies of a dependency must always be in a c-command relation: the higher copy (the antecedent in standard terminology) must c-command the lower copy. It is a fact that any ellipsis phenomenon, together with any movement phenomenon, always obey this condition. As for the apparent counterexamples, they are usually reduced in movement cases to some independent facts, such as other movements internal to the relevant sentence, which somehow obscure this fundamental property. Under the approach adopted here, the natural step is to adopt the same tentative attitude with respect to the apparent counterexamples involving reduplication. To illustrate what I have in mind, in the paradigm in (23) VP-deletion appears to violate the c-command condition: in (23a) the “antecedent” follows, instead of preceding, the deleted copy. (23) a. b. c. d.
Before Mary did [e], John read the book. Before Mary read the book, John did [e]. John read the book before Mary did [e]. *John did [e] before Mary read the book.
There is a fairly natural explanation for these facts which allows to preserve the c-commanding condition on the operation merge copy. Assuming (24) as the basic structure for the sentences in (23), the only possibility is to delete the lower copy, that internal to the adjunct (thus deriving 23c). But the adjunct itself can be leftward-moved (topicalized): The result will be (23a). Finally, (23b) can be derived deleting the lower copy in the structure derived from this topicalization. (24)
IP I
John
VP
I
PP
VP read the book
P before
IP I
Mary I did
VP read the book
Caterina Donati
Interestingly, even in this apparently easy game of movements, no structure whatsoever is able to generate (23d), which coherently results in ungrammaticality. Notice before concluding that this c-command condition on ellipsis, perfectly natural under the syntactic analysis as a condition on merge copy, needs to be necessarily stipulated under the standard analyses of ellipsis as an interface phenomenon: there is no principled reason why a phonological rule or a semantic reconstruction should obey a structural notion like c-command. C-command is a central syntactic notion (perhaps the central notion of syntax), and we thus predict it to be active in the syntactic component. In this section we have dropped some lines about the interesting and complex problem of the locality conditions constraining ellipsis. A real theory of these effects is far beyond these few lines, whose only aim is to explicit an important consequence of the minimalist approach for locality and ellipsis: the Minimal Link Condition, defined as it is in terms of features independently from the operation merge copy, predicts the kind of variation observed in ellipsis, and between ellipsis and movement. Crucially, the only structural condition directly incorporated in the definition of merge copy, i.e. the c-command condition, appears to be universally shared both by movement operations and by ellipsis phenomena.
. Syntactic evidence The syntactic approach to ellipsis pursued here presents some immediate advantages with respect to the standard interface analyses. In particular, it allows us to predict both the phonologic effects (deletion) and the interpretive effects (identity) of ellipsis (i.e. reduplication) with no further cost for the theory, identifying them simply with current interface interpretations on copies. If this is on the right track, it remains to ask whether the syntactic approach, in addition of being more simple and “minimalist”, is also superior or at least different from an empirical point of view, or whether it is a simple notational variant of the traditional interface analyses. In this section I shall provide some empirical evidence in favor of the syntactic theory, showing that ellipsis contexts are to be defined in terms of the overt syntactic structure.
. Verb movement and VP reduplication Consider once again the phenomenon of VP-ellipsis. As is well known, this subclass of ellipsis, systematically available in English in contexts like coordinates and relatives, never holds in other languages, like Italian and French: neither in coordinates nor in relatives.
Merge copy
(25) a. Bill has eaten too much, and Paul has, too. b. Bill has eaten whatever Paul has. (26) a. *Bill Bill b. *Bill Bill
ha has ha has
mangiato eaten mangiato eaten
troppo, too-much, qualunque whichever
e Paolo ha anche. and Paolo has also cosa Paolo ha. thing Paul has
(27) a. *Bill Bill b. *Bill Bill
a has a has
trop mangé et Paul a aussi. too-much eaten and Paul has also mangé tout ce que Paul a. eaten all it that Paul has
One may wonder where such a clear-cut cross-linguistic contrast comes out. Remember that the string subject to ellipsis in English is a VP, as shown by the obligatoriness of do-insertion, or of any other auxiliary in the examples in (25). The corresponding sentences with no auxiliary are just as ungrammatical as the French and Italian examples: (28) a. *Bill eats and Paul. b. *Bill eats whatever Paul. If the relevant constituent is a VP, an obvious possibility for explaining this sharp contrast comes to mind, akin to the syntax of the verb in these languages. As is well known in the literature since Emonds (1978), English on the one hand and French and Italian on the other hand display a fundamental syntactic difference, reducible to verb movement. While the subject moves overtly to the Specifier of IP both in English and in French and Italian, the behavior of the verb diverges: it moves overtly to the inflectional system (call it I◦ for simplicity) in French and Italian, but it remains in situ in English. The examples in (29) through (31) briefly illustrate this difference capitalizing on the standard adverbials test. Assuming that semantically equivalent adverbs in different languages invariably occupy the same position in clause structure, the word order variations observable cross-linguistically in relation to the verb can be interpreted as a clear index of its movement. (29) a. Bill often visits museums. b. *Bill visits often museums. c. [IP Bill I◦ [AdvP often [VP [e] visits museums ]]. (30) a. *Bill spesso visita i musei. b. Bill visita spesso i musei. c. [IP Bill visita [AdvP spesso] [VP [e] [e] i musei]]].
Caterina Donati
(31) a. *Bill souvent visite les musées. b. Bill visite souvent les musées. c. [IP Bill visite [AdvP souvent] [VP [e] [e] les musées]]]. This means that while in English the inflected verb heads an autonomous projection, VP, where the subject is not included, this does not hold for Italian and French: in these languages the verb ends up in I◦ , where it shares a specifier, head relation with its subject. There is in other words no constituent containing the verb without including the subject. As a result, reduplicating the constituent containing the verb, thus IP, necessarily implies also copying its subject. The output of such an operation is a sentence only interpretable as a truism, similar to (32). (32) Bill mangia e Bill mangia. Bill eats and Bill eats a. [[IP Bill mangia [[VP [e] [e]] [e [IP Bill mangia [VP [e] [e] ]]]]]] b. Bill (λ x (x mangia) & Bill (λ x (x mangia)) Reduplicating an IP necessarily amounts to reduplicating a saturated predicative relation subject + predicate. This is true in Italian and French, as well as in English: an IP-ellipsis cannot correlate with two different subjects, since the subjects itself is within IP. English presents however an extra-possibility, due to its syntax, that illustrated in the derivation (13)–(17), summarized below in (33). Since the subject and the verb do not belong to the same constituent, it is possible to reduplicate the verbal projection without yielding a truism: the reduplicated VP is interpreted as an open predicate, to be saturated by an independent subject: (33) Bill eats and Paul does. a. [[IP Bill I◦ [[VP [e] eats ] and [IP Paul [I◦ does [VP [e] eat ]]] b. Bill (λ x (x eats)) & Paul (λ x (x eats)) If this simple explanation of the contrast between English and French/Italian is correct, then we are faced with a clear evidence in favor of the validity of the syntactic approach advocated here: ellipsis proves sensitive to overt syntax. Remember that coherently with the minimalist approach, English and Italian or French only differ in the position of the verb before Spell out: in English as in any language, the inflected verb needs to raise to I◦ in order to check its formal features. In abstract syntax, therefore, there is no difference among these languages which may derive the contrast with respect to VP-ellipsis observed.
Merge copy
. Object ellipsis vs. cliticization Given what we said above, an obvious question comes to mind: what happens with transitive verbs? If ellipsis (hence reduplication) is sensitive to overt syntactic constituency, we predict that languages with overt V-movement exhibit nevertheless cases of VP-ellipsis, although limited to the object NP, being the only lexical material remaining in VP in overt syntax. To clarify, we predict that a derivation like (34) should hold. (34) a. [IP subject verb [VP [e] [e] object]] b. [VP [e] [e] object] [and [[IP subject verb [VP [e] [e] object]]]] c. [IP subject’ verb’ [[VP [e] [e] object] [and [[IP subject verb [VP [e] [e] object]]]]]] d. [IP subject’ verb’ [[VP [e] [e] object] [and [[IP subject verb [VP [e] [e] object]]]]]] The result of this derivation should be that of a sort of “object drop” presenting all the interpretive properties associated with ellipsis briefly mentioned above in §3.2 under the generic label of parallelism effects. Not surprisingly, this phenomenon does exist and is in fact widely attested cross-linguistically: as far as I know it is reported to hold at least in Hebrew (35), in Japanese (36), in Chinese, in Korean and in Irish. (35) Mary koret ha sefer shela ve Dani gam kore. Mary reads-om the book of-her and Dani also reads (a book of Mary/a book of his)
(Hebrew)8
(36) John-wa zibun-no tegami-o suteta. Mary-mo suteta. John-top he-gen letter-acc threw. Mary also threw (a letter of John/ her letter)
(Japanese)9
The very existence of similar facts provides a strong argument in favor of the syntactic approach discussed here. Things however are not so simple, in that the prediction of the theory, namely that a language with overt V-to-I movement should allow for a VP-ellipsis limited to the object along the derivation in (34), appears to be too strong. It is apparently falsified by Romance languages, which do not present VP-ellipsis whatsoever. (37) a.
Paolo prepara i biscotti la domenica e Maria *(li) prepara tutti i giorni. Paul prepares the cookies the Sunday and Mary them (cl.) prepares every day
Caterina Donati
b. Paul prépare les biscuits le dimanche et Marie *(les) prépare tous les jours. Paul prepares the cookies the Sunday and Mary them (cl.) prepares every day c. Pablo prepara los biscochos los domingos y Maria *(los) prepara todos los dias. Paul prepares the cookies the Sundays and Mary them (cl.) prepares every day Notice that in the relevant contexts, illustrated by the examples in (37), these languages obligatorily involve cliticization of the object, instead of a full NP inside the VP. Interestingly, on the other hand, Hebrew and Japanese have no clitic pronouns. How is cliticization related to ellipsis? Under Sportiche’s (1992) approach to cliticization, the defining property of clitics is that of being base-generated in the position where they show up.10 This entails that the clitics in (37) never sit in the VP, at no step of the derivation. This means that in these languages, even with transitive verbs, the VP is indeed totally empty in overt syntax, leaving nothing to be visibly reduplicated.
. IP-reduplication As a matter of fact, the conclusion drawn above about IP-reduplication yielding a truism in (32) does not entail that this option always ends up in ungrammaticality. Rather, the subclass of ellipsis known as sluicing 11 can be traced back to an operation of this kind. Consider the examples in (38), illustrating two typical sluicing sentences. (38) a. Mary surely met somebody, but I don’t know who. b. Mary got married, guess why. The sluicing sentences in (38) can be analyzed as involving an IP-reduplication. The result is in this case significant, because the copied IP contains a gap. (39) sketches the derivation of (38) for illustration. (39) [IP Mary met [e]] [FocP somebody], but I don’t know [FocP who [IP Mary met [e]]] The same derivation involving IP-reduplication may underlie gapping contexts, like (40), where the remnants of ellipsis and their parallels in the first conjunct are clearly in extraposed focus – and topic-related positions, cf. Pestesky (1982).12 (40) Paul loves Mary, and Mary John.
Merge copy
Not surprisingly, French and Italian display no difference whatsoever in the licensing of these two subcases of ellipsis: sluicing and gapping exhibit the same properties and distribution just sketched for English, as shown in (41) through (44). (41) a.
Marie a rencontré quelqu’un, mais je ne sais pas qui. Marie has met someone, but I not know who b. Marie s’est mariée. Devine pourquoi. Marie refl-is married guess why
(42) Paul aime Marie, et Marie Jean. Paul loves Marie and Marie Jean (43) a.
Maria ha incontrato qualcuno, ma non so chi. Maria has met someone but not know who b. Maria si è sposata. Indovina perché! Maria refl. is married guess why
(44) Paolo ama Maria, e Maria Giovanni. Paolo loves Maria and Maria Giovanni The absence of any cross-linguistic variation in ellipsis possibilities for sluicing and gapping confirms that the relevant constituent in this case is IP, and not VP, where a major syntactic difference holds among the languages considered. Concluding this section, the strict correlation between VP-deletion possibilities and overt syntax illustrated above provides a strong argument in favor of the syntactic analysis in terms of reduplication proposed here. It shows in particular that in addition to the phonological effects (deletion) and the semantic effects (identity and parallelism) traditional approaches focus on, ellipsis is crucially constrained by syntactic properties.
. Conclusion Before concluding, there is one last consideration that needs to be sketched. A potential problem for a syntactic account of ellipsis comes from the well known optionality of this phenomenon. It is a fact that any sentence potentially containing ellipsis can be pronounced without any deletion. In other words, (45) appears to have the same interpretation and to be used in exactly the same contexts, independently from the actual spelling out of the VP. (45) Mary will leave tomorrow, and Paul will (leave) the day after. Optionality represents a problem for any minimalist approach, where the computational system and its derivations is seen as the perfect solution to interface
Caterina Donati
conditions. Under such an approach, one would want to eliminate the notion of optionality from narrow syntax as a whole. The apparent optionality of ellipsis can however be overcome capitalizing on the technical notion of numeration. Going back to (45), the sentence can be derived starting from at least two numerations, given in (46). (46) N1: { Mary, Paul, and, tomorrow, the, day, after, will2, leave2 } N2: {Mary, Paul, and, tomorrow, the, day, after, will2, leave} The first numeration, containing two identical verbal elements extracted from the lexicon, will generate the version without ellipsis of the sentence in (45). The second numeration, which does not contain enough lexical elements to construe two interpretable conjuncts, must necessarily rely on the syntactic mechanism of reduplication, eventually generating the version of (45) involving VP-deletion. In both cases, the available derivation is only but one, with no optionality whatsoever. The postulation of two different numerations crucially eliminates the problem of optionality, given that by definition two sentences generated from different numerations cannot be compared on economic grounds.13 Concluding, this paper discusses an embryo of what would be a properly syntactic analysis of ellipsis, in a minimalist perspective. Starting from the observation of the similarity between movement and ellipsis, the first section proposes to derive the gap involved in both contexts from the same copy deletion rule applying at PF. Since the two contexts contrast in a number of syntactic and semantic properties, however, ellipsis cannot be reduced to movement. Section 2 proposes to identify ellipsis with an instance of merge copy not involving feature checking, reduplication. Sections 3 and 4 discuss some consequences of this approach potentially deriving both the interpretive (Section 3) and the locality (Section 4) differences between ellipsis and movement. Finally Section 5 discusses some evidence that ellipsis is indeed a syntactic phenomenon.
Notes . For an introduction to the phonological approach to ellipsis, see Tancredi (1992). On the LF side, see among others Chao (1988), Kitagawa (1991), Fiengo and May (1994). . The derivation in (2) is actually a simplification for ease of exposition. Chomsky’s most recent approach to movement is described in more details in §2.1. . On gapping, see in particular Neijt’s (1979) fundamental study and Pesetsky (1982), plus of course López & Winkler in this volume, and the references quoted there. . The term “occurence” is been used in Chomsky (1999: 32) to refer to the segments of a chain without appealing to the classic movement metaphor.
Merge copy
. I deliberatly abstracted away in the derivation from the well known but poorly understood condition of the obligatoriness of an overt auxiliary preceding the deleted VP, as illustrated in (i). (i) Paul wrote to his parents today, and Mary *(will/does) tomorrow. There is a standard GB analysis of this fact, which essentially interprets this condition as an ECP phenomenon, due to the requirement for the null VP to be properly (i.e. lexically) governed. See Zagona (1988), Doron (1999), Lobeck (1995). This standard account, in addition of incorporating a controversial notion such as government, crucially rely on some very suspicious assumptions, such as that I◦ θ-marks the VP. . The version of MLC cited here is given in Chomsky (1995: 311) and is still defined in terms of Attraction, instead of the probe/goal agree relation, but the substance does not change. . A similar approach is suggested in Chomsky (2001) to account for locality constrains on head dependencies. . These data were given to me by Maya Arad, whom I thank. . These data are taken from Otani & Whitman (1991: 346). See also McCloskey (1991) for the same phenomenon in Irish. . See also Manzini and Savoia (in press) and, on totally different grounds, Jelinek (1984) and Baker (1991). . On sluicing, see Levin (1982) as a classic, and more recently, Romero (1998) and Merchant (2001), plus of course Romero (in this volume) and Schwabe (in this volume). . See also Donati (2000) for an extension to comparative deletion. . See Cecchetto (1999) for an interesting reflection along these lines on some instances of (apparent) optional movement.
Phrase structure paradoxes, movement and ellipsis Winfried Lechner This paper pursues two main objectives: First, it presents evidence for a remnant movement analysis of phrase structure paradoxes (Pesetsky 1995) which arise with VP-fronting, refuting the parsing solution of Phillips (1996). Second, based on a contrastive study of the interpretive properties of adjunct remnants in VP-fronting and VP-ellipsis, it is argued that (VP-) adjuncts can be merged in at least two distinct positions, and that the actual insertion point is determined by economy conditions. This conception entails that (i) not only move, but also merge is (indirectly) regulated by an economy metric, and that (ii) economy cannot be computed locally, but needs to be evaluated on the basis of larger information units such as phases.*
.
Introduction
. Goals and outline The present paper intersects two issues central to the study of movement, ellipsis and economy. First, in current theorizing, empty nodes in movement chains and in contexts of ellipsis are both conceptualized as copies with internal structure, which need to be licensed by a linguistic antecedent under (an appropriate version of) parallelism. But there are also systematic differences between traces and ellipsis copies, which manifest themselves for instance in their varying ability to host reconstruction sites for movement. The first goal of this paper consists in providing evidence that this specific disparity, which will be dealt with in depth below, does not reveal intrinsic properties of the two different exponents of copies, but can be derived from general principles of economy. The second question to be addressed regards an imbalance which can be perceived in the sensitivity of the two structure building operations move and merge to principles minimizing computational cost. In particular, whereas movement processes are widely held to be regulated by economy conditions, no such metric has been identified so far which would choose among competing applications of (root)
Winfried Lechner
merge.1 This is due to the fact that move characteristically enjoys a certain degree of freedom (e.g. in the choice of the landing site), while merge generally introduces categories only in designated positions in the tree, roughly those which support the correct theta and predication relations. Thus, in most cases, competition among derivations which possibly differ only in the position at which a category is merged will simply not arise. Behind this background, I will present evidence drawn from ellipsis and movement phenomena that (i) there are indeed derivations in which a category can be potentially merged into two distinct locations and that (ii) the choice between the competing candidates is determined by economy. This conception entails an interesting consequence for the debate whether economy is computed strictly locally (Collins 1996), or needs to incorporate transderivational constraints. To explicate, suppose that in a given derivation, there is the choice of merging a category α low or high, yielding the two subderivations D1 and D2 schematized in (1). Assume moreover that low attachment of α, as in derivation D1, entails further movement of α at a later point in the derivation (step m), which would not have been necessary if α had been merged high, as in D2: (1)
derivation D1 –derivation D2 Step k: merge α low ...... Step l: ...... merge α high Step m: move α ...... (k < 1 < m)
Finally, suppose that D2 wins over D1 for the reason that D2 employs fewer movement operations than D1. On these assumptions, the grammar must decide at step k whether to introduce α low at step k, as in D1, or whether to delay merging α until Step l, as in D2. However, the critical information that early merger at k leads to additional cost at m, which is indispensable in order to discriminate between D1 and D2, is not yet accessible at k, but only becomes available once the derivation reaches step m. Thus, transderivational rule interactions as in (1) indicate that the economy metric does not operate on a strictly local, step-by-step basis (Collins 1996), but must have access to larger units of information. Empirically, the discussion revolves around the proper analysis of two types of constructions in which a phonetically silent VP is followed by an overt remnant: so-called phrase structure paradoxa involving VP-fronting, as in (2) (Pesetsky 1995; Phillips 1996, to appear), and instances of VP-ellipsis or pseudogapping, exemplified by (3) (Jayaseelan 1990; Johnson 1996; Lasnik 1995b; Levin 1986; (2) adapted from Pesetsky 1995: 230, (570c)): (2) John intended to give the book to the children, and [VP give the books to themi ] he did on each otheri ’s birthdays (3) John gave the book to the children on Monday, and Mary did on Friday.
Phrase structure paradoxes, movement and ellipsis
The paper is structured as follows: Subsequent to some expository remarks on the nature of the paradox in (2), Section 2 reviews a recent proposal for its resolution (Phillips 1996), which will be seen to require rather radical changes in the grammar and which will be discarded on empirical grounds. Section 3 presents an alternative, conservative analysis which likens (2) to the extensively studied group of remnant movement phenomena in Western Germanic. To anticipate, it will be proposed that (2) involves extraction of the remnant PP prior to VP-topicalization. On this view, the analysis of (2) is identical to the one widely adopted for pseudogapping as in (3) (Jayaseelan 1990; Johnson 1996; Lasnik 1995b), the only difference being that the VP in pseudogapping is removed from its base position by ellipsis and not by movement. Section 4, which contains the main theoretical contribution, turns to disparities between pseudogapping and VP-fronting, focusing in particular on the reconstruction behavior of adjunct remnants. These differences will be attributed to the interaction between economy conditions and the assumption that adjuncts may be merged with the root at different stages of the derivation.
. Pesetsky’s Paradox Syntactic constituency tests are hypotheses about the structural organization of surface strings which have been arrived at by generalizations about phenomena including movement, anaphoric dependencies, licensing of NPIs, coordination and ellipsis. Ideally, these hypotheses converge at a single structural description for a given string in a given interpretation. If two or more of the diagnostics lead to contradicting evidence for the constituency of a string, the hypotheses are at conflict, resulting in a phrase structure (PS-)paradox. Pesetsky (1995) observes that English ditransitives represent one instance of such a PS-paradox. On the one side, movement tests indicate that the VP is leftbranching, as illustrated by the fact that any contiguous string of categories including the left edge of the VP may be fronted: (4) VP-fronting John intended to give candy to children in libraries on weekends, and a. . . . [VP give candy to children in libraries on weekends], he did. b. . . . [VP give candy to children in libraries], he did on weekends. c. . . . [VP give candy to children], he did in libraries on weekends. On the other side, c-command sensitive tests, which register the distribution of negative polarity items (NPI), anaphoric dependencies and pronominal variable binding attest to the fact that the VP-shell is right-branching.2 For instance, the scope domain of downward entailing NPs includes NPIs to their right, but does not extend over NPIs to their left (Barss & Lasnik 1986; Larson 1988; Phillips 1996):
Winfried Lechner
(5) NPI-Licensing a. John gave nothing to any of the children in the library on his birthday. b. John gave candy to none of the children in any library on his birthday. c. John gave candy to children in no library on any public holiday. d. *John gave anything to none of the children. e. *John gave candy to any of the children in no library. Thus, different criteria for structure and constituency yield conflicting evidence for the organization of the English VP. At first sight, this finding might be taken as evidence that movement and binding relations can be read off two structurally distinct trees. However, as observed by Pesetsky (1995: 230), it is also possible to find examples like (2), repeated below, which simultaneously exhibit properties of right- and left-branching phrase markers, resulting in what will be referred to as Pesetsky’s Paradox. In (2), the well-formed anaphoric dependency between them and each other indicates that the VP is assigned a right-branching tree, as illustrated by (6a). But the topicalized string give the books to them can strand the adjunct on each other’s birthday only if it is parsed into a left-branching VP, as in (6b):3 (2) Pesetsky’s Paradox John intended to give the book to the children, and [VP give the books to themi ] he did on each otheri ’s birthdays. (6) a. BINDING: right-branching VP
b. TOPICALIZATION: left-branching VP
VP NP the books
VP1 VP-FRONTING F
VP V´
PP to themi
tv
VP2
PP
V´
PP
PP
V
NP
on each otheri’s birthdays
give
the books
on each otheri’s birthdays
to themi
Pesetsky (1995) resolves the paradox by adopting a dual system, which relates a single surface string to two distinct phrase-markers, a right-branching cascade and left-branching trees generated by Layered Syntax. On this view, the conflicting structural requirements of (2) can be distributed between two distinct representations: binding is verified on the basis of cascades, and the satisfaction of principles governing movement relations is delegated to Layered Syntax.
Phrase structure paradoxes, movement and ellipsis
Although empirically adequate and successful in accounting for a wide variety of facts in addition to constituency conflicts,4 the dual system faces two problems. First, it is not obvious why in the dual system the phenomena are distributed the way they are. Why for instance is Binding Theory evaluated on the basis of cascades, and not on the basis of Layered Syntax ? Second, the dual system rejects the standard assumption that (unambiguous) surface strings are bi-uniquely mapped onto graph representations in favor of a weaker, less restricted hypothesis about the relation between terminals and structure. As will be shown in Section 3 and 4, though, PS-paradoxes like (2) also lend themselves to an analysis which relies on orthodox assumptions about the factorization of surface strings, and therefore fail to elicit evidence against the standard view.
. The parsing solution A different approach toward Pesetsky’s Paradox, which rests on a new conception of how trees are assembled by the grammar, is advocated by Phillips (1996, to appear). Section 2 outlines this solution (2.1), discusses an extension to contrasts between VP-fronting (VP-F) and VP-ellipsis (VP-E; pseudogapping), which will be relevant for the further discussion (2.2), and presents three empirical generalizations that pose a serious challenge for Phillips’ account (2.3).
. Incremental merger Phillips (1996) suggests that structure is built incrementally from left to right by a top-down parser. The parser proceeds according to the two principles of the incremental merger in (7): (7) Incremental merger I. Merge right: New items must be introduced at the right edge of a structure. II. Branch right: Merge as low as possible. (8) a.
b. ü {A,{B,C}}
{A,B} A
B
A
c. û
{B,C} B
{{A,B},C} {A,B}
C
A
C B
The schematic derivation of the string ABC in (8) illustrates the algorithm at work. When the parser reaches C, it reads the instructions to attach C at the right edge of
Winfried Lechner
AB (merge right) and to merge C as low as possible (branch right), and accordingly proceeds as in (8b), discarding the representation (8c). Crucially, A and B form a constituent at an early point of the derivation (8a), whereas in the final output, B is grouped together with C, such that C is located within the c-command domain of both A and B (8b). The incremental merger provides now the key to the solution to Pesetsky’s Paradox, because a string of symbols can meet seemingly contradictory constituency requirements by satisfying the individual conditions at different stages of the derivation. As shown by (9), which tracks the derivation of (2), the fronted VP is assembled first (9a). Then, an identical copy of the topicalized VP is inserted into the base position (9b), satisfying the parallelism requirement on movement chains. In (9c), the temporal adjunct is finally merged low at the right edge, supplying the correct configuration for the anaphoric dependency: (9) John intended to give the books to the children, and. . . (= (2)) a. [VP give [the books [to [them]]]] he did b. [VP give [the books [to [them]]]] he did [VP give [the books [to [them]]]] c. [VP give [the books [to [them]]]] he did [VP give [the books [to [themi [on [each otheri ’s birthdays]]]]]] The theory generates a number of interesting predictions (see Phillips 1996 for details). Among them, there is one which is of specific interest for present purposes, as it relates to a contrast between movement copies in VP-F and ellipsis copies in VP-E. I briefly comment on this prediction in the next subsection, turning from there to critical discussion in Section 2.3.
. VP-fronting and VP-ellipsis Phillips notices a further curious property of Pesetsky’s Paradox. Whereas VPinternal categories can bind into the remnant if the VP has been topicalized (2), such a relation cannot be established if the VP has been removed by ellipsis, and not by movement, as in (10). (10) Vp-ellipsis *John [VP gave the books to them on each otheri ’s birthdays] and Mary did on each otheri ’s first day of school. ( = [VP gave the books to themi ]) On the incremental merger account, the contrast between VP-fronting in (2) and VP-ellipsis in (10) follows from two assumptions: First, the parser may alter the constituent structure in course of the derivation only if the changes are compatible
Phrase structure paradoxes, movement and ellipsis
with other, global constraints of the grammar. Second, VP-ellipsis is subject to the global constraint of syntactic parallelism.5 If parallelism is observed, as in (11), the adjunct in the elliptical clause resides outside the c-command domain of its antecedent. If, on the other hand, (10) is parsed into a right-branching VP, as in (12), the structural requirements for binding are met, but a violation of syntactic parallelism ensues. Since the conditions on ellipsis and binding cannot be reconciled in a single structure, the output is ill-formed. (11)
XP
XP
VP
PP
V´ V give
V´ V
NP to them
give
the books
(12)
give VP
NP the books
PP
NP the books
V´ V
PP on each other’s birthdays
PP PP
on each other’s first day of school
NP to them the books
E VP-ELLIPSIS (=(10)) (violation of parallelism)
give VP VP
to them
VP
ü VP-ELLIPSIS F (antecedent = ellipsis)
on each other’s birthdays
PP
(=(10))
VP PP to them
V´ V
PP on each other’s first day of school
Thus, incremental merger successfully accounts for the observation that contradictory constituency effects (Pesetsky’s Paradox) are not attested in contexts of VP-E. The analysis faces problems in other empirical domains, though.
. Criticism The present section lists three problems for the parsing analysis of PS-paradoxes. First, the incremental merger account entails a further prediction for the hierarchical position of remnants in VP-E. Not only should binding from within the elliptical VP into the remnant be blocked (10), but it should equally be impossible to establish an anaphoric link between an NP inside the antecedent VP and remnantinternal anaphors. This prediction is contradicted by the data, though. NPs which
Winfried Lechner
are part of the first conjunct may antecede anaphors inside remnants, indicating that the antecedent VP is right- and not left-branching: (13) 1st conjunct, principle A John gave the books to them on each otheri ’s birthdays and Mary did on their first day of school. ( = [VP gave the books to them]) Notice on the side that the syntactic parallelism condition excludes an alternative derivation in which the remnant is merged low in the antecedent VP but high in the elliptical VP. The conjecture that adjuncts are attached low within the VP in the first, but high in the second conjunct of VP-E is corroborated by data from disjoint reference effects. Phillips (1996: 62, fn. 41) notices that a name inside an adjunct remnant induces a Principle C violation only if it is part of the antecedent clause: (14) 1st vs. 2nd conjunct, principle C a. *?John gave the books to heri on Maryi ’s birthday and Jill did at Christmas. ? b.( )John gave the books to her at Christmas and Jill did on Maryi ’s birthday. ( = give the books to heri ) Similar first vs. second conjunct asymmetries can be observed with pronominal variable binding ((15b) from Phillips 1996: 61, (87b)): (15) 1st vs. 2nd conjunct, variable binding a. *Mary congratulated every boy at his graduation and Sue did at hisi 21st birthday party. ( = congratulated every boyi ) b. Mary congratulated every boyi at hisi graduation and Sue did at her 21st birthday party. Thus, adjuncts are – contrary to Phillips (1996) – merged low if they originate inside the antecedent conjunct of VP-E. Second, on the assumption that the sequential expansion of the tree is regulated by branch right, one is led to expect that categories further to the right should also be more deeply embedded. Although this generalization (by and large) captures the binding theoretic properties of remnants, it fails to provide an accurate description of the scope behavior of quantificational remnants in VP-F. To begin with, it is well-known that in VP-F, quantificational subjects cannot be assigned narrow scope w.r.t. other VP-internal quantifiers (vd. (16b)), their scope domain is ‘frozen’ (see Barss 1986 and Huang 1993, among others):
Phrase structure paradoxes, movement and ellipsis
(16) scope freezing: transitive a. No one will teach every student. b. . . . and teach every student, no one will.
¬∃ > ∀/∀ > ¬∃ ¬∃ > ∀/ * ∀ > ¬∃
The same observation holds for object remnants stranded by VP-topicalization, which have to take scope above quantifiers inside the fronted predicate (Sauerland 1998b: 591): (17) scope freezing: ditransitives a. David planned to give every handout to one of the students. . . ∀ > ∃/∃ > ∀ b. . . . and [VP give every handout] David did to one of the students. *∀ > ∃/∃ > ∀ The proper analysis of Scope Freezing is immaterial for present purposes.6 All that matters is that the incremental merger account implies that the remnant (one of the students) in (17b) is merged low in its surface position, as in (18): (18) . . . and [CP [VP . . . ] [IP David did [VP give every handout [to students]]]]
one of the
But on this assumption, it should – contrary to fact – be possible to construe the indefinite within the scope of the fronted universal every handout. The absence of a narrow scope reading for the indirect object therefore constitutes first hand evidence against the parsing analysis. An additional argument corroborating the view that the surface position of stranded categories is above – and not inside – the silent VP comes from the interaction of VP-F and Antecedent Contained Deletion (ACD). As illustrated by (19a), the remnant may host an ACD-site: (19) VP-fronting and ACD a. Mary asked him to give a book to some of the boys, and [VP give a book] he did [ACD-Host to every boy Mary wanted him to ] b. [VP . . . [VP give a book to t i ] [ACD-Host every boy Mary wanted him to
]i ] ( = [VP give a book to t]) Ellipsis resolution in (19a) is contingent upon the host QP every boy Mary wanted him to being located above the antecedent VP give a book to (in its base position).7 This implies that the ACD-host is VP-external at LF, as schematized in (19b). In principle, such a configuration can be achieved in one of two ways: Either the host undergoes QR out of the lower copy of the VP, or it overtly moves out of the VP prior to VP-topicalization. Evidently, only the latter option is consistent with the
Winfried Lechner
assumptions needed to account for Scope Freezing. The host QP must therefore have reached its VP-external location already in overt syntax, in contradiction to the incremental merger analysis which equates the surface location of the remnant with its base position.8 The third and final problem for incremental merger to be addressed here pertains to the syntactic principles which determine possible configurations for the application of branch right. As documented by the contrast in (20), the remnant in VP-F must not be separated from the position in which it is interpreted by an island: (20) locality She attempted to refute the allegation that theyi met on each otheri ’s birthdays a. . . . and [VP refute the allegation that theyi met on each otheri ’s birthdays] she did. b. *. . . and [VP refute the allegation that theyi met] she did [on each otheri ’s birthdays]. Phillips recognizes this problem and accounts for it by the restriction in (21). However, (21) rather amounts to a restatement of the facts than to a principled analysis of the phenomenon. (21) ‘Clause Constraint on VP-gaps’ (Phillips, to appear, (129)) In the sequence [. . . Aux [VP null] adverbial. . . ] the adverbial is interpreted as a clausemate of the Aux. To recapitulate, the parsing approach towards PS-paradoxes fails to capture three empirical generalizations: (i) first vs. second conjunct asymmetries in VP-E, (ii) Scope Freezing and related effects from ACD, indicating that remnants may be merged high in VP-F, and (iii) locality restrictions on the distance between the remnant and its interpretive position. Section 3 presents an alternative analysis which straightforwardly captures the generalizations (ii) and (iii). In Section 4, I will elaborate on property (i).
. Remnant movement . Remnant topicalization The alternative account of Pesetsky’s Paradox to be pursued here builds on the strategy of remnant topicalization. Remnant topicalization is a phenomenon widely found in Western Germanic (among other languages), which shifts contiguous strings including the right edge of the VP to SpecCP (see e.g. Haider 1993; den Besten & Webelhuth 1990; Müller 1998):
Phrase structure paradoxes, movement and ellipsis
(22) remnant topicalization in german a. Sie wollte [VP einem Freund ein Kamel schenken]. a friend a camel give she wanted b. [CP [VP Einem Freund ein Kamel schenken] wollte sie]. a friend a camel give wanted she c. [CP [VP ti Ein Kamel schenken] wollte sie einem Freundi ]. a camel give wanted she a friend d. [CP [VP ti tk Schenken] wollte sie einem Freundi ein Kamelk ]. friend a camel give wanted she a ‘She wanted to give a friend a camel (as a present).’ According to an influential line of thought, initiated by Thiersch (1985) and den Besten and Webelhuth (1987), remnant topicalization is the product of extracting the remnant out of the VP prior to fronting of the highest VP-projection, as in (22). On an alternative conception, the category that moves is smaller than the topmost VP-shell and does therefore not contain traces of the remnants (Fanselow 1983, 1992; van Riemsdijk 1989; Haider 1993). At the moment, the debate surrounding this issue has not been resolved yet conclusively, both options have arguments in their favor. As will become clear shortly, the synthesis to be advocated here supports the extraction analysis. Applying the remnant movement algorithm to Pesetsky’s Paradox yields the derivation in (23), which is strongly reminiscent of the pseudogapping analysis developed by Jayaseelan (1990), Lasnik (1995b) and Johnson (1996). The PP adjunct originates VP-internally and adjoins to a node above VP prior to VPtopicalization:9 (23) John intended to [VP give the books to the children], and [VP give the books to themi] he did on each otheri ’s birthdays. 2. VP-FRONTING
(=(2))
F
VP VP
give NP the books
PPk
...
on each other’s birthdaysPF E 1. MOVEMENT
(...)VP PP to themi
V´ tv
PPk on each otheri’s birthdaysLF
Winfried Lechner
At LF, the adjunct remnant reconstructs into its base position. Since the pronoun them c-commands the VP-internal copy of the PP, Principle A is observed, and c-command and constituency are no longer at conflict. Before proceeding to further consequences, note that (2) is incompatible with the small constituent approach towards incomplete predicate fronting. Unlike in German, where stranded categories reside at the left edge of a verb-final VP, remnants in English are generated at the right edge of a right-branching VP in which the verb has moved to the left periphery. It follows that English VPs do not contain a node dominating the incomplete VP but excluding the remnant which could serve as the target for ‘small constituent movement’. Hence, the analysis of PSparadoxes advocated here provides evidence against the small constituent movement approach and for the existence of remnant movement. The remnant movement account needs to meet two criteria in order to qualify as a plausible alternative to the parsing solution. First, it has to be clarified which type of dislocation process is involved in the remnant movement step which precedes VP-fronting. For some speculative remarks on this issue, I refer to Appendix I. Second, it has to be demonstrated that the analysis has a wider empirical coverage than the incremental merger account. As it turns out, the remnant movement analysis immediately removes two of the three problems for the incremental merger which were identified in the previous section. To begin with, one is now correctly lead to expect that the remnant and its trace must not be separated by islands, as movement has to proceed locally: (24) Locality She attempted to refute the allegation that theyi met on each otheri ’s birthdays a. . . . and [VP refute the allegation that theyi met on each otheri ’s birthdays] she did. b. *. . . and [VP refute the allegation that theyi met] she did [on each otheri ’s birthdays]. Moreover, the assumption that remnants reach their surface position by overt movement implies that quantificational remnants need to take scope above VPinternal operators (Scope Freezing): (17b) Scope Freezing David planned to give every handout to one of the students and [VP give every handout] David did to one of the students. *∀ > ∃/∃ > ∀ This follows from the assumption that quantifiers which have been overtly moved out of fronted predicates are – for whatever reason – frozen in scope.
Phrase structure paradoxes, movement and ellipsis
The third problem which the incremental merger hypothesis encountered related to a two-way contrast between adjunct remnants in VP-F and VP-E. Recall that remnants appear to be within the binding scope of VP-internal categories if they are generated in the antecedent conjunct of VP-E and in contexts of VP-F, but not if they are part of the elliptical clause of VP-E. Transposing this observation to the remnant movement analysis, one is led to conclude that adjunct remnants reconstruct in VP-F and in the first conjunct of VP-E, but not in the second conjunct of VP-E. So far, the remnant movement account fails to provide an insight into this curious property of adjunct remnants.10 The analysis will for this reason be modified in the following section.
. Adjunct remnants Before proceeding to the proposal for the reconstruction asymmetries in VP-F and VP-E, I will lay out some background assumptions about the licensing mechanism of VP-E (4.1) and about clause structure and the position of VP-adjuncts (4.2).
. Semantic parallelism Unlike Phillips (1996), I adopt a hypothesis more in line with recent research on ellipsis, according to which VP-E is sufficiently licensed by semantic parallelism, and does not require strict LF-identity between the ellipsis and its antecedent. Without going into the details, semantic theories of ellipsis converge on the assumption that in VP-E, the semantic value of the antecedent clause has to be an element of the focus semantic value of ellipsis (see e.g. Fox 1998; Rooth 1992a). This conception has the favorable effect that focused constituents are ignored for the computation of parallelism (Sauerland 1998a). To illustrate, the focus semantic value of the ellipsis clause in (25a) is given by the set of presuppositions (p-set) in (25b), which provides possible alternatives for the expressions containing the focus (Mary and on Monday, respectively, assuming that the focus feature percolates to the PP node).11 Ellipsis is licensed in (25a) despite the fact that the two clauses are strictly speaking not identical, because (25a) satisfies the condition that the meaning of the antecedent clause is contained in the p-set (25b), as can be seen in (25c): (25) a. John [VP gave the books to them on Sunday] and maryF did [VP on mondayF ]. b. P-set = {p∈Dt | ∃x∈De , ∃y∈D<<∈,t>,<∈,t>> [p = x gave the books to them y]} =
Winfried Lechner
= {John gave the books to them on Sunday, John gave the books to them on Monday, Mary gave the books to them on Monday, Mary gave the books to them on Sunday,. . . } c. [[John gave the books to them on Sundays]] ∈ P-set In addition, VP-E is subject to a syntactic requirement that the elided category (gave the books to them) has to form a constituent to the exclusion of the remnants in overt syntax. Hence, remnants in the ellipsis clause cannot be contained inside the ellipsis. This aspect will be taken up again in the discussion of the predictions the system entails for the analysis of remnants in VP-E and VP-F.
. Two positions for VP-adjuncts According to a widely held view first advanced in Larson (1988), and more recently articulated in Haider (1993) and Kayne (1994), postnominal adverbials in English are parsed into a low position within a right-branching VP which is c-commanded by the verbal arguments (see also McConnell-Ginet 1982). Strong evidence in favor of the structural organization of English VPs along these lines comes from the observation that precedence within the VP directly translates into c-command (Barss & Lasnik 1986; Ernst 2001 and references therein): (26) a. She [VP metk [VP no boyi tk [on hisi birthday]]]. b. I [VP visitedk [VP the boysi tk [on each otheri ’s birthdays]]]. c. *We [VP visitedk [VP himi tk [in Sami ’s house]]]. Surface word order in (26) results from overt verb movement into the head of a higher VP-shell to the left of the internal arguments. As verbs are invariantly interpreted in their base position (vd. e.g. Bittner 1994), the verb forms a constituent in semantics with postnominal adjuncts which excludes the internal arguments, though.12 In semantics, the verb therefore has to combine with postnominal adjuncts first, before it can be joined with its internal arguments. But this conception is incompatible with the conjunction of two basic assumptions about the mapping from overt syntax to semantics: (i) the ‘ordered argument approach’ towards argument association (to use the terminology of Dowty 1989), which maintains that the arity of a predicate is reflected in its semantic type, and (ii) the hypothesis that arguments – or, to be precise, at least the internal arguments (see Kratzer 1996 and below) – semantically combine with their predicates by functional application. To illustrate, assume the standard view that verbs extensionally denote relations between individuals (ignoring event arguments for ease of exposition; nothing hinges on this, though). Transitive verbs are then of type <e,<e,t> >. Suppose moreover that VP-adverbs such as in Sam’s house in (26c) are modeled as predicate modifiers of type < <e,t>,<e,t> >:
Phrase structure paradoxes, movement and ellipsis
(27) [[visit in Sam’s house]] = = [[visit]]<e,<e,t> > [[in Sam’s house]]< <e,t>,<e,t>> = Type mismatch It follows that due to type mismatch, the minimal node containing meet and in Sam’s house in (26c) cannot be assigned a meaning. The literature provides two strategies to avoid this complication: On the one side one could give up the assumption that transitive verbs denote two-place relations between individuals in favor of a Neo-Davidsonian semantics (Castañeda 1967; Parsons 1990). On this perspective, verbs and their arguments as well as adverbs denote predicates of eventualities, which can directly combine with each other in any order. The Neo-Davidsonian approach comes however at the cost of losing the means to encode the arity of a predicate, and thereby the distinction between intransitive, transitive and ditransitive verbs. I will for this reason pursue an alternative solution instead, which uses a more orthodox syntax for VP-adverbs by parsing them in a position where they are directly interpretable and which at the same time preserves the insights of the ordered argument approach.13 More specifically, the proposal rests upon the following assumptions: In transitive and ditransitive constructions, the lower (VP) projection of an articulated shell structure embeds the verb and its internal arguments, while the higher projection (vP) hosts the subject and the Case positions for objects (Chomsky 1995, 1998), as shown in (28a): (28) a.
b.
vP<,t> vP<,t> Adv-I<<,t>,<,t>>
vP SUB v
vP
SUB
v´ VP OBJ
v´
OBJi
VP<,t>
v
VP<,t> Adv-II<<,t>,<,t>> ... ti ...
Objects move from inside VP to layered specifiers of vP in course of the derivation to check their Case features. Thus, SpecvP qualifies as an ‘A-position’, which licenses binding relations. Following Kratzer (1996), I assume that this bifurcation in the projection of internal and external arguments is also reflected in the semantics of verbs. For Kratzer, the lexical entry of the verb specifies the number
Winfried Lechner
of internal arguments but does not select for the external argument, which is added by the v◦ (Voice◦ ) head above VP. In addition to argument slots for direct and indirect objects, verbal predicates contain event argument positions (Davidson 1967; Higginbothom 1985). On this view, both VP and vP denote predicates of events (type <∈,t>), as illustrated by the typed tree (28b). Finally, VP-modifiers denote modifiers of predicates of events (<<∈,t>,<∈,t>>). Given these rather standard syntactic and semantic assumptions, it follows that the lowest position into which VP-modifiers can be merged is the VP-adjunction site Adv-II in (28b). However, this is not the only position in which VP-adverbs can be generated, they may also originate in a second designated slot, the vP-adjoined position Adv-I. The next section explores the consequences of the two-adjunct hypothesis for the analysis of reconstruction asymmetries in VP-F and VP-E.
. Adjunct reconstruction in VP-E and VP-F .. VP-ellipsis Turning to VP-ellipsis first, recall that the diagnostics from Principle A, variable binding and disjoint reference effects indicated that adjunct remnants reside in the scope of objects if they are part of the antecedent clause, but not if they modify the ellipsis clause. Two of the relevant paradigms are repeated from above: (29) 1st vs. 2nd conjunct, principle A a. *John [VP gave the books to themi on each other’s birthdays] and (= (10)) Mary did on each otheri ’s first day of school. b. John gave the books to themi on each otheri ’s birthdays and Mary did on their first day of school. (= (13)) (14) 1st vs. 2nd conjunct, principle C a. *?John gave the books to heri on Maryi ’s birthday and Jill did at Christmas. ? b.( )John gave the books to heri at Christmas and Jill did on Maryi ’s birthday. The present system accounts for this disparity without any further substantial additions. On the null-hypothesis to be adopted here, VP-E may target any maximal projection in the VP-shell – i.e. either the lower VP or the higher vP – given that semantic parallelism is met. For reasons of concreteness, suppose that syntactically, the ellipsis operation is encoded as a feature which is assigned to a maximal projection and is interpreted as an instruction to forego phonetic spell-out of this node. Moreover, recall that there are two positions in which adjuncts can be merged. This yields four combinatorial options to derive a string in which an adjunct remnant
Phrase structure paradoxes, movement and ellipsis
follows an elided VP, as documented in (30). To begin with, the desired configuration can be arrived at by merging the adjunct high in Adv-I and phonological suppression of the lower VP-shell, as in (30a). Note that in (30a), the adjunct does not reside within the c-command domain of objects, which move to SpecvP. Next, the same surface string can be derived by merging the adjunct low, in Adv-II, followed by movement of the remnant prior to the application of ellipsis. Ellipsis may target VP, as in (30b), or vP, as in (30c). Since adjunct movement strands a copy in Adv-II, objects in SpecvP may now take scope over low adjuncts. Finally, the fourth possible derivation (30d) combines high merger with ellipsis of vP, and also involves adjunct movement. (30) *. . . and Mary [Ellipsis gave the books to themi] on each other’s first day of school a. ü High Merger with vP (Adv-I) + VP-ellipsis vP vP
vP
Adv-I vP
tSUB
b. û Low Merger with VP (Adv-II) + VP-ellipsis + Adjunct Movement XP
VP E 2. VP-ELLIPSIS
E 1. ADJUNCT MOVEMENT
vP
tSUB
OBJi
Adv-IIPF
VP E 2. VP-ELLIPSIS
OBJi
... ti ...
VP
Adv-IILF
... ti ... c. û Low Merger with VP (Adv-II) + vP-ellipsis + Adjunct Movement XP 2. VP-ELLIPSIS
F
vP
High Merger with vP (Adv-I) + vP-ellipsis + Adjunct
Adv-IIPF vP
tSUB
d. û
VP
OBJi VP ... ti ...
Adv-IILF
E 1. ADJUNCT MOVEMENT
Winfried Lechner
Observe at this point that the four strategies do not harmonize equally well with the general economy metric on movement. In particular, the derivations (30b) and (30c) in terms of low attachment necessitates an additional movement operation which is not required if the adjunct is merged high, as in (30a).14 Low attachment at Adv-II in (30b) and (30c) will for this reason be blocked in favor of high Merger in Adv-I, as in (30a). Thus, economy ensures that whenever the VP elides, adjuncts are introduced in a position which resides outside the scope of objects. The absence of adjunct reconstruction within the elliptical conjunct is thereby reduced to the interaction between two factors: (i) the hypothesis that VP-adjuncts may be inserted wherever they are interpretable and (ii) general economy conditions on movement.15 As already foreshadowed in the introduction, the analysis has an interesting consequence for the definition of the domains within which economy competition is calculated. Observe that in the derivation of the ellipsis clause, movement is triggered by the need of the adjunct to escape the ellipsis site. But for low adjuncts in Adv-II, this need emerges only once the whole vP has been assembled and elided (vd. (30b)). The economy metric can therefore identify the optimal candidate only once the derivation has reached the higher adjunct position vP. It follows as a corollary that economy cannot be computed in a strictly local fashion, but needs to have access to larger information units (such as phases). Next, consider why adjuncts necessarily reconstruct into the antecedent VP, i.e. why they have to be merged into the lower adjunct position Adv-II (as can be seen from the Principle C effect in (14b)). The answer to this question has two parts to it. First, remnants in the antecedent clause can be merged low because they do not need to move to a higher position in course of the derivation (the VP is not elided). Furthermore, since adjunct remnants bear focus, and focused categories are ignored for the computation of semantic parallelism (vd. 4.1), elliptical conjuncts in which the adjunct is parsed low inside the antecedent VP also observe the licensing condition on VP-E. As a second ingredient, the analysis has to be supplemented by a principle which ensures that in the absence of ellipsis, low attachment of adjuncts at AdvII is preferred over high merger into Adv-I. This principle could be related to two independent factors. First, it can be interpreted as an instance of the Earliness Principle (Pesetsky 1989), which holds that operations should be executed as early as possible. Assuming that Earliness also demands that elements in a subarray are merged as early as possible, thereby leading to a rapid exhaustion of the subarray, early – i.e. low – attachment is preferred over late merger. Alternatively, low attachment follows from theories of phrase structure on which precedence is mapped onto c-command (Haider 1993; Kayne 199416 ). Thus, independent factors will ensure that adjunct remnants in the antecedent clause not only can but also have to be
Phrase structure paradoxes, movement and ellipsis
merged low into the antecedent clause, accounting for the disjoint reference effect observed in (14b). (See Appendix II for a speculation along another line.) To recapitulate, adjunct remnants inside the ellipsis clause are merged high (Adv-I), because otherwise, they would have to move overtly to a position above the ellipsis site. It follows that adjuncts are located outside the c-command domain of objects. In the antecedent clause – as well as in regular VPs – adjuncts are generated low and therefore appear to reconstruct for the verification of interpretive principles.
.. VP-fronting Proceeding to remnant reconstruction in VP-F next, recall that VP-F differs from VP-E in that adjuncts reconstruct into the binding scope of VP-internal NPs: (2) VP-fronting, principle A John intended to give the book to the children, and [VP give the books to themi] he did on each otheri ’s birthdays. Moreover, the fact that a remnant-internal name may corefer with a pronoun inside the topicalized VP attests to the fact that adjunct reconstruction in VP-F is not obligatory, but optional: (31) VP-fronting, principle C John promised to give the books to her next year, and give the books to heri he did on Maryi ’s birthday. Thus, the behavior of adjunct remnants stranded by VP-F matches neither that of adjuncts inside the first nor that of adjuncts in the second conjunct of VP-E. This peculiar imbalance in the behavior of adjuncts can be linked to an independent property characteristic of VP-F. VP-F may – in contrast to VP-E – not apply freely to any maximal projection, but has to target the highest verbal projection (i.e. vP). Evidence to this effect comes from Huang’s (1993) interpretation of Barss’s (1986) observation that fronted predicates fail to display Multiple Binding Domain Effects: (32) a. [Which book about herself] does she think he is reading? b. *. . . and [VP tk reading a book about herself] she thinks hek is. Huang suggests that the reflexive herself in (32b) cannot be licensed by the matrix subject she, because the fronted predicate pied-pipes the trace of the subject (tk ), which accordingly serves as a closer potential binder for the anaphor. Since subjects originate in vP (Chomsky 1995; Kratzer 1996), it follows that VP-fronting must target the higher projection vP, and not VP.17
Winfried Lechner
This additional restriction on VP-F provides a straightforward explanation for the optionality of adjunct reconstruction. As shown by (33), the prohibition on topicalization of VP entails that the adjunct remnant has to escape the ellipsis site by overt movement, irrespective whether it is merged high or low. (33) vP-FRONTING
F
vP
vP
Adv-I vP
tSUB
E PRINCIPLE C ... and [VP give the books to heri] he did on Maryi’s birthdays
VP
OBJi VP
Adv-II ... ti ...
E PRINCIPLE A (VARIABLE BINDING, ...) ... and [VP give the books to themi] he did on each otheri’s birthdays
Crucially, since the derivation involves one movement operation in any case, the adjunct remnant can be merged either high or low, resulting in the appearance of optional reconstruction. If the adjunct is merged low, it reconstructs for the assessment of Principle A (see (2)). High attachment leads on the other hand to Principle C obviation in (31). A potential problem has to be addressed at this point. Evidently, the two movement operations in (33) are not of equal length, and one might object that the MLC should prefer high over low attachment. There is however independent evidence that Adv-I and Adv-II count as equidistant from higher attractors, licensing low merger. A first indication supporting this view comes from the observation that adjuncts fail to trigger Superiority violations among each other, signaling that adjuncts can be generated in positions which are equidistant from higher attractors. (34) a. Where did you sleep when. b. When did you sleep where. However, (34) does not yet conclusively demonstrate that Adv-I and Adv-II are equidistant from higher attractors, as where and when could either be both generated in Adv-I or both originate in Adv-II. A more reliable test can be based on examples of VP-F with two adjunct remnants. The remnants have to be chosen in such a way that one of them is merged in Adv-I, whereas the other is generated in Adv-II. This can be achieved by (i) embedding a name inside one of the remnants which is to be construed coreferential with a pronoun inside the fronted VP
Phrase structure paradoxes, movement and ellipsis
(the remnant therefore needs to originate high in Adv-I) and by (ii) embedding an anaphor inside the other remnant which is to be bound by the pronoun inside the fronted VP (ensuring that the remnant is merged low in Adv-II). On the assumption that Adv-I and Adv-II are equidistant from higher attractors, the remnants should now be able to undergo movement in any order. Thus, one is led to expect that the two remnants may surface in either order (on the relation between surface order and order of movement see e.g. Richards 1997). Turning to the data, the control in (35) exemplifies VP-F with two adverbial remnants: (35) John said he would give the books to them in one of the gardens, . . . and [VP give the books to themi] he did [in the garden] [on each otheri ’s birthday]. Pesetsky (1995: 230, (570c)) As illustrated by the examples in (36), a high and a low remnant may surface in the order Remnant-I–Remnant-II (‘Remnant-I’ is mnemonic for ‘generated in Adv-I’): . . . and [VP give the books to themi ] he did [Remnant-I in the childreni ’s garden] [Remnant-II on each otheri ’s birthday]. b. . . . and [VP give the books to themi ] he did [Remnant-I in John and Peteri ’s garden] [Remnant-II on each otheri ’s birthday].
(36) a.
But the two remnants may also reach their surface position by crossing paths, leading to the order preserving linearization (low) Remnant-II – (high) Remnant-I, as in (37): . . . and [VP give the books to themi ] he did [Remnant-II on each otheri ’s birthday] [Remnant-I in the childreni ’s garden]. b. . . . and [VP give the books to themi ] he did [Remnant-II on each otheri ’s birthday] [Remnant-I in John and Peteri ’s garden].
(37) a.
What is important for present purposes is that (36) and (37) can be given a consistent analysis only on the assumption that Adv-I and Adv-II are equidistant.18 It can be concluded that there is no inherent preference for merging adjunct remnants high, as maintained by the account of optional adjunct reconstruction presented above.
Winfried Lechner
. Conclusion The present paper pursued the goal of providing evidence in support of three conclusions pertaining to the analysis of PS-paradoxa and related constructions of pseudogapping. First, the parsing account of Pesetsky’s Paradox and pseudogapping developed in Phillips (1996) fails to account for essential properties of the two constructions (Scope Freezing, binding scope of adjuncts and locality effects). Second, Pesetsky’s Paradox lends itself to a remnant movement approach, which eliminates two of the three problems for the parsing analysis. It lacks however the means to express the complex distribution of adjunct reconstruction in VP-F and VP-E, summarized in the table in (38): (38) Remnants reconstruct for ↓ in context of → Variable binding, inside adjunct remnant Variable binding, inside argument remnant BT-A, inside adjunct remnant BT-A, inside argument remnant BT-C, inside adjunct remnant BT-C, adjunct inside argument remnant BT-C, argument inside argument remnant Scope
VP-F – – – – * * – *
VP-E, 1st – – – – – * – –
VP-E, 2nd * – – – * * – –
Thirdly, the behavior of adjunct remnants falls out from an analysis which combines a flexible syntax for adjuncts with general principles of economy and interpretation. More specifically, I advanced the hypothesis that (temporal and local) VP-adjuncts can originate in at least two discrete positions in the tree. The actual choice of the position into which an adjunct is merged is determined by two factors: (i) independent conditions on VP-F and VP-E and (ii) economy principles, which minimize movement operations.
Appendix I: Remnant movement in English This appendix briefly speculates on which kind of movement operations might be involved in English remnant movement (Pesetsky’s Paradox). I will restrict myself to noting differences and similarities between remnant movement in English and two other types of dislocation processes.
Phrase structure paradoxes, movement and ellipsis
Johnson (1996) finds evidence that the remnant in pseudogapping is evacuated from the VP by the English equivalent of Scrambling. It is tempting to extend this analysis to partial predicate fronting (Pesetsky’s Paradox). To begin with, Johnson notes that Scrambling as well as remnant movement in pseudogapping must not target secondary predicates or particles (a- and b-examples below, respectively). Interestingly, VP-F c-examples) is subject to the same constraint: (39) predicate movement a. *dass Hans Maria [AP interessant]k deshalb tk findet. interesting for this reason finds that John Mary b. *Rona looked annoyed, but she didn’t [AP frustrated]. c. *and [VP consider the book] he did [AP interesting]. (40) particle movement a. *dass Hans den Fernseher ausk immer tk stellt. off all the time puts that John the TV b. *While Perry might switch the TV off, he won’t on. c. *and [VP turn the TV] he did off. There is a however a well-known point of variation between VP-F and pseudogapping (Fanselow 2001; Haider 1993): VP-F cannot strand NPs (which are not heavy; vd. (i) in fn. 7): (41) movement of NP a. dass Hans [NP das Buch]k deshalb tk gelesen hat. the book for this reason read has that John b. (?)Sam read the magazine, while I did [NP the book]. c. *and [VP read] he did [NP the book]. This might suggest that the remnant in VP-F is targeted by an instance of extraposition, which is equally known to ignore NPs (or Heavy NP-shift). On this view, the ill-formedness of (41c) reduces to an independent prohibition on NPextraposition. Moreover, since extraposed categories reconstruct for anaphoric and pronominal variable binding (Büring & Hartmann 1997), the analysis correctly captures the observation that VP-internal categories may bind anaphors and pronominal variables inside the remnant. I have to delegate a more satisfactory account of this problem, which goes beyond these preliminary speculations, to further research, though.
Winfried Lechner
Appendix II: A speculative note on lox attachment This second appendix outlines a possible line of attack for answering the question of why low attachment of adjuncts into Adv-II is preferred over high merger in Adv-I. The proposal appears in an appendix as it is not compatible with the NeoDavidsonian semantics employed in the main text. Assume that – contrary to the view adopted so far – VP-extensions denote sets of individuals (<e,t>), and that VP-adjuncts are predicate modifiers of extensional type <<e,t>,<e,t>>. Maintaining Kratzer’s (1996) hypothesis that the external argument is introduced by v◦ , as well as the hypothesis that VP-adjuncts can be generated in two distinct positions, the typed tree diagram for the VP looks now as in (42a): (42)
a.
b.
vP vPt
Adv-I<<e,t><e,t>>
vPt SUB
VP<e,t>
SUB
VP<e,t>
Adv-II<<e,t><e,t>>
<e,t>
<e,t>
i
VPt vP
OBJ ti
Adv-Ik<<e,t><e,t>>
tk VP<e,t>
If an adjunct is merged low in Adv-II, it can be directly interpreted in its base position, because its sister node (VP) denotes a property. If the adverb originates however in the higher adjunct position of (42a) – Adv-I – it can no longer combine with its sister node (vP), which denotes a proposition, in a meaningful way. Does this mean that adverbs cannot be merged into Adv-I? Not necessarily. Adopting an idea proposed in Nissenbaum (1998: 164) for another type of construction, Adv-I winds up in an interpretable position once the structure is modulated by two string-vacuous movement operations, as shown in (42b): In (42b), the subject does not move to its Case position directly, but leaves an intermediate, vP-adjoined copy first (see Sauerland (to appear) for independent motivation for this movement operation).19 This has the consequence that there is now a node of the appropriate type – the mother node of the λ-binder – which can semantically combine with Adv-I. As detailed by (42b), the adverb may now undergo string-vacuous movement and adjoin to vP, driven by the need to avoid type mismatch. However, since this derivation involves at least one additional movement step (type-driven adjunct
Phrase structure paradoxes, movement and ellipsis
shift from Adv-I), which can be avoided by low attachment, the economy metric prefers low merger in Adv-II over high merger in Adv-I.
Notes * I would like to thank Elena Anagnostopoulou and Uli Sauerland for discussion and helpful comments. . Competition between movement and merge represents the third combinatory option, which has been argued to be resolved in favor of merge (‘merge over move’; vd. Chomsky 1995: 348; 1999). . The same result can be reproduced by using conjunction tests (see Pesetsky 1995 for details). . Pesetsky’s Paradox generalizes to contexts involving VP-fronting and other c-command sensitive tests, as e.g. pronominal variable binding (Phillips 1996: 61, (86b)): (i) Mary said she would congratulate every boy, and [VP congratulate every boyi ] she did at hisi graduation. . It captures for instance the generalization that (most) PPs are invisible for the computation of c-command (Reinhart 1983). . This assumption contrasts with the widely accepted view that VP-ellipsis is sanctioned by semantic identity modulo focus (see e.g. Rooth 1992; Fox 1999 and Section 4.1). . Any account of Scope Freeing must (i) exclude QR out of the lower VP-copy and (ii) prevent reconstruction of overtly moved QPs into the VP-copy (on the latter see Lechner 1998; Sauerland & Elbourne 2000). . Notice that the ACD-site includes the preposition to, whereas the overt antecedent VP in (19a) apparently fails to do so. This appears to signal that the derivation involves an intermediate step. First, the PP remnant to every boy ... moves (ib). Then, every boy ... moves, stranding the preposition (ic). Finally, give a book is fronted. Crucially, since the VP and to are still lower than the ACD-host, the ellipsis site can be matched with give a book to.
(i) a. Base: [VP give a book to every boy Mary wanted him to] b. Move PPi : [[VP give a book ti ] [to every boy Mary wanted him to]i ] c. Move QPk : [[[VP give a book ti ] [to tk ]] [every boy Mary wanted him to]k ] . It is not possible to adopt a global constraint requiring the ACD host to be merged in a position in which ACD can be resolved, forcing high attachment for the remnant in (19). This assumption would e.g. conflict with the observation that ACD displays island sensitivity (Haïk 1987), as remnants could be attached outside the island. . Nothing in the discussion bears on the question whether the adjunct moves to the right or to the left.
Winfried Lechner . Reconstruction of argument remnants is regulated by the same principles governing reconstruction of wh-moved NPs (Freidin 1986; Lebeaux 1988, 1990). For one, remnants reconstruct for Principle C even in the elliptical conjunct of VP-E, but only if the name resides within an argument (Sauerland 1998b: 114, (78)): (i) pseudogapping, principle C a. *While some believed himi everything, others did only the story that Johni had met aliens. b. While some believed himi everything, others did only the story that Johni had evidence for. Moreover, arguments which embed anaphors or bound variable pronouns appear to reconstruct indiscriminately in pseudogapping (i.e. VP-E) as well as in VP-F. The proper binding relations can be repaired in all contexts because the remnant strands a copy in object position. (On the distribution of copies of adjunct remnants see Section 4.) (ii) pseudogapping, variable binding We will introduce every girli to heri Spanish teacher, and you will to heri driving instructor. (iii) pseudogapping, principle A (?)We will introduce the girlsi to each otheri ’s supervisors, and you will to each otheri ’s parents. . Temporal adjuncts such as on Monday are analyzed as modifiers of event predicates (type <∈,t>; ∈ is mnemonic for ‘eventuality’). . In a variant of the low-adjunct theory, postnominal adverbs are generated in-between arguments and the verb, and the verb overtly climbs to a position to the left of the highest adverb (Alexiadou 1997; Cinque 1999). . Two further options come to mind: First, it would be possible to treat adjuncts as arguments, as in McConnell-Ginet (1982). See e.g. Larson and Segal (1995: 468f) for some discussion. Second, adjuncts could be assumed to move to a position where they are interpretable. . These considerations carry over to (30d). . Note that a fifth derivation, in which the adjunct is merged into Adv-II, followed by ellipsis of the lower segment of VP, can be excluded by the plausible condition that VP-E may only affect maximal projections and not just segments. This fifth parse, which does not imply adjunct movement, would illegitimately lead to a configuration in which objects obtain scope over the adjunct. The prohibition against deletion of segments can furthermore be derived from the assumption that adjuncts are (special kinds of) specifiers (Kayne 1994). On this conception, ellipsis must not target a lower segment to the exclusion of adjuncts (i.e. specifiers), because no operation may affect a node to the exclusion of its specifier. . In (30), adjuncts are right-adjoined in order to be able to abstract away from the effects of verb movement. Nothing bears on this issue, though. Adjuncts could also be left-adjoined, in line with Kayne (1994).
Phrase structure paradoxes, movement and ellipsis . The prohibition on movement of VP can be made to follow from the Minimal Link Condition, because vP is closer to the attracting head than VP. . Note that the same conclusion applies if the remnants are assumed to be generated to the left, or if they were to move left-wards. . Following Heim and Kratzer (1998), 1 assume that the binder index (λi ), which creates a λ-abstract, is severed from the NP it is associated with (the subject copy) and syntactically represented.
Unpronounced heads in relative clauses Uli Sauerland This paper argues that English relative clauses exhibit a general structural ambiguity. Based on data from reconstruction, I argue that the relative clause head can either move from a relative clause internal position to its surface position, or be base-generated in its surface position. Furthermore, I argue that when the relative clause head is base-generated in its surface position there is an elided internal head. This result argues for a distinction between the non-pronunciation of lower copies in a movement chain and non-pronunciation of PF-deleted material in ellipsis phenomena.*
.
Introduction
The analysis of English relative clauses is the subject of a long-standing debate (Lees 1960, 1961; Chomsky 1965; Kuroda 1968; Schachter 1973; Vergnaud 1974; Heim 1987; Kayne 1994; Borsley 1997; Grosu & Landman 1998; Hackl & Nissenbaum 1998; Bianchi 1995; and others). Consider the example (1): The question is whether the head of the relative clause – pandas in (1) – stands in a transformational relationship to the relative clause internal argument position occupied by a trace. Though some of the literature also views the determiner the as part of the head, really the central question of the debate is the transformational relation of head and relative clause internal trace. (1) The pandas that we saw t at Ueno were cute. head (NP)
relative clause
Both possible positions – that there’s a transformational relationship and that there isn’t – received support. The latter position, Carlson (1977a) calls the Matching Hypothesis: it claims that there’s no direct transformational relationship between the head NP and the relative clause internal trace position. Instead an phonologically empty operator raises from the relative clause internal position to the initial position of the relative clause, and mediates the semantic relationship between the relative clause internal position and the head. The other position – that there’s a
Uli Sauerland
transformational relationship – together with the generally held assumption that Movement is the only transformational rule amounts to the Raising Hypothesis: The head NP (or sometimes DP) starts out in the relative clause internal position, and moves to its surface position.
. Contributions of this paper A lot of recent work on the syntax and semantics of movement relationships has focussed on properties of reconstruction (Chomsky 1993; Fox 1998, 1999b; Freidin 1986; Heycock 1995; Huang 1993; Lebeaux 1988, 1992, 1998; Sauerland 1998a; Takano 1995; and others). This has led to a better understanding of the properties of movement and a refined concept of reconstruction. This paper attempts to apply these diagnostics to relative clauses, in the hope of shedding new light on the question of the transformational relationship. Based on the study of reconstruction properties, I argue for the following three results. The first result is that Carlson’s (1977a) claim that both raising and matching relative clauses exist in English (also Heim 1987; Grosu & Landman 1998) is supported by evidence from reconstruction in English. However, I also argue that the distinction is drawn along slightly different lines from what Carlson (1977a) proposes. For example, antecedent contained deletion will be necessarily some form of a raising relative according to my analysis. The second result I argue for is that matching relatives just like raising relatives have a complex internal head. For raising relatives, this is expected, since the external head is by hypothesis moved from a relative clause internal position. For matching relatives, however, this result is puzzling, since here the external head didn’t move from a relative clause internal position. The presence of an internal copy of the head argues for some grammatical relationship between the trace position and the external head even in matching relative clauses. My third result concerns the nature of the relationship of the silent internal head and the external head in matching relative clauses. I observe here that research on comparatives (Bresnan 1973, 1975; Lechner 1999) has revealed a similar finding: an internal copy of the head that doesn’t seem to be related by movement to the external head. Bresnan has proposed a transformational rule called comparative deletion to account for this. Following this lead, I propose an analogous process of relative deletion. I analyze this as an obligatory ellipsis process, and show that the standard assumption that vehicle change is possible in ellipsis, but not in movement chains accounts for the special properties of reconstruction into relative and comparative clauses.
Unpronounced heads in relative clauses
In sum, I propose that relative clauses are in general ambiguous between two structures. For example (1), the two structures are those illustrated in (2a) and (2b).1 movement of Op pandas
(2) a. The pandas [Op pandas] we saw [Op pandas] at Ueno. movement of pandas movement of Op pandas b. The pandas [Op pandas] we saw [Op pandas] at Ueno. relative deletion of pandas
Structure (2a) is the raising structure, where, after relative clause internal whmovement, an NP is raised out of the moved wh-phrase to a relative clause external position; namely that of the head of the relative clause. Because lower copies in movement chains are never pronounced, the structure (2a) results in (1). (2b) illustrates the matching structure, I advocate. The relative clause internal wh-phrase is identical to that of the raising structure, and also relative clause internal whmovement takes place in both structures. However, the NP pandas doesn’t move to the external head position in (2b). Rather, the relative clause external head position is filled by some NP that’s a different token from the internal head. However, a special relationship between the two NPs must be satisfied – that of relative/comparative deletion. This relationship, forces phonological deletion of the NP in the relative clause internal position. Moreover, the antecedent licensing this deletion must be the external NP, where I assume that ellipsis licensing requires identity of meaning as first discussed by Sag (1976a) (see also Tancredi 1992; Fox 1999a; and Merchant 2001). (2b) is also pronounced as (1), since again lower copies in movement chains aren’t pronounced, and furthermore PF-deletion of the instance of pandas in the relative clause internal chain head is forced by the relative deletion. This paper raises a number of questions about the syntax and semantics of relative clauses: For example, some of these concern the interpretation of relative clauses. In the conclusions I outline some of the issues that arise, and point at possible solutions to these questions. However, at present, I often don’t know good evidence that would favor one analysis over another.
Uli Sauerland
. Matching and raising relatives In this section, I aim to show that the reconstruction behavior of the relative clause head argues for Carlson’s (1977a) claim that relative clauses are ambiguous between raising and matching relative clauses. The basic contrast leading to this claim is that the head doesn’t show Condition C reconstruction, but allows reconstruction for Condition A and variable binding (Munn 1994: 402).2 (3) a. The relative of Johni that hei likes lives far away. b. The relative of hisi that everybodyi likes lives far away.
. Reconstruction in wh-movement Before addressing reconstruction properties in relative clauses, this section summarizes some of the literature on reconstruction in wh-movement. My goal is to demonstrate that reconstruction here is a well-described phenomenon (Freidin 1986; Heycock 1995; Huang 1993; Lebeaux 1988, 1992, 1998; Takano 1995) and that it can be understood quite well on the basis of the copy theory of movement (Chomsky 1993; Fox 1999b, 2000b; Sauerland 1998a). This literature summary is by no means exhaustive, but rather focuses selectively on the phenomena relevant for the later sections. One property of wh-movement that has been discovered is that reconstruction of the moved NP with its arguments is obligatory. Only the determiner and modifiers adjoined to the NP need not reconstruct. This is demonstrated by (4) and (5). In (4a), the R-expression John is part of the argument of argument and therefore induces a Condition C violation with the pronominal he that c-commands the trace position of wh-movement. In (4b), the relative clause containing John is a modifier to the noun argument, and therefore doesn’t induce a Condition C violation in the trace position. (4) a. *[Which argument that Johni was wrong]j did hei accept t j in the end? b. [Which argument that Johni had criticized]j did hei accept t j in the end? The contrast in (5) makes the same point, but shows more pointedly that only the position of the R-expression John in the wh-phrase determines whether it triggers a Condition C violation or not. (5) a. *[Which argument of Johni ’s that Mary had criticized]j did hei omit t j in the final version? b. [Which argument of Mary’s that Johni had criticized]j did hei omit t j in the final version?
Unpronounced heads in relative clauses
Taking Condition C as a diagnostic for the position of an R-expression at LF, the LF-structure of a wh-chain as revealed by reconstruction is sketched in (6). I use the terms core NP and modifiers to distinguish between the lowest NP-segment of the NP-complement of the Wh-Determiner, and all modifiers adjoined to this NP. In the example (5a), the core NP would be argument of John’s, while the only modifier is the relative clause that Mary had criticized. (6) Wh-Determiner (core NP) modifiers ... core NP Spec of CP
trace
Diagram (6) says that the core NP must occur in the lowest position of a whmovement chain. Modifiers, however, need not. Since modifiers must though be interpreted somewhere, I assume that they must occur in the position of the whdeterminer. For the core NP, it is unclear from the arguments above whether it also occurs in the higher position of a wh-chain. Following Fox (1999b), I will assume that it does. A second factor has been shown to affect the LF-position of modifiers. Namely, a bound variable pronoun in the wh-moved constituent forces reconstruction to a position where the bound variable is c-commanded by its antecedent. So in (7), the relative clause modifying paper must be represented in the trace position t i at LF, where it is c-commanded by every student. (7) [Which paper that hej wrote]i did every studentj plan to revise t i ? The evidence for the reconstruction of bound variables is Lebeaux’s (1992) observation that Variable Binding and Condition C Reconstruction go hand-in-hand (see also Lebeaux 1998; Fox 1999b; Fox 2000b). Hence, a violation of Condition C is observed in (8): Binding of the pronoun hek requires representation of the constituent containing he in the trace position t i . But in that position, the R-expression is c-commanded by she, and therefore Condition C blocks coreference of these two expressions. (8) *[Which paper that hek gave to Bresnanj ]i did shej think that every studentk would like t i ? Lebeaux (1998: (123b))
. Reconstruction of relative clause internal material In the domain of relative clauses, material that is pied-piped internal to the relative clause behaves exactly like the moved material in wh-questions. This observation is, as far as I know, due to Safir (1999). (9) and (10) show that this material exhibits an argument/adjunct distinction just like wh-movement.
Uli Sauerland
Consider first (9).3 The difference between (9a) and (9b) is that in (9a) the R-expression John is part of a prepositional phrase modifying the phrase whose picture, while in (9b) the R-expression is part of an argument of the same phrase (cf. Safir 1999: (30a)). There’s a singer whose picture in Johni ’s office hei ’s very proud of. (Safir 1998: (34b)) b. *There’s a singer whose picture of Johni ’s office hei ’s very proud of.
(9) a.
The contrast in (10) shows essentially the same as the one in (9) under the assumption that the prenominal genitive in (10a) is a modifier of the noun description (cf. also Safir 1999: (28)). (10) a.
Max is a prince Johni ’s description of whom hei varies when spies are around. (Safir 1998: (34c)) b. *Max is a prince whose description of Johni hei varies when spies are around.
Hence, I conclude that whatever is said about the reconstruction of wh-movement in questions carries over directly to movement of the relative clause operator in a relative clause.
. Reconstruction of the relative clause head The external head of a relative clause, however, exhibits a different behavior. The head of the relative clause displays the ambiguous behavior already illustrated by (3) above: Reconstruction effects seem to be absent with respect to Condition C, while other tests for reconstruction show that it must be possible. Consider first Condition C. It’s well known that an R-expression in the head of a relative clause doesn’t trigger a Condition C effect in the relative clause internal trace position, even when it’s an argument as in (11). (11) The relative of Johni that hei likes t lives far away. The following minimal pairs establish that there’s is difference between the head of a relative clause and wh-movement with respect to Condition C. The examples (12a), (13a), and (14a) all show that material of the head of a relative clause doesn’t trigger a Condition C effect in the trace position. The corresponding examples in (12b), (13b), and (14b) establish that, for wh-movement, a Condition C effect is observed under the same circumstances.4 (12) a. Which is the picture of Johni that hei likes? b. *Which picture of Johni does hei like?
Unpronounced heads in relative clauses
(13) a.
The pictures of Marsdeni which hei displays prominently are generally the attractive ones. (Safir 1998: (38a)) b. *Which pictures of Marsdeni does hei display prominently?
(14) a.
I have a report on Bob’s division he won’t like. (Merchant 2000b)
b. *Which report on Bobi ’s division will hei not like? There is also a difference between the relative clause head and material pied-piped internal to the relative clause with respect to Condition C reconstruction. This is shown by (15): (15a) is a case where material pied-piped internal to the relative clause triggers Condition C. (15b) shows that an R-expression in the relative clause head doesn’t trigger Condition C. (15) a. *I respect any writer whose depiction of Johni hei ’ll object to. (Safir 1998: 34a) b. I respect any depiction of Johni hei ’ll object to. The facts from Condition C reconstruction are a challenge to the view that relative clauses allow only the raising analysis. On this view, all relative clauses would be derived by movement of the head from a relative clause internal position. Therefore, relative clause heads would be predicted to behave exactly like other cases of wh-movement with respect to reconstruction. But, this prediction seems incorrect if the facts in (12) to (15) are correct. Therefore, I conclude that at least the raising analysis of relative clauses cannot be the only analysis possible for relative clauses. Safir (1999: 609–614), who discusses reconstructio n into relative clauses and in questions in detail, reaches the opposite conclusion. He cautions that whmovement in questions in some examples doesn’t seem to exhibit a perceivable Condition C reconstruction effect (cf. Kuno 1997; Lasnik 1998), and therefore the generalization about Condition C reconstruction in wh-questions might be more than the core NP vs. modifier distinction of Freidin (1986) and Lebeaux (1988). However, the argument I made here shouldn’t be affected by Safir’s concern: Since the contrast between the head of relative clauses and cases of movement is observed in minimal pairs ((12) to (15)) it is unlikely to be due to some mysterious general property of reconstruction, and therefore should be explained as a difference between relative clause heads and all other cases of movement. Therefore, I conclude that some analysis other than the raising analysis must be possible for relative clauses.5 Now consider arguments that have been given in favor of assuming the raising analysis of relative clauses to be possible. These are cases where the relative clause head must be interpreted only in an internal position. One such case is binding, as already mentioned in (3). (16a) and (16b) are two examples from the literature,
Uli Sauerland
(16c) shows that also a bound variable pronoun in the relative clause head can be bound by a quantifier in the relative clause.6 (16) a.
The interest in each otheri that John and Maryi showed t was fleeting. (Schachter 1973: 43a) b. Une photo de luii que Jeani avait donnée à Marie a été retrouvée hier. (Vergnaud 1974: 256) A photo of him that John has given to Mary has been found again yesterday. c. The book on heri desk (John found out) every professori liked best t concerned model theory.
A second kind of evidence in favor of reconstruction of the relative clause head comes from idiom chunk interpretation. Schachter (1973) attributes this argument to Brame (1968), which I was unable to consult. The fact is that the relative clause head can be part of an idiom chunk with material surrounding the trace. In (17), this is exemplified using the idiom chunks make headway and take pictures. (17) a. The headway John made proved insufficient. b. All the pictures John took showed the baby. This argument based on idiom chunks is in fact more decisive than the one based on binding: For binding, for example Sternefeld (1998) and Sharvit (1999) develop a semantic mechanism that can bring about binding relationships in the absence of c-command. However, the particular mechanism proposed could not bring about idiom chunk interpretation and, I suspect, for principled reasons could not since the parts of the idiom chunk can’t be assigned an interpretation independent of each other which is then brought together by a semantic mechanism. A third case where the relative clause head seems to be interpreted internally is scope interpretation in (18). Namely, it seems that the relative clause head can take scope in a position internal to the relative clause.7 (18) a.
No linguist would read the many books Gina will need for vet school. (need >> many) b. Mary shouldn’t even have the few drinks that she can take. (can >> few)
We’re therefore led to the conclusion that in some cases the head must be interpreted internal to the relative clause, but not in other cases. A potential matching structure is illustrated in (19). Internal to the relative clause an empty operator undergoes movement, and creates semantically an open λ-predicate. This is then intersected with the predicate the head expresses.
Unpronounced heads in relative clauses
(19) a. the picture of Johni hei likes
b. the picture of Johni x hei likes tx (matching)
A potential raising structure is illustrated in (20). Here, the head of the relative clause itself starts out in the relative clause internal position. It moves to the head position, where it also is pronounced. At LF, however, the head is interpreted only in the relative clause internal position, where the variable expressed by himself is bound. (20) a. the picture of himself everybody likes b. the Op everybody likes [picture of himselfi] (raising)
The interpretation of a structure like (20) is by no means straightforward. I take up this question in Section 5. Since my main interest in this paper, however, is the matching structure, it will be sufficient to keep a vague paraphrase of the interpretation intended with representation (20b) in mind. For (20a), such a paraphrase is the function that maps everybody to a picture of himself that he likes.
. Condition C with raising relatives In the previous section I argued that both the raising and matching analysis are required in the analysis of English relative clauses. To explain the obviation of Condition C, I invoked the matching analysis, while I invoked the raising analysis to explain the possibility of binding. The claimed structural ambiguity predicts that Condition C effects should be observed when the raising analysis is forced. This section demonstrates that this prediction is borne out. I show that each of the three factors which I claimed to require the raising analysis induces a Condition C violation when the relevant test is constructed. First consider variable binding. In all examples in (21) and (22), variable binding forces the raising analysis, because the pronoun her is bound by the quantifier every girl only in the relative clause internal position. In the examples (21a) and (22a) the relative clause head contain an R-expression in addition to the bound variable. Furthermore, a pronoun c-commands the trace position in the relative clause. The fact that this pronoun cannot be coreferent with the R-expression in the relative clause head, I claim is due to a violation of Condition C. This is corroborated by the absence of such an effect in (21b) and (22b), where R-expression and pronominal are interchanged. (21) a. *The letters by Johnj to heri that hej told every girli to burn were published.
Uli Sauerland
b. The letters by himj to heri that Johnj told every girli to burn were published. (22) a. *A review of Johni ’s debate with herj that hei wanted every senatorj to read landed in the garbage instead. b. A review of hisi ’s debate with herj that Johni wanted every senatorj to read landed in the garbage instead. This result, of course, resembles Lebeaux’s (1992) observation concerning questions in (8) above. One question that I will put aside here is whether reconstruction to intermediate positions in a relative clause can be established. Sauerland (1998a) presents one datum that I take to support reconstruction to intermediate positions. From the point of view of the raising vs. matching dichotomy, such cases would have to analyzed as raising followed by matching in an intermediate position. The use of idioms is another way to enforce the raising analysis. As Munn (1994) already observes, the prediction that Condition C effects reemerge is borne out. This is shown by the pairs in (23) and (24). In (23), the idiom chunk take picture requires the noun picture to be interpreted in the trace position inside of the relative clause. Therefore, the Condition C violation triggered by the R-expression Bill in this position in (23a) is expected. Again, (23b) shows that coreference is possible when R-expression and pronominal element are exchanged.8 (23) a. *the picture of Billi that hei took. b. the picture of himselfi that Billi took.
(Munn 1994: (15c))
The contrast in (24) is analogous to that in (23). Again, material in the head triggers a violation of Condition C in (24a) confirming the claim that, on the raising analysis, Condition C violations are observed in relative clauses. (24b) provides the relevant contrast, when R-expression and pronoun are exchanged. (24) a. *The headway on Maryi ’s project shei had made pleased the boss. b. The headway on heri project Maryi had made pleased the boss. The third way of forcing the raising analysis was the narrow scope interpretation of material in the relative clause head. In (25), I show that narrow scope of many in (25a) and few in (25b) seems to cause a Condition C effect in the expected fashion. (25) a. *The many books for Ginai ’s vet school that shei needs will be expensive. (need >> many) b. *The few coins from Billi ’s pocket hei could spare weren’t enough for all the needy. (could >> few) Carlson (1977a) and Heim (1987) argue that there are other properties that correlate with the raising vs. matching distinction. In particular, they argue that only
Unpronounced heads in relative clauses
a raising analysis is possible if the relative clause internal trace appears in a thereexistential construction. The data in (26) and (27) indicate that a correlation with Condition C is again confirmed. (26) a. ??I visited all the relatives of Maryi ’s that shei said there are t left. b. I visited all the relatives of heri ’s that Maryi said there are t left. (27) a. *It would have taken us all year to read the letters for Johnj hej expected there would be. b. It would have taken us all year to read the letters for himj Johnj expected there would be. Finally, Carlson (1977a) and Grosu and Landman (1998) point out that a raising analysis seems to be impossible when the head of the relative clause is the complement of an indefinite determiner. Therefore, variable binding is expected to be impossible with an indefinite determiner. This prediction is corroborated by the contrast in (28): 9 (28) a.
(The/Nearly every) picture of herselfi every girli sent angered the teacher. b. ??(A/One picture) of herselfi every girli sent angered the teacher.
The correlation between definiteness and the possibility of the matching analysis is also corroborated by (29), where the head of the relative clause appears in a context that only allows an existential reading. The proposed correlation predicts that variable binding is ruled out in (29), and that therefore the example is ungrammatical. (29) *On the table, there’s (a/one) picture of herself every girl sent. Taken together the facts in this section lend strong support to the claimed structural ambiguity of relative clauses. We have seen that the obviation of Condition C is not observed once the raising analysis of a relative clause is forced by either binding, idiom interpretation, or scope. Therefore, the absence of Condition C effects in other relative clauses cannot be explained based on the raising analysis. Therefore, both analyses – the raising and the matching analysis – are needed. For the rest of the paper, I say nothing more about the raising analysis. For the matching analysis, however, I argue that the analysis proposed above needs to be modified.
Uli Sauerland
. The internal head in matching relatives The straightforward account of matching relatives, already mentioned in (19), would be to assume that an empty λ-operator binds the trace position as sketched in (30b). (30) a. the picture of Johni hei likes
b. the picture of Johni Opx hei likes tx (matching)
In this section, I present two arguments that there’s a more complex representation of the external head in the internal position.
. Double headed ACD The first argument draws on facts and an analysis of Sauerland (1998a, 2000) concerning a particular English construction, which I call here Double Headed Antecedent Contained Deletion, or shorter Double Headed ACD. I present the argument here in an abbreviated form. First recall that ACD is a form of VP-ellipsis inside of a relative clause where the relative clause head seems to be part of the antecedent VP As illustrated by (31), the structure of ACD is such that the apparent antecedent of the elided VP contains the elided VP itself. elided VP
(31)
Polly visited every town Eric did (visit t). antecedent
Sag (1976a), Larson and May (1990), and Kennedy (1997a) strongly argue that ACD is resolved by invisible A-bar movement – quantifier raising – of a DP containing the relative clause. Therefore, (32) is the LF-representation of (31). elided VP
antecedent
(32) [every town, Opy Eric visited [y]] xj Polly visited [x]
In standard examples of ACD like (31) the head of the ACD-relative is also the DP that undergoes quantifier raising. However, this identity of the two DPs isn’t a necessary feature of the construction. If the two DPs are different, I call this construction Double Headed ACD. Double headed ACD is in many cases ungrammatical (Kennedy 1994), as illustrated by (33a). However, (33b) is grammatical. The
Unpronounced heads in relative clauses
difference between (33a) and (33b) is that, in (33a), the head nouns of the two DPs involved in double headed ACD, the head of the relative clause and the DP that undergoes quantifier raising, are different, while they’re identical in (33b). In Sauerland (1998a), I show that more generally double headed ACD is acceptable if the lowest NP-segments of the two DPs involved are identical (or at least very similar), but not otherwise. Of course, single headed ACD of the type illustrated by (31) always satisfies this identity requirement, because the head of the relative clause is identical to the DP that undergoes quantifier raising. (33) a. *Polly visited every town that’s near the lake Eric did visit t. (Kennedy 1994) b. Polly visited every town that’s near the town Eric did visit t. The contrast (33) follows from the structure for ACD proposed in Merchant (2000b) and Sauerland (1998a), and sketched in (34) for (33a) and the general identity requirement on VP-ellipsis. Consider the two trace positions in (34), the trace internal to the relative clause is marked as [y, lake] and the trace left by quantifier raising is marked as [x, town]. Since one of the traces is part of the elided VP while the other is part of the antecedent, we expect that the identity requirement on ellipsis allows ellipsis in (34) if and only if the content of the two trace positions is identical. If both trace positions have the content shown, this predicts ellipsis to be possible only when the content of the two traces is identical. (34) *every [λx x is near the lake [λ y Eric visited [y, lake]]] λx Polly visited [x, town]. But, the prediction only arises if the relative clause internal trace position has as its content the material of the relative clause head. If the relative clause internal position could be contentless in externally headed relatives, the examples in (33) should all have the same status. In this way the paradigm in (33) argues for the assumption that some material of the relative clause head is represented in the relative clause internal trace position even in the case of matching relative clauses. The identity requirement found in double headed ACD is not found in all cases of an elided VP containing a trace where the binders of the trace and the corresponding trace in the antecedent of the elided phrase differ in their lexical content. Both examples in (35) show this. In Sauerland (1998a), I argue that independent factors, in particular sloppy readings and focus, obviate the identity requirement in such cases. (35) a.
I know which cities Mary visited, but I have no idea which lakes she did. b. The cities Mary visited are near the lakes Bill did.
Uli Sauerland
. Crossover The second argument is based on the contrast in (36) from Safir (1998). He observes that a quantifier in the head of the relative can only bind a relative clause internal pronoun if the pronoun is c-commanded by the RC-internal trace in (36). So, the quantifier anyone in (36a) cannot bind the pronoun he in (36a), but in (36b) the binding relation is possible. (36) a. *Pictures of anyonei which hei displays prominently are likely to be attractive ones. (Safir 1999: (66a)) b. Pictures of anyonei that put himi in a good light are likely to be attractive ones. Example (37) corroborates Safir’s empirical claim. In (37), the quantifier every boy occurs in the relative clause head. It cannot bind the pronoun he in (37a) where the relative clause internal trace occupies the object position and therefore doesn’t c-command the pronoun. In (37b), however, the relative clause internal trace occupies the subject position and therefore binding of the pronoun is possible – or at least, only a violation of the Weak Crossover Constraint. (37) a. *Mary exhibited the picture of every boyi that hei /hisi sister brought. b. Mary exhibited the picture of every boyi that was brought by himi /hisi sister. The contrast in (38) shows that even when a matching relative is forced by Condition C a contrast like in (36) is observed: (38) a. *The Times will generally publish pictures of any womani visiting Clintonj that hej told heri about. b. The Times will generally publish pictures of any womani visiting Clintonj that hej thinks will offend heri . As Safir also notes, the matching analysis with an empty internal head doesn’t predict these contrasts. Consider the representation in (39) for (36a). This representation takes into account that the quantifier anyone must be moved out of the relative clause head and adjoined to the clausal level to be interpretable. This has been suggested for inverse linking by May (1977) and recent work of myself has found empirical support for this assumption (Sauerland 1999). (39) *anyonex [pictures of tx whichy hex displays prominently ty ] are likely to be attractive ones. In the representation (39), the quantifier anyone c-commands the pronoun and therefore binding should in principle be possible. The grammaticality of (39) is
Unpronounced heads in relative clauses
predicted to be comparable to other cases of inverse linking where the inversely linked quantifier binds into the matrix clause. But in fact, (36a) is worse. The contrast in (39) is, of course, reminiscent of similar contrasts with whmovement, as Safir also observes who uses the term secondary strong crossover for these constructions. (40) shows that the wh-phrase whom cannot bind a pronoun that c-commands the trace of the bigger wh-phrase in (40a), while it can bind the pronoun in (40b). (40) a. *Which picture of whomi does hei display prominently? b. Which picture of whomi puts himi in a good light? Assuming the copy theory of movement, the ungrammaticality of (40a) is a strong crossover effect or equivalently following Chomsky (1981) a Condition C effect: In the representation (41) for (40a), the unbound wh-trace ty is c-commanded by hey . (41) whom λy which λx does hey display [x, picture of ty ] prominently It’s desirable to reduce the ungrammaticality of (36a) to Condition C in the same way as was done for (40a). But, this requires the extension of the copy theory to matching relative clauses in some way. If we copy the external head of the relative clause into the internal position, the same explanation is available for Safir’s contrast. (42) *anyonex [pictures of tx whichy hex displays prominently [y, pictures of tx ]] are likely to be attractive ones. In (42), hex c-commands the QR-trace tx in the relative clause. Therefore, (42) violates Condition C just like (41) does. However, this solution seems to undermine the motivation for the matching analysis. The observation that led me to propose that the matching analysis is available for relative clauses in addition to the raising analysis was the absence of Condition C effects. If we now adopt the explanation of the ungrammaticality of (36a) as a Condition C violation based on the representation (42), we prima facie predict Condition C violations to occur more generally. In the next section, I show how this paradox is resolved. I’ll argue that the relationship between the internal and external copy of the relative clause head in representation (42) actually allows slight modifications, which obviate Condition C exactly in the cases where it’s in fact obviated.
Uli Sauerland
. Relative deletion . The proposal The relationship of head and the relative clause internal trace position cannot be a direct movement relationship, because that wouldn’t distinguish matching from raising relative clauses. I therefore propose that the material in the trace position is related to the head not by movement, but by ellipsis. More precisely, I propose that the material internal to the relative clause argued for in the previous section is an elided copy of the material in the external position. To exemplify the proposal look at (43). The relative clause in (43a), I propose, receives the matching analysis in (43b): A silent copy of the head book is the complement of the relative clause operator which as shown in (43b). At LF, therefore this copy is represented in the relative clause internal trace position. (43) a. the book which Susi likes b. the
book
which
antecedent
Susi likes t
elided NP
The ellipsis process hypothesized is quite different from VP-ellipsis. One respect in which it’s different is that ellipsis of the NP in (43) is obligatory, while VP-ellipsis is an optional process. A second difference is that the antecedent of the silent internal head in (43) must be the external head of the relative clause. For VP-ellipsis sites, however, any other VP in the discourse can serve as the antecedent. While the hypothesized ellipsis differs substantially from VP-ellipsis, there is another ellipsis process that behaves very much like the ellipsis postulated in (43): comparative deletion. Bresnan (1973, 1975) and Lechner (1999) argue that comparative clauses involve obligatory deletion of an AP or NP Specifically, the AP or NP is obligatorily deleted that contains the trace of the comparative operator. Consider for examples the comparative clause in (44): according to Bresnan’s proposal the subject position of the than-clause in (44) is occupied by a silent copy of the NP a long whale. However, this silent copy cannot be pronounced in (44). Hence, comparative deletion is obligatory exactly like the hypothesized ellipsis in (43). (44) Ahab saw a longer whale than (*a long whale) was ever seen. Furthermore, Williams (1977a: 102) and Kennedy (1997b) show that, in (45), the antecedent of comparative deletion must be the phrase that is the sister of the comparative operator Opd . Hence, an interpretation of the comparative deletion site as wide isn’t available in (45). Again, comparative deletion behaves exactly like the ellipsis postulated in (43).
Unpronounced heads in relative clauses
(45) The table is wider than this rug is, but this rug is longer Opd than the desk is d, long/ * d, wide (Kennedy 1997b: 154) I introduce therefore the term relative deletion to refer to the process that renders the internal head of matching relatives unpronounceable. (46) Relative deletion: In matching relatives, the internal head must not be pronounced. Furthermore, the external head must be the antecedent of the internal head. Lechner (1999) develops an interesting proposal to account for comparative deletion. His idea is that it involves movement without chain formation. As far as I can see, his proposal can also be adopted to relative deletion, but I leave this for future research.
. Vehicle change In this section, I show that the proposed relative deletion solves the problem noted at the end of Section 3.2. To recall the problem, consider (47). (47) shows the apparent conflict between the Condition C evidence and the crossover evidence above. The motivation of the matching analysis was to explain the absence of Condition C effects in examples like (47a), but in Section 3.2 I argued that the matching analysis must then be modified to account for the appearance of strong crossover effects as in (47). To explain (47b), I proposed that the head of the relative clause is in fact represented in the relative clause internal position in matching relatives. This seems to predict that (47a) should violate Condition C. (47) a.
Pictures of Johni which hei displays prominently are likely to be attractive ones. b. *Pictures of anyonei which hei displays prominently are likely to be attractive ones.
In fact, though, the contrast in (47) is predicted by the proposal that the internal head is an elided copy of the external head. The reason is that ellipsis processes have been argued by Fiengo and May (1994) to allow what they call vehicle change. Specifically, Fiengo and May (1994) argue that an R-expression or wh-trace in the antecedent of ellipsis can correspond to a pronoun in the elided material. One piece of evidence for this proposal are data like (48). In (48a) and (48b), the antecedent of the elided VP contains an R-expression. However, only (48a) doesn’t allow coreference between the pronominal subject of the elided VP and this R-expression.
Uli Sauerland
(48) a. *John likes Maryi and shei does like heri , too. b. John likes the story about Maryi and shei knows he does like the story about heri . The difference between (48a) and (48b) is how deeply embedded the R-expression is in the antecedent VP Fiengo and May (1994) argue that Condition B rather than Condition C determines the possibility of coreference in (49). This follows if the Rexpression in the antecedent can correspond to a pronominal in the elided VP The kind of correspondence relation, Fiengo and May (1994) refer to as vehicle change. I show now that vehicle change is at work in comparative and relative deletion as well, and explains the problem mentioned above. The presence of vehicle change corroborates the proposal that ellipsis of the internal head takes place in matching relatives. Consider first the contrast in (49). It shows that vehicle change is observed with comparative deletion. Again, both (49a) and (49b), contain an R-expression in the antecedent of the ellipsis: the comparative AP and a coreferent pronoun c-commands the ellipsis site. In (49a) where coreference between the pronoun he and the position of the R-expression in the ellipsis is blocked by Condition B and C, coreference is in fact blocked. In (49b), however, where Condition B is not violated, coreference is possible. This is exactly the pattern predicted by vehicle change. (49) a. *Mary is more proud of Johni than hei is proud of Johni /himi . (Lechner 1999) b. Mary is more proud of Johni than hei thinks she is proud of Johni / himi . To explain the absence of Condition C effects in matching relatives, I propose that vehicle change of an NP to an NP-anaphor is also possible. Consider (50) under this assumption. If the internal head of the matching relative clause is a one-anaphor referring to the predicate picture of John is possible as indicated in (50b), no violation of Condition C is expected. (50) a. pictures of Johni which hei displays prominently b. [picture of Johni ]j λx which hei displays [x, onej ] What does vehicle change predict for the crossover example Safir’s (1998)? Consider the relevant part of structure in (51a) (repeated from (36a)). In this example, vehicle change to a one-anaphor is blocked. The reason is that the external head contains a variable: it’s the NP pictures of x. But, there is no constant relation a NP-anaphor could refer to that’s coreferent with the external head-NP pictures of x. Hence, in (51 a) vehicle change of the entire NP to an NP-anaphor is blocked.
Unpronounced heads in relative clauses
(51) a. *pictures of anyonei which hei displays prominently b. *anyone λx [pictures of [x][which] λy hex displays prominently [y, picture of [x]]] Consider now vehicle change of the trace [x] to a pronoun. This is predicted to be possible in (51). But, it would not change the status of (51a), however, since the resulting representation would still cause a weak violation of Condition B as shown by (52), even though Condition C wouldn’t be violated.10 (52) ??Johni displays a picture of himi . However, the possibility of this vehicle change predicts that if the trace is more deeply embedded in the antecedent, such that Condition B isn’t violated, the example should become grammatical. The contrast (53) shows that the crossover effect triggered by the internal head exhibits the locality of Condition B. While (53a) doesn’t allow every boy to bind he, binding is possible in (53b), where the quantifier every boy is more deeply embedded in the head of the relative clause. (53) a. *Mary exhibited the picture of every boyi that hei brought. b. Mary exhibited the picture of every boyi ’s mother that hei brought. Note that the locality restriction exhibited in (53) exactly matches Condition B: While coreference of subject and the pronoun him is impossible in (54a), it’s allowed in (54b). (54) a. *Johni brought a picture of himi . b. Johni brought a picture of hisi mother.
. Conclusions In this paper, I provided new arguments for the claim of Carlson (1977a) that there are two structures for relative clauses: the matching and the raising structure (see also Heim 1987; Grosu & Landman 1998; Bhatt 2000). The difference between the two structures is that on the raising structure the relative clause head is moved to its surface position from a relative clause internal position, while on the matching structure it’s not. I furthermore argued that the structure of matching relative clauses involves a silent internal head, that is phonologically deleted by an operation akin to comparative deletion. The work presented here leads to further question about the syntax and semantics of relative clauses. The matching structure of relative clauses hypothesized above raises questions similar to those that have been raised for comparative deletion. By assumption both of these operations involve an PF-deletion operation, just
Uli Sauerland
as ellipsis does. In particular, one of the arguments in Section 4.2 above was drawn from parallels between ellipsis and relative/comparative deletion with respect to the identity of unpronounced material to the antecedent. However, while ellipsis is an optional processes, relative and comparative deletion must be obligatory in most cases. In this respect, the two processes exhibit a behavior more like movement, where also a copy of parts of the moved phrase must not be pronounced. Therefore, an important question for this analysis is how this difference arises? While I cannot answer this question here, I believe that recent work by Lechner (1999) might lead to an account. Lechner (1999) develops an analysis that assigns to comparative deletion a status on which it is predicted to share some properties with ellipsis, but others with movement. The most important question for the analysis of raising relatives is an understanding of their semantics. If my arguments in Section 2 are correct, the head of a raising relative clause must be interpreted in a relative clause internal position. Heim (1987), Bhatt (2000), Hackl and Nissenbaum (1998), and Grosu and Landman (1998) present proposals for the interpretation of several cases of matching relative clauses. However, none of these proposals is intended for the cases like (55) (repeated from (20)), where a relative clause internal quantifier binds a pronoun in the head of the relative clause. (55) the picture of himselfi everybodyi likes Sharvit (1996 and 1999) is the most comprehensive proposal concerning the interpretation of such structures I’m aware of. She points out that, in sentences like (56), the relative clause internal quantifier that binds a pronominal in the relative clause head can also bind a pronoun in the object position of the matrix clause. (56) The picture of himselfi everybody likes is gracing hisi homepage. Variable binding as in (56) would be expected if the raising analysis of relative clauses allowed quantifier raising to take place from a relative clause internal position to a position above the head of the relative clause. Such a structure is shown for (56) in (57). (57)
everybody the Op ti likes [picture of himselfi] is gracing hisi homepage
On the basis of a structures like (57), an account of the interpretation of raising relatives with a bound variable in head is easier than for the structure in (20) above. The interpretive mechanisms required to account for (57) would be the same as those required for (58). But, (58) doesn’t require any special assumptions about
Unpronounced heads in relative clauses
the semantics of reconstruction since all pronouns bound by everybody are in its scope. (58) For everybody: The picture of himself that he likes best is gracing his homepage. Sharvit argues against this analysis. Her main argument is that quantifier raising to a position outside of a relative clause should be blocked by island constraints. At present, I’m not in a position to answer Sharvit’s objections to the quantifier raising analysis properly. But, one possible line to pursue would be to assume that only matching relative clauses are not islands for quantifier raising.
Notes * I would like to thank Noam Chomsky, Danny Fox, Daniel Hardt, Chris Kennedy, Winfried Lechner, Jon Nissenbaum and the editors for giving me very useful comments on the work presented in this paper. This work was presented in a class at Tübingen University in Winter 1999/2000, at the Second Asian GLOW conference at Nanzan University, and at a workshop on relative clauses at Tel Aviv University, I am grateful to the audiences at these venues for their feedback. I gratefully acknowledge the financial support of a postdoctoral fellowship of the Japanese Society for the Promotion of Science and the hospitality of Kanda University during the 1998/99 academic year when parts of this paper were initially written. . My evidence doesn’t bear on the question which one of different versions of the raising analysis proposed in the literature is correct. I assume the structure in (2a) for concreteness. . There seems to be some interesting speaker variation concerning variable binding in relative clauses. Daniel Hardt (p.c.) points out that (3b) is ungrammatical for him. He also dislikes example (i) from the literature, and finds (ii) slightly degraded. At present, I have no analysis of this variation to offer. (i) The picture of himself which every student hated annoyed his friends. (Sharvit 1999: (8a)) (ii) The relative of his every man loves most is his mother. . Several examples in this paper are drawn from the manuscript (Safir 1998), which are not included in the final paper (Safir 1999). However, Safir (p.c.) reports that the reason for omitting these data were not the judgements, and I have independently checked the judgements reported in Safir’s draft paper with several native speakers. . As is the case for most judgements on Condition C with reconstruction, speakers vary in how strong the effect is perceived to be (compared to Condition C effect with surface c-command). What matters for the argument here is the contrast between the a and b examples, and the judgements indicated refer to the contrast. . Fox (2000a) points out another argument for a matching analysis of relative clauses. Namely, the assumption that late merger of relative clauses are merged late (Lebeaux 1988;
Uli Sauerland
Chomsky 1993) is incompatible with the raising analysis because the relative clause head is merged into the structure before the relative clause itself on this analysis. Given the independent support for the late merger of relative clauses, this amounts to an argument for a non-raising analysis of relative clauses. . Interestingly, many speakers find (16c) degraded when the quantifier binding into the relative clause head is more deeply embedded. Possibly, this is related to the suggestion I make in 5 that such examples involve QR of the quantifier to a position outside of the relative clause and the observation that such quantifier movement often seems to be clause-bound. However, also not all speakers find (16) with deeper embedding unacceptable. The effect might plausibly also be a parsing difficulty. . See Heim (1987) and Bhatt (2000) for more discussion concerning the scope of the relative clause head. . Possibly, a Condition C effect in (23) is expected independently of whether take a picture is an idiom chunk or not. Namely, Chomsky (1986: 167) suggests that the noun picture in similar sentences obligatorily takes an implicit pronominal as its agent argument which refers to the creator of the picture. This implicit pronominal could trigger the Condition C effect in (23). . Ruys (2000: 527–528) observes the contrast in (i) and that is similar to the one in (28). If we assume that (ib) requires a matching structure of the relative clause, the ill-formedness of (ia) would indeed be expected. (i) a. ??Some woman that every boyi loved came out to save himi b. The very woman that every boyi loved most came out to save himi . There’s disagreement in the literature as to whether (52) is a Condition B violation. For example, Haegeman (1991: 212) gives the contrast in (i), and ties it to Condition B (see also Fiengo & Higginbotham 1981: 401–402). (i) a. *Poiroti believes any description of himi . b. Poiroti believes Miss Marple’s description of himi . Reinhart and Reuland (1993: (8)) claim that for similar examples both a reflexive and a pronoun are acceptable. My informants generally found an effect in (52) and (i), though the effect was weaker than a Condition B effect when the antecedent and the pronoun are coarguments of a verb.
Variation at the syntax-semantics interface Evidence from gapping Luis López and Susanne Winkler Some languages move wh/focus-phrases to a clause internal position, which can be identified as Spec,v [Bade (Tuller 1992; Kidwai 1999), Aghem (Watters 1979). Interestingly, it may well be the case that English also allows for wh/focus-phrases in Spec,v. Johnson (1994 and subsequent work) has argued that the second conjunct of a gapping construction is a vP as in John bought books and [vP Mary CDs]. If this argument is on the right track, then it is not surprising that you can have wh/focus-phrases in English gapping examples, as in I wonder which books Mary gave to John and which CDs to Mary or in the topicalized construction The beans, Mary cooked, and the potatoes, Henry (examples by Johnson). The challenge is to explain why these sentences are grammatical while wh/focus-phrases in Spec,v are so categorically ungrammatical in any other context. We propose an analysis based on soft rules that can be ignored if the syntactic context is appropriate.*
.
Introduction
The general topic of this paper is cross-linguistic variation. In particular, we want to investigate what role the interpretive systems play in what appears to be syntactic variation. This line of thinking has not been explored extensively in the generative tradition to date. Two recent papers are our immediate predecessors. Pesetsky (1998) proposes that optimality principles at the syntax-phonology interface account for some variation concerning the structure of COMP, arguing, however, that core computational principles are not soft. Chomsky (1999) discusses the absence of Germanic scrambling/object shift in Romance and attributes it to a parameter at the syntax-semantics interface. The idea that we want to pursue is that interpretive rules at the syntax-semantics interface are universal but ranked, which gives rise to variation. This paper is aimed at showing initial evidence that this research program is feasible.
Luis López and Susanne Winkler
It was not until Chomsky (1981) that syntacticians in the generative tradition had the kernel of a theory of syntactic cross-linguistic variation, the so-called Parameter Theory. The most interesting and influential development of parameter theory is Borer’s (1984) proposal that all apparent syntactic variation is to be found in the lexicon: a consequence of specific properties of functional categories. The contrast between (1) and (2) is a celebrated example, from Emonds (1976): the adverb can stand between the verb and the direct object in French because the verb raises to INFL in French, while it cannot in English (where we could have either downwards affix-hopping or covert movement). This movement is triggered by the features of INFL, which under one version of the theory (Chomsky 1993) are “strong” in French but “weak” in English: (1) Pierre joue doucement t(V) la guitare. (2) Peter slowly plays the guitar. Borer in effect proposes that syntax itself is invariant, a hypothesis that we wish to maintain. However, as we will argue, not all variation can be reduced to lexical variation. Take the examples in (3), from Tuller (1992: 302, 305): (3) a.
gafa-n KE viiriidgwar6n? caught who giant-rat ‘WHO caught a giant rat?’ b. zanee]aa, tl6mp6t6-g D6MAAN. gown-my tore wood ‘My gown, WOOD tore it’ (Tuller 1992: 303)
In Western Bade, as well as in other Chadic languages, wh-phrases and focusphrases appear immediately to the right of the verb (apparently raised to INFL).1,2 It is commonly assumed that wh-phrases and contrastive foci appear in Spec,C in English and many other languages, (see Drubig 1994 for arguments that apparent in situ contrastive focus in English actually involves covert movement to Spec,C). However, it seems that wh/focus-phrases in Western Bade must appear on the left edge of the verb phrase (Tuller 1992; Kidwai 1999).3 We suggest that this crosslinguistic difference lies in the syntax-semantic interface and not in properties of functional categories. In Section 4, we discuss another argument aimed in the same direction taken from Chinese focus movement. In Sections 5–7, we change gears and look more closely into the structure of interpretive rules. The crucial piece of evidence that we use is English gapping, exemplified in (4): (4) a. John bought many books and [α Peter many pencils]. b. Bill asked which books we gave to Mary and [α which CDs to Jane].
Variation at the syntax-semantics interface
The constituent α has generally been considered a full clause (Ross 1967). However, Johnson (1994, 1997a) has provided evidence that α is actually a verb phrase. If so, then wh-phrases and focus-phrases in English can appear on the left edge of the verb phrase, making it look more like Western Bade than it seemed at first sight. This will be discussed in Section 8 and a provisional hypothesis, based on ranked rules, will be presented. But first we need to lay out our main assumptions, which we undertake in the following section.
. Assumptions We adopt the main assumptions of the Minimalist Program (MP), particularly those in Chomsky (1995, 2000). Within the MP, what is usually referred to as the module syntax is re-christened as the Computational System of Human Language (CHL ). Two operations form part of CHL : merge and agree. Merge involves selecting two items α and β and combining them to form a set. Additionally, either α or β is to be chosen to be the label for the set. Selection may take place from among items in a numeration – an indexed array of lexical items – or one of the selected items may be a structure. The resulting basic clause structure for a transitive sentence is as follows, omitting items in specs, etc.: (5)
CP TP
C T
vP v
VP
Agree is triggered by a functional category that has a [–interpretable] feature. This feature needs to match and agree with a [+interpretable] version of the same feature in a lexical item. The lexical item itself has a [–interpretable] feature that signals its availability for agreement. Thus, the functional head can probe in its c-command domain until it finds a term with matching features. Displacement occurs if the functional category has an additional feature (strength in Chomsky 1995; EPP in Chomsky 2000) that can only be satisfied if the agreeing term is pied-piped to its spec. Having set up the playing field in these sparse terms, we can ask the question originally posed at the beginning of this article: where does language variation come from? We can discard off the bat the possibility that the operations them-
Luis López and Susanne Winkler
selves can be subject to variation. In other words, merge, agree and move/pied-pipe will take place when they must and their specifications are universal. For instance, it would be highly unlikely that merge in one language would involve two terms while in another it would involve three; or that movement in one language would be triggered by [–interpretable] features while in another language it would be triggered by some other mechanism. However, we can pursue Borer’s proposal, and locate a source of language variation in the feature structure of a functional category: so a certain movement takes place in language A and not in language B because the relevant functional head in A has a certain feature but the equivalent in B does not. This provides an analysis of the English/French contrast in (1) and (2). Consider now wh-movement as analyzed in Chomsky (2000). In his analysis, a head H1 must have a [–interpretable] feature, call it [f1], that matches a [+interpretable] feature in the wh-phrase.4 Additionally, the wh-phrase must have another [–interpretable] feature, call it [wh], that makes the wh-phrase visible for agreement/attraction. For simplicity, we assume that [f1] triggers movement. For Chomsky, H1 is identified with C, but, as we have seen above, wh-phrases do not move to the left periphery in some languages: in Western Bade, wh/focus phrases clearly make a shorter movement. Therefore, it seems that other heads can have the attracting [f1] feature. Let’s see if we can extend this sketchy framework to focus movement. Presumably, there should be a head H2 which must have a [–interpretable] feature [f2] which matches a [+interpretable] version in the focus phrase. Additionally, the focus phrase must have a [–interpretable] feature that signals its receptivity to agree, call it [foc]. Interestingly, it has been argued that at least in some languages the head that licenses focus-phrases is also the head that licenses wh-phrases, so that wh-phrases and focus-phrases may all end up competing for the same position, unless a language allows for multiple instances of Spec,C. For instance, that seems to be the case in Hungarian (Horvath 1986), Basque (Ortiz de Urbina 1987), Spanish (Zubizarreta 1998), Chamorro (Chung 1998) and many others. It also seems to be the case in Western Bade, since focus-phrases and wh-phrases can be found in the same spot. We will adopt then the working hypothesis that wh-phrases and focusphrases are lifted to the checking domain of the same head, although the head itself may vary from language to language. In English, we assume that wh/focus-phrases move to Spec,C. More crucially for our purposes, we would like to claim that one and the same feature triggers wh-movement and focus-movement. In other words, the [–interpretable] [f1] feature that agrees with a wh-phrase and the [–interpretable] [f2] feature that agrees with a focus-phrase are one and the same. This entails that the corresponding [+interpretable] features of the goal must overlap to a considerable extent – not a surprising conclusion, we believe, if both are variants
Variation at the syntax-semantics interface
of the feature [Det]. We present two pieces of evidence in favor of the claim that [f1]=[f2]. First, focus-phrases and wh-phrases create Relativized Minimality effects for each other (see Lasnik & Saito 1992: 81, 96). This is exemplified in (6): (6) a. *The BOOKS Peter said when he didn’t buy t t . b. *When did Peter say that the CDs he didn’t buy t t ? In the MP, Relativized Minimality has been reconceptualized as the Minimal Link Condition (MLC), which became part of the operation Attract in Chomsky (1995: 267). The import of the MLC is that if a head H probes in its c-command domain for a feature [f], then it will agree and attract the first instance of [f]. A probe cannot skip a term with matching features in order to look for a lower term. The MLC is schematized in (7): (7) *[f] > [g] > t[f] where a > b means a c-commands b and [f] and [g] are features of the same type. If the MLC is to be adopted, then it is clear that the ungrammaticality of (6) can only be explained if we assume that the feature that triggers movement of wh-phrases and the feature that triggers movement of focus-phrases are one and the same. The second piece of evidence comes from the analysis of Chamorro whmovement and focus-movement developed by Chung (1998, and earlier work). Chung shows that wh-movement triggers agreement on C and on the lexical verb; interestingly, she shows that focus movement triggers exactly the same sort of agreement (a phenomenon that is not unique to Chamorro, see the references in her book). This we take to be evidence that the feature that triggers focusmovement is the same one that triggers wh-movement.5 In (8), we show two examples indicating nominative wh-agreement on the verb with italics: (8) a.
Haywe fuma’gaswe we kareta. who agr-wash the car ‘Who washed the car?’ (Chung 1998: 236) b. Mañ-odda balutan ya PATGUN humuyung. find bundle and.then child agr-go.out ‘They had found a bundle and A CHILD came out of it’ (Chung 1998: 268)
The last issue to be discussed in this section is interpretation. We adopt the simplest possible model: syntax interfaces with various interpretation modules which, quite naturally, are sensitive to syntax. In other words, the interpretation modules have
Luis López and Susanne Winkler
rules, call them INT, that assign certain interpretations to terms located in certain syntactic positions. For instance, in English, the interpretation module can read that there is a constituent in Spec,C which bears a wh-morphology; depending on the structure, the wh-phrase may get a question, relative or pseudo-question interpretation. Likewise, a constituent in Spec,C can also be interpreted as focus by the relevant module. As we will argue, if the constituent is not in “the right place” at the end of the derivation, the interpretive modules can’t read it. So, if an English wh-phrase is not in Spec,C, it cannot be interpreted. What “the right place” is is what is subject to language variation. The model of grammar that we have can be schematised in the following manner: (9)
lexicon
→
CHL
→
performance
We have a vocabulary of lexical and functional items that combine forming complex structures that must then be interpreted by a number of performance modules: a phonetic module (PF), a logico-semantic module (LF), possibly more. Where is linguistic variation to be found? The combinatory mechanism is invariant. We know that at least some syntactic variation can be traced to the lexicon, but, as we will argue, not all. Instead, we argue that some syntactic variation is to be traced to the performance modules. The structure of the ensuing argument is as follows: we identify a syntactic difference between two languages and show that it can’t be due to properties of a functional category. By Modus Tollens, the culprit must be an interpretive module.
. Wh/focus-movement in Western Bade As we mentioned in the introduction, wh-phrases and foci appear to the immediate right of the verb in Western Bade. Here we repeat the relevant examples for the reader’s convenience: (3) a.
gafa-n KE viiriidgwar6n? caught who giant-rat ‘WHO caught a giant rat?’ b. zanee]aa, tl6mp6t6-g D6MAAN. gown-my tore wood ‘My gown, WOOD tore it’
Variation at the syntax-semantics interface
Tuller argues that the main verb raises to INFL. Further, she argues that there is an A -position in the VP area where wh-phrases and foci are licensed. We assume that wh-phrases and foci in Western Bade are in Spec,v (see also Kidwai 1999: 234):6 TP
(10)
T
SU T+v
vP
wh/focus
v v
t(SU) t(v)
VP
Thus, in English wh-phrases and foci move to Spec,C and in Western Bade they move to Spec,v. We can consider two hypotheses to account for this difference. The first one is a traditional Principles and Parameters formulation: (11) Hypothesis 1: the difference involves a syntactic parameter. a. Feature [f] that triggers movement of wh/focus is in v in Western Bade. b. Feature [f] that triggers movement of wh/focus is in C in English. In other words, (11) postulates that the morpho-syntactic properties of v and C can vary. v is set up in such a way in Western Bade that it includes [f] in its feature matrix while in English [f] is in the feature matrix of C. The second hypothesis suggests that the difference lies not in the lexicon but in how structures are interpreted:7 (12) v can have [f] in English and Western Bade. Hypothesis 2: the difference involves interpretive rules. a. Spec,v receives INT in Western Bade. b. Spec,C receives INT in English. INT should be understood not as a single rule but as a family of rules which take care of wh-phrases and foci and assign them an interpretation. Application of INT can be triggered by the checking of the [f] feature. In the following, we present an argument that hypothesis 1 is false and that English v does have [f]. Since we do not see any alternatives, we will conclude that hypothesis 2 must be right. The argument can be summarized as follows: there is
Luis López and Susanne Winkler
evidence that in English wh/focus-phrases do move to Spec,v. Consequently, the difference between English and Western Bade does not involve availability of [f]. Chomsky (1986a) already hypothesized that wh-phrases adjoin to the verbal phrase on their way to Spec,C (see also Chomsky 1995: 302 and 2000: fn49). More recently, Fox (2000b: Ch. 5) has confirmed Chomsky’s intuition.8 Consider the sentence in (13): (13) [Which of the papers that hei wrote for Mrs. Brownj ] did every studenti get herj to grade t? In (13), the wh-phrase must reconstruct so that the variable, here the pronoun hei , is c-commanded by its binder, the quantifier every student i . The question is to which position should the wh-phrase reconstruct? Reconstruction to the original position t causes a violation of Binding Theory: the pronoun herj would ccommand Mrs Brownj , in violation of Condition C. However, the wh-phrase must reconstruct somewhere, so that the quantifier can bind the pronoun. It follows that the wh-phrase must reconstruct to an intermediate position, higher than the object and lower than the subject. Notice that the same result obtains for focus topicalization in (14): (14) At least ONE of the papers that hei wrote for Mrs. Brownj did every studenti get herj to grade. Let’s assume, essentially following Fox (2000b) and Chomsky (1999), that the whphrase has stopped at the edge of the vP. Thus, we have an A -position in the vP edge for wh-phrases and focus-phrases. This position can’t be the final landing site for A -movement. Notice that since the focus/wh-phrase must reconstruct in Spec,v, we can’t simply maintain the idea that intermediate traces are deletable and irrelevant at the interfaces. But, how did the wh-phrase or focus-phrase get to stop at Spec,v? The inevitable conclusion is that v must have [f] to trigger movement of wh-phrases and foci to Spec,v in English. Consequently, the difference between English and Western Bade does not concern the morpho-syntactic properties of v. Therefore, it must involve the module that interprets structures (hypothesis 2). In other words, wh-phrases and focus-phrases in English can’t be interpreted in Spec,v.9,10
. Focus in Chinese In this section, we use a different data base to provide another piece of evidence that interpretive requirements give rise to surface syntactic structures. The structure of this argument is the following: recall that we concluded that movement of wh-
Variation at the syntax-semantics interface
phrases and focus-phrases is triggered by the same feature [f]. If a head – say v – has [f] in a certain language, then we should be able to find both foci and whphrases in Spec,v. If we find one sort of phrase but not the other, we must conclude that this state of affairs can’t arise because of presence or absence of [f], which would affect both wh and focus-movement. Again, given our paucity of choices, INT should be guilty (and therefore, INT is probably a cover term for two different interpretation rules, as suggested above). As a matter of fact, focus and wh-movement are dissociated in a number of languages. Take Mandarin Chinese. In this language, a position between the subject and the verb is the landing site of focus constituents (Tsai 1994; Gasde 1998; Zhang 2000): (15) Wo ZHE ben shu bu yao. I this CL book not want ‘It is this book that I don’t want’
CL=classifier (Gasde 1998: 59)
Following the standard assumption that the Chinese verb does not move to I overtly (Tsai 1994), this focus position must be Spec,v. Interestingly for our purposes, wh-phrases can’t occupy this same position; instead, as is well known, they must remain in situ: (16) a.
Akiu mai-le shenme? Akiu buy-prf what b. *Akiu shenme mai-le? ‘What did Akiu buy?’
(Niina Zhang, p.c.)
The sentence in (15) shows that Chinese v has [f]. So, the ungrammaticality of (16b) can’t be caused by unavailability of [f]. Where does this ungrammaticality come from? The explanation again must come from the interpretive modules. Let’s assume that in Chinese wh-phrases can’t be interpreted in Spec,v but are interpreted in Spec,C, as in English (further assuming covert raising of wh-phrases as in Huang (1982); Pesetsky (2000: 92), among many others). Moreover, we need to build on a well known generalization: in-situ constituents are ambiguous, since they can be interpreted in any appropriate position that c-commands them; on the other hand, constituents that have moved overtly must be interpreted where they are found. Take for instance the examples in (17): (17) a. Who remembers what John bought? b. Who remembers where John bought what? Let’s focus on ‘what’. In (17a), ‘what’ must be interpreted where it is, whereas in (17b), it can be interpreted as if it were in the matrix CP, giving rise to pair-list readings (a point originally made by Baker 1970). Additionally, notice that ‘where’
Luis López and Susanne Winkler
can’t give rise to pair-list readings. One (sketchy) way of analyzing this could go along the following lines: movement is either entirely overt or entirely covert, but a constituent cannot take an overt step and then a covert one.11 Let’s go back to Chinese. If the Chinese wh-phrase stays in situ, it raises covertly to Spec,C via Spec,v. However, if the Chinese wh-phrase moves overtly to Spec,v, it cannot move covertly to Spec,C. Consequently, the wh-phrase is not in a position where it can be interpreted at LF, and the sentence is ungrammatical. To conclude, the possibility of overt movement of focus-phrases but not of wh-phrases in Chinese provides evidence that some cross-linguistic variation must be traced to the interpretive modules.
. Midway conclusions Let’s recapitulate what we have so far. Our starting point is a model of grammar as in (9), and the original question is what the sources of cross-linguistic syntactic variation are. We immediately discarded CHL as a possible locus of variation. CHL consists of the operations merge and agree (possibly with pied-piping) and its properties should be fixed. One posible source of variation is the lexicon and, in particular, features of functional categories that may or may not trigger movement. This line of research has been fruitfully explored since Borer (1984). We have proposed here that we should also explore the possibility that the rules of interpretation can be subject to variation. In particular, we have argued that the difference concerning wh/focus-movement between Western Bade and English cannot be traced to a property of a functional category and that, therefore, it could plausibly be attributed to interpretation. Likewise, we have proposed that if a language has different surface positions for focus-phrases and wh-phrases, this could be traced to a property of the interpretive module that interprets them. The parameter that distinguishes Western Bade from English has been provisionally summarized as in hypothesis 2 (12). Soon we are going to revisit (12) and find that it needs refinement. What we are going to find is that in gapping constructions, Spec,v can be a place where foci and wh-phrases can be interpreted, in English. This will lead us to reconceptualize (12) in terms of soft, ranked rules. But first we have to clarify what gapping is.
. Gapping A sentence contains a “gap” if the main verb (and possibly other material) is missing, while two other constituents are left overt. One of the overt constituents is
Variation at the syntax-semantics interface
typically (but not necessarily) the subject of the clause (missing material appears in strike-thru): (18) John reads War and Peace and Peter reads The New York Trilogy. (19) John does not read War and Peace or Peter reads The New York Trilogy. We refer to the overt constituents in the second clause as remnants. A remnant must be in a contrastive relation with a constituent in the first clause, which we call the correspondent (more on this in Section 7). It is customary to start a discussion of this phenomenon by referring to Ross’ (1970) seminal article. According to him, the gapped clause is derived from a complete sentence, from which the elements in strike-thru have been deleted. This approach has been adopted in a number of later works (Pesetsky 1982; Kim 1997; Tran 1999; among many others). However, the work of Oehrle (1987), Siegel (1987) and Johnson (1994, 1997a) shows that the Rossian approach has serious problems. Let’s look at Oehrle’s evidence. Consider the sentences in (20) (20) a. Mrs. J. can’t live in Boston and Mr. J. in L.A.. b. Mrs. J. can’t live in Boston and Mr. J. can’t live in L.A.. c. Mrs. J. can’t live in Boston or Mr. J. can’t live in L.A.. As Oehrle explains, if (20a) were derived from (20b), we would expect it to mean [not p & not q]. Instead, it means [not (p & q)], which is equivalent to [not p or not q], that is, (20c). Consider now (21): (21) a. Mrs. J. can’t live in Boston or Mr. J. in L.A.. b. Mrs. J. can’t live in Boston or Mr. J. can’t live in L.A.. c. Mrs. J. can’t live in Boston and Mr. J. can’t live in L.A.. If (21a) were derived from (21b), it would mean [not p or not q]. Instead, what it means is [not (p or q)], which is equivalent to [not p & not q], that is, (21c). The moral of the story is that in (20a) and (21a), it is clear that the negative word in the first conjunct takes scope over the second conjunct. Assuming that semantic scope is syntactically determined, it follows that negation in the first conjunct c-commands the second conjunct – a configuration that does not follow from Ross’ approach. On the basis of this evidence (as well as some constituency data that we skip for space constraints) Johnson has elaborated a theory of gapping in which there is no sentential coordination + deletion. Instead, we have VP coordination + Across The Board raising of the verb to T. The resulting structure of a gapping construction looks in essence like (22):12
Luis López and Susanne Winkler TP
(22)
T
John reads
vP vP v
t t
vP
conj
v
Peter W&P
t
the NYT
We think this sort of analysis can provide an answer to a very puzzling question that has so far not been addressed. Consider the following Japanese sentence with a gap (taken from Abe & Hoshi 1999): (23) John-ga Bill sosite Mary-ga Susan-nituite hanasita. John-nom Bill and Mary-nom Susan-about talked ‘Mary talked about Susan and John about Bill.’ Abe & Hoshi (1999) takes this to be an instance of gapping (and not right node raising), and we follow them in this. Now the question is why the gapped verb is the one in the first conjunct and not the second. Notice that the directionality of anaphoric relations in Japanese discourse is left to right. As we can see in (24), a subject pronoun in the second conjunct can be coreferential with a DP in the first conjunct, but a subject pronoun in the first conjunct can’t be coreferential with a DP in the second conjunct: (24) a.
John-ga hayaku de-te kare-wa eki-ni hasit-te itta. John-nom early left-and he-top station-to ran-and went ‘John left early and he went to the station.’ b. *Kare-ga hayaku de-te John-wa eki-ni hasit-te itta. he-nom early left-and John-top station-to ran-and went (Kunio Nishiyama, p.c.)
If gapping is the result of CP-coordination and subsequent deletion, as in the classic Rossian analysis, this difference between Japanese and English is unaccounted for. But if it is the result of VP coordination + ATB movement of the verb to T, the result is exactly what we would expect in a language in which complements are located to the left of heads:13
Variation at the syntax-semantics interface
(25)
TP vP vP t
conj
T vP t
However, Johnson’s approach has some problems that he discusses. The first one is case assignment to the second conjunct subject in the absence of an inflectional node. Here, he follows Zoerner (1995) in assuming that the conjunction can be a case assigner. A second problem is how or under what conditions the subject of the first conjunct can raise to T without violating the Coordinate Structure Constraint (CSC) (Ross 1967). To this, Johnson replies that A-movement is not subject to the CSC. Thus, assuming an analysis of ECM constructions in which the subject of the subordinate clause raises to an A-position in the main clause (Chomsky 1993; Postal 1974), it can be seen that the ECM subject in (26) must have raised out of a coordination: (26) Julie has believed Liz for some time [t to be honest] and [Scott to be entertaining]. We will not try to make further contributions to Johnson’s approach here. Instead, we simply retain the idea that gapping involves vP coordination and not CP coordination.
. Wh-phrases and focus-phrases in gapping contexts It has frequently been observed in the literature on gapping that the deleted constituents must be contextually given and that the remnants must occur in a contrastive relation to their correspondents (Kuno 1976: 309; Sag 1976a: 280f; Pesetsky 1982: 640ff; Johnson 1994; Kim 1997; Winkler 1997; Hartmann 2000). This contrastivity requirement on gapping is given in (27): (27) In gapping the absent elements must be given. The remnants must occur in a contrastive relation to their correspondents. Evidence of this contrastivity is that gapping sentences are very natural responses to multiple wh-questions:
Luis López and Susanne Winkler
(28) A: B1: B2: B3:
Who bought what? JOHN bought APPLES, and MARY bought BANANAS. JOHN bought APPLES, and /MARY bought BANANAS. John bought APPLES.
In (28B1, B2), the DPs Mary and bananas contrast with John and apples. In (28B3), the simple answer does not require a contrastive reading. If, however, a coordinating conjunction is present, as in the gapped (pair-list) answers, contrastivity results. Thus, we conclude that the remnants of gapping are or can be foci (see Note 1). Alternatively, the remnants can also be wh-phrases: Bill asked which books we gave to Mary and [VP which records to John]. b. Bill asked which books we gave to which students and [VP which bones to which dogs].
(29) a.
Prima facie, (29) looks like evidence that gapping does involve CP coordination (Pesetsky 1982). However, appearances are misleading. In the negated version of (29a), negation takes scope over both conjuncts, as shown in (30a), providing the reading paraphrased in (30b) (see our discussion above of Oehrle’s examples): (30) a.
Bill asked which books we didn’t give to Mary or which records to John. b. Bill asked which books we didn’t give to Mary and which records we didn’t give to John.
So now we are confronted with a paradox: wh-phrases in English must go to Spec,C, but in (30) the wh-phrase is trapped within the vP, so the sentence should be ungrammatical, but it is not. The solution to this paradox is to assume that in gapping contexts English is more like Western Bade, in as much as Spec,v can be the final landing site of a wh-phrase because it can be interpreted in that position: (31) vP
conj wh
v Su
v
Let’s go back now to the other type of remnant, the one that we have defined as focus. The question now is whether these foci are in situ or have raised vP-internally. Evidence suggests the latter. It has been repeatedly observed in literature that topi-
Variation at the syntax-semantics interface
calization is possible in gapping constructions such as in (32) (Sag 1976a: 265; Hankamer 1979: 151; Wilder 1994: 37; vanden Wyngaerd 1998: 8; Johnson 1997a: 66): (32) a. At our house we play poker, at Betsy’s house, bridge. b. The beans, Harry cooked, and the potatoes, Henry. c. During dinner, my father talked to his colleagues from Stuttgart and at lunch time to his boss. Example (33) shows that topicalization is also possible in negated gapping constructions. Under the observation that gapping of negation is not possible in disjunctions, the parse of (33a) is (33b) and not (33c): (33) A: When didn’t your father address whom? a. During dinner he didn’t address his colleagues from Stuttgart or at lunch time his boss, for that matter. b. or at lunch time he addressed his boss, for that matter. c. *or at lunch time he didn’t address his boss, for that matter. Notice that the word order in (33a) can’t be obtained if the remnant constituents are in situ. This can be seen in sentence (34) (pronounced with unmarked intonation). (34) *He didn’t address at lunch time his boss. (34) shows that at least the first remnant in (33a) must have raised. Let’s look at another example. An NPI is possible in first position in the second conjunct. This is exemplified in (35). (35) A1: Was your father in a bad mood last night? B: Why? Did he do anything strange? A2: During dinner he didn’t address his colleagues from Stuttgart or at any time his boss, for that matter. Again, word order confirms that the NPI must be topicalized: (36) *He didn’t address at any time his boss. Incidentally, notice that (35) shows again that the negative word of the first conjunct has scope over the second conjunct. Example (37) shows that the NPI “at any time” cannot occur sentence initially in non-gapped sentences. (37) *At any time my father didn’t address his boss. (also bad with inversion) Thus, (35) can only be grammatical if the NPI has been topicalized within the vP. The NPI in Spec,v can be c-commanded by negation in the first conjunct, but if the NPI had been topicalized to Spec,C, no negative word could c-command
Luis López and Susanne Winkler
it. We conclude that in English Spec,v is a landing site for wh-phrases and foci. Consequently, it must also be a position where INT can be assigned. But only in gapping contexts: (38) conj
vP
wh/focus Su
v v
. Discussion In Section 3, we provisionally concluded that what distinguished English from Western Bade was the following interpretation rule: (39) Hypothesis 2: the difference involves interpretive rules. a. Spec,v receives INT in Western Bade. b. Spec,C receives INT in English. However, what we saw in the previous section showed that this conclusion is not entirely correct, since Spec,v can also receive INT in English. The difference between Western Bade and English cannot be a categorical rule like the one in (39), but must be something more nuanced. Apparently, the interpretive component is free to assign INT to the edges of certain projections, but one position is arbitrarily chosen as the unmarked one, and this choice is what is subject to parametric variation. We suggest the following hypothesis: (40) UG:
(i) Spec,v may receive INT. (ii) Spec,C may receive INT.
Parameters: a. Spec,v > Spec,C (Western Bade) b. Spec,C > Spec,v (English) Rules (i) and (ii) are both part of UG. In a structure with both a vP and a CP, they are clearly in conflict. So we need a parameter that makes one position preferable. The symbol ‘>’ can mean “is ranked higher than” or “is less marked than”. Notice that this parameter only kicks in if the structure to be evaluated contains both a vP and a CP: if we have a structure with only a vP or CP, rules (i) and (ii) of (40) do not conflict.
Variation at the syntax-semantics interface
Let’s suggest a way in which (40) could be developed technically. In order to accomplish this, we need to make some assumptions explicit. First, we need to update our coordinate structures. Let’s follow Munn (1993) and take a coordination to be an adjunction structure, as represented in (41): XP
(41) XP
ConjP Conj
YP
Second, we adopt Chomsky’s (2000) idea of deriving syntactic representations by phase. According to him, a subnumeration including a transitive v or a C constitutes a phase, a structure that can interface with the interpretive modules. Moreover, a phase is evaluated at the next phase level (except for the root node that exhausts the numeration). It seems plausible that adjunct structures, including ConjP, form phases;14 arguments in favor of this assumption are that they are opaque to extraction and they can be merged counter-cyclically. An additional assumption of Chomsky (2000) that we adopt is that a [–interpretable] feature is licensed only if it has an effect on output. Now take wh-movement or focus-movement in English, in an ordinary sentence that includes a vP and a CP. We want v to have [f] so it can attract the wh/focus-phrase to Spec,v. With the wh/focus-phrase in Spec,v, INT may or may not apply. Let’s assume it does not apply. The derivation proceeds and now we have a CP. We need the CP to have [f] to attract the wh/focus-phrase to Spec,C. With the wh/focus-phrase in Spec,C, INT may or may not apply. Assume it does. The wh/focus-phrase is interpreted at Spec,C, parameter (40) is respected and the sentence is acceptable. This is represented in (42). If INT does not apply at C, then the wh/focus-phrase will have to keep moving into the next phase. CP
(42) Spec (INT applies here)
→ structure conforms to parameter (40b) C
C[f]
TP T
vP Spec
v v[f]
Luis López and Susanne Winkler
Alternatively, let’s assume that the wh/focus-phrase is interpreted in Spec,v. This is represented in (43). INT is evaluated at the C phase. In English, assignment of INT to v conflicts with parameter (40b) and the sentence is ungrammatical. CP
(43) Spec
→ application of INT to Spec,v is evaluated here. If parameter (40a) then (WB). If parameter (40b) then * (English) C
C[f]
TP T
vP
Spec (INT applies here)
v v[f]
Let’s now take Western Bade. v has [f], so it attracts a wh/focus-phrase. INT may or may not apply. Let’s assume it applies, as in (43). This is evaluated at the next phase, with C in the structure. Assignment of INT to Spec,v passes parameter (40a). C could also have [f], but this would violate the “have an effect on output” condition because the wh/focus-phrase has already been interpreted; so [f] is not licensed in C. If INT does not apply at the v phase level in Western Bade, the derivation may proceed and construct a new phase with a C with [f] that attracts a wh/focusphrase. If INT applies at the C phase, we have a conflict with parameter (40a), and the result is deviant. So INT can’t apply at the C phase and the wh/focus-phrase must keep raising. Let’s now take a gapping example in English. Recall that the empirical issue here is that wh/focus-phrases can be interpreted in Spec,v. We can plausibly assume INT does not apply to the edge of Conj, since we never find sentences like (44):15 (44) *Bill asked which books I gave to Mary [Spec,Conj which records [Conj and to John]] With this in mind, the derivation may proceed as follows: the wh/focus-phrase raises to Spec,v, attracted by [f] in v. INT may or may not apply; let’s assume it does. The next phase is closed at the ConjP level. Parameters (40a,b) do not apply because there is no C. Either of the UG rules (40i) and (40ii) may apply without conflict. In this case, since ConjP dominates a vP, rule (40i) can apply. Thus the expression is grammatical, as shown in (45). Assume INT does not apply to the wh/focusphrase in Spec,v. Since the next phase does not include a C – indeed, it does not include an element that allows for INT to apply – ungrammaticality results.
Variation at the syntax-semantics interface ConjP
(45)
Conj
vP
Spec [INT applies here]
→ application of INT to Spec, v is evaluated here. Parameters (40a,b) are inapplicable. (40i) can apply without conflict. v
v[f]
David Pesetsky (p.c.) points out a potential problem for our approach. Assume a structure like that in (46), both vP being strong phases: (46)
vP v
VP V
vP
In this structure, if both vP are strong phases, then it should be possible to have a wh/focus phrase in the lower Spec,v, according to the logic of our analysis. The question is whether structures like (46) do exist. As a matter of fact, López (2001a,b) has argued that this is precisely the structure of Romance causatives, like the one in (47): (47) Le hizo a Juan jugar al fútbol. cl.dat made dat Juan play to.the soccer ‘She made Juan play soccer.’ According to López’s proposals, a Juan jugar al fútbol would be a vP small clause directly selected by the causative verb. A wh/focus-word in the spec of the lower v is indeed ungrammatical: (48) *Le hizo a Juan qué jugar. cl.dat made dat Juan what play However, causative verbs in Spanish do not accept a wh/focus-phrase in the complement even when the latter is clearly a CP: (49) a.
Le hizo que trajera dos botellas de vino. dat.cl made that bring.subj two bottles of wine ‘She made him bring two bottles of wine.’ b. *Le hizo que cuándo trajera dos botellas de vino. dat.cl made that when bring.subj two bottles of wine
Therefore, the ungrammaticality of both (48) and (49b) can be accounted for in terms of selection: causative verbs do not select for interrogative complements. It
Luis López and Susanne Winkler
would be interesting to see what would happen with a verb that could select for a vP and could also have interrogative complements. For the time being, our analysis remains unfalsified by structures like (47b).
. Conclusion In this paper, we have made two claims: (i) syntactic variation across languages may involve the interpretive modules, (ii) the interpretive rules are universal but ranked if they are in conflict; variation comes from the alternative rankings of those rules. (40) has the flavor of an optimality type of approach (Prince & Smolensky 1993). Following Pesetsky (1998), we claim that ranked constraints play a role in the mapping from CHL to the interpretive systems, while CHL itself doesn’t include soft rules.
Notes * Acknowledgments: We thank the audiences of a number of workshops for their comments on different parts of this paper, specifically the audience of the “Ellipsis and Information Structure” workshop at the ZAS in Berlin, the “Contrast”-workshop at the University of Leipzig, and the UPenn-Tübingen workshop “Linguistic Data Structure: Theoretical, Methodological, Descriptive Issues” at the University of Tübingen. We particularly thank Bernie Drubig, Winnie Lechner, Jason Merchant, David Pesetsky, Kerstin Schwabe and Nina Zhang for comments on this paper. The first author is also indebted to the ZAS for inviting him as a guest researcher in July 2000 and to the University of Tübingen for inviting him in August 2000. Special thanks goes to Kirsten Brock for brushing up the English, and more than that, for being the best and most competent language consultant one could ever wish for. . A note of caution is in order here. Typically, syntax papers mention that a certain constituent is “focussed” without explaining exactly what they mean by that. However, the term “focus” can refer to several different concepts. In the first place, we have the information focus/contrastive focus distinction (Rochemont 1986 and many others), relevant in all grammatical domains. Second, contrastive topic and contrastive focus are sometimes teased apart, but often they are not. Contrasts are all semantically similar. As argued for by Büring (1999) and Vallduví and Vilkuna (1998), both topics and contrastive foci evoke a set of alternatives. Additionally, in the grammars of many languages like English, foci and topics are generated by movement (but in others topics seem to be base-generated in the left periphery). For our purposes, we can put all contrasts together, following Büring and Vallduví and Vilkuna’s intuitions, while information focus should be considered a separate phenomenon. From simplicity, we use the term “focus” to refer to a contrast that evokes alternatives.
Variation at the syntax-semantics interface . There is variation among the Chadic languages concerning the grammar of wh/focus. See Tuller (1992) for details. . A third possibility, that we will not have space to discuss here, is to have wh/focus-phrases in Spec,T, as argued by Zubizarreta (1998) for Spanish, among others. . Chomsky calls [f1] Q, probably suggesting Question. However, not all wh-phrases are questions, so we prefer to use the more neutral [f1]. . Lechner (p.c.) points out that the phenomenon of participle agreement in French might be relevant in the present discussion. Consider the examples (i) and (ii), from Kayne (1989): (i) La table a été repente. (ii) La table que j’ai repente. In (i), participle agreement is triggered in a passive sentence, in (ii) in a relative clause. Could this be taken as evidence that different types of movement can trigger the same type of agreement? No. It could be argued (it has been argued) that in either case movement to the final position goes in two steps, the first one displacing the constituent to Spec,Agr or Spec,Part. This first movement would be triggered by the same feature in both cases. . This is also the structure proposed by Kennelly (1999) for the grammar of wh/focus in Turkish, minus the directionality of the specifier. Horvath (1986) also proposes that wh/focus-phrases are in Spec,V in Hungarian, but the matter is disputed (see Brody 1990, among others). . A third hypothesis could involve a PF rule that spelled-out intermediate traces. We will not pursue this possibility , mostly because we can’t see what purpose such a rule would accomplish. . See also Merchant (in this volume), whose analysis of VP-ellipsis under SAI lends additional support to the claim that wh-movement out of a VP proceeds via adjunction to that VP. . Additionally, in Kikuyu (Clements 1984) and Chamorro (Chung 1998) wh/focus-phrases agree with every verb along the way, although they surface in Spec,C. This can be taken as additional evidence that even if movement of wh/focus-phrases ends in Spec,C, v is also endowed with the feature [f] and wh/focus phrases must stop at every Spec,v. The analysis, however, is somewhat complex and beyond the limits of this paper. . For the argument to be complete, we should show that in Western Bade long wh/focusmovement also ends in Spec,v. Additionally, our argument would be strengthened if we could show that C in Western Bade also has [f]. We have not yet been able to check whether this is the case or not with a native speaker. . This should be relativized to type of movement, so, e.g., one can have overt Amovement followed by covert A -movement, but not overt A -movement followed by covert A-movement (thanks to Winfried Lechner for pointing this out to us). We also abstract away from the wh-expletive phenomenon in German and other languages. . For the time being, we adopt a “naïve” analysis of coordination for simplicity. In Section 7 we make our assumptions explicit.
Luis López and Susanne Winkler . Obviously, (25) can be suitably translated into Kayne’s (1994) antisymmetry framework, without altering the argument. . Uriagereka (1999) has adjunction structures spell-out before merging with the main clause. This seems to us to be an instantiation of the same idea. . On the other hand, Conj may have [f], given ATB extraction. Obviously, this goes beyond the limits of this project, but see Nunes and Uriagereka (2000) for relevant discussion in the context of parasitic gaps.
III. The semantic component and its connection to focus and discourse structure
Ellipsis and the structure of discourse Daniel Hardt I propose that the matching condition on ellipsis be applied in accordance with discourse structure. I consider two versions of this matching condition, and show that both versions have empirical limitations that are remedied by the discourse structure requirement. The effect of this condition can be clearly observed in examples involving multiple ellipsis, where discourse structure plays a key role in determining possible readings.
.
Introduction
Many authors have suggested that ellipsis and other processes of reduction require a certain similarity or parallelism between the reduced clause and some antecedent clause in discourse. In this paper, I will argue that this matching condition on ellipsis must be applied in accordance with discourse structure. This results in a stronger condition on ellipsis: the matching clauses must also participate in a discourse relation. The effect of this condition can be clearly observed in examples involving multiple ellipsis, where discourse structure plays a key role in determining possible readings. In what follows, I begin by considering two types of approaches to the matching condition – the semantic matching approach, exemplified by Rooth (1992a) and Dalrymple et al. (1991), and an approach based on path identity, originally due to Fiengo and May (1994), and modified by Asher et al. (2001). I propose an account in which semantic matching must be applied according to discourse structure. I consider examples in which there are multiple potential antecedents for ellipsis, where discourse relations determine the correct choice of antecedent. Next, I examine several cases of multiple ellipsis involving sloppy identity, where discourse relations limit the possible readings. These examples show that the matching requirement, however defined, must be applied in accordance with discourse structure. I also present arguments that suggest that the semantic matching requirement is preferable, although more research is required on this issue.
Daniel Hardt
. Background: Two approaches to matching . Semantic matching The focus-based condition of Rooth (1992a) is the following: (1) Ellipsis requires that there be some phrase E containing the ellipsis and some phrase ANT in the discourse, such that [[ANT]]∈ F(E). The operator F produces the focus value of any phrase. The focus value of an expression is a set of similar expressions, in which all possible substitutions have been made for focused elements. In cases involving sloppy identity, there is an additional issue, emphasized by Dalrymple et al. (1991): if a focused element e is replaced by a value x, other elements coreferential with e may optionally be replaced with x. Thus we define F as follows, restricting ourselves to cases in which a single element e is focused: (2) F(P) = {P’|∃x.P’ = P[e/x]} (The notation P[e/x] indicates that e is replaced by x, and other elements coreferential with e in P are optionally replaced by x.) The condition requires that the meaning of ANT matches the meaning of some phrase E containing the ellipsis, except for the focused element(s) of E. Consider: (3) John1 loves his1 cat. BILL2 does too. (love his cat) On the sloppy reading, the focus value of the ellipsis clause is: {P’|∃x.P’ = Bill2 loves his2 cat[Bill2 /x]} This is the set of all propositions where Bill2 (and optionally his2 ) are replaced by some other individual. Thus this set includes the proposition: John1 loves his1 cat.
. Path identity An alternative condition is proposed by Fiengo and May (1994), which relies on the following constraint on sloppy identity (4) Path Identity: paths resolving corresponding dependent elements must be identical. Fiengo and May’s condition permits ellipsis if it is contained by a phrase P that is identical to some phrase ANT, except for changes in index. If an element e changes
Ellipsis and the structure of discourse
index from ANT to P, that is permitted only if there is identical material connecting e to its antecedent in ANT and in P. We consider our simple example again: (5) John1 loves his1 cat. BILL2 does too. (love his2 cat) On the sloppy reading, the requirement is that the path connecting John1 with his1 is identical to the path connection Bill2 with his2 . The paths, represented as sequences of categories, are indeed identical, since they are both NP V NP. This condition will often be equivalent to Rooth’s condition, since the antecedent for a sloppy element is often focused, and Rooth’s condition and the path identity condition both tend to place a requirement that non-focused material is identical.1 Below, we will see that there are certain cases where different predictions are made by these two approaches.
. Discourse structure and matching We have seen that the matching condition plays an important role in capturing the facts about ellipsis. However, it is perhaps misleading to describe it as a condition on ellipsis. It is reminiscent of conditions on Parallel and Contrast discourse relations in the literature on discourse interpretation (Asher 1993, Hobbs 1979). From this perspective, the requirement that two clauses match (modulo focused expressions) results from a communicative intention that the two clauses be understood in a parallel or contrastive fashion. This provides some motivation for the matching condition, and suggests that it applies much more generally. Furthermore, this leads to a sharpening of the semantic matching condition – instead of allowing any other clause in the discourse to match the ellipsis clause, there must be a match with a clause that is related to the ellipsis clause. I state the modified condition as follows: (6) Ellipsis must be contained in a constituent E that participates in a felicitous discourse relation with some ANT, such that MATCH (E,ANT). (Asher 1993; Asher et al. 1997; Asher et al. 2001) (7) Match (E,ANT): [[ANT]]∈ F(E) (8) All related discourse constituents must satisfy Match. In what follows, I show that this discourse-based matching condition correctly captures ellipsis facts that are not captured by the simple matching condition. All the examples examined below are examples of VP-ellipsis (VPE). However, the conditions are stated quite generally, to cover all forms of ellipsis.
Daniel Hardt
. Selection of antecedent We consider first cases in which there are multiple potential antecedents for VPE. (9) If [Agnes arrived after John ate] then [BILL didn’t]. *(eat) / (arrive after John ate) I assume that Bill is focused, and that there is a discourse relation between the antecedent and consequent of a conditional – this explains the preference for the main VP antecedent in this case. To see this, consider the reading BILL didn’t eat. Clearly, the antecedent clause Agnes arrived after John ate does not match. Thus the only reading permitted is BILL didn’t arrive after John ate; now the antecedent clause does match. If semantic matching is permitted to consider any antecedent clause, we then fail to rule out the unacceptable reading, since we can select John ate as the antecedent clause, which would provide a match with BILL didn’t. Similarly, in a sentence with a quantified subject, it is natural to assert a discourse relation between the restrictor (subject relative clause) and nuclear scope (matrix VP). This explains the following observations: (10) Everyone who [could have arrived after John ate]Sl [did]S2 (arrived after John ate) / *(ate) (11) The man who ate salmon left this evening. The man who didn’t (eat salmon), left this afternoon. (12) The man who ate salmon left this evening. John did too. (left this evening) (13) Everyone who wanted to leave before he ate salmon did *(eat salmon) / (left before he ate salmon) Without the requirement that matching follows discourse structure, the semantic matching approach would permit either antecedent in all of these examples. Note that the path identity condition has nothing to say about these examples, since it only applies to cases of sloppy identity.
. The many-clause puzzle Consider the following example, in which we have two ellipsis occurrences preceded by a single antecedent clause. Such examples have been discussed at length in the literature (Dahl 1974; Dalrymple et al. 1991; Fiengo & May 1994). (14) John revised his paper, and Bill did too, although the teacher didn’t.
Ellipsis and the structure of discourse
(15) [John revised his paper]Sl , and [Bill revised his paper]S2 too, although [the teacher didn’t revise his paper]S3 . The three clauses participate in the following discourse relations: AND(S1 , S2 ), ALTHOUGH(S2, S3 ) We draw attention to the interpretation of the pronoun his in the interpretation of the two ellipsis occurrences. It is generally agreed that this example has three readings: both strict, both sloppy, and sloppy/strict (sloppy followed by strict). All three readings are permitted by the semantic matching condition, given the discourse relations described above. Also, note that our approach correctly rejects the strict/sloppy reading (strict followed by sloppy). Readings: – All strict: S1 ∈ F(S2 ), S2 ∈ F(S3 ){P|∃x.P = x revised John’s paper } – All sloppy: S1 ∈ F(S2 ), S2 ∈ F(S3 ){P|∃x.P = x revised x’s paper } – Sloppy/Strict: [John revised John’s paper]Sl ∈ F([BILL revised Bill’s paper]S2 ) = { P|∃x.P = x revised x’s paper } [Bill revised Bill’s paper]S2 ∈ F([the TEACHER revised Bill’s paper]S3 ) = { P|∃x.P = x revised Bill’s paper } – *Strict/Sloppy: [John revised John’s paper]Sl ∈ F([BILL revised John’s paper]S2 ) = { P|∃x.P = x revised John’s paper } [Bill revised John’s paper]S2 ∈ F([the TEACHER revised teacher’s paper]S3 ) This example is problematic for many leading accounts of ellipsis: The account of Sag (1976a) incorrectly rules out the sloppy/strict reading. The semantic matching account incorrectly permits the following reading (as pointed out by Fiengo & May 1994, with respect to Dalrymple et al. 1991): (16) John revised John’s paper, and Bill revised Bill’s paper too, although the teacher didn’t revise John’s paper. [John revised John’s paper]Sl ∈ F([BILL revised Bill’s paper]S2 ) = { P|∃x.P = x revised x’s paper } [John revised John’s paper]Sl ∈ F([the TEACHER revised John’s paper]S3 ) = { P|∃x.P = x revised John’s paper }
Daniel Hardt
This reading is correctly ruled out when semantic matching is controlled by discourse relations: since we have ALTHOUGH (S2, S3), we must have MATCH (S2, S3). With the above reading, S2 and S3 fail to match. [Bill revised Bill’s paper]S2 ∈ F([the TEACHER revised John’s paper]S3 ) This example is problematic for accounts such as Sag’s, or that of Fiengo and May, because these accounts posit an ambiguity in pronouns that is associated with their potential to give rise to a strict or sloppy reading. For Sag, a pronoun is either bound or free, and a bound pronoun is required to have a sloppy reading in a subsequent ellipsis. For Fiengo and May, there is a similar ambiguity between what are termed α and β occurrences of pronouns, and the β occurrences are associated with sloppy readings. The many-clause puzzle suggests that there is no such ambiguity associated with strict and sloppy readings, since there is a reading in which a single pronoun gives rise first to a sloppy reading, and then to a strict. We have seen that this well-known example is problematic for many leading accounts: the ambiguity accounts of Sag and Fiengo and May are insufficiently flexible for such examples, while the semantic matching accounts of Rooth and Dalrymple et al. permit too many readings. By taking account of discourse structure, the proposed account avoids both of those failings.
. A variant of the many-clause puzzle The following example is similar to the previous many-clause example, but here the discourse relations are altered. (17) John revised his paper before Bill did, but after the teacher did. (18) [John revised his paper]Sl before [Bill did revise his paper]S2 , but after [the teacher did revise his paper]S3 . Like the previous many-clause example, we observe relations between clauses S1 and S2. In addition, however this example has a relation, after, between clauses S1 and S3. Thus we have the following discourse relations: BEFORE(S1 , S2 ), BUT(S2, S3 ), AFTER(S1 , S3 ) This difference in discourse relations corresponds with a difference in possible readings. Unlike the previous example, this example does not permit the mixed (sloppy/strict) reading.
Ellipsis and the structure of discourse
Readings: – All strict: S1 ∈ F(S2 ), S2 ∈ F(S3 ), S1 ∈ F(S3 ){P|∃x.P = x revised John’s paper } – All sloppy: S1 ∈ F(S2 ), S2 ∈ F(S3 ), S1 ∈ F(S3 ){P|∃x.P = x revised x’s paper } – *Sloppy/Strict: [John revised John’s paper]Sl ∈ F([BILL revised Bill’s paper]S2 ) = { P|∃x.P = x revised x’s paper } [Bill revised Bill’s paper]S2 ∈[the TEACHER revised Bill’s paper]S3 = { P|∃x.P = x revised Bill’s paper } [John revised John’s paper]Sl ∈ F([the TEACHER revised Bill’s paper]S3 ) We can see that the sloppy/strict reading here is ruled out, because S1 and S3 do not match.
. Another multiple ellipsis example Consider the following well-known example (see discussion in Dalrymple et al. 1991 and references therein). (19) John revised his paper before the teacher did, and Bill did too. [[John revised his paper]Sla before [the teacher did revise his paper]S1b ]S1 , and [[Bill did revise his paper]S2a before [the teacher revise his paper]S2b ]S2 too. Here we have the following discourse relations: AND(S1 , S2 ), BEFORE(S1a , S1b ), BEFORE(S2a , S2b ) As pointed out by Dalrymple et al., the semantic matching approach permits the following six readings for this example:2 Readings: 1. JJJJ 2. JTBT 3. JJBB 4. JTJT 5. *JJBJ 6. *JJJB
Daniel Hardt
We find only the first four readings acceptable.3 The first three readings are straightforwardly permitted on the proposed approach. Consider Reading 3: [John revised John’s paper before the TEACHER did revise John’s paper]Sl ) ∈ [BILL revised Bill’s paper before the teacher did revise Bill’s paper]S2 ) = { P|∃x.P = x revised x’s paper before the teacher revised x’s paper } [John revised John’s paper]S1a ∈ F([the TEACHER did revise John’s paper]S1b ) = { P|∃x.P = x revised John’s paper } [Bill revised Bill’s paper]S2a ∈ F([the TEACHER did revise Bill’s paper]S2b ) = { P|∃x.P = x revised Bill’s paper } Since all three related clauses match, Reading 3 is permitted. On the other hand, readings 5 and 6 are not permitted. Consider Reading 6: [[John revised John’s paper]S1a before [the teacher did revise John’s paper]S1b ]S1 , and [[Bill did revise John’s paper]S2a before [the teacher revise Bill’s paper]S2b ]S2 too. This is incorrectly permitted on the semantic matching approach, since [[John revised John’s paper]S1a ∈ F([the TEACHER did revise John’s paper]S1b ) and [John revised John’s paper before the TEACHER did revise John’s paper]S1 ) ∈ [BILL revised John’s paper before the teacher did revise Bill’s paper]S2 ) = { P|∃x.P = x revised John’s paper before the teacher revised x’s paper } This reading is correctly ruled out on the Discourse Structure Approach, since: [Bill revised John’s paper]S2a ∈ F([the TEACHER did revise Bill’s paper]S2b ) Consider now Reading 4: [John revised John’s paper before the teacherƒ did revise teacher’s paper]S1 ) ∈ [BILL revised John’s paper before the teacher did revise teacher’s paper]S2 ) = { P|∃x.P = x revised John’s paper before the teacher revised teacher’s paper } [John revised John’s paper]S1a ∈ F ([the TEACHER did revise teacher’s paper]S1b ) = { P|∃x.P = x revised x’s paper }
Ellipsis and the structure of discourse
[Bill revised John’s paper]S2a ∈ F([the TEACHER did revise teacher’s paper]S2b ) This reading appears to be incorrectly ruled out on the proposed approach, since S2a and S2b don’t match. Asher et al. (2001) argue that the path identity account would correctly permit this reading. Recall, on this approach, the path from dependent item to antecedent must match path from corresponding dependent item to corresponding antecedent (Fiengo & May 1994; Asher et al. 2001). Asher et al. modify path identity so that what is compared is a sequence of steps through Discourse Representation Structures (for details, the reader is referred to Asher et al. 2001). It is clear that dependent items in S2a and S2b do have matching paths. The corresponding dependent items are represented below as z’ and z”’, and their antecedents are determined by the following equations: z” = Kl .x and z”’ = K2 .x’. Each equation represents the following path to the antecedent; represented by Asher et al. (2001) as: UP, UP, DOWN, DOWN. Reading 4 (JTJT): Table 1.
x,y,z
x’,y’,z’
x’’,y’’,z’’
y’’’,z’’’
john(x) revise(x,y) paper(y) of(x,z) z = K1 x
teacher(x’) revise(x’,y’) paper(y’) of(y’,z’) z’ = K2 x’
bill(x’’) revise(x’’,y’’) paper(y’’) of(x’’,z’’) z’’ = K1 x
revise(x’,y’’’) paper(y’’’) of(x’’’,z’’’) z’’’ = K2 x’
K1
K2
K3
K4
This example clearly shows the need for discourse structure to direct the operation of the matching operation. However, while we observed above that semantic matching is more general than path identity, we have seen in this example that path identity correctly accepts Readings 1–4, while semantic matching incorrectly rejects Reading 4. Below, we will see that there is a different perspective which might permit us to continue to rely on semantic matching rather than path identity.
Daniel Hardt
. Two pronoun puzzle We turn now to the two pronoun puzzle. (20) John said he saw his mother and Bill did too. Of the four potential readings, only three are acceptable: 1. 2. 3. 4.
Bill said John saw John’s mother. Bill said Bill saw Bill mother. Bill said Bill saw John’s mother. *Bill said John saw Bill’s mother.
As shown by Fiengo and May, path identity successfully rules out the fourth reading (see Fiengo & May 1994 for details). Semantic matching permits the fourth reading, since we have: John said John saw John’s mother ∈ BILL said John saw Bill’s mother = { P|∃x.P = x said John saw x’s mother } An alternative view of this example is to suggest that “mixed readings” are, in general, degraded. That is, in cases with two or more coreferential pronouns in the antecedent, it is preferred that either all of them switch, or none of them do.4 This suggests that only Readings 1 and 2 would be permitted. However, Reading 3 could also be represented as follows: (21) John1 said he1 saw [his1 mother]2 and Bill3 did too. [said he3 saw her2 ]. In general, it is quite natural to assume that complex NPs in the antecedent might be “collapsed” in the elliptical expression in this way.5 This is arguably an independently motivated ambiguity present in any adequate dynamic semantics for discourse. In such a system, meanings of NPs, VPs, and other categories are added to context, providing antecedents for pronouns in subsequent discourse. These meanings can be added as intensional (context-sensitive) objects, or extensional (fixed) objects. In the example above if “his mother” is added as an extension, it will be fixed in its reference to John’s mother. If it is added as an intension, its referent can vary. Without describing the mechanics of this, we note that an extension will have all referents fixed within it, while an intension allows all referents within it to change to the new context. Since “his mother” is not represented at the ellipsis site, there is only one switching pronoun, and the reading is permitted. This is still not possible for Reading 4; the antecedent for the strict pronoun cannot be collapsed here, since it is “he”. Thus there is no way to represent (4) without having a mixed reading. Consider now an alternative 2 pronoun example:
Ellipsis and the structure of discourse
(22) John said his mother saw him and Bill did too. Here, all four readings are available: 1. 2. 3. 4.
Bill said John’s mother saw John. Bill said Bill’s mother saw Bill. Bill said Bill’s mother saw John. Bill said John’s mother saw Bill.
For Reading 3, “his mother” is added as an intension, while “saw him” is added as an extension. For Reading 4, “his mother” is added as an extension, while “saw him” is added as an intension. The details of this “collapsing” account remain to be worked out, and its predictions must be tested on a wider range of data. If it can be maintained, it would allow us to maintain the semantic matching account, without resorting to path identity. A similar approach could also solve the problem with example (19). Reading 4, could be represented as analogous to the following: (23) John revised his paper before the teacher did [revise teacher’s paper] and Bill revised John’s paper then too. In effect, we collapse before the teacher did [revise teacher’s paper] from the antecedent to then. Now this reading is permitted, since the previously offending relationship has been removed, namely the relationship between the matrix clause and embedded clause in the second conjunct. This clearly is not possible for Reading 6, which would still be disallowed: (24) John revised his paper before the teacher did [revise John’s paper] and Bill revised John’s paper before the teacher did [revise Bill’s paper]. The second before-clause cannot be collapsed to then here, because it doesn’t represent the same occasion as the previous before-clause. However, Reading 5 is now permitted by semantic matching: (25) John revised his paper before the teacher did [revise John’s paper] and Bill revised Bill’s paper before then. We are left with the following picture: semantic matching incorrectly permits the fourth reading on the two-pronoun example, and it incorrectly rules out the fourth reading on the six-reading example. Path identity corrects both of these problems, but it is less general, and perhaps less well-motivated than the semantic matching approach. Furthermore, it appears that both of these problems might result from ignoring an alternative representation for ellipsis occurrences whose antecedents
Daniel Hardt
contain compound expressions within them. Further research is needed to determine which approach is preferable. However the matching constraint is implemented, there is clear evidence that the matching operation must be directed by the structure of discourse. This is shown in sloppy identity examples with multiple ellipsis, as well as conditionals and other constructions with multiple potential antecedents.
Notes . See Rooth (1992a) for some comparison of path identity with the focus condition on ellipsis. . We represent the four readings for the occurrence of the pronoun his. Thus, for example, JTBT represents this reading: John revised John’s paper before the teacher revised the teacher’s paper, and Bill revised Bill’s paper before the teacher revised the teacher’s paper. . This is controversial: for example, Fiengo and May (1994) argue that reading four is unacceptable, or at least degraded, while Dalrymple et al. argue that not only one through four but also five is acceptable. . Fiengo and May suggest that there is such a preference for avoiding mixed readings. Also, this preference would be a consequence of the dynamic semantics system described in Hardt (1999). . This is suggested, in a quite different framework, by Oku (2000).
Correlate restriction and definiteness effect in ellipsis Maribel Romero This paper is concerned with two characteristics of certain elliptical constructions: a restriction on possible correlates (or antecedent phrases) for the remnants of ellipsis, and a definiteness effect that makes non-definite phrases behave semantically as definites in ellipsis sites. Interestingly, these two properties are found only in some ellipsis construals, namely in German reduced conditionals and in sluicing, and they do not apply to the majority of known ellipsis types, e.g., VP-ellipsis and gapping. The paper presents a unified account of the presence or absence of such characteristics across ellipsis types, taking as the key ingredient the effects of focus in ellipsis and its interaction with the general semantics of each construal (i.e., conditionals, interrogative clauses and declaratives clauses).
.
Introduction
Ellipsis constructions in general impose certain conditions on the nearby discourse. A first requirement is the need of an anaphoric source from which the meaning of the elided constituent (E-constituent) can be retrieved. Hence, typically, there has to be an antecedent constituent (A-constituent) that is identical to the elided constituent in the relevant respects: syntactically identical at LF (except for minor details) according to some theories (Williams 1977a; Fiengo & May 1994 for VP-ellipsis; Neijt 1979 for gapping, but see Johnson 1996; Ross 1969; Rosen 1976; Levin 1982 for sluicing; Schwarz 1996, 1998 for reduced conditionals) or semantically identical according to others (Sag 1976a; Jacobson 1992; Hardt 1993 for VP-ellipsis; Merchant 2001 for sluicing). Take the examples (1)–(4), illustrating VP-ellipsis, gapping, reduced conditionals in German and sluicing. In each case, the preceding discourse provides an A-constituent from which the meaning of the E-site (in strikethrough) is reconstructed: (1) VP-ellipsis: A friend of mine likes turnips, but Ms. Matilde doesn’t like turnips.
Maribel Romero
(2) Gapping: Boys like lentils and girls like potatoes. (3) Reduced conditionals in German: Wenn ich wen besuche, dann besuche ich (immer) den Peter. If I somebody visit then visit I always the Peter ‘If / whenever I visit somebody, then Peter / it’s Peter.’ (4) Sluicing: a. Somebody just left – guess who just left.
(Ross 1969)
A second condition that ellipsis constructions share concerns the interpretation of scopal elements and variables. It has been proposed that the ellipsis clause (Eclause) containing the ellipsis and the antecedent clause (A-clause) containing the A-constituent must be parallel with respect to the scope of their operators and the sloppy/strict readings of pronouns, beyond the elided territory (Hirschbühler 1982; Prüst & Scha 1990; Fiengo & May 1994; Fox 1995, 2000b; Tomioka 1995). To see just one example, note that (5) has the parallel readings (a)–(b), but not the non-parallel readings (c)–(d). (5) A first year student admires every professor, and a second year student does , too. a. ‘There is a first year that admires every professor and there is a second year that admires every professor.’ b. ‘For every professor x, there is a first year that admires x, and, for every professor y, there is a second year that admires y.’ c. ‘There is a first year that admires every professor, and, for every professor y, there is a second year that admires y.’ d. ‘For every professor x, there is a first year that admires x, and there is a second year that admires every professor.’ This parallelism requirement finds an explanation in Rooth (1992a) (but see Fiengo & May 1994 for an alternative account). Rooth exploits the fact that the remnant material in the E-clause typically bears focus stress, a fact widely acknowledged in the ellipsis literature (Fox 2000b; Tancredi 1992; Tomioka 1995; among many others). For example, Ms. Matilde and the negative polarity receive focus pitch in (1), and the remnants girls and potatoes in (2) are stressed as well. According to Rooth, focus on the remnants requires that the A-clause and the E-clause match semantically, differing only in the denotation of the focused expressions. This derives the scope and pronoun parallelism automatically.1 These two properties are common to ellipsis constructions in general. In this paper, we will be concerned with two new requirements that only some ellipsis types share. We will examine two discourse conditions that are required in reduced
Correlate restriction and definiteness effect in ellipsis
conditionals and sluicing and that are obliterated in other elliptical constructions like VP-ellipsis and gapping: (i) the restriction on possible correlates of the remnant(s), and (ii) the definiteness effect. As we will see, these conditions make reduced conditionals and sluicing appear rather idiosyncratic in comparison to other ellipsis types. In fact, analyses have been developed in the literature that apply only to reduced conditionals or only to sluicing and that cannot be extended to other kinds of ellipsis. The goal of this paper is to explain in a uniform manner why such restrictions apply to reduced conditionals and sluicing and why they do not apply to other ellipsis construals. I will argue that the key factor is the focus on the remnant. The felicity conditions of focus interact with the semantics of conditionals and questions to yield the aforementioned restrictions, whereas no such effect arises from the interplay of focus and the semantics of coordinated declaratives. In sum, this paper presents a unified account of the existence and lack of the aforementioned restrictions across ellipsis types, capitalizing on a well-known feature of ellipsis, namely focus. The paper is organized as follows. Section 2 introduces the reader to the restrictions at issue. Section 3 summarizes the focus apparatus assumed in this paper (mostly from Rooth 1985, 1992a, b; adding some insights from Schwarzschild 1999). After that, the two idiosyncratic ellipsis types are examined in turn. Section 4 concentrates on reduced conditionals, presenting and arguing against a previous account and developing a new focus-based approach. Section 5 is devoted to sluicing. It will be shown that the focus analysis of the correlate restriction developed in Romero (1997b, 1998) also covers the definiteness restriction. In Section 6, we turn to VP-ellipsis and gapping to see that the same focus algorithm is inert when combined with the semantics of declaratives. Section 7 concludes.
. Data Let us first examine the restriction on possible correlate phrases. Remnant phrases in the E-clause have a corresponding constituent in the A-clause, their correlate. Take the examples (1)–(4), repeated here. In each case, the relevant remnants in the Eclause are doubly underlined and their correlates in the A-clause are marked with a single line. (6) A friend of mine likes turnips, but Ms. Matilde doesn’t . (7) Boys like lentils and girls potatoes. (8) Wenn ich wen besuche, dann (immer) den Peter. If I somebody visit then always the Peter ‘If / whenever I visit somebody, then Peter / it’s Peter.’
Maribel Romero
(9) Somebody just left – guess who . Note that, so far, all the correlates are indefinite NPs. What happens if we have other NP types as correlates, e.g., referential expressions or non-indefinite quantifiers? For reduced conditionals, Schwarz (1996, 1998) observes that, whereas the indefinite correlate wen (“somebody”) in (8) yields grammaticality, the name in (10b) renders the sentence ungrammatical. Note that the full-fledged version (10a) is fine.2 (10) a.
Wenn If Peter. Peter b. *Wenn If
ich den Karl besuche, dann besuche ich (immer) den I the Karl visit, then visit I always the
ich den Karl besuche, dann (immer) den Peter. then always the Peter I the Karl visit, (Schwarz 1998)
A similar restriction on possible correlates operates in sluicing. Chung et al. (1995) point out that indefinite NPs and wh-phrases (which are sometimes considered indefinite-like) are good correlates, as in (9) and (11), but names and Quantificational NPs (QuNPs) are not, as shown in (12b)–(13b). Note, again, that the full-fledged versions (12a)–(13a) are perfectly grammatical. (11) We know how many papers this reviewer has read, but we don’t know which ones . (Chung et al.) (12) a.
I know that Meg’s attracted to Harry, but they don’t know to who she’s attracted. b. *?I know that Meg’s attracted to Harry, but they don’t know to who . (Chung et al.)
(13) a.
Each of the performers came in. We were sitting so far back that we couldn’t see who came in. b. *Each of the performers came in. We were sitting so far back that we couldn’t see who . (Chung et al.)
These examples show that reduced conditionals and sluicing place a restriction on possible correlates. No such restriction is operative in VP-ellipsis or gapping. This is shown by the examples (14) and (15): names and QuNPs are perfect correlates in VP-ellipsis and gapping. (14) a. Susan likes turnips, but Cecilia doesn’t . b. Most girls like turnips, but Cecilia doesn’t .
Correlate restriction and definiteness effect in ellipsis
(15) a. Susan likes Philadelphia and Patricia Barcelona. b. Every boy sent letters to Sue and every girl to Pat. The second characteristic is a definiteness effect, mandatory in reduced conditionals and sluicing and optional in other elliptical construals. This condition concerns indefinite and QuNPs in the A-clause other than the correlate itself. A definiteness effect obtains if the interpretation of the corresponding expressions in the E-clause is not that of an indefinite or quantificational NPs, but that of a definite, namely that of an E-type or donkey pronoun. The best is to illustrate the phenomenon with an example. Let us take (16) and concentrate on the interpretation of the NP ein Student (“a student”). (16) Reduced Conditional: a. Wenn ein Student was zum Trinken bestellt, dann bestellt If a student something to drink orders, then orders ein Student Bier. a student beer ‘If a student orders something to drink, then a student orders beer.’ b. Wenn ein Student was zum Trinken bestellt, dann Bier. If a student something to drink orders, then beer ‘If a student orders something to drink, then beer.’ (Schwarz 1996) As noted in (Schwarz 1996), the full conditional in (16) is (generally) understood as introducing two students in every relevant situation: one student orders one thing or another, and another student orders beer. The intuition is that, as soon as we reduce (16a) into (16b), the sentence is about one single student per situation. (16b) necessarily means that, if a student orders something, then that same student is ordering beer. That is, the indefinite ein Student in the A-clause corresponds, interpretively, to a definite in the E-clause. The two readings can easily be teased apart in terms of truth conditions. Imagine there is a student in the actual world that ordered whiskey and nothing else. The truth of the full conditional (16a) is compatible with this fact, whereas the reduced (16b) – with a definiteness effect – is false in the depicted case. A definiteness effect also occurs in sluicing. The non-correlate NP in (17) that we are interested in is an article on sluicing. Again, the intuition is that, in (17a), the two full interrogative clauses may be talking about different papers (by different authors); in (17b), the sluiced interrogative necessarily talks about that same paper (by the ZAS student) that was introduced in the A-clause.
Maribel Romero
(17) Sluicing: a. Luckily, Miriam told me that a student at ZAS has written an article on sluicing. I should mention who has written an article on sluicing in my dissertation. b. Luckily, Miriam told me that a student at ZAS has written an article on sluicing. I should mention who in my dissertation. No such definiteness restriction operates in VP-ellipsis or gapping. The examples (18) and (19) show that, though non-correlate indefinites can correspond to E-type pronouns in the E-clause, they do not need to. In (18), the elided VP in (18b) can be understood as introducing a second monastery, as the full version (18a) does. (18) VP-ellipsis: a. Greg visited a monastery, and Mathias visited a monastery, too. b. Greg visited a monastery, and Mathias did , too. Similarly, the gapped (19b) can have the same interpretation as its full counterpart (19a), where two different books may be at stake: (19) Gapping: a. Greg showed a book to Paul and Mathias showed a book to Karl. b. Greg showed a book to Paul and Mathias to Karl. We have seen that ellipsis in conditionals and in interrogative clauses imposes two requirements alien to their full versions: correlates must be of a certain shape (namely, indefinites) and non-definite non-correlates must be interpreted as definites in the E-clause. Other ellipsis constructions lack these requirements. As mentioned above, these data render reduced conditionals and sluicing as rather peculiar ellipsis types. Before we examine them in more detail, some background on focus is presented in the next section.
. Semantics of focus Rooth (1985, 1992b) gives an implementation of the old idea that focused constituents invoke a set of alternatives. This is independent of whether there is an ellipsis site or not. In Rooth’s system, besides the proposition that a sentence ordinarily expresses (its ordinary semantic value, [[.]]), a sentence also introduces a set of focus alternative propositions (its focus semantic value, [[.]]f ): the set of propositions resulting from replacing the denotation of the focused element(s) with another object of the same type. The recursive definition is spelled out in (20):
Correlate restriction and definiteness effect in ellipsis
(20) Computing focus semantic values [[.]]f : (Rooth 1985) a. If α is a terminal node, then [[α]]f = {[[α]]f }. b. If α is a non-branching node with single daughter β, then [[α]]f = [[β]]f . c. If α is a branching node with daughters β and F (Focus feature), then [[α]]f = Dσ , where σ is the type of [[β]]. d. If α is a branching node with daughters β and γ (order irrelevant), and there are types σ and τ such that [[β]] ∈ D<σ,τ> and [[γ]] ∈ Dσ , then [[α]]f = { x∈Dτ : ∃y∃z [ y∈[[β]]f & z∈[[γ]]f & x=y(z) ] } To see some examples, take (21)–(23). The three sentences differ only in the placement of focus stress, marked in capitals. All of them express the same proposition as their ordinary semantic value. But each of them has a different focus semantic value or set of alternatives, depending on the position and number of focus patterns: (21) PETER ate beans. a. Proposition expressed: [[PETER ate beans]] = the proposition λs.Peter ate beans in s. b. Set of focus alternatives: [[PETER ate beans]]f = {p: ∃x [ p = λs.x ate beans in s] } = { λs.Peter ate beans in s, λs.Joan ate beans in s, λs.Sue ate beans in s, . . . } (22) PETER ate BEANS. a. Same proposition expressed. b. Set of alternatives: [[PETER ate BEANS]]f = {p: ∃x,y [ p = λs.x likes y in s] } = { λs.Peter ate beans in s, λs.Peter ate chestnuts in s, λs.Joan ate beans in s, λs.Joan ate chestnuts in s, λs.Sue ate beans in s, λs.Sue ate chestnuts in s, . . . } (23) Peter ate beans. a. Same proposition expressed. b. Set of alternatives: [[Peter ate beans]]f = {p: p = λs.Peter ate beans in s } = { λs.Peter ate beans in s} Sentences containing focused material must obey some felicity condition. This condition involves the relation between the invoked set of alternatives and the nearby discourse. Rooth distinguishes two main cases: (i) the subset condition and (ii) the membership condition.
Maribel Romero
The subset condition typically occurs in question/answer (Q/A) pairs. Let us call the focus licensed by the subset condition in Q/A pairs “free” focus.3 Focused BEANS in (24A) is an example. The subset condition is spelled out in (25): basically, the denotation of the question (in Hamblin’s (1973) style) must be a subset of the focus semantic value of the answer.4 The meaning of the question and the focus semantic value of the answer are given in (26a) and (26b) respectively. Since (26a) is a subset of (26b), the membership condition is fulfilled. (24) Q: What did Peter buy at the store yesterday? A: He bought BEANS (at the store yesterday). (25) Subset Condition: A focus in a sentence S is licensed if there is a question denotation salient (Rooth 1992b) in the discourse that is a subset of [[S]]f. (26) a.
For any situation or world s: [[(24Q)]] (s) = {p: ∃xe [ p = λs’.Peter bought x in s’]} = { λs’.Peter bought beans in s’, λs’.Peter bought rice in s’, λs’.Peter bought blueberries in s’, . . . } b. [[(24A)]]f = {p: ∃xe [ p = λs’. Peter bought x in s’]}
Note that the question denotation does not have to be directly expressed by an interrogative sentence; it suffices to infer it from the surrounding discourse (or to accommodate it), as in (27): (27) Q: Peter bought something at the store yesterday. A: Yes, he bought BEANS. This “free” focus introduces an exhaustivity implicature that will be crucial for the analysis proposed in this paper. The declarative (24A) simply expresses that Peter bought beans at the store yesterday (regardless of whether he bought something else or not), but the Q/A pairs (24) and (27) furthermore imply that Peter did not buy anything else (relevant). That is, the answer is taken to be an exhaustive listing of John’s purchases at the store yesterday. This exhaustivity implicature is derived in the following way. Once the sentence with focus is understood as the answer to a question, we assume that the utterer is competent to answer it and that s/he follows the Gricean conversational maxims. Furthermore, we assume that all the alternative propositions in the question denotation (26a) and their conjunctions are relevant to the subject of the conversation. Adding the Maxim of Quantity (28) to that, the exhaustivity implicature follows: given that the speaker uttered (24A)– (27A) instead of expressing any stronger (conjoined) relevant proposition, s/he must (correctly) believe that no such stronger proposition is true. Hence the negative implicatures in (29a). (29a) plus [[(24A)]] yield the exhaustivity implicature in (29b).5
Correlate restriction and definiteness effect in ellipsis
(28) Maxim of Quantity: Be as informative as possible. (29) a.
Negative implicatures: ‘It is not true that Peter bought beans and rice’, ‘it is not true that Peter bought beans and blueberries’, ‘it is not true that Peter bought beans, rice and blueberries’. b. Exhaustivity implicature: ‘Peter bought beans and nothing else’.
The second type of felicity condition that can license focus is the membership condition. Typical cases involve contrastive focus, as in (30): (30) Tony ate beans last night, (but) PETERF DIDN’TF . Intuitively, focus on PETER marks contrast with its correlate Tony, and focus on negation signals contrast with its correlating polarity in the preceding sentence. It is crucial that the denotation of the correlates be different from the denotation of the focused elements. Otherwise, focus is unfelicitous, as the minimal pair in (31) shows: (31) a. Guess what! Yesterday, Tony ate beans, and PETERF (ate) PAELLAF . b. #Guess what! Yesterday, Tony ate beans, and PETERF (ate) BEANSF. We have seen that a contrastive focus is licensed if there is an alternative to the focused element different from the element itself. This result is guaranteed by the interaction of Rooth’s membership condition and Schwarzschild’s (1999) Avoid Focus constraint: (32) Membership condition:6 (Rooth 1985, 1992a, 1997) A focus in a sentence S is licensed if there is a sentence in the discourse that expresses or implies a member of [[S]]f . (33) Avoid Focus Constraint (Avoid-F): (Schwarzschild 1999: 156) Avoid focus unless needed (to fulfill the membership condition). Let us see these constraints at work. The felicitous (31a), repeated as (34), fulfills both the membership condition and Avoid Focus. Focus stress is necessary in all the contrasting elements in the second conjunctive IP; otherwise, the membership condition is not met, as in (35). Finally, (31b), repeated as (36), is infelicitous because it has an unnecessary focus on beans, and hence it violates Avoid Focus. (34) Guess what! Yesterday Tony ate beans and PETERF ate PAELLAF . a.
√
Membership condition: [[Tony ate beans]] ∈ [[PETERF ate PAELLAF ]]f
Maribel Romero √
b. Avoid Focus constraint: [[Tony ate beans]] ∈ [[PETERF ate paella]]f [[Tony ate beans]] ∈ [[Peter ate PAELLAF ]]f (35) #Guess what! Yesterday Tony ate beans and PETER visited Sue. a. *Membership condition: [[Tony ate beans]] ∈ [[PETER visited Sue]]f √ b. Avoid Focus constraint: [[Tony ate beans]] ∈ [[Peter visited Sue]]f (36) #Guess what! Yesterday Tony ate beans and PETERF ate BEANSF. √ a. Membership condition: [[Tony ate beans]] ∈ [[PETER ate BEANS]]f b. *Avoid Focus constraint: [[Tony ate beans]] ∈ [[PETER ate beans]]f To sum up this section, focus has to be licensed either by the subset condition or by the membership condition. When we have “free” focus as in Q/A pairs, the subset condition plus Gricean principles introduce an exhaustivity implicature. When we have contrastive focus, the membership condition and the economy constraint Avoid Focus apply to ensure that the remnant will contrast with (i.e. will have a different denotation from) its correlate. The up-shot of this section is given in (37): (37) a. A remnant with “free” focus introduces an exhaustivity implicature. b. A remnant with contrastive focus must contrast with its correlate phrase.
. Reduced conditionals The only pre-existent analysis of reduced conditionals (RCs) in the literature can be found in Schwarz’ (1996, 1998, 2000) detailed studies, where the term reduced conditional is coined. Schwarz discovers several grammatical and interpretive differences between a full conditional and its reduced counterpart. Here we will concentrate on the two of them presented above, namely the correlate restriction and the definiteness effect.7 Schwarz develops a proposal in which full and reduced conditionals have different Logical Form representations. In particular, he argues that ellipsis in RCs induces coindexation of two situation variables, namely the situation variable in the if -clause (or A-clause) and the one in the E-clause. Although his analysis derives most of the empirical data correctly, it relies on a coindexation explanation that cannot be extended to other ellipsis constructions (not even to sluicing, which we saw displays similar properties). In this paper, I will present an alternative account where the key factor is a well-known characteristic of ellipsis constructions in general: remnants of ellipsis bare focus stress.
Correlate restriction and definiteness effect in ellipsis
. Schwarz’ analysis of reduced conditionals Let us recall the relevant data. First, we saw that RCs impose a restriction on the shape of the correlate: indefinite phrases are acceptable, as in (38b), but names are not, as in (39b). Full conditionals do not impose such restriction, as shown in (38a)–(39a): (38) a.
Wenn ich wen besuche, dann besuche ich (immer) den If I someone visit, then visit I always the Peter. Peter ‘If I visit somebody, then I (always) visit Peter.’ besuche, dann (immer) den Peter. b. Wenn ich wen then always the Peter If I someone visit, ‘If I visit somebody, then (always) Peter.’ (Schwarz 1998)
(39) a.
Wenn ich den Karl besuche, dann besuche ich (immer) den If I the Karl visit, then visit I always the Peter. Peter ‘If I visit Karl, then I (always) visit Peter.’ b. *Wenn ich den Karl besuche, dann (immer) den Peter. then always the Peter If I the Karl visit, ‘If I visit somebody, then (always) Peter.’ (Schwarz 1998)
The grammaticality status of the two full conditionals follows automatically from the situation-based analysis of adverbs of quantification (Kratzer 1989; von Fintel 1994) and Rothstein’s (1995) application to multi-event quantification. A situation is a spatio-temporal slice of a possible world. For example, besides the situation that comprehends the entire actual world, there is a smaller situation that is part of the actual world and that solely includes the author of this paper writing this paragraph on her computer on the day July 31 in a certain Spanish town. The meaning of a sentence is the set of possible (larger or smaller) situations in which the sentence is true. Adverbs of quantification like always, usually, sometimes, etc., quantify over the situations in those sets. In conditionals, the restrictor of this adverbial quantification is narrowed down by the if -clause, whereas the nuclear scope, with its classical existential closure (Heim 1982; Diesing 1992), is provided by the main clause. This is summarized in the LF schema in (40). What is crucial is that the situation variables of the if -clause and the main clause have different indices and that the latter is bound by the intermediate ∃-closure. (40) Full Conditional: alwaysi . . . [S if -clause ]i . . . ∃j [S main clause ]j
Maribel Romero
Let us see the semantics for this LF schema. It uses the notion of minimal situation (Kratzer 1989; Portner 1992; von Fintel 1994): a situation s is a minimal situation for a given proposition p iff p is true in s and there in no proper subpart of s (ordered by the part-of relation <) in which p is true. (41) A situation s is a minimal situation for a proposition p iff: s∈p and ¬∃s’ [ s’ ∈ p & s’< s ] The standard assumptions in (40) and (41), combined with some insights from Rothstein undertaken in Schwarz, yield the correct semantics for the full conditional examples: (42) Full-fledged, grammatical (38a): “If I visit someone, I (always) visit Peter.” ‘For every minimal situations that makes the proposition [[I visit somebody]] true, there is a situation s’ such that s’ matches (i.e., is related in a particular way to) s and s’ is a minimal situation that makes the proposition [[I visit Peter]] true.’ (43) Full-fledged, grammatical (39a): “If I visit Karl, I (always) visit Peter.” ‘For every minimal situations that makes the proposition [[I visit Karl]] true, there is a situation s’ such that s’ matches (i.e., is related in a particular way to) s and s’ is a minimal situation that makes the proposition [[I visit Peter]] true.’ Now comes Schwarz’ proposal for the reduced conditionals. He proposes that ellipsis in RCs forces the situation variable of the main clause (the E-clause) to be coindexed with that of the if -clause, as sketched below: (44) For any two situation variables i, j, a reduced conditional has the LF in (ii) if its full counterpart has the LF in (i): (Schwarz 1998) (i) alwaysi . . . [S . . . ]i ∃ j . . . [S . . . ]j (ii) alwaysi . . . [S . . . ]i ∃. . . [S . . . ]i When Schwarz’ LF coindexation is interpreted in the semantics, the following truth conditions obtain for the reduced conditional examples: (45) Reduced, grammatical (38b): “If I visit somebody, then Peter.” ‘For every minimal situations that makes the proposition [[I visit somebody]] true, s matches s and s is a minimal situation that makes the proposition [[I visit Peter]] true.’ (46) Reduced, ungrammatical (39b): * “If I visit Karl, then Peter.” #‘For every minimal situations that makes the proposition [[I visit Karl]]
Correlate restriction and definiteness effect in ellipsis
true, s matches s and s is a minimal situation that makes the proposition [[I visit Peter]] true.’ The outcome is that, whereas an indefinite correlate yields the sensible truth conditions in (45), the name makes the resulting truth conditions in (46) deviant. Indeed, (46) amounts to say that the RC with a name as correlate is necessarily false when felicitous, since no minimal situation of me visiting Karl can be, at the same time, a minimal situation of me visiting Peter. These deviant truth conditions, Schwarz argues, are the source of the ungrammaticality of (39b). Schwarz’ coindexed LF not only gives us a way to rule out names as correlates, but it also provides the right reading for RC with indefinite correlates. Note that the full-fledged (38a) does not have the same semantics as its reduced version (38b). This can be seen by comparing the corresponding truth conditions in (42) and (45). The full-fledged (42) simply requires that every visit of mine to somebody be paired with a visit of mine to Peter. The reduced (45) says that every visit of mine is a visit of mine to Peter; that is, the only person that I ever visit is Peter. This interpretive difference between full and reduced conditionals is called exhaustiveness effect in Schwarz (1998). Let us now turn to the second characteristic of RCs, the definiteness effect, and let us see how the LF indexation account derives the semantic difference between full and reduced conditionals. Recall the data: (47) a.
Wenn ein Student was zum Trinken bestellt, dann bestellt If a student something to drink orders, then orders ein Student Bier. a student beer ‘If a student orders something to drink, then a student orders a beer.’ b. Wenn ein Student was zum Trinken bestellt, dann Bier. If a student something to drink orders, then beer ‘If a student orders something to drink, then beer.’ (Schwarz 1996)
Whereas the full (47a) may be understood as talking about two students per case and it is compatible with there being cognac-ordering students, the reduced (47b) talks about the same student in both clauses and it is incompatible with there being cognac-ordering students. Schwarz’ indexation analysis can account for this difference: (48) Full-fledged (37a): “If a student orders something to drink, then a student orders a beer.” ‘For every minimal situations that makes the proposition [[a student orders something to drink]] true, there is a situation s’ such that s’ matches (i.e.,
Maribel Romero
is related in a particular way to) s and s’ is a minimal situation that makes the proposition [[a student orders beer]] true.’ (49) Reduced (37b): “If a student orders something to drink, then a beer.” ‘For every minimal situations that makes the proposition [[a student orders something to drink]] true, s matches s and s is a minimal situation that makes the proposition [[a student orders beer]] true.’ In (48), the fact that the main clause introduces a second situation s’ allows for there to be two students, one in situation s and one in situation s’. In (49), instead, there is only one situation variables, and, since it is a minimal situation, it only contains one student. This derives the feeling that we are talking about the same student in the conditional clause and E-clause. That is, this situation coindexation derives the same semantics as if we had used a full conditional with the definite anaphoric expression the/that student. Hence, this derives the definiteness effect. In sum, Schwarz’ account successfully derives the two characteristics of RCs at issue. However, it does so at a cost: the assumption that ellipsis forces coindexation of situation variables would be operative only in RCs and it would play no role in any other type of ellipsis. In the following subsection, I will show that similar truth conditions follow without the need of LF-coindexation once we take into account the contribution of focus.8
. New proposal for reduced conditionals I propose that the indexation pattern of a reduced conditional is identical to that of its full counterpart. They both follow the LF schema generally assumed for full conditionals, repeated in (50). This LF schema is interpreted according to the general situation semantics for conditionals in von Fintel (1994), as spelled out in (51). I will show that all the interpretive differences between full and reduced conditionals follow from the necessary presence of focus on the remnant material in RCs and its interaction with the general conditional semantics in (51). An empirical advantage of the new approach is examined at the end. (50) alwaysi . . . [S . . . ]i ∃j . . . [S . . . ]j (51) ‘Every minimal situations that makes the proposition [[S1]] true can be extended to a situation s’ that makes the proposition [[S2]] true.’ (von Fintel 1994) Let us start with the full conditional in (52). There is no ellipsis and no necessary focus on Peter. Consequently, (52) has the possible LF representation in (53a), which yields the truth conditions in (53b). These conditions are correct:9
Correlate restriction and definiteness effect in ellipsis
(52) Wenn ich wen besuche, dann besuche ich (immer) den If I somebody visit, then visit I always the Peter (nachher). Peter (afterwards) ‘If I visit somebody, then I visit Peter (afterwards).’ (53) a. LF: Alwayss [IP1 if I visits somebody ] ∃s’ [IP2 then I visits’ Peter ] b. ‘Every minimal situations that makes the proposition [[I visit somebody]] true can be extended to a situation s’ that makes the proposition [[I visit Peter]] true.’ Let us turn now to its reduced version, repeated in (54). The non-elided material, namely the remnant PEter, carries focus stress (stress is marked in capitals), signaling a narrow focus-marking on the direct object: (54) Wenn ich wen besuche, dann (immer) [den PEter]F . If I someone visit, then always the Peter ‘If I visit somebody, then (always) Peter.’ (Schwarz 1998) As we saw in Section 3, focus needs to be licensed, either by the subset condition or by the membership condition in Rooth. I argue that focus on the remnants of RCs is licensed by the subset condition, in a similar way to questions/answer pairs. The intuitive idea is that the protasis of the conditional makes a question relevant and that focus marks that the E-clause functions as the answer to that implied question. Recall our mini-dialog (27), repeated in (55). In (55), the first utterance with an indefinite raises the question “What did Peter buy at the store yesterday?”, and (55A) answers it: (55) Q: Peter bought something at the store yesterday. A: Yes, he bought BEANS. In a similar way, in the RC (54), for each quantified situation s, the protasis [IP I visit s somebody] makes the question denotation [[Who do I visit?]] (s) relevant, and the E-clause answers it. The subset condition between the protasis and the E-clause is spelled out in (56). Since this condition is fulfilled, the focus on the remnant is felicitous: (56) a. b. c. d.
LF: Alwayss [IP1 if I visits somebody ] ∃s’[IP2 then I visits’ [Peter]F ] [[Who do I visit?]](s) = {p: ∃xe [ p = λs’.I visit x in s’]} [[I visit [Peter]F ]]f = {p: ∃xe [ p = λs’.I visit x in s’]} IP1 raises the question denotation (56b), and (56b) ⊆ (56c).
Recall from Section 3 that, once a sentence with focus is understood as the answer to a question salient in the discourse, an exhaustivity implicature obtains
Maribel Romero
via Gricean principles. I propose that the same conversational implicature is computed at the level of the main clause or apodosis of the conditional.10 That is, in the current example, the main clause has the semantic meaning in (57a), it generates the negative implicatures sampled in (57b) and, adding the semantic and the pragmatic contributions, it conversationally implies the stronger, exhaustive proposition in (57c): (57) [IP2 (then) I visits [Peter]F ] a. Semantic meaning: λs”. I visit Peter in s” b. Negative implicatures: λs” it is not true in s” that I visited Pat; λs”. it is not true in s” that I visited Peter and Pat; λs”. it is not true in s” that I visited Peter, Pat and Sue; etc. c. Exhaustivity implicature: λs”. I visited Peter in s” and, for any (relevant) x other than Peter, it is not true in s” that I visited x This last proposition in (57c) is an essential piece of the new analysis. It includes situations where my visiting Peter occurs and my visiting anybody else does not occur (independently of whether there are other situations in the same world where I visit someone else; see Kratzer 1989: 644 on propositions being ‘not true’). That is, although these situations do not have to be minimal (e.g., a situation s100 including me visiting Peter, me buying flowers and nothing else is acceptable), they are “minimal” in the relevant respect: they include my visits to Peter and exclude any possible visit of mine to anybody else. When this proposition is inserted in the conditional interpretive schema, such relative minimality is imposed on the existentially quantified situation s’. This renders the following truth conditions: (58) ‘Every minimal situations that makes the proposition [[I visit somebody]] true can be extended to a situation s’ that makes the proposition [λs”. I visited Peter in s” and, for any (relevant) x other than Peter, it is not true in s” that I visited x] true.’ In short, (58) describes a world in which I always visit Peter and noone else. For, if there was a situation s19 where I visit Sue, s19 could not possibly be extended to a second situation where I visit Peter and noone else, since my visit to Sue would be included in it from the beginning. These truth conditions are correct. Now that we have seen in detail how the pragmatic implicature triggered by the focus and the semantics of conditional are combined, let us examine how the resulting truth conditions derive the two characteristics of RCs at issue. First, the restriction on possible correlates obtains because names as correlates yield, as in Schwarz, deviant truth conditions. Witness the truth conditions of (59)
Correlate restriction and definiteness effect in ellipsis
in (60b): either there is no situation of me visiting Karl (a case in which the sentence would be infelicitous), or there is some situation of me visiting Karl and then the sentence is false: (59) *Wenn ich den Karl besuche, dann (immer) den Peter. If I the Karl visit, then always the Peter ‘If I visit somebody, then (always) Peter.’ (Schwarz 1998) (60) a. LF: Alwayss [IP1 if I visits Karl ] ∃s’[IP2 then I visits’ [Peter]F ] b. ‘Every minimal situations that makes the proposition [[I visit Karl]] true can be extended to a situation s’ that makes the proposition [λs”. I visited Peter in s” and, for any x other than Peter, it is not true in s” that I visited x] true.’ The restriction on possible correlates is related to another property observed in Schwarz (1998), namely presupposition effects. Schwarz noted that the reduced (61b) carries the odd presupposition that my glasses are something to read (again, this presupposition does not necessarily arise in the full-fledged version (61a)). Similarly, (62) presupposes that Simone is a student. (61) a.
Wenn ich was zum Lesen mitnehme, dann nehme ich If I something to read take, then take I immer meine Brille. always my glasses ‘If I take something to read, I always take my glasses.’ b. !!Wenn ich was zum Lesen mitnehme, dann immer meine If I something to read take, then always my Brille. glasses ‘If I take something to read, I always take my glasses.’ (Schwarz 1998)
(62) a.
Wenn ich eine Studentin besuche, dann immer Simone. If I a student visit then always Simone ‘If I visit a student, then always Simone.’
These presupposition effects follow as a corollary of the restriction on correlates. Applying the new proposal to the LF of (62) in (63), we obtain the following truth conditions: (63) LF: Alwayss [IP1 if I visits a student ] ∃s’[IP2 then I visits’ [Simone]F ]
Maribel Romero
(64) ‘Every minimal situations that makes the proposition [[I visit a student]] true can be extended to a situation s’ that makes the proposition [λs”. I visited Simone in s” and, for any x other than Simone, it is not true in s” that I visited x] true.’ Note that the underlined part for any x other than Simone ranges at least over all the students, since the focus semantic value [[I visit SimoneF ]]f , from which the negative implicatures arise, has to be a superset of the question denotation [[which student do I visit]] (s). Hence, (64) guarantees that, whenever I visit a student, I visit Simone and nobody other than her that is a student. But then Simone must be a student herself. For, if Simone was not a student, we would obtain the same deviant truth conditions as in the Karl example. We would get the following: every situation s in which I visit a student can be extended to a situation s’ in which I visit (the non-student) Simone and no student whatsoever. These truth conditions yield either infelicity (if there is no instance of me visiting a student, to begin with) or falsity (if s includes a visit of mine to a student, its extension s’ will, too). In sum, the so-called presupposition effects arise because the assumption that Simone is a student is the only way to avoid fatal truth conditions.11 The second characteristic at issue, the definiteness effect, is derived in the new approach as follows. Take the RC (65). Its LF and truth conditions are the following: (65) Wenn ein Student was zum Trinken bestellt, dann Bier. If a student something to drink orders, then beer ‘If a student orders something to drink, then beer.’ (Schwarz 1998) (66) LF: Alwayss [IP1 if a student orderss something to drink ] ∃s’ [IP2 then a student orderss [BEER]F ] (67) ‘Every minimal situations that makes the proposition [[a student orders something to drink]] true can be extended to a situation s’ that makes the proposition [λs”. a student orders beer in s” and, for any x other than beer, it is not true in s” that a student orders x] true.’ Recall that, by the definiteness effect, (65) should mean the same as if we had used a definite anaphoric that student in the E-clause referring back to the student introduced in the if -clause. (67) achieves this effect: there is a single student x in s, and, since the extended s’ contains beer-ordering students and no student ordering anything else, x must be a beer-ordering student as well. To sum up, I have presented an alternative account to Schwarz’ coindexation analysis. I have shown that the two characteristics at issue (and related properties) can be derived from the presence of focus – understood as “free” focus in Q/A pairs – and its interaction with the general semantics of conditionals. No special
Correlate restriction and definiteness effect in ellipsis
LF is needed to account for the restriction on correlates and the definiteness effect. The desired data are achieved simply by exploiting a feature common to ellipsis constructions in general, namely focus. Let me conclude this section by adding an empirical benefit to the theoretical advantage of the new account. In the focus-based approach, the relative minimality of the extended situation s’ is mediated by the Gricean pragmatics of Q/A pairs. The Gricean maxim that makes us understand (68A) as the exhaustive answer to (68Q) also gives (69) its exhaustive meaning: only Peter was able to solve the problem, if anybody: (68) Q: Who was able to solve the problem? A: PEter. (69) Wenn jemand das Problem lösen kann, dann PEter . If someone the problem solve can, then PEter . ‘If someone can solve the problem, then Peter/it was Peter.’ But it is well-known that there are mention-one Q/A pairs (Beck & Rullmann 1999), where only an example is requested in the question and, by the maxim of relevance, only one example is provided in the answer, as in (70). Interestingly, mention-one readings are also found in RCs, resulting thus in the loss of the exhaustive interpretation of the main clause. For example, the non-exhaustive reading of (71) roughly says that if one can buy Spanish ham anywhere, one can certainly buy it at Sarcone’s, without excluding the possibility that there be other places to buy it. (70) Q: Where can I buy Spanish ham? A: At SarCOne’s. (71) Wenn man (überhaupt) irgendwo Spanischen Schinken kaufen If one ( at all ) anywhere Spanish ham buy (Irene Heim, p.c.) kann, dann (auf jeden Fall) bei SarCOne’s. can then (in any case) at SarCOne’s. ‘If one can buy Spanish ham anywhere, then (for sure) at Sarcone’s.’ It is difficult to see how Schwarz’ indexation account, in which the exhaustivity interpretation is hard-wired in the LF syntax, can allow for the pragmatic flexibility that distinguishes the interpretations of (69) and (71).12
. Sluicing We saw in Section 2 that sluicing is also blessed with the two peculiarities that made RCs such a remarkable ellipsis construction: the restriction on possible correlates
Maribel Romero
and the definiteness effect. The restriction on possible correlates for sluicing was noticed in Chung et al. (1995). There, it motivated an analysis heavily committed to LF-copying, in which the empirical results are distilled from the syntactic nature of several LF components. In this section, we will review their analysis in Subsection 5.1, and we will present Romero’s (1997b, 1998) empirical revision and her alternative focus-based account in Section 5.2. Subsection 5.3 presents more data bearing on the definiteness effect and shows how the analysis in Romero (1997b, 1998) is extended to derive this effect as well. The result is a focus-based approach that obtains the desired data without relying on LF properties of sluicing.
. Chung’s et al. data and analysis on the restriction on correlates Chung et al. (1995) data on possible correlates are represented by the following sentences: (72) Somebody just left – guess who .
(Ross 1969)
(73) We know how many papers this reviewer has read, but we don’t know which ones . (Chung et al.) (74) *?I know that Meg’s attracted to Harry, but they don’t know to who . (Chung et al.) (75) *Each of the performers came in. We were sitting so far back that we couldn’t see who . (Chung et al.) From data like these, Chung et al. conclude that the acceptability of a correlate depends on its NP shape. Namely, their empirical generalization is that (weak) indefinites and wh-phrases are good correlates, but names and quantificational NPs are not. Chung et al. exploit a property that indefinites and wh-phrases have been argued to share in the literature: they both are interpreted as open formulae providing a free variable (Kamp 1981; Heim 1982 for indefinites; Berman 1991 for wh-phrases).13 Chung et al. develop an LF-copying analysis that capitalizes on this property. The Q-operator in C0 has to bind a variable in its IP, since, otherwise, we would run into vacuous quantification. In base-generated interrogative clauses, this variable is provided by the wh-trace. In sluicing, though, no wh-trace is basegenerated. In fact, the entire IP is generated empty and only filled at LF by copying the antecedent IP into the open slot. The place of the trace is, then, occupied by the correlate NP, as in (76a) and (77a) (copied IP in boldface). Here the empirical generalization becomes crucial: indefinites and wh-phrases introduce a free individual variable, whereas names and quantificational NPs do not. Hence, binding by the Qmorpheme succeeds when the antecedent phrase is an indefinite or a wh-phrase, and it results in vacuous quantification otherwise.
Correlate restriction and definiteness effect in ellipsis
(76) Meg is attracted to somebody, but they don’t know to who . a. LF: [CP to who x Q x [IP Joan is attracted to somebodyx ]] (77) *?I know that Meg’s attracted to Harry, but they don’t know to who . a. LF:*[CP to who x Q x [IP Meg is attracted to Harry ]]
. Romero’s revision of the restriction on possible correlates In Romero (1997b, 1998), I showed that Chung’s et al. empirical generalization was not accurate. First, examples like (78)–(79) show that indefinites and wh-phrases as correlates do not always yield grammaticality. Second, names and quantificational phrases are sometimes good correlates, as (80)–(81) demonstrate. In sum, being an indefinite-like NP is neither a sufficient nor a necessary property to be a good correlate in sluicing. (78) *I know that four students came to the party, but they don’t know HOW MANY . (79) a. *We know how many papers this reviewer has read, but they don’t know HOW MANY . b. *We know which papers this reviewer has read, but they don’t know WHICH ones . (80) I know that Joan, Pat, Sam and Paul danced the first tango, but I don’t know WHO with WHO . (81) I know everybody danced the first tango, but I don’t know WHO with WHO. But, if the shape of the NP correlate per se is not the decisive property, what is it then? In Romero (1997b, 1998), I exploited the fact that the default intonational pattern of a sentence with sluicing involves focus stress on the remnant of the ellipsis, that is, on the sluiced wh-phrase. Once this is taken into account, I showed that the crucial factor deciding the acceptability of the correlate was the felicity conditions of focus. To see how the new approach is motivated by the data, let me first regroup all the data into the grammatical examples in (82) and the ungrammatical examples in (83). The common denominator of the bad cases is that the A-clause expresses or answers the same question as the sluiced interrogative expresses. In the good cases in (82), instead, the A-clause enunciates or answers a question, and the sluiced interrogative expresses a different one. This empirical generalization is stated in (84):
Maribel Romero
(82) Grammatical examples: a. Joan ate dinner with someone, but I don’t know with WHOM . b. We know how many papers this reviewer has read, but we don’t know WHICH ones . c. I know that Joan, Pat, Sam and Paul danced the first tango, but I don’t know WHO with WHO. d. I know that everybody danced with someone, but I don’t know WHO with WHO . (83) Ungrammatical examples: a. *I know that four students came to the party, but they don’t know HOW MANY . b. *We know how many papers this reviewer has read, but they don’t know HOW MANY . c. *We know which papers this reviewer has read, but they don’t know WHICH ones . d. *?I know that Meg’s attracted to Harry, but they don’t know to WHOM
. e. *Each of the performers came in. We were sitting so far back that we couldn’t see WHO . (84) An NP is a good correlate for a sluiced focused wh-phrase if its clause expresses (the answer to) a question that contrasts with the question expressed by the sluiced interrogative. I argued that the correlate restriction effect described in (84) is the result of the felicity conditions of focus. As we saw in Section 3, one way to license focus is by the membership condition plus the Avoid Focus principle. This last principle played the important role of ensuring that the focused element and its correlate denote different alternatives in the focus semantic value. The need of contrast was illustrated by (31), repeated in its gapping version in (85a–b). As the reader will recall, focus on BEANS in (85b) was precluded because it was not necessary to fulfill the membership condition between the two IPs and, hence, it had to be avoided, as sketched in (86b): (85) a. Guess what! Yesterday, Tony ate beans and PETERF PAELLAF . b. #Guess what! Yesterday, Tony ate beans, and PETERF BEANSF . (86) a.
√
Membership condition: [[Tony ate beans]] ∈ [[PETER ate BEANS]]f b. *Avoid Focus constraint: [[Tony ate beans]] ∈ [[PETER ate beans]]f
Correlate restriction and definiteness effect in ellipsis
The same licensing algorithm applies to focus in sluicing. All the sluicing examples that we have seen involve this type of contrastive focus on the wh-determiner. Hence, the first step to take is to define a set of alternatives for a wh-determiner. The wh-determiners which and how many are natural alternatives to each other, and they behave as such in the sluicing example (87a). A third alternative is provided by the corresponding whether-question. For example, in (87b) and (87c) respectively, WHICH (papers) and HOW MANY (papers) contrast semantically with the combined semantic contribution of whether. . . any papers: They usually ask how many papers the candidate reviewed for the journal, but they never ask WHICH ones. b. They usually ask whether the candidate reviewed any papers for the journal, but they never ask HOW MANY. c. They usually ask whether the candidate reviewed any papers for the journal, but they never ask WHICH ones.
(87) a.
An appendix at the end of this paper spells out the denotation of each wh-element and the semantic computation of the entire interrogative CP. The crucial result is that the questions denoted by the interrogative clauses in (88) are all alternatives to each other. In other words, the focus semantic value of an interrogative clause with a focused wh-determiner is as in (89): (88) a. Which students came. b. How many students came. c. Whether any/some students came. (89) [[WHICH students came]]f = [[HOW MANY students came]]f = { “which students came”, “how many students came”, “whether any students came” } Once we have this set of alternative questions, Rooth’s membership condition and Schwarzschild’s Avoid Focus apply to yield the new correlate restriction effect. We will illustrate it with only two examples. In (90), the antecedent constituent – with a name as correlate – implies the property of knowing the same question as the one expressed in the sluice; since there is no contrast between the implied question (90a) and the sluiced question (90b), Avoid Focus is violated and the sluice is ungrammatical. Names are perfect correlates in (91) because the questions evoked in the antecedent constituent ((91a)) and in the sluice ((91b)) are different focus alternatives and, hence, the felicity conditions of focus are fulfilled.
Maribel Romero
(90) *?I know that Meg’s attracted to Harry, but they don’t know to WHO (Meg’s attracted). a. Question implied by the A-constituent: “(To know) who Meg is attracted to” b. Question expressed by the E-constituent: “(To know) who Meg is attracted to” √ c. Membership condition, since: [[know that Meg’s attracted to Harry]] implies (if Harry is focused) the property “to know to whom Meg is attracted”, which ∈ [[know WHO Meg’s attracted to]]f . d. *Avoid Focus Principle, since: [[know that Meg’s attracted to Harry]] implies (if Harry is focused) the property “to know to whom Meg is attracted”, which ∈ [[know who Meg’s attracted to]]f . (91) I know that Joan, Pat, Sam and Paul danced the first tango, but I don’t know WHO with WHO. a. Question implied by the A-constituent: “(To know) whether somebody that danced the first tango with somebody.” b. Question expressed by the E-constituent: “(To know) who danced the first tango with whom.” √ c. Membership condition, since: [[know that Joan, Pat, Sam and Paul danced the first tango]] implies “to know whether somebody danced the first tango with someone”, which ∈ [[know WHO danced the first tango with WHOM]]f √ d. Avoid Focus Principle, since: [[know that Joan, Pat, Sam and Paul danced the first tango]] implies “to know whether somebody danced the first tango with someone”, which ∈ [[know who danced the first tango with whom]]f . In sum, the restriction on possible correlates in sluicing depends on the contrast or lack of contrast between two questions. Names are bad correlates for a remnant who because the question that the A-clause answers and the question expressed by the E-clause are the same question. If we have an indefinite correlate and a sluiced who, or if we have a name and a multiple wh-sluice, then the two questions at issue are different and the sluice is grammatical. This need of contrast is motivated by the licensing conditions of contrastive focus on the sluiced wh-phrase. Before turning to the definiteness effect, let me mention a corollary that, as in RC, follows from the analysis of the restriction on correlates. Chung et al. noted
Correlate restriction and definiteness effect in ellipsis
that the sluiced who in (92) seems to inherit its restrictor from its correlate some students. That is, the sluiced interrogative asks which students she talked to, not which professor or which people in general. This phenomenon was dubbed as “inheritance of content effects” in Chung et al., and it is reminiscent of the presupposition effects that we saw in the RCs (61) and (62) above. (92) I know she talked to some students, but I don’t know WHO. a. “I don’t know to who of the students/which students she talked.” b. *“I don’t know to which professor she talked.” c. *“I don’t know to which person she talked.” It is clear by now that the whether-question evoked in the A-constituent belongs to [[WHICH students Q she talked to]]f and not to [[WHICH professors she talked to]]f. That is, the interpretation (92a) fulfills the membership condition and the interpretation (92b) does not, a fact that immediately rules (92b) out. This is sketched in (93) and (94): (93) I know she talked to some students, but I don’t know WHO (students) she talked to. √ a. Membership condition: [[know she talked to some students]] implies “to know whether she talked to some students”, which belongs to [[know WHO (students) she talked to]]f (94) *I know she talked to some students, but I don’t know WHO (professors) she talked to. a. *Membership condition: [[know she talked to some students]] implies “to know whether she talked to some students”, which does NOT belong to [[know WHO (professors) she talked to]]f As for the more general interpretation “which people”, note that to know that she talked to some students implies to know at least a partial answer to the question which people did she talk to. In other words, (95A) is a felicitous partial answer to (95Q), and, hence, to know (95A) entails to at least partially know (95Q). This means that the A-constituent implies the same alternative as the one expressed in the E-constituent. Such configuration is ruled out by Avoid Focus, as in (96): (95) Q: Which people did she talk to? A: She talked to some students. (96) *I know she talked to some students, but I don’t know WHO (people) she talked to.
Maribel Romero
a.
√
Membership condition: [[know she talked to some students]] implies “to (partially) know which people she talked to”, which belongs to [[know WHO (people) she talked to]]f b. *Avoid Focus Principle: [[know she talked to some students]] implies “to (partially) know which people she talked to”, which belongs to [[know who (people) she talked to]]f To sum up this subsection, we have seen that the correlate restriction and the inheritance of content in sluicing are a function of focus and the semantics of know plus a question. As in the case of reduced conditionals, besides the general semantics of the construal, the only necessary apparatus to derive the correlate restriction is the licensing conditions of focus. If in RCs we had “free” focus as in Q/A pairs, in sluicing contrastive focus as in coordinate structures is at stake.
. The definiteness effect in the new proposal As the reader will recall, the definiteness effect in sluicing was illustrated with the example (17), repeated here as (97). The full-fledged (97a) may be talking about different papers (by different authors), whereas the elliptical (97b) talks about the same paper (written by that student at ZAS). That is, interpretively, the noncorrelate indefinite an article on sluicing corresponds to a definite NP – an E-type pronoun (Evans 1980) – in the ellipsis clause. (97) a.
Luckily, Miriam told me that a student at ZAS has written an article on sluicing. I should mention who has written an article on sluicing in my dissertation. b. Luckily, Miriam told me that a student at ZAS has written an article on sluicing. I should mention WHO in my dissertation.
The same occurs regardless of whether the NP at stake is indefinite or quantificational. Take the quantificational NP most professors in (98). In the full version (98a), different majorities of professors may be at issue, but only one majority – the one introduced in the A-clause – is evoked in the elided (98b): (98) a.
I just heard that a student in your class likes most professors in this school. I need to find out who likes most professors here/in this school. b. I just heard that a student in your class likes most professors in this school. I need to find out WHO .
Correlate restriction and definiteness effect in ellipsis
The previous literature had noticed other cases where a non-definite NP corresponds, semantically, to a definite in the sluice. Romero (1998) argues that the modified indefinite at least seven boys from your class corresponds to an elided Etype pronoun they (at least semantically) in (99). This is so because the sluice has a pair-list-like reading, paraphrased in (99a). Pair-list readings arise from the interaction between a wh-phrase and certain quantificational NPs (Chierchia 1993), crucially excluding NPs headed by at least n (Szabolsci 1997). This is shown by (100), which does not have a pair-list reading. Furthermore, Krifka (1992) shows that the interaction of a (plural or unmarked) wh-phrase with a plural definite NP results in a cumulative reading that, by Gricean cooperation principles, is often answered with a list of pairs, as in (101). This means that the pair-list-like reading in the sluice is tributary of a plural definite subject they, not of an indefinite subject at least seven boys from your class. (99) At least seven boys from your class danced the first waltz of the night, but I don’t know with WHOM. √ a. Pair-list-like reading of the sluice: “I don’t know who each of them danced with.” (100) Who did at least seven boys from your class dance the first waltz with? a. *Pair-list answer: “Matts danced it with Sue, Paul with Mary, Stephan with Simone, Bob with Elena, Mark with Tim, Jens with Claudi and Jorn with Pat.” (101) Who did they / the boys dance the first waltz with? √ Pair-list answer, pragmatically derived from cumulative reading. Another case from the literature suggests that indefinite and quantificational NPs are not the only ones that give rise to a definiteness effect. Merchant (1998, 2001) argues that, in (102), the wh-phrase who cannot have a corresponding who in the E-clause (as in (102a)), nor its trace has a corresponding trace (as in (102b)). Instead, the correct LF is achieved if who corresponds to an E-type pronoun he/she, as in (102c): (102) I know who ate, but not what. a. *I know who ate, but I don’t know what did who eat. b. *I know who1 t1 ate, but I don’t know what2 did t1 eat t2 . c. I know who ate, but I don’t know what he ate. We have seen several examples illustrating the definiteness effect in sluicing. We have also seen that this effect does not necessarily arise in full-fledged interrogatives. Now, I would like to point out that the definiteness effect is not a property of ellipsis in interrogatives per se, but a property of their focus. In fact, the effect
Maribel Romero
in the above examples only obtains because of the particular placement of focus there, since, once we move the focus pitch to the NP-part of the wh-phrase, the definiteness effect vanishes. Witness the pair in (103)–(104): when focus marks the wh-determiner, as in (103), we talk about the same majority of professors in both clauses; when focus falls on the NP-restrictor – GIRL in (104) – different majorities of professors may be involved. (103) I know some boy talked to most professors yesterday, but I don’t know WHO (/WHICH boy). (104) I know which boy talked to most professors yesterday, but I don’t know which GIRL. This means that the algorithm that generates the definiteness effect should not be a property of IP-ellipsis, but a product of focus. That is why an extension of Schwarz’ LF coindexation account to sluicing, by itself, would not do the job. In a nutshell, a coindexation analysis for (104) would have to allow for a common situation s’ where there is a boy and a girl (talking to professors), and not just a minimal situation with only a boy. But, then, why not to allow for a situation s” in (103) where there is a boy and, e.g., a parent talking to professors? Then WHO would range over the union of those two individuals and, hence, more than one majority of professors could be at stake. Let us see now how this intermittent definiteness effect in sluicing follows from the interaction of the semantics of questions and focus. Let us start with (98), repeated here as (105). The key intuition is that the wh-phrase who in the full interrogative (105a) may range over a wider set of people than the sluiced WHO in (105b): who may stand for “who in this school”, but WHO has to stand for “which student in your class”. (105) a.
I just heard that a student in your class likes most professors in this school. I need to find out who likes most professors here/in this school. b. I just heard that a student in your class likes most professors in this school. I need to find out WHO .
As the reader will recall, focus is responsible for narrowing down the restrictor of the sluiced WHO, i.e., for the “inheritance of content” discussed in Section 5.2. Basically, focus requires that the question implied by the A-constituent and the question expressed by the E-constituent match semantically and that they only differ in the denotation of the focused material. That gives us (106) for the relevant questions (or VP-properties) in (105b):14
Correlate restriction and definiteness effect in ellipsis
(106) Example (105b): a. Question implied by A-constituent: “(To know) whether there is a student in your class that likes most professors in this school”, i.e., λxλw. ∀w’ ∈ Epix (w) [ λp [p(w’) & p = λw”[a student likes most professors in w”] ] = λp [p(w) & p = λw”[a student likes most professors in w”] ] ] b. Question expressed by the E-constituent: “(To know) which student in your class likes most professors in this school”, i.e., λxλw. ∀w’ ∈ Epix (w) [ λp.∃x [ p(w’) & p = λw”[x is a student in w” & x likes most prof. in w”]] = λp.∃x [ p(w) & p = λw”[x is a student in w” & x likes most prof. in w”]]] The question (106a) asks about the existence of a particular type of student and (106b) asks about the identity of that same type of student. This is required by the subset condition plus Avoid Focus. Since the student is the same in both questions, the (unique) majority of professors that he likes is also the same in both questions. Hence the definiteness effect. Note that the size of the situation involved in the A-clause and in the E-clause does not play a role, since the effect obtains independently with a maximal situation (represented with the world variable w” in the formulae). Let us now turn to the pair (103)–(104). The explanation we just saw applies to (103): the definiteness effect is a by-product of the semantics of the two questions in (107). The interesting case is (104). Here, the two questions have to match semantically except for the wh-phrase’s restrictor, GIRL, which is the locus of focus stress. That gives us (108). Naturally, since the two questions involve different people talking to a majority of professors, the two majorities of professors may be different. This derives the lack of definiteness effect: (107) Example (103): a. Question implied by A-constituent: “(To know) whether there is a boy that talked to a majority of professors yesterday”, i.e., λxλw. ∀w’ ∈ Epix (w) [ λp [p(w’) & p = λw”[a boy talked to most prof. yesterday in w”]] = λp [p(w) & p = λw”[a boy talked to most prof. yesterday in w”]]]
Maribel Romero
b. Question expressed by the E-constituent: “(To know) which boy talked to a majority of professors yesterday”, i.e., λxλw. ∀w’ ∈ Epix (w) [ λp.∃x [ p(w’) & p = λw”[x is a boy in w” & x talked to most prof. yesterday in w”]] = λp.∃x [ p(w) & p = λw”[x is a boy in w” & x talked to most prof. yesterday in w”]]] (108) Example (104): a. Question implied by A-constituent: “(To know) which boy talked to a majority of professors yesterday”, i.e., λxλw. ∀w’ ∈ Epix (w) [ λp.∃x [ p(w’) & p = λw”[x is a boy in w” & x talked to most prof. yesterday in w”]] = λp.∃x [ p(w) & p = λw”[x is a boy in w” & x talked to most prof. yesterday in w”]]] b. Question expressed by the E-constituent: “(To know) which girl talked to a majority of professor yesterday”, i.e. λxλw. ∀w’ ∈ Epix (w) [ λp.∃x [ p(w’) & p = λw”[x is a girl in w” & x talked to most prof. yesterday in w”]] = λp.∃x [ p(w) & p = λw”[x is a girl in w” & x talked to most prof. yesterday in w”]]] In sum, the definiteness effect is not a result of ellipsis in interrogative clauses, but a result of focus in questions. The set of alternatives generated by the focus determines whether the A-constituent and E-constituent need to be about the same participants or not, hence deriving the distribution of the definiteness effect. To sum up Section 5, we have seen that the restriction on correlates and the definiteness effect follow from the requirements that the contrastive focus on the sluiced wh-phrase imposes on two questions. The property – implied in the Aconstituent – of knowing a given question Q must match semantically the property of knowing the sluiced question, differing necessarily in the semantic contribution of the wh-word. In this way, the two effects follow entirely from focus and the semantics of questions; they are not dependent on the LF syntactic representations of different types of NPs, representations that are tangential to the correct empirical generalization.
Correlate restriction and definiteness effect in ellipsis
. Focus in coordinated declaratives: VP-ellipsis and gapping Sections 4 and 5 have spelled out in detail a focus-based analysis that derives the correct data on the correlate restriction and the definiteness effect in reduced conditionals and sluicing, respectively. In RCs, “free” focus, understood as in Q/A pairs, gives rise to an exhaustivity implicature that determines the two effects. In sluicing, contrastive focus requires complete semantic matching between two questions except for the wh-words, a condition from which the correlate restriction and the definiteness effect follow. In this section, we briefly turn to VP-ellipsis and gapping, which lack both characteristics. We will see that contrastive focus, as in sluicing, is involved in these constructions, but that its interaction with the semantics of declaratives does not lead to the aforementioned peculiarities. Let us first recall the VP-ellipsis and gapping data on the correlate restriction. The previous data are repeated in (109a, d) and (110a, d), and other examples are added. These sentences show that, in VP-ellipsis and gapping, names and quantificational NPs are in principle good correlates for a remnant NP:15 (109) VP-ellipsis: a. Susan likes turnips, but CeCIlia DOESN’T . b. Susan likes turnips, but one of the HOSTS DOESN’T . c. Susan likes turnips, but most BOYS DON’T . d. Most girls like turnips, but CeCIlia DOESN’T . e. Most girls like turnips, but one of the HOSTS DOESN’T . f. Most girls like turnips, but most BOYS DON’T . (110) Gapping: a. Susan likes Philadelphia and PaTRIcia BarceLOna. b. Susan prefers Philadelphia and one of the orgaNIzers BarceLOna. c. Susan prefers Philadelphia and most other GIRLS BarceLOna. d. Every boy sent letters to Sue and every GIRL to PAT. In all these cases, the remnants receive a focal intonation that is licensed by the membership condition plus Avoid Focus. I will illustrate this requirement for only two of the examples. For (110a), the membership condition and Avoid Focus are met in (111a,b); furthermore, since the situations introduced in declaratives are not necessarily minimal, no deviant truth conditions arise, as shown in (111c): (111) LF: [IP1 Susan likes Philadelphia] and [IP2 PaTRIcia likes BarceLOna] √ a. Membership condition, since: [[Susan likes Philadelphia]] ∈ [[PaTRIcia likes BarceLOna]]f
Maribel Romero
c.
√
Avoid Focus: [[Susan likes Philadelphia]] ∈ [[Patricia likes BarceLOna]]f [[Susan likes Philadelphia]] ∈ [[PaTRIcia likes Barcelona]]f d. λs [Susan likes Philadelphia in s and Patricia likes Barcelona in s] In (109d), the subjects (as well as the polarities) of the two clauses bear focus pitch and hence must contrast. This means that, in the LF (112), the denotation of the remnant CeCIlia has to find a correlate denotation of the same semantic type as [[Cecilia]] that is different from [[Cecilia]] itself. But Cecilia has type e and its obvious correlate most girls is a generalized quantifier of type <<e,st>,<st>>. Type matching can be achieved by type-lifting CeCIlia to a generalized quantifier (Partee 1987), or by shortly QRing most girls and taking its trace (of type e) to be the actual correlate of CeCIlia.16 Either way, the focus requirements are met and no deviant truth condition arise. (112) LF: [IP1 Most girls like turnips], but [IP2 CeCIlia DOESN’T like turnips] √ a. Membership condition, since: [[Most girls like turnips]] ∈ [[CeCIlia DOESN’T like turnips]]f , which, by type-lifting, equals: {λs.Cecilia likes turnips in s, λs.Cecilia does not like turnips in s, λs.most girls like turnips in s, λs.most girls do not like turnips in s, λs.a bishop likes turnips in s, λs.a bishop does not like turnips in s, ...} √ b. Avoid Focus: [[Most girls like turnips]] ∈ [[Cecilia DOESN’T like turnips]]f [[Most girls like turnips]] ∈ [[CeCIlia doesn’t like turnips]]f c. λs [ most girls like turnips in s and Cecilia does not like turnips in s ] We have seen that non-indefinite correlates are licit in VP-ellipsis and gapping because they fulfill the focus requirements – contrary to the ungrammatical sluicing cases – and they yield appropriate truth conditions – contrary to the deviant semantics arising in RCs. Let us now turn to the definiteness effect. The definiteness effect is also not operative in the constructions at issue given the semantics of coordinated or juxtaposed declaratives, which simply amounts to the intersection of two sets of (not necessarily minimal) situations. Recall the relevant examples, repeated as (113) and (114). Both in the full and in the elliptical versions, Greg and Mathias may have visited different monasteries and shown different books; the truth of the sentences does not require that they visited the same monastery and showed the same book. This follows from the semantics, illustrated for (113b) in (115):
Correlate restriction and definiteness effect in ellipsis
(113) VP-ellipsis: a. Greg visited a monastery, and Mathias visited a monastery, too. b. Greg visited a monastery, and Mathias did , too. (114) Gapping: a. Greg showed a book to Paul and Mathias showed a book to Karl. b. Greg showed a book to Paul and Mathias to Karl. (115) LF: [IP1 Greg visited a monastery], and [IP2 Mathias visited a monastery], too. a. [[IP1]] = λs’. ∃x [x is a monastery in s’ and Greg visited x in s’] b. [[IP2]] = λs”. ∃x [x is a monastery in s” and Mathias visited x in s”] c. [[IP1 and IP2]] = λs [∃x [x is a monastery in s and Greg visited x in s] & ∃x [x is a monastery in s and Mathias visited x in s] ] In conclusion, in VP-ellipsis and gapping, the felicity conditions of focus are satisfied and the semantics of declaratives is computed without leading to a restriction on correlates or a definiteness effect.
. Concluding remarks In this paper, two peculiarities of both reduced conditionals and sluicing have been explained as the result of the interaction of focus with the semantics of conditionals and questions. As an alternative to Schwarz’ (1998) coindexation approach to reduced conditionals, it has been shown that the correlate restriction and the definiteness effect follow as a by-product of the association with focus and the general semantics of Conditionals. Focus is licensed by Rooth’s subset condition as in Question/Answer pairs, and it generates an exhaustivity implicature which, computed within the conditional interpretive schema, is responsible for the aforementioned effects. For sluicing, we saw that, contrary to Chung’s et al. (1995) analysis and following Romero (1997b, 1998), the grammaticality of a given correlate for a sluice does not hinge on the kind of DP itself, but on the contrast between the antecedent phrase and the sluiced wh-phrase, which crucially bears focus stress on the whDeterminer. The desired data are derived by defining the set of alternatives of a wh-Determiner and applying to it Rooth’s (1985, 1992a, b, 1995) focus membership condition and Schwarzschild’s (1999) Avoid Focus constraint. This analysis is extended to cover the definiteness effect in this paper. Two questions must match semantically except for the choice of wh-determiner (in focus); this results in two questions about the same participants, even if the situation variable in the denotations of the questions are world-size.
Maribel Romero
Finally, we have seen that focus is also licensed by the membership condition and Avoid Focus in VP-ellipsis and gapping, but its interaction with the semantics of declaratives is inert. In the bigger picture, the work developed in this paper is a step towards a unified account of certain aspects of reduced conditionals, sluicing, VP-ellipsis, gapping and, potentially, other ellipsis phenomena. The previous accounts specific to reduced conditionals and to sluicing are not needed – nor entirely adequate empirically – to derive the aforementioned data.
Appendix The LF in (116) is assumed for interrogative clauses. A question operator Q sits in C0 , and wh-phrases are reconstructable into IP at different possible depths, leaving only their wh-part in Spec-CP.17 The denotations of the relevant lexical entries are spelled out in (117)–(120), and the complete semantic computation is illustrated in (121)–(123), yielding Hamblin-style question denotations (i.e., we obtain the set of possible – true or false – answers to the question at issue and compute the N’-restrictor of the wh-phrase within the question nucleus). The meaning of the verb know (when applied to a question) is as in (124), slightly modified from Heim (1994) to adapt it from Karttunen-style denotations to Hamblin-style questions.18 (116) (I know) which / how many / whether some students came. CP C’
WH-2 C0 Q
IP IP
WhP1
Wh0
NP
t2- (WH)ICH t2- MANY t2- function j
students
t1 came
(117) [[wh- ]] = f ∈ D<<e,<s,<st,t>>>, <s,<st,t>>> such that, for every P ∈ D<e,<s,<st,t>>>, every w ∈ Ds , p ∈ D<st> f(P)(w)(p) = 1 iff ∃x[P(x)(w)(p)]
Correlate restriction and definiteness effect in ellipsis
(118) [[many]] = f ∈ D<e,δ> such that for every n ∈ De , every P, Q ∈ D<e,st> and every w ∈ Ds , f(n)(P)(Q)(w) = 1 iff ∃n x [P(x)(w) & Q(x)(w)] (119) [[(wh)ich]] = f ∈ D<e,δ> such that, for every x ∈ De , every P, Q ∈ D<e,st> and every w ∈ Ds , f(x)(P)(Q)(w) = 1 iff P(x)(w) & Q(x)(w) (120) Function j: j ∈ D<e,δ> such that for every n ∈ De , every P, Q ∈ D<e,st> and every w ∈ Ds , j(n)(P)(Q)(w) = 1 iff ∃x [P(x)(w) & Q(x)(w)] (121) (I know) how many students came. CP wp.n [ p = w’.nx [student(x)(w’) & came(x)(w’)] ]
C’ nwp.p = w’.nx [student(x)(w’) & came(x)(w’)]
wh-2 Pwp.n[P(n)(w)(p)]
IP w’.nx [student(x)(w’) & came(x)(w’)]
Q qwp.p=q
IP xw’.came(x)(w’)
WhP1 Wh0 t2- MANY
NP
t1 came
students
Qw’.nx [student(x)(w’) & Q(x)(w’)]
(122) (I know) which students came. CP wp.x [ p = w’.student(x)(w’) & came(x)(w’)]
C’ xwp.p = w’.student(x)(w’) & came(x)(w’)
wh-2 Pwp.x[P(x)(w)(p)]
IP w’.student(x)(w’) & came(x)(w’)
Q qwp.p=q
IP xw’.came(x)(w’)
WhP1 Wh0
NP
t2- (WH)ICH
students
t1 came
Qw’.student(x)(w’) & Q(x)(w’)
Maribel Romero
(123) (I know) whether any students came. CP wp.n [ p = w’.x [student(x)(w’) & came(x)(w’)] ]
C’ nwp.p = w’.x [student(x)(w’) & came(x)(w’)]
wh-2 Pwp.n[P(n)(w)(p)]
IP w’.x [student(x)(w’) & came(x)(w’)]
Q qwp.p=q
IP xw’.came(x)(w’)
WhP1 Wh0 t2- function j
NP
t1 came
students
Qw’.x [student(x)(w’) & Q(x)(w’)]
(124) [[know]] = λQλxλw. ∀w’ ∈ Epix (w) [ λp. p(w’) & Q(w’)(p) = λp. p(w) & Q(w)(p) ]
Notes . For a unification of the first LF/semantic condition and Rooth’s focus conditions, see Fox (2000b). . The full-fledged version (10a) is grammatical according to Schwarz (1998). Some other German speakers, though, consider the sentence odd and would rather say auch den Peter (“also Peter”). Judgments do not converge on whether the addition of auch is necessary. But, if we take a Logic class context and utter (i) and (ii), we can see Schwarz’s contrast: the full versions are certainly grammatical in English without too, and the German speakers I consulted accepted them in German as well for these scenarios; the point is that the corresponding German reduced conditional is ungrammatical in these contexts no matter what. (i) This much I can tell you: if I visit Karl, I visit Peter. (ii) Karl and Peter are inseparable. Thus, whenever I visit Karl, I visit Peter. However, there is certainly a preference for adding auch in the full conditional. In fact, the analysis that I will propose does not capitalize on the contrast between ellipsis and nonellipsis, but between on focus on den Peter and lack of focus on den Peter. Focusing den Peter in the full conditional, hence, sometimes gives us the same results as a reduced conditional. See Romero (2000b) for a more detailed discussion of the issue and for the addition of auch. . I take the term “free Focus” from Jacobs (1984) and Krifka (1991). They analyze it as associated with an illocutionary operator ASSERT, but I follow Rooth (1992b) in licensing it through its relation with its question (subset condition). . See the appendix at the end of this paper for examples of the compositional semantics for questions.
Correlate restriction and definiteness effect in ellipsis . This derivation of the exhaustivity implicature is inspired – though not completely faithful to it – by the discussion in von Fintel and Heim (1999). Also, let me note that there are mention-one Q/A pairs where the exhaustivity implicatures do not arise, as in (i). In this case, the speakers are endorsing the Maxim of Quantity only to the extent that the conversation requires it – one answer suffices– and thus the exhaustivity implicature does not arise. We will see that mention-one cases also occur in reduced conditionals. (i) Q: Where in Philadelphia can I buy Spanish ham? A: At SarCOne’s. . As stated in (32), the membership condition can be fulfilled not only by direct membership, but also by indirect membership through implicational bridging. This proviso is needed in order to allow for the focus in (i) (deaccenting in italics): (i) First [IIP1 John told Mary about the budget cuts]I and then [IIP2 SUEF heard about them]. (Rooth 1992a) . Schwarz discusses two other differences: exhaustivity effects and presupposition effects. These are closely related to the correlate restriction property and will be presented here simply as corollaries of the analyses. . In Section 5.3, we will see that, even in sluicing, a situation coindexation analysis plays no role in deriving the two characteristics at issue, since definiteness effects arise and disappear depending on the position of focus. . For the reader’s convenience, I use English words in the following LF representations. I will also spell out the elided constituent syntactically at LF for exposition purposes. Nothing in the present proposal hinges on whether the elided constituent is syntactically recovered at LF or only semantically recovered in the semantic component. . See Chierchia (2000) for a more extensive proposal where conversational implicatures are computed locally per clause and not only at the level of the complete utterance. . This way of deriving the restriction on correlates and the presupposition effects is parallel to Schwarz’ (though I use focus rather than coindexation). However, in the new approach, there is still another possible avenue to explore. First, it can be argued that an if -clause with a name as correlate does not give rise to the necessary implied question for the focus subset condition to be met. Second, the presupposition effects follow from a principle ruling Q/A pairs: given that the question denotation of Q is the set of possible answers to Q (Hamblin 1973), any true answer to Q should belong to the denotation of Q. In this sense, the RC in (62) presupposes that Simone is a student for the same reason that the dialog in (i) does: (i) Q: Which student did you visit? A: Simone. . Romero (2000b) discusses other empirical handicaps of Schwarz’ account and develops an explanation for them within the focus framework. . See Lahiri (1991) and Williams (2000) for a completely different account of the quantificational variability effects that Berman attributed to the open formula status of wh-phrases. . The formulae in the examples below spell out the property of knowing the questions at issue. See the appendix for the lexical meaning of know.
Maribel Romero . Some combinations of correlate NP type and remnant NP type yield marginality in gapping, as in (i). This issue is left to future research. (i) a. ??Every boy sent letters to Sue and PaTRIcia to PAT. b. ??Most boys sent letters to Sue and some GIRLS to PAT. . See Fox (1995) and Tomioka (1995) for application and discussion of the options to scope parallelism. . The NP part of wh-phrases can be reconstructed into IP at LF in order to capture the reconstruction effects found for how many (Heycock 1995; Romero 1997a; Fox 2000b) and for which phrases (though see Reinhart 1992; Rullmann & Beck 1998; and Romero 1999 for a less naïve implementation for which phrases). δ abbreviates <<e, st>, <<e, st>, <st>>>. . Note that, although [[nobody came]] does not belong to [[who came]](w) for a world w where nobody came, the semantics of know guarantee that to know who came entails to know that nobody came in such w. The same idea applies to how many and whether questions. Besides Heim (1994) on know and on weak and strong exhaustivity, see Beck and Rullmann (1999) and an alternative approach in Groenendijk and Stokhof (1984). Nothing in this paper hinges on this choice.
F-marking and specificity in sluicing constructions Kerstin Schwabe This paper shows that in various sluicing types, the wh-phrase in the sluicing sentence as well as its relatum in the antecedent clause must be F-marked, and it explains this observation with Schwarzschild’s (1999) and Merchant’s (2001) focus theory. Complying with the semantics of the wh-phrase, we will argue that the relatum of the wh-phrase is an indefinite expression that must allow a specific interpretation. Following von Heusinger (1997, 2000), specificity will be defined as an anchoring relation between the discourse referent introduced by the indefinite expression and a discourse-given item. Because specific indefinite expressions are always novel, contexts such as the scope of definite DPs, the scope of thematic matrix predicates, and the scope of downward-monotone quantifiers which all exhibit non-novel indefinites do not allow sluicing.*
.
Introduction
Sluicing constructions as in (1) lead to many interesting problems related to ellipsis, specificity, and sentence types. Thus, it is a worthwhile topic to show the interface between syntax, semantics and pragmatics as well as to discuss the status of information structure within these three domains. (1) Peter is reading, but I don’t know what
.
The present paper puts forward an integrated analysis of two factors that play an important role in the semantic licensing of sluicing constructions, namely the focus and the specific-indefinite restriction on the antecedent. The focus analysis is based on Schwarzschild (1999) and Merchant (2001). With respect to the specificindefinite restriction on the antecedent, the paper proposes a solution within the theory of von Heusinger (1997, 2000). It analyses indefinite expressions as modified epsilon terms and shows that the accessibility conditions of the indefinite in sluicing depend on the semantics of wh-phrases. The semantics of the wh-phrases presupposes (i) that the entity denoted by the indefinite in the first conjunct must
Kerstin Schwabe
not be anchored in the discourse representation for a discourse participant a and (ii) that it is anchored for another discourse participant b. The first presupposition prevents the entity from being expressed by a definite. The second prohibits that the entity occurs in contexts that do not allow a specific interpretation. Finally, the paper shows in various applications that the focus restriction and the specific-indefinite restriction are two mutually dependent restrictions. Before we formulate the problems associated with sluicing constructions such as (1) and try to handle them, let’s first see what is meant by the notion of sluicing. A sluicing construction consists of two conjoined sentences with the first one being the antecedent sentence (AS) and the second one the sluicing sentence (SS). The latter is either a mere wh-phrase as in (2) or, as in (1), consists of a matrix clause (MC) and an embedded wh-clause. (2) a. A: What is Hans doing? b. B: Hans is reading a book. c. A: Which one? The former is only possible if the antecedent sentence and the sluicing sentence are conjoined asyndetically or if the antecedent sentence is a wh-interrogative. This is due to the fact that with syndetical coordination both conjuncts have to agree with respect to their sentence type and that the antecedent sentence is always a declarative. What is characteristic for sluicing is that the wh-clause – we call it sluicing clause (SC) – contains merely a wh-phrase. The antecedent sentence includes the antecedent clause which renders the antecedents for the deleted material in the sluicing clause. It also introduces the discourse referent the wh-phrase is related to. We will call this discourse referent relatum. It is the same as Chung’s et al. (1995) “inner antecedent” and Merchant’s (2001) and Romero’s (2000a) “correlate”. The proposition that contains the relatum will be labeled relatum proposition. Usually, but not always, the antecedent clause denotes the relatum proposition. Cases where the relatum of the wh-phrase is not contained in the sentence that immediately precedes the sluicing sentence are the following – cf. Merchant (2001): (3) a.
There was a party yesterday. Do you know who was at this party? beth was there, but I don’t know who else. b. Sheila has some cats and dogs. Do you know how many dogs and cats she has? She has five cats, but I don’t know how many dogs.
Here the antecedent sentences are non-exhaustive answers to contextually given questions that relate to a sentence introducing the relatum of the wh-phrase in the sluicing clause. The stress on the subject in (3a) or on the object in (3b) in the
F-marking and specificity in sluicing constructions
sentence that precedes the sluicing clause indicates that there are alternatives given by the discourse. Many authors dealing with sluicing phenomena, for instance Romero (2000a), have discussed the behavior of sluicing with respect to “weak” (i.e., selective) islands that raise the greatest puzzle for sluicing as in (4). In the sluicing case (4a), the wh-phrase may escape the island whereas it cannot in the corresponding full version (4b). (4) a.
Sandy was trying to work out which students would be able to solve a certain problem, but she wouldn’t tell us which onei she was trying to work out [which students would be able to solve ti] b. *Sandy was trying to work out which students would be able to solve a certain problem, but she wouldn’t tell us which onei she was trying to work out [which students would be able to solve ti ]
That the wh-phrase seems to be channeled or sluiced through syntactic islands within these constructions was the reason that such constructions were labeled as sluicing. Merchant (2001) and Schwabe (2000), on the other hand, are largely devoted to a novel solution to the syntactic island problem. They argue that there is no need to assume islands with respect to sluicing constructions such as (4a). Therefore it would be better to call these constructions wh-ellipsis. As to the characteristics of sluicing we are interested in, we may first mention the commonly known property that the relatum of the wh-phrase is always an indefinite which is either an overt expression as in (1), (2), and (4) or given implicitly by the verb semantics as in (5). (5) Hans left after his mother cooked and he didn’t want to tell us what. Second, we can observe that the indefinite no matter whether explicit as in (6a) or implicit as in (6b) must always be focus-marked or be within a focus-marked constituent. (6) a. Peter has taken a shirt, but I won’t tell you which one. a . *It is peter who has taken a shirt, but I won’t tell you which one. b. Peter is reading, but I won’t tell you what. b . *It is peter who is reading, but I won’t tell you what. Third, the antecedent clause and the sluicing sentence(s) need not be conjoined symmetrically, but the sluicing sentence is always adjacent to the antecedent clause – cf. (Ross 1969). (7) a.
Hans is reading a book, but I don’t know which one. [AC & SS]
Kerstin Schwabe
b. They wanted to hire a linguist who speaks a Balkan language, but they didn’t tell us which. (Merchant 2001) [AS . . . [AC]] & SS c. Paul saw that John killed a girl and because he knew which one, he didn’t go to the police. AC & [[ SS ] . . . ] d. Hans left after his mother cooked and he didn’t want to tell us what. [ . . . [AC & SS] . . . ] e. A girl has got dirty a table-cloth and I want to know which girl and which table-cloth. AC & [SS . . . [ SC & SC ] . . . ] f. This report details what IBM did and why. (Merchant 2001) [SS . . . [ AC & SC] . . . ] Fourth, the indefinite must not occur in contexts that do not allow a specific interpretation of the indefinite. Contexts of this kind are for instance the description of definite DPs (8), complements of thematic matrix predicates (9), the scope of downward-monotone quantifiers (10) (cf. Albert 1993 and Romero 2000a), and the dependency on non-specific indefinite DPs (11). (8) a. *They found the man yesterday who has kissed a woman, but they won’t tell us which one. b. *Yesterday, I bought the book about a politician, but I’ve forgotten about which one. c. *Yesterday, I saw the boy who was reading, but I cannot say what. (9) a. *Ramon is glad that Sally was dancing with a boy, but I don’t remember with which one. b. *They regretted that they were reading, but I don’t know what. (10) a. *They hired few people who spoke a lot of languages – guess how many! (Merchant 2001) b. *Joan rarely read any book, but I don’t know which one. c. *They hired no people who spoke a lot of languages – guess how many! d. *John never makes any joke when he has guests, but I don’t know which one. e. *John rarely sings any song when he has guests, but I don’t know which one. f. *Paul didn’t want to read any book, but I don’t know which one. g. *Few kids ate, but I don’t know what. (Romero 2000a: 200) h. *Joan rarely fed my fish, but I don’t know with which product. i. *They met no people who were reading, but they did not tell us what.
F-marking and specificity in sluicing constructions
j. *John never cooks himself when he has guests, but I don’t know what. k. *John rarely cooks himself when he has guests, but I don’t know what. l. *Paul didn’t want to read, but I don’t know which book. (11) ?They are looking for some linguist who has written a thesis, but they cannot tell you which one. And finally fifth, the relatum must be contained in a proposition that is denoted by the antecedent clause or can be inferred from it. This proposition is supposed to be true by the attitudinal subject of the sluicing sentence. Thus, the interpretation of (12) succeeds if the anticipated proposition ‘the addressee buys a book’ is true in some situation. This can be expressed by then as in (12a) or by the conditional as in (12b). It fails when both conjuncts are interpreted as being only a sequence of imperatives as in (12c). (12) a. Go and buy a book and (then) tell me which one! b. Go and buy a book and (if you buy one,) tell me which one! c. *Go and buy a book and tell me which one! We will see in section two that the focus condition on the indefinite, that the affirmative proposition condition, and the adjacency condition can be derived from Schwarzschild’s (1999) and Merchant’s (2001) focus theory. In section three, we will explain the observation that sluicing is not well-formed if the relatum is in a thematic context or depends on a non-specific DP. There we take the semantics of wh-phrases as a starting point, which demands that the relatum of the wh-phrase must be an indefinite and that this indefinite must allow a specific interpretation. On the basis of von Heusinger’s (1997, 2000) theory on specificity we will explain why the various contexts given in (8) to (11) prohibit sluicing.
. Information structure of the antecedent clause and the sluicing clause According to Schwarzschild (1999), F-markers are freely assigned and subject to constraints such as foc, HeadArg, givenness, and avoidF. foc demands that a F-marked phrase contains an accent if it is not immediately dominated by another F-marked node whereas HeadArg regulates that a head is less prominent than its internal argument. avoidF prevents F-marking more phrases than necessary whereby givenness must not be violated. The latter constraint says that a constituent that is not F-marked must be given. As to Schwarzschild’s definition of given see (13):1
Kerstin Schwabe
(13) i.
Definition of Given (informal version) An utterance U counts as given iff it has a salient antecedent A, and if U is of type e, then A and U corefer; otherwise: modulo ∃–type shifting, A entails the Existential Closure of U. ii. Existential Closure of U (F-clo (U)) The result of replacing F-marked phrases in U with variables and existentially closing the result, modulo existential type shifting
It follows from Schwarzschild’s theory that only given constituents must be licensed and that F-marked constituents may be either novel or given. Turning to the possibility of ellipsis as in the sluicing clause, Merchant (2001) has shown that Schwarzschild’s focus theory must be extended to ensure the semantic identity of the phonologically empty material with the antecedent material it corresponds to. Thus, the IP in the sluicing clause can only be deleted if the sluicing clause satisfies e-givenness. (14) e-givenness (Merchant 2001) An expression E counts as e-given iff E has a salient antecedent A and modulo ∃-type shifting, (i) A entails F-clo(E), and (cf. Schwarzschild 1999) (ii) E entails F-clo(A). Note that ‘F-clo’ corresponds to Schwarzschild’s Existential Closure in (13). As we may see with respect to (15), the matching of the information structural properties of the sluicing and the antecedent clause with e-givenness entails that the whP as well as its relatum must be F-marked and that the antecedent clause must be propositional. (15) They hired a linguist who knows a [balkan language]F but I do not know [which oneF he knows] Here the antecedent clause is the relative clause of the first conjunct where only the object a Balkan language is F-marked. Since the IP in the sluicing clause is given, it must fulfil e-givenness. According to the definition of e-giveness in (14i), the antecedent clause entails the existential F-closure of the sluicing clause (16i). And vice versa, according to (14ii), the proposition derived from the interrogative sluicing clause by existential type shifting entails the existential F-closure of the antecedent clause (16ii). We get the existential F-closure of the sluicing clause by binding the variable that is given by the focused wh-phrase existentially.2 (16) i. He knows a Balkan language → ∃x [ know (he) (x)] (= [[AC]])
F-marking and specificity in sluicing constructions
ii. ∃x [ know (he) (x)] (= [[SC]])
→ ∃x [ know (he) (x)]
In that the antecedent clause must be a proposition, it is a non-restrictive relative clause. This means it cannot be interpreted as a restrictive relative clause, since the latter is of type <<e,t><e,t>>. Additionally, it is a judgement because the adversative sluicing sentence can only be related to a proposition that is true. The next example shows what happens if the whole IP of the antecedent sentence is F-marked. (17) They invited a boy who [knowsF a NorwegianF ]F but I do not know [which oneF ] he knows. (i) He knows a Norwegian → ∃x [ know (he) (x)] (ii) ∃x [ know (he) (x)] → ∃x ∃Q [ Q (he)(x)] IP-ellipsis in the sluicing clause is possible because e-givenness is satisfied. That the relatum of the wh-phrase must be F-marked follows, as we can see in (16ii) and (17ii), from (ii) in e-givenness (14). E-givenness also explains why the VP must be F-marked if the relatum is expressed implicitly. According to (14ii), it must be F-marked so that the existential F-closure of the antecedent clause can be entailed by the sluicing clause. (18) She is writingF , but I can’t imagine whatF . (i) She is writing → ∃x [write (she) (x)] (ii) ∃x [write (she) (x)] → ∃Q [Q (she)] That the relatum of the wh-phrase can also be an unspecified argument of a relational noun can be seen in the next example: (19) Maria has [boughtF ticketsF ]F , but she doesn’t tell us for which filmF . Up to now, the antecedent for the sluicing clause was always a proposition that was expressed by the antecedent clause. But there are cases where the sluicing clause relates to a proposition that must be derived from the antecedent clause of the sluicing clause: (20) a.
Sandy was trying to identify which student solved a certain problem, but she wouldn’t tell us which one. (Merchant 2001) b. Peter told me who Mary met and why. c. Did Peter buy a book and do you also know which one? d. Go to the party, but do not tell me with whom!
The propositions that are to be derived are something like: ‘The student that Sandy has identified solved a problem’ for (20a), ‘Mary met somebody’ for (20b), ‘Pe-
Kerstin Schwabe
ter bought a book’ for (20c), and ‘Hearer goes to the party’ for (20d). Following Schwarzschild (1999: 157), let us try to use existential type shifting to obtain a proposition out of the interrogative antecedent clause in (20a) by binding the free variable by an existential operator and checking whether e-givenness (14) is met – cf. Note 2. ∃x ∃y [student (x) ∧ problem (y) ∧ solve (x) (y)] ∃y ∃x [student (x) ∧ problem (y) ∧ solve (x) (y)] ii. ∃y ∃x [student (x) ∧ problem (y) ∧ solve (x) (y)] ∃y ∃x [student (x) ∧ problem (y) ∧ solve (x) (y)]
(21) i.
→ →
We observe that e-givenness is met in (21), where the subject in the antecedent clause, which is represented similarly to an indefinite, is copied into the sluicing clause. However, (21) does not account for the fact that the subject of the sluicing clause must be an anaphoric expression as indicated in (22): (22) Sandy was trying to identify which student solved a certain problem, a. *but she wouldn’t tell us which (a student solved). b. but she wouldn’t tell us which one (the student she has identified solved). This example as well as (20b) show that we cannot gain the necessary antecedent proposition by existential type shifting of the interrogative antecedent clause, but by accommodating an answer to the question that contains an anaphoric expression such as ‘the student that Sandy has identified solved a certain problem’ or ‘Mary met the person she met’ respectively. These propositions can now be used to check whether e-givenness is met. def. student solved a certain problem → ∃y [problem (y) ∧ solve (def. student) (y)] ii. ∃y [problem (y) ∧ solve (def. student) (y)] → ∃y [problem (y) ∧ solve (def. student) (y)]
(23) i.
Turning to (20c), we notice that there it is also not possible to obtain the antecedent proposition for the sluicing clause by existential type shifting of the yes-no interrogative. ∃f ∃x[f (book (x) ∧ read (peter) (x))] → ∃x [book(x) ∧ read (peter)(x)] ii. ∃x [book(x) ∧ read (peter)(x)] → ∃f ∃x[f (book (x) ∧ read (peter) (x))]
(24) i.
The entailment relation would be invalid if the variable ‘f ’ were instantiated by a function that mapped the proposition onto a negative proposition – cf. Note 2:
F-marking and specificity in sluicing constructions
(25) i. ¬ ∃x[book (x) ∧ read (peter) (x)] → ∃x [book(x) ∧ read (peter)(x)] ii. ∃x [book(x) ∧ read (peter)(x)] → ¬ ∃x[book (x) ∧ read (peter) (x)] Since the antecedent proposition cannot be obtained by existential type shifting, it must be derived in some other way. It can be derived by accommodating the affirmative answer to the question given by the antecedent clause. As to the imperative in (20d), the antecedent is the accommodated proposition that represents the action the addressee is asked to do. Having the necessary antecedent proposition and relating it to the type-shifted sluicing clause, it is easy to work out that e-givenness is met. So far we have shown and explained that and why the relatum in the antecedent as well as the wh-phrase in the sluicing clause must be F-marked. That Merchant’s and Schwarzschild’s theory also holds for all sluicing types given in (7) is easy to work out. Additionally, we have demonstrated that if the antecedent clause is nonpropositional, the antecedent proposition must be derived by accommodation.
. Appropriate and inappropriate contexts for sluicing . The need for specificity Recall that contexts that do not allow for sluicing are the description of definite DPs (see (8) or (26a)), the description of complements of thematic matrix predicates (see (9) or (26b)), the scope of downward-monotone quantifiers (see (10) or (26c)), and the dependency on non-specific indefinite DPs (see (11) or (26d)). (26) a. *They found the man who has kissed a woman, but they won’t tell us which one. b. *Ramon regrets that Sally was dancing with a boy, but I don’t remember with which one. c. *They hired few people who spoke a lot of languages – guess how many! d. ?They are looking for some linguist who has written a thesis, but they cannot tell you which one. With Heim (1982) and Schwarzschild (1999), we regard the referent of a definite DP as an entity which is thematic or given, respectively. But to be given need not mean that it must have been mentioned in the current discourse or that it is prominent in the utterance situation. An entity can also be seen as given if it is anchored in the mental lexicon of the discourse participants. It can be retrieved from there and introduced as a novel discourse referent into the current discourse.3 Let us assume that as the description of definite DPs, also the description of thematic complements and the scope of downward-monotonic quantifiers are thematic,
Kerstin Schwabe
that means given. According to avoidF and givenness, the constituents in these contexts actually need not be F-marked. That according to givenness, non-F-marked constituents must be given does not mean that all F-marked constituents must be non-given. Or to formulate the question in another way: Are there given elements that can be asked for? Schwarzschild (1999: 158ff.) shows that there are cases like (27) where a given constituent must be F-marked to satisfy givenness. (27) Who did John’s mother praise? A: She praised [him]F Here the object in the answer must be F-marked because the existential F-closure of the answer must be entailed by the type-shifted question. If it were not F-marked, existential F-closure could not take place. Now we could ask whether the given definite DP can be F-marked. The answer is yes, as long as it can be asked for and thus a givenness effect can obtain. To demonstrate this, we take (28a) as a contextually given question. With this question, the whole DP in the answer (28b) must be F-marked. (28) a. They have found somebody, but I don’t know who? b. They found [the man who kissed a woman]F. According to Schwarzschild’s (1999: 170) Foc-constraint, Foc-marked material must be accented. Therefore woman carries the pitch accent. The question arising now is why the indefinite in thematic contexts cannot be related to by the wh-phrases in the following sluicing constructions: (29) *They found [the man who has kissed a woman]F , but they won’t tell us which one. (30) *Ramon regrets [that Sally was dancing with a boy]F , but I don’t remember with which one. We suggest that an indefinite in a thematic context cannot be related to by a whphrase if the entity it denotes is interpreted as non-specific by the attitudinal subject of the wh-interrogative. We consider the latter to be the subject that poses the question. It can either be expressed explicitly within the matrix proposition of the sluicing sentence, or it can be the speaker if the sluicing sentence consists only of a wh-phrase like in (2c). That the relatum of the wh-phrase must be an indefinite and that this indefinite must allow for a specific interpretation for the attitudinal subject is presupposed by the wh-phrase. Let’s suppose that a wh-question is something like an instruction to choose a value for a variable out of a value set.4 This value set is denoted by the restriction of the wh-phrase. Thus the wh-phrase presupposes first a
F-marking and specificity in sluicing constructions
value set that is not a singleton. Second the wh-phrase presupposes that the choice of a particular value out of this set is possible. Both conditions are necessary to get a coherent answer for the question. As to the antecedent clause for a question, the value set is denoted by the description of the relatum DP in that clause. This DP can only be an indefinite DP because the value set for an indefinite DP need not be a singleton and because indefinites allow the choice of a particular value for the variable they introduce. If there is a choice of a particular value for a value set, we speak, following Farkas (2002), of a specific interpretation of the indefinite or, to be short, of a specific indefinite. A definite DP, on the other hand, has a value set that is a singleton. This prevents it from serving as the relatum for a wh-phrase. Let us return to contexts as in (26a) that do not allow a specific interpretation of the indefinite a woman for the attitudinal subject they. Notice that the antecedent of the attitudinal subject is not contained in the thematic antecedent clause, but in the non-thematic matrix clause. Now the question arises why the attitudinal subject of the sluicing sentence cannot have access to a discourse referent introduced by an antecedent clause as in (26a–c) which contains given or thematic material. If an indefinite is given, a discourse referent with the same description has been introduced before and has not been assigned a value, and has thus existentially been bound. This happens if the discourse referent is not relevant to the subsequent discourse. If it is not relevant, it, metaphorically speaking, logs out or goes offline, respectively. Then it can go lost and can hardly be retrieved anymore.5 A discourse referent goes online when it is introduced or logged in by an indefinite expression in a particular sentence (see Heim’s (1982) Novelty condition). If the discourse referent is needed for the ongoing discourse as in the sequence of an antecedent clause and a sluicing clause, this means transsententially, it must stay online and thus be anchored to the discourse. It is then anchored to a further discourse referent and thus accessible to the attitudinal subject of the sluicing sentence. As we can see with respect to the complements of the thematic predicates in (26a–c), they only consist of one clause which means that within this thematic context, the discourse is not continued. It follows that the discourse referent introduced by the indefinite is not anchored to the discourse and thus not accessible to the attitudinal subject. But what happens if the discourse proceeds in thematic contexts? The next examples show that sluicing is possible also in thematic contexts. Sluicing there only obtains there if the attitudinal subject is in the same thematic context. They found [the man who has kissed [a woman]F, the man and who didn’t tell us which one]F . b. *They [found the man who has kissed a womanthey ]F and I won’t tell you which one.
(31) a.
Kerstin Schwabe
(32) a.
Ramon regrets [that Sally was dancing with [a boysally ]F and that she didn’t remember with which one]F . b. *Ramon regrets [that Sally was dancing with a boyramon ]F and he doesn’t remember with which one.
In (31a) and (32a), the discourse referent introduced by the indefinite can only be anchored to the subject of the embedded antecedent clause and not to the subject of the matrix clause or to the speaker. If it is anchored to the subject of the embedded relative or complement clause, it can be specific for the attitudinal subject of the sluicing sentence. From this we conclude that the discourse referent introduced by the indefinite in thematic clauses can only be anchored to a discourse referent that is introduced by this thematic proposition. If, on the other hand, the respective proposition is non-thematic, it can be anchored to a discourse referent either introduced by this proposition as in (33a) or by an embedding proposition as in (33b). Or it can even be anchored to the speaker as shown in (33c). Peter told us that Karl kissed a womankarl , but hepeter cannot tell you which one. b. Peter met a boy who kissed a womanpeter , but hepeter cannot tell you which one. c. Peter wants to read a Norwegian novelspeaker , but I don’t tell you which one.
(33) a.
That the relatum of the wh-phrase must allow a specific interpretation for the attitudinal subject also holds for the relatum of the whatP as in (34), which is often thought to be non-specific. (34) A: Peter is reading a book, but I do not know what kind of book (the book he is reading is). B: The book he is reading is a boring one. The whatP asks for a property of a specific DP. This means it asks for a further predication of an online discourse referent. This is attested in (34) by the full-fledged version of the sluicing clause and by the definite expression in the answer. Let’s conclude: On the one hand, the relatum of a wh-phrase must allow a specific interpretation for the attitudinal subject of the sluicing sentence. It can only be specific if it is online for the attitudinal subject. On the other hand, an indefinite DP in a thematic context cannot be interpreted as being specific if the attitudinal subject of the sluicing sentence is outside this thematic context. Then the information-structural status of the indefinite tells the attitudinal subject of the sluicing sentence that there is a given but offline discourse referent. That this discourse referent has gone offline is due to its irrelevance to the discourse. This
F-marking and specificity in sluicing constructions
irrelevance is passed on to the subsequent discourse so that the discourse referent introduced by the indefinite in thematic contexts has no choice but to log out. This contradiction explains why the discourse referent that is introduced by an indefinite in a thematic context is not accessible to an attitudinal subject and thus to the wh-phrase outside the thematic context. In the following section, we will see how the notion of specificity, up to now given pretheoretically, is modeled in von Heusinger’s (1997, 2000) theory.
. The representation of specificity in sluicing As von Heusinger (1997, 2000) explains, indefinite DPs can vary in their referential properties along (at least) two dimensions: scope and specificity. To represent these independent properties appropriately, we take von Heusinger’s (1997, 2000) theory, in which indefinite DPs are represented as indexed epsilon terms. This is illustrated in (35): (35) a painting:
εi x [painting(x)]
The epsilon operator is interpreted as a choice function that assigns to each (nonempty) set one of its elements. In other words, the referent of an indefinite DP is found by the operation of selecting one element out of the set that is described by the description. The selection depends on the context in which the indefinite is located. This treatment is similar to that discourse representation theories (Heim 1982; Kamp 1981), where indefinites introduce new individual variables or discourse referents. One of the main advantages of using choice function variables instead is the following: Indefinites need not be moved or raised to express different dependencies. They remain in situ, whereas the choice function variable can be bound by different operations, e.g adverbs of quantification, existential closure, etc. This causes different scope readings of the indefinites. Specificity is taken as an independent referential property of indefinite DPs (see Fodor & Sag 1982; Enç 1991; Farkas 1995, 2002). Following von Heusinger (2002), we assume that a specific indefinite DP is “referentially anchored” to a discourse item. This can be the speaker or some other index of the utterance context, on the one hand, or some introduced referent, on the other. Since the discourse referent is anchored to some discourse participant, it can stay online and be subject to further linguistic operations. The anchor relation is represented by a function f from that discourse item to a certain choice function. In other words, the function f links the choice of the indefinite to the value of this discourse item. The indefinite receives therefore the same scope as the discourse item it depends on. If the indefinite DP is not anchored and thus goes offline, its context index variable is existentially bound.
Kerstin Schwabe
Example (36) illustrates the different referential options of the indefinite. The example can be assigned a non-specific reading of the indefinite (“There is some painting by Picasso or other such that John likes it”), as in (36a). The more prominent specific reading (36b) can be paraphrased as “I can identify a picture and this picture is such that John admires it”. There is another specific reading of (36), namely (36c) with the paraphrase “John has a particular picture of Picasso in mind, and he admires it, but I cannot tell which one”.6 (36) John admires a painting of Picasso. a. ∃i [admire(john) (εi x [painting(x)])] (non-specific) b. admire(john) (εƒ(speaker) x [painting(x)])] (specific: speaker-anchored) c. admire(john) (εƒ(john) x [painting(x)]) (specific: subject-anchored) (36b) and (36c) differ in that the indefinite is anchored to different discourse items. The different referential properties of indefinite DPs are additionally dependent on the information structure (see Lenerz 2001) and on other constructions, such as coordination (see Schwabe & von Heusinger 2001). Having the two necessary ingredients: the need for specificity and the appropriate representational format, we can now represent the different contextual behavior of antecedent clauses. If the relatum of the wh-phrase in the sluicing sentence must allow a specific interpretation, the context index of the epsilon operator in the semantic representation of the relatum must be substituted by a function ƒ from some discourse item to a certain choice function. This means that the function ƒ assigns to the discourse item a particular choice function, and thus a particular element that is assigned to the given set. In the following example the function ƒ relates the particular choice function to the speaker: (37) Peter is dancing with a girl, but I won’t tell you with which one. peter is dancing with εƒ(speaker) z [girl (z)], but . . . wh (z): girl(z): peter is-dancing-with z If the relatum is in the scope of a universal quantifier as in (38), the function ƒ relates the particular choice function to a particular boy – each boy has his own choice of a particular girl. (38) Every boy was dancing with a girl, but I don’t know with which one! Every(x): boy(x): x was dancing with εƒ(x) z [girl(z)], but . . . wh (z): girl (z): Dist (x): boy (x): x was dancing-with z
F-marking and specificity in sluicing constructions
The answer to such a sluicing sentence would be a pair-list answer such as Peter was dancing with Petra, Paul was dancing with Maria, . . . . This example shows that to get the specific-narrow scope reading in the sluicing clause, there must be an intervening operator between the wh-phrase and its trace. The distributing operator in (38) is necessary to prevent the cumulative reading. It distributes over the set of boys such that each boy dances with a particular girl. Contrary to Romero (2000a: 197ff.), the example (39) shows that also a sluicing clause with a non-overt relatum may contain an operator: (39) Every boy was dancing last night, but I won’t tell you with whom (they were dancing /each/every boy was dancing). She bases her claim on the scope parallelism requirement between the antecedent and the sluicing clause (Chung et al. 1995) and on the observation that implicit indefinites have narrowest always scope (Fodor & Fodor 1980). In her framework, the wh-phrase in the sluicing clause has wide scope, and because the implicit indefinite in the antecedent clause must have narrow scope, the parallelism requirement is not met. If there are any “apparent intervenors” as in (38) between the wh-phrase and its trace, she translates the QP into an E-type pronoun that doesn’t count as an intervenor anymore. But her proposal does not hold because a distributing operator is needed to interpret the predicate in the sluicing clause – see (38). Additionally, it is not true that implicit indefinites must always have narrowest scope. There are cases like in the following example where also implicit indefinites may have wide scope: (40) A: Every child in the kindergarten is dancing, but I do not know with whom. B: With Agnes, I believe. We can also construe a context where the indefinite DP in (38) has wide scope like the implicit indefinite in (40). Then the choice of the indefinite DP depends on the speaker or some other discourse participant: (41) Every(x): boy(x): x was dancing with εƒ(hearer) z [girl(z)], but . . . wh(z): girl(z): they were dancing-with z The relatum however cannot have a non-specific interpretation like the narrow scope one in (42) or the wide scope one in (43) because it would then not be accessible to the wh-phrase in the sluicing sentence. (42) *Every(x): boy(x): ∃i [x was dancing with εi z [girl(z)]], but . . . (43) *∃i [Every(x): boy(x): x was dancing with εi z [girl(z)]], but . . .
Kerstin Schwabe
As we have already mentioned, the specific reading of the relatum cannot obtain if the relatum is in the scope of a definite article or a thematic predicate and the attitudinal subject of the sluicing sentence is not. Since the description of definite DPs as in (44) and the complement of thematic matrix predicates as in (26a-c) are thematic or given, respectively, the indefinite expression in them is also given. To be given means for an indefinite DP that a discourse referent with the same description has previously been introduced, but has gone offline. That it has gone offline indicates that there wasn’t any interest in anchoring it. Because there is no need for its anchoring, the discourse referent that according to Heim’s (1982) Novelty condition is introduced by the indefinite expression in the antecedent clause is also not anchored – cf. (44) and (45). Thus sluicing always fails in such contexts. (44) *∃i [They found the man yesterday who has kissed εi x [woman (x)]] but they won’t tell us which one. (45) *∃i [Ramon is glad that Sally was dancing with εi x [boy (x)]] but I don’t remember with which one. That indefinite DPs in thematic antecedent clauses cannot be specific for attitudinal subjects outside this thematic context explains why their context index cannot be substituted with a function ƒ that relates a particular discourse item to a particular choice function. Their context index can only be bound existentially, which blocks them from being related to by the wh-phrase of the subsequent sluicing clause. That thematic relata are unsuitable antecedents for the wh-phrase outside the thematic contexts can also be attested with respect to downward-monotone quantifiers. Their scope is given by the context as well. Thus, they can only contain non-novel indefinite expressions and not render relata for the wh-phrase. But as Merchant (2001) and Romero (2000a) point out, constructions such as in (46) are evaluated as well-formed by some informants. (46) a. ?They hired few people who spoke a lot of languages – guess how many! b. ?Few kids were reading, but I don’t know what (they were reading each). This becomes possible when these informants interpret the expression few. . . as a plural set and not as a downward-monotone quantifier phrase. The plural set can be related to by an E-type pronoun in the sluicing clause (cf. Evans 1980). But to obtain the correct interpretation of the predicate in the sluicing clause, this set must be distributed. Since the set interpretation does not presuppose given material, the indefinite expression a lot of languages can be non-given and thus specific so that the choice function can be related to a particular discourse item. The following example shows that an indefinite DP is not accessible to a whphrase if this indefinite depends on a non-specific indefinite DP.
F-marking and specificity in sluicing constructions
(47) They are looking for a linguist who speaks a Balkan language, but they cannot tell you which. ∃i [They are looking for εi x [linguist(x)] & εi x [linguist(x)] speak εƒ(x) z [Balkan language (z)]], but . . . If the first indefinite DP a linguist is non-specific and the reference of the second indefinite DP a Balkan language depends on the first indefinite, the DP a Balkan language inherits the non-specificity of this DP. Then sluicing is not possible. The indefinite DP a Balkan language, however, can be specific if it is related to some discourse referent as for instance the speaker (48) or to the linguistically introduced discourse item a linguist which is related by the function ƒ to the subject of the antecedent sentence (49). (48) ∃i [They are looking for εi x [linguist(x) & speak (x)(εƒ(speaker) z [Balkan language (z)])]], but. . . (49) They are looking for εƒ(they) x [linguist(x) & speak (x)(εƒ(x) z [Balkan language (z)])], but . . . To sum up this section, we should record that the antecedent or relatum, respectively, of the wh-phrase must allow a specific interpretation for the attitudinal subject. For this reason, the scope of thematic predicates, the description of definite DPs, the scopi of downward-monotone quantifiers, and the dependency on nonspecific indefinite DPs cannot render the needed relata if the attitudinal subject is not in the scope of thematic predicates, articles and downward-monotone quantifiers as well as of non-specific indefinites. If, on the other hand, the attitudinal subject is in the scope of the above mentioned items, sluicing is obtainable. (50) a. Ramon regrets that Fred kissed a girl and didn’t tell him which one. b. Tom criticized the friend who kissed a girl and didn’t tell him which one. c. No one has read a book and didn’t say which one. d. They are looking for a linguist who knows a Balkan language and doesn’t tell them which one.
. Conclusion The observation that in various sluicing types, the wh-phrase in the sluicing sentence as well as its relatum in the antecedent clause must be F-marked was explained following Schwarzschild’s (1999) and Merchant’s (2001) focus theory. Furthermore, according to the semantics of the wh-phrase, it was argued that the relatum of the wh-phrase must be an indefinite that must allow a specific inter-
Kerstin Schwabe
pretation. According to von Heusinger (1997, 2000), specificity was defined as an anchoring relation between the discourse referent introduced by the indefinite expression and a discourse-given item. It has turned out that specific indefinite expressions are always novel or nongiven and thus F-marked. The reason is that they introduce a new discourse referent that is contextually anchored after its introduction. If there were already a contextually anchored discourse referent, it could not be an indefinite that could be used to pick up this discourse referent, but only a definite expression. Non-specific indefinites, on the other hand, can be given as well as non-given. In both cases, their context index is existentially bound, which means that the discourse referent they denote is not relevant to the discourse. A given indefinite merely indicates that a discourse referent with the same description has been introduced previously, has been considered to be irrelevant, and therefore has been logged out. Since specific indefinite expressions are always non-given, contexts such as the scope of definite articles, the scope of thematic matrix predicates, and the scope of downward-monotonic quantifiers that exhibit given indefinites do not allow sluicing. To stay online, specific discourse referents that are introduced by indefinites must be picked up by an anaphoric expression in the next sentence. This explains why the antecedent clause must be adjacent to the sluicing sentence. Indefinites that are in thematic contexts can be related to by a wh-phrase if the attitudinal subject of the sluicing sentence is identical with the discourse referent the indefinite is anchored to. This discourse referent can only be expressed by the proposition the indefinite is contained in. Since the proposition is a thematic context, there are no discourse referents available to which the indefinite could be anchored to be specific for the discourse outside the thematic context. In that, unlike Chung et al. (1995) and Romero (2000a), we see specificity as decisive for well formed sluicing constructions, we get the possibility of an unified account of sluicing with explicit and implicit relata and a more comprehensive and appropriate account of the failing of sluicing in the above mentioned contexts. Furthermore, we have shown that sluicing is nothing more than a text relation between an antecedent clause and a wh-question where ellipsis is possible because of Merchant’s e-givenness.
Notes * I am grateful to Jason Merchant, Susanne Winkler, Klaus von Heusinger, and John te Velde for initial discussions and for comments on the various written versions. . Schwarzschild (1999) defines existential type shifting as raising expressions to type t, by ∃–binding unfilled arguments.
F-marking and specificity in sluicing constructions . Following von Stechow and Zimmermann (1984) and Krifka (2001a), we consider a question to be a function which results in a proposition if it is mapped onto the meaning of its answer: (i) A: Who does Hans love? λx∈person [love hans, x] B: Anna. anna question mapped onto the answer: λx∈person [love hans, x] (anna) = love hans, anna (ii) A: Does Peter read a book? λf [ f (read (p) (b))] B: Yes. λp [p] question mapped onto the answer: λf [ f (read (p) (b))] (λp [p]) = read (p) (b) (iii) A: Does Peter read a book? λf [ f (read (p) (b))] B: No. λp [¬p] question mapped onto the answer: λf [ f (read (p) (b))] (λp [¬p]) = ¬read (p) (b) . That there are definite DPs that denote discourse referents that are novel with respect to the discourse is also discussed in Umbach (2002). She remarks that such definite DPs contain an accent whereas definite DPs that are given in the discourse do not. An accent in a definite DP indicates that it is either F-marked itself or is dominated by a F-marked constituent. . As to the notion of ‘value set’ see Farkas (2001). . Krifka (2001b) terms given indefinite NPs as “non-novel indefinites”. He discusses them in the context of adverbial quantification and information structure, in examples like (i) and (ii). An indefinite NP in the background is marked as non-novel (=NN). The difference in information structure determines the domain of quantification as illustrated in the paraphrases: (i) [A freshman]NN usually wears a baseball cap. “Most freshmen wear a baseball cap” (ii) A freshman usually wears a [baseball]NN cap. “Most baseball caps are worn by freshmen” . The formulations “has in mind” or “can identify” should motivate the specific reading. However, such formulations are very informal, and in certain contexts even misleading (see von Heusinger 2002 for a detailed discussion).
The semantics of Japanese null pronouns and its cross-linguistic implications Satoshi Tomioka Phonologically silent pronouns in Japanese receive a wide variety of semantic interpretations. This paper relates this semantic diversity to the general fact about Japanese NPs that they do not require determiners and receive a similar range of interpretations. The semantic tools required for the interpretation of such underspecified NPs prove useful to derive the semantic variability of phonologically silent pronouns. Preliminary cross-linguistic examinations suggest that the proposal makes correct predictions on the semantic variability of null arguments in other languages. The paper ends with a conjecture that what has become known as Discourse Pro-drop is a null NP without a determiner.*
.
Semantic functions of Japanese null pronouns
Theories of anaphora have seen significant development since the beginning of Generative Grammar. In recent years, in particular, the major driving force of the investigation has been crosslinguistic examinations of anaphora of various kinds, which show a vast degree of variation from language to language. Whereas syntactic and morphological theories have benefited greatly from such comparative studies of anaphora, it is probably fair to say that semantic theories of anaphora have not been as adventurous, most likely due to the implicit belief shared by many researchers that there is not as much room for crosslinguistic variation in semantics as in the other components of grammar. This view has been challenged, however.1 Particularly interesting for a theory of anaphora are recent studies of NP interpretation across languages (e.g., Chierchia 1997, 1998; Krifka 1995; Dayal 1999; Cheng & Sybesma 1999 among many others), since, except for such types as VP anaphora, anaphora is a kind of NP after all. The present paper is an attempt to shed light on the semantics of anaphora from a crosslinguistic point of view. The main topic of this paper is ‘phonologically silent’ pronouns, with a fair amount of comparison with their overt counterparts. Many of the data in this study are drawn from Japanese.
Satoshi Tomioka
There are at least two distinct types of anaphoric expressions that are recognized in the literature: the referential and the bound variable uses of pronouns. It has been attested (cf. Saito & Hoji 1983) that Japanese null pronouns, which I assume to be small pros, are capable of functioning as either type. (1) Referential use of pro Ken-wa Erika-o saso-tta. Dan-mo pro saso-tta. Ken-top Erika-acc invite-perf Dan-also invite-perf ‘Ken invited Erika. Dan invited (her = Erika), too.’ (2) Bound variable use of pro Dono gakusei-mo [CP Dan-ga pro buzyokushi-ta] to] which student-even Dan-nom insult-perf comp it-ta. say-perf ‘Every student1 said that Dan insulted him1 .’ There is arguably another type of anaphora whose nature has been uncovered in connection with the semantics of indefinite NPs. This type of pronoun shares with bound variable pronouns the property of their antecedents being not referential. Unlike bound variables, however, they need not fall under the scope of their antecedent. In this sense, their syntactic distribution is closer to that of referential pronouns. How to analyze this type of pronoun depends on one’s theoretical inclination (e.g., unselectively bound variables as in Heim 1982 & Kamp 1981, Etype pronouns as in Evans 1977, 1980; Cooper 1979; & Heim 1990). Descriptively speaking, Japanese null pronouns can also function as this type, as shown below. (3) Unselectively bound variable/ E-type pronoun use of pro a. Dareka kita-ra kono-kagi-o pro watasite kudasai. someone came-if this-key-acc give please ‘If someone comes, please give this key to (the person who comes).’ b. Haha-ga ataraii tokei-o katte-kureta-ga boku-wa sugu-ni mother-nom new watch-acc buy-gave-but I-top soon pro nakusite-simatta. lose-perf ‘My mother bought me a new watch, but I lost (the watch she bought me) soon after.’ In addition to these well-documented uses, it has been observed that there are still other instances of pro in Japanese which do not fall into any of the three categories mentioned above. Consider (4).
The semantics of Japanese null pronouns
(4) With an pronominal-containing antecedent Ken-wa zibun-no uti-o utta. Erika-mo pro utta. Ken-top self-gen house-acc sold Erika-also sold ‘Ken1 sold self ’s1 house. Erika2 sold (self ’s2 house), too.’ The pro in (4) produces what is called a ‘sloppy’ interpretation. Under this reading, the pro cannot be referential simply because the house Ken sold is not identical to the one Erika did. Nor can it be construed as a bound variable. First, it is outside the scope of its antecedent. Second, even if we found a way to satisfy the structural condition on variable binding for this case, the resulting interpretation is not correct. Since the antecedent is definite, it is not quite like the third type. A similar, possibly identical, type is found in English, with the popular name of ‘pronouns of laziness’(cf. Karttunen 1969). (5) A man who1 gives his1 paycheck to his wife is wiser than a man who2 gives it (= his2 paycheck) to his cat. So far, each of the four types of Japanese pro has a phonologically overt counterpart in English. However, the parallelism between Japanese and English pronouns is not complete. Consider (6). (6) Ken-wa kuruma-o kat-ta Erika-mo pro kat-tta. Ken-top car-acc buy-perf Erika-also buy-perf ‘Ken bought a car. Erika bought (a car), too.’ In this example, the null pronoun is understood to be a mere indefinite. English has no anaphora that exactly corresponds to this kind. What comes closest is the construction traditionally called N -Deletion.2 (7) a. Ken bought a car, and Erika bought one [= a/one car], too. b. Professor Grimm gave an A to many students and flunked only a few [ = a few students]. Unlike the Japanese indefinite pro in (6), however, the presence of a determiner/numeral is required in N -Deletion cases. As a matter of fact, Japanese also has a construction akin to the English N -Deletion, in which a quantificational expression is stranded. (8) a.
Ken-wa tabako-o ichinichi futa-hako suu-ga Erika-wa Ken-top cigarette-acc a day two packs smoke-but Erika-top pro ip-pon-shika suwan-ai. 1-cl-except smoke-neg ‘Ken smokes two packs of cigarettes a day, but Erika smokes only one (cigarette).’
Satoshi Tomioka
b. Ken-wa furuhon-o zenbu totteoku-ga Erika-wa pro Ken-top used book-acc all keep-but Erika-top hotondo suteteshimau. most throw away ‘Ken keeps all (the) used books, but Erika throw away most (used books).’ In short, Japanese null pronouns have all the semantic functions that the English overt pronouns do, as well as some other uses. The semantic diversity of Japanese null pronouns is summarized as follows.3 (9) a. b. c. d. e. f.
Referential, as in (1) Bound variable, as in (2) Unselectively bound variable or/and E-type pronouns, as in (3) Pronouns with pronoun-containing antecedents, as in (4) Indefinite pronouns, as in (6) Property anaphora, as in (8)
What is the best way to analyze the semantic variability of Japanese null pronouns? To the extent that there is some correspondence with English overt pronouns, we can import successful analyses of English pronouns for analyzing Japanese pro. In the next section, I will try out this strategy. I consider such an approach less than satisfactory, however, since it offers nothing but a list of different meanings for pro and offers no account for cross-linguistic differences. I will also show that there is an empirical problem concerning a mismatch between the two languages with respect to the distribution of pronouns of laziness. In Section 3, I will relate the semantic diversity of Japanese null pronouns to the interpretational freedom that Japanese full-fledged NPs enjoy. Like its neighboring languages, Japanese lacks obligatory marking of definiteness and plurality on NPs, and bare NP arguments get a variety of interpretations. If null pronouns are interpreted by the same mechanisms used for bare NPs, the interpretational variability is reduced to the basic two semantic types of anaphora: individual type (e) or property type (<e,t>). It will also be shown that Korean and Chinese, which also allow bare NP arguments, have exactly the same range of variability in the semantics of null anaphora, as expected under the proposal. In the last section, I will entertain the possibility that all instances of null pronouns in so-called ‘Discourse Pro-Drop’ languages are N -Deletion/NP ellipsis without determiner stranding. The hypothesis captures the otherwise puzzling crosslinguistic generalization that Discourse Pro-Drop languages allow bare NP arguments.
The semantics of Japanese null pronouns
. Analyzing Japanese pro . First try Although the parallelism between Japanese and English pronouns is not complete, it would be a mistake to ignore what is known about English pronouns, as they have been the main beneficiary of the most extensive semantic analyses. So, let us first see what our common wisdom about English pronouns can say about Japanese pro. A referential pro is represented as a free variable with a numerical index. In other words, it is exactly like a bound variable in isolation. (10) a.
Referential pro For any g, [ pro 2 ]g = g(2) b. Bound variable pro For any g, [ pro 3 ]g = g(3)
A bound variable pro is to be bound by its antecedent within the sentence it appears in, whereas a referential pro remains free, and its referent is assigned by an assignment function in the utterance context. For example: (11) For all g such that g: = [2 → Kevin], [ pro 2 ]g = Kevin A wide variety of proposals for pronouns with indefinite antecedents are currently available, and one of the most popular analyses, namely Cooper’s (1979), makes it possible to combine the third and fourth types (i.e., pronouns of laziness). Cooper’s approach is a branch of the E-type pronoun analysis (cf. Evans 1977, 1980; Heim 1990; Heim & Kratzer 1998), in which pronouns anaphoric to indefinites outside the scope of their antecedents are treated as if they were definite descriptions. In Cooper’s theory, the descriptive contents of a definite NP are pragmatically retrieved. λP ∈ D<e,t>. ∃x [∀y [Π(y) ↔ x = y] & P(x)] where Π is a contextually salient property. (The Russellian version) b. ιx. Π(x) where Π is a contextually salient property. (The Fregean version)
(12) a.
Consider (4) again. (4) Ken-wa zibun-no uti-o utta. Erika-mo pro utta. Ken-top self-gen house-acc sold Erika-also sold ‘Ken1 sold self ’s1 house. Erika2 sold (self ’s2 house), too.’
Satoshi Tomioka
Suppose that the contextually salient property in this case is being x’s house (λy. house(y) & own(y)(x)), in which the free variable x corresponds to the reflexive zibun. Then, the null pronoun in the second sentence has the meaning, ιy.[house(y) & own(y)(x)]. Once the variable is bound by the subject, Erika, the sentence means, sold (ιy [house(y) & own(y)(Erika)])(Erika), or in ordinary English, ‘Erika sold Erika’s house.’4 For the use of pro as an indefinite, something new needs to be done since there are no direct correlates to this type in English. We may propose (13), a slight mutation of the meaning of pronouns of laziness. (13) λP ∈ D<e,t>. ∃x [ Π(x) & P(x)], where Π is a contextually salient property This is, in effect, the indefinite version of pronouns of laziness. It is rather obvious that (13) gives the correct interpretations for sentences like (6), so I will not go through the example again. The analysis of N -Deletion is straightforward: it is a property anaphora of type <e,t>. Of course, I am omitting an important issue here. It is debatable whether N -Deletion is really an instance of ellipsis/deletion and thus requires the full-fledged LF structure or is better analyzed as some kind of pro-form. Since I do not know of any convincing argument for one analysis or the other, let us include this type in our discussion. To summarize, the catalog of the semantic functions of pro in Japanese looks like (14). (14) a.
A variable of type e. When it is bound, it functions as a bound variable. When it is free, it receives its reference via assignment functions. b. λP ∈ D<e,t> . ∃x [∀y [Π(y) ↔ x = y] & P(x)] or ιx. Π(x). Both E-type pronouns and pronouns of laziness are of this kind. c. λP ∈ D<e,t> . ∃x [ Π(x) & P(x)]. An indefinite pronoun of laziness. d. A property anaphora, Π, of type <e,t>.
. Why our first try is not satisfactory As far as descriptive adequacy is concerned, our first approximation would do the job. Nevertheless, there is still much to be desired in this ‘catalogue’ analysis. First, it provides no answer to some important questions surrounding the typology of null pronouns. When we compare Japanese with English, we find that the correspondence is only partial. English pronouns have the first four uses (i.e., referential, bound variable, E-type/unselectively bound variable, pronouns of laziness), and they are phonologically overt. There is no pronominal expression in English, overt or covert, that is equivalent to the indefinite use of pro in Japanese. English is also equipped with property anaphora under the name of N -deletion, and the situation is exactly like Japanese in the sense that it is not pronounced. So far as I can see, the
The semantics of Japanese null pronouns
catalog analysis has nothing to say about the similarities/differences between the two languages. Typologically speaking, Japanese is classified, along with Chinese and Korean, as a ‘Discourse Pro-drop language’, in which phonologically null pronouns are licensed when a certain condition on discourse familiarity is satisfied. It contrasts with what is called an ‘Agreement Pro-drop language’, such as Italian or Spanish, in which rich person/number agreement on verbal inflections is believed to be responsible for licensing null pronouns. The catalogue analysis remains totally indifferent of this typological distinction of Pro-drop. Although it is not clear that the semantics of pro should have something to say about it, it would be nice if it did. In addition to these conceptual challenges, the catalogue analysis misses an important generalization within itself. If we look at the list shown in (14), (14b–d) are not so different from each other. The core of their meanings is a contextually salient property (represented as Π). In this sense, it seems rather redundant to list them as separate entries in the semantic functions of pro. Third, the catalogue analysis faces an empirical difficulty concerning the distribution of pronouns of laziness. In English, this type of pronoun is typically found in the ‘paycheck’ environment (as in (7)), but otherwise, its appearance is rather restricted. For instance, pronouns of laziness cannot easily appear in the coordinated structure or in a sentence uttered by a different speaker.5 (15) Conjunction a. Gary1 likes his1 mother. Tim likes her (*Tim’s mother), too. b. Gary1 lost his1 ID in the gym. Tim lost it (???Tim’s ID) in a classroom. c. Gary thinks his teachers are geniuses, but Tim thinks they (???Tim’s teachers) are nuts. (16) Across speakers A: I heard that Gary totaled his car. B: #Did you know his brother also totaled it only a week ago? In none of the examples above does the ‘pronouns of laziness’ reading seem available. The same restriction holds in other languages, such as German and Dutch, which have overt pronouns of laziness. The Japanese null counterparts, on the other hand, make such readings readily available. The following are some examples to illustrate this point. (17) Conjunction a. Ken-wa zibun-no haha-o aisite-iru. Dan-mo pro Ken-top self-gen mother-acc love-be Dan-also aisite-iru. love-be ‘Ken1 loves his1 mother, and Dan2 loves (his2 mother), too.’
Satoshi Tomioka
b. Ken-wa zibun-no saihu-o jimu-de nakus-ita. Dan-wa pro Ken-top self-gen wallet-acc gym-at lose-perf Dan-top kyoositu-de nakus-ita. classroom-in lost-perf ‘Ken1 lost his1 wallet at the gym, and Dan2 lost (his2 wallet) in a bar.’ (18) Across speakers A: Dan-wa zibun-no kuruma-o sugoku taisetu-ni Dan-top self-gen car-acc very much importantly siteru-n-datte. do-hear ‘I hear that Dan1 treasures his1 car.’ B: Aa soo. Oniisan-ga pro somatu-ni siteru-no-to-wa erai Really brother-nom rough do-nominal-with-top very tigai-da. different ‘Really? How different from his brother2 , who abuses (his2 car)!’ Why should there be such a dramatic contrast between the two languages? Since the analysis of English pronouns of laziness has been directly imported into the analysis of Japanese, the catalogue analysis fails to solve this puzzle.
. Pro and the semantics of Japanese NPs . Pro as bare NP anaphora At this point, it is useful to remind ourselves that pro is a kind of nominal expression after all. In the Japanese nominal system, determiners are not of absolute necessity, and a bare NP without a determiner can be in an argument position. The interpretation of such an NP is largely context-dependent (cf. Kuroda 1992: Ch. 1; Chierchia 1997, 1998; Krifka 1995). For instance, the bare NP object in (19) can be indefinite singular, definite singular, indefinite plural, or definite plural. (19) Ken-wa ronbun-o yon-da. Ken-top paper-acc read-past ‘Ken read a paper / papers / the paper / the papers.’ One might try to posit a different LF structure with phonologically null functional heads for each interpretation, but there is no empirical support for such a move. Instead, we can derive those meanings from one basic meaning with independently needed semantic tools, such as type-shifting and Existential Closure.6
The semantics of Japanese null pronouns
(20) Required semantic tools a. Existential Closure (cf. Heim 1982; Diesing 1992): ∃-closure For any P ∈ D<e,t> ∃-closure (P) = ∃x.P(x) b. Type-shifting of a predicate to an individual (cf. Partee 1987): Iota For any x ∈ De , P ∈ D<e,t> Iota (P) = ιx.P(x) ( = the unique x such that P(x))7 One consequence of employing these semantic tools is that the denotation of a bare NP remains as a property-denoting expression up until one of those operations applies. In other words, there is no need to impose ambiguity on the NP itself. The following are some illustrations of how these devices can derive the indefinite singular and the definite singular readings of (19).8 (21) Indefinite singular via object NP-raising and ∃-closure a. Input LF: [IP Ken-wa2 [VP3 ∃ [VP2 ronbun-o1 [VP1 t2 t1 yonda]] b. For any assignment g, [ yonda ]g = λx.λy. read(x)(y) [ VP1 ]g = [ yonda ]g ([ t1 ]g ) ([ t2 ]g ) = read(g(1))(g(2)) [ VP2 ]g = λx. [ [ ronbun-o ]g (x) & λy.[ VP1 ]gy/1 (x)] = λx. [paper(x) & read (x)(g(2))] [ VP3 ]g = ∃([ VP2 ]g ) = ∃x. [paper(x) & read (x)(g(2))] [ IP ]g = λz. [ VP3 ]gz/2 (Ken) = ∃x. [paper(x) & read (x)(Ken)] (22) Definite singular via iota a. Input LF: [IP Ken-wa1 [VP t1 ronbun-o yonda]] b. For any assignment g,[ yonda ]g = λx.λy. read(x)(y) [ ronbun-o ]g = λz.paper(z) iota ([ ronbun-o ]g ) = ιx. paper(x) [ VP ]g = [ yonda ]g (iota[ronbun-o ]g ) ([ t1 ]g ) = read(ιx.paper(x))(g(1))] [ IP ] = λz.[ VP ]gz/1 (Ken) = read(ιx.paper(x))(Ken) The point I would like to make is very simple. If these semantic devices are needed to interpret bare NPs, then there is no good reason for not using them to interpret pro. This reduces the number of pro’s semantic functions to two. While we still need individual variables, the last three types that are listed in (14) are no longer distinct. They are all property anaphora of type <e,t>. A definite pronoun of laziness is the result of the iota operation applied to pro, and the meaning of an indefinite pro is derived from ∃-closure. The kind of analysis that I am proposing has been already anticipated by Hoji (1998), who argues that Japanese pro is a mere NP under the DP structure, and that the descriptive content of the null NP is pragmatically
Satoshi Tomioka
retrieved. In the following subsection, I will show that Hoji’s thesis is successfully incorporated into compositional semantics.
. How to interpret property pro compositionally Even under the new proposal for the semantics of pro, some of its instances stay as they used to be. Referential and bound variable pros are still considered to be variables of individual type, and of course, no semantic manipulation is necessary to interpret pro in the quantifier-stranding cases. It remains as property-denoting and serves as the restriction for the stranded quantifier. Of the remaining two kinds, analyzing an indefinite pro is relatively straightforward. First of all, we need to make a variable assignment function to be type-neutral. (23) A partial function from indices to denotation (of any type) is a (variable ) assignment iff it fulfills the following condition. For any number n and type τ such that ∈ dom (g), g (n,τ) ∈ Dτ . (From Heim & Kratzer 1998: 292)9 The following illustrates how to interpret an indefinite pro. (I only show the computation up to the point where the bare NP pro gets the existential interpretation. The rest is practically the same as in the last section.) (24) Ken-wa kuruma-o kat-ta Erika-mo pro3 kat-tta. Ken-top car-acc buy-perf Erika-also buy-perf ‘Ken bought a car. Erika bought (a car), too.’ Via pro-raising and ∃-closure: ∃ [VP pro1 [VP t2 t1 bought]] (t2 is the subject trace) [ t2 t1 katta ]g = bought(g(1))(g(2)) Assume g := [3 → λy. car (y)] [ pro1 ]g = λy. car(y) [ [ pro1 [ t2 katta t1 ]]]g = λx. [[ pro1 ]g (x) & λz. [ t2 katta t1 ]gz/1 (x))] = λx. [ λy.car(y)(x) & λz.bought(z)(g(2))(x)] = λx. [car(x) & bought(x)(g(2))] ∃([[ pro1 [ t2 katta t1 ]]]g ) = ∃x [car(x) & bought(x)(g(2))] The analysis of (definite) pronouns of laziness needs some work, however. The first approximation is the following. (25) Ken-wa zibun-no uti-o utta. Erika-mo pro utta. Ken-top self-gen house-acc sold Erika-also sold ‘Ken1 sold self ’s1 house. Erika2 sold (self ’s2 house), too.’
The semantics of Japanese null pronouns
(26) The input LF: [IP Erika2 [VP t2 pro3 sold]] Assume g: = [3 → λy. house(y)(g(2))] [ pro3 ]g = λy. house(y)(g(2)) iota ([ pro3 ]g ) = ιy. [house(y)(g(2))] [ VP ]g = sold(ιy. [house(y)(g(2))])(g(2)) [ IP ]g = λx. [ VP ]g x/2 (Erika) = λx. sold(ιy. [house(y)(x)])(x)(Erika) = sold(ιy. [house(y)(Erika)])(Erika) The problem is the content that the assignment functions assign for the denotation of pro (indicated by ). It itself contains a free variable (i.e., g(2)), which appears from nowhere and is eventually bound by the subject in the course of the derivation. This is a very unorthodox operation, to say the least. To avoid this problem, I propose that this type of pro has two indices, one of which corresponds to the otherwise offending individual variable.10 (27) For any g, and natural number i,j, [ pro i (j) ]g = g(i)(g(j)) This strategy is indeed the mixture of Chierchia’s (1993) treatment of functional traces in Wh-questions and Heim and Kratzer’s re-interpretation of Cooper’s Etype pronouns (Heim & Kratzer 1998: Ch. 11).11 The new computation for (25) is shown below. (28) The input LF: [IP Erika2 [VP t2 pro3 (2) sold]] Assume g := [3 → λy.λx. house(y)(x)] [ pro3(2) ]g = λy. house(y)(g(2)) iota ([ pro 3(2) ]g ) = ιy. [house(y)(g(2))] [ VP ]g = sold(ιy. [house(y)(g(2))])(g(2)) [ IP ]g = λx. [ VP ]g x/2 (Erika) = λx. sold(ιy. [house(y)(x)])(x)(Erika) = sold(ιy. [house(y)(Erika)])(Erika) In (28), ‘g(2)’ no longer comes from nowhere (i.e., it has its source in the index in the parenthesis), and there is no free variable in what the assignment functions assign to 3. To sum up this section, Hoji’s (1998) hypothesis and compositional interpretation can be successfully combined.12 One obvious advantage of the current proposal is its generality. The devices used to derive the diverse interpretations of pro are independently needed for ordinary NPs, and some of the interpretational rules specifically tailored for pro are no longer necessary. In addition, the property anaphora hypothesis makes correct crosslinguistic predictions that the catalogue analysis couldn’t. This is the topic of the next section.
Satoshi Tomioka
. Some crosslinguistic predictions . On the distributional mismatch of pronouns of laziness Recall that Japanese and English differ in the distribution of pronouns of laziness, and that the catalog analysis for pro has practically nothing to say about it. Although it is still mysterious why the English pronouns of laziness must obey such a strong restriction, we now have a better understanding of the source of the difference between the two languages. Suppose that the N -Deletion and definite pronouns of laziness are distinct phenomena in English. Assume further that the N -Deletion involves phonologically null property anaphora. I have argued that Japanese pronouns of laziness are simply property anaphora and not marked for definiteness by themselves. Therefore, the previous assumption that pronouns of laziness in English and Japanese are of the same kind was not correct. We should not have compared them in the first place. What we should have done is compare Japanese pro to its closest English counterpart, namely N -Deletion. Indeed, the distribution of N -Deletion is much less restricted than that of pronouns of laziness. (29) Conjunction (Also see (7)) Pablo1 sold many paintings of his1 , but Juan2 sold only a few (= a few paintings of his2 ). (30) Across speakers A: Prof. Grimm1 flunked many of his1 students, I hear. B: It’s not unusual here. Last year Prof. Werner2 flunked even more (= more of his2 students). I can only speculate that the source of the tight restriction on pronouns of laziness in English has something to do with definiteness. Even in Japanese, this restriction seems to hold. Kurafuji (1999) notes that the overt impersonal pronoun sore, which can in principle be construed as an E-type pronoun, cannot replace pro in those environments.13 (31) Ken-wa zibun-no saihu-o jimu-de nakus-ita. Dan-wa sore-o Ken-top self-gen wallet-acc gym-at lose-perf Dan-top it-acc kyoositu-de nakus-ita. classroom-in lost-perf ‘Ken1 lost his1 wallet at the gym, and Dan2 lost (his*2 wallet) in a classroom.’
The semantics of Japanese null pronouns
In this example, the overt pronoun cannot convey the sloppy reading. Therefore, the second sentence is pragmatically very odd since it must mean that the same wallet owned by Ken got lost twice by two different people. Whatever the exact nature of this restriction is, it is not surprising within the current approach to find that Japanese pro is not subject to it.
. Chinese and Korean Since the semantic variability of null pronouns is closely tied to the variability of NP interpretation in general, the current proposal predicts that if a language allows bare NP arguments and has null pronouns, it displays the same property. This prediction is borne out, as far as Chinese and Korean facts are concerned: Both languages have the same range of interpretational possibilities. Here are some examples. I do not include examples with referential, bound variable and E-type pronoun uses. (32) Korean Definite pronouns of laziness a. Chelsu1 -ka caki1 -ui cha-lul pala-tta. Yengmi-to pro Chelsu1 -nom self1 -gen car-acc sold-ind Yengmi-also palat-ta. sold-ind ‘Chelsu1 sold self ’s1 car. Yengmi2 also sold [self ’s2 car].’ Indefinite pronouns of laziness b. Chelsu-ka cha-lul sat-ta. Yengmi-to pro sat-ta. Chelsu-nom car-acc bought-ind Yengmi-also bought-ind ‘Chelsu bought a car. Yengmi also bought [a car].’ Property anaphora (Quantifier Stranding) c. Chelsu-ka cha-lul tu-tae satta. Yengmi-nun pro sae-tae Chelsu-nom car-acc 2-cl bought Yengmi-top 3-cl satta. bought ‘Chelsu bought 2 cars. Yengmi bought three (cars).’ (33) Chinese Definite pronouns of laziness a. Zhangsan1 kanjian ziji1 de xiong le Lisi ye kanjian pro le. Zhangsan1 see self1 gen bear Asp Lisi also see Asp ‘Zhangsan1 saw self ’s1 bear. Lisi2 saw [self ’s2 bear].’
Satoshi Tomioka
Indefinite pronouns of laziness b. Zhangsan kanjian xiong le. Lisi ye kanjian pro le. Zhangsan see bear Asp Lisi also see Asp ‘Zhangsan saw a bear. Lisi saw [a bear].’ Property anaphora (Quantifier stranding) c. Zhangsan kanjian 2-zhi xiong le. Lisi kanjian 3-zhi pro le. Zhangsan see 2-cl bear Asp Lisi see 3-cl Asp ‘Zhangsan saw two bears. Lisi saw three [bears].’
. An intermediate language: Greek While the languages which employ both semantic tools listed in (20) allow the widest range of interpretations for null pronouns, we can reasonably expect that some languages use only one of them. In such languages, only a certain types of nominal expressions are expected to go phonologically silent. Greek seems to fit the bill. Giannakidou & Merchant (1997) present data from Greek which show that ‘weak’ DPs, but not ‘strong DPs’ including definites, can go phonologically null. (34) a.
Efere o Andreas merika / kapja / liga / parapano apo tria vivlia? brought Andreas several / some / a few / more than 3 books? “Did Andreas bring several / some / a few / more than 3 books?” b. Ne, (* ta) efere pro. yes, (them) brought ‘Yes, he brought {several / some / a few / more than 3 books}.’
(35) a.
Efere o Andreas ta vivlia / to vivlio? brought Andreas the books / the book? “Did Andreas bring the books / the book?” b. Ne, *(ta / to) efere pro. yes, (them / it) brought ‘Yes, he brought {them / it}.’
In Giannakidou and Merchant’s analysis, weak quantifiers are not D0 but are simply modifiers like adjectives. Although Giannakidou and Merchant’s analysis is different from the path I have been taking in the sense that they explicitly argue that these null expressions are results of ellipsis, it is relatively easy to import their insights to the current proposal. Let us assume, along with Giannakidou & Merchant, that weak quantifiers are modifiers. Then, the existential readings for them must come from the ∃-closure operation. Then, the null objects in the second sentence of (34) are exactly like an indefinite pro in Japanese. Semantically they stand for properties which include modifier meanings. On the other hand, Greek seems to
The semantics of Japanese null pronouns
do the iota operation only by means of a definite determiner. Therefore, definites cannot go completely null and can only surface in the form of definite clitics. To sum up this section, the new proposal for the semantics of pro in Japanese enables us to make a correct prediction about the distribution of pronouns of laziness in Japanese and English. In addition, it is predicted that the semantic variability of null pronouns is closely related to the way NPs in general are interpreted. This prediction has empirical support from other languages as well.
. Concluding remarks and new questions I started this paper by looking at the surprisingly wide range of semantic interpretations of Japanese null pronouns. Instead of giving an ad-hoc analysis by directly importing the semantic analysis of English pronouns, I analyzed them as the phonologically null version of bare NPs, which requires a limited set of semantic operations to be interpreted appropriately. Some preliminary cross-linguistic examinations indicate that the proposed analysis is on the right track. Given the view that there are cross-linguistic variations in the availability of the semantics tools which play an important role in the interpretation of null pronouns, an entirely new question arises: Does semantics have something to do with what has been known as the Pro-drop Parameter? Under the standard classification of languages in terms of the availability of phonologically null pronouns, languages are categorized as either Pro-Drop or Non-Pro-Drop languages. The former is further divided into two subtypes on the basis of the way in which null pronouns are licensed. In Agreement Pro-Drop languages, such as Spanish or Italian, the strong inflectional morphology on verbs is believed to license pro, whereas the Japanese/Korean/Chinese type uses a condition based on discourse familiarity, hence Discourse Pro-Drop. (36) Pro-Drop languages Agreement Pro-Drop {Italian, Spanish, ...}
Non-Pro-Drop languages {English, German, French, ...}
Discourse Pro-Drop {Chinese, Japanese, Korean, ...}
This classification is a convenient one, but for this model to be a syntactic typology of anaphora, it must answer one challenging question. How do we encode discourse sensitivities in the theory of syntax? I believe that there is indeed a morpho-syntactic property that is shared by all the Discourse Pro-drop languages.
Satoshi Tomioka
(37) Discourse Pro-drop Generalization All languages which allow discourse pro-drop allow (robust) bare NP arguments. The languages which fit the generalization (37) include Japanese, Korean, Mandarin Chinese, Thai, Hindi, Turkish, Brazilian Portuguese, and many others. I haven’t encountered any exceptions so far, so I tentatively conclude that the generalization is valid. What does this generalization tell us, then? Phonologically null NP anaphora, also known as N -Deletion/NP ellipsis, seems to be one of the most widespread forms of null elements across languages. In other words, every language is equipped with some strategy for not overtly expressing a redundant/familiar NP portion of a DP. Even if making NPs phonologically null were universally available, the outcome of this operation would be different from language to language. In languages in which DPs are necessarily projected, we always see leftover D-heads, while entire nominal phrases go silent in languages which allow NPs in argument positions. (38) a. D0 some/all many etc.
b.
DP NP
NP
Under this scheme, phonologically null arguments and N -Deletion/NP ellipsis have the identical outcome in languages with NP arguments. My hypothesis is that they are indeed the same: Null pronouns in Discourse Pro-drop languages are simply the result of N -Deletion/NP ellipsis without determiner stranding. This hypothesis nicely captures the generalization stated in (37). In Chierchia (1998), the availability of bare NP arguments is determined by Nominal Mapping Parameter, which dictates what basic semantic types each language assigns to nouns. Combining the two hypothesis, we arrive at the following conjecture. Discourse Prodrop crucially relies on the availability of bare NPs in argument positions, which is determined by the semantic parameter of nouns and nominal projections. Thus, semantics has a great deal of influence on the Pro-drop Parameter. Although I find the conjecture stated above quite appealing, I am the first one to admit that it raises more questions than it answers. Two of the most intriguing questions are: (39) Although all Discourse Pro-drop languages seem to allow bare NP arguments, not all bare NP languages are Discourse Pro-drop. For instance, English permits bare plurals and bare mass nouns but have no null pronouns even when the antecedents are bare plurals or mass nouns. Why?
The semantics of Japanese null pronouns
(40) What is the nature of this ‘null NP anaphora’? Is it a pro-form or an instance of ellipsis /deletion? Obviously, the present paper cannot offer answers to these questions, and more diverse cross-linguistic examinations are necessary for my conjecture to be validated. It is my hope that I have at least initiated a new direction for research on null arguments across languages.
Notes * In addition to Adding and Omitting Workshop at 1999 DGfS meeting in Konstanz, different versions of this paper were presented at Conference on Semantic Variations at LSA Summer Institute at Cornell University, Conference on Syntax and Pragma-semantics of NP at University of Antwerp, University of Delaware, Universität Tübingen, University of Pennsylvania, University of Durham and University of California-Irvine. In those and many other occasions, I have received comments and suggestions from many, many people, to whom I am grateful. The following individuals deserve special mention for their contribution to this paper: Gennaro Chierchia, Veneeta Dayal, Naoki Fukui, Hajime Hoji, Jim Huang, Kyle Johnson, Chris Kennedy, Yoshihisa Kitagawa, Audrey Li, Jason Merchant, Mike Parker, Maribel Romero, Arnim von Stechow, Wolfgang Sternefeld, and Karina Wilkinson. Special thanks to the editors, Kerstin Schwabe and Susanne Winkler, for suggestions and encouragement. All remaining errors are mine, of course. . The first programmatic study of crosslinguistic semantics is probably Bach et al. (1995). . In recent years, it has become increasingly common to identify maximal nominal projections as D(terminer) P(hrase)s (cf. Abney 1987; Fukui & Speas 1986 among others). Hence, the term ‘N -Deletion’ has become a misnomer since what is missing in this construction can be an NP, rather than N . The change of the terminology has not yet been widespread, however, and I will thus continue to use the original term. . In addition to these 6 uses, Japanese null pronouns can be generic and kind-referring, just as Japanese bare NPs can be. . One problematic aspect of this analysis is that a property which is filled into the meaning of pro has a variable which is free at the time of the substitution but gets bound later in the derivation. See Section 3 for discussion. As far as I can see, this problem is inherent in Cooper’s (1979) analysis of E-type pronouns, in which quantifying into the pragmatically salient property is allowed. . The constraint on pronouns of laziness is still poorly understood. For instance, if the antecedent contains a variable bound by a quantifier, the situation improves. (i) Everyone but John gave his paycheck to his wife. John, on the other hand, gave it to his cat. Whatever the true nature of the constraint is, it is important to note that it doesn’t seem to apply to Japanese null pronouns.
Satoshi Tomioka . The pluralization process in Japanese is not straightforward. Japanese common nouns cannot be directly combined with numerals, and numerals must be attached to classifiers. See Chierchia (1997, 1998), Krifka (1995) for discussion. Since plurality does not play a crucial role in this paper, I will not discuss it. . This operation is category-sensitive. It only applies to nominal categories but not to verbal categories. . I assume the following Predicate Abstraction Rule: (i)
Predicate Abstraction For any α of the following configuration, where i is a natural number and γ is type t, i For any g ∈ G, [α ]g = λx∈D. [γ ]g x/i ([β ]g ) or [α ]g = [β ]g (λx∈D. [γ ]gx/i ) or [α ]g = λ x∈ D.[[β ]g (x ) & λ y∈ D[γ ]gy/i (x)]
The underlined part is a bit unusual. It reflects what is often called ‘Predicate Conjunction’ or ‘Theta Identification’ in Higginbotham (1985) and is needed to interpret NP raising and ∃-closure compositionally. . For simplicity, I will annotate numerical indices only, omitting the type specification. . Another way to avoid this problem is to treat a pronoun as a function from assignments to individuals. Then, the denotation of pro3 becomes λg.g(3), instead of simply being g(3). This type of solution is used for semantic reconstruction in Rullmann (1995), Cresti (1995) and Sternefeld (1997). See Romero (1997c) for comparison between syntactic reconstruction and semantic reconstruction. . In Heim & Kratzer (1998), an E-type pronoun undergoes an LF re-writing rule of some kind, and as a result, the individual variable is explicitly represented at LF. Chierchia’s (1993) double-indexing appears to be different in this respect, but he mentions that the doubleindexing is just an abbreviation of a more complex LF representation of a functional trace. . In Chierchia (1997, 1998) and Krifka (1995), common nouns in languages like Chinese and Japanese start out as kind-denoting. Chierchia derives properties from kinds via predicativization function (see Chierchia 1998: (14)). Can what I have been considering property anaphora be kind anaphora? (i.e., Instead of λx. car(x), simply CAR.) This is an interesting possibility because the types of pros we have seen can be subsumed under the referential use, provided that kinds are entities. However, it seems unlikely that every instance of property pro can be re-analyzed as kind anaphora. Consider: (i) Ken-wa zibun-ga sotugyoo-sita daigaku-ga kirai-da. Erika-mo pro Ken-top self-nom graduate-perf college-nom hateful-be Erika-also kirai-da hateful-be ‘Ken hates the college he graduated from. Erika hates (the college Erika graduated from), too.’
The semantics of Japanese null pronouns
It is difficult to imagine that ‘college x graduated from’ is a kind. It may still be possible to regard this type of anaphora as ‘concept’ anaphora in the sense of Krifka (1995). ‘Concepts’ are like kinds but need not be taxonomically well-defined. It is not clear to me, however, whether we should allow concepts which contain free variables. . This is not a faithful restatement of Kurafuji’s observation. His generalization is that overt pronouns cannot be construed as pronouns of laziness at all. It seems to me, however, that Kurafuji’s generalization is too unforgiving. For instance, I find it quite acceptable to place an overt pronoun in a ‘paycheck’ sentence. (i) zinbun-no saihu-o jimu-de nakusu hito-no-hoo-ga sore-o kyooshitu-de self-gen wallet-acc gym-at lose person-than-nom it-acc classroom-at nakusu hito-yori ooi. person many ‘The number of people who1 lose their1 wallets at the gym is greater than that of people who2 lose them (= their2 wallets) in the classroom.’
Omission impossible? Topic and focus in focal ellipsis Petra Gretsch The standard view on ellipsis entails that only redundant, i.e. backgrounded material can be omitted from a sentence. This is only partly true if we look at the peculiar case of focal ellipsis as e.g. in the following question-answer pair: A: Munich is situated in Bavaria and Kleve in? B: I think in NorthrhineWestphalia. Characteristic for these structures is the omission of (parts of) the syntactic focus-domain. Therefore, FE has to be sharply differentiated from the regular case of background ellipsis. Focal ellipsis is a complementary phenomenon to background ellipsis: The case of background ellipsis may contain parts of the focus but it never allows for the inclusion of the focus exponent in the omitted part, whereas FE precisely requires this fact. This paper concentrates on the syntactic analysis and the semantic/pragmatic interpretation of focal ellipses.*
.
Introduction
The standard view on ellipsis entails that only redundant, i.e. backgrounded material can be omitted from a sentence. This is only partly true if we look at the peculiar case of focal ellipsis (FE1 ). Consider the following questions:2 (1) (geography at issue) A: München liegt in Bayern und Kleve in? ‘Munich is-situated in Bavaria and Kleve in?’ B: Ich glaube, in Nordrhein-Westfalen. ‘I think, in Northrhine-Westphalia.’ (2) (in front of the main door) A: Die Schlüssel sind? ‘The keys are?’ B: Unten drin in meinem Rucksack. ‘At the bottom of my backpack’
Petra Gretsch
Characteristic for these structures is the omission of (parts of) the syntactic focus-domain. To be more precise it is minimally the prominence marked part, i.e. the phrase bearing the primary stress, which gets omitted. Therefore, FE has to be sharply differentiated from the regular case of ellipsis discussed so far in this book which is the classical background ellipsis. If we are to define the term ellipsis broadly as the omission of linguistic material which triggers specific processes of ellipsis resolution in the hearer, the label ellipsis is justified for the above cases, too. As the addition focal in focal ellipsis (henceforth FE) suggests we are dealing with a complementary phenomenon to background ellipsis with regard to the information-structural status of the elided elements. The case of background ellipsis may contain parts of the focus but it never allows for the inclusion of the focus exponent in the omitted part, whereas FE precisely requires this fact. FE exhibit – like background ellipsis – two subcategories, FE which are to be resolved on a structural basis and FE which are to be resolved on a situative basis. But the term resolution is indicating a very different sort of process for background ellipsis and FE. The aim of this paper is to sharpen the notions of topic and focus and to examine the omission of (parts of) the focus in FE with respect to the overall sentence interpretation. This requires in consequence a detailed analysis of the peculiarities of the syntax-semantics interface. Now the reader might ask, why these structures are labeled focal ellipsis although their interpretation seem to suggest a case of dropping a ‘wh in-situ’ element, e.g.. This holds even more if, as claimed above, FE exhibits exclusively an information-question interpretation. But as shown in Section 3, FE bears only a functional and a focus-structural similarity to wh-questions. In this paper, the following claims regarding FE are made: – –
– –
FE are marked with a [–wh] feature in contrast to their use as informationquestion. The gap Λ in FE is associated with focussed and prominence marked material, thus bearing a [+P] and a [+F] feature. Λ contains minimally the focus exponent bearing [+P], maximally a whole Focus Phrase bearing [+F]. The FocusPhrase of the primary focus-structure has to be licensed by a polarity element in INFL, the FocusPhrase of the secondary focus-structure is licensed by the Completion operator. FE denote a propositional function. The question force of FE-utterances is not directly rooted in the grammatical structure as is e.g. the [+wh] feature in wh-questions, but results from a pragmatic derivation process.
Topic and focus in focal ellipsis
–
FE with presentational (i.e. non-categorical) focus structure occur only in languages which allow for unmarked common topics.
. The phenomenon of focal ellipsis Before we have a closer look at the syntactic and semantic/pragmatic analysis of FE this section addresses some terminological and more general points about the phenomenon. To capture the gap at the end of focal elliptical structures the missing elements are symbolized by ‘Λ’ for Latin lacuna ‘gap’. This allows to distinguish between Y/N-questions as under (3) and focal elliptical questions (4) with identical surface structure: (3) A: Marcus segelt? Marcus is sailing B: Ja. Yes (4) A: Marcus segelt Λ? Marcus is sailing B: Er segelt einen Flying Dutchman. He is sailing a Flying Dutchman As alluded to before, the spectrum of background ellipsis spans from the more grammar-oriented cases of omission (gapping, stripping, VP-ellipsis etc.) to the more pragmatic-oriented cases labeled situative ellipsis. The same sort of split is mirrored by FE to the effect that it is more a matter of the grammatical or the semantic-pragmatic structure to derive categorical and selectional restrictions for the gap. Compare (1) and (5) for a primarily grammatical determined gap Λ in coordinated contrastive FE and (2) and (6) for a semantic-pragmatic determination in situative FE: macht [Marcus]DP/Person und das Dessert (5) Den Salat akk makes(=prepares) Marcusnom and the dessertakk The salad macht [Λ]DP/Person ? makes (6) (looking into an office obviously searching for a person) A: Marcus ist Λ? Marcus is
Petra Gretsch
B: Der ist zur Mensa. He is to-the Mensa. ‘He went to the Mensa.’ This kind of ellipsis is strikingly different from the cases discussed in the volume at hand so far. An utterance with the gap Λ does not signal an abbreviated but complete expression as we find in background ellipsis but it reflects a “true” gap in knowledge. Background ellipsis and focal ellipsis share only the general linguistic capacity to “fill in” what is necessary to complete an otherwise incomplete structure. On the other hand, FE shows significant differences on what it is that is filled in and how the gap is resolved in the hearer. The following table summarizes the main points across four linguistic domains which illustrate the differences in more detail and which are selectively discussed in the sections to follow: Table 1. Differences between background ellipsis and Focal ellipsis characteristics common background ellipsis
focal ellipsis
syntactic status – completed structure for structural ellipsis – incomplete structure for situative ellipsis semantic status – completed structure (minimally at LF) with the help of underspecified semantic variables via reconstruction or supplementation accounts – thus it denotes a complete proposition information the information kernel is kernel always overt
always incomplete
functional correlation
no completion possible concerning the content of the gap
denotes a propositional function the information kernel is missing
no specific illocutionary correlation, restricted to informationi.e. it occurs in various functions questions exclusively
The overall structural incompleteness of FE is the main feature that allows the hearer to detect focal elliptical utterances as such and to interpret them as a request for information in the first place. Thus, it is crucial to differentiate between a category bound expectation the hearer of a FE might have due to his or her grammatical knowledge about verb frames or completed sentences on the one hand, and the actual, lexicalized fillers on the other hand. This distinction constitutes the main factor in separating FE from background ellipsis: In FE the element that fills the gap “contentwise” can never belong to the shared knowledge between speaker
Topic and focus in focal ellipsis
and hearer, whereas it is this shared information that constitutes one of the defining features of background ellipsis. That is the reason why FE is always interpreted as an information-question as will be discussed again below, whereas the structure of background ellipsis shows no illocutionary fixation whatsoever. The gap in FE has been “functionalized” to signal a gap in knowledge – an interpretation which can never occur in background ellipsis. The type of incompleteness encountered in FE has therefore to be differentiated from the more or less recoverable incompleteness that background ellipsis exhibits.
. Syntactic analysis of focal ellipsis This section addresses the following questions: Are FE interrogative sentences with a deleted wh-element with corresponding syntactic consequences? What kind of syntactic structure is involved in FE? What is the nature of the gap Λ and which restrictions hold for it?
. Does FE involve an omitted wh-element? For this hypothesis to be correct, the gap has to be a [+wh]-marked constituent and the syntactic structure had to be an interrogative clause bearing a sentence-type [+wh] feature at clause level (excluding the case of wh-exclamatives here where the [+wh]-feature can not percolate to clause level). For a detailed discussion of the issue regarding the wh-feature, see also Gretsch (2000: 29ff.). Consider the following FE under (7) which does not allow for a parallel ordinary wh-construction (8): (7) A: Das Glas ist zur Hälfte Λ? The glass is half B: Voll. / Leer. full / empty (8) a. *Wasi ist das What is the b. *Das Glas ist The glass is
Glas zur Hälfte ti ? glass half zur Hälfte was? half what
The term ordinary wh-construction has been chosen to set it apart from echoquestions which require a strongly accentuated wh-element and underlie very specific interpretational and contextual restrictions.3 (8b) could be interpreted as a well-formed echo-question but lacks the ordinary/plain information-question reading which (7) exhibits. Moreover, FE allows for the questioning of particles in particle verbs (9) which is again impossible with wh-elements (10):
Petra Gretsch
(9) A: Angola führt viel Kaffee Λ? Angola (im-)ports/(ex-)ports a lot of coffee ‘Does Angola import or export a lot of coffee?’ B: Ein. / Aus. im- / ex-(ports) (10) *Angola führt viel Kaffee was? Angola (im-)ports/(ex-)ports a lot of coffee what *Was führt Angola viel Kaffee ? What (im-)ports/(ex-)ports Angola a lot of coffee On the other hand, FE can allow for a variety of interpretations of the gap Λ and thus its constituent status compared to the more or less fixed interpretation and definitely fixed syntactic status of wh-elements, see (11) and (12): (11) A: Und du spielst jetzt Λ? And you play now B: Frisbee. / Bei Schalke. / Morgen. / Dreimal Frisbee / At Schalke (soccer club) / Tomorrow / Three-times wöchentlich. a week (12) Und du spielst jetzt was/ wo/ wann/ wie oft. . . And you play now what/where/when/how often Summing up the respective distributional characteristics of the gap Λ in FE and wh-elements they can not form one homogenous syntactic class. Apart from this, there exists also a severe difference in their topology: Whereas wh-elements as scope-inducing operators occur typically at the left periphery, the gap Λ has to occur at the right periphery. As said before, this is possible for wh-elements, too, but then this induces semantic and pragmatic restrictions which are not encountered in the corresponding FE-structures. An additional argument against an analysis of FE as interrogatives can be found with respect to modal particles. Interrogative sentences typically allow for the modal particle denn in German which signals exclusively interrogative sentence mood. Compare the declarative structure under (13a) and its wh-counterpart under (13b). In FE-structures, this modal particle is excluded (14): (13) a.
(talking about party preparations) *Das Dessert macht denn Johann The dessert makes(=prepares) PRT Johann b. Wer macht denn das Dessert? Who makes PRT the dessert
Topic and focus in focal ellipsis
(14) *Das Dessert macht denn Λ? The dessert makes(=prepares) PRT Despite its illocutional question force FE-structures do not allow for this particle which is otherwise typical for questions. This is a strong indication for the necessity to keep interrogative sentence-type and illocutionary-question force apart in the case of FE. Assuming that FE are wh-interrogative sentences we also expect a parallel behavior under subordination. The matrix predicates to ask, to know are compatible with [+wh]-marked subordinate clauses (in German as well as in English) as demonstrated under (15). But the corresponding structures with subordinated FEclauses are impossible (16). Note also that a parallel structure with an embedded echo-question is also impossible (17): (15) Rosa fragt Karl, wann das Fest beginnt. Rosa is asking Karl, when the party starts (16) *Rosa fragt Karl, das Fest beginnt Λ Rosa is asking Karl, the party starts (17) *Rosa fragt Karl, das Fest beginnt WANN? Rosa is asking Karl, the party starts when This shows on the one hand that FE-structures can not be subsumed under [+wh]marked clauses and on the other hand that the wh-element in echo-questions shares some syntactic characteristics with the gap in FE: Both elements determine the syntactic features and their interpretation at the level of the root clause and both resist thus subordination. Nevertheless, the gap in FE and the echo-wh-constituent show also crucial distributional differences: The echo-wh-constituent exhibits no topological restriction whereas the gap in FE has to appear at the right periphery and the echo-wh-element has to be strongly accented in intonation which is not possible with Λ. Moreover, both elements do not allow for coordination which we would expect if they form a natural syntactic class. This can be confirmed by data from other languages which exhibit overt [+wh] marker as e.g. Korean. Korean employs the interrogativity-indicating morpheme, the Q-marker -ni, for the syntactic marking of questions. Interestingly, this marker is impossible in FE, compare the possible FE cases (18) and (19) with the excluded structure under (20) in contrast to the overt wh-form under (21): (18) Maria*(-nun) cha-ko Peter*(-nun) Λ? Maria-top sleep-and Peter-top ‘Maria is sleeping and Peter Λ?’ / ‘Maria, she is sleeping and Peter (what is he currently doing?)’
Petra Gretsch
(19) Spaghetti*(-nun) Anke-ka matul-ko sauce*(-nun) Λ? Spaghetti-top Anke-nom prepare-and sauce-top ‘Anke will prepare the spaghetti and the sauce (will be prepared by) Λ?’ (20) *Sauce-nun mantu-ni Λ? Sauce-top prepare-q ‘The sauce will be prepared by Λ?’ (21) Sauce-nun nuku-ka mantu-ni? Sauce-top wer-nom prepare-q ‘Who will prepare the sauce?’ The exclusion of -ni marked FE is expected if we assume FE to be marked for [–wh] as argued for above. Chinese, in contrast offers an interesting data puzzle regarding the above sketched account on FE: Whereas presentational FE do not allow for the questionmarking sentence-particle -ne as expected, see (22), contrastive FE allow for it as in (23) and (24): (22) zhe pi’an Wenzhang shi guanyu (*-ne) Λ? this CLASS paper be about (*-ero) ‘This is a paper about Λ?’
(Chinese)
(23) Maliya shui le, Bide *(-ne) Λ? Maria sleep and Peter *(-ero) ‘Maria is sleeping and Peter Λ?’ (24) Maliya zhu mian, tang *(-ne) Λ? Maria cook noodles, soup *(-ero) ‘Maria is cooking the noodles and the soup (will be made by) Λ?’ Why should Chinese allow for the marker -ne in FE whereas Korean does not allow for its related marker -ni? The point here is that the Chinese sentence-particle -ne is not an interrogative marker but a marker for erotetical usage, i.e. for question-function. It results that Chinese -ne and Korean -ni have only a questionfunctionality in common but not a marking of syntactic interrogativity. Only Korean -ni marks sentences as syntactically specified for interrogativity. This is a direct consequence of the behavior of -ne and -ni in FE. Again, the split between presentational and contrastive FE shows structural repercussions in the grammatical distribution of the individual languages. Moreover, the distribution of the Chinese erotetical marker -ne demonstrates that the pragmatic resolution process which re-interprets the completive focus-structure of FE as a question (see Section 4), functions so reliably that at least in presentational cases the marker -ne is superfluous. The case of Chinese shows that this holds even if the language under consideration has erotetical markers in its repertoire. On the
Topic and focus in focal ellipsis
other hand, we see that illocutionary markers are not a priori excluded from FE; the usage of an illocutionary marker as -ne demonstrates the conventional pragmatic question-status that contrastive FE have in Chinese. The use of (erotetical but not interrogative) illocutionary markers in FE in these languages points to an advanced conventionalization of FE as a format for information-questions in a language. Conventionalization effects in FE are further discussed in Gretsch (2000: 94) focussing on the language Taulipang from the Pemon family which allegedly exhibits exclusively FE-structures for questioning with no existing wh-counterpart at all. Summing up, FE-structures and wh-interrogatives have to be differentiated not only in German but in other languages, too. I assume the syntactic [–wh] marking of FE to be universal.
. The syntactic type of FE Let us turn to the second question at the beginning of this section: What kind of syntactic structure is involved in FE if it is not an interrogative one? In other words, are there indications for a [–wh] analysis? One hint could be already obtained by the exclusion of the modal particle denn in FE-structures. In a complementary fashion FE-structures are compatible with declarative sentence type indicating modal particles like ja, doch, and schon. The grammatical restriction for these particles to occur only in sentences with a declarative sentence-type/sentencemood feature has been shown, among others, in Brandt et al. (1992) and OrmeliusSandblom (1997). Note, that the occurrence of one of the latter particles in FE induces a reading as (re-)confirmation-question and includes thus various shadings of presuppositional flavor. Nevertheless, these interpretational specifics do not matter if we are to tackle the question concerning the syntactic type of FE, i.e. its compatibility with syntactic features; compare (25a) with (25b): Das Dessert macht ja/doch/schon Λ? The dessert makes(=prepares) PRT ‘I know, that I should know this but who again (volunteered to) prepare the dessert?’ b. *Wer macht ja/doch/schon das Dessert? Who makes PRT. . . the dessert?
(25) a.
Although the corresponding wh-questions under (25b) distinctly disallow the inclusion of the respective modal particles, the FE-structure is fine. Since the overall meaning of (25a) and (25b) should be in principle similar, this difference in compatibility can not be due to the complex presuppositional nature of these particles. The difference is rooted in the divergent features of the sentence-type for wh-
Petra Gretsch
questions and FE. The exclusive occurrence of ja, doch, and schon in syntactically declarative marked sentences shows the corresponding syntactic feature-marking [–wh] in FE-structures which was the first claim made in the introduction: –
FE are marked with a [–wh] feature in contrast to their use as informationquestion.
This does not imply that FE constitute a completed declarative in interpretational terms. So, FE can neither occur negated nor coordinated or embedded which would be expected if FE were common declarative sentences. Since FE is restricted to main clauses this hints already at the pragmatic nature of the derivation of its question force and the special status of FE as incomplete “declarative” structures.
. The nature of the gap in FE The findings above raise immediately the following the questions: What is the grammatical nature of this peculiar gap and which type of process is responsible for it? The most evident feature of FE-structures is the missing prominence marking, a marking which typically correlates with a particular element of the syntactic focus-phrase, i.e. the focus-exponent as can already be traced from all the examples above.4 It is a well known fact that expletives can not be prominence marked and thus never occur accented. Likewise, it is impossible to ask for them in FE-form: (26) A: *Gestern schneite Λ? Yesterday was-snowing B: Ja, gestern schneite es. Yes, yesterday was-snowing it (27) *Schneite Λ? Was-snowing Here, one could argue that only referential material can be asked for since the parallel wh-questions are excluded, too. The following example avoids this problem and illustrates a case of interaction between background and focus ellipsis with the gap Λ “swallowing” a prominence marked and focussed constituent: (28) (talking about different colors of snowboards) CHRIStian fährt auf einem GELben Snowboard und TObi Λ? CHRIStian rides on a YELlow snowboard and TObi
Topic and focus in focal ellipsis
Turning to the syntactic focus structure of this coordination makes the structuring of the example particularly clear. I follow an account of focus structure as presented in Drubig (1994, 1997) and modified in Drubig (2000) which assumes in a nutshell a free focus-assignment with strong arguments for a relational view on focus. This entails that focus is bound by an operator, the polarity item, abbreviated to pol in the syntactic focus-structures below. Categorically interpreted sentences allow for a split in focus structure between the topicalized phrase and the IP-part in contrast to thetic structures which allow only for a single uniform focus structure. In Drubig (1994) and (1997) it is argued that the focus-binding operator Polarity is related to I0 . Polarity is also manifested in the functional shell of DPs and PPs. This allows for an individual licensing of marked topics as a FocusPhrase5 bearing a [+F] feature in addition to the VP which is marked [+F] qua default. This default-marking of the VP is responsible for the informational value of every sentence and thus for its context-changing potential. There are various syntactic and prosodic devices to mark defocused, i.e. backgrounded material that originate within the VP as such. Due to the interacting background ellipsis, coordinated clauses allow for a controlled parallel structuring of the first and the second conjunct (disregarding the exceptional case of non-homogenous coordination here). This parallelism induces the predictive power with respect to the syntactic and semantic characteristics of the gap Λ. Note that FE can occur combined with background ellipsis only if the gap Λ and the background elided material (crossed-out below) surface both right-peripheral to the overt sentence material, thus constituting one overarching missing constituent: Syntactic focus-structure of (28): [CHRIStian pol ]+F1 [fährtpol [auf einem GELben Snowboard]]+F2 und [TObi pol ]+F1 [fährt pol [auf einem Λ Snowboard.]]+F2 Note that the contrasting elements are the snowboarders {Christian, Tobi} plus the respective colors of their snowboards {yellow, X} whereas their associated minimal FocusPhrases consist of {a yellow snowboard, an X-colored snowboard}. The square brackets delineate [+F] marked FocusPhrases. Compare now the following variant of the snowboard-example where an additional, “fourth” prominence marking occurs (29). Here, Λ can not stand in for a prominence marked element: (29) *CHRIStian CHRIStian auf einem on a
fährt auf einem GELben Snowboard und TObi fährt rides on a YELLow snowboard and TObi rides ROTen Λ? RED
Petra Gretsch
Syntactic focus-structure of (29): [CHRIStian pol ]+F1 [fährtpol auf einem GELben Snowboard]+F2 und [TObi pol ]+F1 [fährt pol auf einem ROTen) Λ]+F2 Here, the semantically contrasting pairs are both already given. The goal of the question is either to be interpreted as asking for a background-elided second snowboard or as asking for another (winter-)sport equipment which is excluded since this would induce a third contrasting pair. The latter reading were only possible under the assumption that the element Snowboard is highlighted, i.e. prominence marked, too, in the first conjunct as exemplified under (30) which is fine: (30) A: Woran erkenne ich Christian und Tobi denn auf der Piste? ‘How do I recognize Christian and Tobi at the ski-run?’ B: Ich weiß nur: CHRIStian fährt auf einem GELben GOOfy-Board und TObi fährt auf einem ROTen Λ? ‘I remember only so much: CHRIStian rides on a YELLow GOOfyboard and TObi rides on a RED Λ?’ In that case, the number of the contrasting sets of alternatives gets shifted from two to three, namely the snowboarders {CHRIStian, TObi}, the colors of their respective snow-equipment {YELLow, RED}, and the kind of snow-equipment themselves {GOOfy-board, Λ}. This shows that Λ always affects and includes a prominence-marked element. The combination of background and focal ellipsis allows also for the occurrence of multiple Λ which exhibits the [+P]-condition on Λ even stronger: (31) A: Der Zug von Gleis DREI fährt um ZWEI Uhr nach The train from platform THREE runs at TWO o’clock to BerLIN. BerLIN B: Und der Zug von Gleis SIEBen Λ? And the train from platform SEVen A: Der fährt um zwei Uhr ZWANzig nach AmsterDAM. That-one runs at two TWENty to AmsterDAM The focus-structuring of (31) corresponds to the following scheme: [Der Zug v. Gl. DREI pol ]+F1 [[fährtpol ]. . . [um ZWEI U. pol ] +F2 [ nach BerLIN pol ] +F3 ] Und [der Zug v. Gl. SIEBen pol ]+F1 [[fährtpol ]...[um Λ Uhr pol ] +F2 [ nach Λpol ] +F3 ]
Topic and focus in focal ellipsis
An interesting mirror image of this claim is that Λ can affect every element that is able to carry a prominence marking. Thus, non-argument phrases or even nonphrases can also form the question-goal of a FE (see (32) and (33)): (32) A: Die Sonne geht im Westen Λ? The sun goes in-the west ‘What does the sun in the west? / Does the sun rise or set?’ B: UNter. Down ‘It sets.’ (33) A: DNS steht für Desoxyribonukleinsäure. DNS stands for deoxyribonucleic acid. B: Wie bitte? DNS steht für Desoxy-Λ? Pardon? DNS stands for deoxy-Λ? As (33) shows, the split in quotational FE might even occur within a word which is, of course, a speciality of the structural flexibility only a quotational usage allows for. These cases illustrate that Λ does not necessarily consist of a phrasal constituent but the elided part bears always (minimally) the [+P] marked element. What happens if the relevant licensing operator for [+F] marking is absent in FE-structures? This is illustrated by infinitival sentence structures: (34) *Fragen Λ? ‘To ask?’ *Heute tun Λ? ‘To do today?’ Interestingly, the corresponding infinitival wh-questions are possible with specific restrictions on their pragmatic potential to deliberate question function: (35) Wen fragen? ‘Whom to ask?’ Was heute tun? ‘What to do today?’ This corroborates additionally the fact that FE-structures can not be analyzed as [+wh]-marked sentences. If we assume (i) that the pol is responsible for the focus to be bound, and (ii) that the pol is closely related to the finiteness element in I0 as a.o. the case of Verum-Fokus indicates, infinitival FE are predicted not to be well-formed. Without the finiteness element a presentational focus-licensing of the VP-material can not take place according to the framework adopted here. In sum, the first part of the second claim from the introductory section holds: –
The gap Λ in FE is associated with focussed and prominence marked material, thus bearing a [+P] and a [+F] feature. Λ contains minimally the focus exponent bearing [+P], maximally a whole Focus Phrase bearing [+F]. The feature [+F] has to be licensed by pol in I0 .
Petra Gretsch
Following the considerations in Drubig (2000), question/answer pairs exhibit an orthogonal secondary focus-background division to the first, contrastive focusstructure considered so far in the examples above. Within the typology of focus as presented below, Drubig (2000) states that A Q/A pair represents a particular type of completive focus construction which consists of an open proposition with one or more variables licensed by a WH-operator, and a syntactically disconnected structure, the answer, which supplies the constituent(s) filling the gap (s) in the open proposition. In order to derive a proposition with a context-changing effect, question and answer (filler and gap) must be connected. (loc. cit., 58) There can be no doubt that all completive focus constructions single out and – in a particular sense – “focus” values that are assigned to variables in open propositions, but I think that this effect is epiphenomenal with respect to information structure. (. . .) Beneath the secondary focus structure, which temporally highlights the value substituting for the variable in the open proposition, the primary focus structure bridges the gap between syntactically discontinuous constituents. (loc. cit., 59f.)
Typology of focus (from Drubig 2000: 58):
– contrast
+ contrast
– completive
presentational focus (spontaneous)
contrastive focus (spontaneous)
+ completive
presentational focus (induced by a question)
contrastive focus (induced by a question)
Although Drubig (2000) does not take into account the case of FE, these considerations can be directly applied to our case at hand. FE is even more prone to constitute a “direct translation” of a completive focus structure avoiding the complexity of the wh-issue. Only FE allow for an isomorphism of completive focus structure and surface expression. It is noteworthy, that “the interpretation of the answer is determined by primary information structure” (loc.cit., 60) which is precisely what we found, especially in the FE-structures with a more complex gap. The only difference is that we can not resort to a WH-operator as licensing element for the variable(s) in FE. Regarding the parallelism of FE/answer pairs and wh-Q/answer pairs in their secondary focus structure presents an argument against the stipulation of a WH-operator as the only licensing element for Q/Ainduced completive focus but points to a more general element – here tentatively labeled Completion operator – which is responsible for the variable binding. For FE structures, this operator is made visible in the prosodic component at PF:
Topic and focus in focal ellipsis
FE-utterances allow only for a rising intonation (high offset) or for a progredient intonation (leveled offset) which both signal incompleteness in a very general sense. For wh-questions then, the [+wh] feature at the root is associated with a Completion operator. The following example and its answer potential demonstrates the interaction of primary and secondary focus-structure very clearly. The resolution of these ambiguities is dependent on the structural context and the situational information as far as common background ellipsis is involved. The dots in B’s utterance in (36) symbolize here different spans of background-elided material: (36) A: Ben Ben B: Und And
kauft ein Geschenk für Wolfgang. is buying a present for Wolfgang. Christine [(. . . ) Λ]? Christine
(37) a.
A: Sie wird sich freuen, daß sie nicht in die Stadt fahren muß. ‘She will be happy that she doesn’t have to drive into town’ b. A: Sie wartet solange in einem Café. ‘She is waiting in the meantime in a Café.’ c. A: Sie kauft Geschenkpapier. ‘She is buying wrapping paper.’ d. A: Sie bastelt ein Geschenk. ‘She is tinkering a present.’ e. A: Sie kauft ein Geschenk für Gaby. ‘She is buying a present for Gaby.’ f. A: Auch./ Sie auch. / Sie kauft auch ein Geschenk für Wolfgang. ‘Too. / She as well. / She is buying also a pr. for Wolfgang.’
The respective background information differs from (37a) to (37f), reflecting the possible interactions of background ellipsis and FE in B’s question Und Christine? and thus their primary focus structure. The following schema under (38) visualizes these differences in elliptical structuring whereby the primary focus information (the presentational or contrastive focus information) is marked by bold print. Note, that the information to be defocused is partly delivered by speaker B, namely the “Christine” part, and partly delivered by the previous utterance from speaker A. This defocused material determines in reverse the primary focussed “rest”. A paraphrase of the respective corresponding questions appears in italics. The term TopicTime (TT), i.e. the time for which an assertion holds, has been introduced in Klein (1994, 1998) and is needed here to depict a particular time span which is the very same time-span for Ben’s and Christine’s activities in the case under (37/38b). The respective syntactic focus structures which are associated with the above interpretations are restricted to the relevant non-topical part of the sentence from
Petra Gretsch
example (38b) onwards. The superscripts at Λ indicate the categorical type aimed at by the question. The movement of the topicalized phrase is not indicated here to avoid confusions. (38) a.
A: Sie wird sich freuen, daß sie nicht in die Stadt fahren muß. (= Topic content Sie as maintained information) Und [CP [Christine] topic [IP pol ΛTenseP ]] What is her attitude towards the proposition expressed by the previous utterance? b. A: Sie wartet solange in einem Café. (= Topic + TT information expressed by solange and the same tense marking) [IP pol ΛVP ]] What is she doing at that time? c. A: Sie kauft Geschenkpapier. (= Topic + TT + content of the finite verb) [IP pol -kauft i [VP ΛDP t i ]]] What is she buying (at that time)? d. A: Sie bastelt ein Geschenk. (= Topic + TT + content of the given (acc-object)) [ IP pol -ΛV i [VP .[DP ein Geschenk] t i ]]] What does she do for a present? e. A: Sie kauft ein Geschenk für Gaby. (= Topic + TT + finite verb + acc-object + specification of the recipient (PP)) [ IP pol -kauft i [VP [PP ΛPP ] [DP ein G.] t i ]]] alternative: [I‘ pol -kauft i [VP [PP für [ΛDP ]] [DP ein G.] t i ]]] For whom does she buy a present? f. A: Sie kauft auch ein Geschenk für Wolfgang. (= Topic + TT + finite verb + acc-object + content of the PP (= the whole proposition with the changed topic element)) [ IP Λpol -kauft i [VP [PP für Wolfgang] [DP ein G.] i ]]] = Λ (true or false for p) Is she buying a present for Wolfgang, too?
The primary focus-structure signals differences in the interpretation of the term Λ in that the example under (38a) requires a non-contrastive, presentational focus interpretation for Λ whereas the examples from (b) to (f) have to be interpreted as involving a contrastive interpretation for the gap involving a specific contrasting element. In contrast, the secondary focus-structure is the same in all cases since all of them exhibit a [+completive] focus-construction.
Topic and focus in focal ellipsis
Now, the nature of the gap Λ in FE can be captured as follows: Λ acts as a placeholder for the variable part in a completive focus construction and has to be bound by a Completion operator. OpCOMPLETION i ( λx [ . . . x . . . ], Λi [+F] ) and Λ= x This renders the second claim from the introduction complete: –
The gap Λ in FE is associated with focussed and prominence marked material, thus bearing a [+P] and a [+F] feature. Λ contains minimally the focus exponent bearing [+P], maximally a whole Focus Phrase bearing [+F]. The FocusPhrase of the primary focus-structure has to be licensed by a polarity element in INFL, the FocusPhrase of the secondary focus-structure is licensed by the Completion operator.
In effect, this framework interprets FE as propositional function which constituted our third claim above: –
FE denote a propositional function.
If we consider only the primary focus-structure, the following two schemes capture the properties of FE: Presentational FE: Contrastive FE:
[XP FINpol [ ... Λ] ]+F [XP pol ]+F1 [FINpol i ... [ ... Λ]i ,] +F2
The first case realizes one focus-domain, the second one realizes two of them. Since we considered already many cases for contrastive FE, I illustrate here only two presentational examples: (39) (talking about the coming holidays) Ah, ihr fliegt also morgen. Es geht nach Λ? Ah, you fly PRT tomorrow. It goes to? ‘Ah, you fly tomorrow. You go to?’ The primary syntactic focus-structure of (39) following the account sketched above is: [Es gehti pol nach Λ]i[+F] (40) Meine Uhr ist kaputt. Es ist jetzt Λ? My watch is broken. It is now? In the same way, the primary focus-structure of (40) can be depicted as: [Es isti pol jetzt Λ]i[+F]
Petra Gretsch
Since clefting constructions are completive focus structures, too (see Drubig 2000: 58ff.) it is predictable that FE and clefts are incompatible structures since they set the variable at different positions.6 (41) *Es war das Messer, das Λ? ‘It was the knife that?’ In clefts the variable occurs left-peripheral whereas in FE it occurs right-peripheral. Therefore, the structure under (41) consists exclusively of two “dangling” variables with no licensing operator at least for the cleft-induced one. What kind of process could be responsible for the gap Λ in FE? One type of process which can be immediately excluded from the discussion is phonological reduction. If no [+wh]-element is involved in FE-structures – neither fronted, nor in situ, nor in its echo-wh variant, then we have to cope with an actual “blank”. Obviously, the answer term itself as given by cooperative interlocutors in reply to a FE is likewise excluded from a phonological reduction process since it is an unknown term at the moment a speaker utters a FE.7 Likewise, any deletion process presupposes some a priori existing linguistic material which can be deleted. This is not given in the case of the gap Λ. It is assumed here that Λ is base-generated at the level of syntax to allow for a transparent logical form. I stipulate that the Completion operator induces this base-generation. The compatibility of the syntactic category is defined via the restrictions the open proposition puts on the variable. In a similar way selectional restrictions are imposed on Λ. Variables in situative ellipsis have also been described as base-generated (see Schwabe 1994) which reflect the unspecific nature of both ellipsis types, whereas deletion or copying accounts rely on exactly predictable, i.e. reconstructable material.
. Bridging the gap from a [–wh] structure to question interpretation With the syntax and focus-structure set out as above, the question interpretation of FE can be derived from the secondary, completive focus-structure. The completion operator splits the structure into a question part and an answer part, whereby the answer is substituted by Λ. This renders the construction type “FE” grammatically complete to allow for an interpretation of these open structures in the first place. Following the Cooperative Principle and other Gricean maximes (Grice 1975) plus Fraser’s Principle of Expression of Desire (Fraser 1975), the cooperative interlocutor is lead to the conclusion that FE-utterances call for an answer – the filler of the variable as expressed by Λ. For a detailed step-by-step derivation of the pragmatics of FE, see Gretsch (2000: 135ff.).
Topic and focus in focal ellipsis
The hearer of an FE-utterance is a.o. guided by the lack the open variable within a [+completive] focus-structure constitutes and the lack of an element bearing the phonologically “topmost” prominence feature since he or she has to figure out why they have been suppressed by the speaker in the first place. The conclusion must be that this failure has been intended. The hearer concludes that he himself or she herself has to close the gap, i.e. present the element or phrase which is associated with the variable. This enables the hearer to interpret FE as a request paraphrasable with “please cooperate and ensure the informativity of the to-becompleted sentence in presenting the information-structural focus, i.e. the content of the variable”.8 Following this approach, the mere existence of FE appeals to a strict conservative interpretation of the principle of informativity that implies for every utterance the existence of an information-structural focus – in contrast to their incomplete grammatical surface appearance. FE illustrate rather a perfect exploitation of precisely this principle. The pragmatically derived question force of FE which stands in contrast to wh-questions is corroborated by the fact that FE occur exclusively in direct and directed question usage. Only wh-questions occur additionally in non-directed functions which call for a characteristic pragmatic re-interpretation e.g. in deliberate or rhetoric function. Moreover, FE do not occur in written language which supports again the lack of a grammaticalized question indication as [+wh] could offer. Summing up, the fourth claim made above in the introduction section holds: –
The question force of FE-utterances is not directly rooted in the grammatical structure as is e.g. the [+wh] feature in wh-questions, but results from a pragmatic derivation process.
. Topic and focus in focal ellipsis . The notion of topic Regarding FE the question arises whether they require generally an overt topical element for the interpretation to be anchored. In the contrastive case the answer is clearly yes, since the disintegrated focus-structure signaled by intonational cues sets a marked topic reliably apart. Typically, it holds that the constituent assigned [. . .]+FOCUS1 has to be interpreted as (potentially marked) topic if the FE is contrastive whereas the constituent with [. . .]+FOCUS2 entails informationstructural material. The presentational case is more intricate. Consider the following example with a typical “non-topical” topicalized element – the expletive pronoun es:
Petra Gretsch
(42) Ah, ihr fahrt also in den Urlaub. Es geht nach Λ? Ah, you go PRT on holiday. It goes to ‘Ah you will go on holiday. You will go to Λ?’ With respect to these non-topical elements in FE the precipitate conclusion that presentational FE do not need any topical element offers itself. But this can not be true. On the one hand, at least a pragmatic sentence anchoring is necessary: es in example (42) refers roughly to the ‘holiday information’ of the preceding sentence whereas in (43) no such referential information can be obtained which renders the FE uninterpretable. On the other hand, topic-drop is excluded from FE banning unlexicalized topics as demonstrated under (44): (43) (asking for the beginning of a fairy tale:) Es war einmal Λ? It was once ‘Once upon a time Λ?’ (44) A: Vitamin D ist wichtig für Säuglinge. Vitamin D is important for babies für Λ? B: *[topic ] sorgt is-responsible for Thus, FE need a topical anchorpoint for interpretation but this anchorpoint need not be necessarily an overt common topic. German does not have an overt morphological topic-marker as e.g. the case of Korean (see examples (18) to (21) from above). Instead, the following factors interact with topic interpretation in German: focus-structure, topological information, and suprasegmental phonological information. None of them is a sufficient marker of topichood. Because of the central status of the focus-structure in determining topichood compared with the other factors, the split between presentational and contrastive focus interpretation has strong repercussions on the topicinterpretation, cf. the focus-structural ambiguous (45) and its different interpretations under (46) and (47). Stressing Die SONne with a so-called “topic-accent” as part of a bridge-accent results in a contrastive reading as shown in (48). The examples are supplemented by crude intonational indices for rise (LH) and fall (HL) accentuation for disambiguation purposes only. The square brackets indicate again the focus phrases within the primary focus-structure: (45) Die Sonne scheint. ‘The sun is-shining.’ (46) thetic reading (no overt partitioning): minimal topical elements: TopicTime (now) and TopicPlace (here)
Topic and focus in focal ellipsis
Das Wetter ist prima. [Die SONnehl scheint]. . . is-shining.’ ‘The weather is fine. The sun (47) categorical reading: minimal topical elements: TopicTime (now) and TopicPlace (here) plus overt topic constituent Die Sonne Wir alle ‘We all damit therefore
lieben die Sonne. [Die Sonne] [SCHEInt] hl und erlaubt is-shining and allows love the sun. The sun Leben. . . life. . . ’
(48) categorical reading with contrastive interpretation: minimal topical elements: TopicTime and TopicPlace plus overt marked, contrasting topic constituent Die Sonne Der Mond leuchtet nur schwach, aber [die SONne]lh ‘The moon is barely gleaming, but the sun hl [SCHEInt.] is-shining.’ Talking about topics in presentational FE as in (46) it seems necessary to point to the differences between the interpretation-pairs ‘presentational vs. contrastive’ and ‘thetic vs. categorical’: To be interpreted as a thetic sentence a presentational focusstructure is a necessary prerequisite but not vice versa. At first sight, it might appear odd to include topics in sentences with a thetic interpretation, an interpretation which is commonly thought of as exactly excluding this possibility. This might be true only for the syntactic surface level. Concerning the notions of TopicTime and TopicPlace, it has been argued for in Dimroth and Gretsch (1999) that the topic component of a sentence comprises minimally of the element TopicTime (TT) as an obligatory anchoring point for the assertion part of every sentence (as already introduced above). And maximally the topic component consists of a tripartite structuring with the coordinates TopicTime and TopicPlace plus a common topic, i.e. the entity the sentence is about if this element does not coincide with TopicTime or TopicPlace in the first place. Consider again the thetic sentence under (46): Even if the assertion made is not interpreted as being an assertion about the sun (which would lead to a categorical interpretation), it is nevertheless an assertion made about a specific time, at which the assertion or the claim made holds. Here, it refers to some TopicTime which comprises the UtteranceTime due to the use of present tense (in a non-habitual, non-future, non-historical manner). If the same statement were to appear in the past, the sentence looses its contextual adequacy: #Das Wetter ist prima. Die Sonne schien. (‘The weather is fine. The sun was-shining.’). This is the reason why the
Petra Gretsch
time an utterance is about has been labeled TopicTime and closely interacts with the assertion force of sentences. Based on these considerations the following terminology is proposed here (for an earlier variant see also Dimroth & Gretsch 1999):
Topic Definition topic is a label for the anchorpoint of an utterance at the level of informationstructure which is an interpretational requirement and thus a necessity for every sentence. topic consists of information needed for the localization of a fixed point on the time/space-axes within a world which is considered as presupposed by speaker and hearer. It can be extended to cover an entity that relates to these time/space-coordinates. Thus, topic is maximally filled by the elements TopicTime, TopicPlace and TopicEntity, i.e. the common topic. Thetic Interpretation topic comprises minimally TopicTime (and TopicPlace) as implicit coordinates for the interpretation of an utterance. The grammatical expression of TopicTime hinges on the finiteness element in relation to tense with a potential collaboration of temporal adverbials or other time-indicating elements in topicalized position. Categorical interpretation An overt or covert syntactic constituent distinguished by grammatical factors (syntactic focus-structure, positioning and intonational cues in German) for being “topicable” and selected by pragmatic and textual factors are part of the topic in addition to TopicTime (and possibly TopicPlace).9 This common topic can either be unmarked or marked. . Topics and FE Turning now again to Korean, more aspects of a grammatical signaling of topichood enter the picture. Korean employs the topic-marker -nun to signal the common topic of a sentence. It is no surprise that Korean contrastive FE exhibit the marker -nun obligatorily, cf. again: (49) Maria*(-nun) cha-ko Peter*(-nun) Λ? Maria-top sleep-and Peter-top ‘Maria is sleeping and Peter Λ?’ / ‘Maria, she is sleeping and Peter (what is he currently doing?)’
Topic and focus in focal ellipsis
(50) Spaghetti*(-nun) Anke-ka matul-ko sauce*(-nun) Λ? Spaghetti-top Anke-nom prepare-and sauce-top ‘Anke will prepare the spaghetti and the sauce (will be prepared by) Λ?’ (51) *Sauce-nun mantu-ni Λ? Sauce-top prepare-q ‘The sauce will be prepared by Λ?’ It is unclear, whether -nun in Korean is actually a topic-marker on the syntactic level, or whether it signals rather contrastiveness on the interpretational level. The latter view is supported by Choi (1997: 548): In distinction from the traditional belief that -nun is a topic marker, it only encodes a contrastive reading of a phrase in the base position and a topic reading is possibly only in a scrambled position. Then, we can hypothesize that the so-called topic marker -nun in Korean is not a topic marker but a ‘contrast’ marker and moreover, that scrambling is the actual encoder of topichood.
With respect to the presentational case of FE, these structures are considered to be odd in Korean: (52) *Sauce mantu/mantu-ni Λ? Sauce prepare ‘The sauce will be prepared by Λ?’ This points to a differentiation among languages concerning the prototypicality a FE-topic must exhibit: Korean does not allow for unmarked, i.e. less prototypical topics, to appear in FE, whereas these are possible in German. This allows for the corroboration of the last, fifth point from the list above: –
FE with presentational (i.e. non-categorical) focus structure occur only in languages which allow for unmarked common topics.
Since FE excludes the occurrence of topic-drop, the following filter on ellipsis in general can be formulated – taking into account background ellipsis and FE: filter on ellipsis The simultaneous omission of common topics and information-structurally focussed material (like Λ) is excluded, even if the topic part is contextually easy recoverable.
. Summary In contrast to the common view on ellipsis, focal constituents can be omitted which results in an erotetical utterance, i.e. an utterance in question-function. Two cases
Petra Gretsch
of FE have to be differentiated according to their focus-structure: the presentational vs. the contrastive focus (-omission). The syntactic nature of the gap Λ in FE is captured as follows: Λ acts as a placeholder for the variable part in a completive focus construction and has to be bound by a Completion operator. OpCOMPLETION i ( λx [ . . . x . . . ], Λi [+F] ) and Λ=x The omitted focus is represented by Λ for lacuna (gap); the subscript i indicates the binding relationship between the focus-binder at the secondary focus-structure, i.e. the completion operator, and the focus phrase. The following claims have been shown to hold for FE: – –
– –
–
FE are marked with a [–wh] feature in contrast to their use as informationquestion. The gap Λ in FE is associated with focussed and prominence marked material, thus bearing a [+P] and a [+F] feature. Λ contains minimally the focus exponent bearing [+P], maximally a whole Focus Phrase bearing [+F]. The FocusPhrase of the primary focus-structure has to be licensed by a polarity element in INFL, the FocusPhrase of the secondary focus-structure is licensed by the Completion operator. FE denote a propositional function. The question force of FE-utterances is not directly rooted in the grammatical structure as is e.g. the [+wh] feature in wh-questions, but results from a pragmatic derivation process. FE with presentational (i.e. non-categorical) focus structure occur only in languages which allow for unmarked common topics.
Notes * I am grateful to Ulrike Heinzel for helpful comments on an earlier version. Moreover, I am indebted to Shin-Sook Kim and Lansun Chen for their translation and discussion of the Korean and Chinese data, respectively. Thanks also to the editors, Kerstin Schwabe and Susanne Winkler, for their suggestions and critical remarks on an earlier version. . We will use the term FE variably for the singular term ‘focal ellipsis’ or the plural one ‘focal ellipses’. . The following notational conventions are used in the examples: Round brackets contain situative information to disambiguate the interpretation of the question goal; if used within an utterance they signal conventional background-elided material. Capital letters signal primary (and secondary) stress where necessary. . For a detailed analysis of echo-questions see Reis (1992).
Topic and focus in focal ellipsis . For a detailed account on the relationship between focus and prominence marking in declaratives and questions see Hetland (1992) and Rosengren (1991). The application of their ideas on FE can be found in Gretsch (2000: 74ff.) where cases of prominence shift in FE and the phenomenon of prompts are discussed. . See also Meinunger (1996: 33) where the notion FocusPhrase is defined as the minimal syntactic constituent which entails the focussed element and which can be moved: “The FocP [=FocusPhrase, PG] is the minimal constituent containing the CE [=contrastive element, PG] that when moving to its associate operator (i.e. its LF scope position) does not violate a movement constraint.” . This does not hold for FE in quotational uses (e.g. in echo-uses) which constitute a special, grammatically more liberal case of focal ellipsis. . This is not true for the case of FE in tension-expanding uses as in (a) which I do not consider any further: (a) Johanna hat gestern ihr Kind bekommen. Ratet mal den Namen! Es heißt Λ? Johanna got her child yesterday. Guess its name! It is-called . Just to enhance clarity with respect to the prior sketched focus-framework: Background ellipsis can contain syntactically focussed material, too, as e.g. in cases like (i) but never the focus-exponent: (i) A: B:
You sure had a great time during your holiday in Spain. Well, what can I say...: rain, rain, rain all over the place.
The elided part, which might be paraphrased with we had rain. . . or it was raining. . . or something similar to that information contribution, belongs to presentationally focussed material but crucially does not affect the prominence marked element of the focus domain (which itself is associated with the IS-focus of a sentence). . Roughly similar to the link in the work of Vallduví (Vallduví 1992).
References
Abbott, B. (1976). Right node raising as a test for constituenthood. Linguistic Inquiry, 7, 639–642. Abe, J. & Hoshi, H. (1999). Directionality of movement in ellipsis resolution in English and Japanese. In S. Lappin & E. Benmamoun (Eds.), Fragments: Studies in Ellipsis and Gapping (pp. 193–226). New York and Oxford: Oxford University Press. Abney, S. P. (1987). The English noun phrase in its sentential aspect. PhD Dissertation, MIT, Cambridge, MA. Akmajian, A. (1968). An interpretive principle for certain anaphoric expressions. Unpublished paper, MIT, Cambridge, MA. Albert, Ch. (1993). Sluicing and weak islands. Ms., University of California, Santa Cruz, CA. Alexiadou, A. (1997). Adverb Placement. Amsterdam: John Benjamins. Alexiadou, A. & Anagnostopoulou, E. (1998). Parametrizing AGR: Word order, V-movement, and EPP-checking. Natural Language and Linguistic Theory, 16, 491–539. Anderson, S. R. (1972). How to get even. Language, 48, 893–906. Aoun, J. & Li, Y.-H.A. (1993). Syntax of Scope. Cambridge, MA: MIT Press. Aoun, J., Hornstein, N., Lightfoot, D. & Weinberg, A. (1987). Two types of locality. Linguistic Inquiry, 18, 537–577. Asher, N. (1993). Reference to Abstract Objects in English. Dordrecht. Asher, N., Hardt, D. & Busquets, J. (1997). Discourse parallelism, scope, and ellipsis. In Proceedings of the Seventh Conference on Semantics and Linguistic Theory. Palo Alto, CA. Asher, N., Hardt, D. & Busquets, J. (2001). Discourse parallelism, scope, and ambiguity. Journal of Semantics, 18, 1–25. Bach, E. (1970). Pronominalization. Linguistic Inquiry, 1, 121–122. Bach, E., Jelinek, E., Kratzer, A. & Partee, B. (Eds.). (1995). Quantification in Natural Language. Dordrecht: Kluwer. Baker, C. L. (1970). Notes on the description of English questions: The role of the abstract question morpheme. Foundations of Language, 6, 197–219. Baker, C. L. (1989). English Syntax. Cambridge, MA: MIT Press. Baker, M. (1991). On some subject/object asymmetries in Mohawk. Natural Language and Linguistic Theory, 9, 537–576. Baltin, M. (1987). Do antecedent-contained deletions exist? Linguistic Inquiry, 18, 537–577. Baltin, M., & Postal, P. M. (1996). More on reanalysis hypothesis. Linguistic Inquiry, 27, 127–145.
References
Barbosa, P., Fox, D., Hagstrom, P., McGinnis, M., & Pesetsky, D. (Eds.). (1998). Is the Best Good Enough? Cambridge, MA: MIT Press. Barss, A. (1986). Chains and anaphoric dependence: On reconstruction and its implications. PhD Dissertation, MIT, Cambridge, MA. Barss, A., & Lasnik, H. (1986). A note on anaphora and double objects. Linguistic Inquiry, 17, 347–354. Beck, S., & Rullmann, H. (1999). A flexible approach to exhaustivity in questions. Natural Language Semantics, 7, 249–298. Beckman, M. E., & Elam, G. A. (1997). Guidelines for ToBI Labelling [Version 3.0, March 1997]. Bennis, H. (1998). Exclamatives! In R. van Bezooijen & R. Kager (Eds.), Linguistics in the Netherlands (pp. 27–40). Amsterdam: John Benjamins. Bennis, H., Corver, N., & den Dikken, M. (1998). Predication in nominal phrases. The Journal of Comparative Germanic Linguistics, 1, 85–117. Berman, S. (1991). On the Semantics and Logical Form of Wh-Clauses. Amherst, MA: GLSA. Besten, H. den (1978). On the presence and absence of wh-elements in Dutch comparatives. Linguistic Inquiry, 9, 641–672. Besten, H. den, & Webelhuth, G. (1990). Stranding. In G. Grewendorf & W. Sternefeld (Eds.), Scrambling and Barriers (pp. 77–92). Amsterdam: John Benjamins. Bhatt, R. (2000). Adjectival modifiers and the raising analysis of relative clauses. In M. Hirotani (Ed.), Proceedings of NELS 30 (pp. 159–173). Amherst, MA: GLSA. Bianchi, V. (1995). Consequences of antisymmetry for the syntax of headed relative clauses. PhD Dissertation, Scuola Normale Superiore, Pisa, Italy. Bittner, M. (1994). Cross-Linguistic Semantics. Linguistics and Philosophy, 17.1, 53–108. Bobaljik, J. (1995). Morphosyntax: The syntax of verbal inflection. PhD Dissertation, MIT, Cambridge, MA. Bolinger, D. (1972). Degree Words. The Hague: Mouton. Booij, G. (1985). Coordination reduction in complex words: a case for prosodic phonology. In H. van der Hulst & N. Smith (Eds.), Advances in Prosodic Phonology (pp. 143–160). Dordrecht: Foris. Borer, H. (1984). Parametric Syntax. Dordrecht: Foris. Borsley, R. D. (1997). Relative clauses and the theory of phrase structure. Linguistic Inquiry, 28, 629–647. Bouchard, D. (1985). Local relations in ‘long distance’ wh-constructions. Ms., University of Texas, Austin, TX. Brame, M. K. (1968). A new analysis of the relative clause: Evidence for an interpretive theory. Ms., MIT, Cambridge, MA. Brandt, M., Reis, M., Rosengren, I., & Zimmermann, I. (1992). Satztyp, Satzmodus und Illokution. In I. Rosengren (Ed.), Satz und Illokution, Vol. 1, (pp. 1–99). Tübingen: Niemeyer. Bresnan, J. (1970). An argument against pronominalization. Linguistic Inquiry, 1, 122–123. Bresnan, J. (1973). Syntax of the comparative clause construction in English. Linguistic Inquiry, 4, 275–343. Bresnan, J. (1974). The position of certain clause-particles in phrase structure. Linguistic Inquiry, 5, 614–619.
References
Bresnan, J. (1975). Comparative deletion and constraints on transformations. Linguistic Analysis, 1, 25–74. Brody, M. (1990). Some remarks on the focus field in Hungarian. University College London working papers in linguistics, 2, 201–225. Büring, D. (1999). Topic. In P. Bosch & R. van der Sandt (Eds.), Focus (pp. 142–165). New York: Cambridge University Press. Büring, D., & Hartmann, K. (1997). Doing the right thing. The Linguistic Review, 14, 1–42. Büring, D., & Hartmann, K. (2001). The syntax and semantics of focus-sensitive particles in German. Natural Language and Linguistic Theory, 19, 229–281. Carlson, G. (1977a). Amount relatives. Language, 58, 520–542. Carlson, G. (1977b). Reference to kinds in English, PhD Dissertation, University of Massachusetts, Amherst, MA. Carlson, G. & Pelletier, J.-F. (Eds.). (1995). The Generic Book. Chicago, IL: University of Chicago Press. Castañeda, H.-N. (1967). Comments on Donald Davidson’s ‘The Logical Form of Action Sentences’. In N. Rescher (Ed.), The Logic of Decision and Action (pp. 104–112). Pittsburgh, PA: University of Pittsburgh Press. Cecchetto, C. (1998). Optionality and directionality. A view from leftward and rightward scrambling in Japanese. Ms., Università di Siena. Chao, W. (1988). On Ellipsis. New York and London: Garland. Cheng, L. L.-S., & Sybesma, R. (1999). Bare and not-so-bare nouns and the structure of NP. Linguistic Inquiry, 30, 509–542. Chierchia, G. (1993). Questions with quantifiers. Natural Language Semantics, 1, 181–234. Chierchia, G. (1997). Plurality of mass nouns and the notion of ‘Semantic Parameter’. Ms., University of Milan. Chierchia, G. (1998). Reference to kinds across languages. Natural Language Semantics, 6, 339–405. Chierchia, G. (2001). Scalar implicatures, polarity phenomena and the syntax/semantic interface. Ms., University Milan. Choi, H.-W. (1997). Topic and focus in Korean: The information partition by phrase structure and morphology. In Japanese/Corean Linguistics (pp. 545–561). Stanford, CA: CSLI. Chomsky, N. (1965). Aspects of the Theory of Syntax. Cambridge, MA: MIT Press. Chomsky, N. (1977a). Conditions on rules of grammar. In N. Chomsky (Ed.), Essays on Form and Interpretation (pp. 163–207). New York: North Holland. Chomsky, N. (1977b). On wh-movement. In P. Culicover, T. Wasow, & A. Akmajian (Eds.), Formal Syntax (pp. 71–132). New York: Academic Press. Chomsky, N. (1981). Lectures on Government and Binding. Dordrecht: Foris. Chomsky, N. (1982). Noam Chomsky on the Generative Enterprise. Dordrecht: Foris. Chomsky, N. (1986a). Barriers. Cambridge, MA: MIT Press. Chomsky, N. (1986b). Knowledge of Language. New York: Praeger. Chomsky, N. (1993). A minimalist program for linguistic theory. In K. Hale & S. J. Keyser (Eds.), The View from Building 20: Essays in Linguistics in Honor of Sylvain Bromberger (pp. 1–52). Cambridge, MA: MIT Press. Chomsky, N. (1995). The Minimalist Program. Cambridge, MA: MIT Press.
References
Chomsky, N. (1998). Minimalist Inquiries: The Framework. MIT Occasional Papers in Linguistics 15 [MIT Working Papers in Linguistics]. Cambridge, MA, MIT. Chomsky, N. (1999). Derivation by Phases [MIT Occasional Papers in Linguistics 18]. Cambridge, MA. Chomsky, N. (2000). Minimalist inquiries: the framework. In R. Martin, D. Michaels & J. Uriagereka (Eds.), Step by Step (pp. 89–156). Cambridge, MA: MIT Press. Chomsky, N. (2001). Derivation by phase. In M. Kenstowicz (Ed.), Ken Hale: A Life in Language (pp. 1–52). Cambridge, MA: MIT Press. Chomsky, N., & Lasnik, H. (1993). The theory of principles and parameters. In J. Jacobs et al. (Eds.), Syntax: An International Handbook of Contemporary Research [Volume 1], (pp. 506–569). Berlin and New York: Walter de Gruyter. Chung, S. (1991). Functional heads and proper government in Chamorro. Lingua, 85, 85– 134. Chung, S. (1998). The Design of Agreement: Evidence from Chamorro. Chicago, IL: University of Chicago Press. Chung, S., Ladusaw, W. A., & McCloskey, J. (1995). Sluicing and logical form. Natural Language Semantics, 3, 239–282. Cinque, G. (1990). Types of A’-dependencies. Cambridge, MA: MIT Press. Cinque, G. (1999). Adverbs and Functional Heads. Oxford: Oxford University Press. Clements, G. N. (1984). Binding domains in Kikuyu. Studies in the Linguistic Sciences, 14, 37–56. Collins, C. (1996). Local Economy. Cambridge, MA: MIT Press. Cooper, R. (1979). The interpretation of pronouns. In F. Heny & H. Schnelle (Eds.), Syntax and Semantics 10: Selections from the 3rd Groningen Roundtable (pp. 61–92). New York: Academic Press. Corver, N. (1990). The syntax of left branch extractions. PhD Dissertation, Tilburg University. Corver, N. (1993). A note on subcomparatives. Linguistic Inquiry, 24, 773–781. Cresti, D. (1995). Extraction and reconstruction. Natural Language Semantics, 3, 79–122. Dahl, O. (1974). How to open a sentence: abstraction in natural language. In Logical Grammar Reports 12. University of Gothenburg, Sweden. Dalrymple, M., Shieber, S. M., & Pereira, F. C. N. (1991). Ellipsis and higher-order unification. Linguistics and Philosophy, 14, 399–452. Daniel, H. (1999). Dynamic interpretation of verb phrase ellipsis. Linguistics and Philosophy, 22, 187–221. Davidson, D. (1967). The Logical Form of Action Sentences. In N. Rescher (Ed.), The Logic of Decision and Action (pp. 81–95). Pittsburgh, PA: University of Pittsburgh Press. Dayal, V. (1999). Bare NP’s, Reference to Kinds, and Incorporation. In T. Matthews & D. Strolovitch (Eds.) Proceedings of SALT IX. Ithaca, NY: CLC, Cornell University. Diesing, M. (1992). Indefinites. Cambridge, MA: MIT Press. Dimroth, C., & Gretsch, P. (1999). Ansätze zum Realismus. MPI-Ms., Max-Planck-Institut für Psycholinguistik, Nijmegen. DiSciullo, A. M., & Williams, E. (1987). On the definition of word. Cambridge, MA: MIT Press. Donati, C. (2000). La sintassi della comparazione. Padova: Unipress.
References
Donati, C. & Tomaselli, A. (1997). Language types in generative grammar. In D. Beerman, D. LeBlanc, & H. van Riemsdijk (Eds.), Rightward Movement (pp. 332–355). Amsterdam: John Benjamins. Doron, E. (1990). V-movement and VP-ellipsis, Ms., Hebrew University of Jerusalem. Doron, E. (1999). V-movement and VP-ellipsis. In S. Lappin & E. Benmamoun (Eds.), Fragments: Studies in Ellipsis and Gapping (pp. 124–140). New York and Oxford: Oxford University Press. Dougherty, R. C. (1969). An Interpretive theory of pronominal reference. Foundations of Language, 5, 488–519. Dowty, D. (1989). On the semantic content of the notion ‘semantic role’. In G. Chierchia, B. Partee, & R. Turner (Eds.), Properties, Types and Meaning. Volume II. Semantic Issues (pp. 69–131). Dordrecht/Boston/London: Kluwer. Drubig, H. B. (1994). Island Constraints and the Syntactic Nature of Focus and Association with Focus. [Universität Tübingen, Seminar für Sprachwissenschaft: Arbeitspapiere des Sonderforschungbereichs 340: Sprachtheoretische Grundlagen für die Computerlinguistik, Bericht Nr. 51], Tübingen. Drubig, B. (1997). Some Cross-Categorical Generalizations of Focus Structure. [Universität Tübingen: Arbeitspapiere des Sonderforschungsbereichs 340], Tübingen. Drubig, B. (2000). Towards a typology of focus and focus constructions. Ms., University of Tübingen. Emonds, J. (1976). A Transformational Approach to Syntax. New York: Academic Press. Emonds, J. (1978). The verbal complex in French. Linguistic Inquiry, 9, 151–175. Enç, M. (1991). The semantics of specificity. Linguistic Inquiry, 22, 1–25. Engdahl, E. (1985). Parasitic gaps, resumptive pronouns, and subject extractions. Linguistics, 23, 3–44. Ernst, T. (2001). The Syntax of Adjuncts. Cambridge, MA: Cambridge University Press. Evans, G. (1977). Pronouns, quantifiers, and relative clauses. Part 1 and 2. In Canadian Journal of Philosophy, 7, 467–536, 777–797. Evans, G. (1980). Pronouns. In Linguistic Inquiry, 11, 337–362. Fanselow, G. (1983). Zu einigen Problemen von Kasus, Rektion and Bindung im Deutschen. Ms., University of Konstanz. Fanselow, G. (1992). Deplazierte Argumente. Ms., University of Stuttgart. Fanselow, G. (2001). Against remnant movement? Ms., University of Potsdam. Paper available at: http://www.ling.uni-potsdam.de/-fanselow/gf-pub.html Farkas, D. (1995). Specificity and scope. In L. Nash & G. Tsoulas (Eds.), Actes du Premier Colloque Langues & Grammaire (pp. 119–137). Paris. Farkas, D. (2002). Specificity Distinctions. To appear in a special issue of the Journal of Semantics. Féry, C., & Hartmann, K. (2001). Prosodic structures of German gapping and right node raising. Ms., Berlin and Potsdam. Fiengo, R., & Higginbotham, J. (1981). Opacity in NP. Linguistic Analysis, 7, 395–421. Fiengo, R., & May, R. (1994). Indices and Identity. Cambridge, MA: MIT Press. Fintel, K. von (1994). Restrictions on Quantifier Domains. Amherst, MA: GLSA. Fintel, K. von, & Heim, I. (1999). Pragmatics in Linguistic Theory. MIT lecture notes, MIT, Cambridge, MA.
References
Fodor, J. A., & Fodor, J. D. (1980). Functional Structure, Quantifiers, and Meaning Postulates. Linguistic Inquiry, 11.4, 759–770. Fodor, J., & Sag, I. (1982). Referential and Quantificational Indefinites. Linguistics and Philosophy, 5, 355–398. Fox, D. (1995). Economy and scope. Natural Language Semantics, 3(3), 283–341. Fox, D. (1998). Economy and semantic interpretation. PhD Dissertation, MIT, Cambridge, MA. Fox, D. (1999a). Focus, parallelism, and accommodation. In T. Matthews & D. Strolovitch (Eds.), Proceedings of SALT IX, (pp. 70–90). Ithaca, NY: CLC, Cornell University. Fox, D. (1999b). Reconstruction, variable binding and the interpretation of chains. Linguistic Inquiry, 30, 157–196. Fox, D. (2000a). Antecedent contained deletion and the copy theory of movement. Ms., Harvard University, Cambridge, MA. Fox, D. (2000b). Economy and Semantic Interpretation. Cambridge, MA: MIT Press. Fraser, B. (1975). Hedged Performatives. In P. Cole & J. L. Morgan (Eds.), Syntax and Semantics Vol.3: Speech Acts (pp. 187–210). New York: Academic Press. Freidin, R. (1986). Fundamental issues in the theory of binding. In B. Lust (Ed.), Studies in the Acquisition of Anaphora, Vol. I (pp. 151–188). Dordrecht: Reidel. Fukui, N., & Speas, M. (1986). Specifiers and Projections. In N. Fukui, T. Rapoport, & E. Sagey (Eds.), MIT Working Papers in Linguistics, 8. Cambridge, MA: MIT. Gasde, H.-D. (1998). Topics, foci and sentence structure in Mandarin Chinese. Sprachtypologie und Universalienforschung, 51(1), 43–94. Geach, P. T. (1962). Reference and Generality. Ithaca, NY: Cornell University Press. Giannakidou, A., & Merchant, J. (1997). On the interpretation of null indefinite objects in Greek. In Studies in Greek Linguistics, 17, (pp. 141–155). Thessaloniki: Aristotle University. Gleitman, L. (1965). Coordinating conjunctions in English. Language, 41, 260–293. Goldberg, A. (1995). Constructions: A Construction Grammar Approach to Argument Structure. Chicago, IL: University of Chicago Press. Goldsmith, J. (1981). Complementizers and root sentences. Linguistic Inquiry, 12, 541–574. Gretsch, P. (2000). Fokale Ellipsen in Ergänzungsfragefunktion in Erwachsenen- und Kindersprache. Tübingen: Niemeyer. Grice, H. P. (1975). Logic and Conversation. In P. Cole & J. L. Morgan (Eds.), Syntax and Semantics, Vol. 3: Speech Acts (pp. 41–58). New York: Academic Press. Grimshaw, J. (1987). Subdeletion. Linguistic Inquiry, 18, 659–669. Grimshaw, J. (1991). Extended projection. Ms., Brandeis University. Grimshaw, J. (1997). Projections, heads and optimality. Linguistic Inquiry, 28, 373–422. Groenendijk, J., & Stokhof, M. (1984). Studies on the semantics of questions and the pragmatics of answers. PhD Dissertation, Universiteit van Amsterdam. Grosu, A. (1974). On the nature of the left branch condition. Linguistic Inquiry, 5, 308–319. Grosu, A. (1976). A note on subject raising to object and Right Node Raising. Linguistic Inquiry, 7, 642–645. Grosu, A. (1994). Three studies in locality and case. London: Routledge. Grosu, A., & Landman, F. (1998). Strange relatives of the third kind. Natural Language Semantics, 6, 125–170.
References
Guéron, J., & May, R. (1984). Extraposition and logical form. Linguistic Inquiry, 15, 1–31. Hackl, M., & Nissenbaum, J. (1998). Variable modal force in for-infinitival relative clauses. Ms., MIT, Cambridge, MA. Haegeman, L. (1991). Introduction to Government and Binding Theory. [Blackwell Textbooks in Linguistics 1]. Cambridge, MA: Blackwell. Haegeman, L., & Zanuttini, R. (1991). Negative heads and the Neg-criterion. The Linguistic Review, 8, 233–251. Haider, H. (1993). Deutsche Syntax – Generativ. Tübingen: Gunter Narr Verlag. Haïk, I. (1987). Bound VPs that need to be. Linguistics and Philosophy, 10, 503–530. Halle, M., & Marantz, A. (1993). Distributed morphology and the pieces of inflection. In K. Hale & S. J. Keyser (Eds.), The View from Building 20: Essays in Linguistics in Honor of Sylvain Bromberger, Chapter 3 (pp. 111–176). Cambridge, MA: MIT Press. Hamblin, C. L. (1973). Questions in Montague English. Foundations of Language, 10, 41–53. Hankamer, J. (1971). Constraints on deletion in syntax, PhD Dissertation, University of Yale, New Haven, CT. Hankamer, J. (1973a). Unacceptable ambiguity. Linguistic Inquiry, 4, 17–68. Hankamer, J. (1973b). Why there are two than’s in English. In C. Corum, T. C. SmithStark, & A. Weiser (Eds.), Papers from the 9th Regional Meeting of the Chicago Linguistics Society (pp. 179–191). Chicago, IL: Chicago Linguistics Society. Hankamer, J. (1979). Deletion in Coordinate Structures [Outstanding Dissertations in Linguistics 16]. New York: Garland. (PhD Dissertation, University of Yale, New Haven, CT, 1971). Hankamer, J., & Sag, I. (1976). Deep and surface anaphora. Linguistic Inquiry, 7, 391–428. Hardt, D. (1992). VP ellipsis and semantic identity. In C. Barker & D. Dowty (Eds.), Proceedings of SALT II [Working Papers in Linguistics] (pp. 145–161). Columbus, OH: Ohio State University, Hardt, D. (1993). Verb phrase ellipsis: Form, meaning and processing. PhD Dissertation, University of Pennsylvania, Philadelphia, PA. Hardt, D. (1999). VPE as a proform: Some consequences for binding. In F. Corblin, C. Dobrovie-Sorin & J.-M. Marandin (Eds.), Empirical Issues in Formal Syntax and Semantics 2: Selected Papers from the Colloque de Syntaxe et Semantique d Paris (pp. 215–232). The Hague: Thesus. Hartmann, K. (2000). Right Node Raising and Gapping: Interface Conditions on Prosodic Deletion. Amsterdam and Philadelphia: John Benjamins. Heim, I. (1982). The semantics of definite and indefinite noun phrases, PhD Dissertation, Amherst, MA, GLSA. (Published by New York: Garland, 1988) Heim, I. (1987). Where does the definiteness restriction apply? Evidence from the definiteness of variables. In E. Reuland & A. ter Meulen (Eds.), The Representation of (In)definiteness, Chapter 2 (pp. 21–42). Cambridge, MA: MIT Press. Heim, I. (1988). The Semantics of Definite and Indefinite Noun Phrases. New York: Garland. (PhD Dissertation, Amherst, MA, GLSA, 1982). Heim, I. (1990). E-type pronoun and donkey anaphora. Linguistics and Philosophy, 14, 137– 177.
References
Heim, I. (1994). Interrogative semantics and Karttunen’s semantics for know. In A. Mittwoch et al. (Eds.), Proceedings of the Israeli Association for Theoretical Linguistics (pp. 128–144). Jerusalem. Heim, I. (1997). Predicates or formulas?: Evidence from ellipsis. In, A. Lawson & E. Cho (Eds.), Proceedings from SALT VII (pp. 197–221). Ithaca, NY: CLC, Cornell University. Heim, I., & Kratzer, A. (1998). Semantics in Generative Grammar. [Blackwell textbooks in linguistics 13]. London: Blackwell. Hendrik, R. (1990). Operator movement in NP. In A. Halpern (Ed.), Proceedings of WCCFL 9 (pp. 249–262). Stanford, CA: CSLI Publications. Hendriks, P., & de Hoop, H. (2001). Optimality theoretic semantics. Linguistics and Philosophy, 24, 1–32. Hestvik, A. (1995). Reflexives and ellipsis. Natural Language Semantics, 3, 211–237. Hetland, J. (1992). Satzadverbien im Fokus. [Studien zur deutschen Grammatik 43]. Tübingen: Narr. Heusinger, K. von (1997). Salienz und Referenz: Der Epsilonoperator in der Semantik der Nominalphrase und anaphorischer Pronomen [Studia grammatica 43]. Berlin: Akademie Verlag. Heusinger, K. von (2000). The reference of indefinites. In K. von Heusinger & U. Egli (Eds.), Reference and Anaphoric Relations (pp. 247–265). Dordrecht: Kluwer. Heusinger, K. von (2002). Specificity and definiteness in sentence and discourse structure. To appear in a special issue of the Journal of Semantics. Heycock, C. (1995). Asymmetries in reconstruction. Linguistic Inquiry, 26, 547–570. Higginbotham, J. (1985). On semantics. Linguistic Inquiry, 16, 547–593. Hirschbühler, P. (1982). VP deletion and across-the-board quantifier scope. In J. Pustejovsky & P. Sells (Eds.), Proceedings of NELS 12 (pp. 132–139). Amherst, MA, GLSA. Hobbs, J. R. (1979). Coherence and coreference. Cognitive Science, 3, 67–90. Hoji, H. (1998). Null object and sloppy identity in Japanese. Linguistic Inquiry, 29, 127–152. Hornstein, N. (1994). An argument for minimalism: The case of Antecedent-Contained Deletion. Linguistic Inquiry, 25, 455–480. Horvath, J. (1986). FOCUS in the Theory of Grammar and the Syntax of Hungarian. Dordrecht: Foris. Huang, J. C.-T. (1982). Logical Relations in Chinese and the Theory of Grammar. PhD Dissertation, MIT, Cambridge, MA. Huang, J. C.-T. (1984). On the distribution and reference of empty pronouns. Linguistic Inquiry, 15, 531–74. Huang, J. C.-T. (1989). Pro-drop in Chinese: A generalized control theory. In O. Jaeggli & K. Safir (Eds.), The Null Subject Parameter (pp. 185–214). Dordrecht: Kluwer. Huang, J. C.-T. (1993). Reconstruction and the structure of VP: Some theoretical consequences. Linguistic Inquiry, 24, 103–138. Izvorksi, R. (1995). A solution to the subcomparative paradox. In J. Camacho, L. Choueiri & M. Watanabe (Eds.), Proceedings of the 14th West Coast Conference on Formal Linguistics (pp. 203–219). Stanford, CA: CSLI Publications Jackendoff, R. S. (1972). Semantic Interpretation in Generative Grammar. Cambridge, MA: MIT Press.
References
Jackendoff, R. (1998). The architecture of the Language Faculty: A neominimalist perspective. In P. Culicover & L. McNally (Eds.), The Limits of Syntax [Syntax and Semantics, Vol. 29] (pp. 19–46). San Diego, CA, and London: Academic Press. Jacobs, J. (1983). Fokus und Skalen. Niemeyer: Tübingen. Jacobs, J. (1984). Funktionale Satzperspektive und Illokutionssemantik. In Linguistische Berichte, 91, (pp. 25–58). Opladen: Westdeutscher Verlag. Jacobs, J. (1986). The syntax of focus and adverbials. In W. Abraham & S. de Meij (Eds.), Topic, Focus and Configurationality (pp. 103–128). Amsterdam and Philadelphia: John Benjamins. Jacobson, P. (1992). Antecedent contained deletion in a variable free semantics. In C. Barker & D. Dowty (Eds.), Proceedings of SALT II [Working Papers in Linguistics], (pp. 193– 213). Columbus, OH: Ohio State University, Jaeggli, O. (1982). Topics in Romance syntax. Dordrecht: Foris. Jayaseelan, K.A. (1990). Incomplete VP deletion and gapping. Linguistic Analysis, 20, 64–81. Jelinek, E. (1984). Empty categories, case and configurationality. Natural Language and Linguistic Theory, 2, 38–70. Johnson, K. (1994). Bridging the gap. Ms., University of Massachusets, Amherst, MA. Johnson, K. (1996). Gapping. In search of the middle field. Ms., University of Massachusets, Amherst, MA. Johnson, K. (1997a). In search of the English middle field. Ms., University of Massachusets, Amherst, MA. Johnson, K. (1997b). When verb phrases go missing. Ms., University of Massachusetts, Amherst, MA. Johnson, K. (2000). Gapping determiners. In K. Schwabe & N. Zhang (Eds.), Ellipsis in Coordination (pp. 95–115). Tübingen: Niemeyer. Johnson, K. (2001). What VP-ellipsis can do and what it can’t, but not why. In M. Baltin & C. Collins (Eds.), The Handbook of Contemporary Syntactic Theory (pp. 439–479). Malden, MA: Blackwell. Kamp, H. (1981). A theory of truth and semantic representation. In J. Groenendijk, T. Janssen & M. Stokhof (Eds.), Formal Methods in the Study of Language (pp. 277–322). Amsterdam: MCT. Kamp, H., & Reyle, U. (1993). From Discourse to Logic. Introduction to Modeltheoretic Semantics of Natural Language, Formal Logic and Discourse Representation Theory. Dordrecht: Kluwer. Karttunen, L. (1969). Pronouns and variables. In Binnick et al. (Eds.), Papers from the 5th Regional Meeting of the Chicago Linguistics Society (pp. 108–115). University of Chicago, Chicago, IL. Karttunen, L. (1977). Syntax and semantics of questions. Linguistics and Philosophy, 1, 3–44. Kay, P., & Fillmore, C. (1999). Grammatical constructions and linguistic generalizations: The What’s X doing Y? construction. Language, 75, 1–33. Kayne, R. S. (1981). ECP extensions. Linguistic Inquiry, 12, 93–133. Kayne, R. S. (1989). Facets of Romance past participle agreement. In P. Benincà (Ed.), Dialect variation and the theory of grammar (pp. 85–103). Dordrecht: Foris. Kayne, R. S. (1994). The Antisymmetry of Syntax. Cambridge, MA: MIT Press. Kayne, R. S. (1998). Overt vs. covert movement. Syntax, 1(2), 128–191.
References
Kehler, A. (1995). Interpreting cohesive forms in the context of discourse inference. PhD Dissertation, Harvard University, Cambridge, MA. Kehler, A. (2000). Coherence and the resolution of ellipsis. Linguistics and Philosophy, 23, 533–575. Kempson, R., Meyer-Viol, W., & Gabbay, D. (1999). VP ellipsis: Toward a dynamic, structural account. In S. Lappin & E. Benmamoun (Eds.), Fragments: Studies in Ellipsis and Gapping (pp. 227–289). New York and Oxford: Oxford University Press. Kennedy, C. (1994). Argument contained ellipsis. [Linguistics Research Center Report LRC94-03], University of California, Santa Cruz, CA. Kennedy, C. (1997a). Antecedent-contained deletion and the syntax of quantification. Linguistic Inquiry, 28, 662–688. Kennedy, C. (1997b). Projecting the adjective: The syntax and semantics of gradability and comparison. PhD Dissertation, University of California, Santa Cruz, CA. Kennedy, C. (1999). Projecting the Adjective: The Syntax and Semantics of Gradability and Comparison. New York: Garland. Kennedy, C. (2002). Comparative deletion and optimality in syntax. Natural Language and Linguistic Theory, 20, 553–621. Kennedy, C., & Lidz, J. (2001). A (covert) long distance anaphor in English. In Proceedings of the 20th West Coast Conference on Formal Linguistics. Cambridge, MA: Cascadilla Press. Kennedy, C., & Merchant, J. (2000). Attributive comparative deletion. Natural Language and Linguistic Theory, 18, 89–146. Kennelly, S. (1999). The syntax of the P-focus position in Turkish. In G. Rebuschi & L. Tuller (Eds.), The Grammar of Focus (pp. 179–212). Philadelphia: John Benjamins. Kidwai, A. (1999). Word order and focus positions in Universal Grammar. In G. Rebuschi & L. Tuller (Eds.), The Grammar of Focus (pp. 213–244). Philadelphia: John Benjamins. Kim, J.-S. (1997). Syntactic focus movement and ellipsis: a minimalist approach. PhD Dissertation, University of Connecticut. Kim, S., & Lyle, J. (1996). Parasitic gaps, multiple questions, and VP ellipsis. In J. Camacho, L. Choueiri, & M. Watanabe (Eds.), Proceedings of the West Coast Conference on Formal Linguistics, Vol. 14 (pp. 287–301). Stanford, CA: CSLI Publications. Kitagawa, Y. (1991). Copying identity. Natural Language and Linguistic Theory, 9, 497–536. Klein, E. (1987). VP ellipsis in DR theory. In J. Groenendijk, D. de Jongh, & M. Stokhof (Eds.), Studies in DRT and the Theory of Generalized Quantifiers (pp. 161–187). Dordrecht: Foris. Klein, W. (1981). Some rules of regular ellipsis in German. In W. Klein & W. Levelt (Eds.), Crossing the Boundaries in Linguistics (pp. 51–78). Dordrecht: Reidel. Klein, W. (1993). Ellipse. In J. Jacobs et al. (Eds.), Syntax: An International Handbook of Contemporary Research [Volume 1], (pp. 763–799). Berlin and New York: de Gruyter. Klein, W. (1994). Time in Language. London and New York: Routledge. Klein, W. (1998). Assertion and finiteness. In N. Dittmar & Z. Penner (Eds.), Issues in the Theory of Language Acquisition: Essays in Honor of Jürgen Weissenborn (pp. 225–245). Bern: Lang. Koster, J. (1978). Locality principles in syntax. Dordrecht: Foris. Kratzer, A. (1989). An investigation of the lumps of thought. Linguistic and Philosopohy, 12, 607–653.
References
Kratzer, A. (1996). Severing the external argument from its verb. In J. Rooryck & L. Zaring (Eds.), Phrase Structure and the Lexicon (pp. 109–138). Dordrecht: Kluwer. Krifka, M. (1991). A compositional semantics for multiple focus constructions. In S. Moore & A. Z. Wyner (Eds.), Proceedings of SALT I (pp. 127–158). Ithaca, NY: CLC Cornell University. Krifka, M. (1992). Definite NPs aren’t quantifiers, Linguistic Inquiry, 23, 156–163. Krifka, M. (1995). Common nouns: A contrastive analysis of Chinese and English. In G. Carlson & F.-J. Pelletier (Eds.), The Generic Book (pp. 398–411). Chicago, IL: University of Chicago Press. Krifka, M. (2001a). For a structured account of questions and answers. In C. Féry & W.Sternefeld (Eds.), Audiatur vox sapientiae (pp. 287–319). Berlin. Krifka, M. (2001b). Non-novel indefinites in adverbial quantification. To appear in C. Condoravdi & G. Renardel (Ed.), Logical Perspectives on Language and Information. Stanford, CA: CSLI. Kuno, S. (1976). Gapping: A functional analysis. Linguistic Inquiry, 7, 300–318. Kuno, S. (1981). The syntax of comparative clauses. In R. Hendrick, C. Masek & M. Miller (Eds.), Papers from the 17th Regional Meeting of the Chicago Linguistics Society (pp. 136– 155). Chicago, IL: Chicago Linguistics Society. Kuno, S. (1997). Binding theory in the minimalist program. Ms., Harvard University, Cambridge, MA. Kurafuji, T. (1999). Japanese pronouns in dymanic semantics: The null/overt contrast. PhD Dissertation, Rutgers University, New Brunswick, NJ. Kuroda, S.-Y. (1968). English relativization and a certain related problem. Language, 44, 244–266. Kuroda, S.-Y. (1992). Japanese Syntax and Semantics. Dordrecht: Kluwer. Ladd, R. (1978). The Structure of Intonational Meaning. Bloomington, IN: Indiana University Press. Lahiri, U. (1991). Embedded interrogatives and predicates that embed them. PhD Dissertation, MIT, Cambridge, MA. Lapointe, S. (1980). A theory of grammatical agreement. PhD Dissertation. University of Massachusetts, Amherst, MA. Lappin, S. (1992). The syntactic basis of ellipsis resolution. In S. Berman & A. Hestvik (Eds.), Proceedings of the Stuttgart Ellipsis Workshop. [Arbeitspapiere des Sonderforschungsbereichs 340: Sprachtheoretische Grundlagen für die Computerlinguistik 29]. Heidelberg: IBM. Lappin, S. (1996). The interpretation of ellipsis. In S. Lappin (Ed.), The Handbook of Contemporary Semantic Theory (pp. 145–175). Oxford: Blackwell. Larson, R. (1988). On the double object construction. Linguistic Inquiry, 19, 335–392. Larson, R. (1990). Double objects revisited: Reply to Jackendoff. Linguistic Inquiry, 21, 563– 589. Larson, R. K., & May, R. (1990). Antecedent containment or vacuous movement: Reply to Baltin. Linguistic Inquiry, 21, 103–122. Larson, R., & Segal, G. (1995). Knowledge of Meaning: An Introduction to Semantic Theory. Cambridge, MA: MIT Press. Lasnik, H. (1995a). Lecture Notes. Berlin.
References
Lasnik, H. (1995b). A note on pseudogapping. In R. Pensalfini & H. Ura (Eds.), Papers on Minimalist Syntax [MIT Working Papers in Linguistics 27] (pp. 143–163). Cambridge, MA: MIT. Lasnik, H. (1998). Some reconstruction riddles. In Al. Dimitriadis et al. (Eds.), Proceeding of the Penn Linguistics Colloquium (pp. 83–98). University of Pennsylvania, Philadelphia, PA. Lasnik, H. (1999a). On feature strength: Three Minimalist approaches to overt movement. Linguistic Inquiry, 30, 197–217. Lasnik, H. (1999b). Pseudogapping puzzles. In S. Lappin & E. Benmamoun (Eds.), Fragments: Studies in Ellipsis and Gapping (pp. 141–175). New York and Oxford: Oxford University Press. Lasnik, H., & Saito, M. (1992). Move-α. Cambridge, MA: MIT Press. Lasnik, H., & Sobin, M. (2000). The who/whom puzzle: On the preservation of an archaic feature. Natural Language and Linguistic Theory, 18, 343–371. Lebeaux, D. (1988). Language acquisition and the form of grammar. PhD Dissertation, University of Massachusetts, Amherst, MA. Lebeaux, D. (1990). Relative clauses, licensing, and the nature of the derivation. In J. Carter et al. (Eds.), Proceedings of NELS 20 (pp. 318–332). Amherst, MA: GLSA. Lebeaux, D. (1991). Relative clauses, licensing, and the nature of the derivation. In S. Rothstein & M. Speas (Eds.), Perspectives on Phrase Structure: Heads and Licensing [Syntax and Semantics 25] (pp. 209–239). New York: Academic Press. Lebeaux, D. (1998). Where does the binding theory apply? [Technical Report 98–044]. NEC Research Institute, Princeton, NJ. Lechner, W. (1998). Two kinds of reconstruction. Studia Linguistica, 52, 276–310. Lechner, W. (1999). Comparatives and DP-structure. PhD Dissertation, University of Massachusetts, Amherst, MA. Lechner, W. (2000). Phrase Structure Paradoxes. Lecture notes for 7th Central European Summer School in Generative Grammar, Blagoevgrad – Bulgaria. Available at http:// www2.sfs.nphil.uni-tuebingen.de/winnie/ Lees, R. B. (1960). The Grammar of English Nominalizations. The Hague: Mouton. Lees, R. B. (1961). The constituent structure of noun phrases. American Speech, 36, 159–168. Lenerz, J. (2001). Word order variation. In G. Müller & W. Sternefeld (Eds.), Competition in Syntax (pp. 249–281). Berlin: de Gruyter. Levin, L. (1982). Sluicing: A lexical interpretation procedure. In J. Bresnan (Ed.), The Mental Representation of Grammatical Relations (pp. 590–654). Cambridge, MA: MIT Press. Levin, N. (1986). Main-Verb Ellipsis in Spoken English. New York: Garland. (PhD Dissertation, Ohio State University, Columbus, OH, 1979.) Levine, R. (1984). A note on Right Node Raising, tough constructions and reanalysis rules. Linguistic Analysis, 13, 159–172. Liberman, M. (1974). On conditioning the rule of Subject Aux Inversion. In J. Aissen & J. Hankamer (Eds.), Papers from the 5th meeting of the North Eastern Linguistics Society (pp. 77–91). Harvard University, Cambridge, MA. Lobeck, A. (1987). Syntactic Constraints on Ellipsis. Bloomington, IN. Lobeck, A. (1991). The phrase structure of ellipsis. In S. Rothstein (Ed.), Syntax and Semantics 25. San Diego, CA: Academic Press.
References
Lobeck, A. (1995). Ellipsis: Functional Heads, Licensing and Identification. New York and Oxford: Oxford University Press. Lobeck, A. (1999). VP Ellipsis and the minimalist program: some speculations and proposals. In S. Lappin & E. Benmamoun (Eds.), Fragments: Studies in Ellipsis and Gapping (pp. 98–124). New York and Oxford: Oxford University Press. López, L. (1995). Polarity and predicate anaphora. PhD Dissertation, Cornell University, Ithaca, NY. López, L. (2001a). The causee and the theory of bare phrase structure. In J. Gutierrez-Rexach & L. Silva-Villar (Eds.), Current Issues in Spanish Syntax and Semantics (pp. 221–242). Berlin: Mouton de Gruyter. López, L. (2001b). On the (non)complementarity of θ-Theory and Checking Theory. Linguistic Inquiry, 32(4), 694–716. López, L., & Winkler, S. (2000). Focus and topic in VP-Anaphora constructions. Linguistics, 38, 623–664. López, L. (in preparation). The Edges of Catalan: On the Syntax/Pragmatics Interface. University of Illinois, Chicago, IL. Manzini, M. R., & Savoia, L. (in press). Morfosintassi dei dialetti italiani. Università di Firenze. Marcus, M. P., Santorini, B., & Marcinkiewicz, M. A. (1993). Building a large annotated corpus of English: The Penn Treebank. Computational Linguistics, 19(2), 313–330. May, R. (1977). The grammar of quantification. PhD Dissertation, MIT, Cambridge, MA. May, R. (1985). Logical Form: Its Structure and Derivation. Cambridge, MA: MIT Press. McCawley, J. D. (1996). The focus and scope of only. In B. H. Partee & P. Sgall (Eds.), Discourse and Meaning: Papers in Honor of Eva Hajicova (pp. 171–193). Amsterdam and Philadelphia: John Benjamins. McCloskey, J. (1979). Transformational Syntax and Model Theoretic Semantics: A Case Study in Modern Irish. Dordrecht: Reidel. McCloskey, J. (1986). Right Node Raising and preposition stranding. Linguistic Inquiry, 17, 183–186. McCloskey, J. (1991). Clause structure, ellipsis and proper government in Irish . Lingua, 85, 259–302. McCloskey, J. (1996). On the scope of verb movement in Irish. Natural Language and Linguistic Theory, 14, 47–104. McConnell-Ginet, S. (1982). Adverbs and logical form. Language, 58(1), 144–184. Meinunger, A. (1996). Focus relations and weak islands. In J. Camacho, L. Choueiri, & M. Watanabe (Eds.), The Proceedings of the Fourteenth West Coast Conference on Formal Linguistics (pp. 335–349). Stanford, CA: CSLI Publication. Merchant, J. (1998). E-type A’-traces under sluicing. In K. N. Shahin, S. Blake, & E.-S. Kim (Eds.), Proceedings of WCCFL 17 (pp. 478–492). Stanford, CA: CSLI Publ. Merchant, J. (2000a). Antecedent-contained deletion in negative polarity items. Syntax, 3, 144–150. Merchant, J. (2000b). Economy, the copy theory, and antecedent-contained deletion. Linguistic Inquiry, 31, 566–575. Merchant, J. (2001). The Syntax of Silence: Sluicing, Islands, and the Theory of Ellipsis. Oxford: Oxford University Press.
References
Miller, P. (1992). Clitics and Constituency in Phrase Structure Grammar. New York: Garland. (PhD Dissertation, Utrecht, 1991) Milner, J.-C. (1978). Cyclicité successive, comparatives, et Cross-over en français (première partie). Linguistic Inquiry, 9, 673–693. Milsark, G. (1977). Towards an explanation of certain peculiarities of the existential construction in English. Linguistic Analysis, 3, 1–29. Molnár, V. (1991). Das TOPIK im Deutschen und im Ungarischen [Lunder germanistische Forschungen 58]. Stockholm: Almqvist and Wiksell. Molnár, V. (1993). Zur Pragmatik und Grammatik des TOPIK-Begriffes. In M. Reis (Ed.), Wortstellung und Informationsstruktur [Linguistische Arbeiten 306] (pp. 155–202). Tübingen: Niemeyer. Müller, G. (1995). On extraposition and succesive cyclicity. In U. Lutz & J. Pafel (Eds.), On Extraction and Extraposition in German (pp. 213–243). Amsterdam and Philadelphia: John Benjamins. Müller, G. (1998). Incomplete Category Fronting. Dordrecht: Kluwer. Munn, A. (1993). Topics in the syntax and semantics of coordinate structures. PhD Dissertation, University of Maryland. Munn, A. (1994). A minimalist account of reconstruction asymmetries. In M. Gonzälez (Ed.), Proceedings of NELS 24 (pp. 397–410). Amherst, MA: GLSA. Neijt, A. (1979). Gapping: A Contribution to Sentence Grammar. Dordrecht: Foris. Nissenbaum, J. (1998). Movement and derived predicates: Evidence from parasitic gaps. In U. Sauerland et al. (Eds.), The Interpretive Tract: Working Papers in Syntax and Semantics [MIT Working Papers in Linguistics 25] (pp. 247–295). Cambridge, MA: MIT. Nunes, J., & Uriagereka, J. (2000). Cyclicity and extraction domains. Syntax, 3, 20–43. Oehrle, R. (1987). Boolean properties in the analysis of gapping. In G. Huck & A. Ojeda (Eds.), Discontinuous Constituency (pp. 201–240). San Diego, CA: Academic Press. Oirsouw, R. van. (1983). Coordinated sentences. Lingua, 60, 135–145. Oirsouw, R. van. (1987). The Syntax of Coordination. London: Croom Helm. Oku, S. (2000). Definite and indefinite strict identity in vp-ellipsis. In K. Schwabe & N. Zhang (Eds.), Ellipsis in Conjunction [Linguistische Arbeiten 418] (pp. 179–194). Tübingen: Niemeyer. Ormelius-Sandblom, E. (1997). Die Modalpartikeln ja, doch und schon – Zu ihrer Syntax, Semantik und Pragmatik [Lunder germanistische Forschungen 61]. Stockholm: Almqvist & Wiksell. Ortiz de Urbina, J. (1987). Parameters in the Grammar of Basque. Dordrecht: Foris. Otani, K., & Whitman, J. (1991). V-raising and VP-ellipsis. Linguistic Inquiry, 22, 345–358. Parsons, T. (1990). Events in the Semantics of English: A Study in Subatomic Semantics. Cambridge, MA: MIT Press. Partee, B. (1987). Noun Phrase interpretation and type-shifting principles. In J. Groenendijk et al. (Eds.), Studies in Discourse Representation Theory and the Theory of Generalized Quantifiers [Groningen-Amsterdam Studies in Semantics 8] (pp. 115–143). Dordrecht: Foris. Pesetsky, D. (1982). Paths and categories. PhD Dissertation, MIT, Cambridge, MA. Pesetsky, D. (1989). Language-particular processses and the Earliness Principle. Ms., MIT, Cambridge, MA. Available at http://mit.edu/linguistics/www/pesetsky.home. html
References
Pesetsky, D. (1995). Zero Syntax. Cambridge, MA: MIT Press. Pesetsky, D. (1998). Some optimality principles of sentence pronunciation. In P. Barbosa et al. (Eds.), Is the Best Good Enough? (pp. 337–384). Cambridge, MA: MIT Press. Pesetsky, D. (2000). Phrasal Movement and its Kin. Cambridge, MA: MIT Press. Pesetsky, D., & Torrego, E. (2001). T-to-C movement: Causes and consequences. In M. Kenstowicz (Ed.), Ken Hale: A life in language (pp. 355–426). Cambridge, MA: MIT Press. Phillips, C. (1996). Order and structure. PhD Dissertation, MIT, Cambridge, MA. Available at http://www.ling.udel.edu/colin/research/diss-abs.html Phillips, C. (to appear). Linear Order and Constituency. Linguistic Inquiry. Paper available at: http://www.ling.udel.edu/colin/research/ftp.html Pilch, H. (1965). Comparative constructions in English. Language, 41, 37–58. Pinkham, J. (1982). The formation of comparative clauses in French and English. PhD Dissertation, Indiana University, Bloomington, IN. Platzack, C. (2000). Multiple interfaces. In U. Nikanne & E. van der Zee (Eds.), Cognitive Interfaces: Constraints on Linking Cognitive Information (pp. 21–53). New York and Oxford: Oxford University Press. Pollard, C., & Sag, I. (1994). Head-Driven Phrase Structure Grammar. Chicago, IL: University of Chicago Press. Postal, P. (1970). On coreferential complement subject deletion. Linguistic Inquiry, 1, 439– 500. Postal, P. (1971). Cross-Over Phenomena. New York: Holt, Rinehart and Winston. Postal, P. (1972). Some further limitations of interpretive theories of anaphora. Linguistic Inquiry, 3, 349–371. Postal, P. (1974). On Raising. Cambridge, MA: MIT Press. Postal, P. (1993). Parasitic gaps and the across-the-board phenomenon. Linguistic Inquiry, 24, 735–754. Postal, P. (1998). Three Investigations of Extraction. Cambridge, MA: MIT Press. Postal, P. (2001). Missing parasitic gaps. In P. Culicover & P. Postal (Eds.), Parasitic Gaps (pp. 401–417). Cambridge, MA: MIT Press. Potsdam, E. (1996). Syntactic issues in the English imperative. PhD Dissertation, University of California, Santa Cruz, CA. Potsdam, E. (1998). Syntactic Issues in the English Imperative. New York: Garland. (PhD Dissertation, University of California, Santa Cruz, CA, 1996). Potts, C. (2002). The syntax and semantics of as-parentheticals. Natural Language and Linguistic Theory, 20, 623–689. Prince, A., & Smolensky, P. (1993). Optimality theory: constraint interaction in generative grammar [Technical Report TR-2, Rutgers Center for Cognitive Sciences]. New Brunswick, NJ: Rutgers University. Prüst, H., & Scha, R. (1990). VP Ellipsis induces clausal parallelism. In R. Bok-Bennema & P. Coopmans (Eds.), Linguistics in the Netherlands (pp. 123–132). Dordrecht: Foris. Quirk, R., Greenbaum, S., Leech, G., & Svartvik, J. (1985). A Comprehensive Grammar of the English Language. London: Longman. Reinhart, T. (1983). Anaphora and Semantic Interpretation. Chicago, IL: University of Chicago Press.
References
Reinhart, T. (1991). Elliptical conjunctions – non-quantificational LF. In A. Kasher (Ed.), The Chomskian Turn (pp. 360–384). Oxford: Blackwell. Reinhart, T. (1992). Wh-in-situ: An apparent paradox. In P. Dekker & M. Stokhof (Eds.), Proceedings of the Eight Amsterdam Colloquium (pp. 483–491). Amsterdam: ILLC, UvA. Reinhart, T., & Reuland, E. (1993). Reflexivity. Linguistic Inquiry, 24, 657–720. Reis, M. (1992). Zur Grammatik und Pragmatik von Echo-w-Fragen. In I. Rosengren (Ed.), Satz und Illokution Bd.1 [Linguistische Arbeiten 278] (pp. 213–261). Tübingen: Niemeyer. Reis, M. (2000). Integrated parenthetical constructions and wh-movement. Paper presented at the 15th Comparative Germanic Syntax Workshop, University of Groningen. To appear in J.-W. Zwart & W. Abraham (Eds.), Studies in Comparative Germanic Syntax. Amsterdam: John Benjamins. Richards, N. (1997). What moves where when in which language? PhD Dissertation, MIT, Cambridge, MA. Richards, N. (2001). Movement in Language: Interactions and Architectures. Oxford: Oxford University Press. Riemsdijk, H. van. (1989). Movement and regeneration. In P. Beninca (Ed.), Dialectal Variation and the Theory of Grammar (pp. 105–136). Dordrecht: Foris. Rizzi, L. (1990). Relativized Minimality. Cambridge, MA: MIT Press. Robert, F., & May, R. (1994). Indices and Identity. Cambridge, MA: MIT Press. Rochemont, M. (1986). Focus in Generative Grammar. Philadelphia: John Benjamins. Romero, M. (1997a). The correlation between scope reconstruction and connectivity effects. In E. Curtis, J. Lyle & G. Webster (Eds.), Proceedings of the XVI West Coast Conference in Formal Linguistics (pp. 351–366). Stanford: CSLI Publications. Romero, M. (1997b). Recoverability conditions for sluicing. In F. Corblin et al. (Eds.), Empirical Issues in Formal Syntax and Semantics: Selected Papers from the Colloque de Syntaxe et de Semantique de Paris (CSSP 95) (pp. 193–216). Bern et al.: Peter Lang. Romero, M. (1997c). Problems for a Semantic Account of Scope Reconstruction. In G. Katz, S.-S. Kim, & H. Winhart (Eds.), Reconstruction: Proceedings of the 1997 Tübingen Workshop (Tübingen, 30–31 May 1997) [Arbeitspapier der Sonderforschungsbereichs 340, Volume 127] (pp. 127–153). Tübingen. Romero, M. (1998). Focus and reconstruction effects in wh-phrases. PhD Dissertation, University of Massachusetts, Amherst, MA. Romero, M. (1999). Intensional choice functions for which-phrases. In T. Matthews & D. Strolovitch (Eds.), Proceedings of SALT IX (pp. 255–272). Ithaca, NY: CLC, Cornell University. Romero, M. (2000a). Antecedentless sluiced wh-phrases and islands. In K. Schwabe & N. Zhang (Eds.), Ellipsis in Conjunction (pp. 195–220). Tübingen: Niemeyer. Romero, M. (2000b). Reduced conditionals and focus. In B. Jackson & T. Matthews (Eds.), Proceedings of SALT X (pp. 149–166). Ithaca, NY: CLC, Cornell University. Rooth, M.E. (1985). Association with focus. PhD Dissertation, University of Massachusetts, Amherst, MA.
References
Rooth, M.E. (1992a). Ellipsis redundancy and reduction redundancy. In S. Berman & A. Hestvik (Eds.), Proceedings of the Stuttgart Ellipsis Workshop. [Arbeitspapiere des Sonderforschungsbereichs 340: Sprachtheoretische Grundlagen für die Computerlinguistik 29]. Heidelberg: IBM. Rooth, M. E. (1992b). A theory of focus interpretation. Natural Language Semantics, 1, 75– 116. Rooth, M. E. (1997). Focus licensing of ellipsis. Paper presented at Jerusalem Conference on Semantics, Dec 1997. Rosen, C. (1976). Guess what about? InA. Ford at al. (Eds.), Proceedings of NELS 6 (pp. 205–211). Amherst, MA: GLSA. Rosengren, I. (1991). Zur Fokus-Hintergrund-Gliederung im Deklarativsatz und im Interrogativsatz. In M. Reis & I. Rosengren (Eds.), Fragesätze und Fragen: Referate anläßlich der 12. Jahrestagung der Deutschen Gesellschaft für Sprachwissenschaft, Saarbrücken 1990 (pp. 175–200). Tübingen: Niemeyer. Ross, J. R. (1967). Constraints on variables in syntax. PhD Dissertation, MIT, Cambridge, MA. Ross, J. R. (1969). Guess who? In R. I. Binnick, A. Davison, G. M. Green, & J. L. Morgan (Eds.), Papers from the Fifth Regional Meeting of the Chicago Linguistic Society (pp. 252– 286). Chicago, IL: University of Chicago. Ross, J.R. (1970). Gapping and the order of constituents. In M. Bierwisch & K. E. Heidolph (Eds.), Progress in Linguistics (pp. 249–259). The Hague: Mouton. Rothstein, S. (1995). Adverbial quantification over events. Natural Language Semantics, 3, 1–31. Rullmann, H. (1995). Maximality in the semantics of wh-constructions, PhD Dissertation, University of Massachusetts, Amherst, MA. Rullmann, H., & Beck, S. (1998). Presupposition projection and the interpretation of whichquestions. In D. Strolovitch & A. Lawson (Eds.), Proceedings of SALT VIII (pp. 215– 232). Ithaca, NY: CLC, Cornell University. Ruys, E. (2000). Weak crossover as a scope phenomenon. Linguistic Inquiry, 31, 513–539. Safir, K. (1998). Reconstruction and bound anaphora: Copy theory without deletion at LF. Ms., Rutgers University, New Brunswick, NJ. Safir, K. (1999). Vehicle change and reconstruction in A-bar chains. Linguistic Inquiry, 30, 587–620. Sag, I. (1976a). Deletion and logical form. PhD Dissertation, MIT, Cambridge, MA. Sag, I. (1976b). A logical theory of verb phrase deletion. Chicago Linguistic Society, 12, 533– 547. Sag, I. (1976c). A note on verb phrase deletion. Linguistic Inquiry, 7, 664–671. Sag, I. (1980). Deletion and logical form. New York: Garland. (PhD Dissertation, MIT, Cambridge, MA, 1976) Sag, I. (1997). English relative clause constructions. Journal of Linguistics, 33, 431–484. Saito, M. & Hoji, H. (1983). Weak crossover and move alpha in Japanese. Natural Language and Linguistic Theory, 1, 245–259. Sauerland, U. (1998a). The meaning of chains. PhD Dissertation, MIT, Cambridge, MA. Sauerland, U. (1998b). Scope Reconstruction without Reconstruction. In S. Kimary, S. Blake, & E.-S. Kim (Eds.), Proceedings of WCCFL 17 (pp. 582–597).
References
Sauerland, U. (1999). Relativized minimality effects with quantifier raising. Paper presented at ESCOL ’99, University of Connecticut, Storrs, CT. Sauerland, U. (2000). Obligatory reconstruction and the meaning of traces. [Universität Tübingen, Sonderforschungsbereich 340: Technical Report 160], Tübingen. (to appear in revised form in Natural Language Semantics). Sauerland, U. (to appear). Syntactic economy and quantifier raising. Paper available at: http://www2.sfs.nphil.uni-tuebingen.de/uli/ Sauerland, U., & Elbourne, P. (2000). Total Reconstruction, PF-movement and derivational order. To appear in Linguistic Inquiry. Available at http://www2.sfs.nphil.unituebingen. de/uli/ Schachter, P. (1973). Focus and relativization. Language, 49, 19–46. Schwabe, K. (1994). Syntax und Semantik situativer Ellipsen. [Studien zur deutschen Grammatik 48.] Tübingen: Gunter Narr. Schwabe, K. (2000). Coordinative ellipsis and information structure. In K. Schwabe & N. Zhang (Eds.), Ellipsis in Conjunction (pp. 247–269). Tübingen: Niemeyer. Schwabe, K., & Heusinger, K. von (2001). On shared indefinite NPs in coordinative structures. To appear in: Journal of Semantics. Schwarz, B. (1998). Reduced conditional in German: Event quantification and definiteness. Natural Language Semantics, 6.3, 271–301. Schwarz, B. (2000). Topics in ellipsis. PhD dissertation, University of Massachusetts, Amherst, MA. Schwarzschild, R. (1997). Interpreting accent. Ms., Rutgers University, New Brunswick, NJ. Schwarzschild, R. (1999). Givenness, avoidF and other constraints on the placement of accent. Natural Language Semantics, 7, 141–177. Selkirk, L. (1995). Sentence prosody: intonation, stress, and phrasing. In J. Goldsmith (Ed.), Handbook of Phonological Theory (pp. 550–569). Oxford: Blackwell. Sharvit, Y. (1996). The syntax and semantics of functional relative clauses. PhD dissertation, Rutgers University, New Brunswick, NJ. Sharvit, Y. (1999). Functional relative clauses. Linguistics and Philosophy, 22, 447–478. Shimada, M. (1999). VP-ellipsis and parasitic gaps. English Linguistics, 16, 145–151. Shopen, T. (1972). A Generative Theory of Ellipsis: A Consideration of the Linguistic Use of Silence. Bloomington, IN: IULC. Siegel, M. (1987). Compositionality, case and the scope of auxiliaries. Linguistics and Philosophy, 10, 53–76. Sobin, N. (1981). On Adv/PP-first reductions. Linguistic Inquiry, 12, 488–491. Sportiche, D. (1992). Clitic constructions. Ms., UCLA. Stechow, A. von (1991). Current issues in the theory of focus. In A. von Stechow & D. Wunderlich (Eds.), Semantik: Ein internationales Handbuch der zeitgenössischen Forschung. [HSK6] (pp. 804–825). Berlin: de Gruyter. Stechow, A. von, & Zimmermann, Th. E. (1984). Term answers and contextual change. Linguistics, 22, 3–40. Sternefeld, W. (1997). The proper treatment of binding in pseudo-cleft constructions. In G. Katz et al. (Eds.), Reconstruction: Proceedings of the 1997 Tübingen Workshop [Universität Tübingen: Arbeitspapier des Sonderforschungsbereichs 340]. Tübingen.
References
Sternefeld, W. (1998). The semantics of reconstruction and connectivity. [Universität Tübingen and Universität Stuttgart, Sonderforschungsbereich 340, Arbeitspapier 97], Tübingen and Stuttgart, Germany. Stowell, T. (1981). The origins of phrase structure. PhD Dissertation, MIT, Cambridge, MA. Svenonius, P. (1992). The structural location of the attributive adjective. In E. Duncan, D. Farkas, & P. Spaelti (Eds.), Proceedings of the 11 th West Coast Conference on Formal Linguistics (438–454). Stanford, CA: CSLI Publications. Swingle, K. (1993). The role of prosody in right node raising. Ms., Santa Cruz, CA. Szabolcsi, A. (1997). Quantifiers in pair-list readings. In A. Szabolcsi (Ed.), Ways of Scope Taking (pp. 311–347). Dordrecht: Kluwer. Takano, Y. (1995). Predicate fronting and internal subjects. Linguistic Inquiry, 26, 327–340. Tancredi, C. (1990). Syntactic association with focus. In D. Meyer, S. Tomioka & L. ZidaniEroglus (Eds.), Proceedings of the First Meeting of the Formal Linguistics Society of Midamerica. Madison: University of Wisconsin. Tancredi, C. (1992). Deletion, deaccenting and presupposition. PhD Dissertation, MIT, Cambridge, MA. Thiersch, C. (1985). Some Notes on Scrambling in the German Mittelfeld, VP and X-bar Theory. University of Connecticut and University of Cologne. Tiedeman, R. (1995). Some remarks on antecedent contained deletion. In S. Haraguchi & M. Funaki (Eds.), Minimalism and Linguistic Theory (pp. 67–103). Tokyo: Hituzi Syobo. Tomioka, S. (1995). [Focus]F restricts scope: Quantifer in VP-ellipsis. In M. Simons & T. Galloway (Eds.), Proceedings of SALT V (pp. 328–345). Ithaca, NY: CLC, Cornell University. Tomioka, S. (1998). Partition semantics for focus. Handout, Seminar für Sprachwissenschaft, Universität Tübingen. Torrego, E. (1984). On inversion in Spanish and some of its effects. Linguistic Inquiry, 15, 103–129. Tran, T. C. (1999). Gapping in Spanish: a minimalist account. PhD Dissertation, UCLA. Truckenbrodt, H. (1995). Prosodie und Ellipse. Ms., Berlin. Tsai, W.-T. D. (1994). On economizing the theory of A’ dependencies. PhD Dissertation, MIT, Cambridge, MA. Tuller, L. (1992). The syntax of postverbal focus constructions in Chadic. Natural Language and Linguistic Theory, 10, 303–334. Umbach, K. (2002). (De)accenting definite descriptions. To appear in a special issue of Theoretical Linguistics. Uriagereka, J. (1999). Multiple spell-out. In S. Epstein & N. Hornstein (Eds.), Working Minimalism (pp. 251–282). Cambridge, MA: MIT Press Vallduví, E. (1992). The Informational Component. [Outstanding Dissertations in Linguistics]. New York: Garland. Vallduví, E., & Vilkuna, M. (1998). On rheme and kontrast. In P. Culicover & L. McNally (Eds.), The Limits of Syntax (pp. 79–108). San Diego, CA: Academic Press. Vergnaud, J. R. (1974). French relative clauses. PhD Dissertation, MIT, Cambridge, MA. Vergnaud, J.-R., & Zubizarreta, M. L. (1992). The definite determiner and the inalienable constructions in French and in English. Linguistic Inquiry, 23, 595–652.
References
Walker, R. (1998). Nasalization, neutral segments, and opacity effects. PhD Dissertation, University of California, Santa Cruz, CA. Walker, R. (2000). Nasalization, neutral segments, and opacity effects. New York: Garland. (PhD Dissertation, University of California, Santa Cruz, CA, 1998). Ward, G. (1988). The semantics and pragmatics of preposing. [Outstanding Dissertations in Linguistics]. New York: Garland. (PhD Dissertation, University of Pennsylvania, 1985). Ward, G., & Birner, B. (1998). Information Status and Noncanonical Word Order in English. Amsterdam: John Benjamins. Wasow, T. (1972). Anaphoric relations in English. PhD Dissertation, MIT, Cambridge, MA. Wasow, T. (1979). Anaphora in Generative Grammar. Gent: Story-Scientia. Watters, J. (1979). Focus in Aghem. In L. M. Hyman (Ed.), Aghem Grammatical Structure. Southern California Occasional Papers in Linguistics 7 (pp. 137–197). Wesche, B. (1992). Symmetric coordination: An alternative theory of phrase structure. PhD Dissertation, Duesseldorf, Germany. Wiese, R. (1993). Tilgung prosodischer Konstituenten. Handout. Wilder, C. (1994). Coordination, ATB and ellipsis: Minimalism and Kayne’s asymmetry hypothesis. In Groninger Arbeiten zur Germanistischen Linguistik, 37, 291–331. Wilder, C. (1995a). Antecedent-containment and ellipsis. In FAS Papers in Linguistics, 4, 132–165. Wilder, C. (1995b). On some properties of ellipsis in coordination. Geneva Generative Papers, 2(2), 23–61. Wilder, C. (1995c). Rightward movement as leftward deletion. In U. Lütz & J. Pafel (Eds.), On Extraction and Extraposition in German (pp. 273–309). Amsterdam: John Benjamins. Wilder, C. (1997a). Phrasal movement in LF: de re readings, binding and ellipsis. In K. Kusumoto (Ed.), Proceedings of NELS 27 (pp. 425–439). Amherst, MA, GLSA. Wilder, C. (1997b). Some properties of ellipsis in coordination. In A. Alexiadou & T. A. Hall (Eds.), Studies in Universal Grammar and Typological Variation [Linguistik Aktuell 13] (pp. 59–107). Amsterdam: John Benjamins. Williams, A. (2000). Adverbial quantification over (interrogative) complements. In Billerey & Lillehaugen (Eds.), Proceedings of WCCFL 19. Somerville, MA: Cascadilla Press. Williams, E. (1977a). Discourse and logical form. Linguistic Inquiry, 8, 101–139. Williams, E. (1977b). On deep and surface anaphora. Linguistic Inquiry, 8, 692–696. Williams, E. (1978). Across-the-board rule application. Linguistic Inquiry, 9, 31–43. Williams, E. (1990). The ATB theory of parasitic gaps. The Linguistic Review, 6, 265–279. Winkler, S. (1997). Ellipsis in English and German: The Phonological Reduction Hypothesis. [Universität Tübingen: Arbeitspapiere des Sonderforschungsbereichs 340: Sprachtheoretische Grundlagen für die Computerlinguistik, Bericht Nr. 121]. Tübingen. Winkler, S., & Göbbel, E. (2002). Focus, p-movement, and the nuclear-stress rule: a view from Germanic and Romance. Linguistics, 40, 1185–1242. Woisetschlager, E. (1981). A note on the autonomy of syntax. Journal of Linguistic Research, 1, 55–70. Wyngærd, G. vanden. (1998). Gapping constituents. Ms., FWO/K.U.Brussel.
References
Wyngærd, G. vanden, & Zwart, J.-W. (1991). Reconstruction and vehicle change. In F. Drijkoningen & A. van Kemenade (Eds.), Linguistics in the Netherlands 1991 [AVTPublications 8] (pp. 151–160). Amsterdam: John Benjamins Zagona, K. (1982). Government and proper government of verbal projections. PhD Dissertation, University of Washington, Seattle, WA. Zagona, K. (1988). Verb Phrase Syntax: A Parametric Account of English and Spanish. Dordrecht: Kluwer. Zhang, N. (2000). Object shift in Mandarin Chinese. Journal of Chinese Linguistics, 28, 201– 246. Zoerner, C. (1995). Coordination: The syntax of andP. PhD Dissertation, University of California, Irvine, CA. Zubizarreta, M. L. (1998). Prosody, Focus and Word Order [Linguistic Inquiry Monograph 33]. Cambridge, MA: MIT Press. Zwart, J.-W. (2001). Syntactic and phonological verb movement. In Syntax.
Name index
A Abbott 147 Abe 238 Abney 52, 337 Akmajian 6 Albert 304 Alexiadou 367, 386 Anagnostopoulou 201 Anderson 11 Aoun 59, 71, 72 Asher 21, 251, 253, 259
B Bach 6 Baker 45, 235 Baltin 80, 84–88, 91–94, 96 Barbosa 56 Barss 179, 184, 190, 195 Beck 281 Beckman 9 Bennis 46, 63 Berman 282 Bhatt 223, 224 Bianchi 205 Birner 51 Bittner 190 Bobaljik 67 Bolinger 45 Booij 147 Borer 20, 228, 230, 236 Borsley 205 Bouchard 76 Brame 212 Brandt 349 Bresnan 6, 45, 206, 220
Brody 247 Büring 11, 13, 115, 199 Busquets 367 C Carlson 19, 205, 206, 208, 214, 215, 223 Castañeda 191 Cecchetto 161 Chao 7 Cheng 321 Chierchia 24, 289, 321, 328, 331, 336 Choi 363 Chomsky 1, 2, 4, 8, 14, 16, 19, 20, 31, 32, 37, 44, 48, 56, 59, 63, 66–68, 71, 74, 81, 88, 96, 156, 157, 159, 161, 191, 195, 205, 206, 208, 219, 227–231, 234, 239, 243 Chung 7, 71, 230, 231, 266, 282, 283, 286, 295, 315, 318 Cinque 38 Clements 247 Collins 178 Cooper 322, 325, 331 Corver 45, 46, 68 Cresti 338 D Dahl 254 Dalrymple 7, 21, 32, 251, 252, 254–257 Daniel 21 Davidson 192 Dayal 321 den Besten 65, 186, 187 den Dikken 115 Diesing 98, 99, 273, 329
Name index
Dimroth 361, 362 DiSciullo 147 Donati 3, 17 Doron 174 Dougherty 6 Dowty 190 Drubig 228, 351, 354, 358 E Elam 9 Elbourne 201 Emonds 169, 228 Enç 313 Engdahl 41 Ernst 190 Evans 288, 316, 322, 325 F Fanselow 115, 187, 199 Farkas 311, 313 Féry 140, 143, 147, 151 Fiengo 21, 30, 38, 93, 221, 222, 251, 252, 254–256, 259, 260, 263, 264 Fillmore 55 Fintel 273, 274, 276 Fodor 313, 315 Fox 19, 51, 59, 66, 73, 189, 206–209, 234, 264 Fraser 358 Freidin 206, 208, 211 Fukui 337 G Gabbay 376 Gasde 235 Geach 372 Giannakidou 334 Gleitman 147 Goldberg 55 Goldsmith 372, 384 Greenbaum 381 Göbbel 3 Gretsch 4, 21, 24, 345, 349, 358, 361, 362 Grice 358
Grimshaw 56, 68 Groenendijk 300 Grosu 49, 50, 205, 206, 215, 223, 224 Guéron 117 H Hackl 205, 224 Haegeman 63 Hagstrom 368 Haider 186, 187, 190, 194, 199 Haïk 30, 33 Halle 49, 60 Hamblin 22, 270, 296 Hankamer 6, 29, 30, 64, 106, 241 Hardt 4, 7, 21, 32, 40, 263 Hartmann 3, 10, 11, 13, 14, 16, 17, 199, 239 Heim 205, 206, 214, 223, 224, 273, 281, 282, 296, 309, 311, 313, 316, 322, 325, 329–331 Hendrik 45 Hendriks 32 Hestvik 34 Hetland 365 Heycock 19, 206, 208 Higginbotham 226, 338 Hirschbühler 264 Hobbs 33, 253 Hoji 322, 329, 331 Hornstein 16, 80, 81, 83, 87–89, 92–95 Horvath 230 Huang 19, 184, 195, 206, 208, 235 I Izvorksi 68 J Jackendoff 2 Jacobs 11, 13 Jacobson 32, 263 Jaeggli 59 Jayaseelan 18, 47, 65, 107, 178, 179, 187 Jelinek 175
Name index
Johnson 4, 13, 18, 20, 47, 65, 178, 179, 187, 199, 227, 229, 237, 239, 241, 263 K Kamp 282, 313, 322 Karttunen 296, 323 Kay 55 Kayne 11, 68, 72, 111, 190, 194, 205 Kehler 33–35, 51 Kempson 65 Kennedy 3, 14–17, 30, 31, 38, 44, 45, 50, 51, 59, 61, 62, 65, 68, 79, 89, 93, 123, 216, 217, 220, 221 Kennelly 247 Kidwai 227, 228, 233 Kim 31, 34–37, 237, 239 Kitagawa 30 Klein 7, 8, 11, 12, 14, 16, 355 Koster 59 Kratzer 190, 191, 195, 200, 273, 274, 278, 325, 330, 331 Krifka 289, 321, 328 Kuno 47, 65, 211, 239 Kurafuji 332 Kuroda 205, 328 L Ladd 137 Ladusaw 370 Lahiri 150, 299 Landman 205, 206, 215, 223, 224 Lapointe 147 Lappin 37 Larson 87, 91, 179, 190, 216 Lasnik 8, 14, 16, 18, 47, 62, 63, 65, 84, 90, 100, 107, 108, 115, 178, 179, 187, 190, 211, 231 Lebeaux 19, 206, 208, 209, 211, 214 Lechner 3, 18, 59, 61, 206, 220–222, 224 Leech 381 Lees 205 Lenerz 314 Levin 47, 65, 178, 263 Levine 129
Li 337 Liberman 63 Lidz 15, 51 Lightfoot 367 Lobeck 7 López 2, 3, 7, 10, 13, 15, 20, 59, 245 Lyle 37 M Manzini 175 Marantz 49, 60 Marcinkiewicz 379 Marcus 9, 39, 343 May 16, 21, 30, 38, 63, 79–83, 87, 91, 93, 98, 101, 115, 216, 218, 221, 222, 251, 252, 254–256, 259, 260, 263, 264 McCawley 11 McCloskey 63, 73 McConnell-Ginet 190 McGinnis 368 Meinunger 365 Merchant 3, 14–17, 23, 30, 31, 44, 45, 50, 51, 61, 62, 65, 66, 68, 123, 207, 211, 217, 263, 289, 301–307, 309, 316–318, 334 Meyer-Viol 376 Miller 47 Milner 68 Milsark 98, 99 Molnár 380 Munn 208, 214, 243 Müller 186 N Neijt 263 Nissenbaum 200, 205, 224 Nunes 248 O Oehrle 237, 240 Oku 262, 322, 324, 328, 353 Ormelius-Sandblom 349 Ortiz de Urbina 230 Otani 175
Name index
P Parsons 191 Partee 294, 329 Pelletier 369, 377 Pereira 370 Pesetsky 18, 177–183, 186, 187, 194, 197–199, 227, 235, 237, 239, 240, 245, 246 Phillips 18, 177–179, 181, 182, 184, 186, 189, 198 Pilch 44 Pinkham 31, 44 Platzack 2, 3 Pollard 75 Postal 6, 16, 38, 40–43, 63, 239 Potsdam 63, 65, 115 Potts 63, 66 Prince 246 Prüst 264
Q Quirk
63
R Reinhart 40 Reis 77, 364 Reuland 40 Reyle 375 Richards 60, 197 Riemsdijk 187 Rizzi 59, 72, 74 Robert 131 Rochemont 246 Romero 4, 7, 21–23, 265, 282, 283, 289, 295, 302–304, 315, 316, 318 Rooth 3, 11, 12, 21, 22, 51, 189, 251–253, 256, 264, 265, 268–271, 277, 285, 295 Rosen 263 Rosengren 364 Ross 6, 7, 9, 10, 14, 20, 30, 31, 44, 50, 229, 237, 239, 263, 264, 282, 303 Rothstein 273, 274
Rullmann 281 Ruys 226 S Safir 209–211, 218, 219, 222 Sag 6, 29, 30, 47, 55, 65–67, 80, 155, 207, 216, 239, 241, 255, 256, 263, 313 Saito 231, 322 Santorini 379 Sauerland 3, 19, 185, 189, 200, 206, 208, 214, 216–218 Savoia 175 Scha 264 Schachter 205, 212 Schwabe 1, 4, 7, 21, 23, 303, 314, 358 Schwarz 263, 266, 267, 272–281, 290, 295 Schwarzschild 3, 16, 22, 23, 133, 134, 146, 265, 271, 285, 295, 301, 305, 306, 308–310, 317 Segal 202 Selkirk 133, 134, 144 Sharvit 212, 224, 225 Shieber 370 Shimada 40 Shopen 6 Siegel 237 Smolensky 246 Sobin 66 Speas 337 Sportiche 172 Sternefeld 212 Stokhof 300 Stowell 73 Svartvik 381 Svenonius 45 Swingle 132, 149 Sybesma 321 Szabolcsi 385 T Takano 206, 208 Tancredi 8, 11, 14, 30, 94, 207, 264 Thiersch 187
Name index
Tiedeman 90 Tomaselli 371 Tomioka 4, 7, 21, 23, 24, 264 Torrego 75, 76 Tran 237 Truckenbrodt 150 Tsai 235 Tuller 227, 228, 233
U Umbach 319 Uriagereka 3
V Vallduví 3 van Oirsouw 147 vanden Wyngaerd 241 Vergnaud 3, 205, 212 Vilkuna 3 von Heusinger 23, 301, 305, 313, 314, 318, 319 von Stechow 11, 22
W Walker 73 Ward 51 Wasow 3, 6, 7, 30, 49 Watters 227 Webelhuth 186, 187 Weinberg 367 Wesche 147 Whitman 175 Wiese 147 Wilder 3, 14, 16, 17, 30, 79, 96, 100, 115, 241 Williams 6–8, 30, 80, 81, 110, 155, 165, 166, 220, 263 Winkler 1, 3, 7, 10, 13, 15, 20, 59, 239 Woisetschlager 45 Z Zagona 7 Zanuttini 63 Zimmermann 150, 319 Zoerner 239 Zubizarreta 3, 230 Zwart 74, 83, 115
Subject index
A across-the-board (ATB) 4, 13, 20, 122, 123, 125, 127 128, 238 adjacency 72, 73, 305 adjunct 35, 42–45, 67, 89, 95–97, 106, 138, 167, 177, 179, 180, 182–184, 187–189, 192–198, 200, 209, 243 agreement 18, 49, 88, 89, 155, 161, 229–231, 327, 335 alternative semantics 3, 4 anaphor 195, 197, 222 Arabic 77 AVOID Focus 22, 271, 272, 284–288, 291, 293–296 B background match 133, 137, 140 binding binding properties 18 binding scope of adjuncts 198 binding theory 14, 29, 30, 39, 40, 128, 129, 181, 234 C Chamorro 230, 231 Chinese 20, 24, 25, 171, 228, 234–236, 324, 327, 333, 335, 336, 348 comparative 3, 7, 14, 15, 20, 44, 48–51, 55–59, 61, 63, 65, 67, 68, 73, 79, 91, 94, 95, 98–100, 102–104, 206, 207, 220–224, 321 Complex NP Constraint 127 Condition A 30, 34, 208 Condition B 30, 35, 39, 40, 44, 222, 223 Condition C 38, 39, 44, 208–211, 213–215, 218, 219, 221–223, 234
contrast 34, 131, 132, 253, 271, 284, 286, 294, 354, 356, 363 copy copy-deletion mechanism 18 copy theory of movement 17, 157, 208, 219 correlate 21–23, 170, 214, 265–268, 271–273, 275, 282–288, 293–295, 302 crossover 38, 39, 41, 43, 44, 130, 131, 218
D deaccenting 94, 137, 138, 140, 142, 146 definite 23, 109, 263, 267, 268, 276, 280, 288, 289, 301, 302, 304, 309–312, 316–318, 323, 325, 328–330, 332, 335 deletion antecedent-contained deletion (ACD) 16, 79, 185 backward deletion (BWD) 16, 79, 105, 110 comparative deletion 3, 14, 44, 48–50, 206, 207, 220, 221–224 copy-deletion mechanism 18 deletion approach 5, 7, 81 deletion theory 2 double headed antecedent contained deletion 216 non-deletion approach 5 phonological deletion 2, 4, 16, 81, 133, 207 relative deletion 19, 206, 207, 220, 221 discourse discourse coherence relations 33 discourse grammar 7, 166
Subject index
discourse pro-drop 24, 321, 324, 327, 335, 336 discourse structure 4, 21, 251, 254, 256, 259 downward-monotone quantifiers 301, 304, 309, 316, 317 Dutch 327
E ECP, see Empty Category Principle e-GIVENness 306–309, 318 economy 3, 18, 22, 136, 146, 166, 177–179, 194, 198, 201, 272 ellipsis background ellipsis 341–344, 355, 363 comparative ellipsis 7 focal ellipsis 24, 341, 343, 359 IP ellipsis 7 NP ellipsis 7, 24, 324, 336 situative ellipsis 343, 344, 358 verb phrase ellipsis (VPE) 80, 106 Empty Category Principle 15, 59 English 2, 4, 13, 15, 16, 19, 20, 32, 37, 48, 49, 51, 56, 58, 62, 64, 65, 69, 72–74, 79, 88, 92, 103, 107, 122, 129, 132, 139, 168–170, 173, 179, 180, 188, 190, 198, 199, 205, 206, 213, 216, 227–230, 232–236, 238, 240, 242–244, 323–326, 328, 332, 335, 336, 347 equidistance 196, 197 existential closure 135, 273, 306, 313, 328, 329 external head 19, 206, 207, 210, 216, 219–222 extraposition 84–88, 91, 92, 94–96, 98, 104, 121–126, 199
F F-marking 133, 135, 136, 138, 139, 141, 142, 144, 146, 305
focus contrastive focus 10, 20, 22, 228, 271, 272, 285, 286, 288, 292, 293, 353–355, 360, 363 focus assignment 133 focus condition 305 focus exponent 144, 341, 342, 353, 357, 364 focus movement 3, 13, 20, 228, 230, 231 focus phrases 230, 360 focus semantic value 189, 268, 269, 270, 280, 284, 285 focus sensitive particle 11 focus structure 10, 17, 51, 121, 131, 137, 146, 342, 350, 354, 355, 363, 364 focus theory 301, 305, 306 317 free focus 22, 350 French 4, 168–170, 173, 228, 230 Full Interpretation 48–50, 158
G gap 9, 25, 33, 37, 38, 40–45, 49, 81, 110, 113, 114, 155–158, 172, 174, 236, 238, 342–347, 350, 351, 353, 354, 356–359, 363, 364 gapping 2, 4–10, 13, 17, 20, 66, 106, 107, 157, 158, 165, 166, 172, 173, 227, 228, 236–242, 244, 263–266, 268, 284, 293–296, 343 German 2, 4, 16, 21, 25, 103–105, 121, 122, 124–126, 129, 143, 187, 188, 263, 264, 327, 346, 347, 349, 360, 362, 363 given 133–142, 144, 145, 239, 274, 292, 295, 301–303, 305, 306, 309–312, 314, 316, 318, 352, 356, 358 GIVENness 3, 16, 133–138, 140, 141, 144, 146, 305, 307, 310 government 59–61, 69–73 Greek 24, 334
Subject index
H Head-Driven Phrase Structure Grammar 55 Hebrew 171, 172 higher-order unification 32 HPSG, see Head-Driven Phrase Structure Grammar
I identy requirement 217 idiom chunk 212, 214, 226 illocution 298, 344, 347–349 incremental merger 181–183, 185, 186, 188, 189 indefinite 23, 90, 98, 100, 128, 132, 185, 215, 266, 267, 273, 275, 277, 283, 286, 288, 289, 294, 301, 303–305, 308–318, 322–326, 328–330, 333, 334 information structure 1, 3, 5, 8, 9, 16, 25, 51, 301, 314, 354 interface syntax-phonology interface 49, 227 syntax-semantics interface 3, 7, 13, 17, 20, 21, 33, 342 internal head 19, 205–207, 216, 218, 220–223 interpretive interpretive module 232, 236 interpretive rules 20, 227, 228, 233, 242, 246 interpretive systems 227, 246 interrogative 21, 22, 137, 245, 263, 267, 268, 270, 282–285, 287, 290, 292, 296, 302, 306, 308, 310, 345–349 Irish 73, 171 island 9, 30, 31, 35, 37–39, 42–44, 63, 123, 127, 158, 165, 166, 186, 225, 303
J Japanese 4, 23, 171, 172, 238, 321–329, 332–336
K Korean 24, 25, 171, 324, 327, 333, 335, 336, 347, 348, 360, 362, 363
L Layered Syntax 180, 181 left branch 14, 31, 44, 45, 50, 69 licensing theory 4 locality 122, 127, 156–158, 165, 166, 168, 174, 186, 188, 198, 223
M matching relative clause 4, 222 merge 17–19, 155–157, 159, 160, 165–168, 177, 178, 181, 182, 229, 230, 236 merge copy 17, 155–157 159, 160, 166, 167 merger 67, 178, 188, 193, 194, 196, 200, 201 Minimal Link Condition (MLC) 166, 168, 203, 231 minimal situation 274–280, 290 missing parasitic gaps 43 modifier 45, 47, 48, 89, 91, 208–211, 334 move 13, 156, 157, 159, 161, 177, 178, 230 movement A-movement 16, 87–90, 92, 93, 95, 239 A’-movement 15, 16, 31, 59, 66, 95, 130, 234 copy theory of movement 17, 157, 208, 219 focus movement 3, 13, 20, 228, 230, 231 remnant movement 179, 187, 198, 199 V-to-I movement 64, 69, 70, 72, 73, 171 wh-movement 3, 15, 17, 19, 55, 56, 59–64, 67, 70, 72, 73, 88, 157, 207–211, 219, 230, 231, 235, 243
Subject index
N negative polarity 123, 128, 129, 179, 264
O Optimality Theory
56
P parallelism 21, 34, 51, 121, 162, 171, 173, 177, 182–184, 189, 192, 194, 251, 264, 315, 323, 325, 351, 354 parameter 20, 24, 227, 228, 233, 236, 242–244, 335, 336 parasitic gaps 14, 29, 37–43 path identity approach 21 periphery 121, 140, 188, 230, 346, 347 PF-deletion 3–8, 10–17, 25, 81, 122, 140, 207, 224 phases 19, 21, 66, 177, 194, 243, 245 phonological reduction hypothesis 14 phrase structure paradoxes (PS-paradox) 18, 179 polarity 129, 271, 342, 351, 357, 364 pragmatic licensing 122, 133, 136, 140, 144, 145, 147 preposition stranding 123, 124 presupposition effect 279, 280, 287, 299 pro 24, 40, 41, 321–337 pro-drop parameter 24 pronoun bound pronoun 256, 321–326, 329, 331, 333 E-type pronoun 268, 322, 324, 326, 331 indefinite pronoun 23, 323–326, 328–330, 333–335 null pronoun 23, 321–326, 337 paycheck pronoun 327, 339 pronoun of laziness 23, 323–330, 332, 333, 335, 337 pseudogapping 14, 16, 18, 47, 48, 50, 64–68, 79, 81, 84, 100, 105–110, 112, 113, 178, 179, 181, 187, 198, 199
Q quantifier raising (QR) 79, 83, 216, 217, 224, 225 question 1–4, 8, 9, 12–14, 16, 17, 22, 24, 25, 29, 36, 51, 56, 59, 69, 71, 72, 74, 79, 91, 105–107, 110, 130, 136–138, 140, 142–146, 159, 171, 177, 194, 200, 205, 206, 213, 214, 223, 224, 229, 232, 234, 236, 238, 240, 245, 270, 277, 280, 281, 283–288, 290–292, 296, 308–311, 335, 341–345, 347–350, 352–355, 358, 359, 363, 364 R R-expression 129, 208–211, 213, 214, 221, 222 raising Raising Hypothesis 19, 206 raising relative clauses 220 raising structure 207, 213, 223 reconstruction 19, 130, 155, 157, 168, 177, 179, 189, 192, 194–198, 205, 206, 208–212, 214, 225, 234, 344 reduced conditional 267, 272, 274, 276 reflexive 30, 40, 195, 326 relative relative clause 19, 33, 83–86, 90–94, 96, 98, 100, 105, 111–114, 205–221, 223–225, 254, 306, 307 relative deletion 19, 221, 222 remnant remnant movement 20, 177, 188, 189, 198 remnant topicalization 186 right node raising 3, 16, 110, 121, 122, 129, 139, 238 Right Roof Constraint 86, 92, 104, 110, 122, 126 Russian 77 S scope 1, 13, 16, 18, 23, 66, 79, 85–103, 105, 110, 122, 128, 129, 146, 179, 184, 185, 188, 189, 192–195, 198,
Subject index
212, 214, 215, 225, 237, 240, 241, 254, 264, 273, 301, 304, 309, 313–318, 322, 323, 325, 346 sentence grammar 6, 7, 166 sentence type 302, 349 sloppy identity 21, 251, 252, 254, 262 sloppy reading 252, 253, 256, 333 sluicing 4, 6, 15, 21–23, 51, 172, 173, 263–268, 272, 281–283, 285, 286, 288–290, 293–296, 301–303, 305–318 Spanish 230, 245, 273, 281, 327, 335 specificity 23, 157–160, 301, 305, 309, 313, 314, 317, 318 strict reading 30, 32, 35, 255, 257 subcomparatives 370 subject-auxiliary inversion (SAI) 15, 55, 74 T topic 11, 12, 51, 172, 227, 301, 321, 331, 356, 359, 362 topicalization 18, 167, 179, 185–187, 196, 234, 241 type-shifting 328, 329 U Universal Grammar (UG) 244, 376 V variable bound variable 33, 135, 209, 212, 213, 224, 322–326, 330, 333
referential variable 322, 325 unselectively bound variable 326 variable binding 179, 184, 192, 199, 208, 213, 215, 224, 323, 354 vehicle change 19, 38, 206, 221–223 VP adjunct 67 VP-deletion 3, 10–17, 25, 81, 122, 140, 207, 224 VP-ellipsis 6–7, 14–16, 21, 55–68, 73–75, 79, 90, 102, 105–107, 116, 118, 147, 156–158, 166, 168, 171, 177–178, 181–183, 192–193, 201, 247, 253, 263–266, 268, 293–296, 395 VPF, see VP-fronting VP-fronting 3, 18, 177–179, 181, 182, 185, 188, 195 W Western Bade 20, 228–230, 232–234, 236, 240, 242, 244 Western Germanic 179, 186 whwh-determiner 209, 285, 290, 295 wh-element 156–158, 285, 345, 347 wh-island constraint 127 wh-marker 347 wh-movement 3, 15, 17, 19, 55, 56, 59–64, 67, 70, 72, 73, 88, 157, 207–211, 219, 230, 231, 235, 243 wh-question 11, 58, 59, 134, 135, 137, 138, 143, 146, 156, 310, 318 wh-trace 9, 59, 72, 130, 219, 221, 282
In the series LINGUISTIK AKTUELL/LINGUISTICS TODAY (LA) the following titles have been published thus far, or are scheduled for publication: 1. KLAPPENBACH, Ruth (1911-1977): Studien zur Modernen Deutschen Lexikographie. Auswahl aus den Lexikographischen Arbeiten von Ruth Klappenbach, erweitert um drei Beiträge von Helene Malige-Klappenbach. 1980. 2. EHLICH, Konrad & Jochen REHBEIN: Augenkommunikation. Methodenreflexion und Beispielanalyse. 1982. 3. ABRAHAM, Werner (ed.): On the Formal Syntax of the Westgermania. Papers from the 3rd Groningen Grammar Talks (3e Groninger Grammatikgespräche), Groningen, January 1981. 1983. 4. ABRAHAM, Werner & Sjaak De MEIJ (eds): Topic, Focus and Configurationality. Papers from the 6th Groningen Grammar Talks, Groningen, 1984. 1986. 5. GREWENDORF, Günther and Wolfgang STERNEFELD (eds): Scrambling and Barriers. 1990. 6. BHATT, Christa, Elisabeth LÖBEL and Claudia SCHMIDT (eds): Syntactic Phrase Structure Phenomena in Noun Phrases and Sentences. 1989. 7. ÅFARLI, Tor A.: The Syntax of Norwegian Passive Constructions. 1992. 8. FANSELOW, Gisbert (ed.): The Parametrization of Universal Grammar. 1993. 9. GELDEREN, Elly van: The Rise of Functional Categories. 1993. 10. CINQUE, Guglielmo and Guiliana GIUSTI (eds): Advances in Roumanian Linguistics. 1995. 11. LUTZ, Uli and Jürgen PAFEL (eds): On Extraction and Extraposition in German. 1995. 12. ABRAHAM, W., S. EPSTEIN, H. THRÁINSSON and C.J.W. ZWART (eds): Minimal Ideas. Linguistic studies in the minimalist framework. 1996. 13. ALEXIADOU Artemis and T. Alan HALL (eds): Studies on Universal Grammar and Typological Variation. 1997. 14. ANAGNOSTOPOULOU, Elena, Henk VAN RIEMSDIJK and Frans ZWARTS (eds): Materials on Left Dislocation. 1997. 15. ROHRBACHER, Bernhard Wolfgang: Morphology-Driven Syntax. A theory of V to I raising and pro-drop. 1999. 16. LIU, FENG-HSI: Scope and Specificity. 1997. 17. BEERMAN, Dorothee, David LEBLANC and Henk van RIEMSDIJK (eds): Rightward Movement. 1997. 18. ALEXIADOU, Artemis: Adverb Placement. A case study in antisymmetric syntax. 1997. 19. JOSEFSSON, Gunlög: Minimal Words in a Minimal Syntax. Word formation in Swedish. 1998. 20. LAENZLINGER, Christopher: Comparative Studies in Word Order Variation. Adverbs, pronouns, and clause structure in Romance and Germanic. 1998. 21. KLEIN, Henny: Adverbs of Degree in Dutch and Related Languages. 1998. 22. ALEXIADOU, Artemis and Chris WILDER (eds): Possessors, Predicates and Movement in the Determiner Phrase. 1998. 23. GIANNAKIDOU, Anastasia: Polarity Sensitivity as (Non)Veridical Dependency. 1998. 24. REBUSCHI, Georges and Laurice TULLER (eds): The Grammar of Focus. 1999. 25. FELSER, Claudia: Verbal Complement Clauses. A minimalist study of direct perception constructions. 1999.
26. ACKEMA, Peter: Issues in Morphosyntax. 1999. ° 27. RUZICKA, Rudolf: Control in Grammar and Pragmatics. A cross-linguistic study. 1999. 28. HERMANS, Ben and Marc van OOSTENDORP (eds): The Derivational Residue in Phonological Optimality Theory. 1999. 29. MIYAMOTO, Tadao: The Light Verb Construction in Japanese. The role of the verbal noun. 1999. 30. BEUKEMA, Frits and Marcel den DIKKEN (eds): Clitic Phenomena in European Languages. 2000. 31. SVENONIUS, Peter (ed.): The Derivation of VO and OV. 2000. 32. ALEXIADOU, Artemis, Paul LAW, André MEINUNGER and Chris WILDER (eds): The Syntax of Relative Clauses. 2000. 33. PUSKÁS, Genoveva: Word Order in Hungarian. The syntax of È-positions. 2000. 34. REULAND, Eric (ed.): Arguments and Case. Explaining Burzio’s Generalization. 2000. 35. HRÓARSDÓTTIR, Thorbjörg. Word Order Change in Icelandic. From OV to VO. 2000. 36. GERLACH, Birgit and Janet GRIJZENHOUT (eds): Clitics in Phonology, Morphology and Syntax. 2000. 37. LUTZ, Uli, Gereon MÜLLER and Arnim von STECHOW (eds): Wh-Scope Marking. 2000. 38. MEINUNGER, André: Syntactic Aspects of Topic and Comment. 2000. 39. GELDEREN, Elly van: A History of English Reflexive Pronouns. Person, ‘‘Self’’, and Interpretability. 2000. 40. HOEKSEMA, Jack, Hotze RULLMANN, Victor SANCHEZ-VALENCIA and Ton van der WOUDEN (eds): Perspectives on Negation and Polarity Items. 2001. 41. ZELLER, Jochen : Particle Verbs and Local Domains. 2001. 42. ALEXIADOU, Artemis : Functional Structure in Nominals. Nominalization and ergativity. 2001. 43. FEATHERSTON, Sam: Empty Categories in Sentence Processing. 2001. 44. TAYLAN, Eser E. (ed.): The Verb in Turkish. 2002. 45. ABRAHAM, Werner and C. Jan-Wouter ZWART (eds): Issues in Formal German(ic) Typology. 2002. 46. PANAGIOTIDIS, Phoevos: Pronouns, Clitics and Empty Nouns. ‘Pronominality’ and licensing in syntax. 2002. 47. BARBIERS, Sjef, Frits BEUKEMA and Wim van der WURFF (eds): Modality and its Interaction with the Verbal System. 2002. 48. ALEXIADOU, Artemis, Elena ANAGNOSTOPOULOU, Sjef BARBIERS and HansMartin GAERTNER (eds): Dimensions of Movement. From features to remnants. 2002 49. ALEXIADOU, Artemis (ed.): Theoretical Approaches to Universals. 2002. 50. STEINBACH, Markus: Middle Voice. A comparative study in the syntax-semantics interface of German. 2002. 51. GERLACH, Birgit: Clitics between Syntax and Lexicon. 2002. 52. SIMON, Horst J. and Heike WIESE (eds): Pronouns – Grammar and Representation. 2002.
53. ZWART, C. Jan-Wouter and Werner ABRAHAM (eds): Studies in Comparative Germanic Syntax. Proceedings from the 15th Workshop on Comparative Germanic Syntax (Groningen, May 26-27, 2000)(Workshop). 2002. 54. BAPTISTA, Marlyse: The Syntax of Cape Verdean Creole. The Sotavento varieties. 2002. 55. COENE, M. and Yves D'HULST (eds): From NP to DP. Volume 1: The syntax and semantics of noun phrases. n.y.p. 56. COENE, M. and Yves D'HULST (eds.): From NP to DP. Volume 2: The expression of possession in noun phrases. n.y.p. 57. DI SCIULLO, Anna-Maria (ed.): Asymmetry in Grammar. Volume 1: Syntax and semantics. n.y.p. 58. DI SCIULLO, Anna-Maria (ed.): Asymmetry in Grammar. Volume 2: Morphology, phonology, acquisition. n.y.p. 59. DEHÉ, Nicole: Particle Verbs in English. Syntax, information structure and intonation. 2002. 60. TRIPS, Carola: From OV to VO in Early Middle English. 2002. 61. SCHWABE, Kerstin and Susanne WINKLER (eds.): The Interfaces. Deriving and interpreting omitted structures. 2003. 62. CARNIE, Andrew, Heidi HARLEY and Mary WILLIE (eds.): Formal Approaches to Function in Grammar. In honor of Eloise Jelinek. 2003.