Explorations of Phase Theory: Features and Arguments 9783110213966, 9783110205206

Citation preview

Explorations of Phase Theory: Features and Arguments

≥

Interface Explorations 18

Editors

Artemis Alexiadou T. Alan Hall

Mouton de Gruyter Berlin · New York

Explorations of Phase Theory: Features and Arguments

edited by

Kleanthes K. Grohmann

Mouton de Gruyter Berlin · New York

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

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 Explorations of phase theory : features and arguments / edited by Kleanthes K. p. cm. ⫺ (Interface explorations ; 18) Includes bibliographical references and index. ISBN 978-3-11-020520-6 (hardcover : alk. paper) 1. Minimalist theory (Linguistics) 2. Generative grammar 3. Grammar, Comparative and general ⫺ Syntax. I. Grohmann, Kleanthes K. P158.28.E97 2009 4151.0182⫺dc22 2009002411

Bibliographic information published by the Deutsche Nationalbibliothek The Deutsche Nationalbibliothek lists this publication in the Deutsche Nationalbibliografie; detailed bibliographic data are available in the Internet at http://dnb.d-nb.de.

ISBN 978-3-11-020520-6 ISSN 1861-4167 ” Copyright 2009 by Walter de Gruyter GmbH & Co. KG, 10785 Berlin All rights reserved, including those of translation into foreign languages. No part of this book may be reproduced or transmitted 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, Laufen. Printed in Germany.

Contents

Exploring features and arguments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Kleanthes K. Grohmann

1

Remarks on features . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21 Halldór Ármann Sigurðsson Feature valuation by sideward movement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53 Bartosz Wiland On Incrementality, overt agreement, theta-roles and Spec,Head relations in the phase-based framework. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87 Iván Ortega-Santos Structural case, locality, and cyclicity. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107 Thomas McFadden PRO, pro, and NP-trace (raising) are interpretations . . . . . . . . . . . . . . . . . . 131 M. Rita Manzini Movement of arguments and negative feature . . . . . . . . . . . . . . . . . . . . . . . . 181 Masaaki Kamiya Inner aspect and phases . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 207 Jonathan E. MacDonald How to become passive . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 231 Berit Gehrke and Nino Grillo Ergativity, accusativity, and the order of Merge and Agree . . . . . . . . . . . 269 Gereon Müller Contributors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 309 Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 311

Exploring features and arguments Kleanthes K. Grohmann

1. Introduction The InterPhases conference, held at Casteliotissa Hall in the Old Town of Nicosia, Cyprus 18–20 May 2006, was an attempt to bring together linguists working on all kinds of interface-related issues (the inter-part of the title) pertaining to current, generative syntactic theory (such as the eponymous phases). It was also many things beyond this narrow aim; in particular, the event sparked a lot of interest and discussion. At the conference, 25 papers were delivered orally (incl. Noam Chomsky’s keynote address and three invited lectures; see also Grohmann 2009b) and another 25 posters presented. All in all, close to 200 linguists got together, interacted lively, and exchanged ideas for a good week, if the Edges in Syntax conference immediately preceding is taken into consideration as well (on this combined Cyprus Syntaxfest, see the introduction to Grohmann & Panagiotidis 2009). Above all, the hope was that the conference was also successful intellectually. The present collection, and its sister volume Explorations of Phase Theory: Interpretation at the Interfaces, is one effort to give testimony to that hope.1 The present compilation is assembled from work accepted for presentation at the InterPhases conference. Apart from an invited speaker’s work (Müller), selected oral (Manzini, Sigurðsson) and poster presentations (Geraci, Gehrke & Grillo, Kamiya, MacDonald, Ortega-Santos), as well as one that unfortunately had to be canceled (McFadden), have been specially prepared for this volume by the authors, having benefited greatly from the feedback received at the conference and afterwards. In addition, one more chapter was solicited explicitly for this volume (Wiland). The selection criterion for this collection is that each chapter explicitly address issues concerning Features and Arguments – the sub-title of this volume – all framed 1

This ‘sister volume’ volume on “Explorations of Phase Theory” also derived from the InterPhases conference (Interface Explorations 17). Note that the two introductions share large parts, in particular sections 1 and 2 as well as the overall arrangement, but, of course, differ in sections 3 and 4, the volume-specific thematic and contributions overviews.

2 Kleanthes K. Grohmann within the leading theme of this volume (and the previous, the abovementioned sister volume), “Explorations of Phase Theory”. Before briefly presenting each contribution in turn, and putting it in perspective to this collection, I will introduce the volume with a sketch of the fundamental properties of Phase Theory and of what kind of interesting or relevant aspects features and arguments provide, also with respect to those contributions that take a more critical stance towards Phase Theory as understood in current research.

2. Phase theory For the purposes of the present collection, I take ‘Phase Theory’ to be representative of the leading developments within the current, generative framework collectively referred to as the Minimalist Program, instigated by Noam Chomsky’s early minimalist work, then published as Chomsky (1995). Phase Theory – employed here sometimes in alternation with the term ‘phasebased approach’ and other synonyms – collectively refers to Chomsky’s (2000) proposal, originally circulated in 1998, his subsequent work over the past 10 years, and various extensions by other scholars, which I will briefly sketch in this section.2 Chomsky’s formulation of the syntactic derivation by phase addresses certain aspects of a ‘dynamic’ approach to the computation, originally advanced by Uriagereka in the mid-1990s (published as Uriagereka 1997, 1999), the so-called ‘Multiple Spell-Out Hypothesis’ (see the sister volume to the present collection for more discussion). The major point of departure from earlier instantiations of the Minimalist Program here lies in the architecture of the grammar. While minimalism as conceived of in Chomsky (1993, 1995) adhered to a slightly modified form of the Y- or T-model from the GB era (Chomsky 1981, 1986), where the interpretive interface levels LF (Logical Form) and PF (Phonetic or Phonological Form) were accessed once through the single application of the operation Spell-Out, as in (1), 2

For textbook introductions, see Adger (2003: ch. 10), Radford (2004: ch. 10), and Hornstein et al. (2005: ch. 10), Lasnik & Uriagereka with Boeckx (2005: section 7.4), and Boeckx (2008: section 3.2), among others. More elaborate expositions, including interesting extensions, of Phase Theory include the recent dissertations by Richards (2004), Hiraiwa (2005), and Gallego (2007), to name but a few, and a host of research monographs and collected volumes (see also den Dikken 2007 and peer responses).

Exploring features and arguments

(1)

3

LEX (qua numeration or lexical array)

PF (instructing the SM system)

Spell-Out

LF (instructing the C-I system) Phase Theory explores a more intricate view of Transfer. The major difference lies in the (new) operation Transfer, that is, from narrow syntax (NS) to the interpretive interface levels LF and PF. Access on all other sides remains as conceived before, that is (finer details aside), NS starts off with some kind of pre-selection from the Lexicon (LEX) in the form of a numeration or, as it is now known in Phase Theory, lexical array and the interpretive – or, as used here (see Grohmann 2009a for discussion), modular – interface levels still feed the linguistic interfaces known as the Sensorimotor (SM) system and Conceptual-Intentional (C-I) system, respectively. Transfer is the ‘super-operation’ (Lasnik & Uriagereka with Boeckx 2005) feeding the modular interfaces, made up of Transfer to LF (Interpret) and Transfer to PF (Spell-Out). Within Phase Theory, Transfer thus understood is assumed to apply more than once, throughout the derivation – which leads to a dynamic evaluation of NS, to use a popular term. This can be captured by the diagram in (2), adapted from Boeckx (2008). (2) lf

pf

lf

pf

lf . . .

pf . . .

4 Kleanthes K. Grohmann From such a dynamic conception of the derivation, several paths can be explored – Uriagereka’s (1999) Multiple Spell-Out model is one, Grohmann’s (2003) framework based on Prolific Domains is another, and further alternatives exist as well. But Chomsky’s (2000 et seq.) Phase Theory seems to be the dominant one these days and constitutes the focus of the present collection, at least as a benchmark or term of comparison. This means, applied to (2), that the relevant unit of the derivation subject to Transfer is the phase – simply put, each phase undergoes Transfer. To be more precise, Chomsky distinguishes strong and weak phases in that only the former are relevant for the computation (Transfer). A phase (ignoring weak phases from now on, this refers to ‘strong phase’) is thus the very local unit for computation from at least three perspectives: (3) a. A phase contains only the lexical array that is needed for its assembly. b. A phase constitutes a local computational domain for narrow syntax. c. A phase undergoes Transfer as soon as possible after its completion. Again taking some shortcuts in the interest of a brief presentation, the Lexi con thus pre-selects several numerations, namely, one lexical array per phase head. Each lexical array is then depleted by subsequent applications of the operation (External) Merge, providing the derivation with a narrow, local domain relevant for subsequent computation. Eventually, the phase acts as a Spell-Out domain of sorts, which means that it undergoes Transfer (to both LF and PF, actually) – and which means that it then becomes impenetrable for further computation, freezing the material contained within it. The freezing part – perhaps a remnant of pre-GB-theoretical concepts, as recently suggested3 – is formulated in terms of the Phase Impenetrability Condition (PIC) in (4), taken from Chomsky (2004), updating earlier version (see also Nissenbaum 2000).

3

Scheer (2008), for example, classifies this condition as a ‘no look-back device’ and traces its history to the Strict Cycle Condition (Chomsky 1973) and its implementation in phonological theory, all the way up to Government Phonology (Kaye 1992, 1993) – but subsequently “forgotten in GB syntax” (see his contribution to the sister volume for more). Note also that Abels (2003) finds a precursor of the PIC in van Riemsdijk’s (1978) Head Constraint, in this way coinciding with Scheer’s assessment that it predates GB.

Exploring features and arguments

(4)

5

Phase Impenetrability Condition (PIC; Chomsky 2004: 108) At the phase ZP containing phase HP, the domain of H is not accessible to operations, but only the edge of HP.

In essence this means that after the phase of some phase head PH1 is complete (HP in (4)) and elements from the next higher phasal lexical array are merged, that phase becomes inaccessible for further computation at the point at which the next higher phase head PH2 is merged (ZP). To be more precise, what becomes inaccessible, or impenetrable, is the domain of the phase head, defined as its sister and everything contained (i.e. c-command) – the so-called edge remains accessible, crucially so. The edge includes the phase head itself and its specifiers, in a way recreating the kind of escape hatch that became popular in the Barriers-framework (Chomsky 1986); concerning (3c) and related issues, see Boeckx & Grohmann (2007) for recent discussion and further references. One immediate effect for the syntactic derivation is the necessity of built-in ‘escape hatches’ (not unlike those from the Barriers-model of Chomsky 1986), through which those syntactic objects must move that need to get out of one phase to target a higher phase. As a simple way of illustration, this can be sketched for wh-questions, as in (5): (5)

a. b. c. d. e. f.

[vP John v [VP kiss who ]] [vP who [vP John kiss-v [VP V who ]] will-T [vP who [vP John kiss-v [VP V who ]] [TP John will-T [vP who [vP John kiss-v [VP V who ]] C [TP John will-T [vP who [vP John kiss-v [VP V who ]] [CP who will-C [TP John T [vP who [vP John kiss-v [VP V who ]]

Once all theta-roles are assigned within the vP (5a), V raises to v and the wh-phrase moves into an outer specifier of v (5b). (Allowing such an additional Spec-position is referred to as the ‘P(eripheral)-feature’ (Chomsky 2000) or ‘EPP-property’ (Chomsky 2001 et seq.) of phase-heads – as many as needed.) The usual T-insertion (5c) and subject raising applies (5d) before the next higher phase head, interrogative C, is merged into the structure (5e). Now the PIC (4) applies and the only way who can move to [Spec,CP] is from the edge of the vP-phase, as in (5f). Had who not moved to [Spec,vP] previously, it would now be frozen in place. That is, an analysis as assumed in earlier versions of minimalism, according to which the object wh-phrase moves straight from its base to [Spec,CP] is not available anymore.

6 Kleanthes K. Grohmann Additional debate surrounds the details of (3b) – just what or which operations exactly cannot apply over a lower phase? This question leads us to another staple of phase-theoretic innovations of the Minimalist Program: grammatical licensing mechanisms. Where in earlier instantiations, Move took care of feature checking through Spec-Head configurations (Checking Theory of Chomsky 1993) or Attract was responsible for displacement (Chomsky 1995), current approaches employ a Probe-Goal system of feature checking known as Agree (Chomsky 2000). Agree checks features between a Probe and a lower Goal with matching features through longdistance, through c-command. Bošković (2007), for example, argues that only Move is subject to the PIC, but not Agree as Chomsky suggests; for additional details and discussion, see the next section (and a number of chapters in the following). (3a) seems to be the least controversial aspect of phases, unless one considers the relation between NS and LEX in more detail, or from the perspective of Distributed Morphology (Halle & Marantz 1993 and much subsequent research), for example, which may lead so quite a different characterization. But this property was one of the core arguments in favour of a cyclic, piecemeal derivation and has in public been discussed as early as Wilder & Gärtner (1997) in the context of a proceedings volume for a conference held in February 1995 – thus even predating any discussion of Multiple Spell-Out models or other aspects of dynamic computations. This concerns data, such as the following:4 (6)

a. There was [a rumor [that a mani was ti in the room]] in the air. b. [A rumor [that there was a man in the room]]i was ti in the air.

This pair of examples offers an apparent case of optionality that is puzzling for a single-numeration view of the derivation and the assumption that Merge is more economical than Move. (6a) and (6b) are presumably derived from the same lexical pre-selection. If now ‘Merge-over-Move’ were real (the focus of Castillo et al. 1999), (6b) should always block the derivation of (6a). If each phase has its own lexical pre-selection (the lexical array), this problem is solved: there is in different lexical arrays (or subarrays) in the two examples.

4

Apart from Wilder & Gärtner’s (1997) presentation and the one independently arrived at in Uriagereka (1999), see also Castillo et al. (1999), Hornstein et al. (2005), and references cited for further discussion.

Exploring features and arguments

7

However, this now leads to yet another question not mentioned so far – and arguably even more controversial than (3c), but certainly (3b) and especially (3a): What constitutes a phase? I will only sketch this issue here.5 Chomsky (2000) originally suggested that v and C, but not V and T, are (strong) phase heads, and later, citing research by other scholars, alludes to the possibility that D (Chomsky 2008), perhaps even P, may also constitute phase heads. This is a hot item of contention within Phase Theory and without. Regarding the latter, for example, Grohmann (2003) suggests an alternative dynamic framework in which vP, TP, and CP are relevant Spell-Out domains (although not subject to the PIC) and Uriagereka’s (1999) original model suggested so-called command units to be impenetrable and frozen (basically, left branches, but again without assuming a condition like the PIC). Within phase-theoretic approaches, Marušič (2005), for example, argues for the concept of ‘non-simultaneous Spell-Out’, where a given phase may undergo Transfer to one modular interface level (such as the phonological component PF) but not the other (such as the semantic component LF). Gallego (2007), to mention another such example, suggests the process of ‘phase sliding’, which may turn TP into a phase under given circumstances. The literature on the identity, and the properties, of phases – and problems the standard view may face as well as solutions how these may be overcome – grows steadily. What all such dynamic approaches have in common is the search for local, economical, and computationally efficient mechanisms in the syntax. Some of these will be addressed throughout this collection (and its sister volume), others will have to be collected elsewhere. Let’s take a closer look at the theme of the present volume. 3. Features and arguments It was alluded to above already that (what I called there) ‘grammatical licensing mechanisms’ have undergone considerable change within the Minimalist Program. Many, if not all, aspects of such grammatical licensing is going to be picked up on the next pages – checking, agreement, and Agree as well as the right structural configurations, with locality concerns, thetaroles, and A-(movement-)constructions at large playing a role in several contributions. The reason for this is obvious: Formal features have, all along 5

Again, I refer the reader to the critical discussion in Boeckx & Grohmann (2007), including the literature cited, beyond basic expositions elsewhere (see fn. 2).

8 Kleanthes K. Grohmann in minimalist investigations, been assumed to be the driving force behind displacement in natural language for one reason or another – in particular, through the specific technical implementation of licensing this grammatical property, also known as feature checking. In this sense, all contributions to this volume say something about features – their nature, their licensing, their role in the syntactic computation, their interface interpretation effects. In this way, the present volume also constitutes a thematic match to the sister volume, especially where the contributors explicitly address interface properties relating to features. Beyond features, arguments throw up similar questions – how can they be defined in a phase-based approach to the derivation, what movement steps and other processes do they underlie in such a model, how are they licensed, i.e. interpreted, and so forth. The next section summarizes each chapter briefly, more or less in isolation. What I intend to do in this section is to capture larger themes that hold some of the chapters more closely together than others. As it turns out, the volume can be conceived of an even tripartition in the order of assembly in this volume. Concerning the nature of features, a major issue has been since the earliest minimalist conceptions (Chomsky 1993) the strength of a feature, and its input for the computation. Originally, Chomsky distinguished ‘strong’ from ‘weak’ features, where the former had to move overtly, prior to SpellOut, and the latter covertly, i.e. at LF. Another property of such formal features (whether they’re categorial or intrinsic) is their interpretability: Some features can be interpreted at an interface level, others cannot. By definition, semantic features can be interpreted at LF, for example, and phonetic features at PF. But the formal features involved in ‘feature checking’ (whatever the exact nature of such grammatical licensing) are presumably more abstract. So it has been argued that the phi-features of nominal expressions do get interpreted at LF (book vs. books or me vs. you, for example), whereas the corresponding phi-features of verbal elements do not – and consequently must be deleted before they reach the interface. Much more can be said about this simplified presentation of affairs, and several colleagues do so on the next few pages (Sigurðsson most clearly, also with respect to interpretation at the interfaces and the nature of interface levels). On the technical side of implementing grammatical licensing mechanisms, Phase Theory has abolished earlier Checking Theory, as briefly sketched above. The ProbeGoal system of Agree or feature valuation is also the concern of some contributors. Here one issue of discussion, if not even contention, is Chomsky’s (2004) recent rejection of the Spec-Head relation as the structural configuration in which features get licensed (discussed by Ortega-Santos). Another

Exploring features and arguments

9

aspect of feature valuation is the role the licensing mechanism plays for displacement, i.e. movement (discussed especially by Wiland). In both these chapters, the other conjunct of the volumes sub-title enjoys some discussion as well, namely, thematically marked arguments and the question how theta-roles are best integrated into the grammar. In all of Chomsky’s writings, theta-roles have always taken a special place in that they are configurationally licensed. An alternative would be to relegate them to the level of a more or less regular formal feature, in the spirit of work by Hornstein (2001) and others. The second implicit part of this volume discusses the connection between features and specific types of arguments and nominal expressions. Just as theta-roles have caused some confusion over the years, so have case properties in languages. Case is arguably the formal feature par excellence, without any tangible interpretive effect. Still, languages differ in whether they mark case morphologically or ‘only’ abstractly. But as a typical property of A-movement – as grammatical constructions such as passive, expletive-associate, raising, and so on have become known in GB – local clauseboundedness has always been around. This leads to a potential problem for the checking replacement of Agree as the one and only grammatical licensing mechanism on grammar (see McFadden’s chapter for a thorough discussion). Other aspects of argument movement and the interplay with other properties in Phase Theory – also from an interface perspective, and extending to A’-syntax – concerns negation, quantifiers, and scope (as Kamiya discusses). Likewise, it has been argued – again by Hornstein (2001) and other works – that a typical rule of construal such as (obligatory) control may boil down to independently needed operations of the computation, namely, Move and as such be the product of displacement rather than construal. This likes control, at first glance, to raising rather than anything else, and has serious implications for empty nominal expressions such as case-less PRO or case-marked pro. The interplay of all of these, and a cross-linguistic angle, is also provided in what follows (Manzini does so in her chapter). The final chapter triplet forms another implicit theme, which can be characterized as aspect at large. At the thematic part of the clause, that is, the vP phase in Phase Theory (or the Θ-Domain in the terminology of Grohmann 2003), so-called ‘inner aspect’ gets computed – the calculation of telicity (the topic of MacDonald’s chapter). Moving up, the thematic part of the derivation interacts with the higher inflectional part – not a phase in Phase Theory (TP), but constituting the Φ -Domain in Grohmann’s framework – when it comes to passivization, again one of the paradigmatic examples of A-movement. Apart from pure formal facets that have been a

10 Kleanthes K. Grohmann staple of passives in generative theorizing for decades, such as movement of the passivized object to the grammatical subject position in a language like English, eventive properties, as exponents of aspect at large, can be observed as well. Working this out shows not only that there is more movement than meets the eye, but allows interesting observations beyond core grammar (done by Gehrke & Grillo). Lastly, there aspect plays an obvious role in determining argument encoding in terms of accusative vs. ergative patterns – or so it has been argued. Phase Theory in both is structure (local domains) and properties (Merge and Agree operations) may lead to a different analysis of this long-standing problem in the grammatical description of languages. Müller, one of the invited speakers at InterPhases (although with a different presentation), argues exactly for that, as summarized below. This chapter forms a nice closure of the volume in incorporating the implicit thematic anchor ‘aspect’ on the one hand, but also in addressing technical issues Phase Theory brings up.

4. Explorations I will now summarize each chapter and put it in perspective with the above introduction to the topic (and, where appropriate, issues that were not mentioned), thereby relating it to other contributions, where relevant. The latter should already be reflected in the arrangement of the contributions to this volume, which is not alphabetical but intended to follow a certain thematic proximity. Instead of dividing the collection further into several distinct parts, it is my hope as the editor that the arrangement makes some sense to the reader, and where it does not, this section should clarify the editorial arrangement decision. Halldór Sigurðsson provides detailed “Remarks on Features”. Chomsky (2000 et seq.) develops an approach where uninterpretable features are deleted prior to (or under) Transfer. In his chapter, Sigurðsson pursues the ‘obvious’ alternative, namely, that such features are not present in syntax, but are instead a product of the interfaces. In particular, he argues that formal features belong to PF only. In support of his approach, Sigurðsson discusses nominal features in Icelandic and some other languages, arguing that gender, number, and case are not operative in syntax, but that they rather are morphological PF interpretations of syntactic correlations – thus assuming and arguing for a sharp distinction between discrete features in morphology and abstract relations in syntax (a radically disentangled morphology approach). Sigurðsson concludes that PF (qua Perceptible Form), including

Exploring features and arguments

11

the ‘sign form’ of sign languages, is not part of or a direct extension of Universal Grammar (i.e. Narrow Syntax + Transfer). Rather, syntax cooperates with motoric systems which provide audible and/or visible means to ‘broadcast’ or externalize complex symbols that are processable as incomplete expressions or translations of syntax. Language externalization thus makes use of conceptual/motoric systems that are not or at least not exclusively species-specific. By scrutinizing externalization and externalized linguistic systems, including complex morphological systems, linguistics can further our understanding of the relations and the borderland between the internal language of homo sapiens sapiens and some of the abilities of humans as motorically skilled animals. It seems to be fundamental property of the internal language faculty to relate to and interpret regular but largely arbitrary external patters (audible and/or visible or even only tactile) as meaningful reflections of the language faculty itself. Bartosz Wiland’s chapter “Feature Valuation by Sideward Movement” considers the status of sideward movement in the theory. Sideward movement – originally motivated by checking a θ-role and Case in the matrix clause – has been proposed to derive a number of constructions (like parasitic gaps or adjunct control). However, relying on a claim that θ-roles are checkable features that trigger movement has been since considerably problematic. Using arguments from asymmetric distribution of structurally specified and deficient DPs and its consequences for donkey anaphora, parasitic gaps, and across-the-board constructions, Wiland argues that inter-arboreal movement should be purified from its dependence on a θ-role. Instead, an element that moves sideward values the full set of phi- and Case-features on a matrix probe, with a θ-role as a trigger being redundant. Consequently, only elements that are present on a syntactic workbench and constitute complete goals for a probe in the matrix clause can undergo sideward movement. This approach can account for the fact that only phi-complete DPs produce parasitic gaps, while certain other θ-marked phrases never do so. Once movement to the θ-position is eliminated, sideward movement turns out to share all the relevant properties of (feature-driven) internal Merge. In the chapter “On Long Distance Agreement, the Spec-Head Configuration and Overt Agreement”, Ivan Ortega-Santos focuses on the status of Spec-Head relations within the phase-based theory. Within this framework, it is argued that minimal search conditions ban the existence of grammatically significant Spec-Head relations or checking relations under m-command (cf. Chomsky 2001 and subsequent work). There are, however, arguments to the contrary. Specifically, a cross-linguistic tendency has been noticed for A-moved elements to trigger agreement as opposed to in situ

12 Kleanthes K. Grohmann ones, a fact that calls for an explanation in any framework. Similar issues arise if theta-roles are features (cf. Bošković 1994; Bošković & Takahashi 1998; Hornstein 2003; and Lasnik 1995). In view of these and related facts, Ortega-Santos argues that a Multiple Spell-Out system (Uriagereka 1999, 2008), when combined with the phase-based system, allows for checking relations in the Spec-Head configuration. The resulting framework is shown to be compatible with minimal search conditions on probing in contrast to an approach in terms of m-command. Furthermore, some remaining issues concerning the cross-linguistic distribution of agreement morphology are argued to follow from the incremental nature of production following Barlow (1992). This would solve the problems that the phase-based framework faces. In view of these and related facts, Ortega-Santos argues that a Multiple Spell-Out system (Uriagereka 1999, 2008), when combined with the phase-based system, allows for checking relations in the Spec-Head configuration. The resulting framework is shown to be compatible with minimal search conditions on probing in contrast to an approach in terms of m-command. Furthermore, some remaining issues concerning the cross-linguistic distribution of agreement morphology are argued to follow from the incremental nature of production following Barlow (1992). This would solve the problems that the phase-based framework faces. In his contribution “Structural Case, Locality and Cyclicity”, Thomas McFadden considers technical issues that arise in the analysis of two morphological case phenomena. The first is the apparent existence of longdistance case-assignment. At least some DPs in embedded object positions seem to get case in a way that violates expectations of locality, even involving unbounded dependencies. The second issue arises with the assignment of structural accusative. This depends, under a number of recent analyses, on information about the case assigned to a higher DP. This would violate expectations of strict cyclicity, which has led Sigurðsson (2006) to propose that subjects first-merge lower than objects. McFadden argues that both of these phenomena can in fact be handled in ways consistent with standard notions of phase-based locality and cyclicity, once the empirical situation is correctly understood. The examples that have been analyzed in terms of long-distance case-assignment all turn out to involve the nominative case, which McFadden argues is appearing in its capacity as the default case, where normal forms of case-assignment have failed. Actual assignment of non-default case, on the other hand, is always phase-local. The accusative case can also be dealt with once we realize that it depends not on a specific case (nominative) on a higher DP, but on the presence of a higher DP with any structural case. Crucially, whether a DP will be assigned a

Exploring features and arguments

13

structural or a non-structural case, McFadden argues, can be determined on the basis of purely syntactic information that is available at a relatively early stage of the derivation. He then shows that the dependency relationship can be formulated in strictly cyclic terms if we assume that case is assigned at a later stage of the derivation, in the post-syntactic morphological component. Case-assignment can thus be handled in strictly local and cyclic terms, and without any unorthodox assumptions about phrase-structure. “PRO, pro and NP-Trace (Raising) Are Interpretations”, the chapter by Rita Manzini, connects to several current debates within minimalist theory. First, there is the proposal by Hornstein (1999) that control (or at least obligatory control) can be reduced to movement – and the ensuing debate with more conservative approaches (Landau 2003). A further debate (in fact not explicitly addressed by Manzini) involves raising, with scholars arguing that movement proceeds in one fell swoop from the embedded thematic position to the matrix sentence, either because there is no EPP (Epstein & Seely 2006) or because the satisfaction of the EPP yields Criterial Freezing (Rizzi & Shlonsky 2007). Finally, these various issues interact with the parameter opposing the best known Germanic and Romance languages to socalled Balkan languages displaying raising and control from finite sentences. This means that the embedded ‘subjunctive’ sentences of Balkan languages must somehow correspond to defective phases – a result which is generally achieved by invoking the defective nature of the embedded temporal specifications. This debate is essential to an understanding of Manzini’s proposal – yet the proposal itself is somewhat eccentric with respect to it, since it assumes a strongly representationalist view of grammar (Brody 2003). The other original aspect of this contribution is that it draws evidence from one of the least familiar Balkan languages, namely Albanian. According to Manzini, Albanian ‘subjunctive’ complements are bona fide tensed sentences, making notions of defectiveness difficult to apply. An important part of Manzini’s chapter is devoted to a close examination of the so-called subjunctive particle of Albanian, të, which she argues to be the same lexical item as the article të. On this basis, she suggests that its role in the economy of the sentence is to provide a lexicalization of the EPP argument. In Manzini’s terms, të acts as a lambda abstractor reopening the EPP argument, otherwise closed by the finite verb inflection. At the LF interface the open position that të introduce can then be bound by an antecedent – yielding raising or control, according to whether they do or do not share a single theta-role. “Movement of Arguments and Negative Feature”, the chapter by Masaaki Kamiya, investigates syntax and phonology interface. Presently, under-

14 Kleanthes K. Grohmann standing syntax and phonology interface is becoming more and more important (Deguchi & Kitagawa 2002; Ishihara 2002, 2004; Kitagawa & Fodor 2006; among many others). For example, Kitagawa & Fodor examine traditionally considered subjacency violation examples in Japanese wh-movement, and claim that unacceptable examples of wh-movement are not due to subjacency violation, but rather assigning wrong prosody to the relevant examples. Phonological information is relevant to interpretations as well as grammatical judgments. Kamiya’s contribution is to show that this is true with the interaction of a universal quantifier and negation in Japanese. First, Kamiya demonstrates that Japanese (either nominative or dative marked) subject must move to [Spec,TP] for EPP reasons, following Miyagawa (2001). This movement is motivated by syntactic side, and the relevant interpretations between a universal quantifier and negation are examined. In a normal reading, a universal quantifier takes scope over negation. However, stressing negative morpheme in the same sentence can produce a reading where negation takes scope over a universal quantifier. Kamiya assumes that negative feature which resides in NegP moves to the left periphery position in which negation takes scope over a universal quantifier. He tests to see if this is the case in other environments such as nominalization. Kamiya claims that as long as C or sentential left edge exits, stressing on negative morpheme produces negation-over-universal-quantifier reading is possible, but this is not true to nominalization (obviously, there is not C). In addition, negative affix, not negative morpheme which projects NegP, cannot undergo a feature movement. In the end, Japanese data show that negative feature can move to the left peripheral position for focus feature checking, and this is the case only to negative morpheme. The key point is that stressing causes this movement. Hence this chapter suggests that phonological information feeds meaning. At the same time, this paper questions if CP and DP are identical with respect to left periphery, which must be more thoroughly investigated in the future. In view of Chomsky’s (2000) proposal that there is a single cycle to syntactic derivations called a phase, Jonathan MacDonald observes in “Inner Aspect and Phases” that the aspectual interpretation of a predicate is conditioned by this single cycle. Since Verkuyl (1972) it has been known that an NP can affect the aspectual interpretation of the predicate. Tenny (1987) showed that only the internal argument NP can do so; in effect, she showed that syntax plays a role in aspectual interpretation. MacDonald notes that Tenny’s observations go beyond the aspectual effect of an NP. He argues for the existence of an aspectual projection between vP and VP, AspP, that has specific aspectual properties dependent on it (cf. Travis 1991,

Exploring features and arguments

15

2000). He provides evidence that for an element (not just an NP) to be able to contribute to the aspectual interpretation of a predicate, it must be below AspP, within a domain of aspectual interpretation. Interestingly, the aspectual effect of the external argument in fake reflexive constructions suggests that these elements cannot be calculated until the external argument merges in the derivation. Observations by Svenonius (2005) regarding idioms and phases together with observations by McGinnis (2002) and Glasbey (2007) suggest that the aspectual properties are not calculated after vP either. MacDonald concludes that, although only elements below AspP can contribute to the aspectual interpretation of a predicate, the calculation of these elements cannot take place until Transfer to the C-I system (via LF) at the vP phase. In their paper “How to Become Passive”, Berit Gehrke & Nino Grillo propose to shift the perspective on passive formation from the commonly held argument structure/DP perspective to an analysis based on event structure. They argue that this shift accounts for several syntactic and semantic properties of the passive construction, some of which remain unexplained under previous proposals. The general idea is implemented through movement of a consequent state sub-event to a discourse-related position at the edge of the VP. The tight relation between the availability of a consequent state and passivization is highlighted by examples showing that the possibility to passivize a predicate depends on its event structure in a crucial way. Evidence from word order in constructions involving secondary resultative predicates, floating quantifiers, ditransitives, and there-expletives strengthen the idea that more than the internal argument moves in passives. Finally, the analysis proposed makes new predictions with respect to impoverished syntactic representation in agrammatic Broca’s aphasia, which are borne out, and unifies the treatment of some of the most typical deficitarian comprehension patterns in this syndrome. Gereon Müller discusses “Ergativity, Accusativity, and the Order of Merge and Agree” in a novel way and presents the core aspects of a minimalist analysis of accusative vs. ergative patterns of argument encoding, be it via Case-marking or through agreement. The central observation is that indeterminacies may arise in the application of the two elementary operations Merge and Agree (Chomsky 2000, 2001), given that they both obey an Earliness requirement (in the sense of Pesetsky 1989). The central claim Müller puts forth is that a principled resolution of one such indeterminacy (on the vP cycle) in one or the other direction yields an accusative or ergative encoding pattern for arguments. Müller also shows how his analysis can be extended so as to cover various types of splits that have been noted

16 Kleanthes K. Grohmann in the literature (among them, tense/aspect-based splits and clause typebased splits).

Acknowledgements Editorial thanks go to all the contributors to this volume for their work, their patience, and their cooperation in the publication process, which was not always easy. In addition, I’m extremely grateful to Artemis Alexiadou, one of the series editors, and Ursula Kleinhenz from Mouton de Gruyter, who not only evaluated the original publication idea as very promising but also guided me along the way with useful advice and immediate responses.

References Abels, Klaus 2003 Successive cyclicity, anti-locality, and adposition stranding. Ph.D. dissertation, University of Connecticut, Storrs. Adger, David 2003 Core Syntax: A Minimalist Approach. Oxford: Oxford University Press. Barlow, Michael 1992 A Situated Theory of Agreement. London: Garland. Boeckx, Cedric 2008 Understanding Minimalist Syntax: Lessons from Locality in LongDistance Dependencies. Malden, MA: Blackwell. Boeckx, Cedric and Kleanthes K. Grohmann 2007 Putting phases in perspective. Syntax 10: 204–222. Bošković, Željko 1994 D-structure, theta criterion, and movement into theta positions. Linguistic Analysis 24: 247–286. 2007 Agree, phases, and intervention effects. Linguistic Analysis 33 (1–2): 54–96. Bošković, Željko and Daiko Takahashi 1998 Scrambling and last resort. Linguistic Inquiry 29: 347–366. Castillo, Juan Carlos, John Drury and Kleanthes K. Grohmann 1999 Merge over Move and the extended projection principle. University of Maryland Working Papers in Linguistics 8: 63–103. Chomsky, Noam 1973 Conditions on transformations. In A Festschrift for Morris Halle, Stephen R. Anderson and Paul Kiparsky (eds.), 232–286. New York: Holt, Reinhart & Winston.

Exploring features and arguments 1981

17

Lectures on Government and Binding: The Pisa Lectures. Dordrecht: Foris. [Subsequent editions published by Berlin /New York: Mouton de Gruyter.] 1986 Barriers. Cambridge, MA: MIT Press. 1993 A minimalist program for linguistic theory. In The View from Building 20: Essays in Linguistics in Honor of Sylvain Bromberger, Kenneth Hale and Samuel Jay Keyser (eds.), 1–52. Cambridge, MA: MIT Press. [Reprinted in Chomsky 1995: 167–217.] 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.), 89–155. Cambridge, MA: MIT Press. 2001 Derivation by phase. In Ken Hale: A Life in Language, Michael Kenstowicz (ed.), 1–52. Cambridge, MA: MIT Press. 2004 Beyond explanatory adequacy. In Structures and Beyond – The Cartography of Syntactic Structures, vol. 3, Adriana Belletti (ed.), 104 – 131. New York: Oxford University Press. 2008 On phases. In Foundational Issues in Linguistic Theory: Essays in Honor of Jean-Roger Vergnaud, Robert Freidin, Carlos P. Otero and Maria Luisa Zubizarreta (eds.), 133–166. Cambridge, MA: MIT Press. Deguchi, Masanori and Yoshihisa Kitagawa 2002 Prosody in syntactic analyses. Ms., Indiana University, Bloomington. den Dikken, Marcel 2007 Phase extension: Contours of a theory of the role of head movement in phrasal extraction. Theoretical Linguistics 33: 133–163. Epstein, Samuel David, Erich M. Groat, Ruriko Kawashima and Hisatsugu Kitahara 1998 A Derivational Approach to Syntactic Relations. New York: Oxford University Press. Gallego, Ángel J. 2007 phase theory and parametric variation. Ph.D. dissertation, Universitat Autònoma de Barcelona. Grohmann, Kleanthes K. 2003 Prolific Domains: On the Anti-Locality of Movement Dependencies. Amsterdam: John Benjamins. 2009a Phases and interfaces. In Grohmann (2009b), 1–22. Grohmann, Kleanthes K. (ed.) 2009b InterPhases: Phase-Theoretic Investigations of Linguistic Interfaces. Oxford: Oxford University Press. Grohmann, Kleanthes K. and Phoevos Panagiotidis (eds.) 2009 Selected Papers from the 2006 Cyprus Syntaxfest. Newcastle-uponTyne: Cambridge Scholars Publishing.

18 Kleanthes K. Grohmann Halle, Morris and Alec Marantz 1993 Distributed morphology and the pieces of inflection. In The View from Building 20: Essays in Linguistics in Honor of Sylvain Bromberger, Kenneth Hale and Samuel Jay Keyser (eds.), 111–176. Cambridge, MA: MIT Press. Hiraiwa, Ken 2005 Dimensions of Symmetry in Syntax: Agreement and Clausal Architecture. Ph.D. dissertation, Massachusetts Institute of Technology, Cambridge. Hornstein, Norbert 1999 Movement and control. Linguistic Inquiry 30: 69–96. 2001 Move! A Minimalist Theory of Construal. Malden, MA: Blackwell. 2003 On control. In Minimalist Syntax, Randall Hendrick (ed.), 6–81. Malden, MA: Blackwell. Hornstein, Norbert, Jairo Nunes and Kleanthes K. Grohmann 2005 Understanding Minimalism. Cambridge: Cambridge University Press. Ishihara, Shinichiro 2002 Invisible but audible Wh-scope marking: Wh-constructions and deaccenting in Japanese. WCCFL 21: Proceedings of the 21st West Coast Conference on Formal Linguistics: 180–193. 2004 Prosody-scope match and mismatch in Tokyo Japanese Wh-questions. Ms., Universität Potsdam. Kaye, Jonathan 1992 On the interaction of theories of lexical phonology and theories of phonological phenomena. In Phonologica 1988, Uli Dressler, Hans Luschützky, Oskar Pfeiffer and John Rennison (eds.), 141–155. Cambridge: Cambridge University Press. 1993 Derivations and interfaces. SOAS Working Papers in Linguistics and Phonetics 3: 90 –126. [Published 1995 in Frontiers of Phonology, Jacques Durand and Francis Katamba (eds.), 289–332. London: Longman.] Kitagawa, Yoshihisa and Janet Dean Fodor 2006 Prosodic influence on syntactic judgments. In Gradience in Grammar: Generative Perspectives, Gisbert Fanselow, Caroline Féry, Matthias Schlesewsky and Ralf Vogel (eds.), ch. 17. Oxford: Oxford University Press. Landau, Idan 2003 Movement out of control. Linguistic Inquiry 34: 471–498. Lasnik, Howard 1995 Last resort and attract F. Proceedings of FLSM 6: 62–81. [Reprinted 2003 in Minimalist Investigations in Linguistic Theory, Howard Lasnik (ed.), ch. 3. London: Routledge.]

Exploring features and arguments

19

Lasnik, Howard and Juan Uriagereka with Cedric Boeckx 2005 A Course in Minimalist Syntax: Foundations and Prospects. Malden, MA: Blackwell. Marušič, Franc Lanko 2005 On non-simultaneous phases. Ph.D. dissertation, Stony Brook University, NY. MacDonald, Jonathan E. 2006 The syntax of inner aspect. Ph.D. dissertation, Stony Brook University, NY. 2008 Domain of aspectual interpretation. Linguistic Inquiry 39: 128–147. McGinnis, Martha 2002 On the systematic aspect of idioms. Linguistic Inquiry 33: 665–672. Miyagawa, Shigeru 2001 The EPP, Scrambling, and Wh-in-situ. In Ken Hale: A Life in Language, Michael Kenstowicz (ed.), 293–338. Cambridge, MA: MIT Press. Nissenbaum, Jonathan W. 2000 Investigation of covert phrase movement. Ph.D. dissertation, MIT, Cambridge, MA. Pesetsky, David 1989 Language-Particular Processes and the Earliness Principle. Ms., MIT, Cambridge, MA. Radford, Andrew 2004 Minimalist Syntax: Exploring the Structure of English. Cambridge: Cambridge University Press. Richards, Marc 2004 Object shift and scrambling in North and West Germanic: A case study in symmetrical syntax. Ph.D. dissertation, University of Cambridge. van Riemsdijk, Henk 1978 A Case Study in Syntactic Markedness: The Binding Nature of Prepositional Phrases. Dordrecht: Foris. Rizzi, Luigi and Ur Shlonsky 2007 Strategies of subject extraction. In Interfaces + Recursion = Language? Chomsky’s Minimalism and the View from Syntax-Semantics, Uli Sauerland and Hans-Martin Gärtner (eds.), 115 –160. Berlin /New York: Mouton de Gruyter. Scheer, Tobias 2008 Intermodular argumentation: Piece-driven phase and one single phonology. Paper presented at WCCFL 27, University of California, Los Angeles (16 –18 May 2008). Sigurðsson, Halldór Ármann 2006 The nominative puzzle and the low nominative hypothesis. Linguistic Inquiry 37: 289 –308.

20 Kleanthes K. Grohmann Svenonius, Peter 2005 Extending the extension condition to discontinuous idioms. Linguistic Variation Yearbook 5: 227–263. Uriagereka, Juan 1997 Multiple spell-out. Groninger Arbeiten zur germanistischen Linguistik 40: 109 –135. 1999 Multiple spell-out. In Working Minimalism, Samuel David Epstein and Norbert Hornstein (eds.), 251–282. Cambridge, MA: MIT Press. 2008 Syntactic Anchors: On Semantic Structuring. Cambridge: Cambridge University Press. Wilder, Chris and Hans-Martin Gärtner 1997 Introduction. In The Role of Economy Principles in Linguistic Theory, Chris Wilder, Hans-Martin Gärtner and Manfred Bierwisch (eds.), 1–35. Berlin: Akademie Verlag.

Remarks on features Halldór Ármann Sigurðsson

This paper pursues the idea that uninterpretable features are not present in syntax, but are instead a product of the interfaces. In particular, it argues that formal feature values belong to PF only, i.e., that they are not syntactic objects but PF ‘translations’ of more abstract syntactic structures and correlations. It follows that case is nonexistent in syntax and it also follows that agreement is a PF copying process, differing radically from abstract, syntactic Agree. Accordingly, much of the ‘labor’ of traditional syntax happens in PF and is thus invisible to the semantic interface, SF, that is, the computation proceeds on the PF side after transfer.

1. Introduction If the syntactic computation proceeds in a single cycle (Chomsky 2000 et seq.), it must be interpretable to both the interfaces, that is, semantic form and perceptible form, SF and PF, for short, where perceptible form refers to PF in a broad sense, including the ‘sign form’ of sign languages. From this general interpretability or legibility condition, it follows that syntax cannot produce any information that is visible but uninterpretable to the interfaces. By necessity, however, linguistic objects contain features that are interpretable to only one of the interfaces, like +HUMAN and [+labial]. Chomsky’s solution to this Interpretablity Puzzle is basically to have uninterpretable features removed or eliminated prior to or under the operation transfer, that hands the derivation over to the interfaces. In this work, I will pursue the ‘obvious’ alternative, namely, that uninterpretable features are not present in syntax, but are instead a product of the interfaces or of their interplay with language external motoric and conceptual subsystems. Such features are functional in a broad sense, but superfluous from a narrow syntactic perspective. Syntax itself contains features that get interpreted or valued through matching, but it contains no features that remain uninterpreted and thus need to be deleted.1 1

See in particular Sigurðsson (2006b), where it is argued that syntax operates with abstract features and roots, ROOT99, etc., that do not get any phonological feature

22 Halldór Ármann Sigurðsson It is a truism that individual languages systematically express categories like Tense, Person, Number, etc., by overt elements. Regardless of whether the elements in question are suprasegmental patterns, morphemes, words, phrases or whole clauses, they are parts of the audible or visible form of language and thus products of PF in the broad sense. It is commonly assumed that these PF products somehow reflect elements of a more abstract and general system traditionally referred to as syntax, and I adopt this traditional view. The central question that arises is how exactly these PF elements relate to the putative underlying syntactic objects. I believe it is fair to say that the most general assumption is that syntactic features are interpreted or translated in a fairly straightforward manner by morphological categories, such that for instance English present and past tense are direct exponents of syntactic Tense, say, simply the two morphological values of the T head of the clausal TP projection that are lexically and parametrically available in English, standing in a direct two-to-one relationship with T. This conception is not often explicitly stated or formulated, and different ideas abound in the literature, but, I believe it is nonetheless fair to say it is the prevailing conception in many or even most generative approaches. I will pursue a different approach here, where the so-called formal features belong to broad PF only. On this view, there is for instance no syntactic masculine feature, no syntactic nominative case feature, and so on. Rather, morphological features of this sort are PF-translations of abstract syntactic relations.2 It follows from the present approach that morphological agreement cannot be a narrowly syntactic phenomenon. Another important consequence is that the standard economy argument (see, e.g., Chomsky 1995) loses much of its force, that is, there is no simple mapping from syntax to morphology and hence also no general economy in PF translations of syntax. Syntax as

2

values (uninterpretable to SF) until on the PF side (much as assumed by proponents of Distributed Morphology (Halle & Marantz 1993, etc.)). A reviewer raises the question of what drives syntactic movement if formal features do not belong to syntax. There are two relevant answers to the question. First, the matching and valuing of syntactic features drives syntactic movement (under the condition of weak or ‘inactive’ intervention). Second, many instances of movement that have traditionally been taken to be syntactic, arguably or even evidently take place in PF. Much work remains to be done on these issues, but see Sigurðsson (2004 a,b, c, 2006 b,c) for some discussion. PF is evidently layered, with several sub-interfaces, including, roughly, Sign Formation, morphophonology, phonology and phonetics (at least in oral languages, see Sigurðsson 2006b: 204).

Remarks on features 23

such is sensitive to economy and so is PF as such, but PF does not heed or preserve syntactic economy, as it were. In view of the variation observed in the languages of the world, it might seem obvious that their common denominator, Universal Grammar or Narrow Syntax, cannot possibly operate with complex entities that are physically present in individual languages, like for instance agreement features in oral languages or eyebrow markers in sign languages. The state of the art is however such that the ‘obvious’ conclusion that Narrow Syntax must be ‘atomic’ and therefore cannot operate with entities overtly expressed in the grammars of individual languages is everything but obvious to most linguists. It must be argued for. In support of the approach pursued here, I will present and discuss data on gender, number and case from Icelandic and some other languages. Many of these data are well-known and simple, but have nonetheless been neglected in mainstream generative approaches. It is high time that these facts be taken seriously and accounted for in some coherent manner.

2. Gender: Some observations Gender is a mixed category. Many languages have relatively transparent gender systems, based on central categories like MALE, FEMALE, ANIMATE, HUMAN. There are however also some languages that base their gender systems on less expected categories, like the NON-FLESH FOOD category in Dyirbal or the LIQUID category in Fula or Fulfulde (Corbett 1991: 30 –31).3 More importantly for our purposes, there are also many languages that have a largely arbitrary gender system. I will illustrate the pervasive arbitrariness of many gender systems with examples from mainly Icelandic and German. Both Icelandic and German have the common three gender system type, consisting of masculine, feminine and neuter. As in most gender languages, many nouns denoting people and domestic animals have natural gender. In a three gender system this will be masculine for adult male beings, feminine for adult female beings and often neuter for young animals, which are thereby treated as not yet sex-differentiable. This is illustrated in (1) for some Icelandic nouns: 3

In Corbett’s approach, these are gender categories in the languages in question since they enter into agreement relations (including pronominal reference), as opposed to classifiers in classifier languages. I adopt this understanding here for convenience, but it is not of any importance for what I have to say.

24 Halldór Ármann Sigurðsson (1)

Masculine

Feminine

Neuter

maður strákur hestur hrútur tarfur etc.

kona stelpa meri ær kýr

barn

‘man’ ‘boy’ ‘horse’ ‘ram’ ‘bull’

‘woman’ ‘girl’ ‘mare’ ‘sheep’ ‘cow’

‘child’

folald ‘foal’ lamb ‘lamb’

Most nouns, however, do not have natural gender. Consider (2): (2)

Masculine

Feminine

Neuter

bátur stóll kafli kofi fótur kappi hlébarði þorskur svanur máni etc.

skúta hilla bók höll hönd hetja gaupa ýsa álft tungl

skip ‘ship’ borð ‘table’ blað ‘(news)paper’ hús ‘house’ læri ‘thigh’ poppgoð ‘pop idol’ ljón ‘lion’ hrognkelsi ‘lumpfish’

‘boat’ ‘chair’ ‘chapter’ ‘hut’ ‘foot’ ‘champion’ ‘leopard’ ‘cod’ ‘swan’ ‘moon’

‘yacht’ ‘shelf’ ‘book’ ‘palace’ ‘hand’ ‘hero’ ‘lynx’ ‘haddock’ ‘swan’ ‘moon’

For these and most other nouns, gender is evidently a plain classificational feature with no semantic import.4 That is, it is like a phonological feature in making a distinction between items without itself adding or reflecting any semantics.5 There are several further kinds of clear evidence that the grammatical gender of most nouns does not belong to or affect their semantics. One simple type of evidence is that one and the same noun may have different genders (see Kvaran 2005: 173):

4

5

In the narrow sense. Gender may have psychological effects, an interesting issue that is however not relevant in the present context. Making overt distinctions of this sort is not only a derivational cost but also a communicative gain, much as it is a gain to phonologically distinguish between e.g. cable and table. See also section 5 on the disambiguating effects of agreement.

Remarks on features 25

(3)

sykur skúr bjúga fress etc.

‘sugar’: ‘shower of rain’: ‘sausage’: ‘tomcat’:

M or N F or M N or F N or M

Another indication of the semantic emptiness of formal gender is that, exceptionally, one and the same noun may have different genders in the singular and plural, a phenomenon known as inquorate genders (Corbett 1991: 170): (4)

foreldri foreldrar fótur fætur

‘parent’, sg.: ‘parents’, pl.: ‘foot’, sg.: ‘feet’, pl.:

N M M F (for some

speakers)

Yet another indication of formal gender’s semantic vacuousness is the fact that even closely related languages, with the same gender system type, show numerous gender contrasts. Consider the German-Icelandic constrasts in (5): (5)

German ‘horse’ ‘lion’ ‘lynx’ ‘sea’ ‘cloud’ ‘autumn’ ‘summer’ ‘table’ ‘shelf’ ‘train’ ‘car’ ‘telephone’ ‘bank’ ‘boat’ ‘book’ etc.

Pferd Löwe Luchs Meer Wolke Herbst Sommer Tisch Regal Zug Auto Telefon Bank Boot Buch

Icelandic N M M N F M M M N M N N F N N

hestur ljón gaupa sjór ský haust sumar borð hilla lest bíll sími banki bátur bók

M N F M N N N N F F M M M M F

(poetic: mar, M)

Gender mismatches of this sort between these two closely related languages are strikingly pervasive, so this list could easily be made much longer.

26 Halldór Ármann Sigurðsson The grammatical gender of most nouns in formal gender languages like Icelandic and German is clearly invisible to the semantic interface. Two interpretations of this fact are conceivable. First, the gender feature might be present in syntax but eliminated under transfer, as are agreeing features in most minimalist approaches. Second, the gender feature might not be present in syntax, in which case it would have to be added after transfer to PF. It is this second position that I am taking here.6 The prevailing assumption, I believe, is that formal gender does belong to syntax (cf. the notion ‘syntactic gender agreement’ in Corbett 1991, see also the approach in e.g. Kayne 2005). However, it is not optimal engineering to first provide all nouns with some specific gender feature and then to delete the same feature of most nouns under transfer to SF. Notice also that this putative feature deletion does not involve agreement, of course. Moreover, if grammar deletes the feminine feature of German Bank ‘bank’ and the masculine feature of Icelandic banki under transfer to SF, it is unclear how it would avoid deleting the gender of natural gender nouns like German Mann and Frau and Icelandic maður ‘man’ and kona ‘woman’. We have to sharply distinguish between the semantic FEMALE/MALE or HE/SHE features and grammatical or formal gender features. Most nouns that have semantic gender, either HE or SHE, have natural formal gender, but there are many exceptions, higlighting that semantic and formal gender are distinct features even in animate nouns. This is illustrated with only a few Icelandic examples in (6):

6

There might seem to be a third alternative here, such that grammatical gender is just a formal classifier, unrelated to or at least independent of semantic gender. On this approach, one might want to say that the formal gender classifier of all nouns is deleted under transfer to SF, whereas their semantic gender is not, and, conversely, that semantic gender is deleted under transfer to PF whereas formal gender is not. As we shall see, however, semantic gender is PF visible in certain cases. Evidently, also, it is not the case that formal gender is generally independent of semantic gender. Rather, it is a ‘PF-translation’ of semantic gender, showing sloppiness that is typical of the overt, socially conventionalized PF side of language. Saying that grammatical features come in pairs in syntax, consisting of a semantic and a formal member, is tantamount to saying that the formmeaning relationship of language must be taken as an unexplainable axiom, which, in turn, raises the question of why the form member of such pairs should vary across languages. In effect, this position would take us back to preChomskyan structuralism, with Universal Grammar as an impossible subject of inquiry.

Remarks on features 27

(6)

Masculine

Feminine

Neuter

kvenmaður ‘woman’

karlugla ‘fool of a man’

karlmenni ‘(strong) man’

kvenskörungur ‘powerful woman’

mannfýla ‘bastard of a man’

fljóð ‘girl’ (poetic)

stelpukjáni ‘fool of a girl’

karlpersóna ‘male person’

naut ‘bull’

Thus, saying that formal gender is a syntactic feature that is generally SF uninterpretable except when it combines with another feature, like ANIMATE or HUMAN, does not help. On the contrary, that approach would make numerous wrong predictions, not only for nouns like the ones in (6). Thus, most Icelandic nouns denoting professions and nationalities are masculine, regardless of the sex of the person referred to, an issue I will return to shortly. Corbett (1991) refers to nouns like the ones in (6) as HYBRID NOUNS. Many such nouns are compounds or suffixed. Derivational morphology usually overrides semantic gender features, as seen in, e.g., the famous German neuter noun for ‘girl’, Mädchen, where -chen is a derivational morpheme deciding the formal, neuter gender of the derived noun, irrespective of the noun’s semantic gender. Formal gender features are normally visible through agreement but they are often only indirectly visible on the noun itself, through its effects on the selection of case/number endings (that is, through its effects on inflectional classification). Thus, feminine ausa ‘scoop, ladle’ is ausu in the oblique singular cases and ausur in nominative and accusative plural, whereas neuter auga ‘eye’ is the same in the other singular cases and augu in nominative and accusative plural: (7)

F ‘scoop’

Sg.Nom Sg.Acc/ Dat / Gen

ausa ausu

‘eye’ auga auga

Pl.Nom/Acc Pl.Dat Pl.Gen

ausur ausum ausna

augu augum augna

N

28 Halldór Ármann Sigurðsson For a handful of kinship nouns, though, no such effects of gender on case/ number endings are observed, that is, bróðir, dóttir, faðir, móðir, systir ‘brother, daughter, father, mother, sister’, which all inflect the same: 7 (8)

‘mother’

‘brother’

Sg.Nom móðir Sg.Acc/ Dat / Gen móður

bróðir bróður

Pl.Nom/Acc Pl.Dat Pl.Gen

bræður bræðrum bræðra

mæður mæðrum mæðra

Thus, these few kinship terms seem to be exceptional in having only a semantic gender feature and no formal gender feature. An alternative way of stating this is to say that these nouns have a zero formal gender feature that is interpreted as formal masculine vs. feminine by agreement morphology. See further below. Most nouns in a formal gender language like Icelandic evidently select and incorporate a formal gender feature (+/– M, +/– F), even when they have semantic gender. As we shall see shortly, this has the effect that semantic gender becomes invisible to PF when a noun has a specified formal gender feature. This incorporation of formal gender may be thought of as a ‘word formation’ process, as it were (cf. Josefsson 1998), but, rather than taking place in Narrow Syntax, it takes place after transfer to PF. That is, lexicalization, combining phonetic material with formal features, is post-syntactic, syntax in contrast operating with only abstract features and roots. Thus, morphology is all post-syntactic (‘radically disentangled’ from syntax).8 We can distinguish between four classes of Icelandic nouns in terms of semantic and formal gender, as sketched in (9): (9)

7

8

A. Nouns that have both semantic and formal gender A1. Natural gender nouns A2. Hybrid nouns, with contrasting semantic and formal genders B. Nouns that have only formal gender C. A handful of kinship terms that have semantic gender and a zero formal gender

As opposed to sonur ‘son’, mamma ‘mom’, pabbi ‘dad’ and many other kinship terms. See further Sigurðsson (2006b).

Remarks on features 29

This is illustrated in (10): (10)

Semantic gender Formal gender A1. A2. B. C.

kona ‘woman’ kvenmaður ‘woman’ ausa ‘scoop’ móðir ‘mother’

SHE SHE

Ø

F M F

SHE

Ø

B-type nouns, with no semantic gender, are by far the most numerous ones, but the natural gender type in A1 is also common, of course. In the rare (Icelandic) case of a zero formal gender feature, agreement morphology interprets it as masculine vs. feminine in accordance with the semantic gender of the noun in question. Thus, móðir ‘mother’ triggers feminine agreement whereas bróðir ‘brother’ triggers masculine agreement: (11) Móðir mín er gáfuð og bróðir minn er líka gáfaður. mother my.F is smart.F and brother my.M is also smart.M ‘My mother is smart and my brother is also smart.’ Similarly, first and second person pronouns have zero formal gender, also triggering this kind of ‘natural agreement’: (12) a. (María sagði:) (Mary said:) b. (Haraldur sagði:) (Harold said:)

Ég I Ég I

er am er am

gáfuð. smart.F gáfaður. smart.M

In formal gender languages, a masculine hybrid noun that is used to refer to a woman may either be referred to as he or she in discourse. A well-known and much cited example is masculine vrač, the Russian noun meaning ‘doctor’, which can be referred to in discourse as either on ‘he’ or ona ‘she’ when denoting a female doctor (Corbett 1991: 232). The same applies to Icelandic læknir ‘doctor’. Similarly, the Icelandic masculine noun forseti ‘president’ is usually referred to with hann ‘he’, but when Iceland’s president was a woman, Vigdís Finnbogadóttir, hún ‘she’ could also be used: (13) a. Núna gengur forsetinn í salinn. Hún er í bláum kjól. now walks president.the in hall.the. she is in blue dress ‘The president now enters the hall. She is wearing a blue dress.’ b. Núna gengur forsetinn í salinn. Hann er í bláum kjól. now walks president.the in hall.the. he is in blue dress

30 Halldór Ármann Sigurðsson There seems to be a rather general preference for the formal gender to control pronominal reference in Icelandic, but in this particular context of a typical female behavior, the feminine pronoun is in fact much preferred. Some languages, including German and most Romance varieties, make numerous derivational sex-distinctions in nouns denoting professions and nationalities. Consider (14): (14) ‘doctor’ ‘lawyer’ ‘baron(ess)’ ‘Italian’ ‘Icelander’ ‘Englishman’ etc.

Italian

German

Icelandic

MALE

FEMALE

MALE

FEMALE

MALE

FEMALE

medico avocato barone italiano islandese inglese

medica avocatessa baronessa italiana = islandese = inglese

Arzt Anwalt Baron Italiener Isländer Engländer

Ärztin Anwältin Baronin Italienerin Isländerin Engländerin

læknir = læknir lögmaður = lögmaður barón barónessa Ítali = Ítali Íslendingur = Íslendingur Englendingur = Englendingur

Grammaticized sex-marking of this sort is a very general trait of German, somewhat less so of Italian and largely absent from Icelandic. One way of analysing this difference is to say that almost all Icelandic nouns and many Italian nouns denoting professions and nationalities of women have a silent feminine marker, call it SHE, corresponding to the overt -in marker in German. If so, one can say that the variation betwen masculine and feminine pronominal reference in (13) reflects this duality, as sketched in (15) (where I do not show the definite article -nn): (15) a. … /forseti+Ø.M/… Hann / *Hún… b. … /forseti+SHE.M/… Hann / Hún… At some level of abstraction, this is presumably the correct analysis. The silent marker SHE can be taken to be the semantic gender feature of any noun that has female semantics, for instance the hybrid noun kvenmaður ‘woman’ in (10)A2 above (literally ‘woman-person’): (16) a. * /kvenmaður+Ø.M/ b. /kvenmaður+SHE.M/ The difference between forseti and kvenmaður, then, is that the silent SHE marker is obligatory in kvenmaður but only optional in forseti, i.e., there are in a sense two forseti nouns but only one kvenmaður noun. In the same

Remarks on features 31

fashion, Italian nouns like islandese and inglese have an optional SHE marker, but at the same time they differ from Icelandic hybrid nouns in only having a zero formal gender feature: (17) a. /islandese+Ø.Ø/ b. /islandese+SHE.Ø/ Accordingly, agreement morphology ‘sees’ the semantic gender feature in Italian, whereas (clause-internal) agreement is controlled by the specified formal gender feature in Icelandic: (18) a. María è una islandese simpatica. Mary is an Icelander sympathetic ‘Mary is a sympathetic Icelander.’ b. Haraldur è un islandese simpatico. (19) a. María er viðkunnanlegur/*viðkunnanleg Íslendingur. Mary is sympathetic.M/*F Icelander.M b. Haraldur er viðkunnanlegur Íslendingur. That is, in the presence of a specified formal gender, as in masculine Íslendingur, semantic gender becomes PF invisible (clause-internally). As we saw above, however, Icelandic first and second person pronouns and a handful of kinship terms are like Italian nouns of the islandese class in not having a specified formal gender feature, cf. the agreement facts in (11) and (12) above. Thus, both languages operate with both semantic and formal gender, that is, they both have nominals that are specified for only semantic gender (HE or SHE), nominals that are specified for only formal gender and nominals that are specified for formal as well as semantic gender. Many languages, including for example Finnish, Hungarian, languages of the Philippines and many languages of the Americas, lack gender as an overt category, whereas Fula or Fulfulde (Niger-Congo) has been claimed to have “about twenty genders” (Corbett 1991: 191), triggering different agreement patterns (thus being grammaticized in a way that differs from classifiers in classifier languages, which do not trigger agreement). Reasonably, however, all languages have access to the semantic categories, on which gender systems are based, including MALE, FEMALE, ANIMATE and HUMAN. Even though Icelandic and Finnish have not grammaticized the category of INSECTS, as has the Rikvani dialect of (the North Caucasian) Andi or the category of LIQUIDS, as has Fula/Fulfulde (Corbett 1991: 30–31), it does not

32 Halldór Ármann Sigurðsson seem to make much sense to assume that languages differ syntactically with respect to these categories. Rather, genders are lexical classification features, serving the general ‘purpose’ of reducing ambiguity on the PF side of language. If so, gender agreement must also belong to PF rather than to Narrow Syntax, an issue I will return to.9 As in many other formal gender languages (see Corbett 1991: 51ff.), gender selection in Icelandic is frequently affected by phonological factors, most roots in -óC or -íC, for instance, combining with feminine gender, most nouns with the -un suffix also combining with feminine, and so on.10 That is, gender selection is affected by phonologial factors, and it shows ‘combinatory tendencies’ that resemble tendencies commonly seen in phonotactics. Moreover, gender feeds phonological processes like the so-called u-mutation, shifting [a] to [œ] or [Y] under certain conditions (see Rögnvaldsson 1981 for a detailed account): (20) a. [a] → [œ] in (mostly) stressed syllables b. [a] → [Y] elsewhere The relevant conditions are somewhat varying, but, strikingly, there are cases where they regularly and productively involve gender information, as illustrated in (21): (21) a. lat- [la:t(h)-] ‘lazy’

→ löt [lœ:t(h)]

in F.SG.NOM & N.PL .NOM /ACC

b. byrjað- [pIrjað-] → byrjuð [pIrjYð] in F.SG.NOM & N.PL.NOM /ACC ‘begun’ Facts of this sort do not provide an unmabiguous argument against gender being syntactic, but they show, at least, that gender is visible to phonological processes. Before proceeding to other features, let us briefly summarize our central findings for gender. A noun may either have or not have semantic gender. 9

10

On the antecedent movement approach to pronominal binding pursued in Kayne (2002), gender agreement can be analyzed as clause-bounded across the board (which it cannot in other approaches, as far as I can see). All other problems remain, though, and Kayne’s approach also has its own, rather serious internal problems, so I will not discuss this here. Bók ‘book’, rót ‘root’, vík ‘small bay’, hlíð ‘slope’, verslun ‘shop’, hönnun ‘design’, and so on.

Remarks on features 33

Limiting ourselves to only FEMALE vs. MALE or HE vs. SHE, leaving, ANIMATE, HUMAN, and so on out of the discussion, this gives us the three possibilities in (22): (22) a. HE b. SHE c. – HE, – SHE Now, if morphology would heed or preserve syntactic economy, this would be all that is needed in three gender languages like Icelandic and German. As we have seen, however, that is very far from being the case. A fact that has not been generally appreciated is that morphology reinterprets these feature values or ‘translates’ them, as it were, into its own terms. Thus, as we have seen, each of the feature settings in (22) may combine with all three formal genders. In addition, a few elements, including the first and second person pronouns, have a zero or an unspecified formal gender feature, their semantic gender thus being interpreted by agreement morphology as if it were formal. The relevant facts are summarized in (23): (23)

Semantic G a. HE

b. SHE

c. – HE, – SHE

Formal G a1. a2. a3. a4. b1. b2. b3. b4. b1. b2. b3.

M F N

Ø M F N

Ø M F N

hrútur mannfýla naut ég ‘I’; bróðir kvenmaður kona fljóð ég ‘I’; móðir bátur skúta skip

‘ram’ ‘bastard of a man’ ‘bull’ ‘brother’ ‘woman’ ‘woman’ ‘girl’ (poetic) ‘mother’ ‘boat’ ‘yacht’ ‘ship’

For formal gender, then, it is evident that morphology is not just a reproduction or preservation of syntax/semantics, as it were, but an autonomous system, translating syntax into its own terms. It is of some interest here to notice that even closely related languages may have very different gender systems. Swedish, for instance, generally makes a two-way distinction in its agreement system, between common and neuter gender, but a four-way distinction in its singular pronominal system, as illustrated in (24) in a somewhat simplified fashion:

34 Halldór Ármann Sigurðsson (24) han ‘he’: hon ‘she’: den ‘it’: det ‘it’:

MALE HUMANS (plus some other small categories) FEMALE HUMANS (plus some other small categories) most non-neuter NON-HUMAN nouns

neuter nouns

Thus, the distinction between +/-HUMAN has been grammaticized in the Swedish pronominal gender system, whereas it has not in the Icelandic system. Generally, gender is a much more ‘natural’ or transparent category in Swedish than it is in Icelandic and German.11 Semantic features like HUMAN, ANIMATE, FEMALE, LIQUID, INSECT, etc., map differently onto different grammars, some languages being grammatically silent (but typically not lexically silent) about certain features that other languages grammaticize. Also, even when two closely related languages opt for grammaticizing two related features such as HUMAN and FEMALE they typically map these features differently onto formal features like masculine and feminine. Such mapping differences are responsible for much of the typological or parametric variation between languages. Plausibly, there is a universal cartography of syntactico-semantic features like HUMAN, ANIMATE, FEMALE, etc., that are the building blocks of nouns, such that a formal feature like masculine is a PF translation or interpretation of a part of a complex feature tree structure, expressing different parts of the tree structure in different languages, often arbitrarily so but sometimes in a more predictable or transparent manner. 3. A note on number Number is usually a ‘better behaved’ category than gender, as it were, that is, most semantically plural nouns are also morphologically plural. Nevertheless, there are also numerous mismatches, illustrating that we need to distinguish between semantic and formal number. Consider the following Icelandic pluralia tantum and singularia tantum nouns, that is to say, ‘plural only’ and ‘singular only’ nouns:

11

The Swedish gender system has other interesting properties, alien to Icelandic, that I will however not address here (but see Källström 1993; Teleman et al. 1999; Josefsson 2006).

Remarks on features 35

(25) Plural only hjón jól páskar lög laun buxur innantökur aftökur

Singular only ‘married couple’ ‘Christmas’ ‘easter’ ‘law’ ‘salary’ ‘trousers’ ‘internal pain(s)’ ‘extremely bad weather’

fólk stóð ös fyndni hugrekki barátta kjöt smjör

‘people’, ‘pack of horses’, ‘crowd, throng’ ‘humor’ ‘courage’ ‘struggle’ ‘meat’ ‘butter’

As for many related languages, this list could be made much longer (cf. Thráinsson 1983, Kvaran 2005: 174 –175; Corbett 2000), but these examples suffice to illustrate the point. It is evident that the plural feature of, e.g., jól ‘Christmas’ or the singular feature of, e.g., fólk ‘people’ are invisible to the semantic interface, that is, these nouns have ‘reverse’ number semantics, as it were. Nonetheless, it is the formal number feature that controls all kinds of agreement, regardless of the actual number of Christmases or people one may be talking about. This is illustrated in (26): 12 (26) a. Lengstu jólin voru skemmtilegust. longest.PL Christmas.the.PL were.PL most-fun.PL ‘The longest Christmas was the most fun.’ b. Gamla fólkið var skemmtilegast. old.SG people.the.SG was.SG most-fun.SG ‘The old people were the most fun.’ Thus, morphology sees formal number while the semantic interface does not.13

12

13

Any other forms would be ungrammatical here. Pronominal reference across clause boundaries is normally also controlled by the formal number, although plural reference to singularia tantum nouns seems to be acceptable to some speakers. However, even morphology may ‘turn a blind eye’ to formal number. Thus, ‘corporate’, formally singular nouns may trigger plural agreement in some languages, including varieties of English (The government are not really fighting crime, etc., cf. Corbett 2000: 187 ff.).

36 Halldór Ármann Sigurðsson Semantic number is a complex phenomenon, not only involving number or plurality as such but also categories like individuation, countability, distributivity, collectivity and definiteness (see Corbett 2000). In relatively simple number systems like the English or the Icelandic one, the grammatical singular / plural distinction most commonly relates to the basic notion of ONE IN NUMBER, 1, and MORE THAN ONE IN NUMBER, >1. Referring to this notion as semantic number for Icelandic, thereby allowing us the simplification of disregarding all other semantic categories, we get the six mappings in (27), exhausting the logical possibilities: (27)

Semantic N

Formal N

a.

1

SG

hestur

‘horse’

b.

>1

SG

fólk

‘people’

c.

Ø

SG

hugrekki

‘courage’

d.

1

PL

jól buxur

‘Christmas’ ‘trousers’

e.

>1

PL

hestar jól buxur

‘horses’ ‘Christmases’ ‘trousers’

f.

Ø

PL

innantökur ‘internal pain(s)’

Again, it is immediately clear that morphology is not a one-to-one reproduction or preservation of syntax/semantics, but an autonomous system, translating syntax into its own terms. The features ONE and MORE THAN ONE are clearly interpretable to the semantic interface, and they are often straigtforwardly translated into the language of morphology. For morphology, however, the most central interest is not to serve or satisfy syntax or processing (although it usually does that), but to see to it that any noun have some value for formal number, sometimes regardless of the noun’s number semantics or even in direct contrast with it. Anomalies of this sort often have a natural historical explanation, but from the point of view of language aquisition that is of course immaterial. It is evident that morphology is selfish, serving its own narrow interests in addition to the interests of processing. The reason why language is organized like this is arguably that “humans are endowed with innate syntactic elements and structures that are independent of whether or how they are expressed” (Sigurðsson 2004a: 251), that is, processing is to a large extent internally given. A certain amount of mismatch between the form of a message and its semantic inter-

Remarks on features 37

pretation is a built in property of natural languages, as opposed to artificial languages.14 An alternative take on the semantics-morphology incongruity in (27) would be to say that formal number is a syntactic feature that may however arbitrarily or chaotically opt for being invisible to the semantic interface or even for being reversely interpreted. If so, it is pointless to discuss the organization of language, so I will not consider this view here.

4. A note on case: The No Case Generalization Icelandic attributive and predicative agreement of adjectives, quantifiers and past participles involves not only gender and number but also case, as illustrated in (28); the suffixed definite article also agrees in the same way, but, for simplicity, I do not indicate this: (28) a. Báðir mennirnir höfðu verið kosnir. both men.the had been elected N.M.PL N.M.PL 3PL N.M.PL b. Ég taldi báða mennina hafa verið kosna. I believed both men.the have been elected A.M.PL A.M.PL INF A.M.PL If gender and number are morphological PF translations of abstract syntactic objects or structures, then case in agreement configurations of this sort must also be post-syntactic, an issue I will return to. In a simple morphophonological sense, this is trivially obvious, but I claim that case is radically non-syntactic in the sense that there are no syntacic features like +/–NOM, nor are there any ‘deep’ case features.15 If case was a property or a feature of Narrow Syntax, we would expect closely related languages to be more or less identical with respect to casemarking, but that expectation is of course not borne out. This is abundantly evident for many case-rich languages, both synchronically and diachronically, e.g., Icelandic and German, Russian and Polish, Latin, Romanian and 14

15

Plausibly, this is the basic reason why language is subject to constant and irresistible change. See also McFadden (2004) and Platzack (2006). I’m here abandoning the assumption in Sigurðsson (2003) that language has some deep cases in the Fillmorian sense.

38 Halldór Ármann Sigurðsson Italian, to mention only a few well-known Indo-European languages. Even closely related case-poor languages like Danish and Swedish show some striking differences with respect to case-marking, as in (29): (29) a. Det er os. it is us.ACC ‘It is us.’ b. Det är vi. it is we.NOM ‘It is us.’

Danish

Swedish

Arguably, there is no Narrow Syntax difference between Danish Det er os and its Swedish translation Det är vi. Rather, it seems that exactly the same syntactic structure gets different interpretations or translations in PF morphology. In recent work on variation of this sort across the Germanic languages (Sigurðsson 2006a), I come to the conclusion that the so-called ‘structural’ cases (regular nominatives and accusatives in accusative systems) must be understood as PF features serving the purpose to overtly distinguish either between the arguments or between the DPs of a predicate, where nominative is simply CASE1 and accusative CASE2.16 The argument distinguishing strategy is prevailing in e.g. Icelandic, German and Swedish, while the DP distinguishing strategy is in e.g. English and Danish. On this approach, it becomes at least partly understandable that many speakers of English, Danish and some other languages have extended the diacritic function of the cases (in colloquial speech) so as to even distinguish between adjacent DPs. Consider the following results from an extensive study on case-marking in New Zealand English (Heidi Quinn p.c., see further Quinn 2005). The 90 informants that took part in the survey were asked to fill the slots with the preferred pronominal forms, in for instance the sentences in (30) and (31): 17

16

17

This captures NOM–ACC conversion phenomena, as in passives. However, the underlying ‘argument structure conversion’ is a syntactic phenomenon. Many thanks to Heidi Quinn for these data (which are not found in this form in her book). 91 and 92 are the numbers of relevant answers given by the 90 informants.

Remarks on features 39

(30) a. b. c. d. e. f. g. h. (31)

[__ and ___] have just taken part in one of these workshops on Asian food. He and I… 45/91 = 49% Him and I… 35/91 = 38% Me and him… 9/91 = 10% Him and me… 2/91 = 2 % I and he… 0/91 I and him… 0/91 Me and he… 0/91 He and me… 0/91 If Morris is late, would you mind taking [__ and __] to the airport?

a. b. c. d. e. f. g. h.

… him and I… … him and me… … me and him… … he and I… … I and he… … I and him… … me and he… … he and me…

30/92 29/92 27/92 4/92 0/92 0/92 0/92 0/92

= 33% = 32% = 29% = 4%

If English subject and object case was decided by syntactic argument features like +NOM and +ACC, this kind of variation would be hard or impossible to explain. According to Quinn (2005), the case variation found in coordinates relates to a morphological distinction between weak and strong pronouns in English and is at least partly influenced by the relative phonological heaviness or ‘robustness’ of strong pronoun forms. Evidently, the ‘structural’ cases, better referred to as the relational cases (see Sigurðsson 2006a), are diactritic features in PF. Case is always a PF interpretation or expression of a complex syntactic correlation, that is, there seems never to be a one-to-one correlation between a particular morphological case and a single feature in syntax. Thus, the Icelandic dative is used for multiple purposes, marking at least the following nine types of elements (Sigurðsson 2003: 230): (32) a. b. c. d. e.

Quirky subjects Indirect objects Direct objects Free benefactives Possessors

(e.g. ‘me feels good’ = ‘I feel good’) (e.g. ‘she gave me the book’) (e.g. ‘she invited me’) (e.g. ‘she wrote me a poem’) (e.g. ‘she looked into eyes me’ = ‘into my eyes’)

40 Halldór Ármann Sigurðsson f. g. h. i.

Prepositional objects Objects of adjectives Instrumental DPs Other DP adverbials

(e.g. ‘she stayed by me’) (e.g. ‘she was me nice’ = ‘nice to me’) (e.g. ‘she stuck him a knife’ = ‘with a knife’) (e.g. ‘she was me older’ = ‘older than me’)

All these elements may however be differently marked, that is, with different cases or with prepositions, depending on various factors. In spite of this, it is possible to formulate many nice generalizations about the distribution of the dative and the other cases in Icelandic (see Jónsson 2003, 2005), as in other case languages, but the relevant point here is that all such generalizations are complex statements, involving many factors. That is, generalizations of this sort basically take the form: A + B + C > CASE X, D + F > CASEX, G + H + I > CASEX, and so on, were A, B, etc., denote syntactic factors and relations and CASEX is some particular case in morphology. Thus, not all agents are nominative in Icelandic, not all subjects are agentive and not all subjects are DPs, but the complex statement in (33) holds true:18 (33) If & & then

a. b. c. d.

X is a DP X is a subject X is an agent X is morphologically nominative

Similarly, one can formulate several (albeit less accurate) complex statements for nominative non-subjects, including nominative objects and nominative predicates (in non-finite as well as finite clauses). However, saying that all these complex statements together in fact are syntactic +NOM is just a pointless word-game, blurring instead of increasing our understanding of grammar. Renaming the cases, saying for instance that nominative “is Tense” in a disguise, is even farther off the track. The core of the matter is that syntax does not operate with case features, that is, the NO CASE GENERALIZATION in (34) holds: (34) Syntax has no case features, that is, it has no rule or process saying “you carry nominative case in morphology if and only if you are +NOM in syntax, you carry dative case in morphology if and only if you are +DAT in syntax”, and so on.

18

But notice that ‘subject’ is not a syntactic primitive, so (33) is actually a metalinguistic statement, on notions of classical grammar and not on syntax.

Remarks on features 41

Notice that this is not to say that case is unlinked to or does not reflect syntactic structures, but it is to say that these structures themselves do not contain or operate with case, neither as features nor as syntactic objects of any other sort. Actually, it seems to be a fundamental property of language that it never applies one-to-one mappings between any levels or derivational stages. Assuming that it does is a contradiction in terms, since it suggests that the derivation is in fact non-derivational, simply reproducing an input as an equivalent albeit a differently looking output. The alleged syntactic effects of nominative case are real, of course, but they are not brought about by case but by matching and valuation of interpretable features, above all Person in accusative systems (Sigurðsson 2003 and subsequent work) but often Aspect, Tense or Focus in different systems (cf. Miyagawa 2005). Nichols (1990) contains an interesting study of dependency marking with respect to the typological notions of A(gent)-S(ubject)-P(atient). In her sample of 155 (relevant) languages, 148 or 95,5% had some such marking, and these in turn split into about equally large groups, with and without case-marking (see Nichols 1990: 90). Proponents of case features in Universal Grammar might want to use statistics like these as an argument in favor of their standpoint, but so might their opponents: – If case is not part of UG, why is it so common in the languages of the

world? – If case is part of UG, why is it absent from so many of the languages of

the world? Case is but one of many available strategies for PF marking the relations between an argument and its linguistic environment. Consider the following list of simple facts: – The most common strategy, across languages and constructions, is not

to make any PF marking at all. – Closely related languages, with the same basic case system, like Ice-

landic and German, often use different cases to mark the (to the best of our knowledge) same syntactic relation.

42 Halldór Ármann Sigurðsson – Even within one and the same language, there may be extensive varia-

tion in case-marking, depending on either linguistic or social variables (constructions, dialects, idiolects,…).19 – Where one language uses case to mark a relation, another language may opt for suprasegmental marking (cf. Sigurðsson 2003: 326 on Swedish) or marking of non-argument members of the relevant syntactic relation (prepositions, particles, verbs, complementizers, adverbs,…). I believe we should look for an understanding of these facts in PF, the medium that ‘broadcasts’ Narrow Syntax. Narrow Syntax itself is a much more abstract or ‘semantic’ system that does not operate with PF visible units like inflectional features in oral languages or facial expressions in signed languages, nor does it have features that stand in simple one-to-one mapping relations to elements in the perceptible form of language. Understandably, it is tempting to seek explanations in Narrow Syntax, since Narrow Syntax explanations should hold generally, across languages, and are thus more ambitious and elegant than alternative accounts could ever be. In hastily seeking such explanations, however, linguistics has too often been overzealous, producing ‘generalizations’ that bear on wishful thinking and other human shortcomings rather than on language. 5. A note on agreement While person evidently differs from formal gender, number and case in being a syntactic category, its values, 1st, 2nd and 3rd person, are not syntactic objects. Rather, they are PF expressions, not of individual objects or features in syntax but of complex matching correlations, as I have argued elsewhere (Sigurðsson 2004a,b). If formal feature values are strictly speaking non-syntactic, then morphological agreement must be non-syntactic as well, and this is what I have argued for in considerable detail in earlier work (e.g. Sigurðsson 2004c, 2006b).20 If so, morphological agreement is only an indirect reflection of abstract, syntactic Agree. Reasonably, there is a syntactic Agree or Selection correlation between any two syntactic elements that are merged. However, 19

20

Jónsson and Eythórsson (2005) study case marking variation in Icelandic and Faroese, presenting evidence that some of the variation is actually based on optionality in the grammar of individual speakers. Bobaljik (2006) comes to the same conclusion, also basing his arguments partly on facts from Icelandic.

Remarks on features 43

as is evident, only some such correlations are overtly expressed by morphological agreement. Consider the following simple facts from a few Germanic languages: (35) a. They would b. Sie würden they would.3PL c. De skulle they would

be gewählt elected bli be

d. Þeir mundu verða they would.3PL be

elected. werden. be valda. elected.PL valdir. elected.NOM.M.PL

English German Swedish Icelandic

English shows no agreement at all here, German has finite verb agreement only, Swedish has only plural agreement of the predicative participle, and Icelandic has finite verb agreement as well as case, gender and number agreement of the participle. There does not seem to be any syntactic motivation for having or not having these agreement types. As will be discussed below, agreement is functional in the sense that it reduces ambiguity, but reducing ambiguity is not the ‘job’ or ‘goal’ of syntax (or else we would not expect grammars to produce large amounts of ambiguity). Rather, various agreement types are different PF options, and the truth of the matter is that we have little understanding or knowledge of why languages choose one over the other (or, in general, of why languages opt for different disambiguating strategies). Even internally to individual languages there is abundant evidence that morphological agreement does not usually relate to semantics or directly to syntax, instead involving processes or ‘adjustments’ that are internal to morphology or broad PF. Consider the variation in (36), discussed by Rögnvaldsson (1982): (36) a. Við hlökkum til jólanna. we.NOM look.1PL to Christmas.the ‘We look forward to Christmas.’ b. Okkur hlakkar til jólanna. us.DAT look.3SG to Christmas.the ‘We look forward to Christmas.’ In the standard (36 a), the subject is nominative, triggering regular, full agreement in person and number of the finite verb. In the common but substandard (36b), on the other hand, the verb hlakka til ‘look forward to’

44 Halldór Ármann Sigurðsson takes a dative subject and this non-nominative case selection automatically ‘switches off’ the person and number agreement of the verb, which thus shows up in the default 3rd person singular instead of 1st person plural. This is in full accordance with the well-known fact that phi-feature specified nominative subjects trigger finite verb agreement in Icelandic, whereas the finite verb regularly shows up in the default 3rd person singular in the absence of a nominative argument (see, e.g., Sigurðsson 2004c and the references cited there). Tellingly, however, both versions in (36) seem to have the same semantics, as highlighted by the fact that many speakers can use them both, not depending on semantics but on social factors (as far as can be judged).21 For these speakers, clauses with the verb hlakka til may either show up with a nominative or a dative subject, and the presence vs. absence of full agreement is an automatic morphological reaction to the case selection, independent of the underlying semantic/syntactic structure. To an extent, this is in fact reminiscent of phonological assimilation processes across word or constituent boundaries: these processes are fed by syntax, leading to adjacency of the sounds involved, but they do not take place in syntax nor are they directly controlled by syntactic factors. Consider also the fact that Modern Icelandic has agreement of participles selected by vera ‘be’, as opposed to participles selected by hafa ‘have’, as illustrated in (37): (37) a. Þær eru ekki lesnar. they.F are not read.NOM.F .PL ‘They are not being read (by somebody). / They are not well-read.’ b. Bækurnar hafa þær ekki lesið. books.the.F have they.F not read.DEFAULT (= NOM /ACC.N.SG) ‘The books, they have not read.’ In Old Norse, on the other hand, participles selected by hafa ‘have’ could but did not have to agree with direct objects (see Nygaard 1906: 188). Again, there does not seem to be any syntactic reason for the variation. In particular, it does not correlate with word order, as related phenomena do in some other languages.22 21

22

Thanks to Eríkur Rögnvaldsson for reminding me of the import of facts of this sort. On speaker-internal case variation, not involving any semantic differences, see Jónsson and Eythórsson (2005: 235 f.). In this respect, Old Norse was similar to Fruilian, a Rhaeto-Romance language spoken in north-eastern Italy (see Paoli 2006).

Remarks on features 45

Similarly, agreement does not spread from finite verbs to non-finite verb forms in Icelandic, or, to my knowledge, in other Indo-European languages, while it does in some Bantu languages. Compare the Swahili example in (38a), here adapted from Carstens (2001: 150), and the Icelandic one in (38b): (38) a. Juma alikuwa amepika Juma 3SG.was 3SG.cooked ‘Juma had cooked food.’ b. Jón hafði gert/*gerti we had.3SG made.Ø/*3SG

chakula. food mat. food

These are only a handful of practically innumerable agreement differences between as well as within languages. It seems most likely to me that differences of this sort should not be explained in syntactic terms but in terms of morphological feature copying processes that apply differently in different languages and language varieties. These copying processes are typically abstract,23 but there is evidence that they are sometimes affected by purely phonological factors. Thus, Icelandic adjectives generally have 144 inflectional possibilities or ‘slots’,24 commonly expressed by 30 distinct forms for each adjective. Strikingly, all this richness is blocked in case the adjective has a stem ending in an unstressed vowel, all such adjectives being entirely indeclinable.25 There are however well-known cases where agreement has semantic correlates, for example certain instances of secondary predicate agreement in Icelandic, as in (39) (from Sigurðsson 2006b: 214): (39) a. Prestarnir hittu biskupinn drukknir. priests.the.N.M.PL met bishop.the.A.M.SG drunk.N.M.PL ‘The priests met the bishop drunk (i.e. the priests were drunk).’ b. Prestarnir hittu biskupinn drukkinn. priests.the.N.M.PL met bishop.the.A.M.SG drunk.A.M.SG ‘The priests met the bishop drunk (i.e. the bishop was drunk).’

23 24

25

Not copying sounds but abstract features like PLURAL and FEMININE, see section 6. 4 cases x 3 genders x 2 numbers x 3 degrees x 2 (indefinite or ‘strong’ vs. definite or ‘weak’). Hissa ‘surprized’, fullvalda ‘sovereign’, hugsi ‘pendent, (deep) in though’, brosandi ‘smiling’, passandi ‘fitting, appropriate’, sveitó ‘provincial’, and so on.

46 Halldór Ármann Sigurðsson However, this does not mean that the underlying syntactico-semantic structures cannot be expressed without agreement morphology in other languages; it only means that individual languages make use of whatever overt tools they happen to have at their disposal, for the purpose of reducing ambiguity in PF.26 Agreement involves copying of information, call it INF, from an item A in a structure or domain to another item B in the same structure/domain. While INF is redundant or uninterpretable on B (INF[–M]), it may or may not be meaningful on A (INF[+/–M]): (40) A[INFi[+/–M]]… B … > A[INFi[+/–M]] … B[INFi[–M]] … As indicated by the indexes, agreement establishes a referential correlation of sorts, between agreeing features and their controlling features (their ‘antecedents’, as it were). Information repetition of this sort is not an essential part of language, as best seen by the simple fact that most or all languages lack agreement in numerous constructions where other languages have agreement. Thus, a language like German can easily do without, say, adjectival/participial predicate agreement, even though such agreement is obligatory in for instance Swedish and Icelandic (and even though these languages all have similar attributive agreement systems). In view of the pervasive redundancy of agreement, it seems paradoxical that it is nonetheless a striking trait of natural language as compared to other communicative systems, like mathematical languages, sign systems and game rules. Even writing systems are largely ‘clean’ of agreement phenomena, that is, we do not usually redundantly repeat graphic information in writing. Evidently, agreement does not ‘make any sense’ from a narrowly syntactic point of view – if it did, we would expect to find it in languages in general and also in other computational systems. From a communicative point of view, however, agreement makes perfect sense, since it radically reduces ambiguity, thereby facilitating processing. For example, pronominal gender agreement across clause boundaries reduces ambiguity in the answers in the Icelandic (41a, b), thus rendering the answers functional and felicitous (F and M = feminine vs. masculine):

26

Thráinsson (1975) is an interesting study of how inflectional morphology radically reduces ambiguity in Icelandic gapping constructions, as compared to English.

Remarks on features 47

(41) a. Var bókin þyngri en boltinn? – Nei, hann var þyngri. was book.the.F heavier than ball.the.M – no, he was heavier ‘Was the book heavier than the ball. – No, it (i.e., the ball) was heavier.’ b. Var boltinn þyngri en bókin? – Nei, hún var þyngri. was ball.the.M heavier than book.the.F – no, she was heavier ‘Was the ball heavier than the book. – No, it (i.e., the book) was heavier.’ Agreeing features evidently function as disambiguating indexes in discourse. What makes this possible is not the semantics of the features, but their distinctiveness. Interestingly, however, agreement features are typically semantically based in at least some part of the lexicon (as recently discussed by Svenonius 2007). An intriguing question is why languages develop seemingly ‘pointless’ properties like inflectional classes. Thus, around 130 different classes or paradigms are needed in order to fully describe Icelandic indefinite noun inflections (see Thomson 1987).27 A related question, raised by Svenonius (2007), is why inflectional classes do not trigger agreement in the same way as for instance gender does. The trivial answer to this latter question is that gender per definition triggers agreement (Corbett 1991), whereas classifiers and inflectional classes do not. However, that does not answer the deeper question of why ‘pointless’ distinctions in morphology do not usually form the basis for agreement. Presumably, a category is the less likely to get repeated or referred to by agreement the less functional or semantic load it carries. I will not try to develop any more precise answers to these intriguing questions here, as they do not bear on the basic issue at stake, namely, which features belong to syntax and which do not.28 Trivially, inflectional classes do not belong to syntax and the evidence presented here suggests that formal features are not syntactic either. Rather, they are more or less inaccurate morphological PF translations of more abstract syntactic categories and correlations. 27

28

Only some of these classes are productive in the modern language, and many of them are ‘historical relics’, containing few members or even only a single one. Svavarsdóttir (1993) has developed a more economic analysis, identifying 27 ‘true’ classes. See Corbett and Baerman (2007) for a general discussion of inflectional classes and other features in morphology that are strictly word bounded and hence nonsyntactic even in the traditional sense.

48 Halldór Ármann Sigurðsson 6. Concluding remarks on the ‘syntactic’ PF Morphological agreement or feature copying involves copying of values like MASCULINE and PLURAL and not necessarily of the phonological shape of these values, as for intance seen by the phonological mismatch between the plural marker -ar of prestar in (39) above and the plural marker -ir of the suffixed article and the secondary predicate adjective. Thus, the processes involved operate with abstract objects, that are later transformed into or translated as physical, phonetic objects by operations in shallow PF. Moreover, these abstract PF agreement processes operate in a ‘syntactic manner’, as it were, applying feature matching, observing minimality and showing intervention effects, as highlighted by recent research on Icelandic dative intervention, for instance in the works of Holmberg and Hróarsdóttir (2003), Hiraiwa (2005), Nomura (2005), Chomsky (2005) and Sigurðsson and Holmberg (2008). These circumstances suggest that agreement morphology is able to ‘see’ syntax, even though it takes place after transfer and thus operates with elements that are out of sight for the semantic interface, such as formal gender, number and case values. If so, the computation (or some computation) proceeds on the PF side after transfer, yielding a reversed Y-model, as it were.29 On the present approach, broad PF cannot be part of Universal Grammar (see Sigurðsson 2004a). Rather, Universal Grammar (or Universal Narrow Syntax) interacts with motoric systems which provide audible and/or visible means to produce or ‘broadcast’ complex symbols that are processable as incomplete expressions or translations of Narrow Syntax. On this view, it might seem unexpected that many morphological and other typological generalizations hold across languages that, to the best of our knowledge, are unrelated or are at least not evidently related. There are however many factors to be taken into consideration here. One is that our knowledge of language history is extremely limited. For all we know, the world’s oral languages might all be the descendants of say 1 or 10 or 100 out of say 5.000 languages that existed say 40.000 years ago, each spoken by only a tiny group of people. The gap between our knowledge and the potential facts in this field is so huge that we can hardly even speculate about how morphologial and typological regularities might have come into being in the remote past. Another factor to consider is that physiological properties may well impose universal constraints and features on individual grammars with29

Indirectly suggesting that covert movement (in ‘LF’) should be eliminated from the model, an issue I shall not address here, though (but see Kayne 1998).

Remarks on features 49

out belonging to Universal Grammar. That is, being universal in oral or even also in signed languages does not entail being part of Universal Grammar as such. A third factor is that given Universal Grammar one would expect some similarities between different translations or mappings from this common system onto individual PF grammars. Thus, if abstract Person, Number and Tense are features of Universal Grammar, as seems reasonable (see Sigurðsson 2004a,b), it is not unexpected that more or less inaccurate indicators or exponents of these abstract categories get marked or highlighted in otherwise unrelated grammars. However, as we have seen, just a glance immediately reveals that the mappings of syntactic/semantic features onto even closely related languages, like the Germanic ones, are amazingly varied (see further Sigurðsson 2004c). Even if broad PF, including the ‘sign form’ of sign languages, is not part of Universal Grammar, it is evidently a highly sophisticated, layered system. Much of our knowledge of grammar or grammars is about the elements and processes of this system. About Narrow Syntax, on the other hand, we know very little, and the main reason why that is so is that we can only study it indirectly, through the interfaces. If it is the case that formal feature values belong to PF only, then we know even less about Narrow Syntax than many linguists have been hoping for or optimistically assuming. However, realizing that one knows less than one previously thought is usually a sign of progress. Acknowledgements For useful discussions, comments and/or help with data, thanks to anonymous reviewers and to Christer Platzack, David Adger, Eiríkur Rögnvaldsson, Greville Corbett, Heidi Quinn, Idan Landau, Klaus von Bremen, Thórhallur Eythórsson and Verner Egerland. The research for this paper was supported by a grant from the Swedish Research Council, VR 421-20062086 (in collaboration with Verner Egerland). References Bobaljik, Jonathan David 2006 Where’s φ? Agreement as a postsyntactic operation. Ms., University of Connecticut. Carstens, Vicki 2001 Multiple agreement and case deletion: Against φ -(in)completeness. Syntax 4: 147–163.

50 Halldór Ármann Sigurðsson Chomsky, Noam 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 Uriagareka (eds.), 89–155. Cambridge, MA: MIT Press. 2008 On phases. In Foundational Issues in Linguistic Theory: Essays in Honour of Jean-Roger Vergnaud, Robert Freidin, Carlos P. Otero and Maria Luisa Zubizaretta (eds.), 133–166. Cambridge, MA: MIT Press. Corbett, Greville 1991 Gender. Cambridge: Cambridge University Press. 2000 Number. Cambridge: Cambridge University Press. Corbett, Greville and Matthew Baerman 2006 Prolegomena to a typology of morphological features. Morphology 16 (2): 231–246. Halle, Morris and Alec Marantz 1993 Distributed morphology. In The View from Building 20, Kenneth Hale and Samuel Jay Keyser (eds.), 111–176. Cambridge, MA: MIT Press. Hiraiwa, Ken 2005 Dimensions of symmetry in syntax: Agreement and clausal architecture. Doctoral dissertation, MIT. Holmberg, Anders and Thorbjörg Hróarsdóttir 2003 Agreement and movement in Icelandic raising constructions. Lingua 113: 997–1019 (see also Lingua 114: 651–673). Jónsson, Jóhannes Gísli 2003 Not so quirky: On subject case in Icelandic. In New Perspectives on Case Theory, Ellen Brandner and Heike Zinsmeister (eds.), 127–163. Stanford: CSLI Publications. 2005 Merkingarhlutverk, rökliðir og fallmörkun [Thematic roles, arguments and case-marking]. In Íslensk tunga III: Setningar, Höskuldur Þráinsson (ed.), 350–409. Reykjavík: Almenna bókafélagið. Jónsson, Jóhannes Gísli and Thórhallur Eythórsson 2005 Variation in subject case marking in Insular Scandinavian. Nordic Journal of Linguistics 28: 223–245. Josefsson, Gunlög 1998 Minimal Words in a Minimal Syntax: Word Formation in Swedish. Amsterdam /Philadelphia: John Benjamins. 2006 Semantic and grammatical genders in Swedish – independent but interacting dimensions. Lingua 116: 1346 –1368. Kayne, Richard S. 2002 Pronouns and their antecedents. In Derivation and Explanation in the Minimalist Program, Samuel David Epstein and T. Daniel Seely (eds.), 133 –166. Oxford: Blackwell Publishing.

Remarks on features 51 2005

On parameters and on principles of pronunciation. Ms., New York University. Kvaran, Guðrún 2005 Íslensk tunga II: Orð. Reykjavík: Almenna bókafélagið. Källström, Roger 1993 Kongruens i svenskan [Agreement in Swedish]. Göteborg: Acta Universitatis Gothoburgensis. McFadden, Thomas 2004 The position of morphological case in the derivation: A study on the syntax-morphology interface. Doctoral dissertation, UPenn. Miyagawa, Shigeru 2005 On the EPP. In Perspective on Phases, Martha McGinnis and Norvin Richards (eds.), 201–236. Cambridge, MA: MIT (Working Papers in Linguistics). Nichols, Johanna 1992 Linguistic Diversity in Space and Time. Chicago: The University of Chicago Press. Nomura, Masashi 2005 Nominative Case and AGREE(ment). Doctoral dissertation, University of Connecticut. Nygaard, Marius 1906 Norrøn syntax. Oslo: H. Aschehoug & Co (W. Nygaard). Paoli, Sandra 2006 On the relation of [gender] agreement: Evidence from Fruilian. In Studies on Agreement, João Costa and Maria Cristina Figueiredo Silva (eds.), 223–242. Amsterdam /Philadelphia: John Benjamins. Platzack, Christer 2006 Case as Agree marker. Working Papers in Scandinavian Syntax 77: 71–99. Quinn, Heidi 2005 The Distribution of Pronoun Case Forms in English. Amsterdam / Philadelphia: John Benjamins. Rögnvaldsson, Eiríkur 1981 U-hljóðvarp og önnur a-ö víxl í nútímaíslensku [U-mutation and other a-ö shifts in Modern Icelandic]. Íslenskt mál og almenn málfræði 3: 25–58. 1982 We need (some kind of) a rule of Conjunction Reduction. Linguistic Inquiry 13: 557–561. [Reprinted 1990 in Modern Icelandic Syntax, Joan Maling and Annie Zaenen (eds.), 349 –353. San Diego: Academic Press.] Sigurðsson, Halldór Ármann 2003 Case: Abstract vs. morphological. In New Perspectives on Case Theory, Ellen Brandner and Heike Zinsmeister (eds.), 223–268. Stanford: CSLI Publications.

52 Halldór Ármann Sigurðsson 2004a

Meaningful silence, meaningless sounds. In Linguistic Variation Yearbook 2004, Vol. 4, Pierre Pica, Johan Rooryck and Jeroen van Craenenbroeck (eds.), 235–259. Amsterdam/Philadelphia: John Benjamins. 2004b The syntax of person, tense, and speech features. Italian Journal of Linguistics / Rivista di Linguistica 16: 219 –251 [special issue, ed. by Valentina Bianchi and Ken Safir]. 2004c Agree and agreement: Evidence from Germanic. In Focus on Germanic Typology, Werner Abraham (ed.), 61–103. Berlin: Akademie Verlag. 2006a The Nom /Acc alternation in Germanic. In Issues in Comparative Germanic Syntax, Jutta Hartmann and Laszlo Molnarfi (ed.), 13–50. Amsterdam /Philadelphia: John Benjamins. 2006b Agree in syntax, agreement in signs. In Agreement Systems, Cedric Boeckx (ed.), 201–237. Amsterdam /Philadelphia: John Benjamins. 2006c The Nominative Puzzle and the Low Nominative Hypothesis. Linguistic Inquiry 37: 289 –308. Sigurðsson, Halldór Ármann and Anders Holmberg 2008 Icelandic dative intervention: Person and number are separate probes. In Agreement Restrictions, Roberta D’Alessandro, Susann Fischer and Gunnar Hrafn Hrafnbjargarson (eds.), 251–280. Berlin /New York: Mouton de Gruyter. Svavarsdóttir, Ásta 1993 Beygingarkerfi nafnorða í nútímaíslensku [The Inflectional System of Nouns in Modern Icelandic]. Reykjavík: Insitute of Linguistics, University of Iceland. Svenonius, Peter 2007 Interpreting uninterpretable features. Linguistic Analysis 33: 375–413. Telemann, Ulf, Staffan Hellberg and Erik Andersson 1999 Svenska Akademiens grammatik. Vols 1–4. Stockholm: Svenska Akademien. Thomson, Colin D. 1987 Íslensk beygingafræði. Isländische Formenlehre. Icelandic Inflections. Hamburg: Helmut Buske Verlag. Thráinsson, Höskuldur 1975 Gapping in Icelandic: Functional explanations and the No-Ambiguity Condition. CLS 11: 604–614. 1983 “Ekki til í fleirtölu” [‘Has no plural form’]. Íslenskt mál og almenn málfræði 5: 175–177.

Feature valuation by sideward movement Bartosz Wiland

This paper offers a reanalysis of sideward movement, as proposed by Nunes (1995, 2001) and developed in Hornstein (2001). I propose that if inter-arboreal Remerge is available in syntax, it can be accessible only to syntactic objects that have a full set of φ-features and case to value. It is argued that sideward movement into θ -positions is applicable, but is only collateral. Consequently, syntactic objects that are φ-incomplete do not undergo interarboreal Remerge, despite the fact that such a movement has been analyzed as being triggered by the Greed of checking a matrix θ -role. The proposal allows to unify inter-arboreal Remerge with other instances of internal Merge which are feature driven and to account for the fact that prepositional and adverbial parasitic gaps are generally impossible.

1. Introduction Bobaljik & Brown (1997) propose that inter-arboreal movement is an available operation in syntax. Hornstein (2001), extending Nunes’ (1995, 2001) analyses, argues that parasitic gaps and adjunct control constructions are licensed by sideward movement. (1)

Principles of sideward movement a. b. c. d. e.

sideward movement is driven by Greed every leg of movement obeys the Extension Condition Ā-to-A movement is licit the head of the chain must c-command every single copy copies are deleted under asymmetric c-command

In Nunes’ analysis of the Copy Theory of Movement, the deletion of the lower copy is necessary for the linearization of a syntactic structure at the PF interface. Given the numeration in (2a), the derivation proceeds as in (2b) and includes two nondistinct copies of Jack. The copy of Jack that is asymmetrically c-commanded by its nondistinct copy is deleted in the phonological component.

54 Bartosz Wiland (2)

a. N = {promoted1, Jack1, was1} b. [Jacki [was [promoted Jacki]]]

In sideward movement, a constituent α of a syntactic object K is copied and merged with an object L, as long as the non-distinct copies of α are in an asymmetric c-command relation at the Spell-out, which facilitates the linearization of the chain CH={α, αi}. Sideward movement proceeds as follows. (3)

a. [K… αi… ] αi  Merge  [L… ] ! : z- Copy -m b. [K… αi… ] [M αi [L… ]]

Nunes and Hornstein argue that sideward movement applies, for instance, in the derivation of parasitic and adjunct control constructions. (4)

Parasitic construction licensed by sideward movement a. Which book did you criticize without reading? b. [CP Which booki did [TP you T [vP [vP ti v [VP criticize ti] [PP without reading ti]]]]]

(5)

Adjunct Control licensed by sideward movement 1 a. Peter lost a tooth before PRO opening the bottle of beer. b. [TP Peteri T [vP [vP ti v [VP lost a tooth] [PP before opening the bottle of beer]]]]

1

In the Movement Theory of Control, as advocated by Hornstein (1999, 2001) and Boeckx and Hornstein (2003, 2004), PRO in AC constructions is a trace (a silent copy) of a category moved overtly into the matrix clause. The movement of Peter from the adjoined clause to the specifier of the matrix vP crosses the adjunct island. It seems plausible to assume that the CP is adjoined to the vP, not to the v’. From the specifier of the CP adjoined to vP, Peter can only move to the specifier of the matrix vP sideways, whereas if the adjunct is a sister of v’, then the movement of Peter from Spec-CP of the adjunct clause to Spec-vP is upward movement. In either case, if AC involves movement at all, the movement of Peter must cross the adjunct island. Since Hornstein, following Nunes (1995), claims that sideward movement is not island-sensitive, the CP in (5b) must be vP-adjoined. Otherwise, the upward movement of Peter from the v’-adjunct would produce the island violation.

Feature valuation by sideward movement 55

According to Hornstein’s and Nunes’ analyses, sideward movement works as follows. Out of the two independent lexical arrays, the adjunct is constructed prior to the matrix VP. Then, an XP moves sideways to the matrix clause in order to check its features. This leg of movement is driven by the Greed of checking the matrix θ-role and structural case. Hornstein takes θroles to be included in the set of features, along with other formal features. The sideward movement of the XP proceeds through the edge of the adjoined CP. If XPs move sideways in order to value their θ -roles and case, then a thematic XP (ideally a nominal) should bear a unanimous set of features (θ and φ) to constitute a goal for the probe. However, if a θ -role is a feature, then it creates a problem of its being valued multiple times and with different values, since sideways movement into incongruent thematic positions is available, as in for instance the AC constructions (cf. (6)), or benefactive constructions (cf. (7)). (6)

Peter[PATIENT] lost a tooth before PRO[AGENT] opening the bottle of beer.

(7)

Which girli did you send a picture of ti[PATIENT] to ti[BENEFACTIVE] ?

The paper is organized as follows. The proposal for the scenario of feature valuation by sideward movement within a Remerge Theory of Movement is given in section 2. In section 3, I argue that CPs do not move sideways, since they are headed by null pronouns that lack the full set of φ-features necessary for inter-arboreal Remerge, which involves Agree. Section 4 uses the discussion in Boeckx (2003) on the derivational approach to donkey anaphora, which, if correct, provides further evidence for the impossibility of moving weak or null pronouns sideways. In section 5, I show that the analysis of across-the-board constructions based on sideward movement has to be rejected, since categories that do not value their φ-features on matrix probes do not undergo inter-arboreal Remerge. Also, I point out the fact that movement for θ -roles does not license inter-arboreal Remerge, which accounts for the fact that locative PPs do not move sideways. Section 6 addresses the issue of how formal features are relevant for inter-arboreal Remerge and the accessibility of syntactic objects with full sets of φ-features to matrix Vs, once they are selected and merged in the adjunct clause. In the conclusion it is shown how the scenario offered in the paper interplays with the inaccessibility of CPs or other “weak” syntactic objects to inter-arboreal Remerge. Section 7 is the conclusion.

56 Bartosz Wiland 2. Proposal The ‘θ-and-case’ version of sideward movement has been advanced to explain the licensing of a number of constructions: parasitic gaps (cf. (8)) (Nunes 2001, 2004; Hornstein 2001); AC constructions (cf. (9)) (Hornstein 2001); relative clauses (cf. (10)) (Nunes 2001: 316-320); or across-theboard constructions (cf. (11)) (Hornstein & Nunes 2002, Nunes 2001: 336339; but see section 5 of this paper for discussion). It fails, however, to explain the derivation of constructions that involve (sideward) movement into incongruent θ-positions, as for instance in (6) and (7). (8)

a. Which articlei did you file without reading ti? b. Which moviei did she try to summarize ti without watching ti?

(9)

Jackj kissed Mary before PROj kissing her sister.

(10) Which claim that Jackk made was hek willing to discuss? (11) Which articlel did Jack recommend tl and Mary read tl? The immediate conjecture is that if inter-arboreal Remerge is available in grammar, it should be accessible under a principle that can provide a unanimous account of all instances of sideward movement. The assumptions relevant for the discussion and the proposal are as follows. (12) The assumptions a. Agree and Move (Remerge/Agree-and-pied-piping) are operations necessary for convergence; syntactic objects undergo feature valuation for convergence reasons. b. Θ-roles are excluded from the set of formal features. c. Syntactic objects do not undergo thematic valuation. (13) The proposal Sideward movement is an accessible operation for syntactic objects that have a full set of φ-features and case to value. Inter-arboreal Remerge into thematic positions is applicable, but is only collateral. In Hornstein’s (2001) analysis, sideward movement is licensed by checking a matrix θ -role and structural case. Hornstein’s approach assumes that θ roles are checkable features that can drive movement, and that an argument can receive more than one θ-role throughout the derivation. These assumptions are at odds with the conservative version of the Θ -Theory which en-

Feature valuation by sideward movement 57

sures that arguments are θ -marked in their base-generated positions (cf. Williams 1980, 1994; Neeleman & van de Koot, forthc.). Hornstein’s treatment of θ-roles as formal features allows one to claim that checking a matrix θ-feature can license sideward movement. The proposal in (13) makes a case for an alternative scenario for sideward movement, which for the purposes of the present discussion I will assume to exist in grammar. Within the Θ-Theory which ensures that θ -roles are excluded from the set of formal features, only syntactic objects that have a full set of φ-features and case to value are legible for sideward movement.2 In the probe and goal system of Chomsky (2000, et seq.), in order to satisfy the Principle of Full Interpretation no unvalued features can survive the derivation. This can be obtained only if the goal is a complete match for the probe, so that the probe’s uninterpretable φ-features are all valued. The following are two fundamental conditions on Agree. (14) a. Probe and goal must both be active. b. α must have a complete set of φ-features (it must be φ-complete) to delete uninterpretable features of the paired matching element β. (Chomsky 2001a) The condition in (14b) is additionally restricted by the Maximize principle (cf. Chomsky 2001a), which states that feature checking is an “all or nothing” process, since Agree does not involve selective deletion of features. Maximize also results in the movement of the goal to the specifier of the matching probe whenever the probe has the EPP property.3 Since remerge into A-positions is driven by checking formal features and only a complete φ-set leads to the valuation of uninterpretable features, inter-arboreal Remerge of φ-incomplete elements is unexpected to take place in the probe and goal system.4 The analysis of sideward movement 2

3

4

In Chomsky’s (2000, 2004, 2005) analysis of formal features, the structural case on DPs is a reflex of uninterpretable φ-features on probes T and v. On the other hand, given that θ-roles have been argued to be realizations of verbal concepts in syntax (cf. Reinhart 2002), the asymmetry between θ-roles and FF becomes even more striking, as, say, a lexical verb assigning a role to its internal argument need not be a probe, at least in the minimalist definition of a probe. Checking probe’s EPP together with its other uninterpretable features by a matching goal is justified even without Maximize, once we think of EPP as “a feature of a feature” (cf. Pesetsky & Torrego 2001). While valuation of uninterpretable features appears to be the trigger for A-movement, Ā-movement need not be driven by formal feature checking. It has been re-

58 Bartosz Wiland that is driven by the valuation of φ-features and case and is not driven by checking a matrix θ-role, can account for the fact that thematic locative PPs (cf. (15)) and thematic adverbs (cf. (16)) do not license parasitic gaps. (15) *This is a tree [under which]i you shouldn’t sit ti after your dog lied ti. (16) *[How sick]i did Jack look ti [without actually feeling ti]? 3.

Pseudo-parasitic gaps and null pronouns

3.1. Parasitic and apparent parasitic gaps The typological distinction between genuine parasitic gaps and pseudoparasitic gaps was first developed in Postal (1994).5 According to Postal’s analysis, pseudo-parasitic gaps are not real parasitic gaps, as they violate universal conditions on parasitic gap licensing.6 Thus, Complex NP Shift

5

6

cently advocated in Neeleman and van de Koot (2007) that movement operations have interpretive effects on phrases in situ and that in this way discourse information is encoded. On the example of Dutch, it is shown that this type of interaction derives topic and focus without the postulation of Rizzi’s (1997) articulate left periphery, which is claimed to provide specific positions in the functional structure responsible for interpretive effects (TopP, FocP) in the cartographic approach to syntax. Ā-movement, however, is irrelevant for the current discussion since inter-arboreal remerge takes place in an A-position, and only subsequent upward movement that enables the head of the chain to c-command lower chain links is Ā-movement. See also Huybregts and van Riemsdijk (1985), Bennis and Hoekstra (1985), and Yatabe (1993) for the Right Node Raising analysis of apparent parasitic gap constructions licensed by Ā-movement; and Levine, Hukari, & Calcagno (2001), Postal (1998, 2001), and Kathol (2001) for subsequent discussion of distribution of true and pseudo-parasitic gaps. The conditions on licensing of parasitic gaps developed in Postal (1994) are as follows. (i) The Dependency Condition: A gap is a parasitic gap only if its existence is dependent on that of another gap, which is a true extraction gap. (ii) The Island Condition: A parasitic gap licensed by a gap G cannot occur internal to an island Σ not containing G, unless Σ is coextensive with the entire parasitic domain.

Feature valuation by sideward movement 59

(cf. (17)), Right Node Raising (cf. (18)), or parenthetical constructions (cf. (19)) which involve CP movement, fail to license parasitic gaps. The parasitic gaps in the associated positions are only apparent, as they fail to observe the Dependency Condition and licensing constraints (see Postal 1994 for discussion). (17) Pseudo-parasitic gaps licensed by CNPS (Authier 1991) a. We suggested ti to our employees [the sort of behavior in question]i. b. We suggested ti to our employees without actually requiring PGi of them [the sort of behavior in question]i. c. We suggested ti to our employees without actually requiring PGi of them [that they wear a tie]i. (18) Pseudo-parasitic gaps licensed by RNR (Postal 1994) a. We could suggest tI to our employees without actually requiring PGi of them, and should suggest ti to them without actually requiring PGi of them [the sort of behavior in question]i.

The condition therefore disallows P-gaps in islands within islands in which there is no real (licensing) gap, as shown below in an example from Postal (1994). (iii) a. an employee whoj your criticism of tj could easily upset tj b. *an employee whoj your criticism of his mother-in-law tj could easily upset tj (iv) The NP Condition: A parasitic gap must be of a category NP. (vi) The Preposition Stranding Condition: A parasitic gap licensed by a T behaves with respect to preposition stranding like an NP-gap and hence, in particular, not like a T-gap. (vi) The Licensing Condition: The licensing category (the extractee from the position of the true gap) of a parasitic gap is an NP. (vii) The Pronominal Condition: Parasitic gaps cannot occur in positions incompatible with definite pronouns. (viii) The Passivizability Condition (see also Postal 1990, 1993): If an NP alternates with a parasitic gap, then the NP must not be inherently unpassivizable. (ix) The Predicate Nominal Condition: Neither a parasitic gap nor its licensing category can be a predicate nominal.

60 Bartosz Wiland b. We could suggest tI to our employees without actually requiring PGi of them, and should suggest ti to them without actually requiring PGi of them [that they wear a tie]i. (19) Pseudo-parasitic gaps in parenthetical constructions (Postal 1994)7 a. Ted is not, asi many who believed PGi fervently tried to convince me ti, a dope fiend and a swindler. b. Ted is not, asi many lawyers who claimed PGi on TV later insisted ti in court, a dope fiend and a swindler. In Postal’s analysis, parasitic gaps licensed by movement of CPs are only apparent and are instances of across-the-board extractions. Thus, the number of related gaps reflects the number of extractions. Postal shows that CNPS can license pseudo-parasitic gaps as in (17), while a that-clause does not produce a gap when topicalized, as shown in (20). (20) *[That they wear a tie]i, we suggested ti to our employees without actually requiring PGi of them.8 Structures like in (20) and other tests, led Postal (1994) to conclude that extractions of that-clauses do not observe conditions on proper parasitic gap 7

8

As-parentheticals parallel with “null”-parentheticals in that they both are construed from main clauses with that-complements. This stems from the fact that also as-parentheticals can appear only with CPs. For instance, expletive constructions can only take CPs as complements in matrix clauses. Examples below are from Postal (1994). (i) a. It appears/seems/turns out that Glen married the nurse. b. Glen has, ei it appears/seems/turns out ti, married the nurse. c. Glen did not, asj it first appeared/seemed tj, marry the nurse. d. Glen has, ask it turns out tk, married the nurse. It appears correct to assume that the ungrammatical construction in (20) (repeated below as (ii)) is not derived from the well-formed construction in (17c) (repeated below as (i)),but that it resists topicalization from its base position ti. (i) We suggested ti to ours employees without actually requiring PGi of them [that they wear a tie]i. (ii) *[That they wear a tie]i, we suggested ti to our employees without actually requiring PGi of them (t’i). The topicalization of the CP from its derived position t’i would have to involve an unjustified vacuous CNPS.

Feature valuation by sideward movement 61

licensing and should not be associated with parasitic gaps of the type given in (4) or (8). Postal specifically claims that pseudo-parasitic gaps are instances of non-coordinate RNR, a conclusion that I will not evaluate here. What is central to the present discussion is that CP-movement fails to license parasitic gaps, which I take to indicate that CPs do not move sideways. While both DPs or CPs can be selected by a number of transitive verbs, only the former ones can be moved sideways from adjuncts. This implies that checking a matrix θ -role does not drive inter-arboreal remerge, in which case both DP and CP arguments would be expected to move sideways. This prediction is borne out by the proposal in (13), according to which only checking φ and case features can license sideward movement. The basis for this is the fact that θ -roles are associated with mapping the predicate-argument relations in syntax (Θ-theory) rather than the computation system (X-bar theory). This helps naturally explain why among many functional differences between φ-features and θ-roles, the latter are blind to probing. Landau (2003: 474) points to the fact that treating thematic relations as formal features on a parity with, say, the carrierhood of morphological features is hardly justifiable, as θ-roles are strongly context dependent, unlike Number, Person, Gender, case, or [±wh]. Instead, Hornstein’s (2001) theory assumes that checking the θ -feature can license movement. However, even accepting Boeckx and Hornstein’s argument that θ-roles are relational just like φ-features on v/T, only the former ones are semantic and context dependent. The hypothesis that thematic considerations suffice to trigger movement demands a more in-depth analysis, since, apparently, θ roles do not participate in Agree, a prerequisite on movement in Chomsky (2000) and much subsequent work in the minimalist syntax.

3.2. DPs, CPs, and null pronouns Postal (1994) offers a number of distributional differences between DPs and CPs, such as, for instance, the asymmetry between CP and DP topicalization. As shown in (21), CPs do not occur in the positions of surface DPs, while topicalized CPs seem to leave their trace in the position where surface DPs are licit. (21) a. I insisted on/*ø that. b. I insisted (*on) [that Sonia attend the interview]. c. Thati, I couldn’t insist on /*ø ti.

62 Bartosz Wiland d. [That Sonia attend the interview]j, I couldn’t insist on/*ø tj. e. Thatk was impossible to insist on/*ø tk. f. [That Sonia attended the interview]l was difficult to insist on/*øtl. These facts indicate that the asymmetric distribution of DPs and CPs must stem from their discrepant sets of default features, which coupled with the fact that they are both recipients of θ-roles limits the possibility to dissimilar values of Person, Number, Gender, and case. Since XP-movement is driven by feature checking, I assume that the asymmetries between CP- and DPmovement stem from the asymmetry in the composition of the φ-features of their heads. If this assumption is on the right track, then DP’s and CP’s differing sets of φ-features can be accounted for by the fact that CPs are headed by null weak pronouns. Crucial evidence in support of this hypothesis is that in a number of languages CPs are headed by overtly realized weak pronouns. For instance, in Polish in a number of contexts CPs must be headed by an overtly realized weak pronoun to “it”. (22) a. [CP To, że Maria nas nie poznała ] zaskoczyło mnie. it COMP Maria us not recognized surprised me ‘That Mary didn’t recognize us, surprised me.’ b. *? [CP Że

Maria nas nie poznała ], zaskoczyło mnie. COMP Maria us not recognized surprised me

(23) a. Przekonało mnie [CP to, że cena była naprawdę korzystna.] it-convinced me it COMP price was really affordable ‘It convinced me that the price was really affordable.’ b. *? Przekonało mnie [CP że cena była naprawdę korzystna.] it-convinced me COMP price was really affordable The idea that null weak pronouns are φ-deficient has been extensively discussed in the literature (see for instance Kayne 1975; Laenzlinger 1998; or Cardinaletti and Starke 1999; among many others). It has been argued that at least one formal feature in the φ-set is underspecified in weak pronouns, most notably it is the Person-feature that is not encoded on the weak pronoun (cf. Grohmann 2000). A further distributional asymmetry between full DPs and CPs is noted in Postal (1994, 2001), and is expected if the latter are null pronominally headed. Namely, as shown in (24), parasitic gaps never occur in the positions of weak definite pronouns.

Feature valuation by sideward movement 63

(24) a. Adam lived in London and Eve lived in London/*it too. b. the city whichi Eve lived in ti c. *the city which Eve lived in tj because Adam lived in PGj The incompatibility between parasitic gaps and what Postal calls “antipronominal contexts” holds true also of any other phrases, not only thatclauses. In the light of the current analysis, the fact that the AdvP as in (25) below does not license a parasitic gap stems from the fact that English adverbials cannot move sideways because they lack the full set of φ-features or case. (25) Extractions from anti-pronominal contexts (Postal 2001: 408) a. Otto dyed his rugs red/*it. b. [What color]i did Otto dye his rugs ti? c. *[That is the color] thati Otto grew to hate ti after dyeing his rugs PG. There is, however, a striking parallelism between the two facts. That-clauses, which are claimed to differ from DPs in a set of their formal features, do not appear in contexts from which they can be remerged sideways.9 Also weak pronouns, which substitute with that-clauses, do not license parasitic gaps. 9

Note that the proposal in (13) opens up the possibility of sideward movement of non-arguments in languages in which such categories have the full φ-set to check. This prediction can receive empirical support from Bondaruk’s (1996: 122 f.) claim that Polish has AP-parasitic gaps. (i) Jak złyi się wydawał ti choć wcale nie był ti ? how angry-3SG.MSC.NOM self seemed though at all not was ‘How angry did he seem to be, though he wasn’t angry at all?’ Polish adjectives inflect for Person, Number, Gender and case, hence analyzing constructions as in (i) in terms of sideward movement cannot be ruled out. However, the alternative analysis of (i) is also possible. Namely, Polish allows for AP deletion in the absence of the Ā-context, as shown in (ii). (ii) a. Wydawał się zły choć wcale nie był zły. seemed self angry though at all not was angry b. % Wydawał się zły, choć wcale nie był zły. c. Wydawał się zły, choć wcale taki (zły) nie był. seemed self angry though at all so (angry) not was. ‘He seemed to be angry, though he wasn’t (angry) at all.’ Though (iic) with the indicative taki ‘so’ which marks the elided AP is better than the construction in (iib) with the elided AP without the indicative, the availability of AP ellipsis in Polish may well show that in Ā-contexts we also

64 Bartosz Wiland In Boeckx’s (2003) sideward movement analysis of donkey anaphora, the lower copy of a bound pronominal variable in the position that corresponds to a parasitic site in PG constructions is weak as well, as shown in (26). (26) a. If John owns a donkey, he beats it. b. If John owns a donkey, he beats a donkey. Note that all resumptive pronouns are weak definite pronouns and English parasitic gaps do not allow for resumptive pronouns (for discussion see Postal 1994: 178). The ill-formedness of (27) was first noted in Chomsky (1982: 57). (27) a. *a man whom everyone who meets him knows someone who likes e b. *a man whom everyone who meets e knows someone who likes him Thus, the parallelism between donkey anaphora and parasitic gaps could seem to be only superficial, yet Boeckx’s analysis shows that both constructions are in fact derived by sideward movement. I will come back to donkey anaphora briefly. The fact that weak pronouns substitute with CPs is significant, as weak and null pronouns do not procrastinate. Consider the distribution of “pluringulars” as analyzed in den Dikken (2001):10 (28) a. The committee has decided to raise the fee. b. The committee have decided to raise the fee. Den Dikken argues that in order to satisfy Full Interpretation, (null) weak pronominal subjects raise to their checking domains in overt syntax and that (null) weak pronouns are invisible to Attract at LF. Along this line, den Dikken accounts for the difference in agreement attraction of the subconstituent in the subject position of the complex DP rather than the DP itself as it is illustrated in (29i–ivb), (30 b) and (31b). Note that den Dikken’s examples in (28) portray plural agreement attraction in casual speech: (29ia,b)

10

deal with the case of the AP ellipsis, analyzed as a parasitic gap in Bondaruk’s work. Since I know of no test that can unambiguously show whether examples as in (i) involve true adjectival parasitic gaps or AP elipses, I accept Bondaruk’s analysis as the support for the proposal in (13) only very cautiously. Den Dikken (2001) calls committee-type complex NPs “pluringulars,” when they are in plural agreement with a finite verb.

Feature valuation by sideward movement 65

is attestable in both American and British English. Examples in (29 ii–v) are poll responses in the study report on number agreement in British and American English offered in Bock et al. (2006: 81).

(iv) a. The dress uniform of the soldiers was hideous. b. The dress uniform of the soldiers were gray.

BrE

(v) a. The jazz player at the nightclubs was paid very badly. b. The jazz player at the nightclubs were very spontaneous.

BrE

(iii) a. The good reputation of the boys was challenged. b. The good reputation of the boys were at stake.

AmE

(ii) a. The relative of the Scottish monarchs disappeared, didn’t he? b. The relative of the Scottish monarchs disappeared, didn’t they?

AmE

(29) (i) a. The identity of the participants is to remain a secret. b. The identity of the participants are to remain a secret.

(30)

a. The identity of the poor is to remain a secret. b. *The identity of the poor are to remain a secret.

(31)

a. The identity of the committee is to remain a secret. b. *The identity of the committee are to remain a secret.

Den Dikken shows that there are NPs in English that are headed by null pronouns, namely nominalized adjectives as the poor in (30) and “pluringulars” as committee in (28) or (31). Agreement attraction is an LF process, in which an embedded DP moves to the agreement domain of the head H and values its features on H.11 Thus, for instance, in (29 i-b) the embedded DP the participants triggers plural agreement with T0 as non-quantificational NPs 11

Bock et al. (2006) extensively discuss noun-verb agreement variations in British and American English. The authors claim that differences between collective agreement in both varieties of English stem from the lexical source. For instance, for educated speakers the word data is correctly analyzed as a plural form, as in (i). (i) the data are very interesting Once data is memorized as a collective noun, the verb can agree with the subconstituent of its subject, as in (ii). (ii) a. I think the interpretation of the data are controversial. b. The last bit of data we collected were on. c. the great thing about these data are (…) (Bock et al. 2006: 99) Unlike committee-type nouns or nominalized adjectives, which lack the full set of φ-features and cannot LF-move, data can attract its FF onto the dominating head.

66 Bartosz Wiland can advance their formal features onto a dominating D-head (den Dikken uses here the ideas put forward by Kayne p.c. and the discussion in Heim and Kratzer 1998 on agreement attraction by QR in wide scope reading). The φ-features of the embedded NP can raise by QR to the D-head of the entire DP subject. Den Dikken takes nominalized adjectives as the poor to be headed by an empty pronoun pro [+Hum; +PL]. (32) [DP the [AP poor [NP pro[+PL]]]] (33) The poor are/*is becoming wealthier in our province. “Pluringulars” such as committee are also pronominally headed. (34) [DP the [NP committee [NP pro[±PL]]]] (35) a. The committee has decided to raise the fee. b. The committee have decided to raise the fee. As both nominalized adjectives and “pluringulars” such as committee-type nouns are headed by null pronouns they cannot move at LF, which accounts for the lack of agreement attraction, as shown in (30b) and (31b). Overt weak pronouns as heads of tensed clausal arguments in languages such as Polish and the distributional parallelism of weak pronouns and CPs, have been taken to imply that CPs are pronominally headed. If this reasoning is on the right track, CPs are expected to resist covert movement on the same grounds as (null) weak pronouns do. This prediction is borne out, for instance, in existential constructions, where CP associates do not move at LF (cf. (36b) and (37b)), while they do move overtly (cf. (36a) and (37a)). (36) a. *There seems to be that you are right beyond doubt. b. That you are right seems to be beyond doubt. (37) a. *There seems to appear that you are telling the truth unconvincing. b. That you are telling the truth seems to appear unconvincing. Pronominally headed CPs lack the full set of φ-features hence they cannot value their features by long-distance Agree, and inasmuch pro’s are unspecified for the Person feature, apparently invariant (underspecified) for Number as in (35), and receive null case. CPs are, thus, illegitimate syntactic objects for sideward movement as they lack the complete φ-set, despite

Feature valuation by sideward movement 67

the fact that they can value the Proposition θ -role on potential probes, or rather be assigned the same θ-role on heads, which follows the scenario that is built on the proposal in (13). Postal’s analysis shows that CP movement does not license genuine parasitic gaps. In the present analysis, the (default) valuation of incomplete uninterpretable features of CPs at LF makes them illegitimate objects for inter-arboreal Remerge. The asymmetries between genuine DP-parasitic gaps and pseudo-parasitic gaps “licensed” by dislocations of that-clauses stemmed from the fact that the latter were supposed to be instances of across-the-board extractions of CPs. However, Hornstein & Nunes (2002) analyze ATB constructions as derived by sideward movement of XPs and verbal heads. I will further show, however, that sideward movement, analyzed as inter-arboreal Remerge, cannot be claimed to license ATB constructions.

4. Donkey anaphora So far it has been advanced that phrases whose heads are underspecified in their φ-features do not qualify for sideward movement, since only φ-complete syntactic objects are goals for matching probes, and feature checking is what drives movement. Since CPs are headed by φ-deficient weak pronouns, they are no match for a probe. In this respect they contrast with φ-complete DPs, which can be remerged in A-positions under Agree, a prerequisite on feature valuation. Since sideward movement is a remerge in an A-position, syntactic objects headed by φ-deficient heads are expected to resist sideward movement. Boeckx’s (2003) analysis of donkey anaphora provides further evidence for the impossibility of moving null and weak pronouns sideways. Donkey sentences as in (38), are instances of bound-variable anaphora in the absence of c-command. The pronoun it is anaphoric to the indefinite NP a donkey, while the indefinite NP does not c-command the pronoun. Thus, the pronoun is outside the scope of the quantifier and cannot be bound. (38) a. If a farmer owns a donkeyi, he beats iti. b. If (∃x,y (x is a farmer, y is a donkey, and x owns y)) → x beats y In Heim’s (1982) analysis, indefinite NPs can be bound at LF by another operator since they are not true quantifiers. Heim argues that conditionals are headed by a universal quantifier corresponding to ‘ALWAYS’, which binds both the indefinite NP and the pronoun, which makes them both cobound variables and secures the universal reading of (38a). However, for

68 Bartosz Wiland the antecedent of a donkey pronoun to be in the restriction of a universal quantifier, it must undergo Quantifier Raising. Otherwise, it effects in the indefinite being bound by the default existential operator that is introduced at the VP level, which results in the existential reading.12 Boeckx (2003) establishes a detailed parallelism between donkey anaphora and parasitic gap constructions and argues that both constructions are derived by sideward movement. Boeckx uses the discussion in Shlonsky (1987) on parallelism between parasitic gaps and donkey anaphors and offers a solution which unifies both phenomena by a derivational rule of construal. Boeckx addresses three parallelisms between both constructions noted in Shlonsky’s work. The first one has to do to with c-command. Namely, the lack of c-command between the antecedent and the donkey pronoun parallels with the anti-c-command configuration between the trace (the true gap) and the parasitic gap. Secondly, both constructions involve strong islands: donkey sentences involve relative or conditional clauses (cf. (39)) and parasitic gaps involve adjunct clauses. Examples in (39)– (42) are from Boeckx (2003: 217). (39) a. *Who did John say that if Mary meets t Bill will be happy? b. *Who did John meet every girl that met t ? Thirdly, in donkey sentences the antecedent is in the restriction of a quantificational head (cf. (40)), while in parasitic gap constructions the head of the relative head noun which embeds the parasitic gap must also be quantificational (cf. (41)). (40) a. Anyone who owns a donkey rides it. b. Some people who own a donkey ride it. c. People who own a donkey ride it. (41) a. He’s a person that anyone who talks to t usually likes t. b. He’s a person that some people who talk to t usually like t. c. He’s a person that people who talk to t usually like t. Boeckx argues that the three parallelisms come naturally if both constructions are derived in a similar manner and when if-conditional clauses are covert (head-initial) relative clauses, as in (42). 12

See Kanazawa (1994) for discussion of the relation between scope and preferences for universal and existential reading in donkey sentences.

Feature valuation by sideward movement 69

(42) a. b.

If John owns a donkey, he beats it. [a donkey [[if John owns t], he beats (it)]

Given the facts above, Boeckx concludes that both constructions involve a sideward movement, and that the donkey pronoun corresponds to a parasitic gap in respective constructions, as shown in (43). (43) a. [a donkey [if John owns t ]] [he beats (it)] : ! z----------m b. [Which paper did [you file t ]] [without reading t ] : ! z------------m Despite the parallelisms noted in Shlonsky’s work, both constructions still differ in three ways: (i) while in donkey sentences the indefinite NP is raised to the restriction of a universal quantifier by QR, the parasitic gap must be licensed by overt Ā -movement; (ii) the antecedent of a donkey pronoun undergoes QR in the if-clause, which is an adjunct (cf. Iatridou 1991), while in parasitic gap constructions, the Ā leg of movement takes place in the matrix clause; (iii) donkey pronouns are weak overt pronouns, while parasitic gaps are traces of sideward movement. Boeckx (2003) convincingly argues that not only are these asymmetries superficial, but they further indicate that both constructions involve similar derivational steps. First, Boeckx takes the raising of the antecedent of the donkey pronoun to be an instance of “overt” QR, along the lines of Fox (2002), Nissenbaum (2000), and Pesetsky (2000), (cf. (44)). (44)

XP 3 if P main clause 3 Spec if ’ : 3 ! if IP ! 6 ! …t… ! ! ! z--------m QR (pronounce t)

70 Bartosz Wiland While the QR targets Spec-if P, it is the lower copy that gets pronounced, which is enforced by the Doubly-Filled Comp Filter (Chomsky & Lasnik 1977). Second, the fact the Ā leg of movement (QR) of the antecedent of a donkey pronoun takes place within the if-clause, which is an adjunct clause, while the wh-movement in parasitic gap constructions takes place within the matrix clause, is accounted for by reprojection. After the antecedent of a donkey pronoun undergoes QR that targets Spec-if P, the if-clause reprojects as a main clause, which is in concert with the Reprojection Principle (Hornstein & Uriagereka 2002). The “main” clause comes out as an adjunct with a silent copy. The derivational steps are shown in (45). (45) If-clauses as “reprojected” main clauses a. if-clause reprojects with the main clause XP XP – main clause 3 ifP X’ ! 3 QP if’ X0 b. Reprojection of the if-clause if P 3 if’ XP – main clause 3 Q if c. the weak NP (a donkey) moves by QR and projects QP 3 Q’ XP – former main clause 3 Q if P Third, while in parasitic gap constructions only the top-most copy is pronounced, two copies of the same item are pronounced in donkey sentences. This is straightforwardly accounted for by the fact that while copy deletion operates under c-command, the antecedent does not c-command the donkey

Feature valuation by sideward movement 71

pronoun, as shown in (44). Since the two copies are not in a c-command relation, its lack does not enforce copy deletion (cf. Nunes 1995). But what seems to be the crux of the matter is that the two copies are dissimilar, the antecedent being a donkey and the foot being lexicalized as it. In Nunes’ (1999) account of the phenomenon, it is argued that this is usually the case whenever multiple pronunciation of a copy takes place. Nunes advances that in such cases only one copy receives a full pronunciation, while the other is reduced, which often results in the cliticization of a deficient copy. Boeckx (2003) rejects Nunes’ analysis, and replaces it with a solution which the proposal in (13) predicts. Boeckx (2003), following Boeckx (2001) and Kayne (2002), points to the fact that the introduction of it into the numeration not only is an instance of an unwelcome counter-cyclic operation, but, what is worse, it violates Inclusiveness.13 Instead, he proposes a configuration in (46) in which resumptive pronouns and their antecedents form an XP constituent at the initial stage of the derivation.14 (46) Floated-quantifier chain (a) and resumptive pronoun chain (b) a. NPi… [QP t’i [ all [ ti ]]] b. NPi… [DP t’i [ D0 [ ti]]] The derivational steps that license parasitic gaps and donkey pronouns are the same: the NP is moved sideways in both derivations, as in (47). (47) a. [Op [If a farmer owns [a donkey]]] [he beats [it [a donkeyi]]] : !: ! z-------------mz---------------m “overt” QR sideward movement b. [Which book] did you read [which booki][before Jack bought [Dnull : [which booki]]] !: z-------------mz------------m wh-movement sideward movement The copies of a donkey and which book are both derived by sideward movement and, given the reprojection and the “overt” character of QR, the asymmetry between donkey anaphora and parasitic gaps narrows down to the 13

14

Inclusiveness is a feature of Chomsky’s (1995a) bare phrase structure theory, which prevents the addition of overt as well as non-overt material (such as features) which is not present in the numeration. The analysis of Ā-binding that involves resumptive pronouns as heads of XP constituents that include their antecedents has been offered in Boeckx (2001).

72 Bartosz Wiland weak overt or null D-head, neither of which moves sideways. The fact that it is the subconstituent that undergoes sideward movement rather than the XP headed by a weak/null pronominal is straightforwardly explained by (13). Since it has been shown that weak and null pronouns are φ-deficient, they are illegitimate candidates for inter-arboreal remerge. This does not, however, disallow their complements from moving sideways. Note that covert movement of CPs, “pluringulars”, or nominalized adjectives is prevented by their inability to value their features on the probe. Such a probe would otherwise have to be φ-deficient itself, and the matrix T0 in tensed clauses is not such a probe. Therefore, no Match and no Agree are possible with φ-incomplete XPs, which therefore do not qualify for sideward movement, an instance of A-movement. 5. Across-the-board constructions ATB, as well as anti-ATB analyses of parasitic gaps have been proposed in numerous studies, for instance in Pesetsky (1982); Huybregts & van Riemsdijk (1985); Haïk (1985); Kardela (1990); Williams (1990); Franks (1992); Munn (1992), (1993), (2001); or Postal (1993). In Postal’s (1994) analysis of pseudo-parasitic gaps, they are treated as instances of ATB extractions of that-clauses. This section shows that just as CPs cannot move sideward, ATB constructions, which canonically may involve the coordination of any XPs, cannot be derived by sideward movement. Again, only those XPs that have the full φ-set are legible for inter-arboreal Remerge, and ATB constructions do not pose such a restriction on the coordinates. Hornstein & Nunes (2002) adopt the approach developed in Nunes (1995, 2001) and Hornstein (2001) where sideward movement licenses both parasitic and ATB constructions. (48) PG construction [Which paper]i did you read ti after Mary recommended ti ? (49) ATB construction [Which paper]i did you read ti and Mary recommend ti ? Postal (1993) and Munn (2001) point out a number of asymmetries between the two constructions, which Hornstein & Nunes account for by arguing that sideward movement licenses parasitic gaps under Last Resort, and ATB constructions by either Last Resort or the Parallelism Constraint.

Feature valuation by sideward movement 73

(50) The derivation of ATB constructions in Hornstein & Nunes (2002): a. b. c. d.

Which paper did you read and Mary recommend? N = {which1, book1, did1, Q1, you1, read1, and1, Mary1, recommend1} K = [andP and [Mary did recommend [which book]] L = read

Which book is copied from K and merges with read of L. Unlike in PG constructions, also did undergoes sideward movement, as both which book and did are attracted by strong features of Q on the matrix clause. (51)

CP 3 [which book]i C’ 3 +Q didk andP 3 [TP you didk read [which book]i] and’ 3 and [TP Mary didk recommend [which book]i]

Hornstein & Nunes (2002) emphasize that unlike in PG constructions, which comprise only referential DPs, any categories can undergo sideward movement in ATB constructions, as in (52). (52) a. b. c. d.

Howi did Deborah cook the pork ti and Jane cook the chicken ti? [How sick]i did John look ti and Betty say he actually felt ti? This is a topic [about which]i you should think ti and I should talk ti. [How many weeks]i did you spend ti in Berlin but want to spend ti in London?15 e. Didi John ti call Mary and Paul ti call Sue?

Under the standard analysis, parasitic gap licensing by sideward movement of DPs takes place under Last Resort. However, Hornstein & Nunes advance that Last Resort is satisfied by checking both formal features and θroles, which accounts for DP copying in PG constructions. On the other 15

Hornstein and Nunes (2002) accept this sentence as grammatical, contrary to Munn (1992).

74 Bartosz Wiland hand, non-DPs that are moved sideways in ATB constructions, as in (53), are not copied under Last Resort as they lack φ and θ. Instead, they are claimed to be moved sideways by the Parallelism Requirement, which does not involve the Greed of θ-role checking. (53) Improper sideward movement in PG constructions under Last Resort (“θ and case”) a. *[How sick]i did John look ti [without actually feeling ti]? {-φ, -case, -θ} b. K = [PRO actually feeling [how sick]] c. L = look d. *L = [look [how sick]i] Hornstein & Nunes (2002: 35) account for improper sideward movement in (53d) in that the verb in L does not assign a θ-role to the AdvP, hence the copying of how sick to L is not licensed by Last Resort. Just as, say, coordinated AdvPs or APs do not receive θ-roles, sideward movement that derives ATB constructions is not licensed by Last Resort but by the Parallelism Requirement (i.e. Ross’ 1967 Constraint on Coordinate Structures). Thus, in Hornstein & Nunes’s analysis, categories that do not undergo formal feature checking or θ -role assignment/checking (including verbs) may still move freely in ATB constructions, as shown in (52). Hornstein & Nunes assume that the Parallelism Requirement is a bare output condition that applies to coordinates, and therefore also XPs that do not move by Last Resort can move by sideward movement to secure the interpretation at the C-I interface. In (53), the copying of neither did nor how sick can be licensed by Last Resort. In (54), the copying of did and how sick is argued to be licensed by the Parallelism Requirement. (54) Sideward movement in ATB’s licensed by the Parallelism Requirement a. [How sick]i didk John look ti and Betty say he actually felt ti? {-φ; -case, -θ} b. K = [andP [TP Betty didk say he actually felt [how sick]i] c. L = [TP John didk look [how sick]i] Hornstein & Nunes conclude that while PG constructions involve the Copying operation licensed by Last Resort, in which case formal feature checking and θ-role assignment/checking takes place, in ATB constructions copying is licensed by either Last Resort or by the Parallelism Requirement,

Feature valuation by sideward movement 75

which does not involve FF checking. The Parallelism Requirement allows for a free application of sideward movement, which accounts for the asymmetries between PG and ATB constructions.16 The analysis of ATB derivations offered by Hornstein & Nunes does not interplay with the proposal in (13), as syntactic objects that undergo sideward movement licensed by the Parallelism Requirement do not value their FF. Within the scenario offered in this paper, for ATB constructions to involve inter-arboreal Remerge, coordinates should be moved by Last Resort, which in turn must be satisfied by exclusive FF checking, as it involves internal Merge. Only DP coordinates could be licensed in this manner, as in (55). (55) a. [Which film]i didk John see ti and Betty criticize ti? {+φ, +case, +θ} b. K = [andP [TP Betty didk criticize [which film]i] c. L = [TP John didk see [which film]i] According to Hornstein & Nunes’s analysis, the copying of PPs in ATB constructions as in (52c) is not licensed by Last Resort but by the Parallelism Requirement, as the verbs in K and L select for a PP, but (“arguably”) do not assign a θ-role to them (p. 35). It is a known fact that PP movement does not license parasitic gaps. Under the current proposal, however, the fact that PPs may or may not be assigned a θ-role directly from the verb that selects them is irrelevant.17 The lack of φ-features of PPs makes them illegitimate objects for inter-arboreal Remerge since they cannot constitute a goal for any probe in the matrix clause. Consequently, no sideward movement can take place, as selectional features do not demand overt checking. In this way, there is no distinction between upward and sideward movement. Hornstein & Nunes (2002: 36) point to the well known fact that even selected phrases can behave like adjuncts, therefore they give rise to weak island effects, as in (56).

16

17

For an extensive discussion of asymmetries between PG and ATB constructions see Postal (1993) and Munn (2001). In fact, the compositional or direct θ-role assignment to PPs depends upon the type of a PP and the argument-predicate relation it enters. For instance, Marantz (1984: 31) has argued that locative PPs receive a θ-role directly from a transitive verb, as they themselves have an internal thematic structure, which is reflected by locative prepositions taking PP complements, as in, say, from under the chair.

76 Bartosz Wiland (56) a. b. c. d.

*Howi do you wonder whether Jane cooked the chicken ti? *[How sick]i do you wonder whether John left ti? *This is a topic [about which]i I wonder whether you want to talk ti? *[How many weeks]i do you wonder whether he spent ti in Berlin ti?

In view of the current analysis, this fact too should not play an important role. Note that, for instance, AdvPs in (56a) and (56b) are post-verbal adverbials, which are not VP-adjoined, but rather occupy a VP internal position (for a discussion on the A-position of adverbs see Stroik 1990).18 They do not have a φ-set to check in a phase higher than their own CP and the Qfeature alone does not suffice to trigger inter-arboreal Remerge. The situation parallels with the impossibility of remerging how sick in (53). As shown in (52), any XPs can be coordinated in ATBs, while only DP arguments can license PGs. (57) a. *This is a tree [under which]i you shouldn’t sit ti after your dog lied ti. {–φ, +θ} b. K = [CP [TP after your dog lied [under which]i] c. L = *[sit [under which]i] (cf. This is a tree under which you shouldn’t sit and your dog lie.) θ-marked non-nominal XPs that fail to license PGs are a challenge to Hornstein & Nunes’s account. In (57) a thematic locative PP is θ-marked directly by the transitive verb that selects it, but it fails to license a PG (cf. Marantz 1984 and Neeleman 1997 on how PPs are selected). There is nothing that disallows copying the thematic PP from K to L in (57) if this operation is licensed by Last Resort. Yet, the PP movement does not give rise to a parasitic gap construction, despite the fact that a locative role should be checked 18

Alternatively, there are good grounds to assume that the A-position of postverbal adverbs is in fact a θ -position. For instance, Larson (1988, 1990) has shown that AdvPs can receive θ-roles once they appear in A-positions, and that adverbials of time are base-generated VP-internally and receive an oblique theta-role. Thematic roles of Manner or Locative (such as there) are mostly assigned to adverbs. Thus, for instance put distributes an internal locative role to either a PP or an AdvP as in (i), and like can assign manner in (ii). (i) a. Jack put the bag down/on the floor/there, etc. b. *Jack put the bag. (ii) a. Jack likes his veal raw. b. *?Jack likes his veal.

Feature valuation by sideward movement 77

by sit in Hornstein & Nunes’s approach. The alternative based on the rule (13) accounts for the ill-formedness of (57) straightforwardly. As already noted, PPs lack the φ-set that needs to be checked overtly in the matrix clause, hence no sideward movement takes place. Since it is possible to coordinate θ-marked but φ-less XPs in ATB’s while the same is not true about PG constructions, an immediate conjecture is that these two constructions do not involve the same derivational steps. While I assume, for the purposes of the present discussion, that PG constructions are licensed by sideward movement, ATB’s are not derived by inter-arboreal remerge but by a different operation. A good candidate here seems to be parallel Merge, which is proposed by Citko (2005) to take place in coordinate structures. I briefly outline Citko’s analysis below to account for the source of asymmetry between parasitic gaps derived by inter-arboreal remerge and ATB’s. In parallel Merge, a syntactic object γ is selected and multi-dominated by α and β: (58) a. External Merge of α (X) with γ (ZP) XP 3 X ZP g g x Z g z b. Parallel Merge of β (Y) with γ (ZP) XP YP 33 X ZP Y g g g x Z y g z The linearization of the terminals x, y, z takes place only at Spell-out, once the ZP has moved overtly from its base position, since there is no reason for linearization of syntactic structures before Transfer. It complies with Chomsky’s (1995b) idea of LCA being responsible for linearization of only pronounceable copies. It also employs Moro’s (2000) proposal that symme-

78 Bartosz Wiland try is legitimate before linearization is required. In ATB constructions, a single interrogative pronoun is selected by parallel VPs as in (59c) and an auxiliary is merged with both vPs, as in (59d). This parallel selection may in fact be the exact instance of the derivationality of the Parallelism Requirement. (59) Derivation of an ATB construction by parallel Merge a. Which paper did you read and Mary recommend? b.

VP 3 V DP 4 g read which paper

c.

VP VP 33 V DP V 4 g g read which paper recommend

d.

TP1 TP2 wo wo vP T vP 3 g 3 DP v’ did v’ DP 4 4 3 3 you v VP VP v Mary 3 3 V DP V 4 g g read which paper recommend

e.

andP 3 TP1 and’ 3 and TP2

Feature valuation by sideward movement 79

f.

CP 3 which C’ paper 3 C andP did 3 TP1 and’ 3 and TP2

Maximize is observed and operation Match takes place as the interrogative DP agrees with both vs and values its φ-set on both heads. Just as the current proposal disallows sideward movement from the derivation in (57), parallel Merge accounts for the grammaticality of the ATB construction in (60). (60) This is a tree [under which]i you shouldn’t sit and I shouldn’t lie ti. While PPs lack the φ-set and cannot undergo inter-arboreal Remerge, they can undergo parallel Merge under standard conditions on movement.19 6. Sideward movement and the features of the probe So far I have been arguing that the availability of XPs to inter-arboreal Remerge is not a matter of selection. Partly for this reason I remain skeptical about the analyses which take sideward movement to be driven by the ΘTheory. If, however, merge-over-move extends to sideward movement, then it seems that sideward movement can be motivated by the fact that it is more economical for a matrix V to merge with a syntactic object that has already been selected and is present on the syntactic workbench (with its features unvalued in the matrix clause). In the minimalist syntax, movement is driven by feature checking. External Merge is driven by selection, whereas internal Merge is an instance of movement, thus it must involve features. Chomsky (1995b) and Bošković & Franks (2000), among others, put forth the idea that LF movement is fea19

Note that the current proposal controls the fact that the PP under which is not going to be merged in parallel with both lie and sit in (57), since parallel Merge licenses coordinate constructions. Instead, the construction in (57) involves an uncoordinated adverbial adjunct clause.

80 Bartosz Wiland ture movement and overt movement is category movement. In the approach that assumes feature movement, category movement is needed independent of feature movement, since, in overt dislocations, categories move for linearization purposes as feature movement itself is invisible for PF. In this sense it is justifiable to claim that category movement is a pied-piping that is necessary for linearization of a syntactic structure. In Chomsky’s (2004) model, features are valued without independent feature movement and Remerge as an instance of internal Merge must involve Agree, while external Merge does not apply to this restriction. I have been assuming in concert with much of the previous work that the valuation of probe’s features is impossible if the goal is φ-incomplete. Likewise, I do not think that the claim that φ-incomplete goals can value probe’s features since the latter co-occurs as φ-incomplete is verifiable, at least not without assuming a massive look-ahead whenever a probe does not select its complement. If, then, A-movement involves full Agree, we may correctly predict that syntactic objects that lack the φ-set resist interarboreal Remerge since sideward movement always takes place from an Āto an A-position. Some empirical evidence for the fact that inter-arboreal Remerge involves full Agree is that covert sideward movement does not exist in grammar. Eliminating φ-incomplete XPs from internal and inter-arboreal Remerge still allows to control for overt movement of PPs to C0 via Spec-vP within a matrix clause in interrogatives. This is made explicit in Chomsky (2008), who claims that such movement is licensed by the Edge Feature (EF) on C0 (and possibly also by the satisfaction of the EPP on v0 en route to C0), and that EF does not undergo valuation. If, as it stems from Richards’ (2007) work, EF differs from EPP, in that the former licenses discourse-interpretive effects and only the latter involves prior Agree, we can correctly predict that φ-less or φ-incomplete XPs do not move to satisfy EF by interarboreal Remerge. For the same reason φ-less XPs never move to Spec-TP, since such phrases do not constitute goals for the probe T0, while they still can move to a clause-initial position higher than TP. Williams (2006), for instance, argues that English T0 has only an optional EPP feature, whose presence demands an agreeing subject in Spec-TP. Otherwise, non-agreeing subjects occupy the specifier of the Logical Phrase (LP), which dominates TP (examples in (61) and (62) are adapted from Williams 2006). The nonagreeing PP-subjects in (61) also fail to undergo Subject-Auxiliary Inversion.

Feature valuation by sideward movement 81

(61) a. Down the hill and over the dale was /*were rolling the ball. b. Down the hill was rolling the ball. c. *Was down the hill rolling the ball? d. In the basement was a good umbrella stand. e. *Was in the basement a good umbrella stand? (62) a. In the basement was a good place to hide. b. In the basement and in the septic tank were good places to hide. b’. In the basement and in the septic tank were good. c. Was in the basement a good place to hide? d. *Was in the basement a good place to hide? (62a) is ambiguous. If in the basement is somewhere which is a good place to hide, agreement holds (cf. (62b)) once the subject involves coordinated DPs (obviously, a single PP will never show agreement). (62b’) shows that the verb in such a context agrees with the subject, not with the postcopular DP, which is here absent. Under this reading, the phrase undergoes SubjectAuxiliary Inversion (cf. (62c)). If the reading is such that ‘in the basement there is a good place to hide’, no Subject-Auxiliary Inversion is possible (cf. (62d)). I take these facts to indicate that T0 does not attract φ-less XPs and if they move to clause-initial position, they move to a projection higher than the TP. 7. Conclusion If sideward movement is a subtype of internal Merge and is available in syntax, it must be subjected to the same general conditions that apply to upward movement. Sideward movement takes place from an Ā to an Aposition, but the question is not whether A-movement can target (potentially) thematic positions, but whether movement is driven by the Θ-theory. I think this is not so, since otherwise the distinction between deriving the generalized argument structure and probing becomes radically blurred, which counters the position that the former is the domain of external Merge, while the latter facilitates Agree and/or dislocation, as argued in Chomsky (2004, et seq.). This problem does not arise once sideward movement is not driven by checking a θ -role (alongside structural case), but by checking the full set of φ-features, which is a prerequisite on Agree. It allows one to account for the fact why only full DPs give rise to parasitic gaps, which generally cannot be licensed by movement of other phrases, including θ-marked CPs or PPs.

82 Bartosz Wiland Acknowledgements I thank Jacek Witkoś for his comments on an earlier draft of this paper. I am also indebted to Kleanthes Grohmann, Malte Zimmermann and the audiences of the DGfS/GLOW student session (Stuttgart 2006), PLM (Poznań 2006), and to an anonymous reviewer for comments and suggestions. Needless to say, all remaining errors remain mine.

References Authier, Jean-Marc 1991 V-governed expletives, case theory, and the Projection Principle. Linguistic Inquiry 22: 721–740. Bennis, Hans and Teun Hoekstra 1985 Gaps and parasitic gaps. The Linguistic Review 4: 29–87. Bobaljik, Jonathan David and Sam Brown 1997 Inter arboreal operations; head movement and the extension requirement. Linguistic Inquiry 28: 345–356. Bock, Kathryn, Anne Cutler, Kathleen M. Eberhard, Sally Butterfield, J. Cooper Cutting, and Karin R. Humphreys 2006 Number agreement in British and American English: disagreeing to agree collectively. Language 82: 64–113. Boeckx, Cedric 2001 Mechanisms of chain formation. Ph.D. dissertation, University of Connecticut. 2003 (In)direct binding. Syntax 6: 213–236. Boeckx, Cedric and Norbert Hornstein 2003 Reply to “Control is not movement”. Linguistic Inquiry 34: 269–380. 2004 Movement under control. Linguistic Inquiry 35: 431–452. Bondaruk, Anna 1996 On parasitic gaps in Polish. Poznań Studies in Contemporary Linguistics 32: 111–125. Bošković, Željko and Steven Franks 2000 Across-the-board movement and LF. Syntax 3: 107–129. 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, Henk van Riemsdijk (ed.), 145–234. Berlin /New York: Mouton de Gruyter. Chomsky, Noam 1982 Some Concepts and Consequences of the Theory of Government and Binding. Cambridge, MA: MIT Press.

Feature valuation by sideward movement 83 1995a

Bare phrase structure. In Government and binding theory and the minimalist program, Gert Webelhuth (ed.), 383 – 439. Cambridge: Blackwell. 1995b The Minimalist Program. Cambridge, MA: MIT Press. 2000 Minimalist inquiries: The framework. In Step by Step: Essays in Minimalist Syntax in Honor of Howard Lasnik, Roger Martin, David Michaels, and Juan Uriagereka (eds.), 89–155. Cambridge, MA: MIT Press. 2001a Derivation by phase. In Ken Hale: A life in language, Michael Kenstowicz (ed.), 1–52. Cambridge, MA: MIT Press. 2004 Beyond explanatory adequacy. In Structures and beyond; The cartography of syntactic structures, Adriana Belletti (ed.), 104 –132. New York: Oxford University Press. 2008 On phases. In Foundational Issues in Linguistic Theory: Essays in Honor of Jean-Roger Vergnaud, Robert Freidin, Carlos P. Otero and Maria Luisa Zubizarreta (eds.), 133–166. Cambridge, MA: MIT Press. Chomsky, Noam and Howard Lasnik 1977 Filters and control. Linguistic Inquiry 8: 425–504. Citko, Barbara 2005 On the nature of Merge: external Merge, internal Merge, and parallel Merge. Linguistic Inquiry 36: 475–497. Den Dikken, Marcel 2001 “Pluringulars,” pronouns and quirky agreement. The Linguistic Review 18: 19–41. Fox, Danny 2002 Antecedent-contained deletion and the copy theory of movement. Linguistic Inquiry 33: 63–96. Franks, Steven 1992 A prominence constraint on null operator constructions. Lingua 88: 1–20. Gärtner, Hans-Martin 2002 General Transformations and Beyond. Berlin: Akademie-Verlag. Grohmann, Kleanthes K. 2000 Towards a syntactic understanding of prosodically reduced pronouns. Theoretical Linguistics 26: 175–210. Haïk, Isabelle 1985 The syntax of operators. Ph.D. dissertation, MIT. Heim, Irene 1982 The semantics of definite and indefinite noun phrases. Ph.D. dissertation, University of Massachusetts, Amherst. Heim, Irene and Angelika Kratzer 1998 Semantics in Generative Grammar. Oxford: Blackwell. Hornstein, Norbert 1999 Movement and control. Linguistic Inquiry 30: 69–96. 2001 Move! A Minimalist Theory of Construal. Oxford: Blackwell.

84 Bartosz Wiland Hornstein, Norbert and Jairo Nunes 2002 On asymmetries between parasitic gap and across-the-board constructions. Syntax 5: 26–54. Hornstein, Norbert and Juan Uriagereka 2002 Reprojections. In Derivational and Explanation in the Minimalist Program, Samuel D. Epstein and T. Daniel Seely (eds.), 106 –132. Oxford: Blackwell. Huybregts, Riny and Henk van Riemsdijk 1985 Parasitic gaps and ATB. In Proceedings of NELS 15: 168 –187. GLSA, University of Massachusetts, Amherst. Iatridou, Sabine 1991 Topics in conditionals. Ph.D. dissertation, MIT. Kardela, Henryk 1990 Paths. An analysis of parasitic gap phenomena in English and Polish. In Further Insights into Contrastive Analysis, Jacek Fisiak (ed.), 389– 410. Amsterdam: John Benjamins. Kathol, Andreas 2001 On the nonexistence of true parasitic gaps in Standard German. In Parasitic Gaps, Peter W. Culicover and Paul M. Postal (eds.), 31–338. Cambridge, MA: MIT Press. Kayne Richard S. 1975 French Syntax. Cambridge, MA: MIT Press. 2002 Pronouns and their antecedents. In Derivation and Explanation in the Minimalist Program, ed. Samuel D. Epstein and T. Daniel Seely, 133–166. Oxford: Blackwell. Kanazawa, Makoto 1994 Weak vs. strong reading of donkey sentences and monotonicity inference in a dynamic setting. Linguistics and Philosophy 17: 109 –158. Laenzlinger, Christopher 1998 Comparative Studies in Word Order Variation: Adverbs, Pronouns and Clause Structure in Romance and Germanic. Amsterdam: John Benjamins. Landau, Idan 2003 Movement out of control. Linguistic Inquiry 34: 471–498. Larson, Richard K. 1988 On the double object construction. Linguistic Inquiry 19: 335–392. 1990 Double objects revisited: Reply to Jackendoff. Linguistic Inquiry 21: 589–632. Levine, Robert D., Thomas E. Hukari, and Michael Calcagno 2001 Parasitic gaps in English: Some overlooked cases and their theoretical implications. In Parasitic Gaps, Peter W. Culicover and Paul M. Postal (eds.), 181–222. Cambridge, MA: MIT Press. Marantz, Alec 1984 On the Nature of Grammatical Relations. Cambridge, MA: MIT Press.

Feature valuation by sideward movement 85 Moro, Andrea 2000 Dynamic Antisymmetry. Cambridge, MA: MIT Press. Munn, Alan 1992 A null operator analysis of ATB gaps. The Linguistic Review 9: 1–26. 1993 Topics in the syntax and semantics of coordinate structures. Ph.D. dissertation, University of Maryland. 2001 Explaining parasitic gap restrictions. In Parasitic Gaps, Peter W. Culicover and Paul M. Postal (eds.), 369–392. Cambridge, MA: MIT Press. Neeleman, Ad 1997 PP-complements. Natural Language & Linguistic Theory 15: 89 –137. Neeleman, Ad and Hans van de Koot 2007 Dutch scrambling and the nature of discourse templates. To appear in Journal of Comparative Germanic Linguistics 11. forthc. Theta theory; In defense of Plan B. Nissenbaum, Jon 2000 Explorations in covert phrase movement. Ph.D. dissertation, MIT. Nunes, Jairo 1995 The copy theory of movement and linearization of chains in the minimalist program. Ph.D. dissertation, University of Maryland. 1999 Linearization of chains and the phonetic realization of chain links. In Working Minimalism, Samuel D. Epstein and Norbert Hornstein (eds.), 217–249. Cambridge, MA: MIT Press. 2001 Sideward movement. Linguistic Inquiry 31: 303–344. 2004 Linearization of Chains and Sideward Movement. Cambridge, MA: MIT Press. Pesetsky, David 1982 Paths and categories. Ph.D. dissertation, MIT. 2000 Phrasal Movement and Its Kin. Cambridge, MA: MIT Press. Pesetsky, David and Esther Torrego 2001 T-to-C movement: causes and consequences. In Ken Hale: A Life in Language, Michael Kenstowicz (ed.), 355–426. Cambridge, MA: MIT Press. Postal, Paul M. 1990 Some unexpected English restrictions. In Grammatical Relations: A Cross-Theoretical Perspective. Katarzyna Dziwirek, Patrick Farrell, and Errapel Mejías-Bikandi (eds.), 365–385. Stanford, CA: CSLI Publications. 1993 Parasitic gaps and the across-the-board phenomenon. Linguistic Inquiry 24: 735 –754. 1994 Parasitic and pseudo-parasitic gaps. Linguistic Inquiry 25: 63–117. 1998 Three Investigations of Extraction. Cambridge, MA: MIT Press. 2001 Missing parasitic gaps. In Parasitic Gaps, Peter W. Culicover and Paul M. Postal (eds.), 403–418. Cambridge, MA: MIT Press.

86 Bartosz Wiland Reinhart, Tanya 2002 The Theta-system: An overview. Theoretical Linguistics 28: 229–290. Richards, Marc 2007 Interpretive asymmetries at the edge. Paper presented at GLOW Colloquium, University of Tromsø, April 2007. Rizzi, Luigi 1997 The fine structure of the left periphery. In Elements of Grammar, Liliane Haegeman (ed.), 281–331. Dordrecht: Kluwer. Ross, John R. 1967 Constraints on variables in syntax. Ph.D. dissertation, MIT. Shlonsky, Ur 1987 Donkey parasites. In Proceedings of NELS 17, 569 –579. Amherst, MA: GLSA Publications. Stroik, Thomas 1990 Adverbs as V-sisters. Linguistic Inquiry 21: 654–661. Williams, Edwin 1980 Predication. Linguistic Inquiry 11: 203–238. 1990 The ATB theory of parasitic gaps. The Linguistic Review 6: 265–279. 1994 Thematic Structure in Syntax. Cambridge, MA: MIT Press. 2006 Subjects of different heights. Proceedings of FASL 14: 409–429. Ann Arbor: Michigan Slavic Publications. Yatabe, Shuichi 1993 Scrambling and Japanese phrase structure. Ph.D. dissertation, Stanford University.

On Incrementality, overt agreement, theta-roles and Spec,Head relations in the phase-based framework Iván Ortega-Santos

Recent developments in syntactic theory posit the existence of a Long Distance Agreement mechanism, arguing that there is no such thing as a grammatically significant Spec,Head configuration (e.g., Chomsky 2004, 2008, etc.). This claim is a hallmark of phase-based syntax and, consequently, its evaluation is relevant to our understanding of this framework. The issue is particularly interesting in light of recent arguments in the opposite direction (e.g., Koopman 2006; or Franck, Lassis, Frauenfelder and Rizzi 2006, a.o.). For instance, there is a crosslinguistic tendency for moved elements to trigger agreement, as opposed to in situ ones, a fact that calls for an explanation within this framework (Chomsky 2008). Furthermore, the issue of how subject Specs check their theta-roles under the assumption that these are features (e.g., Bošković 1994, Bošković and Takahashi 1998, Hornstein 2003 and Lasnik 1995) remains unexplained in a system that dispenses with grammatically significant Spec,Head configurations. In view of these and related facts, here it is argued, that conceptual arguments against Spec,Head relations (e.g., Chomsky (2008) can be circumvented and that Spec,Head relations exist in the system, though not in the traditional guise. In particular, it is argued that under a Multiple Spell-Out system (Uriagereka 1999 and 2008), when combined with the phase-based system, there can be checking relations in the Spec,Head configuration, though not probing of the Spec by the head under m-command. This is shown to solve the problems that the phase-based framework faces. With regard to morphological agreement, the resulting framework predicts that some languages may sanction Spec,Head relations as the domain of morphological agreement. Nonetheless, it is not clear why there appear to be very few cases in natural language where Long-Distance Agreement is sanctioned as said domain. It is argued that within the above framework which adopts grammatically significant Spec,Head configurations, performance factors conspire to achieve this result. In particular, the differences in agreement morphology found across languages, depending on whether the probe goal relation is established locally (in the Spec,Head relation) or via

88 Iván Ortega-Santos a Long-Distance Agreement, are argued to be related to the workings of socalled Incrementality (cf. Barlow 1992). Section 1 presents the relevant crosslinguistic generalizations concerning agreement paradigms and previous approaches to these generalizations within the phase-based system. Section 2 develops the current proposal concerning Spec,Head relations. Section 3 focuses on the role of Incrementality in agreement paradigms across languages.

1. On crosslinguistic variation in agreement paradigms It is often claimed that one can find languages where agreement follows the paradigms in (1) and (2), but not the one in (3) (where this is illustrated in terms of probes (P) and goals (G)); the presence of overt agreement is indicated explicitly, whereas lack of any indication means a corresponding lack of agreement): (1) (2) (3)

Moved goal G Povert agreement G Povert agreement GP

/

In situ goal Povert agreement G PG Povert agreement G

Distribution across languages √ √ *

The crosslinguistic validity of this observation is present in some way or other in Barlow (1992), Chomsky (2004), Corbett (1979), Franck, Lassis, Frauenfelder and Rizzi (2006), Koopman (2006), Manzini and Savoia (1998), Moravcsik (1978), Park (2006) and Samek-Lodovici (2002), among others.1 The following data from the Italian dialect of Ancona and French illustrate the pattern in (2), which will be of particular interest for the present discussion: (4)

1

Italian Dialect of Ancona (Cardinaletti 1997a: 38f.) a. Questo, lo fa / *fanno sempre i bambini. this-ACC it-ACC does / do always the children b. Questo, i bambini lo fanno / *fa sempre. this-ACC the children it-ACC do / does always

Note that the existence of language-particular variation in the realization of overt agreement is beyond question. Still, the correlation in (1–3) seems to be robust, a fact that we would like to explain. Therefore, I do not take such crosslinguistic variation to question the relevance of the observation and, consequently, the general approach explored in this research.

On Incrementality, overt agreement, theta-roles and Spec,Head relations

(5)

89

French (Boeckx 2004: 23) a. Jean a vu / *vue la fille Jean has seen / seen.AGR.FEM the girl ‘Jean saw the girl.’ b. Quelle fille Jean a(-t-il) vue / *vu? which girl Jean has-he seen.AGR.FEM / seen ‘Which girl did Jean see?’ c. Cette fille a été vue / *vu this girl has been seen.AGR.FEM / seen ‘The girl was seen.’

A number of (often unrelated) languages follow this pattern, e.g., Arabic, Hungarian, and some African languages (see Samek-Lodovici 2002 for an overview), a fact that provides further evidence for the validity of the above generalization. Such facts have played a prominent role in syntactic theorizing, because they provide an argument for the role of Spec,Head relations (cf. Kayne 1989). Within an approach to syntax which assumes grammatically significant Spec,Head configurations, the relation between the P and the G at the point of Spell-out is different in the in situ and the movement counterparts. Hence, it is natural to consider the Spec,Head relation the domain of (morphological) agreement (Kayne 1994), at least in the relevant languages. In keeping with this view, closely-related discussions in the literature converge on the necessity for such grammatically significant Spec,Head relations, both in theoretical and experimental research (e.g., Koopman 2006; and Franck, Lassis, Frauenfelder and Rizzi 2006, respectively, among many others). The picture changes once one assumes Long Distance Agreement (LDA; Chomsky 2001 and later work). Within such a system, there is no distinction between the moved and the in situ version in terms of the relation of P and G at the point of Spell-out; things move or do not move, but the agreement relation between P and G is the same no matter what. Therefore, such agreement asymmetries do not follow from the system, in contrast to an approach in terms of grammatically significant Spec,Head configurations or checking under m-command. In fact, as Chomsky (2008: 13) states, as the role of Spec,Head relations is diminished, this calls for a reconsideration of a number of issues, agreement being the most relevant one. Theta roles pose a similar puzzle. Specifically, the literature has provided evidence for the claim that theta roles are features (e.g., Bošković 1994; Bošković and Takahashi 1998; Hornstein 2003; and Lasnik 1995). For instance, Bošković and Takahashi (1998) provide evidence that scrambled

90 Iván Ortega-Santos sentences in Japanese involve obligatory LF movement of arguments basegenerated in non-theta-positions to the positions where they receive thetaroles. Under this view, theta-roles are formal features capable of driving movement (Bošković and Takahashi 1998: 349 –350): (6) a. Canonical word order John-ga [Mary-ga sono hon-o katta to] omotteiru. John-NOM Mary-NOM that book-ACC bought that thinks ‘John thinks that Mary bought that book.’ b. Scrambling Sono hon-o John-ga [Mary-ga t katta to] omotteiru. The derivation corresponding to (6) is the following under this view, where (7) corresponds to narrow syntax, (7) to PF and (7) to LF: (7) a. [IP Sono hon-o [IP John-ga [CP[IP Mary-ga [VP[V katta]]] to] omotteiru]] that book-ACC John-NOM Mary-NOM bought that thinks b. sono hon-o John-ga Mary-ga katta to omotteiru c. [IP John-ga [CP[IP Mary-ga [VP sono hon-o [V katta]]] to] omotteiru] This analysis is particularly appealing in that it is not only empirically wellsupported (see Bošković and Takahashi 1998 for details) but also it makes scrambling consistent with the Last Resort principle. In the present context, an understanding of theta-roles as features is relevant in that it is not clear how a(n externally-merged) subject Spec may check its theta-role within vP. Similar issues arise for any Spec that is not c-commanded by the head expected to license it (e.g., certain phrases base-generated in A-bar positions, e.g. topic phrases or wh-phrases like whether).2 2

Recent trends in syntactic theory put forward the idea that a probing relation is involved in external merge. For instance, Pesetsky and Torrego (2006) follow this view (see their Vehicle Requirement on Merge). Still, it is not clear that this would allow for a licensing/checking relationship between a head and a Specifier-to-be (e.g., a Spec about to be externally merged). Pesetsky and Torrego provide evidence/claim that: i. external merge does not result in feature valuation / agreement (see Pesetsky and Torrego 2006: 2); ii. the attempt at feature valuation takes place after merge has already taken place (see Pesetsky and Torrego 2006: n. 26 and related discussion). Given this, it is not clear how this kind of probing would help a phase-based framework when dealing with licensing relations between Specs and a heads, because the feature valuation process would take place in a Spec,Head configuration even in the cases where the Spec undergoes external merge.

On Incrementality, overt agreement, theta-roles and Spec,Head relations

91

In what follows, I provide an unified approach to the puzzle posed by morphological agreement, theta roles and externally-merged Specs in A-bar positions. Before doing so, I discuss a number of attempts in the literature to deal with said agreement facts. 1.1. Previous approaches to crosslinguistic variation in agreement paradigms within the phase-based LDA framework The literature includes (at least) the following attempts to deal with the above agreement facts in (2) within the LDA-based system: i.

the data are not a reflex of the Spec,Head relation, but of the way this configuration is established: internal merge as opposed to external merge (Chomsky 2004) ii. Spec,Head configurations entail intermediate steps which allow for more direct licensing/checking relations than LDA, a fact reflected in the morphology. E.g., in the case of subject licensing in English existentials, LDA between T and the in situ subject takes place indirectly via agreement between T and the head of the VP phase which, in turn, has agreed with the subject. This indirect agreement would be forced by the Phase Impenetrability Condition.3 In contrast, subjects that end up in a Spec,Head configuration agree directly with the probe when escaping the VP phase (Legate 2001) iii. rich agreement in the movement counterpart correlates with the presence of an agreement projection, absent in the case of the in situ counterpart (Boeckx 2004) iv. Spec,Head configurations entail double-checking the relation between the probe and the goal, hence their stricter agreement requirements (Frank et al. 2006)4 3

4

According to the Phase Impenetrability Condition (PIC), in a phase α with head H, the domain of H is not accessible to operations outside α, only H and its edge are accessible to such operations; the domain of H is the sister of H, and the edge of H is a hierarchy of one or more Specs (Chomsky 2000: 108). The agreement asymmetries under discussion are found in contexts other than verbal agreement morphology, (e.g., internal to DP’s or PP’s as shown by Hornstein et al. 2005: 119). As a consequence, an analysis contingent on expletive subjects (Cardinaletti 1997b), though relevant, is not general enough to capture the paradigm. Still another alternative approach to the agreement paradigms under consideration is to reject LDA and adopt a generalized Spec,Head analysis of (all) agree-

92 Iván Ortega-Santos Although these approaches are valuable, they have some limitations. Specifically, (i) seems to be mere coding. In turn, (ii) faces the challenge that such agreement asymmetries do not seem to correlate with the opacity effects caused by the PIC. For example, in situ subjects of transitive verbs are available for direct agreement according to the PIC and, nonetheless, might correlate with poor agreement in certain languages (cf. (4)). (iii) is at odds with the rejection of agreement projections within the minimalist enterprise, and seems slightly ad hoc. Lastly, the solution in (iv), even though it would generate the data correctly, does not provide any explanation of how such a Spec,Head checking mechanism would work in the context of the conceptual arguments made by Chomsky (2008, et seq.) against this checking configuration (see below for details). So the puzzle that agreement paradigms posit for the phase-based system is real. Somewhat ironically, the traditional view that there are checking/licensing relations in the Spec,Head configuration would solve the problem. The purpose of the next section is to address this concern, showing that Spec,Head configurations can indeed be grammatically significant. 2. Phase-based syntax allows for grammatically significant Spec,Head relations According to Chomsky (2000 et seq.), the licensing of in situ goals takes place via LDA. It is not clear how this approach can be made compatible with crosslinguistic tendencies for moved elements to trigger agreement as opposed to in situ ones, in spite of the fact that the literature includes a number of approaches to this issue (see the previous section). Last but not least, theta-roles understood as features and phrases base-generated in Abar positions pose the same kind of problem. In what follows I develop a unified approach to these issues. Specifically, I argued that there can be checking relations in the Spec,Head configuration, though not the regular probing of the Spec by the head under m-command. The point of departure of this proposal is Chomsky’s (2008: 13) observation that ‘for minimal computation, the Probe should search the smallest domain to find the Goal: its c-command domain. It follows that there should be no m-command, hence no Spec,Head relations, except for the special ment configurations (Koopman 2003). This entails a radical readjustment/reconsideration of a number of standard structures and derivations. As a consequence, I abstract away from this possibility.

On Incrementality, overt agreement, theta-roles and Spec,Head relations

93

case where the Spec itself can be a Probe.’ The same intuition can be found in Chomsky (2005: 14) and in Chomsky’s (2004: 114) analysis of externally merged expletives,5 where the expletive probes the head of the projection hosting it. As hinted already, a system allowing for grammatically significant Spec,Head relations is able to deal with all these puzzles successfully. Chomsky’s framework is open to this possibility, but it is still unclear in exactly which contexts Specs would qualify as probes. Chomsky characterizes probes in the following way: i. probes are/have uninterpretable features (e.g., Chomsky 2001: 6) ii. only heads can be probes (e.g., Chomsky 2004: 109) iii. only phase-heads drive operations (e.g., Chomsky 2008: 11) Under the assumptions that (a.) arguments bear uninterpretable case features and (b.) arguments are phases (e.g., they have a phase head capable of driving operations), it follows that arguments in Spec positions can be probes.6 The only condition missing is that Specs would have to be heads. Uriagereka (1999) provides conceptual reasons in favor of this view. Uriagereka’s (1999) Multiple Spell-Out proposal addresses some shortcomings of the Linear Correspondence Axiom as originally formulated (Kayne 1994). Kayne’s proposal concerning linearization includes a Base step and an Induction step: (8)

Linearization Procedure for Terminal elements a. Base: If X asymmetrically c-commands Y, X precedes Y. b. Induction: If X is dominated by Z, and Z precedes Y, X precedes Y.

According to Uriagereka, the stipulative Induction step is unnecessary. The logic of the Multiple Spell-out proposal is to spell-out ZP prior to connecting it to the structure which is still live in the derivation. As a consequence, the issue that motivates the Induction step does not arise. In Uriagereka’s proposal, all Specs are akin to heads (cf. also van Gelderen 2004 for relevant discussion). For current purposes, this means, if taken

5

6

See Bošković (2002) for evidence that expletives are externally merged into the position where hey surface. See Chomsky (2001: 14) for evidence that DPs are phases, and see Soltan (2003) for evidence concerning PPs. See also Hornstein (1995) for independently motivated arguments that PPs have uninterpretable features (other than those valued by their arguments).

94 Iván Ortega-Santos at face value, that Specs qualify as probes (given that they fulfill all the requisites) and, consequently, Spec,Head relations may exist in the system. Chomsky’s argument against grammatically significant Spec,Head relations is that checking would take place under m-command as opposed to ccommand, in opposition to the conceptually desirable need to minimize the computation (cf. Chomsky 2008: 13 quoted above). However, once one adopts Uriagereka’s proposal as interpreted here, a Spec can probe the head of the projection hosting it under c-command. Crucially, this is consistent with the desire to minimize computations by having probing target the smallest domain, namely, the c-command domain. Note that in a phase-based system it is labels that probe. In determining what a structure’s label is (or whether a syntactic unit is a probe), minimal search is key: ‘The label of an SO must be identifiable with minimal search, by some simple algorithm’ (Chomsky 2008: 11). Specifically, when merge takes place, ‘for optimal computation, one member of the pair must be available with no search. It must, therefore, be the head H of the construction α under consideration, α ={H, XP}. Call H a probe P, which seeks a goal G within XP; P = H c-commands G, but that is a consequence of minimal search.’ (Chomsky 2004: 113). Similarly, Chomsky (2007: 16) approaches the issues as follows, noting that the notion “label” only plays an expository role: ‘In constructions of the form H-XP (H a head), minimal search conditions determine that H is the designated element (label) that enters into further operations; any other choice requires deeper search. At the phase level, H will be the probe, for the same reasons.’ To my knowledge, the idea of minimal search is not made explicit anywhere in Chomsky’s work. However, these passages and others suggest that for phrases to become probes one needs to apply some sort of costly ‘resolution rules’, akin to those found for coordinate structures. When the computational system merges two syntactic units, it needs to take into account all the elements within said units to see what their probing features turn out to be, where new phrases can be merged, etc. By contrast, in the case of heads, one has that information readily available. It is then more costly to have a phrase drive computations, rather than a head –a fact that is taken to underlie why probing is restricted to the latter. This picture changes slightly once an independently motivated Multiple Spell-Out (MSO) framework (Uriagereka 1999) is adopted. While this framework explains why Specifier-internal components become inactive (deriving Huang’s 1982 Condition on Extraction Domains; cf. Hornstein, Lasnik and Uriagereka 2007 for a treatment of apparent counterexamples), the entire objects as such are available to further operations, or their exact

On Incrementality, overt agreement, theta-roles and Spec,Head relations

95

place/role within the phrase-marker would be lost. Specs, that is to say, derived heads, can be argued to undergo independently motivated feature resolution processes, to determine, for instance, specific agreement specifications in corresponding heads.7 This allows the system to have relevant (featural) syntactic information readily available in spite of the fact that these are phrases/Specs. As a consequence of this feature resolution process, the burden of search when merging such Specs is reduced, and so feature search within a spelled-out Spec is minimized. In the spirit of Chomsky’s view that minimization affects labeling/probing, it is then possible to conclude that the label of spelled-out Specs is readily identifiable following the feature resolution mechanism. Due to the fact that operations are driven by labels (which in this system are probes), this means Specs can act as probes – with all the advantages that may bring into the picture (e.g. for the aforementioned crosslinguistic agreement paradigms, theta-roles or for traditional ‘antecedent government’ effects, which at this point have no treatment: the ‘Spec’ can probe into a relevantly local structure to associate, aka ‘γ-mark’, a corresponding trace/copy in familiar conditions, not described in terms of mere cyclicity).8, 9

7

8

9

Independent evidence for this feature resolution process within phrases can be found in agreement attraction errors, e.g., cases like a bunch of aristocrats live(s) here. See Den Dikken (2000) for discussion and references. I thank Juan Uriagereka for pointing this out to me. One counterargument to this view runs as follows: Chomsky (2006: 16) claims that internally merged Specs cannot act as probes due to the fact that the moved XP is a “discontinuous element”, whereas the unit that the Spec will combine with is “unmodified and unary”. Still, heads can also be discontinuous under head-movement, just like phrases. Moreover, even if internally merged Specifiers are discontinous, they will be minimized under the present view. Furthermore, the validity of this reasoning also depends on how one conceives movement (e.g an approach to move in terms or ‘remerge’ a priori could avoid the feature resolution costs associated with “discontinuity”). One may wonder what prevents, within the current system, which boy from probing the embedded CP in the following configuration: i. [vP which boy asked [CP what you eat] Say this probing relations actually takes place, one possibility is to assume that derivational crashes free up other derivational options and, consequently, the sentence finally converges. Still another option is to assume that at the point which boy is introduced in the structure, the embedded CP is already syntactically inactive.

96 Iván Ortega-Santos The probing possibilities of the phase-based framework and of the framework put forward in this research can be illustrated in the following way, where the probing domain of the head/probe is indicated in grey and the probing head is indicated in bold: (9)

Probing possibilities within the phase-based system YP 2 ZP Y’ 1 1 X … Y …

(10) Probing possibilities within the MSO system a.

YP 2 ZP Y’ 1 1 X … Y …

b.

YP 2 ZP Y’ 1 1 X … Y …

The scenario in (10) is made possible by the fact that under the current system ZP is a derived head, which can act as a probe.10 Note, finally, that a Spec might probe a head and, nonetheless, that very head would label the resulting structure. I take this not to introduce a new assumption, but rather to follow from independent factors – namely, the subcategorization restrictions imposed by the head that merges with the resulting structure. In fact such restrictions might in certain cases allow/ force the Spec that is probing a head to label the structure (see Chomsky 2008: 12 or Iatridou et al. 2001 and Donati 2006 for discussion). At the same time, after such restrictions are met (early on) in the course of the 10

An approach in terms of Bare Phrase Structure, where both Specs and the unit formed by a head and its complement are heads predicts that both qualify as probes. E.g., Epstein and Seely (2006) claim that the unit formed by a complement and its head probes its Spec, whereas Heinat (2006) claims that phrases/ Specs probe. Nonetheless, it is not clear how these approaches relate to minimal search considerations, in contrast to the MSO analysis developed above. Furthermore, there is an ongoing discussion in the literature as to whether the phrase / head distinction should exist in the theory or not (e.g., see Carnie 2000). The rendition of the present discussion in terms of minimal search makes such ongoing debate orthogonal to the current concerns. As a consequence, I abstract away from it.

On Incrementality, overt agreement, theta-roles and Spec,Head relations

97

derivation, in principle nothing would prevent a Spec from driving a relabeling process. This is exactly what Hornstein & Uriagereka (1999) argue for, in terms of their independently motivated process of ‘reprojection’.11 To sum up the discussion so far, the LDA mechanism, a hallmark of phase-based syntax (Chomsky 2000, et seq.), fails to provide a proper understanding of agreement asymmetries across languages. The issue is even more dramatic, given that, in general, licensing relations between a Spec and the head of the projection hosting it remain cryptic (e.g., this would be the case of an externally merged subject Specs in need of a theta role). Under the assumption that Specs in fact become heads (cf. Uriagereka 1999 and 2008) and that probes are heads (Chomsky 2004: 119), it follows that Specs can be probes, as suggested, for instance, in Chomsky’s (2004) analysis of externally-merged expletives. In this way, we can successfully address the puzzle that agreement paradigms, theta-role checking and related facts pose for the phase-based-system. To be explicit about the agreement facts in (1)–(3), one could assume derivations of the pattern in (2), that is to say, derivations including elements triggering agreement under movement to proceed as in Frank et al. (2006), where it is argued that there is an LDA relationship followed by a Spec,Head Agreement relationship. This can be illustrated as follows: (11) a. 1st Agreement relation (LDA)

X’ 2 X … DP

b. 2ndAgreement relation (Spec,H) XP 2 DPi X’ 2 X … ti

Given that the latter case corresponds in some sense to double-checking the agreement relationship, the presence of agreement would be more pervasive under movement than under lack of movement.12 This would entail that movement and (some of the components of) LDA overlap. 11

12

According to Hornstein and Uriagereka (2002), reprojection is a process whereby a phrase marker’s label changes in the course of the derivation. For instance, in their system this allows binary quantifiers to take scope (at LF). See Hornstein and Uriagereka (2002) for details. Within the phase-based system adopted in this research, this double-checking relationship would only be possible under the assumption that the first agreement

98 Iván Ortega-Santos Alternatively, it could be the case that movement and LDA are independent from one another (cf. Bobaljik and Wurmbrand 2005; Lidz and Williams 2002; and Park 2006). Under this view, agreement would take place after movement (cf. Park 2006). E.g., a probe P matches a goal G. Subsequently G moves to the Spec position of the P and, afterwards, the P and G agree. If true, the reason why the patterns in (1) –(3) emerge could be the following: Under the minimalist assumption that the Language Faculty is in some sense optimal, it seems reasonable to expect the grammar to sanction the local or optimal Spec,Head configuration as the domain of agreement at least in some languages, as opposed to the non-local LDA relationship, which can be assumed to be less optimal; hence the contrast between (2) and (3), while allowing for (1) (cf. the next section for a detailed proposal). I leave the issue of deciding among these two approaches open, noting that both are plausible alternatives that can deal with the data successfully. Crucially, though, the success of both implementations is contingent on the availability of checking relations under Spec,Head configuration. Within the resulting system, externally-merged subject Specs can check their thetaroles and externally merged A-bar phrases can be licensed by the head of the projection hosting them. In the next section, some remaining issues concerning agreement are addressed. In particular, the logic developed so far provides an argument for the existence of Spec,Head relations in natural language. This allows for the Spec,Head configuration to be sanctioned as the domain of agreement in certain languages. Still, one would like to derive the fact that the LDA configuration does not seem to be singled out or privileged in the same way. I have mentioned briefly above that locatility plays a role in this state of affairs. Below, it will be argued that this follows from the computational dynamics of the interfaces; more specifically, from the incremental nature of the parser/production system (Levelt 1989) and its interaction with the syntax (e.g., Bock et al. 1992; Phillips 1996; and Ferreira 1996).

relation does not result in the valuation of the phi-features on the probe and the case features of the goal (cf. the Activity Condition).

On Incrementality, overt agreement, theta-roles and Spec,Head relations

99

3. On the role of Incrementality in agreement paradigms With regard to regularities in agreement paradigms across languages, it has been argued that such data provide evidence for the redefined Spec,Head relations proposed above. The purpose of this section is to further investigate what underlies the fact that ‘poor’ agreement tends to be associated with LDA. The intuition presented before is that locality conspires to derive these facts. Nonetheless, agreement is agreement, whether local or long-distance in nature. It will be argued that this state of affairs follows from the incremental nature of the production system. Recent research has provided evidence for i. the fact that overt agreement is a morphological phenomenon, not a strictly syntactic one (though it is contingent on the output of syntax, e.g., Sigurðsson 2006); ii. the incremental nature of production (e.g., Bock et al. 1992, Phillips 1996 and Ferreira 1996). With regard to (i.), this literature shows that the realization of overt/morphological agreement is more peripheral than hard core syntax (though it is contingent on the outcome of the syntactic derivation). This is relevant for the discussion below, in that it is consistent with the view that performance factors may play a role in the realization of agreement. In turn, according to the Principle of Incrementality ‘different levels of processing can work on different pieces of an utterance at the same time. Thus, the phonological encoder can work on the early part of the clause while the syntactic encoder works on filling out what remains’ (Ferreira 1996; Schriefers et al. 1998; Levelt 1989; or Phillips 1996).13 This allows for fast/efficient computation, in the sense that the production system does not have to wait for all elements of the sentence to be available before beginning the utterance. The syntactic 13

E.g. Schriefers et al. showed evidence that the verb is not automatically and obligatorily part of the grammatical advance planning unit for finite clauses. In particular, in their experiments, native speakers of German described pictures of simple scenes, while they were presented with verbs as distractor words which were semantically related or unrelated to the verb of the picture description. For utterances with transitive verbs in initial position, utterance onset latencies were longer for the condition with semantically related distractor verbs than for the condition with unrelated distractor verbs. When the target verb did not occur in utterance initial position, the semantic interference effect was not obtained. This means that, at least at some level, the beginning of the sentence is available to the producer before the rest of the sentence is planned or encoded.

100 Iván Ortega-Santos framework that captures the incremental nature of production most naturally is Left-to-Right Syntax (Phillips 1996). I will adopt this framework at this point for the sake of exposition, (though see below for an alternative compatible with bottom-up syntax). Within this framework, agreement is computed from left-to-right and the top of the tree is assembled/made available earlier than the bottom of the tree. Arguably, this state of affairs conspires to derive the above agreement asymmetries, an idea that goes back to Barlow (1992). Specifically, in the P–G order the production system works on P irrespective of whether the rest of the sentence has been coded or not, so as to allow for fast(er) production (cf. Phillips 1996). Nevertheless, a P showing morphological/rich agreement cannot be uttered until G has been coded, because agreement causes P to “wait” for G to become available. Only then can P be spelledout. Inasmuch as such a “wait” goes against the spirit of incremental production, one option the production system has to avoid it is to drop agreement markers or adopt default agreement. Let’s see how the proposal works for the data in (4), repeated here for the sake of exposition. (12) Italian Dialect of Ancona (Cardinaletti 1997a: 38f.) Questo, lo fa /*fanno sempre i bambini. this-ACC it-ACC does / do always the children The crucial factor is whether the element triggering the agreement is already available or not (e.g., whether it has been encoded as part of the advanced planning unit, cf. n. 18), at the point that the element carrying the overt agreement morphology is hit. (13) Course of production of (4) / (12) Production stages 1st stage Questo 2nd stage lo 3rd stage fa(no) † … †

production contingent on the availability of the subject. As a consequence there are two options, to wait or to adopt default agreement.

In turn, in the G–P order (that is to say, in the Spec,Head configuration), the ‘wait’ for G takes place anyway, because it precedes P. Hence, there is nothing to be gained by dropping agreement markers. This is illustrated for (4), repeated here for the sake of exposition:

101

On Incrementality, overt agreement, theta-roles and Spec,Head relations

(14) Italian Dialect of Ancona (Cardinaletti 1997a: 38f.) Questo, i bambini lo fanno / *fa sempre. this-ACC the children it-ACC do / does always (15) Course of production of (4) / (14) Production stages 1st stage Questo 2nd stage i bambini 3rd stage lo 4th stage fanno † †

…

The subject has already been encoded and overt Agreement can be computed accordingly at this stage.

Assuming that such a computation of overt Agreement is incremental and that Spec,Head relations can indeed exist in the cases under discussion, we can derive the mentioned Agreement Asymmetries within the LDA framework (Chomsky 2000, et seq.).14 Going back to the observation in (1)– (3), the present approach suggests the following ranking among Agreement patterns, from the most advantageous system in terms of Incrementality to the less advantageous one. (16) G Povert vs. P G >> G Povert vs. Povert agreement

√ across languages

agreement

G >> G P vs. Povert

agreement

√ across languages

G

agreement

* across languages

Given this, the system at the bottom of the ranking should be fairly uncommon if it exists at all, as seems to be the case. Finally, it is worth mentioning that, within the context of this research, the possibility of agreeing with an in situ goal comes from the fact that the strategies of the parser/production system are defeasible: e.g., center embedding in English is disfavored due to its costly nature, but it is not banned by the parser/production system. In this sense, the present approach succeeds in providing a rationale for the existence of the paradigms under 14

If overt morphological agreement is computed after syntax in the spirit of Distributed Morphology framework (Halle and Marantz 1993), this approach would be compatible not only with left-to-right syntax, but also with standard bottomup syntax.

102 Iván Ortega-Santos consideration. The choice to follow the most incremental pattern or to ‘defeat’ the strategies of the parser/production system would be a languageparticular matter. 15

3.1. On the relationship between Spec,Head relations and Incrementality Given this state of affairs, one wonders whether Spec,Head relations are actually necessary to capture the puzzle posed by morphological agreement for the phase-based system. Could one just blame Incrementality as the sole independent factor underlying the said crosslinguistic variation? This does not seem possible. As discussed, for instance by Lasnik (1999), the interpretation of linguistic data that are consistent with properties of the parser/ production system is not trivial. In particular, Lasnik points out that the properties of the parser/production system may give raise to some of the properties of the grammar (cf., for instance, the seminal work of Berwick and Weinberg 1984 or Wexler and Culicover 1980). For this reason, one cannot conclude that the link between crosslinguistic variation in morphological agreement and Incrementality necessarily excludes the need for checking relations in the Spec,Head configuration. Inasmuch as the need for grammatically significant Spec,Head relations goes beyond morphological agreement and includes at least theta roles if understood as features (Bošković 1994 – and Bošković and Takahashi 1998; Hornstein 2003; Lasnik 1995) and phrases generated in situ in A-bar positions, it follows that checking relations in the Spec,Head configuration are part of the system. 4. Conclusion The present approach explains how evidence for grammatically significant Spec,Head relations can be accommodated within the phase-based Long Distance Agreement framework (Chomsky 2000 and later work). Specifically, the independently motivated proposal of Multiple Spell-Out (Uriagereka 1999, 2008) has been argued to open the door for Specs to establish a probe goal relationship with the head of the projection hosting them. This relation would fulfill the minimalist desiderata of restricting probing to c15

One may wonder whether the pattern in (3) is attested at all as indeed one would expect given that the strategies of the parser are defeasible. Anti-agreement effects exemplify this pattern.

On Incrementality, overt agreement, theta-roles and Spec,Head relations

103

command domains and complying with minimal search conditions on probing. This allows the present system to circumvent conceptual arguments against Spec,Head relations under m-command put forward in Chomsky (2008) and related work. The resulting system is able to address successfully a number of puzzles for the phase-based system, e.g., the fact that across languages moved elements tend to trigger overt agreement as opposed to in situ ones or the issue of how phrases base-generated in A-bar positions are licensed. Furthermore, it has been argued that the differences in agreement morphology found across languages depending on whether the probe goal relation is established locally (cf. Spec,Head relation) or via Long-Distance Agreement are related to optimality considerations as suggested by the Principle of Incrementality. Acknowledgements I would like to thank Tomohiro Fujii, Norbert Hornstein, Howard Lasnik, Juan Uriagereka, Masaya Yoshida, an anonymous reviewer and the audiences at WCCFL25, InterPhases 2006 and BIDE 06 for their comments and suggestions on parts of this paper. All errors are my own. References Barlow, Michael 1992 A Situated Theory of Agreement. London: Garland. Berwick, Robert C. and Amy S. Weinberg 1984 The Grammatical Basis of Linguistic Performance: Language Use and Acquisition. Cambridge, MA: MIT Press. Bobaljik, Jonathan D. and Susi Wurmbrand 2005 The domain of agreement. Natural Language and Linguistic Theory 23: 809–865. Bock, J. Kathryn, Helga Loebell, and Randal Morey 1992 From conceptual roles to structural relations: Bridging the syntactic cleft. Psychological Review 99: 150–171. Boeckx, Cedric 2004 Long-Distance Agreement in Hindi: Some theoretical implications. Studia Linguistica 58: 23–36. Bošković, Željko 1994 D-Structure, Theta Criterion, and movement into theta positions. Linguistic Analysis 24: 247–286. 2002 A–movement and the EPP. Syntax 5: 167–218.

104 Iván Ortega-Santos Bošković, Željko and Daiko Takahashi 1998 Scrambling and Last Resort. Linguistic Inquiry 29: 347–366. Cardinaletti, Anna 1997a Subjects and clause structure. In The new comparative syntax, Liliane Haegeman (ed.), 33–63. New York: Longman. 1997b Agreement and Control in expletive constructions. Linguistic Inquiry 28: 521–533. Chomsky, Noam 2000 Minimalist inquiries: The framework. In Step by Step: Essays on Minimalist Syntax in Honor of Howard Lasnik, Roger Martin, David Michels, and Juan Uriagereka (eds.), 89 –156. Cambridge, MA: MIT Press. 2001 Derivation by phase. In Ken Hale: A Life in Language, Michael Kenstowicz (ed.), 1–52. Cambridge, MA: MIT Press. 2004 Beyond explanatory adequacy. In Structures and Beyond: The Cartography of Syntactic Structures, Vol. 3, Adriana Belletti (ed.), 104 – 191. Oxford: Oxford University Press. 2005 Three factors in language design. Linguistic Inquiry 36: 1–22. 2007 Approaching UG from below. In Interface + Recursion = Language?, Uli Sauerland and Hans-Martin Gärtner (eds.), 1–29. Berlin /New York: Mouton de Gruyter. 2008 On phases. In Foundational Issues in Linguistic Theory: Essays in Honor of Jean-Roger Vergnaud, Robert Freidin, Carlos P. Otero and Maria Luisa Zubizarreta (eds.), 133–166. Cambridge, MA: MIT Press. Corbett, Greville G. 1979 The agreement hierarchy. Journal of Linguistics 15: 203–224. Den Dikken, Marcel 2000 The syntax of features. Journal of Psycholinguistic Research 28: 5–23. Donati, Caterina 2006 On Wh-head movement. In Wh-Movement: Moving on, Lisa Cheng and Norbert Corver (eds.), 21– 46. Cambridge, MA: MIT Press. Epstein, Samuel D. and T. Daniel Seely. 2006 Derivations in Minimalism. Cambridge: Cambridge University Press. Ferreira, Victor S. 1996 Is it better to give than to donate? Syntactic flexibility in language production. Journal of Memory and Language 35: 724 –755. Franck, Julie, Glenda Lassi, Ulrich H. Frauenfelder, and Luigi Rizzi 2006 Agreement and movement: A syntactic analysis of attraction. Cognition 101: 173–216. Gelderen, Elly van 2004 Economy, innovation, and prescriptivism: From Spec to Head and Head to Head. Journal of Comparative Germanic Linguistics 7: 59 – 98.

On Incrementality, overt agreement, theta-roles and Spec,Head relations

105

Halle, Morris and Alec Marantz 1993 Distributed Morphology and the pieces of inflection. In The View from Building 20, Kenneth Hale and Samuel J. Keyser (eds.), 111–176. Cambridge, MA: MIT Press. Heinat, Fredrik 2006 Probes, pronouns, and binding in the Minimalist Program. Ph.D. diss., Lund University. Hornstein, Norbert 1995 Logical Form: From GB to Minimalism. Oxford: Blackwell. 2003 On control. In Minimalist Syntax. Randall Hendrick (ed.), 6 –81. Oxford: Blackwell. Hornstein, Norbert, Jairo Nunes, and Kleanthes K. Grohmann 2005 Understanding Minimalism. Cambridge: Cambridge University Press. Hornstein, Norbert, Howard Lasnik, and Juan Uriagereka 2007 Islands and linearization. Proposal for NSF grant BCS-0722648. University of Maryland. Hornstein, Norbert and Juan Uriagereka 1999 Labels and projections: A note on the syntax of quantifiers. University of Maryland Working Papers in Linguistics 8: 249–270. Huang, C.-T. James 1982 Logical relations in Chinese and the theory of grammar, Ph.D. diss., MIT. Iatridou, Sabine, Elena Anagnostopoulou, and Roumyana Izvorksi 2001 Observations about the form and meaning of the perfect. In Ken Hale: A Life in Language, Michael Kenstowicz (ed.), 189–238. Cambridge, MA: MIT Press. Kayne, Richard S. 1989 Facets of Romance past participle agreement. In Dialect Variation and the Theory of Grammar, P. Beninca (ed.), Dordrecht: Foris. 85– 103. 1994 The Antisymmetry of Syntax. Cambridge, MA: MIT Press. Koopman, Hilda 2006 Agreement configurations: In defense of “Spec head”. In Agreement Systems. Cedric Boeckx (ed.), 159 –199. Amsterdam: John Benjamins. Lasnik, Howard 1995 Last resort and attract F. In Proceedings of FLSM 6: 62–81. Indiana University Linguistics Club, Bloomington. 1999 On the locality of movement: Formalist syntax position paper. In Functionalism and Formalism in Linguistics, Volume 1: General Papers (SLCS 41), M. Darnell, E. Moravscik, F. Newmeyer, M. Noonan, and K. Wheatley (eds.), 33–54. Amsterdam: John Benjamins. Legate, Julie Anne 2001 Phases and Cyclic Agreement. In Perspectives on Phases, Martha McGinnis and Norvin Richards (eds.). MITWPL 49: 147–156.

106 Iván Ortega-Santos Levelt, Willem J. M. 1989 Speaking: From Intention to Articulation. Cambridge, MA: MIT Press. Lidz, Jeffrey and Alexander Williams 2002 Reflexivity and resultatives. In WCCFL 21 Proceedings, L. Mikkelsen and C. Potts (eds.), 250 –263. Somerville, MA: Cascadilla Press. Manzini, M. Rita and Leonardo M. Savoia 1998 Parameters of subject inflection in Italian dialects. Ms., University of Florence, Italy. Moravcsik, Edith A. 1978 Agreement. In Universals of Human Language, Vol. 4, J. H. Greenberg, C. A. Ferguson, E. A. Moravcsik (eds.), 331–374. Stanford: Stanford University Press. Park, Myung-Kwan 2006 Moved elements undergo Agree after Move. Talk at the 36th Annual Meeting of the Michigan Linguistics Society. Pesetsky, David and Esther Torrego 2006 Probes, goals and syntactic categories. Ms., MIT and Massachusetts Boston. Phillips, Colin 1996 Order and structure. Ph.D. diss., MIT. Samek-Lodovici, Vieri 2002 Agreement impoverishment under subject inversion – a crosslinguistic analysis. In Resolving Conflicts in Grammar, Gisbert Fanselow and Caroline Féry (eds.). Linguistische Berichte Sonderheft 11: 49–82. Schriefers, Herbert J., E. Teruel, and R. M. Meinshausen 1998 Producing simple sentences: Results from picture-word interference experiments. Journal of Memory and Language 39: 609–632. Sigurðsson, Halldor Á. 2006 PF is more ‘Syntactic’ than often Assumed. Talk at InterPhases, at the University of Cyprus. Soltan, Usama 2003 A minimalist analysis of locative inversion constructions. Ms., University of Maryland, College Park. Uriagereka, Juan 1999 Multiple spell-out. In Working Minimalism, Samuel D. Epstein and Norbert Hornstein (eds.), 251–282. Cambridge, MA: MIT Press. 2008 Syntactic Anchors. Cambridge: Cambridge University Press. Wexler, Kenneth and Peter W. Culicover 1980 Formal Principles of Language Acquisition. Cambridge, MA: MIT Press.

Structural case, locality and cyclicity Thomas McFadden

It is argued that long-distance case assignment does not actually exist. Unlike long-distance agreement, it would necessarily violate the Phase Impenetrability Condition. All examples that have been analyzed in terms of long-distance case actually involve the appearance of the nominative as the default case. Cases that are actually assigned – the accusative and the nonstructural cases – always obey locality. Yet assignment of the accusative does appear to violate cyclicity. This problem can however be solved as well with the right view on when case is assigned, once the dependency between structural accusatives and other DPs is properly understood. As a result, the alternative solution put forward by Sigurðsson (2006b), requiring that subjects first-merge lower than objects, is shown to be unnecessary.

1. Phases and long-distance agreement Apparent long-distance agreement and case-assignment, as exemplified in the Icelandic example (1a) (from Sigurðsson 2006b) and the English example (1b) (from Legate 2005), raise important questions for the locality of syntactic operations. (1)

a. Það voru taldir hafa verið veiddir fjórir laxar. there were:PL believed have been caught four salmon:N.PL ‘People believed there to have been four salmon caught.’ b. There seem to have arrived ten trains into this station today.

On the one hand, nominative case on the DP fjórir laxar ‘four salmon’ in the embedded non-finite clause would appear to be assigned or licensed from finite T in the matrix clause. On the other, the case and phi-features on fjórir laxar must trigger nominative plural agreement on veiddir ‘caught’, taldir ‘believed’ and voru ‘were’ (similarly in the English example). Crucially, the dependencies here are in principle unbounded, as arbitrarily many non-finite raising clauses can intervene between the finite verb and the clause containing the DP.

108 Thomas McFadden In Chomsky (2000, 2001, 2008) this is handled by defining the units of syntactic representation relevant for locality in such a way that no boundaries intervene between the low nominative DP and the agreeing finite T in such sentences. Specifically, only C and v* (the flavor of v which introduces an external argument) define phases, and neither of these is present at the relevant levels. Crucially, plain v (the version which appears in passive and unaccusative clauses like in (1) and does not introduce an external argument) is not assumed to be a phase head.1 However, the distinction between the two types of v is suspicious, especially if phases are really supposed to be “the closest syntactic counterpart to a proposition: either a verb phrase in which all θ-roles are assigned or a full clause including tense and force” (Chomsky 2000: 106). There is no reason to think that unaccusatives and passives are any less propositional than transitives, and they have certainly had all of their θ-roles assigned in the only sense that can be relevant.2 Indeed, Legate (2003) presents evidence from reconstruction, quantifier raising, parasitic gap licensing and stress assignment suggesting that all vPs do indeed constitute phases, whether they have an external argument or not. If this conclusion about the phasehood of vP is correct, the facts in (1) present a puzzle. The Agree operation between finite matrix T and fjórir laxar in (1a) – or between T and ten trains in (1b) – would have to cross intervening phase boundaries, and this would violate Chomsky’s Phase Impenetrability Condition (henceforth PIC). In order to avoid this problem, Legate (2005) proposes something called Cyclic Agreement, a mechanism parallel to what is standardly assumed to capture the unbounded dependencies of successive-cyclic wh-movement and subject raising.3 These varieties 1

2

3

In some formulations of the theory, vP is considered a phase, but not a strong phase. Since what matters for the locality of operations under those formulations is the strong phase, this point is not relevant for our concerns. That is, once the vP is constructed, passives and unaccusatives have had all of the θ-roles assigned which are going to be assigned, just as transitives do. The other conceivable interpretation of Chomsky’s statement is that phases are verb phrases where all possible θ-roles have been assigned. This, however, cannot be what is relevant, since transitive phrases like [vP John [VP sell his car]] must certainly count as phases, in spite of the fact that the optional recipient θ-role (which we see in examples like John sold Mary his car) has not been assigned. Legate also points out that the locality problem cannot be avoided by assuming with Chomsky (2001) that a given phase is not spelled out until the next higher phase head has been reached. While this would allow agreement to cross one phase boundary, the examples in (1) both involve the intervention of at least two (in (1b) e.g. we have the little v associated with arrive and that with seem).

Structural case, locality and cyclicity

109

of movement have in common that they would be non-local if they were implemented as single steps from base position to landing site. The standard analysis instead breaks up the movement into a series of shorter steps, each of which is individually local. What Legate proposes is that long-distance agreement is exactly parallel, proceeding in local steps through intermediate phase heads until finally reaching an instance of finite T. In a sentence like (1b), ten trains will thus agree first with the v in the embedded clause, i.e. the one connected with arrive. This v will then agree with the matrix v (connected with seem), and it is finally this matrix v which agrees with finite matrix T. A similar analysis will apply to the Icelandic example (1a). 2. Long-distance case? Legate’s proposal of Cyclic Agreement is well-suited to handle the agreement facts in sentences like (1). However, it can’t account for the caseassignment patterns in the same class of examples. The dependency in longdistance agreement is of something in a higher phase – the finite T showing agreement – on something in a lower phase – the DP in the embedded clause. As phases are constructed, information is passed successively further up the tree in standard, cyclic fashion, and the φ-features of the embedded DP can thus (indirectly) influence the form of the matrix verb. The direction of the dependency in case assignment, on the other hand, is crucially reversed. The DP in the lower phase is assumed to depend for the licensing or assignment of its nominative case on finite T in a higher phase. In fact, if we take cyclic, phase-based Spell-out seriously, the phase containing a DP like fjórir laxar would actually be sent to Spell-out before the finite T apparently responsible for its case has even been merged into the derivation.4 Legate considers this issue and suggests two possibilities for getting around it. First, she says we could adopt the view of early Minimalism, according to which DPs enter the derivation with valued but uninterpretable case features, which must be deleted by checking. The idea would be that derivations where the DPs happen to have been inserted with the right case 4

This asymmetry of course applies only to theories that assume a bottom-up derivation. Precisely the opposite would hold in a left-to-right derivation as proposed by Phillips (2003), and there would be no asymmetry in either direction in a theory like Tree Adjoining Grammar, where long-distance dependencies are derived by inserting material between the dependent elements (see e.g. Frank 2006). Evidence regarding the existence and nature of such asymmetries can thus help us to decide between such theories.

110 Thomas McFadden features will eventually converge, once agreement makes its way cyclically up to the case-assigning head. Those where they have the wrong cases will crash and be filtered from consideration. The observed patterns of casemarking would thus emerge more or less by process of elimination. Such a scenario is less than ideal, since it essentially implies that multiple derivations must be considered in parallel. The second possibility she considers is that DPs that engage in long-distance agreement are actually sent off to Spell-out without case. The morphological component can wait until the construction of the phase which contains finite T, at which point it has the information necessary to go back into the previous phase and add in the appropriate case after the fact. Again, such an option is unattractive, as it would violate the spirit of a phase-based approach to syntax. We would ideally like to maintain the idea that, once a phase is sent off to Spell-out, it cannot be affected by subsequent steps of the derivation. The objections to these two approaches to long-distance case assignment may not be serious enough to eliminate them from consideration completely, but they suffice to make it worthwhile to look for alternatives. To this end, we must examine more carefully the conditions under which long-distance case-assignment actually occurs. Alongside (1) above, consider the following examples of long-distance case and agreement ((2a–c) are from Sigurðsson 2006b; (2d) is from Boeckx 2000):5 (2)

a. Þá hafði skyndilega birst maður/*mann. then had:SG suddenly appeared man:N/*A ‘Then a man had suddenly appeared.’ b. Þá höfðu verið lesnir kaflar. then had:PL been read chapters:N ‘Then chapters had been read.’ c. Henni höfðu alltaf leiðst strákarnir. her:D had:PL always found-boring boys-the:N ‘She had always found the boys boring.’ d. Honum eru taldir hafa verið gefnir peningarnir. him:D are:PL thought have been given money:N.PL ‘He is thought to have been given the money.’

5

My discussion in the rest of this paper will focus on Icelandic data for the simple reason that it has rich case marking and agreement and (almost) all of the phenomena that provide crucial tests of locality in case-assignment, viz. ECM, raising and subjects bearing non-structural cases.

Structural case, locality and cyclicity

111

What we find are several variations on the following theme. Some nonnominative element occupies the surface subject position, either an expletive ((2a–b), as well as (1a–b) above) or a quirky subject (2c–d). An argument we would expect to start out low in the VP – an unaccusative subject (1b, 2a), or the underlying object in a passive (1a, 2b and 2d) or in a clause with a quirky subject (2c) – is able to remain low because the expletive or quirky subject satisfies the EPP. In spite of its position, this low argument receives nominative case and triggers agreement on the verb, which may be in the same clause (2a–c) or in a higher clause containing a raising predicate (1b) or passivized ECM verb (1a, 2d). A clear generalization emerges here.6 The case that appears to be assigned in non-local configurations is always the nominative, at least in nominative-accusative languages.7 Now, it is not particularly surprising that non-structural cases should never involve long-distance dependencies: they are generally assumed to be assigned in highly local fashion by specific lexical items or functional heads. What is intriguing is that the structural accusative is not assigned over long distances either. Rather, it is apparently always assigned or checked locally, within the minimal phase. In a simple transitive sentence like (3a) (from Sigurðsson, 2006a), this is standardly assumed to be handled by v. As shown by the relevant portion of the structure shown in (3b), v is indeed local to the accusative DP hana (both in boldface), as both are contained within the same minimal phase, here vP.8 (3)

a. Þeir völdu hana. they:N chose:PL her:A ‘They chose her.’ b. [vP Þeir v [VP völdu hana ] ]

6

7

8

I suspect that I am not the first to notice this generalization, but I present it here without citation because I am not aware of any previous discussion in the literature. It may be that non-nominative cases can be copied over longer distances as part of concord processes in certain languages (see e.g. Richards 2007, for discussion of interesting data from Lardil). The morphophonological mechanisms involved are however poorly understood and are plausibly distinct from those involved in actual case assignment. The structures I give throughout are simplified for clarity’s sake. E.g. movement is not indicated unless relevant to the discussion of the example, and I use a single, undifferentiated v rather than taking a position on whether, e.g., a head introducing the external argument should be distinguished as Voice from heads responsible for eventivity or the verbalization of a category-neutral root.

112 Thomas McFadden (4)

a. Ég tel hana hafa séð myndina. I:N believe her:A have seen picture-the:A ‘I believe her to have seen the picture.’ b. [vP Ég v [VP tel [TP hanai hafa [vP ti v[VP séð myndina ] ] ] ] ]

Even in ECM configurations like (4a) (from Sigurðsson 1992), the embedded subject is local to the matrix v for the purposes of structural accusative assignment, as shown in the structure in (4b). Though it may be first-merged in the phase defined by the embedded v, it is at the edge of that phase and can escape without violating the PIC. Depending on which account one adopts, it either raises to the Spec of the embedded TP (as indicated in (4b)) or even further into a derived object position in the matrix clause. In either case, it ends up within the minimal phase defined by the matrix v, and thus it can receive/check structural accusative there in strictly local fashion. What we do not find are examples which are parallel to 1a, but where the case assigned long distance is accusative. The relevant scenario would be where a DP is able to remain in a low position due to the presence of an expletive or quirky subject, yet still gets structural accusative from an ECM verb further up. In Icelandic, this would look something like (5a) (from Bowers 2002) or (5b) (from Sigurðsson 2006b): (5)

a. *Ég hafði talið það vera villu í essu handriti. I:N had believed there be error:A in this manuscript intended: ‘I had believed there to be an error in this manuscript.’ b. *Við töldum henni hafa leiðst strákana. we:N believed her:D have found-boring boys-the:A intended: ‘We believed her to have found the boys boring.’

Interestingly enough, structures like (5a), with an expletive occupying the embedded subject position under an ECM verb, aren’t possible in Icelandic. A possible explanation for this is that Icelandic expletive það appears in Spec-CP, and thus won’t fit into ECM infinitives, which are only TPs (see e.g. Bowers 2002). Whatever the reason, case is not involved, so sentences of this kind cannot tell us anything about locality in case-assignment.9 The 9

Of course, such structures are possible in English, as shown by the perfectly grammatical translation given for (5a). Unfortunately, they cannot tell us anything about case locality either because the English case system is highly impoverished and organized along different lines than in languages with a more fully developed nominative-accusative system (see McFadden 2007, for discussion of this point).

Structural case, locality and cyclicity

113

interesting example for our purposes is (5b), because the configuration here is indeed possible in Icelandic. The problem with the way that it stands above is that the embedded object is in the accusative. The grammatical version of this structure has nominative case on the lowest DP, as in (6a) (also from Sigurðsson 2006b): (6)

a. Við töldum henni hafa leiðst strákanir. we:N believed her:D have found-boring boys-the:N ‘We believed her to have found the boys boring.’ b. [vP1 Við v [VP töldum [TP hennii hafa [vP2 ti v[VP leiðst strákanir ]]]]]

The structure in (6b) (simplified as above) shows how locality plays out in examples like this. The embedded clause is built around a verb that assigns quirky dative to its subject and thus does not assign accusative to its object. Nonetheless, the embedded clause does involve a vP, here labeled vP2, which constitutes a phase. Now, since the embedded subject gets quirky dative, the matrix ECM verb does not get a chance to assign it structural accusative as it normally would. We expect that in principle it should be able to assign this accusative elsewhere, but as it turns out there is no eligible DP local enough for the assignment to go through. The embedded object strákanir is simply too far away from the matrix v, because it remains within the phase defined by vP2, and does not raise to its edge.10 If the Cyclic Agreement mechanism proposed by Legate (2005) were really responsible for case as well as agreement, we might expect it to allow a long-distance accusative in instances like this, but apparently it only works for the nominative.11

10

11

Note that the impossibility of structural accusative here is not due to an intervention effect caused by the presence of the dative. Accusative is assigned unproblematically across dative arguments in standard ditransitive clauses. This fact would also seem to indicate that the head which introduces indirect object datives in Icelandic does not define a phase boundary. See McGinnis (2005) and earlier work cited there for discussion of the phasal status of various kinds of applicatives cross-linguistically. Halldór Sigurðsson, p.c., has alerted me to another Icelandic construction which might be a legitimate counterexample to the ban on long-distance accusative assignment. Under certain circumstances, subjects of ECM infinitives can surface in post-verbal position without an overt expletive or quirky subject, and they crucially show up accusative, not nominative:

114 Thomas McFadden 3. Nominative case is independent of agreement Now, one might take the facts just presented as confirmation that long-distance case assignment is crucially tied to agreement. After all, in Icelandic and many other languages, we only find overt morphological agreement with nominatives, never with accusatives. One could propose that while nominative is assigned via agreement, accusative comes about differently, and thus long-distance assignment via Cyclic Agreement is available only to the former. However, there is very good reason to doubt the standard assumption that nominative case depends on agreement with finite T. For example, Sigurðsson (2006b) presents a number of Icelandic examples like those in (7) where a nominative object appears in a non-finite clause: (7)

a. Að líka svona fáránleiki/*fáránleika! to like:INF such absurdity:N/*A ‘To like such absurdity!’ b. Ekki ÞIÐ/*YKKUR fara líka! not you:N.PL/*A.PL leave:INF too ‘Please, don’t YOU leave too!’

Note that (6a) above fits into this category as well. What is crucial about these examples is not just that we have nominative in a non-finite embedded clause. After all, 1a has such a thing as well. The difference here is that the embedded nominative does not and cannot trigger agreement on a finite i. Þeir töldu líklega hafa verið veidda fjóra laxa they believed probably have been caught four salmon:A.PL ‘They believed there to have probably been caught four salmon.’ If such subjects never leave the embedded vP phase, as the word order would suggest, then they are not local to matrix v for purposes of accusative assignment, and we would have an instance of long-distance accusative. This construction is, however, mysterious for other reasons. It is an instance of a more general phenomenon which Sigurðsson (2000) calls ‘subject float’, involving indefinite subjects of passives and unaccusatives in clauses where something else fills Spec-CP (expletive það or a topicalized XP) or CP is absent entirely (as in these ECM infinitives). Among other things, if the low subjects really are in situ in such structures, it is not clear why they don’t violate the EPP, which otherwise seems to hold consistently in the language. Judgment on whether sentences like (i) really involve long-distance accusative assignment must thus be postponed until their structure is better understood.

Structural case, locality and cyclicity

115

matrix verb. In (6a) above, the finite matrix verb already agrees with the matrix subject, and in the examples in (7) there is no finite verb. To these we can add a class of examples discussed by Boeckx (2000): (8)

a. Mér fannst/*fundust henni leiðast þeir. me:D seemed:SG/*PL her:D bore they:N ‘I thought she was bored with them.’ b. Mér hefur/*hafa alltaf virst honum líka bækur. me:D has/*have often seemed him:D like books:N.PL ‘It has often seemed to me that he likes books.’

These are more or less parallel to (2d) above, except that there is a quirky dative subject in the embedded clause. This dative intervenes between the embedded object and the finite matrix verb and somehow blocks the expected long-distance agreement. The verb thus shows up in a default 3rd singular form rather than the expected 3rd plural. Crucially however, whatever it is that disrupts agreement here, it has no effect on case assignment: the embedded objects are still nominative, in spite of having nothing to agree with. Finally, in most nominative-accusative languages, we find the nominative in a wide range of contexts which are in some sense extra-sentential, like citation forms and (at least in some languages) vocatives. Particularly instructive are instances of left dislocation, exemplified for German, Russian, Arabic and Icelandic in (9). We might expect a dislocated DP to agree in case with its coreferent resumptive element, and indeed it often does. Sometimes, however, under language-specific circumstances, this fails, and the nominative shows up: (9)

a. Der/*Dem Hans, mit dem spreche ich nicht mehr. the:N/*D Hans with him:D speak I not more ‘Hans, I don’t speak with him anymore.’ b. Vanja/?Vanju, ego ja ne ljublju. John:N/?A him:A I don’t like ‘John, I don’t like him.’ c. al-kitaab-u qarat-u-hu. the-book-N read-1SG-it ‘The book, I read it.’ d. Strákarnir, við þá hafði aldrei verið talað. boys-the:N with them:A had never been spoken ‘The boys, they had never been spoken with.’

(German)

(Russian)

(Arabic)

(Icelandic)

116 Thomas McFadden In instances like these, there is again no question of the nominative showing up as a result of agreement. Not only is there no overt agreement, there is no predicate in the local context that the nominative DP could even be abstractly agreeing with. Schütze (2001) argues convincingly that in instances like these, the nominative is appearing in its capacity as the default case, which surfaces whenever normal rules of case assignment fail.12 Now, if we need to assume that nominative can appear as a default in order to account for examples like these, there is actually nothing to prevent us from arguing that the ‘structural’ instances of the nominative are really default nominatives as well. They are not assigned due to agreement with finite T, as we have seen, but simply appear because the conditions for assignment of the other cases are not met. This interpretation is attractive precisely because it provides a principled explanation for why the nominative – and only the nominative – appears to be assigned long-distance. In fact, there is no such thing as long-distance case assignment. As noted above, unlike long-distance agreement such a thing would necessarily violate the PIC. Accusative and the various non-structural cases are assigned under identifiable syntactic and/or lexical conditions, due to relationships with other elements present in the structure. Because of the involvement of other syntactic elements, such assignment strictly obeys locality conditions, and does not apply when those conditions are not met, as we saw in (6a) above. Nominative, on the other hand, is different, because it is not assigned on the basis of a relationship with an easily identifiable local element. Most of the central examples which have formed the basis for the discussion of case were consistent with the idea that nominative depended on agreement with finite T. This led to the impression that examples like (1a) require long-distance case assignment. However, once we understand that the nominativeagreement connection was illusory, we can see that such examples simply involve a nominative appearing as a default, just like (7) and (9d) where no confusion is possible because there is no agreement. Even in a simple transitive sentence like (3a), where there is local agreement between the nomina12

Note that I am talking here about morphological case, not case in the sense of syntactic DP licensing. It has been amply demonstrated that the two are independent (see e.g. Marantz 1991; Harley 1995; Schütze 1997; Sigurðsson 2003; McFadden 2004; Bobaljik 2006; and also Legate 2008 for a recent dissenting view.), and default case is one of the key pieces of evidence. If morphological case were really responsible for licensing, then the availability of a default case would mean that everything would be licensed, and the Case Filter (or its equivalent) would be vacuous (Schütze 2001; McFadden 2004).

Structural case, locality and cyclicity

117

tive subject and the finite verb, we can assume that the nominative shows up as the default, unmarked case, not because of the agreement.13 We can thus maintain the idea that case-assignment is always local within the minimal phase.14 4. Accusative case and cyclicity Sigurðsson (2006b) argues for a similar view on the independence of the nominative case. The central concern of his paper, however, is a different issue which arises when we consider how the accusative is assigned. Adopting an idea presented in different forms by Yip, Maling, and Jackendoff (1987); Marantz (1991) and others, he proposes that there is a dependency relationship between nominative and accusative. The nominative is the independent case of the structural pair, assigned first in a given domain, while the accusative can only be assigned after a nominative is already present.15 As a simple empirical generalization, this works rather well. If we consider how such a dependency should be implemented in the theory, however, we run into a problem, which Sigurðsson dubs the Nominative Puzzle. If the assignment of structural accusative really depends on the prior assignment of structural nominative, then it is counter-cyclic on standard assumptions about where arguments are first-merged in the structure. In a simple transitive clause, V combines first with the object, then one or more functional heads are merged, and finally the subject enters the picture, yielding a standard vP structure, as depicted in (3b) above. This would lead us to expect, if anything, that properties of the subject would depend on properties of the object, because the object is already present at the point in the derivation when the subject is merged. What we see with the structural cases is, however, precisely the opposite. The case of the object apparently 13

14 15

The inverse claim is not necessarily correct. That is, in Icelandic and many other languages there does indeed seem to be a dependency of agreement on nominative case. See Bobaljik (2006) for recent discussion. See Sigurðsson (2000) and Marantz (2006) for related discussion and conclusions. This is intended as an alternative to Burzio’s Generalization (Burzio 1986), according to which structural accusative depends not on nominative, but on the assignment of an external θ-role. See e.g. Marantz (1991); Woolford (2003); Sigurðsson (2006b); Zwart (to appear) as well as the papers collected in Reuland (2000) for discussion of the relative merits of these and other approaches to the data. See Gehrke (2007) for an attempt to extend the dependency idea to accusative objects of directional prepositions.

118 Thomas McFadden depends on the case of the subject, which should not be possible. At the point when the object is merged, the subject is not yet there, and thus accusative cannot be assigned – we do not yet know whether the clause will be transitive, unaccusative or passive, and even if there is going to be a subject, what case it will get. Sigurðsson (2006b) argues quite rightly that this is out of line with what we expect of the derivation under standard Minimalist assumptions, and that something in our theory must be adjusted so that the inconsistency is avoided. I would like to suggest that the inconsistency arises from the combination of three assumptions: 1. The derivation is strictly cyclic. 2. The assignment of structural accusative depends on the presence of a nominative. 3. Subjects are first-merged higher than objects. In order to solve the Nominative Puzzle, one or more of these assumptions must be abandoned or revised. Sigurðsson chooses to maintain 1 and 2, which leaves him with no other choice but to question the order in which arguments are merged. What he proposes is that the subject originates lower than the object in a simple transitive clause, later moving past for reasons independent of case. This allows the structural case on a DP to be determined in purely cyclic fashion at the point when it enters the derivation. As an example, consider what happens in a simple transitive clause. First the subject is merged with the verb. The nominative is the independent case, always assigned first in a clause, thus the subject is assigned nominative. Then the object is merged, and since the subject is already present and bears nominative, the object can be assigned the dependent accusative. Each DP gets the correct structural case, as soon as it enters the derivation, and in a way that respects cyclicity. Nonetheless, while Sigurðsson’s Low Nominative Hypothesis works as a technical solution to the Nominative Puzzle, it faces significant problems. For one thing, the movement that gets the subject from its low starting position to a place above the object is implemented in a somewhat ad hoc fashion. After the object has entered the derivation and received its accusative case, a series of functional heads are merged, including T, Asp and crucially one or more heads which agree with the object. As a result, all of the active features on the object will have been checked or valued, and the object is “frozen in place”. The subject, on the other hand, still has unchecked φfeatures. At this point, a Num head is merged, which targets these unchecked features on the subject, and since the φ-features on the object have

Structural case, locality and cyclicity

119

already been matched, it intervenes, but only inactively. Sigurðsson proposes then that inactive intervention as in this case does not block agreement entirely, but does disrupt it enough to drive movement: the subject must move past the inactive object in order to match properly with the Num head, and hence we find the subject surfacing above the object in normal transitive clauses. On a technical level this works, but it remains a bit mysterious why inactive intervention should have precisely this movement deriving effect and not another. Note furthermore that the reordering process depends on the stipulation that the Pers and Num heads related to the subject be merged higher in the structure than those related to the object, since otherwise the movement-driving agreement across an inactive intervener would never come about. Considering that the relative order of the arguments at first merge is precisely the opposite, this is not a trivial stipulation. The lack of principled motivation for these two theoretical assumptions is especially suspicious as the mechanism they derive happens to drive precisely the movement that is required to undo the unwanted word-order effects of the Low Nominative Hypothesis. Perhaps more importantly, however, the revision of the first-merge order of subjects and objects is not sufficiently supported. The standard assumption that verbs are more tightly connected to their objects than to their subjects is backed up by considerable evidence (see e.g. the standard discussions in Marantz 1984; Kratzer 1996), which Sigurðsson (2006b) does not really address. The point is not that the standard view is sacrosanct, but that it is well enough supported that we should be hesitant to give it up. I thus submit that an alternative solution to the Nominative Puzzle which allows us to retain a reasonably familiar-looking VP would be preferable to the Low Nominative Hypothesis.

5. Accusative case depends on structural case Let us then reconsider the conundrum presented to us by the combination of cyclicity, case dependency and the relative position of subjects and objects. I would suggest that the most uncertain point in this constellation is the dependency relationship between nominative and accusative case. What makes it clearly problematic, as Sigurðsson formulates it, is that it involves the two cases directly. We cannot determine the case on the object until we have made precisely the same determination on the higher subject, and this will violate cyclicity no matter what we do. However, there is evidence that this formulation is not quite correct. Consider again what happens in ECM sen-

120 Thomas McFadden tences like (4a) above, repeated here as (10a). It is clear that the accusative on the embedded subject hana must depend on the matrix subject Ég, and as noted above, this presents no problem for locality. Matters are not so obvious, however, for the embedded object myndina. One might imagine that it too depends on the matrix subject, which is after all the only nominative argument in the sentence – but this would be non-local. The embedded object never raises out of the embedded vP, or even to its specifier, thus it remains firmly within the lowest phase. The matrix subject, on the other hand, is in a higher phase from the beginning, starting out in the specifier of the matrix vP. Crucially, it does not enter the derivation until after the lower phase has been sent to Spell-out. Without violating the PIC, the accusative on the embedded object myndina thus can only depend on the embedded subject hana – specifically on its first-merge position in the specifier of the embedded vP. (10) a. Ég tel hana hafa séð myndina. I:N believe her:A have seen picture-the:A ‘I believe her to have seen the picture.’ b. Við töldum henni hafa leiðst strákanir/*strákana. we:N believed her:D have found-boring boys-the:N/*A ‘We believed her to have found the boys boring.’ That this is indeed the case can be seen again from the comparison of (5b) and (6a) above, collapsed together here as (10b). The clause embedded under the ECM verb has a quirky dative subject, and under these circumstances the embedded object is marked nominative, not accusative. If the case on embedded objects in ECM clauses depended on the matrix subject, we would expect the same results here as in (10a), since in both the matrix subject is a structural nominative. The fact that we get a difference shows that it is the status of the embedded subject that matters, just as locality predicts. What is crucial for our consideration of the Nominative Puzzle is that in sentences like (10a) we have a structural accusative depending not on a nominative, but on another structural accusative. The correct generalization thus seems to be that structural accusative can only be assigned in the same phase as another, higher DP with structural case. Which of the two structural cases that other DP bears is irrelevant.16 What this means, as I will argue in the following sections, it that the dependency is not necessarily counter-cyclic. 16

Sigurðsson also assumes that the embedded accusative in examples like (10a) depends on the embedded subject, not the matrix subject. However, he assumes

Structural case, locality and cyclicity

121

6. Structural case marking is structurally predictable While the precise identity of the case on the subject should not be known before the identity of the case on the object, some information about the subject could be available at the time that case is determined for the object. This is so if we adopt a realizational view of morphology like Distributed Morphology (Halle and Marantz 1993; Embick and Noyer 2007) and assume that the actual morphological cases are assigned post-syntactically, in the branch of the derivation leading to PF (as Sigurðsson 2006b, himself does: fn. 6, p. 296; see also McFadden 2004). Under such assumptions, we can revise our expectations about exactly when all of the necessary information must be available. There are in principle (at least) two cycles for the construction of each phase. First comes the narrow syntactic cycle, which builds up the structure of each successive phase, sending it off when it is complete to PF and LF for further processing. We then have a second cycle on the PF branch, during which operations take place affecting the form of an utterance, but not its interpretation, e.g. linearization, morphological merger, impoverishment and vocabulary insertion (see especially Embick and Noyer 2001, for relevant discussion). In such a system, we would expect that the determination of morphological cases need not be completed until sometime in the second cycle.17 This means that at the time when case is determined for a particular DP, information about higher portions of the structure will in principle be available, as long as it comes from the first cycle. Information about higher elements from the second cycle will still not be available, thus it should still be impossible for the case on a lower DP to depend directly on the case of a higher DP. However, it should be able to depend on the syntactic status of the higher DP. Crucially, this is sufficient to determine whether a higher DP will ultimately receive a structural case, and thus can serve as the basis for the dependency of structural accusative. A great deal of recent work has argued that DPs bearing various kinds of non-structural cases are associated with specific positions in a syntactically articulated argument structure. For example, dative verbal arguments have been convincingly argued to be introduced by applicative heads in a variety of languages (see e.g. Pylkkänen 2002; Cuervo 2003; Schäfer 2005;

17

that ECM subjects initially have nominative, which is only later “overwritten” by the acussative. The determination of case on the embedded object takes place before the overwriting, so the dependency on the nominative is maintained. Again, it should be noted that syntactic DP licensing is an independent matter and would in fact belong in the first, narrow syntactic cycle.

122 Thomas McFadden Sundaresan 2006; and the papers collected in Hole, Meinunger, and Abraham 2006). The standard external argument position, on the other hand, never seems to host arguments with non-structural cases. Consider that the clearest and least controversial generalization about quirky and inherent cases in Icelandic and German is that they are never assigned to agentive subjects (Yip et al. 1987; Sigurðsson 1992; Fanselow 2000; Jónsson 2003, and many others). The instances of non-structural cases which most look like they are truly lexical – in the sense that they are not predictable from the position where a DP enters the derivation – involve internal arguments (see Woolford 2006, for recent discussion of this issue). E.g., a number of verbs in Icelandic unexpectedly assign dative to what look like typical, internal-argument themes, including transitive kasta ‘throw’ and unaccusative hvolfa ‘capsize’. These are the sorts of examples which could potentially cause problems if we want to determine based on a DP’s syntactic position whether it is going to be assigned a structural case. Arguments of these types would seem to be introduced as normal complements of the verb – a position which is typically associated with structural case. Yet they are marked with non-structural case. Interestingly enough, however, DPs in this position turn out to be irrelevant for the purposes that we are discussing here – they never play a role in the determination of structural case on lower arguments. Consider why this is. First of all, there is no additional position, which is within the VP, yet below the complement of V, where additional DPs could be introduced and be assigned a structural case.18 Verbal complements could thus only potentially affect case-assignment to DPs in embedded clauses, i.e. in ECM configurations. There are, however, apparently no verbs which simultaneously take a DP complement and an ECM complement – presumably because both would be vying for the same structural position. Thus the only DPs that can potentially be relevant for the determination of accusative case on a lower DP are those introduced outside the VP proper, in the vP layer.

18

DPs which do appear to be introduced in such a low position always seem to bear non-structural cases, perhaps because the relevant structures involve PPs. Included in this category would be e.g. the dative with German Acc-Dat verbs like aussetzen ‘expose’ (Haider 1993; McFadden 2004: ch. 4), and the various classes of Icelandic verbs taking dative or genitive objects below the structural accusative direct object (Zaenen, Maling, and Thráinsson 1985; Sigurðsson 1992).

Structural case, locality and cyclicity

123

7. The details If this is correct, the choice between structural nominative and accusative on a DP can indeed depend on whether a higher argument will have a structural case, without violating cyclicity. If there is a higher filled Spec-vP within the same minimal phase, accusative is assigned (see also Marantz 2006). Otherwise, nominative appears as the default.19 To see how this can work, let us consider the derivations of the crucial sentence types discussed above. We begin with our simple transitive (3a), repeated here as (11a), along with the associated structure repeated in (11b) with slight modifications. (11) a. Þeir völdu hana. they:N chose:PL her:A ‘They chose her.’ b. [vP Þeir v [VP völdu hana ] ] When the object is considered during the morphological cycle, the subject DP will already be present in Spec-vP, though nothing will yet be known about its morphological properties, in particular its case. The fact that it is in Spec-vP is sufficient to know that it will receive a structural case and triggers the assignment of structural accusative to the object hana.20 The subject itself will be considered for case in the next phase up, after having moved to Spec-TP (not shown in (11b)). As there is no higher DP of any kind in the structure, accusative assignment will fail, and nominative will appear as the default, correctly yielding þeir. As a contrast, consider the superficially similar sentence (12a) (from Boeckx 2000). The case marking comes out quite differently here, and the structure in (12b) shows my analysis of why: 19

20

If the featural make-up of the v head unambiguously determines whether it will take a specifier or not, this could potentially be the trigger for accusative assignment rather than the presence of the DP itself. In that case, accusative assignment would actually be triggered by features on v, as in Chomsky (2001) and related accounts. However, unlike for Chomsky, agreement would not be involved, and the assignment would crucially occur post-syntactically. Nominative ‘assignment’ as a default in my system of course remains completely different than what is standardly assumed. The actual mechanism assigning the structural accusative is of course blind to the fact that the subject DP will eventually get a structural case and depends solely on its position in Spec-vP.

124 Thomas McFadden (12) a. Henni leiddust þeir. her:D bored they:N ‘She was bored with them.’ b. [vP Ø v[Appl P Henni Appl [VP leiddust þeir ]]] The verb leiðast ‘be bored by’ assigns quirky dative to its subject, which by hypothesis is merged in the specifier of an applicative head. The actual dative will not yet have been assigned to henni at the point in the derivation when the case for the object is considered, yet it will nonetheless be clear from the structure – in particular the fact that henni is in Spec-ApplP – that there will be no higher structural-case-bearing DP in the phase. Crucially, Spec-vP is empty, thus the condition for the assignment of structural accusative to the object is not met, and nominative is assigned as a default, yielding þeir. Consider next our basic ECM example (4a), repeated here one more time as (13a), with the slightly modified structure in (13b): (13) a. Ég tel hana hafa séð myndina. I:N believe her:A have seen picture-the:A ‘I believe her to have seen the picture.’ b. [vP Ég v [VP tel [TP hanai hafa [vP hanai v [VP séð myndina ] ] ] ] ] Accusative on the embedded object myndina will be triggered by the presence of (the lower copy of) the matrix subject hana in the embedded SpecvP, essentially just as in a simple transitive sentence.21 Remember that at the time this occurs, the embedded subject hana will not yet have had its own case determined. So the fact that it will end up accusative, whereas þeir in (11a) ended up nominative, is completely irrelevant. Both subjects started out in Spec-vP, and thus license the assignment of dependent structural accusative to the objects below them. Where things get interesting here is of course in the assignment of case to the embedded subject itself. This will occur on the morphological cycle of the higher phase, after hana has raised to the embedded Spec-TP (or higher). The matrix subject in the matrix SpecvP is sufficiently local, and thus we again get structural accusative. The determination of case for the matrix subject is the same as in (11a).

21

For clarity, I have put the DPs involved in the lower dependency in italics, and those involved in the higher dependency in bold.

Structural case, locality and cyclicity

125

Finally, consider again our example of ECM on top of quirky case, repeated here as (14a), with the revised structure in (14 b): (14) a. Við töldum henni hafa leiðst strákanir. we:N believed her:D have found-boring boys-the:N ‘We believed her to have found the boys boring.’ b. [vP1 Við v [VP töldum [TP henni i hafa [vP2 Ø v[ApplP hennii Appl [VP leiðst strákanir ]]]]]] As in (12a), accusative assignment to the embedded object fails because the embedded subject is introduced as the specifier of an applicative head, and Spec-vP is empty. In this case there will eventually be a higher DP that will receive structural case – the matrix subject – but it will not be merged until the higher phase is constructed, crucially after the stage in the derivation when the embedded object is being considered for case. The embedded subject itself receives dative case on the strength of having been introduced in Spec-ApplP, and simply does not come under consideration for a structural case. It does raises to Spec-TP to satisfy the EPP, but this is unrelated to case. Finally, the matrix subject receives nominative like the others above: it is first-merged in a position that is not associated with a non-structural case, thus it is open to a structural case. There is no higher DP in the structure to trigger assignment of accusative, and thus nominative appears as the default.

8. Conclusion I have argued that it is possible to analyze morphological case-assignment in a purely local, cyclic fashion, without comparison of parallel derivations, violations of the PIC or unorthodox assumptions about phase structure. Crucial to the success of the account are a series of both theoretical and empirical considerations. On the theoretical side, I assumed that case is independent of DP licensing and is assigned on a post-syntactic morphological cycle, and that DPs destined to bear structural case can be identified on a purely syntactic basis. On the empirical side, I argued that nominative is the only case which appears to be assigned long-distance, and that it crucially also appears as the default case on DPs for which other case-assignment fails. If any of these assumptions or generalizations are successfully challenged, the analysis presented here must be altered or abandoned.

126 Thomas McFadden Of course, the discussion here has had a relatively narrow empirical base, restricted primarily to facts from Icelandic. In part this is due to the practical restriction that relatively few languages have all the characteristics that allow clear investigation of potential long-distance case dependencies (as pointed out in fn. 6 above).22 Full-fledged ECM in particular seems to be quite rare.23 However, this is also because my primary goal in this paper has not been to examine the full array of attested case-marking patterns, but to work out what is predicted to be possible if we apply fairly restricted views on locality and cyclicity to the issues of case-assignment. If the generalization I proposed about long-distance case in Section 2 turns out to be incorrect when additional languages are considered, then some revision will be needed – not just in how we handle case-assignment, but also in the implementation of locality restrictions in general.

Acknowledgements I would like to thank Artemis Alexiadou, Berit Gehrke, Kirsten Gengel, Nino Grillo, Kleanthes Grohmann, Alec Marantz, Florian Schäfer, Halldór Sigurðsson and Sandhya Sundaresan for comments and helpful discussion.

22

23

Particularly notable is the absence of any discussion of ergative-absolutive languages. There is good reason to believe that ergative is actually an inherent case (see Woolford 2006; Anand and Nevins 2006, for some recent discussion), while the absolutive may well be a morphological default. Without something like the dependency relationship of accusative assignment, it difficult to even construct examples that would challenge the locality of case-assignment. German for example has an ECM-like construction with causative and perception verbs, but this involves a demonstrably smaller structure than true ECM with different locality properties (Wurmbrand 2002; McFadden 2004). Latin, on the other hand, allows overt accusative subjects of infinitives in a variety of contexts where there is no external source for that case. (vanden Wyngaerd 1994). It is apparently coming from within the non-finite clause and thus again can tell us little about locality.

Structural case, locality and cyclicity

127

References Anand, Pranav and Andrew Nevins 2006 The locus of ergative case assignment: Evidence from scope. In Ergativity: Emerging Issues, Alana Johns, Diane Massam, and Juvenal Ndayiragije (eds.), 3–27. Dordrecht: Kluwer. Bobaljik, Jonathan David 2006 Where’s Ф? Ms., University of Connecticut. Boeckx, Cedric 2000 EPP eliminated. Ms., University of Connecticut. Bowers, John 2002 Transitivity. Linguistic Inquiry 33: 183–224. Burzio, Luigi 1986 Italian Syntax. Boston: Reidel. Chomsky, Noam 2000 Minimalist inquiries: The framework. In Step by Step: Essays on Minimalism in Honor of Howard Lasnik, Roger Martin, David Michaels, and Juan Uriagereka (eds.). Cambridge, MA: MIT Press. 2001 Derivation by phase. In Ken Hale: A Life in Language, Michael Kenstowicz (ed.). Cambridge, MA: MIT Press. 2008 On phases. In Foundational Issues in Linguistic Theory: Essays in Honor of Jean-Roger Vergnaud, Robert Freidin, Carlos P. Otero and Maria Luisa Zubizarreta (eds.), 133–166. Cambridge, MA: MIT Press. Cuervo, María 2003 Datives at large. Doctoral Dissertation, MIT. Embick, David and Rolf Noyer 2001 Movement operations after syntax. Linguistic Inquiry 32. 2007 Distributed Morphology and the syntax/morphology interface. In The Oxford handbook of linguistic interfaces, Gillian Ramchand and Charles Reiss (eds.). Oxford: Oxford University Press. Fanselow, Gisbert 2000 Optimal exceptions. In Lexicon in Focus, Barbara Stiebels and Dieter Wunderlich (eds.), 173–209. Berlin: Akademie Verlag. Frank, Robert 2006 Phase theory and Tree Adjoining Grammar. Lingua 116: 145–202. Gehrke, Berit 2007 On directional readings of locative prepositions. In Proceedings of ConSOLE XIV, 99–120. Haider, Hubert 1993 Deutsche Syntax – generativ. Tübingen: Gunter Narr. Halle, Morris and Alec Marantz 1993 Distributed morphology and the pieces of inflection. In The View from Building 20: Essays in Linguistics in Honor of Sylvain Bromberger, Ken Hale and Samuel Jay Keyser (eds.). Cambridge, MA: MIT Press.

128 Thomas McFadden Harley, Heidi 1995 Subjects, events and licensing. Doctoral Dissertation, MIT. Hole, Daniel, André Meinunger, and Werner Abraham (eds.) 2006 Datives and Other Cases. Amsterdam: John Benjamins. Jónsson, Jóhannes Gíssli 2003 Not so quirky: On subject case in Icelandic. In New Perspectives on Case Theory, Ellen Brandner and Heike Zinsmeister (eds.), 127–163. Stanford, CA: CSLI Publications. Kratzer, Angelika 1996 Severing the external argument from its verb. In Phase Structure and the Lexicon, Johan Rooryck and Laurie Zaring (eds.), 109–137. Dordrecht: Kluwer. Legate, Julie Anne 2003 Some interface properties of the phase. Linguistic Inquiry 34: 506– 515. 2008 Morphological and abstract case. Linguistic Inquiry 39: 55–101. 2005 Phases and cyclic agreement. In Perspectives on Phases, Martha McGinnis and Norvin Richards (eds.), MITWPL 49: 147–156. Marantz, Alec 1984 On the nature of grammatical relations. Cambridge, MA: MIT Press. 1991 Case and licensing. In ESCOL’91: Proceedings of the 8th Eastern States Conference on Linguistics, 234 –253. 2006 Phases and words. Ms., New York University. McFadden, Thomas 2004 The position of morphological case in the derivation: A study on the syntax-morphology interface. Doctoral Dissertation, University of Pennsylvania, Philadelphia. 2007 Default case and the status of compound categories in Distributed Morphology. In Proceedings of the 30th Penn Linguistics Colloquium, Tatjana Scheffler, Joshua Tauberer, Aviad Eilam, and Laia Mayol (eds.), 225–238. McGinnis, Martha 2005 UTAH at Merge: evidence from multiple applicatives. In Perspectives on phases, Martha McGinnis and Norvin Richards (eds.), MITWPL 49: 183–200. Phillips, Colin 2003 Linear order and constituency. Linguistic Inquiry 34: 37–90. Pylkkänen, Liina 2002 Introducing arguments. Doctoral Dissertation, MIT, Cambridge, MA. Reuland, Eric (ed.) 2000 Arguments and Case: Explaining Burzio’s Generalization. Philadelphia: John Benjamins.

Structural case, locality and cyclicity

129

Richards, Norvin 2007 Lardil “case stacking” and the structural / inherent case distinction. Ms., MIT. Schäfer, Florian 2005 The dative causer construction in German. Ms., Universität Stuttgart. Schütze, Carson 1997 Infl in child and adult language: Agreement, case and licensing. Doctoral Dissertation, MIT. 2001 On the nature of default case. Syntax 4: 205–238. Sigurðsson, Halldór Ármann 1992 The case of quirky subjects. Working Papers in Scandanavian Syntax 49: 1–26. 2000 The locus of case and agreement. Working Papers in Scandinavian Syntax 65: 65–108. 2003 Case: Abstract vs. morphological. In New perspectives on case theory, Ellen Brandner and Heike Zinsmeister (eds.). Stanford, CA: CSLI Publications. 2006a The nom/acc alternation in Germanic. In Comparative Studies in Germanic Syntax, Jutta Hartmann and László Molnárfi (eds.), 13–50. Amsterdam: Benjamins. 2006b The nominative puzzle and the low nominative hypothesis. Linguistic Inquiry 37: 289 –308. Sundaresan, Sandhya 2006 The argument structure of verbal alternations in Tamil. In Proceedings of the 25th West Coast Conference on Formal Linguistics, Donald Baumer, David Montero, and Michael Scanlon (eds.), 390 –398. Woolford, Ellen 2003 Burzio’s Generalization, markedness, and locality constraints on nominative objects. In New Perspectives on Case Theory, Ellen Brandner and Heike Zinsmeister (eds.), 301–329. Stanford, CA: CSLI Publications. 2006 Lexical case, inherent case and argument structure. Linguistic Inquiry 37: 111–130. Wurmbrand, Susi 2002 Syntactic vs. semantic control. In Studies in Comparative Germanic Syntax: Proceedings of the 15th Workshop on Comparative Germanic Syntax, C. Jan-Wouter Zwart and Werner Abraham (eds.). Amsterdam: John Benjamins. Vanden Wyngaerd, Guido 1994 PRO-Legomena: Distribution and Reference of Infinitival Subjects. Berlin /New York: Mouton de Gruyter. Yip, Moira, Joan Maling, and Ray Jackendoff 1987 Case in tiers. Language 63: 217–250.

130 Thomas McFadden Zaenen, Annie, Joan Maling, and Höskuldur Thráinsson 1985 Case and grammatical functions: The Icelandic passive. Natural Language and Linguistic Theory 3: 441– 83. Zwart, Jan-Wouter to appear Structural case and dependency marking: A neo-Jakobsonian view. In CLS 42 Panels Volume.

PRO, pro and NP-trace (raising) are interpretations M. Rita Manzini

Generative literature has paid considerable attention to the interpreted nonlexicalized subjects of sentences like English infinitivals or Balkan subjunctives – i.e. to control, raising, null subjects and their interactions (e.g., Davies and Dubinsky 2006). In this article we propose to unify these phenomena under a single syntax at the LF interface; only the interpretations of this syntax make them different. The position we take is not extraneous to previous proposals in the literature, but it does not coincide with any of them. For instance, the unification of raising and control is at the core of Hornstein (1999) – contra Chomsky (1981ff.) and more recently Landau (2000ff.); but because null subjects are also included in the present unification attempt, this cannot involve the classical generative notion of movement – contra Hornstein (1999) and in agreement with Landau (2003 ff.). In general, we argue that control, raising and null subjects are based on the same syntactic configuration at the LF interface, while their differences depend on relations of chain formation, binding, quantification (or lack thereof) defined again at LF. Since we treat the relevant EPP arguments as variables we are close to their conception as pure theta-roles in Manzini and Roussou (2000). But as pointed out by Roussou (2007), the assimilation of EPP variables to theta-roles cannot hold because of Balkan subjunctives, where the EPP argument is independently lexicalized. The theory we propose here is motivated by its ability to provide an adequate analysis not only of the interpretive patterns, but also of the lexicalization patterns that correspond to them, including observed cross-linguistic variation. In this respect, we are also close to Culicover and Jackendoff’s (2005 ff.) criticism of what they call the Interface Uniformity approach, roughly to the effect that different interpretations are isomorphically mapped to different structures – abstractly different, if not overtly so; we further agree with their conclusion that obligatory control is semantically and not syntactically defined. We disagree with them on the overall architecture of the grammar – which in our case is simply a representational variant of the minimalist model (Brody 1997 ff., Manzini and Savoia 2005, 2007). In section 1 we briefly review the relevant literature, focussing on control and its relations to raising and null subjects; in section 2 we introduce

132 M. Rita Manzini relevant data from Albanian and from Italian – again with a critical eye to previous proposals. In section 3 we articulate our main proposal based on the data of Albanian. As anticipated above, our account rests on a close examination of the morpholexical properties of Albanian (including dialectal variation). It is on the basis of a single syntax that takes these properties into accounts that we derive the different control, raising and null subject interpretations (section 3.1). We apply the same general approach to Italian infinitivals in section 4 and to the Geg Albanian paskajore (‘infinitive’) in section 4.1. In the way of a conclusion we briefly discuss how the classical unification between raising and passive (unaccusatives, etc.) achieved in terms of the transformational notion of movement may or may not be maintained within the present framework. The answer that is prospected is that objectto-subject chains also correspond to an interpretive generalization; the conclusion that there is a single, movement syntax underlying it must be evaluated in the light of the different morpholexical structures instantiating it, before the question of its relation to ‘raising’ can be properly asked. Needless to say, we leave this for further work (Manzini and Savoia, to appear). 1. Background The issue of empty categories is central to generative grammar and shapes its various frameworks. In Standard Theory all empty positions depend on transformations; thus a deletion operation, EQUI(valent)-NP Deletion, is proposed by Rosenbaum (1967) to explain control. The GB framework revises this approach, in that the interpreted unpronounced positions that control and pro-drop involve, correspond to base generated empty categories. The movement transformation is affected by this change, in that it itself introduces an empty category, i.e. the trace. An algorithm partitions empty categories into the ±anaphoric, ±pronominal classes (Chomsky 1982). In particular, +anaphoric e.c.’s are subject to Binding Principle A and pronominal ones to Binding Principle B – determining a contradiction for [+pronominal, +anaphoric] empty category, i.e. PRO, which can only be avoided if the latter is in ungoverned position (hence not subject to Binding Theory). Purely pronominal e.c.’s (pro’s) contrast with non-pronominal ones (traces) because of the need for the latter to be governed (ECP). Finally, purely anaphoric e.c.’s (A-traces) differ from other traces and pro’s alike in that the well-formedness of chains requires them to be Case-less – a property which they share with ungoverned PRO’s. In turn this partition determines interpretation.

PRO, pro and NP-trace (raising) are interpretations 133

From the present point of view, the minimalist framework of Chomsky (1995) is a hybrid between Standard Theory and GB. Movement does not introduce an empty category, but is conceived in terms of an operation remerging material already merged – and deleting /not pronouncing its original copy. At the same time, there is no attempt at reverting to a derivational account for PRO and pro, which are kept as base generated empty categories. Because there is no longer an empty category space that can be interestingly partitioned, the properties that determine the insertion of PRO and pro and their interpretation are associated with them in the lexicon. Thus pro must have the phi-features that allow it to check the uninterpretable agreement features of the verb; while PRO must have a null Case matching that of infinitival inflections. At least, this is the case if the idea of Chomsky (1995) that projection is from the lexicon is strictly adhered to (as it will be here) – in other words, it is not the case that feature bundles are freely generated in the syntax to be then spelled out (or not) at the Morphological Structure interface, as under Halle and Marantz’s (1993) Late Insertion. In the GB framework a quite general problem attaches to PRO, namely that control, i.e. the theory responsible for assigning an antecedent to PRO, remains a separate module of the theory, not (entirely) reducible to other modules. For instance Manzini (1983) proposes a reduction of control to Binding Theory, but leaves open the question as to why within a given domain, an argument would be chosen as the obligatory controller rather than another, yielding in particular subject vs. object control. Since Chomsky’s (1995) minimalism adopts the GB theory of PRO and pro, it also inherits these problems. In addition to these, the ad hoc character of PRO and pro within minimalism, notably the fact that they are effectively empty lexical elements, endowed with such features as null Case has often been remarked upon (Manzini and Roussou (2000) on PRO, Manzini and Savoia (2005, 2007) on pro, and references quoted there). More radical solutions to the question of control within the GB framework start with the observation that though PRO and pro supposedly do not overlap, many PRO contexts allow for a pro reading. These are the contexts characterized by Williams (1980) in terms of non-obligatory (as opposed to obligatory) control. Thus Bouchard (1984) associates environments of nonobligatory control to pro. Though the parallel is not drawn by the literature, a similar problem extends to the opposition between PRO and traces as well. Thus it has been known for a long time (Ruwet 1972) that there are environments that support both PRO and raising interpretations, for instance the infinitival complements of verbs like to threaten. In this case there is a standard solution which treats the PRO (control) and A-trace (raising) inter-

134 M. Rita Manzini pretations as corresponding to a structural ambiguity between government into the embedded clause and lack thereof (Rouveret and Vergnaud 1980). As the underlying problems are conceptually close, so are the solutions proposed to them. Thus the ambiguity of infinitival complements between the control and raising readings is imputed to a structural ambiguity between types of infinitival complements (roughly CP’s opaque to government vs. IP’s transparent to it). Similarly, the distinction between obligatory and nonobligatory control is built into a structural ambiguity – since the two readings are imputed to two different empty categories (say PRO and pro). Viceversa Borer (1989) responds to the problems created by the overlapping of pro and PRO properties by concluding in favour of a unique base-generated empty category (pro rather than PRO). Structural unification corresponding to diverging interpretations is also the solution proposed for the control/ raising problem by Hornstein (1999), according to whom the common properties of control and A-movement (raising) are to be unified under a movement derivation. Control is the interpretation of the resulting chain when more than one theta-role is assigned to it. At the other end of the spectrum, Landau (2004, 2006) maintains the rigid partition of underlying structures on which the different PRO, pro and A-trace readings depend – though he abandons any attempt at defining PRO and pro in terms of the classical notions of government or Case, turning instead to interpretive features such as tense(less)ness. In fact, much of the discussion that precedes revolves around what Culicover and Jackendoff (2006: 132) term Interface Uniformity: i.e. “the syntaxsemantics interface is maximally simple, in that meaning maps transparently into syntactic structure; and it is maximally uniform in that the same meaning always maps into the same syntactic structure. If such structure is not present at the surface, it is nevertheless present at some covert level of structure”. Culicover and Jackendoff (2006) are themselves critical of Uniformity and propose a theory of control and raising that reflects this critical stance. For Culicover and Jackendoff (2006), control is a binding of arguments in what they term Conceptual Structure. The arguments are then linked to Grammatical Functions at an intermediate level of representation – with one of the GF’s (conventionally PRO) not linked to any position in Syntactic Structure. Because of this, at Syntactic Structure a control sentence turns out to be identical to a raising one. The difference is that in raising there is no binding of argument positions at Conceptual Structure – rather there is a coindexing of GF’s in the intermediate structure. By contrast, the present work accepts the overall architecture of a minimalist grammar – motivated by what Chomsky (1995) calls ‘bare conceptual

PRO, pro and NP-trace (raising) are interpretations 135

necessity’. Crucially LF representations interface directly with ConceptualIntentional systems without the mediation of Grammatical Function representations. This does not prevent us from sharing the concern with Interface Uniformity. Thus for instance the fact that Landau (2004, 2006) is eventually able to obtain isomorphism between interpretations and structural representations does not necessarily count as an advantage for us. On the contrary we agree with Roussou’s (2007) criticism that ‘by positing abstract features on the embedded C and I heads, this analysis presupposes the interpretation that will be derived at a later stage’ ‘since in morphological terms C and I are alike’. In this perspective, the entire work of Manzini and Savoia (2005, 2007) can be read as an illustration of the theme that enforcing uniformity of the syntax with the interpretation leads to a loss of predictive power at the PF (morpholexical) interface. As we follow Culicover and Jackendoff (2006) in taking issue with Interface Uniformity as a driving principle for the determination of syntactic structures – so we take issue with the complete decoupling of syntax and morphology that typically connects with it. A good example of this is the ‘Radically Disentangled Morphology’ of Sigurðsson (2007: 23), where syntax ‘does not operate with or on morphological entries like NOM(inative) or FEM(inine). That is, syntax exclusively operates with abstract features and abstract roots, ROOT99, etc.’. By contrast, as we already mentioned above, we reject Late Insertion and we assume that syntax is strictly built from actual lexical entries. If abstract predicates are to be postulated, it is much more likely that they’ll be found at Conceptual Structure in the sense of Culicover and Jackendoff. Because of the considerable overlapping between raising and control we share the view that they should be syntactically unified – a point that Culicover and Jackendoff (2006) have in common with Hornstein (1999) as well as with Manzini and Roussou (2000). At the same time, we not only reject Culicover and Jackendoff’s (2006) analysis in terms of Grammatical Functions, but we also agree with Chomsky (2000) and Landau (2003), contra Hornstein (1999), that control cannot be unified with movement within a minimalist grammar, because theta-roles are relations and cannot be treated as features. Finally, in Balkan languages, finite sentences introduced by the so-called subjunctive particle would seem to represent a case in which a pro subject supports the control and raising readings as well. This distribution induces Landau (2004, 2006) to treat control as due to the tenselessness of the relevant sentences, rather than to the lack of agreement, since the latter is overtly lexicalized on the verb. The treatment of Boeckx and Hornstein

136 M. Rita Manzini (2006) is strictly parallel, since they allow for control and raising into finite clauses by assuming that tenselessness gives them (i.e. their C phase) a defective character. Here we shall begin our discussion with Albanian dialects, showing that control /raising sentences are clearly endowed with tensed agreeing verbs. We shall furthermore indicate how problems similar to those of Albanian arise in a more familiar Romance language like Italian, endowed with an infinitive form of the verb.

2.

Some data

2.1. Albanian 1 It is a common property of so-called Balkan languages that they lack a morphological infinitive; nevertheless they do not lack the range of interpretations that are associated with the infinitive in English. Rather control and raising interpretations are associated with a particular subset of finite embedded sentences, introduced by a so-called subjunctive particle. In what follows we shall show that Albanian embedded sentences introduced by the të particle do indeed support the readings associated with pro, PRO and NP-trace (raising). (1) illustrates the pronominal null subject reading of të sentences. Note that the embedded verb is overtly inflected for the subjunctive, as shown by the palatalized final consonant. In this and the following examples the të particle alternates between the syllabic form tə and the elided form t on phonological grounds. (1)

Vena di Maida

u dua t ɛ bɐtʃ. I want Prt it you.do(subj) ‘I want you to do it.’ One of the interesting properties of të is that while it allows for any reference of the embedded subject, it does not allow for its lexicalization in the preverbal position, at least following të; a subject can instead be lexicalized in the postverbal position, as in (2). But under conventional assumptions, 1

The data come from Arbëresh dialects spoken in Southern Italy as well as from varieties in Albania itself (Gjirokastër and Shkodër). This material has been collected in fieldwork by Leonardo Savoia and is reproduced here from Manzini and Savoia (2007) to which reference should be made.

PRO, pro and NP-trace (raising) are interpretations 137

this means that the preverbal subject position in the embedded sentence in (2) supports an expletive pro reading. (2)

Carfizzi dua tə viɲ a» i. I.want Prt comes(subj) he ‘I want him to come.’

As we expect, coreference between the matrix subject and the embedded subject is also possible, as in (3a). What is problematic however is that the coreferential reading is actually obligatory with matrix verbs of obligatory control, such as the (deontic) modal in (3b). If the subject in (3b) is pro and pro has pronominal properties, we expect disjoint reference from the matrix subject to be equally possible – contrary to fact. Only the presence of a PRO in the subject position of the embedded clause can guarantee the obligatory coreference (control) reading. But of course, in terms of the theory of Chomsky (1995), PRO cannot be the subject of a sentence where nominative case is assigned, as in (2), since PRO requires the specialized null Case. (3)

Civita a. jam kuntjɛ nt t ɛ I.am happy Prt him ‘I am glad to see him.’ b. kiʃ / pa I.had(imprf) / I.had(prf) ‘I had to do it.’

ʃɔ x. I.see t ɛ bə ja. Prt it I.did

In the theory of Chomsky (2001 ff.), where a finite inflection automatically implies nominative Case, the problem is if anything starker. Iatridou (1993) argues that Greek na sentences (the counterpart to Albanian të sentences) are non-finite, in the sense that they lack temporal properties; therefore nominative is not necessarily assigned. Iatridou’s (1993) evidence concerns the impossibility of embedding past tense under Greek na. However temporal specifications can be embedded in Albanian, for instance the past tense in (3b). It is true that the embedded past specification in (3b) appears to be in a sequence of tense relation to the matrix one. But this is true also of subjunctives in Romance languages like Italian, where subjunctive sentences do not enter into control configurations and nominative is normally assigned.

138 M. Rita Manzini Terzi (1997) takes the route of claiming that despite appearances, the interpretive ambiguity between PRO and pro readings of na sentences in Greek actually correspond to a structural ambiguity. Thus in particular, the null Case of PRO is licenced by the particle na, taken to be a M(ood) head, while the pro reading is licenced in an altogether different configuration where the na+verb complex has moved to C. The latter structure would make Greek sentences analogous to infinitivals with overt nominative subjects in Italian according to the analysis in Rizzi (1982). However, as we shall argue in section 4, the structure postulated by Rizzi (1997) cannot be correct for Italian – weakening the case for Greek as well. Landau (2004) frankly abandons any attempt at maintaining that the distribution of pro an PRO is determined by notions such as government (GB) or Case (minimalism). In Landau’s (2004) theory the presence of PRO, as opposed to pro, can be determined either by lack of agreement or directly by lack of independent temporal specification, the latter relevant for the cases at hand. However, as pointed out above, in our Albanian examples the lack of temporal specifications does not correspond to any independently observable PF (morpholexical) property; hence it can only be inferred backwards from the control interpretation, leaving us with what is essentially a circular explanation. The sequence of tenses configuration displayed by examples such as (3b) does not indicate tenselessness either – since, as we noted, it is displayed by Italian subjunctives as well, which do not support the (purportedly tenseless) control reading. Another problem yet arises in connection with the fact that an embedded të sentence can receive a raising reading. A relevant example is in (4); since under ordinary assumptions (Moro 1997) the verb ‘to be’ is a raising verb, its argumental subject must be raised from the embedded të sentence. Note that in (4) clitics are normally positioned under the të particle, both verbs are inflected, etc., confirming that a bisentential structure is involved. Though examples with the canonical raising verb to seem are missing from the Manzini and Savoia (2007) data, we know that they are routinely accepted for instance in Greek (Anagnostopoulou 2003). (4)

Portocannone a. jam tə tə srɛ s. I.am Prt you I.call ‘I am (about) to call you.’ b. iʃ i tə mə srisi. he.was Prt me he.called ‘He was (about) to call me.’

PRO, pro and NP-trace (raising) are interpretations 139

In standard minimalist terms, raising depends on the fact that the raised constituent has a Case to be checked. But as already argued in discussing the PRO interpretation, nominative Case is obviously assigned in të sentences; therefore raising the embedded subject should be impossible in (4) and the like. Anagnostopoulou (2003) argues that the Case problem can be circumvented in Greek along the lines of Iatridou (1993), and provides evidence showing that the raising reading is connected to the absence of tense properties on the embedded verb. Once again this appears not to be the case in Albanian, as indicated by the alternation of present and past tense in the two examples in (4). And once again, the sequence of tenses relation in (4) is not sufficient since it is shared with a number of environments (such as Italian subjunctives) where no raising is possible. The line of explanation based on lack of tense is also followed by Boeckx and Hornstein (2006), in connection with their unification of raising and control. As we commented above for Landau (2004, 2006) on control, in the absence of any morpholexical evidence, the supposed tense defectivity of finite raising complements is inferred a posteriori from the raising interpretation itself – in order to maintain consistency with the overall model of grammar. In other words, it is no more and no less than a stipulation. 2.2. Italian Sentences introduced by the so-called subjunctive particles of Balkan languages are not the only environments where pro, PRO and raising interpretations overlap. The same effect is found in languages which have infinitives, such as Italian. If in the case of Albanian të sentences, the occurrence of pro is predicted, and PRO and NP-trace (raising) are problematic, the reverse is true in Italian, on the assumption that infinitives assign null or no Case. In fact the first problem for the analysis of Italian is raised precisely by the fact that nominative is available in infinitival clauses. In particular, as pointed out by Burzio (1986), the controlled subject of an infinitival clause can be doubled by an ‘emphatic’, i.e. focussed, pronoun, which in Italian surfaces in the nominative, as in (5). Borer (1989) puts this fact in the proper cross-linguistic context, of lexicalization of controlled subjects by overt pronouns – as does Szabolcsi (2007) more recently. (5)

Tentai di andarci io. I.tried to go.there I ‘I tried to go there myself.’

140 M. Rita Manzini Belletti (2001) analyzes io in (5) as occupying the ordinary focussed position of a postverbal subject, while assuming that its nominative Case is simply the ‘default’ in Italian. Whatever we make of this analysis, the occurrence of the ‘emphatic’ pronoun in obligatory control contexts leads us to expect that a free pronoun or a full lexical subject is possible in nonobligatory control contexts. Indeed this expectation comes true. In (6) we exemplify three classical contexts in which so-called arbitrary control is possible, namely adjunct sentences in (6a), subject sentences in (6b) and main sentences infinitivals (with an exclamative/ interrogative modal interpretation) as in (6c). In all cases, removing the parenthesized noun phrase will yield the arbitrary reading. Otherwise the parenthesized noun phrase will be read as the (nominative) focussed postverbal subject of the sentence. (6) a. Tutto è successo per aver scioperato (i dirigenti / anche loro). all is happened for to.have struck the heads / also they ‘All happened because department heads/they as well went on strike.’ b. Aver scioperato (i dirigenti / anche loro) è stato un errore. to.have struck the heads / also they is been a mistake ‘(For department heads / them as well) to go on strike was a mistake.’ c. Scioperare (i dirigenti / anche loro)?! Impossibile! to.strike the heads / also they impossible ‘(For department heads / them as well) to go on strike?! Impossible!’ Following the proposal of Belletti (2001) concerning ‘emphatic’ pronouns of the type in (5), it is natural to assign the parenthesized noun phrases in (6) to the ordinary focussed postverbal subject position. But this makes them into the associates of an expletive EPP element, i.e. conventionally a pro. Of course, one may want to say contra Belletti (2001), that infinitival sentences in Italian have a bona fide nominative Case to check, so that their subject can be pro. However such an assumption creates more problems than it solves, since it cannot extend to (5), on pains of not being able to predict control. Furthermore recall that in the contexts in (6) the arbitrary interpretation (conventionally PRO) overlaps with the expletive one (conventionally pro). As we expect, the same contexts in (6) that licence the expletive pro interpretation, also licence the pronominal pro interpretation, as in (7), where the empty subject of the infinitival can refer deictically, a fact noticed by Bresnan (1982) for English.

PRO, pro and NP-trace (raising) are interpretations 141

(7)

a. Tutto è successo per non essermi svegliata in tempo. all is happened for not to.be.myself woken up on time ‘All has happened because I didn’t wake up on time.’ b. Non essermi svegliata in tempo è stato un errore. not to.be.myself woken.up on time is been a mistake ‘(For me) not to wake up on time was a mistake.’ c. Non svegliarmi in tempo?! Impossibile! not to.wake.up.myself on time impossible ‘(For me) not to wake up on time?! Impossible!’

By contrast with pro interpretations, the overlapping of raising and control in infinitival context has routinely been built into generative grammar analyses. A good example of the overlapping is provided by the verb minacciare ‘to threaten’, whose control interpretation is very distinctive, requiring in particular an animate external argument. Yet inanimate and expletive subjects are equally possible with this verb, in a different interpretation, roughly ‘there is a threat that’. Thus (8a) is ambiguous, though the control interpretation is more easily accessible, while (8b–c) must be construed as instances of raising. (8)

a. Gianni minaccia di non imparare. John threatens to not learn ‘John threatens not to learn/ There is a threat of John not learning.’ b. La foresta vergine minaccia di scomparire. the forest virgin threatens to disappear ‘There is a threat of virgin forests disappearing.’ c. Minaccia di piovere. it.threatens to rain ‘There is threat of rain.’

The classical solution to the ambiguities in (8) within generative grammar is to say that they involve two different sentential structures, including a C layer (the control structure) or not including it (the raising structure). Because C is the phase-defining head in minimalism (Chomsky 2001 ff.), its absence implies lack of Case assigning properties on the embedded verb and hence forces (as well as allowing) raising. However, as far as we can see, the proposal concerning the (abstract) presence vs. absence of the C layer does not explain any behaviour other than the difference between the control and raising interpretations themselves. For instance implementing

142 M. Rita Manzini this idea in (5) requires di ‘of’ not to be a complementizer – a conclusion that we may uphold for independent reasons (Manzini and Savoia 2005, 2007; Kayne 2000). But then we are left without any apparent complementizer layer for the control reading as well. In short, the overlapping of control (PRO) and raising (NP-trace) in the same contexts may have an effective description in current frameworks; however this is no more (and no less) than a stipulation, to the extent that no independent behavior follows from it. To summarize, infinitival sentences in Italian display essentially the same referential spread for the EPP argument as we already observed for Albanian të sentences. The overlapping of control and raising readings has long been built into the theory; but the appeal to structural ambiguities, while accurately describing the data, does not really explain it – in the sense that they do not correlate with independent behaviors. As far as we can see, the overlapping of pro and PRO interpretations remains largely unrecognized, to the extent that data involving postverbal nominative subjects (other than ‘emphatic’ pronouns) are not generally discussed in the literature. Their existence in Romance languages other than Italian is documented by Manzini and Savoia (2005, vol. I: 680–683) and references quoted there.

3. Albanian të The literature contains various proposals concerning the status of particles of the type of Albanian të. One conclusion generally agreed upon is that these particles cannot be identified with I, for more than one reason. Thus the verb is itself inflected and hence presumably appears in I; furthermore, the particle always precedes object clitics, which in turn precede the verb (in I). On the other hand, in theories which have only one C position at their disposal the subjunctive particle cannot be assigned to the C position either, because the complementizer can cooccur with it, as illustrated in (9) for the Arbëresh variety of Carfizzi and the Albanian variety of Gjirokastër. Note that there appears to be a specialization of the complementizer for të, which is sa in (9a) or tʃə in (9d) in the presence of të – though it is sɛ in examples like (9b) or (9e) without të. What sa and tʃə appear to have in common is that they are independently attested as wh-elements, as illustrated in (9c) and (9f) a property reminiscent of Romance complementizers, cf. Manzini and Savoia, to appear).

PRO, pro and NP-trace (raising) are interpretations 143

(9)

Carfizzi a. i kɔm prɔmɛtirtu sa t vɛtɛ. to.him I.have promised that Prt I.go ‘I promised him to go.’ b. i kɔm prɔmɛtirtu sɛ vɛtɛ. to.him I.have promised that I.go ‘I promised him to go.’ c. sa gra jɔn? how.many women are ‘How many women are they?’ Gjirokastër d. tə θatʃ tʃə t a laɲɛ. to.you I.said that Prt him/her you.washed ‘I told you to wash him/her.’ e. tə kam prəmtuaɾ sɛ dɔ vi. to.you I.have promised that fut. I.come ‘I promised you that I will come.’ f. tʃə mən»dɔn sɛ tʃə bən a»i what you.think that what do he ‘What do you think that he is doing?’

In order to accommodate subjunctive particles, Rivero (1994) postulates an additional M(ood) node between I and C. This treatment is specifically advocated for Albanian by Turano (1993). On the other hand, as observed by Roberts and Roussou (2003), within an articulated theory of the C field, different C positions are in principle available to host both the subjunctive particle and the that-type complementizer. Roberts and Roussou (2003) propose an analysis of Greek in which the na particle appears in an Op position, forming part of the C field of the left periphery, where it can be preceded by a declarative complementizer like pou inserted in a higher position of the same C field. It is the min negation, when present, which lexicalizes the lowest position of the C field, namely M(ood), following therefore na. Within the same framework Roberts and Roussou (2003) propose that other particles – i.e. the mu particle of (Romance) Southern Italian dialects – appear in the lowest complementizer position M, where they can be preceded by the negation. These analyses agree in associating a modal value to the so-called subjunctive particle, either through direct insertion in a M(ood) position, or

144 M. Rita Manzini more generally through insertion in the modal C field. Thus the combination of the verb with the particle is predicted to yield modal properties, which may not be present in the verb morphology. In particular Greek does not embed a morphological subjunctive under the na particle, but rather an ordinary indicative. There are two types of objections that one can raise against this analysis, at least when applied to Albanian. The first one is that, as we pointed out in the presentation of the data, at least in Albanian the verb independently carries subjunctive morphology; therefore a modal characterization of the particle is not necessary to express the modal character of the construction it enters into. The other objection is more directly pertinent to the topic under discussion here – namely that a modal construal of the particle is not sufficient to explain why obligatory control and raising interpretations are licensed in the contexts where it is present, contrary to other finite contexts both in Albanian and in other languages. In order to overcome these difficulties, here we shall take a different approach to të. As it turns out, there is another major occurrence of the morpheme të in Albanian, namely as an article. Albanian nouns and adjectives generally bear an inflection that is specified not only for number and gender but also for Case and for definiteness. What is relevant for us is that certain subclasses of nouns and adjectives also require a preposed article. Thus in the standard, të is the preadjectival article for, among others, the indefinite plural (Camaj 1995: 82). An example of this is provided in (10a), taken from the Geg dialect of Shkodër, but reflecting the conditions described for the standard.2 A similar inflectional structure characterizes a subset of nouns including kinship terms, where the prenominal article appears in the definite paradigm (Camaj 1995: 62). The example in (10b), again from the Geg dialect of Shkodër, illustrates të in the accusative feminine singular. (10) Shkodër a. jan t kutʃ / t kutʃ-ɛ. they.are Art red / Art red-Infl ‘They are red(m./f.).’ b. pɒ t mɔtɾ-ɛn. I.saw Art sister-Infl ‘I saw my sister.’ 2

A major distinction is made in Albanian linguistics between Tosk dialects, including the standard variety and Arbëresh dialects, and Geg dialects of Northern Albania and the Kosovo.

PRO, pro and NP-trace (raising) are interpretations 145

Finally të shows up as the article in front of nominalized adjectives and participles (Camaj 1995: 64). We exemplify this with a nominalized participle in (11) for the standard variety spoken in Gjirokastër; specifically (11a) illustrates the nominative, (11b) the accusative and (11c) the oblique (genitive). Note that in the genitive (11c), the second article is the one we are concerned with; the first article agrees with the head of the noun phrase of which the genitive is a complement, as if in English one were to say ‘the time, that of chatting’. Note also that in the accusative in (11b) the nominalized participle is doubled by the ɛ clitic; clitic doubling is not specific to the construction at hand but a quite general phenomenon of the language. (11) Gjirokastër a. mə pəl»cɛn tə ða»fɔsuɾ-it. to.me likes Art chatted-Infl ‘I like a chat.’ b. ɛ dua tə ða»fɔsuɾ-it. it I.want Art chatted-Infl ‘I like (it,) a chat.’ c. əʃt kɔh-a ɛ tə ða»fɔsuɾ-it. it.is time Art Art chatted-Infl ‘It is the time for a chat.’ We can assign the article that appears in front of adjectives and nouns to the D position of the noun phrase, as shown in (12) for the examples in (10). The position of the (inflected) adjective or noun is taken to be I, on a par with the position of the inflected verb – assuming a complete parallelism between the basic skeleton of verbal and nominal structures. Note that following Manzini and Savoia (2005, 2007), we provide labels only for heads and for their complements/specifiers, assuming that some algorithm (that we do not want to commit ourselves to) eventually labels other constituents. In any event, we adhere to minimalist principles of projection from the lexicon and Inclusiveness – on the basis of which labels exist only as abstractions from the properties (feature sets) of lexical items. (12) Shkodër a.

wo D wo t I kutʃ /kutʃɛ

b.

wo D wo t I mɔtɾ-ɛn

146 M. Rita Manzini A similar structure will hold for the nominalized participle in (11). In (13) we provide a structure for the example in (11b), where we assume that the participial expression is the accusative object of the main verb. The label of the nominalized participle as well as of the ɛ object clitic doubling it is N(P) is, because we follow Manzini and Savoia (2005, 2007) in reducing the accusative property to the N(ominal class) property. More conventional labelling, i.e. the usual D(P) label annotated with accusative Case, can be substituted throughout without prejudice for the present argument. (13) Gjirokastër wo N wo ɛ I wo dua N(P) wo D wo tə I ðafɔsuɾit Interestingly from the point of view of the present discussion, there are other contexts where të introduces a participle – and it cannot be assimilated to the traditional article. Thus in a standard(-like) variety like Gjirokastër, non-finite purpose clauses and relatives have the preposition për followed by the particle të followed in turn by the participle. In (14a), the latter recognizably has the same participial inflection -ɾ as the nominalization in (11), though it does not have the -it nominal inflection. Clitics are inserted between të and the participle, exactly as in subjunctive and other finite sentences; in particular the examples in (14) introduce the middle/reflexive (M/R) u clitic (similar to Italian si). (14) Gjirokastër a. aɾðtʃ pəɾ t u laɾə. I.came for Prt M/R washed ‘I came to wash myself.’ b. əʃt ɲə Ôə pəɾ t u bənə. it.is a thing for Prt M/R made ‘It is a thing to be done.’

PRO, pro and NP-trace (raising) are interpretations 147

A further link between nominalized participles and subjunctive sentences is provided by complements of a small class of verbs, including ‘to finish’. As can be seen in (15a), in the Geg variety of Shkodër this is constructed with a nominalized participle of the type in (11). Interestingly, in the Arbëresh dialect of Civita a të + participle complement with a non nominalized participle is equally attested, as shown in (15b). The fact that the verb form in (15b) is not nominalized is evident from the lack of the -it suffix present in (15a) – which leaves only the ordinary participial inflection in -ɾ. (15) a. Shkodër maɾɔva t hɒ)ŋɾn-it. I.finished Art eaten-Infl ‘I finished (the) eating.’ b. Civita fiɾnɔva tə bənuɾ ət ʃuɾ»bɛs. I.finished Art done that thing ‘I finished doing that thing.’ In short, in a number of cases where të appears in front of participles, the divide between article të, briefly analyzed above, and subjunctive particle të appears to be quite fluid. This kind of observation establishes a certain degree of plausibility for the proposal we are about to make – namely that the phonological identity between the article të and the so-called subjunctive particle të reflects the existence of a common lexical entry. Consider the simple example in (1), repeated here in (16a). In the structure in (16b) we position the verb, as usual, in the I head of the sentence and the accusative clitic in an immediately superordinate position, notated N(P) in accordance with the labelling hypothesis introduced in the discussion surrounding (13). Suppose next we assign të to a D position, as also indicated in (16b). As it turns out, there is a natural interpretation for D(P) as an argument of the embedded sentence. In particular following Chomsky (1995), Manzini and Savoia (2005, 2007) reduce the EPP property to the D(efiniteness) property. Therefore assigning the structure in (16b) to the sentence in (16a) amounts to saying that the so-called subjunctive particle të is in reality a subject – specifically a subject clitic.

148 M. Rita Manzini (16) Vena di Maida a. u dua t ɛ bɐtʃ I want Prt it you.do(subj) b. ei D ei t N ei ɛ I bɐtʃ Intuitively, the D characterization of occurrences of të should be understood on a par with that assigned to both subject clitics and definite articles in Romance. Like the Romance subject clitic/definite article, të lexicalizes a nominal element that closes off the argumental domain of a sentence or a noun phrase. Therefore particle të is not directly connected to the modal properties of the sentence, but rather to its EPP argument – and to its interpretation, which directly concerns us here. Interestingly Roussou (2007), while maintaining with Roberts and Roussou (2003) that Greek na is inserted in a C position, also concludes that it associates with the D argument. The structural claim in (16) is consistent with the distribution of të, in particular with respect to the (modal) negation, that it can either follow or precede, and to other material in the left periphery of the sentence (whphrases, etc.) – as detailed by Manzini and Savoia (2007). To be more precise, Manzini and Savoia (2007) conclude that the të particle does not occur in the clitic sequence immediately above I, but rather lexicalizes one of the clitic positions that must be available above C (or higher) on the independent evidence of Romance languages. If the structure of the noun phrase is parallel to that of the sentence, this will mean that the article also appears in the C field. But this conclusion is plausible anyway, given the literature assimilating the determiner to the complementizer (Szabolcsi 1994). Therefore, structures like (12b) or (16) should be revised as in (17). (17) a. Shkodër ei D ei t (C) ei ɛ I mɔtɾ-ɛn

PRO, pro and NP-trace (raising) are interpretations 149

b. Vena di Maida ei D ei t (C) ei N ei ɛ I bɐtʃ As the C field of the sentence is associated with modal properties, so the C field of the noun phrase is associated with quantificational properties. The two can be unified on the natural assumption that modal properties are quantificational properties ranging over situations/possible worlds (Chierchia and McConnell-Ginet 1990). Under this analysis, the modal character that previous theories have commonly assumed for të is structurally captured not by the category it projects, but rather by the high domain it is inserted in. So far, our analysis structurally unifies what we may take to be the two end points of the range of constructions with të, namely the subjunctives in section 2.1 and the nominalizations in this section. Along the same lines we can analyze descriptively intermediate cases such as the (për) të construction, where të associates with a participle which does not bear nominal inflection, as in (14) or (15b) – hence is presumably sentential. In fact, the distributional evidence reviewed by Manzini and Savoia (2007) indicates that the position of të in participial sentences is the same as in finite sentences; not only pronominal clitics appear immediately before the participle and after të, as already noted in connection with (14), but të can either precede or follow the modal negation, and so on. We shall return to the tëparticiple construction at the beginning of section 4.1.

3.1. The interpretation of të The connection between the analysis of të just provided and the problems as to the interpretation of the EPP argument reviewed in section 2 should be obvious. Given the structure in (16) or (17b), where të itself represents a lexicalization of D, the original question about the interpretation of the EPP argument becomes a question about the interpretation of të. In sentences other than those introduced by të, whether matrix or embedded ones, Albanian null subjects have the same range of interpretations as null subjects in more familiar languages like Italian. Thus they allow for

150 M. Rita Manzini the classical pro readings (expletive and argumental) – to the exclusion in particular of the bound control readings. This is illustrated in (18) with the matrix verb ‘I say’, whose control and non-control readings are sharply distinguished in Albanian as in English. As shown in (18a–d) complements introduced by të induce the typical object control reading, close to that of verbs like ‘to order’ etc. and associated with infinitival complements in English. By contrast complements introduced by the se complementizer, as in (18a’–d’) are associated with the declarative reading typical of that complements in English and characterized by the absence of control properties. (18) Civita a. tə θaʃ tə viʃ mbjatu. to.you I.said Prt you.come quickly ‘I told you to come quickly.’ a’. mə θa:n sɛ vijin mə»nat. to.me they.said that they.came tomorrow ‘They told to me that they came the day after.’ Carfizzi b. tə θɔm t ɛ bɔ:tʃ. to.you I.say Prt it you.do ‘I am telling you to do it.’ b’. mə kɔn θɔ:n sɛ viɛn. to.me they.have said that he.comes ‘They told me that he comes.’ Portocannone c. tə θatʃa t ɛ bəjɛ. to.you I.said Prt it you.did ‘I told you to do it.’ c’. mə θa:n kɛ ti u zÔuovɛ. to.me they.said that you M/R woke.up ‘They told me that you woke up.’ Gjirokastër d. i θatʃ tə mə fliɾte. to.him I.said Prt me he.called ‘I told him to call me.’

PRO, pro and NP-trace (raising) are interpretations 151

d’. »θuai mə sɛ tʃ ha. say to.me that what s/he.eats ‘Tell what s/he eats.’ Accounting for the range of interpretations conventionally associated with pro depends on some preliminary choices about the nature of the EPP argument. According to the view endorsed by Chomsky’s (1995 ff.) minimalism, the EPP is satisfied in any case by a noun phrase projection on the sentential tree, hence by an empty category (pro) in the absence of independent lexicalization. An alternative view, put in minimalist terms by Pollock (1996), allows for the EPP to be satisfied by the verb inflection. Manzini and Savoia (2005, 2007), in advocating this second view, argue in particular, contra Chomsky (1995 ff.), that there is no reason to take the phi-features of the verb to be uninterpretable. A perfectly viable alternative theory treats the morphological constituent carrying the phi-features of the verb as interpretable, namely as a lexicalization of the EPP argument, hence of the internal to the verb itself. This conception yields structures of the type in (19) for the embedded verb in (17b). (19) Vena di Maida ei I D bɐ tʃ If the inflection in (19) has pronominal properties, we can assume that the pronominal pro interpretation simply arises when no independent (syntacticlevel) lexicalization of the EPP argument is present. Suppose on the other hand that a syntactic EPP argument is lexicalized, as for instance in (18c’) – partially repeated in (20a) – where the -ɛ 2nd person singular inflection of the past of ‘I wake up’ doubles the lexical subject ti ‘you’. Within the present framework, the relation between the morphological-level EPP argument (i.e. the verb inflection) and the syntactic level one (i.e. the lexical subject) can be an instance of chain-formation – i.e. sharing of one theta-role by two arguments, as schematically indicated in (20b). (20) Portocannone a. … ti u zÔuovɛ you M/R woke.up b. (ti, -ɛ )

152 M. Rita Manzini Needless to say, there cannot be a derivational process relating the two positions in the chain (20b), which are lexicalized independently of one another at the syntactic and morphological level respectively. This implies that chain is an interpretive relation defined at the interface. In this picture agreement, or to be more precise compatibility in referential (and Case) properties, is itself not a derivational process but a relation that must hold between the members of a chain for the latter to be properly defined. The view that chains are primitive LF interface constructs, rather than the result of a derivational history, is defended by Brody (1997 ff.), who also holds a non-derivational, symmetric view of agreement. Nevertheless, care should be taken not to impute to Brody (1997 ff.) objects like (20b) – since his chains are sequences of identical copies, exactly like Chomsky’s (1995 ff.), nor do they include morphological-level constituents. Let us then turn to the expletive pro reading. This is connected with the focussing of the postverbal subject, briefly discussed in section 2.2 for Italian and true of Albanian as well. We take it that focus is a predication relation (Manzini and Savoia 2005, 2007). Thus in the embedded sentence in (2), repeated in (21a) below, the focussed postverbal subject is predicated of the pronominal inflection, roughly as in the pseudo-cleft in (21b). Indeed Frascarelli (2007) reviews many languages where cleft-like structures are the ordinary way of expressing focus. In the case at hand, predication requires compatibility of referential properties, i.e. agreement, of the verb inflection and the postverbal subject. (21) Carfizzi a. … tə viɲ a»i … Prt comes he b. … the one who comes is him. When a null subject sentence is embedded under të, the pronominal/expletive interpretation of the EPP argument remains in general possible, but a bound interpretation (control and raising) also becomes available. As we argued on the basis of (18), the bound readings are not possible in null subject sentences not introduced by të. In the discussion that precedes, we suggested that të itself represents a lexicalization of the EPP argument, i.e. D, of the embedded sentence. This implies that të and the D inflection of the verb share the same theta-role (i.e. form a chain) – which in turn implies their compatibility in referential properties. Therefore the intrinsic referential context of të must be impoverished enough to be compatible with the refer-

PRO, pro and NP-trace (raising) are interpretations 153

ential features associated with any finite inflection. Correspondingly, we suggest that what të introduces is a variable. Control, as in (3) or in (18a–d), can be obtained simply by having the variable bound by a matrix argument. Therefore in an example like (18b), repeated below in (22a), the t variable and the -tʃ inflection share the same theta-role, forming a chain, as in (22b), while the matrix argument tə ‘you’ and the embedded variable are in a referential binding relation, namely (22c). Compatibility of referential properties (i.e. agreement) must hold throughout, hence ultimately of the embedded inflection and of the matrix controller as well. (22) Carfizzi a. tə θɔm t ɛ bɔ:tʃ to.you I.say Prt it you.do b. (t, -tʃ ) c. (tə , t) As Butler (2004) notices, Manzini and Roussou (2000) effectively treat theta-roles as variables; if so, control for Manzini and Roussou (2000) is essentially a variable binding operation. Butler (2004) also proposes a treatment of this sort, noticing that it recalls semantic treatments where there is no syntactically represented controlled argument (PRO in conventional terms) such as William’s (1980) or Chierchia’s (1984). For Sigurðsson (2007) it is ‘a plain fact, not a theory’ that ‘what is there is a thematic element… that is simultaneously a reference variable’. At the same time Roussou (2007) correctly points out that control in Balkan languages cannot be a relation involving theta-variables à la Manzini and Roussou (2000) since the EPP argument is independently lexicalized at least by the verb inflection. Similarly, Landau (2004) argues that the existence of control into inflected Balkan sentences provides a definitive argument against the predicational theory of control of Williams (1980) or Chierchia (1984) – as opposed to the propositional theory maintained by Chomsky (1981) and much related work. The present work does not fully belong to either of the traditions mentioned. On the one hand the variable bound in the control relation of a language like Albanian is syntactically introduced by the specialized të element, yielding a syntactic representation for the controlled element – thus taking Landau (2004) and Roussou (2007) into account. At the same time like Manzini and Roussou (2000), we see control as a relation not involving

154 M. Rita Manzini syntactic empty categories in the sense of Chomsky (1995) and Landau (2000) but simply variable interpretations at the LF interface. Further elaboration on the LF representation of the të variable is certainly possible – and perhaps necessary. If we assume that variables correspond to theta-roles, i.e. argument slots of predicates, it is natural to construe të as introducing not so much a variable as a lambda-operator; the latter binds the variable corresponding to the EPP argument slot and the embedded sentence is turned into a predicate-abstract assigned a value by a matrix argument. Suppose that following Adger and Ramchand (2005), Butler (2004) we assume that arguments are generally introduce in grammar to value lambda-abstractions over variables corresponding to argument slots. If so, within the present framework finite verb inflections are also introduced to value a lambda abstract. The (të, verb inflection) chains can then correspond to dependencies between two instances of the lambda operator, abstracting over the same argument slot. In control on the other hand, an argument such as tə ’you’ in (22) values two lambda abstracts at once, corresponding to the matrix and the embedded predicate’s EPP slots. At the same time keep in mind that this article is concerned essentially with morphosyntactic evidence, which hardly provides any insight on the exact format of the interpretation. Therefore we consider it sufficient to have provided an indication as to what a possible such format may look like – and we shall not pursue the matter further. The approach that we are taking to control suggests that raising, as in (4), is just another interpretation of the të variable. In fact, the licencing of the variable in raising is essentially the same proposed for control, namely binding of the variable (or valuation of the predicate construed by lambdaabstracting over it) by a superordinate argument. The difference is in the distribution of the theta-roles /argumental slots. In a control sentence like (22) the të variable and the matrix subject that binds it have two independent theta-roles. In the raising case they share a single theta-role, in other words a chain relation holds between them. For instance, in (4b), repeated below in (23a), the two lexicalizations of the EPP argument in the embedded sentence, i.e. the -i 3rd person singular inflection of srisi and the të (tə) variable share the same theta-role – hence form a chain, as in (23b). But the same is true also of the matrix EPP argument, represented by the -i 3rd person singular inflection of iʃi, and of the të (tə) variable, as in (23c).

PRO, pro and NP-trace (raising) are interpretations 155

(23) Portocannone a. iʃ i tə mə srisi he.was Prt me he.called b. (tə, -i) c. (-i, tə) That the object in (23c) is a not a control, but a chain dependency does not depend on any syntactic, i.e. computational/derivational, property of the structures involved; indeed it would be hard to distinguish (23) from, say, (3b) on such grounds. Rather the distinction depends on the thematic properties of the matrix verb. Thus the control verb ‘to have’ in (3) has an external theta-role which is satisfied by its EPP argument, while the verb ‘to be’ in (23) does not have any theta-role to assign to its EPP argument. Examples that are ambiguous between raising and control, as in Italian (8), are predicted by the present theory at what appears to be the minimum cost – i.e. that of assuming that verbs like ‘to threaten’ have an optional (external) thetarole. We shall return to this point in connection with Italian in section 4. We have already commented on the fact that objects such as (20b) are not chains for Chomsky (1995 ff.) – but at best agreement pairs. An object like (23c) is not a chain either for Chomsky (1995 ff.) – since the ‘trace’ element, i.e. të, is not a copy created by movement, but an independently inserted element. However under the present theory a chain is defined simply by the assignment of the same theta-role to two elements, say të and the matrix EPP argument. What is more, since each element is introduced in the chain by its satisfaction of a given theta-role, this same theta-role must be satisfied by the foot of the chain as well – yielding the generalization that the foot of the chain is theta-marked (Chomsky 1981, 1995). In this sense, (23c) can indeed be identified with a raising chain. It remains for us to consider how our proposals concerning control and raising are compatible with our initial discussion of null subjects – in other words how the ordinary null subject readings can be maintained once a finite sentence is embedded under të, as seen for instance in the embedded sentence in (1), repeated below as (24a). If the discussion that precedes is correct, a chain is formed in (24a) between the embedded -tʃ inflection and the të variable, as in (24b). Control examples like (22) or raising examples like (23) are characterized by an additional dependency between a matrix argument and the të variable. However, no such dependency is present in (24). Therefore the value of the EPP argument is simply fixed by the embedded inflection. In effect, the chain in (24 b) is like an expletive chain in

156 M. Rita Manzini which the referential value is attached to the lower, rather than the higher member of the chain. (24) Vena di Maida a. … t ɛ bɐtʃ Prt it you.do(subj) b. (t, -tʃ ) Similarly, in (21) we considered expletive environments with a postverbal focussed subject, concluding that the latter is predicated of the pronominal inflection. The same remains true once we take into account the të variable, and the fact that it forms a chain with the -tʃ inflection. In null subject sentences not involving raising or control, the EPP argument embedded under të can also receive a coreferential reading. For instance coreference rather than, say, control must be implicated in (25), where të introduces the hypothetical sentence, and the matrix and embedded EPP argument, though coreferential, do not c-command each other. In this and similar cases there is of course no assignment of a value to the të variable by a matrix argument, unlike what happens in control and raising. Rather the coreference relation is established between the matrix and embedded representing the EPP arguments represented here by the pronominal verb inflections. (25) Gjirokastër pɔ t a haʃ dɔ jɛtʃ mə miɾ. if Prt it you.eat fut. you.feel more good ‘If you eat it, you’ll feel better.’ Finally, a lexical subject can also appear in the left periphery of a të sentence, above të and below the complementizer, as in (26) (where hɛ is a phonological alternant of the sɛ complementizer). In this case, we can assume that the lexical subject, i.e. ti in (26), values the të variable. Note that quite independently of the discussion that precedes, we do not expect a lexical subject to be able appear between të and the verb – since preverbal lexical subjects in null subject languages are generally topics, inserted higher in the C field than the të variable. (26) Vena di Maida u dua hɛ ti t ɛ bɐtʃ. I want that you Prt it do ‘I want you to do it.’

PRO, pro and NP-trace (raising) are interpretations 157

This account of the null subject reading of të sentences implies that raising and control readings are not forced by any property of the syntax of të sentences; for, the latter turn out to be well-formed and interpretable in the absence of any higher binder – precisely as null subject sentences. Therefore the presence of a control or raising reading, very much like the choice between control and raising discussed in connection with (22)–(23) above, must ultimately depend on the properties of the selecting matrix verb. The raising reading is forced with verbs which do not assign a theta-role to their EPP argument, so that the latter must inherit a theta-role from another predicate. Some similar consideration must be at play in cases of obligatory control. We shall return to this question in discussing the obligatory vs. non-obligatory control distinction for Italian in section 4. It is also worth noting that one of the interpretations that is supposed to associate with PRO, namely the so-called arbitrary interpretation is not found in Albanian të sentences, and is in fact missing from the data in section 2. But this is only to be expected, independently of whether the të variable is compatible with the arbitrary interpretation per se, i.e. with binding by a generic operator or the like. For, the embedded finite inflection, interpreted as a definite pronoun and forming a chain with the të variable, effectively blocks the generic reading. Again we shall return to this reading in connection with Italian in section 4. In short, if what precedes is on the right track, the various interpretations of the EPP argument of të sentences in Albanian can all be derived under the assumption that they represent the possible readings of a variable introduced by të. This eliminates all problems connected with the idea that the different interpretations depend (in part) on the insertion in the EPP position of empty categories whose context is lexically or computationally defined. Recall now that at the beginning of this section the plausibility of a nominal treatment for particle të was established through a comparison with the occurrences of të as an article. On the basis of the structural unification achieved in section 3, an interpretive unification of the two also appears to be within our grasp. In particular, the adjectival article in (10a), repeated below in (27a), seems to be a perfect match to raising të in an example like (23) – even the matrix verb is the same. Therefore we can take it that të in (27) introduces a variable. This on the one hand satisfies the same thetarole as the inflection – in this case the noun’s, i.e. -ɛ, forming a chain with it, as in (27b). On the other hand the value of the variable introduced by të is fixed by the matrix EPP argument, represented in (27) by the verb inflection -n, with which it shares the same theta-role, i.e. it forms a chain, as in (27c).

158 M. Rita Manzini (27) Shkodër a. jan t kutʃ-ɛ they.are Art red-Infl b. (t, -ɛ ) c. (-n, t) An asymmetry between particle të and article të arises in that, as we mentioned in introducing the data, the article depends on the Case, number, gender, and definiteness properties of the adjective. Thus for the masculine and feminine singular (nominative, indefinite) the article will be i and ɛ respectively, as illustrated in (28), rather than të. Accounting for the paradigm of adjectival (and nominal) articles (and inflections) in Albanian is beyond the scope of the present article. However, the general form of the solution to the restrictions observed on të in adjectival (and nominal) contexts can be briefly indicated. What we will have to assume is that the pure variable instantiation of the adjectival article, namely të, is compatible only with certain types of reference (for instance the plural); other references require instantiations by definite pronoun-like elements, such as i and ɛ (roughly corresponding to gender, i.e. nominal class, specifications). (28) Shkodër

ɐʃt i kutʃ / ɛ kutʃ-ɛ it.is Art red / Art red-Infl ‘It is red (m./f.)’ Another major instance of article të is the one that introduces noun phrases such as (10b), repeated below in (29a). Here again, as in (27), të and the embedded inflection, -ɛn in (29), satisfy the same theta-role of the predicate ‘sister’, hence form a chain in present terms, as in (29b). At the same time (t, -ɛn) also satisfies the internal argument of the predicate ‘to see’ – which is what it means for t mɔtɾ-ɛn to be the internal argument of pɒ. Therefore (29) is much closer to a case of control, as described above in terms of the same argument valuing two different theta-roles. In particular, because the argument (t, -ɛn) is lexicalized within the constituent formed by the more embedded predicate ‘sister’, the configuration is not dissimilar from a case of backward control (Polinsky and Potsdam 2006). Whether this is correct or not, të is compatible with only a subset of referential, definiteness and Case properties. For instance, taking Case, the nominative counterpart of the noun phrase in (29) would be e motr-a (Camaj 1995: 63). In present terms this means that the pure variable instantiation of the article, i.e. të, is

PRO, pro and NP-trace (raising) are interpretations 159

compatible with binding of the internal theta-role of the verb (accusative Case) but not with binding of the EPP argument (nominative) – which requires instead lexicalization of nominal class (gender) properties by ɛ. (29) Shkodër a. pɒ t mɔtɾ-ɛn I.saw Art sister-Infl b. (t, -ɛn) If the descriptions we provided in section 3 are on the right track, furthermore, nominalized participles have essentially the properties of noun phrases, while non nominalized participles have sentential structure. In conclusion, it seems to us that no major obstacles stand in the way of a unification of article and particle të. Even if this proved not feasible, note that the nominal treatment of particle të, which is what we are directly interested in here, does not require its unification with article të – though vice versa the unification of the two elements does entail the nominal treatment of të. Therefore the present analysis of të sentences is strengthened by the unification with article të, but does not depend on it.

4. Italian -re Having provided an analysis of Albanian të sentences we can now turn to the infinitival sentences of Romance languages, as described in section 2.2. By analogy with Albanian, we shall work on the hypothesis that the infinitival inflection, -re in Italian, is a morphological level counterpart of a syntactic element like të. Roughly speaking, therefore, -re has the property of introducing a variable EPP argument. Since one generally speaks of the infinitive as of a mood and infinitival inflections have not been connected before to the nature of the EPP argument (with the exception of Manzini and Savoia 2005), we are reshaping conceptions about the nature of Romance infinitivals along the same lines as we did for Balkan so-called subjunctives. The idea that the Romance infinitival inflection is nominal in nature reflects traditional ideas about the infinitive as the nominal form of the verb. Indeed in Romance, the infinitive is the form of the verb appearing in nominalizations, much as the participle is in Albanian (11) or the gerund in English. In Italian (30) for instance the matrix verb can equally well embed an infinitive without determiners, i.e. a sentence, or the same infinitive preceded by determiners, i.e. a nominalization.

160 M. Rita Manzini (30) Detesto (questo continuo) criticare il governo. I.hate this continuous to.criticize the government ‘I hate (this continuous) criticizing the government.’ Recall now that for Albanian we have assumed that the finite verb inflection, with definite pronominal reference, occupies the D position within the verb, as in (19), where it provides a lexicalization for the EPP argument of the sentence. If we pursue the idea that Italian -re also lexicalizes the EPP argument, we will want to assign to it a structure where it also fills the D position of the verbal constituent. This yields a structure like (31b) for the andar(e) form in (5), repeated below in (31a). At the same time, it is evident that -re will not have the interpretive content of a finite inflection. Rather, if our hypothesis is correct, it will have the same content as të, in other words that of a variable. (31) a. Tentai di anda-r-ci io I.tried to go-Infl-there I b. wo I wo anda D r(e) Note that in the example in (31) the infinitival verb precedes the object clitic ci ‘there’ – though the finite verb would follow it. This distribution can best be derived if the infinitive inserts in a C position (Kayne 1991), while object clitics remain in their ordinary position in the inflectional domain of the sentence, as schematized in (32); ci in particular is assigned to a Loc(ative) position in accordance with the labelling of Manzini and Savoia (2005, 2007). (32)

wo C wo andar Loc ci

From an interpretive point of view, the sentence in (31) is an example of control. Given the hypothesis that -re introduces the EPP argument as a variable, control corresponds to the binding of the variable by a matrix argument, as already discussed for Albanian të. Nominative Case, as seen on

PRO, pro and NP-trace (raising) are interpretations 161

the postverbal subject in (31) is a problem if one assumes that control requires PRO and PRO in turn requires null Case. But under the present theory the control interpretation implies neither PRO nor null Case. Therefore we are free to conclude that infinitival sentences support nominative Case (in Romance) exactly as finite sentences do. In (31), in particular, the subject appears in focus, inviting an account along the lines of Albanian (21). This means that in (31) the EPP variable introduced by -re, beside being bound by a matrix antecedent, also enters a predication relation with the postverbal subject; the latter is restricted to a pronominal element simply by the presence of a binder for the EPP variable – which binds the postverbal subject as well. In the raising reading, the EPP variable introduced by -re not only has an antecedent in the matrix clause, as in control, but it also shares the same theta-role with it – in other words the chain property holds of them. As we mentioned in the previous section, there is nothing in the syntax that indicates which of the two interpretations (raising or control) holds; rather their alternation is determined entirely by the distribution of the theta-roles in the matrix predicate. This theory is ideally suited to predict cases like (8a), repeated here as (33), where the lack of apparent differences between the raising and control reading translates into a single syntactic structure. The two readings are a matter of ambiguity at the LF interface, driven by the ambiguity of the matrix predicate between a control argumental frame (with a theta-role for the EPP argument) and a raising one (without a thetarole). (33) Gianni minaccia di non imparare John threatens to not learn A further interpretation that can be straightforwardly obtained in the present framework is the so-called arbitrary reading, that arises for instance if we remove the overtly lexicalized subject in (6), repeated here in (34). Quite simply the so-called arbitrary reading corresponds to quantificational binding of the EPP variable, i.e. -re, say by a generic operator. (34) Tutto è successo per aver scioperato everything has happened for to.have gone.on.strike Before we sharpen our approach to arbitrary control, it is useful however to consider the last set of interpretations possible for -re sentences, roughly what we may call pro or null subject interpretations. Thus in the absence of

162 M. Rita Manzini a controlling antecedent, the postverbal nominative subject can take the form of a full noun phrase – yielding for instance (6a), repeated below as (35a), which can be described as an instance of expletive pro. The other reading available for the EPP variable in the same contexts is the argumental, i.e. pronominal, pro reading, salient in the examples in (7) – or below in (35b). (35) a. Tutto all b. Tutto all

è successo per aver scioperato gli statali is happened for to.have gone.on.strike the public.employees è successo per non essermi svegliata in tempo is happened for not to.be.myself woken on time

Examples like (35a) are generally not discussed in the literature and are presumably implied to be ungrammatical because of the unavailability of nominative Case in infinitivals. However we have independently concluded for (31) that nominative is available in infinitival sentences, so that the occurrence of the full postverbal noun phrase is as predicted. On the other hand, examples like (35b) are standardly accepted – as are their English counterparts. In fact contexts like (35) are discussed by the literature under the general heading of non-obligatory control. For Hornstein (1999) the key to analyzing the range of interpretation in (34)–(35) is in the pro (rather than PRO or trace) status of the empty category EPP argument of the infinitive – in other words in a structural difference between (34)–(35) and, say, (31). But we could repeat for Italian an argument of the same general form as the one we ran for Albanian in the previous section – namely that there are no independently demonstrable structural differences between control (/raising) sentences and non-(obligatory) control ones that would allow us to derive the different status of their empty category EPP argument and hence their different interpretations. Rather it is the different interpretations of the EPP argument that justify in a backward fashion the different structures or feature assignments imputed to the sentences containing them. To be more precise, a potential structural correlate of the obligatory vs. non-obligatory control distinction is that, as implied by Manzini (1983), classical non-obligatory control, as in (34)–(35), has the property of occurring inside islands. For instance the infinitival sentence is an adjunct in (34)–(35). However, islandhood is not a necessary condition for non-obligatory control readings. Thus Rizzi (1982) discusses at least one set of matrix verbs – roughly believe-type verbs – with which control is not obligatory into an infinitival complement. For Rizzi (1982) this coincides with the

PRO, pro and NP-trace (raising) are interpretations 163

presence of V-to-C movement and hence of a lexical subject in the position associated with V2 in Germanic languages (as seen for instance in English questions). We on the other hand agree with Mensching (2000) that this is not necessary. In particular, in (36a) the lexical subject is in the ordinary focussed position at the right periphery of the sentence. To the extent that (36 a) is acceptable, so is the pronominal reading of the embedded subject in (36b) and the control, or to be more precise coreferential reading in (36 c). (36) a. Ritenevo aver partecipato alla riunione tutti i capi I.believed to.have taken.part to.the meeting all the heads dipartimento. department ‘I believed all of the department heads to have taken part to the meeting.’ b. Ritenevo doverci comportare secondo coscienza. I.believed to.need.ourselves to.behave according conscience ‘I believed we should behave according to our conscience.’ c. Ritenevo dovermi comportare secondo coscienza. I.believed to.need.myself to.behave according conscience ‘I believed that I should behave according to my conscience.’ Incidentally, sentences like (36a) provide not only a counterexample to the connection between nominative Case in infinitivals and V2 of the Germanic type proposed by Rizzi (1992), but also indirectly argue against the approach to nominative Case in Balkan subjunctive particle sentences proposed by Terzi (1997) – as we already mentioned in section 1. We have no specific argument to offer against Terzi’s (1997) analysis for Greek – which is not one of the languages studied here. However an analysis based on Germanic-type V2 for Greek will not at this point have the independent support of Italian. If given the evidence in (36), the presence of islands is not a necessary for non-obligatory control, still it could be the case that their absence is necessary for obligatory control. Thus Manzini (1983) proposes that the domain for the binding of PRO is computed not on the basis of PRO itself, but rather on the basis of the whole control sentence. The effect of this is to make the control sentence transparent for PRO in those cases where the sentence is governed (i.e. a complement of the verb) – forcing obligatory control. On the contrary, in the cases where the sentence is not governed (i.e. a subject or an adjunct), no locality domain is defined in terms of

164 M. Rita Manzini Chomsky’s (1981) binding theory, yielding non-obligatory control. In this way, while the distribution of PRO is not sensitive to binding domains (because of the so-called PRO theorem), the control relation becomes sensitive to them. Similarly, theories that treat obligatory control as movement (Hornstein 1999) predict that it is sensitive to islands, as all instances of movement are. To be more precise, control, being an instance of raising, is subject to a sharper constraint, since it cannot make use of escape hatches (Chomsky 1973) or edge positions (Chomsky 2001 ff.) in the C field; therefore it can only apply in contexts where the embedded sentential boundary does not count as a subjacency node (Chomsky 1973) or phase (Chomsky 2001 ff.). But the line of thought correlating obligatory control with transparent environments does not appear to be empirically correct. In particular, there are adjunct sentences that are effectively obligatory control environments. Thus the control reading of the purpose clause in (37a) is obligatory; neither definite pronominal reference, as in (37b) nor arbitrary (generic) reference as in (37c) are possible. (37) a. Mi sono alzato per andarmene. myself I.am stood.up for to.go.myself ‘I stood up to go.’ b. *Mi sono alzato per andartene. myself I.am stood.up for to.go.yourself ‘I stood up for you to go.’ c. *Mi sono alzato per andarsene tutti. myself I.am stood.up for to.go.themselves all ‘I stood up for everybody to go.’ This leaves only one current proposal open, namely the one advocated by Culicover and Jackendoff (2006) – that obligatory control properties are determined by what they call Conceptual Structure. As anticipated in section 1, we do not agree with Culicover and Jackendoff (2006) on the place that Conceptual Structure has in the overall architecture of the grammar. Even accepting that there is a conceptual structure, where computation takes place over abstract predicates, in the present approach there is no need either for the intermediate Grammatical Functions level nor for the linking rules mapping Conceptual Structure to Grammatical Functions and the latter to Syntactic Structure. It seems to us that the architecture implied by these various levels not only leads to a richer (as opposed to a simpler) grammar, but also ends up reproducing the empty categories of the tradi-

PRO, pro and NP-trace (raising) are interpretations 165

tional generative approach – and in fact its structural ambiguity approach to different interpretations. Thus for Culicover and Jackendoff (2006) control is defined by the fact that a Grammatical Function, namely that corresponding to the embedded EPP argument, remains unlinked to any Syntactic Structure constituent; but one could argue that at the Grammatical Function level the unlinked GF ends up providing a counterpart to the empty category (PRO) that is eliminated from the SS level. In turn, for Culicover and Jackendoff (2006) raising is the result of binding of a Grammatical Function by another in GF Structure. In other words, the control and raising interpretations correspond to two different computations – effectively upholding Interface Uniformity. By contrast, under present assumptions the variable bound in control and raising can be read directly off the LF interface structure, from elements such Albanian të or Italian -re. Furthermore, the control/binding divide is interpretive, essentially an ambiguity in LF/ conceptual structure. What is crucial for us in the proposal of Culicover and Jackendoff (2006) is that the distinction between obligatory and non-obligatory control is taken charge of by the interpretive properties of the predicates involved – some of which in particular determine obligatory control. Because this article is not devoted to control per se – but rather to structures involving (in present terms) EPP variables – we will not dwell at any length on a number of problems in (obligatory) control that have not been mentioned so far. As Culicover and Jackendoff (2006) note, their theory derives cases which have been taken to involve control by implicit arguments on a par with cases where overt controllers are involved. This is because control is a Conceptual Structure relation which does not necessarily hinge on a particular Syntactic Structure. Relevant examples are control of the subject of a purpose clause by the implicit agent of passives as in (38a) or by an implicit argument of nouns as in (38b). We are sympathetic to the proposal of Williams (1985), Brody and Manzini (1989), Butler (2004) that implicit arguments are nothing but free thematic roles/variables; in terms of the proposal of Culicover and Jackendoff (2006) this is all that is needed for the linking (at Conceptual – i.e. predicate-argument – Structure) to apply. (38) a. Il battello è stato affondato per riscuotere l’assicurazione. The boat is been sunk to get the insurance ‘The boat has been sunk to get the insurance.’ b. La promessa di andarsene è irrilevante. The promise to go is irrelevant ‘The promise to go is irrelevant.’

166 M. Rita Manzini Other classical problems in (obligatory) control may not have much real substance. At least from the standpoint of Italian, there is a single obligatory control phenomenon – of which, say, split antecedents are just a manifestation. Thus any control verb with more than one argument does allow for joint control, as in (39). The judgements in the English literature appear to be needlessly restrictive – in particular those of Hornstein (1999 ff.) for whom the existence of split antecedents represents one of the main empirical problems. (39) a. Ho persuaso Mario a incontrarci. I.have persuaded Mario to meet.ourselves ‘I persuaded Mario to meet.’ b. Ho promesso a Mario di incontrarci. I.have promised to Mario to meet.ourselves ‘I promised Mario to meet.’ If the discussion of obligatory control that precedes is on the right track, we still have the non-obligatory control issue to settle. The theory presented here commits us to the conclusion that ‘infinitival’ -re, exactly like ‘subjunctive’ të of Albanian introduces a variable EPP argument. If so, we are obviously committed to the conclusion that the readings of non-obligatory control environments (whether arbitrary, pronominal or expletive) are possible readings of this variable. Let us begin with the arbitrary reading, which is presumably quantificational, as hinted in the discussion of (34). In (34) we have chosen an example with perfect temporal specification on the matrix and embedded verb. If we turn to a present specification, as in (40), it seems evident that the generic reading reduces to a universal one, as originally proposed by Lebeaux (1984). (40) Bisogna amare il prossimo. It.is.necessary to.love the next.one ‘It is necessary to love one’s neighbor’ i.e. ‘Everybody must love his neighbor.’ At the same time, the strict parallelism between arbitrary control and the overt generic of Italian, namely (impersonal) si (cf. already Chomsky 1981 on their similar agreement properties), suggests that many of the considerations developed by Chierchia (1995) for impersonal si can be extended to arbitrary control. In particular, Chierchia (1995) distinguishes the generic

PRO, pro and NP-trace (raising) are interpretations 167

reading proper from what he calls ‘episodic’ reading. Applying this distinction to arbitrary control we may contrast (40) with an example like (41). In (41) the contextual information about the immediate nature of the required action and the specific recipient of it (Gianni) means that the most plausible interpretation is one in which just a restricted set of individuals is referred to by the EPP argument, possibly just the speaker and the listener(s). In fact, while the reading is that of a universal in (40), it is if anything that of an existential in (41). (41) Bisogna dirlo subito a Gianni. It.is.necessary to.say.it immediately to John ‘It is necessary to tell John immediately’ i.e. ‘Somebody must tell John immediately.’ We propose that the definite pronoun reading of the EPP argument of infinitivals results simply from stronger contextual (pragmatic) restrictions along the scale also including (40)–(41). This approach predicts that nonobligatory control contexts are systematically ambiguous between the generic reading, the episodic reading and the definite pronoun reading – and can sometimes be disambiguated by non-linguistic context. This is correct, as (42) admits of interpretations that range from universal quantification to the reference to a specific individual with contextual salience. (42) Sposarsi è un errore. To.marry.self is a mistake ‘To marry is a mistake.’ Finally, it is worth noting that existential binding of the variable introduced by -re also allows for the expletive reading of the EPP argument of Italian infinitivals, as seen in (35a) or (36a), with a focussed postverbal subject. Comparable sentences considered for Albanian in the previous section had a finite inflection – so that the postverbal subject agreed with it. As is wellknown however, agreement in expletive-associate constructions is parametrized, so that in many languages even a finite verb presents an invariable form independent of the referential properties of the postverbal subject (or of some of them). Though Chomsky’s (1995 ff.) discussion ties this parameter to the different morphologies of expletives, it can also be seen in null subject languages. Thus Manzini and Savoia (2005, 2007) provide examples of null subject Italian dialects where the verb is in the invariable 3rd person singular with a plural postverbal subject, as in (43b).

168 M. Rita Manzini (43) Urbino a. ki burdɛi dɔrm-ne de la those children sleep-3PL there ‘Those children sleep there.’ b. de la dɔrm-e ki burdɛi there sleeps those children ‘Those children sleep there.’ Manzini and Savoia (2005, 2007) argue that in examples like (43b) the invariable 3rd person singular verb inflection introduces the EPP argument as a variable. By hypothesis the same is true of the infinitival -re inflection studied here. If we assume that the postverbal subject has a predication relation to the subject variable, given an example like (36a), repeated here as (44a), we obtain an LF representation of the type in (44b) – i.e. a good match to the observed focussed interpretation. (44) a. … aver partecipato alla riunione tutti i capi dipartimento to.have taken.part in.the meeting all the heads department b. there is some x such that x has taken part in the meeting, x = all department heads

4.1. Participles and bare verb stems in Albanian The final set of data to be introduced here, from Geg varieties of Albanian, shows that the same readings that are found with subjunctives introduced by the të particle in standard Albanian or with the -re infinitive in Italian can be found with a different form of the verb yet, i.e. what could be described as a participle, yielding the so-called paskajore (‘infinitive’). The basic empirical argument in favour of the present proposals, namely that these various readings cannot be tied up with properties such as null Case connected with particular verb morphologies, is therefore further strengthened. In order to introduce the Geg data, it is useful to go back first to the (non-nominalized) participles of Tosk varieties of Albanian, which are preceded by the të particle both in adjuncts, as in (14a) repeated as (45a) below, and in complements to verbs like ‘to finish’, as in (15b) repeated as (45b) below. In comparable contexts, languages like Italian or English have an infinitive, while other Balkan-type languages – for instance the Romance Southern Italian dialects exemplified by Manzini and Savoia (2005) – have a sentence introduced by the subjunctive particle.

PRO, pro and NP-trace (raising) are interpretations 169

(45) a. Gjirokastër aɾðtʃ pəɾ t u laɾə I.came for Prt M/R washed b. Civita fiɾnɔva tə bənuɾ ət ʃuɾ»bɛs I.finished Prt done that thing The embedded verbs in (45) are clearly inflected with the -ɾ participle suffix. The same -ɾ form is embedded under the auxiliary, yielding the perfect, both in the active with kam ‘I have’, as in (46a), and in the middle-reflexive with jam ‘I am’, as in (46b). In adjectival contexts the participle is further inflected for number and nominal class (gender) features, according to the pattern illustrated in (10a) for the adjective, and here in (46c). (46) Gjirokastër a. ɛ kam zÔuaɾ him-her I.have woken.up ‘I have woken him/her up.’ b. jan vɛʃuɾ they.are dressed.up ‘They have dressed up.’ c. jan tə vɛʃuɾ / »vɛʃuɾa they.are Art dressed.up / dressed.up-Infl ‘They are dressed up (m./f.).’ As mentioned in section 3, the occurrences of të in (45) are closer to the particle than to the article ones, since their complement has sentential properties, as seen from the possibility of clitics, modal negation, etc. For all of this, the form of the verb that të embeds is a participle in (45) – providing us with perhaps the clearest argument that ‘particle’ të does not have anything to do with ‘subjunctive’. Furthermore, both complements of the ‘to finish’ class of verbs and purpose clauses do alternate between të – participle sentences, and të – subjunctive sentences, as in (47). Note that even in the same dialect, Civita, the complement to ‘to finish’ can be a participial sentence, as in (45b) or a subjunctive, as in (47b). Apparently, what is at stake is simply selection of the participial or of the finite verb/sentence by the matrix predicates ‘to finish’, ‘for’ (‘in order to’), etc. Remember also that alternation of të – participle sentences with nominalized participles is illustrated in (15).

170 M. Rita Manzini (47) a. Vena di Maida ɛrða pə t ɛ ʃɔrə I.came for Prt him I.see ‘I came to see him.’ b. Civita fiɾnɔva t ɛ diɛvasja I.finished Prt it I.read ‘I finished reading it.’ Crucially, given the present theory, nothing needs to be added to the discussion in section 3 in order to derive the (obligatory) control readings in (45). The lack of pronominal inflections on the verb simply means that the EPP argument is directly introduced as a variable by të. This variable is then bound by an antecedent, while the obligatoriness of the binding relation and its interpretation as a referential dependency (rather than as chain) depend essentially on the properties of the matrix predicate. Turning next to Geg varieties, these differ from Tosk in that the verb form embedded by the auxiliary, descriptively the participle, corresponds to the bare verb stem, as exemplified in the active (48a) with a stem ending in vowel, and in the middle-reflexive (48b) with a stem ending in consonant – the latter presenting a lengthening of the stressed nucleus instead of the thematic vowel. Geg dialects do have a participle suffix, which however turns up only in cases in which the participle also bears a nominal/adjectival inflection. Thus in the adjectival context in (48c), the participial suffix is -un (for the stem ending in consonant), followed by -a in the feminine. The same -n morphology of the participle can be recognized in the nominalization in (15a). (48) Shkodër a. ɛ kan mlu him they.have covered ‘They have covered him.’ b. ɐʃt ve:ʃ he.is dressed ‘He has dressed / He has been dressed.’ c. jena t veʃ-un / t »veʃ-un-a we.are Art dress-ed / Art dress-ed-Infl ‘We are dressed (m./f.).’

PRO, pro and NP-trace (raising) are interpretations 171

What interests us here is that in Geg dialects the bare verb stem in (48a–b) also appears preceded by the preposition me (literally ‘with’) in contexts where we would find të followed by the finite verb in Tosk dialects – or indeed the infinitive in Italian and English. Relevant examples are provided in (49); in particular (49a) illustrates the obligatory control reading of the EPP argument with matrix ‘to try’ – while (49b) illustrates the possibility for the control/coreference reading to combine with a nominative focused (postverbal) pronoun. In (49c) a full lexical subject is instead positioned between the me particle and the tʃ i complementizer – incidentally showing that me sentences can be embedded under complementizers exactly like të sentences are. Since we have already assumed that the full lexical subject that appears between the complementizer and the të particle in (25) is a topic in the left periphery of the sentence, it is natural to extend this analysis to (49c). (49) Shkodër a. munɔhɛn mɛ ɛ bɒ̃ they.try to it do ‘They try to do it.’ b. doin mɛ ɒ:ɾð a»ta they.want to come they ‘They want to come.’ c. du tʃi vɬa-i jat mɛ ɛ bɒ̃ I.want that brother-the yours to it do ‘I want your brother to do it.’ As noticed at the outset, the paskajore of Geg dialects in (49) provides yet another illustration of the problem for current generative theories that originally motivated this article – namely that exactly the same morphosyntactic structure is compatible with what in classical generative theories would be incompatible constructs, in this case the bound reading, associated with PRO, and the presence of a full nominative subject. Conversely the parallelism with Tosk të sentences or with Italian infinitivals suggest a similar treatment for the paskajore. In particular we are led to propose that the EPP argument of bare verb stems is read as a variable, predicting in this way both the antecedent-bound reading (obligatory control/raising), as in (49a), and the null subject/ pronominal reading, as in (49c). The questions that remain open concern precisely the morphosyntactic properties whereby the Geg paskajore differs from the infinitive/subjunctive sentences considered so far. Consider first the role of me in sentences like (49). The characterization of me cannot abstract from the fact that it also

172 M. Rita Manzini appears with noun phrase complements (rather than sentential ones), with the meaning of ‘with’, as illustrated in (50). The alternation between preposition and particle occurrences of the same lexical item are independently known from languages like English (up etc.) – and it appears to depend simply on whether the relevant element takes a noun phrase complement or not. (50) Shkodër

ɛ kam bɒ̃ mɛ kry:p it I.have made with salt ‘I made it with salt.’ The connection of prepositions to the nominal system (in particular to Case) suggests that their particle occurrences can also be treated as nominal in nature. In particular, Manzini and Savoia (2007) assume that me fills the same D (subject clitic-like) position as të, and like të it introduces the EPP argument as a variable. A different analysis however might more consistent with the observation that preposition me basically introduces an instrumental – and that this kind of specification is associated in the Manzini and Savoia’s (2005, 2007) clitic/nominal hierarchy with the Loc(ative). Thus me could be analyzed as a Loc, rather than a D clitic. For a connection between Loc and D one needs to look no further than English there, which is argued to be the real locative deictic even in expletive contexts both by Manzini and Savoia (2005, 2007) and by Kayne (2006), though within quite different overall analyses. The other obvious question raised by the paskajore in (49) concerns the morphological-level analysis of what we have described as a bare verb stem. Manzini and Savoia (2007) treat both the thematic vowel found in vocalic stems and the lengthening of the stressed nucleus found in consonantal stems as (pro)nominal inflections relating to an argumental position of the verb. As before, their exact nature depends on the other contexts where they occur, including the participial contexts in (48). The structure provided by Manzini and Savoia (2007) has the thematic vowel and equivalently the prosodic lengthening of the stress vowel as N inflections, i.e. as a lexicalization of the internal argument of the verb, which might be expected for participles. On the other hand, there is no reason why the same element couldn’t also be construed as a lexicalization of the D argument in other contexts. Under this latter hypothesis, in sentences such (49) it could be the inflection itself of the verb that introduces the EPP argument as a variable, in the manner of Italian -re. This would then leave the me particle free to play the role of ‘expletive’ Loc element.

PRO, pro and NP-trace (raising) are interpretations 173

Whatever the correct analysis of the morphosyntax of Geg Albanian will turn out to be, what is crucial here is that that each of the three language types that we have reviewed has its own way of introducing EPP variable – namely the -re inflection in Italian, the të (P)article in Tosk Albanian and a different type of particle or of inflection (according to the exact analysis that we want to give of them) in Geg Albanian. We could certainly impose on this different lexical material an abstract structural grille, treating inflections an particles as accidental elements in a derivation played out by abstract elements such as the empty categories (PRO, pro, trace) of classical generative theory – thereby complying with what Culicover and Jackendoff (2006) would call Interface Uniformity. But if the discussion that precedes is correct, Interface Uniformity does not yield any advantage at the LF interface – where the relevant interpretations are adequately treated in terms of variable dependencies. Vice versa our contention throughout has been that Interface Uniformity leads to a loss of generalizations concerning the lexicon (including the morphology) – which in turn undermines any attempt at a proper understanding of linguistic variation given the hypothesis that the lexicon is the seat of parametrization. About the latter, remember that according to the Principle and Parameters framework of Chomsky (1981, 1995), the theory of parameters is not the periphery of an independently existing Universal Grammar core; on the contrary the parametric space, however defined, is an integral part of Universal Grammar. In this sense, an Interface Uniformity approach, far from strengthening the Universal Grammar program, may ultimately weaken it in some of its central components.

5. Further theoretical prospects The discussion that precedes leaves a major theoretical question open. Since raising structures are unified with control and null subjects ones, differing from them only in interpretation, one may wonder what happens to the unification of raising with passives and unaccusatives in terms of movement. Now, at several points in the present discussion we mentioned that thetaroles, i.e. argument slots of the verb, are probably best treated as variables, valued by arguments in the syntax, eventually through an intermediate step of lambda abstraction of the type suggested by Adger and Ramchand (2005). If so, chains can correspond to dependencies between two instances of the lambda operator, with the effect that all of them abstract over the same argument slot. This is compatible with our discussion of raising and it

174 M. Rita Manzini is also what Butler (2004) suggests for passive. Therefore from a purely formal point of view, all that is needed to unify raising with passives and unaccusatives is the (obvious) assumption that the latter introduce the internal argument as a variable (/lambda abstractor). The problem however is empirical. It is a defining property of the present treatment of raising that it is unified with control. Thus in the case of EPP arguments the same formal set of operations (variable introduction, lambda abstraction, valuation by an argument) can yield either the raising or the control reading. On the contrary, to put the generalization in the GB terms in which it was first formulated, there appears to be no object PRO; in other words no verbs whose empty category /variable internal argument has the control relation to the EPP argument – i.e. it is interpreted as a reflexive. Standard generative theory provides answers to this problem in the form of structural constraints (government and its successors, say null Case) – but these are precisely the notions that we have argued against in what precedes, on empirical as well as on simplicity grounds. Consider lexical unaccusatives. A possible alternative is to deny that there is chain formation/movement. Standard generative theory is built on the assumption that there is a canonical linking of theta-roles to syntactic positions, as formalized relatively late by Baker’s (1988) UTAH; therefore non-canonical linkings involve chain formation – or movement, on the assumption that chains are derived from movement. But abandoning the idea of a canonical linking, nothing prevents us from assigning the internal theta-role directly to the EPP argument, without the mediation of chains/ empty categories /variables. This translates into the binding of the argumental variable (or of a lambda abstract on it) directly by the EPP argument – without the further variable (or lambda abstract) that Butler (2004) associates with T. In other words there is no movement and no chain, and ultimately no unification with raising either. At the same time many languages, including those considered here (Albanian and Italian), do in fact have morphological entries ambiguous between the reflexive and the passive. In particular both Albanian and Italian (more generally Romance) can lexicalize both passives and reflexives through a clitic, si in Italian and u in Albanian. Manzini and Savoia (2005, 2007) argue at great length that si and u are bona fide object clitics, and that they introduce a variable, exactly as we argued here for the ‘subjunctive’ particle or the ‘infinitival’ inflection. Furthermore Manzini and Savoia (2005, 2007) argue that the passive/unaccusative and reflexive readings all correspond to the binding of the si or u variable by the EPP argument; in fact si also allows for a quantified (‘arbitrary’) reading, the so-called imper-

PRO, pro and NP-trace (raising) are interpretations 175

sonal reading. Thus si /u constructions indeed parallel të/ -re constructions studied here, with which they can after all be unified, in the formal terms suggested above. The same unification cannot be achieved in terms of movement. For what it is worth, Manzini and Savoia (2005, 2007) run an extensive argument to the effect that approaches to Italian si that assume syntactic movement (Marantz 1984; Reinhart 1997; and many others) cannot provide a unified lexical entry for si/u – nor can they account for its syntactic distribution as an object clitic. We will not try to review the several other ways of forming so-called passives (and reflexives); even only keeping to Albanian and Italian, these include auxiliary – participle forms (as in English) and specialized inflections on the verb (in Albanian only). In general, the aim of this article has been to show that – contra what Culicover and Jackendoff (2006) would call Interface Uniformity – different structures/lexicalizations can support the same meaning (as in the case of Albanian subjunctives and Italian infinitives) – and the same structure/ lexicalization can support different meanings (as in the case of control, raising and null subject interpretations in the same Albanian or Italian embedded sentences). Similarly, for passives/reflexives it appears that the same clusters of meanings can be associated with very different morphosyntactic structures (in Albanian a specialized clitic, a specialized verb inflection and an auxiliary construction, alternating according to tense specifications). On the other hand, it is clusters of meanings that are associated with each given morphosyntax. Attempting to apportion these meanings to different abstracts structures (say, an unaccusative one for passives and an accusative one for reflexives) leads to a loss of predictive power with respect to the actually observed morpholexical forms. But the same is true if a unification is attempted on the basis of the syntactic notion of movement, as in the references just quoted (paralleling Hornstein (1999) on control/raising). If this article is correct, many difficulties can be overcome by having morpholexical structures directly interpreted at the LF interface without passing through a derivational component. Even if this proved wrong, the arguments here and in Manzini and Savoia (2005, 2007) point to the conclusion that the construal of interpreted non-lexicalized arguments/argument slots provided by standard generative theory, specifically in terms of the notion of movement, is not a matter of conceptual necessity. Rather it is an empirical matter, so that other construals (for instance the one proposed here) may turn out to be compatible with the data or even better at capturing some of them – and what is ultimate evidence from linguistic variation may be able to modify our overall perspective on the theory.

176 M. Rita Manzini References Adger, David and Gillian Ramchand 2005 Merge and Move: Wh-dependencies revisit(ed.) Linguistic Inquiry 36: 161–193. Anagnostopoulou, Elena 2003 The Syntax of Ditransitives. Berlin /New York: Mouton de Gruyter. Baker, Mark C. 1988 Incorporation. Chicago: Chicago University Press. Belletti, Adriana 2001 ‘Inversion’ as focalization. In Inversion in Romance and the Theory of Universal Grammar, Aafke Hulk and Jean-Yves Pollock (eds.), 60–90. Oxford /New York: Oxford University Press. Boeckx, Cedric and Norbert Hornstein 2006 The virtues of control as movement. In William D. Davies and Stanley Dubinsky (eds.), 118–130. Borer, Hagit 1989 Anaphoric Agr. In The Null Subject Parameter, Osvald Jaeggli and Ken Safir (eds.), 69–109. Dordrecht: Kluwer. Bouchard, Denis 1984 On the Content of Empty Categories. Dordrecht: Foris. Bresnan, Joan 1982 Control and complementation. Linguistic Inquiry 13: 343–434. Brody, Michael 1997 Towards perfect chains. In Handbook of Syntax, Liliane Haegeman (ed.), 139 –167. Dordrecht: Kluwer. 1999 Relating syntactic elements: Remarks on Norbert Hornstein’s “Movement and chains”. Syntax 2: 210 –226. Brody, Michael and M. Rita Manzini 1988 On implicit arguments. In Mental Representations: The Interface Between Language and Reality, Ruth Kempson (ed.), 105–130. Cambridge: Cambridge University Press. Burzio, Luigi 1986 Italian Syntax. Dordrecht: Kluwer. Butler, Jonny 2004 On having arguments and agreeing: Semantic EPP. In York Papers in Linguistics Series 2 (1): 1–27. Camaj, Martin 1995 Grammatica albanese. Cosenza: Edizioni Brenner. Chierchia, Gennaro 1984 Topics in the syntax and semantics of infinitives and gerunds. Ph.D. dissertation, University of Massachusetts at Amherst

PRO, pro and NP-trace (raising) are interpretations 177 1995

Impersonal subjects. In Quantification in natural languages, Emmon Bach, Eloisa Jellinek, Angelika Kratz, and Barbara Hall Partee (eds.), 107–143. Dordrecht: Kluwer. Chierchia, Gennaro and Sally McConnell-Ginet 1990 Meaning and Grammar. Cambridge, MA: MIT Press. Chomsky, Noam 1973 Conditions on transformations. In A Festschrift for Morris Halle, Stephen R. Anderson and Paul Kiparsky (eds.), 236 –286. New York: Holt, Rinehart and Winston. 1981 Lectures on Government and Binding. Dordrecht: Foris. 1982 Some concepts and consequences of the theory of government and binding. Cambridge, MA: MIT Press. 1995 The Minimalist Program. Cambridge MA: MIT Press. 2000 Minimalist inquiries: The framework. In Step by Step, R. Martin, D. Michaels and J. Uriagereka (eds.), Cambridge, MA: MIT Press. 2001 Derivation by phase. In Ken Hale: A life in language, M. Kenstowicz (ed.), 1–52. Cambridge, MA: MIT Press. 2008 On phases. In Foundational Issues in Linguistic Theory: Essays in Honor of Jean-Roger Vergnaud, Robert Freidin, Carlos P. Otero and Maria Luisa Zubizarreta (eds.), 133–166. Cambridge, MA: MIT Press. Culicover, Peter W. and Ray Jackendoff 2005 Simpler Syntax. Oxford: Oxford University Press. 2006 Turn over control to the semantics! In William D. Davies and Stanley Dubinsky (eds.), 131–152. Frascarelli, Mara 2007 Narrow Focus, Clefting and Predicate inversion. Paper presented at GLOW XXX, University of Tromsø, Norway. Davies, William D. and Stanley Dubinsky (eds.) 2006 Special Issue on Raising and Control. Syntax 9 (2). Halle, Morris and Alec Marantz 1993 Distributed morphology and the pieces of inflection. In The View from Building 20, Kenneth Hale and Samuel Jay Keyser (eds.), 111–176. Cambridge, MA: MIT Press. Hornstein, Norbert 1999 Movement and Control. Linguistic Inquiry 30: 69–96. Iatridou, Sabine 1993 On nominative Case and a few related things. Papers on Case and Agreement II. MIT Working Papers in Linguistics 19: 175–196. Kayne, Richard S. 1991 Romance Clitics, Verb Movement and PRO. Linguistic Inquiry 22: 647– 686. 2000 Parameters and Universals. Oxford: Oxford University Press. 2006 Expletives, datives and the tension between morphology and syntax. Ms, NYU, New York.

178 M. Rita Manzini Landau, Idan 2000 Elements of Control. Dordrecht: Kluwer. 2003 Movement out of control. Linguistic Inquiry 34: 471– 498. 2004 The scale of finiteness and the calculus of control. Natural Language and Linguistic Theory 22: 811– 877. 2006 Severing the distribution of PRO from Case. In William D. Davies and Stanley Dubinsky (eds.), 153 –170. Lebeaux, David 1984 Anaphoric binding and the definition of PRO. In NELS 14, Charles Jones and Peter Sells (eds.) GLSA, University of Massachusetts, Amherst, MA. Manzini, M. Rita 1983 On control and control theory. Linguistic Inquiry 14: 421– 446. Manzini, M. Rita and Anna Roussou 2000 A minimalist theory of A-movement and control. Lingua 110: 409 – 447. Manzini, M. Rita and Leonardo M. Savoia 2005 I dialetti italiani e romanci. Morfosintassi generativa, 3 Vols. Alessandria: Edizioni dell’Orso. 2007 A Unification of Morphology and Syntax. Studies in Romance and Albanian Dialects. London: Routledge to appear Grammatical Categories. Cambridge: Cambridge University Press. Marantz, Alec 1984 On the Nature of Grammatical Relations. Cambridge, MA: MIT Press. Mensching, Guido 2000 Infinitive Constructions with Specified Subjects. New York: Oxford University Press. Moro, Andrea 1997 The Raising of Predicates. Cambridge: Cambridge University Press. Polinsky, Maria and Eric Potsdam 2006 Expanding the scope of control and raising. In William D. Davies and Stanley Dubinsky (eds.), 171–192. Pollock, Jean-Yves 1996 Langage et cognition: Introduction au programme minimaliste de la grammaire generative. Paris: Presses Universitaires de France. Reinhart, Tanya 1997 Syntactic effects of lexical operations: Reflexives and unaccusatives. OTS Working Papers in Linguistics, University of Utrecht. Rivero, María Lusia 1994 Clause structure and V-Movement in the languages of the Balkans. Natural Language and Linguistic Theory 12: 63–120.

PRO, pro and NP-trace (raising) are interpretations 179 Rizzi, Luigi 1982 Issues in Italian Syntax. Dordrecht: Foris. Roberts, Ian G. and Anna Roussou 2003 Syntactic Change. Cambridge: Cambridge University Press. Rosenbaum, Peter S. 1967 The Grammar of English Complement Constructions. Cambridge, MA: MIT Press. Roussou, Anna 2007 In the mood for control. Ms., University of Patras. To appear in Lingua, special issue on Mood. J. Quer (ed.). Rouveret, Alain and Jean-Roger Vergnaud 1980 Specifying reference to the subject: French causatives and conditions on representations. Linguistic Inquiry 11: 97–202. Ruwet, Nicolas 1972 Théorie syntaxique et syntaxe du français. Paris: Seuil. Sigurðsson, Halldór Á. 2007 The case of PRO. Ms., Lund University (http://ling.auf.net/lingBuzz/ 000395). Szabolcsi, Anna 1994 The Noun Phrase. In The Syntactic Structure of Hungarian. Syntax and Semantics 27, Katalin E. Kiss and Ferenc Kiefer (eds.), 179 –274 New York: Academic Press. 2007 Hidden in plain sight: Overt subjects in infinitival control and raising complements. Ms., New York University (http://ling.auf.net/lingBuzz/ 000445). Terzi, Arhonto 1997 PRO and null Case in finite clauses. The Linguistic Review 14: 335 – 360. Turano, Giuseppina 1993 Subjunctive Constructions in Arbëresh and standard Albanian. Rivista di Grammatica Generativa 18: 101–133. Williams, Edwin 1981 Predication. Linguistic Inquiry 11: 203–238. 1985 PRO and the subject of NP. Natural Language and Linguistic Theory 3: 297–315.

Movement of arguments and negative feature Masaaki Kamiya

1. Introduction Woolford (2006) splits nonstructural cases (traditionally called inherent cases) into two classes: inherent case and lexical case. Inherent case is regular and predictable, associated with particular theta-positions (e.g. inherent dative case associated with the experiencer theta-role), whereas lexical case is idiosyncratically licensed by certain lexical heads (verbs and prepositions). Woolford explains the syntactic configuration where both lexical case and inherent case are licensed (Woolford 2006: 113): (1)

a. Lexical heads (e.g. V, P) license idiosyncratic lexical case. b. Little/light v heads license inherent (dative) case (experiencer argument).1

Building on Woolford’s insight, this paper will investigate A-movement of dative argument (which is not relevant for case; EPP at T) and interpretations between universal quantifier and negation. The relevant data that we are concerned with is in (2) and (3): (2)

Taroo-ga zenin-ni sigoto-o atae-nakat-ta. Taroo-NOM all-DAT job-ACC give-NEG-PAST ‘Taroo did not give all jobs.’

all > not, not > all

Note that universal quantifier appears with dative case in (2). On the one hand, ‘all’ in (2) can be interpreted above and below the scope of negation. On the other hand, it is hard to obtain not > all reading when the universal quantifier appears at the subject position (see Miyagawa 2001 for the judgments). (3)

1

Zenin-ni mizu-ga kakar-anakat-ta. all-NOM water-NOM pour-NEG-PAST ‘All did not get water on.’

all > not, *not > all

I have added (dative) and (experiencer argument) to Woolford’s original statement.

182 Masaaki Kamiya Examples in (2) and (3) show the subject and non-subject asymmetry with respect to the interpretations of negation and universal quantifier even if they are dative-marked. Although all > not reading is easily obtained in (3), we found that not > all reading is also possible only when the negative morpheme (underlined) is stressed: (4)

Zenin-ni mizu-ga

kaKAR-ANAkat-ta.

not > all

Then, one may wonder what stressing does for interpretation. In this paper, I propose that stressing on negative morpheme causes its feature to undergo focus movement (see Kato 2000 for negative feature movement in the embedded contexts). Putting details aside for now, the relevant two readings (i.e., all > not, not > all) are obtained as in (5): (5)

a. [CP [TP zenini [NegP [vP zenini [VP V] v] Neg] T] C]2 all > not : ! z--------m b. [CP [TP zenin [NegP [vP … V… ] Neg] T] C] not > all : ! z---------------m Neg-feature

In (5a), negative morpheme stays in-situ (‘all’ takes scope over negation), while in (5b), negative feature undergoes focus movement to C domain where not > all reading is derived. Given that the relevant reading in (5b) is available, I claim that prosody (intonation or stress) is an important factor for interpretations (see Deguchi and Kitagawa 2002, Ishihara 2002, 2004, Kitagawa and Fodor 2006 and among many others). I maintain that stress is a part of focus feature that triggers a relevant feature movement. This paper is organized as follows: Section 2 lays out the treatment of dative case and experiencer arguments in Woolford’s system. Section 3 introduces Japanese inchoative constructions, where subject is also dative marked. Section 4 explains how scope ambiguity and scope freezing occur when subject is dative case marked. We will also explore what stressing on the predicate does to interpretations. Section 5 investigates negative-affixes and their interpretations with respect to the universal quantifier all. I will propose that the negative morpheme that projects itself can undergo negative feature movement. Section 6 concludes this paper. 2

Double-strikethrough means that the trace (or copy) is not accessible for reconstruction, following Boeckx (2001), Chomsky (1995), Lasnik (1999, 2003), Nevin and Anand (2003), Wurmbrand (2003) among others. I will come back to this issue later section.

Movement of arguments and negative feature

183

2. Background: Inherent dative case and experiencer Let us start with background on Woolford’s system. Woolford examines the so-called inherent cases and splits them into inherent case and lexical case. Inherent case is regular and predictable, associated with particular thetapositions, while lexical case is idiosyncratic and is licensed by certain verbs or prepositions. The crucial data that Woolford relies on are DP goals and PP goals in ditransitive verbs from Icelandic and German. For example, in Icelandic, DP goals typically take inherent dative case and this is licensed by a particular configuration in vP, whereas PP goals take whatever case the particular P licenses its complement (i.e. idiosyncratic case). The following Icelandic examples are cited from Woolford (2006: 114):3 (6)

a. Ég skilaði henni peningunum. I returned her-DAT the money-DAT b. Ég skilaði peningunum til hennar. I returned the money-DAT to her-GEN

On the one hand, in (6a), the DP goal henni takes inherent dative case. On the other hand, in (6b), the goal inside the PP gets genitive case since it is the lexical case that the preposition til ‘to’ licenses. With respect to the particular place where inherent dative case is licensed, Woolford proposes that a light verb (vG in her terminology) between VP proper and v*P (in the sense of Chomsky 2000) is responsible. (7)

3

v*P ru External v* Argument ru v* vGP ru a-------> DP vG Licensing inherent dative ru z----------- vG VP ru V Theme/ Internal Argument

According to Woolford, these data are originally from Zaenen, Maling, and Thráinsson (1985)

184 Masaaki Kamiya To diagnose whether or not a DP obtains inherent dative case, Woolford utilizes constructions involving A-movement, such as passive and raising, since inherent case can be preserved under A-movement. The following examples are cited from Woolford (2006: 118): (8)

a. Þeir skiluðu Maríu bókinni. they returned Mary-DAT book-the-DAT ‘They returned the book to Mary.’ b. Maríu var skilað þessari bók. Mary-DAT was returned this book-DAT

In (8), Maríu appears in the same form both in the active sentence (8a) and the passive (8b). This indicates that the DP goal maintains inherent case, according to Woolford. The same observation carries over to Japanese dative case -ni. Woolford shows that the DP goal in Japanese gets inherent dative case in passive constructions (Woolford 2006: 120): (9)

a. John-ga Mary-ni sono hon-o okutta. John-NOM Mary-DAT that book-ACC sent ‘John sent Mary that book.’ b. Mary-ni sono hon-ga okur-are-ta. Mary-DAT that book-NOM send-PASS-PAST ‘Mary was sent that book.’

In (9), the DP goal Mary in active and passive obtains the dative case -ni, which proves Woolford’s point. Woolford also presents raising constructions in Icelandic to show that dative case (i.e. Barninu ‘child-Dat in 10 a and 10 b) is preserved (Woolford 2006: 120): 4 (10) a. Barninu batnaði veikin child-DAT recovered-from disease-NOM ‘The child recovered from the disease.’ b. Barninu virðst [t hafa batnað veikin]. child-DAT seems [t to-have recovered-from disease-NOM] ‘The child seems [t to have recovered from the disease].’ Based on A-movement tests such as passivization and raising, languages such as Icelandic, German, and Japanese seem to possess inherent dative case. With this much background in mind, we now turn to inchoative variants and experiencer subjects in Japanese. 4

Originally, these examples are from Yip, Maling, and Jackendoff (1987: 233).

Movement of arguments and negative feature

3.

185

Inchoative variants of Japanese ditransitive verbs and experiencer subject

3.1. Subject in inchoative variants Japanese has a certain class of ditransitive verbs that show causative – inchoative alternations. Consider the following examples: (11) a. Taroo-ga Hanako-ni mizu-o kaketa. Taroo-NOM Hanako-DAT water-ACC put on ‘Taroo put water on Hanako.’ b. Hanako-ni Mizu-ga kakatta. Hanak-DAT water-NOM put on (lit.) ‘Hanakoi experiences pouring water on (heri).’ (12) a. Taroo-ga Hanako-ni booru-o ateta. Taroo-NOM Hanako-DAT ball-ACC hit ‘Taroo hit a ball to Hanako.’ b. Hanako-ni booru-ga atatta. Hanako-DAT ball-NOM hit (lit.) ‘Hanakoi experiences ball hitting on (heri).’ Examples (11) and (12) are causative – inchoative pairs of ditransitive verbs. The verbs kaketa ‘put on’ in (11a) and ateta ‘hit’ in (12a) are causative, while kakatta ‘put on’ in (11b) and atatta ‘hit’ in (12b) are inchoative variants. ‘Hanako’ in (11b) and (12b) appears with the dative case marker -ni, and ‘Hanako’ denotes an experiencer in the sense of Matsumoto (2000) and Matsuoka (2003): “the dative argument refers to the individual that is affected and undergoes a change of state in some way” (Matsuoka 2003: 189). I will follow this definition of experiencer in this paper. With this definition, ‘Hanako’ in (11b) experienced getting water on her, and as a result, she is wet.5 The dative case -ni in these examples shows inherent status under the passivization test in Woolford:

5

One may wonder if Hanako in (11) and (12) is a location. However, the following example clarifies Hanako as an experiencer, and asi ‘foot’ is a location: (i) Hanako-ni mizu-ga asi-ni kakatta. Hanako-DAT water-NOM foot-on poured ‘(lit.) Hanako experienced water pouring on her foot.’

186 Masaaki Kamiya (13) Hanako-ni mizu-ga kaker-are-ta. Hanako-DAT water-NOM pour-PASS-PAST (lit.) ‘Hanako was poured water on.’ / ‘Hanako got water poured on her.’ (14) Hanako-ni booru-ga ater-are-ta. Hanako-DAT ball-NOM hit-PASS-PAST (lit.) ‘Hanako was hit a ball.’ / ‘Hanako hand a ball hit her.’ The passive test shows that ‘Hanako’ gets inherent dative case. In addition, the following raising construction supports that -ni on ‘Hanako’ is inherent dative case: (15) Hanako-ni mizu-ga kakari kaketa. Hanako-DAT water-NOM pour about-to was (lit.) ‘Hanako was about to get water on.’ In (15), under raising construction kaketa ‘was about to’, the dative case -ni is preserved. While the passive and raising tests show that ‘Hanako’ in (11) and (12) has inherent dative case, it is also a subject, as shown by well-known diagnostics in Japanese: zibun-binding and subject-honorification. First, in the literature, it is known that the subject-oriented anaphor zibun ‘self’ and zibun-zisin ‘self-self’ cannot be coreferential with any non-subject (Shibatani 1977; Ura 1996). This applies to experiencer dative subjects: (16) a. Hanakoi -ni zibuni /*j -no mizu-ga kakatta. Hanako-DAT self-GEN paint-NOM poured (lit.) ‘Hanakoi experienced heri paint poured on heri.’ b. Hanakoi -ni zibun-zisini /*j no penki-ga kakatta. Hanako-DAT self-self-GEN paint-NOM poured (lit.) ‘Hanakoi experienced heri paint poured on heri.’ In this example, the subject-oriented anaphor zibun ‘self’ and zibun-zisin ‘self-self’ take ‘Hanako’ as its antecedent. This shows that ‘Hanako’ is a subject in these sentences. Second, in Japanese, an honorific form of a verb agrees with the subject (Shibatani 1977). (17) Yamada-sensee-ni sono mondai-ga o-wakari-ni nar-u Yamada-professor-DAT that problem-NOM HON-understand-to become -PRES ‘Professor Yamada understands that problem.’

Movement of arguments and negative feature

187

In (17), the honorific verbal form semantically agrees with the subject ‘Yamada-sensee’. In (17), the subject ‘Yamada-sensee’ is dative-marked, and the subject and honorific verbal form agree semantically. The following example is the same as (17): (18) Yamada-sensee-ni mizu-ga o-kakari-ni nat-ta. Yamada-professor-DAT water-NOM HON-pour-to become-PAST ‘Professor Yamada got water (on him).’ The subject-oriented anaphor and honorification show that the dative experiencer argument has subject status. In sum, passivization and raising tests prove that -ni in (11) and (12) is an inherent case marker which is assigned to ‘Hanako’ (experiencer). In addition, zibun-binding and honorification shows the subject status of inherent dative marked DPs in Japanese. So, just like languages that Woolford examines, I assume that EPP triggers A-movement to T in Japanese. Now, although Woolford states that inherent dative case is licensed at SPEC-vG, she does not specify where the experiencer argument starts. She is neutral as to whether or not the experiencer (or DP goal in her terminology) is base-generated at SPEC-vG in line with McGinnis (1996, 1998, 2001) or moved from lower position in line with Baker (1988, 1997). In the next subsection, following an insight of Miyagawa and Tsujioka (2004), I will demonstrate that the experiencer subject in inchoative variants is thematically higher than the theme and goal/location, and that the experiencer is base-generated at SPEC-vG. 3.2. Base generation of the experiencer Miyagawa and Tsujioka (2004) attempt to show that there are two goals in Japanese ditransitive verbs, the high goal and the low goal. Although the terminology is different, these are essentially the same as the DP goal and PP goal in Woolford (2006). Observe (19). There are two -ni marked DPs: Hanako-ni and uti-ni. (19) Taroo-ga Hanako-ni kozutumi-o uti-ni okutta. Taroo-NOM Hanako-DAT package-ACC house-to sent ‘Taroo sent Hanako a package to her house.’ Miyagawa and Tsujioka utilize numeral quantifier float as a diagnostic of the categorial status of Hanako-ni and uti-ni. What this test shows is that a

188 Masaaki Kamiya numeral quantifier can float off from its host DP if it is marked with dative case, while it cannot if it occurs with the postposition ni. (20) a. Taroo-ga [san-nin-no gakusee]-ni hon-o okutta. Taroo-NOM three-CL-GEN student-DAT book-ACC sent ‘Taroo sent three students a book.’ b. Taroo-ga [ti gakusee]-ni san-nini hon-o okutta. (21) a. Daitooryoo-ga [huta-tu-no kokkyoo]-ni heetai-o okutta. president-NOM two-CL-GEN border-to solders-ACC sent ‘The president sent solders to two borders.’ b. *Daitooryoo-ga [ti kokkyoo]-ni huta-tui heetai-o okutta. (Miyagawa and Tsujioka 2004: 7) In (20b), the numeral quantifier san-nin ‘three-CL’ floats off from its host DP gakusee ‘student’, while in (21b) the numeral quantifier huta-tu ‘twoCL’ cannot. This behavior distinguishes case-markers from postpositions in Japanese in general. In addition, Miyagawa and Tsujioka show that the thematic hierarchy determines scope interactions (Miyagawa and Tsujioka 2004: 15): (22) Taroo-ga dareka-ni subete-no-basyo-ni nimotu-o okutta. Taroo-NOM someone-DAT all-GEN-place-to package-ACC sent ‘Taroo sent someone a package to every place.’ some > every, *every > some (23) Taroo-ga dareka-ni Tookyoo-ni subete-no-nimotu-o okutta. some > every, *every > some These examples show that experiencer dareka ‘someone’ takes wide scope over both location ‘Tokyo’ and theme nimotu ‘package’. However, Miyagawa and Tsujioka note that theme and location are not hierarchical with respect to each other due to the fact that inverse scope is possible (Miyagawa and Tsujioka 2004: 15): (24) Taroo-ga Hanako-ni dokoka-no-basyo-ni subete-no nimotu-o okutta. Taroo-NOM Hanako-DAT some-GEN-place-to all-GEN-package-ACC sent ‘Taroo sent Hanako every package to some place.’ some > every, every > some

Movement of arguments and negative feature

189

Miyagawa and Tsujioka assume that the relevant thematic hierarchy is agent > experiencer (high goal) > theme, goal/location (low goal). I assume that they are correct. Miyagawa and Tsujioka’s tests show that the experiencer subject (DP goal) is higher than the location (or PP goal). (25)

vP ru Taroo v ru VP1 v ru Hanako-DAT V1 ru VP2 V1 (vG) wo PP V2 ru ru Tokyo to a package V2 (send)

Let us turn our attention to the experiencer subject in inchoative variants. The following examples use the numeral quantifier test and scope test in inchoative constructions. (26) a. San-nin-no gakusee-ni booru-ga atatta. three-CL-GEN student-DAT ball-NOM hit b. [ti Gakusee]-ni san-nini booru-ga atatta. (27) Dareka-ni subete-no-booru-ga atatta. someone-DAT all-GEN-ball-NOM hit some > every, * every> some Just as in Miyagawa and Tsujioka’s examples, the numeral quantifier can float off from its host in (26) and scope interaction is restricted to some > every (not every > some) in (27). These tests indicate that -ni is a dative marker, and the thematic hierarchy is experiencer > theme. Following Miyagawa and Tsujioka’s tree diagram in (25), I assume that the experiencer is generated at SPEC-vG.6 6

Miyagawa and Tsujioka’s tree diagram is essentially the same as Woolford (2006) since both authors assume Marantz (1993) and McGinnis (1996, 1998, 2001).

190 Masaaki Kamiya In sum, the previous subsection and the current subsection examine the type of case on experiencer and the location of it in vP structure. We found that dative case is licensed at SPEC-vG and undergoes A-movement to check an EPP feature at T in passive and raising constructions. Now, although experiencer starts at SPEC-vG, its interpretations with respect to universal quantifier and negation differ, as shown in the introduction: (28) Taroo-ga zenin-ni sigoto-o atae-nakat-ta. Taroo-NOM all-DAT job-ACC give-NEG-PAST ‘Taroo did not give all jobs.’ (29) Zenin-ni mizu-ga kakar-anakat-ta. all-NOM water-NOM pour-NEG-PAST ‘All did not get water on.’

all > not, not > all

all > not, *not > all

What causes different interpretations? In the next section, we will investigate how the relevant readings are obtained. 4. A-movement, anti-reconstruction, and focus movement In the previous section, we have observed that dative argument in double object constructions is inherent case in the sense of Woolford. We also observed that dative subject in inchoative constructions passes typical A-movement diagnostic in Woolford. (30) a. John-ga Mary-ni sono hon-o okutta. John-NOM Mary-DAT that book-ACC sent ‘John sent Mary that book.’ b. Mary-ni sono hon-ga okur-are-ta. Mary-DAT that book-NOM send-PASS-PAST ‘Mary was sent that book.’ (31) Hanako-ni mizu-ga kakari kaketa. Hanako-DAT water-NOM pour about-to was ‘(lit.) Hanako was about to get water on.’ Following Woolford, I assume that the motivation of A-movement is to check an EPP feature at T, not for case. Then, how come only one reading (i.e., all > not) becomes available in the subject position (without predicate

Movement of arguments and negative feature

191

stress)? In the next subsection, we will explain the relevant derivations. In addition, we will account for what predicate stress does for interpretations.

4.1. Anti-reconstruction In the literature, researchers have been examining the correlations between A-movement and anti-reconstruction (Boeckx 2001, Chomsky 1995, Lasnik 1999, 2003, Nevin and Anand 2003, Wurmbrand 2003 among others). Among them, Nevin and Anand report interesting paradigm in Hindi: (32) a. kisii šaayer-ne har ghazal likhii some poet-ERG every song-NOM write.f-PERF ‘Some poet wrote every song.’ (some > all, * all > some) b. koi šaayer har ghazal likhtaa hai some poet-NOM every song-ACC write.m-IMPF be-PRES ‘Some poet writes every song.’ (some > all, all > some) In Hindi, ergative case is a lexical case, which does not require movement to obtain case. What is interesting in this data is that while the nominative marked subject in (32b) allows inverse scope reading, this is not the case in (32a). Nevin and Anand analyze that T in Hindi has an EPP feature and that the movement only for checking off an EPP feature cannot reconstruct. (33) Purely EPP Eliminates Reconstruction (PEPPER): A-movement only for EPP does not reconstruct (Nevin and Anand 2003: 370) The relevant phenomena are found in other languages (examples are cited from Nevin and Anand). (34) Russian a. puli ranili soldat bullets-NOM wound-PAST soldiers-ACC ‘Bullets wounded soldiers.’ b. soldat ranilo puljami soldiers-ACC wound-PAST bullets-INST ‘Bullets wounded soldiers.’ There are two types of adversity construction in Russian and as (34) shows, the non-agentive logical subject is either nominative (34a) or instrumental

192 Masaaki Kamiya (34b) case marked. The scope freezing is observed only when the nonagentive subject is instrumental case marked as in (35a). (35) a. dvux soldat ranilo kazhdoj pulej two-ACC soldiers-ACC would-PAST every-INST bullet-INST ‘Every bullet wounded two soldiers.’ *every > two, two > every b. dve puli ranili kazhdogo soldata two-NOM bullet-NOM wounded every-ACC soldier-ACC ‘Two bullets wounded every soldier.’ two > every, every > two Dvux in (35a) moves to T only to check an EPP feature, not case. Therefore, scope freezing takes place, just like Hindi. As another example, passive construction in Greek shows the same phenomenon. When the goal argument in Greek double object constructions undergoes A-movement in passive, the goal argument appears either in nominative or PP form. When the goal argument with PP appears as a subject in passive construction, scope freezing is observed as in (36 b, c): (36) Greek: Dative or PP in passive shows the scope freezing. a. Fronted Nom DP passive kapjo vivlio tu stalthike kathe ekdhoti some book-NOM CL was-sent every reviewer-GEN ‘Some book was sent to every reviewer.’ some > every, every > some b. Fronted Dat DP passive enos ekdhoti (tu) stalthike kathe vivlio some reviewer-GEN CL was-sent every article-NOM ‘Some reviewer was sent every article.’ some > every, *every > some c. Fronted PP goal sena ekdhoti stalthike kathe vivlio to-some reviewer was-sent every article-NOM ‘To some reviewer was sent every article.’ some > every, * every > some These crosslinguistic data show the striking similarities with Japanese dative subject when it appears with universal quantifier in negative context. For all > not reading, I assume that dative subject moves to SPEC-TP only to check an EPP feature, and, hence, inverse scope reading is not available just as in Hindi, Russian, and Greek.

Movement of arguments and negative feature

193

(37) [TP zenini [NegP[vP zenini [VP V] v] Neg] T]7 : ! z--------m In (37), zenin ‘all’ moves to SPEC-TP for EPP checking. Note that zenin ‘all’ in vP is double-strikethrough which means that the trace is not accessible for reconstruction.8 Thus, Japanese dative subject construction supports Nevin and Anand (2003). Hence, inverse scope cannot be obtained. However, as I mentioned in the introduction, inverse scope is available only when the predicate part is stressed. How can I account for this reading? In the next section, I will explain how the inverse scope is obtained. 4.2. Negative Feature Movement To investigate negative feature and interpretations, Kato (2000) finds that certain verbs with negative meaning induce ambiguous readings regarding universal quantifier. The following example is cited from Kato (2000: 83): (38) Zenin-ga konai-to komaru. all-NOM come-NEG-if be in trouble (i) (We will be) in trouble, if nobody comes. (ii) (We will be) in trouble, if it is not the case that everyone comes. Kato suggests that complement taking verbs like komaru ‘be in trouble’ attracts the negative feature on nai ‘not’. (39) [VP [CP [IP …nai ] [C]] komar] : ! z------m +neg-feature Kato explains that (ii)-reading in (38) is obtained by neg-feature raising to verb komar ‘be in trouble’. Kato assumes that neg-attraction is only triggered by a certain class of verbs with negative meanings. Although Kato supposes that neg-feature raising is only applied to certain class of verbs, I utilize his idea to our example by assuming (null) CP projection. In the lit7 8

I assume that Neg has its own projection as Watanabe (2004). See Lasnik (1999, 2003) regarding A-movement and anti-reconstruction. According to Lasnik, A-movement does not leave a trace, and hence, scope freezing results.

194 Masaaki Kamiya erature, it is reported that quantifier-scope interpretations are susceptible to focus intonation (Beghelli and Stowell 1997 for English and Miyagawa and Arikawa 2007 among others). Although the meaning shift by focus (or stressing certain words) is noticed, the details have not been discussed.9 Here, I try to explain how the relevant reading is obtained. (40) Zenin-ni mizu-ga kaKAR-ANAkat-ta. all-DAT water-NOM pour-NEG-PAST Recall that not > all reading is obtained only when the predicate is stressed (otherwise, the relevant reading is all > not). I assume that stressing triggers focus movement of negative feature and this is coded at CP domain. (41) [CP [TP zenin [vP … V… ] na T] C] : ! z------------m Neg-feature In (41), neg-feature undergoes focus movement to SPEC-CP where negfeature takes scope over zenin ‘all’. I assume that this is a derivation for not > all reading. As I suppose in (41), neg-feature focus movement is crucially dependent on the existence of CP where focus is coded. This means that neg-feature raising does not take place where CP (or its equivalent which codes focus) does not exist. This is actually borne out. A case in point is nominalization. In Japanese, the morpheme -sa nominalizes adjectival. (42) taka-i → taka-sa high height Japanese desiderative form and negation -na ‘not’ are also categorially adjectives and the morpheme -sa can nominalize the relevant sentences: (43) a. Taroo-wa hasiri-ta-i. Taroo-TOP run-want-PRES ‘Taroo wants to run.’ 9

For example, Beghelli and Stowell (1997: 96) report that (i) only has all > not reading under the neutral intonation, although they imply that changing intonation patterns changes interpretation. (i) All the boys didn’t leave. Miyagawa and Arikawa (2007) admit that sentence like (40) could be ambiguous by stressing the predicate, although the details are not discussed.

Movement of arguments and negative feature

195

b. [NP Taroo-no hasiri-ta-sa] Taroo-GEN run-want-NML ‘(How much) Taroo wants to run’10 (44) a. Taroo-wa hasiri-ta-ku-na-i. Taroo-TOP run-want-KU-NEG-PRES ‘Taroo does not wan to run.’ b. [NP Taroo-no hasiri-ta-ku-na-sa] Taroo-GEN run-want-KU-NEG-NML11 ‘(How much) Taroo does not want to run.’ Both (43b) and (44b) are nominalized forms of (43a) and (44a), respectively. Now, observe the interaction between universal quantifier and negation in the nominalized form of (45a): (45) a. Zenin-ga hasiri-ta-ku-na-i. all-NOM run-want-KU-NEG-PRES ‘All do not want to run.’ b. [NP Zenin-no hasiri-ta-ku-na-sa]-wa minnna-ni wakatta. all-GEN run-want-KU-NEG-NML-TOP everyone-DAT understood ‘Everyone understood (how much) all did not want to run.’ According to my informants, (45a) is ambiguous with respect to universal quantifier and negation. Note that stressing on the predicate is the key to obtain not > all reading. However, the only reading that is available in (45b) is all > not. One of my informants reported that stressing on the predicate does not produce not > all reading.12 I take this as independent evidence that lacking of CP (where focus is coded) is a source of impossible reading (not > all). 10

11

12

Note that even if the English equivalent is sentential, this is a nominalized form. In addition, -sa denotes degree, so I added ‘how much’ for its English equivalent in example (43b). Aoyagi (2006) proposes that -ku is a head of AP. Whatever the categorial status of -ku is, this does not affect our argument. When ‘all’ appears in non-subject position in nominalizations, both readings (i.e., all > not and not > all) are available. (i) [NP Taroo-no zen-gaku-no tukai-taku-na-sa]-wa watasi-ni wakatta. Taroo-GEN all-sum-GEN use-want-NEG-NML-TOP I-DAT understood ‘I understand how much Taroo did not want to spend all sum.’ (all > not and not > all)

196 Masaaki Kamiya (46)

[NP Zenin-no hasiri-ta-ku-na-sa] : z------=----- Neg-feature

((46) shows that negative morpheme na ‘not’ is stressed for focus feature checking.) Although negation is stressed, nominalization does not contain CP where focus is coded. Therefore, the relevant meaning (i.e., not > all) cannot be obtained. Thus, just like Kato’s (2000) work, the current paper argues for neg-feature raising. However, unlike Kato, I claim that negfeature undergoes focus movement even in matrix sentences.

4.3. Universal Quantifier in Non-Subject Position In the previous two subsections, I show how the relevant readings (i.e., all > not and not > all) are derived when the universal quantifier is in the subject position. As I showed, I assume that A-movement only for EPP does not reconstruct, and therefore only all > not reading is available. On the other hand, not > all reading is available on stressing predicate position. I explained that this has to do with focus neg-feature movement in the same line as Kato. Now, I will account for how ambiguous readings are obtained when the universal quantifier appears in non-subject position. Recall the relevant example: (47) Taroo-ga zenin-ni sigoto-o atae-nakat-ta. Taroo-NOM all-DAT job-ACC give-NEG-PAST ‘Taroo did not give all jobs.’

all > not, not > all

According to Diesing (1992), a quantifier like ‘all’ must move out of VP to take scope (or QR).13 I assume that ‘all’ in Japanese undergoes QR, and its landing cite is outer SPEC-TP (Heim and Kratzer 1988): (48) [TP zenin-nij [TP Taroo-gai [vP ti tj sigoto-o atae] nakat ta]] (48) is a LF representation of (47). I suppose that all > not reading becomes available when ‘all’ undergoes QR where it takes scope over negation as in (47). On the other hand, not > all reading is available either at in-situ position or the trace of ‘all’. This is hard to decide, but either way the relevant 13

See Kawashima (1998) for universal quantifier extraction in nominal construction.

Movement of arguments and negative feature

197

reading is available. Since my informants did not report to me that stressing predicate is not necessary to obtain relevant readings, I assume that focus neg-feature movement does not occur when ‘all’ appears in non-subject positions. 5.

Further examinations of neg-feature movement

5.1. Lexical negative prefix In the previous section, I showed that neg-feature moves CP domain where focus is coded. I suggested that neg-feature movement is a part of focus movements and as a result negation takes scope over the universal quantifier (i.e., not > all reading). In this section, I will examine whether or not negative-feature always moves as a focus movement. I will conclude that only the negative morpheme that projects its own can undergo neg-feature movement. In the previous section, we investigated the interactions between negative morpheme -na ‘not’ and the universal quantifier all. However, there are other types of negative morphemes in Japanese; hu-, mi-, or mu- ‘not’. Negative prefixes hu-, mi-, or mu- ‘not’ has a few characteristics. First, just as negative morpheme -na ‘not’, they have a negative feature (Kato 2000: 72–73). Second, they are not completely productive (Kageyama 1993: 24). Kageyama (1993) shows that these negative prefixes hu-, mi-, or mu- have selectional restriction; they only select lexical items with nominal features. (49) a. hu-sinsetu NEG-kind ‘not kind’ b. mu-kansin NEG-interest ‘not interest’ c. mi-kaihatu NEG-development ‘not develop’

hu-tyuui NEG-attention ‘not (paying) attention’ mu-kiryoku NEG-energy ‘not (have) energy’ mi-teesyutu NEG-submission ‘not submission’

(50) a. *hu-umai NEG-tasty ‘not tasty’

*hu-erai NEG-great ‘not great’

198 Masaaki Kamiya b. *hu-awu NEG-fit ‘not fit’

*hu-nareru NEG-used to ‘not used to’

(Examples in (50) are cited from Kageyama (1993: 24). English equivalents are mine.) The examples in (49) are so-called adjectival nouns; they show both nominal and adjectival characters. Since the negative morphemes such as mu-, hu-, or mi- can only be attached to nominals, (49) is fine. On the other hand, examples in (50) are either adjectives (50a) or verbs (50b); hence affixation of these negative morphemes is not possible. Third, unlike negative morpheme -na ‘not’, negative prefixes mu-, hu-, or mi- cannot be separated by a focus particle such as -mo ‘also, even’. (51) Taroo-wa keeki-o tabe-mo-si-nakat-ta. Taroo-TOP cake-ACC eat-even-do-NEG-PAST ‘Taroo did not even eat a cake.’ (52) *Taroo-wa kodomo-ni hu-mo-sinsetu da. Taroo-TOP children-DAT NEG-even-kind is ‘Taroo is not even kind to children.’ As in (51), focus particle mo ‘even’ appears between verb tabe ‘eat’ and negative morpheme -na ‘not’; this is not the case for negative prefix hu- as in (52). Due to the limited productivity of negative affixes, Martin (1975) calls these affixes lexical negative. In the next section, we will observe how these lexical negatives interact with universal quantifiers.

5.2. Interactions between universal quantifier and negative affixes Consider the following sentences: (53) a. Taroo-wa kono kamoku-ni mu-kansin-da. Taroo-TOP this subject-DAT NEG-interest-is ‘Taroo is not interested in this subject.’ b. Taroo-wa kono kamoku-ni kansin-ga na-i. Taroo-TOP this subject-DAT interest-NOM NEG-PRE ‘Taroo does not have interest in this subject.’

Movement of arguments and negative feature

199

(54) a. Taroo-wa roozin-ni hu-sinsetu-da. Taroo-TOP elders-DAT NEG-kind-is ‘Taroo is not kind to the elders.’ b. Taroo-wa roozin-ni sinsetu-zya na-i. Taroo-TOP elders-DAT kind-zya NEG-PRE ‘Taroo is not kind to the elders.’ These examples are minimal pairs of negations. In the (a)-examples, the negative prefix negates the sentences while in the (b)-examples negative morpheme na- ‘not’ negates the sentences. Now, let us consider the interactions between negation and the universal quantifier all. (55) a. Zenin-ga kono kamoku-ni mu-kansin-da. all-NOM this subject-DAT NEG-interest-is ‘All is not interested in this subject.’

all > not, *not > all

b. Zenin-ga kono kamoku-ni kansin-ga na-i. all-NOM this subject-DAT interest-NOM NEG-PRE ‘All does not have interest in this subject.’ all > not, not > all (56) a. Zenin-ga roozin-ni hu-sinsetu-da. all-NOM elders-DAT NEG-kind-is ‘All is not kind to the elders.’ b. Zenin-ga roozin-ni sinsetu-zya na-i. all-NOM elders-DAT kind-zya NEG-PRE ‘All is not kind to the elders.’

all > not, *not > all

all > not, not > all

As the (a)-examples show, when the universal quantifier appears with a negative prefix, only all > not reading is available. In examples (55a) and (56a), the universal quantifier all appears in the subject position. However, only all > not interpretation is found in the non-subject position. (57) a. Taroo-wa zen-kamoku-ni mu-kansin-da. Taroo-TOP all-subject-DAT NEG-interest-is ‘Taroo is not interested in all subjects.’ b. Taroo-wa zen-roozin-ni hu-sinsetu-da. Taroo-TOP all-elder-DAT NEG-kind-is ‘Taroo is not kind to all the elders.’

all > not, * not > all

all > not, * not > all

As the examples in (57) show, the only all > not reading is available when the universal quantifier appears in the non-subject position.

200 Masaaki Kamiya On the other hand, as we observed in section 4.2, the negative morpheme can take a wide scope over the universal quantifier when the negative part is stressed. In order to account for the wide scope reading of negation, I suggested the following derivations: (58) [CP [TP [vP …V …] na T] C] : ! z---------m Neg-feature The neg-feature undergoes a movement to CP domain where focus is coded. Recall that this reading is available only when the predicate part is stressed. However, the same strategy does not work when the negative prefix is added: (59) a. Zenin-ga kono keekaku-ni MU-kansin-da. all-NOM this project-DAT NEG-interest-is ‘All is not interested in this project.’

all > not, *not > all

b. Zenin-ga roozin-ni HU-sinsetu-da. all-NOM elders-DAT NEG-kind-is ‘All is not kind to the elders.’

all > not, *not > all

As example (59) shows, the only available reading is all > not. Then, the question is why can’t the negative feature of negative affixes such as hu-, mu-, or mi- undergo movement? There are two crucial differences between negative morpheme na- ‘not’ and negative affixes. First, while negative morpheme na- appears with any predicate, negative affixes have a limited environment. In other words, these negative sentences with negative affixes are not productive. Second, while negative morpheme na- can be separated by a focus particle, this is not the case with negative affixes. In the literature, negative morpheme na- has its own projection (Aoyagi and Ishii 1994; Kishimoto 2007; Watanabe 2004, among many others). Following these authors, I assume that the negative morpheme has its own projection. Based on the observations, I propose that neg-feature movement is more restricted to the type of negation that has its own projection and appears with any predicate. Negative affixes which do not have its own projection do not undergo negative feature movement. So, negative feature movement is limited to the negative morpheme that has its own projection. As a matter of fact, once negative-affixed words are negated by negative morpheme -na, then the inverse scope reading becomes available.

Movement of arguments and negative feature

201

(60) a. Zenin-ga kono keekaku-ni mu-kansin-zya na-i. all-NOM this project-DAT NEG-interest-zya NEG-PRE ‘All is not interested in this project.’ all > not, not > all b. Zenin-ga roozin-ni hu-sinsetu-zya na-i. all-NOM elders-DAT NEG-kind-zya NEG-PRE ‘All is not kind to the elders.’

all > not, not > all

As example (60) shows, not > all reading becomes possible (when the predicate part is stressed). Thus, based on these examples, the negative feature movement is limited to the negative morpheme which has its own projection.

6. Conclusions In this paper, we explore the interpretations between dative subject with universal quantifier and negation. Following Woolford’s diagnostic of (inherent) dative case, I showed that Japanese dative case also belongs to inherent dative case. This is also applied to dative subject construction in Japanese (inchoative variant of causative – inchoative pairs). Therein, we assume that dative case is licensed at vG, and the only motivation for dative subject to undergo A-movement (at SPEC-TP) is for EPP. This assumption is squared with crosslinguistic facts (Hindi, Russian, and Greek). That is, movement only for EPP does not reconstruct (Nevin and Anand 2003). I further explored that the inverse scope reading is obtained when the predicate is stressed. I explained that neg-feature undergoes focus movement, and this is only observed when CP is projected (recall that negfeature focus movement does not occur when negation appears in nominalization). Furthermore, I examined that only the negative morpheme -na ‘not’, which has its own projection, can undergo neg-feature movement. Now, if the current observations are on the right track, it raises a question on recent research on CP and DP parallelism (Abney 1987; Hiraiwa 2005; Larson 1999; Roeper and van Hout 2006; Szabolcsi 1994, among others). If DP is an equivalent to CP to a certain degree, what prevents DP from coding focus feature? How much is it similar between CP and DP? I will leave these questions for future research.

202 Masaaki Kamiya Acknowledgements An earlier version of this paper was presented at the InterPhases Conference in Nicocia, Cyprus on May 18 –20, 2006. I would like to thank Kleanthes Grohmann for organizing a stimulating conference. I also thank an anonymous reviewer for giving me valuable comments. As usual, I am alone responsible for any remaining errors and oversights. References Abney, Steven 1987 The English noun phrase in its sentential aspect. Doctoral dissertation, MIT. Adger, David and Cécile De Cat 2004 Core questions about the edge. In Peripheries: Syntactic Edges and their Effects, David Adger, Cécile De Cat, and George Tsoulas (eds.), 1–18. Dordrecht: Kluwer. Aoyagi, Hiroshi 2006 Nihongo no zyosi to kinoo hantyuu (Japanese particles and functional projections). Tokyo: Hitsuji syoboo. Aoyagi, Hiroshi and Toru Ishii 1994 On agreement-inducing vs. non-agreement-inducing NPIs. In Proceedings of the 24th North East Linguistic Society, 1–15. GLSA. Baker, Mark 1988 Incorporation: A Theory of Grammatical Function Changing. Chicago: University of Chicago Press. 1997 Thematic roles and syntactic structure. In Elements of Grammar, L. Haegeman (ed), 73–137. Dordrecht: Kluwer. Beghelli, Filippo and Tim Stowell 1997 Distributivity and negation: The syntax of each and every. In Ways of Scope Taking, Anna Szabolcsi (ed.), 71–107. Dordrecht: Kluwer. Boeckx, Cedric and Kleanthes K. Grohmann 2004 Submove: Towards a unified account of scrambling and D-linking. In Peripheries: Syntactic Edges and their Effects, D. Adger, C. Cat, and G. Tsoulas (eds.), 241–257. Dordrecht: Kluwer. Chomsky, Noam 1995 The Minimalist Program. Cambridge, MA: MIT Press. 2000 Minimalist inquiries: The framework. In Step by Step, R. Martin, D. Michaels, and J. Uriagereka (eds.), 89–155. Cambridge, MA: MIT Press. 2005 On phases. Ms., MIT. 2006 Approaching UG from below. Ms., MIT.

Movement of arguments and negative feature

203

Deguchi, Masanori and Yoshihisa Kitagawa 2002 Prosody in syntactic analyses. Ms., Indiana University. Ishihara, Shinichiro 2002 Invisible but audible wh-scope marking: wh-constructions and deaccenting in Japanese. In Proceedings of the 21st West Coast Conference on Formal Linguistics, 180–193. Stanford, CA: CSLI Publications. Ishihara, Shinichiro 2004 Prosody-scope match and mismatch in Tokyo Japanese wh-questions. Ms., Universität Potsdam. Diesing, Molly 1992 Indefinites. Cambridge, MA: MIT Press. É. Kiss, Katalin 1998 Identificational focus versus informational focus. Language 74: 245– 273. Hiraiwa, Ken 2005 Dimensions of symmetry in syntax: agreement and clausal architecture. Doctoral dissertation, MIT. Hiraiwa, Ken and Shinichiro Ishihara 2001 Missing links: cleft, slucing, and ‘no da’ construction in Japanese. Proceedings of HUMIT 2001, MIT Working Papers in Linguistics 43: 35–54. Kageyama, Taroo 1993 Bunpoo to gokeesei (Grammar and word formation). Tokyo: Hitsuzi Syoboo. Kamiya, Masaaki 2006 DP goal, PP goal, and vP internal structure in Japanese. Proceedings of the 8th Seoul International Conference on Generative Grammar: 99–116. 2007 Negation, Universal Quantifier, and A-Movement in nominalization in Japanese. Ms., Hamilton College Kato, Yasuhiko 2000 Interpretive Asymmetries of Negation. In Negation and Polarity: Syntactic and Semantic Perspectives, Laurence R. Horn and Yasuhiko Kato (eds.), 62–87. Oxford University Press. Kawashima, Ruriko 1998 The structure of extended nominal phrases: The scrambling of numerals, approximate numerals, and quantifiers in Japanese. Journal of East Asian Linguistics 7: 1–26. Kishimoto, Hideki 2007 Negative scope and head raising in Japanese. Lingua 117: 247–288. Kitagawa, Yoshihisa and Janet Dean Fodor 2006 Prosodic influence on syntactic judgments. In Gradience in Grammar: Generative Perspectives, Gisbert Fanselow, Cécile Fery, Matthias Schlesewsky, and Ralf Vogel (eds.), 336 –358. Oxford: Oxford University Press.

204 Masaaki Kamiya Kuno, Susumu 1973 The Structure of the Japanese Language. Cambridge, MA: MIT Press. Larson, Richard 1999 Semantics of adjectival modification. Lectures at the LOT Winter School. 1999 Chains of arguments. In Working Minimalism, Samuel D. Epstein and Norbert Hornstein (eds.), 189 –215. Cambridge, MA: MIT Press. 2003 Minimalist Investigations in Linguistic Theory. New York: Routledge. Marantz, Alec 1993 Implications of asymmetries in double object construction, Theoretical Aspect of Bantu Grammar, 113–150. Stanford, CA: CSLI Publications. Matsumoto, Yo 2000a Nihonga-ni-okeru tadoosi/nizyuudoosi-pea-to nitieigo-no siekidkootai (pairs of transitive /ditransitive verbs in Japanese and causative alternations in Japanese and English). In Nitieigo-no zita-no kootai, T. Maruta and K. Suga (eds.), 167–207. Tokyo: Ituzi Syobo. 2000b Osowaru/osieru nadono tadoosi/nizyuutadoosi-pea-no imiteki seisitu (semantic properties of pairs of transitive/ditransitive verbs such as osowaru/osieru ‘get taught/teach). Susumu Yamada, Yasuto Kikuchi, and Yosuke Momiyama (eds.), Nihongo: imi-to-bunpoo-no fuukei, 79–95. Tokyo: Hituzi Syobo. Matsuoka, Mikinari 2003 Two types of ditransitive constructions in Japanese. Journal of East Asian Linguistics 12: 171–203. McGinnis, Martha 1996 Projection and position. Proceedings of ConSole IV, 203–220. Leiden: HIL. 1998 Case and locality in L-syntax: evidence from Georgian. Papers from the UPenn /MIT Roundtable on Argument Structure and Aspect, MIT Working Papers in Linguistics 32: 139–159. 2001 Semantic and morphological restrictions in experiencer predicates. Proceedings of the 200 CLA Annual Conference: 245–256. Miyagawa, Shigeru 1997 Against optional scrambling. Linguistic Inquiry 28: 1–25. 2001 The EPP, scrambling, and wh-in-situ. Ken Hale – a Life in Language, Michael Kenstowicz (ed.), 293–338. Cambridge, MA: MIT Press. Miyagawa, Shigeru and Koji Arikawa 2007 Locality in syntax and floated numeral quantifiers. Linguistic Inquiry 38: 645–670. Miyagawa, Shigeru and Tsujioka Takae 2004 Argument structure and ditransitive verbs in Japanese. Journal of East Asian Linguistics 13: 1–38.

Movement of arguments and negative feature

205

Nevin, Andrew and Panav Anand 2003 Some AGREE(ment) Matters. In Proceedings of the 22nd West Coast Conference on Formal Linguistics, 370 –383. Stanford, CA: CSLI Publications. Rizzi, Luigi 1997 The fine structure of the left periphery. In Elements of Grammar, L. Haegeman (ed.), 281–337. Dordrecht: Kluwer. Roeper, Thomas and Angeliek van Hout 2006 The representation of movement in –ability nominalizations: evidence for covert category movement, edge phenomena, and local LF. Paper presented at Saarbrücken Conference on Nominanlizations and Quantifications. Saito, Mamoru 2004 Japanese scrambling in a comparative perspective. In Peripheries: Syntactic Edges and their Effects, David Adger, Cécile De Cat, and George Tsoulas (eds.), 143 –163. Dordrecht: Kluwer. Shibatani, Masayoshi 1977 Grammatical relations and surface cases. Language 53: 789–809. Szabolcsi, Anna 1994 The noun phrase. In Syntax and Semantics 27: The Syntactic Structure of Hungarian, Ferenc Kiefer and Katalin É. Kiss (eds.), 179 –274. San Diego, CA: Academic Press. Ura, Hiroyuki 1996 Multiple feature checking: A theory of grammatical function splitting. Doctoral dissertation, MIT. Woolford, Ellen 2006 Lexical case, inherent case, and argument structure. Linguistic Inquiry 37: 111–130. Wurmbrand, Susi 2003 A-movement to the point of no return. In Proceedings of the 33rd North East Linguistic Society, 463–474. GLSA. Watanabe, Akira 2004 The genesis of negative concord: Syntax and morphology of negative doubling. Linguistic Inquiry 35: 559 – 612.

Inner aspect and phases Jonathan E. MacDonald

Introduction In this paper I discuss several syntactic properties of inner aspect 1 and what these properties can tell us about derivations in narrow syntax. I argue for the presence of an aspectual projection (AspP) between vP and VP (see also Travis 1991, 2000) that syntactically instantiates an object-to-event mapping via an Agree relation. Furthermore, we see that AspP and everything dominated by AspP defines a syntactic space in which elements must appear in order to contribute to the aspectual interpretation of the predicate (see also MacDonald 2008a, 2008b). Interestingly, however, the time at which these elements are calculated is not until later in the derivation, at the phase. This study contributes to our understanding of narrow syntactic derivations by providing alternative support from inner aspectual phenomena for a cyclic interpretation of the verbal predicate. In section 1, I discuss what aspectual information the durative phrase and the time span adverbial actually provide about a predicate. In section 2, I discuss the distinct aspectual interpretations and distributions of bare plurals and mass nouns; we will see that a proper account of them includes an aspectual projection (AspP) between vP and VP. In section 3, I provide evidence for a domain of aspectual interpretation which defines a space in which elements must appear in order to contribute to the aspectual interpretation of the predicate. In section 4, I discuss properties of stative predicates; I argue that statives lack AspP and that this explains the systematic lack of properties argued in previous sections to be dependent on the presence of AspP. In section 5, we see evidence that the time at which the aspect of the predicate is determined in the narrow syntactic derivation is at Transfer to CI at the vP phase.

1

I use the terms inner aspect and (a)telicity interchangeably here to refer to the same phenomenon.

208 Jonathan E. MacDonald 1. The durative phrase and time span adverbial The durative phrase (e.g. for an hour) and time span adverbial (e.g. in an hour) are widely used in the literature on inner aspect as a test to determine the (a)telicity of a predicate. For this reason it is important to clarify exactly what they tell us about the (a)telicity of a predicate. I take as a telic predicate a predicate that describes an event as having an endpoint. I take as an atelic predicate a predicate that describes an event that lacks an endpoint. Standardly, the aspectual relevance of the durative phrase is only considered under a single-event (or semelfactive) interpretation of the event described by a predicate (Borer 2005; Ramchand 2003; Ritter & Rosen 1998; Schmitt 1996; Vendler 1967; Verkuyl 1993, among others). The following statement from Verkuyl (2000: 170) is indicative of this: “An adverbial such as for an hour applied to [Judith ate a sandwich] excludes the single-event interpretation of [Judith ate a sandwich], the result being some other form of aspectuality (a queer sort of repetition)…” Thus, on a single-event interpretation of a predicate the durative is incompatible with a telic predicate, as in (1a), and compatible with an atelic predicate, as in (1b). (1)

a. John drank a beer # for an hour. b. John drank beer for an hour.

The predicate in (1a) describes an event that has an endpoint; when the beer is finished, the event is finished. The durative is incompatible with this predicate. The predicate in (1b) describes an event that does not have an endpoint.2 The durative is compatible with this predicate. Although this single-event approach to the interpretation of the durative phrase, as just outlined above, is the standard way to approach data of these sort, a large 2

Note that because (1b) is in the past, there is an interpretation that the event ends. This is not to be confused with an event having an endpoint. For note that there is no entailment that the event described in (1b) actually does end, there is only an implicature: (i) John drank beer last night and in fact he is still drinking beer now. The beer drinking event continues to the utterance time; it therefore cannot have an endpoint per se. Note, as expected, that with the telic predicate in (1a) this is not possible: #John drank a beer last night, and in fact he is still drinking it now. Since there is an endpoint to the event, and it takes place in the past, it cannot continue to the utterance time. In MacDonald (2008b) the interaction of inner aspect and tense, and its significance for understanding event structure is discussed.

Inner aspect and phases 209

number of authors have noted that a durative phrase is in fact compatible with a telic predicate under an iterative interpretation (Alsina 1999; Jackendoff 1996; Moltmann 1991; Schmitt 1996; Smith 1991; Tenny 1987; Verkuyl 1972; and Vanden Wyngaerd 2001).3 The data in (2) illustrate this. (2)

a. John spotted a plane for an hour. b. John carried a goat into the barn for an hour.

The interpretation of (2a) is that John spotted a plane over and over for an hour. There was an indefinite number of plan-spotting subevents that took place for an hour. The interpretation of (2b) is that John carried a goat into the barn over and over for an hour. There was an indefinite number of goatinto-the-barn-carrying subevents that took place for an hour. Additionally, observe that, as noted in Moltmann (1991), the object undergoing the action of the verb is required to be the same object in each of the iterated subevents. That is, in (2a), the same plane must be spotted, and in (2b) the same goat must be carried into the barn over and over again. I refer to this type of telic iterative interpretation in which the same object undergoes the action expressed by the verb in each of the iterated subevents as a Sequence of Identical Events (SIE) interpretation.4 Considering that an SIE interpretation is elicited by the durative in these utterances, we can explain straightforwardly why the durative in (1a) is incompatible; once a beer is drunk, under normal pragmatic circumstances, it cannot be drunk again (Jackendoff 1996 makes this point). For the same pragmatic reasons, the durative is incompatible with the sentences in (3) below; the object undergoing the action expressed by the verb cannot undergo the action more than once. (3)

a. John ate a cake # for ten minutes. b. John built a house # for a month.

The durative is only incompatible, then, when an SIE interpretation is pragmatically odd, therefore, I conclude that syntactically the durative is compatible with all aspectual predicate types.5 Moreover, I assume that the durative adjoins to vP (or at an EP above vP (Borer 2005; Travis 2000), 3

4 5

Cross-linguistically, there is variation with respect to whether an iterative interpretation is available in the presence of a durative phrase. See footnote 5. Filip (1999) refers to this as the “resettability of the denoted happening” (114). This is the case for English, but not for Spanish or Russian in which the durative does not elicit an SIE interpretation (MacDonald 2006, 2008b).

210 Jonathan E. MacDonald and modifies the event described by the predicate (Alsina 1999; Larson 2003).6 Observe in (4) that the durative is grammatical in the do so construction, which I take as evidence that it is adjoined higher up in the verb phrase. (4) a. John drank beer for an hour and Frank did so for two. b. Frank played soccer for ten minutes and John did so for twelve. The interpretation elicited by the durative depends on the telicity of the predicate (Tenny 1987; Alsina 1999). An event described by a telic predicate is interpreted as having an end. The durative forces an interpretation in which the event must continue for the time that the durative specifies. Thus, a telic event with a durative is interpreted as reaching an end over and over for a specified amount of time. The result is an indefinite number of telic subevents that repeat for the amount of time specified by the durative.7 Observe that the time span adverbial can target the end of each of these iterated subevents: (5) a. John carried a goat into the barn in 30 seconds (for an hour straight). b. John dragged a log into the shed in 10 seconds (for an hour straight). In a situation in which a goat keeps running out of the barn, the sentence in (5a) is grammatical under the interpretation that each time the goat ran out it took John 30 seconds to carry it back into the barn and this occurred for an hour straight. A similar interpretation is available for (5b). Atelic predicates describe events as not having an end. Observe that the time span adverbial is incompatible with atelic predicates (Borer 2005; Dowty 1979, etc): 8 6

7

8

Following Alsina (1999) and Larson (2003) I assume that the durative phrase modifies event stuff. Dowty (1979) and Moltmann (1991) assume that the durative phrase modifies time stuff. If the durative modifies temporal intervals it is not immediately clear how to account for the SIE interpretation of telic predicates vs. the lack of SIE interpretation of atelic predicates. Nevertheless, whether the durative modifies events or time intervals, it is not immediately clear why there should be an obligatory wide scope effect of singular indefinite NPs as noted in (2). The SIE interpretation results only when the NPs are singular. With a bare plural a different type of iterative interpretation results. See section 2 for a discussion. Strictly speaking, the time span adverbial is not incompatible with atelic predicates, since it can target the beginning of the event. Dowty (1979), Filip (1999),

Inner aspect and phases 211

(6)

a. John drank beer # in ten minutes. b. John carried the goat # in ten minutes.

In the presence of the durative an atelic event is interpreted as continuing essentially uninterrupted (i.e. without reaching an end) for the amount of time specified by the durative. Observe this in the activities in (7).9 (7)

a. John drank beer b. John drove the car

for an hour. for an hour.

I conclude that the durative phrase is syntactically compatible with all aspectual predicate types and depending on the telicity of the predicate it elicits different interpretations. With atelic predicates, an uninterrupted interpretation results, and with telic predicates an SIE interpretation results, in which an indefinite number of iterated subevents continue for the amount of time specified by the durative. Additionally, as is standardly assumed, the time span adverbial targets the end of the event expressing the amount of time that passes before the end takes place. Consequently, the time span adverbial is only compatible with telic predicates. (But see footnote 8.)

2. Bare plurals and mass nouns aspectually In this section, I discuss the distinct aspectual interpretations and distributions of bare plurals (BPs) and mass nouns (MNs). To account for BPs and MNs aspectually, I propose that there is an aspectual projection (AspP) between vP and VP with which they establish distinct relations. The aspectual effect of BPs and MNs on a predicate is standardly taken to be the same; they turn a telic predicate into an atelic predicate (Borer 2005; Dowty 1979; Pustejovsky 1991; Thompson 2006; Verkuyl 1972, etc.). Consider data that seem to support this BP-MN assumption in (8–9). (8)

9

a. John ate a pizza b. John drank a soda

#for an hour. #for an hour.

and Thompson (2006) note this. This fact about the time span adverbial is not immediately relevant to the present discussion, thus when I point out the incompatibility of the time span adverbial, I mean that it cannot target the end of the event. This is also true for stative predicates, which are atelic. See section 4 for a discussion of some of the properties of stative predicates.

212 Jonathan E. MacDonald (9)

a. John ate pizza/pizzas b. John drank soda/sodas

for an hour. for an hour.

The sentences in (8–9) show that the presence of a BP or MN internal argument makes the durative phrase compatible where it once was not. However, the presence of a time span adverbial shows that there is a difference in aspectual interpretation elicited by BPs and MNs: (10) a. John ate pizzas b. John drank sodas (11) a. John ate pizza b. John drank soda

in ten minutes for an hour straight. in three minutes for an hour straight. # in ten minutes for an hour straight. # in three minutes for an hour straight.

With a BP internal argument, as in (10a), the time span adverbial is compatible under an interpretation that for each pizza John ate, he ate it in ten minutes, and he did this for an hour straight.10 This is reminiscent of the SIE interpretation discussed in the previous section, in which the time span adverbial targeted the end of each iterating subevent. Here the time span is playing the same role, thus the predicates in (10) with a BP internal argument are telic. The durative forces the telic event to iterate an indefinite number of times and the BP provides an indefinite number of similar objects (i.e. different pizzas) to undergo the action expressed by the verb in each of the iterated subevents. In the presence of the BP there is a Sequence of Similar Events (SSE) interpretation: in (10a) John ate one pizza, then another pizza and so on. An SSE interpretation is available with the BP sodas in (10b) as well. No such interpretation is available in (11) in the presence of the MN. The MN elicits only an atelic interpretation, as is standardly assumed, and therefore, the time span adverbial is incompatible. Given these facts, I conclude that BPs and MNs have distinct aspectual interpretations. MNs elicit an atelic interpretation of the predicate, and BPs elicit an SSE 10

Thanks to John Bailyn for pointing out to me this interpretation of the BP with a time span adverbial. Note also that Filip (1999: 66), referencing Fillmore and Kay (1991), observes a similar fact about the following datum: Pat built houses (*) in six months. She notes that it is “acceptable if it has a generic (habitual) interpretation…whereby each [building event] is associated with a different house whose construction took six months.” (10) above illustrates that the time span adverbial is compatible with a BP under an non-habitual interpretation as well, resulting in one type of iterative interpretation.

Inner aspect and phases 213

interpretation of the predicate.11 Let us consider the aspectual distributions of MNs and BPs. Consider the ditransitive structures in (12–13). (12) a. John carried goats into the barn in ten minutes (for an hour straight) b. John pushed carts into the store in three minutes (for an hour straight) (13) a. John carried mud into the barn # in ten minutes (for an hour straight) b. John pushed ice into the store # in ten minutes (for an hour straight) In (12a) there is a BP internal argument and as expected an SSE interpretation is available in which one goat after another was carried into the barn in ten minutes for an hour straight. An SSE interpretation is also available in (12b). No such interpretation is available for the MN in (13), and consequently the time span adverbial is out. The MN, as expected, elicits only an atelic interpretation of the predicate. Consider BPs and MNs as the complements of a goal preposition (14–15).12 (14) a. John carried a goat into barns for an hour. b. John pushed a cart into stores for an hour. 11

12

The SSE interpretation elicited by BPs is a multiple events interpretation of a predicate. It is worth noting that not all multiple events interpretations have the same aspectual effect. For example, note that a distributive interpretation does not make the durative phrase compatible: (i) Every boy drank a beer #for an hour. This does not mean that one boy drank a beer, then another boy drank a beer and so on for an hour. The durative phrase can only take scope within the distributive operator resulting in the interpretation that one boy drank a beer for an hour, another boy drank a beer for an hour etc. This does not license the durative like the SSE interpretation does. Consider the multiple events interpretation elicited under an habitual interpretation: (ii) John ate an apple (for breakfast) for a year. Under an habitual interpretation, there is no longer the requirement that the same object undergo the action expressed by the verb. In (ii), it cannot be the same apple in each of the iterated apple-for-breakfast-eating events. Different properties arise under this habitual multiple events interpretation. Note that what I refer to as a goal preposition is a preposition that typically expresses that a goal is reached, e.g. to, into, onto. These prepositions induce a telic interpretation when surfacing with a transitive activity, although, see footnote 16. I do not consider prepositions such as toward or at (on the non-location interpretation of at) a ‘goal’ prepositions per se, because they never entail that a goal is reached. I consider them ‘directional’ prepositions; they express the direction of the event. Note that directional prepositions do not induce telicity when they surface with transitive activities: John pushed the car (toward/at the building) for an hour /#in an hour.

214 Jonathan E. MacDonald (15) a. John carried a goat into water for an hour. b. John pushed a goat onto mud for an hour. The BP in (14a) elicits an SSE interpretation in which a goat was carried into one barn, then another barn and so on for an hour. The MN in (15a) does not elicit an atelic interpretation; the only interpretation available is one in which a goat was carried into water, then back out, and back into it again for an hour. This is an SIE interpretation. Observe, as expected, that the time span adverbial together with the durative is compatible with both sets of sentences from (14 –15) as illustrated in (16 –17). (16) a. John carried a goat into barns in ten minutes (for an hour straight). b. John pushed a cart into stores in ten minutes (for an hour straight). (17) a. John carried a goat into water in ten minutes (for an hour straight). b. John pushed a cart into mud in ten minutes (for an hour straight). These sentences are a bit pragmatically odd because the same goat and cart undergo the action expressed by their respective verbs in each of the iterated events; however, as long as there is a situation in which the same goat or cart is removed from the barn / water or store/mud respectively (perhaps in a competition in which John is timed to see how fast, or how many times, he can repeat these actions), the sentences are pragmatically fine. Once the contexts are set up, these predicates are perfectly compatible with the time span adverbial in conjunction with the durative phrase. Thus, as a complement of a goal preposition MNs do not elicit an atelic interpretation, while BPs do elicit an SSE interpretation. BPs and MNs have distinct aspectual distributions. In order to account for the distinct aspectual interpretations and distributions of BPs and MNs, I claim that there is an aspectual head (AspP) between vP and VP with which BPs and MNs establish distinct relations (18). (18)

…vP ru v AspP ru Asp VP ru V …

Inner aspect and phases 215

I claim that BPs move to Spec, AspP and MNs Agree with Asp. This straightforwardly accounts for their distributions. The most immediate consequence of this proposal is that neither BP nor MN external arguments can affect the aspectual interpretation of the predicate (Tenny 1987), because they are structurally higher than AspP. This expectation is shown to be borne out in (19–20). (19) a. Wildlife ate a sheep in ten minutes / # for ten minutes. b. Livestock destroyed the barn in ten minutes / # for ten minutes. (20) a. Bears ate a sheep in ten minutes (# for an hour straight). b. Animals destroyed the barn in ten minutes (# for an hour straight). Observe in (19) that in the presence of a MN external argument the durative phrase is not compatible with these predicates. Recall that in the presence of a MN internal argument with predicates of this type (see 9) the durative becomes compatible; if the MN had an aspectual effect, we would expect the same results. Moreover, observe that the time span adverbial is compatible, and in (19a) it expresses that ten minutes passed before wildlife ate (and finished) a sheep. Just as a MN external argument does not affect the telicity of the predicate, a BP external argument does not either. The BPs in (20) do not elicit an SSE interpretation of the predicates. Ignoring the time span adverbial for the moment, observe that the durative is simply ungrammatical. Recall from (9) that, like MN internal arguments, in the presence of a BP internal argument, the durative phrase becomes compatible. In the presence of the BP external argument in (20), however, the durative is not compatible. Moreover, note that crucially (20 a) does not mean that one bear ate a sheep, then another bear ate a sheep and so on for an hour straight. Finally, note that the time span adverbial targets the end of the event, but there is only one event end to target; after ten minutes passed the single sheep eating event was over, and there were no more. Thus, MN external arguments do not elicit an atelic interpretation of the predicate, and BP external arguments do not elicit an SSE interpretation of the predicate. Under the hypothesis that MNs Agree with Asp and BPs move into Spec,AspP to elicit an SSE interpretation, these facts follow straightforwardly. BP and MN external arguments cannot establish the necessary kind of relation with AspP because they are structurally higher than AspP. In addition, observe that the only subjects that can contribute to aspectual interpretation are derived subjects. The derived subjects of passives and unaccusatives exemplify this in (21–22) respectively.

216 Jonathan E. MacDonald (21) a. The bottle of beer was drunk # for an hour b. Bottles of beer were drunk in three minutes for an hour straight. c. Beer was drunk # in three minutes/for an hour. (22) a. An animal escaped for an hour. b. Animals escaped for an hour. c. Wildlife escaped # in three minutes/for an hour. In the (a) examples the predicates are telic, as evidenced by the incompatibility of the durative with the passive (see 21a) and by the SIE interpretation elicited by the durative with the unaccusative in (22a). In the presence of a BP (b examples), an SSE interpretation is elicited,13 and in the presence of a MN (c examples) an atelic interpretation results. Let us consider the movement account of BPs in more detail. I assume that on an SSE interpretation BPs behave like existential quantifiers.14 I claim that they must bind a variable inside a syntactic domain of aspectual interpretation defined as everything dominated by AspP in order to elicit the SSE interpretation. For now I simply assume the existence of this domain, but in section 3 I provide motivation for it. Evidence for the movement of BPs comes from what appears to be an island for BP movement in (23). (23) a. # John destroyed a row of houses b. # John ate a box of cookies

in ten minutes in ten seconds

for an hour. for an hour.

The BPs in (23) do not elicit an SSE interpretation. (23a) does not mean that John destroyed one house in ten minutes, then another in ten minutes and so on for an hour. Likewise, (23b) does not mean that John ate one cookie in ten seconds, then another in ten seconds and so on for an hour. 13

14

Note that the time span adverbial is grammatical with the passivized accomplishment in (21b), but not with the achievement in (22 b). Essentially, achievements pattern with activities with respect to the time span adverbial (Dowty 1979) and as such only the beginning of the event can be modified. See MacDonald (2006, 2008b) for an syntactic explanation for why achievements pattern with activities in this way. There does seem to be another reading of BPs available, most likely what Carlson (1977) refers to as a group reading of the BP in which the BP’s denotation is vague. I claim that on an SSE interpretation BPs are existential quantifiers, and consequently do not have the group reading.

Inner aspect and phases 217

The lack of an SSE interpretation can be explained if we assume that the complex NPs in (23) do not allow the BP to move out to Spec, AspP to elicit the SSE interpretation. Let us consider the Agree account of MNs in more detail. I claim that the Agree relation with Asp is the syntactic instantiation of the object-to-event mapping well-known in studies on inner aspect (Verkuyl 1972; Krifka 1989).15 The object-to-event mapping occurs when a property of the internal argument affects the aspectual interpretation of the entire predicate. This is illustrated in (24). (24) a. John drank a pitcher of beer # for ten minutes / in ten minutes. b. John drank beer for ten minutes / # in ten minutes. The noun phrase in (24a) a pitcher of beer has a property that elicits a telic interpretation of the predicate. The noun phrase in (24 b) beer has a property that elicits an atelic interpretation of the predicate. This is the object-toevent mapping. I refer to the property of an internal argument NP that participates in this object-to-event mapping as a [q] feature ([q] for specified quantity of A (Verkuyl 1972)). If the NP that Agrees with and values Asp is [+q] the predicate can be interpreted as telic.16 If the NP that Agrees with 15

16

I do not assume a Krifkian homomorphism. The exact nature of this object-toevent mapping is tangential to the main focus of this paper. See Borer (2005), Filip (1999), Hay, Kennedy and Levin (1999), MacDonald (2006, forthcoming), Tenny (1987), and Verkuyl (1972) for different implementations of the object-toevent mapping. I say can be interpreted as telic because of the existence of transitive activity predicates in which the [+/– q] feature of the internal argument does not affect the aspectual interpretation of the predicate: John pushed the car/stereo equipment for an hour/#in an hour. Although note that when a goal PP is added, these predicates behave exactly like the predicates in (24): John pushed the car into the garage #for an hour/in an hour. John pushed equipment into the garage for an hour/#in an hour. MacDonald (2006) argues that the goal PP here adds a property that the predicates in (24) already possess: an event feature that expresses that the event has an end (cf. Snyder’s (1995) null telic morpheme). Transitive activities project AspP, but the object-to-event mapping is irrelevant without this extra property. Note that the property introduced by the goal PP cannot be the same property of the internal argument entering into the object-to-event mapping; otherwise, the goal PP would not be required for a telic interpretation of the predicate; what is more, we would not expect that a [–q] NP could overrule the aspectual effect of the goal PP. This asymmetry strongly suggests that two independent properties are responsible for the range of aspectual phenomenon in

218 Jonathan E. MacDonald and values Asp is [–q] (e.g. a MN), the predicate will be interpreted as atelic.17 The Agree relation captures a local relation that an NP has with the verb phrase in which the core aspectual interpretation of the predicate is affected. The core aspectual interpretation of a predicate is the basic telic/atelic distinction. This local relation, and its affect on the core interpretation of the predicate, is reminiscent of the effect that the internal argument has on the core lexical meaning of a predicate (Marantz 1984).

3. The domain of aspectual interpretation We have seen that external arguments cannot contribute to aspectual interpretation, while internal arguments can. This distribution hints at a syntactic space below AspP in which elements must appear in order to contribute to the aspectual interpretation of the predicate. In this section, I consider more elements structurally higher and lower than AspP and their effect on the aspectual interpretation of the predicate. Hay, Kennedy and Levin (1999) observe that the CAUSE head that introduces the external argument causer in causative-inchoative alternations does not contribute to the aspectual interpretation of the predicate. As has been observed (Borer 2005; Dowty 1979; Hay, Kennedy and Levin 1999 among others), the predicates in (25) are ambiguous between a telic and an atelic interpretation.18

17

18

English. One property is involved to the object-to-event mapping (and implicates AspP) and the other is directly involved event structure (and implicates event features). See MacDonald (2008b) in which the consequences of these two independent properties of English inner aspect are explored. Borer (2005) puts forth sentences similar to John found water in ten minutes/#for ten minutes. as an example of a predicate that takes a MN (i.e. a [–q]NP) internal argument but yet still surfaces as telic. MacDonald (2006, 2008b) argues that a sentence like this falls into a class of predicates he terms psych-achievements in which the surface subject is an experiencer and derived from a position below AspP yet higher than the surface direct object. This predicts that the argument that Agrees with Asp, and enters into the object-to-event mapping, is the derived subject. Observe that in the presence of a MN subject the durative phrase improves: Wildlife found the body of water for an hour. Also a BP subject elicits an SSE interpretation as well: Animals found the body of water for an hour. Dowty (1979) calls these degree-achievements and claims that there is an ambiguity between an achievement and an activity interpretation. Hay, Kennedy, and Levin (1999) claim that the ambiguity is between and accomplishment and an

Inner aspect and phases 219

(25) a. The soup cooled for an hour/in an hour. b. The kingdom expanded for a week /in a week. Hay, Kennedy and Levin (1999) observe that when the external argument is added, the ambiguity is not affected, as illustrated in (26). (26) a. Neal cooled the soup for an hour/in an hour. b. Neal expanded the kingdom for an hour/in an hour. They rightly conclude that the CAUSE head does not contribute to the aspectual interpretation of the predicate. Consider more evidence for the domain of aspectual interpretation. As we have seen above, transitive activities (27) can be turned into accomplishments by the addition of a goal PP (28); this is a widely observed fact (Borer 2005; Dowty 1979; Pustejovsky 1991; Verkuyl 1972). (27) a. John carried the goat # in an hour/ for an hour. b. John pushed the cart # in an hour/ for an hour. (28) a. John carried the goat into the barn in an hour/for an hour. b. John pushed the cart into the store in an hour/for an hour. The time span adverbial in (27) is out because the predicate is atelic. In the presence of the durative in (27), there is an uninterrupted event interpretation. In (28), on the other hand the time span adverbial is compatible, and the durative elicits an SIE interpretation of the predicate. In the presence of the goal PP, the predicate is interpreted as telic. Consider the lack of aspectual effect of location PPs when added to the sentences from (27): (29) a. John carried the goat (in the barn) #in an hour/ for an hour. b. John pushed the cart (in the store) #in an hour/ for an hour. In contrast to goal PPs, location PPs do not affect the (a)telicity of the predicate; they remain atelic. Moreover, observe a structural difference between location PPs and goal PPs indicated by the ungrammaticality of goal PPs and the grammaticality of the location PPs in the do so construction: activity interpretation. The exact nature of the ambiguity is irrelevant here. What is important is that there is an ambiguity and that the causer does not affect this ambiguity. Nevertheless, note that if Dowty (1979) is correct, the time span adverbial is modifying the beginning of the event here. See footnote 8.

220 Jonathan E. MacDonald (30) a. ?? John carried the goat into the barn and Frank did so into the church. b. ?? John pushed the stool into the store and Frank did so into the church. (31) a. John carried the goat in the barn and Frank did so in the church. b. John pushed the cart in the store and Frank did so in the church. That the goal PPs are ungrammatical suggests that goal PPs are lower in the verb phrase (see also Larson 1988), and that location PPs are grammatical suggests that they are adjoined to vP. A difference in ability to contribute to the telicity of the predicate corresponds to a difference in structural position.19 The CAUSE head that introduces an external argument does not contribute to the aspectual interpretation of the predicate. Location PPs, which are adjoined to vP, do not contribute to the aspectual interpretation of the predicate. BP and MN external arguments do not contribute to the aspectual interpretation of the predicate either. The only elements that contribute to the aspectual interpretation of the predicate are internal arguments (BPs and MNs), complements of goal prepositions (BPs) and goal PPs themselves. These findings are summarized in the tree in (32).

19

Note that being low in the verb phrase is a necessary but not sufficient condition for a PP to contribute to aspectual interpretation. Observe that directional PPs are not perfectly grammatical in the do so construction, suggesting that they are low in the verb phrase: ?John carried the bag toward the store and Frank did so toward the church. Nevertheless, they do not contribute to the aspectual interpretation of the predicate, as noted in footnote 12. See also footnote 16.

Inner aspect and phases 221

(32)

…vP ei vP PP ru 4 Ext. v’ Loc. Outside domain of Arg. ru aspectual interpretation v AspP (Cause) ru Asp VP Domain of aspectual interpretation ru Int. V’ Arg. ru V PP ru P Goal (Goal) Comp.

From the tree in (32), it is apparent that there is a limited syntactic space within which an element must appear in order to contribute to the aspectual interpretation of the predicate. I refer to this space as the domain of aspectual interpretation. It is defined as the aspectual projection AspP and everything AspP dominates. If this domain exists, we can explain straightforwardly why neither external arguments, location PPs, nor CAUSE can contribute to the aspectual interpretation of the predicate. They are outside of the domain of aspectual interpretation. 4. Statives The previous sections have argued for the existence of an aspectual projection between vP and VP that has specific properties associated with it: 1. Agree with Asp syntactically instantiates the object-to-event mapping; 2. Movement to Spec, AspP of a BP elicits an SSE interpretation; and 3. AspP defines a domain of aspectual interpretation. In this section, as further support that these properties are associated with AspP I discuss stative predicates and claim that statives differ minimally from eventives syntactically by lacking this AspP projection in their phrase structure. There are three straightforward predictions to be made from this claim. Statives should 1. Lack the object-to-event mapping, 2. BPs should not elicit an object-toevent mapping, and 3. There should be no domain of aspectual interpretation; i.e. goal PPs should not elicit a telic interpretation when combined with

222 Jonathan E. MacDonald a stative predicate. We will see that these predictions are borne out, and that furthermore, by assuming that statives lack AspP, the present analysis extends straightforwardly to a recent analysis of do so by Hallman (2004). Observe that the nature of the internal arguments in the stative predicates in (33) varies from a [+q]NP to a [–q]NP without affecting the telicity of the predicate. (33) a. John owed a car/money for a week /#in a week. b. John owned a stereo /stereo equipment for a week /#in a week. Statives are atelic irrespective of the [+/–q] nature of the internal argument NP as evidenced by the uninterrupted interpretation of the durative phrase and the incompatibility of the time span adverbial.20 There is no object-toevent mapping with stative predicates. The first prediction is borne out. Now consider the interpretation of BPs with statives in (34). (34) a. John owed cars for a month. b. John owned books for a month. There is an interpretation of these sentences in which John owed one car, then another and then another for a month (34a) and in which John owned one book then another and then another for month (34b) which is arguably an SSE interpretation. Nevertheless, I suggest that this is a result of the availability of a vague denotation of BPs (see footnote 14) and that in fact there is no SSE interpretation. Recall from section 2 above that under an SSE interpretation, BPs are compatible with both a time span adverbial and a durative phrase. Observe that this is not the case with BPs appearing with stative predicates (35). (35) a. John owed cars (#in a day) for a month straight. b. John owned books (#in a day) for a month straight.

20

There may be some interpretation of the time span adverbial with stative predicates. The time span adverbial seems to only be able to express the amount of time before some event occurs that results in the state expressed by the stative. So, John owned a car in a week, if there is an interpretation, only seems to mean that one week passed before something happened resulting in the state John owning the car. This is different from the interpretations elicited with eventive predicates.

Inner aspect and phases 223

There is no interpretation available in which John owed one car in a day, then another in a day and so on for a month straight (35a). Likewise in (35b) there is no comparable interpretation. The second prediction is borne out. Consider now the lack of aspectual effect that the presence of a goal PP has on the predicate headed by a stative (36). (36) a. John owed a car (to the bank) for a week /#in a week. b. John loved the game (to the core) for a year /#in a year. Irrespective of the presence of the goal PP the predicates are atelic. If there is no AspP in statives, then there is no domain of aspectual interpretation either. If there is no domain of aspectual interpretation, then the goal preposition is interpreted outside the domain and cannot contribute to the aspectual interpretation of the predicate; it cannot turn the atelic statives into telic predicates. These three sets of facts can be explained quite straightforwardly if we assume that they result from the lack of AspP in the syntax of stative predicates. This proposal for stative predicates also fits in well with an account of do so from Hallman (2004). Hallman (2004: 304) argues, on independent grounds, that the do of do so is an “…overt reflex of a functional head that licenses the external argument in eventive VPs…”, and so replaces the constituent selected by this do which “…must be marked as [+eventive].” (Hallman 2004: 306). Essentially, eventive predicates (i.e. activities, accomplishments and achievements) can participate in do so constructions, and statives cannot. This is shown in (37a–c) for activities, accomplishments and achievements respectively and in (38) for statives. (37) a. John drove the car and Frank did so too. b. John ate a cake and Frank did so too. c. John caught a raccoon and Frank did so too. (38) a. ?John owns a book and Frank does so too. b. ?John owes money to the bank and Frank does so too. Note that the statives in the do so construction are odd on the reading that is made more salient in the following constructions, as noted in Jackendoff (1983). (39) a. ?? What John did was own a book. b. ?? What John did was owe money to the bank.

224 Jonathan E. MacDonald This fact about eventive predicates vs. stative predicates can be explained quite straightforwardly if we assume that statives do not have an AspP, while eventive predicates do. The do in do so selects for AspP (i.e. the “[+eventive]” VP of Hallman). I take this as independent support for the proposal that eventive predicates project AspP and statives do not. 5. Aspectual calculation at the phase We concluded above that there is an aspectual projection between vP and VP in English eventives, and that this aspectual projection creates a domain of aspectual interpretation. Interestingly, although there is a syntactic space in which elements must be located in order to be able to contribute to the aspectual interpretation of the predicate, we will see in this section that the time at which these elements are calculated is higher up in the syntactic structure, at vP, at Transfer to CI. Consider the aspectual properties of the fake reflexive construction illustrated in (40). (40) a. A bear walked itself into the barn for ten minutes/ in ten minutes. b. Wildlife walked itself into the barn for ten minutes/#in ten minutes. Observe that when the external argument is [+q], as in (40a), there is only an SIE interpretation elicited with the durative, and the time span adverbial can target the end of the event. Observe that when the external argument is [–q], as in (40b), there is no SIE interpretation with the durative, and the time span adverbial cannot target the end of the event. The predicate in (40a) is telic and the predicate in (40 b) is atelic. The data in (40) give the impression that the external argument is participating in the object-to-event mapping, contrary to the conclusion from section 3 that external arguments are outside the domain of aspectual interpretation. If, in fact, the external argument were directly participating in the object-to-event mapping, we might expect it to do so when the direct object is not a fake reflexive. This is not the case, however, as illustrated in (41). (41) a. John walked a letter/Joe into a post office for ten minutes/in ten minutes. b. John walked wildlife into a barn for ten minutes/#in ten minutes. In each of the sentences of (41) the external argument is [+q]. In (41a) the direct object is [+q] and the predicate is telic. In (41b) the direct object is [–q] and the predicate is atelic. The [q] feature of the direct object “over-

Inner aspect and phases 225

rides” the [q] feature of the external argument. If the external argument could directly contribute to the aspectual interpretation of the predicate we would not expect the direct object to affect the aspectual interpretation of the predicate as it does. Consequently, I assume that the external arguments in (40) indirectly participate in the object-to-event mapping through the fake reflexive. That is, the value of the [q] feature of the fake reflexive is determined by the external argument. Once, the value of the [q] feature of the fake reflexive is determined, the reflexive establishes an Agree relation with Asp instantiating the object-to-event mapping. This entails that the telicity of the predicate cannot be calculated minimally until the external argument merges; otherwise the [q] feature of the reflexive would not have a value with which to value Asp. Although AspP defines a domain in which elements must appear to contribute to the aspectual interpretation of the predicate, the aspect of the predicate is not calculated minimally until the external argument merges, until vP. Moreover, consider evidence that suggests that the aspect of the predicate is not calculated any higher than vP either. That is, evidence that suggests that the aspect of a predicate is calculated at the phase, where arguably all syntactic relations take place, as suggested by Chomsky (2007). Recall from above that degree-achievements can appear with both the time span adverbial and the durative phrase as a result of their aspectual ambiguity, illustrated again in (42). (42) a. John cooled the soup b. The king expanded the empire

in ten minutes / for ten minutes. in a month / for a month.

Interestingly, nevertheless, this aspectual ambiguity is lost when the same verbs are found in idioms (43). (43) a. John cooled his jets in ten minutes /#for ten minutes. b. John cooled his heels # in ten minutes / for ten minutes. The idiom in (43a) roughly means to relax; only the time span adverbial is compatible. The idiom in (43b) roughly means to wait; only the durative phrase is compatible. Observe a similar set of facts with the verb read, which has been noted to be aspectually ambiguous as well (Dowty 1979; Borer 2005), as evidenced by the compatibility of both the time span adverbial and durative phrase, illustrated in (44a). (44) a. John read her the newspaper in ten minutes/for ten minutes. b. John read her the riot act # in ten minutes/for ten minutes.

226 Jonathan E. MacDonald Although both the time span adverbial and durative phrase are compatible with the predicate in (44a), when read is used in the idiom in (44b) (which roughly means to give a harsh warning), only the durative phrase is compatible. At the very least the data in (43–44) indicate that there are idioms whose aspect is specified (McGinnis 2002; Glasbey 2007). Independently, Svenonius (2005) observes a relation between phases and idioms; idioms do not seem to cross phase boundaries. If we take this observation together with the observation that the aspectual class of an idiom is specified, we can conclude that aspectual class is determined within the phase. Assuming this to be correct and taking it together with the conclusions from above that the aspect of a predicate cannot be calculated until vP, we can draw a more general conclusion that the aspect of a predicate is calculated at the phase, when Transfer to CI takes place. This conclusion adds support to a phase-based account of syntactic derivations (Chomsky 2007). Moreover, it provides evidence that the phenomenon of inner aspect is constrained by narrow syntax, and, by extension, inner aspect can contribute to our understanding of the nature of narrow syntax.

6. Conclusions I have argued that there is an aspectual projection (AspP) between vP and VP. One important consequence of the presence of this projection is a domain of aspectual interpretation defined as AspP and everything dominated by AspP. Only elements within the domain of aspectual interpretation can contribute to the aspectual interpretation of the predicate. Interestingly, nevertheless, these elements are not calculated until vP, until the phase. That is, the aspect of a predicate is not determined until Transfer to CI. In this way, inner aspect provides alternative support to a phased-based account of narrow syntactic derivation. It provides support that the interpretation of predicates occurs cyclically.

References Alsina, Alex 1999 On the representation of event structure. In Grammatical Semantics: Evidence for Structure in Meaning. Tara Mohanan and Lionel Wee (eds.), 77–122. Stanford, CA: CSLI Publications.

Inner aspect and phases 227 Borer, Hagit 2005 Structuring Sense II: The Normal Course of Events. Oxford: Oxford University Press. Carlson, Greg N. 1977 A unified analysis of the English bare plural. Linguistics and Philosophy 1: 413–457. Chomsky, Noam 2007 Approaching UG from below. In Interfaces + Recursion = Language?: Chomsky’s Minimalism and the View from Syntax-Semantics. Uli Sauerland and Hans-Martin Gartner (ed.), 1–29. Berlin /New York: Mouton de Gruyter. Dowty, David 1979 Word Meaning and Montague Grammar. Dordrecht: Reidel. Filip, Hana 1999 Aspect, Eventuality Types, and Nominal Reference. New York: Garland. Fillmore, Charles J. and Paul Kay 1991 Construction Grammar Textbook. Department of Linguistics, The University of California at Berkeley. Revised versions, 1993 and 1994. Glasbey, Sheila R. 2007 Aspectual composition in idioms. In Recent Advances in the Syntax and Semantics of Tense, Aspect, and Modality [Trends in Linguistics Series] Louis de Saussure, Jaques Moeschler and Genoveva Puskas (eds.) 71–88. Berlin /New York: Mouton de Gruyter. Hallman, Peter 2004 Constituency and agency in VP. In Proceedings of WCCFL 23, Gina Garding and Mimu Tsujimura (eds.), 304 –317. Somerville, MA: Cascadilla Press. Hay, Jennifer, Christopher Kennedy, and Beth Levin 1999 Scalar structure underlies telicity in “Degree Achievements”. In Proceedings of SALT IX, Tanya Mathews and Devon Strolovitch (ed.), 127–144. Ithaca, New York: CLC Publications. Jackendoff, Ray 1983 Semantics and Cognition. Cambridge, MA: MIT Press. 1996 The proper treatment of measuring out, telicity, and perhaps even quantification in English. Natural Language and Linguistic Theory 14 (2): 305–354. Krifka, Manfred 1989 Nominal reference, temporal constitution and quantification in event semantics. In Semantics and Contextual Expression, Renate Bartsch, Johann van Benthem, and Peter van Emde Boas (ed.), 75–115. Dordrecht: Foris Publications.

228 Jonathan E. MacDonald Larson, Richard K. 1988 On the double object construction. Linguistic Inquiry 19 (3): 335–391. 2003 Time and Event Measure. In Philosophical Perspectives 17, Language and Philosophical Linguistics. J. Hawthorne and D. Zimmerman (eds.) 247–258. Oxford: Oxford Univ. Press MacDonald, Jonathan E. 2006 The syntax of inner aspect. Ph.D. dissertation, Stony Brook University, New York. 2008a Domain of aspectual interpretation. Linguistic Inquiry 39 (1): 128–147. 2008b The Syntactic Nature of Inner Aspect: A Minimalist Perspective. [Linguistik Aktuell / Linguistics Today series] Amsterdam: John Benjamins. Marantz, Alec 1984 On the Nature of Grammatical Relations. Cambridge, MA: MIT Press. McGinnis, Martha 2002 On the systematic aspect of idioms. Linguistic Inquiry 33:665–672. Moltmann, Friederike 1991 Measure adverbs. Linguistics and Philosophy 14: 629–660. Pustejovsky, James 1991 The syntax of event structure. Cognition 41: 47– 81. Schmitt, Cristina J. 1996 Aspect and the syntax of noun phrases. Ph.D. dissertation, University of Maryland at College Park. Smith, Carlota S. 1991 The Parameter of Aspect. Dordrecht: Kluwer. Snyder, William 1995 A Neo-Davidsonian approach to resultatives, particles, and datives. In Proceedings of NELS 25, vol. 1: Papers from the Main Session, Jill Beckman (ed.), 457– 471. Amherst: University of Massachusetts, GLSA. Svenonius, Peter 2005 Extending the extension condition to discontinuous idioms. Linguistic Variation Yearbook 5: 227–263. Tenny, Carol Lee 1987 Grammaticalizing aspect and affectedness. Ph.D. dissertation, MIT, Boston, MA. Thompson, Ellen 2006 The structure of bounded events. Linguistic Inquiry 37 (2): 211–228. Travis, Lisa 1991 Inner aspect and the structure of VP. Ms., McGill University, Montreal.

Inner aspect and phases 229 2000

Event Structure in Syntax. In Events as Grammatical Objects: The Converging Perspectives of Lexical Semantics and Syntax, Carol Tenny and James Pustejovsky (eds.), 145–185. Stanford, CA: CSLI. Vanden Wyngaerd, Guido 2001 Measuring Events. Language 77 (1): 61–90. Vendler, Zeno 1967 Verbs and Times. In Linguistics in Philosophy. Zeno Vendler, 97–121. Ithaca, New York: Cornell University Press. Verkuyl, Henk J. 1972 On the Compositional Nature of the Aspects. Dordrecht: Reidel. 1993 A Theory of Aspectuality: The Interaction between Temporal and Atemporal Structure. Cambridge: Cambridge University Press. 2000 Events as Dividuals: Aspectual Composition and Event Semantics. In Speaking of Events James Higginbotham, Fabio Pianesi and Achille C. Varzi (eds.), 169–205. New York /Oxford: Oxford University Press.

How to become passive Berit Gehrke and Nino Grillo

In this paper, we propose that movement of a stative subevent of a structurally complex event to a discourse-related position at the edge of the verb phrase is the fundamental characteristics of passive constructions. This assumption is supported not only by the semantics of passives but also by the fact that it provides a natural account of many of their syntactic properties some of which are left unaccounted for in previous approaches. More generally we give a principled explanation, based on the availability of a consequent state reading, of why some predicates do not form good passives. Psycholinguistic data provide further arguments to support our hypothesis.

1. Introduction Since early works in generative syntax (see Chomsky 1957) passivisation has been analysed as an operation on argument structure. Such analyses single out the most typical property of this construction, namely the inversion in the mapping of argument type and syntactic relation in actives and passives. The internal argument (the understood object) appears in the (syntactic) subject position, whereas the logical subject is demoted and (optionally) surfaces in a by-phrase. In this paper, we will defend a different perspective on passives. It places the complex structure of events at the centre of this transformation and takes it to be responsible for determining not only core properties of passive formation but also its availability in general. This change in perspective allows us to distinguish predicates that can form good passives from those that cannot. In particular, we argue that passivisation is an operation on event structure, more precisely a secondary predication referring to a transition into a consequent (result or inchoative) state. We propose that a semantic requirement, some kind of topicalisation, singles out this consequent state and assigns it a feature that will determine its movement to Spec, VoiceP, projected by by at the edge of the verb phase, which we take to be endowed with discourse-related properties reminiscent of the low focal projection proposed by Belletti (2004a). We support this claim with

232 Berit Gehrke and Nino Grillo evidence from the syntactic and semantic properties of passives, some of which are unaccounted for in previous approaches. The paper is structured as follows. Section 2 discusses previous NPbased approaches to passive formation that treat it as an operation on argument structure and points out some of their disadvantages. Section 3 outlines our own proposal, according to which a stative subevent moves to a position above VP in passive constructions. To support this analysis, empirical data and psycholinguistic data from comprehension patterns in agrammatic Broca’s aphasics are provided in section 4. In section 5, we propose that the position which the stative subevent moves to in passives, is needed independently also for actives, since it allows the creation of a link between the genuinely atemporal event structure and the temporal and discourse domains of the clause. Finally, section 6 concludes. 2. NP movement approaches to passives Strong Crossover effects (1a), the availability of subject-controlled infinitival clauses (1b) and subject-oriented modifiers (1c), depictives (1d), binding (1e,f), and purpose adverbials (1g) provide empirical evidence for the assumption that the external argument is still present in verbal passives.1 (1)

Presence of the external argument in verbal passives a. b. c. d. e.

*Theyi were killed by themselves i. The book was written to collect the money. (Manzini 1980) The book was written deliberately. (Roeper 1983) The book was written drunk. (Baker 1988) Damaging testimony is always given about oneself in secret trials. (Roberts 1987) f. Such privileges should be kept to oneself. (Baker et al. 1989) g. The book was written on purpose. 1

This holds for eventive passives. Stative passives behave differently cross-linguistically. For instance, German does not allow an external argument whereas Greek does (see Kratzer 2000; Anagnostopoulou 2003). Although recent works on the semantic properties of different kinds of participles have payed increasing attention to subtle differences in event structure (see, for example, Embick 2004; Anagnostopoulou 2003; Kratzer 2000; Travis 2005), an alternative analysis of passivisation as an operation on event structure has to our knowledge not yet been proposed. A more careful analysis of the issues discussed in these works has to be left for future research.

How to become passive

233

On the basis of observations like these, Baker et al. (1989), elaborating on Jaeggli (1986b), propose that the passive participle morphology is the external argument in passives. More precisely the -en morpheme of the passive participle is claimed to be a clitic base-generated in the IP head and later on in the derivation lowered down to adjoin to the verbal stem. This operation is argued to ‘absorb’ the case assignment capacity of the verb making it necessary, given Theta Theory and legibility requirements, for the internal argument to move to the subject position. Such an analysis allows us to derive the transformation of passives without having to make use of any special rule. With some fairly simple assumptions and the interaction of syntactic principles and parameters, the transformation could be seen as a necessity. While recognising these merits, we think there are reasons to criticise these assumptions at their base. One issue to be raised, for instance, is that it is not clear on which basis the passive participle morpheme should be distinguished from the active past participle morpheme, which it is homophonous to. Why should only the former but not the latter be analysed as a clitic, whose lowering absorbs the verb’s case assignment capacity? Here we reject this assumption as unmotivated. Moreover, under our analysis there is no need to establish such a distinction. Even more problematic is the assumption that the external theta role is assigned to the passive morpheme, since it poses a problem for the explanation of how the NP in the by-phrase (the logical subject) receives its theta role. As shown by Marantz (1984) and Roberts (1987), it is clear that this NP is not assigned its theta role by the preposition but that it receives it compositionally from the VP. To solve this problem, Jaeggli (1986a) proposes a (fairly complex) mechanism of theta transmission while Baker et al. (1989) argue that the NP in the by-phrase receives its theta role from the clitic via a non-movement chain like the one found in clitic doubling. For reasons of economy of the system we assume that there is a strict mapping between syntax and semantics along the lines of the Uniformity of Theta Assignment Hypothesis (UTAH) (Baker 1988). Given UTAH, both Jaeggli’s (1986a) and Baker et al.’s (1989) solutions are problematic since the external theta-role in these two approaches is assigned in two different fashions in active and passive constructions. On the basis of this argument, Collins (2005) concludes that passive morphology does not absorb the external theta-role or accusative case. Instead, the external theta role is assigned in Spec vP in line with UTAH and accusative case is checked by the by-phrase (by itself being the head of VoiceP) merged directly above vP. An immediate problem that arises under this account is one of locality. Under current assumptions, the movement of the in-

234 Berit Gehrke and Nino Grillo ternal argument over the external argument should raise a minimality effect. Collins provides the following solution to this problem. Smuggling of the VP over the vP makes the internal argument the closest to Spec TP allowing for its promotion to subjecthood without any violation of Relativised Minimality or its derivational counterpart. Smuggling is defined in (2). (2)

Smuggling (Collins 2005) a. Suppose a constituent YP contains XP. Furthermore, XP is inaccessible to Z because of the presence of W, some kind of intervener that blocks any syntactic relation between Z and XP. If YP moves to a position c-commanding W, we say that YP smuggles XP past W. b. Z … [YP XP ] …W… ! # z-----m OK z--------------m x

To derive the right word order, Collins proposes that it is actually the participle that moves to the left of the by-phrase. This movement is argued to be phrasal, dragging along the internal argument. Collins (2005) argues convincingly against alternative analyses in terms of right specifiers or extraposition of the by-phrase to the right.2 Evidence supporting an XP-movement analysis over a head movement analysis comes from (3). (3)

a. The coach summed up the argument. The coach summed the argument up. b. The argument was summed up by the coach. *The argument was summed by the coach up.

Example (3a) shows that in English active verb-particle constructions the particle can appear before or after an internal argument. In the passive, however, only the order where the particle directly follows the participle is grammatical (3b). Hence, the particle has to move along with the participle which can be taken as evidence that more than just the participle head moves in passive constructions. In (4) and (5) we provide examples that make a similar point with respect to different kinds of goal phrases. 2

See Collins (2005) for details on this point and on the formation of the participle. Collins assumes that the participle morpheme -en heads a PartP and that the head of V raises and adjoins to Parto.

How to become passive

(4)

a. Jutta was spoken to by Eric. b. *Jutta was spoken by Eric to.

(5)

a. Tom zipped the sleeping bag all the way up to the top. b. *The sleeping bag was zipped by Tom all the way up to the top.

235

Even though Collins’ account seemingly solves the locality issue, it raises several new ones, the most severe of which is that it poses serious lookahead problems. The computational system is supposed to be able to apply an operation with an unclear status in order for the internal argument to move to the subject position without violating minimality. This leads to several questions such as those in (6). (6)

Questions raised by a smuggling approach to passive formation a. What is the status of smuggling in the theory? b. What are the limits of smuggling and more generally of look-ahead computations? Doesn’t smuggling massively over-generate? Can it be used to avoid minimality effects with other potential interveners, for example in A’-movement? c. How do we explain sentences like (7), where passivisation applies independently of movement of the internal argument to the subject position?

(7)

There was a Swabian killed.

Given these questions and problems and the overall ad-hoc flavour of this solution, we will leave Collins’ proposal aside and develop a different approach by shifting the perspective from argument structure to event structure. This shift will provide new predictions for and insights into the nature of passivisation and will solve the locality issue in a natural and less stipulative way. Unlike Collins, we will dissociate the movement of part of a complex event structure, which is taken to be the essential characteristics of passivisation, from the movement of some argument DP to Spec TP to satisfy EPP. By doing so, we will provide a semantically motivated trigger for the movement involved in passive formation and will also be able to account for cases like (7). In essence, then, our proposal will not be a smuggling approach, and the questions in (6) do not arise. It is important to stress that the distance between Collins’ analysis and the one to be proposed here lies also in the fun-

236 Berit Gehrke and Nino Grillo damentally traditional view that he pursues. In his explanation passivisation is still treated as an operation on argument structure and the whole mechanism of smuggling is motivated by the necessity to bring the internal argument closer to the subject position than the external argument. The novelty of the present approach is that we put event structure at the core of passivisation. Nevertheless, we will see that the picture that emerges from the analysis we propose is similar in many, especially technical points to that of Collins (2005). We will therefore refer to important observations of his work and at times integrate them in our proposal.

3. Shifting the perspective on passives Grounding our analysis on the semantic and syntactic properties of passive sentences we propose that the promotion of a consequent state subevent of a complex event to a position above VP is a fundamental ingredient of the passive. In the spirit of Travis (2000, and subsequent work), we employ a VP shell account for the syntactic representation of event structure. In Travis’ model, V2 introduces the theme argument (DPint) as well as the endpoint of the event, whereas V1 corresponds to the causing sub-event and introduces the external argument (DPext). Thus, a consequent state is structurally represented as a lower VP shell with the VP-internal argument DP in its specifier. In section 5, we will propose that the position the lower VP moves to is independently needed for actives as well, since it forms a basis for the event time that subsequently serves as the internal argument of Asp (in the sense of Demirdache and Uribe-Etxebarria 2000). In passives, then, the event time falls within this stative subevent. The syntactic tree in (8) exemplifies the proposal.3

3

XP in this structure represents further elements like, for instance, PPs or APs, which can be complements to V2.

How to become passive

(8)

237

AspP ei ASST-T Asp' ei Asp VoiceP ei EVT-T Voice' ei Voice VP1 ei DPext V1' ei V1 VP2 ei DPint V2' ei V2 (XP)

We propose that a semantic requirement, some kind of topicalisation, singles out this consequent state and assigns it a feature that will determine the movement of the lower VP to a discourse-related projection at the edge of the VP phase, represented as VoiceP. Voice is responsible for grounding the event time in a particular way. In the case of passives the event time is anchored within the consequent state subevent. The feature that triggers movement to VoiceP has two properties, a discourse-related and a quantificational one. The discourse-related part chooses the element of the complex event that needs to be singled out whereas the quantificational part makes it readable to the next phase. Thus, the main job of this feature is to single out an element of the atemporal event structure associated with the VP phase and to enrich its semantics by introducing temporality, thereby making it available to the next phase, the temporal domain (and ultimately the discourse domain) of the clause. Contrary to Collins, this operation is completely independent of the promotion of the internal argument to subject position. This is supported by the fact that the internal argument does not necessarily land in Spec TP in passives (see (7) and section 4.2). Hence, we take movement of VP2 to be the only necessary condition to define passivisation. Let us make precise what it means for an event to be structurally complex, particularly focusing on the formation of consequent states. We are following Dowty (1979) in assuming that verbal predicates, associated with

238 Berit Gehrke and Nino Grillo Vendler’s (1967) event types, can be decomposed into particular atomic predicates CAUSE, DO and BECOME and combinations of these. Here we will concentrate on event types that involve the BECOME predicate, namely accomplishments (9) and achievements (10), and we will abstract away from the other predicates DO and CAUSE (see Dowty 1979, for discussion and formal definitions).4 (9)

Accomplishments (Dowty 1979: 124) DO (α1, [πn (α1,

…, αn)])] CAUSE [ BECOME ρm (β1, … , βn)]]]. e.g. John broke the window. (10) Achievements (Dowty 1979: 124) BECOME [πn (α1, … , αn)]. e.g. John discovered the solution. The definition of the predicate BECOME relative to an interval I is given in (11). (11) Original definition of BECOME (Dowty 1979: 140) [BECOME φ] is true at I iff there is an interval J containing the initial bound of I such that ¬φ is true at J and there is an interval K containing the final bound of I such that φ is true at K. With this definition, then, event types containing become such as accomplishments and achievements are associated with definite change of state predicates. The state denoted by πn in (9) and (10) comes into existence, since it is false at interval J and true at some later interval K. Dowty’s (1979) idea of decomposing predicates has been reformulated in event semantic terms, where an event (the macroevent) is viewed as potentially structurally complex and decomposable into particular subevents. Such subevents, in turn, are associated with CAUSE, DO or BECOME predications, or related notions such as preparatory phase, process, transition, culmination, consequent or result state and the like (Moens and Steedman 1988; Parsons 1990; Pustejovsky 1991; von Stechow 1995; Rapp and von Stechow 1999; Higginbotham 2000; Kratzer 2000, 2005; Rothstein 2004; Ramchand 2004; Beck 2005, among many others). 4

In the following, αi and βi stand for arbitrary individual terms, πn and ρn stand for arbitrary n-place (stative) predicates, and φ and ψ are arbitrary formulas, either atomic or complex.

How to become passive

239

What event structure approaches have in common is that they capture the semantics of predicates involving change, a (durative or instantaneous) change from ¬φ to φ, by assuming an ontology which contains a transition into a state. Let us call this state a consequent state, using the terminology of Moens and Steedman (1988). This state is directly related to (an atemporal version of) Dowty’s (1979) BECOME-operator. In (12) we provide McIntyre’s (2006) (informal) reformulation in event semantics terms. (12) BECOME in event semantic terms (McIntyre 2006) λeλsλP BECOME [P(s)](e) ‘e is an event of coming-into-existence of a situation s with property P, where ‘coming-into-existence’ is a conceptualised entry/arrival of s in the domain of existing things’ In the following, we will focus on accomplishments in particular, since these are most relevant for the issues discussed in this paper. According to Dowty, all accomplishments have the logical structure [φ cause ψ] where φ in most cases contains an activity predicate and ψ is a BECOME sentence. The underlying activity (φ) can be quite general, which is the case with lexical accomplishments such as kill (13). (13) John kills Bill.

(Dowty 1979: 91)

[[John does something] CAUSE [BECOME ¬ [Bill is alive]]] It can also be quite specific, especially with accomplishments that are syntactically created. Syntactically created accomplishments are generally assumed to involve some kind of secondary resultative predication. Among the constructions that have been analysed in terms of secondary resultative predication are those involving verbal prefixes (e.g. German, Slavic) or verbal particles in, for instance, English (14a), resultative adjectives (14b), and (some) directional PPs (14c).5 (14) Secondary (resultative) predicates a. Boban took off his hat. b. Mai hammered the metal flat. c. Kriszta and Maaike danced into the house. 5

There is some debate in the literature as to whether directional PPs should be treated on a par with other secondary resultative predicates or rather as mere (e.g. Davidsonian) event modifiers (see Gehrke, 2008b, for discussion).

240 Berit Gehrke and Nino Grillo Note that we are not necessarily concerned here with telic events but rather with complex event structures that rely on some form of the BECOME operator.6 For example, Dowty (1979) provides the following semantic representation of a syntactically created accomplishment: (15) He sweeps the floor clean.

(Dowty 1979: 93)

[[He sweeps the floor] CAUSE [ BECOME [the floor is clean]]] Here, the floor undergoes two predications, first the verbal predication (the floor is being swept) and the consequent state predication (the floor is clean). Both predications are connected by the predicates CAUSE and BECOME, namely the sweeping of the floor causes the floor to become clean. With resultative secondary predicates, the main (verbal) and the secondary predicate form a complex predicate (at least from a semantic point of view). It is generally the internal argument of the verbal predicate that undergoes the secondary predication.7 There are different approaches in the literature as to how the two predicates in syntactically created accomplishments are combined semantically to form one complex predicate and refer to a single event. Something extra is needed to make this link such as von Stechow’s (1995) Principle (R) adding a cause become component that glues the two predications together, Doetjes’ (1997) inchoative auxiliary mediating between the two predications, Rothstein’s (2004) accomplishment type shifting operation, or Snyder’s (2005) Rule C. For the purpose of this paper, it is not relevant which one we choose so we will leave it open (see also Gehrke, 2008b, for discussion). Our proposal for passivisation, then, is that this formation necessarily involves the zooming in on a consequent state subevent, which is the result of a transition associated with the BECOME component. Syntactically, the 6

7

In general, we follow Rothstein (2004) and others in separating a theory of event types and the creation of such types at the VP level, from the effect that the quantificational properties of the internal argument DP can have on the interpretation of the VP as bounded or unbounded. Secondary predication has been analysed as small clause structures (e.g. Stowell 1981; Kayne 1985; den Dikken 1995; von Stechow 1995; Doetjes 1997) or in terms of complex predicate formation in syntax (e.g. Williams 1980; Baker 1988; Neeleman 1994; Zeller 2001). This paper will not commit itself to either position but rather tries to remain agnostic in this respect by employing a VP shell account.

How to become passive

241

BECOME component is associated with the lower VP shell (VP2), whereas the CAUSE component correlates with VP1. The prediction of our proposal is that only with event structures that contain a BECOME component, repre-

sented by means of a lower VP shell, passivisation is possible. Supporting evidence that this is the case will be provided in section 4. There are several important issues we will not address in the present paper, of which the exact status of the by-phrase and an account of accusative case ‘absorption’ are surely the most important ones. Nevertheless, the similarities between the syntactic configurations that surface in Collins’ (2005) analysis and our own, though in many respects simply epiphenomenal, allow us to make reference to that work and provide a formal characterisation of many aspects of passivisation left aside here. Collins’ account of the status of short passives and the PRO nature of the external argument in these constructions as well as his formal explanation of case absorption or the status of the by-phrase can all be assumed here for the time being, even if some of them might need slight adaptation. For example, throughout this paper we follow Collins’ analysis of by as the head of VoiceP which is responsible for assigning case to the external argument. The following section provides empirical evidence for our proposal.

4.

Empirical evidence

Having outlined the main ideas underlying our approach to passivisation, we now move on to the empirical evidence in favour of the complex structure involving BECOME and the interpretation associated with consequent states.

4.1. Consequent states in passive constructions An important prediction that our analysis of passive formation in terms of the obligatory movement of a consequent state subevent (the lower VP shell) makes is that only those predicates which involve a BECOME component (verbal or complex) should allow passivisation. Hence, all predicates with some kind of resultative semantics should allow passivisation. This is straightforward in examples involving secondary resultative predication, as those provided by Collins (2005) and others (16), (17).

242 Berit Gehrke and Nino Grillo (16) a. The argument was summed up by the coach. (= (3), Collins 2005) *The argument was summed by the coach up. b. Jutta was spoken to by Eric. *Jutta was spoken by Eric to. (17) a. The table was wiped clean by John. ??The table was wiped by John clean. b. The metal was hammered flat by John. ??The metal was hammered by John flat.

(= (4)) (from Postal 2004)

Under our account, the lower VP shell in these examples, which represents the consequent state, moves obligatorily taking along the particle in verbparticle constructions (16a), to-phrases in ditransitives (16b), and resultatives (17). These elements are part of the lower VP shell because they relate to the BECOME event (either they are in complement to V2 or they are directly inserted in V2), and there is no way of moving VP2 over VP1 without moving the particle, the to-phrase, or the resultative along with it. This is why the by-phrase cannot intervene between the verb and the secondary predicate in the passive sentences. Postal (2004), among others, shows that not all transitive verbs can form passives. The analysis proposed allows us to make important predictions with respect to which transitive predicates can form passives and which ones cannot. Under our proposal, it is predicted that unless a secondary predicate is supplied syntactically, only those transitive verbs that are associated with an accomplishment or achievement event structure (involving BECOME) should be able to form passives. This prediction is borne out when we compare the examples in (18), which are generally assumed to be accomplishment structures, with those in (19).8 (18) Transitive transition structures allow passives a. The lion killed the antilope.
 The antilope was killed (by the lion). b. He put the card on the table.
 The card was put on the table (by him).

8

See Postal (2004) for more equivalent examples.

How to become passive

243

(19) Transitive verbs associated with simple event structures do not allow passives a. This laptop weighed two kilos.
 *Two kilos were weighed (by this laptop). b. This chair cost 50 euro.
 *50 euro were costed (by this chair). It is commonly assumed that transitive verbs like the ones in (19) are not associated with an event structure containing a transition into a state. Hence, our approach correctly predicts that passive formation is not possible with these verbs.

4.2. Passives of existential constructions Recall from example (7) in section 2 that the behaviour of passives in the presence of there-expletives poses a problem for Collins’ (2005) smuggling approach. The data, however, support our proposal, which shifts the perspective from argument structure to event structure (20).9 (20) a. There was a Swabian killed. b. *There was killed a Swabian. Given a traditional analysis of passives, it is not clear why the internal argument has to appear in preverbal position, whereas the postverbal position is ungrammatical. If we assume instead that in passive constructions VP2 moves to some position above VP1, dragging along the internal argument, the word order in there-passives is accounted for as follows. Our proposal assumes that regular passives involve two independent operations. First, the lower VP shell moves to Spec VoiceP to form a basis for the event time, and second, a DP moves to Spec TP to satisfy the EPP. In there-constructions (both active and passive), this second movement does not take place but an expletive is inserted instead to satisfy the EPP. However, the first movement of VP2 still takes place in passive sentences since it is completely independent of the DP-movement to Spec TP. Under standard assumptions, EPP requirements on T can be satisfied in two ways: movement of the closest argument to Spec TP or expletive insertion. We propose that the same options are available in passives. If EPP is 9

Thanks to Jutta Hartmann for pointing these facts out to us.

244 Berit Gehrke and Nino Grillo satisfied via movement, the closest argument (the internal argument given prior movement of VP2) will be attracted. If EPP is satisfied by an expletive we obtain (7). As noted in the beginning, it is not clear how Collins (2005) can account for these data since under his approach the participle moves in order to smuggle the internal argument to get it closer to Spec DP than the external argument. So in that sense, his approach is not much different from traditional accounts where the perspective lies on the DP which in the end has to move to Spec TP. However, if there is no subsequent movement of a DP to Spec TP the movement of the participle should also not take place in Collins’ approach. This would predict the word order with a by-phrase in (21). (21) *There was by the police a Swabian killed. We take the ungrammaticality of this example as additional evidence for our account where movement of VP2 in passive constructions takes place for reasons independent of the need to satisfy the EPP. There is cross-linguistic variation with respect to the word order in expletive passives. In French, for example, the internal argument has to follow the participle (22). (22) a. Il a été tué un Souabe. it has been killed a Swabian ‘There was a Swabian killed.’ b. *Il a été un Souabe tué. it has been a Swabian killed The present analysis can be extended to cover these facts by the natural assumption that also in this case, as is standard for verb movement in active sentences, French participles raise higher than English ones. In fact, the French passive participle possibly raises higher than the active participle. The past participle used in French perfect tenses can be shown to appear quite low in the inflectional field, which contrasts directly with Italian, for instance (23).

How to become passive

245

(23) French active participle is lower than Italian one a. Tanya a (tout) compris (*tout). Tanya has (all) understood (*all) ‘Tanya has understood everything.’ b. Andrea ha (*tutto) capito (tutto). Andrea has (*all) understood (all) ‘Andrea has understood everything.’

FRENCH

ITALIAN

In French, the past participle obligatorily follows floated quantifiers and other adverbial material (23a), whereas in Italian it is the other way around (23b). In addition, past participle agreement is not required with a clitic left dislocated object (24a), whereas in passives, the agreement is obligatory (24b) (examples from Guasti and Rizzi 2002). (24) French active participle is lower than passive participle a. La voiture, il l’a mise /mis dans le garage. the.FEM car, he it.FEM-has put-FEM /-MASC in the garage ‘What concerns the car, he put it into the garage.’ b. La voiture a été mise / *mis dans le garage. the.FEM car has been put-FEM/ put-*MASC in the garage ‘The car was put into the garage.’ We take the obligatory presence of passive participle agreement in French to explain the word order difference between English and French. In French, the passive participle has to raise higher than in English, at least as high as AgrOP to check agreement and this explains the word order difference. Independently of the position of the participle, the crucial facts concerning the relative position of by-phrases, which cannot intervene between the verb and the DP, still remain the same (25). (25) a. *Il a été tué par la police un Souabe. it has been killed by the police a Swabian b. Il a été tué un Souabe par la police. it has been killed a Swabian by the police ‘There was a Swabian killed by the police.’

246 Berit Gehrke and Nino Grillo 4.3. Floating quantifiers A long-lasting problem for the analysis of floating quantifiers (at least since Sportiche 1988) comes from the observation that these are banned from the post-verbal position in passives (26), whereas both word orders are grammatical in the active counterparts (27). (26) a. The boys were both given a good talking to. b. *The boys were given both a good talking to. (27) a. Ad gave the boys both a good talking to. b. Ad gave both the boys a good talking to. This behaviour of floating Qs is unexpected under previous approaches to passives, since the quantifier should be able to be stranded in postverbal position where it is originally merged. However, if there is additional movement of the consequent state (VP2), independent of any DP-movement to satisfy the EPP, the word order is straightforwardly accounted for as follows. The floating Q moves together with the internal argument in Spec VP2 and remains stranded after the movement of the internal argument to Spec TP.10 4.4. Passives involving states A potential problem for our account is that a number of states can still form passives. These are generally considered not to involve BECOME, since they consist of only one sub-event, namely a state (or alternatively they have been considered not to be eventive at all). This is the case with stative predicates like know, surround, own, believe, among others (28). (28) States that passivise a. The house is owned / surrounded by the army. b. The answer / myth is known / believed by the pupils. c. Adriana is loved (by Luigi). The availability of passivisation, however, is not a common property of stative predicates in general. The difference between predicates that are otherwise similar in their stative characteristics opens up a possible solution 10

Thanks to Ad Neeleman for pointing out these facts to us.

How to become passive

247

to this problem. Belletti and Rizzi (1988), for example, show that there are three kinds of psych-verbs, which are generally considered stative verbs, namely the fear-type (temere), the worry-type (preoccupare) and the appeal-type (piacere). Only fear-verbs can undergo verbal passivisation (29a). Worry-verbs, on the other hand, can only derive adjectival passives, whereas appeal-verbs cannot form passives at all (29b). (29) Different kinds of psych-verbs (examples from Reinhart 2002) a. The news surprised / excited Max. Max was surprised /excited (by the news). b. The solution appeals to me / escapes me. *I am appealed / escaped (by the solution). There is a clear intuitive difference between the class of stative predicates above and at least one class of psych-verbs (the piacere/appeal class) in the sense that only the former can have an inchoative meaning of the state denoted by the verb. We can say, for instance, Max got to know the answer / into a knowing state, Max got to own the house / into an owning state. Appeal-verbs do not lend themselves to this inchoative reading and we cannot have examples like *I got to escape the solution / into an escaping state. To put it in different terms, know-verbs allow a reading where the state denoted by the verb is re-interpreted as a consequent state, a state having come into existence, and this state is predicated over the internal argument. We propose that this reading is derived by means of some kind of type shift from states to achievements, which adds a BECOME predication with the interpretation defined in (12). As a consequence, there is a secondary predication over the internal argument, represented by a lower VP shell with the internal argument in its specifier. Adding BECOME to an otherwise structurally simple state allows this predicate to passivise. A semantic restriction on the possibility to add BECOME is that the state has to be interpretable as a consequent state, otherwise the addition of become is not possible. Passive formation, then, is possible if it involves promoting this kind of consequent state. Appeal-type verbs, on the other hand, cannot involve such a secondary predication by BECOME and as a result cannot form passives. Similar cases of type shift operations are discussed in the literature. For example, de Swart (1998) argues for a coercion operator in e.g. French, in case stative predicates are combined with the passé simple, whose input cannot be a state or a process (activity) but has to be an ‘event’ (subsuming what we have labeled accomplishment or achievement) by definition. A semantic effect of this type shift is that the state is interpreted as an in-

248 Berit Gehrke and Nino Grillo choative state, i.e. a transition into a state. Following upon de Swart’s analysis of this type shift as involving covert aspectual operators, Travis (in preparation) proposes to represent these operators in the syntax by means of a VP shell account. Furthermore, it is often assumed that accomplishments differ from achievements in their ability to derive progressives (Vendler 1967, among many others). The fact that there are nevertheless many cases where achievements actually form progressives (30), in turn, has been taken as an argument for collapsing both accomplishments and achievements into one group (e.g. Bach 1981; Verkuyl 1993). (30) Progressive achievements Jonathan was reaching the summit. Rothstein (2004), however, makes a point for maintaining the distinction between achievements (which merely contain BECOME) and accomplishments (which contain DO, CAUSE and BECOME under Dowty’s system) by discussing semantic pecularities that clearly distinguish progressive achievements from progressive accomplishments. As a result, she assumes that in order for an achievement to combine with the progressive, some kind of type shift has to take place that shifts the achievement into an accomplishment by adding an activity (associated with Dowty’s DO predicate), which is interpreted as an activity that can be the preparatory process of e.g. reaching a summit. Again, this addition is semantically constrained in the sense that it has to be possible to construe some appropriate activity. In this spirit, we assume that there is a more general mechanism to add subevents to the event structure in order to enable certain operations such as passives (adding BECOME to states) or progressives (adding DO to achievements), and that this mechanism is subject to strict (semantic) constraints. The precise implementation of this idea has to be left for future research.

4.5. Passivisation in the DP domain Additional evidence that a more complex structure (i.e. one containing a BECOME predicate) needs to be provided in the syntax for stative predicates comes from the restricted availability of passivisation in the DP domain. Relevant examples are given in (31).

How to become passive

249

(31) Restricted passivisation in the DP domain a. Giorgos feared Roberta. b. Roberta was feared by Giorgos. c. Giorgos’ fear of Roberta d. *Roberta’s fear by Giorgos e. the enemy’s destruction of the city f. the city’s destruction by the enemy The crucial difference between the nominals in (31d) and (31f) is that only the latter is lexically specified with a complex event semantics which crucially involves some sort of BECOME predication (similar to kill). For pure stative elements like fear, however, the only possible way to introduce this semantics is the formation of a complex predicate in the syntactic structure via merge of additional verbal structure with the nominal head. However, we know independently that nominals do not allow this option. As (32) shows, nominals are banned from taking SC as complements, a possibility which is nevertheless granted to verbal heads.11 (32) a. I consider [Sharon a good writer]. b. *My consideration [Sharon a good writer]. This asymmetry between Ns and Vs is only one example from a set of constructions and has been associated with verbs having a tighter (selectional) link with their complements. Nominal heads are not able to assign Case to a DP in the Spec of their complement, nor to incorporate the complement’s head, as illustrated by (33). (33) a. I consider [Sharon to be a good writer]. b. *My consideration (of) [Sharon to be a good writer]. (33b) shows that a nominal head is not able to assign case to a DP in the Spec of its complement. The inability of nominals to combine with particles (34) supports the assumption that a complex nominal with the same properties of complex verbs is not a grammatical option in English. (34) a. Hamida gave the book away. b. Hamida gave away the book. c. *the gift away of the book d. *the gift of the book away 11

On this point see Haegeman and Guéron (1999).

250 Berit Gehrke and Nino Grillo Given the impossibility for nominals to combine with a SC, which is in many ways similar to a lower VP shell, we take it that the only option available for them to undergo nominal passivisation is to be endowed with a complex eventive semantics already in the lexicon. The need for stative nominals to combine with a BECOME operator can only be satisfied via a categorical change, their only option is to be projected as verbal elements. This solution is maximally simple in supporting our assumptions about stative predicates and in explaining the asymmetric behaviour of nominals with respect to passivisation.

4.6. On PRO as the external argument of short passives Drawing an interesting parallel between the ability to assign null case of the complementiser for and by in passives, Collins (2005) proposes that PRO is the external argument in short passives. Baker et al. (1989) already attempt to define the external argument in short passives as being of the same kind as PRO. However, commenting on data like (35), they notice that while PRO can relate to the 1st person in infinitives, silent arguments in passives cannot. (35) a. PRO to shave ourselves is fun. b. *Love letters were written to ourselves. This fact is surprising considering that PRO and the silent argument of passives seem to share all other properties. For example, their nature of arbitrary pronouns explains their inability to bind a non-arbitrary pronominal. Baker et al. (1989) do not have much to say about this fact and are obliged to stipulate that the two elements are different despite most evidence. We believe the present account can hint at a possible explanation for this problem. We could speculate that at least part of the inflectional field is present in infinitive clauses, namely some of the agreement features, and that furthermore the local relation between PRO and TP (or AgrS for that matter) can provide PRO with the relevant person feature in active sentences (say, via Agree for concreteness). In passives, however, given VP2 movement to Voice, this option is not available since all the features in TP are checked by the internal argument that intervenes between PRO in VP1 (vP in Baker et al.) and the functional field. In sum, there is ample empirical data to support an approach to passive formation that involves promoting a consequent state subevent. In the next

How to become passive

251

section, we discuss evidence in favour of our proposal from the comprehension patterns in agrammatic Broca’s aphasics.

4.7. Agrammatic Broca’s aphasia In this section, we provide additional psycholinguistic evidence in support of our analysis of passives. Agrammatic Broca’s aphasics have been reported to have particular troubles in comprehending sentences in which the ‘canonical order’ of thematic roles has been inverted. Deficitarian comprehension of passive sentences in this population is a prototypical case at hand which, among other syntactic structures, has been widely studied and important discussion has been raised on this topic at the intersection between neurolinguistic data and theoretical syntax (see, for instance, Grodzinsky 1990: ch. 4). We believe that our account of passive formation combined with the approach to this deficitarian comprehension pattern proposed in Grillo (2005, 2008, forthc.) opens new intriguing perspectives on the interpretation of these facts. 4.7.1. Comprehension patterns and Relativised Minimality A turning point in the study of Broca’s aphasia came about with the discovery in the mid seventies that, contrary to what was generally assumed at the time, their linguistic deficit is not restricted to production (e.g. the prototypical telegraphic/elliptic speech characterised by omission and substitution of functional material) but extends to comprehension. In particular they have a deficit in the ability to comprehend movement derived sentences displaying a non-canonical order of thematic roles (Caramazza and Zurif 1976) (for an overview of these and other issues connected to agrammatic aphasia see Zurif and Swinney 1994; Kolk 1998; Grodzinsky 2000; Avrutin 2001, among many others). A summary of some of the crucial comprehension asymmetries is given in Table 1.12 12

Table 1 should be read as a comparison between two structures at a time. In each row a comparison is made between two structures, the crucial difference between which is that only the structures in the right column involve movement of a DP over an intervening DP. Note that the hypothesis developed in Grillo (2005, forthcoming) is based on the assumption that a processing problem is at the base of the comprehension deficit described here (see Avrutin 2006, among others,

252 Berit Gehrke and Nino Grillo Table 1. Agrammatic aphasia comprehension patterns (Grodzinsky 2000) Above Chance Performance

Chance Performance

Subject Relatives Subject Clefts Adjectival Passives Unaccusatives SVO Hebrew Actives Unscrambled Object

Object Relatives Object Clefts Verbal Passives Verbal Passives OSV/OVS Hebrew Actives Scrambled Object

Grillo (2005) argues that this selective comprehension problem is the consequence of minimality effects that arise when a dependency has to be built over an intervening element which shares part of its featural make-up with the probe. Grillo takes some recent developments of the Relativised Minimality (RM) approach to locality as a starting point (Rizzi 1990, 2004b; Starke 2001). A formal definition of RM is given in (36) (from Rizzi 2004a). (36) MINIMAL CONFIGURATION: … X … Z …Y…Y is in a Minimal Configuration (MC) with X iff there is no Z such that i. Z is of the same structural type as X, and ii. Z intervenes between X and Y. Intervention in (36ii), is defined in terms of c-command. The exact characterisation of ‘sameness’ in (36i) has been the object of debate and modifications since the first discussion in Rizzi (1990). The definition of ‘same structural type’ we adopt is taken from Rizzi (2004a). Rizzi’s definition of ‘sameness’ is based on the empirical observation that only features belonging to the same class, in the sense defined below, block each other. Different classes of features are identified on the basis on this issue). This assumption allows us to deal with variation in the degree of severity of the deficit which can significantly modify the performance of an agrammatic patient with respect to the pattern indicated in the table. Nevertheless, notice that, with the important exception of Druks and Marshall (1995) (on which, however, see Zurif 1996), the inverse pattern was never found. Variation could be more complex to handle for ‘knowledge based approaches’, i.e. approaches that take part of the linguistic knowledge to be lost in agrammatism. In depth discussion of this and other important issues on agrammatic comprehension in relation to the present hypothesis can be found in Grillo (2008).

How to become passive

253

of their intrinsic properties and their distribution by taking as a point of departure the hierarchy of positions in the syntactic tree as studied in detail in recent cartographic studies (see Rizzi 1997, 2004b; Cinque 1999, 2002; Belletti 2004b, among others). Each of these positions, in fact, can be defined by its particular set of morphosyntactic features, and such features can be catalogued in virtue of the ‘class’ they belong to (37). (37) a. Argumental: person, gender, number, case b. Quantificational: wh-, Neg, measure, focus… c. Modifiers: evaluative, epistemic, Neg, frequentative, celerative, measure, manner… d. Topic By virtue of this classification, we can define ‘same structural type’ as in (38). (38) ‘Same structural type’ = Specifier licensed by features of the same class in (37). Given the above formulation, we expect RM effects to be generated by intervening elements whose set of features belong to the same class, but not by feature sets that belong to a different class.13 4.7.2. Agrammatic Minimality Grillo (2005, 2008, forthc.) hypothesises that a (temporal or permanent) processing deficit can lead to an underspecification/impoverishment of the morphosyntactic feature sets normally associated with the elements in the syntactic tree. A slower than normal activation of the syntactic information associated with lexical items (Zurif et al. 1993; Piñango 1999, among others), a slowed-down building up of this information into well formed syntactic constituents, or a faster than normal decay of syntactic representations (see Haarmann and Kolk 1991; Kolk 1995, 1998) can be at the base of this im13

Note that the definition in (37) allows to avoid the excess of restriction on movement generated by the simple A/Ā-distinction (of the system developed in Rizzi 1990) on the one hand, and the excessive freedom generated by the Minimal Link Condition (see Chomsky 1995) on the other. See Rizzi (2004a) for extensive discussion of this point.

254 Berit Gehrke and Nino Grillo poverishment. Grillo claims that features belonging to the Quantificational class, and more generally those related to the periphery of the clause and of the VP, are more likely to be compromised in this kind of scenario. Selective minimality effects can be expected to arise as a natural consequence of this underspecification. Given the structure in (39) in which every node is associated with a particular feature set, RM should permit the formation of a relation Q between X and Y. The presence of the element δ, say, a whfeature changing the class of the set from Argumental to Quantificational, suffices for RM to see the difference between X and Z and therefore to authorise the movement of Y over Z. (39) Successful extraction X (α, β, γ, δ)ClassQ … Z (α, β, γ)ClassA …Y (α, β, γ, δ) ClassQ z--------------m Q

(40) Minimality Effect X (α, β, γ)ClassA … Z(α, β, γ)ClassA …Y(α, β, γ)ClassA z--------------m *

Given an impoverished structure as that in (40), RM fails to see any relevant distinction between X and Z and therefore disallows a relation between X and Y. A more concrete example will be helpful in clarifying the hypothesis. In (41) the wh-feature, which defines as a member of the Operator’s class and thus as distinct from the Argumental class to which belongs, is present in the normal representation. (41) Normal representation (D,N,θ2,φs,acc,wh) ClassQ (D,N,θ1,φs,nom) ClassA (D,N,θ2,φs,acc,wh) ClassQ It is the boyi [whoi [the girl]j [ j kissed i]] z--------------m Q

In the agrammatic representation, on the other hand, the feature Q is absent (42). (42) Agrammatic representation (D,N,θ?,φs, …) ClassA (D,N,θ?,φs, …) ClassA (D,N,θ2,φs, …)ClassA It is the boyi [whoi [the girl]j [< … >? kissed < … >?]] z--------------m *

How to become passive

255

The impoverishment of the set of features (and subsequent the change of class of the moved element from Quantificational to Argumental) makes it impossible for agrammatic patients to distinguish the moved element from the intervening subject and RM blocks chain formation. Given that the chain is not built, it becomes impossible to assign the correct theta-role to each argument. This in turn leads to poor comprehension. This analysis correctly predicts a different pattern to arise with subject relatives, which are in fact correctly interpreted by agrammatic patients (see Table 1). In these structures, no DP intervenes between the moved constituent and its trace, hence no RM effects arise (43). (43) Subject clefts It is the boyi [whoi [ i loved the girl]] z----m The same analysis can be extended to cover the other asymmetries in Table 1 (see Grillo 2005, forthc., in prep., for further discussion).14 For our actual purposes, it is important to stress how the analysis sketched above interacts in a productive way with the present discussion of passivisation. We propose that agrammatic patients have troubles with activating (or maintaining active for long enough) the discourse-related feature that triggers movement of the consequent state due to a limitation of their processing capacities. This assumption is supported by the general difficulty of this population in dealing with discourse-related and quantificational features (see Avrutin 2006), both in comprehension and production. Additional work is needed to identify exactly which features are more problematic. However, evidence from omission/substitution of Det or Tense by Broca’s aphasics as well as their problems with embedding, wh-questions and other elements typically related to the CP field (on which see Friedmann and Grodzinsky 1997) are open for a unified explanation which takes edge/peripheral material to be more problematic to compute. Combining this assumption with our account of passivisation allows us to make the correct prediction that agrammatic comprehension of (eventive) 14

Grillo (in prep.) extends the present explanation to comprehension difficulties in language development and discusses its more general application to the case of normal adult speakers in stressful situations. Both populations have often been compared to agrammatic Broca’s aphasia for the similarities in comprehension deficits (see Grodzinsky 1990: ch. 4, among others Avrutin 1999; Dick et al. 2001).

256 Berit Gehrke and Nino Grillo passives with and without an overt by-phrase should be equally problematic whereas comprehension of unaccusatives and adjectival passives should not be (see Table 1 and Piñango 1999; Grodzinsky 2000, among many others). Under our proposal only eventive passives involve movement of a complex subevent (the lower VP shell) across another verbal projection, VP2. In the standard situation grammaticality is ensured by the presence of the discourse-related feature that allows the system to distinguish between the two subevents (the two VP shells) thus avoiding minimality. If the activation of this feature is delayed or decays faster than normally, a minimality effect between the moved VP2 and its trace arises. As a final note we should add that passives involving psych-verbs have been shown to be more difficult for agrammatics to comprehend than those involving actional predicate (see Grodzinsky 1995). We take this data to further support our analysis. As pointed out in section 4.4 passivisation of a psych-verb is an option available only through the construction of a complex event structure on top of a stative predicate. This step is necessary in order to generate a BECOME operator. We will assume that this additional requirement raises the overall complexity associated with the derivation of psychverb passives with respect to their actional counterparts. This increment in processing complexity, then, is expected to be observable in populations suffering from syntactic processing deficit.15 Hence, the deficitarian comprehension patterns of agrammatic Broca aphasia patients support our account of passivisation as involving the movement of a verbal projection associated with a consequent state subevent to some projection right above the higher VP shell. In the final section of this paper, which has a more speculative character than the previous sections, we propose that the movement is driven by some quantificational or discourse feature in order to create a link between the VP (the atemporal event structure) and the TP/CP domain. This link is assumed to independently be needed also in active sentences. 15

As noted above, the present explanation can in principle be extended to child language, under the natural assumption that the child processing system is not fully developed. The data from psych-verbs are particularly interesting from this perspective. It is a fairly established characteristics of child language that actional passives are mastered much earlier than passives of non-actional/stative predicates (on passives in child language see Borer 1989; Fox and Grodzinsky 1998, among others). A proper discussion of these facts would lead us too far from the scope of the present paper. It should be noted, however, that these asymmetries could be amenable to a natural explanation along the lines sketched above (see also Grillo, 2008, on this issue).

How to become passive

257

5. Anchoring the event structure in the temporal domain Roughly following Moens and Steedman (1988) who argue that the basic components that make up an event (the subevents in our terms) are not connected via temporal relations but rather by contingency (similar points are made in Rothstein 2004; Ramchand 2004), we assume that event structure itself is atemporal in nature in the sense that there are no times associated with any of the subevents.16 As a consequence, there is no immediate link between the (atemporal) event and the temporal domain of the clause. In this section, we propose that the position VP2 in passive constructions moves to is independently needed, also for active sentences, to form a basis for the event time that subsequently serves as the internal argument of Asp (in the sense of Demirdache and Uribe-Etxebarria 2000). As a result of this movement, the interpretation is that the event time falls within the consequent state. The syntax and semantics of tenses and aspects are commonly thought of as involving some reference to points or intervals in time (Partee 1984; Zagona 1990; Stowell 1996; Giorgi and Pianesi 1997; Demirdache and UribeEtxebarria 2000, among others). The point of departure for these accounts is usually Reichenbach (1947) who employs three temporal points, namely event time (E), speech time (S), and reference time (R). In his system, English simple tenses relate E to S with E before S with the past tense, E simultanous to S with the present tense, and E after S with the future tense. English complex tenses additionally express a relation between E and R, with E before R in perfect tenses and E simultaneous to R with the progressive. Klein’s (1994) model is similar to Reichenbach’s but it uses intervals instead of points labelled event time (EV-T), assertion time (AST-T), and utterance time (UTT-T). Demirdache and Uribe-Etxebarria (2000) use Klein’s terminology to capture the syntax of tenses and aspects in the following way (44).

16

Recently, Zwarts (2006) argues for the need of an atemporal account for properties of events like, for instance, telicity in terms of generalised paths.

258 Berit Gehrke and Nino Grillo (44) The syntax of Tense and Aspect (Demirdache & Uribe-Etxebarria 2000) TP ei UT-T T' ei T AspP WITHIN /AFTER ei AST-T Asp' ei Asp VP WITHIN /AFTER ei EVT-T V' ei V VP In this framework, tenses and aspects are predicates that take temporal arguments (following Zagona 1990). An aspect head takes the event time as its internal argument and the assertion time as its external argument. With the imperfective aspect the assertion time lies within the event time (WITHIN), with the perfective aspect it lies after the event time ( AFTER). Similarly, a tense head takes the assertion time as its internal argument and the utterance time as its external one. The utterance time can be placed within the assertion time (present), after the assertion time (past) or before the assertion time (future) (the latter is argued for in Demirdache 2005). In discussing a similar model, namely Stowell’s (1996), Ramchand (2004) notes that there is a ‘crucial phase boundary between vP and the temporal phrase structural domain’ which ‘requires the establishment of a relation between the extended event topology which makes no direct reference to times, and the actual time variable which is only introduced at Asp’ (Ramchand 2004: 333). In other words, there is no event time in her model since vP is crucially atemporal in nature. Rather, aspect introduces a time variable that is related to the event structure in a particular way. We can think of this time variable as being the counterpart to Demirdache and Uribe-Etxebarria’s (2000) assertion time. Ramchand argues that the minimal denotation of Asp in the language she discusses in her paper, Russian, is the one given in (45). (45) Minimal denotation of Asp (Ramchand 2004) [[Asp]] = λP λt ∃e:[P(e) and t ∈τ(e)]

How to become passive

259

She relies on Krifka’s (1998) temporal trace function τ, which is defined in Krifka as a function from E (the domain of events) into the extension of T (the domain of times) to map an event to its temporal trace (its ‘run time’). According to Ramchand’s analysis, in cases where there is no particular aspect head in Russian,17 t (or the assertion time) falls somewhere within the entire time the event takes, which is provided by the temporal trace function. The tree and the semantics of tense and aspect that Ramchand assumes are quite similar to those of Demirdache and Uribe-Etxebarria (2000). The crucial difference between the two proposals, however, is that the event time is non-existent in Ramchand’s approach but is more or less replaced by Krifka’s temporal trace function. An issue that arises under Ramchand’s proposal is that it still remains unclear which part of the complex event the temporal trace function actually picks out. Furthermore, under the assumption that there is a strict mapping between syntax and semantics (which is generally assumed by Ramchand), AspP problematically seems to provide both the temporal trace function (or alternatively the event time in Demirdache and Uribe-Etxebarria 2000) as well as the assertion time (t in Ramchand’s approach). To solve these problems, we propose to dissociate the introduction of the assertion time from some counterpart of Krifka’s temporal trace function or the precise conditions on how the event time is related to the complex event. In addition, we view the event time to be necessary, since Asp needs a temporal internal argument. We propose that the event time is provided by Voice at the point of transition from the atemporal domain of the event structure to the temporal one, i.e. from one phase to the other. Voice is responsible for grounding the event time in a particular way. In the case of passives the event time is anchored within the consequent state subevent, because VP2, which is semantically associated with the transition into a consequent state, moves to Spec VoiceP where it is assigned temporal properties. The feature that triggers movement to VoiceP has two properties, a discourse-related and a quantificational one. The discourse-related part chooses the element of the complex event that needs to be singled out whereas the quantificational part makes it readable to the next phase. Thus, the main job of this feature is to single out an element of the atemporal event structure 17

Russian productively expresses grammatical imperfective and perfective aspect by verbal prefixes and suffixes (see Gehrke 2008, for discussion). We assume, however, that this approach can also be carried over to other languages, also those that do not have a morphological category Aspect.

260 Berit Gehrke and Nino Grillo associated with the VP phase and to enrich its semantics by introducing temporality, thereby making it available to the next phase, the temporal domain (and ultimately the discourse domain) of the clause. In (46), we summarise the ingredients that we view necessary to account for the way the atemporal event structure is linked to the temporal domain. (46) Ingredients for creating the temporal link with events a. Events can be complex and consist of (atemporal) subevents. b. Aspect and Tense heads project argument structure with the relevant arguments utterance time,18 assertion time, event time (Demirdache and Uribe-Etxebarria 2000). c. Voice introduces the event time. d. Voice provides an additional landing site for the part of the event structure that the event time is related to. The precise implementation, especially the compositional account, still needs to be worked out but the general idea should be clear. We assume that these mechanisms are also needed for active sentences. We could think, then, that in the default case, nothing moves to the position above VP1 and the event time is assigned locally, or alternatively, the entire VP moves. In other cases, e.g. where the event time has to be placed within a certain subevent, the projection associated with this subevent moves up. For example, we could think of the progressive as focusing on the process of an event rather than on its initial or final state. We will leave this for future research. What is crucial for our approach to passives, then, is the point in (46d), i.e. the additional landing site for the part of the event structure that the event time is related to. In this way, Voice is reminiscent of the low focal projection proposed by Belletti (2004a). This could also be thought of in terms of some quantificational phrase such as Borer’s (2005) QP since it clearly involves some kind of quantification over events. Similarly, Arsenijević (2006) argues that verbal predicates have some functional projection that basically picks out that part of the complex event structure that something is asserted about. In any case, whatever is asserted about the event or that part of the event the focus lies on has to move up. In the case of passives, the consequent state moves up to serve as a basis for the event time.

18

Or in any case, some reference time which in many cases is the utterance time (see Stowell 1996, for discussion).

How to become passive

261

6. Conclusion This paper argued for a shift from an argument structure/DP perspective on passive formation to an analysis based on event structure. We showed that this shift accounts for several syntactic and semantic properties of the passive construction, some of which remained unexplained under previous proposals. The general idea is implemented through movement of a consequent state subevent to a discourse related position at the edge of the VP. From this position the internal argument can further move to the subject position, though this is not a necessary feature of passivisation, as clearly indicated by there-expletive passives. The tight relation between the availability of a consequent state and passivisation was highlighted by examples showing that the possibility to passivise a predicate depends on its event structure in a crucial way. Evidence from word order in constructions involving secondary resultative predicates, floating quantifiers, ditransitives, and there-expletives strengthened the idea that more than the internal argument moves in passives. Finally, the analysis proposed allowed us to make new predictions with respect to impoverished syntactic representation in agrammatic Broca’s aphasia, which were borne out, and to unify the treatment of some of the most typical deficitarian comprehension patterns in this syndrome. In a more speculative vein, we proposed that the position the lower VP shell moves to is independently needed for active derivations. In general, the movement of (part of) the atemporal and structurally complex event is necessary to single out an element of the verbal domain (the consequent state subevent in the case of passives), associated with the VP, and to enrich its semantics by introducing temporality, thus making it available to the next phase, the temporal (and eventually to the discourse) domain. Let us conclude with another speculation. We said in the beginning of section 2 that one problem of particular NP approaches to passivisation is that they draw a formal distinction between the past participle used in active sentences (in perfect tenses) and that in passives, although from a morphological and semantic point of view this distinction is not to be expected. Under our approach, both participles can in principle be treated the same, since reference to a consequent state is often also used in the analysis of perfect tenses. In discourse theories like Kamp and Reyle (1993), for instance, perfect tenses are generally analysed in terms of a state that abuts an event, which is similar to the idea of a consequent state coming into existence. We will leave this for future research.

262 Berit Gehrke and Nino Grillo Acknowledgements We would like to thank the following people for suggestions and comments on earlier drafts of this paper: Boban Arsenijević, Adriana Belletti, Valentina Bianchi, Hamida Demirdache, Jutta Hartmann, Jonathan MacDonald, Marjo van Koppen, Ad Neeleman, Eric Reuland, Luigi Rizzi, Tom Roeper, Andrea Santi, Maaike Schoorlemmer, Giorgos Spathas, Kriszta Szendrői, Naoko Tomioka, Mai Tungseth.

References Anagnostopoulou, Elena 2003 Participle and voice. In Perfect Explorations, Artemis Alexiadou, Monika Rather, and Armin von Stechow (eds.), 1–36. Berlin /New York: Mouton de Gruyter. Arsenijević, Boban 2006 Inner Aspect and Telicity: The Decompositional and Quantificational Nature of Eventualities at the Syntax-Semantics Interface. Doctoral Dissertation, University of Leiden. LOT Dissertation Series 142. Avrutin, Sergey 1999 Development of the Syntax Discourse Interface. Dordrecht: Kluwer. 2001 Linguistics and agrammatism. GLOT International 5: 1–11. 2006 Weak syntax. In Broca’s Region, chapter 4, Katrin Amunts and Yosef Grodzinsky (eds.). New York: Oxford University Press. Bach, Emmon 1981 On time, tense and aspect: An essay in English metaphysics. In Radical pragmatics, Peter Cole (ed.), 63–81. New York: Academic Press. Baker, Mark 1988 Incorporation: A Theory of Grammatical Function Changing. Chicago: University of Chicago Press. Baker, Mark, Kyle Johnson, and Ian Roberts 1989 Passive argument raised. Linguistic Inquiry 20: 219–251. Beck, Sigrid 2005 There and back again: A semantic analysis. Journal of Semantics 22: 3–51. Belletti, Adriana 2004a Aspects of the low IP area. In The Structure of CP and IP, Luigi Rizzi (ed.), 16–51. New York: Oxford University Press. 2004b Structures and Beyond: The Cartography of Syntactic Structures, Vol. 3. New York: Oxford University Press.

How to become passive

263

Belletti, Adriana and Luigi Rizzi 1988 Psych verbs and θ-theory. Natural Language and Linguistic Theory 6: 291–352. Borer, Hagit 1989 Anaphoric AGR. In The null subject parameter, Osvaldo Jaeggli and Kenneth J. Safir (eds.), 69–109. Kluwer. 2005 Structuring Sense. Oxford: Oxford University Press. Caramazza, Alfonso and Edgar Zurif 1976 Dissociation of algorithmic and heuristic processes in language comprehension: Evidence from aphasia. Brain and Language 3: 572–582. Chomsky, Noam 1957 Syntactic Structures. The Hague: Mouton. 1995 The Minimalist Program. Cambridge, MA: MIT Press. Cinque, Guglielmo 1999 Adverbs and Functional Heads. New York: Oxford University Press. 2002 Functional Structure in DP and IP: The Cartography of Syntactic Structures, Vol. 1. New York: Oxford University Press. Collins, Chris 2005 A smuggling approach to the passive in English. Linguistic Inquiry 36: 289–297. Demirdache, Hamida 2005 Scope and anaphora with time arguments: The case of ‘perfect modals’. Paper presented at the LAGB satellite workshop on Perfectivity and Telicity, Cambridge, UK. Demirdache, Hamida and Myriam Uribe-Etxebarria 2000 The primitives of temporal relations. In Step by Step: Essays on Minimalist Syntax in Honour of Howard Lasnik, Roger Martin, David Michaels, and Juan Uriagereka (eds.), 157–186. Cambridge, MA: MIT Press. Dick, Frederic, Elizabeth Bates, Beverly Wulfeck, Jennifer Utman, Nina Dronkers, and Morton Ann Gernsbacher 2001 Language deficits, localization and grammar: Evidence for a distributive model of language breakdown in aphasics and normals. Psychological Review 108: 755–788. den Dikken, Marcel 1995 Particles: On the Syntax of Verb-Particle, Triadic, and Causative Constructions. New York: Oxford University Press. Doetjes, Jenny 1997 Quantifiers and Selection. Doctoral Dissertation, Leiden University / HIL. Dowty, David 1979 Word meaning and Montague Grammar: The Semantics of Verbs and Times in Generative Semantics and in Montague’s PTQ. Dordrecht: Reidel.

264 Berit Gehrke and Nino Grillo Druks, Judit and John C. Marshall 1995 When passives are easier than actives: two case studies of aphasic comprehension. Cognition 55: 311–331. Embick, David 2004 On the structure of resultative participles in English. Linguistic Inquiry 35 (3): 355–392. Fox, Danny and Yosef Grodzinsky 1998 Children’s passive: A view from the by-phrase. Linguistic Inquiry 29: 311–332. Friedmann, Na’ama and Yosef Grodzinsky 1997 Tense and agreement in agrammatic production: Pruning the syntactic tree. Brain and Language 56: 397–425. Gehrke, Berit 2008a Goals and sources are aspectually equal: Evidence from Czech and Russian prefixes. Lingua 118: 1664 –1689. 2008b Ps in Motion: On the Semantics and Syntax of P Elements and Motion Events. Doctoral Dissertation, Utrecht University. LOT Dissertation Series 184. Giorgi, Alessandra, and Fabio Pianesi 1997 Tense and Aspect: From Semantics to Morphosyntax. Oxford: Oxford University Press. Grillo, Nino 2005 Minimality effects in agrammatic comprehension. In Proceedings of ConSOLE XIII, Sylvia Blaho, Eric Schoorlemmer, and Luis Vicente (eds.), 106–120. 2008 Generalized minimality: Syntactic underspecification in Broca’s Aphasia. Doctoral Dissertation, Utrecht University. LOT Dissertation Series 186. forthc. Structural impoverishment and minimality effects. Lingua. Grodzinsky, Yosef 1990 Theoretical Perspectives on Language Deficit. Cambridge, MA: MIT Press. 1995 Trace deletion θ-roles and cognitive strategies. Brain and Language 51: 469–497. 2000 The neurology of syntax: Language use without Broca’s area. Behavioral and Brain Sciences 23: 1–71. Guasti, Maria Teresa and Luigi Rizzi 2002 Agr and tense as distinctive syntactic projections: Evidence from acquisition. In The Cartography of Syntactic Structures, Guglielmo Cinque (ed.), 167–194. New York: Oxford University Press. Haarmann, Henk J. and Herman H. J. Kolk 1991 A computer model of the temporal course of agrammatic sentence understanding: The effects of variation in severity and sentence complexity. Cognitive Science 15: 49–87.

How to become passive

265

Haegeman, Liliane and Jaqueline Guéron 1999 English Grammar: A Generative Perspective. Oxford: Blackwell. Higginbotham, James 2000 Accomplishments. Ms., University of Southern California and University of Oxford. Jaeggli, Osvaldo A. 1986a Three issues in the theory of clitics: Case, doubled NPs, and extraction. In The Syntax of Pronominal Clitics, Syntax and Semantics 19, Hagit Borer (ed.), 15–42. New York: Academic Press. 1986b Passive. Linguistic Inquiry 17: 587–622. Kamp, Hans and Uwe Reyle 1993 From Discourse to Logic: Introduction to Modeltheoretic Semantics of Natural Language, Formal Logic and Discourse Representation Theory. Studies in Linguistics and Philosophy 42. Dordrecht: Kluwer. Kayne, Richard S. 1985 Principles of particle constructions. In Grammatical representation, ch. 5, Studies in Generative Grammar 22, Jacqueline Guéron, HansGeorg Obenauer, and Jean-Yves Pollock (eds.), 101–140. Dordrecht: Foris. Klein, Wolfgang 1994 Time in Language. Germanic Linguistics. New York: Routledge. Kolk, Herman H. J. 1995 A time-based approach to agrammatic production. Brain and Language 50: 282–303. 1998 Disorders of syntax in aphasia, linguistic-descriptive and processing approaches. In Handbook of Neurolinguistics, Brigitte Stemmer and Harry A. Whitaker (eds.), 249–260. San Diego: Academic Press. Kratzer, Angelika 2000 Building statives. In Proceedings of the 26th Annual Meeting of the Berkeley Linguistics Society, Lisa J. Conathan (ed.), 385–399. Berkeley, CA: Berkely Linguistics Society. 2005 Building resultatives. In Event Arguments: Foundations and Applications, Claudia Maienborn and Angelika Wöllstein (eds.), 177–212. Tübingen: Niemeyer. Krifka, Manfred 1998 The origins of telicity. In Events and Grammar, Susan Rothstein (ed.), 197–235. Dordrecht: Kluwer. Manzini, Maria Rita 1980 On Control. Ms., MIT. Marantz, Alec 1984 On the Nature of Grammatical Relations. Linguistic Inquiry Monograph 10. Cambridge, MA: MIT Press.

266 Berit Gehrke and Nino Grillo McIntyre, Andrew 2006 The integration of directional PPs: Thoughts on the way to getting towards knowing if I’m (be)coming or going. Paper presented at the conference ‘Syntax and semantics of spatial P’ in Utrecht, June 2006. Moens, Marc and Mark Steedman 1988 Temporal ontology and temporal reference. Computational Linguistics 14: 15–28. Neeleman, Ad 1994 Complex Predicates. Doctoral Dissertation, Utrecht University. Parsons, Terence 1990 Events in the Semantics of English: A Study in Subatomic Semantics. Cambridge, MA: MIT Press. Partee, Barbara 1984 Nominal and temporal anaphora. Linguistics and Philosophy 7:243– 286. Piñango, Maria Mercedes 1999 Syntactic displacement in Broca’s aphasia comprehension. In Grammatical Disorders in Aphasia: A Neurolinguistic Perspective, Yosef Grodzinsky and Roelien Bastiaanse (eds.), 75–87. London: Whurr. Postal, Paul M. 2004 Skeptical Linguistic Essays. Oxford /New York: Oxford University Press. Pustejovsky, James 1991 The syntax of event structure. Cognition 41: 47–81. Ramchand, Gillian 2004 Time and the event: The semantics of Russian prefixes. Norlyd 32: 323–361. Rapp, Irene and Arnim von Stechow 1999 Fast ‘almost’ and the visibility parameter for functional adverbs. Journal of Semantics 16 (2): 149–204. Reichenbach, Hans 1947 Elements of Symbolic Logic. London: MacMillan. Reinhart, Tanya 2002 The theta system: An overview. Theoretical Linguistics 28: 229–290. Rizzi, Luigi 1990 Relativized Minimality. Linguistic Inquiry Monograph 16. Cambridge, MA: MIT Press. 1997 The fine structure of the left periphery. In Elements of Grammar, Liliane Haegeman (ed.), 281–337. Dordrecht: Kluwer. 2004a Locality and the left periphery. In Structure and Beyond, ch. 7, Adriana Belletti (eds.), 223–251. New York: Oxford University Press. 2004b The Structure of CP and IP: The Cartography of Syntactic Structures, Vol. 2. New York: Oxford University Press.

How to become passive

267

Roberts, Ian G. 1987 The Representation of Implicit and Dethematized Subjects. Dordrecht: Foris. 32 Roeper, Thomas 1983 Implicit arguments and the head-complement relation. Linguistic Inquiry 18: 267–310. Rothstein, Susan 2004 Structuring Events: A Study in the Semantics of Lexical Aspect. Oxford: Blackwell. Snyder, William 2005 Motion predicates and the compounding parameter: A new approach. Paper presented at the University of Maryland, April 2005. Sportiche, Dominique 1988 A theory of floating quantifiers and its corollaries for constituent structure. Linguistic Inquiry 19: 425–449. Starke, Michal 2001 Move Dissolves into Merge: A Theory of Locality. Doctoral Dissertation, University of Geneva. von Stechow, Arnim 1995 Lexical decomposition in syntax. In The Lexicon and the Organization of Language, Urs Egli, Peter E. Pause, Christoph Schwarze, Arnim von Stechow, and Götz Wienhold (eds.), 81–118. Amsterdam: John Benjamins. Stowell, Tim 1981 Origins of Phrase Structure. Doctoral Dissertation, MIT. 1996 The phrase structure of Tense. In Phrase Structure and the Lexicon, J. Rooryck and L. Zaring (eds.), 277–291. Dordrecht: Kluwer. de Swart, Henriëtte 1998 Aspect shift and coercion. Natural Language and Linguistic Theory 16 (2): 347–85. Travis, Lisa 2000 Event structure in syntax. In Events as Grammatical Objects: The Converging Perspectives of Lexical Semantics and Syntax, Carol Tenny and James Pustejovsky (eds.), 145–185. Stanford, CA: CSLI Publications. Travis, Lisa deMena 2005 Agents and causes in Malagasy and Tagalog. In The Syntax of Aspect: Deriving Thematic and Aspectual Interpretation, Nomi Erteschik-Shir and Tova Rapoport (eds.), 174–189. New York: Oxford University Press. in prep. Inner Aspect. Dordrecht: Kluwer. Vendler, Zeno 1967 Linguistics in Philosophy. Ithaca, NY: Cornell University Press.

268 Berit Gehrke and Nino Grillo Verkuyl, Henk 1993 A Theory of Aspectuality: The Interaction between Temporal and Atemporal Structure. Cambridge: Cambridge University Press. Williams, Edwin 1980 Predication. Linguistic Inquiry 11: 203–238. Zagona, Karen 1990 Times as temporal argument structure. Paper presented at the conference ‘Time in Language’, MIT, Cambridge, MA. Zeller, Jochen 2001 Lexical particles, semi-lexical postpositions. In Semi-Lexical Categories, Norbert Corver and Henk van Riemsdijk (eds.), 505–549. Berlin /New York: Mouton de Gruyter. Zurif, Edgar 1996 Grammatical theory and the study of sentence comprehension in aphasia: Comments on Druks and Marshall (1995). Cognition 58: 271–279. Zurif, Edgar, and David Swinney 1994 The neuropsychology of language. In Handbook of Psycholinguistics, Morton Ann Gernsbacher (ed.), 1055 –1074. New York: Academic Press. Zurif, Edgar, David Swinney, Penny Prather, Julie Solomon, and Camille Bushell 1993 An on-line analysis of syntactic processing in Broca’s and Wernike’s aphasia. Brain and Language 45: 48–464. Zwarts, Joost 2006 Event shape: Paths in the semantics of verbs. Ms. Radboud University Nijmegen.

Ergativity, accusativity, and the order of Merge and Agree Gereon Müller

The goal of this paper is to outline the core of a new minimalist analysis of accusative vs. ergative patterns of argument encoding (via case marking or agreement). The central observation is that indeterminacies may arise in the application of the two elementary operations Merge and Agree (see Chomsky 2000, 2001), given that they both obey an Earliness requirement (see Pesetsky 1989); and the central claim I would like to put forward is that a principled resolution of one such indeterminacy (on the vP cycle) in one or the other direction yields an accusative or ergative encoding pattern for arguments. The analysis can be extended so as to cover various types of splits that have been noted in the literature (among them tense/aspect-based splits and clause type-based splits).

1. Argument encoding patterns There are two basic encoding patterns for external and internal argument DPs (DPext, DPint) of transitive and intransitive verbs (Vt , Vi ) that are not lexically marked. In an accusative pattern, DPint of Vt is encoded by accusative morphology; DPext of Vt and Vi, and DPint of Vi are encoded by nominative morphology. In contrast, in a (pure) ergative pattern, DPext of Vt is encoded by ergative morphology; DPext of Vi, and DPint of Vt and Vi are encoded by absolutive morphology. This is illustrated schematically in (1) (see Plank 1995). (1)

a. Accusative marking

b. Ergative marking

DPext – Vi DPext – Vt

DPint – Vi DPint – Vt

DPext – Vi DPext – Vt

DPint – Vi DPint – Vt

NOM

ACC

ERG

ABS

Argument encoding can proceed by case-marking on the DP argument (‘dependent-marking’) or by agreement-marking on the verb (‘head-marking’);

270 Gereon Müller see Nichols (1986), Baker (1996). I take this issue to be orthogonal to the choice of encoding pattern as such, and will use the terms ‘accusative’, ‘nominative’, ‘ergative’, and ‘absolutive’ indiscriminately for case-and agreement-marking throughout this paper, with CASE as a cover term for both.1 In what follows, I give examples instantiating each of the four language types that result from cross-classifying type (accusative vs. ergative) and place (case vs. agreement) of argument encoding. The Icelandic examples in (2) illustrate an accusative case-marking pattern. DPint of Vt is marked by accusative; other primary arguments receive nominative. (2)

a. Sól-Ø=in skín-Ø sun-SG.NOM=det.SG.FEM.NOM shine-3.SG ‘The sun is shining.’

(Kress 1982: 263)

b. Ólaf-ur byrja-ð-i of sein-t Olaf-SG.NOM begin-PAST-3.SG too late-3.SG.NEUT ‘Olaf began too late.’ c. Ólaf-ur las-Ø bók-Ø=ina Olaf-SG.NOM read.PAST-3.SG book-SG.ACC=det.SG.FEM.ACC ‘Olaf read the book.’ (Sigurðsson 2002: 698) An accusative pattern that relies on agreement-marking can be found in Navajo. An intransitive context is given in (3a), transitive contexts are shown in (3b–c). Again, DPint of Vt is encoded by one type of morphological marker (accusative), whereas all other primary arguments are encoded by another marker (nominative).2

1

2

This extension of the traditional case terminology to agreement-marking may be more common for ergative systems than for accusative systems; see, e.g., Bickel & Nichols (2001). Overt argument DPs are usually optional in head-marking languages; I assume that primary arguments are nevertheless present in the syntax here, in the form of empty DP pronouns (see Baker (1996), Bruening (2001) for some of the options that arise under this general view). The Navajo agreement markers are usually called ‘subject’ and ‘object’ markers in the literature, and glossed here with the labels NOM and ACC; they are fusional and encode person and number in addition to CASE.

Ergativity, accusativity, and the order of Merge and Agree

(3)

271

a. (Y)i-sh-cha Ø-1.SG.NOM-cry ‘I am crying.’ b. Ni-sh-ch’id 2.SG.ACC-1.SG.NOM-scratch ‘I am scratching you.’ c. Shí-í-ní-gháád 1.SG.ACC-PERF-2.SG.NOM-shake ‘You shook me.’

(Speas 1990: 209)

A language that instantiates an ergative case-marking pattern in a relatively pure way is Archi (North-Caucasian, Daghestanian; see Kibrik 1979). DPext of Vt is singled out and marked by ergative case; all other primary arguments receive absolutive case. (4a) provides an intransitive context, (4b) a transitive one.3 (4)

a. Dija-Ø w-irx̅ in ° father:I-SG.ABS I.SG-work ‘Father is working.’ b. Dija-mu x°alli-Ø b-ar-ši b-i father:I-SG.ERG bread:III-SG.ABS III.SG-bake-GER III.SG-Aux ‘Father is baking the bread.’ (Kibrik 1979: 67)

Finally, Sierra Popoluca (Meso-American, Mixe-Zoque; see Elson 1960; Marlett 1986) exhibits an ergative encoding pattern that relies on agreement-marking. DPext of Vt is encoded by one type of agreement marker (ergative); all other primary arguments are encoded by another type of agreement marker (absolutive).4 (5a–b) illustrate intransitive contexts; (5c–d) provide transitive contexts. (5)

3

4

a. A-nɨk-pa 1.ABS-go-INC ‘I am going.’

(Marlett 1986: 364)

I, III are noun classes (there are eight); case markers bear number information (Kibrik 2003: 53ff.). The agreement markers also indicate person, but not number; the latter plays a minor role in Sierra Popoluca morphology (Elson 1960: 209/218).

272 Gereon Müller b. A-pɨːšiñ 1.ABS-man ‘I am a man.’ c. A-Ø-koˀc-pa 1.ABS-3.ERG-hit-INC ‘He is hitting me.’ d. Ø-Aŋ-koˀc-pa 3.ABS-1.ERG-hit-INC ‘I am hitting him.’

(Elson 1960: 208)

These examples may suffice as an illustration of the two basic argument encoding patterns, by case-marking and by agreement-marking. Needless to say, closer scrutiny reveals the actual situation to be more involved in all four languages, with various interfering factors emerging that blur the simple picture arising on the basis of the data presented here. Still, the core of the systems of argument encoding in these languages is either (1a) or (1b), and these two patterns need to be derived in a simple and general way. What follows is an attempt to do this by resolving an indeterminacy in the application of elementary minimalist operations that can independently be observed.

2. A case of indeterminacy on the vP cycle Let me begin with some background assumptions (based on Chomsky 2000, 2001, 2005a,b). Assume that syntactic structure is created incrementally, bottom-up, by the elementary operations Merge and Agree, and by Move (which may or may not be a special case of Merge – e.g., “internal” Merge –, and which will not play a major role in what follows). For present purposes, the operations Merge and Agree can be understood as in (6) and (7), respectively.5 (6)

Merge: α can be merged with β, forming a projection of α, if α bears a subcategorization feature [•F•] and F is the label of β.

According to (6), Merge operations do not come for free (as assumed by Chomsky (2001, 2005a), among others). Rather, they are driven by sub5

In (6) and in (7), α, β, δ stand for categories.

Ergativity, accusativity, and the order of Merge and Agree

273

categorization features (which I note here as [•F•], following Heck & Müller 2007); see Svenonius 1994; Collins 2003; Adger 2003; Lechner 2004; Kobele 2006; Sternefeld 2006; and Pesetsky & Torrego 2006).6 (7)

Agree: α can agree with β with respect to a feature bundle Γ iff (a), (b), and (c) hold: a. α bears a probe feature [*F*] in Γ, β bears a matching goal feature [F] in Γ. 7 b. α m-commands β.8 c. There is no δ such that (i) and (ii) hold: (i) δ is closer to α than β.9 (ii) δ bears a feature [F] that has not yet participated in Agree.

Thus, Agree operations are driven by a probe seeking a goal (7a), require mcommand (7b), and obey minimality (7c). I assume that Agree needs only identity of probe and goal features to apply, and can thus apply if probe and goal differ in their feature value (i.e., if the feature specifications are different); but if it does so, it is unsuccessful and creates a crash of the derivation.10 6

7

8

9

10

Merge of adverbs may or may not require additional assumptions on this approach; not much depends on this in the present context, though. Starring a feature indicates its probe status, here and in what follows; see Roberts & Roussou (2002), Adger (2003), and Sternefeld (2006) for this kind of notational device. This permits an Agree relation between a head and its specifier, as seems natural. This assumtion will turn out to be crucial for the analysis. It should be noted, though, that Chomsky (2001, 2005a) explicitly rejects such a possibility. δ is closer to α than β if the path from δ to α is shorter than the path from β to α. The path from X to Y is the set of categories Z such that (a) and (b) hold: (a) Z is reflexively dominated by the minimal XP that dominates both X and Y. (b) Z dominates X or Y. (See Müller 1998: 130; also cf. Pesetsky 1982: 289; Collins 1994: 56.) The length of a path is determined by its cardinality. It follows that the specifier and the complement of a head qualify as equally close to the head; and that the specifier of a head is closer to the head than a category that is further embedded in the complement of the head. This implies a checking (rather than valuation) approach to Agree. However, all that follows would in principle also be compatible with a valuation approach; see Heck & Müller (2007).

274 Gereon Müller Suppose further (following Chomsky 2000, 2001) that the basic clause structure consists of CP, TP, vP, and VP; that lexical items that are to participate in derivations are selected from the lexicon pre-syntactically, and assembled in a numeration N (or lexical array); that DPint is merged in VP, whereas DPext is merged in vP, as a specifier; and, finally, that T and v are involved in the structural encoding of primary arguments (i.e., DPext and DPint arguments for which no inherent/lexical CASE is specified), by bearing features that act as probes and thus trigger Agree operations. More specifically, I will assume that that there is only one structural argument encoding feature, CASE, which can have two values: ext(ernal) and int(ernal) (determined with respect to vP, the predicate domain). The feature specifications [CASE:ext] and [CASE:int] replace the traditional feature specifications [CASE:nom], [CASE:acc], [CASE:abs], [CASE:erg]. [CASE] features figure in Agree relations involving T/v and DP, where T bears a probe [*CASE:ext*] requiring a matching [CASE:ext] goal on DP, and v bears a probe [*CASE:int*] requiring a matching [CASE:int] goal on DP.11 Casemarking and agreement-marking both depend on an Agree relation between T/v and DP, and thus qualify as two sides of the same coin (see, e.g., Bobaljik & Wurmbrand (2003)): Argument encoding proceeds by case marking if [CASE:α] is morphologically realized on DP; it proceeds by agreementmarking if [*CASE:α*] is morphologically realized on T/v.12,13 11

12

13

This deviates from Chomsky (2001: 6). Still, what follows would in its essentials also be compatible with the assumption that some other features (e.g., Φ-features) on T/v act as probes forcing Agree with DP, and CASE is part of the feature bundle Γ in the sense of (7). However, Chomsky further assumes that CASE is ‘not a feature of […] T, v’ (even though ‘the value assigned depends on the probe: nominative for T, accusative for v’), which might raise questions concerning agreement-marking under present assumptions. The two options are not mutually incompatible and often co-exist to some extent in a single language (cf., e.g., nominative encoding via case and agreement in Icelandic); but see Nichols (1986) for some distributional asymmetries. In languages that employ both case and agreement, the encoding patterns can sometimes be mixed, such that, e.g., case marking involves an ergative encoding pattern, and agreement involves an accusative encoding pattern. In these cases, I will take the case marking pattern to be primary. For now, I will have to leave open the question of which additional assumptions are necessary (in the approach to be developed below) to derive the diverging agreement pattern. Morphological agreement marking on T/v requires complex head formation (involving T, v, V), at least in the languages under consideration. I assume that this is accomplished by head movement (and, perhaps, head lowering), followed by

Ergativity, accusativity, and the order of Merge and Agree

275

Independently of these specific assumptions about the values of CASE, it is a conspicuous property of the overall system that v (unlike T or V) plays a dual role: It participates in a (first) Merge operation with a DP, and it also participates in an Agree relation with a DP. I will now argue that this dual role has far-reaching consequences for the nature of argument encoding in a language. Consider first a simple transitive context, with two arguments DPint, DPext. Suppose that the derivation has reached a stage Σ where v has been merged with a VP containing DPint , with DPext waiting to be merged with v in the workspace of the derivation.14 At this point, an indeterminacy in rule application arises: The next operation could be either Agree(v,DPint) or Merge(DPext ,v). Based on economy considerations, Chomsky (2001) proposes an Earliness requirement for syntactic operations (see Pesetsky 1989).15 Given Earliness, operations like Merge and Agree must apply as soon as their structural conditions are met, which they both are at stage Σ in the derivation under consideration. Consequently, there is a dilemma: Only one operation can apply first, as required by Earliness. In view of this, one might conclude that an Earliness requirement should be abandoned for either Merge or Agree, so as to resolve the indeterminacy. However, such a step would make it necessary to deny all empirical relevance of Earliness for one of the two operations, and it would also be at variance with the economy-based motivation.16 In contrast, I would like to contend that conflicts of this type are real, and must be resolved in a language by giving one

14

15

16

post-syntactic insertion of agreement markers into T/v (Halle & Marantz 1993). See Embick & Noyer (2001), Bobaljik (2002) for discussion of the role of head movement in this; and Müller (2006) for an account of the morphology of agreement-marking in Sierra Popoluca along these lines. The workspace of the derivation comprises items in the numeration and phrases that have been created independently; DPext belongs to the former if it is a bare lexical item, and to the latter if it has been created via Merge. See, among others, Frampton & Gutman (1999) and Hornstein (2001) for relevant discussion. See Chomsky (2001: 15): “With the motivation for Procrastinate gone, considerations of efficient computation would lead us to expect something like the opposite: Perform computations as quickly as possible.” Rezac (2003: 158), in discussing stage Σ, tacitly presupposes that Earliness holds only for Agree (even though he envisages Agree(v,DPext) as a general possibility underlying cases of ‘agreement displacement’). However, there is empirical evidence for an Earliness requirement for Merge, and such a requirement has been assumed in practice; cf. the discussion of Merge before Move in Chomsky (1995, 2000, 2001).

276 Gereon Müller Earliness requirement priority over the other in the case of conflict – in other words, by ranking the two requirements.17 It turns out that this not only resolves the indeterminacy encountered on the vP cycle; it is in fact all that needs to be assumed to derive the core difference between accusative and ergative encoding patterns. 3. Transitive contexts: The order of elementary operations Suppose first that a language gives priority to the Earliness condition on Agree in the case of conflict. Then, an accusative pattern arises (see (8a) below): At stage Σ, Agree(v,DPint) applies first (step (i)). Since v is marked [*CASE:int*], this ensures a [CASE:int] specification on DPint. If DPint is specified as [CASE:ext], Agree(v,DPint) applies unsuccessfully, and the derivation crashes.18 DPext is merged in Specv in the next step (step (ii)). The derivation continues, merging T and vP, and then carrying out Agree(T,DPext), which requires [CASE:ext] on DPext (step (iii)). The morphological realization of an internal encoding feature [(*)CASE:int(*)] with Agree(v,DPint) (by case or agreement) can be called accusative; the morphological realization of an external encoding feature [(*)CASE:ext(*)] with Agree(T,DPext) can be called nominative. This accounts for argument encoding in transitive contexts in languages like Icelandic and Navajo: The internal argument is marked by the internal CASE, the external argument is marked by the external CASE.19

17

18

19

This amounts to an optimization procedure, with minimal violability of the lower-ranked requirement; see Prince & Smolensky (1993). However, the optimization involved here is extremely local (competing candidates are derivational steps), which avoids the complexity problems incurred by standard optimization procedures; see Heck & Müller (2000, 2003). Note also that the same kind of local optimization procedure underlies the Merge before Move principle. Recall that I have adopted a checking approach to Agree relations like CASE assignment (and not a valuation approach). Under a valuation approach, a v marked [*CASE:int*] would automatically instantiate [CASE:int] on DPint (thereby raising questions for correlations between interpretability and valuation that are standardly assumed), so that the problematic (non-matching) situation could not arise in the first place. As noted, the issue of checking vs. valuation is irrelevant for the approach to variation in argument encoding adopted here. In all languages discussed in this section, Agree triggered by [*CASE:α*] also affects Φ-features.

Ergativity, accusativity, and the order of Merge and Agree

277

In contrast, suppose now that a language gives priority to the Earliness condition on Merge. Then, an ergative pattern arises (see (8b)): At stage Σ, Merge(DPext,v) must apply first (step (i)). Crucially, DPext is now closer to v than DPint (cf. note 9), and given that Agree relations are subject to a minimality requirement and require only m-command by the probe (see (7)), the next operation will have to be Agree(v,DPext), in a specifier/head configuration (step (ii)). This requires[CASE:int] on DPext. Subsequently, T is merged, and Agree(T,DPint) is carried out (step (iii)), with [CASE:ext] for DPint .20 The morphological realization of an internal encoding feature [(*)CASE:int(*)] with Agree(v,DPext) can be called ergative; the morphological realization of an external encoding feature [(*)CASE:ext(*)] with Agree(T,DPint) can be called absolutive. This accounts for argument encoding in transitive contexts in languages like Archi and Sierra Popoluca: The internal argument is marked by the external CASE; the external argument is marked by the internal CASE.21

20

21

This Agree operation is just local enough to be in accordance with the liberal version of the Phase Impenetrability Condition (PIC) in Chomsky (2001: 14) (or with the ‘alternative’ interpretation of the PIC that is available to languages; see Richards 2004, 2007). (Also, DPext does not intervene, given (7c).) Some remarks are due on how the present approach is related to other analyses that identify a difference in structural case assignment as the source of the two basic argument encoding types (as opposed to differences in projection, as in Marantz 1984, or differences in lexical case assignment, as in Nash 1996; Alexiadou 2001; and Woolford 2001). The analyses in Levin & Massam (1985), Chomsky (1995: ch.3), Bobaljik (1993), Laka (1993), Rezac (2003), and Bobaljik & Branigan (2006) differ from the one given here in that ergative is identified with nominative, and absolutive with accusative. Bittner & Hale (1996b) identify absolutive and nominative but treat ergative and accusative differently. The closest predecessor of the present proposal is the analysis in Murasugi (1992) (also see Jelinek 1993; Ura 2000, 2006), where nominative is identified with absolutive, and ergative with accusative.

278 Gereon Müller (8)

a. Agree before Merge: accusative

b. Merge before Agree: ergative

TP 3 Tʹ 3 T[*c:ext*] vP 3 DPext vʹ 3 (iii) v[*c:int*] VP 3 (ii) V DPint

TP 3 Tʹ 3 T[*c:ext*] vP 3 DPext vʹ 3 (i) v[*c:int*] VP 3 (iii) (ii) V DPint

(i)

To sum up so far, I have proposed an account of the difference between ergative and accusative systems of argument encoding in transitive contexts that has the following properties (which, taken together, set it apart from all alternative analyses I am aware of; see footnote 21): First, the projection of arguments is identical in both encoding systems. Second, no semantically irrelevant projections (like AgrS P, AgrO P, etc.) need to be assumed, in line with the program laid out in Chomsky (1995). Third, CASE determination is independent of movement. Fourth, the analysis identifies ERG with ACC, and ABS with 22 NOM. The former cases are internal (to vP); the latter ones are external. 22

Independent evidence for this treatment of ergative and absolutive as essentially identical cases may come from the fact that nominative and absolutive are generally morphologically less marked than ergative and accusative (see Comrie (1989: 126), Dixon (1994: 11); and Cysouw et al. (2006) for the exception to the rule). In the present approach, this can be interpreted as a preference for morphological realization of internal structural CASE (i.e., v-related case) over external structural CASE (i.e., T-related case). However, this argument may not be conclusive. Adopting an approach where T is considered responsible for nominative and ergative, and v for accusative and absolutive, Chomsky (1993) suggests that the correct generalization with respect to morphological realization might be that “the ‘active’ element typically assigns a less-marked Case to its Spec.” A similar kind of approach is developed by Bobaljik (1993), based on an Obligatory Case Parameter. According to this parameter, one of the two core cases is obligatory. In accusative languages, the obligatory case is the case assigned by AgrS /T (nominative/ergative); in ergative languages, it is the case assigned by AGRO /v (accusative/absolutive). With respect to the morphological realization asymmetry, Bobaljik (1993: 51) states: “Presumably, the observed morphological tendency towards null morphology for these Cases [i.e., nominative and absolutive] is a reflection of this obligatory status.”

Ergativity, accusativity, and the order of Merge and Agree

279

Finally, it is worth pointing out that no recourse is made to arbitrary parametrizations – the indeterminacy in rule application on the vP cycle arises independently, and Earliness requirements are independently motivated for syntactic operations, so some resolution of the kind sketched here must be assumed anyway.23

4. Intransitive contexts: Case feature specifications in numerations Consider next intransitive contexts. Unchecked probes lead to a crash of the derivation; hence, [*CASE:α*] must be absent on either T or v in the derivation if only one DP is present that provides a matching goal feature specification [CASE:α]. Thus, again, there is an indeterminacy, and again, the problem arises independently of specific assumptions about CASE feature values.24 I will suggest that several principled solutions are available. Consider a derivation that fails to provide a matching goal feature specification for each probe feature specification that it employs. Such a derivation is doomed from the start. It should therefore be excluded in some general way. A place where this can be ensured straightforwardly is the numeration – a component of grammar that does not yet involve structure (outside of individual lexical items) but provides just enough information to formulate constraints that reduce the number of unsuccessful derivations. For concreteness, I assume the following constraint on numerations:25 23

24

25

In contrast, existing analyses that share the same basic characterization of cases (ergative as accusative, and absolutive as nominative) need to resort to specific parameters without independent motivation. Thus, Murasugi (1992) assumes that the basic argument encoding parameter can ultimately be reduced to the question of whether T or v (or, in her terms: AgrS or AgrO ) is “strong” (also recall Chomsky’s (1995) notion of “active” functional head); but there is no morphological correlate on the functional head associated with this notion of “strength”. Alternatively, Ura (2006) assumes that the ergative parameter consists in whether or not an external argument in Specv can undergo Agree in a base-generated position – if it can, case is assigned to it by v, if not, by T. A faithfully updated version of Burzio’s generalization (“v can bear [*CASE:int*] iff V takes a DPext”, cf. Burzio 1986: 185) would still require an auxiliary assumption to derive the patterns in (1) (the problem is to ensure that DPext of Vi cannot be marked by accusative/ergative). This constraint may possibly be conceived of as a special case of a more general requirement for numerations, which, e.g., might also ensure the presence of exactly n lexical items that can serve as arguments for every n-place predicate (es-

280 Gereon Müller (9)

Feature Balance: For every feature specification [*F:α*], there must be a matching feature specification [F:α].

It follows that either [*CASE:ext*] on T or [*CASE:int*] on v must be absent if there is only one D with a CASE feature in the numeration. But which of the two? One possibility is suggested by markedness considerations: [(*)CASE:ext(*)] (nominative/absolutive) is unmarked, [(*)CASE:int(*)] (accusative/ergative) is marked, both from a syntactic and from a morphological point of view. Syntactically, nominative/absolutive is the type of CASE used in default contexts. Morphologically, at least as a robust tendency, nominative and absolutive markers are segmentally less complex than accusative and ergative markers (see footnote 22). Thus, assume that (9) is respected in intransitive contexts by maintaining the unmarked feature specification. Then, [*CASE:ext*] on T has to show up, and [*CASE:int*] cannot be instantiated on v in the numeration.26 Consequently, the sole argument of an intransitive predicate (DPext or Dint) is predicted to be encoded by [(*)CASE:ext(*)] (nominative/absolutive), after Agree(T,DPext) or Agree(T,DPint), which captures the situation in the types of language discussed so far. However, suppose now that there is a second way for a language to respect (9): In the numeration, a CASE feature specification must be matched by the argument type of a D with respect to markedness; this can be viewed as an iconicity constraint on numerations. The unmarked situation for an argument of a predicate is to be merged in that predicate’s projection; ‘externalization’ of an argument can be viewed as a special operation in argument structures (see, e.g., Williams 1981) which has the effect that the subcategorization feature [•D •] introducing an external argument is located on v rather than on V. Consequently, with respect to argument type, DPext is

26

sentially, a version of the Θ-Criterion). Also compare Stabler’s (1996) discussion of count invariants. Alternatively: [*CASE:int*] is deleted on v in the numeration. One may either assume that (finite) T and v bear CASE features as a lexical property, which can then be deleted in the numeration if constraints like Feature Balance force this; or that CASE features are inserted on T and v in the numeration by redundancy rules, which can then be blocked by constraints like Feature Balance. It is hard to see how the two options could make a difference empirically; for the sake of concreteness, I presuppose the second (arguably somewhat simpler) approach, here and in what follows.

Ergativity, accusativity, and the order of Merge and Agree

281

inherently more marked than a DPint. Under this assumption, a marked feature specification ([*CASE:int*] on v) must show up in the numeration in the presence of a V taking an marked argument (DPext), and an unmarked feature specification ([*CASE:ext*] on T) occurs in the presence of a V taking an unmarked argument (DPint). A language that chooses this option does not differ from the languages considered thus far in transitive contexts, but it does in intransitive contexts. Suppose that such a language exhibits an ergative marking pattern (by giving priority to Merge over Agree on the vP cycle). Then, DPint undergoes an Agree(T,DPint) operation and is encoded by absolutive in intransitive contexts, whereas DPext participates in Agree(v,DPext) and is encoded by ergative. This way, an ‘active’ system of split ergativity arises; see (10). (10)

Active ergative marking DPext –Vi DPext –Vt

DPint –Vi DPint –Vt

ERG

ABS

Again restricting attention to the core system, (10) is instantiated in languages like Basque (with case) and Guaraní (with agreement); see (11), (12), respectively. (11) a. Jon-Ø etorri da Jon-ABS come:PTCP.PRF is:3.SG.INTR ‘Jon came.’ b. Jon-ek saltatu du Jon-ERG jump:PTCP.PRF have:3.SG.TR ‘Jon jumped.’ c. Jon-ek ardo-a-Ø ekarri du Jon-ERG wine-DET-ABS bring:PTCP.PRF have:3.SG.TR ‘Jon brought the wine.’ (Hualde & Ortiz de Urbina 2003: 364) (12) a. Še-manuˀa 1.SG.ABS-remember ‘I remember.’ b. A-ma.apo 1.SG.ERG-work ‘I work.’

282 Gereon Müller c. Ø-Ai-pete 3.SG.ABS-1.SG.ERG-hit ‘I hit him.’

(Gregores & Suárez 1967)

The present analysis does not per se exclude an ‘anti-active’ pattern, as in (13). Anti-active marking would arise in an accusative system that instantiates (or preserves) the CASE feature specification matching the argument type in markedness in intransitive contexts (rather than the CASE feature specification that is unmarked); it differs from the accusative pattern in (1a) in encoding DPext of Vi by accusative. This type of encoding pattern does not seem to occur. However, there is an obvious problem with (13) (see Bechert 1979): In an anti-active pattern, there is not a single implicational relation between CASE feature specification and argument type. Consequently, the pattern is extremely dysfunctional and poses severe difficulties for language acquisition.27 (13)

Anti-active accusative marking DPext –Vi

DPint –Vi

DPext –Vt

DPint –Vt

NOM

ACC

Finally, there could be a third principled way of selecting a unique CASE feature in intransitive contexts in order to respect (9) (next to instantiation of the unmarked CASE feature, and instantiation of the iconic CASE feature). Suppose that, in an accusative system (that gives priority to Agree over Merge on the vP level), the internal CASE feature shows up with internal arguments, and the external CASE feature shows up with external arguments; i.e., there is a matching requirement not with respect to markedness (marked argument type corresponds to marked CASE feature), but with respect to status (external argument corresponds to external CASE feature). This will give rise to an active accusative pattern, as in (14): v’s CASE feature not only determines the CASE of the internal argument in transitive contexts; it is also selected for internal arguments in intransitive contexts. In contrast, T’s CASE feature encodes the external argument both in transi27

This may imply that the pattern is in principle available to the language faculty but unusable in practice, or that optimal design restrictions prevent it from being available in the first place (e.g., by requiring unambiguity of argument identification); I will leave this open.

Ergativity, accusativity, and the order of Merge and Agree

283

tive contexts (where v’s CASE feature has already been discharged when the external argument is merged), and in intransitive contexts (the external CASE feature is preserved for an external argument). (14) Active accusative marking DPext –Vi DPext –Vt

DPint –Vi DPint –Vt

NOM

ACC

Such an encoding pattern is distributionally identical to an active ergative pattern. Evidence distinguishing an active ergative from an active accusative pattern will then have to come from independent sources (among them morphological realization and syntactic ergativity). Plausible candidates for languages instantiating such an active accusative pattern include Eastern Pomo (extinct; Hokan, California) and Acehnese (Austronesian, Northern Sumatra); see Bittner & Hale (1996b). Relevant examples from Eastern Pomo are given in (15). (15) a. Míip míip-al sáaka. he.NOM him-ACC killed ‘He killed him.’ b. Míip-al xáa baakúma. him-ACC in the water fell ‘He fell in the water (accidentally).’ c. Míip káluhuya. he.NOM went home ‘He went home.’ We have seen above that the preservation of the iconic CASE feature leads to a dysfunctional anti-active encoding pattern in accusative language types. Similarly, the question arises as to what happens with a preservation of the status-matching CASE feature in ergative language types. The answer is straightforward: This way, an anti-active ergative encoding pattern will arise that is just as dysfunctional as its accusative counterpart; see (16). (16)

Anti-active ergative marking DPext –Vi

DPint –Vi

DPext –Vt

DPint –Vt

ERG

ABS

284 Gereon Müller Thus, we may conclude that there are three major strategies to comply with requirement (9) in intransitive contexts: Either the unmarked CASE feature is preserved in numerations (which derives the canonical ergative and accusative marking patterns); or the CASE feature is maintained that matches the sole remaining argument DP in either iconicity (yielding an active ergative pattern and a dysfunctional anti-active accusative patten) or status (yielding an active accusative pattern and a dysfunctional anti-active ergative pattern). One may speculate that the latter two strategies represent more marked options.28 5. Further cases of split ergativity There are other instances of split ergativity, in addition to an active marking pattern – most notably, tense/aspect-based split ergativitiy, as in Burushaski (see Dixon 1994: 97–101), Georgian (see Comrie 1978; Nash 1996; Ura 2006), and Hindi (see Mahajan 1990); clause-type based split ergativity, as in Sierra Popoluca (see Elson 1960); and person-based split ergativity, as in Dyirbal (see Dixon 1994). The goal of this section is to show how these further cases of split ergativity can be integrated into the present analysis. To this end, I will first present a minimally different version of the approach 28

It has sometimes been suggested that active encoding patterns do not actually exist but rather represent instances of systematic, hidden transitivity. See, e.g., Bobaljik (1993), Laka (1993), Nash (1996), and Bittner & Hale (1996a,b), among others. On this view, what looks like an intransitive context with ergative marking in languages like Basque, Guaraní, Hindi, and Georgian is really a transitive context with a hidden internal argument (which can, e.g., be overtly realized as a cognate object); sometimes the presence of the internal argument is suggested by overt agreement morphology (e.g., in Basque). Pure ergative encoding patterns, as in languages like Archi, Yupʹik, and Sierra Popoluca, may then arise because the internal argument of the relevant verbs is incorporated into V; as a result, V would become intransitive. I am skeptical that such an approach can cover all cases of active encoding patterns. (It is also instructive to note that a concept of hidden transitivity is not invoked in optimality-theoretic approaches like Kiparsky (1999), Stiebels (2000; 2002), Wunderlich (2000), Woolford (2001), and Lee (2003), where deriving active patterns from the theory does not pose a problem but rather follows straightforwardly from factorial typology.) However, it is worth pointing out that a hidden transitivity approach to active encoding patterns would be compatible with the present approach without further ado if we were to assume that only the option of preserving the unmarked CASE feature (i.e., the CASE feature of T) is available in intransitive contexts in the numeration.

Ergativity, accusativity, and the order of Merge and Agree

285

laid out in section 2 above; the reason for this will become clear when I turn to tense/aspect-based split ergativity in Hindi after that.

5.1. Feature hierarchies Recall the Agree/Merge indeterminacy problem at stage Σ on the vP level: v has both a subcategorization feature [•D •] that introduces the external argument, and a CASE feature [*CASE:int*] that is responsible for the realization of internal CASE (ergative/accusative). Given Earliness requirements for Merge and Agree, a conflict arises, and this can be resolved by giving priority to one of the two Earliness requirements – which one, is subject to language-specific parametrization.29 As noted, this amounts to a local optimization procedure in syntax: Either the Earliness requirement for Merge outranks the Earliness requirement for Agree, or vice versa; and the lowerranked requirement can be minimally violated (by postponing the respective operation by one derivational step). However, there is an alternative way of implementing the two possible solutions to the intederminacy problem at stage Σ that does without constraint ranking and constraint violability; one that may in fact be independently motivated. Consider the question of how linking is brought about in current derivational syntactic theories, i.e., how the order of Θ-roles in argument structures of predicates in the lexicon is mirrored by argument realization in syntax.30 For examples like (17), the task is to derive that John1 is interpreted as AGENT, a book2 as THEME, and Mary3 as GOAL . (17) John1 gave a book2 to Mary3 One possible strategy is to assume that syntactic positions are provided with fixed argument structure correlates, such that, e.g., a DP daughter of vP is interpreted as AGENT, a DP daughter of VP is interpreted as THEME , and a 29

30

Or, as argued in Heck & Müller (2007), phase-specific parametrization – i.e., the resolution of the indeterminacy on the vP level may differ from the resolution of an analogous indeterminacy on the CP level, or on the DP level, within a single language. I am assuming here a standard approach to argument structure that relies on ordered lexical entries which predict a hierarchy of Θ-roles (see, e.g., Bierwisch 1988; Wunderlich 1997; and Heim & Kratzer 1998). Things would be different under radical approaches like the one developed in Borer (2004), where argument structures are not basic, but are read off syntactic representations.

286 Gereon Müller PP daughter of V ʹ is interpreted as GOAL; such an approach is adopted in Adger (2003). On this view, an example like (18) is classified as ungrammatical because it emerges as uninterpretable: The VP-internal arguments cannot be associated with Θ-roles. (18) *[vP John gave+v [VP to Mary tV a book ]] AGENT ?? ?? However, it is not really clear that this is the right approach to take: A sentence like (18) can perfectly well be understood by speakers of English – it just happens to violate formal rules of grammar. I would therefore like to adopt another approach to linking in minimalist grammars, one that has its roots in categorial grammar and has a long tradition in transformational and phrase structure grammar (see Lewis 1972; Pollard & Sag 1994; Wunderlich 1997; and Lechner 2004, among many others). According to this alternative approach to linking, Θ-roles are ordered in lexical entries of predicates, and they are mapped to a list of categorial subcategorization features (i.e., structure-building features: [•F•]) in reverse order, as schematically depicted in (19). There must then be a condition ensuring that subcategorization features are discharged one after the other, beginning at the top of the hierarchy (see below for a specific suggestion). (19) a. Θ-roles: Θ2 ≫Θ2≫Θ3

(Agent≫Theme≫Goal)

b. Subcategorization features: [•P•]3 ≻ [•D•]2 ≻ [•D•]1 In addition, I assume that there can also be subcategorization features that do not go back to Θ-grids. These kinds of subcategorization features play a role in structure-building operations involving functional categories, and they also enter argument lists of lexical items. For example, VP is not Θmarked by v, but v has a subcategorization feature [•V•] that permits Merge of v and VP; and similar reasonings apply in the case of Merge (T,vP) and Merge (C,TP). At this point, the question arises of how an external argument DP comes to be base-generated in the specifier of v. Based on Larson (1988) and subsequent work, two basic options have been pursued. One possibility (adopted in approaches that correlate v with causation or some related semantic concept; see Hale & Keyser (2002), Harley (1995), Kratzer (1996), Adger (2003), Ramchand (2003), Schäfer (2007), and references cited in these works) is that AGENT DPs are introduced by v. In the present

Ergativity, accusativity, and the order of Merge and Agree

287

approach, this would amount to a feature list [•V•] ≻ [•D •] on v, with V discharged by Merge (v,VP) before D is discharged by Merge (vʹ,DP). On this view, transitive verbs take only one argument; and ditransitive verbs subcategorize for two (rather than three) arguments. Alternatively, v is motivated solely by considerations pertaining to phrase structure geometry; external arguments remain true arguments of V, even though they are merged in the specifier of v (see Haider 2000, 2006, for an analysis along these lines, and Georgi & Müller 2007, for an implementation of this idea in terms of head movement by reprojection). Here I adopt the former view. It can then be concluded that v has two subcategorization features that trigger Merge – one for the VP, and one for the external argument DP –, and it has a probe feature that triggers Agree with an argument that gets encoded by v.31 Subcategorization features ([•F•]) must be ordered on lexical items (like v), i.e., they are part of a stack. How do probe features ([* F*]) of a lexical item (say, v) fit into this? Suppose that they can in principle be interleaved with structure-building features ([•F•]) on a lexical item; i.e., it is not necessarily the case that all subcategorization features must be checked before the first probe feature of a lexical item can be checked (however, a probe feature on top of feature stack is always going to lead to a crash of the derivation because it can never be discharged – at the point where a topmost probe feature must be checked, there is no complement – let alone specifier – yet in which it could possibly find a goal). As noted, given the existence of feature stacks associated with lexical items (possibly involving a mix of subcategorization features and probe features), there has to be a condition that states that only the highest feature on the feature hierarchy is visible for Agree or Merge operations at any given stage of the derivation. This condition can be formulated as part of a general Earliness requirement; cf. (20). (20) Earliness: a. An operation-inducing feature ([•F•] or [*F*]) must be checked immediately. b. Only the topmost operation-inducing feature ([•F•] or [*F*]) of a lexical item is visible. Thus, given (20), only the highest feature on the feature hierarchy is visible for Agree and Merge at any given stage of the derivation. The concept of 31

Of course, v has two such subcategorization features only in clauses in which an external argument is present, not in intransitive unaccusative contexts. See Chomsky (2001) for a systematic distinction between the two types of v.

288 Gereon Müller feature checking that is required for this analysis involves the notion of discharge: A probe or subcategorization feature disappears (or becomes inert) after having triggered an operation (Merge or Agree), and thus makes room for the next-lower feature on the hierarchy. On the basis of these assumptions, the ergative/accusative parameter does not rely on different rankings of the Earliness requirements for Merge and Agree; rather, there is only one Earliness requirement, and the parameter is related to the order in which v’s [*CASE:int*] probe feature and [•D •] subcategorization feature show up on v’s stack of operation-inducing features. This is shown in (21).32 (21) Ergative/accusative parameter: a. b.

[*CASE:int*] ≻ [•D •] on v: accusative encoding pattern [•D •] ≻ [*CASE:int*] on v: ergative encoding pattern

If some language is equipped with a v on which the order is [*CASE:int*] ≻ [•D •], Agree with a VP-internal DP takes place before the external argument DP is merged, and an accusative pattern arises. If, on the other hand, the order on v is [•D •] ≻ [*CASE:int*], an external argument DP is merged first (after v has been merged with VP), and the newly formed specifier turns out to provide the closest goal for v’s [*CASE:int*] feature whose turn it is next to be discharged (see (7c)), yielding an ergative encoding pattern. So far, this reanalysis might look like a notational variant of the approach laid out in section 3 above.33 However, unlike the earlier approach, the new 32

33

Note that these hierarchies are only partial. The subcategorization feature that triggers Merge of v and VP (i.e., [•V•]) must invariantly show up on top of both [*CASE:int *] and [•D•] in v’s feature stack. This is so for two reasons. First, VP must be merged before an external argument DP is merged in vP. And second (as noted above), probe features cannot be topmost for principled reasons on a lexical item. This excludes four of the six logically possible hierarchies (indicated here by #), leaving only the two hierarchies which correspond to those in (21). See (i:a–b) vs. (i: c–f). (i) a. [•V•] ≻ [*CASE:int *] ≻ [•D•] b. [•V•] ≻ [•D•] ≻ [*CASE:int *] c. # [*CASE:int *] ≻ [•V•] ≻ [•D•] d. # [•D•] ≻ [•V•] ≻ [*CASE:int *] e. # [*CASE:int *] ≻ [•D•] ≻ [•V•] f. # [•D•] ≻ [*CASE:int *] ≻ [•V•] Still, it might be worth noting that this view of the basic ergative/accusative parameter would be more straightforwardly compatible with approaches that do not envisage constraint ranking as a legitimate concept than the analysis I have

Ergativity, accusativity, and the order of Merge and Agree

289

analysis turns out to make a simple account of tense/aspect-based split ergativity possible.

5.2. Tense/aspect-based split ergativity In Hindi, a nominative-accusative pattern shows up in the imperfective aspect, whereas an ergative-absolutive pattern occurs in the perfective aspect; see Mahajan (1990), Mohanan (1994), Dixon (1994), Woolford (2001), Anand & Nevins (2006), Ura (2006) and Keine (2007), among many others. The following data are taken from Mahajan (1990).34 (22) a. Raam toTii khaataa thaa. Ram.MASC-NOM bread.FEM-ACC eat.IMP.MASC be.PAST.MASC ‘Ram (habitually) ate bread.’

34

adopted so far. Basically, under the present perspective, all parametrization could be assumed to be located in the properties of functional heads; see Borer (1984), Chomsky (1995). In addition, Hindi exhibits a version of active alignment in perfective contexts: In intransitive contexts, sole arguments of unaccusative verbs are absolutivemarked; in contrast, sole arguments of unergative verbs are ergative-marked – but only (according to standard wisdom) if the external argument can be interpreted as volitional. Compare (i: a) with (i:b) (volititional vs. non-volitional interpretation with an unergative verb), and with (i: c) (unaccusative verb); see Mohanan (1994), Lee (2003). (i) a. (Raam-do acaanak šer dikhaa.) Vah/*us-ne cillaayaa. (Ram-DAT suddenly lion-NOM appear.PERF) he-NOM/*he-ERG scream.PERF ‘Ram suddenly saw a lion. He screamed.’ b. Us-ne/*vah jaanbuujhkar cillaayaa. he-ERG/*he-NOM deliberately shout-PERF ‘He shouted deliberately.’ c. Raam/*Raam-ne giraa. Ram-NOM/*Ram- ERG fall-PERF ‘Ram fell hard.’ I have nothing to say here about the question of whether the data in (i) could be integrated directly into what is said about active ergative marking in section 4 (e.g., by assuming the non-volitional subject in (i: a) to be VP-internal after all), or whether they require some (perhaps non-trivial) extension of the present system.

290 Gereon Müller b. Raam-ne roTii khaayii thii. Ram-ERG bread-ABS eat.perf.FEM be.past.FEM ‘Ram had eaten bread.’ Assuming that the ergative/accusative parameter does not reduce to language-specific ranking of Earliness requirements for Merge and Agree, but different feature hierarchies for [•D •] and [*CASE:int*] on v in a language, a simple solution offers itself for aspect-based split ergativity in Hindi (and related phenomena in other languages): Suppose that a language is equipped with two types of v – one on which the order is [•D •] ≻ [*CASE:int*] (v1, giving rise to an ergative encoding pattern), and one that has a feature stack including the order [*CASE:int*] ≻ [•D •] (v2, which produces an accusative encoding pattern). Then, the co-occurrence of the two patterns in a single language is to be expected. The only thing that remains to be ensured for Hindi is that the right kind of v is chosen in the right kind of aspect. This is a simple matter of selection. Given that aspectual information is located on T (though one could just as well postulate a separate functional category Asp between T and v), a perfective T selects v1, and an imperfective T selects v2. For the sake of concreteness, this can be achieved if it is assumed that T has a subcategorization feature [•v•], and this feature is accompanied by a probe feature [*1*] if T is [+perf], and by a probe feature [*2*] if T is [–perf]; corresponding goal features [1], [2] characterize the two v items of the language.35 In contrast, note that the version of the present approach to ergative vs. accusative patterns of argument encoding developed in the previous part of this paper does not permit such a simple approach to tense/aspect-based split ergativity: One would have to assume that Earliness requirements are not 35

This is essentially as in Gazdar et al. (1985). Alternatively, instead of using diacritics that mediate between T’s aspectual feature and the order of probe and subcategorization features on v’s feature stack, one might pursue the hypothesis that T can directly “see” what v’s feature stack looks like, and select v appropriately. If one assumes that operation-inducing features on v are discharged by the operation they trigger, selection of v by T will presumably have to be handled in the numeration, rather than in the syntax (when T merges with vP, v has already lost its operation-inducing features); e.g., via co-occurrence restrictions. Alternatively, operation-inducing features may simply become inert after having been checked, in the sense that they make the next-lower feature accessible (see (20)) but are still visible for selection. Resolving these technical issues in one or the other way does not bear on the basic analysis, though.

Ergativity, accusativity, and the order of Merge and Agree

291

associated with grammars as a whole, or with phases (see Heck & Müller 2007 and footnote 29); rather, they would have to be specified for individual lexical items, distinguishing between two types of v in the case at hand. Such an assumption would certainly be far from impossible, but it might be viewed as somewhat less attractive than the present solution. Under the present approach, then, ergative and accusative turn out to be realizations of the very same CASE feature in Hindi. Still, this does not imply that they will have to be realized in the same way. Indeed, they are realized differently in Hindi – by -ne in one case, and by Ø or -ko in the other case.36 One and the same abstract CASE feature [*CASE:int*] may well be realized as x in context A (external argument DP), and as y in context B (internal argument DP), given that morphological spell-out can depend on contextual information (see Halle & Marantz 1993; Bobaljik 2000; and Trommer 2003, among many others). It seems likely that this kind of approach can be generalized so as to cover other instances of tense/aspect-based split ergativity; however, the question requires close investigation for each case that has been suggested in the literature.37 36

37

The conditions for choosing among the latter allomorphs are related to general laws of differential object marking. In this context, it is worth noting that it might in fact ultimately be possible to reduce all variation in CASE marking in Hindi to issues of morphological realization, on the basis of the assumption that there is but a single syntactic encoding pattern. See Keine (2007) for an analysis along these lines, with impoverishment operations in a Distributed Morphology approach to morphological realization taking over much of the work that is standardly, and also in the present section, attributed to differences in syntactic CASE assignment. For instance, tense-based split ergativity in Georgian or Burushaski would follow under basically the same kinds of assumptions: If T is [+past], a v is selected that exhibits the order [•D•] ≻ [*CASE:int*]; if T is [–past], it requires a v with the reverse hierarchy [*CASE:int*] ≻ [•D•], which predicts an ergative pattern in the former case, and an accusative pattern in the latter. This analysis may be straightforward for Georgian, but it turns out that the situation might not be quite as simple in Burushaski: In Burushaski, there is no marking whatsoever in (what would be predicted as) the accusative encoding pattern (Dixon 1994: 100). Therefore, I take it to be an open question whether Burushaski does indeed involve a true syntactic split in encoding patterns, or whether postulating a uniform ergative pattern, with systematic non-realization of the ergative CASE feature in non-past contexts in the morphological component, might offer a more adequate analysis. See the previous footnote.

292 Gereon Müller From a more general point of view, it can be noted that the present analysis leaves open the question of whether there is an inherent relation between, say, past tense (or perfective aspect) and an ergative encoding pattern on the one hand, and non-past tense (or imperfective aspect) and accusative encoding pattern on the other. If such a correlation (as it is suggested, e.g., by Dixon 1994) can ultimately be shown to exist (to determine this, a large sample of languages would have to be investigated), this might arguably imply that the present approach would have to be modified so as to capture the generalization. For the time being, it may suffice to note that existing approaches to tense/aspect-based split ergativity generally do not seem to fare much better with respect to the issue of deriving the putative generalization.38 38

Dixon’s (1994) own account remains largely impressionistic, drawing analogies to the typical structure of crime novels. In optimality-theoretic approaches, a uniform pattern arises: Taking the case of Hindi as a representative example, all existing analyses seem to agree that the effect is due to a high-ranked constraint that is active in perfective contexts only, and that ensures that an (otherwise expected) accusative encoding pattern is suppressed in these contexts in favour of an ergative encoding pattern. Thus, Woolford’s (2001) high-ranked constraint FAITH-LEXperf demands that ergative case (assumed to be inherent on a certain class of verbs) is realized in an optimal output; this constraint outranks an opposite constraint *ERG that requires ergative to be absent throughout (it does so successfully in imperfective contexts for which there is no high-ranked FAITH-LEX constraint). Similarly, in Kiparsky’s (1999) analysis there is a high-ranked constraint Max([+lr])/[+perf] with the same effect: [+lr] stands for the ergative case, and the constraint demands that this CASE be realized in perfective environments. (Kiparsky’s inventory of case features is actually slightly different, but this does not affect the point to be made here.) Given that the general constraint against ergative CASE is ranked lower, ergative is successfully imposed in perfective contexts. Yet another version of this optimality-theoretic reasoning can be found in Lee (2003): A ranking ERGperf ≫ *ERG confines ergative CASE to perfective environments in Hindi. Finally, Stiebels (2000, 2002) presents an analysis that is minimally different in that the basic logic is reversed: *[+lr] /[– perf] (which prohibits ergative marking in imperfective contexts) outranks a general constraint stating that external arguments in transitive clauses are always marked ergative; this latter, lower-ranked constraint can then only spring into action, and induce ergative CASE, in perfective contexts. As for alternative minimalist approaches, Ura (2006) proposes an analysis that looks fairly simple at first sight: The relevant tense/aspect information responsible for ergative CASE is assumed to be located on v (rather than on T). As a matter of fact, Ura suggests that the CASE feature associated with v is the tense/aspect-related feature; so ergative/accusative in effect is tense/aspect of v on DP (see Williams 1994 and Pesetsky & Torrego 2001 for related ideas). In my

Ergativity, accusativity, and the order of Merge and Agree

293

5.3. Clause type-based split ergativity As shown above (see (5)), Sierra Popoluca exemplifies a relatively canonical system of ergative CASE encoding by agreement (head marking). Still, as noted by Elson (1960: 208) and Marlett (1986: 364), in certain kinds of clauses this system is replaced with an accusative pattern of argument encoding. More specifically, this affects temporal adjunct clauses that do not have a Spanish complementizer (the native temporal complementizer is mu (‘when’), which alternates with Spanish cuando (‘when’)); and to some extent also clauses embedded under intransitive verbs. Consider (23a,b), which involves an accusative pattern in embedded clauses introduced by mu. (23) a. mu an-nɨk when 1.NOM-go ‘when I went’ b. mu an-iš caːñ when 1.NOM-see snake ‘when I saw a snake’ Here, an (internal or external) argument of an intransitive predicate is marked in the same way as the external argument of a transitive verb. This can be taken to imply that, as we have just seen for Hindi (Georgian, Burushaski), two types of v are present in the language: As the default case, v1 is view, the association of inherent tense/aspect features with v may be considered somewhat dubious; in a system of clausal structure that envisages only two functional heads below C, it should surely be T, not v, where this kind of information is located (also see footnote 7 of Ura 2006: 141). Alternatively, one might assume that whereas tense/aspect information is inherently located on T, the relevant features may also show up on v as a result of Agree (see, e.g., Adger 2003, who argues for this on the basis of morphological realization). However, one would then either have to give up the assumption that probes are non-interpretable (tense/aspect is interpretable on T, and it would have to be uninterpretable on v), or one would have to abandon the hypothesis that probes c-command goals (if v acts as the probe); note that these are assumptions that can be maintained in the present analysis despite adopting a checking (as opposed to valuation) approach. Furthermore, if tense/aspect on v is agoal feature (with tense/aspect onT an interpretable probe), this feature would have to have a dual status; for CASE assignment to a DP, it would have to be a probe in addition. – Finally, in addition to these problems, which are primarily of a conceptual and/or technical nature, there is the empirical problem that Ura’s analysis requires additional assumptions for languages where ergative CASE does not depend on perfective aspect or past tense.

294 Gereon Müller chosen from the numeration in Sierra Popoluca; on v1, the relevant operationinducing subcategorization and probe features are ordered in such a way that they produce an ergative encoding pattern (viz., [•D•] ≻ [*CASE:int*]). However, the language is also equipped with a second v, v2, which induces an accusative marking pattern (via the feature hierarchy [*CASE:int*] ≻ [•D•]); v2 is selected by certain kinds of C (viz., either one with the features [+native, +temporal], or one that can be embedded under intransitive verbs). Thus, the account is fully parallel to the one given for tense/aspect-based split ergativity in the preceding section; the only relevant difference is that this time, it is the type of C node (rather than the type of T node) that determines which v is selected – one that induces ergative marking or one that induces accusative marking. At this point, the technical issue arises of how selection of v by C can be accomplished. However, since several plausible possibilities suggest themselves (options include selection mediated by T; selection via feature co-occurrence restrictions that apply in the numeration; and direct selection of one phase head by another, cf. Chomsky 2005a), I will not dwell on this issue here.39 Similar clause type-based splits have been reported for other languages (see Dixon 1994: 101–104), and similar analyses can presumably be given on the basis of the present approach.40 39

40

Note incidentally that the markers that Sierra Popoluca uses for ergative CASE in default contexts correspond to those it uses for nominative CASE in embedded temporal clauses with a native complementizer; similarly, the markers for absolutive CASE in standard environments and for accusative CASE in the restricted context are morphologically identical. At first sight, this might seem to favour theories of argument encoding in which ergative is the same case as nominative, and absolutive the same case as accusative; see Levin & Massam (1985), Chomsky (1995: ch.3), Bobaljik (1993), and other references in footnote 21. This potential problems disappears if morphological realization can take into account contextual information (in addition to the respective case features themselves). In addition, note that the markers used for ergative CASE and for nominative CASE also show up DP-internally as exponents of possession; so they are clearly not confined to a single syntactic domain (like the vP). Still, integrating this morphological evidence would require a reanalysis of the approach to exponence of argument encoding given in Müller (2006). For instance, in Päri (Nilotic), there is generally an ergative argument encoding pattern, but in purposive clauses, an accusative pattern arises, with the sole (internal or external) argument of an intransitive verb marked in the same way as the external argument of a transitive verb. Again, it seems likely that the relevant difference can be tied to a property of C.

Ergativity, accusativity, and the order of Merge and Agree

295

5.4. Three-way systems Before I turn to person-based split ergativity, “three-way” languages need to be addressed. Here, ergative CASE and accusative CASE seem to co-exist together with nominative (absolutive) CASE; see, for instance, Dixon (1994), Bittner & Hale (1996b), Woolford (1997), Kiparsky (1999), and Stiebels (2002). The following data from Antekerrepenhe (Central Australia) in (24) and Nez Perce (Penutian) in (25) come from Bittner & Hale (1996a: 51–52) (also see Woolford (1997) for a detailed discussion of Nez Perce). (24) a. Arengke-le aye-nhe ke-ke. dog-ERG me-ACC bite-PST ‘The dog bit me.’ b. Apwerte-le athe arengke-nhe we-ke. stones-INS I-ERG dog-ACC pelt-PST ‘I pelted the dog with stones.’ c. Arengke nterre-ke. dog-NOM run-PST ‘The dog ran.’ (25) a. Wewúkiye-ne pée-’wi-ye háama-nm. elk-ACCj 3i.3j -shoot-PRF man-ERGi ‘The man shot an elk.’ b. Hi-páayn-a háama. 3i-arrive-PRF man-NOM ‘The man arrived.’ Since the present approach locates the source for ergative and accusative in the CASE feature of one and the same head (viz., v), it might seem at first sight as though three-way languages pose a severe problem for a syntactic analysis (that identifies two distinct cases, ergative and accusative) in the present framework.41 In view of this state of affairs, one might contemplate 41

In contrast, optimality-theoretic approaches like those mentioned above predict three-way languages to be entirely unmarked, on a par with regular ergative and accusative argument encoding patterns. Simplifying a bit, these kinds of analyses typically envisage two constraints inducing CASE marking – one that requires ergative marking for external arguments of transitive verbs, and one that requires accusative marking for internal arguments of transitive verbs (call these constraints E and A). In addition, there is a conflicting constraint that blocks re-

296 Gereon Müller assigning the languages in question a syntactically simple (ergative or accusative) marking pattern, where either the ergative is reanalyzed as a nominative in an accusative system, or the accusative is reanalyzed as an absolutive in an ergative system, and the apparent ergative/nominative alternation or accusative/absolutive alternation is accounted for in the morphological component (see, e.g., Murasugi 1992). For the sake of concreteness, suppose that three-way systems are basically accusative systems (recall that an absolutive of ergative encoding patterns is typically morphologically unmarked on a DP, whereas a nominative of accusative encoding patterns can be marked on a DP). Then, we have to assume that there are nominative allomorphs, e.g., le and Ø in Antekerrepenhe. In Distributed Morphology (Halle & Marantz 1993), inflectional morphology is post-syntactic: It realizes morpho-syntactic feature bundles that are present in X0 positions in the syntax. Between syntax and morphology, morpho-syntactic features are sometimes deleted; such a deletion is brought about by impoverishment rules. Suppose now that Antekerrepenhe has an impoverishment rule like (26). (26) [CASE:EXT] → Ø/__v[¬∗case∗] According to (26), nominative case cannot be morphologically realized on an external or internal argument of v[¬∗case∗], i.e, of a v that does not have a 42 CASE feature to assign. Note that this would imply that checked probe features become inert but do not delete – otherwise a post-syntactic morphology could not distinguish between a v that does not assign CASE in the first place because it shows up in an intransitive structure, and a v that has checked CASE with an internal argument. One might argue that such an analysis gives rise to a potential conceptual problem in that what arguably

42

alization of CASE marking (*C). Consquently, factorial typology implies that there can be four rankings that produce four basic grammar types: (i) A ranking E ≫ *C ≫ A produces a standard ergative pattern; (ii) a ranking A ≫ *C ≫ E gives rise to an accusative pattern; (iii) *C ≫ E, A results in no CASE marking for either external or internal argument of a transitive verb; (iv), and finally, a ranking E, A ≫ *C creates a three-way language. Arguably, such a treatment fails to reflect the highly marked status of three-way languages cross-linguistically. Thus, such an approach to three-way language types would be similar in all relevant respects to the impoverishment-based approach to argument encoding in Hindi developed in Keine (2007) that was mentioned above.

Ergativity, accusativity, and the order of Merge and Agree

297

looks like a functional motivation for ergative encoding in the syntax is mimicked in morphology. I do not think that this problem is a real one, though; quite on the contrary, such an approach would give a functional motivation to impoverishment operations, and thereby strengthen the overall concept. However, a morphological approach may raise other problems that are real; Legate (2005; 2006) (for Warlpiri, Djapu, and Kugu Nganheara) and Bierkandt (2006) (for Diyari) argue that there is evidence for the syntactic presence of three cases. For this reason, I take it to be worth pursuing the question of whether the present approach could be extended to cover three-way systems in syntax after all. As it turns out, this is very well possible. Proceeding on the assumption that three-way languages are a marked phenomenon, we can assume that what is marked is that v can exceptionally be equipped with two [*CASE:int*] features. In transitive contexts, there are thus three potential CASE features: [*CASE:ext*] on T, and two identical CASE features on v: [*CASE:int*], [*CASE:int*]. In such a scenario, the very same dilemma arises that we have seen to arise with two CASE features ([*CASE:ext*] on T, [*CASE:int*] on v) in intransitive contexts: There is one more CASE feature than there are argument DPs for checking. Again, one of the CASE features has to disappear. As before, suppression of the surplus feature is forced by a general constraint on numerations. Feature Balance is repeated here from (9). (27) Feature Balance: For every feature specification [*F:α*], there must be a matching feature specification [F:α]. Which of the three features (two of them identical) has to go? It seems plausible to assume that [*CASE:int*] on v can never go away if there are both DPext and DPint in the numeration (see Marantz 1991, Bittner & Hale 1996b, and Wunderlich 1997, where [*CASE:int*] is in fact defined by the presence of an external argument c-commanding an internal argument). Consequently, if there are two [*CASE:int*] features present on v, [*CASE:ext*] cannot be instantiated on T in the numeration in transitive contexts (alternatively, it is deleted; see above). As a result, in accusative as well as ergative systems, DPext and DPint must both bear [CASE:int] in transitive contexts. However, in intransitive contexts, only the unmarked case feature is instantiated: [*CASE:ext*] shows up on T (and no [*CASE:int*] feature can show up on v, let alone two of them). To sum up this subsection, three-way languages can either be integrated into the present approach by pursuing the hypothesis that closer inspection

298 Gereon Müller may reveal the underlying systems to involve canonical ergative or accusative encoding patterns after all (with the apparent three-way distinctions accounted for morphologically, e.g., by impoverishment); or they can be handled by assuming that v may exceptionally bear two CASE features. Needless to say, the two possibilities are not mutually exclusive: The morphological approach may well be correct for some languages, and the syntactic approach for others.

5.5. Person-based split ergativity Let me end this paper with some brief speculations on how to analyze person-based split ergativity. The phenomenon that has received most attention in the literature is person-based split ergativity in Dyirbal (see Dixon 1972, 1994). In Dyirbal, an external argument DP of a transitive verb is marked ergative if it is a 3rd person pronoun or an item to the right of it on the person/animacy scale in (28) (see Silverstein 1976 for the original scale, and Aissen 1999, 2003, for a systematic reconstruction). In contrast, an internal argument DP of a transitive verb is marked accusative if it is a 1st or 2nd person pronoun. All other types of argument DP fail to be encoded by an overt marker. (28) Person/animacy scale: 1st person pronoun > 2nd person pronoun > 3rd person pronoun > proper name > common noun, human > common noun, animate > commoun noun, inanimate At least at first sight, Dyirbal would thus seem to belong to the class of languages that exhibit three-way systems, as discussed in the previous section. In line with this, the strategies for analysis are comparable: A person-based split may in principle be either a syntactic or a morphological phenomenon. Such splits often seem to be functionally motivated in the sense that only unexpected, atypical configurations are marked. A morphological analysis could look as follows: There is a zero allomorph Ø that results from impoverishment, where impoverishment rules might be motivated by functional considerations incorporating the Silverstein hierarchy in (28). Such an approach would seem to suggest itself for simpler cases of person-based argument encoding systems, among them most of the cases of differential subject and object marking discussed in Aissen (1999, 2003). For Dyirbal, it would imply postulating a two-case system, with impoverishment opera-

Ergativity, accusativity, and the order of Merge and Agree

299

tions that preclude morphological realization of CASE features on prototypical external and internal arguments; on this view, the CASE features are only retained (for morphological realization) on 3rd person pronouns (and lowerranked items on the person/animacy scale) that are external arguments of transitive verbs, and on 1st or 2nd person pronouns that act as internal arguments of transitive verbs; in all other environments, CASE features would be deleted. Alternatively, one may adopt a syntactic approach that envisages three structural CASES (ergative, accusative, and nominative/absolutive). This would basically be as in the cases discussed in the previous section. However, there would be an additional complication: With simple three-way systems, the only relevant factor determining which CASE features are present is whether the verb is transitive or intransitive. In addition to this difference between transitive and intransitive verbs, CASE feature instantiation in numerations in Dyirbal would have to be sensitive to properties of the arguments (i.e., their rank on the person/animacy scale). Technically, the required rules for CASE feature instantiations in the numeration in languages like Dyirbal do not pose any problem whatsoever; but since I take it to be far from clear that a simpler, purely morphological approach is excluded, I will not pursue the matter any more at this point.

6. Concluding remarks Let me sum up the main claims of this paper. I have suggested two parameters for argument encoding which are active in different components of grammar: Whether a language employs accusative marking or ergative marking is decided in the syntax, by resolving an indeterminacy in the order of the elementary operations Merge and Agree on the vP cycle. Whether a language employs active marking or not is decided in the numeration, i.e., pre-syntactically, by resolving an indeterminacy in the preservation of CASE feature specifications in intransitive contexts via markedness (choice of the unmarked feature specification), iconicity (choice of the feature specification with the same markedness status as the argument) or status-matching (choice of the feature specification matching the status – external vs. internal – of the argument). To the extent that this approach succeeds in capturing the main argument encoding patterns in a simple way, it has potential repercussions on the theory of parametrization: In the original version, it suggests that parametric variation cannot exclusively be due to variation in the (functional)

300 Gereon Müller lexicon, and that there must be a limited decision space in the application of elementary operations in the syntax and in the numeration. However, it has turned out that extending the analysis to cases of tense/aspect-based (or clause-based) split ergativity might suggest a minimal modification of the general approach that ties all parametric variation to properties of lexical items (v, in the case at hand) after all. If this latter approach is on the right track, a more conservative approach to cross-linguistic variation can be maintained. Furthermore, and independently of which of the two basic versions of the general hypothesis is adopted (ranking Earliness requirements for Merge and Agree vs. diverging feature hierarchies on v), it is worth emphasizing that the present analysis crucially depends on an incremental-derivational approach to syntax; the difference between an accusative and an ergative pattern boils down to whether DPext is or is not yet part of the structure when the CASE feature of v triggers Agree. To end this paper, let me stress that what precedes is not intended to be a comprehensive theory of accusative vs. ergative patterns of argument encoding; for this, the analysis would have to be extended in various directions. I will here confine myself to mentioning two of them. First, in this paper I have only been concerned with ‘morphological ergativity’ (i.e., formal argument encoding by case or agreement), and not at all with ‘syntactic ergativity’ (i.e., cases where DPext and DPint of Vi, and DPint of Vt are systematically treated on a par in the syntax by processes different from argument encoding); see Comrie (1989), Bobaljik (1993), Dixon (1994), and Bittner & Hale (1996a), among others. And second, movement to SpecT has been neglected in the preceding discussion; but this operation seems to be related to the nature of argument encoding to some extent. Possibly, these two issues ultimately reduce to the question of which argument qualifies as most ‘prominent’ (i.e., which argument acquires ‘subject’ or ‘pivot’ properties). Relevant factors here include (a) external CASE and (b) external argument status. With accusative marking patterns, the two properties usually converge (with notable exceptions like Icelandic quirky case subjects); with ergative marking patterns, they usually diverge, and this may plausibly be taken to underlie the substantial variation in promotion to subject status that can be observed in the are of languages employing an ergative argument encoding pattern.

Ergativity, accusativity, and the order of Merge and Agree

301

Acknowledgements I would like to thank Josef Bayer, Lennart Bierkandt, Chris Collins, Gisbert Fanselow, Fabian Heck, Riny Huybregts, and Bernd Wiese for helpful comments. I would also like to thank the audiences of the Tools in Linguistic Theory (TiLT) workshop in Budapest, and the 27th GLOW Colloquium in Thessaloniki, both in 2004. An earlier, much shorter version of the present paper has been available under the title “Argument Encoding and the Order of Elementary Operations” since 2004 (www.uni-leipzig.de/~muellerg/ mu56.pdf). The research documented here was supported by a grant to the project Argument Encoding in Morphology and Syntax, as part of DFG-Forschergruppe 742 (‘Grammar and Processing of Verbal Arguments’). References Adger, David 2003 Core Syntax. Oxford /New York: Oxford University Press. Aissen, Judith 1999 Markedness and subject choice in Optimality Theory. Natural Language and Linguistic Theory 17: 673–711. 2003 Differential object marking: Iconicity vs. economy. Natural Language and Linguistic Theory 21: 435–483. Alexiadou, Artemis 2001 Functional Structure in Nominals. Amsterdam: John Benjamins. Anand, Pranav and Andrew Nevins 2006 The locus of ergative case assignment: Evidence from Scope. In Ergativity, A. Johns, D. Massam and J. Ndayiragije (eds.), 3–25. Dordrecht: Springer. Baker, Mark 1996 The Polysynthesis Parameter. New York /Oxford: Oxford University Press. Bechert, Johannes 1979 Ergativity and the constitution of grammatical relations. In Ergativity, F. Plank (ed.), 45–59. London: Academic Press. Bickel, Balthasar and Johanna Nichols 2001 Inflectional morphology. Ms., Universität Leipzig. To appear in T. Shopen (ed.), Language Typology and Syntactic Description. Cambridge: Cambridge University Press. Bierkandt, Lennart 2006 Kasusmorphologie des Diyari. Ein Ansatz im Rahmen der distribuierten Morphologie. In Subanalysis of Argument Encoding in Dis-

302 Gereon Müller tributed Morphology, Vol. 84 of Linguistische Arbeitsberichte, G. Müller and J. Trommer (eds.), 43–62. Leipzig: University of Leipzig. Bierwisch, Manfred 1988 On the grammar of local prepositions. In Syntax, Semantik und Lexikon. Rudolf Růžička zum 65. Geburtstag, M. Bierwisch, W. Motsch and I. Zimmermann (eds.), 1– 65. Berlin: Akademie-Verlag. Bittner, Maria and Ken Hale 1996a Ergativity: Toward a theory of a heterogeneous class. Linguistic Inquiry 27: 531– 604. 1996b The structural determination of Case and Agreement, Linguistic Inquiry 27: 1–68. Bobaljik, Jonathan 1993 Ergativity and ergative unergatives. In Papers on Case and Agreement II. Vol. 19 of MIT Working Papers in Linguistics, MITWPL, C. Phillips (ed.), 45–88. Cambridge, MA: MIT. 2000 The ins and outs of Contextual Allomorphy. In University of Maryland Working Papers in Linguistics, Vol. 10, K. Grohmann and C. Struijke (eds.), 35–71. College Park: University of Maryland. 2002 Realizing Germanic inflection: Why morphology does not drive syntax. Journal of Comparative Germanic Linguistics 6: 129 –167. Bobaljik, Jonathan and Phil Branigan 2006 Eccentric agreement and multiple case checking. In Ergativity, A. Johns, D. Massam & J. Ndayiragije (eds.), 47–77. Dordrecht: Springer. Bobaljik, Jonathan and Susi Wurmbrand 2003 Relativized phases. Ms., University of Connecticut, Storrs. Borer, Hagit 1984 Parametric Syntax. Dordrecht: Foris. 2004 Structuring Sense. An Exo-Skeletal Trilogy. Vols 1 and 2. Oxford: Oxford University Press. Bruening, Benjamin 2001 Syntax at the edge: Cross-clausal phenomena and the syntax of Passamaquoddy. Ph.D. thesis, MIT, Cambridge, MA. Burzio, Luigi 1986 Italian Syntax. Dordrecht: Kluwer. Chomsky, Noam 1993 A minimalist program for syntactic theory. In The View from Building 20, K. Hale and S. J. Keyser (eds.), 1–52. Cambridge, MA: MIT Press. 1995 The Minimalist Program. Cambridge, MA: MIT Press 2000 Minimalist inquiries: The framework. In Step by Step, R. Martin, D. Michaels and J. Uriagereka (eds.), 89 –155. Cambridge, MA: MIT Press. 2001 Derivation by phase. In Ken Hale. A Life in Language, M. Kenstowicz (ed.), 1–52. Cambridge, MA: MIT Press.

Ergativity, accusativity, and the order of Merge and Agree

303

2005a On phases. Ms., MIT, Cambridge, MA 2005b Three factors in language design, Linguistic Inquiry 36: 1–22. Collins, Chris 1994 Economy of derivation and the generalized proper binding condition, Linguistic Inquiry 25: 45–61. 2003 Eliminating labels. In Derivation and Explanation in the Minimalist Program. S. D. Epstein & T. D. Seely (eds.), 42–64. Oxford: Blackwell. Comrie, Bernard 1978 Ergativity. In Syntactic Typology, W. Lehmann (ed.), 329–394. Austin, TX: Texas University Press. 1989 Language Universals and Linguistic Typology. 2. ed.. Oxford: Blackwell. Cysouw, Michael, Søren Wichmann, and Corinna Handschuh 2006 Marked nominative or: The search for S. Ms., Max-Planck-Institute for Evolutionary Anthropology, Leipzig. Dixon, Robert M. W. 1972 The Dyirbal Language of North Queensland. Cambridge: Cambridge University Press. 1994 Ergativity. Cambridge: Cambridge University Press. Elson, Ben 1960 Sierra Popoluca morphology. International Journal of American Linguistics 20: 206 –223. Embick, David and Rolf Noyer 2001 Movement operations after syntax, Linguistic Inquiry 32: 555–595. Frampton, John and Sam Gutman 1999 Cyclic computation. Syntax 2 (1): 1–27. Gazdar, Gerald, Ewan Klein, Geoffrey Pullum, and Ivan Sag 1985 Generalized Phrase Structure Grammar. Oxford: Blackwell. Georgi, Doreen and Gereon Müller 2007 Noun phrase structure by reprojection. In 1 2 Many, Vol. 85 of Linguistische Arbeitsberichte, J. Trommer and A. Opitz (eds.), 263–301. Leipzig: University of Leipzig. Revised version to appear in Syntax. Gregores, Emma and Jorge Suárez 1967 A Description of Colloquial Guaraní. The Hague: Mouton. Haider, Hubert 2000 Branching and discharge. In Lexical Specification and Insertion, P. Coopmans, M. Everaert and J. Grimshaw (eds.), 135–164. Amsterdam: John Benjamins. 2006 Zufall und Notwendigkeit bei der germanischen VO/OV-Spaltung. Ms., Universität Salzburg. Hale, Ken and Samuel Jay Keyser 2002 Prolegomenon to a Theory of Argument Structure. Cambridge, MA: MIT Press.

304 Gereon Müller Halle, Morris and Alec Marantz 1993 Distributed morphology and the pieces of inflection. In The View from Building 20, K. Hale and S. J. Keyser (eds.), 111–176. Cambridge, MA: MIT Press. Harley, Heidi 1995 Subjects, events, and licensing. Ph.D. thesis, Cambridge, MA, MIT. Heck, Fabian and Gereon Müller 2000 Successive cyclicity, long-distance superiority, and local optimization. In Proceedings of WCCFL. Vol. 19, R. Billerey and B. D. Lillehaugen (eds.), 218–231. Somerville, MA: Cascadilla Press. 2003 Derivational optimization of Wh-movement, Linguistic Analysis 33: 97–148. 2007 Extremely local optimization. Proceedings of WECOL 2006. California State University, Fresno. Heim, Irene and Angelika Kratzer 1998 Semantics in Generative Grammar. Oxford: Blackwell. Hornstein, Norbert 2001 Move. A Minimalist Theory of Construal. Oxford: Blackwell. Hualde, José Ignacio, and Jon Ortiz de Urbina (eds.) 2003 A Grammar of Basque. Berlin /New York: Mouton de Gruyter. Jelinek, Eloise 1993 Ergative ‘Splits’ and argument type. In Papers on Case and Agreement I. Vol. 18 of MIT Working Papers in Linguistics, J. Bobaljik and C. Phillips (eds.), 15–42. Cambridge, MA: MITWPL, MIT. Keine, Stefan 2007 Reanalysing Hindi split ergativity as a morphological phenomenon. In 1 2 Many. Vol. 85 of Linguistische Arbeitsberichte, J. Trommer and A. Opitz (eds.), 73–127. Leipzig: University of Leipzig. Kibrik, Aleksandr 1979 Canonical ergativity and Daghestan languages. In Ergativity, F. Plank (ed.), 61–77. London: Academic Press. 2003 Nominal inflection galore: Daghestanian, with side glances at Europe and the world. In Noun Phrase Structure in the Languages of Europe, F. Plank (ed.), 37–112. Berlin /New York: Mouton de Gruyter. Kiparsky, Paul 1999 Analogy and OT: Morphological Change as Emergence of the Unmarked. Ms., Stanford University. Kobele, Greg 2006 Generating copies. Ph.D. thesis, UCLA, Los Angeles. Kratzer, Angelika 1996 Severing the external argument from its verb. In Phrase Structure and the Lexicon, J. Rooryck and L. Zaring (eds.), 109 –137. Dordrecht: Kluwer.

Ergativity, accusativity, and the order of Merge and Agree

305

Kress, Bruno 1982 Isländische Grammatik. 1. ed.. Leipzig: VEB Verlag Enzyklopädie. Laka, Itziar 1993 Unergatives that assign ergative, unaccusatives that assign accusative. In Papers on Case and Agreement I. Vol. 18 of MIT Working Papers in Linguistics, J. Bobaljik and C. Phillips (eds.), 149–172. Cambridge, MA: MITWPL, MIT. Larson, Richard 1988 On the double object construction. Linguistic Inquiry 19: 335–391. Lechner, Winfried 2004 Extending and reducing the MLC. In Minimality Eff ects in Syntax, A. Stepanov, G. Fanselow and R. Vogel (eds.), 205–240. Berlin /New York: Mouton de Gruyter. Lee, Hanjung 2003 Parallel optimization in case systems. Ms., University of Minnesota, Twin Cities. Legate, Julie Anne 2005 The absence of absolutive case. In Proceedings of the 24th WCCFL, J. Alderete, C. Han and A. Kochetov (eds.), Cascadilla Press. 2006 Split Absolutive. In Ergativity, A. Johns, D. Massam and J. Ndayiragije (eds.), 143–171. Dordrecht: Springer. Levin, Juliette and Diane Massam 1985 Surface ergativity: Case/theta relations reexamined. In Proceedings of NELS 15, S. Berman (ed.), 286–301. Amherst, MA: BLS. Lewis, David 1972 General semantics. In Semantics of Natural Language, D. Davidson and G. Harman (eds.), 169–218. Dordrecht: Reidel. Mahajan, Anoop 1990 The A/A-bar distinction and movement theory. Ph.D. thesis, MIT, Cambridge, MA Marantz, Alec 1984 On the Nature of Grammatical Relations. Cambridge, MA: MIT Press 1991 Case and licensing. In Proceedings of the Eigth Eastern States Conference on Linguistics, G. Westphal, B. Ao and H. R. Chae (eds.), 234– 253. University of Maryland. Marlett, Stephen 1986 Syntactic levels and multiattachment in Sierra Popoluca. International Journal of American Linguistics 52: 359–387. Mohanan, Tara 1994 Argument Structure in Hindi. Stanford: CSLI Publications. Müller, Gereon 1998 Incomplete Category Fronting. Dordrecht: Kluwer.

306 Gereon Müller Müller, Gereon 2006 Global impoverishment in Sierra Popoluca. In Subanalysis of Argument Encoding in Distributed Morphology. Vol. 84 of Linguistische Arbeitsberichte, G. Müller and J. Trommer (eds.), 23–42. Leipzig: University of Leipzig. Murasugi, Kumiko 1992 Crossing and nested paths. Ph.D. thesis, MIT, Cambridge, MA Nash, Léa 1996 The internal ergative subject hypothesis. In Proceedings of NELS 26, K. Kusumoto (ed.), 195–209. Amherst, MA: GLSA. Nichols, Johnanna 1986 Head-marking and dependent-marking grammar. Language 62: 56– 119. Pesetsky, David 1982 Paths and categories. Ph.D. thesis, MIT, Cambridge, MA 1989 Language-Particular Processes and the Earliness Principle. Ms., MIT, Cambridge, MA. Pesetsky, David and Esther Torrego 2001 T-to-C movement: Causes and consequences. In Ken Hale. A Life in Language, M. Kenstowicz (ed.), 355–426. Cambridge, MA: MIT Press. 2006 Probes, goals and syntactic categories. Ms., MIT. Lingbuzz/000321. Plank, Frans 1995 Ergativity. In Syntax. Ein internationales Handbuch zeitgenössischer Forschung. Vol. 2, J. Jacobs, A. von Stechow, W. Sternefeld and T. Vennemann (eds.), 1184–1199. Berlin /New York: Mouton de Gruyter. Pollard, Carl J. and Ivan A. Sag 1994 Head-Driven Phrase Structure Grammar. University of Chicago Press, Chicago. Prince, Alan and Paul Smolensky 1993 Optimality theory. Constraint interaction in generative grammar. Book ms., Rutgers University. Ramchand, Gillian 2003 First phase syntax. Book ms., Oxford University. Rezac, Milan 2003 The fine structure of cyclic agree. Syntax 6: 156 –182. Richards, Marc 2004 Object shift and scrambling in North and West Germanic: A case study in symmetrical syntax. Ph.D. thesis, University of Cambridge, Cambridge, UK. 2007 Deriving the edge: What’s in a phase? Ms., Universität Leipzig.

Ergativity, accusativity, and the order of Merge and Agree

307

Roberts, Ian and Anna Roussou 2002 The extended projection principle as a condition for the tense-dependency. In Subjects, Expletives, and the EPP, P. Svenonius (ed.), 125– 155. Amsterdam: John Benjamins. Schäfer, Florian 2007 On the nature of anticausative morphology. Ph.D. thesis, Universität Stuttgart. Sigurðsson, Halldór 2002 To be an oblique subject: Russian vs. Icelandic. Natural Language and Linguistic Theory 20 (4): 691–724. Silverstein, Michael 1976 Hierarchy of features and ergativity. In Grammatical Categories in Australian Languages, R. Dixon (ed.), 112–171. Canberra: Australian Institute of Aboriginal Studies,. Speas, Margaret 1990 Phrase Structure in Natural Language. Dordrecht: Kluwer. Stabler, Edward 1996 Acquiring and parsing languages with movement. Book ms., UCLA. Sternefeld, Wolfgang 2006 Syntax. 2 Vols. Tübingen: Stauffenburg. Stiebels, Barbara 2000 Linker inventories, linking splits and lexical economy. In Lexicon in Focus, B. Stiebels and D. Wunderlich (eds.), 211–245. Berlin: Akademie Verlag. 2002 Typologie des Argumentlinkings: Ökonomie und Expressivität. Berlin: Akademie Verlag. Svenonius, Peter 1994 C-selection as feature-checking. Studia Linguistica 48: 133–155. Trommer, Jochen 2003 Feature (non-)insertion in a minimalist approach to spellout. In The Main Session. Papers from the 39th Meeting. Vol. 1. Chicago: Chicago Linguistic Society. www.ling.uni-osnabrueck.de/trommer/papers.html. Ura, Hiroyuki 2000 Checking Theory and Grammatical Functions in Universal Grammar. New York: Oxford University Press. 2006 A parametric syntax of aspectually conditioned split-ergativity. In Ergativity, A. Johns, D. Massam and J. Ndayiragije (eds.), 111–141. Dordrecht: Springer. Williams, Edwin 1981 Argument Structure and Morphology. The Linguistic Review 1: 81–114. 1994 Thematic Structure in Syntax. Cambridge, MA: MIT Press.

308 Gereon Müller Woolford, Ellen 1997 Four-way case systems: Ergative, nominative, objective and accusative. Natural Language and Linguistic Theory 15: 181–227. 2001 Case patterns. In Optimality-Theoretic Syntax, G. Legendre, J. Grimshaw and S. Vikner (eds.), 509–543. Cambridge, MA: MIT Press. Wunderlich, Dieter 1997 Cause and the structure of verbs. Linguistic Inquiry 27: 27– 68. 2000 Optimal Case in Hindi. Ms., Universität Düsseldorf.

Contributors Berit Gehrke

Jonathan MacDonald

Universitat Pompeu Fabra – Departament de Traducció i Ciències del Llenguatge Carrer Roc Boronat, 138 08018 Barcelona Spain

Universitat Autònoma de Barcelona – Centre de Lingüística Teòrica Departament de Filologia Espanyola – Facultat de Lletres Edifici B 08193 Bellaterra (Barcelona) Spain

[email protected]

[email protected] Nino Grillo Consejo Superior de Investigaciones Científicas – ILLA – Grupo de Lingüística y Ciencia Cognitiva c. Albasanz 26 –28 28037 Madrid España

Thomas McFadden

[email protected]

[email protected]

Kleanthes K. Grohmann

Rita Manzini

University of Cyprus Department of English Studies 75 Kallipoleos P.O. Box 20537 1678 Nicosia Cyprus

Università di Firenze Dipartimento di Linguistica Piazza Brunelleschi 4 50121 Firenze Italy

Universität Stuttgart Institut für Linguistik: Anglistik Heilbronnerstr. 7 70176 Stuttgart Germany

[email protected]

[email protected] Masaaki Kamiya

Gereon Müller

Hamilton College, Department for East Asian Languages and Literature 198 College Hill Road Clinton, NY 13323 USA

Universität Leipzig Institut für Linguistik Beethovenstr. 15 04107 Leipzig Germany

[email protected]

[email protected]

310 Contributors Ivan Ortega-Santos

Bartosz Wiland

University of Memphis Foreign Languages & Literatures 390 Dunn Hall Memphis, TN 38152 USA

University of Poznań School of English 61-874 Poznań Poland

[email protected] Halldár Armann Sigurðsson Lund University SOL, Centre for Language and Literature Box 201 221 00 Lund Sweden [email protected]

[email protected]

Index

across-the-board construction, 11, 55–56, 72 active marking, 284, 299 agrammatism, 251 Agree, 6–10, 15, 21, 42, 44, 55–57, 61, 65–67, 72, 80–81, 91, 98, 108–109, 115, 118, 131, 135, 143, 163–164, 187, 207, 215–218, 221, 225, 250, 269, 272–293, 299–300 agreement, 7, 11–12, 15, 21–48, 64–66, 81, 87–92, 95–103, 107–119, 123, 131–135, 138, 152–155, 166–167, 245, 250, 269–271, 274–276, 281, 284, 293, 300 cyclic ~, 108–109, 113–114 Long Distance ~ (LDA), 11, 87, 89, 91–92, 97–102 morphological ~, 22, 42–43, 48, 87, 91, 99–102, 114, 274 Albanian, 13, 132, 136–139, 142–144, 149–153, 157–162, 165–168, 173–175 ambiguity, 32, 43, 46, 134, 138, 141– 142, 161, 165, 182, 218–219, 225 A-movement, 9, 53, 57, 72, 80–81, 134, 181, 184, 187, 190–193, 196, 200–201 Antekerrepenhe, 295–296 aphasia, 15, 251, 252, 255, 256, 261 Archi, 271, 277, 284 argument structure, 15, 38, 81, 121, 231– 236, 243, 260–261, 280, 285 articulatory-perceptual interface (AP), 63, 66, 175, 195 atelic (see also telic), 208, 21–219, 222– 224 bare plural, 207, 210–211 Basque, 281, 284 BECOME-operator, 239

broad PF (see also Phonological /Phonetic Form), 22, 43, 48–49 Case, 6, 9–14, 21, 23, 26–29, 32–33, 37– 45, 48–49, 53–58, 61–63, 66, 71, 74– 75, 87, 90–98, 101, 107–126, 131–141, 144–175, 181–186, 190–194, 198, 200– 201, 209, 222, 224, 233–241, 244– 255, 259–261, 269–287, 291–300 accusative ~, 12, 38, 117–119, 122, 233, 241, 270, 294–295 dative ~, 12, 40, 107, 116, 125, 181– 190, 201, 260, 293 ergative ~ (see also split), 191, 271, 291–295 inherent ~, 122, 126, 181–184, 187, 190 nominative ~ (see also Nominative Puzzle/Low Nominative Hypothesis), 12, 22, 40–41, 107–114, 117, 137, 160, 294, 296 structural ~, 12, 55–57, 81, 107, 117– 125, 277–278, 299 causative-inchoative, 185, 201, 218 chain, 53–54, 57, 71, 131–134, 152–158, 161, 170, 173–174, 233, 255 checking (see also feature), 6–14, 53, 55–58, 61, 64, 73–75, 81, 87–94, 97– 98, 102, 109, 191, 193, 196, 273, 276, 293, 297 clause types, 16, 269, 284, 293–294 complementizer, 142–143, 148–150, 156, 171, 293–294 Complementizer Phrase (CP), 5, 7, 14, 54–55, 59–62, 66–67, 73, 76, 79, 90, 95, 113, 134, 182, 193–197, 200–201, 255–256, 274, 285 Conceptual-Intentional interface, 3, 135

312 Index conceptual-intentional interface (CI), 160, 207, 224, 226 control, 9, 13, 30, 35, 54, 79, 80, 131– 144, 150–167, 170–175 adjunct ~, 11, 53–54 dependency marking, 41 determiner, 148 Determiner Phrase (DP), 12–15, 38, 40, 61–66, 71–81, 91, 97, 107–125, 183– 184, 187–189, 192, 201, 235–236, 240, 243–245, 248–251, 255, 261, 269–270, 274–275, 279, 284–288, 291–293, 296, 298 Distributed Morphology (DM), 6, 21, 101, 121, 291, 296 ditransitive verb, 183, 185, 187, 287 do so-construction, 210, 219–223 durative, 207–219, 222–226, 239 earliness, 15, 269, 275–279, 285, 287– 288, 290, 300 empty category, 132–134, 151, 154, 157, 162–165, 173–174 endpoint, 208, 236 event, 1, 15, 145, 208–219, 222, 224, 231– 232, 235–243, 247–249, 256–261 ~ structure, 15, 208, 217, 231–243, 248, 256–261 ~ time, 236–237, 243, 257–260 Exceptional Case-Marking (ECM), 110– 113, 119–126 experiencer, 181–190, 218 Extended Projection Principle (EPP), 13–14, 57, 80, 111, 113, 125, 131, 140, 142, 147–162, 165–174, 181, 187, 190–193, 196, 201, 235, 243–246 external argument, 15, 108, 111, 122, 141, 215, 218–225, 232–236, 241, 244, 250, 258, 276–300 extraction, 58, 60, 63, 67, 72, 94, 196, 254 fake reflexive, 15, 224–225

feature, 6–14, 21–39, 45, 48, 53–57, 62, 66–67, 71, 79–80, 90, 95, 133, 145, 162, 181–182, 190–194, 197, 200– 201, 217, 224–225, 231, 237, 250, 253–256, 259, 261, 272–300 ~ balance, 280, 297 ~ checking, 6, 8, 57, 62, 67, 79, 288 ~ hierarchies, 285, 287, 290, 294, 300 ~-driven movement, 11, 53, 55, 57– 58, 62, 79, 81, 256 case- ~, 11 focus- ~, 14, 182, 196, 201 formal ~, 7–10, 21–22, 28, 34, 42, 47, 49, 55–57, 61–63, 66, 73–74, 90 interpretable ~, 41 phi- ~, 8, 44, 97, 107, 133, 151 uninterpretable ~, 10, 21, 57, 67, 93 gender, 10, 23–34, 37, 42–48, 61–63, 144, 158–159, 169, 253 inquorate ~, 25 semantic ~, 26–33 gender- (see also gender), 31 Guaraní, 281, 284 Hindi, 191–192, 201, 284–285, 289–293, 296 hybrid nouns, 27–31 Icelandic, 10, 23–48, 107–117, 122, 126, 183–184, 270, 274, 276, 300 idiom, 15, 225–226 incrementality, 87–88, 99, 101–103 infinitive, 132, 136, 139, 159–162, 168, 171, 175 inflectional classes, 27, 47 Interface Uniformity, 131, 134–135, 165, 173, 175 Italian, 30–31, 38, 88, 100–101, 132, 136– 143, 146, 149, 152, 155–175, 244–245 iterative interpretation, 209–210, 212 Japanese, 14, 90, 182–188, 192–197, 201

Index 313 left edge, 14 Low Nominative Hypothesis, 118–119 mass noun, 207, 211 Merge (see also Remerge), 4, 6, 10–11, 15, 53–54, 75, 77–81, 90–91, 94, 119, 121, 249, 269, 272–278, 281–282, 285–290, 299–300 parallel ~, 77–79 minimal search, 11, 12, 94, 96, 103 model, 4, 6–8, 48, 80, 131, 139, 236, 257– 258 Y- ~, 2, 48 movement, 7–15, 21, 32, 48, 53–61, 66– 76, 79–81, 89–91, 95–98, 109, 111, 118–119, 131–135, 155, 163–164, 173–175, 181–182, 190–197, 200– 201, 216, 221, 231–237, 241–246, 250–261, 274, 278, 287, 300 Navajo, 270, 276 negation, 9, 14, 143, 148–149, 169, 181– 182, 190, 194–201 Nez Perce, 295 No Case Generalization, The, 37, 40 Nominative Puzzle, 117–120 number, 6, 10–12, 22–23, 27, 28, 34–38, 42–45, 48–49, 56, 60–66, 72, 89–92, 103, 114, 122, 139, 144, 147, 158, 165, 169, 209–212, 246, 253, 270– 271, 279 semantic ~, 36 number-~ (see number), 31 numeration, 3–4, 53, 71, 274–275, 279– 281, 284, 290, 294, 297, 299–300 parasitic gaps, 11, 53, 56, 58–64, 67–77, 81, 108 pseudo-~, 58–61, 67, 72 participle, 37, 43–44, 145–149, 159, 168– 170, 172, 175, 232–234, 244–245, 261 passive/passivization, 9, 10, 15, 38, 108, 111, 113, 118, 132, 165, 173–175,

184–187, 190, 192, 215–216, 231– 237, 241–252, 256, 257–261 perceptible form, 10, 21, 42 phase, 1–10, 14–15, 76, 91–94, 107–113, 117, 120–125, 136, 164, 207, 224– 226, 231, 237–238, 258–261, 277, 294 ~ head, 4–5, 7, 93, 108–109, 294 VP ~, 5, 91, 237, 260 Phase Impenetrability Condition (PIC), 4–7, 91–92, 107–108, 112, 116, 120, 125, 277 Phonological/Phoentic Form (PF) (see also perceptible form), 2–4, 7–10, 21–23, 26, 28, 31–34, 37–43, 46–49, 53, 80, 90, 121, 135, 138 broad ~, 21–22, 43, 48–49 pluralia tantum (see also singularia tantum), 34 Pomo, Eastern, 283 preposition, 40, 42, 58, 75, 117, 146, 171–172, 181–183, 213–214, 220, 223, 233 Preposition Phrase (PP), 54, 75–76, 79, 81, 91, 183, 187–192, 217–223, 286 ~ goal, 183, 187, 189, 192 ~ location, 185, 187–190, 213 PRO, 9, 13, 54–55, 66, 74, 131–142, 150– 153, 157, 161–165, 171–174, 241, 250 Probe, 6, 11, 55, 57, 67, 72, 75, 79–80, 87, 91–98, 102–103, 252, 273–274, 277, 279, 287–290, 293–294, 296 pronoun, 29–33, 39, 55, 58, 62, 64, 66– 71, 78, 139–140, 142, 157, 167, 171, 250, 270, 298–299 clitic ~, 71, 138, 142, 145–149, 160, 169, 172, 174–175, 233, 245 null ~, 55, 58, 61, 64–66, 72 resumptive ~, 64, 71 weak ~, 62–64, 66, 67 quirky, 39, 111–113, 115, 120, 122, 124– 125, 300

314 Index raising, 5, 9, 13, 58, 59, 68–69, 107–111, 131–144, 152–157, 161–165, 171–175, 184–187, 190, 193–194, 196, 276 Relativized Minimality (RM), 252–255 Remerge, 53–57, 61, 67, 72, 75–80, 95 theory of movement, 53–55 reprojection, 70–71, 97, 287 resultativity, 15, 239–242, 261 scope, 9, 14, 66–68, 97, 158, 181–182, 188–197, 200–201, 210, 213, 256 scrambling, 90 secondary predication, 231, 240, 247 sex-marking, 30 sideward movement, 11, 53–57, 61–81 Sierra Popoluca, 271, 274, 277, 284, 293–294 singularia tantum (see also pluralia tantum), 34–35 Spec-Head configuration, 6, 11–12 Spell-Out, 2–4, 7–8, 12, 54, 77, 89, 93, 109–110, 120, 291 Multiple ~ (MSO), 2–6, 12, 87, 93– 96, 102 split, 41, 166, 281, 284, 291, 293, 298, 300 ~ ergativity, 281, 284, 293, 298, 300 clause-type based ~, 284 person-based ~, 284, 295, 298

tense/aspect-based ~, 16, 269, 284– 285, 289–292, 294 stative, 207, 211, 221–224, 231–232, 236, 238, 246–250, 256 stress, 108, 172, 182, 190–191, 235, 255, 300 subevent, 209–212, 231–232, 236–241, 248, 250, 256–261 subjunctive particle, 13, 135–136, 139, 142–143, 147, 163, 168 telic (see also atelic), 208 –224, 240 theta-role, 5, 7, 9, 12–13, 76, 87, 90, 92, 95–98, 131, 134–135, 151–161, 173– 174, 181, 233, 255 three-way systems, 295–299 time span adverbial, 207–219, 222–226 topic/topicalization, 9–10, 57, 60–61, 73, 76, 90, 144, 156, 171, 251, 253 Transfer, 3–4, 7, 10–11, 15, 21, 26, 28, 48, 77, 207, 224, 226 unaccusative, 108, 111, 113, 118, 122, 132, 173–175, 215–216, 252, 256, 287, 289 universal quantifier, 14, 67–69, 181–182, 190–201 valuation, 8–9, 11, 41, 53–58, 67, 80, 90, 97, 154, 174, 273, 276, 293