Phases of Interpretation

Phases of Interpretation

≥

Studies in Generative Grammar 91

Editors

Henk van Riemsdijk Harry van der Hulst Jan Koster

Mouton de Gruyter Berlin · New York

Phases of Interpretation

Edited by

Mara Frascarelli

Mouton de Gruyter Berlin · New York

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

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

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

Library of Congress Cataloging-in-Publication Data Phases of interpretation / edited by Mara Frascarelli. p. cm. ⫺ (Studies in generative grammar ; 91) Includes bibliographical references and index. ISBN-13: 978-3-11-018684-0 (cloth : alk. paper) ISBN-10: 3-11-018684-5 (cloth : alk. paper) 1. Grammar, Comparative and general ⫺ Syntax. Mara. P295.P45 2006 415⫺dc22

I. Frascarelli,

2006018390

Bibliographic information published by Die Deutsche Bibliothek Die Deutsche Bibliothek lists this publication in the Deutsche Nationalbibliografie; detailed bibliographic data is available in the Internet at ⬍http://dnb.ddb.de⬎.

ISBN-13: 978-3-11-018684-0 ISBN-10: 3-11-018684-5 ISSN 0167-4331 쑔 Copyright 2006 by Walter de Gruyter GmbH & Co. KG, D-10785 Berlin. All rights reserved, including those of translation into foreign languages. No part of this book may be reproduced in any form or by any means, electronic or mechanical, including photocopy, recording, or any information storage and retrieval system, without permission in writing from the publisher. Cover design: Christopher Schneider, Berlin. Printed in Germany.

Contents

List of contributors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

vii

Introduction Phases and interpretation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Mara Frascarelli

1

Chapter 1 – Interpretation and structural conditions Grafts follow from merge . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17 Henk van Riemsdijk An interpretive effect of head movement . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45 Winfried Lechner When we do that and when we don’t: A contrastive analysis of VP ellipsis and VP anaphora . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71 Carlo Cecchetto and Orin Percus Chapter 2 – Interpretation in the DP-phase HAVE = BE + PREP (osition): New evidence for the preposition incorporation analysis of clausal possession . . . . . . . . . . . . . . . . . . . . . . . . . . 107 Chris H. Reintges and Anikó Lipták

Northern Norwegian degree questions and the syntax of measurement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133 Peter Svenonius and Christopher Kennedy Parallels in clausal and nominal periphery . . . . . . . . . . . . . . . . . . . . . . . . . . . 163 Giuliana Giusti

vi

Contents

Chapter 3 – Functional projections in the vP-phase The properties of anticausatives crosslingustically . . . . . . . . . . . . . . . . . . . . 187 Artemis Alexiadou, Elena Anagnostopoulou and Florian Schäfer Number agreement and event pluralization: A case study. . . . . . . . . . . . . 213 Valentina Bianchi The Phase Condition and cyclic Spell-out: Evidence from VP-topicalization. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 237 Roland Hinterhölzl Parallel phases: a study on the high and low left periphery of Old Italian . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 261 Cecilia Poletto Chapter 4 – The CP-phase and Subject Licensing Obviation in subjunctive argument clauses and the first-personal interpretation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 295 Francesco Costantini Who is lui? Reference of Italian overt and covert subject pronouns . . . . 321 Alessandra Trecci Satisfying the Subject Criterion by a non subject: English Locative Inversion and Heavy NP Shift . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 341 Luigi Rizzi and Ur Shlonsky Informational focus in Sicilian and the left periphery. . . . . . . . . . . . . . . . . 363 Silvio Cruschina Subject index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 387 Language index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 390

List of contributors

Artemis Alexiadou

Francesco Costantini

Institut für Linguistik – Anglistik Universität Stuttgart Keplerstrasse 17 70174 Stuttgart Germany

Università Ca’ Foscari – Venezia Dipartimento di Scienze del Linguaggio Ca’ Bembo – Dorsoduro 1075 30123 Venezia Italy

[email protected]

[email protected]

Helena Anagnostopoulou

Silvio Cruschina

University of Crete Department of Philology 74 100 – Rethymno Greece

University of Cambridge Department of Italian St. John’s College CB2 1TP – Cambridge UK

[email protected]

[email protected] Valentina Bianchi

Mara Frascarelli

Università degli studi di Siena Centro Interdipartimentale di Studi Cognitivi sul Linguaggio P.zza S.Francesco 8 53100 Siena Italy

Università degli Studi Roma Tre Dipartimento di Linguistica Via Ostiense 236 00146 Roma Italy [email protected]

[email protected] Carlo Cecchetto

Giuliana Giusti

Università degli Studi di Milano – Bicocca, Dipartimento di Psicologia Piazza dell'Ateneo Nuovo 1 20126 Milano Italy

Università Ca’ Foscari – Venezia Dipartimento di Scienze del Linguaggio Ca’ Bembo – Dorsoduro 1075 30123 Venezia Italy

[email protected]

[email protected]

viii List of contributors Roland Hinterhölzl

Orin Percus

Institut für deutsche Sprache und Linguistik Humboldt Universität zu Berlin Unter den Linden 6 10099 Berlin Germany

Département de Sciences du Langage UFR Lettres et Langages Université de Nantes Chemin de la Censive du Tertre BP 81227 44312 Nantes Cedex 3 France

[email protected]

[email protected] Chris Kennedy

Cecilia Poletto

Department of Linguistics University of Chicago 1010 E 59th St Chicago IL 60637 USA

Università Ca’ Foscari – Venezia Dipartimento di Scienze del Linguaggio Ca’ Bembo – Dorsoduro 1075 30123 Venezia Italy

[email protected]

[email protected]

Winfried Lechner

Chris Reintges

Institut für Linguistik – Anglistik Universität Stuttgart Keplerstrasse 17 70174 Stuttgart Germany

Universiteit Leiden, Faculteit Letteren C.I.F/NINO/Hebreeuws Matthias de Vrieshof 4 2311 BZ Leiden The Netherlands

[email protected]

[email protected]

Anikó Lipták

Luigi Rizzi

Universiteit Leiden, Faculteit Letteren Franse Taal- en Letterkunde van Wijkplaats 2 2311 BX Leiden The Netherlands

Università degli studi di Siena Centro Interdipartimentale di Studi Cognitivi sul Linguaggio P.zza S.Francesco 8 53100 Siena Italy

[email protected]

[email protected]

List of contributors Florian Schäfer

Alessandra Trecci

Institut für Linguistik – Anglistik Universität Stuttgart Keplerstrasse 17 70174 Stuttgart Germany

Università degli Studi Roma Tre Dipartimento di Linguistica Via Ostiense 236 00146 Roma Italy

[email protected]

[email protected]

Ur Shlonsky

Henk van Riemsdijk

Département de Linguistique Faculté des Lettres Université de Genève Rue de Candolle 2 1211 Genève 4 Suisse

Faculteit Communicatie en Cultuur Grammaticamodellen Universiteit van Tilburg P.O.Box 90153 5000 LE Tilburg The Netherlands

[email protected]

[email protected]

Peter Svenonius Universitetet i Tromsø Humanistiske Fakultet CASTL 9037 Tromsø Norway [email protected]

ix

Phases and interpretation Mara Frascarelli

1. Introduction In the last decade syntactic research within the Minimalist Program was focussed mainly on three basic issues: a) How the set of minimally necessary syntactic operations is to be characterized (including questions about the exact nature of Copy and Merge, the status of remnant movement, the role of head movement in the grammar), b) How the set of minimally necessary functional heads can be defined that determine the built-up and the interpretation of syntactic objects, and c) How these syntactic operations and objects interact with principles and requirements that are thought to hold at the two interfaces. The concept of “phase” and the notion that derivations are built-up cyclically interspersed with phases of interpretation on both the PF- and the LFside plays an important role in defining these crucial issues. The concept of phase has also implications for the research on the functional make-up of syntactic objects, as it is undertaken by the “cartographic approach”, implying that functional projections not only apply in a (universally given) hierarchy but split up in various phases pertaining to the head they are related to. This volume provides a contribution to the ongoing discussion, investigating these issues in a variety of different languages (among others, Coptic, Dutch, English, German, Greek, Hungarian, Italian, Norwegian and West Flemish), combining the analysis of empirical data with the theoretical insights of the last years.

2. Phases in the Minimalist Program Roughly ten years after the Pisa Lectures and the publication of the Government and Binding theory (Chomsky 1981) – which imposed modularization as a crucial device to account for language acquisition and grammatical properties – Chomsky (1993, 1995) proposed a further development of the theory with the so-called “Minimalist Program”. In the constant effort to

2 Mara Frascarelli discover concepts of computational efficiency that carry us “beyond explanatory adequacy”, the model of grammar is reduced to the interface levels, since they are the only “conceptually necessary” ones, consisting of legitimate objects that can receive an interpretation. In this model language is claimed to be an “optimal object”, based on simple and economical principles and derivations are restricted to four types of operations, namely Select, Merge, Agree and Move.1 The (costly) operation Move has been subject to particular scrutiny in the last ten years. Under the pressure of Greed, it is initially reduced to feature-checking requirements: “Move raises
only if morphological properties of
itself would not otherwise be satisfied in the derivation” (Chomsky 1995: 261). In subsequent works, however, movement is also licensed by the necessity to meet interface conditions (Reinhart 1998) and Chomsky himself proposes that “well-designed languages have a dislocation property”, connected with discourse requirements in the C-domain (Rizzi’s 1997 “left periphery”) and optional EPP features (cf. Chomsky 2000, 2002). Derivations triggered by information-structural features were thus eventually included in the Minimalist Program. With Derivation by Phase (1999/2001) a further reduction of the theoretical apparatus is proposed and the notion of phase is defined as the source of locality and the domain for cyclic interpretation. The motivations provided by Chomsky for this assumption are primarily conceptual and can be summarised as follows. In the minimalist model the access to the lexicon is a one-time selection of a Lexical Array (LA). To reduce the computational burden, it is assumed that the initial LA enters the derivations in different steps. The syntactic derivation thus proceeds in incremental chunks, with each phase built from a separate lexical subarray that satisfies the principle of full interpretation. In each step, the derivation must exhaust one subarray (that is put in active memory) before returning to LA and extract another subarray (up to the moment in which LA is totally exhausted). For minimal computation, information will be forgotten as soon as it is transferred to the interfaces: “the computation will not have to look back at earlier phases as it proceeds and cyclicity is preserved in a very strong sense” (Chomsky 2005: 9). Derivation is therefore submitted to phases: once these domains have been built, they are shipped immediately to the interfaces for interpretation. Besides these conceptual reasons, phases immediately appear advantageous on empirical grounds. Take the defective nature of those constituents that exclude extraction (Ross’ 1967 “islands”). In the Government and

Phases and interpretation

3

Binding framework, in which different operations were submitted to specific conditions, Subjacency would account for the fact that movement cannot pass more than one “bounding node”, where bounding nodes are NP and IP. This could account for some (well known) asymmetries like the following: (1)

a. [IP John claimed [CP that [IP Bill hit that man]]] b. [CP whom did [IP John claim [CP t’whom that [IP Bill hit twhom ]]]]?

(2)

a. [IP John made [NP the claim [CP that [IP Bill hit that man]]]] b. *[CP whom did [IP John make [NP the claim [CP t’whom that [IP Bill hit twhom ]]]]]?

Subjacency, however, cannot explain the difference between subject and object extraction (cf. (3)–(4) below). This contrast required reference to another (major) condition, namely the Empty Category Principle (ECP): objects traces are always “properly governed” (and marginality in (3) is due to Subjacency), while subjects move from a non L-marked position: (3)

a. ?What do you wonder when John bought? b. [CP what do [IP you wonder [CP when [IP John bought twhat ]]]]

(4)

a. *Who do you wonder when bought this book? b. [CP who do [IP you wonder [CP when [IP twho bought this book]]]]

With Barriers Chomsky (1986) provides a unified treatment of islands, avoiding the necessity of “combined conditions” to block derivations. However, the definition of a barrier itself requires a combination of notions and structural conditions (e.g., dominance, government, L-marking) and could not be satisfactorily assumed as a “primitive” of language acquisition. Moreover, the defective behavior of IP still requires an ad hoc stipulation (namely, its “inherited” blocking quality). In this respect, an important step further was provided by Rizzi’s (1990) seminal work on Relativized Minimality, which inspired the formulation of minimal constraints (like the Minimal Link Condition) and provided a principled explanation for many crucial asymmetries. A unified account of syntactic derivation was still missing, though. Through the notion of phase, Chomsky discards specific conditions and derives syntactic derivation from a unique, comprehensive and powerful device, namely the Phase Impenetrability Condition (PIC) that determines interface interpretation:

4 Mara Frascarelli (5)

Phase Impenetrability Condition (Chomsky 2000: 108) In phase
with head H, the domain of H [= complement of H] is not accessible to operations outside
, but only H and its edge [= H plus any/all of its specifiers]

The PIC capitalizes on the observation that some categories at a certain point of the derivation stop being accessible to syntactic operations because they have been transferred to the phonological and semantic components. However, the edge and the head of a phase remain accessible at the next higher phase. Extraction from a phase is thus permitted by the presence of an edge-feature (EF) that triggers Move (i.e., Internal Merge,2 under the “copy theory” of movement) of a constituent. At this point the question of what counts as a phase is crucial. Chomsky (2000, 1999/2001, 2004) takes vP3 and CP to be phases, since these syntactic objects have a degree of independence both on the phonetic and on the meaning side. DP is also (tentatively) acknowledged as a phase in later works (see discussion below). NP and IP, on the other hand, are not phases of interpretation. This independent property in combination with the PIC immediately derives their islandhood: since they are not endowed with an edge position, extraction from these constituent is blocked. The PIC can therefore be used to explain barriers, subjacency effects as well as other grammatical constraints (e.g., remnant movement constraints) in a unified and principled way. The possibility of subject extraction in null subject languages, for instance, can be reduced to the possibility of moving the subject from a lower position (i.e., the edge of the vP phase), with the subject position filled by an expletive pro which satisfies the “Subject Criterion” (Rizzi 2003; Rizzi and Shlonsky this volume). The notion of phase (applied to Kayne’s 1994 proposal) can also account for asymmetries on PP-extraction as is shown in (6a–b) below. Possessive PPs are copied in the Specifier position of the DP, which provides an escape hatch for extraction, while this possibility is excluded for other PPs that are “trapped” within the DP-phase. (6)

a. di chi hai comprato un libro t ? ‘of whom did you buy a book?’ b. *di cosa hai comprato un tavolo t ? ‘of what (material) did you buy a table?’

Phases and interpretation

5

Given this model, complements and modifiers appear as unaccessible elements, while heads and specifiers of phases permit Internal Merge. However, material inside the internal domain of a phase can bypass the effects of the PIC by moving into the edge of the relevant phase, via “(indirect) feature driven movement”. This movement, that Chomsky (1999/ 2001) defines as a generalized EPP, is understood either as a feature in the technical sense or simply as a grammatical property of some sort. This is a crucial point: since Specifiers can be iterated in Chomsky’s approach, the question is how to limit the power of a potentially infinite recursive projection and avoid the drawback of an “adjunction pattern”. Linearization thus represents a major issue for current research and different solutions have been proposed in the recent literature.4 As said above, the operation Move – which was initially considered an “imperfection” of language – assumes an outstanding importance as a consequence of the phase theory. Contrary to previous assumptions, the displacement property is considered to be “simply a fact” of natural language: expressions are commonly pronounced in one place and interpreted in others as well, hence “I[nternal] M[erge] is as free as E[xternal] M[erge]; it can only be blocked by stipulation” (Chomsky 2005: 7). In particular, movement is a copying operation that does not (indeed, cannot) eliminate a LI in its merge position (that is part of a different phase) but simply copies the relevant syntactic object in the edge position of a higher phase. Hence, all the copies are carried to the semantic interface, eliminating lowering operations of reconstruction and making processing easier. Of course, on the phonetic-side minimization of computation calls for erasure of all but one copy, by an independent PF-principle (something like “avoid redundant information”). This approach allows for a unified explanation of movement, QR interpretation and ellipsis (cf. Cecchetto and Percus this volume), as is illustrated below: (7)

a. <what> did you read <what>? b. what did you read <what> about Chomsky?

(8)

a. John bought and read the newspaper b. <John> bought and <John> read

Syntactic derivation thus involves Merge and Re-merge in incremental phases, with the consequence that structure is built from bottom to top. This also means that mapping between syntax and the interface levels (Spell-out) takes place at various points in the course of derivation and the

6 Mara Frascarelli global properties of phonology are assumed to be superimposed on the outcome of a cyclic operation. A further question thus arises concerning the properties of phases: How can a phase be defined in a principled way? Chomsky assumes that v and C are phases because they are -complete and “isolable” at the interfaces. While for CPs phonetic independence is obvious, since these objects are propositional units, evidence for the phase status of vPs is provided by pseudoclefting (9), fronting (10) and response fragments (11) (from Chomsky 2000: 16): (9)

what John did was [vP insult the dean]

(10) John said he would insult the dean and [vP insult the dean] he did (11) [vP me insult the dean]?! On the LF-side, CP is the space in which Tense and Force of the sentence are determined,5 and vP is the space in which -roles are assigned. This means that passive and unaccusative vPs cannot be considered as phases proper, since they lack the external -role. Although incomplete, on the other hand, they are still phonetically independent (cf. Legate 2002) and it seems desirable to maintain them as phases of interpretation as well. Chomsky (1999/2001) proposes that phases can be strong or weak (this is exactly the case of unaccusative VPs), with the application of the PIC restricted to strong phases. As for DPs, their similarities with CPs (cf. Szabolsci 1989 and much related works) lead to the assumption that DP is a phase as well (Chomsky 2005: 9), though this question is still debated in the literature. Evidence in this direction is provided in Svenonius (2004), Legate (2002) and Hiraiwa (2005). Matushansky (to appear), however, observes that Legate’s diagnostics for phase-hood support a treatment of DPs as phases at the PF interface, but crucially fail on the semantic side. The interpretation of discourse categories also shows that DPs are not fully independent LF-phases: Giusti (this volume) shows that the interpretation of information Focus must be checked against the relevant feature in the C-domain – even if the DPphase is completed.6 In fact, as is pointed out in Boeckx and Grohmann (2004: 7), interfaces appear to need full representations for specific processes. For example, LF needs full clauses to see multiply spelled out chunks for pronominal binding7 and PF quite possibly needs full clauses to determine intonational

Phases and interpretation

7

patterns (mutatis mutandis, a DP that is given as a fragment answer clearly shows the same intonational contour as a full clause). In the same spirit, Bianchi (this volume) suggests that some semantic features of the vP should remain visible at the LF-interface. These considerations apparently lead to the definition of a major role for the CP-phase: full interpretation cannot do without Tense and Force specifications, that are encoded in the left periphery of the sentence, and the intonation of fragments is a direct reflex of clausal Force.8 “Signals” of the crucial importance attributed to the C-domain (with respect to vP and, possibly, DP) can be found in different parts of Chomsky’s writings. Chomsky (2001) suggests that movement can violate minimality as long as violation is repaired at the phase-level at which locality is checked and, crucially, locality is checked at the CP-phase. Chomsky (2005) seems to proceed a step further in this direction and defines the phase head C as the source of phi-features (Case included) and Tense, that are transferred to T through the operation Agree. The interpretation of unaccusative VPs is also strictly connected with the C-domain: since VPs are not complete, they cannot proceed to Spell-out. Hence, Case assignment must “wait for” the CP-phase to be complete and, at that point, only Nominative Case from the T head is available.9 In the cartographic approach (Rizzi 1997), investigation on the C-domain plays a major role. Indeed, there are many points of convergence between the Minimalist Program and this approach: Chomsky himself acknowledges that C is “a shorthand of the region that Rizzi calls the ‘left periphery’, possibly involving feature spread from fewer functional heads (maybe only one)” (Chomsky 2005: 9). Accordingly, it is assumed that EPP features may be assigned to a functional head to ensure the checking of scopal or discourse related properties that are relevant for interpretation (Chomsky 2002, 2004). Authors have therefore explored the structure of the C-domain, discussed its phase status and defined its edges and functional properties. To conclude, the notion of phase is no doubt an important step forward in the Minimalist Program: it derives the effects of diverse phenomena, provides a unification of different Government-Binding modules and allows a substantial reduction of the theoretical burden. However, the identity and exact properties of phases is still an open issue for research and further elaboration. The papers collected in this volume provide a contribution in this direction, discussing data and providing original proposals for some of the major questions addressed in this introduction.

8 Mara Frascarelli 3. The structure of the book This book includes fourteen articles, grouped into four chapters. The first Chapter is dedicated to the discussion of structural conditions in the phase theory. Particular attention is paid to linearization after Merge and to conditions on movement, ellipsis and deletion in the copy theory. The Chapters 2, 3 and 4 are dedicated to discussing the properties and interpretation of the three phases: the DP, the vP and the CP, respectively. The most important issues discussed also include the functional structure of weak phases (IP and NP), the status of optional EPP features (employed for checking discourse requirements) and the licensing of subjects. Based on empirical evidence from different languages, the authors critically discuss previous analyses and endorse the notion of phase as the syntactic object that is transferred to the interfaces for interpretation. At the same time, they propose new insight in the structural definition of phases, their edges, properties and peripheries, pushing a step further the current research in the Minimalist Program. In the first Chapter, Henk van Riemsdijk’s contribution addresses linearization – a crucial issue for syntactic derivation – and shows that the concept of “grafts” (presented in previous works) can be given a natural interpretation in a framework that makes use of Internal and External Merge. The author argues that when all the logical combinations of Internal Merge and External Merge are taken into account, grafts must exist and are not, as hitherto believed, an odd problem for phrase structure. A number of constructions are therefore expected to have their properties. Interpretation in movement is, on the other hand, the main concern of Lechner’s paper. In a framework that is extensively dominated by (remnant) movement of XPcategories, the author offers arguments from the semantic side that certain instances of head movement must be computed in the syntactic component. The search for evidence also provides new support to particular analyses of three phenomena: first, the interaction of PPI licensing and negation is seen to provide a new argument for generating modals low and moving them to a higher head position in overt syntax; second, the discussion yields new diagnostics from NPIs for identifying the position in which subject NPs are submitted to interpretation; third, the deliberations result in a novel way for expressing restrictions on scope diminishment. Finally, Cecchetto and Percus compare the properties of VP anaphora with the properties of VP ellipsis and focus on facts showing that VP ellipsis admits a kind of analysis that VP anaphora does not. The authors thus argue that a “semantic copying”

Phases and interpretation

9

analysis is the right one for the cases of VP anaphora considered, while a “PF deletion” approach is correct for (most cases of) VP ellipsis. The second Chapter is dedicated to the DP-phase, with particular attention to the connection between DP and CP. In Reintges and Lipták’s contribution, the parallelism between clausal and nominal ph(r)ases is dealt with through the interpretation of possession. Basing on original evidence (from Coptic), the authors explore a configurational analysis and derive the transparent semantic relation from a shared predicational structure (viz. a small clause). The relevance of the phase notion is shown by cyclicity in the derivation of possessive HAVE sentences: sentential possession is argued to start from DP-internal possession, thus pushing the mirror analysis a step further, in a derivative perspective. A connection between the DPphase and the C-domain is also proposed in Svenonius and Kennedy’s work: null degree questions in Northern Norwegian are derived by moving a phonological null operator from a degree argument position inside a gradable predicate to Spec,CP. The paper also provides interesting data about the prosodic interpretation of degree questions: since the positive value is what can be presupposed, the property (the degree of which is questioned) is totally destressed (like a right-dislocated constituent). The connection between DP and information structure is, finally, the concern of Giusti’s investigation in the left periphery of the D-domain. The author proposes that the crucial difference between clauses and noun phrases is the lack of Tense in the latter. This proposal is in line with Alexiadou (2001) who claims that an event expressed by a noun phrase is not asserted, as the assertion properties of a clause are related to the higher layer, in particular to Force in the CP system (strictly related to Fin, which is in turn related to Tense). The third Chapter investigates External Merge, functional projections in the V-domain and the left periphery of the vP-phase. Challenging the current approach, Alexiadou, Anagnostopoulou and Schäfer provide evidence against a derivational account of the causative/anticausative alternation. Based on English, German and Greek data, the authors put forth an analysis of change of state verbs which builds on the idea that these are decomposed and contain, in addition to the root they are derived from, a Voice and a Cause component. Valentina Bianchi discusses the Italian reciprocal modifier uno dopo l’altro (lit. ‘one after the other’) and argues that this modifier requires an antecedent that syntactically agrees for plural number with the modified predicate. It is proposed that the reciprocal modifier requires a particular type of event pluralization, and that this obtains via syntactic

10 Mara Frascarelli number agreement in a dedicated functional projection in the immediate periphery of the verb phrase. This analysis implies that at least some instances of verbal number agreement are visible at the interface with the Conceptual-Intentional system(s). Hinterhölzl investigates functional projections in connection with remnant movement. Based on evidence from verb clusters in West Germanic, the author argues that the V-domain comprises a number of functional heads which serve to license finite and different types of nonfinite complements of the verb. The discussion on phase condition effects provides important support for the V-domain as a strong phase and – differently from standard assumptions – shows that the leftedge of this phase is formed by the Aspect phrase. Poletto deals with the left periphery of the vP phase in connection with Internal Merge (scrambling) for discourse requirements. Following recent proposals, it is argued for the existence of a low left periphery that includes Focus and Topic positions in a cartographic approach. Working on data from Old Italian, the author shows interesting parallels with the high left periphery (i.e., the Cdomain), endorsing the possibility that the features of a functional head (like Foc) can be parametrized as phase independent properties. The fourth Chapter is dedicated to the CP-phase, its connections with Force and discourse requirements on the one hand, and with Tense and (subject) Agreement specifications on the other. Internal Merge, licensing conditions and the checking (valuation) of optional EPP features are therefore major concerns of these contributions, in which a cartographic approach is generally assumed. In his work on subjunctive obviation, Costantini works out a syntaxsemantics interface approach, which exploits a series of independently supported principles. Crucially, he shows that the notion of “environment” that accounts for first-personal interpretation is rendered more precise through the notion of phase, so that in phases devoid of the speaker’s coordinate a de se reading may be conveyed only by means of an implicit argument. The distribution of overt and covert subjects in Italian is, on the other hand, the concern of Trecci’s contribution, in which an attempt is made to formalize the binding properties and interpretation of subjects in the theory of phases. The author proposes an A’ (topic) position for preverbal subjects (due to an extended EPP feature) and shows that the interpretation of null subjects depends on a specific configuration with their antecedent that relies on the notion of phase-edge. Rizzi and Shlonsky extend the approach to clausal subjects originally put forth in Rizzi (2003) to English Locative Inversion. Given the existence of a functional head, Subj, in

Phases and interpretation

11

which the Subject Criterion is satisfied, the authors argue that Locative Inversion indirectly satisfies this criterion through the intermediary device of a nominal Fin (showing A properties). Subject-like properties of inverted locatives are thus connected with the valuation of features in the Cdomain, in which the connection between C and T plays a crucial role. Finally, the interpretation of Focus is addressed in Cruschina’s paper on Sicilian. The author shows that information Focus undergoes movement to the left periphery only when related to specific pragmatic and illocutionary features (i.e., relevance and unexpectedness of the new information). This analysis bears out the cartographic and minimalist claim that Internal Merge in the CP-phase must be motivated by interpretive or discourserelated effects. A comparison with the syntactic properties of contrastive Focus finally leads the author to assume the necessity of two distinct projections in the left periphery, in line with recent interface analyses.

Acknowledgements A number of friends and colleagues have helped me in this project, in different ways. I thank all of them. In particular, I wish to thank Annarita Puglielli for helpful advice and suggestions, Roland Hinterhölzl for constant discussion and precious comments, and Henk van Riemsdijk, who liked this project from the very beginning and followed my work with kind support. I acknowledge Raffaele Simone and the secreterial staff of the Department of Linguistics at the University of Roma Tre, who permitted a perfect organization for the XXXI “Incontro di Grammatica Generativa” (February 2005), the presentations of which inspired the project of this book. Thanks to all the contributors of this volume: their collaboration was the most punctual and effective an editor can dream of, to Francesca Ramaglia, for her reading help and to Frank Benno Junghanns, for careful and fast typesetting. Finally, it was a pleasure to collaborate with the Publishing Editor of Mouton, my dear friend Ursula Kleinhenz. A final, special thanks is dedicated to Marcello, Lidia and Roland, because they are always with me.

12 Mara Frascarelli Notes 1. Through Select elements in the numeration are introduced as lexical items in the derivation, while Merge combines lexical items together to form a syntactic object in accordance with the “No-Tampering Condition” (i.e., Merge of X and Y cannot break up X or Y, or add new features to them). The operation Move is triggered by checking requirements of (strong) [-interpretable] EPP features, which are otherwise satisfied through Agree (a relation at a distance between a probe and a goal) (Chomsky 2000). 2. Two types of Merge are indeed assumed: External Merge, connected with argument structure, and Internal Merge that “yields the ubiquitous displacement phenomenon” (Chomsky 2005: 7). 3. The label vP indicates the larsonian verb phrase, that is, the functional projection headed by the light verb that selects for the VP headed by the lexical verb. 4. See, among others, Fox and Pesetsky (2004), Cecchetto (2004), Hinterhölzl (2006, this volume). For a critical approach, see Boeckx and Grohmann (2004); for a different perspective to linearization, see Van Riemsdijk (this volume). 5. IPs, on the other hand, have to do with Tense and Aspect but Tense features are inherited by C (see discussion below). 6. A possible solution is to assume that transfer to PF and LF is not simultaneous (cf. Hicks 2005) or, more drastically, that clausal interpretation requires a topdown parsing (as proposed by Bianchi and Chesi 2006). These hypotheses are however impossible to maintain in the current model and require further elaboration. 7. A proper treatment of binding conditions in the phase theory is still an open issue. For discussion see, among others, Reuland (2001) and Lee-Schoenfeld (2004). 8. In this respect, note that full interpretation of the vP in a question like (11) can be also considered as complete only after checking of the interrogative Force, yielding a typical rising intonation. 9. Chomsky (2005) crucially connects Structural Case assignment with (Subject) Agreement. So, in the presence of an expletive (EXPL) the phi-set of Tense remains intact because EXPL is incomplete. This can provide an explanation for a number of mismatches in agreement that are typical of unaccusative constructions.

References Alexiadou, Artemis 2001 Functional Structure in Nominals. Nominalization and Ergativity. Amsterdam: John Benjamins.

Phases and interpretation

13

Bianchi, Valentina and Cristiano Chesi 2006 Phases, left branch islands and computational nesting. Penn Working Papers in Linguistics 12 (1): 15–28. Boeckx, Cedric and Kleanthes K. Grohmann 2004 Putting phases into perspective. Paper presented at TiLT 2004: Tools in Linguistic Theory (Budapest, May 2004). Cecchetto, Carlo 2004 Remnant movement in the theory of phases. In The Structure of CP and IP. The Cartography of Syntactic Structures, Vol. 2, Luigi Rizzi (ed.), 166–189. Oxford: Oxford University Press. Chomsky, Noam 1981 Lectures on Government and Binding: The Pisa Lectures. Dordrecht: Foris. 1993 A Minimalist Program for linguistic theory. In The View from Building 20: Essays in Honor of Sylvain Bromberger, Ken Hale and Samuel J. Keyser (ed.), 1–52. Cambridge, MA: MIT Press. 1995 The Minimalist Program. Cambridge, MA: MIT Press. 1999 Derivation by phase. MIT Occasional Papers in Linguistics 18. Cambridge, MA: MIT Press. 2000 Minimalist inquiries: The framework. In Step by Step: Essays in Minimalist Syntax in Honor of Howard Lasnik, Robert Martin, David Michaels and Juan Uriagereka (eds.), 89–155. Cambridge, MA: MIT Press. 2001 Derivation by phase. In Kale Hale: A Life in Language, Michael Kenstowicz (ed.), 1–50. Cambridge, MA: MIT Press. 2002 An interview on minimalism. In On Nature and Language, Adriana Belletti and Luigi Rizzi (eds.), 92–161. Cambridge: Cambridge University Press. 2004 Beyond explanatory adequacy. In Structures and Beyond. The Cartography of Syntactic Structures, Vol. 3, Adriana Belletti (ed.), 104– 131. Oxford: Oxford University Press. 2005 On phases. Ms., Cambridge, MA: MIT. Fox, Danny and David Pesetsky 2004 Cyclic linearization of syntactic structure. To appear in Theoretical Linguistics: Special Issue on Object Shift in Scandinavian Languages, Katalin É. Kiss (ed.). Berlin: Mouton. Hicks, Glyn 2005 Binding theory and its consequences for features, phases, and derivation. Ms., University of York. Hinterhölzl, Roland 2006 Restructuring, Remnant Movement and Scrambling in West Germanic. Oxford: Oxford University Press.

14 Mara Frascarelli Hiraiwa, Ken 2005 Dimensions of Symmetry in Syntax: Agreement and Clausal Architecture. Ph.D. diss., MIT. Kayne, Richard S. 1994 The Antisymmetry of Syntax. Cambridge, MA: MIT Press. Lee-Schoenfeld, Vera 2004 Binding by phase: (Non-)complementarity in German. Journal of German Linguistics 16: 111–171. Legate, Julie 2002 Some interface properties of the phase. Linguistic Inquiry 34: 506– 516. Matushansky, Ora to appear Going through a Phase. In Proceedings of the MIT Workshop on EPP and Phases, Martha Mc Ginnis and Norvin Ricahrds (eds.). Cambridge, MA: MIT Press. Reinhart, Tanya 1998 Wh-in-situ in the framework of the Minimalist Program. Natural Language Semantics 6: 29–56. Reuland, Eric 2001 Primitives of binding. Linguistic Inquiry 32: 439–492. Rizzi, Luigi 1990 Relativized Minimality. Cambridge, MA: MIT Press. 1997 The fine structure of the left periphery. In Elements of Grammar: Handbook in Generative Syntax, Liliane Haegeman (ed.), 281–337. Dordrecht: Kluwer. 2003 On the form of chains: Criterial positions and ECP effects. To appear in On Wh Movement, Lisa L. Cheng and Norbert Corver (eds.). Cambridge, MA: MIT Press. Ross, John R. 1967 Constraints on Variables in Syntax. Ph.D. diss., MIT. Svenonius, Peter 2004 On the edge. In Peripheries: Syntactic Edges and their Effects, David Adger, Cécile de Cat and George Tsoulas (eds.), 259–287. Dordrecht: Kluwer Szabolsci, Anna 1989 Noun phrase and clauses: is DP analogous to IP or CP? In The structure of noun phrases, John R. Payne (ed.). Amsterdam: Mouton.

Chapter 1 Interpretation and structural conditions

Grafts follow from merge Henk van Riemsdijk

In this article I intend to show that the concept of “grafts” that I have argued for in a number of articles (cf. Van Riemsdijk, 1998b, 2000, 2001a) can be given a natural interpretation in a framework that makes use of internal and external merge. More specifically, I will argue that when all the logical combinations of internal and external merge are taken into account, grafts must exist and are not, as hitherto believed, an odd problem for phrase structure. It follows also that, in addition to my prime examples of grafts (free relatives (FRs) and transparent free relatives (TFRs)), we should expect there to be more constructions with the telltale properties of grafts. One such construction, “Horn Amalgams” will be introduced here in combination with an argument that Grosu’s (2003) attempt at reducing TFRs to regular FRs fails. 1. Introductory remarks1 In a number of articles I have argued that there are constructions in which classical syntactic structures in terms of so-called well-formed trees are insufficient.2 A typical simple example is (1). (1)

a far from simple matter

Clearly this is a noun phrase. The question is, what is its left-hand modifier? Is it the adjective simple, or is it the adjective far? Semantically, simple seems to be the head, but then what modifies it? Far from is not really a constituent. If far were the head, then it might be thought to have a PPdependent in the form of from simple. But the latter option is unlikely anyway since adjectival heads in prenominal position must be adjacent to the noun.3 This means we have a paradoxical situation. Under the assumption of grafts, trees can “grow together” in unconventional ways, specifically in sharing terminal elements. Under such an analysis, (1) would consist of two

18 Henk van Riemsdijk simple trees, one corresponding to a simple matter, the other one representing far from simple. The adjective simple is the shared element, the “callus” in my botanical metaphor. The paradox is avoided by the idea that the root node of the PP is not connected to any position in the host tree (the DP). Even in a simple case like this, interesting evidence exists in favor of an approach like this. Consider the fact that in Dutch attributive adjectives inflect, while predicative adjectives do not. In the equivalent of (1), the adjective is inflected by means of schwa: (2)

a. een alles behalve eenvoudig-*(e) oplossing a anything but simple solution b. deze oplossing is alles behalve eenvoudig-(*e) this solution is anything but simple

On the graft approach we can say that the uninflected adjective is the shared element, while the inflection marker remains outside the callus. The most compelling evidence for grafts comes from the construction called “transparent free relatives” which I will return to in section 4. Before doing so, however, consider the question of whether the introduction of grafts into the arsenal of descriptive tools in linguistic theory4 does not constitute an excessive extension of power. The point of the present article is to argue that, on the contrary, the existence of grafts follows directly from the logic of external and internal merge, as developed in Chomsky (2001, 2005).

2. The logical necessity of displacement In two recent lectures,5 Chomsky has sharpened the reasoning concerning what used to be called the displacement problem. Until recently, the existence of movement (displacement) was seen as a problem: why does it exist?6 Chomsky now argues that this is a misguided question. Recursive merge is now thought to be the defining property of the human language faculty (cf. Hauser et al. 2002). What Merge does is, it takes two elements, call them A and B, and puts them together. Now, in the simplest case, A and B are both simplex. That would be a case of “very first merge”. Apart from this case, however, A and B will generally be syntactic trees that have been formed by previous instances of merge. Consider the following situation.

Grafts follow from merge

(3)

A

19

B

C In a configuration like this, there are only two ways in which merge could apply to B: either B is merged with A, resulting in (4a), or it is merged with C, yielding (4b).7 (4)




a. A

b. B


B

C

C

As Chomsky puts it, we would need a stipulation to prevent merge from applying to the pair {B, C}, therefore a theory that avoids this stipulation and hence admits internal merge (or remerge) is the simplest one. Therefore it is not the existence of displacement that constitutes a problem. The real problem would be if displacement did not exist.

3. Grafts as the missing case Pursuing this line of thinking, consider now a situation that is minimally different from (3), the only difference being that the internal structure of the tree dominated by A is partly specified. (5)

A

D

B

C

Using the same style of reasoning as Chomsky’s, we may say that a stipulation would be required to prevent merge from applying to the pair {B, D}. The result of such a merger operation can be depicted as in (6).

20 Henk van Riemsdijk (6)

A

D




B

C The resulting structure, of course, is exactly what I have been calling a graft. D is the shared element which, in the host tree
, is dominated by
and is a sister of B, while it preserves the structural relations it had in its source tree A.8 It might be said that some fundamental principle blocks access to D in (6), see for example the discussion on possible interactions between grafting and phase theory in section 4.3. below. But notice that this is equally true for Chomsky’s argument concerning internal merge: whatever fundamental principle is invoked to block internal merge would be a stipulation in Chomsky’s sense. I am not, of course, denying that the availability of grafting could lead to massive overgeneration – it does. And one of the major tasks will be to severely constrain this particular instantiation of merge. But this necessity to constrain grafting is not different, in principle, from the necessity of constraining, say, internal merge. The only difference is that we have worked on constraining internal merge (formerly move) for several decades now, while the program to constrain external-internal merge (that is, graft) is only just starting.9 I conclude that we should not be surprised to find phenomena in natural language that point in the direction of grafts. On the contrary, if we did not find such phenomena, we should be seriously worried. With this in mind, let us first reexamine two major cases for the existence of grafts: free relatives (FRs) and transparent free relatives (TFRs). Subsequently, in section 6, I will discuss a third construction, “Horn Amalgams” (HAs), and show that this construction has all the properties that we expect to find in grafts. Moreover, the analysis of HAs as grafts will be shown to invalidate the arguments of Grosu (2003) against my analysis of TFRs as grafts.

Grafts follow from merge

4.

21

More thoughts on grafts: Free relatives and transparent free relatives revisited

4.1. Free relatives One of the most interesting properties of free relatives in many languages is the so-called case matching property, cf. Groos and Van Riemsdijk (1981) and Van Riemsdijk (2006). In German, for example, a free relative in which a dative indirect object has been relativized must be in a position in the matrix clause to which dative case is assigned.10 (7)

a. Ich gebeDAT die Belohnung wemDAT eine gebührtDAT I give the reward whom one deserves ‘I give the reward (to) who(m) deserves one’ b. Ich gebeDAT die Belohnung *werNOM /*wemDAT eine verdientNOM I give the reward who whom one deserves c. *WemDAT /*werNOM eine Belohnung gebührtDAT bekommtNOM eine whom who a reward deserves receives one

In (7b) we have a case conflict: the matrix verb requires a dative on the whword, while the verb in the relative clause (unlike the verb in (7a)) requires a nominative. Similarly, in (7c) we also have a case conflict because the opposite situation is found: the matrix requires a nominative while the relative clause requires a dative. Observations like these have led some researchers to propose that the fronted wh-element in the free relative clause is not in the usual position of wh-phrases in relative clauses, the Specifier of CP, but in the position of the (apparently missing) head of the free relative (cf. Bresnan and Grimshaw, 1978). Such an analysis, similar in many ways to the so-called raising analysis of headed relative clauses,11 would account for case matching in a straightforward way. What might be considered problematic, however, is that under such an analysis there is a single element, the wh-word, that is assigned a -role both in the relative clause and in the matrix clause, which constitutes a violation of the Theta Criterion of Chomsky (1981). This is, however, a problem that also attends the graft approach to free relative clauses that I now turn to (cf. also footnote 9). In Van Riemsdijk (2006) I propose to treat free relatives as grafts. The above example (7a) would accordingly be derived in the following manner.12

22 Henk van Riemsdijk (8)

input tree A (matrix/host):

V’ ei DP V 6 g die Belohnung geb-

graft

input tree B (grafted):

2

CP 3 u IP u vP 3 1 DP VP g 3 eine DP V g g wem gebühr-

VP u V’ 3 DP V 5 g die B. geb-

Three steps involving merger are depicted in (8). Step 1, which is not directly relevant to our purposes but included for completeness, is the internal merge of the subject of the relative clause by which the bold-faced IP is created. Step 2 is the internal (re-)merger (wh-movement) of wem inside the relative clause, creating the bold-faced CP. The crucial third step is the external merge of an internal element in the dependent relative clause tree to the partial matrix tree, creating the bold-faced VP. This third step is what I call the process of grafting.

4.2. Transparent free relatives Transparent free relatives are exemplified by the following sentences (cf. footnote 2 for references). (9)

a. I ate what they euphemistically referred to as a steak b. There is what I suspect is a meteorite on the front lawn

These examples deviate from “normal” free relatives in a number of ways. First, free relatives generally have either a definite interpretation or a free

Grafts follow from merge

23

choice (universal) interpretation.13 Take (10), which either means that I eat the thing that is on the table, or that I eat whatever is on the table, no matter what it is. (10) I eat what is on the table In contrast, the examples in (9) are indefinite. For example, the most plausible paraphrase of (a) is something like ‘I ate a steak – at least they called it a steak’. The indefiniteness of such examples is also shown by the fact that in (9b) the free relative occurs in a there–insertion context. It appears, in fact, that what determines the indefiniteness of the free relative in (9) is the predicate nominal (a steak, a meteorite). This is a first indication that it may be not the properties of the wh-element but the properties of the predicate nominal (or AP) that determine the behavior of the free relative. This is precisely why they are called ‘transparent’. If we treat them as regular free relative clauses, the predicate XP is deeply embedded inside the relative clause. Yet, it seems to act as if it were the head of the relative clause construction. The most direct or radical approach to account for this observation would seem to be to take the predicate XP to be the shared element, the callus. It turns out that there is ample evidence for such an approach. I will first review a number of the more salient pieces of evidence, and then present a few new considerations supporting the same point.14 The wh-word what normally dictates singular agreement, as shown in (11a), both inside the relative clause, if it is a subject, and also as the virtual head of the free relative. (11) a. What pleases/*please me most adorns/*adorn the living room wall b. What *seems/seem to be some meteorites *was/were lying there c. What seems/*seem to be a meteorite was/*were lying there In (11b) the situation is precisely the reverse. The plural agreement appears to be dictated by the predicate nominal, not by what, as (11c) confirms. Elements inside a (free) relative clause resist extraction, that is, they are subject to the Complex Noun Phrase Constraint. But if the predicate nominal is a shared element, it is also part of the matrix clause and should therefore allow extraction from it. This is precisely what we find. (12) a. *Who did they copy a photograph that was identified as [a picture of [ei]] ?

24 Henk van Riemsdijk b. *Who did they copy whatever was identified as [a picture of [ei]]? c. Who did they copy what was identified as [a picture of [ei]]? (12a) shows that extraction from the predicate nominal is not possible out of a headed relative clause. (12b) shows that the same fact holds for regular free relatives. And (12c) shows that extraction is OK (or at least significantly better) from a TFR. Consider next the fact that the adjectival agreement facts in Dutch that we discussed above in connection with example (2) hold with equal force for TFRs: (13) een wat ik zou noemen eenvoudig-*(e) oplossing a what I would call simple solution Inside the relative clause the adjective is a predicative adjective which should not be inflected. But in the matrix noun phrase it is an attributive adjective that requires the schwa. Note furthermore that the position between the article and the head noun is not a position where we would expect to find any kind of relative clause.15 The arguments that have been used to support the raising analysis of headed relative clauses can be applied to the predicate XP of a TFR as well. Take the idiom chunk argument. The noun headway can only exist as part of the idiom in which it is the object of the verb make. In this light (14) seems odd. (14) The headway they made was impressive But if the head noun is raised from the position of the direct object inside the relative clause into the head position, then this type of reconstructive behavior is explained. Observe now, that we have the same effect in (15). (15) They didn't make what can reasonably be considered headway If headway is a shared element, that is, if it is simultaneously the predicate of consider and the object of make, the problem disappears. With bound anaphors, typically used to illustrate reconstructive behavior, we get the same pattern. (16) a. They live in what is often referred to as each other’s backyard b. She was what can only be interpreted as proud of herself

Grafts follow from merge

25

In both cases it is the subject of the matrix clause that binds the anaphor in the predicative XP. This should not be possible if the relative clause were a ‘real’ relative clause. But if the predicative XP is shared with the matrix clause, the binding relation between they and each other and between she and herself comes as no surprise.16 In this example, the binder is in the matrix clause. But if the callus, the DP containing the bound anaphor is truly shared, we might expect to find cases in which the bound anaphor is jointly bound by elements both in the TFR and in the matrix clause. Such cases are not easy to construct, but the following examples are candidates. (17) a. Bushi would never acknowledge what Cheneyj refers to as [each other’s]i+j mistakes b. Johni hates to discuss what Maryj calls [each other’ s]i+j sexual deficiencies The next fact we will briefly elucidate concerns case matching. If TFRs are grafted at the predicative XP, we expect case matching effects to show up when the XP is a noun phrase. That is, case matching should be found at the predicate nominal, the callus, instead of at the wh-word. This is indeed what we find (cf. Van Riemsdijk, 2000, 2001a). Predicate nominals agree in case with their subject, which may either be a nominative or an accusative. Accordingly the TFR will have to be in a nominative or accusative position in the matrix clause.17 (18) a. Ich habe was man einenACC schnellenACC Wagen nennt gekauft I have what one a fast car calls bought ‘I have bought what one calls a fast car’




b. Ich habe was  *einNOM schnellerNOM Wagen  genannt wird gekauft  *einenACC schnellenACC Wagen 

I have what a fast car ‘I have bought what is called a fast car’


called

is

bought



*einNOM schnellerNOM Wagen  nennen wird selten gekauft c. Was viele   *einenACC schnellenACC Wagen 

what many a fast car call ‘What many call a fast car is rarely bought’

is

rarely bought

26 Henk van Riemsdijk d. Was einNOM schnellerNOM Wagen genannt wird wird selten gekauft what a fast car called is is rarely bought ‘What is called a fast car is rarely bought’ In the active variants (both in the matrix clause and in the TFR) we have a small clause construction in which the subject of the predicate is marked accusative by the verb nennen (‘call’). If the construction is passivized, the small clause subject appears in the nominative. In the b- and c-examples we have a conflict since the matrix requires an accusative while the TFR requires a nominative (18b), and vice-versa (18c).

4.3. Grafts and phases The complex derivation that was presented as (8) above, and which I repeat here as (19) for convenience, presents a problem. (19) input tree A (matrix/host):

V’ ei DP V 6 g die Belohnung geb-

graft

input tree B (grafted):

2

CP 3 u IP u vP 3 1 DP VP g 3 eine DP V g g wem gebühr-

VP u V’ 3 DP V 5 g die B. geb-

The problem is that if we take the numbering of steps 2 and 3 literally, the graft (step 3) seems to be in conflict with Phase Theory (cf. Chomsky, 2001). Following Chomsky, vP and CP may be thought to be the phases

Grafts follow from merge

27

(essentially what used to be called cycles). In building up tree B, once we leave the domain of vP and start building the IP, the vP is immediately shipped off to the PF-interface. That is, any constituent inside the vP (such as the DP wem) is no longer accessible to operations of syntax. Therefore, step 3 appears to be incompatible with Phase Theory. In this particular case, however, this reasoning does not apply. This is so because the DP wem is also dominated directly by the CP and is therefore also a member of the higher phase. But notice that this easy way out is not available for TFRs, since in TFRs the predicative XP that is shared does not move to a higher position inside its clause. In order to see this, consider the derivation of (a simplified variant of) (9a): (20) I ate what they called a steak (21) input tree A (matrix/host):

V ! eatV’ e V ! eat-

input tree B (TFR):

2

graft

CP i IP i vP ri 1

DP

! they

3

VP

ri V SC ! ri call- DP DP ! 6 what a steak

Step 1 in the derivation is, again, immaterial to the issue at hand. The crucial difference between (8/19) and (21) is that in (8/19) steps 2 and 3 affect the same element (wem) while in (21) it is what that gets remerged (wh-moved to Spec,CP) and the predicative DP a steak that is grafted to become the direct object of eat. Now, if we complete the building of tree B before applying graft (step 3), we do have a problem with Phase Theory because the DP a steak has already been transferred to PF. However, there is no reason

28 Henk van Riemsdijk why we should assume that tree B is already completed when graft applies. Specifically, graft may apply at any stage (DP, SC, VP, vP) until the vP is sent off to spell-out (PF). I conclude that external merge of any kind, that is including graft, can apply to a pair of nodes that each belong to their respective trees, where every one of these trees can be at some intermediate stage of its growth. Phase Theory can be thought of as one of the conditions that partly regulate this process.18

4.4. Preliminary conclusions Grafts are by no means the marginal and exotic creatures that most linguists like to ignore. Once we recognize that the most non-stipulative theory of merge includes graft as one of the subcases of external merge, the study of grafts assumes considerable importance. Many questions impose themselves. For example, TFRs are quite similar to parentheticals in many ways: both constructions are mostly used as hedges, creating a certain distance between the speaker and the choice of a specific term in a sentence. The main difference is that parentheticals do not seem to share any material with the matrix clause. Is there a way to treat parentheticals on a par with TFRs? A second question, already mentioned in footnote 11, is whether headed relative clauses, and perhaps adjuncts in general, could be conceived of as grafts.

5.

Grosu’s critique

In a very substantial study, Grosu (2003) argues against my analysis of TFRs in terms of grafts. Within the size limitations of the present article, I cannot do full justice to an 85 page long article, obviously, so I will have to limit myself in the present section to a few remarks on those passages that deal specifically with the arguments that I present in favor of the graft analysis. In the next section, however, I will show that Grosu’s line of argument is seriously on the wrong track in one crucial respect. By and large, Grosu accepts the notion that the predicative XP in TFRs, which he calls the “transparent nucleus” (TN), displays some properties that indeed make it look transparent. By criticizing my arguments, Grosu wants to support his idea that TFRs are really the same thing as regular FRs in which the transparent properties of the TN are passed along, presumably

Grafts follow from merge

29

as a kind of free riders on the wh-word what, into a position where they are indeed accessible to the matrix clause in very much the same way that fronted wh-elements are in regular FRs. Unfortunately, Grosu remains rather vague about the precise way the relevant features are carried along. Consider first the adjectival agreement facts illustrated in (13) above. I consider this a strong argument in favor of the graft approach. If the adjective (or AP) is purely a predicative constituent inside a free relative, we would not expect to find any agreement morphology, since predicative adjectives never inflect in Dutch. Prenominal attributive adjectives, however, do inflect. More specifically, attributive adjectives carry a schwa-suffix unless the head noun is a singular neuter indefinite. The number, gender and definiteness features in question are unlikely to be passed along by the whword wat because wh-pronouns have such features of their own. Since wat is a neuter singular pronoun, we would not expect it to be able to pass along the feature non-neuter as would be required in the case of example (13), for example. Grosu does not deal with this argument as such, but he discusses some relevant facts under the heading “right edge constraint” in his section 7.5. (Grosu 2003: 311ff). By right edge constraint Grosu refers to Williams’ (1982) Head Final Filter discussed above in connection with example (1) (cf. note 3).19 What Grosu apparently fails to appreciate is that these adjectival inflection phenomena, which are absent in German due to the Head Final Filter (the verb will always follow the predicative adjective inside the TFR due to the SOV-property of German), do occur in Dutch by virtue of the (quite exceptional) fact that the predicative adjective can follow the verb in TFRs as illustrated in (13). As a result of this exceptional AP-extraposition, the Head Final Filter is obeyed. Grosu also objects to my idiom chunk argument. He uses my own observation that idiom chunk licensing is virtually suspended in a small clause predicate whose subject is this/that (the non-wh variant of what): (22) I would not call this significant headway (Grosu’s example (76b)) But thinking about what this means in an example like this, it would be quite plausible to say that this stands for what we made. At any rate, this observation does not make the argument go away because there still is a quite appreciable contrast between (23a) and (23b). (23) a. Nick has made what one may call significant headway b. ?*Nick has achieved what one may call significant headway

30 Henk van Riemsdijk Grosu is justified in observing that some of my arguments are difficult to evaluate because I did not fully elaborate the theoretical framework in which the concept of graft could be formalized. This deficit on my part is now, at least partly, remedied by the interpretation of grafts as internal/external remerges as illustrated in (21) and by the works of Guimarães (2004) and De Vries (2004, 2005). Nevertheless, Grosu is to be forgiven for assuming that in my analysis “the semantic nucleus qua element of the matrix c-commands the relative CP (and thus, itself, qua element of that CP)” (2003: 288). Still, he is wrong. (21) shows that this is not the case. In addition, of course, there are the usual empirical discussions which I will go into only marginally here. It is true, clearly, that the there-insertion argument is always somewhat complicated by the fact that sometimes, under certain specific conditions, definite DPs can also show up in this construction. One example Grosu mentions is (24). (24) In this vat, there is just the {kind, amount} of wine that I consider ideal But consider again the example (9b), repeated here as (25). (25) There is what I suspect is a meteorite on the front lawn It is still a fact that (25) contrasts with (26), despite the fact that both could be rephrased by means of some expression like the kind of thing, as (27) shows. (26) *There is what originates in outer space on the front lawn (27) a. #There is the kind of thing that I suspect is a meteorite on the front lawn b. #There is the kind of thing that originates in outer space on the front lawn The examples in (27) may or may not be fully felicitous (hence the #mark), but there is no denying that there is a clear contrast between (25) and (26). Another area where the facts are contested is that of case matching phenomena discussed above in connection with (18). It is true that judgments are notoriously difficult, especially in the domain of case morphology, where many German dialects have deficient systems compared to the standard

Grafts follow from merge

31

language. The fact remains, however, that there is a discernible matching effect there. Furthermore, this effect cannot be explained away by assuming that case features are carried along by was since was syncretizes the nominative and accusative (see note 17 on this issue), a fact that is also significant in the analysis of regular FRs. A last issue that I want to briefly touch upon here is that of extraposed free relatives. This is important both for the graft analysis of FRs and that of TFRs and it involves an observation that has played an important role in the debates about FRs ever since it was pointed out in Groos and Van Riemsdijk (1981). The point is this. In Dutch and German, by and large, CPs can be extraposed (to the postverbal position, these languages being SOV) but DPs cannot. In FRs the wh-word must extrapose along with the rest of the FR, hence the wh-word must be inside the CP and cannot be in the position of the head of the relative clause, as claimed by Bresnan and Grimshaw (1978). In a graft analysis the situation is a bit different. A FR like (28) would roughly be represented as (29). (28) Wij hebben wat hij zei opgenomen we have what he said recorded ‘We have recorded what he said’ (29) input tree A: (matrix/host)

V ! opgenomen

input tree B: (FR graft)

wh-mvt

VP

VP p graft V ! opgenomen

CP u IP rp DP VP ! rp hij DP V ! ! wat zei

Extraposition is possible: (30) Wij hebben opgenomen wat hij zei

Extraposition

32 Henk van Riemsdijk The question then is what structure we should attribute to (30). Observe that much is unclear about this question in more or less standard versions of minimalism. Many hold that rightward movement does not exist (cf. e.g. Beermann et al. 1997; Kayne 1994), and if it does exist, it is difficult to determine where the CP moves to. In (29), I have assumed that extraposition is external merge to the right. On this assumption there is no surprise: it truly is the CP that moves into a postverbal position. This does not affect the fact that the DP wat is multiply dominated both by the matrix VP, by the FR-VP, and by the FR-CP. 20 There is also a second way of looking at things, however. Suppose that extraposition is indeed external merge to the right of VP. But suppose that extraposition applies blindly to any XP. In that case we could also say that, indeed, it is the DP wat of the FR-tree that is extraposed. The corresponding derivation is given in (31), which is minimally different from (29) in precisely this point. (31) input tree A: (matrix/host)

input tree B: (FR graft)

wh-mvt

V ! opgenomen

VP

VP p graft V ! opgenomen

CP u IP rp DP VP ! rp hij DP V ! ! wat zei

Extraposition

Such a derivation would seemingly lead to a violation of the generalization that DPs may not extrapose in Dutch and German, as Grosu objects. But now consider the situation from the perspective of linearization, cf. note 8 above. No matter how exactly the linearization algorithm is formulated, it will have to ensure that all elements of the graft are spelled out contiguously to the left and to the right of the shared element. That is, all elements of the matrix are pronounced (wij hebben opgenomen) until we get to the shared element wat, which is then spelled out preceded by what precedes it

Grafts follow from merge

33

in the graft, which is nothing, and followed by what follows it in the graft, which is hij zei.21 Thereby (30) results. Looking at the linear sequence of elements, now, we can see that what follows the matrix verb (opgenomen) is wat hij zei, which is a CP and not a DP and hence is expected to be grammatical. Observe now that indeed TFRs pattern identically: (32) a. Wij hebben wat hij een genie noemt benoemd we have what he a genius calls appointed ‘We have appointed what he calls a genius’ b. Wij hebben benoemd wat hij een genie noemt The structure/derivation of (32b) is identical in all relevant respects to that of (29) except that the predicative DP een genie is now multiply dominated only by the two VPs. This difference does not affect the point that it is the CP that is externally merged to the right of the matrix VP. I tentatively conclude that the original argument from extraposition stands, and that it does not stand in the way of interpreting the wh-word in FRs or the predicative DP (TN) in TFRs as an external head from the perspective of the matrix clause. Much more remains to be said about Grosu’s (2003) article, but all I can do in the remainder of this article is discuss one very general point of logic in his line of thinking.

6. Lakoff’s Horn-amalgams What drives Grosu, witness the title of his article (“A unified theory of “standard” and “transparent” free relatives”), is the desire to provide a uniform analysis for every syntactic construct that superficially looks like a headless relative clause. Unification is, of course, always a noble goal. Still, sometimes you have to think about what should be unified with what. Faced with three constructions, call them A, B and C, there might be one set of properties (for example absence of a head and appearance of a whclause) that A and B share to the exclusion of C. But there might be another property (for example transparency phenomena on the predicative XP) that is shared by B and C to the exclusion of A. In such a case we have to be careful to make the right choice. What I wish to show in the present section

34 Henk van Riemsdijk is that there is a construction that indeed shares with TFRs, but not with regular FRs, the property that the predicative XP exhibits transparency phenomena. The construction has been brought to the attention of the (generative) linguistic world by George Lakoff at a time when the first observations about TFRs were seeing the light and led Lakoff to (informal) conclusions that can be considered antecedents of my graft theory (cf. Lakoff, 1974). More specifically I propose to discuss a construction apparently pointed out to Lakoff by Larry Horn.22 The following is an example: (33) John is going to, I think it’s Chicago on Saturday (Lakoff, 1974: 324 ex (13a)) I will use the term Horn Amalgams (HAs) to refer to this construction. Right away we can observe that the semantic or pragmatic function of HAs is very similar to that of TFRs: the predicate XP is the semantic nucleus and the rest is a hedge by means of which the speaker distances him/herself from the choice of the term or directly calls it into doubt. Putting it in Grosu’s terms (Grosu, 2003: 279), “the small clause whose predicate is the TN is felicitous just in case it is in the scope of a TFR-internal intensional operator”. For ‘TFR-internal’ we can also read ‘HA-internal’. Lakoff’s examples show the contrast between intensional and non-intensional contexts quite clearly (Lakoff, 1974: 324): 23 (34) a. John is going to, is it Chicago? on Saturday b. John is going to, I’m sorry to say it’s Chicago on Saturday c. *John is going to, God knows it’s Chicago on Saturday d. *John is going to, it’s odd that it’s Chicago on Saturday The c- and d-examples are not hedges but rather statements of fact that are embellished or modified by some additional qualification. Putting things in a somewhat wider context, the shared element (callus), both in TFRs and HAs, seems to have the status of a quote (cf. i.a. Abbott 2003; Geurts and Maier 2003; Potts, to appear; Recanati 2001; von Fintel 2004). Observe that there is one quite obvious difference between TFRs and HAs. HAs are not subject to an indefiniteness restriction, unlike TFRs. This is already apparent from (33), and also shown in the following example. (35) John is taking did he say his daughter? out today

Grafts follow from merge

35

A second difference, also illustrated in (28) is that HA’s can be questions. TFRs involve (relative) wh-movement, hence interrogative wh-movement is excluded. But nothing precludes wh-question formation in HAs. I will now proceed to show that the predicative XP in HAs does indeed exhibit the transparency phenomena that we know from TFRs. Consider first the issue of bound anaphors. Alongside (16) we have the corresponding HA (36). (36) a. They live in, don’t the Americans call it each other's backyard? b. She was, I think you might call it proud of herself Similarly, we find idiom chuncks inside HAs that are licensed by some element of the matrix clause: (37) a. They didn’t make a lot of, I think the correct term is headway b. Bill kicked, I seem to remember you call it the bucket The extraction facts can also be reproduced. (38) a. Whoi did they publish, I believe it was a dirty picture of ei? b. What conversationi did John make, I think it very probably was an unauthorized recording of ei? Turning now to the morphological properties, observe that Dutch adjectival agreement in HAs functions in exactly the way that we found in TFRs, cf. (13) above. (39) Dit is een, ik denk dat je het zou mogen noemen this is a I think that you it would may call eenvoudig-*(e) oplossing simple solution ‘This is a, I think you might call it simple solution’ As we saw before, the adjective must carry the schwa-inflection characteristic of attributive adjectives even though, from the point of view of the inserted clause, it is a predicative adjective which should not inflect.24 Turning now to case matching, things ought to be rather simpler since there is no wh-element to take into consideration. The question simply is, does the predicate nominal in a German HA have to satisfy the HA case

36 Henk van Riemsdijk requirement, or the matrix case requirement, or both? The following examples show that indeed both requirements must be satisfied, that is, case matching is obligatory:25
*einem DAT   NOM  

(40) a. Er hat sich, ich glaube das nennt NOM sich  *ein

he has refl. I believe that calls itself a Wahrsager anvertraut DAT soothsayer trusted ‘He entrusted himself to, I believe it is called a soothsayer’ b. Er hat sich, ich glaube das nennt ACC man einenACC Wahrsager he has refl. I believe that calls one a soothsayer ACC angelacht engaged ‘He has gotten himself, I believe you call it a soothsayer’ While space prevents me from going through the complete range of cases, I believe these facts are incontrovertible. Consequently, I feel that I have shown that HAs truly pattern like TFRs with respect to the ambiguous status of the shared element, the TN (or the ‘callus’ as I call it, extending the botanical metaphor).26 I conclude that the true unification lies in developing a theory that will account for the properties of both TFRs and HAs.27 I believe the theory of grafts goes a fair part of the way in the right direction. And consequently I also believe that Grosu’s (2003) attempt to unify TFRs with regular FRs is on the wrong track.

7. Conclusion On the basis of the above considerations, I conclude that it is time to take the notion of merge seriously. This implies that the theory of grafts, hitherto considered a dangerous oddity from a theoretical perspective and “only” concerned with completely marginal and exotic constructions such as TFRs and HAs, now finds itself right in the middle of the very core or narrow syntax. And the empirical range of linguistic facts that the theory of narrow syntax is to be held accountable for turns out not to be quite as narrow as some would like us to believe.

Grafts follow from merge

37

Notes 1. The ideas presented here have benefited from discussions with the students in my classes at the University of Vienna in June 2004, and from thoughtful comments by Edwin Williams. Thanks are also due to audiences at Tilburg University and at the Incontro di Grammatica Generativa in Rome, February 2005, as well as subsequent comments by Alex Grosu, Viola Schmitt and an anonymous reviewer. Parts of the present article appear in Van Riemsdijk (2004, to appear). 2. See Van Riemsdijk (1998b, 2000, 2001a) with antecedents in Nakau (1971), Kajita (1977), Lakoff (1974), and McCawley (1982, 1988). Lakoff coined the term ‘syntactic amalgams’ for what I call grafts. My own work grew out of a critique of Wilder (1998, 1999). 3. Cf. Williams’ (1982) Head Final Filter, going back to the Surface Recursion Restriction of Emonds (1976: 19, Emonds, 1985: 131). See also Van Riemsdijk (1998a) for more recent discussion. 4. A formalization of grafts should be possible along the lines of the theory developed in Moltmann (1992). This is not to say that the task is trivial. For more recent discussion, see Guimarães (2004) and De Vries (2004, 2005). These authors share a number of the basic insights that I have been pursuing in the present work. Linearization, clearly a potential problem for any theory of grafts, is insightfully discussed in Guimarães (2004) and De Vries (2005). 5. The lecture at the GLOW conference in Thessaloniki, April 2004, and the lectures at the TiLT conference in Budapest, May 2004. 6. See, e.g., Chomsky (2004:164f). 7. I am assuming without further argument here that internal merge must be conceived of as remerge, that is as multiple dominance (cf. Gärtner 2002) rather than as copy+deletion, as proposed in Chomsky (1995). The remerger of C with B under
is expressed as a dotted line here. For proposals arguing this point, see also Bobaljik (1995) Epstein et al. (1998) Guimarães (2004) De Vries (2004) and Zhang (2004). 8. Linearization seems rather straightforward. There is an asymmetry between the host tree and the graft tree; pronouncing elements of the host tree from left to right has precedence; as soon as we hit a shared element, here D, anything that precedes D in the graft tree A is pronounced first, followed by D itself, followed by whatever follows D in A. After D is exhausted, the remainder of
(that is, B) is pronounced. De Vries (to appear) shows, however, that the formalization of the linearization algorithm is far from trivial. One might raise the question, as did a reviewer, how we know which one of the two trees, floating in what Jean-Roger Vergnaud once, using a metaphor from molecular biology, called “the soup”, is the matrix or host tree. The most preferable take on this, I feel, is to say that the operation as such is blind to this choice: all trees are equal. At some point in the derivation, however, one part of the tree will have to assume

38 Henk van Riemsdijk

9.

10. 11.

12.

13. 14. 15.

root character in not having an overt complementizer, for example, and in having the verb in second position in languages like German. Thereby that part of a complex tree that projects to the root will be defined as the host. Future research will have to show whether such an approach is tenable. In Van Riemsdijk (2000) I speculate that it is not an accident that the shared element (the callus) in amalgamated constructions is typically a predicative constituent. For example, argument XPs are subject to the theta-criterion, which would mean that if some DP is an argument both in the host and in the graft a theta-criterion violation results. Predicate XPs are not subject to the theta-criterion. They typically are also less rigid in their morphological behavior, often fluctuating between case agreement and inherent case, often showing deviations from strict number agreement, etc. Pursuing such matters would be well beyond the scope of the present article, but we should take the program of constraining graft seriously. Note, however, that the above may well be overly restrictive. The analysis of TFRs, discussed below draws a close parallel between TFRs and internally headed relative clauses of the type found in languages such as Japanese (cf. also Van Riemsdijk 2000). The latter may well be treated as grafts, but they are not limited to predicative XPs. In the following examples, I use superscripts for case government properties of verbs and other case assigners, and subscripts for actual case forms. See Vergnaud (1974, 1985), Kayne (1994), Bianchi (1999). Pursuing the arguments given for these raising approaches, it is natural to investigate the possibility that headed relative clauses might be analyzed as grafts in ways similar to what I propose for free relatives in the text. Undertaking an attempt at such an analysis would go beyond the limits of the present article, however. Still, contemplating a derivation analogous to (8) for headed relatives is not hard, and it sheds considerable light on the relationship between head-external and head-internal relative clauses. In particular, it is easy to see that reconstruction effects with bound anaphors contained by the head of the relative clause being bound either in the matrix clause or in the relative clause are straightforwardly accounted for. The structures are greatly simplified. No inflectional functional structure has been taken into account. I remain agnostic as to the proper way of handling datives. Also, I stick to pre-bare-phrase-structure labeling. The final steps of the derivation of (7a) have also been omitted: merging the matrix subject and internal merge of the finite verb (Verb Second). This is uncontested for argument free relatives. With adjunct free relatives, however, the situation is less clear, cf. Caponigro (2004). See Wilder (1998, 1999) and Van Riemsdijk (1998b, 2000, 2001a) for extensive discussion of these and other properties. Recall from the discussion of (2) above that I am assuming that the inflection marker is external to the callus, and hence an element of the host tree. It is fair to ask, as does an anonymous reviewer, why case inflection could not be simi-

Grafts follow from merge

39

larly treated, thereby circumventing the case matching requirement in examples like (7). The answer is that case, at least in languages like German, is fusional and hence cannot be kept outside the callus. 16. It is important to keep in mind that TFRs generally also allow a construal as standard free relatives. In trying to keep the two apart in establishing introspective judgements, it is helpful to take into account the fact that TFRs are used to create an intensional context, see section 6 for some discussion. It is not easy to construct direct tests, but one possibility could be based on the question of how c-command works in grafts, as suggested to me by Viola Schmitt. Take bound pronouns in regular free relatives. (i)

Every Germani likes what hei considers peasant food

On standard assumptions every German should c-command the bound pronoun he. In my analysis of regular free relatives this should not be a problem (though a formal account remains to be given) since every German c-commands what and what c-commands he. By transitivity every German c-commands he. On a TFR construal, however, things should be different since every German ccommands peasant food, but peasant food does not c-command he. Does this correspond to your judgment, in other words is it true that (i) lacks the intensional interpretation typical of TFRs? I find this very difficult to ascertain. But notice that we can force a TFR situation by invoking idiom chunks. Consider the following pair, taken from German this time. (ii) a. Jeder Teilnehmer hat die unterschiedlichen Standpunkte auf [was every participant has the distinct points-of-view on what manche für einen gemeinsamen Nenner halten] gebracht some for a common denominator take brought ’Every participant has reduced the various points of view to what some take to be a common denominator’ b. Jeder Teilnehmeri hat die unterschiedlichen Standpunkte auf [was eri für einen gemeinsamen Nenner hält] gebracht ’Every participant has reduced the various points of view to what he takes to be a common denominator’ While judgements are still not crystal clear, I get a discernible contrast between (iia) and (iib). In (iib) the idiomatic reading of auf einen gemeinsamen Nenner bringen is harder to get and the intensional interpretation is largely lost. Needless to say, these are complicated issues that deserve to be pursued elsewhere. 17. It should be noted that the judgments require some thinking. This is so because a TFR generally also allows for a construal as a regular free relative clause. On such a construal, the free relative will mean something like ‘the thing that we call X’ or ‘the thing that is called X’. Of course, in a regular free relative we expect the matching effect to occur, but the form was is syncretic: the nomina-

40 Henk van Riemsdijk tive and the accusative of the neuter wh-word have the same form (was). As I argue in Groos and Van Riemsdijk (1981) and Van Riemsdijk (2006), it is the forms, not the features that determine the matching effect. Indeed, if instead of a masculine noun as in (18) we choose a feminine or a neuter noun, all four combinations of active and passive are grammatical since these case forms are also syncretic. In the following examples, the noun Wagen, which is masculine, is replaced by Auto, which is neuter. (i) a. b. c. d.

Ich habe was man einacc schnellesacc Auto nennt gekauft Ich habe mir was einnom/acc schnellesnom/acc Auto genannt wird gekauft Was viele einnom/acc schnellesnom/acc Auto nennen wird selten gekauft Was einnom schnellesnom Auto genannt wird wird selten gekauft

18. Exactly how Phase Theory interacts with principles of linearization is rather unclear. In view of the fact that, very roughly speaking, trees tend to be higher on the left and lower on the right, the first phases to be shipped off to spell-out will be among the last to be pronounced, while the first chunks of a sentence to be pronounced are most likely among the last phases that the derivation reaches. Hence some mechanism must be assumed to keep track of all these chunks of phonetic material waiting to be arranged in some sequence in order to be pronounced as a complex sentence. In lectures in Thessaloniki and Budapest, Chomsky addressed this problem (jokingly?) by saying that there had to be a kind of humunculus taking care of this. This humunculus would have to be like a superb parking lot attendant with a very good survey of all the parked vehicles and an excellent memory of where they came from and who they belong to. 19. Grosu’s choice of terminology is somewhat unfortunate here because Wilder (1998, 1999) had suggested to treat TFRs as “right edge phenomena” on a par with the right node raising construction. 20. On the assumption that rightward movement does not exist a more complicated situation arises which I will not attempt to work out here. 21. Note, incidentally, that this presupposes that the linearization algorithm is formulated in such a way that the “highest” merge site determines the locus of spell-out. In other words, wat is both the left hand sister of the matrix V and the right hand sister of the matrix VP in (31). But the latter is a location “higher” up in the tree and therefore wins. As a consequence wat is part of the extraposed material. 22. There is a second construction that Lakoff discusses which I believe to be amenable to an analysis in terms of grafts. Lakoff credits Avery Andrews with the discovery of the relevance of this construction, which is exemplified by cases of the following kind: (i) John invited you’ll never guess how many people to his party (Lakoff, 1974: 321 ex (1)) Space prevents me from pursuing this line of investigation here, but see Van Riemsdijk (2001b) for some discussion.

Grafts follow from merge

41

23. I deviate somewhat from Lakoff’s own typographical devices to identify the amalgam. 24. Recall that, in order to satisfy the Head Final Filter the adjective is (exceptionally) extraposed inside the inserted clause. I use the dotted underline on the inflection marker to reflect the fact that it is not, properly speaking, part of the graft but of the matrix clause. 25. The superscripts indicate the case that the respective predicates govern. 26. I do note in passing one potential problem. HAs do not extrapose the way TFRs do. Given what I say about extraposition of TFRs above, this is somewhat unexpected. I believe the answer must be that HAs have the structure of root clauses: verb second order, no complementizer, etc. But I am not, at this point, prepared to present a more formal account of this fact. 27. Alex Grosu (p.c.) points out an interesting asymmetry between TFRs and HAs. He observes that HAs cannot modify (be grafted onto) subjects; (i) a. *I think it is Chicago is a large city. b. *Is it Chicago? is a large city. c. *It’s Chicago isn’t it? is a large city. I feel, however, that these facts have nothing to do with subjects. In fact any subject that is not clause initial will tolerate a HA. Consider the following examples: (ii) a. There were I believe it was five hijackers on the plane b. Many Italians resent that isn’t it Turin? is organizing the Winter Olympics. c. Under no circumstances will I believe her name is Faye Lovsky ever win the European Song Contest. The contrast between the examples in (i) and (ii) is due, I believe, to perceptual clause typing effects: the HAs in (i) could also be root clauses. Apparently this is a garden path situation that must be avoided. A similar confusion is avoided in the examples in (ii).

References Abbott, Barbara 2003 Some notes on quotation. Belgian Journal of Linguistics 17: 13–26. Beermann, Dorothee, David LeBlanc and Henk C. van Riemsdijk, 1997 Rightward Movement. Amsterdam: John Benjamins. Bianchi, Valentina 1999 Consequences of Antisymmetry: Headed Relative Clauses. Berlin / New York: Mouton de Gruyter. Bobaljik, Jonathan D. 1995 In terms of merge: copy and head movement. MIT Working Papers in Linguistics 27: 41–64.

42 Henk van Riemsdijk Bresnan, Joan and Jane Grimshaw 1978 The syntax of free relatives in English. Linguistic Inquiry 9: 331– 391. Caponigro, Ivano 2004 The semantic contribution of wh-words and type shifts: Evidence from free relatives crosslinguistically. In Proceedings from Semantics and Linguistic Theory (SALT) 14, Robert B. Young (ed.), 38–55. Ithaca, NY: CLC Publications Cornell University. Chomsky, Noam 1981 Lectures on Government and Binding. Dordrecht: Foris Publications. 1995 The Minimalist Program. Cambridge, MA: The MIT Press. 2001 Derivation by phase. In Ken Hale: A Life in Language, Michael Kenstowicz (ed.), 1–52. Cambridge, MA: MIT Press. 2004 The Generative Enterprise Revisited. Discussions with Riny Huijbregts, Henk van Riemsdijk, Naoki Fukui and Mihoko Zushi. Berlin / New York: Mouton de Gruyter. 2005 Three factors in language design. Linguistic Inquiry 36: 1–22. Emonds, Joseph E. 1976 A Transformational Approach to English Syntax: Root, StructurePreserving, and Local Transformations. New York: Academic Press. 1985 A Unified Theory of Syntactic Categories. Dordrecht: Foris. Epstein, Samuel D., Eric M. Groat, Ruriko Kawashima and Hisatsugu Kitahara 1998 A Derivational Approach to Syntactic Relations. Oxford: Oxford University Press. Gärtner, Hans-Martin 2002 Generalized Transformations and Beyond: Reflections on Minimalism. Berlin: Akademie Verlag. Geurts, Bart, and Maier, Emar 2003 Quotation in context. Belgian Journal of Linguistics 17: 109–128. Groos, Anneke and Henk C. van Riemsdijk 1981 Matching effects in free relatives: a parameter of core grammar. In Theory of Markedness in Generative Grammar. Proceedings of the 1979 GLOW Conference, Adriana Belletti, Luciana Brandi and Luigi Rizzi (eds.), 171–216. Pisa: Scuola Normale Superiore. Grosu, Alexander 2003 A unified theory of ‘standard’ and ‘transparent’ free relatives. Natural Language and Linguistic Theory 21: 247–331. Guimarães, Maximiliano 2004 Derivation and representation of syntactic amalgams. Ph.D. diss., University of Maryland. Hauser, Marc D., Noam Chomsky and W. Tecumseh Fitch 2002 The faculty of language: what is it, who has it, and how did it evolve? Science 298: 1569–1579.

Grafts follow from merge

43

Kajita, Masaru 1977 Towards a dynamic model of syntax. Studies in English Linguistics 5: 44–66. Kayne, Richard 1994 The Antisymmetry of Syntax. Cambridge, MA: MIT Press. Lakoff, George 1974 Syntactic amalgams. In Papers from the Tenth Regional Meeting of the Chicago Linguistic Society, Michael W. La Galy, Robert A. Fox and Anthony Bruck (eds.), 321–344. Chicago: Chicago Linguistic Society. McCawley, James D. 1982 Parentheticals and discontinuous constituent structure. Linguistic Inquiry 13: 91–106. 1988 The Major Syntactic Phenomena of English. Chicago: Chicago University Press. Moltmann, Friederike 1992 Coordination and comparatives. Ms., dept. of Linguistics and Philosophy, MIT. Nakau, Minoru 1971 The grammar of the pseudo-free relative pronoun what. English Linguistics 6: 2–47. Potts, Christopher to appear The dimensions of quotation. In Proceedings of the Workshop on Direct Compositionality, Chris Barker and Pauline I. Jacobson (eds.). Cambridge, MA: MIT Press. Recanati, François 2001 Open quotation. Mind 110 (439): 637–687. Riemsdijk, Henk C. van 1998 a Head movement and adjacency. Natural Language and Linguistic Theory 16: 633–678. 1998 b Trees and scions – science and trees. In Fest-Web-Page for Noam Chomsky, 32pp. Cambridge, MA: MIT Press. 2000 Free relatives inside out. Transparent free relatives as grafts. In PASE Papers in Language Studies. Proceedings of the 8th Annual Conference of the Polish Association for the Study of English, Bozena Rozwadowska (ed.), 223–233. 2001 a A far from simple matter. Syntactic reflexes of syntax-pragmatics misalignments. In Perspectives on Semantics, Pragmatics, and Discourse, István Kenesei and Robert M. Harnish (eds.), 21–41. Amsterdam: John Benjamins. 2001 b WH-prefixes: the case of wäisch in Swiss German. In Naturally! Linguistic Studies in Honour of Wolfgang Ulrich Dressler Presented on the Occasion of his 60th Birthday, Chris Schaner-Wolles, John

44 Henk van Riemsdijk Rennison and Friedrich Neubarth (eds.), 423–431. Torino: Rosenberg & Sellier. 2004 Graft is the logically missing case of merge. Visnyk of the Kiev National Linguistic University 7: 5–13. 2006 Free relatives. In The Blackwell Companion to Syntax, Vol. 2, Martin Everaert and Henk C. van Riemsdijk (eds.), 338–382. Oxford: Basil Blackwell. to appear Towards a unified theory of wh- and non-wh-amalgams. In Festschrift for NN. Vergnaud, Jean-Roger 1974 French relative clauses. Ph.D. diss., MIT. 1985 Dépendances et niveaux de représentation en syntaxe. Amsterdam : John Benjamins. von Fintel, Kai Uwe 2004 How multi-dimensional is quotation? Ms. Cambridge, MA. Vries, Mark de 2004 Parataxis as a different type of asymmetric merge. In Proceedings of IPSI-2004. Pescara Conference on Interfaces, hosted by the International Conference on Advances in the Internet, Processing, Systems, and Interdisciplinary Research. Pescara: CD-ROM (ISBN 86-7466117-3). 2005 Merge: properties and boundary conditions. In Linguistics in the Netherlands 22: 219–230. Amsterdam: John Benjamins. Wilder, Chris 1998 Transparent free relatives. ZAS Papers in Linguistics 10: 191–199. 1999 Transparent free relatives. In Proceedings of WCCFL 17, Kimary N. Shahin, Susan Blake and Eun-Sok Kim (eds.), 685–699. Cambridge: Cambridge University Press (Stanford: CSLI). Williams, Edwin 1982 Another argument that passive is transformational. Linguistic Inquiry 13: 160–163. Zhang, Ning (Niina) 2004 Move is remerge. Language and Linguistics 5: 189–209.

An interpretive effect of head movement Winfried Lechner

1. Introduction In this paper, I explicate an argument for the view that head movement (HM) can have an effect on semantic interpretation, as expressed by the SAHM-conjecture in (1): (1)

SAHM conjecture: There are instances of semantically active head movement.

The argument is structured as follows. In certain contexts involving modals, schematized in (2a), the modal can take scope above a clause-mate quantifier, resulting in inverted scope order, as in (2b). The position in which QP is interpreted (L in (2b)) can moreover be shown to be located above the position in which the modal is base generated (K in (2b)). It follows that the modal has to be interpreted in a derived location.1 (2)

a. Overt syntax: 3 QP XP 3 Modal

b. Interpretation: 3 QP XP 3 ƒModal„ YP 3 ZP L ƒtQP„ wo K tModal (base position) WP 6 tQP

This result is of theoretical interest inasmuch as a demonstration that HM can affect interpretation, or can be affected by principles of interpretation, generates a strong argument for computing HM in syntax, and not at PF, as recently

46 Winfried Lechner suggested in the literature2 (Boeckx and Stjepanovic 2001; Chomsky 2000, 2001; Harley 2005). More precisely, the combination of (1) and the assumption implicit in the current model that PF-operations do not influence LF representations entails (by contraposition3) that HM has to apply in the stem of the derivation, and cannot be relocated into the phonological component. In addition, only an orthodox account of HM, according to which HM consists in syntactic displacement of terminals, correctly predicts that HM operations have the potential of shifting the scope of the moved category. This finding supplies an argument against alternative analyses of HM which derive its effect by iterative applications of phrasal movement (on remnant movement see Hinterhölzl 1997; Koopman and Szabolcsi 2000; Mahajan 2000; Müller 2004; Nilsen 2003). If the modal in (2b) were embedded inside a phrasal category, as implied by the remnant movement analysis, it should not be possible to generate new interpretations. This is so because the node containing the modal (XP in (3a)) would either prevent the modal from c-commanding quantifiers it has crossed over (3a), or XP would have to reconstruct (see (3b)). None of these options derives the scope order modal ™ QP (for details see Lechner 2005b and Lechner, in preparation). (3)

a. [[XP …Modal…] … [[QP… tXP … (where XP is not a projection of the modal) b.

… [[QP… [XP …Modal…]

Finally, SAHM also directly entails that HM cannot be epiphenomenal, as maintained by Brody (2000). Below, I will present a chain of evidence embedded in a discussion of socalled scope splitting phenomena in English that supports the SAHM conjecture. Section 2 introduces the core data. In section 3, I present some background assumptions concerning the LF position of nominal quantifiers and the syntax of modality. Section 4 and 5 assemble the argument, while section 6 expands on independent support for parts of the specific analysis to be advanced.

2. Scope splitting Empirically, the argument for SAHM is centered around a particular group of modal constructions exemplified in (4) to (6) below.4 The most prominent reading of these sentences denies the compatibility of a universal proposition

An interpretive effect of head movement

47

with a circumstantial modal background. For instance, (4) means that it is not possible that every pearl is above average size, a proposition which is analytically true given the logical impossibility of mapping all pearls to a degree above the mean. This reading correlates with the scope order 5
™ œ, in which the negation is separated from its surface host, the quantifier every. (In addition, the examples can be construed de re, and lack de dicto readings.) (4)

Not every pearl can be above average size. “It is not possible that every pearl is above average size”

5
™ œ

(5)

Not everyone can be an orphan. (André Gide) “It is not possible that everyone is an orphan”

5
™ œ

(6)

Not every boy can make the basketball team. 5
™ œ “It is not possible that every boy makes the basketball team”

(4) to (6) represent instances of the so-called scope splitting construction. In the literature5, scope splitting (or ‘negative split’) has been extensively discussed on the basis of examples such as (7), which differ from (4) to (6) in that negation is combined with a quantifier that carries existential and not universal force. (The negative QP in (7) is also construed as the object in order to avoid further complications with subject reconstruction; see 3.1.) (7)

Sam can find no solution. “It is not possible that Sam finds a solution”

5
™ ›

For reasons of concreteness, I will adopt the analysis of scope splitting formulated in von Stechow (1993) and Penka (2002),6 according to which all negative indefinites bear a syntactic feature [+neg] which has to be licensed in the immediate scope of a possibly abstract semantic negation (NOT). The morphologically negative NPs themselves are assigned the meaning of their contradictories (e.g. solution for no solution and everyone for not everyone). Scope splitting is induced by configurations in which the abstract negation NOT is separated from the negative NP by another operator at LF. To exemplify, (7) can be parsed as in (8). In the LF (8a), the feature [+neg] is licensed by the abstract clausal negation NOT in SpecNegP. Since the modal intervenes between NOT and [+neg], interpreting (8a) consequently leads to the split scope order (8b), in which the morphologically negative object is translated as an indefinite. (Acc denotes the accessibility relation):

48 Winfried Lechner (8)

a. Sam [NegP NOT [can find [no solution][+neg]]] 5
™ › b. w5›w’›x[Acc(w)(w’) v solution(x)(w’) v find(x)(Sam)(w’)]

As already mentioned, (4) to (6) essentially differ from the classic instances of splitting such as (7) in that the negation associates with a universal, and not with an indefinite. This particular contrast will become important below, as universal QPs are subject to different, stricter conditions on where they can be interpreted in the tree (see 3.1).7 Following a discussion of the relevant background assumptions in section 3, I will proceed to argue in section 4 that mapping the split readings of (4) to (6) onto a syntactic structure has nontrivial consequences for the analysis of HM.

3.

Mapping syntax to interpretation

The present section will introduce two specific properties of the mapping procedure from syntax to interpretation, both of them involving empirical generalizations about the way in which movement interacts with interpretation. In section 3.1, I will comment on differences in the reconstruction behavior of different logical types of quantifiers. These findings will be used to set the lower structural bound for the interpretation of subjects and (by transitivity) categories that scope over subjects. Section 3.2 addresses the dissociation between the surface position and the base position of modals, presenting evidence for the view that modals are generated below the position in which they surface. According to this conception, the ordering of modals and other categories in the tree is derived by movement, and not by the availability of alternative insertion points for the modal or certain adverbs and negation. These results form the basis of the argument for semantically active HM (SAHM) to be spelled out in section 4.

3.1. The Strong Constraint When quantified terms (QPs) surface in A-positions that do not correspond to their points of origin, these QPs can sometimes be assigned scope below their overt positions. Although the exact mechanisms underlying ‘scope diminishment’ – borrowing a term from von Fintel and Iatridou (2004) – are poorly understood, there is an emerging consensus that the logical properties of the moved QP co-determine its ability to reconstruct. More specifically, NPs that

An interpretive effect of head movement

49

have undergone A-movement fall into one of three groups: negative NPs, strong NPs (in the sense of Milsark 1974) and indefinites. Postponing the discussion of negative NPs until section 6, indefinites and strong NPs differ in that raised indefinite subjects may be assigned a narrow scope interpretation w.r.t. the raising predicate (see (9)), while strong quantifiers cannot be construed with scope below the intensional operator, as illustrated by (10), and probably more clearly by (11) (see Iatridou 2002; Lasnik 1999; Lechner 1996, 1998; Wurmbrand and Bobaljik 1999, among many others). (9)

a. A critic seemed to like the movie. b. It seemed that a critic liked the movie.

de re/de dicto de dicto

(10) a. Every critic seemed to like the movie. b. It seemed that every critic liked the movie.

de re/*de dicto de dicto

(11) a. Every movie which was promoted by a critic seemed to impress the jury. b. It seemed that every movie which was promoted by a critic impressed the jury.

de re/*de dicto de dicto

For (11a) to be true, the individuals promoting the movie must be actual critics in the evaluation world, whereas (11b) leaves open the option that these individuals only appeared to be critics – it could turn out that in fact, they were radical Christians. Provided that the absent de dicto interpretation of (11a) is contingent upon reconstruction of every movie along with the relative clause, it can be concluded that universals do not reconstruct below seem. Further confirmation for the assumption that strong quantifiers resist reconstruction into the subordinate clause comes from contexts involving non-verbal intensional operators as in (12) (Lasnik 1999: 93). Sentence (12a) contradicts the laws of probability, while (12b) is evaluated as true in a situation with five events of tossing (
5 = 1/32 . 3%). Again, the absence of an equivalent reading for the raising construction (12a) indicates that strong QPs have only limited access to scope diminishment: (12) a. Every coin is 3% likely to land heads. œ ™ likely/*œ ™ likely b. It is 3 % likely that every coin will land heads. A first version of the condition on strong NPs that was seen to be active in (10) to (12) can be formulated in terms of the Strong Constraint in (13):

50 Winfried Lechner (13) Strong Constraint (1st version) Strong QPs do not reconstruct below raising predicates. The specifics of (13) still require a minor amendment, though. In particular, the same constraint which is responsible for prohibiting reconstruction in the raising-to-subject constructions (10) to (12) should presumably also operate in contexts in which raising arguably proceeds to an object position. Following Lasnik (1995), such environments are prominently exemplified by ECMconfigurations like (14). As documented by (14a), the ECM-subject has raised into the matrix sentence in overt syntax (but see also Lasnik 2005 for a diverging view): (14) I expected everyone not to be there. œ ™ 5/5 ™ œ a. I expected1 [XP everyone2 [VP t1 [NegP not [TP t2 T° to be …]]]] b. I [XP everyone [VP expected [NegP not [TP everyone to be …]]]] Still, the universal retains the ability to be construed within the scope of the negation. This is unexpected from the perspective of (13) inasmuch as in order to generate the inverse reading (14b), the ECM subject would have to reconstruct across the raising predicate expect, which itself is restored into its base position at LF. Thus, the Strong Constraint in (13) is too restrictive as it rules out the inverse reading (14b), and will therefore be revised as in (15): (15) Strong Constraint A strong NP cannot reconstruct below T°. According to (15), reconstruction in (14b) is licit because the universal does not cross T°. Thus, (15) tolerates limited applications of reconstruction as in (14), while the standard manifestations in (10) to (12) are not negatively affected by the changes in the revised version. In order to parse the scope splitting constructions in (4) to (6) into a tree, one last ingredient is still missing. Section 3.2 expands on this issue by providing a strategy for determining the structural position of modals, while section 4 will finally present the synthesis of the argument for SAHM.

3.2. The position of modals There is good reason to believe that English modals are generated in a VPexternal position, from where they move into a higher head which is located

An interpretive effect of head movement

51

above clausal negation and (some aspectual) adverbs. It is arguably the effect of this movement which carries the modal to the left of not, always and never in (16) to (18). In what follows, I will focus on examples involving negation, as in (16); the generalization carries over to contexts with adverbials, though. (16) John can1 not t1 come along today. (17) He can1 always t1 count on me. (18) He can1 never t1 do that.

5 ™
/??
™ 5 always ™
/*
™ always never ™
/*
™ never

A first indication that modals are indeed generated in a position below negation and (certain aspectual) adverbs comes from the observation that modals precede these operators yet display a strong preference for narrow scope (Lerner and Sternefeld 1984; Öhlschläger 1989). This dissociation of surface position and scope is straightforwardly captured by an analysis that adopts low base-generation, movement and reconstruction. For ease of reference, the base and the derived position of the modal will be identified with T° and AgrS°, respectively.8 The choice of labels does however not play a substantive role in the development of the argument. In order to establish that the scope order in (16) is actually the product of modal raising and reconstruction, it must be ascertained that scope reversal does not result from an alternative derivation in terms of covert movement of negation (‘Neg-Shift’) across the modal. A strong argument against Neg-Shift is furnished by examples such as (19). (19) includes a positive polarity item (PPI) (sometimes) which is assigned wide scope w.r.t. the modal, which in turn takes scope below not: (19) It can sometimes not be avoided to confront the enemy. sometimes ™ 5
The critical property of the PPI in (19) consists in its ability to introduce two additional scope criteria which will be seen to exclude a Neg-Shift analysis. First, the PPI must stay outside the scope of negation. Combined with the narrow scope tendency of circumstantial modals, this requirement leads to the scope order sometimes ™ 5
. The LF underlying this reading can now either be attributed to reconstruction of the modal, as documented by (20), or to Neg-Shift followed by covert movement of the PPI sometimes, as in (21):

52 Winfried Lechner (20) Derivation A: modal reconstruction a. Surface order: [AgrSP it can1 [sometimes [NegP not [TP t1 b. Reconstruction [sometimes [NegP not [TP can (21) Derivation B: Neg-Shift a. Surface order: [AgrSP it can1 [sometimes [NegP not [TP t1 b. Neg-Shift: [XP not2 [AgrSP it can1 [sometimes [NegP t2 [TP t1 c. QR of ‘sometimes’: [YP sometimes3 [XP not2 [AgrSP it can1 [t3 [NegP t2 [TP t1 Second, Szabolcsi (2002) discusses a property of an intriguing class of PPIs which makes it possible to adjudicate between the two competing analyses in (20) and in (21). She observes that the weak indefinite PPI somewhat in (22) has to satisfy two conflicting requirements simultaneously. As a PPI, somewhat would have to escape the scope of negation by Neg-Shift. However, being a weak indefinite, somewhat must not cross the negative island established by negation, resulting in an ill-formed output string: (22) *John doesn’t appreciate this somewhat Sometimes in (19) behaves now just like somewhat, as illustrated by (23). This is expected since sometimes is also interpreted as a weak indefinite. (23) *John didn’t sometimes come to class. But given that sometimes must not covertly move across negation in (23), it should not be able to do so in (19), either. Moreover, since the success of the Neg-Shift derivation in (21) crucially depends on the ability of the adverb to cross negation subsequent to Neg-Shift (21c), it also follows that (19) cannot be the result of Neg-Shift. Thus, modals must be allowed to reconstruct, as implied by the derivation in (20). With these assumptions in the background, the next section proceeds to the argument for the view that HM may affect interpretation (SAHM).

4. Analysis Above, it was seen that strong quantifiers do not reconstruct freely and that negated universals partake in scope splitting in examples such as (6):

An interpretive effect of head movement

(6)

Not every boy can make the basketball team.

53

5
™ œ

These two observations supply evidence for the view that the LF-position of the subject in (6) c-commands the node in which the modal originates. It follows that the modal, which takes scope over the universal, has to be interpreted above its base position (T°), generating an argument in support of SAHM. Recall that negative NPs bear a feature [+neg] which has to be locally ccommanded by the abstract negation NOT (see (24a)): (24) a. Syntax:

[+neg] must be in the local scope of the possibly abstract negation NOT. b. Semantics: ƒ[Not every NP]„ = ƒ[every NP]„ (adopted from Penka 2002; Penka and von Stechow 2001; von Stechow 1993)

If the NP[+neg] precedes its licensing head, as is the case in (6), the subject must reconstruct in order to satisfy (24a), and a lower copy is submitted to interpretation (25a). (25) a. [Not every boy]PF NOT can [not every boy][+neg] make the team. b. ƒNOT„ can ƒnot every boy„ make the team. c. ƒNOT„ can ƒevery boy„ make the team. 5
™ œ After the negative feature has been eliminated, the interpretive convention (24b) further regulates the transition from the morphologically negative QP in (25b) to its contradictory in (25c), yielding the split reading. The representations in (25) do not reveal yet the exact location in which the subject is interpreted. It can be shown, though, that the subject is parsed into a position at LF that c-commands the node in which the modal originates, and that the modal therefore has to be interpreted in a derived position. The tree in (26) below provides the relevant details for (6). Starting bottom-up, not every boy is submitted to PF in SpecArgSP. In order to conform with the licensing conditions on its negative feature, not every boy cannot be interpreted in its surface position, but has to reconstruct into the scope of the abstract negation NOT.9 However, the subject must not reconstruct too low, either, as scope diminishment of strong NPs below T° is prohibited by the Strong Constraint. The lowest interpretable subject copy is consequently located above T°. On the most parsimonious clausal architecture, this position can be identified with SpecTP. Thus, the subject is interpreted in

54 Winfried Lechner SpecTP10 (marked by L in (26)). Since TP is above the base position of the modal, and the modal takes scope above the universal in the split reading (scope order 5
™ œ), it follows now that the split reading derives from interpreting can in a derived head (Neg°).11 Assuming that the trace of HM in (26) is semantically vacuous, the raising modal in Neg°, which denotes a propositional operator of type <<s,t>, <s,t>>, can furthermore directly combine with its sister node (TP). (26) a. Not every boy can make the team.

5
™ œ

b.

AgrSP<s,t> ei [not every boy]PF NegP<s,t> ei ƒNOT„ Neg’<s,t> ei ƒcan2„<<s,t><s,t>> TP<s,t> ei L ƒnot every boy1„ TP<e<s,t>>> 3 1 T<e,t™ 3 t2 vP/VP<s,t> 3 Strong constraint º *ƒ(not) every boy1„ 6 ƒt1„ make the team 5
œx[boy(x) ÿ make_the_team(x)] On this particular interpretation of the data, HM may have an effect on interpretation, eliciting a first piece of evidence in support of SAHM. Moreover, the analysis contradicts a PF-account of HM, because HM operations that apply at PF should not be able to alter scope relations. Thus, (26) provides an argument for the conservative conception that HM is produced by displacement of terminals in the syntactic component. 5. The LF-position of the subject The validity of the argument in favor of SAHM laid out in the section 4 rises and falls with the accuracy of the tools that are used to locate the interpretive position of the subject in (26). In particular, the evidence is conclusive only if

An interpretive effect of head movement

55

it can be shown that the LF position of the subject is above the position in which the modal originates. In (26), this conclusion was secured by the additional assumption that the vP is immediately merged with TP, with no other functional projections disrupting the spine of the tree. If, contrary to this assumption, vP and TP were separated by further categories (AspP, PerfP, AuxP; see e.g. Cinque 1999) which may host additional intermediate copies of the subject, it evidently has to be demonstrated that the subject is not interpreted in one of these intermediate landing sites. Otherwise, the modal could be translated in its base position, invalidating the evidence for SAHM.

5.1. Preliminary considerations A first argument in this direction can be derived from the selectional properties of raising modals. On a widely shared assumption, raising modals embed small clauses (Stowell 1983, 2004). Moreover, as initially observed in Stowell (1981) and Williams (1983), small clauses are scope islands, they minimally contrast with clausal complements in that their subjects cannot be construed de dicto (see Johnson 2001 for an account): (27) a. A linguist seems to be unhappy. b. A linguist seems unhappy.

de re/de dicto de re/*de dicto

Turning to the scope splitting example (26), the combination of these two premises is now sufficient in order to exclude subject reconstruction into specifiers inbetween TP and vP. More precisely, transposing the small clause analysis to raising modal constructions entail that all nodes c-commanded by the base position of the modal (K) are scope islands for the subject: (28) No reconstruction: Subjectk … [Modalbase position K[Small Clause tk …]] On this view, the presence of other projections between TP and vP in (26) turns out to be immaterial for the strength of the argument. In fact, adopting the analysis above has the even more radical consequence of rendering reconstruction below the base position of raising modals generally impossible. If correct, it would therefore follow that all de dicto readings below circumstantial modals derive from interpreting the modal in a derived position. Although this at first sight looks like an attractive feature of the analysis that provides further support for SAHM, the generality of the claim makes it hard to be falsified, and therefore weak in its empirical foundation.

56 Winfried Lechner Moreover, the analogy between small clauses and complements of raising modals is less straightforward than one might hope. First, raising modals allow de dicto readings for weak subjects, while it was the absence of such readings in (27b) which formed the basis of the scope island hypothesis. Second, the evidence for treating complements of raising modals as small clauses is, as far as I know, not very strong, and requires further empirical justification. For at least these reasons, it would be advantageous if it were possible to find independent support for the claim that the subject in (26) is interpreted no lower than in SpecTP. There is an additional piece of evidence indicating that this view might be correct, which will be outlined in the following section.

5.2. Negative polarity licensing A strong argument in support of interpreting subjects high can be distilled from split scope configurations which include Negative Polarity Items (NPIs). Observe to begin with that negated universals license NPIs if the QP is assigned surface scope ((29) from Horn 2000: (49b); see Penka 2002: fn. 37): (29) Not everyone who works on negation has ever read any Jespersen. 5œ ™ NPI Interestingly, scope splitting appears to conflict with NPI-licensing, as demonstrated by the deviance of (30). The relative scope of the NPI w.r.t. the universal and the modal (to the degree that they are logically independent) does not affect acceptability judgements; the example is ill-formed in all split interpretations. (30) *Not everyone can ever be on the team. 5(™NPI) ™
(™NPI) ™œ (™NPI) Moreover, (31) testifies to the fact that modals (and split indefinites) do not interfere with NPI-licensing, implying that the degraded status of (30) should not be blamed on the presence of can (see also (33) below; see also von Fintel and Iatridou 2005: 21f, who reach the same conclusion). (31) Nobody can ever be on the team.

5› ™
™ NPI/5
™ › ™ NPI

A split reading for (30) is arguable unavailable for the same reason that the paraphrase of (30) given in (32) is ill-formed:12

An interpretive effect of head movement

57

(32) *It is not possible that everyone will ever be on the team. 5
™ œ ™ NPI In both cases, a universal intervenes between negation and the NPI. Removing the quantifier salvages (32), as shown by (33): (33) It is not possible that you will ever be on the team.

5
™ NPI

Thus, it is tempting to relate the absence of the split interpretation in (30) to the same condition which is usually evoked in handling contrasts such as (32) vs. (33), or those illustrated in (34) and (35): Linebarger’s (1980) Immediate Scope Constraint (see also Guerzoni, to appear). (34) a. He didn’t like anything. b.*He didn’t always like anything. (35) a. I didn't want her to eat any cheese. b.*I didn't want every boy to eat any cheese.

(Linebarger 1987) * 5 ™ œ ™ NPI (Linebarger 1980: 29) * 5 ™ œ ™ NPI

For this analysis to succeed, the subject of (30) has to reconstruct into a position above the NPI, from where it can disrupt the relation between the negation and ever, triggering a violation of the Immediate Scope Constraint. But before the subject can be linked to a specific node in the tree that meets this condition, it is necessary to identify the attachment site of the NPI ever. Ever and always are both aspectual modifiers, but they are not in strict complementary distribution. If they cooccur, ever needs to precede always, as the Immediate Scope Constraint might lead one to expect: (36) a. No one source is ever always authoritative. b.*No one source is always ever authoritative. (37) a. Where in the world is it ever always easy? b.*Where in the world is it always ever easy? Furthermore, always takes scope above modals to its left (see section 3.2), indicating that always originates as a TP-adjunct, and that modals optionally reconstruct below always: (17) He [AgrSP can1 [TP always [T’ t1 count on me]]]

always™
/*
™always

58 Winfried Lechner According to the ordering generalization extracted from (36) and (37), ever is located higher than always. Together with the scope fact (17) this entails that ever is generated as a TP-adjunct or as an adjunct above TP. It also follows now that the subject in (30) has to be squeezed inbetween the negation and the TP-adjunct ever in order to induce a violation of the Immediate Scope Constraint. One way to arrive at the desired structural configuration for the application of the Immediate Scope Constraint consists in parsing the subject copy into an outer specifier of TP, from where it impedes NPI licensing, as shown by (38). This derivation excludes (30) by assuming that the split reading represents the scope order 5 ™
™ œ ™ NPI: (38)

*AgrSP * 5 ™
™ œ ™ NPI wo not everyone1 NegP wo ƒNOT„ Neg’ wo ƒcan2„ TP wo ƒnot everyone1„ TP wp Y Immediate Scope Constraint º ƒeverNPI„ … t2 … t1…

Next, the absence of the alternative split scope order 5 ™ NPI ™
™ œ encapsulates the crucial (reductio) argument against long subject reconstruction (and for SAHM). Suppose that the subject in (30) had the option to be interpreted below TP, in the specifier of some intermediate XP, as shown by (39). In this alternative derivation, the modal is located in its base position T° at LF, and no category intervenes between the NPI and its licensing negation. As a result, (39) observes the Immediate Scope Constraint. Hence, if the surface string (30) could be parsed as in (39), one would wrongly be led to expect that (30) can be assigned a split reading (5 ™ NPI ™
™ œ):

An interpretive effect of head movement

(39)

59

*AgrSP * 5 ™ NPI ™
™ œ wo NegP not everyone1 wo ƒNOT„ TP wp TImmediate Scope Constraint º ƒeverNPI„ T’ wp ƒcan„ XP wo Y Strong Constraint º ƒeveryone1„

The unavailability of scope splitting for (30) therefore furnishes evidence for the view that the universal subject is interpreted in SpecTP, as in (38). The illformedness of (39) can then be attributed to the Strong Constraint, which blocks reconstruction of strong NPs below T°. Considerations similar to the ones above also exclude the third logically possible scope order for (30) (5 ™
™ NPI ™ œ). This last available derivation differs from (39) minimally in that the modal is interpreted in its derived position Neg°. Just like (39), this reading cannot be produced due to illicit long subject reconstruction into SpecXP. Note in passing that the argument above has the objective of securing the LF-position of the subject, and is not concerned with extracting from the data direct support for SAHM. If structures that violate a syntactic constraint, but are otherwise well-formed (such as (38)) are nonetheless assigned a semantic value, the modal in (38) might indeed be interpreted in a derived position. Whether this view turns out to be correct or not is orthogonal for the soundness of the main argument, though. To summarize, the interaction of scope splitting and NPIs provides independent evidence for the claim that universal subjects cannot be interpreted below TP. As explicated in section 4, this finding furthermore implies that some modal contexts manifest instances of SAHM.

6. Negative Quantifiers This final section elaborates on some consequences that the analysis for negative universals presented in sections 3 and 4 entails for the treatment of reconstruction in related constructions involving negative indefinites.

60 Winfried Lechner Negative quantifiers are generally held to resist reconstruction, as demonstrated by the fact that the subject of the proposition expressed by (40) can only be understood de re (see von Fintel and Iatridou 2004; Iatridou 2002; Lasnik 1999; Wurmbrand and Bobaljik 1999, a.o.): (40) a. No critic is certain13 to like the movie. b. It is certain that no critic likes the movie.

de re/*de dicto de dicto

The contrast manifest in (41) and (42) indicates that the scope options of negative NPs are even more restricted than those of strong NPs. Unlike strong NPs, which may undergo short reconstruction below negation (41), negative quantifiers seem to lack inverse scope readings all together (42): (41) a. Every guest didn’t show up. b. All that glitters isn’t gold. (42)

No guest didn’t show up.

œ ™ 5/5 ™œ (Lasnik 1972) 5›5 ] œ/*55› ] ›

While scope diminishment with negative subjects is severely limited, the present analysis must on the other side not entirely prohibit reconstruction of negative NPs, either. More precisely, on current assumptions, reconstruction of negative subjects is indispensable in order to satisfy the local scope requirement of their [+neg]-feature at LF.14 In the scope splitting example (26), for one, the [+neg]-feature on the subject can only be licensed subsequent to reconstruction into SpecTP. The maximal depth at which a negative subject may lower is determined by independent principles. Negative universals fall under the reign of the Strong Constraint (see (26)), and are therefore interpreted no lower than in SpecTP. Moreover, it must be ensured that negative NPs do not reconstruct to a position below other scope bearing categories. In particular two instances of scope diminishment have to be excluded in this domain: lowering of a negative indefinite into a subordinate clause (40) and reconstruction across negation (42). As will be demonstrated below, the [+neg]-feature analysis can also be extended to these cases. Turning to the monoclausal structures involving negation first, the unattested narrow scope reading for the subject in (42), repeated in (43a), is contingent upon a representation in which the subject reconstructs to a position below NOT, as in (43b). Since not occupies Neg°, both the head and the specifier of NegP are now filled:

An interpretive effect of head movement

61

(43) a. No guest didn’t show up. *55› ] ›/5›5 ] œ *55› ] › b. [NegP NOT [Neg’ not [TP [no guest][+neg] show up But such ‘doubly-filled NegP’ configurations fail to satisfy the well-formedness conditions on negative contexts in English. More specifically, in double negation languages such as (standard) English, the relation between morphologically negative lexical expressions (negative NPs, the negative particle not, …) and logical negations (NOT in SpecNegP) is biunique.15 That is, every negative expression has to be associated with a NOT of its own. A single logical negation (NOT) in (43b) can therefore not simultaneously license two morphologically negative expressions (no guest and not), blocking the reconstructed interpretation.16 It also follows from the above that for the derivation to converge, both no guest and not have to be supplied with an individual licensor. Representation (44) satisfies this requirement by projecting an additional NegP (NegP2) that holds a second occurrence of the logical negation NOT. In (44), NegP2 is located above NegP1,17 and each negatively marked expression is matched with a logical negation (a higher one in NegP2 for no guest, and a lower one in NegP1 for not ), resulting in the attested scope order 5›5: (44) [NegP2 NOT [[no guest][+neg] [NegP1 NOT [Neg1’ do4-not [TP t4 … 5›5 ] œ In sum, the absence of reconstruction of negative NPs across negation can be made to follow from the specifics of the licensing condition on [+neg] and the syntactic well-formed conditions on negation. Turning to long reconstruction of negative NPs next, the subject of (40), repeated below as (45), can in principle be interpreted in three different positions: (45) No critic[+Neg] is certain to like the movie.

= (40)

a. [NegP NOT is [TP [no critic[+Neg]] [VP/AP certain to like the movie 5› ™ certain b.*[NegP NOT is [TP [VP/AP certain [TP [no critic[+Neg]] to like the movie 5 ™ certain ™ › c. *is [TP [VP/AP certain [NegP NOT [TP [no critic[+Neg]] to like the movie certain ™ 5› In the well-formed representation (45a), which underlies the surface scope reading, no critic undergoes short reconstruction to the matrix SpecTP, licensing the [+neg]-feature which now resides within the scope of NOT at LF. The

62 Winfried Lechner parse in (45b) minimally differs from (45a) in that the subject has been reconstructed into the lower clause, resulting in an unattested split reading across a raising predicate. The structure is excluded by the locality requirement on [+neg] licensing, which demands that the feature reside within the scope of clause-mate negation. Finally, in (45c), the subject reconstructs into the lower SpecTP again. But this time, the abstract NOT is generated in the embedded clause, too, in accordance with the locality conditions on [+neg]. At first impression, one is therefore led to expect that the derivation should converge, providing an LF for the unattested narrow-scope de dicto interpretation. The LF in (45c) falls short of satisfying an independent criterion on derivations, though, which is usually referred to as the Improper Movement Constraint, and which rules out certain combinations of A- and }- dependencies. (46) provides a version that transposes the traditional concept of A/}-movement into the currently more popular Agree-based system: (46) Improper movement constraint (Agree-based version) If a category C partakes in an }-Agree dependency at node n, it must not enter into an A-dependency at a node that dominates n. Classic instances of improper movement prototypically involve wh-movement, where the constraint e.g. excludes subsequent applications of wh-movement and raising of one and the same category. Moreover, on a widely accepted view, the distribution of [+neg] features is governed by principles similar to the ones which are thought to be responsible for the licensing of wh-phrases. Haegeman and Zanuttini (1996), for one, express various restrictions on negative NPs by appealing to the Neg-Criterion, which they define in analogy to the wh-criterion of Rizzi (1991).18 It is therefore only natural to expect that the application of (46) also has empirical manifestations in the domain of negative licensing. I would like to suggest that exactly such a case has been identified above in the guise of (45c). As detailed by (47), the subject of (45c) enters both an }-dependency ([+neg]-licensing) and an A-dependency (raising). Moreover, the node containing the }-relation (L) is dominated by the node which demarcates raising (K). Thus, the derivation violates the improper movement constraint, and the missing narrow scope reading (45c) cannot be generated: (47) a. [+neg]-licensing: [TP [certain [NegP NOT L [TP [no critic[+Neg]] b. Raising: K [AgrSP [no critic[+Neg]] is [TP [certain [NegP NOT [TP to …

An interpretive effect of head movement

63

Notice in this context that the system has enough flexibility for adjustments in order to counter the potential objection that the creation of the }-dependency in (47) does not derivationally precede raising, but is delayed at LF. One way to remove this apparent disparity between Improper Movement with [+neg]features and wh-phrases consists in assuming that negative NPs check their [+neg]-features already in overt syntax (as e.g. in Penka 2002), but that they also need to satisfy an independent scope requirement at LF which then drives reconstruction. The latter might be similar to that found with NPIs. On an alternative implementation, negative NPs are endowed with two features, which have to be eliminated in overt syntax and at LF, respectively. To recapitulate, negative NPs can in principle reconstruct, but the effects of lowering are detectable only in a limited set of contexts involving scope splitting (see (26)). The general resistance of negative NPs to partake in scope diminishment can furthermore be linked to the interaction between the [+neg]licensing criterion and syntactic factors (e.g. improper movement constraint).

7. Conclusion The present paper presented an attempt at isolating an argument in favor of the view that certain instances of HM must be computed in the syntactic component (SAHM). If HM can be shown to uniformly display the same behavior w.r.t. its defining characteristics, this implies that a PF-analysis of HM is not viable (contra e.g. Boeckx and Stjepanovic 2001; Chomsky 2000, 2001; Harley 2005). The search for SAHM also produced some new evidence for particular analyses of three phenomena. First, the interaction of PPI licensing and negation was seen to provide a new argument for generating modals low and moving them to a higher head position in overt syntax (contra non-derivational approaches such as Cormack and Smith 1998, 1999, a.o.). Second, the discussion yielded new diagnostics from NPIs for identifying the position in which subject NPs are submitted to interpretation. Third, the deliberations resulted in a novel way for expressing restrictions on scope diminishment. On the one hand, the inability of negative NPs to undergo reconstruction was attributed to an LF-licensing requirement on the [+neg]-feature. This approach not only offers the advantage of deriving (substantial parts of) the behavior of negative NPs from independent principles, but also supports the hypothesis that negative NPs are semantically decomposed into their contradictories and an abstract negative symbol NOT (von

64 Winfried Lechner Stechow and Penka 2001; Penka 2002). On the other hand, the reconstruction properties of strong NPs led to the formulation of a specific descriptive generalization blocking scope diminishment below T° for this group of NPs.

Acknowledgements I am indebted to Elena Anagnostopoulou, Marcel den Dikken, Kleanthes Grohmann, Idan Landau, Howard Lasnik, Roland Hinterhölzl, Henk van Riemsdijk, Wolfgang Sternefeld, Uli Sauerland and in particular Orin Percus for discussion, and the audiences of GACL 1 (Cyprus), Incontro di Grammatica Generativa XXXI (Rome), and Glow 2005 (Geneva) for comments. Kyle Johnson, Chris Kennedy, Jason Merchant and Orin Percus kindly provided judgements. A long version of this paper circulates as Lechner (2005a).

Notes 1. For any , denotation brackets (‘ƒ„’) signal where  is interpreted. 2. For evidence that HM can be affected by semantic principles see Lechner (2001, 2004: ch. 3.4). For further discussion of arguments pro and contra HM see Fanselow (2002); Matushansky (2005); vanRiemsdijk (1998); Roberts (2004); Zwart (2001), among others. For the claim that V2 has semantic effects see Benedicto (1994). 3. If  is a PF-operation, then  does not have an impact on interpretation.  has an impact on interpretation. Hence,  is not a PF-operation. 4. The use of negative universals instead of universals or negative indefinites in the design of the argument is motivated in Lechner (2005a). 5. See Bech (1954/57: § 80); von Fintel and Iatridou (2005); Heim (2000); Kratzer (1995); de Swart (2000); Penka (2002); Zeijlstra (2004) and references. 6. See also Kratzer (1995: 144) and Penka and von Stechow (2001). Von Stechow and Geuder (1997: 19) credit Ede Zimmermann with the first explicit semantics. 7. For evidence that the split reading cannot be logically subsumed under the independently available de re interpretation (5œ ™
) see Lechner (2005a). 8. Johnson (2003: 68ff) supplies further motivation for locating NegP between TP and AgrSP and allowing modals to move from T° to AgrS°. For discussion of the syntax of negation see Zeijlstra (2004: 167ff). Cormack and Smith (1998, 1999, a.o.) present a non-derivational account for mismatches between scope and order. 9. For evidence that not every boy forms a constituent see Lechner (2005a).

An interpretive effect of head movement

65

10. Sauerland (2003) presents evidence that subjects can be reconstructed into a low A-position he identifies with vP. As Sauerland demonstrates, this position must be below negation (see (i)), but still high enough to bind an experiencer (see (ii)): (i) Every student mustn’t get an A. ~ ™ 5œ (Sauerland 2003: 309; (4)) (ii) Every child2 doesn’t seem to his2 father to be smart. 5 ™ œ (ibid, 310; (7)) Sauerland’s insight can be maintained in the present system if it is assumed that the modal in (i) is interpreted in AgrSP, the subject in SpecTP, and negation in NegP. 11. The surface interpretation 5œ ™
arises from interpreting the subject in SpecTP and the modal in-situ, i.e. in T°. 12. It is orthogonal for present purposes whether (32) is well-formed on a non-split interpretation. According to informants, this is not the case, raising the additional question why (32) contrasts with (29). 13. seem treats clause-mate negation semantically as if it were part of the lower proposition (‘Neg-raising’; (i)). Neg-raising is also attested with negative subjects, as in (ii) (from Kayne 1998: fn. 26): (i) ƒJohn does not seem to be there„ ] ƒIt seems that John is not there„ (ii) ƒNobody seems to be there„ ] ƒIt seems that nobody is there„

14.

15. 16.

17.

The change from seem to is certain in the examples in the main text makes it possible to avoid interference from Neg-raising in the scope judgements. Penka and von Stechow (2001), who focus on German, restrict (24b) to surface syntax. This difference application might be due to the general scope rigidity of German. In negative concord languages, a single NOT may check more than one [+neg]. I adopt the standard premise that each SpecNegP can hold only a single NOT - i.e. each NegP may be marked for polarity only once. The negative marker not can be taken to represent the spell-out of NOT, or can be assumed to bear a [+neg] of its own. The representation (i), in which NegP2 is projected below NegP1 can arguably be independently excluded. For instance, do in (i) would have to skip the lower negative head on its way to NegP1, in violation of the head movement constraint. (i) *[NegP1 NOT [do4-not [NegP2 NOT [Neg1° [TP [no guest][+neg] t4 ...

55› ] ›

Further support for the assumption that NegP2 is above NegP1 can be derived from typological evidence, which indicates that additional negative projections are located higher in the tree than classic sentence negation (see Zanuttini 1997). 18. For discussion see Zeijlstra (2004). The Neg-Criterion requires specifier-head relations, instead of scope (i.e. c-command at LF) to apply between the [+neg-]feature and the semantic negation. This difference is immaterial for present purposes, though.

66 Winfried Lechner References Bech, Gunnar 1955/57 Studien über das Deutsche Verbum Infinitum. Kopenhagen: Danske Videnskabernes Selskab. Benedicto, Elena E. 1994 AGR, phi-features, and V-movement: Identifying pro. In Functional Projections, Elena E. Benedicto and Jeffrey T. Runner (eds.), UMOP 17, 1–18. Amherst: Graduate Linguistic Student Association. Boeckx, Cedric and Sandra Stjepanovic 2001 Heading towards PF. Linguistic Inquiry 32 (2): 345–355. Brody, Michael 2000 Mirror theory: Syntactic representation in perfect syntax. Linguistic Inquiry 31 (1): 29–57. Chomsky, Noam 2000 Minimalist inquiries: The framework. In Step by Step: Essays on Minimalist Syntax in Honor of Howard Lasnik, Roger Martin, David Michaels and Juan Uriagereka (eds.), 89–155. Cambridge: MIT Press. 2001 Beyond explanatory adequacy. MIT Occasional Papers in Linguistics 20. Cinque, Guglielmo 1999 Adverbs and Functional Heads. Oxford: Oxford University Press. Cormack, Annabel and Neil Smith 1998 Negation, polarity and V-positions. Working Papers in Linguistics 10, 285–322. London: University College London. Cormack, Annabel and Neil Smith 1999 Where is a sign merged. GLOT International 4 (6): 20. Fanselow, Gisbert 2002 Münchhausen-style head movement and the analysis of verb second. Ms., University of Potsdam. von Fintel, Kai and Sabine Iatridou 2004 Epistemic containment. Linguistic Inquiry 34 (2): 173–198. 2005 Anatomy of a modal. Ms., Cambridge: MIT. Haegeman, Liliane and Raffaella Zanuttini 1996 Negative concord in West Flemish. In Parameters and Functional Heads, Adriana Belletti and Luigi Rizzi (eds.), 117–180. Oxford: Oxford University Press. Harley, Heidi 2005 Merge, conflation and head movement: The First Sister Principle revisited. In Proceedings of NELS 34, Keir Moulton and Matthew Wolf (eds.). Amherst: Graduate Linguistic Student Association.

An interpretive effect of head movement

67

Heim, Irene 2000 Degree operators and scope. In Proceedings of SALT X, Jackson Brendan and Tanya Matthews (eds.), 40–64. Ithaca, NY: CLC Publications. Hinterhölzl, Roland 1997 A VO-based approach to verb raising. In Proceedings of NELS 27, Kiyomi Kusumoto (ed.), 187–202. Amherst: Graduate Linguistic Student Association. Horn, Lawrence 2000 Pick a theory (not just any theory): Indiscriminatives and the freechoice indefinite. In Negation and Polarity, Larry Horn and Yasuhiko Kato (eds.), 147–193. Oxford: Oxford University Press. Iatridou, Sabine 2002 What I’ve learned about reconstruction in A-chains, Part I. Class handout, February 11, 2002, MIT. Johnson, Kyle 2001 Clausal edges and their effects on scope. Ms., University of Massachusetts, Amherst. 2003 In search of the middle field. Ms., University of Massachusetts, Amherst. Johnson, Kyle and Satoshi Tomioka 1997 Lowering and mid-size clauses. In Proceedings of the Tübingen Workshop on Reconstruction, Graham Katz, Shin-Sook Kim and Heike Winhart (eds.), SFB 340, Bericht Nr. 127, 185–206. Tübingen: University of Tübingen. Kayne, Richard 1998 Overt vs. covert movement. Syntax 1 (2): 128–191. Koopman, Hilda and Anna Szabolcsi 2000 Verbal Complexes. Cambridge, MA: MIT Press. Kratzer, Angelika 1995 Stage-level and individual-level predicates. In The Generic Book, Gregory Carlson and Francis Pelletier (eds.), 125–175. Chicago: The University of Chicago Press. Lasnik, Howard 1972 Analyses of negation in English. Ph.D. diss., MIT. 1995 Verbal morphology: Syntactic structures meets the Minimalist Program. In Evolution and Revolution in Linguistic Theory, Hector Campos and Paula Kempchinsky (eds.), 251–275. Georgetown: Georgetown University Press. 1999 Chain of arguments. In Working Minimalism, Samuel Epstein and Norbert Hornstein (eds.), 189–215. Cambride, MA: MIT Press. 2005 What is the Extended Projection Principle. Invited talk at GLOW 2005, April 2 2005. University of Geneva.

68 Winfried Lechner Lechner, Winfried 1996 On semantic and syntactic reconstruction. Wiener Linguistische Gazette 57–59. 1998 Two kinds of reconstruction. Studia Linguistica 52 (3): 276–310. 2001 Reduced and phrasal comparatives. Natural Language and Linguistic Theory 19 (4): 683–735. 2004 Ellipsis in Comparatives. Berlin, New York: Mouton de Gruyter. 2005a Interpretive effects of head movement. Ms., University of Stuttgart (available for download at LingBuzz). 2005b Syntactic and semantic effects of head movement. Talk at GLOW 2005, March 31–April 2 2005, University of Geneva. in prep. Two arguments against remnant movement. Lerner, Jean-Yves and Wolfgang Sternefeld 1984 Zum Skopus der Negation im komplexen Satz des Deutschen. Zeitschrift für Sprachwissenschaft 3 (2): 159–202. Linebarger, Marcia 1980 The grammar of negative polarity. Ph.D. diss., MIT. 1987 Negative polarity and grammatical representation. Linguistic Inquiry 1 (2): 325–387. Mahajan, Anoup 2000 Eliminating head movement. The GLOW Newsletter 44: 44–45. Matushansky, Ora 2005 Head-movement in linguistic theory. Linguistic Inquiry 37 (1): 69–109. Milsark, Gary 1974 Existential sentences in English. Ph.D. diss., MIT. Müller, Gereon 2004 Verb second as vP-first. Journal of Comparative Germanic Linguistics 7: 179–234. Nilsen, Øystein 2003 Eliminating positions. Ph.D. diss., Utrecht: OTS. Öhlschläger, Günther 1989 Zur Syntax und Semantik der Modalverben des Deutschen. Tübingen: Niemeyer. Penka, Doris 2002 ‘Kein’ muss kein Problem sein. MA-thesis, University of Tübingen. Penka, Doris and Arnim von Stechow 2001 Negative Indefinita unter Modalverben. In Modalität und Modalverben im Deutschen, Reimar Müller and Marga Reis (eds.), 263–286. Hamburg: Helmut Buske. van Riemsdijk, Henk 1998 Head movement and adjacency. Natural Language and Linguistic Theory 16 (3): 633–678.

An interpretive effect of head movement

69

Rizzi, Luigi 1991 Residual verb second and the Wh Criterion. Geneva Working Papers on Formal and Computational Linguistics 2. Republished 1996 in: Parameters and Functional Heads: Essays in Comparative Syntax, Adriana Belletti and Luigi Rizzi (eds.). 63–90. Oxford: Oxford University Press. Roberts, Ian 2004 Bare Head Movement. Talk presented at the ‘Incontro di Grammatica Generativa XXX’, University of Venice. Sauerland, Uli 2003 Intermediate adjunction with A-movement. Linguistic Inquiry 34 (2): 308–314. von Stechow, Arnim 1993 Die Aufgaben der Syntax. In Handbuch Syntax, Joachim Jacobs, Arnim von Stechow, Wolfgang Sternefeld and Theo Vennemann (eds.), 1–88. Berlin, New York: de Gruyter. von Stechow, Arnim and Wilhelm Geuder 1997 Sind manche Sprachen präziser als andere? Talk presented at Studium Generale, November 5, 1997. Stowell, Timothy 1981 Origins of phrase structure. Ph.D. diss., MIT. 1983 Subjects across categories, The Linguistic Review 2: 285–312. 2004 Tense and modals. In The Syntax of Time, Jacqueline Guéron and Jacqueline Lecarm (eds.), 621–637. Cambride, MA: MIT Press. de Swart, Henrëtte 2000 Scope ambiguities with negative quantifiers. In Reference and Anaphoric Relations, Klaus von Heusinger and Urs Egli (eds.), 109–132. Dordrecht: Kluwer. Szabolcsi, Anna 2002 Positive polarity – negative polarity. Ms., NYU. Williams, Edwin 1983 Against Small Clauses. Linguistic Inquiry 14 (1): 287–308. Wurmbrand, Susi and Jonathan Bobaljik 1999 Modals, raising and A-reconstruction. Talk given at Leiden University (October 1999) and at the University of Salzburg (December 1999). Zeijlstra, Hedde 2004 Sentential negation and negative concord. Ph.D. diss., University of Amsterdam. Zwart, C. Jan-Wouter 2001 Syntactic and phonological verb movement. Syntax 4 (1): 34–62. Zanuttini, Rafaella 1997 Negation and Clausal Structure. Oxford: Oxford University Press.

When we do that and when we don’t: A contrastive analysis of VP ellipsis and VP anaphora Carlo Cecchetto and Orin Percus

We compare the properties of VP anaphora with the properties of VP ellipsis. We focus on facts showing that VP ellipsis admits a kind of analysis that VP anaphora does not. We argue that a “semantic copying” analysis is the right one for the cases of VP anaphora we consider, while a “PF deletion” approach is correct for (most cases of) VP ellipsis.

1. Introduction VP ellipsis, exemplified by sentence (1), has been extensively studied since the seminal work of Sag (1976) and Williams (1977)1. A central question about VP ellipsis concerns the status of the VP category that seems to be missing in sentences like (1). One might group the existing proposals into three families: (i) the “PF deletion” approach; (ii) the “LF copying” approach; (iii) the “semantic copying” approach. (1)

If you order a pizza then I will order a pizza too

According to the PF deletion approach, a full-fledged VP like order a pizza is generated in the normal way in the “elided VP position” but goes unpronounced at PF. According to this approach, at every stage of the syntactic derivation the “missing VP” is actually present, so its presence should be detected. According to the LF copying approach, an unpronounced element with no internal structure (one could view it as a silent proform) is generated in the “elided VP position” and is replaced at the syntactic level LF by more structured syntactic material like order a pizza. According to this approach, the “missing VP” is present in the late stage of the syntactic derivation, so its presence should be detected at this stage but not in the earlier steps. Finally, according to the semantic copying approach, an unpronounced element with no internal structure is generated in the “elided VP

72 Carlo Cecchetto and Orin Percus position” and is simply interpreted in the semantic component as having the same meaning that a constituent like order a pizza happens to have. This means that in no step of the syntactic derivation is the “missing VP” present. Given that the goal in uttering a sentence like (1) is to communicate a full proposition, it is reasonable to assume that VP ellipsis occurs only when the missing material is in some sense recoverable. However, Hankamer and Sag (1976) have argued, using an independent diagnostic for what material is recoverable, that there are also further conditions on what the missing material in a VP ellipsis sentence can be: the “missing VP” must have the meaning of a VP that is identical or nearly so to a VP in the immediate context. The different approaches could see these further conditions in different ways. The PF deletion approach could say that a condition on PF deletion of a VP is that the VP must be (near-) identical to another VP in the immediate context. The LF copying approach could say that there is a comparable condition on what VPs can be copied at LF. The semantic copying approach could say that there is a comparable condition on how the unpronounced element can be interpreted. It is fair to say that, notwithstanding the substantial knowledge that has been accumulated on VP ellipsis in more than two decades of research, the choice among these three approaches is still open, at least to a certain extent. Our goal in this paper is to contribute to this debate. However, our approach will be oblique. Instead of addressing VP ellipsis directly, we will focus on a related, but distinct, construction, namely VP anaphora, which is exemplified in (2). The relevant interpretation of (2) is the one on which I order a pizza too. (2)

If you order a pizza then I will do that (too)

VP anaphora has played a role in the discussion of VP ellipsis. In particular, Hankamer and Sag used it in their argument that conditions on the meanings of “missing VPs” go beyond the recoverability condition. For Hankamer and Sag, the set of recoverable meanings was the set of possible meanings for a VP anaphor, and they showed that more meanings are available for VP anaphors like do that in (2) than for “missing VPs” like the one in (1). But VP anaphora has not received the kind of attention that VP ellipsis has, and this probably reflects the view that determining the meaning of a VP anaphor is entirely a matter of pragmatics.2 We will show in this paper (following Cecchetto 2004) that one can be more precise about what meanings

A contrastive analysis of VP ellipsis and VP anaphora 73

are available for VP anaphors, and more specifically that syntax constrains the available meanings. We will then show that, by comparing the properties of VP anaphora with the properties of VP ellipsis, we can draw conclusions about both constructions. Building on a discussion in Fiengo and May (1994), we will focus on the fact that not all VP ellipsis sentences have VP anaphora counterparts, and that not all meanings for VP ellipsis sentences are available for VP anaphora counterparts. These differences will clearly show that VP ellipsis admits a kind of analysis that VP anaphora does not. This paper is organized as follows. In section 2, we establish a strong similarity between VP anaphora and VP ellipsis: in the absence of a clear extralinguistic source for VP anaphora resolution, VP anaphors must have the meaning of another VP present at LF. (Our appendix reviews this claim and modifies it slightly.) In section 3, we discuss cases in which VP ellipsis is more liberal than VP anaphora and conclude that they argue against a semantic copying analysis for VP ellipsis and support this analysis for VP anaphora. In section 4, we compare the PF deletion and the LF copying approaches for VP ellipsis and we (tentatively) adopt the former. In section 5, we focus on cases in which VP ellipsis is less liberal than VP anaphora and note a consequence for the identity condition that holds between the antecedent and the missing VP. Section 6 briefly explores why VP ellipsis is not allowed in Italian. Section 7 concludes the paper, mentioning another lesson that can be learned from considering VP anaphora and VP ellipsis side by side.

2. VP anaphora is syntactically constrained VP anaphors can get resolved in a variety of ways. Take for instance the sentence in (2). On the one hand, the meaning of the VP anaphor can be identified with a salient action in the discourse context, one that does not relate to anything that has been explicitly said. For example, suppose Mary utters (2) (without “too”) in a context in which John has dropped all of his papers on the restaurant floor and is trying to pick them up while the waiter is patiently waiting for his order. It would be natural to take (2) to mean “If you order a pizza, then I will pick up your stuff.” On the other hand, the meaning of the VP anaphor can be identified with the meaning of another VP. For instance, we can take the VP anaphor in (2) to “mean the same thing” as the VP in the antecedent clause – on this reading, we would understand (2) as “If you order a pizza, then I will order a pizza.”

74 Carlo Cecchetto and Orin Percus One might wonder whether there is a unified explanation for how we resolve VP anaphors. Maybe, one might think, the second case can be reduced to the first somehow. Maybe the fact that we understand the VP anaphor as “order a pizza” is due to the fact that the utterance of the sentence induces us to think about the action of ordering a pizza, in the same way that John’s picking up his stuff induces us to think about the action of picking up John’s stuff.3 As we will mention below, we think that this approach is on the wrong track. But be that as it may, there are some facts that any theory of VP anaphora resolution will have to explain. Cecchetto (2004) has argued that any attempt to explain VP anaphora resolution that does not take the syntax of the sentence itself into account is doomed to failure. Cecchetto argues that, in cases where there is no clear extralinguistic source for VP anaphora resolution, VP anaphors must have the meaning of another VP present at LF. Cecchetto’s argument is modeled after an argument in the ellipsis literature (Hornstein 1995). Cecchetto starts with examples like (3a) – his actual example is the less awkward Italian (3b) – and notices that these sentences naturally admit certain readings. (3a) can be read as saying that, for every x such that x imagined I would declare x to be present, I declared x to be present. (This is the reading that the sentence I declared to be present everyone who imagined I would declare him to be present has, if him is understood as “bound within the relative clause.”) (3b) can be read as saying that, for every boy x such that x asked me to observe x dancing the tango, I observed x dancing the tango. (This is the reading that the sentence I observed dancing the tango every boy who had asked me to observe him dancing the tango has, if him is understood as “bound within the relative clause.”) Note that the relative clause here is being “filled in” in such a way that it describes a property that any individual x has if x imagined that I would declare x to be present, of if x asked me to observe x dancing the tango.4 Cecchetto then considers examples like those in (4), where the ECM clause has been replaced by a finite clause. He notices that these examples are not naturally read as saying that I declared that everyone who imagined I would declare him to be present was present, or that I observed that every boy who had asked me to observe him dancing the tango was dancing the tango. This, on the surface, is surprising. After all, the material that would be “filled in” on these impossible readings seems like just the material that would be “filled in” on the possible readings of (3). The relative clause is understood in exactly the same way.5

A contrastive analysis of VP ellipsis and VP anaphora 75

(3)

a. I declared everyone who imagined I would do that to be present b. Ho osservato attentamente ogni ragazzo che mi aveva chiesto di farlo danzare il tango ‘(I) have observed carefully every boy who had asked me to do it dancing the tango’

(4)

a. # I declared that everyone who imagined I would do that was present b. # Ho osservato attentamente che ogni ragazzo che mi aveva chiesto di farlo danzava il tango ‘(I) have observed carefully that every boy who had asked me to do it was dancing the tango’

Why does (4) lack these readings? Because, says Cecchetto, there is no VP in the sentence which, if substituted for the VP anaphor, would yield the readings in question. The readings of (3) arise from an LF in which the quantifier has undergone QR, leaving behind a VP with just the meaning that the VP anaphor would have to have in order to obtain these paraphrases. The idea here (look at the LFs in (3’)) is that the trace of the quantifier is a variable, and that a quantifier of the form everyone who imagined I would declare X to be present, where X is a variable, can mean just what is meant by everyone who imagined I would declare him to be present. So, when the VP anaphor has the same meaning as the remnant VP, the quantifier as a whole will mean what is meant by everyone who imagined I would declare him to be present. But now consider (4). If we want to have a chance of obtaining the “impossible readings” of (4) that we described, we are going to have to leave the quantifier within its clause: our readings are readings on which the matrix verb has scope over the quantifier. But, if the quantifier remains within its clause (see (4’)), there will be no VP around with the meaning that the VP anaphor would have to have on the desired reading. (3’) a. LF: [IP [DP everyone who imagined I would do that ] [IP I [VP declared t to be present ]]] b. LF: [IP [DP ogni ragazzo che mi aveva chiesto di farlo] [IP pro [VP ho osservato t danzare il tango ]]] (4’) a. LF: [IP I declared that [IP [DP everyone who imagined I would do that ] [IP t was present ]]]

76 Carlo Cecchetto and Orin Percus b. LF: [IP Ho osservato attentamente che [IP [DP ogni ragazzo che mi aveva chiesto di farlo ] [IP t danzava il tango]]] On this explanation, then, what is important in allowing the VP anaphor to mean declare him to be present is that we have a constituent like declare t to be present elsewhere at LF. In sentences like those in (3), such a constituent is formed when the quantifier QRs to a position above declare. In sentences like (4), such a constituent is not present if the quantifier remains low, and this is why the “impossible readings” we described do not arise.6 The result, in fact, is that, in the case of a sentence like (4), there will be no reading possible on which the VP anaphor contributes something like declare him to be present – given that a quantifier subject of a finite clause can never QR to a position outside the clause,7 there is no way of producing a constituent with this meaning. In the discussion that follows, we accept this argument that, in the absence of extrasentential cues to salient actions, VP anaphors have the meaning of another VP present at LF. (We will return to it in the appendix.) Let us pause for two technical points before we enter into the core of our paper. We have not presented a detailed analysis of (3) and (4). But it is essential to Cecchetto’s story that a sentence like (3) admit LFs in which the variable in the trace position is the kind of variable that can be bound within the relative clause, and that it only admit such LFs. To be more precise, assuming with Heim and Kratzer (1998) that LFs contain adjoined binder indices (which function as “lambda abstractors”), Cecchetto’s account of (3) is based on the assumption that the index at the top of the relative clause is always the same as the index on the trace. (3’’) indicates the way semantic composition then works8 – the VP anaphor, whose denotation is the same as that of the remnant VP, “behaves” as though it contained a variable bound by the binder index at the top. (And it might contain a variable like this, if something like the LF copying approach is appropriate for VP anaphors. But it also might not.) If the index at the top of the relative clause were different from the index on the trace, as in (5), the VP anaphor would be predicted to “behave” as though it contained a free variable, and on standard assumptions we would predict an impossible9 reading for a sentence like (3) – a reading in which I declared to be present everyone who imagined I would declare a certain salient person to be present. This kind of point is familiar from the ellipsis literature (see Kennedy 2004 for a recent summary), where the same issue has come up.

A contrastive analysis of VP ellipsis and VP anaphora 77

(3’’) a. LF: [IP [DP everyone who [ 1 [IP t1 imagined I would do that ]] ] [ 1 [IP I declared t1 to be present ]]] i. [[declare t1 to be present]] = g. the property of declaring g(1) to be present ii. [[do that]] = the above iii. [[t1 imagined I would do that]] = g. the proposition that g(1) imagines I declare g(1) to be present iv. [[1 t1 imagined I would do that]] = g. the property that an individual x has as long as x imagines I declare x to be present ( = [[1 t1 imagined I would declare t1 to be present]] ) b. LF: [IP [DP ogni ragazzo che [ 1 [IP t1 mi aveva chiesto di farlo]] ] [ 1 [IP pro ho osservato t1 danzare il tango ]]] i. [[osservato t1 danzare il tango]] = g. the property of observing g(1) dance the tango ii. [[farlo]] = the above iii. [[t1 mi aveva chiesto di farlo]] = g. the proposition that g(1) asks me to observe g(1) dance the tango iv. [[1 t1 mi aveva chiesto di farlo]] = g. the property that an individual x has as long as x asks me to observe x dance the tango ( = [[1 t1 mi aveva chiesto di osservare t1 danzare il tango ]] ) (5)

LF: [IP [DP everyone who [ 1 [IP t1 imagined I would do that ]] ] [ 2 [IP I declared t2 to be present ]]] i. [[declare t2 to be present]] = g. the property of declaring g(2) to be present ii. [[do that]] = the above iii. [[t1 imagined I would do that]] = g. the proposition that g(1) imagines I declare g(2) to be present iv. [[1 t1 imagined I would do that]] = g. the property that an individual x has as long as x imagines I declare g(2) to be present

We can now explain why we are suspicious of the view that VP anaphora resolution always starts from our thinking about a particular action or property. In simple cases like If you will order a pizza, then I will do that too, perhaps one can imagine how we would go from the action of ordering a

78 Carlo Cecchetto and Orin Percus pizza to the denotation of the VP anaphor, which is just (a function from assignments to) the property of performing that action. But in the case of a sentence like (3), the denotation of the VP anaphor is something more complicated whose identity depends on the identity of a particular index – an index that must be the same as the binder index at the top of the relative clause and the index of the trace of the relative pronoun. Even if we could claim that processing the sentence leads us to think about a property like the property of being declared to be present, getting from there to the denotation we need seems less straightforward.10 (Incidentally, complex examples like (3) are not needed to make this point. There are many simpler examples where a VP anaphor is resolved in an index-dependent way. The examples we will consider just below are like that.)

3. When VP ellipsis is possible and VP anaphora is not Our discussion so far suggests that, typically, when we take a sentence like (6) (or Cecchetto’s Italian version in (7)) to mean that I examine every boy who asks me to examine him, this is because in some manner the VP anaphor comes to have the meaning of the VP that QR produces. VP anaphora thus seems to bear an important similarity to VP ellipsis, if common beliefs about VP ellipsis are correct: some process or other can insure that its meaning is the same as the meaning of another VP present at LF. (6)

a. I examine every boy who asks me to do that b. [IP [DP every boy who t asks me to do that] [IP I [VP examine t ]]]

(7)

a. Interrogo ogni ragazzo che mi chiede di farlo ‘(I) examine every boy who asks me to do it’ b. [IP [NP ogni ragazzo che t mi chiede di farlo] [IP pro [VP interrogo t]]

In fact, one might see sentences like (6) or our earlier sentence in (3) as a cognate case of “A(ntecedent)-C(ontained) D(eletion)” sentences like (8). Arguments just like our argument in the last section have been given that support the idea that QR is what enables the “missing VP” here to have the meaning it does. (8)

a. I examine every student John does examine t b. [IP [every student OP John does [VP2 examine t] ] [IP I [VP examine t] ]]

A contrastive analysis of VP ellipsis and VP anaphora 79

The same questions that are asked about VP ellipsis, then, might reasonably be asked about VP anaphora. What is the process by which a VP anaphor comes to have the meaning of another VP present at LF? In these cases, is do that the pronunciation of a more complex VP, following a kind of pronunciation rule that applies only when this VP is identical to another one? Is do that replaced at LF by another VP in the same structure? Or is do that for some reason just interpreted as having the same meaning that another VP has?11 We will now consider some differences between VP ellipsis and VP anaphora, which suggest that the two kinds of phenomena call for different kinds of analyses. One difference – which as far as we know has not been noticed before – can be appreciated by looking at sentences like (9). As Hirschbühler (1982) observed, the “ellipsis” clause in sentences like (9) can have the “inverse scope” reading of its pronounced counterpart in (10), a reading compatible with the planes being inspected by different technicians.12 (9)

A security agent inspected every plane and a technician did too

(10) A security agent inspected every plane and a technician inspected every plane too The same reading, however, is not available for sentences like (11) or the Italian (12), in which the “missing VP” has been replaced by a VP anaphor (note that the Italian sentence Un addetto alla sicurezza ha controllato ogni aereo, just like its English counterpart A security agent inspected every plane, does allow inverse scope). (11) A security agent inspected every plane and a technician did that too (12) Un addetto alla sicurezza ha controllato ogni aereo, e anche un tecnico l’ha fatto. ‘A security agent inspected every plane and a technician did it too’ A second difference – which was noticed by Fiengo and May (1994) – involves ACD sentences like (8). We just discussed our VP anaphor sentences (6) and (7) as cognate cases of ACD, and one might imagine that treatments of ACD in the literature could have considered VP anaphor sentences of this kind, especially given that VP anaphora has a wider distribution cross-linguistically than VP ellipsis. But discussions of ACD have

80 Carlo Cecchetto and Orin Percus usually been based on English data and have steered clear of VP anaphora. And this is for a good reason. If one tries to reproduce a classical ACD example like (8) by using VP anaphora, the result is sharply ungrammatical. This is indicated by (13a) for English and (13b) for Italian: (8)

I examine every student John does

(13) a. *I examine every student (that) John does that b. *Interrogo ogni ragazzo che Gianni lo fa ‘(I) examine every student that Gianni it does’ What do these differences suggest? To begin with, what is going wrong in these last examples with VP anaphors? Here, we will essentially adopt the diagnosis of Fiengo and May (1994: 247). Examples like (6), which we started with, show that the problem is not that do that can’t mean what examine t means: after all, in (6), do that seems to mean just that. Rather, we take it that the problem in (13) is basically syntactic. Even if do that were to mean what examine t means, we would still need a relative operator13 at the edge of (that) John does that in order to make it into the kind of constituent that can combine with NP. We take the problem to be this: a relative operator needs to be associated with a trace, but do that cannot literally contain a trace for the relative operator to be associated with. Note by the way that in sentences like (6), this detail that do that cannot contain the trace of a relative operator does not cause problems: there, the trace of the relative operator appears outside do that, as the subject of ask. But in (13), where there is no place outside do that where we could locate a trace of a relative operator, it does cause problems. In general, we will only be able to produce ACD-like examples with VP anaphora when the relative clause is big enough to accommodate the trace of a relative operator. Simple as it is, this explanation for the ungrammaticality of (13) has consequences for the choice among our three approaches, both as regards VP ellipsis and as regards VP anaphora. Important to this explanation was that, if a constituent like (that) John VP is to function as a relative clause, VP must contain the trace of a relative operator. If this is the case, however, then in ACD cases like (8), where the parallel constituent John does VP functions as a relative clause, it must be the case that the elided VP material contains the trace of a relative operator. This is incompatible with the semantic copying approach to VP ellipsis, which would treat the elided VP as an item with no internal structure. So

A contrastive analysis of VP ellipsis and VP anaphora 81

the facts we considered have consequences for the analysis of VP ellipsis: they force us to reject the semantic copying approach. At the same time, if do that cannot contain the trace of a relative operator, this suggests that a semantic copying approach is promising for do that: on this view, do that (or maybe just that) is an unanalyzed expression whose meaning can be identified with the meanings of other LF constituents, including of course constituents that contain variables14. This approach to do that, as it happens, is also well suited to account for the unavailability of the “inverse scope reading” of the do that clause in (11)– (12). If we restrict ourselves to situations where the meaning of do that is identified with the meaning of another LF constituent, it is clear why we would never arrive at an “inverse scope reading” for the do that clause. Assuming that the antecedent clause only has LFs of the kind in (14), no constituent of the antecedent clause has a meaning that would compose with a technician to yield the proposition that every plane got inspected by at least one technician. More assumptions are required to derive that A technician did that can never mean that every plane got inspected by at least one technician, but (to our minds) nothing particularly outrageous. If do that is further constrained as to its semantic type, for example, that would do the trick. Suppose that do that is constrained so that it can only take a property, i.e. an object of type <e,st>, as its value (or, more precisely, a function from assignments to properties). No property can combine with the existential quantifier a technician to yield the proposition that every plane got inspected by technicians, so if do that is constrained to take a property as its value, no LF in which do that combines with a technician could possibly yield that proposition.15, 16 Alternatively, to the extent that the discourse context makes certain values for anaphors “salient,” it might be that we are built in such a way that the discourse context never makes salient to us the kind of value that we would need for do that if we wanted to produce the “inverse scope reading” – a function that takes quantifiers and yields propositions. (14) a. [ [DP a security agent] [ 1 [ [DP every plane] [ 2 [ t1 inspected t2]] ] ] ] b. [ [DP every plane] [ 2 [ [DP a security agent] [ 1 [ t1 inspected t2]] ] ] ] To summarize, we have argued in this section that two differences between VP anaphora and VP ellipsis can be made sense of if a semantic copying approach is correct for the former, but not the latter, construction.17

82 Carlo Cecchetto and Orin Percus 4. The PF deletion approach to VP ellipsis vs. the LF copying approach Having excluded the semantic copying approach for VP ellipsis, we are left with the choice between the PF deletion approach and the LF copying approach. Is there any basis for making this further choice? We suspect that sentences like (15), which involve apparent extraction from an ellipsis site, could be relevant to this issue.18 (15) I know which person Mary talked to t and which person Bill didn’t. The two approaches would treat cases like (15) in different ways. The PF deletion approach can maintain that (15) is just a run-of-the-mill case of extraction from the VP talk to which person. The VP from which extraction has taken place, talk to t, can then go unpronounced, given that this VP is syntactically and semantically identical to its pronounced counterpart (assuming that the two traces can have the same index). By contrast, for the LF copying approach, overt extraction cannot be involved, since, by assumption, the VP in the second clause does not have any structure at the point when overt extraction would take place. Adopting the LF copying approach thus forces us to assume that which person is base-generated in Spec,CP. The trace that which person binds then gets supplied in the covert component, after LF copying. Conceptual considerations might on their own militate against the assumptions that the LF copying approach is forced to here. If there are not so many cases where base-generation is needed, perhaps it is better to have an analysis that doesn’t assume it. And adding the possibility of base-generating wh-phrases in Spec,CP potentially introduces redundancies into the theory, if conditions that otherwise seem to regulate movement have to be extended to include the relation between a base-generated wh-phrase and its trace.19 There may also be empirical reasons for favoring the PF deletion approach. Suppose you think that the only way of interpreting I know whose mother Mary talked to t is by reconstructing the pied-piped material, and suppose you believe that material can only reconstruct to a position that it originally occupied. Then the LF copying approach would force you to treat an example like (16a) as follows. (16) a. I know whose mother Mary talked to t, and whose mother Bill didn’t b. desired LF: I know who Mary talked to t’s mother, and who Bill didn’t talk to t’s mother

A contrastive analysis of VP ellipsis and VP anaphora 83

(17) a. *I know whose mother Mary talked to t, and who Bill didn’t. b. not a desired LF: I know who Mary talked to t’s mother, and who Bill didn’t talk to t’s mother First, reconstruction would take place in the first conjunct, forming the VP talk to t’s mother. Then this VP would be copied into the second conjunct, forming whose mother Bill talked to t’s mother. Then, interpretation would work in such a way that the only part of the wh-phrase that gets interpreted in the second conjunct is who. But this approach would seem to make wrong predictions for cases like (17a). (17a) should mean just what (16a) does, since we would copy just the same material, and who is all that we find in the wh-phrase of the second conjunct. The PF deletion approach, by contrast, does not make this prediction. Arguably, for the second conjunct of (17a) to have the LF of the second conjunct in (17b), we would have to produce it starting from a complete clause of the form Bill didn’t talk to whose mother, and independent morpho-phonological constraints prevent us from moving who alone in a clause like that. Note that (17a) also cannot mean that I know who Bill didn’t talk to. This is expected too: on the LF required for this, the VP in the first conjunct would be talk to t’s mother and the VP in the second conjunct would be talk to t, and it is natural to think that such a representation would not satisfy the identity condition on ellipsis. More generally, as Merchant (2001) has observed for parallel cases of sluicing, we can only create ellipsis examples with two wh- extractions when the wh-phrase in the ellipsis clause “matches” the wh-phrase in the antecedent clause (here, we have seen that we cannot have a who-phrase in the ellipsis clause and a whose-phrase in the antecedent clause). The PF deletion approach seems better suited to deal with matching effects. We thus tentatively adopt the PF deletion approach to VP ellipsis. A well known consequence is that, in some manner, information about what gets deleted at PF must be available at LF. ACD examples like (18) lead naturally to the conclusion that the identity condition on ellipsis applies at LF: at spell-out, there is no VP that the deleted VP is identical to, but at LF – thanks to QR – there is. So apparently there is a condition that applies at LF and that determines whether PF-deleted material is identical to other material. For our purposes here, we endorse Merchant’s (2001) approach to this issue.20 On this approach, there is a silent head E that combines with the VP in all VP ellipsis sentences – the syntactic representations of (18) thus look more as in (19’) than as in (19). E’s role at PF is to issue an instruction to the articulatory component to leave its complement unpronounced; E’s

84 Carlo Cecchetto and Orin Percus role at LF is to convey that its complement is subject to an identity condition. In Section 6, we will return to this silent head E, and suggest that it might contribute to the explanation of the cross-linguistic distribution of VP ellipsis. (18) I examine every student (that) John does (19) a. spell-out: I examine [DP every student [CP OP that [IP John does [VP examine t ]]]] b. LF: [DP every student [CP OP that [IP John does [VP examine t ]]] I [VP examine t] (19’) a. spell-out: I examine [DP every student [CP OP that [IP John does E [VP examine t ]]]] b. LF: [DP every student [CP OP that [IP John does E [VP examine t ]]] I [VP examine t] We would like to conclude this section with some remarks about a peculiar property of ACD that often goes unnoticed (and for which no explanation has been given, to the best of our knowledge). Apparently, in ACD constructions, the relative operator can never be overtly expressed. For example, there is a clear contrast between (18) and (20). A similar contrast is found in other pairs of sentences, like (21) and (22). (20) ?? I examine every student who(m) John does (21) I talk to every boy (that) you do (22) ?? I talk to every boy who(m) you do We would like to suggest that we can get a handle on this phenomenon if we assume that in examples like (18) – contrary to what we indicated in (19a) – the relative operator moves covertly.21, 22 On this way of looking at things, it is specifically overt movement of a relative operator that creates problems in ACD constructions. Covert movement saves a relative operator. A reviewer suggested an interesting explanation for this that is inspired by Lasnik’s (1999) account of pseudogapping and roughly amounts to the following: relative operators are required to move overtly unless they are contained in an unpronounced constituent; overt movement doesn’t occur when it doesn’t have to.23 On this explanation, what is wrong with overt

A contrastive analysis of VP ellipsis and VP anaphora 85

movement of the relative operator in ACD examples is that it is unnecessary – it occurs out of an unpronounced constituent. (Note that an explanation of this kind doesn’t imply that overt extraction from ellipsis sites will always be impossible: it will be impossible only if there are no additional factors forcing overt movement. We have seen that, in examples like (15), an interrogative phrase can be extracted. There, one could say that movement is forced by the condition that Spec, CP of an interrogative clause must be overtly filled in English.) These are speculations about why it is that covert movement would save a relative operator. But the main point is that the diagnosis that covert movement saves a relative operator makes sense only on the PF deletion approach: on the LF copying approach (or at least its simplest implementation), the relative operator in (18) could not be in its final position at LF without being there before as well. To summarize, at this point we have: (i) established that the semantic copying approach is inadequate for VP ellipsis but is promising for VP anaphora; (ii) suggested that the PF approach to VP ellipsis seems preferable to the LF copying approach. In passing we have indicated the kinds of assumptions that a PF approach must make in order to treat VP ellipsis in ACD examples. 5. When VP anaphora is possible and VP ellipsis is not In section 3, we were concerned with the fact that, if we start with a VP ellipsis sentence and create a counterpart sentence in which a VP anaphor “occupies the ellipsis site,” we do not always wind up with a grammatical sentence with the same interpretations. This suggests that VP ellipsis admits a kind of analysis that VP anaphora does not. The reverse is also the case, however. If we start with a VP anaphor sentence and create a counterpart sentence in which the VP is elided, we do not always wind up with a grammatical sentence with the same interpretations. This suggests that VP anaphora admits a kind of analysis that VP ellipsis does not. There is one well known way in which VP anaphora is “more permissive” than VP ellipsis. The former allows a “non-linguistic antecedent” much more freely – a speaker could say after coughing violently “I did that to catch your attention” but not “I did to catch your attention.” There seem to be other ways as well. One, we have noticed, is subject to some variation across speakers. Speakers like the native English-speaking author of this paper (O.P.) do not seem to admit VP ellipsis counterparts to the kinds of VP anaphora examples that we focused on in the previous sections. For these

86 Carlo Cecchetto and Orin Percus speakers, (23a) – unlike (23b) – cannot mean that I examined every boy who asked me to examine him, and (24a) – unlike (24b) – cannot mean that I introduced her to everyone who wanted me to introduce her to him. These judgments contradict other judgments that have been reported in the literature (e.g. Haïk 1987), and that hold for some other speakers. (23) a. I examined every boy who asked me to. b. I examined every boy who asked me to do that. (24) a. I introduced her to everyone who wanted me to. b. I introduced her to everyone who wanted me to do that. The behavior of speakers like O.P. is somewhat complicated to describe. The generalization can be expressed if we assume the PF deletion approach to VP ellipsis, and if we assume that the “identity” condition on unpronounced VPs is sensitive to syntax – let us say that an elided VP must be “sufficiently identical” syntactically to another VP. For O.P., pronouns do not count as “sufficiently identical” to traces in cases where their binders are not the same – though they do count as “sufficiently identical” to traces in cases where their binders are the same. For O.P., what is going wrong in (23a)/(24a), then, is that the intended meaning would come from a sentence in which a VP with a pronoun goes unpronounced ((23’a)/(24’a)), but where the pronoun does not have the same binder as the trace in the parallel position of the “antecedent VP.” (The relevant LFs are given in (23’b)/(24’b). Note that, while the pronoun and the trace left by QR both have the same index 1, the binder index that binds the pronoun is at the top of the relative clause, while the binder index that binds the trace left by QR is at the top of the matrix clause.) (23’) a. I examined every boy who t asked me to examine him. b. required LF: [IP [DP every boy who [ 1 [IP t1 asked me to examine him1 ]] ] [ 1 [IP I examined t1 ]]] (24’) a. I introduced her to everyone who t wanted me to introduce her to him. b. required LF: [IP [DP everyone who [ 1 [IP t1 wanted me to introduce her to him1 ]] ] [ 1 [IP I introduced her to t1 ]]] The reason for saying that pronouns do count as “sufficiently identical” to traces in cases where their binders are the same is that, for speakers like

A contrastive analysis of VP ellipsis and VP anaphora 87

O.P., sentences like (25a) and (26a) behave differently. (25a) can naturally be understood as saying that I examined every boy before you examined him, and (26a) can naturally be understood as saying that I introduced every woman to everyone she wanted me to introduce her to. We can explain why if we say that these sentences involve ellipsis of a VP with a pronoun, and, in that case, in the relevant LFs, the pronoun would have the same binder as the trace in the parallel position of the “antecedent VP.” (Note that, in the LFs in (25c)/(26c), both are bound by the binder index below the quantifier at the top.) (25) a. I examined every boy before you did. b. I examined every boy before you examined him. c. required LF: [IP [DP every boy ] [ 1 [IP I examined t1 before you examined him1 ]] ] (26) a. I introduced every woman to everyone she wanted me to. b. I introduced every woman to everyone she wanted me to introduce her to. c. required LF: [IP [DP every woman] [ 1 [ [DP everyone 2 she1 wanted me to introduce her1 to t2 ] [ 2 [IP I introduced t1 to t2 ]]]] ] Why should there be these apparent differences across speakers? A speculation might be that in fact all speakers behave identically where phonetic deletion of complex VPs is concerned – all speakers apply the same identity condition and have the same criteria for “sufficient identity” – but that the speakers unlike O.P. have in their lexicon a “null” VP anaphor that receives no pronunciation at all. These latter speakers would allow (23a) to mean what (23b) means because they would allow a structure for (23a) with the silent anaphor in the place of do that, interpreted in just the way do that is interpreted. Essentially, these speakers, for some sentences with unpronounced VPs, would allow a semantic copying analysis in addition to a PF deletion analysis.24, 25 This analysis raises obvious questions – do the same speakers that allow (23a) to mean (23b) also allow “non-linguistic antecedents” for examples like (23a)? are there any unpronounced VPs for which these speakers do not allow a semantic copying analysis in addition to a PF deletion analysis? However, whether or not our speculation is correct here, we can think of another case where this kind of approach seems plausible. It concerns Italian.

88 Carlo Cecchetto and Orin Percus While Italian contains a VP anaphor that behaves very like English do that, Italian is different from English in that it does not make productive use of VP ellipsis. Since Italian does not have do-support at all, there is nothing like VP ellipsis with auxiliary do. And sentences like those in (27) are sharply out. The only sentences that are potential candidates for VP ellipsis of the English type are sentences like those in (28), where the “missing structure” is introduced by a modal verb26. (27) a. *Tu hai mangiato la pizza e anche io ho. ‘You have eaten the pizza and also I have’ b. *Tu sei arrivato tardi e anche io sono. ‘You are arrived late and also I am’ (28) a. Vorrei mangiare la pizza e anche Gianni vorrebbe mangiare la pizza ‘(I) want to eat the pizza and also Gianni wants’ b. Dovrei mangiare la pizza e anche Gianni dovrebbe mangiare la pizza ‘(I) must eat the pizza and also Gianni must’ c. Potrei mangiare la pizza e anche Gianni potrebbe mangiare la pizza ‘(I) can eat the pizza and also Gianni can’ We think it likely, however, that these sentences actually contain something like a silent do that – so, really, there is no VP ellipsis in Italian at all. Our reasons are as follows. First, the sentences in (28) all admit a counterpart with the overt VP anaphor farlo in the position of the “null” VP. Second, if the sentences in (28) involved VP ellipsis of the English kind, we would expect to find ACD configurations of the English kind, but in fact sentences like (29) are ungrammatical.27 If the apparent cases of ellipsis actually involve a silent anaphor, it is clear why: the silent anaphor blocks a source position for the relative operator, as was the case for the examples we considered in section 3. This diagnosis implies that sentences of this kind should become grammatical if we modify them by adding a piece of structure that can accommodate the trace of the relative operator, and this prediction seems correct. (30a) is acceptable, and has the interpretation “I want to examine every boy who asked if I wanted to examine him,” which is unsurprising if its structure is basically as in (30b).28 (29) a. *Vorrei interrogare ogni ragazzo che Gianni vorrebbe interrogare b. *Potrei mangiare ogni pizza che Gianni potrebbe mangiare c. *Dovrei mangiare ogni pizza che Gianni dovrebbe mangiare

A contrastive analysis of VP ellipsis and VP anaphora 89

(30) a. Vorrei interrogare ogni ragazzo che ha chiesto se io volessi. ‘I want to examine every boy who has asked if I wanted’ b. Vorrei interrogare ogni ragazzo OP che t ha chiesto se io volessi Øfarlo. The Italian candidates for English-type ellipsis, then, are better analyzed as containing a silent VP anaphor. In sum, we have suggested here that a few special cases of “null” VPs should be exceptionally analyzed as involving the presence of an anaphor, rather than PF deletion (or LF copying). This idea in turn reinforces the hypothesis that PF deletion is blocked when a pronoun in the elided VP and a trace in the corresponding position in the antecedent VP do not have the same binder. This further identity condition (if we are on the right track) should be taken into consideration in future work on when PF deletion is possible.

6. Why is VP ellipsis absent in Italian? In the previous section, we argued that Italian does not exhibit VP ellipsis of the English type. Here we briefly address why this is so. Earlier, we mentioned the hypothesis (essentially due to Merchant) that PF deletion of VPs is due to the presence of a silent head E which combines with the VP and tells the articulatory component “Don’t pronounce my complement.” E is a possible lexical item, and part of the lexicon of English. This hypothesis can help to account for the restrictions on where VP ellipsis can occur in English. Assuming that E blocks verb movement – as should follow from some version of the Head Movement Constraint – a language that has E should allow VP ellipsis only in cases where the verb does not need to move to the INFL node. As far as English is concerned, this seems to be what happens. In English, the verb can remain inside the VP when do-support is present, or when the INFL node is occupied by an auxiliary, and here we see that ellipsis is possible: (31) a. I like everyone Bill does. b. I like everyone Bill does E [like t] (32) a. I have eaten pizza and John has, too. b. I have eaten pizza and John has E [eaten pizza], too.

90 Carlo Cecchetto and Orin Percus By contrast, ellipsis is impossible in cases where there is no support for tense and agreement features: (33) a. * I like everyone Bill b. * I like everyone Bill INFL E [likes t] Now imagine a language that, unlike English, typically requires the verb to move out of the VP. Such a language, even if it had E in its lexicon, would rarely use it, since the use of E would be incompatible with the presence of verb movement. Since evidence for the presence of E would be sparse in languages that make extensive use of verb movement, acquisition considerations suggest that E will not appear in such languages. Italian is a language of this kind. In Italian, the verb must move out of the VP (arguably due to its morphological richness). Accordingly, we wouldn’t expect to see much evidence of VP ellipsis even if it had E, and acquisition considerations suggest that it is unlikely that Italian would have E at all. Take an example like (27a) above (repeated as (34a)). Arguably, even if Italian had E, we would be unable to generate this sentence. The only way of doing so would be with a structure like (34b), but it has been argued (cf. Belletti 1990 among others) that structures like this are unavailable: past participles like mangiato cannot stay confined to the VP. Evidence that the Italian past participle overtly moves out of the VP comes from the fact that it appears to the left of VP-peripheral elements like low adverbs (cf. (35)) and floating quantifiers (cf. (36)).29 (34) a. *Tu hai mangiato la pizza e anche io ho. b. Tu hai mangiato la pizza e anche io ho E [VP mangiato la pizza] (35) I ragazzi hanno mangiato spesso la pizza ‘The boys have eaten often the pizza’ (36) I ragazzi hanno mangiato tutti la pizza ‘The boys have eaten all the pizza’ If Italian had E, where would we expect to find ellipsis? There are a few cases where verb movement is not obviously necessary in Italian. In the last section, we considered one: infinitival complements of modal verbs. But even there, we argued, although it seems on the surface that ellipsis is possible, this phenomenon is not PF deletion of the English type. Italian, then,

A contrastive analysis of VP ellipsis and VP anaphora 91

does not have E, and the language acquisition considerations we cited above make this natural. On the approach we have just sketched, the reason why Italian lacks VP ellipsis is that the null head triggering PF deletion interferes with the (generally) obligatory movement of the head of the VP. This account is appealingly simple. It links the fact that VP ellipsis cannot occur in Italian in any context to the fact that VP ellipsis cannot occur in English in some contexts (those where the verb moves). And, obviously, it depends on the idea that PF deletion can only be caused by the null head that blocks verb movement.30 Perhaps it is too simple. McCloskey (1991) and Goldberg (2005) have argued that VP ellipsis of the English type is present in some languages that, like Italian, display overt verb movement out of the VP. In Hebrew and Irish, they argue, verbs outside the VP can be pronounced while material that remains within the VP gets deleted, giving rise to questionanswer sequences like (37). This poses a clear problem for our account, which implies that VP ellipsis should not be possible in sentences with verb movement. It is true that superficially examples of this kind resemble null object constructions, but Goldberg discusses a set of tests that indicate that the relevant construction in Hebrew and Irish is the same construction that is called VP ellipsis in English. If Goldberg’s conclusions are right, then the source of PF deletion can be something other than a null head that blocks verb movement. And this reopens the issue of why Italian lacks VP ellipsis. Evidently, more research is needed to settle this question. (37) a. Will you invite John? b. I already invited [VP John ]

7. Conclusion and loose ends In this paper, we reconsidered the issue of what process underlies VP ellipsis in English. By systematically comparing VP ellipsis with VP anaphora, we reached the conclusion that a semantic copying approach is inadequate for VP ellipsis but works pretty well for VP anaphora. We also suggested that a PF deletion theory of VP ellipsis will turn out to be more adequate than an LF copying theory, and we cast our explanation for the absence of VP ellipsis in Italian in terms of the PF deletion theory of VP ellipsis. Since VP ellipsis has been investigated much more extensively than VP anaphora, we were interested in investigating whether the properties of VP

92 Carlo Cecchetto and Orin Percus ellipsis that have been described in the literature hold for VP anaphora, and whether the explanations proposed for the former construction can be extended to the latter. We presented cases where VP ellipsis seems to be more permissive than VP anaphora, and this allowed us to draw some conclusions about the semantic copying mechanism that underlies VP anaphora. We then discussed cases that go in the opposite direction. We saw that VP anaphora is more liberal than VP ellipsis not only because it admits a non-linguistic antecedent more freely, but also in more subtle ways, that sharpen our understanding of the syntactic “identity” condition on unpronounced VPs. This paper did not offer a comprehensive theory of VP ellipsis and VP anaphora. It did add some new data and some new ideas to the debates concerning these phenomena. And, more generally, it showed that examining VP anaphora can provide a useful perspective on issues that arise in the study of VP ellipsis. We would like to conclude this paper by recalling another case where a consideration of VP anaphora is useful.31 Pairs of sentences like those in (38) have received a fair amount of attention in the ACD literature. (38) can be understood as saying that I introduced Mary to everyone she wanted me to introduce her to, while (39) cannot be understood as saying that I introduced Mary to everyone she wanted me to introduce herself to. It thus seems that the elided VP in (39) cannot have the same meaning that the elided VP in (38) does. (38) I introduced Mary to everyone she wanted me to (OK interpretation: I introduced Mary to everyone she wanted me to introduce her to.) (39) # I introduced Mary to everyone I wanted her to. (impossible interpretation: I introduced Mary to everyone I wanted her to introduce herself to.) This contrast has received attention in the context of discussions of “Vehicle Change.”32 This discussion assumes as background that elided VP material must be syntactically identical to antecedent VP material. On this assumption, the acceptability of (38) on the relevant reading is not expected: the elided VP in (38) would have to contain the R-expression Mary, and Condition C should then block the interpretation on which she and Mary are coreferential: (38’) LF: [DP everyone [CP OP [IP she wanted me to [VP introduce Mary to t ] ]] I [VP introduced Mary to t]

A contrastive analysis of VP ellipsis and VP anaphora 93

To circumvent this problem, Fiengo and May (1994) propose essentially that after the identity condition is satisfied, a proper name can be replaced by a pronoun. (The process by which this takes place is called “Vehicle Change”.) After this replacement, the LF of (38) is (38’’), Condition C is no longer relevant, she and her can be coreferential, and we arrive at the desired interpretation. (38’) [DP everyone [CP OP [IP she wanted me to [VP introduce her to t ] ]] I [VP introduced Mary to t] (39) is relevant to this discussion in that, if Vehicle Change can only change a name to a pronoun, (39) is correctly predicted to lack the reading we mentioned. This interpretation cannot arise from an LF to which Vehicle Change has not applied ((39’)), without incurring a Condition C violation. And it cannot arise from an LF to which Vehicle Change has applied ((39’’)), without incurring a Condition B violation: (39’) *[DP everyone [CP OP [IP I wanted her to [VP introduce Mary to t ] ]] I [VP introduced Mary to t] (39’’) *[DP everyone [CP OP [IP I wanted her to [VP introduce her to t ] ]] I [VP introduced Mary to t] The idea of Vehicle Change is subject to criticism – for one thing, this mechanism must be restricted to VP ellipsis constructions, since we do not want to be able to use it to save canonical Condition C violations like *Shei said that Maryi saw John. However, the idea behind this proposal could be expressed easily enough in other terms. One might just say that the identity condition on ellipsis is loose enough to allow pronouns in those positions of elided VPs where we find names in the antecedent VP. The basic approach to these facts, then, would be that in (38) and (39) independent factors allow the elided VP to be introduce her to t, but not, say, introduce herself to t. Conditions on pronominal reference then permit the desired interpretation for (38) but not for (39). The point we want to make is that a look at VP anaphora suggests another perspective on these facts. Parallel to the pairs involving VP ellipsis, we find similar pairs involving VP anaphors. (40) can be understood as saying that I introduced Mary to everyone who wanted me to introduce her to him, but (41) cannot be understood as saying that I introduced Mary to everyone

94 Carlo Cecchetto and Orin Percus who wanted her to introduce herself to him. It thus seems that the VP anaphor in (41) cannot have the same meaning that the VP anaphor in (40) does. However, since we have seen that VP anaphors are unanalyzed items, we certainly cannot attribute this effect to the existence of a pronoun within VP. It seems, then, that something else is going on. We conjecture that, when it comes to expressing what we would like (39) and (41) to express, there is just something wrong with using a VP with the kind of denotation that we would get from a VP with a pronoun coreferential with Mary. The right explanation for the puzzle regarding (39) and (41) should probably be sought in the theoretical domain that deals with effects like “Rule I” effects.33 (40) I introduced Mary to everyone who wanted me to do that. (OK interpretation: I introduced Mary to everyone who wanted me to introduce her to him.) (41) # I introduced Mary to everyone who wanted her to do that. (impossible interpretation: I introduced Mary to everyone who wanted her to introduce herself to him.) Here, then, is another case where approaches to VP ellipsis facts have to be revised, if they are to be general enough to cover parallel cases in the VP anaphora domain.

Appendix In our initial discussion of VP anaphora, we adopted Cecchetto’s proposal that, in cases where there is no clear extralinguistic source for VP anaphora resolution, VP anaphors must have the meaning of another VP present at LF. This proposal, together with some assumptions about what the LFs for quantificational structures with relative clauses look like, was enough to explain the pattern we considered – in a sentence like (42), when we imagine that the quantifier takes scope outside the clause, we naturally understand do that in a way that we do not when we hear a sentence like (43) and imagine that the quantifier takes scope inside its clause. However, this proposal needs revision: it is too restrictive as it stands. (42) I declared everyone who imagined I would do that to be present (43) #I declared that everyone who imagined I would do that was present

A contrastive analysis of VP ellipsis and VP anaphora 95

Sometimes it seems that, in the absence of extralinguistic cues, we admit for VP anaphors certain meanings that do not constitute the denotations of other VPs around. Two examples are in (44) and (45).34 In (44) (based on an example from Rooth 1992 without VP anaphora), do that is naturally understood as “divides into itself exactly once.” In (45), do that is naturally taken as meaning “gave her son a stuffed animal.” Neither of these meanings comes from an LF constituent. The only appropriate meanings that derive from LF constituents seem to be meanings like “divides into 7 exactly once,” or “gave Carlo a stuffed animal.” What does seem to be the case, though, is that there is another clause around that amounts to saying that some object (or group of objects) has the property in question. For instance, the first clause of (45) amounts to saying that Carlo’s mother has the property we get from “gave her son a stuffed animal.” (44) 7 divides into 7 exactly once. 5 does that too, of course. (45) Carlo’s mother gave Carlo a stuffed animal on his birthday. Then, on my birthday, my mother did that. It must be a fad. Our earlier remarks suggested that the values for VP anaphora come from two different kinds of places. On the one hand, we determine values on the basis of actions that are psychologically salient. On the other hand, we determine values on the basis of what syntactic constituents are present at LF. Conceivably, some actions can become psychologically salient just as a result of processing linguistic material. A position that one could take, then, is that what we said about our “syntax-oriented” way of locating values is right – we only use the denotations of surrounding constituents – and that (44) and (45) are cases that involve the other way of locating values. On this view, in adopting Cecchetto’s proposal, we were just a little imprecise: we neglected that sometimes VP anaphors can be resolved in the non“syntax-oriented” way without true extralinguistic cues. But we suspect that, in fact, what happens in (44) and (45) is compatible with our “syntaxoriented” way of locating values, and that this way of locating values is different from what we have been assuming – that it does not in fact restrict us to the denotations of surrounding constituents. One reason for this suspicion is as follows. To the extent that it is natural to understand (45) as saying that my mother gave me a stuffed animal, it seems equally natural to understand (46) as saying that Carlo’s mother introduced Carlo to all those people who previously had asked my mother to introduce me to them. Now, in order to obtain this reading, do that must

96 Carlo Cecchetto and Orin Percus have a value that is index-dependent – it must behave as though it contains a variable bound within the relative clause. (That is, the analysis must run roughly along the lines of (47)). But recall from earlier that we feel that, in cases where the value of an anaphor is index-dependent, it is implausible to maintain that we determined the value in the non-“syntax-oriented” way. We conclude, then, that something about our “syntax-oriented” procedure must be providing the index-dependent value for the anaphor in (46), and this is just the kind of value that we would need to account for (45) as well. (46) Carlo’s mother introduced Carlo to everyone who previously had asked my mother to do that. (47) a. [IP [DP everyone who [ 1 [IP t1 had asked my mother to do that ]] ] [ 1 [IP Carlo’s mother introduced Carlo to t1 ]]] b. [[do that ]] = g. the property of introducing one’s son to g(1) What kind of “syntax-oriented” procedure would admit values for do that like those we need in (44)–(46)? A first guess might be that we assign values to VP anaphors in such a way as to satisfy the following condition. Even if the VP anaphor itself doesn’t have the same denotation as another LF constituent, you must be able to obtain something that (informally speaking) conveys the same information as another LF constituent by performing an operation like this: you take a constituent containing the VP anaphor, and you replace pieces of that constituent (other than the VP anaphor) with other entities of the same category. For example, in (47a), we can obtain an expression that carries the same information as the constituent Carlo’s mother introduced Carlo to t1 by taking the complement of ask, my mother to do that, and replacing my with Carlo’s. Interestingly, this is basically the “contrast condition” that has been discussed in the VP ellipsis literature (cf. Heim 1997, Kennedy 2004). We can’t go into the many remarks that we would like to make in this regard.35 What is most relevant to us here is just that revising Cecchetto’s proposal in this way would not lead to other substantial changes in what we said, and would still allow us to distinguish (42) from (43).36 It is also worth noting in passing that the VP ellipsis counterparts of (44) and (45) do not allow the interpretations that we have been discussing (cf. Rooth 1992), so here we have another phenomenon that suggests that VP anaphora admits a kind of analysis that VP ellipsis does not.

A contrastive analysis of VP ellipsis and VP anaphora 97

Acknowledgements Earlier versions of this paper were presented at seminars at the Universities of Milan-Bicocca and Siena and at the XXXI Incontro di Grammatica Generativa at the University of Roma Tre. We thank the audiences there for their comments. For very helpful suggestions we are grateful to Chris Kennedy, an anonymous reviewer and an audience member in Rome.

Notes 1. In this paper, we will sometimes strike through a VP to indicate that the sentence is interpreted as though it had a VP with the relevant meaning, but is pronounced as though it had no VP at all. This is a notational device and has no theoretical significance. 2. Hoji (2003) is an important exception. Our approach is very similar to Hoji’s, and this paper complements his in many ways. Unfortunately, we learned of Hoji’s paper only as our paper was nearing completion. 3. See Heim and Kratzer (1998: 239–241), for relevant discussion. 4. The judgment about the availability of the relevant reading does not change if the subject of the ECM structure is postverbal. In (i) the intended interpretation is as natural as in (3a), and in (ii) it is as natural as in (3b). ((ii) feels slightly better than (3b), probably because heavy subjects like to be postverbal in Italian.) (i) I declared to be present everyone who imagined I would do that (ii) Ho osservato attentamente danzare il tango ogni ragazzo che mi aveva chiesto di farlo ‘(I) have observed carefully dancing the tango every boy who had asked me to do it’ 5. Note that, when we described the readings of the sentences in (3), we chose paraphrases in which the quantifier has scope over the matrix verb. By contrast, here we have described readings on which the quantifier takes scope within its clause. It is not surprising that the examples in (4) lack the very same readings that we mentioned for (3): it is well known that a quantifier subject of a finite clause has difficulty taking scope outside the clause. But it is on the surface surprising that they don’t have readings of the kind we described where the quantifier takes scope within its clause. 6. This kind of explanation predicts that we also won’t allow a reading for (3) in which the VP anaphor is understood in the same way and in which the quantifier remains below declare. Assuming that leaving the quantifier below in I declared every X to be present would yield a reading on which my declaration was that every X is present, it seems to us that we prefer not to leave the quantifier be-

98 Carlo Cecchetto and Orin Percus

7.

8. 9. 10.

11.

12.

13. 14. 15.

16.

low to begin with. (In a situation in which my declaration was that every professor was present but in which I declared that John – a professor who I did not know to be a professor – was absent, O.P. finds it difficult to take I declared every professor to be present to be true.) But in any case it is clear to us that the sentence cannot be read as saying that my declaration was that everyone who imagined that I would declare him to be present was present. We won’t discuss the relevant evidence here. A sentence like (4a) thus cannot be read with a “wide scope” reading for the quantifier, no matter what the interpretation of do that is taken to be. QR out of finite clauses is commonly held to be impossible. See Cecchetto (2004) for a specific formulation of the clause-boundedness constraint on QR from which it follows that in cases like (4) extraction of the quantifier is impossible. We assume here that denotations are functions from assignments. Where Heim and Kratzer would write [[ X ]]g =
, we would write [[ X ]] = g.
. Impossible in the absence of cues to salient actions, that is. This is not a watertight argument, of course. The best way to counter it would be to find an actual case where a VP anaphor can be resolved in an indexdependent way, but where (as in our pizza example earlier) it is impossible to maintain that the resolution is triggered by linguistically present material. Similar questions could be posed about other kinds of anaphora. For instance, Elbourne (2001) analyzes e-type pronouns as pronunciations, subject to an identity requirement, of more complex expressions. So similar questions arise there. A reviewer points this out, and sees our paper as “relevant more generally to analyses … that seek to create an underlying … equivalence between a pro-form and a fully spelled-out syntactic representation.” Typically, we understand the “ellipsis” clause in this way when we understand the “antecedent” clause in the same kind of way, as compatible with the planes being inspected by different security agents. Fox (1999) observed moreover that if instead the “ellipsis” clause is understood as saying that a single technician did all the inspections, the “antecedent” clause cannot have the inverse scope reading. See Fox (1999) for discussion and an explanation. Or at least a binder index, if what matters is that we produce the kind of constituent that can semantically modify NP. In this respect, do that seems akin to pronouns on their “paycheck” uses. If you don’t see this point, try to come up with a property such that, if we find just one technician with that property, that is enough to conclude that every plane got inspected by at least one technician. (Good luck.) In fact, to treat do that properly, we clearly need a more sophisticated analysis. We have frequently been talking about do that as describing the performance of an action, and indeed do that generally seems to be limited to the kinds of properties that agents have (though there are some apparent exceptions). This could fit with an event-based approach on which VPs describe properties of events: one might say, on this approach, that do that is limited to events that

A contrastive analysis of VP ellipsis and VP anaphora 99 involve agents (if “involving an agent” is a coherent notion). More specifically, one might imagine that that carries no restrictions on its own, and that it is do that imposes requirements on what kind of property of events its sister can be. 17. A reviewer reminds us of another difference between VP anaphora and VP ellipsis, points out that it is reminiscent of the “inverse scope reading” difference that we discussed, and suggests that it naturally receives the same kind of explanation. Hankamer and Sag (1976) discuss a phenomenon involving “Russell’s ambiguity” in comparatives: (i) is ambiguous between a reading on which Kim’s claim is incorrect (they call it a “sensible” reading) and a reading on which Kim makes a contradictory claim (they call it a “stupid” reading). They observe that while ellipsis as in (iia) preserves the ambiguity, VP anaphora as in (iib) forces the “stupid” reading. Our reviewer notices that, if we arrive at the “sensible” reading of (i) by moving something out of the scope of claimed (cf. von Stechow 1984), then it is clear why the VP anaphora clause should lack a “sensible” reading. No constituent of the antecedent clause will have a meaning that could compose with Pat to yield the “sensible” proposition. (i) Kim claimed that Lee was older than he was... (ii) a. … and Pat did too. (ambiguous) b. … and Pat did that too. (“stupid” reading only) 18. See Chao (1987), Fiengo and May (1994) and Tancredi (1992) for the observation that it is possible to extract a wh-phrase from an ellipsis site. 19. Of course, if it turned out that conditions on movement do not hold between allegedly base-generated wh-phrases and their traces, that would be support for the base-generation view. One might wonder, for example, whether islands can intervene between allegedly base-generated wh-phrases and their traces. In fact, we think that there are some cases of island repair (or amelioration) under VP ellipsis, as shown by the contrast between (i) and (ii): (i) ? I examined every student that John asked when Mary would (ii) * I examined every student that John asked when Mary would examine However, these data have to be treated carefully. There is a large literature (Ross 1969; Chomsky 1972; Chung et al. 1995; Merchant 2001; Fox and Lasnik 2003, among others) that shows that island repair under deletion occurs even in very clear cases of movement. This implies that the absence of island effects in deletion contexts is not a reliable diagnostic for the absence of movement. 20. See Lobeck (1995) for further discussion of the issue. 21. It suffices for us to assume that null categories can move covertly, and this seems to be the null hypothesis. 22. A very similar idea is defended in Kennedy (2002). Kennedy argues that, in comparative clauses, a kind of movement (of an unpronounced item) that normally occurs before spell-out does not occur before spell-out when the source position is inside an elided VP. Kennedy presents his analysis in Optimality Theoretic terms.

100 Carlo Cecchetto and Orin Percus 23. The reviewer suggests the following implementation, which assumes, in the spirit of Chomsky (1995), that covert movement is feature movement, instead of movement of a full category. “All things being equal, it’s most economical to just move features, but most of the time (in English), you also have to move the lexical material associated with these features or you run into some sort of violation at PF. The way to avoid this PF violation is to not pronounce the lexical material from which the features have moved; one way to achieve this is to ensure that it’s part of a deleted constituent.” (20) and (22), which exhibit category movement, are uneconomical, since we could have moved features only and escaped without a PF violation. By contrast, category movement is motivated in sentences like I talk to every boy to whom you talk or I examine every student who(m) John examines, since we couldn’t have moved features only without incurring a PF violation. 24. Hoji (2003) argues for a similar conclusion on the basis of different evidence. For him, too, while most sentences with unpronounced VPs involve copying or deletion, a few seem to contain a silent anaphor instead. A reviewer notes that, if the silent anaphor we have posited is really interpreted in just the way do that is interpreted, speakers unlike O.P. should not allow a sentence like (ia) to mean that I admire most of the people who think I admire them. This is because a parallel sentence with do that ((iia)) cannot convey this (for reasons relating to the stativity of think, cf. note 16). Similar remarks apply to (ib). The reviewer, who is a speaker unlike O.P., reports that in this case these predictions are correct. (i) a. I admire most of the people who think I do. b. I’m proud of every student who thought I would (be). (ii) a. # … who think I do that. b. # … who thought I would do that. 25. There are other respects in which O.P.’s judgments are more restrictive than those sometimes reported. For instance, O.P. does not allow the often cited reading for (i) on which neither of them can realize his desire. A natural conclusion is that readings like this are also due to the presence of a null anaphor, although one might question here whether the anaphor is really interpreted in just the way do that is (it seems to us that “neither of them can do that” expresses this meaning awkwardly at best). We thank an anonymous reviewer for bringing up these examples. (i) Wendy is eager to sail around the world and Bruce is eager to climb Kilimanjaro, but neither of them can because money is too tight. (Webber 1978) 26. In fact, modal verbs might take a structure bigger than VP as a complement (an IP or a functional projection in the IP field). We will not go into this issue here: the most important aspect of our claim is that the missing structure in the sentences we are discussing is a null anaphor. However, the question should be

A contrastive analysis of VP ellipsis and VP anaphora

101

investigated in order to understand which portion of the syntactic structure the anaphor occupies. 27. We thank Fabio del Prete for this observation. 28. For reasons that are not completely clear to us, with other modal verbs the interpretation we would expect is not fully natural. For instance, (i) and (ii) do not naturally give rise to the interpretations “I should examine every boy who asked if I had to examine him” and “I could examine every boy who asked if I could examine him.” However, (i) and (ii) on these interpretations are significantly better than the sentences in (29), which are totally unacceptable.

29.

30.

31. 32. 33.

(i) ? Dovrei interrogare ogni ragazzo che t ha chiesto se io dovessi (farlo) ‘(I) must examine every boy that has asked if I must (do that)’ (ii) ? Potrei interrogare ogni ragazzo che t ha chiesto se io potessi (farlo) ‘(I) can examine every boy that has asked if I can (do that)’ Note that there is a clear contrast here with English. In English, the past participle does not move out of the VP, as indicated by the ungrammaticality of the English counterparts of (35) and (36): (i) *The boys have eaten often pizza (ii) *The boys have eaten all pizza See Donati (2002) for a different explanation for why VP ellipsis of the English type is impossible in Italian. Donati capitalizes on the fact that V moves overtly in Italian but covertly in English. In many respects this discussion echoes points that have been made before. See especially Dalrymple (1991). Cf. Fiengo and May (1994: 275–280) for data presentation and for an early discussion of Vehicle Change. Guglielmo Cinque brought to our attention examples like (i). We seem to wrongly predict that (i) could have the meaning that I examined everyone who promised me to examine himself. We would obtain this meaning from an LF like (ii), where the semantic value of do that is identified with the semantic value of the matrix VP. We conjecture that this puzzle belongs to the same category as the puzzle we just considered. There is just something wrong with a sentence that has the meaning we would arrive at in this way, and whose semantics is built up compositionally in this way. We have speculations as to precisely what is going wrong here, but explaining them would take us too far into issues regarding “Rule I”-type effects, indexing constraints at LF, and the semantics of control. (i) I examined everyone who promised me to do that. (ii) [IP [DP everyone who [ 1 [IP t1 promised me to do that ]] ] [ 1 [IP I examined t1 ]]]

34. See Hoji (2003) for similar examples. Hoji’s examples, however, make use of do the same thing, which we suspect has different properties from do that.

102 Carlo Cecchetto and Orin Percus 35. One worth mentioning, in light of the kinds of directions pursued in the ellipsis literature, is that any such condition on VP anaphora could not be reduced to a condition on when a VP with a given meaning can be destressed. Examples with “implicational bridging” like First John sang an aria, and then Sue performed allow destressing of the second VP, but the second VP cannot be replaced with do that and interpreted the same way. 36. Followers of the VP ellipsis literature will suspect that, once we accept this condition on VP anaphora, we can also dispense with the assumptions about possible LF indexing patterns that were important to Cecchetto’s account of (42) and (43). We believe this suspicion is correct, but cannot justify this here. If this suspicion is correct, and you don’t believe that the indexing assumptions are tenable (see Kennedy 2004 for relevant discussion), then you have another reason for preferring this condition on VP anaphora to the simple semantic identity condition Cecchetto assumed.

References Belletti, Adriana 1990 Generalized Verb Movement. Turin: Rosenberg e Sellier. Cecchetto, Carlo 2004 Explaining the locality condition of QR. Consequences for the theory of phases. Natural Language Semantics 12 (4): 345–397. Chao, Wynn 1987 On ellipsis. Ph.D. diss., University of Massachusetts at Amherst. Chomsky, Noam 1972 Some empirical issues in the theory of transformational grammar. In Goals of Linguistic Theory, Paul Stanley Peters (ed.), 63–130. Englewood Cliffs: Prentice-Hall. 1995 The Minimalist Program. Cambridge, MA: MIT Press. Chung, Sandra, William Ladusaw and James McCloskey 1995 Sluicing and Logical Form. Natural Language Semantics 3: 239–82. Dalrymple, Mary 1991 Against reconstruction in ellipsis. Unpublished ms., Xerox-PARC and CSLI. Donati, Caterina 2002 Merge copy. In The Interfaces, Deriving and Interpreting Omitted Structures, Kerstin Schwabe and Susanne Winkler (eds.), 155–175. Amsterdam: John Benjamins. Elbourne, Paul 2001 E-type anaphora as NP-deletion. Natural Language Semantics 9: 241–288.

A contrastive analysis of VP ellipsis and VP anaphora

103

Fiengo, Robert and Robert May 1994 Indices and Identity. Cambridge, MA: MIT Press. Fox, Danny 1999 Economy and Semantic Interpretation. Cambridge, MA: MIT Press. Fox, Danny and Howard Lasnik 2003 Successive-cyclic movement and island repair: the difference between sluicing and VP-ellipsis. Linguistic Inquiry 34: 143–154. Goldberg, Lotus 2005 Verb-stranding VP-ellipsis: A cross-linguistic study. Ph.D. diss., McGill University. Haïk, Isabelle 1987 Bound VPs that need to be. Linguistics and Philosophy 10: 503–530. Hankamer, Jorge and Ivan Sag 1976 Deep and surface anaphora. Linguistic Inquiry 7: 391–428. Heim, Irene 1997 Predicates or formulas? Evidence from ellipsis. In Proceedings of SALT VII, A. Lawson (ed.), 197–221. Cornell University: CLC Publications. Heim, Irene and Angelika Kratzer 1998 Semantics in Generative Grammar. Oxford: Basil Blackwell. Hoji, Hajime 2003 Surface and deep anaphora, sloppy identity and experiments in syntax. In Anaphora: A Reference Guide, Andrew Barss (ed.), 172–236. Oxford: Basil Blackwell. Hirschbühler, Paul 1982 VP deletion and across-the-board quantifier scope. In Proceedings of NELS 12, James Pustejovsky and Peter Sells (eds.), 132–139. University of Massachusetts at Amherst: GLSA. Hornstein, Norbert 1995 Logical Form. Oxford: Basil Blackwell. Kennedy, Christopher 2002 Comparative deletion and optimality in syntax. Natural Language and Linguistic Theory 20: 553–621. 2004 Argument contained ellipsis revisited. Ms., Northwestern University. Lasnik, Howard 1999 On feature strength: three minimalist approaches to movement. Linguistic Inquiry 30: 197–218. Lobeck, Ann 1995 Ellipsis: Functional Heads, Licensing and Identification. Oxford: Oxford University Press. McCloskey, James 1991 Clause structure, ellipsis and proper government in Irish. Lingua 85: 259–302.

104 Carlo Cecchetto and Orin Percus Merchant, Jason 2001 The Syntax of Silence. Oxford: Oxford University Press. Ross, John R. 1967 Constraints on variables in syntax, Ph.D. diss., MIT. Rooth, Mats 1992 Ellipsis redundancy and reduction redundancy. In Proceedings of Stuttgart Ellipsis Workshop. Arbeitspapiere des Sonderforschungsbereichs 340, Bericht Nr. 29, Steve Berman and Arild Hestvik (eds.). University of Stuttgart. Sag, Ivan 1976 Deletion and Logical Form. Ph.D. diss., MIT. von Stechow, Arnim 1984 Comparing semantic theories of comparison. Journal of Semantics 3: 1–77. Tancredi, Christopher 1992 Deletion, deaccenting and presupposition. Ph.D. diss., MIT. Webber, Bonnie 1978 A formal approach to discourse anaphora. Ph.D. diss., Harvard University. Williams, Edwin 1977 Discourse and Logical Form. Linguistic Inquiry 8: 101–139.

Chapter 2 Interpretation in the DP-phase

‘HAVE’ = ‘BE’ + PREP(osition): New evidence for the preposition incorporation analysis of clausal possession Chris H. Reintges and Anikó Lipták

Freeze (1992) and Kayne (1993, 1994) present a uniform syntactic analysis for the crosslinguistically well attested similarities between locative and possessive sentences, in which possessive ‘HAVE’ is transformationally derived by incorporating a dative-benefactive or locative-comitative preposition into an existential copular verb ‘BE’. Here we focus on possessive constructions in Coptic Egyptian, which provide first-hand evidence for the ‘HAVE’/‘BE’ alteration. The Coptic evidence also shows that clausal possession cannot fully be accounted for by a mechanical application of the preposition incorporation analysis.

1. Introduction In this paper we explore a configurational analysis for the connection between locative (‘BE’ + PREP(osition)) and possessive (‘HAVE’) sentences on the one hand, and possessive noun phrases on the other hand. That there is, indeed, such a relation – both crosslinguistically and within languages – has been extensively documented in Freeze (1992). Based on newly disclosed evidence, we will derive the transparent semantic relations between clausal and nominal possession from a shared predication structure (viz., a small clause). The idea that there is a common syntactic source for possessive and locative predication goes back to classic work by Benveniste (1966, section 15). Kayne (1993, 1994), elaborating on Freeze (1992), presents a decompositional analysis of ‘HAVE’ as the lexicalization of the copular verb ‘BE’ plus a dative-benefactive (‘TO’) or locative-comitative (‘WITH’) preposition. Crucial to this analysis is the co-occurrence of an existential copular verb ‘BE’ and a dative-benefactive or locative-comitative preposition in existential-locative as well as possessive sentence constructions. Prior to

108 Chris H. Reintges and Anikó Lipták now, no language has been documented where the possessive verb ‘HAVE’ is formed by a transparent ‘BE’ + PREP(osition) complex. As we will see in this paper, Coptic Egyptian is such a language.1 To be more precise, the possessive verb wnte ‘to have’ and its negative counterpart mnte ‘to have not’ can be decomposed into a form of the copular verb wn ‘to be’ and mn ‘not to be’ and a locative-comitative preposition nte ‘with’. But the preposition nte is also used as a linkage device in complex noun phrases, where it marks the possessor noun. For this reason, Coptic possessives provide prima facie evidence for the syntactic relatedness of possession in noun phrases and in clauses. On the other hand, the Coptic facts also show that the preposition incorporation analysis of ‘HAVE’ needs to be revised in two respects. First, possessive ‘BE’ in the ‘BE’ + PREP(osition) complex must be analyzed as a functional category, since it displays substantial differences from the existential copular verb BE. Second, something more has to be said about the reassignment of Cases in the course of wnte ‘HAVE’ formation: the erstwhile prepositional object of nte ‘with’, which designates the possessor, becomes the clausal subject which is assigned nominative Case, while the erstwhile subject becomes the direct object of wnte ‘to have’ and is assigned objective Case. The syntactic derivation of possessive ‘HAVE’ sentences proceeds in a cyclic fashion or phases in the sense of Chomsky (2000, 2001). Sentential possession starts out from DP-internal possession. Since the syntactic derivation of possessive noun phrases is accomplished by interacting movement operations, nominal possessives are the clearest cases for the DP Phase (Svenonius 2004). To express possessive relation at the clausal level, the DP phase must be further embedded into a CP phase, which is brought about by the incorporation of the preposition nte into the unaccusative ‘BE’-copula wn. The organization of the paper is as follows. Section 2 discusses the theoretical background assumptions that underlie our analysis of Coptic possessives. Section 3 reviews the evidence for the correlation between existentiallocative and possessive sentences in several Afro-asiatic languages. Section 4 discusses the main descriptive facts concerning clausal possession in Coptic Egyptian, which provides prima facie evidence for a decompositional analysis of the possessive auxiliary ‘HAVE’. Section 5 examines the working of copula support in Coptic sentences with indefinite subjects. Section 6 discusses the syntactic and semantic differences between existential and possessive ‘BE’ in the language. Section 7 examines DP-internal possession,

‘HAVE’ = ‘BE’ + PREP(osition)

109

which supports the view of the parallel structure of nominal and clausal possession. Section 8 summarizes the main results of the paper. 2. A decompositional analysis of possessive ‘HAVE’ Before entering the empirical discussion on Coptic possessives, this section reviews the key issues concerning the syntax of the possessive auxiliary verb ‘HAVE’. Extending Benveniste’s (1966) influential analysis of possessive ‘HAVE’, Freeze (1992: 585–589) and Kayne (1993: 6–7, 1994: 102) take the structural similarities between locatives and possessives as point of departure for a syntactic derivation of possessive sentences from existential ‘BE’ constructions. On this analysis, ‘HAVE’ sentences are underlyingly ‘BE’ sentences, with a syntactic element being incorporated into ‘BE’. For Kayne, the incorporated element is considered to be a D(eterminer) head, which originates from a possessive nominal expression that forms the core of possessive sentences. The idea that possessive DPs and ‘HAVE’ sentences have a common source goes back to Szabolcsi’s (1983, 1994) extant work on Hungarian possessive constructions. In possessive noun phrases, the agreement between the possessor and the possessed item is registered by morphological markers that are by and large identical to subject-verb agreement inflection in transitive clauses with definite direct objects. Both possessor and subject-verb agreement is obligatory. Thus, consider (1 a–b), in which the first person singular subject agreement marker -m appears in both possessive noun phrases and clauses.2 (1)

a. az én vendég-e-m (Hungarian) the I guest-POSS-1SG ‘my guest’ b. (Én) alud-t-am. I sleep- PAST-1SG ‘I slept’

Consider next the corresponding third person singular examples, in which agreement has a null form. (2)

a. Mari vendég-e Mari guest-POSS(-3SG) ‘Mari’s guest’

110 Chris H. Reintges and Anikó Lipták b. Mari alud-t. Mari sleep- PAST(-3SG) ‘Mari slept’ Based on the parallelism between possessor and subject-verb agreement, Szabolcsi (1983, 1994) argues that the structure of possessive noun phrases mirrors clausal structure. This analysis is taken a step further by deriving possessive sentences transformationally from DP-internal possession: the possessor (POSS-or) DP in ‘HAVE’ sentences originates from the possessor position of a possessive DP. The first step of the syntactic derivation involves the extraction of the POSS-or DP from the possessive DP. Due to locality, extraction requires that POSS-or land in Spec, DP, which serves as an escape hatch for movement. The Spec, DP position is associated with dative Case, witness the obligatory dative Case of the subject in possessive sentences. (3)

Step 1: the derivation of possessive DPs [DP Marinaki [D' a [AgrP ti Agr0 [ vendég-e ]]]] Mari-DAT the guest-POSS(-3SG) ‘Mari’s guest’

The second derivational step involves the formation of possessive sentences from DP-internal possessives. The possessed (POSS-ed) DP agrees with and receives Case from an existential verb ‘BE’ after the POSS-or DP has been extracted out of the nominal possessive configuration. Since the main verb of possessive sentences is existential van ‘to be’, its subject (the POSS-ed DP) can only be indefinite. This is why the determiner that is obligatory in (3) must be missing in (4). Definitely determined subjects render the possessive sentence ungrammatical (*Marinak van a vendége, Mari-DAT is the guest, ‘Mari has the guest’). (4)

Step 2: the derivation of possessive ‘HAVE’ sentences Marinak van [DP t'i D0 [AgrP ti Agr0 [ vendég-e ]]] Mari-DAT is guest-POSS(-3SG) ‘Mari has a guest’

Crucial to the Szabolcsi-Kayne approach to possessive ‘HAVE’ is the assumption that clausal possession originates from an existential structure in which the unaccusative copular verb ‘BE’ selects a possessive DP as its

‘HAVE’ = ‘BE’ + PREP(osition)

111

complement. Subsequently, the D0 head of the possessive DP moves to and incorporates into the initial copula ‘BE’, which thus becomes complex. In Kayne’s system, the incorporation of the D0 head licenses the movement of the POSS-or DP into subject position [Spec, ‘BE’], saving the derivation from improper movement violations. Keeping the case-assigning potential of Hungarian D0 in view, the corresponding incorporating D0 in English is said to be prepositional in nature. This preposition is covert in English, yet can assign dative Case to the possessor to its left and incorporate into ‘BE’ just like a D0 head. The derived ‘BE’ + D0/P0 complex is spelled out as a separate lexical item ‘HAVE’ in English. See the diagrams in (5 a–b) for further illustration. (5)

The Szabolcsi-Kayne incorporation analysis of ‘HAVE’ a. … ‘BE’ [DP [D' D0 [AgrP DPPOSS-or [Agr' NPPOSS-ed ]]]] b. [DPPOSS-or,j ‘BE’ + D0i [DP [D' ti [AgrP tj [Agr' NPPOSS-ed ]]]]]

Although den Dikken (1995, 1997) generally agrees with the SzabolcsiKayne analysis of clausal possession, he takes a different stance on the construction of the DP-internal structure: the incorporation of the dative preposition into ‘BE’ is coupled with an additional step of predicate inversion that reverses the order of the POSS-ed and the POSS-or DPs, which represent the subject and the predicate of a possessive relation. (6)

Predicate inversion in the derivation of possessive DPs a. … ‘BE’ [AgrP DPPOSS-ed [Agr' Agr [PP PDAT DPPOSS-or ]]] b. [pp tj DPPOSS-or ]i Pi + ‘BE’ [AgrP DPPOSS-ed [Agr' tj ti ]]

While Coptic possessives provide empirical support for the ‘HAVE’/‘BE’ alteration, they cannot fully be accounted for by the mechanical application of the preposition incorporation analysis, since existential and possessive ‘BE’ show distinctive morpho-syntactic behaviour.

3. Copula support in possessive sentences The empirical evidence adduced for the decompositional analysis of possessive ‘HAVE’ as a ‘BE’ + PREP composite comes from languages where the copula BE and a locative preposition express clausal possession, but occur as distinct syntactic elements. Yet, in several of the Afro-asiatic languages

112 Chris H. Reintges and Anikó Lipták we discuss here, the presence of such a ‘BE’-copula is by no means obligatory. Rather, a copula is introduced to support tense distinctions. The notion of possession therefore hinges entirely on the selected dative or locative preposition. 3.1. The dative-benefactive pattern In Classical and Modern Standard Arabic, the preposition li- ‘to, for’ is widely used for the expression of clausal possession. As we can see from the contrast between examples (7a) and (7b), the dative-benefactive pattern may designate permanent as well as temporary possession (see, among various others, Fischer 2002: 157, §295; Badawi, Carter and Gulli 2004: 190, §2.6.10) (7)

a. wa-hiyya waHad-h la-h (Standard Arabic) and-she alone-POSS.3 F.SG to-3 F.SG akl-un x ss-un mumayyaz-u-n. appearance-NOM-INDEF special-NOM-INDEF peculiar-NOM-INDEF ‘and she alone has a special distinguished face.’ (Badawi, Carter and Gulli 2004: 190) b. /al-ban- t-u la-hunna ul-u-n kaTr-u-n. the-girl-F.PL-NOM to-3 F.PL work-NOM-INDEF much-NOM-INDEF ‘The girls have much work.’ (Cowan 1958: 52)

In present tense sentences, the dative-benefactive pattern lacks an overt copula. By contrast, the auxiliary verb k na ‘to be’ must appear in the past or future tense contexts (Badawi, Carter and Gulli 2004: 404, §3.16.3.1). The dative preposition li- does not overtly incorporate into the auxiliary k na, whose sole purpose is to import temporal semantics. Thus, the copular verb ‘BE’ has no role at all in bringing about the possessive reading. (8)

a. kna li-l-?abd-i Himar-u-n. be.PERF.3M.SG to-the-servant-GEN donkey-NOM-INDEF ‘The servant had a donkey.’ (Fischer 2002: 157) la-hum ul-/awwaliyyat-u. b. wa-li-D lika sa-taknu and-for-this FUT-3 F.SG.IMPERF.be for-3M.PL the-priority-NOM ‘Therefore, they will have priority’ (Badawi, Carter and Gulli 2004: 404)

‘HAVE’ = ‘BE’ + PREP(osition)

113

The decompositional approach of possessive ‘HAVE’ as a syntactically derived ‘BE’ + PREP composite would be strengthened empirically, if a language could be found where the possessive verb ‘HAVE’ still shows the effects of preposition incorporation. According to Ouhalla (2000), the auxiliary verb all-ä in Amharic is just that, namely a copula verb BE plus an incorporated dative preposition lä- ‘to’. The possessive use of all-ä ‘there is’ is exemplified in (9a). On his analysis, the ungrammaticality of (9b) follows from economy considerations: since the auxiliary allä already contains the dative preposition lä-, the marking of the possessor noun with another instance of that preposition is redundant and therefore excluded. (9)

a. Aster ÈhÈt all-ä-at. (Amharic) Aster sister there.is.PERF-3M.SG-3 F.SG ‘Aster has a sister.’ (Yimam 1997: 628–629) b. *lä Kassa däbtar all-ä-w. to Kassa notebook there-is-PERF-3M.SG-3M.SG ‘Kassa has a notebook.’ (Adapted from Ouhalla 2002: 223)

However, there is no ‘BE’-type copula al in the language that could serve as a host for preposition incorporation. The Amharic auxiliary all- has Classic Ethiopian and Tigre cognates with the root
hlw ‘to exist, be present’. As expected, it functions as copular verb in existential sentences (Leslau 1995: 527–528, §§83.1–83.2). (10) bä-gäbäya bzu säw allä. in-market many man there-is ‘There are many people in the market.’

(Leslau 1995: 528)

Moreover, the auxiliary all- has present tense reference. In past tense contexts, the auxiliary näbbar-ä ‘there was’ must be chosen, which has a cognate root in Classic Ethiopian
nbr ‘to sit’ (Yimam 1997: 621–626). (11) Aster gänzäb näbbär-ä-at. Aster money there.was.PERF-3M.SG-3 F.SG ‘Aster had money’

(Yimam 1997: 624)

See Leslau (1995: 531, §83.12) and Yimam (1997: 628–634) for relevant discussion on the appearance of accusative/dative clitics on the auxiliary verb all-.

114 Chris H. Reintges and Anikó Lipták 3.2. The comitative-locative pattern For Moroccan Arabic, Ouhalla (2000: 227ff.) argues that the locative-comitative pattern has a possessive interpretation only if the POSS-ed subject is indefinite, as in (12a). The possessive reading is no longer available when the POSS-ed subject is a definite noun phrase, as in (12b). In this case, the locative-comitative pattern describes temporary possession. (The corresponding English sentences have the same properties.) (12) a. Nadia ?nd-ha ktab. Nadia with-3 F.SG book ‘Nadia has a book.’ l-ktab. b. Nadia ?nd-ha Nadia with-3 F.SG the-book ‘Nadia has the book.’

(Morrocan Arabic) (Ouhalla 2002: 228)

(Ouhalla 2002: 228)

Akin to Moroccan Arabic, possessive sentences in Hausa (Chadic, Nigeria) are formed with a locative preposition dà ‘with’, but unlike Modern Standard and Moroccan Arabic, the presence of the imperfective auxiliary is obligatory. Moreover, the subject of the resulting construction designates the POSS-or and the DP object of dà, the POSS-ed noun (Newman 2000: ch. 33). (13) a. y riny  tan

dà zb. girl 3 F.SG.IMPERF with ring ‘The girl has a ring.’ b. kan

dà mt ? 2M.SG.IMPERF with car ‘Do you have a car?’

(Hausa) (Newman 2000: 222)

(Jaggar 2001: 470)

This brief typological review has shown that an overt ‘BE’-copula is not an indispensable part of clausal possession. In Arabic, such a copula only supports temporal and aspectual semantics. In Amharic, copula support in possessive clauses is mandatory, but does not necessarily involve ‘BE’.

4. The Coptic wnte- ‘HAVE’ construction The Coptic wnte ‘HAVE’-construction provides the ‘missing link’ for the decompositional approach to the ‘HAVE’/‘BE’ alteration. The possessive

‘HAVE’ = ‘BE’ + PREP(osition)

115

auxiliary verbs wnte ‘to have’ and mnte ‘to have not’ consist of the copular items wn ‘to be’ and mn ‘not to be’ and the comitative preposition nte ‘with’ (e.g., Layton 2000: 305, §308). Thus, compare the existentiallocative in (14) with the corresponding possessive sentence in (15). In (14), the ‘BE’-copula wn precedes the partitive wh-phrase w r n-oeik ‘how many (loaves) of bread’, while the locative-comitative phrase nte-t utn ‘with you (plural)’ appear in clause-final position. (14) wn w r n-oeik nte-t utn? BE how.many LINK-bread with-2PL ‘How many (loaves) of bread do you have (literally ‘are there with you’)?’ (Matthew 15: 34) By contrast, the copula wn and the preposition nte form a single word wnte in possessive sentences like (15). (15) ne-wnte p-rro salpigks snte n-nu n-tath. PRET-HAVE DEF.M.SG-king trumpet two LINK-gold LINK-refined ‘The King had two trumpets of refined gold.’ (Eudoxia 60: 13) In Coptic, possessive predication is computed on the basis of existentiallocative predication: HAVE (wnte) = ‘BE’ (wn) + WITH (nte). Further evidence for the locative source of possessive predication comes from the optional presence of the locative expletive mmau ‘there’ (Reintges 2004: 400, §10.2.3; see also Freeze 1992: 581–582).3 e-mnt-u nune mmau DEM.PL PCL REL-HAVE.NOT-3PL root there ‘But as for them, who have no root’

(16) nai

de

(Luke 8: 13)

In possessive ‘HAVE’ sentences, the incorporated preposition nte is adjacent to the BE-element wn- and appears, as a consequence, to the left of the possessor subject and the possessed direct object. The resulting VSO structure involves a departure from the canonical SVO word order (Layton 2000: 306, §373; Reintges 2004: 401, §10.2.3.2). (17) Verb Subject (POSS-or) Object (POSS-ed) wnte n-halate n-t-pe ne.u-mah. HAVE DEF.PL-birds LINK-DEF. F.SG-sky DEF.PL.3PL-nest ‘And the birds of the sky have their nests.’ (Luke 9: 58)

116 Chris H. Reintges and Anikó Lipták To accommodate the VSO word order of possessive clauses, two types of analyses may be envisaged, which are schematically represented in (18a) and (18b). On one analysis, the incorporation of the preposition nte ‘with’ into the copular verb wn and the raising of the POSS-or DP to the postverbal subject position are two distinct though related movement operations. Alternatively, one could derive the surface order from the PP-fronting of the entire nte-phrase in front of the copula wn without applying further movement operations that would break up the structure. (18) a. Preposition incorporation and possessor DP raising [ … [ ‘BE’-nte [ [ DP ] … ]]]] b. PP-fronting of the nte-phrase [ … [ ‘BE’ [PP nte [ DP ]] … ]]]] The PP-fronting analysis in (18b) looks more economical than the competing preposition incorporation analysis in (18a) above, since it involves a single movement operation. Despite its conceptual appeal, the PP-fronting analysis is, however, not viable. If there were, indeed, a phrasal boundary between the copula verb wn and the locative-comitative preposition nte, one might expect to find examples with material intervening between these elements. Such examples are, however, systematically absent in our documentation of Coptic Egyptian. The strict adjacency requirement that holds between wn- and nte can readily be explained under the preposition incorporation analysis. Moreover, prepositional phrases appear either clausefinally after the verb and its complements or clause-initially, preceding both the tense-aspect-mood marker and the subject. The preposed placement of locative phrases is exemplified in (19), where the PP nnahrn p-nte ‘before God’ precedes both the existential verb mmn ‘not to be’ and the indefinite subject laau n-at-kyom ‘anything impossible’. (19) nnahrm p-nute de mmn laau n-at-kyom. before DEF.M.SG-god PCL BE.NOT any LINK-NEG.PFX-power ‘Before God, (there) is not anything impossible.’ (Matthew 19: 26) Finally, an alternative word order pattern can be observed if the POSS-or subject is a DP and the POSS-ed object an enclitic pronoun. As we can see from (20), the POSS-ed pronoun 3rd SING feminine -s ‘it’ cliticizes to the right of the preposition nte. As a result, it precedes the POSS-or pa-eiOt ‘my father’.

‘HAVE’ = ‘BE’ + PREP(osition)

pa-eiOt ] (20) nka nim [ete wnte-s thing every CREL HAVE-3 F.SG DEF.M.SG.1SG-father ‘everything that my Father has’

117

(John 16: 15)

Since nte originates from the POSS-or DP, we can be sure that it moves to the position adjacent to wn- ‘BE’ by itself, i.e. independently of the POSSor DP.

5. Copula support and the definiteness restriction Our point of departure for the analysis of Coptic possessives is existential and existential-locative sentences of the kind in (21) and (22). In both existential constructions, the copular verbs wn ‘to be’ and mn ‘not to be’ appear to the left of an indefinite subject. The main difference between the ‘bare’ existential constructions in (21) and existential-locative constructions in (22) is that existentials have an implicit locative adjunct, whereas existential-locative sentences contain an overt locative phrase. (21) a. wn

aggelos. angel ‘Angels exist.’

BE

b. mn laau. BE.NOT someone ‘(There) is no one (here).’

(Acts 23: 8)

(V. Pach. 1: 7)

(22) a. ne-wn hen-oos de hm p-ma et-mmau. PRET-BE INDEF.PL-shepherd PCL in DEF.M.SG-place that ‘(There) were shepherds in that place.’ (Luke 2: 8) m-pei-ma nmma-i. b. mn laau n-rOme BE.NOT someone LINK-man in-DEM.M.SG-place with-1SG ‘(There) is noone here with me.’ (Mena 14b: 29–31) What interests us here is that both types of existentials demonstrate a ‘definiteness restriction’, meaning that only indefinite subjects are licensed in the post-copular subject position (see, among various others, Milsark 1977; Heim 1987; McNally 1998). Significantly, the copulaless version of English a book is under the table is unavailable in Coptic. Instead, an existentiallocative construction with the ‘BE’-copula is used. That copula support is,

118 Chris H. Reintges and Anikó Lipták indeed, contingent on the definiteness restriction is evident from locative sentences with definite subjects in which copula support no longer applies. (23) ta-me mn pa-na nmma-f. DEF. F.SG.1SG-truth and DEFM.SG.1SG-mercy with-3M.SG ‘My truth and my compassion (are) with him.’ (Psalm 88: 24) Copula support is not confined to existential-locative sentences, but also applies to Present tense and Near Future sentence constructions with indefinite subjects. As expected, copula support is blocked in the context of definite subjects, as shown by the corresponding b-examples. (24) a. m wn meewe polymei nmma-k? (Present) Q BE thought fight(-INF) with-2M.SG ‘Are (there) any thoughts troubling you?’ (AP Chaîne no. 181: 44, 16–17) b. eye ere ne.tn- re nute eol hn nim? Q REL(-PRES) DEF.PL.2PL-son cast(-INF) PCL in who? ‘In whom are your sons casting out (demons)?’ (Luke 11: 19) (25) a. wn u-mnt-ebei n na-taho-u. (Future) BE INDEF.SG-NOMINAL-misery AUX-come.upon-3PL ‘A misery will come upon them.’ (V. Pach. 90: 28–91, 1) b. se-na-tsa o ero-f nkyi n-eol 3PL-AUX-teach(-2SG.F) about-3SG.M FOC DEF.PL-out hn ta-phyl. from DEF.SG.M.1SG-tribe ‘The people of my tribe will inform you about it (the tomb).’ (Eudoxia 58: 25–26) In other tense-aspect-mood conjugations, for instance the Perfect, the presence of an indefinite subject does not trigger copula support. (26) a

u-son

tne Apa Sarapion (…) (Perfect) ask(-INF) Apa Sarapion ‘A brother asked Apa Sarapion (…)’ (AP Chaîne no. 28: 5, 24)

PERF INDEF.SG-brother

At present stage of research, we do not have an explanation to offer for the restriction of copula support to present and future tense contexts. However,

‘HAVE’ = ‘BE’ + PREP(osition)

119

it is clear that in these contexts the copulas wn and mn no longer function as existential verbs, since there is a verbal predicate. Possessive sentences provide yet another context, in which the ‘BE’-copulas are deprived of their existential semantics. In not showing definiteness effects, they differ, however, systematically from both existential-locative and Present tense sentences.

6. Asymmetries between possessive and copular ‘BE’ The Coptic wnte- ‘HAVE’ pattern is of considerable theoretical interest, showing that the preposition incorporation analysis alone cannot accommodate the semantically and syntactically contrastive behavior of existential and possessive ‘BE’.

6.1. No definiteness restriction While underlyingly similar, existential and possessive sentences display striking differences, most notably, the absence of the definiteness restriction on the postverbal subject. Thus, consider: (27) a. p-hof wnte te.f-matu ne.s.i. DEF.M.SG-snake HAVE DEF. F.SG-venom DEF.PL.3 F.SG-limit ‘As for the snake, its venom has its limits.’ (Sh.Chass. 28: 24–26) b. hoson wnte-tn p-woein (…) HAVE-2PL DEF.M.SG-light C ‘As long as you have the light (…)’

(John 12: 36)

In existential-locative sentences, the indefinite subject cannot undergo subject-verb inversion or clitic-left-dislocation, because such discourse-driven reordering processes would require the insertion of a co-referential resumptive pronoun in the subject position and hence, violate the definiteness restriction (Reintges 2004: ch. 10). Since the definiteness restriction is not operative in possessive sentences, their subject has a greater positional freedom. The inversion of the possessor p- re m-p-rOme ‘the Son of Man’ is shown in (28), in which the new information status of the inverted subject is indicated by the focus particle nkyi.

120 Chris H. Reintges and Anikó Lipták (28) wnt-f eksusia mmau nkyi p- re HAVE-3M.SG authority there FOC DEF.M.SG-son m-p-rOme e-ka noe eol. LINK-DEF.M.SG-man to-put(-INF) sin PCL ‘The Son of Man has authority to forgive sins.’ (Mark 2: 10, ed. Quecke) If we were to derive possessives from the incorporation of a locative preposition into existential BE, the loss of the definiteness restriction in the resulting possessive construction would be mysterious. We thus conclude that the wn- ‘BE’ component of wnte ‘HAVE’ has a different specification and syntactic distribution than the copula wn- ‘BE’ in existential-locative sentences.4

6.2. Optional deletion of the wn-‘BE’ element The contrastive behavior between existential and possessive ‘BE’ are the result of categorial reanalysis of the erstwhile existential verb as a functional element in the wnte ‘HAVE’-construction. More specifically, possessive wn- expresses tense and finiteness, while the locative-comitative preposition nte is the predicator. On this analysis, we understand why it is possible to delete possessive wn- from the surface structure of possessive clauses, while nte cannot be so elided. (29) nka nim [ et-nta-f ] thing every CREL-WITH-3M.SG ‘everything that he has’

(Matthew 13: 44)

The possibility of deletion becomes available because the relative complementizer et- itself is endowed with tense and finiteness features. The encoding of these features by means of wn-component is therefore not strictly necessary.

6.3. Objective Case and ‘HAVE’ raising While the copular verbs wn and mn in existential-locative constructions are bona fide unaccusative verbs, the possessive counterparts wnte and mnte are transitive verbs that are capable of assigning objective Case to

‘HAVE’ = ‘BE’ + PREP(osition)

121

the POSS-ed DP. There are two objective Cases in Coptic, accusative and oblique case, where accusative Case requires strict adjacency. For this reason, the clause-internal negation adverb an ‘not’ in (30a) comes after the transitive verb tnnu ‘to send’ and the direct object pe.f- re ‘his son’. No such adjacency requirement is at work in the oblique Case pattern, where the direct object is marked by the semantically vacuous preposition n-. As seen in (30b), the negation an intervenes between the verb soof ‘to defile’ and the oblique object m-p-rOme ‘the man’ (Reintges 2001). (30) a. nt-a p-nute gar tnneu pe.f- re an REL-PERF DEF.M.SG-god PCL send DEF.M.SG.3M.SG-son not e-p-kosmos (…) to-DEF.M.SG-world ‘God has not sent his son to the world (…)’ (John 3: 17) b. e-wom de -mpe-k-ja toot-k sOOf to-eat PCL REL-NEG.PERF-2M.SG-wash hand-2M.SG defile an m-p-rOme. not PREP-DEF.M.SG-man ‘To eat without having washed your hands does not defile the man.’ (Matthew 15: 20) The possessive auxiliary wnte- ‘HAVE’ is compatible with both accusative and oblique Case (cf. also Bach 1967; Freeze 1992; Moro 1997). The accusative object u-sOne ‘a sister’ in (31a) appears to the left of the locative expletive mmau ‘there’, whereas the oblique object n-u-m e n-khr ma ‘a lot of money’ in (31b) appears to its right. mmau (31) a. ne-wnte p-rro KOstantinos u-sOne PRET-HAVE DEF.M.SG-king Constantine INDEF.SG-sister there m-parthenos. LINK-virgin ‘King Constantine had a virgin sister.’ (Eudoxia 50: 3–4) n-khr ma. b. ne-wnta-f mmau n-u-m e PRET-HAVE-3M.SG there PREP-INDEF.SG-mass LINK-money ‘He had a lot of money.’ (Mena 13a: 8–10) In contradistinction to run-of-the-mill transitive verbs, there is no adjacency requirement between the possessive verb and the direct object, given that possessive clauses have VSO order. Moreover, discourse particles may

122 Chris H. Reintges and Anikó Lipták intervene between the POSS-or subject and the accusative case-marked POSS-ed object (e.g., Layton 2000: 310, §309a). At present, it is not clear which factors underlie the selection of accusative and oblique case marking in possessive sentences. (32) ne-wnte t-orp men hen-dikaiOma n-me. PRET-HAVE DEF. F.SG-first PCL INDEF.PL-rules LINK-worship ‘The first one had regulations for worship.’ (Hebrews 9: 1) While lexical verbs never leave the IP domain, the possessive verb wnte occupies a pre-IP position, which can be identified with Rizzi (1997)’s FIN(inteness) Phase. As a result, wnte is in complementary distribution with other tense/aspect/mood markers that occupy the FIN0-node.5

6.4. Interim summary In the previous sections we studied the morphosyntax of Coptic possessive constructions in the clausal domain. We observed the following facts and properties: (i) The possessive verbs wnte ‘to have’ and mnte ‘to have not’ transparently reflect the incorporation of a locative-comitative preposition nte into a copula element wn ‘to be’ and mn ‘not to be’. This incorporation step involves the preposition alone rather than the entire prepositional phrase. (ii) There exist some surprising asymmetries between existential and possessive ‘BE’, most notably the disappearance of the definiteness restriction and the availability of objective Case in the latter. In the following section, we will outline a syntactic analysis for the observed facts, in which the connection between nominal and clausal possession plays a crucial role.

7. The configurationality of clausal possession Our path towards a configurational analysis of possessive sentences leads us through DP-internal possession. This is because possession in noun phrases and in clauses share important structural properties in the languages

‘HAVE’ = ‘BE’ + PREP(osition)

123

we study. The exact nature of the relatedness between nominal and clausal possessives is controversial. That there exists, indeed, such a derivational relationship between both types of possessives is challenged Ouhalla (2000), who argues that even the semantic relation between the possessor and the possessed DP is different in each of the two constructions. Our findings support the Szabolcsi-Kaynean view that clausal possession is syntactically derived from nominal possession. This is, we think, a warranted result, since it offers a principled syntactic explanation for the semantic similarities between nominal and clausal possessives, both of which express the notional category of ‘possession’. For the configurational analysis of clausal possession we combine two strands of current minimalist research. On the one hand, we assume a locality theory along the line of Chomsky (2000, 2001), according to which syntactic derivation proceeds successively through distinct cyclic domains or ‘phases’. Phases represent relatively autonomous derivational units, containing propositional content. Given the conception of the phase as a propositional unit, possessive DPs are perhaps the clearest instances of the DPphase, since the relation between the POSS-or and the POSS-ed DP is one of predication (Svenonius 2004). Following den Dikken (1995), we assume that possessive DPs are built around a small clause kernel with propositionlike qualities. We furthermore follow den Dikken and Singhapreecha (2004) and den Dikken (2006) in carefully distinguishing linkers from relators. While relators are predicative expressions, linkers are void of semantic content and merely signal the application of a DP-internal movement operation (‘predicate inversion’).

7.1. DP-internal possession and linkers We include attributive modifiers in our discussion of DP-internal possession, since one type of possessive DPs employs the same linking device, viz. the proclitic particle n-/m-, as seen in (33a) and (33b). The other type of complex possessive DPs uses the by now familiar locative-comitative preposition nte ‘with’, which is exemplified in (33c) (Reintges 2004: 90– 96, §§3.1.3–3.1.4).6 (33) a. t-diath k n-rre DEF. F.SG-covenant LINK-new ‘the new convenant’

(attributive n-/m-) (Matthew 26: 28)

124 Chris H. Reintges and Anikó Lipták b. t-epsitol m-p-rro (possessive n-/m-) DEF. F.SG-letter LINK-DEF.M.SG-king ‘the letter of the king’ (Hilaria 10: 32) c. nife nte p-nute breath WITH DEF.M.SG-god ‘breath of God’

(possessive nte) (2 Timothy 3: 16)

Following den Dikken and Singhapreecha (2004) and den Dikken (2006), we analyze the linker n- as a semantically void linker, whose presence signals the application of predicate inversion inside the DP. The inverted order AP n-NP is retained in examples of the following kind. (34) p-noky n-nute DEF.M.SG-great LINK-god ‘the great God’

(attributive n-/m-) (Titus 2: 13)

That the adjective has, indeed, undergone predicate inversion is also evident from postnominal attributives that appear without any such linker (see Borghouts 1980 for the pre-Coptic precursors of possessive nte-). (35) t-eere  m DEF. F.SG-girl little ‘the little girl’

(Matthew 9: 24)

To arrive at the canonical order DP n-AP like t-diathk n-
rre ‘the new convenant’ in (33a) above, we, again, follow den Dikken and Singhapreecha (2004: 20 ff.) in assuming that the modified DP moves past the inverted attributive modifier to the specifier of a higher functional projection, whose head is instantiated by the linker n-. The syntactic derivation of the different n-marked attributive constructions is schematically represented in (36). The structure in (36b) corresponds to the more marked order with pre-nominal attributes AP n-DP (as in (34)), while structure in (36c) describes the canonical post-nominal order DP n-AP (cf. (33a–b)). Granted that nominal possessives have an underlying small clause structure with the possessor as the predicate, the syntactic derivation of nominal possessives of the form DP n-DP in examples like t-epsitol m-p-rro in (33b) above therefore runs entirely parallel to the one of attributive modification.

‘HAVE’ = ‘BE’ + PREP(osition)

125

(36) Deriving modification with the linker na. [SC NP AP] merging functional structure F1P; AP-to-Spec,F1P (predicate inversion), spelling out F0 as n-  b. [F1P APi [F1’ n- [SC NP ti ]]] merging another functional FP; NP-to-Spec,F2 P, n- head moves to F2  c. [F2P NPj [F2’ n- [F1P APi [F’ tlink [ SC tj ti ]]]]] Unlike the linker n-, the preposition nte ‘with’ does not appear with attributive modifiers. Although the contrastive behavior of the linker n- and the preposition nte needs more research, it generally seems to hold that the linker n- is selected when the POSS-or and the POSS-ed agree in definiteness, as in (33b) above, while the preposition nte is chosen when there is a mismatch in definiteness, as in (33c) above. When both n- and nte are combined, n- appears in a higher structural position. (37) hah n-remao nte t-polis many LINK-rich.man WITH DEF. F.SG-city ‘many rich men of the city’

(KHML I 72: 1)

The DP-internal markers n- and nte are fundamentally different categories: the former is a functional category that signals inversion, while the latter is a relational category with locative semantics. Therefore, possessive DPs with nte contain a prepositional small clause, with the possessor as complement to P0 head.7 (38) [PP DP/NP POSS-ed

[P nte [DP POSS-or]]]

Clausal possession was shown to be derived by preposition incorporation. Since the ‘BE’-component of wnte ‘HAVE’ does not qualify as an existential verbal predicate, one may plausibly assume that the preposition nte constitutes the semantically meaningful predicate of the possessive clause. Since the linker n- has no such relational or predicative meaning, it couldn’t possibly contribute semantic content to the BE-copula to turn it into a predicate.

126 Chris H. Reintges and Anikó Lipták 7.2. Deriving clausal possession from DP-internal possession The derivation of clausal possession basically involves the construction of a CP phase on top of a DP phase. The DP phase starts out from a small clause configuration with the locative-comitative preposition nte as its head. On top of the small clause/PP, there is a DP-layer which serves as the complement of the copula element ‘BE’. Within this DP phase, minimally two movement operations must be synchronized. On the one hand, there is the raising of the POSS-or nominal from the complement position of the ntePP to the Spec, DP. On the other hand, there is the raising and subsequent incorporation of the small clause head nte into the D0 head. Both movement steps are required, because the possessor and the preposition will move out of the DP, into the higher clausal structure. To allow the derivation to proceed further, both elements must move to the edge of the DP phase. The derivation proceeds further by merging the copula ‘BE’ in a V0node on top of the DP-structure.8 Subsequently, ‘BE’ provides a host for the raised P0 + D0 complex. For us, this derivational step is what distinguishes clausal from nominal possession in this language. The newly created ‘BE’ + P0 + D0 composite has Case-assigning potential, yet differs from lexical transitive verbs in that it does not stay in-situ in the IP-domain, but rather moves to the FIN0-node, which constitutes the lowest left-peripheral heads on top of the IP domain (see Rizzi 1997 and much related research). In this position, ‘BE’+ P0+D0 is in complementary distribution with other pre-IP tense/aspect/mood markers. Akin to such markers, the possessive auxiliary wnte ‘have’ does not show any definiteness restrictions. The derivation of possessive 'HAVE' from preposition incorporation is indicated by the arrows in the following tree diagram. The tree structure in (39) also captures the additional movement operations in possessive clauses, which are motivated by Case-requirements. The POSS-or always moves to the subject position Spec,IP to receive nominative Case. In this position, it appears to the right of the ‘BE’ + P0+D0 complex in Fin0. For the DP residue containing the POSS-ed DP, there are two Case assigning possibilities, viz. accusative and oblique Case (see, above, section 6.4). When assigned accusative Case, the DP residue moves to the Spec,VP position. When the possessed DP remains in-situ in the specifier of the small clause, the semantically vacuous locative preposition n- is introduced into the structure to avoid a Case Filter violation.

‘HAVE’ = ‘BE’ + PREP(osition)

(39)

127

FINP

3

‘BE’+P0+D0

IP

3 DPPOSS-or

I

3

tV‘BE’+ P0+ D0

VP

3

V

3

tV‘BE’+ P0+D0

DP

3

tDPPOSS-or

D

3

t P0+D0

PP=SC

3 DPPOSS-ed

P

3

tP0

tDPPOSS-or

A question remains with respect to the Case role of the incorporated preposition nte. Since this preposition incorporates into ‘BE’, and thus provides ‘BE’ with case-assigning capacities, its trace in the small clause kernel is case-impaired. Therefore, the erstwhile prepositional object, the possessor DP is forced to move to the Spec,IP position for gaining nominative Case.

8. Concluding remarks The Coptic language facts shed new light on the relation between existential-locative and possessive sentences. The possessive verb wnte ‘HAVE’ comprises a ‘BE’ element wn and a locative-comitative preposition nte ‘with’ and thus bears prima facie evidence for Szabolcsi’s (1983, 1994) and Kayne’s (1993, 1994) decompositional approach to the ‘HAVE’/‘BE’ alteration. Yet, the Coptic evidence also shows that existential-locative and possessive constructions differ systematically from one another with respect to the operativeness of the definiteness restriction. We have shown that next to incorporation, the prerequisite for deriving possessive ‘HAVE’ from

128 Chris H. Reintges and Anikó Lipták ‘BE+WITH’ is the semantic bleaching of existential ‘BE’ into a functional category. In agreement with the Szabolcsi-Kaynean view, we argued that Coptic clausal possession cannot be derived from an existential-locative construction. Rather, it represents a CP structure that is built on possessive DP structure, with a ‘BE’-element mediating between the DP and the CP phases.

Notes 1. Coptic Egyptian is the indigenous language of late-antique and early medieval Christian Egypt (from about the third to the eleventh century CE) and represents the final development stage of Ancient Egyptian (Afro-asiatic). Coptic Egyptian is actually a dialect cluster with at least six regional varieties, two of which gained supra-regional importance: Sahidic Coptic, the vernacular of Upper Egypt, and Bohairic Coptic, the vernacular of Lower Egypt. The data in this article are taken from Sahidic Coptic, which due to its early records and rich literature represents the main reference dialect for Coptic Egyptian. For the textual sources of the Sahidic Coptic examples, see Reintges (2004: 597– 600). 2. The following abbreviations are used in the glosses: AGR ‘agreement’; CAUS.INF ‘causative infinitive’; C ‘subordinating complementizer’; CREL ‘relative complementizer’; COND ‘conditional’; CONJ ‘conjunctive’; DEM ‘demonstrative article’; IMPERF ‘imperfective’; INF ‘infinitive’; LINK ‘linker’; NEG.PFX ‘negative prefix’; OBL ‘oblique Case’; PCL ‘particle’; PERF ‘perfect’; POSS ‘possessive’; PREP ‘preposition’; Q ‘yes/no question particle’; REL ‘relative marker’. Glosses are given in parentheses for morphemes that have no surface-segmental shape. 3. Similar facts have been reported for Italian, where possessive ‘HAVE’ constructions may include the expletive pronoun ci ‘there’ (Moro 1997: 236–242). Although ci can be omitted at a high stylistic level, as seen in (ia), it becomes obligatory when the object of avere ‘to have’ is cliticized, as seen in (ib). Crucially, ci cannot occur with the possessive verb possedere ‘to possess’. (i) a. i professori (c’)-hanno molti libri (Italian) the professors there-have many book ‘Professors have many books’ b. i professori ??(ce)-lii hanno ti the professors there-them have ‘Professors have them’ c. i professori (*ci) possiedono molti libri the professors there have many book ‘Professors have many books’ (Moro 1997: 237 (50)) See Moro (1997: ch. 2, 2000: 109ff.) for further discussion of the syntax of ci.

‘HAVE’ = ‘BE’ + PREP(osition)

129

4. Further support for the distinctiveness of possessive ‘BE’ from existential ‘BE’ comes from the availability of two locative arguments (not adjuncts). Consider the possessive sentence in (i), in which both the POSS-or and the POSS-ed DPs appear in the complement position of a locative preposition, viz. the comitative preposition nte ‘with’ and the directional preposition e- ‘to, against’. It is hard to imagine what this sentence would mean if the BE element was still a fullyfledged existential verb, given that asserting the existence of a location is semantically anomalous. (Consider, for instance, the ungrammaticality of *There is in the garden). PREP1 POSS-or PREP2 POSS-ED (i) BE ne-w(n)- nte- u-danist s erOme snau pe PRET-BE- WITH INDEF.SG-debtor TO man two COP ‘A certain creditor had two debtors.’ (Luke 7: 41) Layton (2000: 306, §383) posits an implicit indefinite pronoun denoting the unexpressed possessed item, corresponding to the paraphrase ‘A creditor had something against two people’. The problem with this analysis is that in Coptic indefinite pronouns are never pro-dropped. Our understanding of this passage is different. For us, the POSS-ed DP is encoded as a directional phrase to express the idea that the possessive relationship has a negative effect on its referent, i.e. he designates a negatively affected entity designated by POSS-ed DP. On this analysis, the ‘BE’-component of wnte- ‘HAVE’ has by itself no existential force but rather provides a syntactic host only for the raised preposition. 5. For British English, den Dikken (1997: 131) observes that possessive have can raise past a negation or a subject, as seen in (i a–b). It thus behaves differently from non-possessive have, as seen in (ii a–b): (i) a. John hadn’t the faintest idea. b. Have you a daughter? (ii) a. *John hadn’t Mary throw up on him. b. *Had John Mary throw up on him? 6. Complex possessive noun phrases with the linker n- lack the inverted alternant DP POSS-or n- DP POSS-ed. We can safely assume that attributive modifiers the linker n- are not compounds, because such compounds involve the phonological reduction of the modified head noun, e.g. rp n-as ‘old wine’ vs. rp-as ‘vintage’ wine (Layton 2000: 90, § 112). 7. Den Dikken (2006) argues that any predication relationship involves a RELATOR element between the subject and the predicate. The RELATOR is a functional category, which permits either the subject or the predicate to occupy its specifier or complement position. This RELATOR therefore constitutes the head of the small clause. Our view of predication is not formulated in terms of a ‘RELATOR phrase’. 8. See Moro (1997: 241) for a related view. In Moro’s system, the possessive verb avere ‘to have’ is analyzed as an auxiliary verb, as in our proposal, but

130 Chris H. Reintges and Anikó Lipták not further decomposed into a ‘BE’ plus a preposition. Rather, he regards the auxiliary essere ‘to be’ as support for the V0 that incorporates into a single agreement node, while avere ‘to have’ is support for V0 that incorporates into two AGR nodes.

References Bach, Emmon 1967 Have and be in English syntax. Language 43: 462–485. Badawi, Elsaid, Michael G. Carter and Adrian Gully. 2004 Modern Written Arabic: A Comprehensive Grammar. Routledge Comprehensive Grammars. London /New York: Routledge. Benveniste, Emil 1966 Problèmes de Linguistique Génerale. Paris: Editions Gallimard. Borghouts, Joris F. 1980 Late Egyptian precursors of Coptic ‘genitival’ nte-/nta=. Studien zur altägyptischen Kultur 8: 65–78. Chomsky, Noam 2000 Minimalist inquiries: The framework. In Step by Step: Essays on Minimalist Syntax in Honor of Howard Lasnik, Roger Martin, David Michaels and Juan Uriagereka (eds.), 89 –155. Cambridge, MA: MIT Press. 2001 Derivation by phase. In Ken Hale: A Life in Language, Michael Kenstowicz (ed.), 1–52. Current Studies in Linguistics 36. Cambridge, MA: MIT Press. Cowan, David 1958 An Introduction to Modern Literary Arabic. Cambridge: Cambridge University Press. Dikken, Marcel den 1995 Copulas. Ms., Free University of Amsterdam. 1997 The syntax of possession and the verb ‘have’. Lingua 101: 129–150. 2006 Relators and Linkers: The Syntax of Predication, Predicate Inversion, and Copulas. Linguistic Inquiry Monograph 47. Cambridge, MA: MIT Press. Dikken, Marcel den and Pornsiri Singhapreecha 1997 Complex noun phrases and linkers. Syntax 7: 1–54. Fischer, Wolfdietrich 2002 A Grammar of Classical Arabic. Translated by Jonathan Rodgers. Yale: Yale University Press. Freeze, Ray 1992 Existentials and other locatives. Language 68: 552–595.

‘HAVE’ = ‘BE’ + PREP(osition)

131

Heim, Irene 1987 Where does the definiteness restriction apply? Evidence from the definiteness of variables. In The Representation of (In)definiteness, Eric J. Reuland and Alice G.B. ter Meulen (eds.), 21–42. Cambridge, MA: MIT Press. Jaggar, Philip J. 2001 Hausa. London Oriental and African Language Library 7. Amsterdam/ Philadelphia: John Benjamins. Kayne, Richard S. 1993 Towards a modular theory of auxiliary selection. Studia Linguistica 47: 3–31. 1994 The Antisymmetry of Syntax. Linguistic Inquiry Monograph 25. Cambridge, MA: MIT Press. Layton, Bentley 2000 A Coptic Grammar with Chrestomathy and Glossary. Porta Linguarum Orientalium N.S. 20. Wiesbaden: Harrassowitz Leslau, Wolf 1995 Reference Grammar of Amharic. Wiesbaden: Harrassowitz. McNally, Louisa 1998 Existential sentences without existential quantification. Linguistics and Philosophy 21: 353–392. Milsark, Gary L. 1977 Toward an explanation of certain peculiarities of the existential construction in English. Linguistic Analysis: 1–29. Moro, Andrea 1997 The Raising of Predicates: Predicative Noun Phrases and the Theory of Clause Structure. Cambridge Studies in Linguistics 80. Cambridge: Cambridge University Press. Newman, Paul 2000 The Hausa Language: An Encyclopedic Reference Grammar. New Haven /London: Yale University Press. Ouhalla, Jamal 2000 Possession in sentences and noun phrases. In Research in Afroasiatic Grammar. Papers from the Third Conference on Afroasiatic Languages (Sophia Antipolis, France, 1996), Jacqueline Lecarme, Jean Lowenstamm and Ur Shlonsky (eds.), 221–242. Current Issues in Linguistic Theory 202. Amsterdam/Philadelphia: John Benjamins. Reintges, Chris. H. 2001 Aspects of the morphosyntax of subjects and objects in Coptic Egyptian. 2001. In Linguistics in the Netherlands 2001, Hans Broekhuis and Ton van der Wouden (eds.), 177–188. Amsterdam/Philadelphia: John Benjamins. 2004 Coptic Egyptian (Sahidic Dialect) – A Learner’s Grammar. Cologne: Rüdiger Köppe Verlag.

132 Chris H. Reintges and Anikó Lipták Rizzi, Luigi 1997 The fine structure of the left periphery. In Elements of Grammar: Handbook in Generative Syntax, Liliane Haegeman (ed.), 281–337. Dordrecht: Kluwer. Svenonius, Peter 2004 On the edge. In Peripheries: Syntactic Edges and their Effects. David Adger, Cécile de Cat and George Tsoulas (eds.), 259–287. Studies in Natural Language and Linguistic Theory 59. Dordrecht: Kluwer. Szabolcsi, Anna 1983 The possessor that ran away from home. The Linguistic Review 3: 89–102. 1994 The noun phrase. In The Syntactic Structure of Hungarian, Ferenz Kiefer and Katalin É. Kiss (eds.), 179–274. San Diego: Academic Press. Yimam, Baye 1997 The verb to have in Amharic. In Ethiopia in Broader Perspective. Papers of the XIIIth International Conference of Ethiopian Studies. (Kyoto, 12–17 December), vol. 1, Katsuyoshi Fukui, Eisei Kurimoto and Masayoshi Shigeta (eds.), 619–636. Kyoto: Shokado Book Sellers.

Northern Norwegian degree questions and the syntax of measurement Peter Svenonius and Christopher Kennedy

Norwegian dialects, including Northern Norwegian (NN), make use of degree questions with no overt degree operator (Null Degree Questions, NDQs). These questions have a gradable adjective in situ and subject-verb inversion, for example Er du gammel?, literally ‘Are you old?’, has the interpretation “How old are you?” In this paper we provide a detailed syntactic and semantic analysis for NDQs in NN which provides new insight into the decompositional semantics of adjectives more generally. We show how our analysis fits in with current research into the syntax and semantics of measure phrases and comparatives.

1. Degree questions in Northern Norwegian 1.1. Various Scandinavian degree question operators Scandinavian languages generally form degree questions in the same way as English does, where a question word (English how) pied-pipes a degree adjective. (1)

a. How old are you? b. How far is it to Alaska?

Here we illustrate with Icelandic in (2) and with Norwegian in (3); we gloss Icelandic hversu as ‘how much’ because it is only used as a degree operator, lacking the manner adverbial use of English how, and we gloss Norwegian kor as ‘where’ because it has a locative meaning in the absence of an adjective (as in Kor er han? ‘Where is he?’). (2)

a. Hversu gammall ertu? how.much old are.you ‘How old are you’?

134 Peter Svenonius and Christopher Kennedy b. Hversu langt er til ingvalla? how.much far is to Thingvellir ‘How far is it to Thingvellir?’ (3)

(Icelandic)

a. Kor gammel er du? where old are you ‘How old are you?’ b. Kor langt er det til Nordkapp? where far is it to North.cape ‘How far it is to the North Cape?’

(Norwegian)

In addition to the above strategy, Icelandic can front a different question word, corresponding approximately to English what, and leave the adjective in situ. (4)

a. Hva ertu gammall? what are.you old ‘How old are you?’ b. Hva er langt til ingvalla? what is far to Thingvellir ‘How far is it to Thingvellir?’

(Icelandic)

Norwegian dialects in the northwest and north of Norway (stretching from Nordmøre and Trøndelag northwards; Endresen 1985) also have an alternative to (3), which like the Icelandic option has the adjective in situ. Unlike the Icelandic construction, there is no overt question word. Instead, as we will demonstrate in section 2, there is a phonologically null counterpart to Icelandic hva: (5)

a. Er du gammel? are you old ‘How old are you?’ b. Er det langt til Nordkapp? is it far to north.cape ‘How far is it to the North Cape?’

(Northern Norwegian)

The questions in (5) are string identical to yes-no questions in these dialects, but they are intonationally distinct (see the next subsection). They are un-

Northern Norwegian degree questions and the syntax of measurement

135

ambiguously interpreted as degree questions by speakers of these dialects, though speakers in the southeast and south of Norway do not recognize them as such and often find them confusing or mistake them for yes-no questions (as noted by Endresen 1985). In this section, we show that the degree interpretation is not a pragmatic reinterpretation of a yes-no question, and point out some restrictions on the construction. We draw our data from Northern Norway, in particular around Tromsø and Narvik; henceforth we will refer to these dialects as NN.1 There is some small variation in the way the construction is used from one dialect to the next (see Bull 1987 for some examples), but to the best of our knowledge our remarks hold for dialects as far south as Nordmøre, and for the Icelandic construction, apart from the fact that the Icelandic counterpart of the operator has phonological content. We will suggest, then, that degree questions can vary at least along two parameters: whether the degree question operator pied-pipes the adjective with which it is associated, and whether the degree question operator is pronounced. We have found in Scandinavian languages three of the four possible combinations of these two points of variation. It is an open question whether a language could also have a null degree operator which piedpiped an adjective, yielding degree questions roughly like “Old are you?”2

1.2. Syntax/Semantics versus Pragmatics Previous work on what we will call Null Degree Questions, or NDQs, in Norwegian dialects has concurred that they cannot be analyzed in terms of a pragmatic interpretation of a formal yes-no question (Endresen 1985; Bull 1987; Midtgård 1995), and we agree with this assessment. First, as noted above, the construction is systematically intonationally distinct from a yes-no question. Yes-no questions tend to have their prosodic peak on the most deeply embedded part of the sentence, which may be a predicative adjective, for example cold in Are you cold? NDQs never have an intonational peak on the adjective, but typically have the stress further to the left (Endresen 1985). Here we use underlining to represent intonation peak. (6)

a. Vil du ha mange? will you have many ‘Do you want many?’

136 Peter Svenonius and Christopher Kennedy b. Vil du ha mange? will you have many ‘How many would you like?’ (7)

a. Har du vært her lenge? have you been here long ‘Have you been here long?’ b. Har du vært her lenge? have you been here long ‘How long have you been here?’

As Endresen notes, the yes-no questions above (the first member of each pair) may pragmatically elicit more informative responses than a simple yes or no, depending on context. But NN speakers interpret the NDQs (the second member of each pair) as unambiguously asking for degrees. This leads to a sharp difference of intuition with southern speakers (and English speakers, for the translations). Unless southern speakers are claimed to have a different pragmatics, this suggests that the difference is grammatical, not pragmatic. A second argument that the construction is not simply pragmatic comes from the difference between the presuppositions of degree questions and yes-no questions involving adjectival predications. A yes-no question is based on the semantics of the ‘positive’ (morphologically unmarked) form, which involves reference to a contextual standard of comparison: Is he old? asks whether an object’s age is greater than the prevailing standard of age (for things like X) or not, and crucially presupposes that the answer to that question is not part of the common ground. In contrast, a degree question asks for the degree to which an object possesses some property: How old is X asks for an object’s age, and presupposes that this value is unknown, but it is completely indifferent as to whether X is old or not. The prediction is that in a context in which it is part of the common ground that X is old, only a degree question will be felicitous. This is illustrated by the English translations in the NN discourse in (8); the felicity of B’s question shows that it must have the semantics of a degree question. (8)

A: Læreren min er skikkelig gammel. the.teacher my is properly old ‘My teacher is really old.’

Northern Norwegian degree questions and the syntax of measurement

137

B: Er han gammel? is he old ‘How old is he?’ / ‘??Is he old?’ A third indication that the degree interpretation is not simply pragmatic is that various degree modifiers can be used in NDQs which are infelicitous in yes-no questions, for example sånn cirka, ‘approximately,’ as pointed out by Midtgård (1995). (9)

a. Er den tung, sånn cirka? is it heavy such approximately ‘How heavy is it, approximately?’ b. *Er den tung, sånn cirka? is it heavy such approximately (*‘Is it heavy, approximately?’)

We conclude that the NDQ is semantically a degree question, not a yes-no question.

1.3. The Salient Measure Restriction As noted by Midtgård (1995), NDQs are not possible with all gradable adjectives. (10) a. Er han ung? is he young ‘Is he young?’ (NOT: ‘How young is he?’) b. Er ho flink? is she talented ‘Is she talented?’ (NOT: ‘How talented is she?’) For such adjectives, NN uses the overt degree question operator kor (kor is the Nynorsk spelling, corresponding to hvor in Bokmål; the Northern pronunciation is /kur/). (11) a. Kor ung er han? where young is he ‘How young is he?’

138 Peter Svenonius and Christopher Kennedy b. Kor flink er ho? where talented is she ‘How talented is she?’ The general pattern is that NDQs are possible with adjectives for which there is a salient measurement system; measurement systems are typically only used with the ‘positive’ member of antonymic adjective pairs, e.g. old not young, long not short, deep not shallow. Other gradable adjectives simply lack a salient measurement system, e.g. talented, beautiful, tired, hungry, etc. are gradable in that one can be very talented or more beautiful and so on, but not associated with a salient measurement system. In the table in (12), only the adjectives in the leftmost column support NDQs.3 (12)

GRADABLE MEASURE POSITIVE

NEGATIVE

long deep far old expensive heavy

short shallow near young cheap light

NOT GRADABLE NO MEASURE

beautiful intelligent friendly tired happy anxious

atomic wooden binary quadrilateral human Norwegian

When a scale is referred to, it is usually the positive values of antonymic pairs which are used; thus, in degree questions where what is at issue is the place on a scale, the positive value is used. The questions in (13) can be posed neutrally to elicit information about the length of a fjord or the intelligence of moles, but the questions in (14) presuppose that Sognefjord is relatively short and that moles are relatively stupid. (13) a. How long is Sognefjord? b. How intelligent are moles? (14) a. How short is Sognefjord? b. How stupid are moles?

Northern Norwegian degree questions and the syntax of measurement

139

The restriction of NDQs to positive adjectives associated with salient measure scales is also true of Icelandic degree questions with hva ‘what’; otherwise hversu ‘how much’ must be used. (15) a. Hva ertu stór? what are.you big ‘How tall are you?’ b. Hversu stór ertu? how.much big are.you ‘How tall are you?’ (16) a. Hva er bókin dr? what is the.book expensive ‘How expensive is the book?’ b. Hversu dr er bókin? how.much expensive is the.book ‘How expensive is the book?’ (17) a. *Hva er hann dugarlegur? what is he clever b. Hversu dugarlegur er hann? how.much clever is he ‘How clever is he?’ (18) a. *Hva er hann hávær? what is he loud b. Hversu hávær er hann? how.much loud is he‘ How loud is he?’ There is a strong (though not perfect, see below) correlation between the possibility of overt measure phrases and the possibility of NDQs (or Icelandic degree questions with hva). For example, overt measure phrases are illustrated below with the adjectives lang ‘long,’ tung ‘heavy,’ høj ‘tall,’ gammel ‘old,’ and dyr ‘expensive.’

140 Peter Svenonius and Christopher Kennedy (19) a. en 100m lang fjord a 100.meter long fjord ‘a 100 meter long fjord’ b. en 100 kilo tung sekk a 100 kilo heavy bag ‘a 100 kilo bag’ c. en 160 cm høj mann a 160 cm tall man ‘a 160 cm tall man’ d. et 100 år gammelt hus a 100 year old house ‘a 100 year old house’ e. det 5 millioner kroner dyre huset the 5 millions crowns expensive house ‘the five million kroner house’ Note that an adjective can generally be used with a measure even if the thing being measured does not achieve the minimal standard implied by the adjective in the absence of a measure phrase: a hundred meter long fjord is not a long fjord, and a 160 centimeter tall man is not a tall man. No such measure phrases are possible with the negative adjectives kort ‘short,’ lett ‘light,’ ny ‘new,’ etc. (20) a. *en 100m kort fjord a 100.meter short fjord b. *en 100g lett sekk a 100.gram light bag c. *en 160 cm kort mann a 160 cm short man Nor are measure phrases possible with adjectives that are not associated with a measure scale. (21) a. *to søvnløse dager trøtt two sleepless days tired b. *syttifem bekymringer engstelig seventy-five worries anxious

Northern Norwegian degree questions and the syntax of measurement

141

These facts lead to the empirical generalization stated in (22). (22) The Salient Measure Restriction Only gradable predicates that are associated with a salient measurement system give rise to NDQs Midtgård (1995) presents the minimal pair in (23)–(24); the predicate vanskelig ‘difficult’ is not associated with a salient scale, and cannot support a NDQ, as indicated in (23a); the overt degree question operator must be used as seen in (23b). (23) a. Er det vanskelig å komme inn på den skolen? is it difficult to come in on that school ‘Is it difficult to get into that school?’ (NOT: ‘How difficult…’) b. Kor vanskelig er det å komme inn på den skolen? where difficult is it to come in on that school ‘How difficult is it to get into that school?’ If the question is formed over grade point averages, which provide a salient scale, as in (24a), then the NDQ reading is possible (again, with the characteristic intonation, here stress on ha), and equivalent to the question with the overt question operator in (24b). (24) a. Må man ha mye i snitt for å komme inn på den skolen? must one have much in average for to come in on that school ‘Does one need a high average to get into that school?’ OR ‘How high an average does one need to get into that school?’ b. Kor mye må man ha i snitt for å komme inn på den how much must one have in average for to come in on that skolen? school ‘How high an average does one need to get into that school?’ The pattern suggests that the restriction to a salient scale has a syntactic dimension: knowledge that grade-point averages are what is at stake does not make (23a) felicitous as a NDQ; grade-point averages are not a direct measure of difficulty. They do, however, directly quantify ‘how high an average’ (literally “where much in average”), which licenses the NDQ

142 Peter Svenonius and Christopher Kennedy reading of (24a). A word like mange ‘many’ always licenses a NDQ, and mye (roughly, ‘much’) does so whenever the substance referred to is associated with a salient countable unit of measure. (25) a. Vil du ha mange poteter? will you have many potatoes ‘How many potatoes do you want?’ b. Skal du låne mye penger? shall you borrow much money ‘How much money do you intend to borrow?’ The salience of the measurement system associated with a particular adjective can be affected by context and world knowledge, however. For example, when speakers are presented with the context of a game of Limbo, in which the objective is to walk under a bar held at successively lower heights, they accept (26a) as a degree question (with the right intonation, i.e. stress on gikk), and (26b) with a measure phrase. (26) a. Gikk du lavt? went you low ‘How low did you go?’ b. Eg gikk førti cm lavt. I went forty cm low ‘I went forty centimeters low’ Similarly, when speakers are presented with a context in which clipboardwielding intelligence researchers are discussing an experimental subject, they are willing to accept (27a) as a degree question, and (27b) with a measure phrase. (27) a. Var han intelligent? was he intelligent ‘How intelligent was he?’ (i.e. how many intelligence points) b. Ho var ti poeng mer intelligent enn han. she was ten points more intelligent than he ‘She was ten points more intelligent than he was’ Such examples are regarded as creative and/or marginal. As a final remark in this subsection, we note that the match between overt measure phrases

Northern Norwegian degree questions and the syntax of measurement

143

and NDQs is not perfect. For example, NDQs are perfectly good when asking the cost of something, but overt measure phrases are not consistently good with dyr ‘expensive’ (cf. English a forty-dollar (*expensive) watch). We believe that such mismatches have to do with idiosyncracies in the system of overt measure phrases, and concentrate in this paper on detailing the NDQ, though we return briefly to the matter below.

1.4. Comparatives Comparatives of adjectives permit measure phrases, if the scale is saliently associated with a measurement, regardless of whether the adjective is the positive or negative member of the antonymic pair. For example, though dyr ‘expensive’ but not billig ‘cheap’ allows a measure phrase, both the comparative dyrere and the comparative billigere do. (28) a. Tunfisk er 100 kroner dyrere / billigere enn den var før. tuna is 100 crowns expensiver / cheaper than it was before ‘Tuna is 100 kroner more expensive/cheaper than it used to be’ b. Han Julian er 8 måneder eldre / yngre enn han Sam. he Julian is 8 months older / younger than he Sam ‘Julian is 8 months older/younger than Sam’ Similarly, comparatives license NDQs, in NN (the following strings also have yes-no question interpretations, with a different intonation, as usual). (29) a. Er den mye dyrere / billigere enn den var før? is it much expensiver / cheaper than it was before ‘How much more expensive/cheaper is it than it used to be?’ b. Er han mye eldre / yngre enn deg? is he much older / younger than you? ‘How much older/younger is he than you?’ Neither NDQs nor measure phrases are possible with comparatives based on adjectives not associated with a salient measurement system. (30) a. Eg er (*fem timer) trøttere enn deg. I am five hours tireder than you ‘I am (*five hours) more tired than you are’

144 Peter Svenonius and Christopher Kennedy b. Er han trøttere enn deg? is he tireder than you ‘Is he more tired than you?’ (NOT: ‘How much more tired is he than you?’) With these basic empirical facts in place, we can detail our analysis.

2. A null degree operator The existence in Icelandic of parallel constructions with an overt operator hva lends immediate plausibility to the postulation of a null operator in NN NDQs. For the basic syntax, we assume something like (31a) for NN; cf. (31b) for Icelandic, identical except for the overtness of the operator. (31) a. [CP Op1 er2 [IP du3 t2 [VP t2 [AP t3 t1 gammel ]]]] Op are you old ‘How old are you?’ b. [CP Hva1 er2 [IP -tu3 t2 [VP t2 [AP t3 t1 gammall ]]]] what are you old ‘How old are you?’ Evidence for a null operator comes from the fact that NDQs exhibit the same locality conditions that hold of overt wh-movement. For example, the gradable predicate that supports the NDQ can be embedded inside a complement clause, but not a subject, as pointed out by Endresen (1985). (32a) and (33a) (based on Endresen’s examples) form a minimal pair: (32a) is impersonal, with the subject not occupying the canonical subject position, and is acceptable as an NDQ. (33a) has the subject in subject position, and is impossible as an NDQ (though the same string is grammatical as a yes-no question). (32) a. Tror du det kommer mange? think you it come many ‘How many do you think will come?’ b. Mente du pinnen skulle være lang? thought you the.stick should be long ‘How long did you think the stick should be?’

Northern Norwegian degree questions and the syntax of measurement

145

(33) a. Tror du mange kommer? think you many come ‘Do you think many will come?’ (NOT: ‘How many do you think will come?’) b. Kor mange tror du kommer? where many think you come ‘How many do you think will come?’ Subjects can be questioned, as indicated in (33b), but it is difficult to subextract from them. Hence the pattern in (32)–(33) is explicable given that NDQs involve the movement of a null operator. This also correctly predicts that NDQs should not escape adjunct islands, as illustrated in (34). (34) a. Er det viktig at vi har fest fordi han er gammel? is it important that we have party because he is old ‘Is it important that we throw a party because he is old?’ (NOT: ‘How old is it that his being that old makes it important that we throw a party?’) b. *Kor gammel er det viktig at vi har fest fordi han er? where old is it important that we have party because he is The null operator analysis is also compatible with the fact that bare noun phrases, but not noun phrases with overt possessors or determiners, are transparent to wh-movement.4 (35) a. Treng du langt tau? need you long rope ‘How long a rope do you need?’ b. Kor langt tau treng du? where long rope need you ‘How long a rope do you need?’ (36) a. Treng du et langt tau? need you a long rope ‘Do you need a long rope?’ (NOT: ‘How long a rope do you need?’) b. *Kor langt et tau treng du? where long a rope need you

146 Peter Svenonius and Christopher Kennedy As already noted, the NN NDQ operator has a syntax distinct from the overt degree morpheme kor, as evident from the fact that kor never strands its associated predicate (for completeness we include an example without inversion, since NN has non-inverted wh-questions, cf. e.g. Vangsnes to appear). (37) a. Kor gammel er du? where old are you ‘How old are you?’ b. *Kor er du gammel? where are you old (Only as: # ‘Where are you old?’) c. *Kor du er gammel? where you are old (Only as: # ‘Where are you old?’) Following Corver (1990), we can explain facts like (37b–c) by assuming that kor (like English how) is a head which combines directly with a gradable adjective (a Deg0 head, presupposing the syntactic analysis to be developed in the next section), and so does not move away from it. The NDQ, like the Icelandic questions with hva, must therefore contain a phrasal operator in a specifier position, which we will claim occupies the same position as overt measure phrases occupy. In section 3, we develop a detailed syntactic analysis along these lines, which explains the Salient Measure Restriction (stated in (22) above) in terms of the syntax and semantics of the extended adjectival projection. Before turning to this analysis, however, we must first consider a potentially more direct explanation for this restriction based on the semantics of the null degree operator. The intuition underlying this approach (as well as the one we develop below) is that whatever principles prevent measure phrases from combining with negative adjectives and adjectives without measurement systems should also prevent the null operator from combining with such adjectives. The fact that measure phrases are based on nouns that denote units (see Lehrer 1986) suggests that they presuppose that the degrees they denote (or quantify over) can be mapped onto a finite set of countable units. We might therefore posit a semantics for null degree questions along the lines of (38). (38) [Op1 V DP t1 A] How many units is the maximal degree d such that DP V d-A?

Northern Norwegian degree questions and the syntax of measurement

147

Only adjectives that support measure phrases could support this operator, since the meaning of the operator presupposes the existence of discrete units of measurement. An adjective like young is infelicitous because negative degrees are not countable (Seuren 1978; von Stechow 1984b; Kennedy 2001), as we discuss in more detail below, and an adjective like intelligent is infelicitous because there is no measurement system associated with the intelligence scale (though in specific contexts, such as the IQ test discussed above, one can be imposed, rendering measure phrases and NDQs acceptable). However, we believe this simple and straightforward account to be incorrect. For one thing, it conflicts with an independently motivated semantics for adjectival constructions which we will elaborate in the next section. For another, it makes incorrect empirical predictions regarding the range of acceptable answers to NDQs. The predictions can be seen most easily by examining comparatives. The analysis sketched in (38) assigns the interpretation given in (40a) to (39), which is distinct from that in (40b). (39) Er han mye eldre enn deg? is he much older than you ‘How much older is he than you?’ (40) a. How many units is the maximal degree d such that his age ex ceeds your age by at least d? b. What is the maximal degree d such that his age exceeds your age by at least d? The difference gives rise to a subtle difference in prediction regarding the range of felicitous answers to questions like (39). If (40a) is the correct representation, then (41a–b), but not (41c), should be acceptable answers ((41c) should give rise to the reaction, Why are you avoiding my question?). On the other hand, if (40b) is the correct representation, then all of (41a–c) should be felicitous answers. (41) a. Eight months older. b. Many months older./A few months older/Not as many months older as I thought. c. Much older./A little bit older./Not as much older as I thought. The difference is not simply one of informativity: much older is nearly as informative as many months older. The difference is that (41a–b) are framed

148 Peter Svenonius and Christopher Kennedy explicitly in terms of the information that is requested in (40a): units of agemeasurement. Turning back to the NN question in (39), we find that all of the following are regarded as felicitous answers (though not a simple ‘yes’ or ‘no’). (42) a. Åtte måneder eldre. eight months older ‘Eight months older’ b. Mange måneder eldre. / Noen få måneder eldre. / Ikke så mange many months older / some few months older / not so many måneder eldre. months older ‘Many months older/A few months older/Not many months older’ c. Mye eldre./ Litt eldre./ Ikke så mye eldre som eg trodde. much older / little older / not so much older as I thought ‘Much older/A little bit older/Not as much older as I thought’ We conclude that the NN NDQ operator cannot be an implicit version of how many units, as in (40a), but must have a more general meaning as an underspecified quantifier over degrees. (43) [Op1 V DP t1 A] What is the maximal degree d such that DP V d-A? If this is correct, we can maintain our account of the impossibility of NDQs with negative adjectives, given that this operator only quantifies over positive degrees, but it is not obvious from (43) how it is to be restricted to adjectives with salient measurement systems. In fact, the general operator postulated in (43) manifests exactly the kind of quantification argued to be involved in the interpretation of comparative clauses like that in (44) (see von Stechow 1984a; Heim 1985; 2000; Rullmann 1995), clearly not restricted by the Salient Measure Restriction in (22). (44) a. The pasta is harder in Rome than it is in Tromsø. b. … than [CP Op1 the pasta is [AP t1 d-hard in Tromsø ]] c. … the maximal d such that pasta is d-hard in Tromsø We will argue below that the Salient Measure Restriction should be derived from the semantics of the predicate, rather than the operator.

Northern Norwegian degree questions and the syntax of measurement

149

3. The syntax of measurement The analysis of gradable predicates which we wish to adopt here is one in which degree morphology does not fill an argument position, but rather heads the extended functional projection of the lexical head (Abney 1987; Corver 1990; Grimshaw 1991; Kennedy 1999). (45)

DegP V Deg AP

Kennedy (1999) argues in favor of a decompositional semantic analysis of gradable adjectives, in which adjectives do not have degree arguments, but rather denote measure functions—functions from individuals to degrees (type
e,d). For example, the adjective old denotes the function in (46): it takes an object and returns its (positive) degree of age. (46) [[old]] = x.the degree to which x is old A consequence of this analysis is that gradable adjectives must combine with some other expression in order to be converted into properties of individuals; this is the function of degree morphology. For example, the unmarked ‘positive’ form of an adjectival predicate is derived by combining an adjective with a null Deg head pos (possibly overt in some languages; see Sybesma 1999 on Mandarin hen) which has the (simplified) semantics in (47).5 (47) [[ [Deg pos] ]]c = gx.g(x)f ds(g)(c) Here ds(g)(c) represents the ‘standard of comparison’ for a context of utterance c: the degree that is required to count as having the property measured by g in c. Combination of pos with old results in the property in (48), which is true of an object if it has a degree of age that exceeds a contextually determined standard of age. (48) x.the degree to which x is old f
ds(old)(c) Kennedy (1999) shows how this approach extends to comparatives and other complex degree constructions; here we focus on the analysis of measure phrases. If measure phrases denote (or quantify over) degrees, but

150 Peter Svenonius and Christopher Kennedy gradable adjectives do not themselves have degree arguments, then some other element of the structure must provide this position for the measure phrases in expressions like two meters tall, eight months old, five fathoms deep, and so on. We propose that the measure phrase is introduced by a Deg head that we will refer to as Meas, so that an example like eight months old has the structure in (49). (49)

DegP wo NumP Deg 6 V eight months Deg AP # 5 Meas old

We further suggest that Meas is constrained to combine only with adjectives that denote functions that map their arguments onto measurable degrees. This is stated as a domain (selectional) restriction on the adjectival argument of Meas in (50).6 (50) [[[Deg Meas] ]] = g:g is a function from objects to measurable degrees d x.g(x) f d


Whether a degree is measurable depends on two factors. The first is whether the scale that it comes from is associated with a measurement system in the first place: AGE is; FATIGUE is not, thus Meas is compatible with the adjective old (which denotes a function from objects to degrees of age) but not with the adjective tired (which denotes a function from objects to degrees of fatigue). Some adjectives may also permit contextual accommodation of a measurement system (as we saw in section 1.3 for intelligent in the context of an IQ test); the semantics of Meas requires that they must take on this kind of meaning whenever they project a degree argument. The second factor is a purely structural one: whether the degree is bounded or unbounded. This distinction is relevant to the characterization of adjectival polarity. According to Seuren (1978), von Stechow (1984b) and Kennedy (2001), both members of an antonymous pair like old and young measure objects according to the same general scale (AGE in this case), but differ in that old maps its arguments onto bounded, measurable ‘positive’ degrees (degrees that originate at the zero point of a scale), while

Northern Norwegian degree questions and the syntax of measurement

151

young maps its arguments onto unbounded, unmeasurable ‘negative’ degrees (degrees that range from some value to the upper reaches of the scale). The end result is that the set of adjectives that can combine with Meas to project a degree argument is a fairly restricted one: adjectives that map their arguments onto bounded intervals of scales associated with a measurement system. This set includes the positive dimensional adjectives in the leftmost column of the table in (12), and, as we will show below, comparative forms of both positive and negative dimensional adjectives. This is the set of adjectives that permit overt measure phrases in Northern Norwegian, and as we saw in section 1.3, the set of adjectives that permit NDQs. Below we demonstrate how our analysis derives this correlation. We begin by assuming a very general semantics for questions formed using the null NN degree quantifier (and its overt counterpart in Icelandic), along the lines of (51). (51) [Op1 V DP [DegP t1 Meas A]] What is the maximal degree d such that DP V d-A? The operator in (51) does not quantify over units of measurement, as was the case with (38), our first attempt at deriving the Salient Measure Restriction, and so does not run into the problem of incorrect expectations about answers that we documented above. Instead, the operator in (51) simply asks for a (maximal) degree, and so may in principle target any degree variable. As observed above, this analysis of the null degree quantifier runs into problems if all gradable adjectives introduce a degree argument, as is standardly assumed (i.e., under standard assumptions the semantic type of a gradable adjective is
d,
e,t), since it would be impossible to ensure that the operator in (51) combines only with adjectives that are associated with measurement systems. Under the analysis proposed here, however, the degree argument is introduced by Meas rather than by a gradable adjective (which is type
e,d). Since the null degree operator originates in this position (it quantifies over the degree variable introduced by Meas, the same variable that is restricted by an overt measure phrase), it follows that only those adjectives that can independently combine with Meas will form null degree questions. For example, in the case of (52a–c), only (52a) is well-formed, since only gammel ‘old’ has a meaning that permits combination with Meas and projection of a position for the operator.7

152 Peter Svenonius and Christopher Kennedy (52) a. Op1 er du [DegP t1 Meas [AP gammel]]? Op are you old ‘How old are you?’ b. *Op1 er du [DegP t1 Meas [AP ung]]? Op are you young ‘How young are you?’ c. *Op1 er du [DegP t1 Meas [AP trøtt]]? Op are you tired ‘How tired are you?’ Neither ung ‘young’ nor trøtt ‘tired’ map their arguments onto measurable degrees, so combination with Meas as in (52b–c) is precluded, rendering a degree question parse impossible. The Salient Measure Restriction on Northern Norwegian degree questions thus follows from the selectional restrictions on Meas. Turning now to comparatives, we saw earlier that comparative forms of both positive and negative adjectives permit measure phrases and give rise to NDQs even when the non-comparative forms (of the negative adjectives) do not. (53) a. Han Julian er åtte måneder yngre enn han Sam. he Julian is eight months younger than he Sam ‘Julian is eight months younger than Sam’ b. Er han mye yngre enn deg? is he much younger than you ‘How much younger is he than you?’ One potential (but, we will argue, wrong) explanation of this fact would be that the comparative degree morphology is like Meas in that it introduces a degree argument, namely the ‘differential degree’ that measures the distance between the compared objects (Hellan 1981; von Stechow 1984a; Kennedy 2001; Schwarzschild and Wilkinson 2002; Schwarzschild to appear): (54) a. [[-er/more]]= g d1 d2 x.g(x)d1 =d2 b. [[ [DegP two meters longer than Y]]]= x.the degree to which x is long minus the degree to which Y is long = two meters But if our claims about the semantics of Meas and the NN NDQ operator are correct, then (54a) can’t be the right analysis of comparatives; it would

Northern Norwegian degree questions and the syntax of measurement

153

wrongly predict that the null degree operator could combine with any comparative, regardless of the semantic properties of the adjectival base. That this is incorrect is demonstrated in (55). (55) a. Er han mye trøttere enn deg? is he much tireder than you ‘Is he much more tired than you?’/*‘How much more tired is he than you?’ b. Eg er (*fem timer) trøttere enn deg. I am five hours tireder than you ‘I am (*five hours) more tired than you’ (55) shows that adjectives that are not associated with measurement systems do not have NDQ readings in comparatives; but if we adopt the analysis of comparatives described above, we run into the same problem we had with the standard analysis of gradable adjectives: we provide a predicate (the comparative form) with a degree argument when it shouldn’t have one, i.e., even when the adjectival base does not use a scale that supports measurement. A second problem for this analysis comes from the fact that mye ‘much’ is crucial to deriving the DQ interpetation of a comparative: without it, comparatives have only yes-no interpretations, regardless of polarity: (56) Er han (mye) eldre / yngre enn deg? is he (much) older / younger than you ‘Is he (much) older/younger than you?’ ‘How much older/younger than you is he?’ (requires mye) If comparative morphology introduced a degree argument, there would be no syntactic or semantic distinction between it and Meas (in the relevant respects), and no obvious reason to require the presence of mye ‘much’ to get the DQ interpretation. We can account for this data if we adopt the analysis of comparatives proposed by Kennedy and McNally (2005a), in which comparatives are not true degree morphemes, but rather expressions that map adjective meanings to new adjective meanings (i.e., adjectival modifiers; cf. Neeleman et al. 2004). Kennedy and McNally propose that comparative morphology (the degree head plus the comparative clause) takes a gradable adjective meaning (a function from individuals to degrees) and returns a new one that is

154 Peter Svenonius and Christopher Kennedy just like the old one except that it uses a scale whose zero point is the degree denoted by the than-clause. (57) a. [[old]]: a function from individuals to: AGE: 0 ---------------------------->
b. [[older than Sam]]: a function from individuals to the squarebracketed part of: AGE: 0 - - - - Sam’s age - - - - [ --------------->
] ° Since the comparative+adjective constituent denotes a function from individuals to degrees, it must combine with a Deg head to derive a property of individuals. A very similar proposal has been made on independent syntactic grounds by Corver (1997a,b), who argues that comparative morphology fills a functional head between AP and DegP: (58)

DegP wo Deg QP # 3 pos Q
PP V 5 Q AP than Sam # # er old

In Kennedy and McNally’s analysis, simple comparatives involve combination with pos, and are assigned a ‘minimum standard’ interpretation: the structure in (58) denotes a property that is true of an object if it has a nonzero degree of ‘older-than-Sam-ness’; i.e., if its degree of age exceeds Sam’s (maximal) degree of age. Here we propose that differential comparatives involve combination with Meas, as in (59). What is crucial for our proposal is that the semantics of comparatives will entail that both positive and negative comparatives map their arguments onto bounded and measurable degrees, since both involve measurement from a derived ‘zero point’: the degree denoted by the comparative clause. This is illustrated for the case of older/younger than Sam in (60)– (61).

Northern Norwegian degree questions and the syntax of measurement

(59)

155

DegP wo 8 months Deg wo Deg QP # ei Meas Q
PP V 5 Q AP than Sam # # er old

(60) a. [DegP 8 months older than Sam] b. AGE: 0 - - - - Sam’s agepos - - - - [° -- 8 mos -- ----->
]

•

(61) a. [DegP 8 months younger than Sam] b. AGE: [0 <-- - 8 mos - °] - - - - Sam’s ageneg - - - - --->


•

It follows that both older than Sam and younger than Sam can combine with Meas and introduce a degree argument for the NN null degree operator to bind. In contrast, comparatives formed out of adjectives like tired still make use of scales without measurement systems. Combination with Meas is therefore impossible, and no degree argument is projected.8 Finally, this analysis also provides a basis for explaining the fact that mye ‘much’ is required to obtain a DQ interpretation in NN, since it posits a syntactic distinction between comparatives and lexical adjectives: as illustrated by (62a–b), the former contain a layer of functional structure between meas and A; the latter do not. (62) a.

DegP wo Deg QP # 3 Meas Q



AP # # COMP A

b.

DegP 3 Deg AP # # Meas A

The exact nature of the constraint that forces the appearance of mye remains a question for future work. For now, we hypothesize that the Meas

156 Peter Svenonius and Christopher Kennedy head, like the overt word kor, its English counterpart how, and modifiers like very, combines with AP, not QP (cf. *very taller, *how taller, etc.). The word mye ‘much’ is an A (or a category-changing morpheme; cf. very much, how much) inserted to ensure that the syntactic selectional requirements of Meas are satisfied: (63)

DegP ei Deg AP # 3 Meas mye
QP 2 Q AP # # COMP A

4. Conclusions This paper has proposed that Null Degree Questions in Northern Norwegian are derived by moving a phonologically null operator from a degree argument position inside a gradable predicate – the same position occupied by a measure phrase – to SpecCP. We have claimed that the restriction of such questions to predicates that are associated with salient measurement scales (the Salient Measure Restriction) follows from a more general hypothesis about the projection of degree arguments of gradable predicates. Such arguments are not specified in the lexical entry of the predicate, as is standardly assumed, but are rather introduced by functional morphology in the extended projection of the adjective (our Meas head). Whether a degree argument can be projected or not depends on the semantic properties of the lexical (adjectival) head: whether the scale it is assocated with comes with a measurement system (or whether one can be contextually accommodated), and whether the degrees onto which it maps its argument are measurable (the positive vs. negative). A consequence of this analysis is that gradable adjectives must be analyzed as measure functions (type
e,d ), rather than as relations between individuals and degrees (type

d,
e,t), and that the lexical adjective must project extended functional structure.

Northern Norwegian degree questions and the syntax of measurement

157

Acknowledgements Kristine Bentzen contributed importantly to the collection of the data used throughout this paper; thanks to her and to the other consultants, including Marit Richardsen Westergaard and Merete Anderssen. We are also grateful to the audience at IGG XXXI, and to Marit Richardsen Westergaard and an anonymous reviewer for insightful comments on earlier versions of this work. This material is based upon work supported by the National Science Foundation under Grant No. 0094263 and 0618917 and by the American Council of Learned Societies through a Charles A. Ryskamp Fellowship to Chris Kennedy.

Notes 1. Except where noted, all examples in this paper are NN, in a standardized orthography, lightly modified to reflect salient dialectal features. In some cases we have based examples on Endresen (1985) or Midtgård (1995) but standardized the orthography. 2. In fact, such a structure is attested in Spanish in a related construction: the comparative clause of ‘comparative subdeletion’ structures (see Rivero 1981; Kennedy 2002): (i) a. La mesa es más large que ancha es la puerta. the table is more long than wide is the door ‘The table is longer than the door is wide.’ b. *La mesa es más large que la puerta es ancha. the table is more long than the door is wide It is well established that a comparative clause is an Ā-movement structure, so the existence of comparatives like (ia) suggests that degree questions with the same form should indeed be possible. 3. The arrangement of the table does not reflect the fact that gradable adjectives not associated with salient measurement scales can often be arranged into antonymic pairs, e.g. beautiful– ugly, intelligent– stupid, etc. 4. To illustrate: (i) a. Korsen bil kjenner du en passasjer i? which car know you a passenger in ‘Which car do you know a passenger in?’ b. ??Korsen bil kjenner du passasjeren i? which car know you the.passenger in

158 Peter Svenonius and Christopher Kennedy

5.

6.

7.

8.

Examples like (35a) seem to be incompatible with Ross’ Left Branch Condition (LBC); note that in (35b) the adjective is pied-piped, rather than stranded, in keeping with the LBC. We leave this matter unresolved here. Most degree morphology will have a semantic type similar to that of pos, and will occupy the Deg0 position. The intensifier very, for example, can be analyzed as a function from adjectives to properties of individuals along the lines of (i), where the context is modified in such a way that just the individuals that ‘count as’ having the property in question are considered when computing the standard of comparison (Wheeler 1972; Klein 1980; Kennedy and McNally 2005a). (i) [[ [Deg very] ]]c = g x.g(x) f d s(g)(c[{x |pos(g)(x)(c) = 1}]) On this view, to be very tall, for example, is to be tall relative to the tall objects in the context. See Kennedy and McNally (2005b) for discussion of the semantic and syntactic differences between different types of degree modifiers. Note that the semantics of Meas is distinct from that of pos, and as Deg heads, the two morphemes are in complementary distribution. This explains the fact that a predication involving a measure phrase generates no entailments to the positive form: a 10 month old baby need not be an old baby. A relevant question is how the NDQ operator escapes the DegP, given that measure expressions are not ordinarily extractable: (i) *Tretti år1 er han [DegP t1 Meas [AP gammel ]] thirty years is he old We will assume that this is because measure phrases do not have independent reference, and so are invisible to topicalization and scrambling operations. The NDQ operator, in contrast, has wh-features and so is attracted by the interrogative C. A reviewer asks whether the standard syntactic analysis of comparative clauses as Ā-movement constructions is in conflict with our claims that gradable adjectives are type
e,d
and that a degree argument position is projected only through the mediation of Meas. As the reviewer correctly points out, if the comparative clause operator has the same syntax as the NN null degree operator – if it binds a degree variable position in the comparative clause, projected by Meas – then our analysis would seem to predict that adjectives like tired would fail to form comparative clauses. This is obviously wrong, assuming the comparative clause in (i) contains an occurrence of the adjective. (i) I am tireder these days than I have ever been before. In fact, our assumptions are consistent, if we adopt the syntactic analysis of the comparative clause proposed in Kennedy (2002), in which the comparative clause operator is not a phrasal constituent parallel to the NN null degree operator, but rather a Degree head parallel to Meas or how. In this analysis, the entire DegP moves (as in the Spanish subdeletion constructions discussed in note 2), and then undergoes deletion under identity with the comparative adjective (cf. the ‘matching’ analysis of relative clauses):

Northern Norwegian degree questions and the syntax of measurement

159

(ii) I am tireder these days than [CP [DegP Comp tired] I have ever been before t] Unlike Meas, Comp is not restricted to combine only with a subset of gradable adjectives; instead, it is like all comparative degree morphology in merely requiring its adjectival complement to be gradable. See Kennedy (2002) for a semantic analysis of Comp that ensures that structures like (ii) are assigned the correct interpretation.

References Abney, Steven 1987 The English noun phrase in its sentential aspect. Ph.D. diss., MIT. Bull, Tove 1987 Eit nordlig syntaktisk drag i norsk. Maal og Minne 1–2 /3–4: 137– 141. Corver, Norbert 1990 The syntax of left branch extractions. Ph.D. diss., Tilburg University. 1997a The internal syntax of the Dutch adjectival projection. Natural Language and Linguistic Theory 15: 289–368. 1997b Much-support as a last resort. Linguistic Inquiry 28: 119–164. Endresen, Øyalf 1985 E a mang, klokka? Maal og Minne 3–4: 249–262. Grimshaw, Jane 1991 Extended projection. Unpublished ms., Brandeis University. Heim, Irene 1985 Notes on comparatives and related matters. Ms., University of Texas, Austin, TX. 2000 Degree operators and scope. In Semantics and Linguistic Theory 10, Brendan Jackson and Tanya Matthews (ed.), 40–64. Ithaca, NY: CLC Publications. Hellan, Lars 1981 Towards an Integrated Analysis of Comparatives. Tübingen: Narr. Kennedy, Christopher 1999 Projecting the Adjective: The Syntax and Semantics of Gradability and Comparison. New York: Garland. (1997 UCSC Ph.D thesis). 2001 Polar opposition and the ontology of ‘degrees’. Linguistics and Philosophy 24: 33–70. 2002 Comparative deletion and optimality in syntax. Natural Language and Linguistic Theory 20 (3): 553–621. Kennedy, Christopher and Louise McNally 2005 a Scale structure and the semantic typology of gradable predicates. Language 81: 345–381.

160 Peter Svenonius and Christopher Kennedy 2005 b

The syntax and semantics of multiple degree modification in English. In The Proceedings of the 12th International Conference on HeadDriven Phrase Structure Grammar, Department of Informatics, University of Lisbon, Stefan Müller (ed.), 178–191. Stanford: CSLI Publications.

Klein, Ewan 1980 A semantics for positive and comparative adjectives. Linguistics and Philosophy 4: 1–45. Lehrer, Adrienne 1986 English classifier constructions. Lingua 68: 109–148. Midtgård, Tone 1995 Yes/no interrogatives as wh-questions: The syntax and intonation of a Norwegian grammatical construction. In Papers from the XVth Scandinavian Conference of Linguistics, Inger Moen, Hanne Gram Simonsen and Helge Lødrup (eds.), 374–380. Oslo: Department of Linguistics, University of Oslo. Neeleman, Ad, Hans van de Koot and Jenny Doetjes 2004 Degree expressions. The Linguistic Review 21:1–66. Rivero, María-Luisa 1981 Wh-movement in comparatives in Spanish. In Linguistic Symposium on Romance Languages 9, William Cressey and Donna Jo Napoli (eds.), 177–196. Washington, DC: Georgetown University Press. Rullmann, Hotze 1995 Maximality in the semantics of wh-constructions. Ph.D. diss., University of Amherst, MA. Schwarzschild, Roger to appear Measure phrases as modifiers of adjectives. Recherches Linguistiques de Vincennes. Schwarzschild, Roger and Karina Wilkinson 2002 Quantifiers in comparatives: A semantics of degree based on intervals. Natural Language Semantics 10: 1–41. Seuren, Pieter A. M 1978 The structure and selection of positive and negative gradable adjectives. In Papers from the Parasession on the Lexicon, Chicago Linguistics Society, Donka Farkas, Wesley M. Jacobsen and Karol W. Todrys (eds.), 336–346. Chicago: Chicago Linguistic Society. von Stechow, Arnim 1984 a Comparing semantic theories of comparison. Journal of Semantics 3: 1–77. 1984 b My reply to Cresswell’s, Hellan’s, Hoeksema’s and Seuren’s comments. Journal of Semantics 3: 183–199. Sybesma, Rynt 1999 The Mandarin VP. Boston: Kluwer.

Northern Norwegian degree questions and the syntax of measurement

161

Vangsnes, Øystein Alexander to appear Microparameters for Norwegian wh-grammars. Linguistic Variation Yearbook 5. Wheeler, Samuel 1972 Attributives and their modifiers. Noûs 6: 310–334.

Parallels in clausal and nominal periphery Giuliana Giusti

1.

Introduction

This paper argues for a nominal periphery in terms of a split DP-system, parallel to the clausal split CP of Rizzi (1997). This issue arises naturally from the interaction of two main lines of research in generative grammar:1 the possibility to displace constituents in the clause and the parallels between noun phrase and clause structure.2 In this section, I give a very brief overview of previous literature. In 2, I state the unmarked structure of noun phrases in UG and emphasize the parallelism in the tripartite structure of nominal and clausal structure in a lexical NP-layer, an inflectional AgrP-layer and a complementation DP-layer. While the former two layers have been shown to be “split” in multiple projections, the latter has not yet been studied in detail in this respect. In 3, I present evidence for a split DP from Italian, Albanian and Serbo-Croatian. In 4, I outline some theoretical details of my proposal.

1.1. Parallelisms between noun phrases and clauses The parallelism between clausal and nominal structure is apparently imperfect but could be reduced to perfection in the sense of Chomsky (2005) if lack of sentential properties such as the presence of wh- and Tense features, and clitics is reduced to one property that languages do not realize in noun phrases.3 Luigi Rizzi (p.c.) points out that wh- and Tense features are strictly related and lack of one can directly cause lack of the other. This is also the case for the landing site of clitics, notably related to Tense morphology in the clause. I propose that the crucial difference between clauses and noun phrases is the lack of Tense in the latter. This proposal is in line with Alexiadou (2001) who claims, following Pustejowski (1995), that an event expressed by a noun phrase is not asserted, as the assertion properties of a clause are related to the higher layer, in particular to Force. In the CP

164 Giuliana Giusti system, Force is strictly related to Fin, which is in turn related to Tense (belonging to the IP system). 4 Maintaining the hypothesis of a perfect parallelism, I propose that mutatis mutandis, the DP is structured in three layers, with a major difference with reference to the clause which is directly related to the lexico-semantic properties of the lexical head N, and reflected in its inflectional properties (and EPP-features): V is inflected for Tense, which has a reflex on Fin; while N is inflected for gender and number. More precisely, parallel to Fin which is related to Tense, a low functional head in the DP-system (which I call [small] d) is related to Number in the inflectional layer of the noun phrase.

1.2. The nominal periphery Previous literature has already observed that nominal structures have some kind of left peripheral elements. For example, (genitival) arguments may be fronted and resumed by weak possessives (cf. Gavrouseva 2000 for a crosslinguistic overview). There are clear examples of derived positions for possessors either moved to the peripheral position (cf. Giusti 1996), as in Greek, which does not allow a resumptive element (cf. Horrocks and Stavrou 1987; Giusti and Stavrou in progress), or directly merged as topical elements, as in Bulgarian (cf. Dimitrova-Vulchanova and Giusti 1999) and Germanic (cf. Haegeman 2004), with crucial differences between the two languages as to the resumptive strategy (possessive clitic in the former, possessive adjective in the latter). Apart from the seminal work of Horrocks and Stavrou (1987) – too early to be set in a split-DP framework – the other studies assume a number of projections in the nominal complementation area, which host moved possessors and modifiers, but the discourse function of these movements are somehow only hinted at. A split DP projection has also been proposed by Bernstein (2001) for English and Romance, Ihsane and Puskás (2001) for French and Hungarian, Aboh (2004) for Gungbe, and recently Svenonius (2004) on the notion of “edge”. But these studies diverge sensibly from what I am going to investigate here, in that they correlate peripheral positions inside the noun phrase with a Topic/Focus interpretation that the noun phrase is assigned in the clause. A consequence of their claim is the assumption that a split DP is always projected, at least in the languages they investigate, and that checking of some features in D including definiteness/referentiality is triggered to satisfy the requirements of criterial positions in the clause.

Parallels in clausal and nominal periphery

165

Quite differently, in this paper I concentrate on DP-internal movements which create a dichotomy inside the noun phrase, parallel to the Topic/ Comment distinction found in the clause. This gives rise to a marked order of the constituents in the noun phrase, somewhat regardless of the information structure in the clause. This is not to claim that there may not be an interaction between information structure in a noun phrase and the clause containing it, especially in the computation of the clausal Focus. But for the time being, I abstract away from this aspect to avoid side effects.5 Mimicking Rizzi’s (1997) split CP (1a), in Giusti (2005) I assume a split DP (1b), where the Top and Foc are sandwiched between D (the highest) and d (the lowest functional head in DP): (1)

a. ForceP > TopP* > FocP > TopP* > FinP b. DP > TopP* > FocP > TopP* > dP

(Rizzi 1997) (Giusti 2005)

In this paper, I refine Giusti’s (2005) proposal concentrating on interface properties. Regarding the interpretive level, I substitute the Top and Foc features in the DP with a Topic position combined with a Contrast [Kon]feature (cf. Molnár 2002 for the sentence). Molnár proposes that [Kon] is a superimposed feature on Topic or Focus, and the highest one in the clause in order to guarantee the “discourse connection” in the sense of Haegeman and Guéron (1999). I assume that in the noun phrase, it can only be superimposed on Topic. For minimalist requirements I also assume that features can be conflated in one and the same head, to the result that the left periphery of the noun phrase (when projected) is reduced to a unique projection (namely, DP/TopP) which may contain a [±] specification for [Kon] (cf. also section 4). This is consistent with the general observation of the uniqueness of Focus in the clause, which straightforwardly excludes the possibility of a Focus in the noun phrase, as pointed out to me by Puglielli (p.c.). It is also consistent with the discourse function of fronted elements, at least in the languages under consideration. Regarding the spell-out level, following Frascarelli (2000), I propose an analysis of the syntax-phonology interface according to which the noun phrase structure is reinterpreted by prosodic structure into phonological constituents , along the lines of Nespor and Vogel (1986). The relevance of prosodic structure for the interpretation of discourse features is mediated by the computational system that encodes discourse features in syntactic positions, triggering the fronting of contrasted modifiers.

166 Giuliana Giusti 2. Three layers of nominal projection and the Possessor Criterion The structure in (2) represents the three layers of nominal projection, parallel to VP, IP, and CP in the clause:6 (2)

DP (Complementation layer) 2 AgrP (Inflectional layer) 2 NP (lexical layer) 2

In (3) below we observe a complex noun phrase with three arguments and three adjectival modifiers. The possessor sua, bearing the agent -role, is higher than the other APs7, and shares with them the property of agreeing for gender and number with the noun. In the present framework, it is implausible to assume that a -role is assigned as high in the structure, as the highest SpecAgrP. Consequently, I maintain the internal merge of the possessor inside NP and propose a Possessor Criterion, parallel to the Subject Criterion of Rizzi and Shlonsky (2004), which forces the subject of the noun to move out of the lexical layer to check EPP-features present in the DP as well as in the clause. This constitutes an important parallelism in the functional structure of noun phrases and clauses.8 I also maintain that the internal merge of adjectival modifiers is hierarchically ordered (as extensively argued for in Cinque 1994 and many others after him, including Giusti 1993). The final merge of N is also parametrized along the lines of Cinque (1994). In Italian, it targets the Agr head immediately dominating the AgrP where the lowest AP (e.g. appassionata) is merged: (3)

a. la sua seconda improbabile lettura appassionata the her second improbable reading passionate di Harry Potter a suo figlio of Harry Potter to her son ‘Her second improbable passionate reading of H.P. to her son’ b. [DP la [Agr-layer [AP sua] Agr° [seconda Agr° [incredibile [Agr°+N lettura] [appassionata [Agr°+N lettura] [NP-layer [AP sua] [N lettura] [di H. P. [N lettura] a suo figlio ]]]]]]]

Parallels in clausal and nominal periphery

167

Following Giusti (2002), I take this internal merge of N to trigger feature sharing of number and gender features on nominal modifiers. I assume that the Agr heads in (3b) are all copies of the same u-features [Gender], [Number], [Case] which are individually associated with different heads: Gender is associated with N in the lexical entry; Number is related to the DP-layer ([small] d, as argued for in sect.4), while Case is assigned to the DP by an external head and is realized in the highest projection of the DP-layer.9 The copies of these features are bundled together, as shown by the unique agreement marker on all nominal elements. In a minimalist fashion, I assume that features are bundled together (when possible) due to the notion “Economize structure”. I take Number, Gender and Case to be uninterpretable on Adjectives. As such, they must therefore be checked before Spellout. Following Chomsky (2000), I assume a Merge operation that creates structure in a minimal way, to the extent that no empty projections are merged. I propose to interpret the cartographic hierarchy as a UG principle guiding the phrase structure building operation Merge. According to this principle, Merge combines in hierarchical structure all and only the items (including features) that are in the array. Under this view, empty projections cannot be generated. The Agr-heads do not have semantic labels but are copies of u-features that must be checked against their specifiers before Spell-Out and deleted before LF. A language-specific principle of Italian requires that the head-N move at least to an Agr higher than the first (lowest) AgrP in which an adjectival modifier is merged. This N-movement (which is the result of a parametric choice along the lines of Cinque 1994) triggers u-feature copying in all Agr-heads that are subsequently merged. It is common ground to assume that the highest projection is DP which contains the article, e.g. la in (3). The question is whether there are reasons to believe that it can contain more than one or two elements (one in D and one in Spec). The core issue of this paper is to show that it can. 3. Marked orders in noun phrases across languages This section concerns A-bar movements inside the noun phrase. It is based on Giusti (1996), who claims that unexpected orders of the modifier hierarchy are related to a marked discourse interpretation. I elaborate on that proposal along the lines suggested by Frascarelli (2000) for the clause. Following Molnár (2002) for clause structure, I propose that the relevant discourse feature in the noun phrase under consideration is [±Kon], which

168 Giuliana Giusti combines with a Topic. I leave the possibility open for multiple Topics, especially for cases of multiple fronting. For reasons of space, I cannot get into that matter here. Molnár (2002) conceives [±Kon] as a feature that combines with either Topic or Focus. In the bare phrase structure perspective I have just outlined, I propose that in the noun phrase we can only have [+Kon] or [-Kon] Topics but no Focus, since Focus can only be checked in the main clause.

3.1. Marked orders in Italian In (3) above, I have claimed that the possessive adjective marks the highest AgrP. In (4) we observe the unmarked order of the adjectives in which long precedes blond and the possessive adjective precedes both. This order cannot be violated, as in (4b), unless a descriptive adjective is fronted, as in (5): (4)

a. le sue lunghe trecce bionde b. *le sue bionde trecce lunghe

(5)

a. le lunghe sue trecce bionde b. le bionde sue lunghe trecce

The intonation attributed to the marked orders in (5) rises on the fronted adjective followed by a slight pause after which the intonation of the rest of the noun phrase is identical to the unmarked intonation, with the typical fall on the last word: (6)

L H H L a. le lunghe # sue trecce bionde b. le bionde # sue lunghe trecce

The accented, rising intonation on the fronted adjective is different from the low tone that is specifically associated with familiar and continuing topics and is parallel to the high tone of shifting and contrastive topics, as reported by Frascarelli and Hinterhölzl (to appear). (For a definition of contrastive topics, cf. Büring 1999). There is therefore one feature in common in the prosody of a fronted adjective and that of a focused constituent: both are accented, differently from a [-Kon] Topic, usually de-stressed. As for interpretation, while Focus is related to new information, the fronted adjective is related to given information. Furthermore, Focus relates

Parallels in clausal and nominal periphery

169

to an open set of possible alternatives, while the contrasted modifier in (5) relates to a closed set of alternatives, in that only an adjective which expresses a property salient for N, or otherwise known as associated to N can be fronted. I therefore propose that the feature triggering fronting of the adjective in Italian is [+Kon], combined with a Topic, and heads a single head in the split DP (due to “Economize structure”). The structure proposed for (5) is in (7), with the fronted adjectives externally merged in the inflectional layer and further merged in a projection sandwiched between the highest AgrP and D°:10 (7)

a. [DP le [KonP [AP lunghe] Kon [AgrP [sue] Agr [AgrP [APlunghe] [Agr+N trecce] [AgrP [bionde] Agr [NP]]]]]] b. [DP le [KonP [AP bionde] Agr [AgrP [APsue] Agr [AgrP [APlunghe] [Agr+N trecce] [AgrP [bionde] Agr [NP]]]]]]

The spell-out structure in (7) feeds PF, which maps the linear order into two constituents, as in (8), strictly respecting the highest AgrP as a boundary, giving rise to the pause # reported in (6): (8)

a. [le lunghe] [sue trecce bionde] b. [le bionde] [sue lunghe trecce]

In the nominal domain, we are dealing with prosodic constituents no higher than the -domain (the Phonological Phrase). Partially parallel to the prosodification of Topic in the sentence, as proposed by Frascarelli (2000) and reported here as (9), I propose the Nominal Contrast Prosodic Domain and the Nominal Contrast Restructuring Rule in (10): (9)

a. Topic Prosodic Domain (Frascarelli 2000: 63, (117)) A Topic is minimally and exhaustively contained in an I b. Topic Restructuring (Frascarelli 2000: 63, (118)) If non-branching, a Topic may restructure into the adjacent constituent on either side.

(10) a. Nominal Contrast Prosodic Domain A Nominal Contrast is minimally and exhaustively contained in a . b. Nominal Contrast Restructuring If non-branching, a Nominal Contrast can restructure on its left side (in normal speech), or on either side (in fast speech).

170 Giuliana Giusti According to (10b), all elements preceding the DP can form a  together with a contrasted Adjective, not just the articles that are proclitic by nature. This is the case in (11):11 (11) a. [quei lunghi] [suoi monologhi interiori] those long his monologues internal b. [tutte quelle incredibili] [sue manie di grandezza] all those incredible his manias of grandeur c. [molti dei disgustosi] [suoi ultimi comportamenti] many of the disgusting his last behaviours In current minimalist terms, PF-rules are sensitive to phases. I suggest that the prosodification observed in (8) indicates that the noun phrase has two phases, at least when the higher phase, the DP, is realized in more than one projection, as in the cases under discussion. KonP in (7) is part of the splitDP. The low dP is not projected according to “Economize structure” (cf. section 4).

3.2. Prenominal adjectives in Albanian In Albanian, as in other Balkan languages, the syntax of possession is rather complex and the position of the possessive cannot be taken to mark the highest projection in the inflectional layer. For this purpose, I take the adjective meaning ‘other’ and the ordinal ‘first’. The unmarked order of adjectival modification in Albanian complies with Cinque’s (1994) hierarchy. In particular, we observe that ‘other’ in (12)– (13) and ‘first’ in (14)– (15) precede a descriptive adjective as ‘nice’ or ‘good’, irrespective of the presence of the nature of the article: proclitic indef. një or enclitic def. -(j)a /-i: 12 (12) a. një grua tjetër e bukur a woman other nice ‘another nice woman’ b. gruaja tjetër e bukur woman-the other nice ‘the other nice woman’

Parallels in clausal and nominal periphery

171

(13) a. *një grua e bukur tjetër b. *gruaja e bukur tjetër (14) a. një djalë i parë i mirë a boy first good ‘a first nice boy’ b. djali i parë i mirë boy-the first good ‘the first nice boy’ (15) a. *një djalë i mirë i parë b. *djali i mirë i parë The data above show that in the unmarked order, the head N in Albanian moves very high, crossing all adjectival modifiers, supposedly to the highest possible Agr, or even to [small] d, if this is merged at all (as suggested by an anonymous reviewer). No prenominal adjective is expected, as is in fact the case for unmarked orders. A prenominal adjective is however possible if it is emphasized, but not necessarily focused. This also happens irrespective of the kind of article: (16) a. një e bukur grua a nice woman

b. e bukura grua nice-the woman

(17) a. një tjetër grua a other woman

b. tjetra grua other-the woman

(18) a. një i parë djalë a first boy

b. i pari djalë first-the boy

(19) a. një i mirë djalë a good boy

b. i miri djalë good-the boy

If two adjectives are merged, the prenominal position is available for either one, yielding a violation of the hierarchy, as in (20)–(21).13 (20) a. një e bukur grua tjetër b. e bukura grua tjetër ‘a/the nice woman other’

172 Giuliana Giusti (21) a. një i mirë djalë i parë a good boy first b. i miri djalë i parë good-the boy first ‘a/the first good boy’ The fact that the hierarchy is only violated if the offending adjective is prenominal, straightforwardly suggests that its prenominal position is derived by (A-bar) movement. This movement is discourse-driven and triggers a contrastive interpretation of the prenominal modifier. This movement displays a prosodic reflex that is parallel to what we observe in Italian; namely, the displaced adjective prosodifies with the preceding article (if there is one) but is separated from what follows by a pause: (22) a. b. c. d.

[e bukura] [grua tjetër] [një e bukur] [grua tjetër] [i miri] [djalë i parë] [një i mirë] [djalë i parë]

From the results above, I suggest a [+Kon] projection in the Albanian noun phrase immediately lower than the DP: (23) a. [DP një [KonP [AP e bukur]i Agr [AgrP [Agr/N° grua][AgrP tjetër [Agr/N°grua] [AgrP [AP e bukur]i [Agr/N°grua] [NP …]]]]]] b. [DP një [KonP [AP i mirë]i [AgrP [Agr/N° djalë][AgrP i parë [Agr/N°djalë] [AgrP [AP i mirë]i [Agr/N°djalë] [NP …]]]]]] (24) a. [DP [e bukura]i D [KonP [AP e bukur]i [AgrP [Agr/N° grua][AgrP tjetër N [AgrP [AP e bukur]i N [NP N..]]]]]] b. [DP [i miri]i D [KonP [AP i mirë]i [AgrP [Agr/N° djalë][AgrP i parë [Agr/N°djalë] [AgrP [AP i mirë]i [Agr/N° djalë] [NP …]]]]]] When KonP is merged, N cannot move to D through Kon, unless it receives a [+Kon] feature. For this reason it is the AP with [+Kon] in SpecKonP that continues its way to SpecDP and checks the u-features expressed by the definite article, namely Case.14

Parallels in clausal and nominal periphery

173

For our parallel with the clausal periphery, it is relevant to point out that only one contrastive topic is allowed in the CP-phase (cf. Frascarelli and Hinterhölzl). The parallel therefore comes out nicely from the Albanian data, showing that only one prenominal [+Kon] adjective is allowed in the DP-phase. 3.3. Long and short adjectives in Serbo-Croatian Serbo-Croatian has the well-known property of allowing, for some adjectives, two different inflections (so-called “long” and “short” inflection), which are often related to the referential properties of the noun phrase, as indicated by the glosses in (25) (cf. Leko 1988, 1999; Aljovi 2002; Trenki 2004; Cinque 2005b). (25) a. siromaan poor-INDEF (= short) ‘a poor boy’ b. siromani poor-DEF (= long) ‘the poor boy’

djeak boy djeak boy

I believe that the proposal of a left periphery in the noun phrase can provide a fresh view to this intricate empirical domain. Attributing to the long adjectival inflection the semantic property of definiteness or specificity and the opposite referential value to the short form obscures one important fact, namely that adjectives with different endings can cooccur in one and the same noun phrase – as in (26) – in dedicated positions, so that the order AP-DEF > AP-INDEF in (27) is ungrammatical: (26) a. siromaan, poor-INDEF b. bolestan, sick-INDEF

bolesni sick-DEF siromani poor-DEF

djeak boy djeak boy

(27) a. *siromani(,) bolestan djeak b. *bolesni(,) siromaan djeak The cooccurrence of a definite and an indefinite adjective shows that the socalled definite/indefinite inflection of the adjective is not directly relevant

174 Giuliana Giusti to the interpretation of the noun as the glosses suggest. This observation is confirmed by the fact that a sequence such as (26) can be preceded by a demonstrative, resulting in the apparent contradiction that a supposedly indefinite/non-specific AP would be sandwiched between a demonstrative and a definite/specific AP in (28): (28) a. onaj this b. onaj this

siromaan, poor-INDEF bolestan, sick-INDEF

bolesni sick-DEF siromani poor-DEF

djeak boy djeak boy

Note that the short form (labelled as indefinite) appears in predicate position, either combined with a copula (29a), or adjoined to the noun in postnominal position as a secondary predication, as in (29b): (29) a. onaj this b. onaj this

djeak boy djeak, boy

je siromaan /*siromani is poor-INDEF/*DEF siromaan i bolestan poor and sick

Leko (1999) and Cinque (2005b) propose that this is the position for an appositive modifier. I propose that short adjectives in prenominal position are merged in a left peripheral [-Kon] Topic position, which is external to the inflectional layer where long-agreement takes place. The two hypotheses are not incompatible, in that it is not inconceivable that the non-restrictive interpretation is actually derived by the [-Kon] Topic feature. In other words, short adjectives would be “root”, predicative (appositive) APs, that are merged in TopP and specified for [-Kon]. Hence iterable, as expected. The example in (25a) thus means “poor, a boy/ a boy, who is poor” (also parallel to the Hungarian case in (34) below). The position of KonP is lower than DP, where the demonstrative is inserted (30a–b) and higher than a focalized adjective in (30c), which I assume to be in situ: (30) a. onaj siromaan, bolesni djeak this poor-INDEF sick-DEF boy b. *siromaan, onaj bolesni djeak poor-INDEF this sick-DEF boy c. onaj siromaan, BOLESNI djeak

Parallels in clausal and nominal periphery

175

The fronting analysis is confirmed by the comma intonation which is obligatory even if what follows is another indefinite adjective as in (31). This comma-intonation is not found between two adjectives inflected for definiteness, as in (32): (31) a. siromaan, bolestan djeak b. bolestan, siromaan djeak A-INDEF – A-INDEF N (32) a. siromani bolesni djeak b. bolesni siromani djeak A-DEF – A-DEF N In Serbo-Croatian, KonP cannot be prosodified with a constituent to its right unless this is light, as is the bare noun djeak. I tentatively propose the structure in (33a) with recursive KonP (as indicated by the star)15 mapped into the prosodic structure in (33b), or (33c): (33) a. [DP (dem) [KonP* AP-short [AgrP AP-long [NP N] b. [onaj siromaan] [bolesni djeak] c. [onaj siromaan] [bolestan djeak] Contrary to what I have proposed for Italian and Albanian (e.g., external merge of AP in the AgrP-layer and further internal merge into the split DP area), for Serbo-Croatian I propose external merge of the modifier in KonP to account for the reduced agreement on short forms. Notice however that not all adjectives have the complete declination (long and short). It is therefore plausible that inflectional morphology on A is undergoing a reduction and that short adjectives are externally merged in the AgrP-layer and later moved to the left periphery in Serbo-Croatian as well.

3.4. Possible extensions In the preceding sections, I have analyzed the marked orders of prenominal adjectives in Italian, Albanian and Serbo-Croatian,16 proposing a fronting analysis of the offending modifier into the highest layer of noun phrase structure namely the DP. As a consequence, I have proposed a split of the DP-layer into two projections, the lower of which has proven to be silent (and for minimal requirements not independently projected, at least in the

176 Giuliana Giusti cases under scrutiny), which sandwich the projection of the discourse feature [Kon]. I have also proposed two phonological rules that map the obtained structure into phonological constituents. I have observed that in normal speech there is a pause between the contrasted modifier and the rest of the noun phrase in the three languages. If this phenomenon is present in three unrelated languages, it is expectedly found across languages. From a typological perspective, Rijkhoff (2002: 267–268) argues for adjective displacement in Turkish and Hungarian in cases in which the adjective is emphasised, as the following in (34): (34) a. gyönyör egy állat! beautiful an animal ‘what a beautiful animal!’ b. büyük bir ev big a house ‘a big house’

(Hungarian, from Tompa 1968: 77)

(Turkish, from Krámsk 1972: 112)

In both languages the unmarked order is Art AP N and in both cases it is the emphasis on the adjective that triggers the marked order by fronting the adjective. In these two languages we find evidence for the low d which follows KonP. Apparently, if the low d is overt the high D is covert in (34)17. Cases of redundancy are expected and can be exemplified by the case reported by Rijkhoff (op. cit.) for Turkish: (35) a. bir güzel bir kiz BIR beautyful BIR girl

(Turkish, from Krámsk 1972: 112)

b. [DP bir [KonP güzel [dP bir [ ….. kiz]]]] The two occurrences of bir in (35a), which is glossed as the indefinite article in (34), can be analysed in our framework as in (35b) with the split DP sandwiching KonP overtly marked in both heads. More research is needed to support this claim, which we cannot carry over here for reasons of space.18 Many questions remain open in this preliminary proposal. In the next section, I raise possible questions and provide tentative answers.

Parallels in clausal and nominal periphery

177

4. The split DP layer If the structure of DP is the one proposed in (36), the first question to ask is why the lower dP is rarely merged: (36) DP > KonP > dP A possible answer comes from what is proposed for the split CP-system by Rizzi (1997), as stated in (37): (37) a. TopP and FocP are merged only if necessary. b. If no TopP or FocP is merged, ForceP and FinP can be realized in a unique projection. c. If ForceP and FinP are split, it is not often the case that both heads are realized by an overt element, in some cases one is zero and the other is overt. Given (37), we expect that if [Kon] is not present in the array, DP and dP are realized as one and the same projection. I therefore attribute the properties in (38) to the Split DP in (36): (38) a. The projection of discourse features are merged only if necessary. b. If no discourse feature is merged, DP and dP can be realized in a unique projection, and they must be (for reasons of Economy). c. If DP and dP are split, it is often the case that one is zero and the other is overt (for reasons of Economy). However, if D and [small] d can split, they obviously realize two (or more) features which can/must be merged in one and the same head if no [Kon] is present in the array. I propose that these features are Case (for the high D) and Number (for the low d). These features are realized in one and the same head in syncretic languages, but are obviously split in analytic languages. With respect to the clause, D realizing Case is parallel to Force which realizes the selection of a higher head, while d realizing Number is parallel to Fin. In the same fashion as C selects for Fin which in turn selects for Tense, I propose that D (reference) selects for (semantic) Number which in turn selects for morphological number in Agr.

178 Giuliana Giusti The parallelism and the relative complementarity between Force and Case is clear. It has often been noticed that clauses cannot appear in Case assignment positions (cf. Stowell 1982). Vice versa, noun phrases cannot have the [+wh] features that are possible in clauses. The Force of an embedded clause either matches the selectional requirements of the selecting head [±wh], or the features of an element that they modify [+rel], as is the case of relative clauses. Root clauses in some languages lack the highest CP projection because they do not have to comply with any of the previously listed cases. In other languages they must comply with requirements that give rise to so called “root-phenomena”. In a parallel fashion, Case is the feature that relates the noun phrase to the syntactic context.19 It is either assigned by a selecting head, (verbal or prepositional), or by a functional projection of the selecting head. “Root noun phrases” can be exemplified by vocative Case or exclamative noun phrases as (34) above, both of which are not embedded into a larger constituent. The Fin of an embedded clause may also be dependent on the requirements of the selecting head: there are verbs that only select for finite or infinitival complements, many others can select for either. A parallel case is represented by the feature Number of noun phrases. There are verbs (such as gather) that select for the Number of the subject or object, while others accept either singular or plural arguments. Semantic Number is realized in d. In turn, d determines the morphological number in Agr according to the inflectional properties of N. In conclusion, there are good reasons to believe that the “imperfect” parallelisms between the noun phrase and the clause can be reduced to a single feature, lack of inflectional Tense in the noun phrase (at least in the languages under scrutiny) and consequent lack of Fin in the higher (DP) layer.

Acknowledgements For discussion, suggestions, and stimulating criticism, I am indebted to Anna Cardinaletti, Guglielmo Cinque, Mila Dimitrova-Vulchanova, Verner Egerland, Mara Frascarelli, Alessandra Giorgi, Nedad Leko, Valeria Molnár, Annarita Puglielli, Luigi Rizzi, and a peer reviewer. All remaining errors are mine. I also thank Megan Rae for revising the English.

Parallels in clausal and nominal periphery

179

Notes 1. Since Ross (1967), informationally marked orders are derived by means of syntactic movements, to be interpreted at the interfaces. In the meantime a fine graded array of constructions have been formalized, e.g. various kinds of leftor right- dislocation with or without resumptive clitics in different languages. 2. In the last twenty years, a great number of studies have been dedicated to the parallelisms between the noun phrase and the clause that cannot be reported here. In these studies, however, a reduced capacity of expansion is often observed for the noun phrase as opposed to the clause: less adjectives than adverbs are supported in the modification area; only one structural Case (genitive) is usually assigned in the noun phrase as opposed to two structural cases in the sentence; only declarative complementation is realized by the noun phrase (cf. English *I wonder what answer). Finally, very few languages have tense features and only a few languages have argument clitics in the noun phrase. 3. Clitics in the noun phrase have a more restricted distribution across languages than in the clause, (cf. Giusti and Stavrou in progress). 4. Lack of Tense appears to prevent wh-checking in infinitival clauses in Germanic languages (cf. Giusti 1991) and in Romanian (cf. Giusti 1995b). If lack of Tense is parametrized, we expect that DPs in a language such as Somali (which is argued to have Tense in noun phrases by Lecarme 1996) could also have wh-features, clitics and propositional value. 5. I assume that a Focus feature on a modifier, e.g. blue in Give me the BLUE pen is an instance of in situ Focus which is checked (after spell-out) in the FocP of the clause, through pied-piping, in the same fashion as the wh-modifier in which problem is checked against wh-features in a SpecCP. I thank the anonymous reviewer for pointing out the potential problematicity of this case. 6. Here I disregard the behaviour of quantifiers. They are merged outside the DPsystem and build an independent lexical projection, which in turn projects two functional layers, as extensively argued for in Cardinaletti and Giusti (2006), who also review previous literature. 7. I assume possessives such as mio (‘my/mine – Ms.sg.’), tuo (‘your(s) – Ms.sg.’), etc. in Italian to be APs due to their morphological property of agreeing with the head N for gender and number. A possessive such as loro (‘their(s)’) is however a pronoun (cf. Cardinaletti 1998) parallel to his/her in English. Regardless of their adjectival or pronominal nature, possessives in Italian appear to have the same distribution. In Giusti (1993) and in following work I propose that they receive a -role in a “low” position (SpecNP or immediately higher) and move to the “high” position for reasons which I formalize here as an extension of the Subject Criterion found in the clause. 8. A possessive AP is found in postnominal position if it bears stress (cf. Cardinaletti 1998), as in (ia). I take this to be perfectly parallel to a clausal subject

180 Giuliana Giusti

9.

10. 11.

12.

13.

14.

15. 16.

(ib), which appears in a low position if focused (procrastinating feature checking against AgrP): (i) a. un’altra esecrabile dichiarazione immediata SUA alla stampa another hateful declaration immediate HIS to the press ‘Another hateful immediate declaration of his to the press’ b. Ha esecrabilmente dichiarato subito LUI alla stampa che… has hatefully declared immediately HIM[SELF] to the press that… ‘He has himself hatefully immediately declared to the press that...’ Apparently, the Possessor Criterion can be satisfied at a later stage in the derivation leaving the possessive adjective low until Spell-Out. I abstract away from the postnominal position of the possessor and use the prenominal position to mark the highest AgrP. Where this is not possible for language-internal reasons, I use the high adjective altro, ordinals or numerals. For the position of these adjectives cf. Giusti (1993, 2002). In this sense it is very similar to the F head of Alexiadou (2001). A crucial difference, though, is that instead of being a single head which heads a projection with multiple adjunctions, the bundle of features is copied in multiple Agrheads, each of which hosts a unique specifier. Only in this way can the specifiers be merged according to the hierarchy, while adjunctions can be taken to be insensitive to ordering (as Cinque 1999 claims for adverbials). The internal merging position of sue and trecce is irrelevant here and is not indicated in the NP. The quantifiers in (11b,c) are not in the DP, but restructure into the adjacent  at PF. This also happens with a one-word V, as Sciolse le lunghe # sue bionde trecce (lit. ‘[she] undid the long her blond braids’). The pre-adjectival morpheme e/i is a sort of agreement for -features (including definiteness/Case) and appears with some adjectives and not others, independently of the (in)definiteness features of the noun phrase, cf. Bucholtz (1987: 319–324). For this reason this morpheme is not glossed. When two adjectives are present, fronting is only possible for some speakers. I cannot establish whether it is a matter of dialectal or idiolectal variation. I thank Marinela Sotiri (and her family) for discussing this data with me. Sotiri (p.c.) suggests that these structures are strongly banned by prescriptive grammar and that the normative pressure can be so strong as to influence their acceptability. See Dimitrova-Vulchanova and Giusti (1998) for AP-to-SpecDP versus A-toD movement in Albanian, Bulgarian, and Rumanian. Also see Giusti (1995a, 2002) for Case features on articles and Dimitrova-Vulchanova (2003) for an alternative analysis of the AN vs. NA order. The head N does not move, or moves to a very low Agr position. This alleged movement is irrelevant to our topic. The fact that both Serbo-Croatian and Albanian belong to the Balkan Sprachbund does not seem to be relevant in this respect, given that the former has no

Parallels in clausal and nominal periphery

181

article, no (or very limited) N-movement, and the long/short declension of adjectives, while the latter has an enclitic article, N-movement targeting the highest Agr of the inflectional layer and no trace of a long/short distinction. 17. Valeria Molnár (p.c.) suggests that the examples in (34) make good candidates for Focus movement inside the noun phrase. This is not in contrast with my proposal to ban Focus in DPs. There are two possible ways to interpret the lack of an explicit clause in (34). Either is the DP in the CP of an elliptic exclamative clause, and in this case, the movement of the modifier inside DP piedpipes the Focus feature on the whole DP which is checked at the clausal level, also cf. footnote 5. Or is the DP an independent discourse unit, and is entitled to bear the unique Focus feature of root clauses. 18. For more cases of DP-internal AP-movement cross-linguistically, see Cinque (2005a). Dimitrova-Vuchanova and Mieska-Tomi (to appear) extend my hypothesis to cases of “double definiteness” in Bulgarian where a demonstrative precedes an adjective inflected with the definite article. 19. For evidence of the relation between Case and articles in a diachronic perspective cf. Giusti (1995a).

References Aboh, Enoch 2004 Topic and Focus within D. Linguistics in the Netherlands 21: 1–12. Alexiadou, Artemis 2001 Functional Structure in Nominals. Nominalization and Ergativity. Amsterdam: Benjamins. Aljovi , Nadira 2002 Long adjectival inflection and specificity in Serbo-Croatian. Recherches Linguistiques de Vincennes 31: 27–42. Bernstein, Judy 2001 Focussing the ‘right’ way in Romance Determiner Phrase. Probus 13 (1): 1–29. Buchholtz, Oda and Wilfried Fiedler 1987 Albanische Grammatik. Leipzig: Verlag Enzyklopedie. Büring, Daniel 1999 Topic. In Focus – Linguistic, Cognitive and Computational Perspectives, Peter Bosch and Rob van der Sand (eds.), 142–165. Cambridge: Cambridge University Press. Cardinaletti, Anna 1998 On the deficient/strong opposition in possessive systems. In Possessors, Predicates and Movement in the Determiner Phrase, Artemis Alexiadou and Chistopher Wilder (eds.), 17–53. Amsterdam: John Benjamins.

182 Giuliana Giusti Cardinaletti, Anna and Giuliana Giusti 2006 The syntax of partitive phrases and quantitative clitics. In The Blackwell Companion to Syntax, vol. V, Martin Everaert and Henk van Riemsdijk (eds.), 23–93. London: Basil Blackwell. Chomsky, Noam 2000 Minimalist inquiries. In Step by Step. Essays on Minimalist Syntax in Honor of Howard Lasnik, Roger Martin, Daniel Michaels and Juan Uriagereka (eds.), 89–155. Cambridge, MA: MIT Press. 2005 Three factors in language design. Linguistic Inquiry 36 (1): 1–22. Cinque, Guglielmo 1994 On the evidence for partial N-movement in the Romance DP. In Paths to Universal Grammar. Studies in Honor of R. Kayne, Cinque, Guglielmo, Jan Koster, Jean-Yves Pollock, Luigi Rizzi and Raffaella Zanuttini (eds.), 85–110. Washington, D.C.: Georgetown University Press. 1999 Adverbs and Functional Heads. A Cross-linguistic Perspective. Oxford: Oxford University Press. 2005a Deriving Greenberg’s Universal 20 and its exceptions. Linguistic Inquiry 36 (3): 315–332. 2005b The dual source of adjectives and phrasal movement in the Romance DP. Ms. University of Venice. Dimitrova-Vulchanova, Mila 2003 Modification in the Balkan nominal expression: an account of the (A)NA : AN(*A) order contrast. In From NP to DP, Coene Martine and Yves D’Hulst (eds.), 91–118. Amsterdam: John Benjamins. Dimitrova-Vulchanova, Mila and Giuliana Giusti 1998 Fragments of Balkan nominal syntax. In Possessors, Predicates and Movement in the Determiner Phrase, Artemis Alexiadou and Christopher Wilder (eds.), 333–360. Amsterdam: Benjamins. 1999 Possessors in the Bulgarian noun phrase. In Topics in South Slavic Syntax and Semantics, Mila Dimitrova-Vulchanova and Lars Hellan (eds.), 163–192. Amsterdam: Benjamins. Dimitrova-Vulchanova, Mila and Olga Mieska Tomi

to appear The structure of the Bulgarian and Macedonian nominal expression: Introduction. In Investigations in the Bulgarian and Macedonian Nominal Expression, Dimitrova-Vulchanova, Mila, Olga Mieska Tomi and Zuzanna Topolinska (eds.). Holland: Rodopi. Frascarelli, Mara 2000 The Syntax-Phonology Interface in Focus and Topic Constructions in Italian. Dordrecht: Kluwer. Frascarelli, Mara and Roland Hinterhölzl to appear Types of topics in German and Italian. In On Information Structure, Meaning and Form, Kerstin Schwabe and Susanne Winkler (eds.). Amsterdam: John Benjamins.

Parallels in clausal and nominal periphery

183

Gavrouseva, Elisabeth 2000 On the syntax of possessor extraction. Lingua 110: 830 –772. Giusti, Giuliana 1991 Zu-infinitivals and sentencial structure in German. Rivista di Linguistica 3 (2): 211–234. 1993 La Sintassi dei Determinanti. Padova: Unipress. 1995a A unified structural representation of (abstract) case and articles. Evidence from Germanic. In Studies in Comparative Germanic Syntax, Hubert Haider, Susan Olsen and Sten Vikner (eds.), 77–93. Dordrecht: Kluwer. 1995b Lo statuto categoriale del morfema infinitivale a in rumeno. In Studi Rumeni e Romanzi. Omaggio a Florica Dimitrescue, Alexandru Niculescu, Coman Lupu and Lorenzo Renzi (eds.), 132–148. Padova: Unipress. 1996 Is there a FocusP and a TopicP in the noun phrase structure? University of Venice Working Papers in Linguistics 6: 105–128. 2002 The functional structure of noun phrases. A bare phrase structure approach. In Functional Structure in DP and IP: The Cartography of Syntactic Structures, vol. 1, Guglielmo Cinque (ed.), 54–90. Oxford: Oxford University Press. 2005 At the left periphery of the Rumanian noun phrase. To appear in Proceedings of the Tense and Aspect Conference (Antwerp 19–22 September 2004), Martine Coene and Liliane Tasmowski (eds.). Cluj (Romania): Clusium. Giusti, Giuliana and Melita Stavrou in prep. Possessive clitics in the DP: doubling or dislocation? Ms., University of Venice and Thessaloniki. Haegeman, Liliane 2004 The remote possessor construction in West Flemish. Ms., University of Lille III. http://ling.auf.net/lingBuzz. Haegeman, Liliane and Jaqueline Guéron 1999 English Grammar. A Generative Perspective. Oxford: Blackwell. Horrocks, Geoffrey and Melita Stavrou 1987 Bounding theory and Greek syntax: Evidence for wh-movement in NP. Journal of Linguistics 23: 79–108. Ihsane, Tabea and Genoveva Puskás 2001 Specific is not definite. Generative Grammar in Geneva 2: 39–54. Krámsk , Jiri 1972 The Article and the Concept of Definiteness in Language. Janua Linguarum (Series Minor) 125. The Hague: Mouton. Lecarme, Jaqueline 1996 Tense in the nominal system: the Somali DP. In Studies in Afroasiatic Grammar, Jaqueline Lecarme, Jean Lowestamm, Ur Shlonsky (eds.), 159–178. The Hague: Holland Academic Graphics.

184 Giuliana Giusti Leko, Nedad 1988 X-bar theory and internal structure of NPs. Lingua 75, 135–169. 1999 Functional categories and the structure of the DP in Bosnian. In Topics in South Slavic Syntax and Semantics, Mila DimitrovaVulchanova and Lars Hellan (eds.), 229–252. Amsterdam: John Benjamins. Molnár, Valeria 2002 Contrast – from a contrastive perspective. In Information Structure in a Cross-linguistic Perspective, Hilde Hasselgård, Stog Johansson, Bergljot Behrens and Cathrine Fabricious-Hansen (eds.), 147–162. Amsterdam /New York: Rodopi. Nespor, Marina and Irene Vogel 1986 Prosodic Phonology. Dordrecht: Foris. Pustejowski, James 1995 The Generative Lexicon. Cambridge, MA: MIT Press. Rijkhoff, Jan 2002 The Noun Phrase. Oxford: Oxford University Press. Rizzi, Luigi 1997 The fine structure of the left periphery. In Elements of Grammar. Handbook in Generative Syntax, Liliane Haegeman (ed.), 281–337. Dordrecht: Kluwer. Rizzi, Luigi and Ur Shlonsky 2004 Strategies of subject extraction. Ms., University of Siena. Ross, John R. 1967 Constraints on variables in syntax. Ph.D. diss., MIT. Stowell, Timothy 1982 Subjects across categories. The Linguistic Review 2: 285–312. Svenonius, Peter 2004 On the edge. In Peripheries, David Adger, Cecile de Cat and George Tsoulas (eds.), 261–287. Dordrecht: Kluwer. Tompa, Jósef 1968 Ungarische Grammatik. Janua Linguarum (Series Practica 96). The Hague: Mouton. Trenki Danijela 2004 Definiteness in Serbian/Croatian/Bosnian and some implications for the general structure of the nominal phrase. Lingua 114: 1401–1427.

Chapter 3 Functional projections in the vP-phase

The properties of anticausatives crosslinguistically Artemis Alexiadou, Elena Anagnostopoulou and Florian Schäfer

In this paper we provide evidence against derivational approaches to the causative/anticausative alternation. On the basis of the behavior of anticausatives in English, German and Greek we put forth an analysis of change of state verbs which builds on the idea that these are decomposed and contain in addition to the root they are derived from a Voice and a Cause component.

1. Goals The causative/anticausative alternation has been the topic of much typological and theoretical discussion in the linguistic literature. This alternation is characterized by verbs with transitive and intransitive uses, such that the transitive use of a verb V means roughly ‘cause to V-intransitive’ (see Levin 1993). The discussion revolves around two issues: the first one concerns the similarities and differences between the anticausative and the passive, and the second one concerns the derivational relationship, if any, between the transitive and intransitive variant. With respect to the second issue, a number of approaches have been developed. Judging the approach conceptually unsatisfactory, according to which each variant is assigned an independent lexical entry, it was concluded that the two variants have to be derivationally related. The question then is which one of the two is basic and where this derivation takes place in the grammar. Our contribution to this discussion is to argue against derivational approaches to the causative/anticausative alternation. We focus on the distribution of PPs related to external arguments (agent, causer, instrument, causing event) in passives and anticausatives of English, German and Greek and the set of verbs undergoing the causative/anticausative alternation in these languages. We argue that the crosslinguistic differences in these two domains provide evidence against both causativization and detransitivization

188 Artemis Alexiadou, Elena Anagnostopoulou and Florian Schäfer analyses of the causative/anticausative alternation. We offer an approach to this alternation which builds on a syntactic decomposition of change of state verbs into a Voice and a CAUS component. Crosslinguistic variation in passives and anticausatives depends on properties of Voice and its combinations with CAUS and various types of roots.

2. Distinguishing between passives and anticausatives Passives and anticausatives in English differ in two well-known aspects (Manzini 1983; Marantz 1984; Jaeggli 1986; Roeper 1987; Baker, Johnson and Roberts 1989; Levin and Rappaport Hovav 1995; Reinhart 2000; Chierchia 1989/2004, among many others): (i) Modification / Control: Passives but not anticausatives can be modified by by-phrases, agent-oriented adverbs, and allow control into purpose clauses, as illustrated in the examples (1–3): (1)

a. The boat was sunk by Bill b. *The boat sank by Bill

(2)

a. The boat was sunk on purpose b. *The boat sank on purpose

(3)

a. The boat was sunk to collect the insurance b. *The boat sank to collect the insurance

(ii) Verb Restrictions: Virtually any transitive verb can be passivized, but only a subset of transitive verbs form anticausatives: (4)

a. The baker cut the bread b. The bread was cut by the baker c. *The bread cut

(5)

a. Bill broke the glass b. The glass was broken by Bill c. The glass broke

The properties of anticausatives crosslinguistically

189

2.1. Previous explanations As far as the first difference is concerned (modification and control), the consensus has been reached that this is due to the presence vs. absence of an implicit external argument in passives and anticausatives respectively. While passives contain such an implicit external argument which can be accessed by modification (by-phrases, agent-oriented adverbs) and can control into purpose clauses, anticausatives lack such an argument and therefore modification and control are impossible (see (1)–(3)). Two issues of controversy remain, though: the level at which the implicit external argument is expressed in the passive1 and why anticausatives lack an implicit external argument. To answer the latter question, two influental views have been proposed. According to one view, anticausatives lack an implicit external argument because they are basically monadic. The causative alternant is derived from the anticausative/inchoative via causativization (Lakoff 1968, 1970; Dowty 1979; Williams 1981; Brousseau and Ritter 1991; Pesetsky 1995, among others). This is illustrated in (6) below taken from Dowty (1979, section 4.3): (6)

a. breakincho : b. breakcaus :

x [Become BROKEN (x)] yx [P [P (x) Cause Become BROKEN (y) ]]

According to the second view, alternating verbs are inherently dyadic predicates. Anticausatives lack an implicit external argument due to a lexical process of detransitivization that creates an intransitive entry from the transitive one. There are two recent implementations of this general idea, which we briefly summarize below: (i) Levin and Rappaport Hovav (1995: 83, 108; henceforth L&RH) propose a bi-eventive analysis of causative verbs. Their lexical semantic representation (LSR) of such verbs involves the predicate ‘cause’ which takes two arguments: the causing subevent and the central subevent (the latter specifying the change associated with the verb). The cause argument is associated with the causing subevent and the theme is associated with the central subevent. In transitive break the cause and the theme are projected from the LSR into argument structure (AS) (and from AS onto the syntax) as shown in (7):

190 Artemis Alexiadou, Elena Anagnostopoulou and Florian Schäfer (7)

Transitive break: LSR Linking rules AS

[[x do-something] cause [y become BROKEN]] ? ? x

In intransitive break the cause is lexically bound in the mapping from LSR to AS thereby being prevented from being projected into the syntax (cf. 8): (8)

Intransitive break: LSR Lexical binding Linking rules AS

[[x do-something] cause [y become BROKEN]] ? ! !  ! ?

(ii) Reinhart (2000, 2002), building on Chierchia (1989/2004), proposes that causation is coded through a lexical cause [+c] feature defining a set of theta-roles that cause change, namely cause, agent, instrument.2 Alternating verbs are inherently transitive. They select a [+c] external argument (and a [-c-m] theme internal argument). Anticausatives are derived from the transitive entry in the lexicon by a reduction operation (called “expletivization” and shown in (9)) that reduces the external [+c] role. The output of expletivization is a one place (intransitive) verb entry: (9)

Expletivization: Reduction of an external [+c] role a. Vacc (1[+c],2)
Re (V) (2) b. Re (V) (2) = V (2)

2.2.

Problems for previous analyses3

2.2.1. Morphological marking While the causative alternation is a semantically quite well-defined crosslinguistic phenomenon, languages show substantial variation in the morphological shape of the alternation (Haspelmath 1993) to which neither of the derivational accounts can do full justice. Any derivational approach that derives one version of the causative/anticausative alternation from the other states that the derived version is more complex, since it is formed by an extra

The properties of anticausatives crosslinguistically

191

operation on some computational level of grammar. But the morphological variation found with the alternation does not support any direction of derivation in a compelling way; both views discussed above are challenged by languages with special morphological marking on what is assumed to be the basic version of the alternation, i.e. they are challenged by a mismatch of assumed derivational and overt morphological complexity. The causativization view faces the problem that it leaves unexplained the fact that in many languages the anticausative and not the causative variant of the alternation is marked by special morphology ((10), see also Haspelmath 1993; Chierchia 1989/2004; L&RH 1995; Reinhart 2000, 2002; Piñon 2001b). (10) Anticausative Marking: a. Russian: katat’-sja katat’ b. Polish: zama -si zama

‘roll (intr)’ ‘roll (tr)’ ‘break (intr)’ ‘break (tr)’

(Haspelmath 1993: 91) (Piñon 2001b: 2)

On the other hand, as Piñon (2001b) points out, the postulation of a detransitivization process faces a similar problem with languages that mark the causative variant of the alternation (11). Furthermore, there are also languages with non-directed alternations which do not fit easily any of the above views: equipollent when both forms are derived from a common stem (12a), suppletive when different roots are used (12b), labile when the same form is used (12c). (11) Causative Marking: a. Georgian:

du -s ‘cook (intr) a-du -ebs ‘cook (tr)’ b. Khalka Mongolian: ongoj-x ‘open (intr)’ ongoj-lg-ox ‘open (tr)’

(Haspelmath 1993) (Piñon 2001)

(12) Non-directed Alternations: a. Japanese: atum-aru atum-eru b. Russian: goret’ e

c. English: open open

‘gather (intr)’ ‘gather (tr)’ ‘burn (intr)’ ‘burn (tr)’ ‘intr’ ‘tr’

(Haspelmath 1993)

192 Artemis Alexiadou, Elena Anagnostopoulou and Florian Schäfer 2.2.2. Verb restrictions and selection restrictions Both accounts face the same logical problem that sometimes they have to derive change of state verbs from a corresponding non-existent base. The causativization view faces this problem with the derivation of causatives from non-existing anticausatives; this is the case in the context of the verb restrictions discussed above, i.e. the difference between “break” and “cut”: the former can form an anticausative but the latter lacks an anticausative. The same problem also emerges with change of state verbs that show selection restrictions. More specifically, some verbs have intransitive uses only for certain choices of internal arguments, as is illustrated below with examples taken from L&RH (1995, 85–86). Causatives impose no such selection restrictions. (13) a. He broke his promise / the contract / the world record b. *His promise / the contract / the world record broke c. He broke the vase d. The vase broke L&RH (1995: 105–106) argue that verb restrictions and selection restrictions are related and both can be handled by the detransitivization view on the basis of the following generalization (Smith 1970; L&RH 1995; see also Reinhart 2000, 2002): (14) The transitive verbs that cannot form anticausatives restrict their subjects to agents or agents and instruments and disallow causers. As shown in (15)–(16), the non-alternating “cut” selects an agent or an instrument as a subject but disallows a causer, while the alternating “break” is compatible with an agent, an instrument and a causer subject, in accordance with generalization (14). (15) a. The baker / the knife cut the bread b. *The lightning cut the clothesline c. *The bread cut (16) a. The vandals / the rocks / the storm broke the window b. The window broke

The properties of anticausatives crosslinguistically

193

The same explanation can also account for the selection restrictions in (13) above because for certain choices of objects the nature of the external argument is specified. The eventuality cannot come about without the intervention of an agent in (13b). The rationale behind (14) in the detransitivization view is that a causative verb can leave its external argument unexpressed, if its thematic nature is left underspecified (agent or causer or instrument). If the verb lexically specifies something about the nature of the external argument, then the external argument position cannot be “lexically bound” or “reduced”. But the detransitivization view also encounters the logical problem that sometimes it would have to derive something from a non-existing base. This is the case with change of state unaccusatives which have no causative counterpart (e.g., bloom, blossom, decay, flower). The examples in (17) are taken from L&RH (1995: 97). (17) a. The cactus blossomed early b. *The gardener blossomed the cactus c. *The warm weather blossomed the cactus The crucial property of these verbs is that they describe changes of state that are internally caused, i.e. the cause of the change of state event is linked to properties inherent to the argument undergoing change. In contrast, verbs that have a causative counterpart can be externally caused, i.e. can be brought about by an external cause (cf. L&RH 1995). In the next sections, we discuss two further problems for a detransitivization approach towards anticausatives. The first one (discussed in section 3) concerns the types of arguments that can be introduced by PPs in anticausatives. The second one (discussed in section 4) has to do with crosslinguistic differences in verbs licensing the alternations and selectional restrictions. Our factual discussion will concentrate on data from English, German and Greek. 3.

PP Modification in passives and anticausatives crosslinguistically

3.1. English As already mentioned in section 2.2.1, English causatives license all types of external arguments, namely agents, causers (18), causing events (19a) and instruments (20a). Note that causing events and instruments can also be introduced as PPs, coocurring with agent subjects (19b /20b).

194 Artemis Alexiadou, Elena Anagnostopoulou and Florian Schäfer (18) John / The earthquake broke the vase (19) a. Will’s banging shattered the window b. I cooled the soup by lowering the temperature (20) a. A stone broke the window b. I broke the window with a stone In the English passive, PPs bearing all of the above thematic roles are licit. (21) The window was broken by John / by the storm / with a stone (22) The window was shattered by Will’s banging Turning to English anticausatives, notice first that they license the phrase by itself in the interpretation “without outside help”: (23) The plate broke by itself Chierchia (1989/2004) and L&RH (1995) argue that this modifier reflects the presence of a cause component in the LSR of anticausatives, providing evidence for the detransitivization analysis. Anticausatives do not license agents, instruments and causers/causing events introduced by the preposition by, as shown in (24) and (25): (24) *The window broke by John / with a stone (25) a. *The window broke by the storm b. *The window shattered by Will’s banging However, they do license causers and causing events if these are introduced by the preposition from, as has been discussed in the literature (DeLancey 1984; Piñon 2001a; L&RH 2005; Kallulli 2005). (26) The window cracked / broke from the pressure4 (27) The window cracked / broke from the explosion (28) *The door opened from Mary / from the key

The properties of anticausatives crosslinguistically

195

The distribution of PPs in English passives is correctly predicted by the ditransitivization approach. As mentioned in section 2.1, passives (of causatives) contain a thematically unspecified implicit external argument (resulting from saturation in Reinhart’s system, or it is present in A-structure and is bound from the mapping from A-structure into syntax in L&RH’s terms). This implicit external argument can be modified by PPs denoting agents, instruments, causers/causing events, i.e. the three theta-roles that are also licensed in the corresponding causatives. Anticausatives, on the other hand, are taken not to contain a thematically unspecified implicit external argument. Therefore, PPs denoting agents, instruments, causers / causing events, i.e. the three theta-roles that are licensed in the corresponding causatives, are expected not to be licensed. This prediction seems to be borne out if one concentrates on the by-phrases in (24) and (25), but not if one takes into consideration the well-formedness of the causer from-phrases in (26)–(28). On the detransitivization view, these examples are expected to be ungrammatical, contrary to fact. In the next two sections we will show that the same problem arises in German and Greek. 3.2. German Before we proceed to the discussion of the argument realisation in German causatives, passives and anticausatives, a note on the prepositions associated with the different thematic roles under discussion is in order. In German agents are introduced by von (32), instruments by mit (31b)–(32) causers/natural forces by durch (32),5 and causing events by durch (see (30b), (33)). As in English, causatives license all of the above external arguments (see (29), (30a), (31a)), and agent-subjects can co-occur with causing-event/instrument PPs in causatives ((30b) and (31b)). (29) Hans / der Erdstoß zerbrach die Vase Hans / the earth-tremor broke the vase (30) a. Das Rauchen von Zigaretten verschlechtert die Luftqualität the smoking of cigarettes worsens the air-quality im Raum. in-the room ‘Smoking cigarettes worsens the air quality in the room’

196 Artemis Alexiadou, Elena Anagnostopoulou and Florian Schäfer b. Peter verschlechtert die Luftqualität im Raum Peter worsens the air-quality in-the room durch das Rauchen von Zigaretten. through the smoking of cigarettes ‘Peter worsens the air quality in the room by smoking cigarettes’ (31) a. Die Medizin the medicine b. Der Arzt the doctor

heilt cures heilt cures

den Patienten the patient den Patienten mit der Medizin 6 the patient with the medicine

German passives behave similarly to their English counterparts. They permit agents, causers/forces, instruments (32) and causing events (33): (32) Die Vase wurde von Peter / durch den Erdstoß / mit dem the vase was by Peter / through-the earth tremor / with the Hammer zerbrochen hammer broken ‘The vase was broken by Peter / by the earth tremor / with the Hammer’ (33) Die Luftqualität im Raum wird durch das Rauchen the air-quality in-the room is through the smoking von Zigaretten verschlechtert of cigarettes worsened ‘The air quality in the room is worsened by the smoking of cigarettes’ German anticausatives do not license agents and instruments (34) but license causers and causing events if these are introduced by durch (35)– (36). Thus they behave exactly like their English counterparts. Note that there are two morphologically distinct types of anticausatives in German (± reflexive morphology; compare e.g. (34a) to (34b) below). These morphological differences do not influence the distribution of the PPs. (34) a. Die the b. Die the

Vase vase Tür door

zerbrach *vonPeter / *mit dem Hammer broke *by Peter / *with the hammer öffnete sich *von Peter / *mit dem Schlüssel opened REFL by Peter / with the key

The properties of anticausatives crosslinguistically

197

(35) a. Die Vase zerbrach durch ein Erdbeben the vase broke through an earthquake b. Die Tür öffnete sich durch einen Windstoß the door opened REFL through a blast-of-wind (36) Die Luftqualität im Raum verschlechtert sich the air-quality in-the room worsens REFL durch das Rauchen von Zigaretten massiv. through the smoking of cigarettes severely Finally, the German counterpart of the English by-itself phrase is licensed in anticausatives. (37) a. Die Vase zerbrach von selbst the vase broke by itself b. Die Tür öffnete sich von selbst the door opened REFL by itself 3.3. Greek Again, before we proceed to the discussion of the Greek data, we need to introduce the PPs associated with the thematic roles under discussion. In Greek, agents are introduced by apo (38a), instruments by me (38a), cau7 sers/natural forces are introduced by either apo or me (41) and causing events are introduced by me (42). Since Greek causatives behave similarly to their English and German counterparts, we do not present the relevant data here. Turning to the Greek passive, let us first note that unlike English and German the Greek passive is synthetic and is characterized by the presence of non-active (Nact) morphology. As the following data show, the Greek passive licenses agents and instruments (38a) but not causers (38b) and causing events (38c), and therefore it crucially differs from the English and German passsive (see also Zombolou 2004). (38) a. Ta mallia mu stegnothikan apo tin komotria / the hair my dried-Nact by the hairdresser / me to pistolaki with the hair-dryer ‘My hair was dried by the hairdresser / with the hair dryer’

198 Artemis Alexiadou, Elena Anagnostopoulou and Florian Schäfer b. ?*Ta ruxa stegnothikan apo ton ilio / me ton ilio the clothes dried-Nact by the sun / with the sun ‘The clothes were dried by the sun’ c. ?*Ta ruxa stegnothikan me toaploma ston ilio the clothes dried-Nact with the hanging-up under the sun ‘The clothes were dried by hanging them up under the sun’ Greek anticausatives are like their English and German counterparts in that they do not license agents (39) but do license causers and causing events (cf. (41)–(42); see also Zombolou 2004). Unlike English and German, Greek anticausatives license instruments (cf. (40); but see notes 7 and 8 and section 5 for refinements). As in German, there are two morphologically distinct types of anticausatives in Greek (± active morphology) and this difference does not influence the distribution of the PPs (compare (39a) to (39b), (40a) to (40b), (41a) to (41b) and (42a) to (42b)). However, an issue arises concerning the verbs that form both the passive and the anticausative via nonactive morphology e.g. katastrefo ‘destroy’ (in (39b)) or skizo ‘tear’ (in (40b)); these are ambiguous between the two interpretations.8 For those verbs, modification by an agent PP yields a passive interpretation (see (39b)). (39) a. *Ta mallia mu stegnosan apo tin komotria the hair my dried-Act by the hairdresser ‘*My hair dried by the hairdresser.’ b. (*)To hirografo katastrafike apo tin ipalilo the manuscript destroyed-Nact by the employee ‘*The manuscript destroyed by the employee.’ (40) a. Ta mallia mu stegnosan me to pistolaki9 the hair my dried-Act with the hair-dryer ‘*My hair dried with the hair dryer.’ b. To pani skistike me to psalidi the cloth tore-Nact with the scissors ‘*The clothes tore with the scissors.’ (41) a. Ta ruxa stegnosan apo / me ton ilio the clothes dried-Act by / with the sun ‘*The clothes dried by the sun’ b. To hirografo katastrafike apo / me tin pirkagia the manuscript destroyed-Nact by / with the fire ‘The manuscript got destroyed by the fire’

The properties of anticausatives crosslinguistically

199

(42) a. Ta ruxa stegnosan me to aploma ston ilio10 the clothes dried-Act with the hanging-up under the sun ‘*The clothes dried by hanging them up under the sun’ b. Me tin afksisi tis igrasias to hirografo katastrafike with the rising the humidity-gen the manuscript destroyed-Nact ‘*The manuscript destroyed by the rising of humidity’ Finally, as in the other two languages, the by-itself phrase is also licensed in Greek anticausatives. (43) a. I porta anikse apo moni tis the door opened-Act by alone-sg hers ‘The door opened by itself’ b. To pani skistike apo mono tu the cloth tore-Nact by alone-sg its ‘The cloth tore by itself’ Summarizing, assuming that the grammaticality of from-PPs, durch-PPs and apo/me-PPs points to the presence of an implicit causer in anticausatives, then the difference between passives and anticausatives in English, German and Greek cannot be expressed in terms of the presence (in passives) vs. absence (in anticausatives) of implicit arguments. Moreover, the fact that agents are licensed exclusively in passives and not in anticausatives suggests that the difference between the two constructions has to do with the presence of agentivity only in the former. Furthermore, the observation that the passive in Greek can only be modified by an agent or an instrument (and not a causer/causing event) leads to the conclusion that the implicit argument in Greek passives is an agent and never an unspecified external argument.

4. Crosslinguistic differences in verb and selection restrictions The core of verbs that undergo the causative alternation is stable across languages. There is, however, interesting variation in two domains, namely verb restrictions and selection restrictions. With respect to the first domain, there are verbs that are predicted by L&RH and Reinhart to allow the alternation but don’t in English (and German), while they do in Greek, e.g. destroy and kill: 11

200 Artemis Alexiadou, Elena Anagnostopoulou and Florian Schäfer (44) a. John / the fire / the bomb destroyed the manuscript a’. *The manuscript destroyed b. John / the fire / the bomb killed Mary b’. *Mary killed (45) a. O Petros / i fotia / i vomva katestrepse to paketo Peter / the fire / the bomb destroyed the package b. To paketo katastrafike apo / me tin fotia / me tin vomva the package destroyed-Nact by / with the fire / with the bomb (46) a. O Petros / o sismos / i vomva skotose ti Maria Peter / the earthquake / the bomb killed Mary b. I Maria skotothike apo / me ton sismo / me tin vomva Mary killed-Nact by / with the earthquake / with the bomb With respect to the second domain, certain V+Obj combinations predicted by L&RH and Reinhart to not allow the alternation do in Greek. The Greek vs. English/German contrast is illustrated in (47)–(49). (47) a. He broke his promise / the contract/ the world record. b. *His promise / the contract / the world record broke c. The dressmaker lengthened the skirt / *The skirt lengthened (48) a. Er brach sein Versprechen / den Weltrekord he broke his promise / the world-record b. *Sein Versprechen / der Weltrekord brach his promise / the world-record broke (49) a. O athlitis espase to simvolaio / to pagkosmio record the athlete broke the contract / the world record b. To simvolaio / to pagkosmio record espase the contract / the world record broke-Act Recall that contrasts like the ones in (47) and (48) have been taken to show that when the subject is necessarily an agent anticausativization of verbs normally entering the alternation is impossible (see L&RH 1995: 85–88, 105). But this does not explain why the Greek examples are grammatical.

The properties of anticausatives crosslinguistically

201

5. Towards an account 12 As argued for in the previous sections, derivational analyses of the causative alternation have several drawbacks. First, the crosslinguistic variation in morphological marking found in the alternation does not provide conclusive evidence for either direction. Second, both views do not fare satisfactorily with respect to the issue of verb/selection restrictions within a language and across languages. As pointed out in section 2, the causativization approach has nothing to say on this issue. Moreover, we showed in section 4 that the class of alternating verbs is not stable across languages. The verbs that alternate in English and German form a subset of the verbs alternating in Greek, a variation not expected by the detransitivization approach. Even more importantly, the thematic restriction on the Greek passive discussed in section 3 (i.e. the fact that the implicit subject is necessarily an agent) is in conflict with the assumption made in L&RH (1995) and Reinhart (2000, 2002) that the implicit external argument of alternating verbs can optionally be an agent or a causer. In turn, this suggests that the class of alternating verbs cannot be defined in terms of the nature of the external theta-role (unspecified external theta-role can be suppressed vs. specified external theta-role cannot) arguing against detransitivization. Third, as shown in section 3, the generally accepted view that anticausatives lack an implicit external argument is challenged by PP-modification in the languages under discussion. If we take the grammaticality of fromPPs, durch-PPs and apo/me-PPs to point to the presence of an implicit causer in anticausatives, then the difference between passives and anticausatives cannot be expressed in terms of implicit arguments. The fact that agents are licensed only in passives and not in anticausatives suggests that the difference between the two has to do with agentivity. These considerations lead us to suggest that agentivity and causation should be syntactically represented in terms of distinct functional heads (see also Pylkkänen 2002). In what follows we will outline what we take to be a rough sketch of a solution to the discussed phenomena. Building on and modifying Kratzer (2005), we adopt a syntactic decomposition of change of state verbs into a Voice and a CAUS component, as in (50), which we take to be the core structure of all (i.e., causative, anticausative and passive) change of state verbs: (50) [Voice [ CAUS [ Root ]]]

202 Artemis Alexiadou, Elena Anagnostopoulou and Florian Schäfer CAUS introduces a causal relation between a causing event (the implicit argument of CAUS) and the resultant state denoted by the verbal root + theme.13 Voice is responsible for the introduction of the external argument and bears features relating to agentivity, and manner. Different features of Voice are involved in the formation of causatives, passives and anticausatives.14 The presence of +/-agentive features is responsible for the licensing of agent and causer external arguments in active and passive constructions. Specifically, agentive Voice (Voice [+AG]) licenses agents (and instrumental PPs); non-agentive Voice (Voice [–AG]) licenses causers. If a Voice head is active then the relevant thematic role is realised in its specifier; if it is passive, the relevant thematic role is implicit.15 In anticausatives, in principle there are two options: Voice might be totally absent or realised as Voice [-AG] with an implicit causer argument. We propose that the first option is available in all of the languages under discussion. Where languages differ is the availability of the second option. We expect languages to show the following two patterns of variation: (i) in a languages where the Voice [-AG] head is possible in passives, anticausatives must appear without Voice; (ii) in a language where the passive is necessarily agentive, the Voice [-AG] head is free to be used in an anticausative interpretation. We propose that English and German instantiate pattern (i), while Greek realises pattern (ii) (see Alexiadou and Anagnostopoulou 2004; Embick 2004). On the present view, there is no direction in the causative/anticausative alternation, as none of the two constructions is directly derived from the other. The last component involved in the structure in (50) is the verbal root. We propose that roots fall into different classes depending on their encyclopedic semantics (cf. also Bhatt and Embick in progress): (51)
agentive (murder, assassinate)
internally caused (blossom, wilt)16
externally caused (destroy, kill)
cause unspecified (break, open) All of the above roots combine with CAUS. “Internal vs. external causation” categorization of the root influences the combinations of roots with particular types of Voice heads. Languages differ in how they treat externally caused roots. In German and English they form only the passive. In Greek (possibly also Hebrew, cf. Doron 2003) they can form anticausatives.

The properties of anticausatives crosslinguistically

203

As will be proposed below, this depends on the parametric presence or absence of a particular type of Voice head in anticausatives. Predicates like murder are based on roots that are externally caused but also agentive. For this reason they can only appear in the context of Voice[+AG], and hence cannot form anticausatives in any of the languages under discussion.17 Roots like blossom are internally caused and hence combine exclusively with CAUS in all the languages under discussion (i.e. no thematic Voice head can be present), as they cannot be brought about by an external argument. It is crucial to note that they are nevertheless causative. A convincing piece of evidence is provided by modification by PPs, illustrated here with an English example, and similar observations can be made for German and Greek. (52) The flowers wilted from the heat In (52) the PP necessarily introduces an indirect causer.18 We assume that this follows from the encyclopedic meaning of internally caused roots which tells us that properties of the internal argument are highly involved in the bringing about the change of state. Therefore, whenever these roots are combined with causers these can only be interpreted as indirectly facilitating the change of state of the theme. Note that while internally caused verbs do not transitivize they nevertheless can be causativized in the periphrastic causative construction, which expresses an indirect causation (Piñón 2001b). This means that the concept of internal causation does not exclude any type of causativization per se, but only direct causation which is necessarily expressed by the Voice + CAUS combination. Externally caused roots require an external argument and hence the presence of Voice, just like agentive roots. Unlike agentive roots which require Voice [+AG], externally caused roots can also combine with Voice [-AG]. We have pointed out that such roots (i.e. destroy and kill) do form anticausatives in Greek but not in English and German. We propose to account for this variation in terms of the proposal that Voice [-AG] can be present in the anticausative structure in Greek while this is not possible in the other languages. Finally, roots like break and open are unspecified for the type of causation involved (in the languages under discussion). This allows them to show up both with and without an external argument, i.e. they alternate.19

204 Artemis Alexiadou, Elena Anagnostopoulou and Florian Schäfer Turning to the licensing of PPs, we assume that adjunct PPs are licensed by structural layers that contain the relevant semantic features. The decomposition in (50) involves two types of licensing heads, Voice and CAUS, for the PPs under discussion. We propose that passive Voice [+AG] licenses agent and true instrument PPs, and CAUS licenses causative from-, durchand apo-, me- PPs. Note that the from/durch/apo/me phrases are interpreted causatively only in constructions in which CAUS is available. In constructions where such a head is not available (e.g. noun phrases), the prepositions have a different meaning (e.g. temporal, locative, source etc). Passive Voice [-AG] licenses causer by-PPs in English (and causer von-PPs in German). Recall, finally, that the Greek passive differs from its English/German counterpart in that it allows only for agent PPs but not for Ccauser PPs. This suggests that the passive Voice head in Greek necessarily carries the feature [+AG] in the line with the proposal made above. What explains the second difference between Greek and German anticausatives with respect to instrumental PPs (licensed in Greek but not in German)? We believe that this has to do with the distinction between ‘pure instrument’ and ‘instrument-causer’ proposed in Kamp and Rossdeutscher (1994), see also the discussion in L&RH (2005: 147 and references therein). Pure instruments presuppose either agentivity or volition both located in Voice, which is not present in anticausatives. They are licensed by Voice [+AG] obligatorily. Instrument-causers are ‘Instruments which can be conceived as acting on their own, once the agent has applied or introduced them’ (from Kamp and Rossdeutscher (1994: 144). They are licensed by CAUS. The German preposition mit introduces exclusively instruments, which we take as evidence that they are associated with the pure instrument role. Therefore mit-PPs are compatible only with Voice [+AG]. The Greek preposition me introduces instruments but also causers and causing events, which we take as evidence that they are associated with the instrument-causer role. Therefore me-PPs are also compatible with CAUS.20 Finally, the by itself phrase asserts that there is no external argument that is responsible for the bringing about of the event hence modifying a CAUS head in structures that do not have an (implicit) external argument. This phrase is most comfortably used with verbal roots that are unspecified for causation, e.g. break, open, asserting that they are brought about without external causation. With predicates that are agentive or necessarily externally caused, they are ruled out because they lead to a contradiction,21 and with predicates that are internally caused they are marginal because they are redundant. We argued that Greek can have anticausatives with destroy,

The properties of anticausatives crosslinguistically

205

kill. On the basis of our interpretation of the by-itself test, we predict that anticausatives of externally caused roots are not compatible with by-itself. The prediction is borne out: (53) To paketo katastrafike (*apo mono tu) the parcel destroyed-Nact by alone its ‘*The parcel destroyed by itself’ 6. Summary In this paper we developed an approach towards the causative/anticausative alternation which builds on a syntactic decomposition of verbs into a Voice and a CAUS component. We argued that the crosslinguistic variation in passives and anticausatives depends on the type and the availability of all or a subset of the above heads across languages. In particular, this variation depends on different types of Voice heads, a CAUS head and four ontological types of roots and various possible/impossible combinations thereof. The distribution of PPs in passives and anticausatives of different languages provides evidence for the presence of these heads.

Acknowledgements Versions of this paper were presented at the XXXI Incontro di Grammatica Generativa in Rome (February 2005), at the 28th GLOW Colloquium in Geneva (April 2005), at the Linguistics Seminar at the University of Venice (May 2005), at the CGSW 20 at the University of Tilburg (June 2005) and at the 36th NELS conference at the University of Massachusetts at Amherst (October 2005). We would like to thank these audiences and an anonymous reviewer for comments and discussion. Notes 1. Some assume that the implicit argument is present in the lexical syntactic representation of the verb, i.e. its argument structure (e.g. Levin and Rappaport Hovav 1995), some that it is present in the semantic representation of verbs (e.g. Reinhart 2002) and others that the implicit argument is even realized in the syntax (e.g. Baker, Johnson and Roberts 1989; Kratzer 1994).

206 Artemis Alexiadou, Elena Anagnostopoulou and Florian Schäfer 2. In Reinhart’s (2002) theta-system, agent is positively specified for the feature m (mental state), i.e. agent is [+c+m]. Instrument is [+c-m]. Its presence implicates the existence of an agent due to a lexical generalization. Cause is characterized as [+c], which makes it consistent with the [+c+m] and the [+c-m] construal (capturing the generalization that verbs selecting for cause arguments also select instruments or agents). 3. Some of the following arguments have already been brought up in Doron (2003). 4. The causing event can sometimes/for some speakers be introduced by ‘through’ instead of ‘from’: (i) a. John's smoking ((of) cheap cigars) worsened the air quality in the room
b. ?The air quality worsened through John's smoking 5. Many German speakers allow both von and durch to introduce natural forces. 6. German does not allow instruments in subject position but only ‘instrumentcausers’ (“Instruments which can be conceived as acting on their own, once the agent has applied or introduced them”, cf. Kamp and Rossdeutscher 1994: 144): (i) a. Der Arzt heilt den Patienten mit der Kamille / dem Skalpell the doctor cures the patient with the camomile / the scalpel b. Die Kamille / *Das Skalpell heilt den Patienten the camomile / *the scalpel cured the patient In the example (ii) below “the hammer” is ungrammatical unless it is contextually construed as an instrument-causer: (ii) (?*)Der Hammer zerbrach die Vase the hammer broke the vase 7. Choice of apo vs. me seems to correlate with “direct” vs. “indirect” causation (Bittner 1999; Kratzer 2005). In contexts where the causal relation between the causer and the change of state is semantically indirect (the causal chain includes intermediate causes) me is favored and apo is dispreferred (in examples (a, b) below apo is licensed only in a temporal interpretation corresponding to since): (i) a. I times afksithikan me tin krisi tu petreleu / ??apo tin krisi tu petreleu ‘The prises increased with the petrol crisis / by the petrol crisis’ b. I dimosia sinkinonia alakse me tus Olimbiakus agones / ??apo tus Olimbiakus agones ‘Public transportation changed with the Olympic games / by the Olympic games’ 8. Note that katastrefo ‘destroy’ forms anticausatives in Greek, unlike English and German. See section 4 for extensive discussion. 9. Not all anticausatives allow instruments, as shown in (i-ii): (i) *O tixos asprise me to pinelo the wall whitened with the paint-brush

The properties of anticausatives crosslinguistically

207

(ii) *To psigio ksepagose me to maxeri the refrigerator defroze with a knife It seems that instruments are licensed when they can surface as subjects in the corresponding transitives (compare the well-formed transitive (iiia) to its wellformed counterpart (40a) and the ill-formed (iiib) to (i)) and they are not licensed when they cannot be subjects of transitives: (iii) a. To pistolaki stegnose ta mallia the hair-dryer dried the hair b. *To pinelo asprise ton tixo the paint-brush whitened the wall The above seems to relate to the distinction between instruments and instrument-causers in German; see note 7. 10. It seems that sometimes apo can introduce a causing event when this is understood as a “direct causer”, as in (ia). This is possible only when the causing event is expressed though a process nominal, not when it is expressed through a nominalized clause; see (ib): (i) a. I porta espase apo to apotomo klisimo the door broke by the abrupt closing b. Me / *apo to na kliso apotoma tin porta tin espasa with / *by the SUBJ-close-1sg abruptly the door Cl-acc broke-1sg ‘I broke the door by closing it abruptly’

11. 12. 13.

14.

15.

The contrast between (ia) and (ib) as well as the contrast between (ia) and the examples in (42) are left to further research. As pointed out by Reinhart (2002), ‘destroy’ has an unaccusative variant in Hebrew (neheras) and French (se-detruire). Special thanks to Hagit Borer, Edit Doron, Hans Kamp, Gillian Ramchand and Peter Svenonius for their insightful comments on earlier versions of this part. On this view, the postulation of a BECOME operator in the structural representations of anticausative (and causative) predicates (Dowty 1979) becomes superfluous (see Kratzer 2005 for discussion). Note that CAUS could also simply be seen as an eventive v of the type proposed in Marantz (2005), if we can ensure that this head can license causative PPs. In this case the causative semantics would not be directly encoded on any verbal head but results from the combination of an activity v and its stative complement (see Ramchand 2003 for related ideas). Cf. Kallulli (2005) for a proposal, according to which it is the presence of features such as [+intent] and [+caus] on v that distinguish between the agentive vs. causative interpretation of the external argument. We remain agnostic with respect to the specific syntactic implementation of implicit arguments, i.e. whether they are present in the form of a covert pronoun or just in terms of features on Voice.

208 Artemis Alexiadou, Elena Anagnostopoulou and Florian Schäfer 16. For us, unergative predictes are not causatives, and hence cannot be classified as internally caused, contra L&RH (1995) and in line with Marantz (1997). 17. Evidence for this comes from the fact that agentive roots are not licensed in derivational processes which cannot include Voice, for example adjectival passives in German (see Anagnostopoulou 2003). 18. This is supported by the fact that in the Greek counterpart of (52) only me is acceptable and not apo (see note 7). 19. There is a complication, though. Predicates like break the world record are necessarily agentive, hence they should behave like murder. It seems that the system deals with agentivity specified on the root (murder) and agentivity on the VP level (V+Obj combination, break the world record) differently. We leave this for further research. 20. Note that in German instrument-causers can also be introduced by durch in anticausatives (cf. note 6): (i) Der Wunde heilte durch die Medizin the wound cured through the medicine ‘*The wound got cured through the medicine’ 21. It was pointed out to us by an anonymous reviewer that there is a grammaticality contrast between (ia) and (ib): (i) a. The window broke by itself b. *The window was broken by itself This is captured by our interpretation of the by itself phrase, which is in conflict with the presence of the implicit agent in passives. Tom Roeper (p.c.) noted examples where the by itself phrase can appear in the passive as in e.g. The FBI was investigated by itself. In this case the by itself phrase modifies the implicit agent. This option is presumably related to the fact that the NP FBI is a group noun so that we interpret this sentence as involving parts of the FBI investigating other parts of the organisation.

References Alexiadou, Artemis and Elena Anagnostopoulou 2004 Voice morphology in the causative-inchoative alternation: Evidence for a non unified structural analysis of unaccusatives. In The Unaccusativity Puzzle: Explorations of the Syntax-Lexicon Interface, Artemis Alexiadou, Elena Anagnostopoulou and Martin Everaert (eds.), 115– 136. Oxford: Oxford University Press. Anagnostopoulou, Elena 2003 Participles and Voice. In Perfect Explorations, Artemis Alexiadou, Monika Rathert and Arnim von Stechow (eds.), 1–36. Berlin /New York: Mouton de Gruyter.

The properties of anticausatives crosslinguistically

209

Baker, Mark, Kyle Johnson and Ian Roberts 1989 Passive arguments raised. Linguistic Inquiry 20 (2): 219–252. Bhatt, Rajesh and David Embick In prep. Causative derivations in Hindi. Ms., University of Massachusetts at Amherst and University of Pennsylvania. Bittner, Maria 1999 Concealed causatives. Natural Language Semantics 7 (1): 1–78. Brousseau, Anne-Marie and Elisabeth Ritter 1991 A non-unified analysis of agentive verbs. In Proceedings of the 10th West Coast Conference on Formal Linguistics, Dawn Bates (ed.), 53–64. Stanford: CSLI Publications. Chierchia, Gennaro 1989 A semantics for unaccusatives and its syntactic consequences. Ms., Cornell University, Ithaca, New York. 2004 A semantics for unaccusatives and its syntactic consequences. In The Unaccusativity Puzzle: Explorations of the Syntax-Lexicon Interface, Artemis Alexiadou, Elena Anagnostopoulou and Martin Everaert (eds.), 22–59. Oxford: Oxford University Press. DeLancey, Scott 1984 Notes on agentivity and causation. Studies in Language 8: 181–213. Doron, Edit 2003 Agency and Voice: The semantics of the Semitic templates. Natural Language Semantics 11 (1): 1–67. Dowty, David 1979 Word meaning and Montague Grammar – The Semantics of Verbs and Times in Generative Semantics and in Montague's PTQ. Dordrecht: Reidel. Embick, David 2004 Unaccusative syntax and verbal alternations. In The Unaccusativity Puzzle: Explorations of the Syntax-Lexicon Interface, Artemis Alexiadou, Elena Anagnostopoulou and Martin Everaert (eds.), 137–158. Oxford: Oxford University Press. Haspelmath, Martin 1993 More on the typology of inchoative/causative verb alternations. In Causatives and Transitivity, Bernard Comrie and Maria Polinsky (eds.), 87–120. Amsterdam: John Benjamins. Jaeggli, Oswaldo 1986 Passive. Linguistic Inquiry 17: 587–622. Kallulli, Dalina 2005 Unaccusatives with overt causers and experiencers: A unified account. To appear in Datives and Similar Cases, Daniel Hole and André Meinunger (eds.). Amsterdam: John Benjamins.

210 Artemis Alexiadou, Elena Anagnostopoulou and Florian Schäfer Kamp, Hans and Antje Rossdeutscher 1994 Remarks on lexical structure and DRS construction. Theoretical Linguistics 20 (1): 97–164. Kratzer, Angelika 1994 The event argument and the semantics of voice. Ms., University of Massachusetts, Amherst. 2005 Building resultatives. In Event Arguments in Syntax, Semantics, and Discourse, Claudia Maienborn and Angelika Wöllstein-Leisten (eds.), 178–212. Tübingen: Niemeyer. Lakoff, George 1968 Some verbs of change and causation. In Mathematical Linguistics and Automatic Translation, Report NSF-20, Susumu Kuno (ed.). Cambridge, MA: Aiken Computation Laboratory, Harvard University. 1970 Irregularities in Syntax. New York: Holt, Rinehart and Winston. Levin, Beth 1993 English Verb Classes and Alternations: A Preliminary Investigation. Chicago: University of Chicago Press. Levin, Beth and Malka Rappaport Hovav 1995 Unaccusativity. At the Syntax-Lexical Semantics Interface. Cambridge, MA: MIT Press. 2005 Argument Realisation. Cambridge: Cambridge University Press. Manzini, Rita 1983 On Control and Control Theory. Linguistic Inquiry 14 (3): 421–446. Marantz, Alec 1984 On the Nature of Grammatical Relations. Cambridge, MA: MIT Press. 1997 No escape from syntax: Don’t try morphological analysis in the privacy of your own lexicon. In University of Pennsylvania Working Papers in Linguistics, Vol. 4.2, Alexis Dimitriadis, Laura Siegel, Clarissa Surek-Clark and Alexander Williams (eds.), 201–225. Philadelphia. 2005 Objects out of the lexicon: objects as event. Talk presented at the University of Vienna. Pesetsky, David 1995 Zero Syntax. Cambridge, MA: MIT Press. Pylkkänen, Liina 2002 Introducing Arguments. Ph.D. diss., Department of Linguistics, MIT. Piñón, Christopher 2001a A finer look at the causative-inchoative alternation. Abstract for Semantics and Linguistic Theory 11. 2001b A finer look at the causative-inchoative alternation. In Procedings of Semantics and Linguistic Theory 11, Rachel Hastings, Brendan Jackson and Zsofia Zvolenszky (eds.). Ithaca, NY: CLC Publications.

The properties of anticausatives crosslinguistically

211

Ramchand, Gillian 2003 First phase syntax. Ms., University of Tromsø. Reinhart, Tanya 2000 The Theta System: Syntactic realization of verbal concepts, OTS Working papers, 00.01/TL. 2002 The Theta System – An overview. Theoretical Linguistics 28 (3): 229–290. Roeper, Tom 1987 Implicit arguments and the head-complement relation. Linguistic Inquiry 18 (2): 267–310. Smith, Carlotta 1970 Jespersen’s ‘move and change’ class and causative verbs. In English. Linguistic and Literary Studies in Honor of Archibald A. Hill, Vol. 2: Descriptive Linguistics, Mohammad Ali Jazayery, Edgar Polomé and Werner Winter (eds.), 101–109. The Hague: Mouton. Williams, Edwin 1981 Argument structure and morphology. The Linguistic Review 1 (1): 81–114. Zombolou, Katerina 2004 Verbal alternations in Greek: a semantic approach. Ph.D. diss., Department of Linguistics, University of Reading.

Number agreement and event pluralization: A case study Valentina Bianchi

1. Introduction In the recent minimalist framework, the number feature is taken to be interpretable when it is specified on a nominal category, but not when it is specified on a verbal category. This is actually a twofold assumption. On the one hand, the number feature on a verbal category is unvalued, and it must get a value by means of an Agree relation with a nominal category endowed with an interpretable (valued) number feature. On the other hand, even after it has got a value, the verbal number feature remains semantically irrelevant, i.e. it is not visible at the interface with the so-called Conceptual-Intentional system(s). In this paper I wish to suggest that verbal number agreement is semantically relevant at least in a well-defined set of cases, and that its semantic import is event pluralization. The evidence that I will be discussing involves the Italian reciprocal modifier uno dopo l’altro (lit. ‘one after the other’), exemplified in (1): (1)

I soldati spararono uno dopo l’altro. the soldiers shoot- PAST.3PL one after the other ‘The soldiers shot one after another.’

This modifier contains a form of the reciprocal expression (l)’uno…l’altro,1 and it is interpretively related to an antecedent (the subject in (1)), with which it agrees in gender. Roughly speaking, the modifier expresses a temporal serialization of a set of events: in (1), there is a series of shooting events, each one having a different soldier as its Agent. This multiplicity of events will play a crucial role in my proposal. As a first step, I will point out three generalizations on the distribution of the reciprocal modifier.

214 Valentina Bianchi 2.

Three generalizations

2.1. Syntactic plurality of the antecedent Current approaches to reciprocal expressions require a semantically plural antecedent for them, due to the inherent distributivity of the reciprocal interpretation (see, among many others, Heim, Lasnik and May 1991; Sternefeld 1998; Beck 2001). However, it can be shown that the reciprocal modifier requires an antecedent that is not only semantically, but also syntactically plural. The relevant evidence comes from certain quantifiers featuring a mismatch between syntactic and semantic number. A clear example is the quantifier più di un NP ‘more than one NP’, which is semantically plural but syntactically singular (Partee 1995): it cannot antecede the reciprocal modifier, contrary to the syntactically and semantically plural quantifier almeno due NP ‘at least two NPs’: (2) a. *Più di un soldato sparò uno dopo l’altro. more than one soldier shoot- PAST.3SG one after the other b. Almeno due soldati spararono uno dopo l’altro. at least two soldiers shoot- PAST.3PL one after the other ‘At least two soldiers shot one after the other.’ Another example is the quantifier la maggior parte dei NP (‘most NP’, lit. ‘the major part of the NPs’), which for some speakers can trigger either singular or plural verb agreement. For these speakers, the reciprocal modifier is incompatible with singular agreement, compare (3a) to (3b):2 (3)

a. La maggior parte dei soldati hanno sparato uno dopo l’altro. the major part of the soldiers have-3PL shot one after the other ‘Most soldiers shot one after another.’ b. *La maggior parte dei soldati ha sparato uno dopo l’altro. the major part of the soldiers has-3SG shot one after the other

A third example is the quantifier phrase qualche NP ‘some NP-SG’, discussed by Zamparelli (2004): though syntactically singular, it is semantically compatible either with a singular or with a plural interpretation – depending on the context – and allows for plural inter-sentential anaphora. However, its potential semantic plurality is not sufficient to license the reciprocal modifier, as shown in (4).

Number agreement and event pluralization: A case study 215

(4) * Qualche soldato ha sparato uno dopo l’altro. one after the other some soldier has-3SG shot Finally, the relative pronoun chi ‘who’ in free relatives can receive a quasiuniversal interpretation, and it supports a reciprocal interpretation for the si clitic in (5a); however, it is syntactically singular, and it cannot antecede the reciprocal modifier in (5b):3 (5) a. Chi si ama si perdona. who SI love-3SG SI forgive-3SG ‘Those who love each other forgive each other.’ b. *Chi arriva uno dopo l’altro si mette in fila. who arrive-3SG one after the other SI stand-3SG in a row These data suggest that syntactic plurality of the antecedent is a necessary condition to license the reciprocal modifier.4

2.2. Grammatical role of the antecedent A second constraint on the reciprocal modifier concerns the possible grammatical role of the antecedent. In all of the above examples, the antecedent is the subject of the sentence. The antecedent can also be a direct object, as shown in (6): (6)

Maria ha ingoiato tre panini uno dopo l’altro. Maria has swallowed three sandwiches one after the other ‘Maria swallowed three sandwiches one after another.’

But the antecedent cannot be an indirect object or another prepositional object: (7)

a. ?*Ho telefonato ai miei colleghi uno dopo l’altro. have-1SG phoned to my colleagues one after the other b. ?*Ho parlato con gli studenti uno dopo l’altro. have-1SG spoken with the students one after the other

This constraint cannot be derived from the anaphoric nature of the reciprocal expressions, for anaphors in general can be bound by prepositional comple-

216 Valentina Bianchi ments (see Belletti 1982: 117–118 on this point). Belletti (1982) reduces this constraint to her movement analysis, whereby the first part of the reciprocal expression (l’uno) moves and adjoins to the antecedent in LF: if the antecedent is embedded in a PP, the moved part does not c-command the empty category left by movement, violating the Proper Binding Condition. (In current terms, raised (l’)uno could not countercyclically adjoin to a DP that is embedded in a PP.)5 (8)

ho telefonato [PP a [DP (l’)uno [i miei colleghi]] [l’uno [dopo l’altro]]

: # z-----------m

One problem with this idea is that it does not extend to English, where the reciprocal each other can be anteceded by a prepositional complement, although each-movement is standardly assumed.6 Anyway, the problem with c-command does not even arise in (7) if one adopts Kayne’s (2003) approach to prepositions as probes, attracting their purported complement to a specifier position c-commanding the base VP-internal position.7 I wish to suggest an alternative view of (7), capitalizing on the observation that in Italian, indirect (and generally prepositional) objects can never trigger number agreement on the verb. Note that subjects regularly trigger finite verb agreement, which involves a verbal functional projection (T or Agr-S). As for direct objects, in contemporary Standard Italian they only trigger a residual form of past participle agreement when they are realised as clitics;8 however, under standard assumptions direct objects too establish an Agree relation with a low verbal head endowed with phi-features (little v or Agr-O), in order to license/value their structural Accusative Case feature. On the contrary, indirect and prepositional objects can never trigger participle agreement – strikingly, not even when they are morphologically identical to the Accusative clitics, as in the first and second person (compare (9) to (10)): (9)

Ci hanno invitate. 1PL have-3PL invited-F.PL ‘They invited us’ (female speaker and addressee(s))

(10) *Ci hanno parlate. 1PL have-3PL spoken-F.PL ‘They spoke to us’ (female speaker and addressee(s))

Number agreement and event pluralization: A case study 217

Even if the indirect object is assumed to bear a structural Dative Case (see Kayne 2003), it is far from clear that this Case licensing involves agreement for phi-features; on the contrary, it appears that the presence of the preposition somehow blocks agreement (see Belletti and Rizzi 1995 for discussion).9 I wish to propose that this is the crucial property that accounts for (7): indirect/prepositional objects, contrary to subjects and direct objects, do not establish a number agreement relation with any functional head associated to the verb. There is also some further evidence for the relevance of number agreement in licensing the reciprocal modifier. As discussed by Cinque (1988: 554–558), in transitive clauses involving the impersonal clitic si the verb usually agrees in number with the thematic object (11a), but it may sometimes fail to agree with it, as an (idiolectally restricted) option (11b): (11) a. Da qui si vedono le montagne. from here SI see-3PL the mountains b. %Da qui si vede le montagne. from here SI see-3SG the mountains ‘One can see the mountains from here.’ Although the judgements are delicate, a correlation seems to emerge: when the reciprocal modifier is anteceded by the impersonal subject, the nonagreeing version is preferred (for the speakers who accept this option). (12) a. ?*Per fare l’effetto-onda, si alzano le mani uno dopo l’altro. to make the wave-effect SI raise-3PL the hands one after the other b. ?Per fare l’effetto-onda, si alza le mani uno dopo l’altro. to make the wave-effect SI raise-3SG the hands one after the other ‘To make the wave-effect, people raise their hands one after another.’ In (12a), number agreement with the thematic object seems prevent the impersonal subject from anteceding the reciprocal modifier.10 An obvious objection is that even in (12b), the impersonal subject does not seem to trigger plural number agreement on the finite verb, which appears in a default (third person singular) form. Note however that the impersonal subject does trigger plural agreement on participial and adjectival predicates, as shown in (13):

218 Valentina Bianchi (13) Si viene criticati. SI get-3SG criticized-M.PL ‘One gets criticised.’ I will return below to this interesting split in number agreement. On the basis of these data, I wish to propose the following hypothesis: in order to license the reciprocal modifier uno dopo l’altro, the antecedent must trigger number agreement on the modified predicate.

2.3. Aktionsart constraints A third constraint on the distribution of the reciprocal modifier concerns the nature of the modified predicate. In all of the above examples, the predicate is stage-level. The reciprocal modifier is impossible if the modified predicate is individual-level (14), or more generally stative (15): (14) ?*I miei antenati sono stati alti e biondi uno dopo l’altro. my ancestors are-3PL been-M.PL tall and blonde one after the other (15) a. ?*I soldati restarono immobili uno dopo l’altro. the soldiers remained-3PL still one after the other b. *I soldati erano spaventati uno dopo l’altro. one after the other the soldiers were-IPF.3PL afraid There is an even stricter Aktionsart constraint: durative atelic predicates, as in (16), are much less felicitous than a telic predicate like ‘shoot’: (16) a. ?I soldati camminarono uno dopo l’altro. the soldiers walked-3PL one after the other b. ?I ragazzi dormirono uno dopo l’altro. the boys slept-3PL one after the other c. *I ragazzi russarono uno dopo l’altro. the boys snored-3PL one after the other (16a–b) are marginally acceptable only if the atelic predicates receive an inchoative interpretation (denoting the inception of the event) or if they denote contextually bounded events (e.g. a walk along a well defined path,

Number agreement and event pluralization: A case study 219

or a fixed sleeping interval); no such interpretation is available in the case of (16c). This suggest the requirement for an inherently bounded event. Bach (1986) proposed an interesting parallel between the mass-count distinction in the nominal domain and the event-process distinction in the “aspectual” (Aktionsart) domain. Under his view, telic predicates denote atomic eventualities (events), with the property of indivisivity (no proper part of a telic event can be a telic event falling under the same predicate). Atomic events can be i-joined (à la Link 1983) to form plural events; they allow for count modifiers (17a), but not for mass modifiers (17b): (17) a. John fell asleep three times last night. b. * John found a unicorn a lot last night. On the contrary, stative – and generally, atelic – predicates denote “mass”, non-atomic eventualities, with the property of divisivity (a proper part of an atelic eventuality can be an atelic eventuality falling under the same predicate; e.g. a proper part of a walk can be a walk). Atelic eventualities cannot be i-joined to form a plural eventuality; they do not allow for count modifiers (18a), but only for mass modifiers (18b): (18) a. * John slept three times last night. b. John slept a lot last night. Capitalizing on this parallelism, we can draw the following conclusion: the reciprocal modifier uno dopo l’altro can only be licensed in a clause that contains an event variable ranging over atomic (countable) events.

2.4. Summary To summarize the preceding discussion, I have proposed the following three generalizations on the distribution of the reciprocal modifier: (i) there must be a syntactically and semantically plural antecedent in the clause; (ii) the modified predicate must agree in number with the antecedent; (iii) the modified predicate must denote atomic events. The idea that I will explore is that via syntactic number agreement, the plural feature of the antecedent can be transmitted to the predicate, so that the

220 Valentina Bianchi latter comes to denote a plural event, i.e. a sum of individual events that can be distributed over and be temporally ordered one after the other.

3. Verbal plurality Starting from Link’s (1983) seminal work, pluralization has been conceived of as an operation (dubbed *) that applies to the denotation of a count noun (a set of individuals) to yield a structured domain constituted by all the possible sums of these individuals into pluralities.11 Recently, Heycock and Zamparelli (2004) have proposed a syntactic implementation of the pluralization operation as a functional head (dubbed Pl) selecting the NP projection, and taking the denotation of the latter as its argument: (19) [[ [NP soldato] ]] = { {al}, {ben}, {carl} } (20) [[ [PlP soldati] ]] = [[ Pl ]] ([[ [NP soldato] ]] ) = { {a}, {b}, {c}, {a,b}, {a,c}, {b,c}, {a,b,c} } I propose that a parallel Pl head selects the verb phrase: (21)

PlP

rp Pl

#

[plural]

vP

rp

DP

#

Carl

v’

3

v

VP

# shot Under current assumptions, the vP denotes (the characteristic function of) a set of shooting events whose Agent is Carl (see e.g. Kratzer 1996). If the Pl head has a [plural] feature, the PlP will denote a set of plural shooting events whose Agent is Carl (e.g. in Carl shot three times). Consider however our initial example, repeated here as (22): (22) I soldati spararono uno dopo l’altro. the soldiers shoot- PAST.3PL one after the other ‘The soldiers shot one after another.’

Number agreement and event pluralization: A case study 221

Here we do not want to obtain a set of plural events each of which has all the soldiers (i.e. the whole DP denotation) as the Agent, but rather, a plurality of events each of which has an individual part of the denotation of [[ [DP the soldiers] ]] as its Agent. This can be obtained by pluralizing a relation between individuals and events via the double star ** operator (Sternefeld 1998; Beck 2001): (23) ** is that function such that for any two place relation R, x, y:[**R](y)(x) iff (i) R(y)(x), or (ii)
x1x2y1y2 [x=(x1x2) & y=(y1y2) & **R(y1)(x1) & **R(y2)(x2)] (where x1x2 is the smallest x such that x1
x & x2
x) Suppose that we take the vP in (22) to denote a two-place relation between individuals and events, rather than a set of events: (24) x.e.(shoot(x,e)) (24) is a relation holding between an individual x and an event e if and only if x is the Agent of e and e is a shooting event. Suppose that the Pl head above vP encodes the ** operator, applying to this two-place relation between individuals and events. The double-starred relation will hold between an overall plural shooting event E and the soldiers set denoted by the subject DP if and only if for every individual part of the soldiers set there is an atomic shooting event of which he is the Agent, and vice versa: (25) <E, [[ [DP the soldiers] ]] >
**x.e.(shoot(x,e))



(26) (x
i [[ [DP the soldiers] ]] ) (
e
i E) (shoot(x,e)) & (e
i E) (
x
i [[ [DP the soldiers] ]] ) (shoot(x,e)) (where
i is the individual part relation) 12 In order to obtain the two-place vP denotation (24), I propose that a Pl head encoding the ** operator attracts a [+plural] DP argument to its Spec, in order to check/value its own [+plural] feature. This leaves in the vP an individual variable that can be abstracted over to yield the required two-place relation:

222 Valentina Bianchi (27)

PlP ri DP1 [pl: +] Pl’ ri Pl[pl: +] vP

# **

 ** x1.e.(shoot(x1,e))  x1.e.(shoot(x1,e))

6 t1

Number agreement in the PlP layer pluralizes the individual-event relation: hence, it must be visible at the interface with the C-I system(s). A full compositional analysis of the reciprocal modifier uno dopo l’altro would exceed the limits of the present discussion. An obvious problem is the fact that the syntactic antecedent of the reciprocal is a DP, but semantically, the modifier does not temporally order the individuals in the DP denotation (rather, the atomic subevents that these individuals are involved in).13 I will not try to develop a semantic analysis; on the syntactic side, I will simply assume that the reciprocal modifier is hosted in a functional projection immediately above the [+plural] PlP layer. Having proposed a syntactic implementation of verbal plurality, I can now go back to the three generalizations discussed in §2. First, there must be a syntactically (as well as semantically) [+plural] argument inside the vP, because only such an argument can check/value the [+plural] feature on the verbal Pl head. Second, the [+plural] argument must be able to establish a number agreement relation with the verbal functional head Pl, in order to license the [+plural] feature of the latter, in the way described in (27). This is possible for a DP argument (a subject or a direct object); on the contrary, a prepositional argument cannot establish such an agreement relation, for reasons that remain to be understood. Third, the modified predicate must be eventive and telic because only such a predicate has a Davidsonian event position ranging over atomic events, which allow for pluralization (giving rise to sums of atomic events).14

4. The syntactic position of the reciprocal modifier In (21) and (27), I have proposed a very low position for the Pl projection – in the immediate periphery of the verb phrase – essentially for reasons of semantic compositionality: just as in the case of the noun phrase (see (20)), here too the Pl head applies to the lexical layer in order to yield a plural

Number agreement and event pluralization: A case study 223

denotation. Zamparelli (2000) and Heycock and Zamparelli (2004) argue that semantic compositionality is a valid criterion to establish the relative order of functional layers, a view that I fully endorse. The criterion of semantic compositionality can also be used to establish the position of the proposed PlP layer with respect to some higher layers. Consider for instance the following examples: (28) a. Quando entrava il preside, gli studenti si alzavano when came-in-IPF.3SG the dean the students SI stood-IPF.3PL uno dopo l’altro. one after the other ‘When the dean came in, the students used to stand up one after another.’ b. Maria stava divorando i miei panini uno dopo l’altro. Maria was-IPF.3SG swallowing my sandwiches one after the other ‘Maria was swallowing my sandwiches one after another.’ (28a) is a habitual sentence, expressing a quasi-universal quantification over events (cf. among others Bonomi 1997). Note that this quasi-universal quantification affects the overall plural event in which all the students stand up one after the other: this means that the plural event is available for aspect to operate on. Similarly, in (28b) it is the overall plural event that is seen in its progress: for the sentence to be true, there must be a non-null sequence of individual events (each involving a single sandwich) preceding the focalization point of the progressive aspect. These considerations suggest that the Pl projection and the reciprocal modifier are placed below the aspectual projection Asp(generic/progressive) proposed by Cinque (1999): this confirms the low syntactic position proposed above. As an anonymous reviewer pointed out, the position of the PlP projection in the immediate periphery of the verb phrase suggests that it might be assimilated to an “aspectual” projection in the sense of Borer (2005), i.e. one of the projections that build up the event structure. This possibility exceeds the limits of the present discussion; for my present purposes, the crucial point is that this projection must be sensitive to the number feature of the argument that it hosts. I will retain the label ‘PlP’ in order to stress the potential parallelism with the noun phrase functional structure, and leave the suggested assimilation for future research. Besides these compositionality arguments, it is also useful to take a closer look at the syntactic positions in which the reciprocal modifier can

224 Valentina Bianchi appear.15 I am making the crucial assumption that the reciprocal modifier uno dopo l’altro sits in a functional layer immediately adjacent to PlP; under this assumption, the syntactic evidence that I am going to discuss will also indirectly bear on the syntactic position of PlP. The unmarked position of the modifier follows all the verb complements and the lowest adverbs of Cinque’s (1999) hierarchy. However, this rightmost linear position would be compatible with a relatively high structural position if we assume that the reciprocal modifier can be right-adjoined. If we exclude this possibility, it is still possible that the modifier sit in a lefthand specifier and be crossed over by remnant VP movement (à la Kayne 2003). Thus, the rightmost position is not really informative w.r.t. the hierarchical placement of the modifier. More interestingly, in compound tenses the reciprocal modifier can appear in an intermediate clausal position. Here we find a striking contrast: a reciprocal modifier anteceded by a transitive subject can marginally appear in between the auxiliary and the participle (29a); this position is instead unavailable for a modifier anteceded by the direct object (29b): (29) a. ?Gli studenti hanno [uno dopo l’altro] eseguito the students have-3PL one after the other played il brano musicale. the piece of music ‘The students played the piece of music one after another.’ b. *Mio cugino ha [uno dopo l’altro] eseguito my cousin has-3SG one after the other played tutti i brani musicali. all the pieces of music We can account for this contrast in terms of head-to-head movement of the participle (cf. Cinque 1999). Suppose that there is a higher position for the subject-related modifier and a lower position for the object-related one: the past participle necessarily moves beyond the lower layer (as indicated by the solid arrow in (30)), whereas movement across the higher layer is optional (as indicated by the dotted arrow): (30) X° …[ MODS [ Y° [MODO … [vP Vparticiple …]

: #: # z-------mz--------m

There is some more evidence in support of two functional layers hosting the reciprocal modifier. The modifier can appear in a clause-initial position:

Number agreement and event pluralization: A case study 225

(31) a. Uno dopo l’altro, i soldati abbassarono il fucile. one after the other the soldiers laid-down- PAST.3PL the gun ‘The soldiers laid down their guns one after another.’ b. ?Uno dopo l’altro, ha ingoiato tutti i panini. one aftert he other has-3SG swallowed all the sandwiches ‘(S)he swallowed all the sandwiches one after another.’ I assume that this position is derived by some sort of adverbial fronting. Interestingly, it is also possible to have two co-occurring reciprocal modifiers, one in the clause-initial position and another one in the clausefinal position, each having a distinct antecedent: (32) Uno dopo l’altro, i ragazzi alzano le due bandiere one-M after the other-M the boys.M lift-3PL the two flags-F una dopo l’altra. one-F after the other-F ‘One by one, the boys lift each his two flags one after the other.’ The possibility of two reciprocal modifiers suggests a recursion of the relevant functional layer, but this is not an unordered recursion. In fact, there is a sharp contrast between (32), where the fronted modifier is anteceded by the subject and the clause-final one by the direct object, and (33), where the modifier anteceded by the direct object is fronted across the one anteceded by the subject: (33) *Una dopo l’altra, i ragazzi alzano le due bandiere one-F after the other-F the boys.M lift-3PL the two flags-F uno dopo l’altro. one-M after the other-M ‘One by one, the boys lift each his two flags one after the other.’ This looks like a Relativized Minimality effect. Assume, as above, that the subject-related modifier is inserted in a higher layer than the object-related one. If the subject-related modifier is fronted, no intervention effects arise: (34) [TopP MODS [IP I ragazziS .. le due bandiereO … [tMOD-S .. MODO VP ]

: # z-------------------m

226 Valentina Bianchi On the contrary, if the lower object-related modifier is fronted, it moves across the higher subject-related one, which creates an intervention effect: (35) *[TopP MODO [IP I ragazziS .. le due bandiereO .. [MODS .. tMOD-O ..]

: # z-----------------------m

The evidence in (29)–(35) suggests two hypotheses: first, there can be recursion of the layer hosting the reciprocal modifier (and of the related PlP layer); second, this recursion is intrinsically ordered: the higher layer is reserved for the subject-related modifier (and event pluralization triggered by a plural subject), whereas the lower layer is reserved for the object-related modifier (and event pluralization triggered by a plural direct object). Although the distinction between a subject-related and an object-related functional layer is reminiscent of the Agr-S vs. Agr-O architecture, I am actually claiming that both these layers are quite low in the functional structure of the clause, in the periphery of the verb phrase. One way to implement the distinction would be to assume that the higher reciprocal modifier and the corresponding PlP projection are placed above vP – the layer which introduces the transitive subject – whereas the lower reciprocal modifier and PlP are placed below little v and above VP, at a structural level which only c-commands the merge position of the direct object, but not that of the subject. Unfortunately, it is quite difficult to find empirical evidence in support of this specific functional structure.

5. Back to interpretability: “high” vs. “low” number The evidence discussed in § 4 supports the following conclusions: there are two distinct functional layers responsible for event pluralization in the periphery of the verb phrase and – whatever their exact position in the functional sequence – the higher one is reserved for the transitive subject, the lower one for the direct object. These conclusions imply an apparently redundant duplication of the number feature in the clausal functional structure: this would be checked/valued by the subject in the vP-peripheral Pl projection, but also in the phi-feature bundle of the functional head that is responsible for Nominative Case licensing (Tense or Agr-S), which is quite high in the functional structure. Why should Universal Grammar allow for such a double checking of the number feature?

Number agreement and event pluralization: A case study 227

From a minimalist perspective, a possible anwer is that vP and CP are distinct phases, and the subject is the distinguished argument that is involved in the computation of both, since it originates within the vP phase but it must be Case-licensed in the higher CP phase. This is, at best, a theoryinternal answer. However, there is also some independent empirical evidence that supports this double representation of number. As is well known from the literature, there are some interesting structures featuring a number agreement mismatch, whereby number agreement on the finite verb form and non-finite (participial/adjectival) number agreement differ, even though syntactically one and the same argument is involved (see, among others, Cinque 1988; Kayne 1989; Wechsler 2002). One relevant example is the impersonal si construction in Italian, exemplified in (13) above and in (37): here the finite verb is in the singular, whereas all non-finite predicates appear in the plural. (37) Si è stati criticati. SI be-3SG been-M.PL criticized-M.PL ‘One got criticised.’ The reverse pattern emerges with the well known politeness plural second person, which is commonly used in French and in earlier stages of Italian to formally address a single interlocutor. Here the finite verb appears in the plural second person, but all non-finite predicates appear in the singular.16 (38) Siete stato dichiarato colpevole. are-2.PL been-M.SG declared-M.SG guilty-SG ‘You were found guilty.’ These mismatches could be reduced to an opposition between grammatical and semantic agreement, roughly in the sense of Wechsler (2002) (see also Pollard and Sag 1994: 60–99). Simplifying considerably, under this view certain instances of morphological agreement are driven by a purely grammatical number feature, whereas others are triggered by a semantic number feature, corresponding to the intrinsic (or pragmatically relevant) properties of the referent, that is, its individuation as an aggregate or nonaggregate entity. In (37), we could assume that the impersonal subject lacks a grammatical [plural] feature, but it is semantically plural; the finite verb undergoes grammatical agreement, whereas the participles and the nominal predicate undergo semantic agreement. Conversely, in (38) the politeness second person subject may be formally [plural], but semantically singular in that it

228 Valentina Bianchi refers to a single addressee; the finite verb undergoes grammatical agreement, and the non-finite predicates semantic agreement. It is fair to say that these agreement mismatches can hardly be expressed by the syntactic mechanism of Agree. The interface question of the relationship between syntactic and semantic number is extremely complex;17 nevertheless, I believe that the data in (37) and (38) show a regularity that we can try to make sense of. The crucial observation is that the distribution of formal vs. semantic number agreement is not random, but it correlates with the property of finiteness. Both in (37) and in (38), finite agreement is formal/syntactic, whereas non-finite agreement is semantic. The next question is, then, what distinguishes finite from non-finite agreement. In Bianchi (2003), it is argued that the crucial difference is the presence vs. absence of the [person] feature. Capitalizing on an insight by Wechsler (2002), we can speculate that finite number agreement is somehow parasitic on person agreement. In (37), the impersonal subject of the si construction plausibly bears a [+plural] feature and a defective person feature: the latter fails to trigger full person agreement, which would in turn entail the impossibility of parasitic number agreement, yielding a default inflection on the finite verb form (cf. Cinque 1988: 537). In (38), instead, the apparent plural number agreement can be though of as a consequence of person agreement with an “augmented” second person, i.e. as a form of honorific agreement.18 This dependency of number agreement can be directly captured in an approach where both person and “high” number are realised on one and the same phi-complete head, T or AgrS. However, Shlonsky (1989), Ritter (1995), and SigurDsson (2004), among many others, have proposed a split projection of person and number in separate functional heads; from this perspective, the dependency should be encoded in a more indirect way, by allowing the feature specification of one head to affect the possible values of the other, in a way that remains to be understood.19 Be this as it may, in both (37) and (38) the finite number agreement that is syntactically expressed in the higher part of the clausal functional structure is parasitic on person and, crucially, blind to semantic number. On the contrary, non-finite number agreement is insensitive to person, whereas it is sensitive to semantic number. We can now assume that non-finite number agreement is the morphological manifestation of the proposed Pl layers in the lower part of the clausal functional structure: the number feature that they express is semantically relevant at the interface with the C-I systems, hence they are sensitive to the semantic plurality of the agreement trigger.

Number agreement and event pluralization: A case study 229

This brings us back to the question that we started with, namely, the interpretability of the number feature in the clausal functional structure. As mentioned above, the number feature that expresses event pluralization must be relatively low, in particular, lower than the functional layers that quantify over (bind) the Davidsonian event variable. On the other hand, the “high” number feature is projected above the aspectual/Tense layers that bind the event variable, and it could not possibly yield event pluralization; thus, it cannot be semantically interpretable in the same way as “low” number is. The different status of the number feature in the two cases is entirely expected from the viewpoint of semantic compositionality.

6. Concluding remarks In this paper I have provided some evidence that verbal number agreement can be semantically relevant and yield event pluralization. The evidence only consists in one specific case study; it is impossible here to draw any more general conclusion, beyond the obvious implication that such possibility is at least allowed by Universal Grammar.20 Despite its admittedly limited empirical scope, this proposal has some interesting implications on the methodological side. Starting from Abney’s (1987) seminal work, it is customary to draw a parallelism between the functional structure of the noun phrase and that of the clause. This usually consists in importing in the noun phrase such clausal categories as Complementizer, Inflection etc. (see Giusti 2006 for recent discussion). In my analysis I have proposed an assimilation in the opposite direction: that is, I have imported a functional head of the noun phrase structure (Pl) in the clause. From this perspective, the functional structure above NP and vP is even more strictly parallel than it is usually thought. I believe that this is entirely reasonable in the light of the parallelism in the structure of the nominal and verbal denotation domains pursued by Bach (1986), Krifka (1992), and others. Secondly, the proposed analysis shows a welcome convergence between syntactic arguments and semantic compositionality arguments in individuating the functional architecture of a phase. To the extent that it is correct, my proposal further corroborates Heycock and Zamparelli’s approach. I believe that both these issues – the parallelism of nominal and clausal projections, and the issue of the semantic compositionality of functional layers – deserve a thorough investigation in the cartographic approach to phase structure.

230 Valentina Bianchi Acknowledgements I wish to thank the audience at the XXXI Incontro di Grammatica Generativa (Rome, February 2005), and in particular Adriana Belletti, Alex Grosu, Luigi Rizzi, Halldór SigurDsson, Roberto Zamparelli, and an anonymous reviewer for insightful comments. I am fully responsible for any remaining errors or inaccuracies.

Notes 1. See Belletti (1982) for general discussion. 2. A possible assumption is that when syntactically singular, the quantifier actually introduces a group denotation, similar to collective NPs like the platoon. If we analyse a group as a special atomic individual (related to a plurality of ordinary individuals: see a.o. Landman 1989; Chierchia 1998), then the subject DP in (3b) would be also semantically non-plural. However, note that the quantifier triggering singular agreement does not necessarily take on a collective interpretation, e.g.: (i) La maggior parte degli studenti ha risolto l’esercizio a casa propria. the major part of the students has-SG solved the exercise at home self ‘Most students correctly solved the exercise at home.’ 3. Thanks to Alex Grosu (p.c.) for pointing out these data. 4. Syntactic plurality is not a sufficient condition without semantic plurality: a syntactically plural argument cannot license the reciprocal modifier when the predicate requires a group interpretation for it: (i) * I soldati si incontrarono nel cortile uno dopo l’altro. the soldiers SI met-PAST.3PL in the yard one after the other Collective noun phrases also cannot antecede the reciprocal modifier in Italian; I take them to be both semantically and syntactically singular (cf. note 2). Pollard and Sag (1994: 70–71) show that in certain varieties of English collectives can trigger plural agreement; the plural vs. singular agreement correspond to an aggregate vs. nonaggregate (atomic) mode of individuation of the referent. H. SigurDsson (p.c.) informs me that in Icelandic the reciprocal modifier hver á fætur ödhrum, lit. ‘each on feet other’s’’ allows for a collective antecedent triggering singular agreement on the predicate (e.g. the collective noun fólk). 5. The same problem would arise in a doubling derivation (Belletti 2003; Kayne 2002), whereby the antecedent would move away, stranding the reciprocal: (i) Ho telefonato [PP a [DP i miei colleghi]] [(l’)uno [i miei colleghi] [dopo l’altro]] 6. See the following examples (from Jackendoff 1990, (14)):

Number agreement and event pluralization: A case study 231

7. 8. 9.

10.

11.

12.

13. 14.

(i) I talked to John and Bill about each other. (ii) I heard from John and Bill about each other. Belletti (1982) argues that English each other is not subject to LF extraction of each, contrary to Heim, Lasnik and May (1991), among others. With subsequent remnant VP movement to a higher Specifier. On participial agreement see Kayne (1989, 1993); Friedeman and Siloni (1997); on Italian, see especially Egerland (1996). Or possibly Dative Case does, even when it is not (overtly) prepositional: cf. the impossibility for Dative quirky subjects to control verbal number agreement in Icelandic (see SigurDsson 2000; Taraldsen 1995, among others). When the antecedent is the thematic object, the agreeing version is preferable; but since this is the least marked option, this contrast is not really informative: (i) In questo esercizio di ginnastica, si alzano/?? alza le mani una dopo l’altra. In this gym exercise si raise-3PL/?? 3SG the hands one after the other ‘In this gym exercise, you have to raise your hands one after the other.’ Cinque (1988: 554–558) argues that the agreeing option in (12a) involves a [+argument] clitic si, whereas the non-agreeing option involves [-argument] si and an argumental pro subject; the former is incompatible with reciprocal l’uno …l’altro because there is no [+plural] NP antecedent to which l’uno could possibly adjoin (Cinque 1988: 540–541; cf. (8) above). In the account to be proposed here, the [+plural] NP is required for different reasons (see below § 3). The resulting algebraic structure is a complete join semilattice minus the bottom element. The crucial property of such a structure is that it is closed under join, i.e. the join of any two individuals yields another individual belonging in the same structure. The relevant literature is too vaste to be summarized here; see Landman (1996) for an overview. Beck (2001) argues that the ** operator is sensitive to a contextually relevant cover, i.e. a partition of the plural denotations to be distributed over. A plural reciprocal expression in the modifier is felicitous only if the context provides a way to partition the set of the soldiers in at least two subgroups: (i) I soldati spararono gli uni dopo gli altri. the soldiers shot- PAST.3PL the ones after the others The singular reciprocal in (22) instead requires a partition into individual parts.The role in determining a cover for the DP is evident in the case of the cognate modifiers uno a uno ‘one by one’, due alla volta ‘two at a time’, etc. Possibly, the temporal preposition dopo take as a complement an elliptical clause rather than a simple DP. I will not try to explore this possibility here. It is beyond the scope of this paper to explore to what extent the generalizations hold of other instances of the reciprocal expression in Italian. L’uno…l’altro seems to obey the first two generalizations in all of its instances; it is possible to speculate that all these reciprocal structures involve pluralization of a Davidsonian e-position (whereby each pair of individuals in a reciprocal relation

232 Valentina Bianchi

15.

16. 17. 18.

19.

20.

would correspond to a distinct subevent or subsituation). See Beck (2001) for an application of the e-position to certain reciprocal interpretations. One structure that I will not discuss is the apparently DP-internal reciprocal modifier exemplified in (i): (i) Leggeva un libro dopo l’altro. read-IPF.3SG one book after the other ‘(s)he read one book after another.’ Agreement with politeness forms is actually subject to cross-linguistic variation (see e.g. Comrie 1975).The topic is too vaste to be fully addressed here. See Heycock and Zamparelli (2004) on plurality in the noun phrase. According to Wechsler (2002), second person plural agreement is actually pure person agreement. I cannot adopt Wechsler’s radical hypothesis that the second person plural pronoun lacks a formal [+plural] feature in all of its instances, because a semantically plural second person pronoun can antecede the reciprocal modifier, which I have argued requires a syntactically as well as semantically plural antecedent. In a top-down perspective, this could be thought of as a sort of “selectional restriction” imposed by the higher person head on the lower number head. Another possibility is that in cases like (37) and (38), the person and number features are exceptionally projected in a single syncretic head. The present proposal is inconsistent with some claims put forth by Sauerland (2003, 2005): first, that phi-features are semantically interpretable only in the DP (in particular, in a dedicated head at the top of the DP structure); second, that the * and ** pluralization operators do not correspond to morphological plural marking (but here there might be some cross-linguistic variation); third, that the plural feature is semantically unmarked in all languages, even though it tends to be morphologically marked cross-linguistically. I have to leave a detailed discussion for future research.

References Abney, Steven 1987 The English noun phrase in its sentential aspects. Ph.D. diss., MIT. Bach, Emmon 1986 The algebra of events. Linguistics and Philosophy 9: 5–16. Beck, Sigrid 2001 Reciprocals are definites. Natural Language Semantics 9: 69–138. Belletti, Adriana 1982 On the anaphoric status of the reciprocal construction in Italian. The Linguistic Review 2: 101–137. 2003 Extended doubling and the VP periphery. Ms., University of Siena.

Number agreement and event pluralization: A case study 233 Belletti, Adriana and Luigi Rizzi 1995 Su alcuni casi di accordo del participio passato in francese e in italiano. Ms., University of Siena. Bianchi, Valentina 2003 On finiteness as logophoric anchoring. In Temps et point de vue / Tense and Point of View, Jacqueline Guéron and Liliane Tasmovski (eds.), 213–246. Paris: Université Paris X – Nanterre. Bonomi, Andrea 1997 Aspect, quantification, and when-clauses in Italian. Linguistics and Philosophy 20: 469–514. Borer, Hagit 2005 Structuring Sense, Vol. II. New York/Oxford: Oxford University Press. Chierchia, Gennaro 1998 Plurality of mass nouns and the notion of “semantic parameter”. In Events and Grammar, Susan Rothstein (ed.), 53–104. Dordrecht: Kluwer. Cinque, Guglielmo 1988 On si constructions and the theory of arb. Linguistic Inquiry 19: 521–581. 1999 Adverbs and Functional Heads. Oxford: Oxford University Press Comrie, Bernard 1975 Polite plurals and predicate agreement. Language 51: 406–418. Egerland, Verner 1996 The Syntax of Past Participles. A Generative Study of Nonfinite Constructions in Ancient and Modern Italian. Lund: Lund University Press Friedeman, Marc Ariel and Tal Siloni 1997 Agrobj is not Agrparticiple. The Linguistic Review 14: 69–96. Giusti, Giuliana this vol. Parallels in clausal and nominal periphery. Heim, Irene, Howard Lasnik and Robert May 1991 Reciprocity and plurality. Linguistic Inquiry 22: 63–101. Heycock, Caroline and Roberto Zamparelli 2004 Friends and colleagues: plurality, coordination, and the structure of DP. Ms., University of Edinburgh and University of Bergamo. To appear in Natural Language Semantics. Jackendoff, Ray 1990 On Larson’s treatment of the double object construction. Linguistic Inquiry 21: 427–456. Kayne, Richard S. 1989 Facets of Romance past participle agreement. In Dialect Variation and the Theory of Grammar, Paola Benincà (ed.), 55–84. Dordrecht: Foris.

234 Valentina Bianchi 1993

Toward a modular theory of auxiliary selection. Studia Linguistica 47: 3–31. 2002 Pronouns and their antecedents. In Derivation and Explanation in the Minimalist Program, Samuel Epstein and Daniel T. Seely (eds.), 133– 166. London: Blackwell. 2003 Prepositions as probes. In Structures and Beyond. The Cartography of Syntactic Structures, vol. 3, Adriana Belletti (ed.), 192–212. Oxford: Oxford University Press. Kratzer, Angelika 1996 Severing the external argument from its verb. In Phrase Structure and the Lexicon, Johan Rooryck and Laurie Zaring (eds.), 109–137. Dordrecht: Kluwer. Krifka, Manfred 1992 Thematic relations as links between nominal reference and temporal constitution. In Lexical Matters, Ivan Sag and Anna Szabolsci, (eds.), 29–53 Stanford: Stanford University Press. Landman, Fred 1989 Groups I. Linguistics and Philosophy 12: 559–605 1996 Plurality. In The Handbook of Contemporary Semantic Theory, Shalom Lappin (ed.), 425–457. London: Basil Blackwell. Link, Godehard 1983 The logical analysis of plural and mass terms: a lattice-theoretic approach. In Meaning, Use and Interpretation of Language, Bauerle Rainer, Christoph Schwartze and Arnim von Stechow (eds.), 302– 323. Berlin: Walter de Gruyter. Partee, Barbara 1995 Quantificational structures and compositionality. In Quantification in Natural Languages, Emmon Bach, Eloise Jelinek, Angelika Kratzer and Barbara H. Partee (eds.), 541–601. Dordrecht: Kluwer. Pollard, Carl and Ivan Sag 1994 Head-driven Phrase Structure Grammar. Stanford: CSLI Publications. Ritter, Elizabeth 1995 On the syntactic category of pronouns and agreement. Natural Language and Linguistic Theory 12: 405–443 Sternefeld, Wolfgang 1998 Reciprocity and cumulative predication. Natural Language Semantics 6: 303–337. Sauerland, Uli 2003 A new semantics for number. Proceedings of SALT 13, 258–275. Cornell University, Ithaca, NY: CLC Publications. 2005 A comprehensive semantics for agreement. Available from: http:// www.zas.gwz-berlin.de/mitarb/homepage/sauerland/downloads.html

Number agreement and event pluralization: A case study 235 Shlonsky, Ur 1989 The hierarchical representation of subject verb agreement. Ms., University of Haifa. Available from: http://www.unige.ch/lettres/linguistique/shlonsky/readings/hierarchical.pdf Taraldsen, Knut T. 1995 On agreement and nominative objects in Icelandic. In Studies in Comparative Germanic Syntax II, Hubert Haider, Susan Olsen and Sten Vikner (eds.), 307–327. Dordrecht: Kluwer SigurDsson, Halldór Ármann 2000 The locus of case and agreement. Working Papers in Scandinavian Syntax 65: 65–108. 2004 Agree in syntax, agreement in signs. Ms., University of Lund. to appear The nominative puzzle and the low nominative hypothesis. Ms., University of Lund. To appear in Linguistic Inquiry. Wechsler, Steven 2002 Number as person. In Empirical Issues in Syntax and Semantics 5, Olivier Bonami and Patricia Cabredo Hofherr (eds.), 255–274. http:// www.cssp.cnrs.fr/eiss5/wechsler/wechsler-eiss5.pdf Zamparelli, Roberto 2000 Layers in the Determiner Phrase. New York: Garland. 2004 On singular existential quantifiers in Italian. Ms., University of Bergamo.

The Phase Condition and cyclic Spell-out: Evidence from VP-topicalization Roland Hinterhölzl

In this paper, I argue that next to the C-domain, which constitutes the leftperiphery of the TP, there is a V-domain which forms the left-periphery of the VP. This V-domain comprises a number of functional heads which serve to license finite and different types of nonfinite complements of the verb. As is evidenced by phase condition effects, this V-domain constitutes a strong phase the left-edge of which is formed by the Aspect phrase. The empirical evidence for this proposal comes from the formation of verb clusters in West Germanic.

1. Introduction There are two long standing unsolved problems with restructuring in West Germanic and I argue that their solution involves assuming that extraction out of verb clusters is subject to the Phase Impenetrability condition. Furthermore, I will argue that the Aspect phrase constitutes the left edge of the VP-phase and provide evidence that Spell-out and the application of prosodic conditions are tied to phases. The empirical issue concerns verb cluster formation, a morphological effect related with it and VP-topicalization in West Germanic (WG). Verb clusters in German, Dutch and West Flemish (WF) give rise to the so-called IPP-effect, in which the expected participle in the perfect construction of a restructuring verb is replaced with a bare infinitive, the Infinitivus Pro Participio, as is illustrated in (1a-b) for Dutch and in (1c-d) for German. (1)

a. *dat Elsje hem een brief heeft gewild schrijven that Elsje him a letter has wanted-PP write b. dat Elsje hem een brief heeft willen that Elsje him a letter has want-IPP

schrijven write

238 Roland Hinterhölzl c. *dass Else ihm einen Brief schreiben gewollt hat that Elsje him a letter write wanted-PP has d. dass Else ihm einen Brief hat schreiben wollen1 that Elsje him a letter has write want-IPP The IPP-effect is generally voided in cases of VP-topicalization in Dutch and WF (2). In German, the IPP-effect also shows up in cases of VPtopicalization (3a) and is only voided with perception verbs (3b). This raises the question of why the IPP-effect is voided in Dutch and WF but only in a subcase in German. I will argue that these differences follow from the fine structure of the verb clusters in these languages plus the Phase Impenetrability Condition. (2)

a. een boek lezen heeft hij niet a book read has he not b. in nen bank werken ee se in a bank work has she

gewild / *willen (Dutch) wanted-PP / want-IPP niet gewild / *willen (WF) not wanted-PP / want-IPP

(3)

a. in einer Bank arbeiten hat sie nicht wollen / *gewollt in a bank work has she not want-IPP / wanted-PP b. kommen hat Hans die Maria nicht gesehen / *sehen come has Hans the Maria not seen / see-IPP

VP-topicalization in Dutch is also special since it can involve the dependent infinitive plus its direct object, as is also illustrated in (2a). Note that this is unexpected since after restructuring the dependent infinitive and its arguments and adjuncts modifying it do not form a constituent anymore. Instead, the verbs form a verb cluster which is thought to be the result of Verb Raising (VR) and all the other constituents of the infinitival clause join up in the middle field of the matrix clause. VR may only ‘pied-pipe’ additional material, such as the direct object of the dependent infinitive, if the language or dialect allows for Verb Projection Raising (VPR), as is the case in WF, Swiss Germain and many other WG dialects. Parallel cases of VR and VPR in WF are given in (4). However, as is illustrated in (5), Dutch, contrary to German and WF, does not allow for VPR, neither with a participle, nor with an IPP-infinitive.

The Phase Condition and cyclic Spell-out

(4)

(5)

239

a. da Valere [ze morgen nen boek] wilt [geven] that Valere her tomorrow a book wants give ‘that Valere wants to give her a book tomorrow’

(VR)

b. da Valere [ze morgen] wilt [nen boek geven] that Valere her tomorrow want a book give

(VPR)

a. *hij heeft niet gewild een boek lezen he has not wanted a book read

(VPR)

b. *hij heeft niet willen een boek lezen

(VPR)

c. hij heeft een boek niet willen lezen he has (a book) not want-IPP (a book) read

(VR)

Given that verb cluster formation in Dutch necessarily involves VR, this raises the question of why VR can be dispensed with in cases of VPtopicalization. Or put in more general terms, the question arises of what the derivational source of VP-topicalization in Dutch is. Haider (1990) and Zwart (1993) argue that topicalized VPs are base-generated in the C-domain. I have argued in Hinterhölzl (1999) that such an analysis runs into a number of technical difficulties and should therefore be avoided if possible. In this paper, I argue that VP-topicalization structures and VR clusters can be derived from the same derivational source, employing competition and blocking between the regular infinitive and the gerund as complements of restructuring verbs. The paper is organized in the following way. In Section 2, I address the bleeding or rather the lack of the bleeding of the IPP-effect in German. I will first provide an account of the IPP-effect and then argue that the solution to this intralinguistic variation involves assuming that there are two types of infinitives in WG, one giving rise to the IPP-effect in verb cluster formation and one failing to do so. In Section 3, I provide an account of the properties of restructuring infinitives in terms of remnant movement that derives the necessary movement operations from a general theory of sentential complementation. Furthermore, I provide a motivation for the formation of verb clusters that explains why infinitives induce an IPP-effect, while gerunds fail to do so. In Section 4, I show that the differences in VP-topicalization between German on the one hand, and Dutch and WF on the other hand, follow from the different fine structure of the verb clusters in these languages plus the Phase Impenetrability Condition (PIC). I argue that VR- and VPR-

240 Roland Hinterhölzl structures in Dutch result from competition between the regular infinitive and the gerund, where the gerund is blocked by the more economic derivation involving the infinitive, unless extraction of the infinitive out of the verb cluster is blocked by the PIC in cases of VP-topicalization. In Section 5, I discuss the quirky properties of topicalized right-branching verb clusters in German and argue that these properties follow straightforwardly if it is assumed that Spell-out applies in a cyclic fashion. Evidence for this proposal is provided by showing that there is a prosodic condition that only applies in the V-domain in German.

2. The IPP-effect and two types of infinitives in West Germanic

In this section, I address the (non-)voiding of the IPP-effect in VP-topicalization in German and argue that the appearance of the IPP-effect signals the availability of two types of infinitives in WG. I will argue that the IPP-effect is a syntactic effect of verb cluster formation which can be avoided by nominalized infinitives since they are licensed in a different structural position in the V-domain. The interesting question that verb clusters displaying the IPP-effect raise is the issue of whether IPP-infinitives are real infinitives or hidden participles of some sort. Already Grimm (1898/1969) put forth the hypothesis that IPP-infinitives are prefixless participles. I will adopt the hidden participle hypothesis, since it allows us to treat IPP-infinitives as participles for the purpose of checking the subcategorisation of the auxiliary as well as for the purpose of the temporal interpretation of these verb clusters.2 Second, there is distributional evidence from WF and Afrikaans showing that IPPinfinitives and participles pattern alike (cf. Hinterhölzl 1999, to appear) for the details. Let us now address the issue of how to account for the IPP-effect (cf. also van den Wyngaerd 1994). In Hinterhölzl (1999), I have argued that the IPP-effect is due to a structural incompatibility between the participial prefix (of the selecting verb) and the dependent infinitive in the formation of verb clusters on account of the fact, that those WG languages and dialects (namely Frisian and Low German) that never display an IPP-effect build the participle without a prefix. In the WG languages displaying the IPP-effect, the participle is formed by affixation of the prefix ge and the suffix t/d. I follow Halle and Marantz

The Phase Condition and cyclic Spell-out

241

(1993) in assuming that inflected forms are (partially) derived in the syntax. In particular, I have proposed in Hinterhölzl (1999) that the prefix ge is inserted in [Spec,F2P], while the suffix is inserted in the head position of the Aspect phrase. The verb will then first move to F2, to check its prefix, and then up to Asp0 to adjoin to its suffix. The prefix then undergoes head movement and left-adjoins to the complex of verb and suffix to form the participle before Spell-out. This is illustrated in (6). (6)

[AspP -t [F2P [ ge ] [ F2 [VP V ]]]]

If we assume that dependent infinitives are licensed in [Spec,F2P] of the selecting verb, then it follows that a verb in participial form and a dependent infinitive selected by such a verb rule each other out. In Hinterhölzl (1999), I propose that in this case the prefix is blocked and that the suffix selected by it is dropped. Instead, a zero-affix is inserted in Asp0 that comprises the relevant features and the verb is spelled-out with the default morphology of an infinitive in F20, as is illustrated in (7). (7)

AspP1

lesen

Asp 0 [+ participle] [+ past]

F2P t wollen

F3P

Let us now address the bleeding of the IPP-effect in VP-topicalizations involving perception verbs. It is important to note that verb clusters with perception verbs give rise to the IPP-effect only optionally, as is illustrated in (8). (8)

a. weil Hans die Maria kommen gesehen hat since Hans the Maria come seen has b. weil Hans die Maria hat kommen sehen since Hans the Maria has come see-IPP

How can we explain the optionality in the appearance of the IPP-effect with perception verbs? In order to avoid assuming optionality in the triggering of the IPP-effect, I propose that there are two types of infinitives in German,

242 Roland Hinterhölzl the regular infinitive, here simply called infinitive, and a nominal infinitive, called gerund. Given these assumptions, we can assume that the licensing of the infinitive in the verb cluster always gives rise to the IPP-effect, as discussed above, while the gerund fails to induce the IPP-effect due to being licensed in a different position in the verb cluster. In this scenario, the optionality in the triggering of the IPP-effect with perception verbs is explained by the assumption that perception verbs are special in that they can select both types of infinitives, similarly to perception verbs in English which can either take the infinitive or the gerund as in He saw her come versus He saw her coming. Though there is no morphological evidence for the proposed distinction in (modern) German, there is comparative and diachronic evidence for the existence of two types of infinitives in WG. First, Frisian morphologically distinguishes two types of infinitives, called the Doelfoarm (ending in -n) and the Nammefoarm (ending in -e) which are in complementary distribution. According to Wolf (1996), Doelfoarms occur after the infinitival marker te and in the complement of perception verbs, as is illustrated in (9), while Nammefoarms occur in most other positions including the complement of litte (‘let’, ‘make’), as is illustrated in (10). Furthermore, it is important to note that the class of verbs that select the Doelfoarm in Frisian largely overlaps with the class of verbs that fail to trigger the IPP-effect in German (cf. Hinterhölzl 1999, to appear). (9)

Ik kin har der rinnen /*rinne sjen I can her there run see

(10) Ik sil har mar restich *lezen / leze litte I will her just calmly read let Secondly, Old High German had, next to the regular infinitive, an inflected infinitive which displayed both nominal and verbal properties, matching the envisaged category of a gerund. For example, in sentence (11) (taken from Demske 2001) the inflected infinitive (tuonne) is marked with the Dative Case ending -e (since it is selected by the preposition zi (‘to’)), a typical nominal property, and at the same time displays a typical verbal property in licensing the adverb ubelo (‘badly’). (11) uuanda imo lussam uuas ubelo zi tuonne since to him desirable was badly to act

(N Ps 413.28)

The Phase Condition and cyclic Spell-out

243

The morphological distinction between infinitive and gerund was lost in the Middle High German period, but I assume that the gerund is still available in the grammar of WG. I propose that the gerund in (modern) German, Dutch and WF is analyzed as a zero-affix on the infinitive with nominal properties. To explain the absence/presence of the IPP-effect, I propose that the gerund and the infinitive are licensed in different positions in the verb cluster, as is illustrated in (12). Infinitives are licensed in [Spec,F2P], thereby blocking the participial prefix and giving rise to the IPP-effect, while gerunds are licensed in [Spec,AspP] without interfering with the participle morphology of the selecting verb. Furthermore, I assume, a fact that will become important for the analysis of verb clusters below, that CPcomplements are licensed in [Spec,F3P] of the selecting verb. (12) [AspP (=F1P) gerund [F2P dependent infinitive [F3P CP [VP V ]]]] Given the existence of two types of infinitives in WG, the assumptions about their licensing in (12) and the account of the IPP-effect decribed above, we do not need to assume that VP-topicalization can somehow circumvent verb cluster formation. In this scenario, we can treat restructuring and verb cluster formation as obligatory assuming that VP-topicalization that bleeds the IPP-effect involves the gerund, while VP-topicalization that fails to bleed the IPP-effect must involve the infinitive. I will argue in Section 4 that the choice between the infinitive and the gerund, or between the bleeding and the lack of bleeding of the IPP-effect in VP-topicalization follow from the PIC and the fine structure of the verb clusters in German, Dutch and WF. But before we take a closer look at verb clusters, it is important to understand what motivates the formation of verb clusters in restructuring infinitives. 3. Restructuring and the formation of verb clusters Restructuring is a notoriously difficult phenomenon that raises several questions. First, there is the issue of what forces the formation of verb clusters. Second, there is the issue of how to explain the transparency of restructuring infinitives that displays itself in different phenomena like clitic climbing in Romance and extended scope possibilities of the infinitival complement in Germanic. There are two types of approaches to explaining the properties of restructuring infinitives, monoclausal approaches and biclausal approaches.

244 Roland Hinterhölzl In monoclausal approaches the issue of transparency disappears trivially but reappears in the form of the question under which conditions two (main) verbs can project a single clause (cf. Cinque 2001 and Wurmbrand 2001, for interesting new answers to this question). In particular, Cinque (2001) put forth the hypothesis that restructuring verbs in Italian – essentially modal and aspectual verbs in this language – are those verbs that can be taken to occupy functional positions in the clausal skeleton above VP. Wurmbrand (2004b) proposed an important distinction between lexical and functional restructuring verbs, arguing that the majority of restructuring verbs in German are of the lexical type, since they are not subject to order restrictions that are typical of functional restructuring verbs. Biclausal approaches, on the other hand, assume that restructuring infinitives are reduced clauses or employ special operations (like VR) that make them transparent (cf. Hinterhölzl 1997; Koopman and Szabolcsi 2000, for accounts in terms of remnant movement). However, as is also pointed out by Wurmbrand (to appear), what is still missing is a satisfactory answer to the question of what motivates restructuring and the formation of verb clusters (for some discussion of this question see also the papers in Kiss and Van Riemsdijk 2004). I argue that the complements of restructuring verbs are full CPs and that the formation of verb clusters is due to the presence of a deficient complementizer (cf. Hinterhölzl 1999; to appear). This account is embedded in a general theory of sentential complementation in which the complementizer a) is essential for rendering a sentential complement (a TP) into an argument and b) plays a crucial role in linking the embedded TP to the matrix event time (cf. Den Besten 1977/1983; Enç 1986; Guéron and Hoekstra 1988). In this approach, the complementizer acts as place holder for the selectional restrictions of the matrix verb (cf. Hinterhölzl to appear for further details). It is inserted in the head position of the Status phrase3 to check the subcategorization of the matrix verb against the Aspect phrase of the embedded clause. Then it moves through MoodP – where it assigns a temporal index to the embedded TP – to the head of ForceP, which constitutes the highest head position in the C-domain, as is illustrated in (13a). The features of the complementizer are then checked by movement of the entire CP (=ForceP) into [Spec,F3P] of the selecting verb. In this account, not only arguments need to be licensed, but also the main phases of the clause, namely TP and AspP are licensed by moving into dedicated licensing positions in the C-domain, as is illustrated in (13b).

The Phase Condition and cyclic Spell-out

245

(13) a. functional heads licensing sentential complements [CP Force [MoodP M [StatP S [TP …]]]] b. licensing movements in an embedded clause MoodP Force StatP M TP S

AspP

dass c. licensing movements in a restructuring clause [CP1 [TP1 [PredP1 [TP2] [AspP1 [AspP2] V1 [F3P [CP2 0 tTP2 tAspP2] [VP1]]]]] In restructuring infinitives the embedded complementizer, being defective, fails to value the infinitival TP and AspP completely, hence they undergo, in a parallel manner to subjects in subject raising constructions, licensing movement into dedicated positions in the matrix clause, as is illustrated in (13c). The infinitival TP, not being temporally linked, fails to denote an event token, hence does not qualify as an argument and is licensed as predicate in [Spec,PredP] of the selecting verb (for the relevance of PredP see also Bowers (1993) and Koster (1995)). This movement accounts for the transparency and the monoclausal middle field of restructuring infinitives. The infinitival AspP in this approach is taken to move into a Specifier in the higher V-domain to check the subcategorisation of the selecting verb and to link the embedded event to the matrix event time. In this approach, movement of the infinitival AspP leads to VR-constructions, whereas VPR-constructions are derived, if the infinitival AspP pied-pipes additional structure, for instance the infinitival AgrOP, when it is moved into the V-domain of the matrix verb. To conclude, verb cluster formation has two motivations. It serves to temporally link the embedded event and to check the subcategorization of the matrix verb.

246 Roland Hinterhölzl 4. The fine structure of verb clusters in West Germanic and the Phase Condition In this section, I will first explain why the IPP-effect in VP-topicalizations with perceptions verbs in German is voided, that is to say, why only the gerund is available for VP-topicalization in this context. This account will then provide us with a clue for tackling the voiding of the IPP-effect in Dutch and WF, as well as for the constituency paradox in Dutch. 4.1. Verb clusters in German First, it is important to note that nonfinite verbs in German, contrary to nonfinite verbs in WF, obligatorily precede IPP-infinitives, as is illustrated in (14). Given that IPP-infinitives occupy [Spec,F2P] of the selecting verb, it follows that the selecting verb in German, whether finite or nonfinite, always moves into the highest head position in the V-domain, that is, into Asp0 (cf. Hinterhölzl 1999, to appear for further discussion). (14) a. Else wird ihm einen Brief haben schreiben wollen Else will him a letter have write want-IPP b. *Else wird ihm einen Brief schreiben wollen haben Else will him a letter write want-IPP have ‘Else will have wanted to write him a letter’ c. dan-ze kosten willen dienen boek kuopen een that they could want-IPP that boek buy have d. *dan-ze kosten een willen dienen boek kuopen that they could have want-IPP that boek buy ‘that they could have wanted to buy that book’ This in turn implies that in left branching verb clusters, which are prevalent in German, the dependent infinitive preceding the selecting verb occupies [Spec,AspP]. Given that infinitives first move into [Spec,F2P] of the selecting verb thereby inducing the IPP-effect, it follows that infinitives in German touch down twice in a verb cluster, raising the question of what the motivation for this double checking is. Returning to the motivations for the formation of verb clusters discussed in the previous section, I propose that [Spec,F2P] is responsible for the temporal linking of nonfinite verbs and that, possibly depending on the

The Phase Condition and cyclic Spell-out

247

extent of verb movement in a language, both [Spec,AspP] and [Spec,F2P] can check the subcategorisation of the selecting verb. In German, where the (selecting) verb moves up to Asp0, a dependent infinitive is temporally linked in [Spec,F2P] (triggering the IPP-effect) and moved on into [Spec,AspP] to check the subcategorisation of the selecting verb. This raises the question about the temporal licensing of the second infinitive, that is, the gerund. In (12), we have assumed that gerunds move directly into [Spec,AspP], thus failing to give rise to the IPP-effect. However, this may not be a problem given that the gerund is a nominal category. Enç (1986) argued that only verbs must be temporally anchored, while nominals are interpreted independently of tense: if nominals were bound by tense, then sentences like Fugitives will be put in prison should be contradictory, contrary to what is the case. In conclusion, this analysis of the licensing of the two types of infinitives strengthens our assumption that the second type of infinitive has nominal properties. The same analysis can be applied to verb clusters involving participles. The participle being a verb first moves into [Spec,F2P] of the selecting auxiliary to be temporally licensed. Then the participle moves on to [Spec,AspP] to check the subcategorisation of the auxiliary. In the case of the participle, however, there is an additional motivation for the latter step in the derivation. It is plausible to assume that movement of the participle phrase into [Spec,AspP] of the selecting auxiliary also serves to make the temporal information contained in the participial morphology visible for further computation (this will follow from the Phase condition to be introduced below). (15) shows a verb cluster involving a participle of a perception verb selecting a gerund. In this case, the gerund moves directly into [Spec,AspP2] without giving rise to an IPP-effect. Given this analysis, I propose that the Spell-out of the infinitive in the higher Specifier yields left-branching verb clusters, while Spell-out of the infinitive in the lower Specifier yields right-branching verb clusters in German, as is illustrated in (16) and (17).4 (15) verb cluster involving participle + gerund [AspP1 [AspP2 [AspP3 lesen-GER ] gesehen ] hat]]]

248 Roland Hinterhölzl (16) left branching verb cluster: lesen wollen wird (read want shall) AspP1 AspP2 F2P

wird AspP3 wollen

AspP2 tAspP3

lesen <wollen> (17) right branching verb cluster: wird lesen wollen (shall read want) AspP1 AspP2 F2P

wird AspP3 <wollen> tAspP3

AspP2

wollen lesen Verb clusters with IPP-infinitives are obligatorily right-branching, as we have seen in (14) above. In the case of an IPP-infinitive, all the relevant information is contained in the empty participial morpheme, as is illustrated in (18a). I propose that in IPP-infinitives the zero morpheme undergoes head movement to the Aspect head of the auxiliary to check its subcategorisation and to make the temporal information accessible for further computation. Head movement in this case would follow, if we make the plausible assumption that an empty morpheme cannot induce pied-piping of the entire Aspect phrase. This analysis implies that the highest Specifier of a verb cluster with an IPP-infinitive is empty and available as an escape hatch. This will become important when we talk about extractions out of verb clusters in the following section. Some indirect evidence for this analysis comes from IPP-infinitives in Bavarian dialects which regularly fill the highest empty Specifier with the dependent infinitive yielding verb clusters displaying the order V3-V1-V2, as is illustrated in (18b).

The Phase Condition and cyclic Spell-out

249

(18) a. The analysis of infinitive + IPP-infinitive [AspP1 0-hat [F2P1 [AspP2 [AspP3 lesen ] t0 [F2P2 wollen ]]]] b. IPP-infinitive in Bavarian [AspP1 [AspP3 lesen] 0-hat [F2P1 [AspP2 tAspP3 t0 [F2P2 wollen ]]]]

4.2. The Phase Condition and VP-topicalization In this section, I argue that the VP-topicalization facts discussed in Section 1 follow if we assume that the Aspect phrase, not the vP itself, as proposed in Chomsky (1998, 2001), constitutes a (strong) phase, implying that extraction out of a verb cluster is only possible via [Spec,AspP], given that under these assumptions this Specifier constitutes the left-edge of the relevant phase. In Chomsky’s theory of phases, the access to the lexicon is a one-time selection of a lexical array LA. LA enters the derivation in different steps. In each step a subarray of LA is put in active memory. The syntactic object that is formed when a subarray is exhausted is called a phase. Furthermore, Chomsky assumes that vPs and CPs, but crucially not IPs, are strong phases. A derivation by phases involves a cyclic Spell-out of (sub)structures, the point of which is determined by (19). Computation is strictly local and constrained by the Phase Impenetrability Condition as given in (20). (19) Phase Condition (Chomsky 2001) Evaluation for a phase is done at the level of the next highest strong phase (20) Phase Impenetrability Condition (PIC) The complement of a strong phase a is not accessible to operations at the level of the next highest strong phase b, but only the head and the edge of a are Since VP-topicalization involves movement of a constituent contained in a strong phase, namely vP/AspP, to the next highest strong phase, namely the C-domain, it is expected that the Phase condition constrains VP-topicalization. (21) again illustrates the facts of VP-topicalization that we need to account for in German. Extraction out of a verb cluster involving a perception

250 Roland Hinterhölzl verb leads to a bleeding of the IPP-effect (21b), while extraction out of all other verb clusters including modal verbs, fails to induce a bleeding of the IPP-effect (21a). (21a) also indicates that extraction out of a verb cluster involving an IPP-infinitive must be licit in principle. (21) a. lesen hat er das Buch wollen /*gewollt read has he the book want-IPP / wanted-PP b. lesen hat er ihn das Buch gesehen /*sehen read has he him the book seen / see-IPP Let us start with verb clusters involving perception verbs. Perception verbs can optionally select an infinitive or a gerund. If the gerund is selected, the verb cluster that results from restructuring is given in (15). Given that the Aspect phrase constitutes the left-edge of the vP-phase, the gerund occupying a specifier of the Aspect phrase of the auxiliary, that is, the highest verb in the verb cluster, can be extracted without further ado. On the other hand, if the infinitive is selected, the verb cluster that results from restructuring is parallel to the structure illustrated in (18a). In this structure, the dependent infinitive, as is evidenced by the grammaticality of (21a) is extractable, since the escape hatch, the left-edge of the verb cluster, that is [Spec,AspP] of the auxiliary, is empty. Note, however, that in this structure the dependent infinitive does not occupy the left-edge of the Aspect phase, but is contained in a lower Specifier namely [Spec,F2P] of the auxiliary, from which position it cannot extract without violating the PIC at this point of the derivation. To insure extractability, the dependent infinitive must undergo last resort movement to the edge of the phase in the previous cycle, parallel to the movement of dependent infinitives in Bavarian illustrated in (18b). This movement of the dependent infinitive to the left-edge of verb clusters involving IPP-infinitives must be taken to be a last resort operation, since it does not serve any licensing purposes and parallels the movement of a wh-element to the local [Spec,CP] in cases of long distance wh-movement. This last resort operation will only then be licit, if the grammar does not provide another option leading to a convergent derivation, as is the case with verb clusters involving modal verbs in German. With verb clusters involving perception verbs, however, there is an alternative derivation, since the gerund could have been selected instead. In other words, VP-topicalization with perception verbs leads to a bleeding of the IPP-effect since the derivation involving the gerund is more eco-

The Phase Condition and cyclic Spell-out

251

nomic: it requires one movement step less, than the derivation involving the infinitive.

4.3. Verb clusters in Dutch and West Flemish While verb clusters in German are (predominantly) left-branching, verb clusters in Dutch and WF are generally right-branching. Furthermore, as we have also seen in (14) above, nonfinite verbs in Dutch and WF, contrary to German, fail to move to the highest head in the V-domain. This implies that a dependent infinitive, while moving into [Spec,F2P] of the selecting verb and thereby inducing the IPP-effect, must be taken to be spelled-out in [Spec,F3P] below, that is to say, within the containing CP that is licensed in [Spec,F3P] (cf. (23)). Another feature that distinguishes verb clusters in German and Dutch, is the fact that particles can climb in the verb cluster in Dutch but not in German, as is illustrated in (22).5 I propose that particle-climbing is analyzed as movement of the particle into [Spec,AspP] of the higher verb, as is shown in (24). Since infinitives, gerunds and participles in German check the subcategorization in [Spec,AspP] of the selecting verb, particle climbing is correctly predicted to be ruled out in this language. On the other hand, this implies that the subcategorisation of nonfinite verbs in Dutch and WF is also checked in [Spec,F2P]. This is in accordance with the amount of verb movement in the V-domain in these languages. (22) a. dat hij mij (weg) kan (weg) horen (weg) rijden that he me (away) can (away) hear (away) drive ‘that he can hear me drive away’

(Dutch)

b. dass er mich (*an) wird (*an) haben (an) rufen wollen that he me (up) will (up) have (up) call (up) want-IPP ‘that he may very well have wanted to call me up’ (23) right branching dependent infinitive in Dutch: (zal) willen uitlezen [AspP1 [F2P1 [AspP2 ] willen [F3P1 [AspP2 uit lezen ]]]] (24) verb cluster + particle climbing in Dutch: (zal) uit willen lezen [AspP1 uit [F2P1 [AspP2 ] willen [F3P1 [AspP2 lezen ]]]]

(G)

252 Roland Hinterhölzl Let us now look at VP-topicalization in the two languages. The relevant data are given again in (25a–b). Topicalization of the dependent infinitive, either alone or with its direct object, leads to a bleeding of the IPP-effect. The structure of the verb cluster feeding VP-topicalization in (25a) is given in (25c). (25) a. lezen heeft Jan het boek niet gewild read-INF has Jan the book not wanted b. een boek lezen heeft Jan niet gewild a book read-INF has Jan not wanted c. [AspP heeft [F2P [AspP 0 [F2P willen [F3P lezen ]]]]] Like in German IPP-infinitives, the left-edge of the verb cluster is available as an escape hatch. However, in this case the dependent infinitive has to undergo last resort movement to the left-edge crossing its own unspelledout copy in [Spec,F2P], which we may assume is excluded in principle. Thus, since extraction of the infinitive is blocked in Dutch and WF, the gerund is inserted in the course of the derivation as a means of last resort. The gerund being licensed in [Spec,AspP] of the selecting verb will not induce an IPP-effect and occupying the left-edge of the verb cluster is accessible for extraction to be topicalized in the final step of the derivation. Note that for this account to work, it is crucial that Spell-out decisions in the verb cluster are fixed before the derivation reaches the C-domain. If Spell-out of the verb cluster could be deferred to the point where VPtopicalization applies, then the question arises why the dependent infinitive could not be spelled-out (as a last resort operation) in [Spec,F2P] in Dutch and WF. In this case, the dependent infinitive could be extracted out of the verb cluster and, contrary to the fact, the IPP-effect would not be voided. Thus it is important that Spell-out proceeds in a cyclic fashion. In Section 5, I will provide some evidence for the assumption that Spell-out is phasebased, by showing that also prosodic constraints must be taken to apply to specific phases only. Before this, however, we must discuss how (25b) can be derived, since Dutch does not allow for VPR.

4.4. The syntax of gerunds The proposal is that the derivation of (25b) in Dutch involves the insertion of the gerund as a means of last resort. I analyze the gerund as a phrasal

The Phase Condition and cyclic Spell-out

253

affix that morphologically selects for an infinitive. To satisfy its selection, the gerund affix can attach to the infinitive or to any extended projection of it, including a projection in which the direct object of the dependent infinitive is licensed. That is to say, at the point of the derivation at which the direct object of the infinitive is Case-licensed in [Spec,AgrOP] (and VPR applies to this constituent in German or in WF), a functional head containing the empty affix is inserted (cf. (26a)). This affix will attract its complement into its Specifier and fuses with the adjacent infinitive at Morphological Form (cf. Halle and Marantz 1993) to fulfil its morphological selection, as is illustrated in (26b). (26) a. [GP [G´ 0 [AgrOP een boek [AspP lezen [VP ]]]] Insertion of a functional head b. [GP [AgrOP een boek [AspP lezen [VP ]]] [G´ 0 t AgrOP ]] Adjacency allowing the affix to be fused with the infinitive This constituent is then moved in the very same fashion as a VPR-constituent into a licensing position in the V-domain of the selecting verb. However, since the derivation involving the gerund is less economic or contains more steps than the parallel derivation involving the infinitive, the alternative derivation with the gerund will only be available if the derivation with the infinitive does not converge, as is the case in VP-topicalization. In all other environments a derivation involving the insertion of a gerund will be blocked in Dutch by the less complex derivation involving an infinitive which will give rise to the IPP-effect and to VR structures only.

5. Cyclic Spell-out and phase specific prosodic constraints In this section, I will argue that Spell-out decisions are tied to phases by showing that also prosodic constraints apply phasewise. The evidence comes from the properties of topicalized right-branching verb clusters in German. It will be shown that these verb clusters are subject to a prosodic constraint that does not apply in the other phases of the clause. Right-branching verb clusters in German are possible as long as the most deeply embedded verb cluster is left-branching. This means that as soon as we have a verb cluster with at least three verbs, a right-branching verb cluster is possible, as is illustrated in (27a). In this case the verb cluster has the structure V1-V3-V2. As discussed in Section 4.1, I assume that a right-

254 Roland Hinterhölzl branching verb cluster is derived in that the dependent infinitive is moved into [Spec,AspP] of the selecting verb but spelled-out in [Spec,F2P] below. (27) a. weil er den Text muß lesen können since he the text must read can ‘since he must be able to read the text’ b. ??weil er den Text [[müssen [lesen können]] wird] since he the text must read can will c. weil er den Text [wird [müssen [lesen können]] since he the text will must read can ‘since he will have to be able to read the text’ Right-branching verb clusters are also subject to the following restriction. If a right-branching verb cluster is introduced at one cycle it has to be rightbranching in the next cycle as well, as is illustrated by the contrast in (27b– c). This restriction, however, cannot be taken to be due to a hard syntactictype of condition in WG, since it is violated regularly by IPP-infinitives in WF and Afrikaans (cf. Hinterhölzl 1999, to appear). Thus I would like to propose that this restriction is the effect of an interface requirement on the mapping between syntactic and prosodic structures.6 Assuming that left- and right-branching verb clusters are mapped onto left- and right-headed phonological phrases, this restriction can be formulated as given in (28a). (28b) is taken to account for the ungrammaticality of purely right-branching verb clusters in the standard language. (28) Prosodic Constraint a. A right-headed phonological phrase in a verb cluster must sit on a right branch with respect to the non-head b. The most deeply embedded phonological phrase in a verb cluster must be left-headed Furthermore, I assume that the violation of interface requirements leads to marked structures that count as ungrammatical only if there is no alternative derivation which does not violate the given interface condition. Therefore, (27b) is prosodically marked since it violates the prosodic condition in (28a) and counts as ungrammatical since there is an alternative derivation that involves a Spell-out option that does not violate it, namely the derivation yielding (27c).

The Phase Condition and cyclic Spell-out

255

In this context, it is interesting to note that the topicalization of rightbranching verb clusters in German gives rise to rather subtle grammatical differences. As is illustrated in (29), while the topicalization of a rightbranching verb cluster without an IPP-infinitive is ungrammatical, the topicalization of a right-branching verb cluster including an IPP-infinitive is slightly marked but fully grammatical. (29) a. ?*[müssen [lesen können]] wird er den Text must read can will he the text b. ?[haben [lesen wollen]] wird er den Text have read want-IPP will he the text ‘he will have wanted to read the text’ How can we account for the above contrast? I will argue that the subtle differences in (29a–b) follow from the interaction between the PIC and the prosodic condition in (28). Given the PIC, the verb cluster [müssen [lesen können]] must be extracted from [Spec,AspP] of the finite verb. In this position it will incur a violation of the prosodic constraint in (28a). This violation induces ungrammaticality, since there is an alternative derivation that involves a Spell-out option that does not violate (28a). This is the derivation in which the infinitives are spelled-out in the highest Specifier of the V-domain in the previous cycle yielding the purely left-branching verb cluster [[lesen können] müssen]. In the case of the verb cluster involving the IPP-infinitive in (29b), however, there is no alternative derivation that does not violate (28a), since IPP-infinitive only allow for right-branching verb clusters. Therefore (29b) is judged as prosodically marked but fully grammatical. It is important to note that the violation of this prosodic constraint must be incurred at the level of the V-domain, since at the CP level no such constraint can be taken at work, as is evidenced by the right-branching phrases occupying [Spec,CP] in (30). (30) a. [CP [behauptet dass die Erde rund ist] hat [IP keiner t]] claimed that the earth round is has noone b. [CP [der Bruder von Peter] hat [IP Maria getroffen]] the brother of Peter has Mary met The above data therefore show that the violation of prosodic constraints can be incurred in the course of the derivation. This implies that Spell-out must

256 Roland Hinterhölzl be cyclic proceeding from phase to phase. Furthermore, the above discussion has shown that prosodic constraints can be relevant in one domain but fail to apply in another domain. Therefore, we can conclude that prosodic constraints not only apply phase-wise but can also be tied to specific phases. We have seen evidence that the German V-domain is subject to a prosodic constraint that is typical of VO-languages, but does not apply in other domains, that is, in any other phase in German.

Acknowledgements I would like to thank Mara Frascarelli and an anonymous reviewer for helpful suggestions and comments on this paper. This paper was presented in different versions at the 31st IGG in February 2005 in Rome and at the 28th GLOW Colloquium in March 2005 in Geneva. I thank these audiences for insightful comments and discussions. All remaining errors are mine.

Notes 1. An anonymous reviewer points out that IPP-infinitives can also show up in main verb uses of modals in Swiss German, as is illustrated in (i). In this dialect, the IPP-infinitive replaces a participle that is lost in the lexicon. The important point of the data in (1), however, is that in this environment the IPPeffect shows up, even when the participle is still available in the lexicon of a variety, as is the case in the Standard language. (i) Das han ich nöd wele (Swiss German) Das habe ich nicht gewollt (Standard) this have I not want-IPP/wanted-PP 2. An anonymous reviewer points out that this approach is supported by the occurrence of prefixless participles in certain IPP-environments in Dutch.: instead of the expected form zijn (‘be’) we find the strong participle wezen. A similar case occurs in the German passive, as is illustrated in (i). The regular appearance of infinitives in IPP-context with modals can be related to the fact that modals lost their strong participles in late Middle High German (cf. Hinterhölzl (to appear)). (i) weil die Maria geliebt (*ge) worden ist since the Maria loved been is ‘since Maria has been being loved’

The Phase Condition and cyclic Spell-out

257

3. The name Status is taken from Bech (1955/1983) who compares the status of non-finite verbs with Cases in the nominal domain (StatP can be taken as a generalization of Rizzi’s (1997) FinP). 4. An anonymous reviewer raises the question, why tAspP3 cannot be rendered via alternative Spell-out as lesen in (16) and (17). In section 5, I will discuss prosodic restrictions on Spell-out options in verb clusters. In the standard language the above option is excluded by the condition in (28b). 5. See also Ackema (2004) for a discussion of the issue of whether verb particles can climb within the verb cluster. 6. The idea that prosodic factors condition the linearization of verb clusters at least goes back to Haegeman and van Riemsdijk (1986), where it is argued that the heaviness (of a branching node in the cluster) determines whether flip is optional or obligatory. See also Williams (2004) and Wurmbrand (2004a) for more discussion on the linearization of verb clusters.

References Ackema, Peter 2004 Do preverbs climb? In Verb Clusters – A study of Hungarian, German and Dutch, Katalin E. Kiss and Henk van Riemsdijk (eds.), 359–393. Amsterdam: John Benjamins. Bech, Gunnar 1955/83 Studien über das Deutsche Verbum Infinitum. Tübingen: Niemeyer. Cinque, Guglielmo 2001 Restructuring and the order of aspectual and root modal heads. In Current Studies in Italian Syntax: Essays Offered to Lorenzo Renzi, Guglielmo Cinque and Giampaolo Salvi (eds.), 137–155. Amsterdam: Elsevier. Chomsky, Noam 1998 Minimalist explorations. Ms., MIT. Cambridge, MA. 2001 Derivation by phase. In Ken Hale: A Life in Language, Michael Kenstowicz (ed.), 1–52. Cambridge, MA: MIT Press. Demske, Ulrike 2001 Zur Distribution von Infinitivkomplementen im Althochdeutschen. Linguistische Berichte 9: 61–86. Den Besten, Hans 1977/83 On the interaction of root transformation and lexical deletive rules. In On the Formal Syntax of the Westgermania, Werner Abraham (ed.), 47–131. Amsterdam: John Benjamins. Enç, Mürvet 1986 Towards a referential analysis of temporal expressions. Linguistics and Philosophy 9: 405–426.

258 Roland Hinterhölzl Guéron, J. and T. Hoekstra 1988 T-chains and the constituent structure of auxiliaries. In Constituent Structure, Anna Cardinaletti, Guglielmo Cinque and Giuliana Giusti (eds.), 35–99. Dordrecht: Foris. Grimm, Jacob 1819 Deutsche Grammatik. [Reprint 1898/1969] Göttingen: Gütersloh. Haegeman, Liliane and Henk van Riemsdijk 1986 Verb projection raising, scope, and the typology of rules affecting verbs. Linguistic Inquiry 17: 417–466. Haider, Hubert 1990 Topicalization and other puzzles of German syntax. In Scrambling and Barriers, Günther Grewendorf and Wolfgang Sternefeld (eds.), 93–112. Amsterdam: John Benjamins. Halle, Morris and Alec Marantz 1993 Distributed Morphology and the pieces of inflection. In The View from Building 20, Ken Hale and Jay Keyser (eds.), 53–110. Cambridge, Massachusetts: MIT Press. Hinterhölzl, Roland 1997 Infinitival tense, verb raising and verb projection raising. In NELS 27: Proceedings of the North East Linguistic Society, Kiyomi Kusumoto (ed.), 187–201. Amherst: GSLA. 1999 Restructuring infinitives and the theory of complementation. Ph.D. diss., University of Southern California. to appear Scrambling, Remnant Movement and Restructuring in West Germanic. Oxford: Oxford University Press. Kiss, Katalin E. and Henk van Riemsdijk (eds.) 2004 Verb Clusters – A study of Hungarian, German and Dutch. Amsterdam: John Benjamins. Koopman, Hilda and Anna Szabolcsi. 2000 Verbal Complexes. Cambridge, MA: MIT Press. Rizzi, Luigi 1997 The fine structure of the left periphery. In Elements of Grammar. Handbook in Generative Syntax, Liliane Haegeman (ed.), 281–337. Dordrecht: Kluwer. Williams, Edwin 2004 The structure of clusters. In Verb Clusters – A study of Hungarian, German and Dutch, Katalin E. Kiss and Henk van Riemsdijk (eds.), 173–201. Amsterdam: John Benjamins. Wyngaerd, Guido van den 1994 IPP and the structure of participles. Groninger Arbeiten zur Germanistischen Linguistik. 37: 265–276.

The Phase Condition and cyclic Spell-out

259

Wolf, Henk 1996 Changes in the Frisian verbal complex. Ms., Fryske Akademy, Leeuwarden. Wurmbrand, Susi 2001 Infinitives: Restructuring and Clause Structure. Berlin /New York: Mouton de Gruyter. 2004a West Germanic verb clusters: the empirical domain. In Verb Clusters – A study of Hungarian, German and Dutch, Katalin E. Kiss and Henk van Riemsdijk (eds.), 43–85. Amsterdam: John Benjamins. 2004b Two types of restructuring – lexical vs. functional. Lingua 114: 991– 1014. 2006 Verb clusters, verb raising and restructuring. In The Blackwell Companion to Syntax, Martin Everaert and Henk van Riemsdijk (eds.), 229–343. Oxford: Basil Blackwell. Zwart, Jan-Wouter 1993 Dutch Syntax. Ph.D. diss., University of Groningen.

Parallel phases: a study on the high and low left periphery of Old Italian Cecilia Poletto

1. Introduction In this work I will investigate a set of OV constructions in Old Italian (OI), a phenomenon which has been up to now neglected in the literature on Old Romance languages. Some cases of OV structures can be analyzed as V2 cases, with the Object in (some) SpecC position and the inflected verb in the corresponding C° head. There are however further cases which require a different treatment as the verbal form involved is a past participle, not an inflected verb, as shown in (1): (1)

Allora il cavalero,che ’n sì alto mestero avea la mente misa Then the knight that in so high work had.3sg his mind set (BL, Tes., v. 1975)

I intend to put forth an analysis in terms of movement of the XP to a low Focus (or Topic) position located to the left of the low vP phase, similar to (at least one type of) scrambling studied in the Germanic languages (see among others Grewendorf 2005). That such a low left periphery exists and that it includes Focus and Topic positions has been proposed in recent work by a number of authors (notably Jayaseelan 2001; Belletti and Shlonsky 1995; Belletti 2004 for modern Italian; Paul 2002 for Chinese; among others). I will adopt this proposal and show that OI displays an interesting parallel between the high left periphery located in the CP layer at the external border of the high CP phase and the low left periphery, so that it is possible to hypothesize that the features of a functional head as Focus are parametrized as phase independent properties. Whenever a Focus head is inserted in the syntactic structure (whether in the high or in the low phase) it is bound to be either strong or weak depending on the language, yielding uniformity of behavior across phases. If this view is correct, it predicts that also in the DP phase similar phenomena should be found, and that whenever the

262 Cecilia Poletto property is weakened and then finally lost, all the phenomena connected to it in each phase are weakened and lost in the same way as well: we will see that V2, scrambling and DP internal movement in the Renaissance period are weakened1 in a parallel way. Hence, our prediction is borne out. If the general view is correct, it could open up interesting possibilities to translate the old notion of parameter of the GB approach The article is divided as follows: in section 2 I briefly sketch the peculiar type of V2 displayed by OI, in section 3 I present the phenomenon of scrambling. Section 4 contains an explanation of Egerland’s (1996) generalization concerning the relation between OV and past participle agreement patterns and the core proposal that properties of functional heads must be stated as phase independent properties. Section 5 contains a discussion on quantifiers, which behave differently from definite DPs. Section 6 shows that the high and the low left periphery behave alike with respect to three phenomena already discussed for the CP in section 2. In section 7 I show that the same type of scrambling is found as expected also within the DP phase and in section 8 that Renaissance Italian (RI) has a restricted usage of V2, scrambling in IP and scrambling in DP as expected and that these phenomena are sensitive to the same type of factors, like the presence of a modal verb or the presence of another operator, like a wh-item in relative clauses. Section 9 concludes the article.

2. The high left periphery of the clause As is well known since Benincà (1984), OI shows some (but not all) of the typical correlates traditionally associated with the V2 property, namely subject inversion between the auxiliary and the past participle (not to be confused with free subject inversion, which is still possible in Modern Italian (MI) and occurs after the past participle).2 This property has been lost in the Renaissance and is not found in MI. (2)

a. quali denari avea Baldovino lasciati loro which money had.3sg Baldovino left them

(Doc. fior.: 437)

b. …primieramente avea ella fatta a llui ingiuria …for first had.3sg she done to him injury (BL, Rett.: 116) Cases like those in (2) are standardly analyzed as movement of the inflected verb to a C° position, as also assumed for Germanic V2.

Parallel phases: a study on the high and low left periphery of Old Italian

263

The licensing of pro drop is submitted to different requirements in OI with respect to MI and shows that the inflected verb is moving higher in OI than in modern Romance: pro drop in OI is sensitive to the main versus embedded asymmetry: pro drop is found in main clauses but not in embedded ones, where a subject pronoun is always realized (see Vanelli 1987 for a detailed discussion on subject pronouns in embedded contexts in Old Italian). (3)

E così ne provò _ de’ più cari ch’elli avea. and so of-it tested.3sg _ of-the most dear that-he had.3sg (Testi fior.: 74)

The standard analysis of this asymmetry is that pro can only be licensed when the verb has moved to the CP layer (cf. Benincà 1984 for OI, and Roberts 1993 for Old French). On the other side, OI does not display any linear V2 restriction typical of V2 languages, as V3 cases are frequently attested3. (4)

a. E dall’ altra parte Aiaces era uno cavaliere franco and on the other side A. was a knight courageous (BL, Rett.: 94, r. 7) b. E la reina Artemidora di Alicarnasso,che in adiuto di Serses and the queen A. of Alicarnasso who in help of Serses era venuta, francamente si mescolò nella battaglia was come courageously herself mingled.3sg into the battle (BG, Or.: 92, r. 1)

I will adopt here Benincà’s (to appear) proposal that in OI the verb moves to the head of an operator projection labeled as FocusP in (5) and located in the lower portion of the CP layer, leaving the higher Topic positions available for other XPs yielding cases of V3.4 The structure of the left periphery of OI according to Benincà (to appear) is the following: (5)

[Force C°[Relwh C°]/{Frame [ScSett][HT] C°}{TOPIC[LD] [LI] C°} {Focus [I Focus][II Focus]/[Interrwh ] C°}[Fin C° ] }

The target of verb movement in OI is a projection inside the Focus field, higher positions can only be occupied by Topics (no more than one Focus can be realized due to minimality reasons). Another interesting peculiarity of OI is that it makes extensive use of V1 constructions (cf. (6a)), which were much more widespread also in Old

264 Cecilia Poletto Germanic languages with respect to their modern counterpart (see Fuss 2005). However these cases are analyzed – whether there is a narrative null operator (corresponding to ‘then’) as proposed for their Germanic counterpart, or whether V1 is due to movement of the verb to the higher Topic field – it is well know that in these cases clitics are obligatorily found after the inflected verb in enclisis (this is known as a case of the Tobler Mussafia law), as shown in (6b):5 (6)

a. Avemo ditto che è rettorica have.1pl told what is rhetorics b. Leggesi di Salamone che… Reads-one of Salomone that

(BL, Rett.: 5, r. 17) (Novellino: 138, r. 1)

Therefore, OI displays the following properties connected to the CP left periphery of the clause: a) V2 constructions in which an XP is preposed to the inflected verb b) V3 constructions in which a number of Topics could precede the V2 structure c) V1 with subject inversion d) Pro drop licensed by V to C movement e) Obligatory enclisis in V1 constructions We will see that similar properties also hold for the scrambling phenomenon.

3. XP V as scrambling to Focus Once we factor out the cases of XP V ordering triggered by V2, a number of residual cases are still to be explained, the clearest cases being those with the object located between an auxiliary verb (in Focus°) and the past participle. Examples like those in (7) cannot be analyzed as movement of the object to the high Focus position, which is in this case occupied by the subject preceding the auxiliary verb:6 (7)

a. i nimici avessero già il passo pigliato the enemies had.subj.3pl already the pace taken (BG, Or.: 88, r. 15)

Parallel phases: a study on the high and low left periphery of Old Italian

265

b. ch’egli avea il maleficio commesso that he had.3sg the crime committed (BG, F. di rett.: 31, r. 12-13) c. dice che poi àe molto de ben fatto in guerra et in pace. says that then has a lot of good done in war and in peace (BL, Rett.: 26, r. 22) d. il quale da che ebbe tutto Egitto vinto… whom since had.3sg all Egypt won… (BG, Or.: 83, r. 15) On the other side, it is not possible to assume that OI was an OV language as the unmarked word order found in the texts is identical to that of MI and it is typical of modern Romance languages, which are all VO: (8) a. tenea un savio greco in pregione kept.3sg a wise Greek in prison (Novellino: 125, r. 6) b. fece menare il destriere al campo let.3sg lead the horse to-the camp (Novellino:126, r. 13) c. Molto onoroe la donna nel parto A lot honoured.3sg the woman in-the childbirth (Novellino: 234, r. 7) d. Torquato, consolo di Roma, fece per iustizia tagliare Torquato, consul of Rome, had.3sg for justice cut la testa al figliuolo the head to-the son (F.V.F.: 113, r. 2–3 ) A revealing feature of the OV construction is that not only direct objects, but any type of internal argument (including passive subjects) can be found to the left of the past participle: (9)

a. Ed essendo dell’ unico guernimento già ispogliato and being of-the only ornament already stripped (BG, Or.: 411, r. 1) b. quello che per uso è già dagli antichi servato what that usually is already from-the ancients kept (BG, Veg.: 108, r. 25–26)

266 Cecilia Poletto c. Non crederei che fosse per voi rotto. not would-believe.1sg that was by you broken (Fiore.: 442, r. 11) d. ch’elli è a fine venuto that he is to end come (Trist. Ricc.: 397, r. 17) e. avegna che neuno possa buono advocato essere né perfetto happens that noone can.subj.3sg good advocate be nor perfect (BL, Rett.: 147, r. 1) f. perciò che quelli cui conviene udire sono già udendo fatigati for that those that have to hear are already listening tired (BL, Rett.: 193, r. 19) g. se l’avessi a mente tenuto if it-had.subj.1sg in mind kept (BG, VeV: 16, r. 5) h. Assa’bene, quando sono di te acompagnata very well when am by you accompanied (BG, VeV: 33, r. 2) i. E quand’ebbi cosí chiaramente a ogni cosa risposto and when had.1sg so clearly to everything answered (BG, VeV: 37, r. 24) j. comandò questo giovane che fossero tutte quelle genti menate ordered.3sg this young that were all those people led (Novellino: 143. r. 24) The same is true for adverbials (either formed by a single word or by a complex constituent) and verbal modifiers in general which in their unmarked order usually occur after the past participle:7 (10) a. e holla già molte volte letta nella Bibbia and have-it already many times read in-the bible (BG, VeV: 15, r. 22) b. a quelli che sono già avanti iti to those that are already forward gone (Tes. volg.: c350, r. 2) c. da tutta la gente sarai scarso tenuto of all the people will-be.2sg poorly considered (BL, Tes.: v. 1561 ) d. Poi lo fece fuori trarre that him made.3sg outside take (Novellino: 158, rr. 6–7)

Parallel phases: a study on the high and low left periphery of Old Italian

267

e. il cavaliere era molto bene costumato the knight was so well educated

(Novellino: 311, r. 3)

f. Quand’ ebbero così ordinato when had.3pl so ordered

(Novellino: 349, r. 1)

Given that this is not the unmarked word order of OI, I propose that the ordering found in (7), (9) and (10) is due to a movement operation similar to scrambling targeting a Focus position located in the low IP area. Following Belletti (2004) I take this position to be in the left periphery of the low vP phase.8 This is clearly a position dedicated to XPs (no clitic or weak elements can occur there as they cannot be focalized), and it hosts virtually any type of constituent, all types of arguments as well as adverbials and verbal modifiers. This lack of “specialization” in the element that they can host is a feature typical of left peripheral positions, where any type of XP can be moved. Suppose further that the “left periphery” of each phase is construed in the same way, namely by merging a “Topic-Focus” field before the highest projection “closing up” the phase (corresponding to Force in the CP system proposed by Rizzi 1997; see again Belletti 2004 for MI, among others). Therefore, all examples in (7) (9) and (10) are to be analyzed as cases of movement to a SpecFocus position in front of the past participle but lower than the subject position (SpecAgrS or SpecT in a strict minimalist framework).

4.1. The relation between scrambling and past participle agreement Egerland (1996) examines only cases of direct object scrambling and notices that the OV order is diachronically related to the possibility of past participle agreement with post-participial objects, when post-participial agreement is lost, the OV order is lost as well: (11) a. quando egli avea già fatti molti miracoli when he had.3sg already done.pl many miracles.pl (Tes. volg.: a258, vv. 3–4) b. E quando il notaio ha letta la proposta and when the notare has read.f the proposal.f dinanzi a’ consiglieri to the counselors (Tes. volg.: d335, v. 17)

268 Cecilia Poletto c. c’ ha rifiutata la nobile città di Giadres et ha that has refused.fem the noble city.fem of Giadres and has preso li marchi taken.m the money.pl (Novellino: 133, r. 3) Moreover, he shows that if the order is VO, past participial agreement is optional (as is shown by preso in (11c)) while it is obligatory when the order is OV. This observation is confirmed by modern languages, which still have OV orderings, like Friulian. As is shown below, only three of the four logical possibilities are attested: the sequence OV (12d) without agreement is not possible (see Paoli 1997, for a detailed analysis of this phenomenon in modern Friulian). (12) a. O ai lis sigaretis dismenteadis I have the cigarettes.pl forgotten.pl b. O ai

dismenteadis lis sigaretis

c. O ai dismentea:t lis sigaretis I have forgotten.sg the cigarettes.pl d. *O ai lis sigaretis dismentea:t We can reformulate Egerland’s observations in the form of two descriptive generalizations as in (13) and (14):9 (13) When past participial agreement is reduced (and finally lost), OV is reduced (and finally lost) as well (14) Past participle agreement is obligatory with the OV order, not with the VO order Both generalizations can be instrumental to shed light on the account of scrambling in OI. Generalization (14) is to be considered a special case of a descriptive generalization proposed by Guasti and Rizzi (2002): basing on cases of subject agreement they note that movement and morphological richness are connected and propose the following generalization: (15) If a feature is checked in the overt syntax, then it is expressed in the morphology10 Generalization (15) captures all cases in which a preverbal subject shows stronger agreement with respect to a postverbal subject: when the subject

Parallel phases: a study on the high and low left periphery of Old Italian

269

moves to the specifier of AgrS thus checking the AgrS feature in overt syntax, morphology reflects this process, hence the fullest agreement pattern is always selected. This is not always the case for postverbal subjects, where there is no overt checking of the feature in the syntax. Thus, postverbal subjects differ cross-linguistically as they can either have full agreement, a reduced form of it, or no agreement at all. Given Egerland’s generalization, confirmed by Friulian data, this is also true for objects, when the object is checking a given feature in overt syntax, this is obligatorily reflected in the past participle agreement morphology. Therefore, OV structures are to be intended as cases of overt movement through a position encoding strong features for object agreement11 on their way to Focus. The derivation of OV structure is given below (for sentence (7b)): (16) [CP che [AgrS [SpecAgrS egli] [AgrS° avea] [FocusP [SpecFocus il maleficioj] [Focus° commessoi] [AgrOP [SpecAgrO tj] [AgrO ti] …[VP [V° ti] [tj]]]]]] VO structures, on the contrary, are ambiguous between a movement and a non movement analysis. When there is no agreement, the null hypothesis is that no movement has applied to the object DP: given that movement to Focus is not obligatory, the object DP can stay in its base position. On the other hand, when there is object agreement, I assume that the object DP moves to the SpecAgrO position. This means in turn that the past participle must have moved higher than the SpecAgrO position where the DP has moved, in order to derive the ordering Participle+agr Object. I assume that this position is precisely the head of the Focus projection located in the low left periphery. Hence, the derivation of VO orders is (17a) for non agreeing cases and (17b) for agreeing cases (I take (11b) as a sample sentence): (17) a. [CP quando [AgrS [SpecAgrS il notaio] [AgrS° ha] [FocusP [Focus° letto] [AgrOP [SpecAgrO ] [AgrO ti] …[VP [V° ti] [la proposta]]]]]] b. [CP quando [AgrS [SpecAgrS il notaio] [AgrS° ha] [FocusP [Focus° lettai] [AgrOP [SpecAgrO la propostaj] [AgrO ti] …[VP [V° ti] [tj]]]]]] In (17a) only the past participle moves by virtue of the head movement constraint (or any minimalist version of it) through the head of AgrO to reach Focus°, the object does not move. Hence, no agreement is possible. In (17b) on the contrary, not only the participle moves to AgrO, but also the

270 Cecilia Poletto object, triggering agreement. The object is then not further moved to Focus, while the participle does. Notice that this derives straightforwardly not only generalization (14) but also (13): if a strong agreement is always an instance of movement in OI, then loss of movement directly accounts for loss of strong agreement. Alternatively, in a minimalist framework where the agreement relation is no more conceived as deriving from the structural relation between a head and its specifier, the agreement pattern with a post-participial object might be derived via an Agree operation applying freely between the features of the object and those of the past participle.12 The main difficulty of this approach is that the Agree operation seems as such too unconstrained to account for the fact that Italian has lost at the same time OV and post-participle agreement (namely generalization (13)) and for the fact that the Agree operation is now blocked in MI for postparticipial objects, but not for other types of objects like clitics or passive subjects. In other words, the operation Agree as it is usually stated is too liberal to account for the diachronic path we observe here. One possible way to constrain the operation is to assume that there is only an indirect link between agreement and syntactic movement and hypothesize that Agree only applies to an in situ object when the language also has a corresponding movement operation for the object. However, we have to assume that agreement with a clitic element or with a passive subject in MI is not sufficient to license Agree with post-participial DPs: only scrambling is. Moreover, the pattern of MI, where agreement is maintained only with those elements that have been moved to the left clearly shows that movement and agreement are directly and not only indirectly linked. Therefore I will discard this analysis here and will continue to assume that strong agreement and movement are directly related (as shown in (17)). Notice that if we are on the right track with the idea that scrambling in OI is movement to Focus, there is an interesting parallel between V2 structures and OV structures: in both cases an XP is moved to a SpecFocus and the verb is moved to the head of Focus. If both Focus heads have the same strong property we can hypothesize that the properties of the low Focus position in IP and in the high CP phase remain constant across phases. It is a fairly standard assumption that the old notion of “parameter” has to be formulated in terms of properties of functional heads. This, however, often leads to postulating a parameter for a single syntactic construction, while the original notion of parameter has proved extremely powerful in the GB framework because it accounts for a number of syntactic constructions

Parallel phases: a study on the high and low left periphery of Old Italian

271

under the same abstract property. Assuming that properties of functional heads remain constant across phases is a new way to account for different phenomena (in our case V2 and scrambling) on the basis of the same abstract feature. I will therefore assume the strongest version of this proposal, namely that a parameter corresponds to the “activation” of a given functional head F°, whose features must be checked in the computational component wherever it is merged. Hence, Focus in OI maintains the same property throughout all the phases where it occurs: more specifically it must be filled by a verbal head in all phases, the inflected verb fills the Focus° of the high CP phase, the past participle fills the Focus° of the low vP phase.13 In the high CP-phase it triggers the subject inversion phenomenon typical of V2 contexts (and the others phenomena seen in section 2). In the vP phase it triggers post-participial agreement, OV and more generally XPV. This means that the high and the low left periphery should behave alike in all respects and leads to a number of expectations concerning the low left periphery on the basis of the constructions illustrated in section 2 for the high left periphery. More specifically, I examine here the predictions this hypothesis makes concerning three constructions already discussed above: a) as there are several left dislocated items in the high left periphery, this analysis predicts that more than one scrambled element above the past participle should be possible as well. b) as there exists V1 in the high left periphery, this should be the case also in the low left periphery. c) as V1 constructions always trigger enclisis in the high phase (see section 2), it should also be the case in the low phase. Two additional predictions are made by the analysis proposed here, the first one concerns OI itself: if the strong feature of Focus is phase-independent we should find similar phenomena in the DP as well, given that DP is also a phase (see Giusti this volume). The second prediction is a diachronic one, namely that if V2 and XP V are effects of the same abstract property, they should behave alike in the diachronic process of progressive loss.14 We will show in section 6, 7 and 8 that these predictions are correct.

272 Cecilia Poletto 5. Quantifier movement Not all cases of OV constructions have to be treated as movement to a SpecFocus (or SpecTopic) position. Analyzing the frequencies of the occurrences of OV orderings it appears evident that there is a major difference concerning the type of element moved, as QPs behave differently with respect to DPs. Reading the texts it seems at first sight that the QP tutto ‘all, everything’ is found much more frequently to the left of the past participle than definite object DPs. This however remains just a frequency effect unless various cases of tutto are distinguished. A further inquiry has been done considering 2000 relevant occurrences of tutto in the whole corpus.15 Within the sample I have split the cases of bare tutto with respect to the cases of tutto modified by a relative clause or followed by a DP. The results are that when tutto is modified by a relative clause it never raises before the past participle: (18) a. e ffue fatto tutto e ccioe che lo ree comandoe and was done all that that the king required.3sg (Trist. Ricc.: 25) b. bene servita di tutto cioe ch’ ella comanda well served of everything that that she requires (Trist. Ricc.: 136) When tutto is followed by a DP (preposing seems more frequent with a pronoun like questo ‘this’) it can be found to the left of the past participle, as shown in (19), but in the majority of the cases, it is found to the right of the past participle, as in (20): (19) a. e poi ch’ elli àe tutto questo trovato per lo suo pensame and then that he has all this found for his own thinking (BL, Rett.: 74) b. quale da che ebbe tutto Egitto vinto since that had.3sg all Egypt won (BG, Orosio: 83, r. 15) (20) a. àe insegnato per tutto il libro insine a questo luogo has taught for all the book until to this place (BL, Rett: 140) b. à(n)no ve(n)duto tutto i loro podere have.3pl sold all the their property (Doc. fior.: 226)

Parallel phases: a study on the high and low left periphery of Old Italian

273

c. e hannovi messo tutto loro ingegno e forza and have-there put all their intelligence and force (BG, Orosio: 24) Finally, when tutto is a bare quantifier, it is always preposed if it corresponds to the direct object (as shown in (21a–c)), whereas it can be (less frequently) found after the past participle when it is introduced by the preposition di (as in (21d)): (21) a. chi ’l tene del tutto in sé celato who it-keeps of-the all in himself concealed

(BL: 178)

b. ’avrei sovente a tutto tuo conforto scritto would-have.1sg frequently to all your comfort written e risposto and answered (BL, Somm.: 197) c. seguire Idio chi à tutto venduto follow God who has everything sold d. e ben fornito di tutto and well furnished of everything

(Fiore XIII: 232) (Tes. volg.: 98)

Thus, it seems that not only the status of tutto as bare vs. non bare quantifier influences the frequency of preposing, but also the presence of the preposition di. No cases of tutto as direct or indirect object after the past participle have been found. It is well known that there are positions dedicated to quantifiers to the left of the VP in languages like German and Dutch. We can therefore hypothesize that bare tutto has to raise to a specific position to the left of the past participle which is specified for checking the features of a universal quantifier. Similar proposals for quantifiers have been made by Beghelli (1995) and Beghelli and Stowell (1997). A possible alternative would be to consider that the position in question is a case position, given that movement is obligatory with direct and indirect object, but optional with a PP introduced by di ‘of’. This however, would imply that all direct and indirect objects have to raise to the case position, which is clearly not true (see above the discussion on scrambling). The only solution I see to account for the fact that movement is obligatory only a) when the XP moved is a bare quantifier and b) when it is a direct or indirect object, is to assume that the field in which QPs are hosted is sensitive to case distinctions, so that there

274 Cecilia Poletto are distinct positions dedicated to universal quantifiers according to their case. The picture on quantifiers is complicated by the fact that, if we examine another bare quantifier, namely niente/nulla ‘nothing’, we see that it behaves exactly in the opposite way: when the cases of niente as emphatic negation which could be translated with ‘at all’ (illustrated in (22)) are excluded from the sample, the results are that the quantifier is never moved to the left of the past participle.16 (22) a. Non ti annoi niente a visitare i malati not yourself bother.2sg nothing to visit the sick

(ZB: 69)

b. che Dio non ode niente l’orazione that God not hears nothing the preach

(ZB: 86)

c. falso legno non è niente buono false wood not is nothing good

(ZB: 93)

(23) a. te n’avea detto niente to-you of-it had.3sg told nothing

(Paolino P.: 31)

b. non avea saputo niente not had.3sg known nothing

(Tav. rit: 182)

One might think that the basic difference between tutto and niente lies in the movement of the quantifier: while tutto moves, niente does not. This seems at first sight rather strange, given the hypothesis made here, namely that a QP moves to a specialized checking position. The question is, if there is a specialized position for universal quantifiers, why is there none for negative QPs? Further data showing that niente does not probably remain in the merge position of the direct object are the following: (24) a. e non vede in lui niente perchè sia degno and not sees in him nothing because be.subj.3sg worth del pane of-the bread b. dicessono a llui niente tell.subj.3pl to him nothing

(ZB: 25) (ZB: 37)

c. sapendo di Paolo niente, venneli uno cotale pensiero knowing of Paolo nothing came.3sg-to-him a such thought (GC, Vite: 88)

Parallel phases: a study on the high and low left periphery of Old Italian

275

We have seen above (see (8)) that the unmarked word order of the internal arguments in OI is the same found in MI, namely accusative-dative. However, when niente is the direct object, it occurs after the dative or genitive PP, as shown in (24). Hence, we have to explain two facts: a) the fact that niente occurs after the past participle and b) the fact that niente occurs after the indirect object PP (which is not the case for non quantified objects). A tentative way to analyze these data would be to consider the two facts as related and assume remnant movement of the VP to the left of the position to which niente has moved in the first step of the derivation. If this is correct, both tutto and niente move to the left of the thematic position, and probably both target a position specialized for universal and negative quantifiers respectively. The movement of niente however, triggers remnant movement of the VP, while tutto does not. Whether this is due to the negative status of niente is left to future research. Although the exact derivation of these cases still remains to be investigated, there is a fairly uncontroversial conclusion that we can draw, namely that the movement of bare quantifiers is different from the type of XP movement that we have analyzed in scrambling to the low left periphery in the previous sections. I will therefore leave aside all cases of bare quantifiers in the inquiry on scrambling and discuss the predictions considering the parallelism among phases.

6. First prediction: the analogy of the high and the low phase As seen in section 2, OI displays frequent V3 constructions, where one (or several) Topic precedes the Focus element. This is, as expected, also found in the low left periphery: (25) a. ed ha’mi la cosa molte volte ridetta and has-to-me the thing many times retold

(BG, Tratt.: 131)

b. E quand’ebbi cosí chiaramente a ogni cosa risposto and when had.3sg so clearly to everything answered (BG, VeV: 37, r. 24) Examples in (25) show that two elements occur to the left of the past participle. In (25a) they are the direct object la cosa and the adverbial XP molte volte, in (25b) the indirect object a ogni cosa and the adverbial XP così chiaramente. Notice that this type of adverbial XPs usually occur after the

276 Cecilia Poletto past participle, as shown in (26), hence cases like (25) are genuine cases of multiple scrambling of an argument and an adverbial:17 (26) Tullio dice, che ‘l fatto è contato chiaramente T. says that the deed is told clearly (Tes. volg.: 138, r. 6–7) The second type of construction frequently found in the high left periphery of the clause is V1. Trivially, cases of V1 are also found in the low left periphery, where all objects and lowest adverbs18 are located after the past participle, which is then the first element of the phase. In other words, cases of V1 in the low phase are “normal” cases of VO: (27) a. ciò che savi avevano detto intorno alla retorica what that wise had.3pl said about the rhetorics (BL, Rett.: 7 r. 19) b. …fue isbandito della terra …was banned from-the earth (BL, Rett.: 7 r. 9) c. poi che Tullio ae advisati li mali since T. has seen the evils

(BL, Rett.: 12 r. 7)

The third construction found in the high left periphery is the one with V1 and enclisis of a past participle. As for the low phase, it seems difficult to test this possibility as in general it is not possible to leave the clitic inside the low phase, i.e. OI has obligatory clitic climbing to the high phase.19 However, absolute participial clauses of non-unaccusative verbs have been analyzed by Belletti (1990) as truncated structures corresponding to AspP in MI. Hence, we can use those cases as a test for our hypothesis: (28) a. trovò l’ arme del re Meliadus, che lli avea fatta sì found the weapons of king M. that he had.3sg done so bella deliberanza, e donatogli: et era suo mortale nemico nice disposal and given-to-him and was his mortal enemy (Novellino: 268 r. 21) b. Fatto ha chiamare made has call faccia chiamare le make call the

Licomede re, e dettogli che L. king and told-him that donne women (Arm., Fior.: 546)

Parallel phases: a study on the high and low left periphery of Old Italian

277

As expected, in these cases enclisis is the rule just like in the high phase. We can conclude that the high and the low left periphery really behave alike with respect to the phenomena observed as predicted by the parallelphases hypothesis.

7. Second Prediction: the left periphery of the DP phase The second prediction made by the hypothesis that functional properties are phase independent is that within the DP area scrambling phenomena are also possible (and are ungrammatical in MI, which has also lost V2 and IP scrambling). One interesting fact about OI is that it is possible to have modified adjectives in prenominal position (contrary to MI): (29) a. domandò se avesse più care pietre asked if had.subj.3sg more valuable stones (Novellino: 123, 54) b. qual ti sembra di più ricca valuta? which to-you seem of more rich value? (Novellino: 127, 28) c. Democrito fue molto grande filosofo. Democrito was very great philosopher

(F.V.F.: 106, r. 2)

How to interpret examples like (29) becomes clear when we notice that adjectives only occurring in postnominal position in MI can occur to the left of the noun: (30) a. la quale guardava al figliuolo piccolo del morto fratello, whom looked.3sg at-the child young of-the dead brother (BG, Or.:148, r. 7) b. e dagli usati uomini and from-the experienced men (BG, Veg.: 167, r. 167) c. il ben usato cavaliere disidera battaglia the well behaved knight wants battle (BG, Veg.: 70, r. 6) We can hypothesize that adjectives that are located in a postnominal position in MI can move to the left of the noun due to a DP internal scrambling process, and that modified adjectives do the same in OI. Empirical evidence that the basic position of the adjective is much lower in the structure than the one where the adjective appears and that, in other words, we do not have

278 Cecilia Poletto to do here with a language in which the noun raises very little and most adjectives are found to its left (like Germanic languages for instance) is provided by examples like the following: (31) a. e di gentile aspetto molto and of kind appearance very b. e ciò non è propia natura di cavallo and this not is own nature of horse

(Dante, Vita Nuova: cap. 8, par. 1, v. 11) (Novellino: 128, r. 67)

In (31a) the adjectival modifier molto ‘very’ has remained to the right of the head noun aspetto ‘appearance’ and only the bare adjective gentile ‘kind’ has been extracted out of the complex adjectival phrase and scrambled to the left of the noun. In (31b) the portion of the AdjP left to the right of the noun is its complement PP di cavallo. Cases like (31) clearly show that the prenominal position is not a base generated one, but it is due to movement of the entire modified AdjP or of part of it. An even more interesting case is provided by examples like (32): (32) a. quando vi dissi del cavallo cosa così meravigliosa when you told.1sg of-the horse thing so wonderful (Novellino: 120, r. 14) b. Sì so le the

come as cose things

quando ordino di ritrarre dell’antiche scritte when order.1sg to draw of-the ancient writings che… that (BL, Rett: 11, r. 18)

These examples show that it is possible to move even the complement PP of a noun (in (32a) del cavallo, in (32b) dell’antiche scritte) to its left. Given that in general object PPs are generated to the right of the noun (OI is not a head-final language), (32) clearly shows that OI had a scrambling process in the DP as well as in IP, thus confirming our prediction.20 The scrambling process can be interpreted as movement to either a Focus or a Topic position in the left periphery of the DP area similar to the one of the other phases .21 8. Third prediction: the loss of the medieval system As discussed above, the hypothesis that IP-scrambling, V2, and also DPscrambling depends in OI from the same abstract property concerning the

Parallel phases: a study on the high and low left periphery of Old Italian

279

strong properties of a Focus head located in the left periphery of each phase predicts that all these properties are lost together at the same time and at the same rate. In order to test the prediction, a XVI century text has been examined and screened for all the cases of subject inversion, IP scrambling, agreement with a post-participial object, and DP scrambling. The text chosen is “Il Principe” by Machiavelli, of which the first X chapters and chapters XX–XIII have been analyzed. The first phenomenon considered is V2: all cases of subject inversion and cases in which new information focus (ungrammatical in MI) is found in first position have been marked. In the whole sample there are only three cases of subject inversion with an auxiliary verb (cf. (33)) and eight cases with modals (six with potere ‘can’ and two with dovere ‘must’). Cases with modal verbs are illustrated in (34): 22 (33) a. Spenti adunque questi capi, e ridotti i partigiani loro blown off then these bosses and reduced the partisan their amici suoi, aveva il duca gittato assai buoni fondamenti friends his had the duke thrown very good foundations alla potenza sua to-the power his (p. 221) b. Aveva adunque Luigi fatto questi cinque errori had then L. made these five mistakes

(p. 197)

c. Mentre che durò la memoria, sempre furono i Romani while that lasted.3sg the memory always were the Romans incerti di quella possessione unsure of that possession (p. 203) (34) a. E deve soprattutto uno principe vivere con i suoi sudditi and must.3sg overall a prince live with the his subjects in modo che … so that… (p. 237) b. E con più facilità se le può un principe and with more easyness for-himself them can.3sg a prince guadagnare gain (p. 205) As for cases in which Focus appears in first position, a feature typical of OI syntax, there exist only four cases of this type (shown in (35)) if we factor

280 Cecilia Poletto out the two adverbs sempre ‘always’ and mai ‘(n)ever’, which are very frequently placed in first position even in contexts where they do not seem to be contrastively focalized (in which case they are still possible in MI as well): (35) a. Confido assai che per sua umanità gli debba essere accetta trust.1sg a lot that for his umanity to-him must be accepted (p. 173) b. Più facilmente si tiene more easily one keeps con il mezzo dei suoi with the means of-the its

una città a city cittadini, citiziens

usa a vivere libera adapted to live free che in altro modo that in other way (p. 204)

c. Come di sopra si disse as above one said.3sg (36) a. Sempre si trova dei malcontenti always one finds some grudging b. Mai si troverà ingannato da lui never himself will-find.3sg cheated by him

(p. 213) (p. 202) (p. 243)

Notice that in non V2 languages like Spanish and Catalan adverbs like those corresponding to OI sempre and mai are always found to the left of the inflected verb and are probably located in a dedicated position inside the Focus layer (see Grava 2005), so they cannot be taken as real indications of a V2 grammar. It has already been noted for Old French (see among others Roberts 1993) that some elements “resist” more than others to the loss of the V2 system, and are found in V2 constructions also in texts which do not present obligatory V2 anymore. Apparently this is also the case in OI.23 Relative clauses deserve particular attention, as V2 cases embedded under the relative complementizer or a wh-item are the overwhelming majority with respect to relative clauses that do not present a V2 effect. (37) a. Quelli i quali per vie virtuose simili a costoro diventano those who along ways virtuous similar to them become.3pl principi princes (p. 210) b. Quelli che di sua qualità gli avevano invidia the ones that of his qualities to-him had.3pl envy

(p. 211)

Parallel phases: a study on the high and low left periphery of Old Italian

c. Quelle armi che vicine lo potevano offendere those weapons that close him could.3pl offend

281

(p. 223)

Apparently, the context of relative clauses is special in maintaining V2 longer with respect to other clause types, and this is probably due to the fact that in these cases the CP layer is already activated by the wh item sitting in a higher relative position (see Rizzi 1997, on the position of relative elements) which nevertheless leave the Focus position free for V2.24 Egerland (1996) proposed that at this stage two grammars were available, one simulating the OI syntax, which is less frequently used and the new one, in which V2 has already been lost. If we are on the right track, it seems that the V2 option still available is subject to specific conditions and very limited in its use. As for IP scrambling, we have a parallel situation: very few cases of scrambling are found in the sample and are all restricted to compound tenses with the auxiliary essere ‘be’ or copular clauses. (38) a. Non è, oltre a questo, la provincia spogliata da’ tuoi ufficiali not is, beyond this, the province stripped by your officers (p. 187) b. E benché dai Cartaginesi fusse due volte rotto and although by C. was twice broken (p. 232) c. Da coloro che saranno in quella malcontenti by those who will-be.3pl in that unhappy

(p. 190)

d. Non consentono che sia tra gli eccellentissimi not consent.3pl that be.subj.3sg among the best uomini celebrato men celebrated (p. 233) e. che senza l’una e l’altra fu da lui conseguito that without the one and the other was by him achieved (p. 233) Two cases with have have been found, both in a construction with a bare noun: (39) a. Che esso abbia con loro obbligo that it have.subj.3sg with them obligation

(p. 246)

b. Altrimenti non ha nelle avversità rimedio otherwise not has in-the misfortune remedy

(p. 240)

282 Cecilia Poletto Most of the exampes are with modals (recall that the same observation has been made for subject inversion in the case of V2) or in causative constructions: (40) a. Lo può con grandissima difficoltà perdere it can.3sg with greatest difficulty loose b. Si possono con più difficultà corrompere one can.3pl with much difficulty bribe

(p. 187) (p. 202)

c. Tanto potette in su tale fondamento edificare ogni edificio somuch could.3sg on this foundation build every building (p. 212) d. Fece da’ suoi soldati uccidere tutti i senatori e made by his soldiers kill all the senators and i più ricchi del popolo the richest of-the folk (p. 232) e. Non abbiano ancora mai potuto ne’ tempi pacifici mantener not have.1pl yet ever could in times peaceful keep lo Stato the state (p. 235) f. Che non fa i fondamenti prima, li potrebbe con una that not does the foundations before them could.3sg with a gran virtù fare dipoi great virtue do afterwards (p. 213) g. Per potere con quelli tenere il papa in freno for can with those keep the pope at bay

(p. 224)

Post-participial agreement is also extremely rare, only five cases are found in the sample: (41) a. Per aver tenuta più lunga possessione in Italia for have kept.f more long possession.f in Italy b. Basta avere spenta la linea del principe che enough have turned.f off the line of-the prince who li dominava them dominated

(p. 93)

(p. 86)

Parallel phases: a study on the high and low left periphery of Old Italian

283

c. Se egli avesse osservate le regole sopradette if he had.subj.3sg observed.f.pl the rules.f.pl aforementioned e tenuti sicuri e difesi tutti quelli suoi amici and kept.m.pl safe and protected all those his friends.m.pl (p. 195) d. Avrebbero sempre tenuti gli altri discosto dalla had always kept.m.pl the others away from-the impresa di Lombardia enterprise of Lombardy (p. 197) e. Si vedrà lui aversi fatti gran fondamenti one will-see.3sg him have-himself done.m.pl great foundations alla futura potenza to-the future might (p. 213) Interestingly, past participle agreement is extremely frequent in relative clauses, which are exactly the same construction in which V2 is still quite consistently found, as discussed above: (42) a. La quale opera io non ho ornata né ripiena di clausole the which deed I not have adorned.f nor filled.f of clauses ample big (p. 173) b. E and che that

quelli fondamenti che gli altri hanno fatti avanti those foundations that the others have done.m.pl before diventino principi become princes (NM, Princ.: p. 213)

c. Di quelli cardinali che lui avesse offesi of those cardinals that he had.subj.3sg offended.m.pl

(p. 229)

We can conclude that scrambling and post-participial agreement are also extremely restricted and are more frequently found in the same contexts still triggering V2, namely with modals and relative clauses. DP scrambling, as expected, is also quite restricted but still present: (43) a. Ed uno de maggiori rimedii e più vivi and one of main remedies and more powerful b. Troverà difficultà grande will-find.3sg difficulty great

(p. 187) (p. 202)

284 Cecilia Poletto Therefore, I conclude that the Renaissance system treats consistently all the phenomena I claim to depend from the same abstract property of a Focus head. The phenomena are all still present, though apparently only in some constructions, which, at least for V2, scrambling and past participle agreement seem to be the same. Further investigation is needed to exactly determine which elements maintain for a longer period the ability of moving to the Focus head (be it in the DP, vP or CP phase), but what can already be seen is that the three phenomena of V2, IP-scrambling and DP scrambling are weakened in a parallel way.

9. A comparative note If the idea that phases are built in a uniform way and that the formal properties of the functional projections are constant across phases is correct, we predict that the phenomena discussed in this work always go together, a prediction which is plainly wrong. As already noted above, there are indeed languages in which V2 and scrambling go together, as German and Dutch, but there are languages like Korean that exhibit scrambling but not V2. However, given a highly split left periphery of each phase, we have to establish which projection V2 (and scrambling) targets, as these are phenomena that can involve different projections, although all located in the peripheral area of a phase. The idea that V2 is a complex phenomenon and can be analyzed as involving different projections is generally accepted in the literature: it is well known that even inside the domain of Germanic languages the projection targeted by V2 can be different (see on this the so called “symmetrical V2 languages” like Icelandic and Yiddish, or the distinction between Mainland Scandinavian languages and German in embedded V2 contexts under bridge verbs, where the complementizers can be expressed or not (see Vikner 1995). Moreover, Zwart (1993) has shown that subject initial sentences are different from sentences beginning with a focalized XP. Whatever the analysis of these differences turns out to be, it is clear that V2 is not a uniform phenomenon in the sense that it always targets a single projection.25 The same is true for scrambling, in the literature we find distinctions on long and short scrambling (languages like German only exhibit short scrambling, while Japanese for instance also exhibits long scrambling outside the simple clause (see Grewendorf and Sternefeld 1990); moreover, sometimes scrambling seems to have the typical properties of A-movement,

Parallel phases: a study on the high and low left periphery of Old Italian

285

in other cases it behaves as A’-movement (and there are cases that display mixed properties as well). Going through the literature on scrambling, the reader gets the impression that different phenomena have been gathered under the same label and that scrambling does not always target the same functional projection. Therefore, before considering the cases mentioned above as real counterexamples to the idea that the same phenomenon can occur across phases (and be termed differently), we need a very detailed analysis of scrambling and V2 phenomena in all languages that allow for them. Another possible problem for the parallel phases idea turns out in connection with MI: how come MI does not allow either scrambling or V2 but the Focus and Topic projections can host XPs both in the high and the low left periphery (see on this Rizzi 1997; Benincà and Poletto 2004; Belletti 2004)? The distinction between OI and MI does not concern the specifier position, which can clearly be filled in both languages, but the head of the Focus (and Topic) projections involved. Following Benincà (2005) I assume that in OI the head of these projections (both in the high and the low phases, and, as we have seen also in the DP phase) is filled by the lexical head (V or N), when an element is realized in its specifier, while this is not the case in MI, where no element is required in Focus° or Topic° and only the specifier of these projections is realized (apart from wh-main contexts, where, following Rizzi (1991) I assume that there is indeed verb movement). The difference between the two languages is precisely a difference in the classical terms of V2, namely whether the head contains strong features or not.

10. Conclusion In this work I have tried to show that OI is not only a V2 language, but also has scrambling to a Focus position internal to the IP layer. Scrambling is both diachronically and synchronically connected to past participle agreement: when scrambling occurs agreement is obligatory, while it is optional when the object occurs in a post-participial position; moreover the two phenomena disappear at the same time. I have adopted an analysis of past participial agreement in terms of syntactic movement (and rejected an analysis in terms of Agree with in situ objects) of the DP to a SpecAgrO position and further movement of the past participle to Focus. If OV orders are analyzed as such, new perspectives open up for a unitary treatment of functional projections that occur in different phases: if

286 Cecilia Poletto Focus is marked strong in OI independently of the phase where it is merged, V2 and scrambling are the two sides of the same coin, only one occurring in the high, the other in the low phase. We have seen that the two phenomena have a number of properties in common, as the possibility of V1 and V3 on a par with V2. The hypothesis of a feature uniformity in all phases further predicts that the same type of reordering phenomena is also found within the DP, which is confirmed by the cases of prenominal adjectives and PPs present in the sample. The last prediction is that all reordering phenomena: V2, IP scrambling, past participle agreement, and DP scrambling are lost at the same time. A close examination of a Renaissance text shows that they all occur mainly in restricted contexts, as relative clauses and modals and very sporadically in other contexts. Such a far reaching hypothesis as the one assuming the uniformity of features across phases must clearly be tested in other functional domains, and in other languages. The first domain of inquiry are other Old Romance languages, to see whether the connection between V2 on one side and IP and DP scrambling on the other is confirmed.26

Acknowledgements This work originally stems from my participation in the Italant project for an OI Grammar directed by L. Renzi and G. Salvi. A version of this work has been presented at the Giornata sull’Italiano antico at the University of Padova, at the research seminar of the Georgetown University and at the XXXI IGG conference in Rome. I thank the audiences of these conferences and Paola Benincà, Verner Egerland, Nicola Munaro, Nicoletta Penello, for insightful comments and discussion. The usual disclaimers apply.

Notes 1. What is meant here by weakened is the fact that the Focus position is not obligatorily occupied anymore, but can only be used by a set of elements with operator properties (as for instance adverbials like sempre ‘always’ and mai ‘never’), which is nonetheless larger than the set of operators possible in SpecFocus in MI (a set which includes essentially only wh-items and contrastively focalized XPs)

Parallel phases: a study on the high and low left periphery of Old Italian

287

2. All examples are taken from the OVI (Opera del Vocabolario Italiano) data base which contains all OI texts from 1200 to 1350. I follow here Lorenzo Renzi and Giampaolo Salvi (forth.) in defining OI as the Florentine variety written in this period. 3. Cases of V4, V5 and so forth are also attested, this is expected under the hypothesis that there are several topic projections located higher than FocusP (cf. Benincà 1995, 2005). Here I will refer to V3 meaning also V*. 4. A number of other phenomena are connected to V2, as null topics and past participle fronting. They are left aside here because the purpose of this section is to illustrate the similarities between the CP left periphery and the low left periphery. 5. In this work I will use enclisis of the clitic pronoun merely as an empirical test to provide evidence for the similarity between the high and the low left periphery, therefore I will leave aside the complex matter of explaining V1 and deriving enclisis. 6. Given that OV cases with a simple verb can always be ambiguous between V2 and scrambling I will restrict the data to cases of compound tenses. 7. The same phenomenon is found with modal verbs and infinitival complements and with causative constructions. 8. Differently from Belletti (2004) and on a par with Paul (2002), I will not assume that the low left periphery is located lower than “lower adverbials” (in the sense of Cinque 1999). The position of this set of projections still requires further research, however see section 8 for some discussion on this issue. 9. For data supporting this generalizations, see Egerland (1996) and section 6. 10. The generalization has to be interpreted in the following way: whenever a language has strong morphology, this has to be used when the DP precedes the verbal head, it is clearly vacuous with respect to languages that do not have any morphology at all. 11. This position could be AgrOP, as originally proposed by Kayne (1991) or another functional projection which has other features, as AspP. I will leave this question open here, although I use the label AgrO for the sake of concreteness. 12. Note that such an hypothesis can also explain Guasti and Rizzi’s generalization: given that movement is a complex operation composed of “match”, “agree” and “pied-pipe”, the fact that agreement is obligatory with movement is straightforward in this account. 13. Note that this is also consistent with the HMC 14. An apparent problem is provided by the equation between V2 and scrambling: if the two phenomena are always connected we should expect them to co-occur in all languages. This is clearly not the case, as there are languages with scrambling (for instance Korean) which do not display V2. The problem here is that the definition of V2 and scrambling have to be qualified in terms of properties of a functional projection. The peculiar type of V2 found in OI is not the “standard” one described for Germanic languages, and as such it probably

288 Cecilia Poletto

15. 16.

17. 18. 19. 20.

21.

22.

targets a much lower position in the left periphery. In other words, V2 is a complex phenomenon that always involves the left periphery but not always the same FP. The same is true of scrambling, as different types of scrambling (notably A and A’) are known in different languages (see Grewendorf and Sternefeld 1990; Grewendorf 2005). Hence, given that V2 and scrambling do most probably not target the same projection in all languages, this analysis does not predict that the two phenomena always go together. What is meant here with relevant occurrences are cases with a past participle. The only text where we find preposing of niente is the Decamerone by Boccaccio, which I leave aside, because it is clearly a more artificial language (this text has also been left out in the original selection of texts for the OI Grammar edited by L. Renzi and G. Salvi (to appear) precisely for this problem). Notice that chiaramente in OI can only be a low manner adverb, not a high one as in MI (see Poletto 2004). By lowest adverbs I intend here those adverbs that are usually found after the past participle both in MI and OI when they are not focalized. There are Romance languages such as Piedmontese that leave the clitic on the past participle, but this is not the case for OI. A speculation left for future work is the possibility to connect DP scrambling to the fact that OI had very frequent determinerless DPs, in contexts which are completely excluded in MI as in (i). If the parallel between the three phases is complete, the head noun also has to raise to the head of the DP internal Focus position and through this possibly to the higher D° position. Evidence for N° to D° movement (through Focus°) in OI is provided by the frequent cases of determinerless nouns with postnominal possessive adjectives, as illustrated in (ii) (i) a. esser figliolo di pastore to be son of shepherd (Novellino: 128) b. che uomo vecchio dicesse così grande villania that man old said.subj.3sg so big rudeness (Novellino: 129) c. con moltitudine di gente in assedio with lot of people in siege (Novellino: 130) d. donami cavallo e somiere e dispendio… give-me horse and helmet and money (Novellino: 131) e. mandò per maestri sent.3sg for masters (Novellino: 127) (ii) a nostra magione to our home (Novellino: 128) Contrary to Giusti (this volume)and following Giusti (1996) the hypothesis that phases are construed in a parallel way leads to the assumption that Focus is present also in the DP phase. The difference between modal verbs and auxiliaries might simply be due to the fact that compound tenses are less frequent in OI than in MI. I will leave the investigation of this difference to future research.

Parallel phases: a study on the high and low left periphery of Old Italian

289

23. I leave a list of the elements that undergo V2 longer than others to future research, as the main topic of the present discussion is to establish a parallel among phases. 24. One might wonder why this is not the case in interrogative clauses, where a wh-item activates the CP layer. However, as already shown by Rizzi (1997) interrogative wh-items are located low in the CP structure, in the same Focus field where V2 occurs in OI. Therefore, interrogative wh-items and V2 are not compatible. 25. For an analysis that treats Romance and Germanic V2 as stemming from different CP projections see Poletto (2002). 26. In Ledgeway (to appear), he notes that Old Neapolitan has the same features that OI displays, and Old French seems to confirm the connection as well.

Abbreviations for sources used in the text Arm.

Armannino: Fior. = Fiorita

BG

Bono Giamboni: F. di rett. = Fiore di rettorica, Or. = Orosio, Tratt. = Trattato di Virtù e di Vizi, Veg. = Vegezio, VeV = Vizi e Virtudi

BL

Brunetto Latini: Rett. = Rettorica, Somm. = Sommetta ad amaestramento di componere volgarmente lettere Tes. = Tesoretto

GC

Guido Cavalcanti: Vite Doc. fior. = Documenti fiorentini

Fiore

Il Fiore e il Detto d'Amore attribuibili a Dante Alighieri

F.V.F.

Fiori e vita di filosafi

Paolino

Paolino Pieri, Merlino

Tav. rit.

Tavola ritonda o l’Istoria di Tristano

Tes. volg. Tesoro di Brunetto Latini volgarizzato da Bono Giamboni Testi fior. Testi fiorentini Trist. Ricc. Tristano Riccardiano ZB

Zucchero Bencivenni

290 Cecilia Poletto References Beghelli, Filippo 1995 The phrase structure of quantifier scope. Ph.D. Thesis, UCLA. Beghelli, Filippo and Tim Stowell 1997 Distributivity and negation: the syntax of each and every. In Ways of Scope Taking, A. Szabolcsi (ed.), 72–107. Dordrecht: Kluwer. Belletti, Adriana 1990 Generalized Verb Movement. Torino: Rosenberg and Sellier. 2004 Aspects of the low IP area. In The Structure of CP and IP, Luigi Rizzi (ed.), 16–51. Oxford: Oxford University Press. Belletti, Adriana and Ur Shlonksy 1995 The order of verbal complements: A comparative study. Natural Language and Linguistic Theory 13: 489–526. Benincà, Paola 1984 Un’ipotesi sulla sintassi delle lingue romanze medievali. Quaderni Patavini di Linguistica 4: 3–19. 1995 Complement clitics in medieval romance: The Tobler-Mussafia law. In Clause Structure and Language Change, Adrian Battye and Ian Roberts (eds.), 325–344. Oxford: Oxford University Press. 2005 A detailed map of the left periphery of medieval Romance. To appear in Negation, Tense and Clausal Architecture: Crosslinguistics Investigations, Raffaella Zanuttini, Héctor Campos, Elena Herburger and Paul Portner (eds.). Washington: Georgetown University Press. Benincà, Paola and Cecilia Poletto 2004 Topic, Focus and V2: defining the CP sublayers. In The Structure of CP and IP, Luigi Rizzi (ed.), 52–75. Oxford: Oxford University Press. Cinque, Guglielmo 1999 Adverbs and Functional Heads. Oxford: Oxford University Press. Egerland, Verner 1996 The Syntax of Past Participles. Lund: Lund University Press. Fuss, Eric 2005 Kongruenzphenomene. Ph.D. diss., University of Frankfurt. Giusti, Giuliana 1996 Is there a FocusP and a TopicP in the Noun Phrase structure? University of Venice Working Papers in Linguistics 6.2: 105–128. this vol. Parallels in clausal and nominal periphery. Grava, Roberta 2005 Estudio comparativo entre italiano espanol y catalan sobre la posicion de los denominados adverbios “bajos”. Graduation thesis, University of Venice.

Parallel phases: a study on the high and low left periphery of Old Italian

291

Grewendorf, Günther 2005 The discourse configurationality of scrambling. In The Free Word Order Phenomenon: Its Syntactic Sources and Diversity, Joachim Sabel and Mamoru Saito (eds.), 75–135. Berlin /New York: Mouton de Gruyter. Grewendorf, Günther and Wolfgang Sternefeld 1990 Scrambling and Barriers. Amsterdam: John Benjamins. Guasti, Maria Teresa and Luigi Rizzi 2002 Agreement and tense as distinct syntactic positions: Evidence from acquisition. In Functional Structure in DP and IP, Guglielmo Cinque (ed.), 167–194. Oxford: Oxford University Press. Jayaseelan, Karattuparambil A. 2001 IP-internal topic and focus phrases. Studia Linguistica 55 (1): 39–75. Kayne, Richard 1991 Romance clitics, verb movement and PRO. Linguistic Inquiry 22: 647–686. Ledgeway, Adam to appear Old Neapolitan word order: Some initial observations. In Histories of the Italian Languages, Anna Laura Lepschy and Arturo Tosi (eds.). Oxford: Legenda. Paoli, Sandra 1997 Agreement: A relational property or a functional projection? Evidence from part participle agreement in Friulian. Master thesis, University of York. Paul, Waltraud 2002 Sentence-internal topics in Mandarine Chinese: The case of object preposing. Language and Linguistics 3 (4): 695–714. Poletto, Cecilia 2002 On V2 and V3 sequences in Rhaetoromance. In Syntactic Microvariation, Sjef Barbiers, Leonie Cornips and Susanne van der Klei (eds.). http://www.meertens.nl/books/synmic/pdf/poletto.pdf Poletto, Cecilia 2004 La struttura della frase in italiano antico. Ms., University of Padua. Renzi,Lorenzo and Giampaolo Salvi forthc. Grammatica dell´Italiano Antico. Bologna, Il Mulino. Rizzi, Luigi 1997 The fine structure of the left periphery. In Elements of Grammar. Handbook in Generative Syntax, Liliane Haegeman (ed.), 281–337. Dordrecht: Kluwer. Roberts, Ian 1993 Verbs and Diachronic Syntax: A Comparative History of English and French. Dordrecht: Kluwer.

292 Cecilia Poletto Vanelli, Laura 1987 I pronomi soggetto nei dialetti italiani settentrionali dal Medio Evo ad oggi. Medioevo Romanzo XIII: 173–211. Vikner,Sten 1995 Verb Movement and Expletive Subjects in the Germanic Languages. Oxford: Oxford University Press. Zwart, Jan-Wouter 1993 Dutch Syntax. A minimalist approach. Ph.D. diss., University of Groningen.

Chapter 4 The CP-phase and Subject Licensing

Obviation in subjunctive argument clauses and the first-personal interpretation Francesco Costantini

1. Introduction The phenomenon of obviation in Romance subjunctive clauses, which is sometimes referred to as “subjunctive disjoint reference effect” (Kempchinsky 1987), has been discussed at least since the early 1980s. Generally speaking, these terms, “obviation” and “subjunctive disjoint reference effect” refer to the fact that a null (in Catalan, Italian, Portuguese, and Spanish) or a clitic (in French) subject in a subjunctive embedded clause cannot be coindexed with the subject of the embedding clause. Most of the hypotheses worked out over the years claim that the phenomenon at issue follows from the interpretative properties of subjunctive mood (particularly, its nature of “anaphoric” mood) and the Binding Theory. Particularly, obviation is supposed to stem from a violation of Condition B – the pronominal (null or clitic) subject is claimed to be bound in its binding domain, contrary to Condition B. Proposals based on Binding Theory are able to account for a crosslinguistically uniform set of data – subjunctive obviation has been studied in many Romance languages: Catalan (Picallo 1985), French (Bouchard 1982, 1983; Farkas 1992; Ruwet 1984; Tsoulas 1996; Schlenker 2005), Italian (Manzini 2000; Rizzi 2000), Portuguese (Raposo 1985), Spanish (Kempchinsky 1987, 1998; Suñer 1986). Furthermore, it has been studied in non-Romance languages, such as Dutch (Everaert 1986) and Russian (Avrutin 1994; Avrutin and Babyonyshev 1997). Alternative proposals, based on the observation that (necessarily obviative) subjunctive clauses are complementary to (necessarily non-obviative) infinitive clauses, are also able to account for a cross-linguistically uniform set of data. This notwithstanding, well known examples, originally discussed by Ruwet (1984), are still unaccounted for. Some recent proposals on binding properties of subjunctive argument clauses – particularly, those concerning long distance anaphors (Giorgi 2004) – seem to cast new light on the phe-

296 Francesco Costantini nomenon at issue. Particularly, the hypothesis will be explored that obviation in subjunctive clauses can be traced back to a grammatical principle which prevents a syntactic formative serving as the subject in a subjunctive clausal argument from being interpreted first-personally in the sense of Castañeda (1966, 1968) and Higginbotham (2003), while allowing only implicit arguments to be interpreted first-personally. In a way, this proposal is reminiscent of Farkas’ (1992) and Schlenker’s (2004). But in contrast to these proposals, it tries to account for obviation in subjunctive argument clauses not in terms of mood competition, rather in terms of explicit (i.e. syntactically projected arguments) versus implicit arguments (i.e. unassigned
-positions) in certain embedded clauses – the ones in which the speaker is not represented (Giorgi and Pianesi 2001). This proposal predicts (a) the contrast between subjunctive clauses, which show obviation effects, and indicative clauses, which do not; (b) the contrast between subjunctive clauses, which cannot generally be interpreted first-personally, and infinitive clauses, which must be interpreted first-personally; (c) the contrast between subjunctive clauses in which the lexical verb bears subjunctive morphology, which instantiate obviation effects, and subjunctive clauses in which the lexical verb is an infinitive, a participle, or a gerund, which do not instantiate obviation effects. For reasons of space I will describe only sentences presenting the following properties: (a) the subjunctive clause is an argument clause, not an adverbial clause (for a discussion on obviation in some adverbial subjunctive clauses I refer to Manzini 2000); (b) the subject of the subjunctive clause is third person singular; furthermore, it is phonologically null1; (c) the matrix predicate is epistemic; emotive-factive contexts will not be considered. I will focus on an empirical framework based on Italian data, and will leave the comparative aspects of the phenomenon at issue for future research. I will proceed as follows: in section 2 the phenomenon at issue is described in a general way. In section 3 the analysis will be extended and the most important hypothesis will be summed up. Finally, in section 4, an alternative proposal, based on some properties of subjunctive clauses and of non-finite verbal forms, will be outlined.

Obviation in subjunctive argument clauses

297

2. Basic Data The puzzle of obviative effects in Romance subjunctive clauses has been widely discussed by generative linguists since the early 1980s. The sentence in (1) exemplifies this phenomenon in Italian: (1)

Gianni vuole che parta. Gianni wants that (she/he) leaves-SUBJ ‘Gianni wants her/him to leave’.

I will give the following representation of sentence (1)2: (2)

Giannii vuole che pro*i/j parta. Giannii wants that pro*i/j leaves-SUBJ ‘Gianni wants *himself/her/him to leave’.

The representation by indexes in (2) means that in sentence (1) Gianni and pro cannot be coreferent – i.e., they are disjoint in reference or obviative3. In the following sentence the embedded subject can be coreferent with the matrix one: (3)

Giannii ha deciso che proi/j partirà domani. Giannii has decided that proi/j will-leave-IND tomorrow ‘Gianni decided that he will leave tomorrow’.

In sentence (1) the embedded verb is a subjunctive form, whereas in sentence (3) it is an indicative. Moreover, in the following sentences the embedded and the matrix subjects must be coindexed: (4)

a. Giannii vuole PROi/*j partire. Giannii wants PROi/*j leave-INF ‘Gianni wants to leave’. b. Giannii ha deciso di PROi/*j partire domani. Giannii has decided of PROi/*j leave-INF tomorrow ‘Gianni decided to leave tomorrow’.

In sentences (4a) and (4b) the verb in the embedded clause is an infinitive and Gianni controls PRO.

298 Francesco Costantini Thus, subjunctive clauses are different from indicative and infinitive clauses, in that only in the former type of clause is the coindexation forbidden between the matrix and the embedded subject. Given these data, the following question arises: (5)

3.

In sentences like (2), why is coindexation ruled out between the matrix subject and the embedded subject?

Other data and existing theories

3.1. Subjunctive vs. infinitive? Existing accounts of this problem split into two types of an approach: the “competition” approach, and the Binding-Theoretical approach. The first type of account stems from the tenet that obviation is the consequence of the “competition” between subjunctive and infinitive forms (Bouchard 1982, 1983; Farkas 1992; and Schlenker 2004). The basic hypothesis is as follows: if coreference between the matrix and the embedded subject is expressed by means of an infinitive clause, it cannot be expressed by means of a subjunctive complement clause. This hypothesis immediately accounts for the contrast between sentences (2) and (4). This hypothesis predicts that subjunctive and infinitive are complementary: when infinitive is available, the subjunctive is not, and vice versa. However, contexts in which subjunctive and infinitive are not complementary do exist. Take for instance the following examples in Italian: (6)

a. Giannii sperava che proi/j potesse partire il giorno dopo. Giannii hoped that proi/j was-SUBJ able leave-INF the day after ‘Gianni hoped that he was able to leave on the following day’. partire il giorno dopo. b. Giannii sperava di PROi/*j poter Giannii hoped of PROi/*j be-able-INF leave-INF the day after ‘Gianni hoped to be able to leave on the following day’.

Notice that in (6b) the matrix subject, Gianni, obligatorily controls PRO. Hence, Gianni and PRO corefer. The fact that in sentence (6a) Gianni and pro may carry the same referential index is unexpected under competition theories4. Example (6a) also shows that question (5) is too general, since contexts in which the coindexation between the matrix and the embedded

Obviation in subjunctive argument clauses

299

subject is available do exist. A more appropriate question, descriptive in nature, would be the following one: (7)

In which sentences with a subjunctive clausal complement is coindexation forbidden between the matrix and the embedded subject?

3.1.1. Other non-obviative subjunctive argument clauses In example (6a) the subjunctive verb is a modal. This property distinguishes it from example (2). In other sentences in which the coindexation between the matrix and the embedded subject is allowed the subjunctive verb is not lexical5. In embedded clauses with an auxiliary, the coindexation between the matrix and the embedded subject is not completely ruled out6: (8)

Giannii sperava che pro?i/j avesse fatto pochi errori. Giannii hoped that pro?i/j had-SUBJ made few mistakes ‘Gianni hoped that he had made few mistakes’.

Although the coreferential interpretation might be marginal, still it does not seem to be completely excluded. Notice that the infinitive is available as well: (9)

Giannii sperava di PROi/*j aver fatto pochi errori. Giannii hoped of PROi/*j have-INF made few mistakes ‘Gianni was afraid to have made many mistakes’.

Another example of a non-obviative subjunctive clause is given by embedded clauses whose subjunctive verb is passive. Like example (8), these examples involve a subjunctive (passive) auxiliary: (10) Giannii sperava che proi/j fosse autorizzato a partire. Giannii hoped that proi/j was-SUBJ authorized to leave ‘Gianni hoped that he was authorized to leave’. Again, coreference can also be expressed by means of an infinitival clause7: (11) Giannii sperava di PROi/*j essere autorizzato a partire. Giannii hoped of PROi/*j be-INF authorized to leave ‘Gianni hoped to be authorized to leave’.

300 Francesco Costantini Finally, it is also possible to interpret subjunctive clauses with a progressive auxiliary via a coreferential interpretation: (12) a. #Giannii credeva che proi/j stesse facendo molti errori. Giannii thought that proi/j was-SUBJ making many mistakes ‘Gianni thought that he was making many mistakes’. b. *Giannii credeva di PROi star facendo molti errori. Giannii thought of PROi be-INF making many mistakes ‘Gianni thought that he was making many mistakes’. Notice that judgments may vary with respect to sentence (12a). However, the coreferential reading is more acceptable in sentence (12a) than in (1). In any of the examples in which obviation does not occur despite the presence of a subjunctive, a modal or an auxiliary verb appears – let us include modal and auxiliary verbs under the label of “functional” verbs (in the sense of Cinque 1999, 2004). In these examples the full (or lexical) verb has a non-finite form: infinitive, participle, or gerund. This distinction seems to be crucial. Hence, the following generalization seems to hold: (13) Coindexation between the matrix and the embedded subject is ruled out if the verb carrying subjunctive morphology is a full verb; it is allowed if the verb carrying subjunctive morphology is a functional verb and the full verb is non-finite.

3.2. Obviation, subjunctive mood and binding domain Generalization (13) is unexpected both under competition hypotheses and under hypotheses based on the Binding Theory. The latter type of approach is based on the claim that the binding domain for a subject pro in a subjunctive clause includes the subject of the main clause. Suppose that in sentence (2) the matrix and the embedded subject were coindexed. The matrix subject c-commands the embedded subject. Furthermore, suppose that the binding domain of the embedded subject includes the matrix subject. Then, the latter would bind the former. In other words, pro would not be free in its binding domain. This violates Principle B of Binding Theory: (14) A pronominal is free in its binding domain.

Obviation in subjunctive argument clauses

301

Many hypotheses have been worked out on how the binding domain of the embedded subject happens to include the matrix subject. Generally, some interface properties of the subjunctive mood in dependent clauses are claimed to be responsible for such an operation. The crucial property is that subjunctive has no temporal reference of its own – that is, it does not set an event with respect to the time of the utterance. This property distinguishes indicative and subjunctive mood 8. Let us go back to the main question, i.e. why the binding domain of the embedded subject includes the matrix subject. This question has been a matter of discussion and many proposals have been worked out. Some of them claim that the binding domain of the embedded subject is the whole sentence – this may be due to the temporal properties of subjunctive (Picallo 1985; Rizzi 2000) or to the properties of the complementizer of subjunctive clauses (Kempchinsky 1987, 1998; Suñer 1986); some others claim that only the matrix and the embedded AgrS are included in the same binding domain (Avrutin 1994; Avrutin and Babyonyshev 1997; Manzini 2000). The former approach seems to be more powerful, in that it is able to account for the following kind of example: (15) a. Preoccupava Giannii che pro*i/j partisse il giorno dopo. worried Giannii that proi/*j left-SUBJ the day after. ‘That he/she left on the following day worried Gianni.’ b. A Giannii sembrava che pro*i/j partisse il giorno dopo. to Giannii seemed that pro*i/j left-SUBJ the day after. ‘It seemed to Gianni that he/she would have left on the following day.’ In sentences (15a) and (15b) obviation occurs between the embedded subject and a matrix object that does not agree with the matrix verb. Theories that claim that the binding domain of the embedded subject is the whole sentence can generally account for these examples; theories that crucially refer to AgrS cannot; Gianni does not indeed agree with the matrix verb9. In terms of checking theory, this is to say that [Spec; AgrS] is not an appropriate landing site for (a) Gianni. Most importantly, however, the theories based on the Binding Theory are unable to explain generalization (13)10. Examples (15) show that generalization (13) is not precise enough. It defines obviation as a phenomenon occurring between the matrix and the embedded subject. But in sentences (15) the argument with respect to which the embedded subject cannot be

302 Francesco Costantini coindexed is not the subject: in (15a) it is the direct object, in (15b) it is the indirect object. It seems that the argument to which the experiencer
-role is assigned, rather that the subject, is involved in the phenomenon at issue11. Alternatively, one may hypothesize that in sentences (2), (15a), and (15b), Gianni occupies the same syntactic position at an appropriate level of representation. Such a position would be dedicated to the attitude bearer – that is, the person who has some attitude (will, belief, hope, etc.) towards a proposition. As we will see, this claim will turn out to be useful. Then, generalization (13) may be reformulated as follows: (16) Coindexation between the matrix argument referring to the attitude bearer and the embedded subject is ruled out if the verb carrying subjunctive morphology is a full verb; it is allowed if the verb carrying subjunctive morphology is a functional verb and the full verb is nonfinite. Notice that generalization (16) is compatible with another set of data which seems to be problematic for existing theories on subjunctive obviation, namely sentences involving directive verbs – to order, to persuade, etc. These verbs differ from epistemic verbs in that they have adicity three. Thus, they have two individual arguments, and one propositional argument. Of the two individual arguments, the first may be considered as the attitude bearer, provided that an order or a request expresses a certain sort of will. Hence, one expects that the null subject of the argument clause cannot be coindexed with it, but it is free to corefer with the second individual argument. This expectation seems to be correct. Consider, for instance, the following sentence: (17) Giannii chiese a Mariaj che pro*i/j partisse presto. Giannii asked to Maria that pro*i/j left-SUBJ soon ‘Gianni asked Maria to leave soon’. In sentence (17) pro cannot be coindexed with the argument referring to the attitude bearer, Gianni, although it can corefer with the argument referring to the addressee of the request, Maria12. Thus, generalization (16) is able to include example (17). Conclusively, given the generalization in (16), a theory of obviation should be able to answer the following questions:

Obviation in subjunctive argument clauses

303

(18) Why is coindexation ruled out between the subject of an argument clause whose lexical verb is subjunctive, and the argument referring to the attitude bearer?

4. Towards a hypothesis Given the generalization in (18), the following desiderata for a hypothesis on obviation in subjunctive clauses arise: – such a hypothesis should be able to explain why subjunctive, indicative and infinitive argument clauses are different with respect to the referential properties of their subject; – it should be able to explain why full verbs and functional (modal, auxiliary, etc.) verbs are different with respect to the referential properties of the subject of a subjunctive clause; – it should be able to explain why obviation involves only matrix arguments referring to the attitude holder.

4.1. Again on obviation and subjunctive mood As for the first question, all the existing theories on subjunctive obviation based on the Binding Theory (Picallo 1985; Kempchinsky 1987, 1998; Raposo 1985; Rizzi 2000; Progovac 1993, 1994; Avrutin 1994; Avrutin and Babyonyshev 1997) claim that the interpretative properties of the subjunctive are responsible for obviation effects. Particularly, the embedded subjunctive is such that the eventuality it refers to is evaluated with respect to time frame of the matrix predicate. Because of this property, the subjunctive mood, which is sometimes said to be an “anaphoric” mood, is in contrast to the indicative, which is said to be “deictic” (Picallo 1985) – even though it cannot be simply “deictic”13. If it were only deictic, an embedded indicative should be interpreted only with respect to the utterance time. However, it has been noticed in the literature on Double Access Reading (henceforth, DAR) that this is not correct. Consider, for instance, the following sentence: (19) Gianni ha detto che Maria è incinta. Gianni has said that Maria is-IND pregnant ‘Gianni said that Maria is pregnant’.

304 Francesco Costantini Sentences (19) can be considered true if the state of pregnancy holds both at the time of Gianni’s saying and at the time of the utterance – if the embedded indicative were “deictic”, the time of embedded eventuality should not be evaluated with respect to the attitude episode only. Let us now consider more in detail the differences between indicative and subjunctive. When DAR obtains, the eventuality of an embedded clause is evaluated both with respect to the eventuality of the matrix clause (the attitude episode), and with respect to the utterance time. Hence, if the embedded verb is future indicative, a past indexical adverb induces ungrammaticality, and vice versa, since the indicative mood instantiates DAR: (20) a. Gianni ha detto che Maria è partita ieri / *domani. Gianni has said that Maria is-IND left yesterday / *tomorrow ‘Gianni said that Maria left yesterday/*tomorrow’. b. Gianni ha detto che Maria partirà domani / *ieri. Gianni has said that Maria leaves-IND.FUT tomorrow / *yesterday ‘Gianni has said that Maria will leave tomorrow/*yesterday’. Subjunctive clauses do not instantiate DAR – that is, the embedded eventuality is not evaluated with respect to the utterance time. Hence, any indexical adverbial is compatible with a subjunctive: (21) Gianni credeva che Maria partisse ieri / domani. Gianni believed that Maria left-SUBJ yesterday / tomorrow ‘Gianni believed that Maria would leave yesterday /tomorrow’. Notice that infinitive clauses are similar to subjunctive clauses in that the eventuality referred to by an infinitive clause is not evaluated with respect to the utterance time. Infinitive clauses are indeed compatible with any indexical adverbial: (22) Gianni pensava di partire ieri / domani. Gianni thought of leave-INF yesterday / tomorrow ‘Gianni thought he would leave yesterday /tomorrow’. Following Higginbotham (1995), Giorgi and Pianesi (2001) hypothesize that the facts in (20) and (21) can be formally characterized as follows: the temporal coordinates of the attitude episode (that is, the temporal coordinates of the attitude bearer – ‘Subject-’) are always represented in an em-

Obviation in subjunctive argument clauses

305

bedded clause – the time of an embedded eventuality must always be evaluated with respect to the time of the attitude episode. The temporal coordinates of the speaker (‘Speaker-’) are instead represented in an embedded clause only when the DAR obtains – that is, only when the embedded eventuality must be evaluated with respect to the temporal coordinates of the speaker (the utterance time). Particularly, analyzing data from Complementizer Deletion (henceforth, CD) in Italian14, they propose that the temporal coordinates of the subject of the attitude may be located in the embedded T, whereas the temporal coordinates of the speaker appear in the complementizer15: (23) a. Indicative: V [CP Cspeaker- … [TP Tsubject-… b. Subjunctive: V [MOODP … [TP Tsubject-… Generalization (16) may then be restated as follows: (24) Coindexation between the matrix argument referring to the attitude bearer and the embedded subject is ruled out iff a. the coordinates of the speaker are not represented in the embedded clause, and b. the embedded verb is a full verb.

4.2. Subjunctive vs. infinitive or finite vs. non-finite? Let us go back to the basic paradigm, which I repeat here: (25) a. Giannii sperava che pro*i/j partisse. Gianni hoped that pro*i/j left-SUBJ ‘Gianni hoped that he left’. b. Giannii sperava di PROi/*j partire presto. Giannii hoped of PROi/*j leave-INF soon ‘Gianni hoped to leave soon’. partire il giorno dopo. c. Giannii sperava che proi/j potesse Giannii hoped that proi/j was-SUBJ able to-leave the day after ‘Gianni hoped that he was able to leave on the following day’. As already observed, in sentences like (25a) the verb carrying subjunctive morphology is a full verb; in sentences like (25c) the verb carrying sub-

306 Francesco Costantini junctive morphology is a functional verb; in sentences like (25b) the full verb is non-finite; the same holds for sentences like (25c), in which the full verb is infinitive. Let us suppose, then, that the presence of a non-finite form of a full verb is responsible for the coindexation between the matrix and the embedded subject both in (25b) and in (25c). Notice that in example (25b) coindexation is obligatory. Moreover, the only available interpretation for sentence (25b) is a first-personal (Castañeda 1966, 1968) or de se (Lewis 1979) interpretation. Let us consider the following example: (26) John thinks he is a war hero. By means of sentence (26) the speaker might say that John has a de re belief about a certain person who is John himself. But John might have two distinct thoughts. (a) Imagine that John is amnesiac; he might be reading about the deeds of a war hero, who is himself, actually, and exclaims: “this man is a war hero!”, without realizing that it is himself the person who he is talking about. The speaker might then tell what John said by means of sentence (26). She is aware of the identity of the war hero and coindexation between John and he would be adequate from her point of view. This would be a non de se interpretation. (b) In another scenario, John consciously says about himself: “I am a war hero”; even in this case the speaker might tell what John said by means of sentence (26), and even in this interpretation coindexation would occur, from the point of view of the speaker. This would be a de se, or first-personal interpretation16. The ambiguity of sentence (26) does not obtain when the embedded clause is infinitival: (27) John expects to win. By means of sentence (27) the speaker can only mean that John is perfectly aware that the person whom he has an expectation about is himself. Then, PRO can only be interpreted de se. As for the sentences in (25), in (25a) coindexation is not available; then the problem whether the sentence is ambiguous between the de se reading and the non de se reading does not arise. Sentence (25b) is strictly de se. Finally, in sentence (25c) coindexation is an option; but even under the coreferential interpretation the sentence can only be de se17. Generalization (24) may then be restated as follows:

Obviation in subjunctive argument clauses

307

(28) The first-personal reading is ruled out iff a. the coordinates of the speaker are not represented in the embedded clause, and b. the embedded verb is a full verb.

4.3. Non-finite moods and first-personal interpretation Given the observations above, let us hypothesize that the same mechanism is responsible for the first-personal interpretation both if the complement clause has a subjunctive verb and if it has an infinitive verb. In order to implement this hypothesis, I will resort Giorgi’s (2004) account for the nature and distribution of proprio, an Italian possessive anaphor, which may be bound by a constituent outside its local domain. The relevant properties of long distance anaphora (henceforth, LDAs) like proprio are the following: (29) a. LDAs are strictly first personal, and b. LDAs can appear only if the coordinates of the speaker are not represented. Notice that property (29b) defines the same syntactic environment as obviation18. Secondly, LDAs and obviative subjects have opposite properties in that the former formatives are strictly first-personal, whereas the latter cannot be first-personal. Notice, finally, that the LDA proprio exhibits the same properties as PRO in attitude contexts (Chierchia 1989). Particularly, they are both strictly first-personal. Given generalization (29), Giorgi proposes the following hypothesis: (30) Long distance anaphoric binding: a. a LDA is the spell-out of an unsaturated position; b. an unsaturated position can be
-identified i. with a co-argument, or ii. with the bearer of the attitude. This hypothesis exploits the notion of “unsaturated position”, which may be considered as an analogous of implicit argument (in the sense of Higginbotham 1997; Williams 1985, 1987, 1989, 1994 – that is, an open
-position

308 Francesco Costantini in a
-grid, i.e. a
-position that is not realized overtly by
-marking) and “
-identification” with the attitude holder19. Notice that the mechanism of
-identification is local. We have seen, however, that the attitude holder is present in the embedded clause by means of its temporal coordinates, as Giorgi and Pianesi (2001) suggest, or by means of one of the highest functional projections (see note 15). Hence,
-identification between an implicit argument and the attitude bearer can obtain within the embedded clause20. Notice that the machinery Giorgi resorts is not ad hoc, since it has been claimed for independent reasons that implicit arguments can be involved in anaphoric relations21. Notice also that Higginbotham (1997) claims that Control itself may be understood as a process of
-identification between two
-positions, one of which is the external argument of an infinitive verb22. Moreover, if Giorgi’s hypothesis is feasible, to reduce Control to
-identification would be a natural step, given that similar interpretative properties characterize both LDAs and PRO, as observed by Chierchia (1989). Suppose, furthermore, that not only infinitives, but any non-finite form, that is infinitives, gerunds, and participles, is associated with an external implicit argument that may be
-identified either with a co-argument or with the attitude bearer, as hypothesized in (30) with respect to LDAs. Then, the hypothesis that arises from the present discussion is the following: (31) If a first-personal interpretation occurs in environments devoid of the speaker’s coordinate, that interpretation is induced by an implicit argument that is
-identified with the attitude bearer. We may assume that the notion of ‘environment’ in (31) may be rendered more precise through the notion of ‘phase’ (Chomsky 2000, 2001), so that in phases devoid of the speaker’s coordinate a de se reading may be conveyed only by means of an implicit argument. Notice that an approach in the spirit of Giorgi (2004) would be able to explain the contrast between phases including the speaker’s coordinate, corresponding to indicative clauses, on the one hand, and, on the other, phases devoid of the speaker’s coordinate, corresponding to subjunctive and infinitive clauses, which are here claimed to need an implicit argument in order to be first personal. Giorgi claims indeed that an implicit argument percolates until it is satisfied through
-identification (see note 20). But if the speaker’s coordinate is introduced in the computation, that is, if the indicative complementizer C is merged, all
-positions must be satisfied23, and an implicit argument cannot percolate any further.

Obviation in subjunctive argument clauses

309

A further implementation of this mechanism may assume that when the speaker’s coordinate, which is associated to the speaker’s assignment sequence, is introduced in the computation, all pronominal and anaphoric elements must be assigned a value with respect to the speaker’s sequence. Hence, in indicative argument clauses pro is evaluated relative to the speaker assignment sequence, and is free to corefer to the attitude holder, since it does not belong to the same local domain of the argument referring to the attitude holder herself. If the speaker’s coordinate is not introduced in the embedded clause, as in subjunctive and infinitive clauses, the relevant assignment sequence is that of the attitude bearer (or a modified speaker’s sequence24). Suppose, then, that this sequence specifies the first personal reading and that pro cannot be associated with this position. Cross-linguistic evidence is supposed to show that this hypothesis may be feasible25. Furthermore, suppose that only implicit arguments can be associated with the de se position. An explanation would arise for the difference between indicative and subjunctive clausal arguments with respect to obviation, as exemplified by examples (1) and (3). 4.4. Some predictions Let us now go back to the paradigm in (25). Given hypothesis (31), sentences (25) follow directly. In sentence (25) the subject of the subjunctive clause is not an implicit argument – though being phonetically unrealized. By definition, indeed, implicit arguments are devoid of any feature characterizing overt formatives, that is phonetic features, categorial features and -features (Chomsky 1995). The pronominal pro, however, is not devoid of these features – it enters indeed in agreement relations with the verb in the subjunctive: (32) a. pro legge molti libri. pro reads many books ‘He/she reads a lot of books’. b. pro leggono molti libri. pro read many books ‘They read a lot of books’. Thus, pro must have -features. This implies that it is a syntactic formative, not an implicit argument. Hence, sentence (25) cannot be interpreted firstpersonally.

310 Francesco Costantini In sentence (25) the subject of the argument clause has properties compatible with those of an implicit argument, since no features are detectable26. Hence, the first-personal interpretation is available. Finally, let us consider sentence (25). Suppose that the following simplified structure is a viable representation of sentence (25) (where the notation ‘‹1›’ indicates the (relevant) unassigned
-role of the infinitive): (33) Gianni sperava [che [AgrSP pro [TP [T, subject- potesse] [FP, ‹1› partire]]]] Notice that the
-role of the infinitive, ‹1›, is not assigned via
-marking, extending the ideas of Higginbotham (1997), since it is an implicit argument. Accordingly, it can be
-identified with the attitude bearer. Hence, the first-personal interpretation is expected to be available. Furthermore, the fact that pro is in the specifier of [AgrSP] and that an implicit argument may be
-identified with a co-argument, predicts that the null subject of the subjunctive clause may refer to someone else than the attitude bearer. This seems to be a correct prediction – sentence (25) is ambiguous between a coreferential reading and the disjoint one.

5. Conclusion Hypothesis (31) seems to be able to generate the data discussed here. It also seems to be able to account for the complementary distribution of infinitive and subjunctive argument clauses, which turned out to be only partial. If the hypothesis worked out here were correct, it would follow that only apparently would obviation in subjunctive clausal arguments occur between the matrix argument referring to the attitude bearer and the embedded subject pro. Rather, obviation would be the result of a mechanism occurring within a subjunctive clause. Moreover, if the hypothesis explored here were on the right track, Binding Theory would not be involved in subjunctive obviation – at least, not in the way supposed by existing theories; indeed, the impossibility of coindexation between the arguments involved in the phenomenon at issue would be a consequence of more general principles, supported by independent evidence, which are supposed to account for many phenomena concerning Italian subjunctive complements – temporal anchoring, sequence of tense, CD, and LDAs.

Obviation in subjunctive argument clauses

311

Acknowledgements I would like to thank Alessandra Giorgi for the substantial discussions, her careful reading and her suggestions, Guglielmo Cinque and the faculty at the Dipartimento di Scienze del Linguaggio – Venice for thoughtful advice. I am grateful for comments and suggestions to the audience of the XXXI IGG (particularly, I would like to thank Luigi Rizzi, Ur Shlonsky, and Carlo Cecchetto), to the audience of the seminar held in Padua on 9th June 2005 (in particular, Paola Benincà, Cecilia Poletto, and Federico Damonte), to Vieri Samek-Lodovici, Walter Schweikert, Luigi Zennaro, Vesselina Laskova for insightful discussion, an anonymous reviewer for valuable remarks, Mara Frascarelli and John Bleasdale for careful reading.

Notes 1. For a discussion on the properties of phonologically realized subjects as opposed to those of phonologically unrealized ones in Italian, see Calabrese (1986) and Samek-Lodovici (1996). 2. I will assume the existence of a null pronoun, pro, as a syntactic constituent (for an alternative proposal, see Alexiadou and Anagnostopoulou 1998) and adopt the customary notation by referential indexes – the same index is assigned to coreferent NPs, different indexes are assigned to NPs which do not corefer. 3. I will exclude from the analysis the case of accidental coreference. 4. Example (6) parallels one of the examples discussed by Ruwet (1984): dès demain. (i) ?Je veux que je puisse partir I want that I can-SUBJ leave-INF by tomorrow Notice that in sentence (6) the coreferential reading is not gained via accidental coreference. This can be proved by means of sentences like the following one: (ii) Ogni ragazzoi sperava che proi/j potesse partire il giorno dopo. hoped that proi/j could-SUBJ leave-INF the day after Every boyi ‘Every boy hoped that he was able to leave on the following day’. In the sentence above the matrix subject is a quantifier binding pro. 5. See also Ruwet (1991) for examples in French, Raposo (1985) for examples in Portuguese, and Quer (in press) for examples in Spanish. 6. The judgments on this sentence may vary. However, all the Italian native speakers required to give grammaticality judgments on the sentences discussed here agree that in sentence (8) the coreferential reading can be marginally available, whereas in sentence (1) it cannot be available at all.

312 Francesco Costantini 7. If the subjunctive verb in the clause is a passive auxiliary and the participle is a past form, the coindexation between the matrix and the embedded subject is even more acceptable than in example (10): stato autorizzato a partire. (i) a. Giannii sperava che proi/j fosse Gianni hoped that proi/j was-SUBJ been authorized to leave ‘Gianni hoped that he had been authorized to leave’. b. Giannii sperava di PROi/*j essere stato autorizzato a partire. Gianni hoped of PROi/*j be-INF been authorized to leave ‘Gianni hoped to have been authorized to leave’. Moreover, if the subjunctive verb is modal and the infinitive is passive, the coindexation is also possible. essere autorizzato a partire. (ii) a. Giannii sperava che proi/j potesse Gianni hoped that proi/j was-able-SUBJ be-INF authorized to leave ‘Gianni hoped that he could be authorized to leave’. b. Giannii sperava di PROi/*j poter essere autorizzato a partire. Gianni hoped of PROi/*j be-able-INF be-INF authorized to leave ‘Gianni hoped to be able to be authorized to leave’. 8. On this point, see paragraph 4.1. 9. If the experiencer of the matrix clause were plural, the verb would be singular as well: (i) a. Preoccupava tutti che stesse piovendo da più di worried everyone that was-SUBJ raining for more than cinque giorni. five days ‘That it had been raining for more than five days worried everyone’. b. A tutti sembrava che Maria partisse il giorno dopo. after to everyone seemed that Maria left-SUBJ the day ‘It seemed to everyone that Maria would have left on the following day’. 10. Only Raposo (1985) discusses the examples involving a ‘functional’ verb carrying subjunctive morphology. He hypothesizes that in Romance languages a binding domain may be characterized in two ways: it can be the c-command domain of a subject, or the c-command domain of a “verbal operator”. He includes among the verbal operator modal verbs, auxiliary verbs, the operator [+TENSE], which characterizes the complementizer of argument clauses whose verb has independent time reference, i.e. indicative argument clauses. This hypothesis is able to predict generalization (13), but is not supported by independent evidence. 11. See Belletti and Rizzi (1988). 12. Notice that the interpretation of sentence (17) according to which pro is coindexed with Maria, is problematic for some of the previously mentioned theories. As for the competition theories, the infinitive counterpart to sentence (17) is available:

Obviation in subjunctive argument clauses

13. 14.

15.

16. 17.

313

(i) Giannii chiese a Mariaj di PROi/j partire. Gianni asked to Maria of PRO leave-INF ‘Gianni asked Maria to leave’. Notice that two interpretations are available for the above sentence (control shift – see Landau 2001): Gianni controls PRO; Maria controls PRO. The former interpretation is unavailable if the argument clause has a subjunctive verb. This is correctly predicted by competition theories. The second interpretation is available both if the argument clause has a verb in the subjunctive (cf. (17)), and if the argument clause has a verb in the infinitive, which is also the most natural alternative, although (17) is grammatical as well. Competition theories are not able to account for the availability of both the subjunctive and the infinitival argument clause. According to Picallo’s (1985) the coindexation between Maria and pro in (17) should not be available, whereas the one between Gianni and pro should be, unless it was the case that Maria did not ccommand pro. This is indeed what she stipulates. As for those theories according to which the binding domain of pro is the whole sentence, coindexation should be excluded both between the matrix and the embedded subject and between the matrix indirect object and the embedded subject. Moreover, another interpretation is available: (ii) Giannii chiese a Mariaj che pro*i/j/k partisse presto. Gianni asked to Maria that pro*i/j/k left-SUBJ soon ‘Gianni asked Maria to leave soon’. See Giorgi and Pianesi (2001) for a discussion on this point. Giorgi and Pianesi observe indeed that DAR contexts overlap with contexts that do not allow CD. Moreover, they observe that CD is possible only in subjunctive complements (with the exception of complements of emotive-factive predicates and those of ipotizzare-class predicates). They propose, then, that the Italian complementizer che lexicalizes two sets of features, those of the indicative complementizer C, and those of the subjunctive complementizer, which they call MOOD. The features of MOOD, but not those of C, may eventually be lexicalized together with the -features of AGR, which determines CD. Given that the indicative C cannot be “deleted” and that DAR obtains only if C appears, they propose that the interpretative properties distinguishing DAR contexts from non-DAR ones (that is, the speaker’s temporal coordinate, with respect to which the embedded eventuality has to be interpreted) may be localized in C. For alternative proposals on the syntactic representation of the speaker and the attitude bearer, see also Cinque’s (1999) discussion on the highest functional heads in the IP-domain – speech act mood, evaluative mood, and evidential mood. See also Speas (2004). I assume that a de se reading is a peculiar kind of de re reading (see Higginbotham 1991). The non de se may be obtained if the embedded subject were a ‘strong’ pronoun, e.g. lui:

314 Francesco Costantini

18. 19.

20.

21.

22.

partire il giorno dopo. (i) Giannii sperava che luii potesse Gianni hoped that hei could-SUBJ leave-INF the day after ‘John hoped that he could leave on the following day’. Sentence (25) might be appropriate in a scenario in which Gianni says: “I hope I will be able to leave tomorrow”; but it would not be appropriate in a scenario in which Gianni says: “I hope he will leave tomorrow”, without realizing that the person he is referring to is himself. According to most of Italian native speakers, only sentence (i) may have an interpretation (that is, the non de se reading) compatible with this second scenario. I will not try to explain this puzzle here. A promising starting point could be the fact that ‘strong’ pronouns and pro have different interpretative properties, as observed by Cardinaletti and Starke (1999). For alike observations, see Manzini (2000), and Progovac (1993, 1994). On implicit arguments, see Bhatt and Pancheva 2004; Brody and Manzini 1991; Chomsky 1986; Higginbotham 1985, 1997; Roeper 1987; Safir 1991; Williams 1987, 1989, 1994. On
-identification, see Higginbotham (1985). Particularly,
-identification is supposed to obtain under sisterhood (Higginbotham 1985), and unassigned
-roles are able to percolate to a dominating maximal projection until they are assigned (Williams 1987, 1989). Take for instance the following example (Higginbotham 1997): (i) self-starting (motor) Higginbotham argues that start
-marks neither an internal argument nor an external argument. This notwithstanding, the external argument can bind the internal argument, because an available interpretation is that a self-starting motor is ‘a motor x such that x starts x’. Given such an interpretation, an implicit anaphora is claimed to be involved. Furthermore, consider the following example: (ii) Every participant had to defeat an enemy The theta-grid of the word enemy contains two slots, the second of which is
identified with the external role of defeat, i.e. every participant – the sentence may be indeed interpreted ‘for any x, x had to defeat an enemy of x’. Other pieces of evidence that implicit (external) arguments interact with the three Principles of Binding Theory are the following (see Williams 1987): (iii) a. Respect for oneself is important b. Admiration of him c. The realization that John was unpopular I refer to Williams (1987) for a discussion of the examples in (iii). For evidence that the external arguments in the above examples are in fact implicit (rather than PRO, which is a formative) see Higginbotham (1997). Thus, PRO should not be considered as a syntactic constituent (at least in attitude contexts), as traditionally done in the Government and Binding framework, but an implicit argument. Alternatively, it may even be considered as a syntactic

Obviation in subjunctive argument clauses

23.

24.

25.

26.

315

constituent, but it should be conceived as a sort of “syntactic reflex of Control, rather than the controlled element itself” (Higginbotham 1997: 192), that is, a formative that may be merged when Control has already been established. Giorgi claims that the speaker’s assignment sequence, which includes the temporal coordinate of the speaker, anchors an eventuality to the utterance time, so that the eventuality is evaluated with respect to the world the speaker thinks he is in. This is supposed to explain why assertive sentences are in the indicative and cannot be in the subjunctive. In Italian, as in English, the first person pronoun in embedded clauses refers to the speaker, but this is not a universal property – see Schlenker’s (2001) discussion on Amharic first person pronouns, which refers to the attitude bearer. Schlenker (2003) observes that the Amharic counterpart of English sentence “John says that he is a hero” under the de se reading would exploit a first personal pronoun in the embedded sentence – “John says that I am a hero”. For different proposals, see Chomsky and Lasnik (1993), Landau (2001, 2004), Martin (2001).

References Alexiadou, Artemis and Elena Anagnostopoulou 1998 Parametrizing Agr: word order, V-movement and EPP-checking. Natural Language and Linguistic Theory 16: 491–539. Avrutin, Sergey 1994 Psycholinguistics investigations in the theory of reference. Ph.D. diss., MIT. Avrutin, Sergey and Maria Babyonyshev 1997 Obviation in subjunctive clauses and Agr: evidence from Russian. Natural Language and Linguistic Theory 15: 229–267. Belletti, Adriana and Luigi Rizzi 1988 Psych-verbs and
-theory. Natural Language and Linguistic Theory 6: 291–352. Bhatt, Rajesh and Roumyana Pancheva 2004 Implicit arguments. Ms., University of Texas, Austin, and USC. Bouchard, Denis 1982 On the content of empty categories. Ph.D. diss., MIT. 1983 The Avoid Pronoun Principle and the Elsewhere Principle. In Proceedings of ALNE 13/NELS 13, Peter Sells and Charles Jones (eds.), 29–36. Amherst, MA: GLSA. Brody, Michael and Rita Manzini 1991 On implicit arguments. In Mental Representations. The Interface between Language and Reality, Ruth Kempson (ed.), 105–130. Cambridge: Cambridge University Press.

316 Francesco Costantini Calabrese, Andrea 1986 Pronomina. MIT Working Papers in Linguistics. Vol. 8, Naoki Fukui, Tova R. Rapoport and Elizabeth Sagey (eds.), 1–46. Cardinaletti, Anna and Michal Starke 1999 The typology of structural deficiency. A case study of the three classes of pronouns. In Clitics in the Languages of Europe, EALT/ EUROTYP 20–5, Henk van Riemsdijk (ed.), 145–233. Berlin /New York: Mouton de Gruyter. Castañeda, Hector-Neri 1966 He*: a study in the logic of self-consciousness. Ratio 8: 130–157. Castañeda, Hector-Neri 1968 On the logic of attributions of self-knowledge to others. The Journal of Philosophy 65 (15): 439–456. Chierchia, Gennaro 1989 Anaphora and attitude de se. In Semantics and Contextual Expressions, Renate Bartsch, Johan van Benthem and Peter van Emde (eds.), 1–31. Dordrecht: Foris. Chomsky, Noam 1986 Knowledge of Language: Its Nature, Origin and Use. New York: Praeger. 1995 The Minimalist Program. Cambridge, MA: MIT Press. 2000 Minimalist inquiries: The framework. In Step by Step. Essays on Minimalist Syntax in Honor of Howard Lasnik, Roger Martin, David Michaels and Juan Uriagereka (eds.), 89–135. Cambridge, MA: MIT Press. 2001 Derivation by phase. In Ken Hale. A Life in Language, Michael Kenstowicz (ed.), 1–52. Cambridge, MA: MIT Press. Chomsky, Noam and Howard Lasnik 1993 The theory of principles and parameters. In Syntax: An International Handbook of Contemporary Research, Joachim Jacobs, Arnim von Stechow, Wolfgang Sternefeld and Theo Vennemann (eds.), 506– 569. Berlin /New York: Mouton De Gruyter. Cinque, Guglielmo 1999 Adverbs and Functional Heads. A Cross-Linguistic Perspective. Oxford: Oxford University Press. 2004 ‘Restructuring’ and functional structure. In Structures and Beyond. The Cartography of Syntactic Structures, Adriana Belletti (ed.), 132– 191. Oxford: Oxford University Press. Everaert, Martin 1986 Long reflexivization and obviation in the romance languages. In Formal Parameters of Generative Grammar, Peter Coopmans, Ivonne Bordelois and Bill Dotson Smith (eds.), 51–71. Dordrecht: Foris.

Obviation in subjunctive argument clauses

317

Farkas, Donka 1992 On obviation. In Lexical Matters, Ivan A. Sag and Anna Szabolcsi (eds.), 85–109. Stanford University: CSLI. Giorgi, Alessandra 2004 From temporal anchoring to long distance anaphors. University of Venice Working Papers in Linguistics 14: 107–172. Giorgi, Alessandra and Fabio Pianesi 2001 Tense, attitudes, and subjects. In Proceedings of the SALT XI, Rachel Hastings, Brendan Jackson and Zsofia Zvolenszky (eds.), 212–230. Cornell University: CLC Publications. Higginbotham, James 1985 On semantics. Linguistic Inquiry 16: 547–593. 1992 Reference and control. In Control and Grammar, Richard K. Larson, Sabine Iatridu, Utpal Lahiri and James Higginbotham (eds.): 79–108. Dordrecht: Kluwer. 1995 Tensed thoughts. Mind and Language 10: 226–249. 1997 A plea for implicit anaphora. In Atomism and Binding, Hans Bennis, Pierre Pica and Johan Rooryck (eds.), 182–203. Dordrecht: Foris. 2003 Remembering, imagining, and the first person. In Epistemology of Language, Alex Barber (ed.), 496–533. Dordrecht: Foris. Kempchinsky, Paula 1987 The subjunctive disjoint reference effect. In Studies in Romance Linguistics, Carol Neidle and Rafael Nuñez Cedeño (eds.), 123–140. Dordrecht: Foris. 1998 Mood phrase, Case checking and obviation. In Romance Linguistics. Theoretical Perspectives, Arnim Schwegler, Bernard Tranel and Myriam Uribe-Etxebarria (eds.), 143–154. Amsterdam /Philadelphia: John Benjamins. Landau, Idan 2001 Elements of Control: Structure and Meaning in Infinitival Constructions. Dordrecht: Kluwer. 2004 The scale of finiteness and the calculus of control. Natural language and Linguistic Theory 22: 811–877. Lewis, David 1979 Attitudes de dicto and attitudes de se. Philosophical Review 88: 513– 543. Manzini, Maria Rita 2000 Sentential complementation. The subjunctive. In Lexical Specification and Insertion, Peter Coopmans, Martin Everaert, Jane Grimshaw (eds.), 241–268. Amsterdam /Philadelphia: John Benjamins. Martin, Roger 2001 Null case and the distribution of PRO. Linguistic Inquiry 32: 141– 166.

318 Francesco Costantini Picallo, Carme 1985 Opaque domains. Ph.D. diss., CUNY. Progovac, Ljiljana 1993 Subjunctive: The (mis)behavior of anaphora and negative polarity. The Linguistic Review 10: 37–59. 1994 Negative and Positive Polarity: a Binding Approach. Cambridge: Cambridge University Press. Quer, Josep in press Subjunctives. In The Blackwell Companion to Syntax, vol. IV, Martin Everaert and Henk van Riemsdijk (eds.). London: Basil Blackwell. Raposo, Eduardo 1985 Some asymmetries in the Binding Theory in Romance. The Linguistic Review 5: 75–110. Rizzi, Luigi 2000 On the anaphor-agreement effect. In Comparative Syntax and Language Acquisition, Luigi Rizzi (ed.), 158–173. London/New York: Routledge. Roeper, Thomas 1987 Implicit arguments and the head-complement relation. Linguistic Inquiry 18: 267–310. Ruwet, Nicolas 1984 Je veux partir / *Je veux que je parte: on the distribution of finite complements and infinitival complements in French. Cahiers de Grammaire 7: 75–138. Safir, Ken 1991 Evaluative predicates and the representation of implicit arguments. In Principles and Parameters in Comparative Grammar, Robert Freidin (ed.), 99–131. Cambridge, MA: MIT Press. Samek-Lodovici, Vieri 1996 Constraints on subjects. An optimality theoretic analysis. Ph.D. diss., Rutgers University. Schlenker, Philippe 2003 A plea for monsters. Linguistic and Philosophy 26: 29–120. 2005 The lazy Frenchman’s approach to the subjunctive (speculations on reference to worlds, presuppositions, and semantic defaults in the analysis of mood). In Romance Languages and Linguistic Theory 2003, Twan Geerts, Ivo van Ginneken and Haike Jacobs (eds.), 269– 309. Amsterdam /Philadelphia: John Benjamins. Speas, Margaret 2004 Evidentiality, logophoricity and the syntactic representation of pragmatic features. Lingua 114: 255–276.

Obviation in subjunctive argument clauses

319

Suñer, Margarita 1986 On the referential properties of embedded finite clause subjects. In Generative Studies in Spanish Syntax, Ivonne Bordelois, Helas Contreras and Karen Zagona (eds.), 183–196. Dordrecht: Foris. Tsoulas, George 1996 The nature of the subjunctive and the formal grammar of obviation. In Grammatical Theory and Romance Languages, Karen Zagona (ed.), 293–306. Amsterdam/Philadelphia: John Benjamins. Williams, Edwin 1985 PRO in NP. Natural Language and Linguistic Theory 3: 277–295. 1987 Implicit arguments, the Binding Theory, and Control. Natural Language and Linguistic Theory 5: 151–180. 1989 The anaphoric nature of
-roles. Linguistic Inquiry 20: 425–456. 1994 Thematic Structure in Syntax. Cambridge, MA: MIT Press.

Who is lui? Reference of Italian overt and covert subject pronouns Alessandra Trecci

Introduction It is a well-known fact that Condition B of Binding Theory (together with various other constraints such as Ross’ (1986) Background Pronominalization) determines the occurrence of pronominal elements. Thus, Condition B is assumed to apply to strong pronouns (Cardinaletti and Starke 1999) as well as to clitics. In Italian, a Null Subject Language (henceforth NSL), subject pronouns fall into two types: the null category and the overt paradigm (which fits the category of strong pronouns). The generative approach, both in its GB formulation (Chomsky 1981) and in the Minimalist framework (Chomsky 1995) has always ascribed to pro exactly the same properties as the overt paradigm (apart from phonological features). However, the occurrence of either type is not in complementary distribution and their different binding properties have not yet been determined. This issue is systematically addressed by Calabrese (1986), who makes a survey of the different anaphoric relations relevant for the overt and the null category and proposes an explanation for the formal differences of these two types. A very accurate analysis of pronominal asymmetries is also provided in Cardinaletti and Starke’s (1999) work, in which pro is defined as a weak pronoun without phonological material and data are finely described. However, the Principle of Economy, to which they ascribe the occurrence of pro, seems too general to explain the distinctions determining the use of covert vs. overt (weak) forms that is crucial in the present analysis. Indeed, Economy can explain the distribution of pro in a number of syntactic contexts (coordination, distribution, range and so on), though not in relation with information structure requirements. This is exactly the aim of this paper. The present results have therefore no impact on the validity of Cardinaletti and Starke’s analysis. In this paper I reconsider some of Calabrese’s arguments in the light of the theory of phases as formulated in Chomsky (2001, 2003, 2005) and I attempt

322 Alessandra Trecci to formalize the binding properties and the identification devices concerning the two subject pronoun categories. To achieve this purpose, I assume: (1)

a. subjects in NSLs sit in A’-position when in preverbal position (Alexiadou and Anagnostopoulou 1998); b. the TopP projection is not freely recursive: different types of Topics have different formal properties connected with their different discourse role. In a cartographic approach, this means that they sit in different projections in the C-domain, which are appropriate to their role (for the discussion of different kinds of Topics cf., Büring 1999; for a cartographic approach cf. Frascarelli and Hinterhölzl to appear).

1. Pro: licensing and interpreting conditions Despite the fact that pro is a pronominal category, it shows different characteristics with respect to overt pronouns. In particular, the analysis of data shows that: (2)

a. pro cannot start a Topic chain though it is a topical element; b. pro cannot perform a deictic function.

(2a) and (2b) are problematic for a model which does not posit formal differences between overt and covert pronouns apart from phonological features. Indeed, overt pronouns can convey given or new information in the discourse, while pro can only be given or presupposed. In this sense, it is a “topical” element.1 However, what kind of topical element is pro? I propose that the limitation in (2a) is due to the fact that pro cannot be merged in a TopP projection nor can it move there as a consequence of its licensing conditions (Rizzi 1986): pro cannot move away from Spec-AgrSP, because it must satisfy EPP requirements and receive its
-features. Nonetheless pro, as given information, must be endowed with a [+TOP] feature to be made visible at LF for interpretation (pro being an empty category, there is no requirement of visibility at PF). Interpretative problems do not arise for overt subject pronouns. First of all, they are full DPs and do not need to be licensed by an antecedent DP when they are “active” or “semi-active” (in the sense of Chafe 1987);2 besides – following (1a)) – they are either Topics or Foci, hence they do not sit in Spec-AgrSP.3

Who is lui? Reference of Italian overt and covert subject pronouns 323

My proposal to account for the interpretability of pro [+TOP] feature is based on a reconsideration of Rizzi’s (1986) assumption that both licensing conditions and the interpretation of pro are obtained through Spec-Head agreement in AgrSP. Specifically, I will show that AgrS° can only license pro (for person, number and gender features), but it cannot provide an interpretation (i.e., a referent), to the null pronoun (cf. also Frascarelli 2005). A sentence such as (3) gives evidence of the split between licensing and interpreting conditions: (3)

pro mangia. 0 eat.PRES.3SG ‘He/she/it eats’

When (3) is pronounced out of the blue, it is grammatical, but there is no possibility for the reader to identify to what pro refers, in that we do not know who is eating: pro is licensed, but not interpreted. The issue is to understand at which point of the derivation pro is interpreted. An initial tentative explanation is to assume that the Agree-feature and the External-feature (henceforth EF; cf. Chomsky 2005) of C° create two copies of pro: one in Spec-TP4 for EPP requirements and one in Spec-CP for the interpretation of pro as a given element (EF assuming the characteristics of a [+TOP] feature in this case). However, if this line of reasoning is sound for an overt DP, it misses the point for an empty element such as pro. A constituent with no phonological content cannot be identified through its place in the derivation (since pro is not a trace) because the copy in the left periphery, such as the copy in Spec-TP, cannot be spelled out. Indeed, as already observed above, pro can only be identified through an antecedent DP, since it is not fully referential. Therefore, I propose that [+TOP] “default” feature of pro is made visible at the C-I interface through an index which links pro, in Spec-TP, with a specific TopP position in the sentence, as (4) shows for a complex sentence (a more detailed structure will be provided in section 2). Indices are taken to be a sort of expository convention, a way to describe the relation with the antecedent DP illustrating how pro is identified while allowing for the unchecked [+TOP] default feature to be recognised: (4)

[TopP DPi [TopP […[TP [vP [+TOP]

]] … ]]] [TopP … [TP proi]…] [+TOP] (default)

324 Alessandra Trecci Assuming the structure given in (4), the behaviour of the null category in (2a) is no longer a controversial point. Its apparent contradiction is due to the fact that the lack of phonological material obliges pro to be licensed in Spec-TP, but such a lack also provides pro with a default [+TOP] feature available for interpretation.5 This feature must be checked through indices due to the impossibility for pro to be spelled out.6 As I will show in sections 2 and 3.1, what I propose is not a check proper: the relevant relation is not between indices, but between edges of particular phases. Pro needs this mechanism to be interpreted, since the verb cannot provide a referent to the null subject. Hence, referential properties are defined through an antecedent DP, which conveys information about the main subject matter of the discourse. Following the structure in (4) and adding some context (i.e., a Topic constituent) to sentence (3), pro can now be given a referent: (5)

Quando Carloi torna a casa dal lavoro, When C. come back.PRES.3SG to home from.DET work proi mangia. 0 eat.PRES.3SG ‘When Charles comes back home form work, he eats’

Cardinaletti (1997, 2004) analyses similar data and argues for two subject positions, SubjP and AgrsP, dedicated to full DPs/strong pronouns and weak pronouns, respectively. She considers the higher position to be the place in which the subject-of-predication feature is checked, while grammatical subjects check Case and phi-features in the low (AgrsP) position. Thus, she establishes a relation between full DPs/strong pronouns and “subjects of the predication”, a notion that is very similar to the discourse property of “topicality” that we assume as relevant here.7 However, Cardinaletti rejects the possibility that strong subjects are Topics and argues instead that SubjP is part of the Infl domain. This claim implies that PPs of psych-verbs (“quirky subjects”) are in subject position, allowing psych-verbs to select indirect subjects such as they select indirect objects. Cardinaletti (2004) supports her claim with a number of arguments that cannot be addressed here. However, in considering Poletto’s (2000) and Northern Italian data, the author also concludes that subjects can appear in the C-domain, though they need not. This is exactly in the spirit of the present proposal: some subjects are Topics, while others (all types of weak pronouns8 and non-referential DPs) are not. The claim that DPs are Topics in NSLs is therefore in line with the idea that the Infl domain is the place where morpho-syntactic features are licensed,

Who is lui? Reference of Italian overt and covert subject pronouns 325

while the C-domain is the place in which syntactic derivation establishes the link with the discourse (cf. Rizzi 1997; also in Cardinaletti 2004). And when an element is a Topic, it can be linked with a weak element in the Infl domain (e.g., pro) as I will show in the next sections. Indeed, arguing that non pronominal DPs are subjects obscures the relationship that the antecedent DP (in the discourse, hence in the CP-phase) establishes with pro in Infl.

2. Different kinds of Topic and different kinds of binding In a language like Italian, multiple Topic constructions are allowed. However, Topics do not share the same syntactic, semantic and prosodic features. I will show below how coreference relations are affected by those differences. I will follow Frascarelli and Hinterhölzl (to appear) in distinguishing three different kinds of Topic, namely: Aboutness/Shift, Contrastive and Familiar Topics. The authors distinguish each type according to its semantic, syntactic and prosodic features and propose the following hierarchy in the C-domain: (6)

[ABOUTP [CONTRP [FOCP [FAMP* [IP

Aboutness Topics are identified as the matter of concern in the discourse: they sit in the highest Topic position in the C-domain, are non-recursive and show a raising (L*+H)9 pitch contour; Contrastive Topics, which convey some contrast between two or more Topics, sit in a position between Aboutness Topics and the Focus Phrase, are non-recursive and show a H* tone; and Familiar Topics, which are established Topics and part of the background information, sit lower than the Focus Phrase (thus, after whelements), can be recursive and show a L* pitch contour. Here are some example of Aboutness (7a), Contrastive (7b) and Familiar Topics (7c): (7)

a. La casa, Laura la comprerà a maggio. DET house L. 3SG.F.ACC buy.FUT.3SG at May ‘As for her house, Laura is going to buy it in May’ mangio ora, mentre b. La pasta, la DET pasta 3SG.F.ACC eat.PRES.1SG now while la torta la mangerò più tardi. DET cake 3SG.F.ACC eat.FUT.1SG more late ‘I will eat pasta now, while I will eat the cake later’

326 Alessandra Trecci c. Loro, gli amici di Leo, sono simpatici. 3PL.NOM DET friend.PL of L. be.PRES.3PL nice ‘They – Leo’s friends – are nice’ According to the present proposal, pro is coreferent with the highest Topic in the C-domain, i.e., the Aboutness Topic, as is illustrated in (8): (8)

A: [AboutP Leoj, [FocP quando [AgrSP ha L. when have.PRES.3SG conosciuto Marco? know.PP.SG.M M. ‘When did Leo meet Mark’ B: [FamP Marcoi, [AgrSP pro*i/j li/*j’ ha M. 0 3SG.M.ACC have.PRES.3SG conosciuto tempo fa know.PP.SG.M time ago ‘Mark, he met him some time ago’

In (8) Leo is what the discourse is about; hence it sits in the highest position in the C-domain (AboutP). In the B-sentence, Marco is also introduced as a Topic. However, this is a different kind of Topic with respect to Leo i.e., Marco is a Familiar Topic. Therefore, the null pronoun in the B-sentence is linked with the Aboutness Topic Leo in the question,10 while the object clitic is coreferent with the Familiar Topic, Marco. The reverse interpretation is ungrammatical as (8) shows. Examples like (8) demonstrate that pro does not allow for a Topic shift and that it always refers to the (highest) discourse Topic. This line of reasoning is all the more evident if we stress the fact that the sole function of pro is to allow the speakers to continue talking about the same referent i.e., pro allows for topic continuity. Therefore, we may assume the existence of a long distance relation among the ultimate edges (CP’s strong phase) of the sentences in a discourse and, as we will see below, this Topic chain exclude the presence of overt pronouns. The long distance binding is justified by the fact that I assume a link for the edges of two phases: Spec-AboutP, which is the edge of the strong CPphase and Spec-TP, which is the edge of a weak phase, but made visible by the fact that pro still has a feature to be interpreted, i.e. [+TOP]. This is consistent with the current Minimalist approach that views edges as the sole

Who is lui? Reference of Italian overt and covert subject pronouns 327

visible (“accessible”, in Chomsky’s terms) positions for syntactic operations and interface interpretation. A hypothesis worth exploring is to assume that Aboutness Topics are copied (silent copies) in the appropriate AboutP position in every sentence of a discourse till Topic is changed, so that a short distance binding could be assumed between pro and the silent Topic in the same sentence. The idea is challenging and its verification is worthy of future study. Overt pronouns can be Topic or Focus elements and, being full DPs, they do not need to be licensed in any given configuration. Hence they sit in the appropriate CP position in order to make their [+TOP] or [+FOC] features available at the interfaces. From their A’-positions, overt pronouns can be bound by Contrastive and Familiar, but never by Aboutness Topics, or by embedded constituents. As we have seen in (7c), an overt pronoun can corefer with a Familiar Topic, which occurs in the C-domain of its same sentence. Still, this coreference can be established when the overt pronoun is focalized: (9)

LORO sono simpatici – gli amici di Leo. 3PL.NOM be.PRES.3PL nice DET friend.PL of L. ‘THEY are nice – Leo’s friends’

In 3.2. the behaviour of overt pronouns will be examined in detail.

3.

Data

All sentences presented in the following paragraphs are to be considered as uttered out of the blue, unless a plausible context is explicitly provided.

3.1. Topics and syntactic subjects In standard Italian, morphosyntax offers no evidence to distinguish a syntactic subject from a Topic. While object Topics (DO/IO) are easily discernible by the presence of a clitic in argument position, subject Topics are not resumed in the sentence. Nonetheless, following Alexiadou and Anagnostopoulou’s (1998) argument, I assume that in Italian a constituent, which is taken to be the syntactic subject of a sentence, is in fact a Topic. That is to say, I assume a discourse requirement by which every sentence must have a

328 Alessandra Trecci Topic just as in syntax the EPP states that every sentence must have a subject. Let us consider the following sentences: (10) a. Giannij dice che pro*i/j/*k… G. say.PRES.3SG COMP 0 studierà di più il prossimo mese. study.FUT.3SG of more DET next month che luii/??j/k… b. Giannij dice G. say.pres.3sg comp 3sg.m studierà di più il prossimo mese. study.FUT.3SG of more DET next month ‘John says he will study more next month’ I propose that in the (out-of-the-blue) examples (10a) and (10b), Gianni is not the syntactic subject (as is generally assumed). If this was the case, these sentences would be topicless, since no other element can perform this discourse function. Considerations on coreference relations support the analysis that Gianni is a Topic. In (10a), as uttered out of the blue, pro is obligatorily coreferent with Gianni and since the null pronoun can never introduce new referents, Gianni is the sole plausible antecedent for the interpretation of the reference of the pronoun. In (10b), the overt pronoun can be coreferent with Gianni; however, it is more likely to introduce a new referent (favourite interpretation). In fact, since topic continuity is realised through pro, the coreference between lui and Gianni conveys additional discourse information than simple coreference: the sentence means that only Gianni and no one else will study more next month. But if we do not want to add any additional context maintaining the sentence as uttered out of the blue, we must reject this interpretation. Thus, the most plausible function for the overt pronoun in (10b) is to introduce a new referent and be new information. Likewise, we can conclude that the most plausible interpretation of lui is that of a disjoint reference with Gianni.

3.2. Embedded contexts Let us now consider overt pronouns. Their main function is to shift the reference of the established Topic or, in a more common terminology, to introduce a new referent and provide new information, as in (10b). This charac-

Who is lui? Reference of Italian overt and covert subject pronouns 329

teristic allows overt pronouns to refer to embedded constituents, as is shown by the contrast between (11a) and (11b) in which pro is ungrammatical under the intended coreferential interpretation: quadri di Leoi che luii/j… picture.PL of L. COMP 3SG.M ama sono molto costosi. love.PRES.3SG be.PRES.3PL very expensive.PL

(11) a. I

DET.PL

b. I

quadri di Leoi che pro*i/j… picture.PL of L. COMP 0 ama sono molto costosi. love.PRES.3SG be.PRES.3PL very expensive.PL ‘The pictures of Leo that he loves are very expensive’ DET.PL

In (11a) and (11b), ‘the pictures’ is the established Topic.11 In (11a) the overt pronoun can be coreferent with Leo or it can introduce a new referent (external to the text). In both cases it shifts the attention of the reader from the established Topic to another constituent. On the other hand, pro in (11b) cannot be coreferent with Leo. This provides crucial evidence in favour of the present analysis since it shows that pro only allows for topic continuity. Indeed, Leo is not a Topic in (11), but a modifier of it. Hence, in (11b) pro has no possible binder. The only interpretation available for pro is one in which the null pronoun is coreferent with some other constituent in the discourse (i.e., a previously established Aboutness Topic). These examples challenge the general assumption that in Italian pro is the “unmarked choice” and that the overt category surfaces only for reasons of emphasis or contrast. The data show that the two categories have different discourse functions, hence different structural positions and different mechanisms of interpretation. Besides, they provide clear evidence that the grammatical device for pro licensing is not one and the same with the interpretation of its reference (cf. 2). An analysis of spoken corpora (Bonvino 2005) also shows that the realisation of overt pronouns is not necessarily due to “distance”. Consider the following: (12) L1: era un tizio di Formia che ha fatto, che lavora nell’ambito della moda però come hobby pro è fotografo ed pro è molto bravo. pro ha fatto una specie di raccolta di- di- cioè pro ha fatto diverse mostre. C’era un vernissage appunto una mostra sui barbieri. L2: ma dai.

330 Alessandra Trecci L1: barbieri di tutto il mondo; questo c’ha la fissa dei barbieri L2: carino. L1: pro fa più tipo reportage, proprio il suo tipo diciamo di fotografi, quindi molto classico e tutto quanto L2: e pro è bravo? L1: sì molto bravo/tra l’altro io l’ho detto al capo e lui ha mandato un redattore ‘L1: there was this guy from Formia that works in the fashion business, but as a hobby he is a photographer and he is very good. He published a sort of collection of- of-, indeed he had a few exhibitions. There was a vernissage, an exhibition on barbers. L2: really? L1: barbers from all over the world, he is fond of barbers. L2: it sounds nice. L1: he is more of a photo-reporter, this is its way of taking pictures, a classical style. L2: is he good? L1: yes, very good, by the way I told my boss and he sent a reporter’ io l’ ho detto [al capo]i e luii 1SG 3SG.M.ACC have.PRES.1SG say.PP.SG to.DET boss.SG and 3SG.M ha mandato un redattore (…) have.PRES.3SG send.PP INDEF reporter.SG ‘I said it to my boss and he sent a reporter’ First of all, note that the Aboutness Topic, ‘this guy from Formia’, is always resumed by null pronouns in accordance with the proposal that the interpretation of pro depends on the established Aboutness Topic. When the speaker, L1, needs to shift Topic (and to pronominalize a referent embedded in a PP) the overt pronoun surfaces, in spite of the fact that the pronoun is strictly adjacent to the embedded phrase.12 This is evidence that adjacency has no compelling role in determining the choice of the pronoun type. The choice of pronoun type is determined by discourse grammar: pro corefers with Aboutness Topics and the overt category introduces all other kinds of referents: Familiar Topics as in (7c) and (9), referents external to the text as in (10b) or embedded constituents as in (11a) and (12).13

3.3. Quirky subjects Calabrese (1986) points out that the surface subject of a psych-verb cannot be coreferent with pro. His examples are given in (13):

Who is lui? Reference of Italian overt and covert subject pronouns 331

Maria, proi vuole (13) a. Poiché a Giovannii piace Since to G. like.PRES.3SG M. 0 want.PRES.3SG rendersi interessante ai suoi occhi. make.INF.REFL interesting to.DET.PL POSS.3SG.PL eye.PL ‘Since John likes Mary, he wants to make himself interesting’ l’ iconografia, b. Poiché ad Andreai interessa Since to A. interest.PRES.3SG DET iconography proi abbandonerà la linguistica con piacere. 0 abandon.FUT.3SG DET linguistics with pleasure ‘Since Andrew is interested in iconography, he will abandon linguistics with pleasure’ If we assume for psych-verbs Belletti and Rizzi’s (1988) structure, where the Experiencer is in a higher position than the surface subject, it can be plausibly assumed that the PPs that contain the DPs Giovanni and Andrea are Topic constituents in (13).14 In particular, these DPs are the referents under discussion (i.e., the “Thema” of the predication, in Calabrese’s words, the Aboutness Topic in the terminology I assume), hence, according to my proposal, pro in the main sentence must be coreferent with the DP inside the PP and not with the syntactic subject (which is not a Topic). Indeed, if we force a coreference with the syntactic subject – such as in (14a) where the (adjectival) predicate shows feminine agreement – the sentence that we obtain is ungrammatical. On the other hand, in (14b) lei can corefer with the subject since, as expected, the overt pronoun is free to refer to all elements that are not Aboutness Topics: (14) a. *Poiché a Giovanni piace Mariaj, pro*j è Since to G. like.PRES.3SG M. 0 be.PRES.3SG contentissima. overjoyed.SG.F b. Poiché a Giovanni piace Mariaj, leij è Since to G. like.PRES.3SG M. 3SG.F be.PRES.3SG contentissima. overjoyed.SG.F ‘Since John likes Mary, she is overjoyed’

332 Alessandra Trecci Calabrese also argues that PPs of non-psych verbs cannot be coreferent with pro since they cannot serve as “Thema” of the predication. His example is given in (15): ha dato Giacomo, (15) Poiché a Carloi glielo Since to C. 3SG.DAT.3SG.ACC have.PRES.3SG give.PP G. pro*i è stato molto contento. 0 be.PRES.3SG stay.PP.SG.M very happy.SG.M ‘Since James gave it to Charles, he was very happy’ In (15) a Carlo is merged lower in the vP phase than a Giovanni in (13a) consistent with its theta-role. Thus, we must assume that a Carlo cannot reach the Aboutness Topic position and that it cannot be available as a Topic. Surface position is immaterial in the present analysis. These sentences are evidence that the discourse grammar is encoded in syntax and that the relations between the two are fundamental for the establishment of coreference relations. It is, however, important to notice that multiple topics in the left periphery can create opacity effects. Consider the following example, in which an object clitic pronoun occurs: (16) (Per quanto riguarda) Leoi, Giannij dice che (As for) L. G. say.PRES.3SG COMP proj/ ?i loi/ ?j inviterà alla festa. 0 3SG.M.ACC invite.FUT.3SG to.DET party ‘As for Leo, John says that he will invite him to the party’ As we can see, in one interpretation the antecedents of the pronouns maintain the same syntactic relations: the subject (Gianni) is the antecedent of the subject pro and the Topic, as an object, is the antecedent of the object pronoun. However, the reverse interpretation is also possible (though marked for some speakers). Indeed, without a proper context, the relevant ambiguity cannot be solved. I leave the analysis of these structures for future research. 3.4. Restrictives vs. appositive clauses In relative clauses a binding asymmetry can be observed between restrictives and appositives: pro is allowed to corefer with the subject within the

Who is lui? Reference of Italian overt and covert subject pronouns 333

relative clause only in a restrictive context (17a), but not in an appositive, in which the intended coreference is ungrammatical (17b): invitato, pro?i (17) a. La ragazza che Leoi ha DET girl COMP L. have.PRES.3SG invite.PP 0 l’ ha conosciuta tempo fa. 3SG.F.ACC have.PRES.3SG know.PP.SG.F time ago ‘The girl that Leo invited, he knew her some time ago’ invitato, pro*i b. Maria, che Leoi ha M. COMP L. have.PRES.3SG invite.PP 0 l’ ha conosciuta tempo fa. 3SG.F.ACC have.PRES.3SG know.PP.SG.F time ago ‘Mary, who Leo invited, he met her some time ago’ The asymmetry noted in (17a) and (17b) can be explained if we assume a different structural analysis for the two types of relative clauses, as argued in recent works (cf., among others, De Vries 2004; Del Gobbo 2003; Frascarelli and Puglielli 2005). The head of an appositive clause is a fully referential DP, while in the restrictive the reference of the head is always dependent on the description provided in the relative clause proper. Hence, in (17a) the content of the relative clause is a specification of the person we are talking about while in (17b) the appositive sentence gives only additional information about the DP, Maria. Following Kayne’s (1994) raising analysis, the behaviour of restrictives is straightforward: the restrictive clause is the complement of the D° head la in (17a), so that the NP-head (ragazza) and the restrictive clause are part of the same DP, sitting in the C-domain of the matrix clause. Thus, the Topic in (17a) is not ‘the girl’, but ‘the girl that Leo invited’. This “complex” Topic constituent is coreferent with the (obligatory) object clitic and this coreference is also shown by feminine agreement on the past participle. Since the subject of the relative clause (i.e, Leo) is part of this topical element, pro can pick up Leo as a possible antecedent (the sole available in the sentence) and the coreference is allowed: (18) [AboutP [DP [D° la [CP [NP ragazza] [C° che [TP Leoi ha invitato]]]]]] [TP pro?i l’ha conosciuta tempo fa]

334 Alessandra Trecci As is shown in (18), the structure proposed is one in which the relative DP is merged in AboutP as Topic of the sentence ‘pro l’ha conosciuta tempo fa’. The intended coreference is marginal, but it is allowed by the fact that the subject of the relative is part of the Topic constituent. On the other hand, in (17b) the appositive clause is not part of the DP Maria that is merged independently as the Aboutness Topic of the sentence. Since appositive clauses convey additional information about the DP-head, we must assume that Leo is introduced as a new referent. In section 3.2 I have shown that new referents are coreferential with overt pronouns, hence pro cannot be coreferent with the subject of an appositive clause and the intended coreference relation cannot be allowed. (19) [AboutP [DP Maria] [CP [C° che [TP Leoi ha invitato]]]] [TP pro*i l’ha conosciuta tempo fa] The intended coreference could be obtained only by the introduction of an overt pronoun: (20) [AboutP [DP Maria] [CP [C° che [TP Leoi ha invitato]]]] [TP luii l’ha conosciuta tempo fa] To conclude, pro in (17a) can be (marginally) interpreted as coreferential with the subject of the restrictive clause because it is part of the Aboutness Topic. Remember, however, that this possibility is excluded in a sentence such as (11a), repeated here as (21), where both the antecedent and pro are part of the Aboutness Topic and the overt DP is embedded in a PP: quadri [di Leo]i che pro*i/j… / luii/j… ama picture.PL of L. COMP 0 3SG.M love.PRES.3SG sono molto costosi. be.PRES.3PL very expensive.PL ‘The pictures of Leo that he loves are very expensive’

(21) I

DET.PL

We can tentatively attribute the ungrammaticality of (21) to a restriction concerning syntactic embedding, so that when the antecedent is embedded, resumption is only allowed through an overt pronoun. Further investigation is needed to understand the nature of such a restrction. This issue will be the subject of future research.

Who is lui? Reference of Italian overt and covert subject pronouns 335

4. Conclusions This paper has provided evidence that the interpretative device for pro and the overt category in Italian is defined by discourse grammar. I assumed that in NSLs preverbal subjects sit in A’-position due to a sort of extended EPP condition: discourse grammar requires an Aboutness-Topic just like a sentence must have a syntactic subject; if no other constituent in the sentence assumes this function, the alleged subject must be considered as such. I showed that licensing and interpreting conditions must be distinguished, depending on different formal requirements; in fact, pro is not interpreted through its position in the structure, but is identified through an antecedent DP. The antecedent and pro are bound by a particular configuration in the sentence: the DP must sit at the edge of CP (strong phase), hence it must be a Topic, and pro must sit in Spec-TP to receive its
-features. These two positions are the eligible places for the interpretation of pro. Further studies on topicality should determine the “distance” required to establish the coreference between pro and its antecedent: either the link is “strong” enough for the grammar of discourse to allow a long distance binding from the AboutP position in a C-domain which is different from the domain of pro; or the long distance binding could be obviated assuming the occurrence of silent copies of the established Aboutness Topic in every local C-domain, till the Topic is changed (for this line of analysis, cf., Frascarelli 2005). As regards overt pronouns, it is argued that, since the function of pro is to allow Topic continuity, overt pronouns, as free elements, can be coreferent with any constituent. They also surface when there is a need for a disjoint reference. In fact, they can bind Focus constituents and all nonAboutness Topics. Moreover, they can “reach into” embedded constituents to retrieve a referent.

Acknowledgements Part of this paper was presented at the IGG XXXI and I thank the audience for their helpful observations. Special thanks to Mara Frascarelli for all her precious comments and insightful suggestions and to Annarita Puglielli for her support. I am also grateful to Anna Cardinaletti and to an anonymous reviewer for their valuable remarks. I take responsibility for all errors.

336 Alessandra Trecci Notes 1. I would like to remind the reader that in the definition of pro as a topical element I do not include its non-referential use. Expletive pro is therefore not addressed here. 2. As for the internal licensing of strong pronouns, cf. Cardinaletti and Starke (1999: 195) 3. It is a general assumption that subject Foci do not move through Spec-AgrSP position to reach FocP. Furthermore, I follow Frascarelli (2000) in assuming that Topics do not move from an IP-internal position, but merge in TopP, hence they cannot sit in Spec-AgrSP at any time in the derivation. 4. Chomsky (2005) assumes Spec-TP and not Spec-AgrSP to be the licensing place of the verb external argument. 5. Cardinaletti (2004) argues that all weak pronouns behave alike as far as topic continuity is concerned. I will not pursue this issue here and I will focus only on the referential occurrences of pro and the use of strong pronouns. 6. Lack of phonological material implies the impossibility for pro to be identified through morpho-syntactic cues. Only traces can be identified by their position in the sentence, since they are silent copies of a constituent merged in a different phase. 7. Among other things, Cardinaletti’s (2004) refusal of Topic-subjects is based on the assumption that Topics are always given elements (contrary to subjects). However, the analysis of Topicalization in spoken corpora (Frascarelli 2000; Frascarelli and Hinterhölzl to appear) shows this connection is not a compelling one: topical elements can be newly introduced in the discourse and be used to propose a shift in the conversation. Of course, they must be part of the shared knowledge of participants, but this condition does not imply their “givenness” in the discourse. Hence, the distinction between Topics and subjects-of-predication is not so clear-cut as is argued in Cardinaletti’s work. The claim that referential preverbal subjects in Italian are Topics (or Foci) is also supported by intonational analysis, that shows their non-inclusion in the prosodic unit of a broad Focus sentence (cf. Frascarelli 2005 for discussion). 8. Thus, egli and esso are equivalent to pro in this respect, according to the fact that they are weak pronouns. 9. Tonal events are described through ToBI’s framework (cf. Pierrehumbert 1980 and subsequent works): L*+H describes a complex tone that starts Low on the tonic vowel and reaches a High tone on the post-tonic; H* is defined by a High tone on the tonic and L* by a Low tone on the tonic vowel. 10. The relevant Topic chain also includes the subject pro in the question. 11. The Aboutness Topic is actually the whole restrictive relative (cf., 3.4.), i.e., ‘the pictures of Leo that he loves’. This is however immaterial for the present analysis.

Who is lui? Reference of Italian overt and covert subject pronouns 337 12. For a complete survey of this specific phenomenon cf. Trecci (2004). 13. Anna Cardinaletti (p.c.) points out that a sentence like: (i) pro sono simpatici, gli amici di Leo DET friend.PL of L. pro be.PRES.3PL nice is also possible, showing that pro can also be coreferent with Familiar Topics. For space limitations, I cannot discuss this issue here and refer to Frascarelli (2005) in which evidence is provided that a sentence like (i) is only possible in a context in which both pro and the Familiar Topic are coindexed with the local Aboutness Topic. 14. Remember that the clitic resumptive pronoun with oblique Topics is optional.

References Alexiadou, Artemis and Elena Anagnostopoulou 1998 Parametrizing AGR: word order, V-movement and EPP checking. Natural Language and Linguistic Theory 16: 491–539. Belletti, Adriana and Luigi Rizzi 1988 Psych-verbs and theta-theory. Natural Language and Linguistic Theory 6: 291–352. Bonvino, Elisabetta 2005 Le Sujet Postverbal en Italien Parlé: Syntaxe, Zones et Intonation, Biblioteque des faits des langues, Paris: Ophrys. Büring, Daniel 1999 Topic. In Focus – Linguistic Cognitive and Computational Perspective, Peter Bosch and Rob van der Sandt (eds.), 142–165. Cambridge: Cambridge University Press. Calabrese, Andrea 1986 Pronomina. Some properties of the Italian pronominal system. In Papers in Theoretical Linguistics, Naoki Fukui, Tova R. Rapoport and Elizabeth Sagey (eds.), 1–46. (MIT Working Papers in Linguistics 8). Cambridge, MA: MIT Press. Cardinaletti, Anna 1997 Subjects and clause structure. In The New Comparative Syntax, Liliane Haegeman (ed.), 33–63. London /New York: Longman. 2004 Towards a cartography of subject positions. In The Structure of CP and IP: The Cartography of Syntactic Structures, Vol. 2., Luigi Rizzi (ed.), 33–63. Oxford: Oxford University Press. Cardinaletti, Anna and Michael Starke 1999 The typology of structural deficiency: A case study of the three classes of pronouns. In Clitics in the Language of Europe, Henk van Riemsdijk (ed.), 145–233. Berlin /New York: Mouton de Gruyter.

338 Alessandra Trecci Chafe, Wallace 1987 Cognitive constraints on information flow. In Coherence and Grounding in Discourse, Russel Tomlin (ed.), 21–51. Amsterdam: John Benjamins. Chomsky, Noam 1981 Lectures on Government and Binding: The Pisa Lectures. Dordrecht: Foris. 1995 The Minimalist Program. Cambridge, MA: MIT Press. 2001 Derivation by phase. In Ken Hale. A Life in Language, Michael Kenstowicz (ed.), 1–52. Cambridge, MA: MIT Press. 2003 Beyond explanatory adequacy. In Structures and Beyond. The Cartography of Syntactic Structures, Vol. 2, Adriana Belletti (ed.), 104– 131. Oxford: Oxford University Press. 2005 On phases. Ms., MIT. Cinque, Guglielmo 1990 Types of A’-dependencies. Cambridge, MA: MIT Press. Del Gobbo, Francesca 2003 Appositives at the interface, Ph.D. diss., University of California, Irvine. De Vries, Mark 2004 Head-internal relative clauses in Dutch? Linguistics in the Netherlands 21: 193–204. Frascarelli, Mara 2000 The Syntax-Phonology Interface in Focus and Topic Constructions in Italian. Dordrecht: Kluwer. 2005 Aboutness Topics and the interpretation of pro. A new approach to the Null Subject Parameter. Ms., Università degli Studi Roma Tre, Rome. Frascarelli, Mara and Roland Hinterhölzl to appear Types of topics in German and Italian. In On Information Structure, Meaning and Form, Susanne Winkler and Kerstin Schwabe (eds.). Amsterdam: John Benjamins. Frascarelli, Mara and Annarita Puglielli 2005 A comparative analysis of restrictive and appositive relative clauses in Cushitic languages. In Contribution to the IGG XXX, Laura Brugè, Giuliana Giusti, Nicola Munaro, Walter Schweikert and Giuseppina Turano (eds.), 279–303. Venice: Cafoscarina. Kayne, Richard S. 1994 The Antisymmetry of Syntax. Cambridge, MA: MIT Press. Pierrehumbert, Janet 1980 The phonology and phonetics of English intonation. Ph.D. diss., MIT, Cambridge, MA.

Who is lui? Reference of Italian overt and covert subject pronouns 339 Poletto, Cecilia 2000 The Higher Functional Field: Evidence from Northern Italian Dialects. Oxford: Oxford University Press. Rizzi, Luigi 1986 Null objects in Italian and the theory of pro. Linguistic Inquiry 17 (3): 501–557. 1997 The fine structure of the left periphery. In Elements of Grammar. Handbook in Generative Syntax, Liliane Haegeman (ed.), 281–337. Dordrecht: Kluwer. Ross, John R. 1986 Infinite Syntax. Norwood, NJ: Ablex Publishing Corporation. Trecci, Alessandra 2004 Distribuzione e referenza del pronome soggetto in italiano, Ph.D. diss., Università degli Studi Roma Tre, Rome.

Satisfying the Subject Criterion by a non subject: English Locative Inversion and Heavy NP Shift Luigi Rizzi and Ur Shlonsky

Introduction This paper extends the approach to clausal subjects of Rizzi (2003) and Rizzi and Shlonsky (2006) to English Locative Inversion, a construction in which a locative PP seems to play a subject-like role, while remaining distinct in other respects from regular subject DP’s. Section 1 summarizes the framework based on the Subject Criterion. Section 2 presents the salient properties of Locative Inversion, building on Stowell’s (1981) analysis. Section 3 motivates the analysis of Locative Inversion in terms of indirect satisfaction of the Subject Criterion. Section 4 discusses one familiar subject-like property of inverted locatives, namely, their sensitivity to thattrace effects. Section 5 briefly discusses a variety of clause-initial locative PPs with genuine subject properties. Section 6 takes up the relationship between Locative Inversion and Heavy NP Shift, building on work by Culicover and Levine (2001). Section 7 tries to explain why locative inversion may not feed Raising and Section 8 is the conclusion.

1. Criterial Freezing and the Subject Criterion Our point of departure, developed in the papers cited above, is that the Government and Binding notion of the EPP, “clauses must have subjects”, should be thought of as a Criterial requirement in the sense of Rizzi (2003). It is assumed that a functional head Subj, distinct from and higher than T, must be locally c-commanded by an element (a specifier or a head) bearing the formal features of Subj, which we take to be Phi-features. Unlike feature checking or valuation, the satisfaction of a Criterion creates a freezing configuration: an element satisfying a Criterion can not be moved further (e.g., to satisfy another Criterion). Thus, a subject DP moved to Spec,Subj to satisfy the Subject Criterion is expected to be frozen in place, unable to

342

Luigi Rizzi and Ur Shlonsky

move on. This explains complementizer-trace effects, e.g., the que-trace effect in French, in a way alternative to the traditional Empty Category Principle approach: (1)

*Qui crois-tu que t Subj gagnera la course? ‘Who do you think that will win the race?’

The Subject Criterion straightforwardly captures the fact that subjects are harder to move than objects and other phrases. Still, (thematic) subjects are not generally unmovable: languages can have subject interrogatives, relatives etc. all of which implicate subject movement of some sort. Rizzi and Shlonsky (2006) addresses this problem and argue that subject movement can be made licit if some other device can be found to satisfy the Subject Criterion. Put differently, subject DPs can move out of IP only if they can avoid moving into Spec,Subj, the freezing position. One familiar case of a subject extraction strategy along these lines is the one generally used by Null Subject languages (cf. Rizzi 1982, 1990): word by word equivalents of (1) are grammatical in e.g., Italian, because the Subject Criterion can be satisfied by expletive pro, and the thematic subject can be extracted from a lower, non criterial position, thus escaping Criterial Freezing (the exact position of t is immaterial to our argument here). (2)

Chi credi che pro Subj vincerà t la corsa? ‘Who do you think that will win the race?’

2. Properties of Locative Inversion The following examples from Stowell (1981) illustrate Locative Inversion in English. In this construction, a locative PP occurs in clause-initial position, apparently satisfying the EPP requirement and allowing the thematic subject to remain in a lower, presumably predicate-internal position. (The original example number in Stowell (1981) is indicated as “St#”) (3)

a. Into the room walked my brother Jack b. On the table was put a valuable book c. Down the stairs fell the baby.

(St 31)

From the perspective of our analysis, a fundamental question raised by locative inversion is how the preposed PP can satisfy the Subject Criterion.

English Locative Inversion and Heavy NP Shift 343

One possibility is that the locative PP moves to Spec,Subj and satisfies the criterion directly. Another possibility is that criterial satisfaction is implemented indirectly, through the kind of devices which we postulate for subject movement in Rizzi and Shlonsky (2006). For example, it could be that the Fin(iteness) head in the left periphery of the clause satisfies the Subject Criterion, much as in cases of subject extraction. Movement of the locative phrase can be taken to serve the licensing requirements of the special featural endowment which allows Fin to satisfy the Subject Criterion (see below for a more detailed illustration of this strategy of subject extraction). Stowell (1981) provided clear evidence for the necessity of a more complex derivation of Locative Inversion than simple movement of the locative to subject position. In his terms, the locative PP could not occupy the EPP position at S-structure and had to be moved to a Topic position. One piece of evidence is distributional: locative inversion is infelicitous in structures which disallow embedded topicalization. Sentential subjects are a case in point. The similar status of (4a) and (4b) is immediately explained if the inverted locative occupies a topic position, while it would be unexpected if the locative could remain in subject position. (4)

a. *That in the chair was sitting my old brother is obvious b. *That this book, you should read is obvious

(St 38a)

Locative Inversion is disallowed in ECM structures which lack a Top position in the left periphery as the C system is truncated in such environments. Compare (5a) and (5b), (5)

a. *I expect [in the room to be sitting my old brother] b. *I expect [this book John to read]

(St 35a)

Stowell also noted that the preposed PP, like topics in English, appears to create an island for wh extraction of the theme argument. If the locative PP were in subject position, no island effect would be expected. (6)

John says that near his house lies a buried treasure

(St 33a)

(7)

*What does John say that near his house lies t?

(St 34b)

This argument in favour of the topic status of preposed locatives is perhaps less persuasive than the others because the ungrammaticality illustrated in (7) could be amenable to other properties which make the inverted subject

344

Luigi Rizzi and Ur Shlonsky

unmovable, a possibility which is supported by the analogous impossibility of inverted subject extraction in the presentational there construction.1 (8)

*How many students did John say that there arrived t?

However, long extraction of other types of elements, a temporal adjunct for instance, is equally ill-formed over a topic and over a preposed locative while being perfectly grammatical over an embedded subject. (9)

a. *When did he say that into the room Jack walked t? b. *When did he say that into the room walked Jack t? c. When did he say that Jack walked into the room t?

We may add to these pieces of evidence the observation that I to C movement cannot apply across locative inversion. This would be unexpected if the PP were spelled out in subject position: (10) a. *Is in the room sitting my old brother? b. *Did down the hill roll the baby? It should also be noticed that not just topicalization, but also other kinds of movement to the left periphery, such as wh movement (11) or focus movement (12), license locative inversion: (11) In what room is sitting my old brother? (12) a. IN THE LIVING ROOM is sitting my old brother (, not in the bedroom) b. IN THE LIVING ROOM, but not in the bedroom, were hanging portraits of GWB So, in line with Stowell’s analysis, the intermediate conclusion seems to be that the preposed locative plays a critical role in the satisfaction of the Subject Criterion, but that it can do so only parasitically, as it were, and in passing, when moving to a final destination in the left periphery. Clearly, if the locative PP could directly satisfy the Subject Criterion, the necessity of further movement to the left periphery would be unexpected. Stowell dealt with this necessary further step by invoking the Case Resistance Principle. In his analysis, the PP can move to subject position to

English Locative Inversion and Heavy NP Shift 345

satisfy the EPP because heads cannot select for specifiers belonging to particular categories, so that the inflectional head bearing the EPP feature could not restrict its specifier to the category NP (or DP); nevertheless, as P is itself a Case assigner, P and its projection cannot remain in a case assignment position due to Case Resistance. Since the EPP position is the locus of Nominative assignment, the locative PP must evacuate it. 3. Locative inversion and indirect satisfaction of the Subject Criterion This Case resistance analysis is based on the assumption that Case assignment or checking apply at S-structure, after movement, so that the locative PP first moves to satisfy the EPP, and then evacuates the position, voiding a violation of Case Resistance. This particular ordering assumption was standard in various versions of the Government-Binding framework, but it is not naturally expressible in Minimalist terms. In the Minimalist approach, Case properties are checked derivationally, as soon as the relevant configuration is created by Merge and Move. Therefore, if a constraint like Case Resistance were translated into this framework, it would amount to disallowing the creation of a configuration in which a Case assigner/checker ends up in a Case assigning/checking position. Let us implement and strengthen this conclusion by means of the working hypothesis that in English, T moves to Subj (namely, the position checking nominative Case is incorporated into the Subject Criterion position). Thus, an XP moving to Spec,Subj must not only satisfy the Subject Criterion, it must also be able to check the nominative Case features of T. If PPs cannot check nominative (in English) then, mutatis mutandis, they are barred from Spec,Subj. If movement of the locative PP to the EPP position is banned, then its contribution to the satisfaction of the Subject Criterion must be indirect, exploiting some other mechanism. The formal aspects of this mechanism are those developed for the analysis of subject extraction in Rizzi and Shlonsky (2006). Fin is the lowest head of the complementizer system, at the junction between the I and C systems (as in Rizzi 1997). Normally, Fin is directly merged on top of the SubjP layer, which terminates the IP system. The Subject Criterion must usually be satisfied by moving a nominal expression to Spec,Subj before Fin is merged. Thus, the direct merger of the Fin layer on top of the Subj layer, with no nominal expression satisfying the Subject Criterion in Spec,Subj gives rise to an ill-formed structure such as (13a), illustrated by (13b).

346

Luigi Rizzi and Ur Shlonsky

(13) a. *(I think that) Fin Subj will come a man b. *(I think that) will come a man Being non-nominal, Fin is normally unable to satisfy the Subject Criterion; hence movement of a nominal (argument or expletive) must take place to yield: (14) (I think that) a man will come / there will come a man There are, however, cases of nominal Fin, i.e., of Fin endowed with Phifeatures. Elaborating on Taraldsen (2001), Rizzi and Shlonsky (2006) propose that such a nominal Fin is spelled out as the -i of qui in French. It can be directly merged on top of the Subj layer and can satisfy the subject Criterion in the absence of a nominal in Spec,Subj (assuming that the critical configuration for criterial satisfaction is local c-command, encompassing Spec-head and local head-head relations). The derivation of a subject relative construction like l’homme qui viendra (‘the man who will come’) in French goes through the following intermediate stage (OP being the relative operator, merged in the appropriate thematic position, the object position in (15), as the verb is unaccusative): (15) -iPhi Subj … viendra OP The nominal Fin -i satisfies the Subject Criterion, so that the thematic subject OP can move without incurring a violation of Criterial Freezing (Rizzi 2003). However, the nominal features of Fin are uninterpretable, so they must be licensed (valued and checked, under standard Minimalist assumptions) by the subject OP passing through Spec,Fin, (viz. t’ in (16)) on its way to its final destination, Spec,Rel or Spec,Force. (Here as elsewhere, we mark traces with the traditional “t”. Everything we say is consistent with the copy theory of movement.) (16) L’homme OP qu- t’ -iPhi Subj viendra t This analysis can be extended to Locative Inversion, assuming that, in English-like languages, the relevant feature which renders Fin nominal is Loc, a particular kind of Phi-feature (though not one that enters into agreement or nominative Case checking in English; see also Landau 2005). Consider the following derivational level:

English Locative Inversion and Heavy NP Shift 347

(17) Subj T be [sitting [my old brother] [in the room]] The argument my old brother can be moved to Spec,Subj, satisfying the Subject Criterion, and yielding the uninverted order My old brother was sitting in the room. But suppose that in this configuration, the two constituents my old brother and in the room are “equidistant” from an external attractor (either directly from their thematic position, or because of some “smuggling” mechanism of the kind proposed in Collins (2005); see below for discussion), so that either one could move. The PP cannot move to Spec,Subj, as we have tentatively assumed; but Fin with special nominal features ( [+Loc] in this case ) may be merged, satisfying the Subject Criterion. (18) Fin+Loc Subj T be [sitting [my old brother] [in the room]] Things cannot stop at this point because the Loc feature in Fin (or whatever other feature determines the nominal content of Fin) is uninterpretable, and must be valued. Thus, the locative PP moves to Spec,Fin and values the Loc feature: (19) In the room Fin+Loc Subj T be [sitting my old brother t ] The derivation cannot stop at this point either, though. Fin is not a criterial head; it is not a head which assigns any special interpretive property to its Spec. In order to comply with the Last Resort guideline on movement which, following Rizzi (2003), we interpret in terms of criterion satisfaction (the formal expression of a scope-discourse interpretive property, in the terms of Chomsky 2004), the PP must move to a Top position, or to any other criterial position (e.g., Q or Foc in (11)–(12)) which would determine a well-formed chain. We thus capture Stowell’s observation that the locative PP must proceed to a left-peripheral position, Topic or the like. This accounts for the distributional properties of locative inversion, notably its absence from configurations characterized by a defective or truncated left periphery (4a) and (5a), and its incompatibility with auxiliary movement to C (10a,b). The revision of Stowell’s analysis is now consistent with the derivational approach to Case: The locative PP can’t move directly to Spec,Subj, but it can move to Spec,Fin+Loc, valuing the Loc feature which allows Fin to satisfy Criterial Freezing. The locative must move further from this position to

348

Luigi Rizzi and Ur Shlonsky

a scope-discourse (criterial) position in the left periphery, in order to satisfy Last Resort guidelines. The interplay of these different factors accounts for the peculiar distributional properties of the construction.

4. Locative inversion and some ECP effects Another important property of Locative Inversion is that it gives rise to that-trace type phenomena (cf. Bresnan 1977), much as subject extraction does in uninverted clauses (at least in commonly analyzed varieties of English. See Sobin (2002) and the discussion in Rizzi and Shlonsky (2006) for the analysis of dialectal variations concerning this point). The pattern is illustrated by the examples in (20)–(22), adapted from Bresnan’s work. (20) In which villages do you believe (*that) can be found the best examples of this cuisine? (21) In which villages do you believe (that) the best examples of this cuisine can be found? (22) The best examples of which cuisine do you believe (*that) can be found in these villages? In Rizzi and Shlonsky (2006), (22) is analysed on the basis of the assumption that the declarative complementizer that, expressing force as well as finiteness and functioning as the head of the embedded clause, is incompatible with the expletive-like status of nominal Fin in the successful cases of subject movement and extraction. As locative inversion also crucially involves the nominal Fin strategy, we expect it to be incompatible with that, as is the case in (20). No problem arises for locative extraction in uninverted clauses like (21) which do not depend on any special properties of the Fin head. The parallel between locative inversion and subject extraction holds in other environments as well. Recall from the ungrammaticality of (10a,b) that I to C cannot move the auxiliary past the inverted locative, as is expected if the locative occupies a left-peripheral position. In fact, locative inversion is incompatible with I to C movement, whatever the order: (23) a. *Did into the room walk my brother Jack? b. *Into the room did walk my brother Jack?

English Locative Inversion and Heavy NP Shift 349

The impossibility of (23b) is reminiscent of the incompatibility of I to C movement with subject questions, as in (24). (24) *Who did come? The similarity can be captured through the assumption that the nominal Fin head, acting as an expletive-like element which satisfies the Subject Criterion, is crucially involved in both locative inversion and wh movement of the subject (Rizzi and Shlonsky 2006). In I to C movement, on the other hand, Fin has the role of an attractor of a verbal element, much as the other functional heads of the inflectional space (Asp, T, Mood, etc.). It appears that Fin can be either nominal or verbal, but not both at the same time. Subject extraction and I to C movement, which require conflicting characteristics of Fin, are thus incompatible. In a similar vein, the impossibility of (24) is mirrored by the impossibility of (23b), if locative inversion also crucially involves the nominal Fin head. (23a) is independently excluded by the fact that the locative PP cannot occupy Spec,Subj in English, for reasons discussed previously.

5. A different “locative inversion” construction As Stowell (1981) observed, none of the above facts hold for the locative subject PPs in copular constructions like the following: (25) a. b. c. d. e.

Under the stars is a nice place to sleep How nice a place to sleep did John say that under the stars is? The fact that under the stars is a nice place to sleep is obvious I expected under the stars to be a nice place to sleep Is under the stars a nice place to sleep?

In the examples in (25), the PP does not create an island to extraction (25b), it is compatible with contexts disallowing topicalization (25c), and ECM (25d), and it does not conflict with I to C movement (25e). So, it looks as if in this case the phrase under the stars really is in subject position. Following Bresnan (1994), we may assume that the constituent under the stars in (25) is not a PP but a DP, headed by a null PLACE noun, a special structure presumably licensed by the copula in ways that remain to be elucidated (see also Kayne 2006). This categorial difference would allow under the stars to check nominative Case and control number agreement on the

350

Luigi Rizzi and Ur Shlonsky

verb, the latter point illustrated by the following contrast between a locative ‘subject’ and an inverted locative PP (from Bresnan 1994 citing Levine 1989). (26) a. Under the bed and in the fireplace are/*is not the best (combination of) places to leave your toys. b. Down through the hills and into the forest *flow/flows the little brook.

6. Heavy NP Shift and Locative Inversion Heavy NP Shift can affect objects, but not subjects in English: (27) Bill will give __ to John – a book which I decided to recommend for the literary prize (28) *__will give a book to John – the author whom I decided to recommend for the literary prize This asymmetry, ascribed to the ECP in Rizzi (1990), immediately follows from the Subject Criterion and the notion of Criterial Freezing: The DP the author whom I decided to recommend for the literary prize satisfies the Subject Criterion at some point in the derivation of (28), and is then rendered unmovable by Criterial Freezing. No problem arises for the object in (27) as there is no Object Criterion.2 In a recent paper, Culicover and Levine (2001) (henceforth C&L) argue that Heavy NP Shift may affect subjects in special constructions in English, such as Locative Inversion. In this section, we would like to show that this is expected under our analysis for the following reason: in the relevant construction, the Subject Criterion is satisfied by a preposed locative (in an indirect manner, as we have seen). Hence the “subject” DP (the DP which determines the agreement morphology on the inflected verb) is exempted from Criterial Freezing and can undergo Heavy NP Shift. C&L argue that English Locative Inversion really corresponds to two distinct constructions. The first one is restricted to sentences with unaccusative verbs and is compatible with ‘light’ subjects which remain in situ in predicate-internal position (e.g., they can precede low adverbs, as in (29) vs. (30)); the second is also consistent with unergative verbs and requires

English Locative Inversion and Heavy NP Shift 351

heavy subjects which end up in clause-final position, following all predicate internal material, as in (31) and (32): (29) Into the room walked Robin carefully (30) *In the room slept Robin fitfully (31) Into the room walked carefully – the students in the class who had heard about the social psych experiment that we were about to perpetrate (32) In the room slept fitfully – the students in the class who had heard about the social psych experiment that we were about to perpetrate C&L also show that the inversion construction with Heavy NP Shift may even involve some transitive sentences (see their (57)): (33) a. Outside in the still upright hangar, were having deep sighs of relief the few remaining pilots who had not been chosen to fly in the worst hurricane since hurricanes had names. b. *Outside in the still upright hangar, were having deep sighs of relief the pilots. C&L plausibly argue that Heavy NP Shift of the subject is crucially involved in the derivation of (31), (32) and (33). But why is the light subject construction restricted to unaccusatives? And why is Heavy NP Shift instrumental in permitting the unrestricted kind of inversion? Building on their analysis, and expressing it in terms of our conception of the clausal structure, we would like to propose that the configuration underlying (29) and its uninverted variant Robin walked into the room carefully is roughly as in (34). (The VP-final position of the adverb is presumably derived through leftward VP movement, see Cinque 1999). (34) Subj T+Phi [ Robin walk into the room carefully ] We assume, with Levin and Rappaport Hovav (1995: 221), that manner of motion verbs like walk display an unaccusative behaviour when they select a directional PP. We bypass various questions concerning the details of the first merge position of the arguments by assuming that the subject and the PP are direct dependents of the same head V with these verbs, while with

352

Luigi Rizzi and Ur Shlonsky

unergative and transitive verbs the subject is always first-merged as the specifier of a separate head, v (see Hale and Keyser 1993, and much related work). So, the subject of verbs which are incompatible with “light subject” locative inversion, e.g., sleep, is dependent on a little v head which does not itself select the locative. We also assume that dependents of the same head are equidistant from an external attractor in the sense of Chomsky (1995) and subsequent work.3 Phi in T can establish an AGREE relation with Robin in (29), thus determining the agreement morphology on the verb. The DP can be attracted to Spec,T and then to Spec,Subj, giving rise to the uninverted order Robin walked into the room carefully (recall that T+Phi raises to Subj via head movement. This technical assumption is needed to explain why a (non-nominative) element like a locative PP cannot be attracted to Spec,Subj in English). But there is another option: T+Phi can agree with Robin without attracting it (see section 7 for how the lack of attraction may be permissible). While Subj is unable to attract the locative PP, as we have assumed following Stowell, locative preposing can be instrumental for the satisfaction of the Subject Criterion in the indirect way that we have proposed in the previous section: The nominal Fin is merged on top of the SubjP layer and it satisfies the Subject Criterion. The locative phrase is attracted to Spec,Fin and then continues to move to a Criterial position (Topic, etc.). This yields (29) from (34). Verbs such as sleep, which are incompatible with this kind of locative inversion, are associated with an underlying structure in which the subject is merged in the specifier of a higher verbal head. (30) is underlyingly roughly as in (35). (35) Subj T+Phi [ Robin v [ sleep in the room fitfully ]] The uninverted order can come about through AGREE between T+Phi and Robin, and attraction of the latter to Spec,Subj (possibly via Spec,T+Phi), just as in (29). The inverted order, however, cannot be derived from (35). To see why this is so, consider what happens at the point at which the nominal Fin, endowed with Loc features is merged with (35) to yield (36). (36) Fin+Loc Subj T+Phi [ Robin v [ sleep in the room fitfully ]] While the Subject Criterion can be satisfied by the nominal Fin, the nominal Fin+Loc is unable to attract the locative across the thematic subject. Assuming Rizzi’s (2004) conception of Relativized Minimality, according to

English Locative Inversion and Heavy NP Shift 353

which locality effects arise between positions bearing features of the same class, and continuing to hold that Loc belongs to the class of Phi-features, it follows that attraction of the locative is blocked by Relativized Minimality in (36): Robin intervenes between Fin+Loc and the locative PP as it asymmetrically c-commands the locative and is the dependent of a higher head. An AGREE relation cannot be established and movement of the locative to Spec,Fin fails to take place. This analysis requires that the movement of the locative PP to Fin+Loc be assimilated to A-movement. If it were not, no locality effect would be induced by an intervening subject in (36). In fact, C&L provide interesting evidence that the position targeted by the locative PP in Locative Inversion (Spec,Fin in our analysis) is indeed an A-position. They show that Locative inversion eliminates a Weak-Crossover effect that arises in an uninverted structure. Contrast the examples in (37):4 (37) a. *Into every dogi ’s cage itsi owner peered (Topicalization,WCO) b. Into every dogi ’s cage peered itsi owner (PP-preposing, no WCO) We know that Weak-Crossover effects are alleviated when the variable of the operator c-commands the pronoun. If the PP into every dog’s cage moves through an A-position higher than the pronoun, Spec of Fin+Loc, under our assumptions, the WCO alleviation is explained. Consider now the “heavy” subject examples in (31) and (32). (31) is derived exactly like (29), except that the heavy subject DP is shifted to the right periphery. In (32), the subject DP first enters an AGREE relation with T+Phi and, differently from (29) moves to Spec,T: (38) Subj [DP the students of the class who had heard…] T+Phi [ tDP v [ sleep in the room fitfully ]] At this point, it is possible to satisfy the Subject Criterion by moving the heavy NP to the Spec,Subj where it would be would be subject to Criterial Freezing. Alternatively, the Subject Criterion can be satisfied by directly merging the nominal Fin+Loc, which would yield the following: (39) Fin+Loc Subj [DP the students of the class who had heard…] T+Phi [ tDP v [ sleep in the room fitfully ]]

354

Luigi Rizzi and Ur Shlonsky

The next step in the derivation is the valuation of the uninterpretable Loc feature in Fin, achieved by moving the locative PP to Spec,Fin: (40) [PP in the room] Fin+Loc Subj [DP the students of the class who had heard…] T+Phi [ tDP v [ sleep tPP fitfully ]] Why does this step of movement not violate locality, as it crosses the subject in Spec, T? To answer this, we must take a position with respect to the landing site of Heavy NP Shift. For concreteness, let us assume that the shifted phrase is right-adjoined to FinP (but see note 2), giving: (41) [PP in the room] Fin+Loc Subj t’DP T+Phi [ tDP v [ sleep tPP fitfully ]] – [DP the students of the class who had heard…] With Chomsky (2001), we assume that the traces of the heavy NP-shifted nominal, namely t’DP and tDP in (41) are not “visible” and don’t count in the computation of locality on the locative PP chain (in the room, tPP). Alternatively, and perhaps more plausibly, we can consider only “whole chains”, and not just positions as relevant for the calculation of locality. The possibility, indeed, the requirement that chains cross – with apparent multiple violations of locality – is also suggested in Chomsky (2001) and extended to the A’-system by Krapova and Cinque (2004). So, no intervention effect is determined on the chain (in the room, tPP) by intervening traces tDP and t’DP because only part of the chain of the shifted phrase intervenes, but not the whole chain, the heavy subject ending up, by our current assumptions, in a position higher than Spec,Fin. It should be noticed that this analysis crucially requires a representational view of locality, with Relativized Minimality evaluated after movement (at the interface or at the end of each phase). In a strictly derivational view, Relativized Minimality would be violated in step (40), because the subject would intervene when the locative is moved to Fin. The aspect of this analysis which we would like to stress is that a subject (at least in the sense of an element which enters into an AGREE relation with T+Phi) can undergo Heavy NP Shift when it is allowed not to move to Spec,Subj. This happens when the Subject Criterion can be satisfied by the nominal Fin, which is in turn valued by the locative PP under Locative Inversion. If the subject itself has to move to Spec,Subj to satisfy the Subject Criterion, as in (28), it could not subsequently undergo Heavy NP Shift because of Criterial Freezing. Locative Inversion and Heavy NP Shift thus interestingly interact in the derivation of (31) and (32): Locative inversion

English Locative Inversion and Heavy NP Shift 355

indirectly satisfies the Subject Criterion (through the intermediary device of nominal Fin), thus freeing the subject DP (the DP entering into an AGREE relation with the inflected verb) from Criterial Freezing and making it available to undergoing Heavy NP Shift. Conversely, Heavy NP Shift of the subject DP is instrumental in rendering the locative phrase accessible to attraction from the nominal Fin without violating locality. No circularity or chicken-egg problem arises if a representational view of locality is adopted. 7. Locative inversion and raising There is another aspect of C&L’s analysis which is immediately relevant to us. They observe that the locative PP is unable to undergo Subject to Subject Raising in the light subject construction. Compare the simple and the raising case: (42) a. Into the room walked Robin slowly b. *Into the room appeared to be walking Robin slowly C&L’s suggestion that this may be a semantic effect (cf. their fn. 4) remains unclear to us. The set of assumptions we are working with offers a structural explanation for this contrast. Under the Subject Criterion approach, Raising must involve a defective clausal complement, truncated below Subj. Otherwise, subject raising would never be allowed because of Criterial Freezing. Consequently, the clause embedded under a raising predicate is just the following reduced structure, with some sort of defective Phi associated to T (Chomsky 2001, 2004): (43) T+Phi be [Robin walking into the room slowly] Locative inversion cannot take place in (43), as it critically requires the nominal Fin head and Fin is not expressed in the structurally defective Raising infinitive, which lacks the C-system altogether. Following the establishment of the AGREE relation between T+Phi and Robin, the latter moves to Spec,T+Phi, yielding (44). (See our discussion below on why movement of the DP is obligatory here, while it appears to be optional in (34)). (44) Robin T+Phi be [ t walking into the room slowly] At this point, the raising verb is merged and the main clause VP is constituted. Then, the main Subj- T+Phi complex is merged. From (45), the only

356

Luigi Rizzi and Ur Shlonsky

possible candidate for movement into the main Spec,Subj is Robin. Direct attraction of the locative is blocked by Robin in the Spec of the embedded T+Phi. (42b) is hence underivable and the only possible output is the uninverted order (46): (45) Subj T+Phi appear [ Robin T+Phi be [ t walking into the room slowly]]  (46) Robin Subj T+Phi appear [t’ T+Phi be [ t walking into the room slowly]] C&L also observe that the equivalent of (42b) with a heavy subject DP moved to the right edge is grammatical: (47) Into the room appeared to be walking a very large caterpillar In our terms, (47) is derivable as follows: a very large caterpillar enters into an AGREE relation with T+Phi in the embedded clause and then it is attracted to its Spec. Successive merger of the main clause material yields the following representation: (48) Subj T+Phi appear [ a very large caterpillar T+Phi be [ t walking into the room]] At this point, a very large caterpillar enters into a second AGREE relation with the main clause T+Phi, and possibly moves to its Spec. The crucial point is that, being sufficiently heavy, it can undergo Heavy NP Shift and end up in the right periphery of the main clause: (49) Subj t’’ T+Phi appear [ t’ T+Phi be [ t walking into the room]] – a very large caterpillar Now, nothing prevents the direct merger of Fin+Loc, satisfying the subject criterion in the main clause: (50) Fin+Loc Subj t’’ T+Phi appear [ t’ T+Phi be [ t walking into the room]] – a very large caterpillar Finally, Fin+Loc can directly attract the locative phrase (neither t’, t’’ nor the heavy shifted DP count as interveners under the adopted approach, see the discussion around example (41)), yielding (47) (and incorporating further movement of the locative to Top), as desired. We are now left with the following question: The ill-formedness of (42b) requires that T+Phi in the embedded clause attract the nominal it

English Locative Inversion and Heavy NP Shift 357

agrees with, yielding (45). If the DP could remain in situ in the embedded VP, main clause Subj should be able to directly attract the locative under equidistance, and (42b) would be incorrectly derived. The obligatoriness of attraction by the embedded T+Phi is in apparent contradiction with the fact that main T+Phi can choose to attract the DP it agrees with in the simple case (29), (structure (35)): if attraction of DP were obligatory here, one could not derive the inverted order at all, since the DP Robin, moved to Spec,T+Phi, would block attraction of the locative to Spec,Fin, a necessary step for the licensing of the inversion structure under our analysis. One obvious difference between (35) and (44) is that the former has a SubjP layer which the latter crucially lacks (the hallmark of Raising, to recall, is a truncated clausal complement, as in (37)). One could, then, attempt to explain the difference between optional DP movement to Spec, T+Phi in (35) and its obligatoriness in (44), by relating it to this independent difference. Yet it is not clear that this structural difference harbours the key to this mystery. In both (35) and (44), AGREE is established with the DP in situ after the merger of T+Phi. In (35), Subj is then merged, T+Phi head-moves to Subj, Fin+Loc is merged to satisfy the Subject Criterion and finally, the locative PP is moved to Spec, Fin+Loc. Note, now, that there is no obvious reason why, after the establishment of AGREE in (44), the DP could not also remain in situ and wait until the main clause Subj layer is merged. Then, (42b) would be incorrectly derived. Put differently, nothing in the system introduced thus far can force movement of a nominal to Spec,T. We would like to propose that the embedded subject must move to Spec,T in (44) because if it didn’t, the main clause heads T+Phi and Fin+Loc would be unable to attract either the subject or the locative from the embedded clause. Our suggestion is that the embedded T, endowed with a (defective) Phi specification would constitute an intervener for attraction by a higher head. The structure would then be ill-formed because there would be no way to satisfy the Subject Criterion in the main clause. On the other hand, if the embedded subject is attracted to the embedded T+Phi, as in (44), the Subject Criterion in the main clause can be satisfied in one of two ways: Either by continuing to move the embedded subject, which ultimately yields the uninverted structure (46), or through heavy NP shift of the subject followed by attraction of the locative to the main Fin+Loc, yielding the inverted structure (47). Crucially, movement of the DP to the Spec of the embedded T+Phi checks the (defective) Phi-features on T, thus eliminating it as a potential intervener.

358

Luigi Rizzi and Ur Shlonsky

This analysis still assumes an asymmetry between the defective Phi specification of a raising structure, which gives rise to a Relativized Minimality effect and the full Phi specification of T in a complete clausal structure, which does not prevent an AGREE relation from being established between Fin+Loc and a predicate-internal locative. If it did, no case of simple locative inversion with a light subject would ever be possible. This is somewhat counterintuitive: Why should a defective Phi specification in a raising construction count in the calculus of locality from a main clause Fin+Loc, while the full Phi specification of the main clause would not? We believe a possible answer to this question may be based on Chomsky’s (2006) idea that the complete Phi specification on T in a full clause is a mere duplication of a specification in the C system: T displays a “second occurrence”, as it were of the Phi specification occurring in the C-system. So, if the locality restriction only arises from the intervention of distinct elements, not from different occurrences of the same element, we do not expect Phi in T to block Phi attraction from an element in the adjacent Csystem. On the other hand, the defective Phi specification of a raising infinitive is not inherited from a C system, which is radically absent from that kind of structure. Therefore, it clearly involves distinct features from the Csystem of the main clause, and as such it gives rise to Relativized Minimality effects, along the lines we have assumed. 8. Conclusion Inverted locatives in English possess a number of subject-like properties: they suffice to satisfy EPP requirements, give rise to that-trace effects, alleviate Weak-Crossover effects, and seem to be able to undergo subject raising; on the other hand, preposed locatives have other properties which show that they cannot occur in subject position, and must reach one of the scope-discourse related positions in the left periphery: they are incompatible with different contexts not licensing left-peripheral positions (sentential subjects, ECM environments, etc.), and with T to C movement. We have tried to show that these apparently mixed subject and non-subject properties can be reconciled under an analysis which capitalizes on a device for subject extraction independently argued for in Rizzi and Shlonsky (2006). If, in the relevant structures, the Fin head can satisfy the Subject Criterion by being endowed with appropriate nominal features, locative features in this case, it will not be necessary to move the thematic subject to the criterial position; preposing of the locative will be required to check the locative

English Locative Inversion and Heavy NP Shift 359

features in Fin, and further movement of the preposed locative to a criterial position will be required by Movement as Last Resort guidelines. Heavy NP shift of the thematic subject, normally blocked by Criterial Freezing as other kinds of subject movement, becomes possible if the Subject Criterion is indirectly fulfilled by the preposed locative (via the locative Fin), a configuration which liberates the thematic subject and makes it available for undergoing Heavy NP Shift.

Notes 1. One possibility is that the post verbal subject in Locative Inversion constructions is focalized, as Culicover and Rochemont (1990) have argued and that new information focus is structurally encoded, see Belletti (2004). Extraction of what in (7) would then violate criterial freezing: The focalized post verbal subject would be frozen in the VP-area focus position and could not undergo further movement. 2. If Heavy NP Shift is reinterpreted à la Kayne (1994), as involving leftward movement of the heavy constituent followed by further leftward movement of the remnant, the same conclusion should follow: the first step of moving leftward an element satisfying the Subject Criterion should be precluded by Criterial Freezing as before, and the further remnant movement could do nothing to salvage the structure. We will not discuss the exact mechanics of Heavy NP Shift here, and the auxiliary hypotheses that are needed (e.g., why no device normally permitting wh extraction of the subject is available if the subject undergoes Heavy NP Shift). 3. Alternatively, accessibility of the subject to an external attractor may result from a “smuggling” operation, in the sense of Collins (2005), scrambling some sort of small VP over the locative, as in Rizzi’s (2003) proposal for certain quirky subject constructions. 4. C&L argue that the locative PP occupies distinct landing sites in the light and heavy subject constructions (29) and (31), but they rely on subtle weak crossover evidence which is not uncontroversial (see their fn. 8). They also argue that the preposed PP cannot undergo further movement in the light inversion construction (their point (vi), p. 301), while it can be extracted from an embedded clause in the heavy inversion construction. The data don’t seem to straightforwardly point in this direction. For instance, their example (iv)a, fn. 20 Out of which room did you claim came who?, appears to allow wh-extraction, and still plausibly involves inversion with a light subject. In our discussion we adopt C&L’s finding that two separate inversion constructions should be distinguished, but will continue to assume the same landing site for the locative PP.

360

Luigi Rizzi and Ur Shlonsky

References Belletti, Adriana 2004 Aspects of the low IP area. In The Structure of IP and CP, Luigi Rizzi (ed.), 16–51. Oxford: Oxford University Press. Bresnan, Joan 1977 Variables in the theory of transformations. In Formal Syntax, Peter W. Culicover, Thomas Wasow and Adrian Akmajian (eds.), 157–196. New York: Academic Press. 1994 Locative inversion and the architecture of universal grammar. Language 70 (1): 1–131. Chomsky, Noam 1995 The Minimalist Program. Cambridge, MA: MIT Press. 2001 Derivation by phase. In Ken Hale: A Life in Language, Michael Kenstowicz (ed.), 1–50. Cambridge, MA: MIT Press. 2004 Beyond explanatory adequacy. In Structures and Beyond: The Cartography of Syntactic Structure, Adriana Belletti (ed.), 104–131. Oxford: Oxford University Press. 2006 On phases. To appear in Interfaces + Recursion = Language? Chomsky’s Minimalism and the View from Syntax-Semantics, Uli Sauerland and Hans M. Gärtner (eds.), Berlin /New York: Mouton de Gruyter. Cinque, Guglielmo 1999 Adverbs and Functional Heads: A Cross-Linguistic Perspective. Oxford: Oxford University Press. Collins, Chris 2005 A smuggling approach to raising in English. Linguistic Inquiry 36 (2): 289–298. Culicover, Peter W. and Robert D. Levine 2001 Stylistic inversion in English: A reconsideration. Natural Language and Linguistic Theory 19 (2): 283–310. Culicover, Peter W. and Michael S. Rochemont 1990 English Focus Constructions and the Theory of Grammar. Cambridge: Cambridge University Press. Hale, Kenneth and Samuel J. Keyser 1993 On argument structure and the lexical expression of syntactic relations. In The View From Building 20, Kenneth Hale and Samuel J. Keyser (eds.), (2): 53–109. Cambridge, MA: MIT Press. Kayne, Richard S. 1994 The Antisymmetry of Syntax. Cambridge, MA: MIT Press. 2006 A short note on where vs. place. Ms., New York University.

English Locative Inversion and Heavy NP Shift

361

Krapova, Ilyana and Guglielmo Cinque 2004 On the order of wh-phrases in Bulgarian multiple wh-fronting. Ms., University of Venice. Landau, Idan 2005 The locative syntax of experiencers. Ms., Beer Sheva: Ben Gurion University. Levin, Beth and Malka Rappaport Hovav 1995 Unaccusativity: At the Syntax-Semantics Interface. Cambridge, MA: MIT Press. Levine, Robert D. 1989 On focus inversion: Syntactic valence and the role of a SUBCAT list. Linguistics 17: 1013–1055. Rizzi, Luigi 1982 Issues in Italian Syntax. Dordrecht: Foris. 1990 Relativized Minimality. Cambridge, MA: MIT Press. 1997 The fine structure of the left periphery. In Elements of Grammar. Handbook in Generative Syntax, Liliane Haegeman (ed.), 281–337. Dordrecht: Kluwer. 2003 On the form of chains: Criterial positions and ECP effects. To appear in On Wh Movement, Lisa L. Cheng and Norbert Corver (eds.). Cambridge, MA: MIT Press. 2004 Locality and left periphery. In Structures and Beyond, Adriana Belletti (ed.), 223–252. Oxford: Oxford University Press. Rizzi, Luigi and Ur Shlonsky 2006 Strategies of subject extraction. To appear in Interfaces + Recursion = Language? Chomsky’s Minimalism and the View From SyntaxSemantics, Hans M. Gärtner and Uli Sauerland (eds.). Berlin /New York: Mouton de Gruyter. Sobin, Nicholas 2002 The Comp-trace effect, the adverb effect and minimal CP. Journal of Linguistics 38: 527–560. Stowell, Tim 1981 Origins of phrase structure. Ph.D. diss., Cambridge, MA: MIT. Taraldsen, Knut T. 2001 Subject extraction, the distribution of expletives and stylistic inversion. In Subject Inversion in Romance and the Theory of Universal Grammar, Aafke Hulk and Jean-Yves Pollock (eds.), 163–182. Oxford: Oxford University Press.

Informational focus in Sicilian and the left periphery Silvio Cruschina

In Sicilian the informational focus of the sentence (IF) can be realised in two different positions: in situ or in the left periphery of the sentence. This paper examines the different interpretations associated with these two positions and shows that IF undergoes movement to the left periphery only when related to specific pragmatic and illocutionary features, i.e. relevance and unexpectedness of the new information. This analysis bears out the cartographic and minimalist claim that movement to the left periphery/edge of CP must be motivated by interpretive or discourse-related effects. Moreover, this paper investigates the syntactic and distributional properties of IF in comparison with contrastive focus (CF). It will be shown that although IF and CF share many syntactic properties (i.e. focus properties), they differ in their position with respect to the verb. A further difference is that IF appears to be present only in matrix C-domain. We shall account for these differences by suggesting that they involve two distinct projections within the left periphery.

1.

Introduction

1.1. Data and structure of the paper This paper attempts to account for a particular word order commonly found in Sicilian and the regional variety of Italian spoken in Sicily.1 The word order in question differs from that found in Italian and other Romance languages resulting in a kind of OV order where the verb is preceded by the direct object (or XP constituent) and hence frequently found in clause-final position. Despite being one of the most salient features of Sicilian, this phenomenon has hitherto been neglected in the literature. We take the informational structure of the sentence as the starting point of this investigation, distinguishing between those constituents expressing new information and those expressing given information. We shall claim that the particular word order under consideration involves a case of sentential

364 Silvio Cruschina IF, inasmuch as informationally new constituents are focalised in Sicilian by moving them into the left periphery of the sentence. This idea is not entirely new; Benincà and Poletto (2004) in their detailed investigation of the left periphery quote the following examples from Sicilian: (1)

a. Un libro comprasti? a book buy. PAST.2SG ‘Did you buy a book?’ b. Antonio sono. Antonio be.PRES.1SG ‘It’s Antonio.’

In light of such examples, they argue that in Sicilian, as well as in some medieval varieties of Italo-Romance, an IF position can be freely activated within the left periphery, whereas in Italian the same position is parasitic on a CF projection first being activated. The paper is organised as follows. We begin with a description of the informational structure of the Sicilian sentence, first briefly analysing the properties of topicalisation (section 2) and then how new information is introduced in the clause (section 3). On the basis of data highlighting the informational structure of the sentence, it will be shown in section 4 that IF in Sicilian shares the same syntactic properties identified for CF by Rizzi (1997). Nonetheless, the differences between IF and CF, discussed in section 5, will lead us to the conclusion that the two types of focus occupy distinct positions in the left periphery.

1.2. Focus, the left periphery and the CP phase For this work we adopt a cartographic framework (Cinque 1999, 2002; Rizzi 2004; Belletti 2004b). In particular we assume a split-CP system, whereby the CP domain is articulated into a series of functional heads and projections, as originally proposed by Rizzi (1997): (2)

ForceP

TopP*

FocP

TopP*

FinP

While ForceP expresses the illocutionary force of the sentence (e.g. declarative, interrogative, exclamative), FinP encodes those specifications related to the finiteness and the mood of the sentence. As we shall see below, TopP

Informational focus in Sicilian and the left periphery

365

and FocP are the projections hosting topicalised and focalised constituents respectively in their specifier position. Although superficially somewhat distinct, some points of convergence between the Cartographic Approach and the Minimalist Program (MP) have recently been pointed out by Chomsky (cf. Chomsky 2001, 2002, 2004, 2005). In particular, the role of interface properties, which according to the Cartographic Approach are encoded in an articulated system of functional projections, has also become central in the MP: EPP-features may be assigned to a functional head to ensure scopal or discourse-related properties relevant for specific interpretation. Phases, namely CP and vP, are those phrases headed by functional heads associated with the kind of features that trigger movement (internal merge) and give rise to so-called edge-effects. The extra edge position required by internal merge to a phase head bearing an EPP-feature is optional. Indeed, Chomsky distinguishes between obligatory EPP-features, which are inherently associated with a given functional head, and optional EPP-features, which trigger obligatory movement (internal merge) in order to check and eliminate (value) them, and which must be associated with an interpretive effect on outcome (Chomsky 2001: 34). This is perfectly compatible with Rizzi’s (1997, 2001) assumption that a projection within the left periphery is present only when activated, and Chomsky himself acknowledges that the phase head C “is a shorthand of the region that Rizzi (1997) calls “left periphery”, possibly involving feature spread from fewer functional heads (maybe only one)” (Chomsky 2005: 9). EPP-features therefore represent a locus of parametric variation, and whether the EPP-feature, or the [+focus] feature in non-strict minimalist terminology, responsible for movement of the focus constituent is optional or not is indeed subject to cross-linguistic variation.

2. Informational structure of the sentence: old information In terms of marking old information, Sicilian differs from Italian with respect to at least two closely interrelated properties: 1) constituents conveying given information must be topicalised (in Spec/ TopP);2 2) a resumptive pronoun is obligatory for all dislocated arguments.

366 Silvio Cruschina It is well known that in Italian resumptive clitics of left-dislocated elements are obligatory only if these elements are direct objects and partitives (Benincà 1988; Cinque 1990), while they prove optional in other cases.3 If we look at examples (3)–(6),4 we can see that unlike in Italian where the resumptive clitic is optional with PPs (a-sentences), in Sicilian the absence of the resumptive clitic in the same contexts yields ungrammatical results (b-sentences):5 (3)

a. Maria, a Salvo, (gli)

regalerà

un libro.

b. Maria, a Salvo, *(ci) av’à regalari un libbru. Maria to Salvo to-him.CL give.FUT.3SG a book ‘Maria will give a book to Salvo.’ (4)

a. Che cosa (le)

ha dato,

Salvo, alla

sua ragazza?

b. Chi *(ci) detti, Salvo, a sa zita? what to-her.CL give. PAST.3SG Salvo to-the his girlfriend ‘What did Salvo give his girlfriend?’ (5)

a. Al mare, Maria e Salvo, quest’ anno, (ci) sono andati poche volte. b. A mari, Maria e Salvo, st’ annu,*(ci) hannu jutu picca voti. to-the seaside Maria and Salvo this year there.CL AUX go.PP few times ‘Maria and Salvo (only) went to the seaside a few times this year.’

(6)

a. Di quella partita, Salvo ed io (ne) abbiamo parlato per tutta la giornata. b. Di dda partita, Salvo e ìa *(nni) parlammu pì tutta a jurnata. of that match Salvo and I of-it.CL talk. PAST.1PL for all the day ‘Salvo and I talked about that match all day.’

Clitic resumption thus proves obligatory with left-dislocation (cf. (3), (5), (6)), as well as with right-dislocation (cf. (4)), independently of the grammatical category of the dislocated constituent.

Informational focus in Sicilian and the left periphery

367

In Sicilian we do not find emarginazione ‘marginalisation’ with VSO and VOS orders, which are possible in Italian under specific prosodic and intonational conditions (cf. Antinucci and Cinque 1977; Calabrese 1982, 1992; Cardinaletti 2001; Belletti 2004a), as shown below:6 (7) *Liggì Maria/MARIA, u giornali. read. PAST.3SG Maria the newspaper ‘Maria read the newspaper.’ Clitic-resumed topicalisation, therefore, is the only syntactic construction possible for constituents conveying old information. 3.

Informational structure of the sentence: new information

While a topic constituent expresses old information, the focus of a sentence conveys new information with respect to a given state/event or, in the case of contrastive focus, it negates or corrects a former presupposition. In Italian the new element in the sentence, i.e. IF, tends to appear in post-verbal position (cf. Antinucci and Cinque 1977; Calabrese 1982, 1992; Frascarelli 2000; Belletti 2001, 2004a). New constituents may only appear in the left periphery if they bear a contrastive interpretation (cf. Benincà 1988; Rizzi 1997, a.o.). In Sicilian it is also possible for them to appear in the left periphery when combined with specific interpretive features, such as relevance and unexpectedness. These two features are in general associated with answer-to-question and exclamative contexts respectively.

3.1. New and relevant information In the following examples, we shall make use of question-answer pairs to establish discourse roles. Traditionally, in fact, IF identifies the constituent which answers a wh-question. However, we will use the term IF also to refer to Sicilian (emphatic) IF in the left periphery of the sentence. In accordance with common practice, we will indicate IF in bold and CF in capitals.7 We begin our investigation by considering the licensing of OV (or more generally, XP-V) orders. The distinction between narrow and wide focus proves fundamental in this respect. Specifically, the OV order is only possible under narrow focus, in which case new information is conveyed by a single constituent alone (cf. 8), whereas under wide Focus (in which the

368 Silvio Cruschina whole sentence expresses new information), only the unmarked VO order is possible (cf. (9)):8 (8)

a. Chi scrivisti airi? what write. PAST.2SG yesterday ‘What did you write yesterday?’ b. N’ articulu scrissi. an article write. PAST.1SG ‘I wrote an article.’

(9)

a. Chi successi? what happen. PAST.3SG ‘What happened?’ b. Salvo arristà un latru. Salvo arrest. PAST.3SG a thief ‘Salvo arrested a thief.’

The VO order in the same context of (8b) is also acceptable in Sicilian, but with the difference that it does not imply any specific intonational emphasis on the DP object.9 The emphasis, therefore, is not associated with IF per se, but with the relevance of the new information constituent and with the pragmatic intention of the speaker to foreground it. There are no restrictions on the phrasal categories that can be moved to the left periphery as IF: the focus position in the left periphery, in fact, can also host indirect objects (cf. (10)), prepositional phrases such as locatives (cf. (11)), and DPs and APs functioning as predicative complements (cf. (12), (13)): (10) a. A cu a scrivisti a littira? to whom it.CL write. PAST.2SG the letter ‘To whom did you write the letter?’ b. A Maria a scrissi (a littira). to Maria it.CL write. PAST.1SG (the letter) ‘I wrote it to Maria.’ (11) a. Unni u mittisti u libbru? where it.CL put. PAST.2SG the book ‘Where did you put the book?’

Informational focus in Sicilian and the left periphery

369

b. ’Ncapu u tavulu u misi (u libbru). Above the table it.CL put. PAST.1SG (the book) ‘I put it on the table.’ (12) a. Prontu, cu parla? / Cu jè? hello who speak.PRES.3SG / who be.PRES.3SG ‘Hello, who’s speaking?’ / ‘Who’s that?’ b. Salvo sugnu. Salvo be.PRES.1SG ‘It’s Salvo.’ (13) a. Cumu sapiva a carni? how taste. PAST.3SG the meat ‘How did the meat taste?’ b. Bona sapiva. good taste. PAST.3SG ‘It tasted good.’ As for subject fronting to Spec/FocP, this operation is only evident in those cases where a post-verbal subject would be expected in Italian, namely, with unaccusative verbs or generally when the subject expresses new information (Belletti 2001): (14) a. Cu partì? who leave. PAST.3SG ‘Who left?’ b. Salvo partì. Salvo leave. PAST.3SG ‘Salvo left.’ An embedded clause can also convey new information and hence be focalised. In (15) the answer to the question is the indirect interrogative clause headed by si ‘if’ in (15b), which constitutes the object of the verb spiari ‘to ask’. Given that the latter expresses new information, it can be fronted: (15) a. Chi ci spiasti a Salvo? what to-him.CL ask. PAST.2SG to Salvo ‘What did you ask to Mario?’

370 Silvio Cruschina b. Si voli viniri a mari cu nuantri ci If want.PRES.3SG come.INF to seaside with us to-him.CL spiavu. ask. PAST.1SG ‘I asked Mario if he wants to come to the seaside with us.’ Along the same lines, if we question a constituent contained within an embedded clause, this can be extracted and moved into the left periphery of the main clause, as in (16): (16) a. Chi dicisti ca s’ accattà Maria? what say. PAST.2SG that REFL.CL buy. PAST.3SG Maria ‘What did you say that Maria bought? b. Na machina dissi ca s’ accattà. a car say. PAST.1SG that REFL.CL buy. PAST.3SG ‘I said that she bought a car.’ As verified by (16b), in these cases the constituent expressing new (relevant) information appears in the left periphery of the matrix clause. In fact, focalisation within the left periphery of the embedded clause is not possible in such contexts (17a), unless the constituent in question bears a contrastive interpretation (17b): (17) a. *Dissi ca na machina s’ accattà. say. PAST.1SG that a car REFL.CL buy. PAST.3SG ‘I said that she bought a car.’ b. Dissi ca NA MACHINA s’accattà, no un muturinu. A restriction therefore appears to operate which rules out IF, but not CF, in the left periphery of embedded clauses. This matrix-embedded asymmetry supports the recent claims of a structural difference between root and nonroot clauses pointing to a structure reduction of the embedded C-domains (cf. Bayer 2001; Haegeman 2002, 2004). In light of this, we postulate that IFocP is only present in matrix clauses.10

Informational focus in Sicilian and the left periphery

371

3.2. New and surprising/unexpected information The new constituent of the sentence can also express surprising and/or unexpected information. The illocutionary force of the sentences containing IF of this type is that of an exclamative sentence:11 this is another feature associated with movement of IF to the left periphery of the sentence. (18) Na casa s’ accattà! a house REFL.CL buy. PAST.3SG ‘He bought a house!’ (19) U suli niscì! the sun go-out. PAST.3SG ‘The sun came out!’ Evidently, when IF in answers to questions expresses new and unexpected information, the utterance may possess an exclamative illocutionary force. However, this is not always the case: in simple requests for identification, for example, the answer does not imply surprising/unexpected information (cf. (12)). In these contexts, we consider the relevance of IF the feature associated with focalisation into the left periphery. In exclamative contexts, fronting of adjectives (functioning as adverbs) proves very common in Sicilian, as in the following examples: (20) a. Bunu jè! good be.PRES.3SG ‘That’s good!’ b. Bunu facisti! good do. PAST.2SG ‘You did well!’ c. Veru ti dicu! true to-you.CL say.PRES.1SG ‘I really mean it!’ Focalisation is also possible with yes/no questions. Whereas in Italian interrogatives of this kind do not exhibit any overt syntactic marking, yes/no questions in Sicilian can optionally be introduced by the element chi (regional Italian che), as shown by the following examples:

372 Silvio Cruschina (21) a. Chi ci a addumannasti dda cosa a Mario? INT to-him.CL it.CL ask. PAST.2SG that thing to Mario ‘Did you ask Mario that thing?’ b. Chi ci isti nn’u dutturi? INT there.CL go. PAST.2SG in the doctor ‘Did you go to the doctor’s?’ As can be seen in (22), the left periphery of yes/no questions can host IF, irrespective of whether the interrogative marker chi is present or not: 12 (22) a. Chi viglianti sì? INT awake be.PRES.2SG ‘Are you awake?’ b. Vossia chi dutturi jè? You.POLITE INT doctor be.PRES.3SG ‘Are you a doctor?’ c. Chi a Maria salutasti? INT to Maria greet. PAST.2SG ‘Did you say hello to Maria?’ The emphasis related to IF in the left periphery corresponds to a difference in interpretation:13 without focalisation (as in (21)) the yes/no question is read as a canonical, genuine request for new information, whereas with focalised preverbal IF (as in (22)) the question represents a request for confirmation (of unexpected information) or a rhetorical question (expressing surprise). Evidently, in this type of yes/no question the two illocutionary forces, i.e. exclamative and interrogative, combine. As we have seen, IF mainly occurs in the left periphery in answers to questions and in sentences with an exclamatory force. It must also be noted that emphatic IF proves quite common with predicative categories, such as APs dependent on a be auxiliary, and with the complements of lexical have (expressing possession) and lexical be (with locative meaning) (cf. Kayne 1994 for a connection between these structures). The following examples are from Rohlfs (1969: 323): (23) a. Iddu picciliddu è. he child be.PRES.3SG ‘He is a child.’

Informational focus in Sicilian and the left periphery

373

b. A frevi aju. the fever have.PRES.1SG ‘I have a temperature.’ This explains the observation that in Sicilian essiri ‘to be’ and aviri ‘to have’ forms appear at the end of the sentence with surprising regularity (Rohlfs 1969).

4.

Syntactic properties of IF shared with CF

On the basis of the observed data, it is a natural progression to ask whether Sicilian preverbal IF exhibits similar or different syntactic properties with respect to CF. In our discussion below, we shall demonstrate that IF does indeed share the same syntactic properties that Rizzi (1997) identifies for CF (i.e. absence of resumptive clitic, Weak-Crossover effects, focalisation of bare quantifiers, uniqueness, and incompatibility with wh-elements) supporting Frascarelli’s (2000) claim that the focus constituent displays the same structural properties, namely operator properties, irrespective of the “type of the newness” that it expresses. Note that all the examples of IF in this section could be appropriate both in exclamatives or as answers to wh-questions.

4.1. Resumptive clitics Regarding its occurrence with resumptive clitics, we acknowledge that IF patterns in an identical fashion to CF: (24) a. UN LIBBRU ci (*u) detti. b. Un libbru ci (*u) detti. a book to-him.CL it.CL give. PAST.1SG ‘I gave him a book.’ As we can observe in (24b), IF proves incompatible with resumptive clitics, as well as CF (24a).

374 Silvio Cruschina 4.2. Weak Crossover As shown by Rizzi (1997), the FocP projection targeted by CF exhibits WCO effects. This sensitivity constitutes the main proof of the quantificational nature of CF and further confirms its status as an operator binding a syntactic variable.14 The absence of a resumptive clitic in the case of IF considered above also highlights that it is an Operator-Variable construction, a conclusion further confirmed by the following examples: (25) a. ??/*A Salvoi vasà sai suru. to Salvo kiss. PAST.3SG his/her sister ‘His/her sister kissed Salvo.’ sai matri. b. ??/*A Marioi vitti to Mario see. PAST.3SG his/her mother ‘His/her mother saw Mario.’ These sentences, in fact, exemplify WCO effects in conjunction with IF.

4.3. Bare quantificational elements While bare quantifiers cannot be topicalised, they do undergo focalisation from their argumental positions, witness the following examples with preverbal IF: (26) a. Nenti ci fa. nothing there.CL do.PRES.3SG ‘It doesn’t matter.’ b. Nenti jè chissu. nothing be. PAST.3SG that ‘That’s nothing.’ c. Nenti (*u) fici. nothing it.CL do. PAST.1SG ‘I did nothing.’ (27) a. Nuddu c’ è. nobody there.CL be.PRES.3SG ‘There is nobody there.’

Informational focus in Sicilian and the left periphery

375

b. A nuddu vitti. to nobody see. PAST.1SG ‘I didn’t see anybody.’ c. Tutti cosi si mangià. all things REFL.CL eat. PAST.3SG ‘He/she ate everything.’ With quantifiers, emphatic IF proves very common, and in some cases, like (26a,b) post-verbal IF would sound rather odd. In (26c) we see that clitic resumption of the preverbal IF would yield ungrammatical results.

4.4. Uniqueness Only one IF (either preverbal or post-verbal) at most can be licensed in any given sentence. A multiple wh-question represents a context in which two IFs could potentially co-occur. Even if a multiple wh-question is assumed to be well-formed (although in practice it is not readily accepted in either Sicilian or Italian), an answer with two fronted IFs proves ungrammatical: (28) a. ??Cu vasà a cu? who kiss. PAST.3SG to whom ‘Who kissed who?’ b. *Salvo a Lucia vasà. Salvo to Lucia kiss. PAST.3SG ‘Salvo kissed Lucia.’ b’. *A Lucia Salvo vasà. If one of the two questioned constituents is stranded in situ, only the preverbal one is associated with emphasis and hence interpreted as IF. Indeed, the informational articulation of the sentence is subject to the restriction that there can only be one informationally new constituent in each sentence (see a.o. Calabrese 1982, 1992; Rizzi 1997; Belletti 2001).15

4.5. Incompatibility with Wh On a par with CF, IF is also not compatible with wh-operators. The incompatibility of CF with wh-operators is subject to an asymmetry between root

376 Silvio Cruschina and embedded interrogatives. Contrastively focalised constituents and whoperators are not compatible in root interrogatives and this, in turn, is considered evidence for assuming that these elements occupy the same position, namely Spec/FocP (Rizzi 1997). In embedded interrogatives, however, a wh-element can appear in a different and lower position, which Rizzi (2001) calls Wh. In Sicilian, by contrast, IF cannot co-occur with a wh-element either in root or embedded interrogatives. This difference is less surprising if we consider that Sicilian IF is not licensed in the left periphery of an embedded clause (cf. the discussion of examples (16), (17)).

5.

CF and IF in the left periphery

5.1. The distinction Kiss (1998) assumes a clear-cut distinction between CF and IF. In her analysis, based principally on Hungarian data, the two types of focus differ in many respects: syntactically, semantically and prosodically. With regard to syntactic differences, she claims that only identificational focus, which according to her corresponds to Italian CF,16 has a quantificational nature (insofar as it displays operator properties) and involves a movement to the Spec of a functional projection within the left periphery. This is not the case for Sicilian. Indeed we saw that, when combined with specific pragmatic/discourse-related features, IF also involves movement to the left periphery of the sentence and exhibits operator properties. However, in §3 we discussed an important difference between the two types of focus, i.e. the matrix-embedded asymmetry. A further significant difference concerns the adjacency of the focus constituent with respect to the verb. It has been observed (Calabrese 1992; Frascarelli 2000) that Italian (post-verbal) IF must be adjacent to the verb. This requirement does not hold for CF when it occurs post-verbally (cf. Calabrese 1982) nor when it occurs in the C-domain where it can be followed by a Topic (cf. (2)). In Sicilian emphatic IF in the left periphery (as well as IF in post-verbal position) needs to be adjacent to the verb. A focus constituent followed by a Topic in the left periphery is grammatical, but it is best read as contrastive: (29) a. Chi ci scrivisti a Maria? what to-her.CL write. PAST.2SG to Maria ‘What did you write to Maria?’

Informational focus in Sicilian and the left periphery

377

b. A Maria na littira ci scrissi. to Maria a letter to-her.CL write. PAST.1SG ‘I wrote a letter to Maria.’ b’.

??

Na littira

a Maria ci scrissi.

b”. #NA LITTIRA a Maria ci scrissi. In (29) we can see that while a Topic can precede IF (29b), the sentence in (29b’) with a post-focal Topic is marginal; (29b”) shows that the same sentence with CF followed by a Topic is perfectly grammatical, but inappropriate (as indicated by the symbol #) as an answer to the question in (29a). This adjacency requirement also holds when IF expresses surprising/ unexpected information: (30) a. ??A Salvo ta suru vasà! to Salvo your sister kiss. PAST.3SG ‘You sister kissed Salvo!’ b. A SALVO ta suru vasà. Several studies also argue for prosodic and intonational differences between CF and IF (cf. Zubizarreta 1998; Nespor and Guasti 2002; Donati and Nespor 2003). Sicilian, however, presents the case of a focus category which can be defined as ‘informational’ from an interpretive point of view, but which exhibits intonational and prosodic properties different from those found in CF and “traditional” (non-emphatic) IF. These particular prosodic and intonational features, which signal leftward focalisation of informationally new constituents, are dependent on a particular pragmatic facet, namely, a major saliency or relevance of focalised constituents with respect to the discourse.

5.2. Two distinct positions in the left periphery We have seen so far that in Sicilian under specific conditions IF can occupy a position within the left periphery of the sentence and that it exhibits the same syntactic properties as CF (cf. §4). Thus, two possible hypotheses present themselves: (i) (ii)

IF in the left periphery occupies the same position as CF; CF and IF occupy two distinct positions within the left periphery.

378 Silvio Cruschina We note that CF and IF are not compatible since they cannot co-occur. 17 This incompatibility naturally follows from the observation that CF is not acceptable in the same contexts in which IF is licensed (cf. note 8). However, their incompatibility in itself is not valid and independent proof that a single FocP projection is involved, in that other factors might be involved such as a semantico-pragmatic constraint which imposes a limit of one focalised constituent per sentence (cf. §4.4). Benincà and Poletto (2004) assume that CF and IF occupy distinct positions within the left periphery.18 In their analysis, FocP is not a single projection, but a “field” of projections in the same way that Topic is, and no Topic can occur after or in the middle of the focus field. We adopt Benincà and Poletto’s (2004) idea of two distinct positions, but with the difference that we recognise the existence of a Top projection between the two FocPs, as illustrated in (31): (31) CFocP

TopP

IFocP

We have seen, indeed, that a Topic constituent can grammatically follow a CF, but not IF (cf. (29), (30)). Another piece of evidence supporting this assumption is that in Sicilian post-focal elements must be resumed by a clitic: (32) a. A SALVO ssu libbru *(u) detti. to Salvo that book it.CL give. PAST.1SG ‘I gave that book to Salvo.’ b. A MARIA na littira *(a) scrissi. To Maria a letter it.CL write. PAST.1SG ‘I wrote a letter to Maria.’ Sentences (32a) and (32b) would be ill-formed without the resumptive clitic and, identifying the post-focal element as a Topic, this is consistent with property [2] above regarding old information and topicalisation (cf. §2), which states that in Sicilian a resumptive pronoun is obligatory for all dislocated arguments.

Informational focus in Sicilian and the left periphery

379

6. Conclusion Sicilian provides evidence for the existence of two distinct and dedicated focus positions within the left periphery of the sentence: one for contrastive new information and the other for emphatic new information. The emphasis of IF is connected with its pragmatic relevance or with an exclamative illocutionary feature. We have seen that IF is absent in the C-domain of embedded clauses, that IF must be adjacent to the verb but CF need not be, and that IF and CF have different prosodic and intonational properties. These findings support the proposal of two distinct projections. Moreover, this paper shows that the movement of IF into the left periphery/edge of CP is associated with interpretive effects. The features responsible for focalisation can be considered in minimalist terms as EPP-features on the functional head of the CP-phase. Whether an EPP-feature is or is not associated with the C-phase is optional and indeed is subject to parametric variation, but if the EPP-feature is present, movement proves obligatory and edge-effects must be associated with the outcome. There are languages with one structural ex-situ focus position, such as Gungbe (Aboh 1999, 2004), and languages with focus in-situ (cf. Rebuschi and Tuller 1999, and Frascarelli 2005 for a survey of cross-linguistic strategies to realise Focus). In addition, there are languages in which focus constituents can appear in different positions and in many cases some interpretive restrictions hold: in Italian, for example, a focus constituent can appear in the left periphery only if it has a contrastive interpretation (Rizzi 1997, a.o.), whereas in Hungarian only an exhaustive identificational interpretation allows a focus constituent to undergo movement to the Spec/FocP (Kiss 1998, a.o.). Sicilian is another language in which the focus constituent can appear either in post-verbal position or – as result of intonational and syntactic focalisation – in the left periphery: here too the two positions are associated with different interpretations.

Acknowledgements I would like to thank Theresa Biberauer, Roberta D’Alessandro, Adam Ledgeway, Luigi Rizzi, the editor of this volume and an anonymous reviewer for helpful comments and suggestions on the present version of the paper.

380 Silvio Cruschina Notes 1. The variety presented here is that of Mussomeli, a small town in the province of Caltanissetta. However, some preliminary informal investigations among Sicilians from other areas yielded identical results, thereby suggesting a high degree of homogeneity in word order patterns across Sicilian dialects. Given the syntactic nature of this study, the data have been transcribed in normal orthography with no attempt to indicate many of the phonetic characteristics of the dialect. 2. This property has also been claimed for Italian (cf. Calabrese 1992; Frascarelli 2000). 3. The analysis of corpora, however, shows that in spoken language clitic resumption is usually present, also in the case of elements other than direct objects and partitives (cf. Frascarelli 2000: 157). 4. The translation of these examples uses standard English word order. 5. In examples (3) and (4) an indirect object is topicalised. In these cases it is possible to consider the presence of the dative clitic a case of “clitic doubling”, typical of certain Spanish varieties (Jaeggli 1982). Some Italian dialects, in fact, show this kind of doubling restricted to indirect objects (Benincà and Vanelli 1984; Cordin 1993). However this is not the case in Sicilian, in which a resumptive clitic is only licensed when it is bound by a topicalised constituent. Evidence for this claim comes from structures like (i) where a focalised indirect object proves unacceptable if doubled by a dative clitic: u detti u libbru. (i) A SALVO (*/??ci) to Salvo to-him.CL it.CL give.PAST.1SG the book ‘I gave the book to Mario.’ Additional evidence comes from wh-questions. When the questioned constituent is not the indirect object (which is therefore given information), the dative clitic proves obligatory (cf. ii); whereas if the indirect object expressing new information is questioned, the presence of the clitic yields ungrammatical results, as in (iii): (ii) Chi ci scrivisti a Maria? what to-her.CL write.PAST.2SG to Maria ‘What did you write to Maria?’ (iii) A cu (*ci) scrivisti? to whom to-him/her.CL write.PAST.2SG ‘To whom did you write?’ 6. Among the Romance languages, Catalan displays similar behaviour. Vallduvì (1995) and Vallduvì and Vilkuna (1998) claim that in Catalan all constituents conveying old information are moved to positions outside IP. Thus, also in Catalan marginalisation is not possible and clitic resumption is obligatory for dislocated arguments.

Informational focus in Sicilian and the left periphery

381

7. The emphasis associated with IF and CF will not be marked in the English translations of the examples. 8. Note that neither in Italian nor in Sicilian can contrastive focalisation answer a wh-question, which represents a request for purely new information and does not offer an explicit antecedent for contrast or correction. 9. However, the preferred strategy for answering questions in Sicilian is focalisation into the left periphery of the sentence (cf. Belletti 2005 for different strategies in other languages). 10. See Frascarelli (2005) for the same claim based on cross-linguistic evidence. 11. Equally, Echepare (1997) observes that the illocutionary force associated with emphatic focus in Basque is that of an exclamatory sentence expressing something surprising or unexpected (Echepare 1997: 114–115). Using data from Basque, he makes the distinction between two types of focus: contrastive focus (CF) and emphatic focus (EF). Semantically, CF is characterised by an existential presupposition and an exhaustive interpretation, while EF is not interpreted exhaustively and an existential presupposition does not arise. Syntactically, the two types of focus differ in that only EF undergoes overt operator-movement to an A’-position, while CF undergoes movement to an A-position. 12. We assume that this interrogative particle occupies a specific projection of the C-domain higher than FocP (22a,c) and lower than a TopP (22b), namely the position IntP identified by Rizzi (2001). 13. Note that the interrogative particle cannot be considered responsible for this different interpretation, insofar as it is optional and can also be realised in yes/no questions without focalisation (cf. (21a,b)). 14. Although throughout this paper we are generally assuming a phase framework, we are here using a pre-minimalist terminology. 15. Although we do not have a specific analysis for this case, it should be noted that in Sicilian there is one case in which this restriction seems to be violated. This occurs when the two negative quantifiers nenti ‘nothing’ and nuddu ‘nobody’, both arguments of the verb, are simultaneously focalised: (i) Nuddu nenti sapi. nobody nothing know.PRES.3SG ‘Nobody knows anything.’ As said before, quantifiers of this type cannot be topicalised. Furthermore, in the construction in question neither of the quantifiers shows prosodically or interpretively contrastive features. Two observations might be made concerning this structure: a) inversion of the two quantifiers can freely occur (cf. ii); b) it is not possible to focalise just one of the two negative quantifiers as IF (cf. iii): (ii) Nenti nuddu sapi. nothing nobody know.PRES.3SG (iii) a. *Nuddu nenti sapi. c. *Nenti nuddu sapi. b. *Nuddu sapi nenti. d. *Nenti sapi nuddu. For the moment we leave this aside for future research.

382 Silvio Cruschina 16. Note that in Italian the semantic property which characterises Hungarian identificational focus, i.e. exhaustivity, is not an inherent feature of either CF or IF, as shown by Frascarelli (2000) and Brunetti (2004). This is also valid for Sicilian. 17. This incompatibility holds not only for CF and IF in the left periphery, but also for CF in the left periphery in conjunction with post-verbal IF (see Belletti 2004a). 18. Belletti (2001, 2004a) also proposes two distinct positions for the two types of focus in Italian: a clause-external position in the left periphery of the sentence and a clause-internal position in the left periphery of vP. She assumes that in Italian these positions are specialised, in the sense that the higher focus in the left periphery can only host contrastively focalised constituents, whereas the lower focus position can only host informationally focalised elements.

References Aboh, Enoch 1999 From the syntax of Gungbe to the grammar of Gbe. Ph.D. diss., University of Geneva. 2004 The Morphosyntax of Complement-Head-Sequences. Oxford: Oxford University Press. Antinucci, Francesco and Guglielmo Cinque 1977 Sull’ordine delle parole in Italiano. L’emarginazione. Studi di Grammatica Italiana 6: 121–146. Bayer, Josef 2001 Asymmetry in emphatic topicalization. In Studia Grammatica 52: Audiatur Vox Sapientiae, Caroline Féry and Wolfgang Sternefeld (eds.), 15–47. Berlin: Akademie Verlag. Belletti, Adriana 2001 Inversion as focalization. In Subject Inversion in Romance and the Theory of Universal Grammar, Aafke Hulk and Jean-Yves Pollock (eds.), 60–90. Oxford: Oxford University Press. 2004a Aspects of the low IP area. In The Structure of IP and CP. The Cartography of Syntactic Structures, Vol. 2, Luigi Rizzi (ed.), 16–51. Oxford: Oxford University Press. 2004b (Ed.). Structures and Beyond. The Cartography of Syntactic Structures, Vol. 3. Oxford: Oxford University Press. 2005 Answering with a “cleft”: the role of the null subject parameter and the vP periphery. In Proceedings of the Thirtieth “Incontro di Grammatica Generativa”, Laura Brugè, Giuliana Giusti, Nicola Munaro, Walter Schweikert and Giuseppina Turano (eds.), 63–82. Venezia: Cafoscarina.

Informational focus in Sicilian and the left periphery

383

Benincà, Paola 1988 L’ordine degli elementi della frase e le costruzioni marcate. In Grande Grammatica Italiana di Consultazione, Vol. 1, Lorenzo Renzi and Giampaolo Salvi (ed.), 129–194. Bologna: Il Mulino. Benincà, Paola and Cecilia Poletto 2004 Topic, focus and V2: defining the CP sublayers. In The Structure of IP and CP. The Cartography of Syntactic Structures, Vol. 2, Luigi Rizzi (ed.), 52–75. Oxford: Oxford University Press. Benincà, Paola and Laura Vanelli 1984 Italiano, veneto, friulano: fenomeni sintattici a confronto. Rivista Italiana di Dialettologia 8: 165–194. Brunetti, Lisa 2004 A Unification of Focus. Padova: Unipress. Calabrese, Andrea 1982 Alcune ipotesi sulla struttura informazionale della frase in italiano e sul suo rapporto con la struttura fonologica. Rivista di Grammatica Generativa, 5: 65–116. 1992 Some remarks on focus and logical structures in Italian. In Harward Working Paper in Linguistics 1: 91–127. Cardinaletti, Anna 2001 A second thought on emarginazione: Destressing vs. ‘Right Dislocation’. In Current Studies in Italian Syntax. Essays Offered to Lorenzo Renzi, Guglielmo Cinque and Giampaolo Salvi (eds.), 117–135. Amsterdam: Elsevier. Cinque, Guglielmo 1990 Types of A’ Dependencies. Cambridge MA, MIT Press. 1999 Adverbs and Functional Heads. Oxford: Oxford University Press. 2002 (Ed.). The Structure of DP and IP. The Cartography of Syntactic Structures, Vol. 1. Oxford: Oxford University Press. Chomsky, Noam 2001 Derivation by phase. In Kale Hale: A Life in Language, Michael Kenstowicz (ed.), 1–52. Cambridge, MA: MIT Press. 2002 On Language and Nature. Cambridge: Cambridge University Press. 2004 Beyond explanatory adequacy. In Structures and Beyond. The Cartography of Syntactic Structures, Vol. 3, Adriana Belletti (ed.), 104– 131. Oxford: Oxford University Press. 2005 On phases. Ms., Cambridge, MA: MIT. Cordin, Patrizia 1993 Dative clitic doubling in Trentino. In Syntactic Theory and the Dialects of Italy, Adriana Belletti (ed.), 130–154. Torino: Rosenberg and Sellier. Donati, Caterina and Marina Nespor 2003 From focus to syntax. Lingua 113: 1119–1142.

384 Silvio Cruschina Echepare, Ricardo 1997 Two types of focus in Basque. In Proceedings of the Fifteenth West Coast Conference on Formal Linguistics, Brian Agbayani and SzeWing Tang (eds.), 113–127. Stanford Linguistics Association: Center for the Study of Language and Information. Frascarelli, Mara 2000 The Syntax-Phonology Interface in Focus and Topic Constructions in Italian. Dordrecht: Kluwer. 2005 Focus as predicate and Spell-Out conditions. Deriving cross-linguistic strategies from a unified account. Ms., Università degli Studi Roma Tre, Rome. Haegeman, Liliane 2002 Anchoring to the speaker, adverbial clauses and the structure of CP. In Georgetown University Working Papers in Theoretical Linguistics 2: 117–180. Department of Linguistics, Georgetown University, Washington DC. 2004 Topicalization, CLLD and the left periphery. In Proceedings of the Dislocated Elements Workshop ZAS Berlin, November 2003, Benjamin Shaer, Werner Frey and Claudia Maienborn (eds.), 157–192. Berlin: ZAS Papers in Linguistics 35. Jaeggli, Osvaldo 1982 Topics in Romance Syntax. Dordrecht: Foris. Kayne, Richard 1994 The Antisymmetry of Syntax. Cambridge, MA: MIT Pres. Kiss, Katalin É. 1998 Identificational focus versus information focus. Language 74: 245– 273. Nespor, Marina and Maria Teresa Guasti 2002 Focus to stress alignment and its consequences for acquisition. Lingue e Linguaggio 1: 79–106. Rebuschi, Georges and Laurice Tuller (eds.) 1999 The Grammar of Focus. Amsterdam: John Benjamins. Rizzi, Luigi 1997 The fine structure of the left periphery. In Elements of Grammar. Handbook in Generative Syntax, Liliane Haegeman (ed.), 281–337. Dordrecht: Kluwer. 2001 On the position “Int(errogative)” in the left periphery of the clause. In Current Studies in Italian Syntax. Essays Offered to Lorenzo Renzi, Guglielmo Cinque and Giampaolo Salvi (eds.), 287–296. Amsterdam: Elsevier. 2004 (Ed.). The Structure of IP and CP. The Cartography of Syntactic Structures, Vol. 2. Oxford: Oxford University Press.

Informational focus in Sicilian and the left periphery

385

Rohlfs, Gerhard 1969 Grammatica Storica della Lingua Italiana e dei suoi Dialetti, Vol. 3, Sintassi e Formazione delle Parole. Torino: Einaudi. Vallduvì, Enric 1995 Structural properties of information packaging in Catalan. In Discourse-Configurational Languages, Katalin. É. Kiss (ed.), 122–152. Oxford: Oxford University Press. Vallduvì, Enric and Maria Vilkuna 1998 On rheme and kontrast. In Syntax and Semantics 29: The Limits of Syntax, Peter W. Culicover and Louise McNally (eds.), 79–108. San Diego: Academic Press. Zubizarreta, Maria Luisa 1988 Prosody, Focus, and Word Order. Cambridge, MA: MIT Press.

Subject index

adjectives, 18, 29, 35, 133, 137, 138– 140, 143, 146–153, 155–160, 168– 171, 173–175, 179–182, 277, 286, 288, 371 Agree, 2, 7, 12, 62, 213, 216, 228, 270, 285, 323 anaphora, 8–9, 71–74, 77–81, 85, 91– 99, 102, 214, 307, 314, 317–318 anticausatives, 187–89, 192–199, 201– 206, 208 AP, 29, 124, 154–156, 166, 172–176, 179, 180–181 be, 107–120, 122, 125–130, 372 binding theory, 6, 10, 12, 25, 190, 295, 300–301, 307, 311–313, 321–322, 326–327, 332, 335, 374 causative(s), 9, 128, 187, 189–193, 195, 197, 199, 201, 202, 203, 204, 205, 207–208, 282, 287 comparative(s), 99, 133, 143, 147–149, 151–155, 157–159 CP-phase, 7, 10–11, 108, 126, 128, 173, 227, 261, 270–271, 284, 293, 325, 326, 364 cyclic spell-out, 237, 249 de se, 10, 306, 308–309, 313, 314–315 definiteness, 29, 117, 119–120, 122, 125–127, 164, 173, 175, 180–181 degree questions, 9, 133, 135, 136, 138– 139, 146, 151–152, 157 discourse grammar, 330, 332, 335 disjoint reference, 295, 328, 335 DP-phase, 4, 6, 9, 105, 108, 123, 126, 173, 261–262, 285, 288 Economize structure, 167, 169–170

edge(s), 4, 5, 7, 8, 10, 29, 40, 80, 126, 164, 237, 249–250, 252, 324, 326, 335–356, 363, 365, 379 ellipsis, 5, 8, 71–74, 76, 78–83, 85–94, 96, 98–99, 101–102 EPP, 2, 5, 7, 8, 10, 12, 164, 166, 322– 323, 328, 335, 341–343, 345, 358, 365, 379 event, 9, 98, 163, 187, 193, 195, 199, 202, 204, 206–207, 213, 218–223, 226, 229, 244–245, 301, 367 existential, 47, 81, 107–111, 113, 115– 120, 122, 125, 127, 129, 381 extraposition, 29, 32–33, 41 feature(s), 2, 4–8, 10–12, 29, 31, 40, 47, 53, 60–61, 63, 65, 90, 100, 120, 158, 163–167, 169, 172, 174, 176–181, 190, 202, 204, 206, 207, 213, 216– 217, 219–222, 226–229, 232, 241, 244, 261, 268–271, 273, 285–287, 309–310, 313, 321–327, 341, 345– 347, 352–354, 358–359, 363, 365– 367, 371, 376–377, 379, 381–382 free relative clauses (FR), 17–18, 20– 22, 24, 31–33, 38–39, 215 Focus, 6, 10–11, 164–165, 168, 179, 181, 261, 263–264, 267, 269–272, 275, 279–281, 284–288, 325, 327, 335–336, 364, 367, 379 contrastive ~, 11, 363, 367, 381 informational ~, 363 gerund, 239–240, 242, 243, 246–247, 250, 252–253, 296, 300 gradable predicate(s), 9, 141, 144, 149, 156 graft, 18, 20–22, 26–32, 34, 37, 38, 41

388 Subject index have, 9, 107–111, 113–115, 117, 119– 122, 125–130 head, head movement, 1, 41, 45, 65, 125, 224, 241, 248, 269, 352 functional head, 1, 7, 10, 154, 164– 165, 201, 217, 220, 222, 226, 228– 229, 237, 245, 253, 261–262, 270– 271, 313, 341, 349, 364–265, 379 lexical head, 149, 164, 285 silent head, 83–84, 89, 91 head noun, 21, 23–24, 29, 31, 33, 38, 129, 167, 171, 179, 278, 288 head phase, 1, 4, 7–8, 10, 365 heavy NP shift, 388 Horn Amalgam (HA), 17, 20, 33–35, 41, 56 index, 76, 78, 82, 86, 87, 96, 98, 244, 298, 311, 323, 388 interpretive effects, 379 IPP-effect, 237, 238, 239, 240, 241, 242, 243, 246, 247, 250, 251, 252, 253, 256 left periphery, 2, 7, 9, 11, 165, 173, 175, 261–264, 267, 269, 271, 275–279, 284–285, 287–288, 323, 332, 343– 344, 347–348, 358, 363–365, 367– 372, 376–379, 381–382 linearization, 8, 12, 32, 37, 40, 257 locative inversion, 341–342, 344, 347– 349, 352, 358 measure phrase(s), 133, 139, 140, 142, 143, 146, 147, 149–152, 156, 158, 388 Merge, 1–5, 8, 12, 17–19, 28, 36, 40, 167, 226, 274, 345, 351 Internal ~, 4–5, 8, 10–12, 18–20, 22, 37–38, 166–167, 175, 365 External ~, 8, 12, 17, 22, 28, 32, 175 modals, 8, 45, 48, 50–52, 55–57, 63–64, 256, 279, 282–283 Move, 2, 4, 5, 12, 20, 345

negation, 8, 47–48, 50–53, 56–58, 60– 65, 121, 129, 274 null subject language(s), 4, 321, 342 number agreement, 10, 38, 213, 216– 219, 222, 227–229, 231, 349 obviation, 10, 295–296, 298, 300, 301– 303, 307, 309–310 operator(s), 34, 47, 49, 51, 54, 133, 135, 137, 141, 144, 146–148, 151–155, 158, 207, 221, 231–232, 262–264, 286, 312, 353, 373–376, 381 null ~, 9, 144–146, 156, 264 relative ~, 80–81, 84–85, 88, 346 passive(s), 40, 187–189, 193–199, 201– 202, 204–205, 208, 256, 265, 270, 299, 312 Past Participle Agreement, 2166, 262, 268–269, 283–286, 333 PF deletion, 9, 71–73, 82–83, 85, 86–91 Phase Impenetrability Condition (PIC), 3–6, 238–239, 243, 249–250, 255 phrasal affix, 252–253 pluralization, 9, 213, 220, 222, 226, 229, 231–232 possessive, 9, 107–116, 119–129, 164, 168, 170, 179–180, 288, 307 PP modification, 193, 201, 203 pro, 4, 64, 75, 77–78, 98, 129, 231, 263, 297–299, 300–302, 305, 309–314, 321–337, 342 pronouns, 29, 39, 86, 93, 98, 129, 263, 314–315, 321–322, 324, 326–330, 332, 334–336 prosodic constraint(s), 252–253, 255– 256 quantifier scope, 46–51, 60–61, 79, 81, 94, 97–99 quantifier(s), 45–46, 48–49, 52, 57, 60, 75–76, 81, 90, 98, 179–180, 214, 230, 262, 273–275, 311, 373–375, 381

Subject index reciprocal, 9, 213–219, 222–226, 230– 232 reconstruction, 5, 38, 47–52, 55, 58–61, 63, 83 relative clause(s), 21–25, 28, 31, 33, 38– 39, 49, 49, 74, 76, 78, 80, 86, 94, 96, 158, 178, 262, 272, 280–281, 283, 286, 332–333 headed relative clauses, 21 headed relatives, 38 restructuring, 237–239, 243–245, 250 root clauses, 41, 181, 370 scope, 8, 34, 38, 45–65, 75, 79, 81, 94, 97–99, 229, 231, 243, 347–348, 358 scope splitting, 47, 60 scrambling, 10, 158, 261–262, 264, 267– 268, 270–271, 273, 275–278, 281, 283–359 Spell-out, 6, 7, 167, 237, 240–241, 247, 249, 252–255, 257 subject, 3, 4, 8, 10, 22–23, 25–26, 29, 41, 47, 50, 53–63, 65, 76, 80, 97, 108– 111, 114–119, 122, 126, 129, 133, 142, 144, 166, 178–179, 192, 200– 201, 206, 213, 215, 217, 221–222,

389

224–228, 230–231, 237, 245, 262– 267–268, 270–271, 279, 282, 284, 295–303, 305, 306, 309–313, 321– 322, 324, 327–328, 330–336, 341– 346, 348–359, 369 subjunctive, 10, 295–305, 307–310, 312–313, 315 syntactic decomposition, 188, 201, 205 temporal anchoring, 310 transparent free relatives (TFR), 24–27, 29, 34, 39 Topic, 10, 164–165, 168–169, 174, 261, 263–264, 267, 275, 278, 285, 322, 324–337, 343, 347, 352, 376–378 Topic, contrastive, 168, 173, 325 verb cluster, 237–240, 242–243, 245– 255, 257 VP anaphor, 72–79, 85, 87–89, 94, 96– 98 VP ellipsis, 8, 71–73, 78–79, 81, 83, 85, 88–89, 91–92, 96 vP-phase, 4, 6–10, 12, 185, 227, 271, 332 low ~, 261, 267, 271

Language index

Albanian, 163, 170, 171–173, 175, 180 Basque, 381, 383 Catalan, 280, 295, 380, 383 Chinese, 261 Coptic, 1, 9, 107–109, 111, 114–117, 119, 121, 122, 124, 127–129 Dutch, 1, 18, 24, 29, 31–32, 35, 237– 239, 243, 246, 251–253, 256, 273, 284, 295 English, 1, 9, 10, 46, 50, 61, 79–80, 85, 88–91, 100–101, 111, 114, 117, 129, 133–134, 136, 143, 146, 156, 164, 178–179, 187, 188, 191, 193–204, 206, 216, 230–231, 242, 315, 341–343, 345– 346, 348–350, 352, 358, 380–381 French, 164, 207, 227, 263, 280, 289, 295, 311, 342, 346 Frisian, 240, 242 Friulian, 268–269 Greek, 1, 9, 164, 187, 193, 195, 197–204, 206, 208, 265 Gungbe, 164, 379 Hungarian, 1, 109, 111, 164, 174, 176, 376, 379, 382 Icelandic, 133–135, 139, 144, 146, 151, 230–231, 284

Italian, 1, 9–10, 73–74, 78–80, 87–91, 97, 101, 128, 163, 166–169, 172, 175, 179, 213, 216, 227, 230–231, 244, 261– 263, 270, 295–298, 305, 307, 310– 311, 313–315, 321, 324–325, 327, 329, 335–336, 342, 363–367, 369, 371, 375–376, 379–382 Japanese, 38, 191, 284 Korean, 284, 287 Middle High German, 243, 256 Northern Norwegian, 9, 133, 134, 151, 152, 156 Norwegian, 1, 133, 134, 135, 138 Old High German, 242 Old Italian, 10 Portuguese, 295, 311 Russian, 191, 295 Serbo-Croatian, 163, 173, 175, 180 Sicilian, 11, 363–368, 371, 373, 375–382 Somali, 179 Spanish, 157–158, 280, 295, 311, 380 Turkish, 176 West Flemish, 1, 237, 251 Yiddish, 284