Explorations of Phase Theory: Interpretation at the Interfaces

Explorations of Phase Theory: Interpretation at the Interfaces edited by Kleanthes K. Grohmann Mouton de Gruyter Exp...

Author: Grohmann | Kleanthes K.

16 downloads 711 Views 2MB Size Report

This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!

Report copyright / DMCA form

DOWNLOAD PDF

Explorations of Phase Theory: Interpretation at the Interfaces

edited by Kleanthes K. Grohmann

Mouton de Gruyter

Explorations of Phase Theory: Interpretation at the Interfaces

≥

Interface Explorations 17

Editors

Artemis Alexiadou T. Alan Hall

Mouton de Gruyter Berlin · New York

Explorations of Phase Theory: Interpretation at the Interfaces

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 : interpretation at the interfaces / edited by Kleanthes K. Grohmann. p. cm. ⫺ (Interface explorations ; 17) Includes bibliographical references and index. ISBN 978-3-11-020521-3 (hardcover : alk. paper) 1. Minimalist theory (Linguistics) 2. Generative grammar 3. Grammar, Comparative and general ⫺ Syntax. I. Grohmann, Kleanthes K. P158.28.E973 2009 4151.0182⫺dc22 2009002390

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 978-3-11-020521-3 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 interfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Kleanthes K. Grohmann

1

Intermodular argumentation and the word-spell-out-mystery . . . . . . . . . . . 23 Tobias Scheer On bare prosodic structure and the spell-out of features . . . . . . . . . . . . . . . . 67 Martin Haiden Spell out before you merge . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95 Hisao Tokizaki On the derivation of the relation between givenness and deaccentuation: A best-case model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115 Dalina Kallulli Phase theory, linearization, and zig-zag movement . . . . . . . . . . . . . . . . . . . 133 Carlo Geraci Surviving reconstruction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161 Thomas Stroik and Michael Putnam On the interface(s) between syntax and meaning . . . . . . . . . . . . . . . . . . . . . 181 Anjum Saleemi Dynamic economy of derivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 211 Takashi Toyoshima The conceptual necessity of phases: Some remarks on the minimalist enterprise . . . . . . . . . . . . . . . . . . . . . . . . . . . 253 Dennis Ott Contributors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 277 Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 279

Exploring interfaces 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: Features and Arguments, is one effort to give testimony to that hope.1 The present compilation is assembled from work accepted for presentation at the InterPhases conference. Selected oral (Kallulli, Stroik & Putnam, Tokizaki, Toyoshima) and poster presentations (Geraci), as well as two that unfortunately had to be canceled (Haiden, Saleemi), have been specially prepared for this volume by the authors, having benefited greatly from the feedback received at the conference and afterwards. In addition, two more chapters were solicited explicitly for this volume (Ott, who also participated at InterPhases, and Scheer). The selection criterion for this collection is that each chapter explicitly address issues concerning Interpretation at the Interfaces – the sub-title of this volume – all framed within the leading

1

This ‘sister volume’ volume on “Explorations of Phase Theory” also derived from the InterPhases conference (Interface Explorations 18). 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 theme of this volume (and the next, the above-mentioned 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 interface interpretations might be interesting or relevant, 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 also next section). 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 interfaces

(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 Lexicon 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 this 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 interfaces

(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 Chomsky suggests. The sister volume to the present collection addresses such issues in more detail. (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 sub-arrays) in the two examples. However, this now leads to yet another question not mentioned so far – and arguably even more controversial than (3c), but certainly (3b) and es4

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 interfaces

7

pecially (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. Interface interpretation The important role of the interfaces within Phase Theory has been alluded to above already (for further specific treatments along these lines, see also Grohmann 2007a–c, 2008, 2009a, but also many contributions in those respective volumes and many other current work). In fact, I would go so far as to say that the most interesting and computationally relevant aspects of syntax lie here, in the interfaces (as argued for in Grohmann 2008 and follow-up research). With respect to phases, one might want to investigate further, for example, what aspects of NS are relevant for LF and how this could help find a definition of ‘phasehood’ (as Ott does) and PF (see espe5

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 cially the chapters by Haiden and Scheer); ultimately, this is what the present volume is concerned with. While this is not the only possible perspective, it (and many other alternatives) have recently been collected in Ramchand & Reiss (2007), paying tribute to the important role that the study of interfaces plays for linguistic theorizing from more perspectives than the rather narrow ones pursued here, namely: How is LEX/NS to be mapped to the modular interfaces LF/ PF? I already introduced my terms linguistic vs. modular interfaces, which are only new by name, not by concept (see Grohmann 2009a for more) – where the linguistic interfaces are those systems in the human brain that take the linguistic input and construe true interpretation (meaning and sound) and where the modular interfaces are those well-known modules within the language faculty linguists have explored for decades (LF and PF).6 But recall that under the dynamic approach to the computation, as understood here, it is not so clear anymore that LF and PF are ‘discrete’ levels of representations, and the interface levels used to be known. In fact, it is not at all clear that they are to be conceived as ‘levels’. In a series of works, Juan Uriagereka has suggested the term ‘component’ instead (Uriagereka & Martin 1999 and other work from the late 1990s, some of which published in Uriagereka 2002, 2008) and he also makes the case for additional components, specifically, one that interacts between LEX and NS, a kind of remnant or modern conception of the old D-structure level of representation (Uriagereka 2008). Another question arises as to what happens on the meaning side of the computation, and relating to ‘non-core semantic’ aspects, such as pragmatics or information structure; or, on the sound side, one may wonder how prosody, a typical syntax-phonology interface aspect of language, can be computed. Perhaps the relevant information can be read of a combination of NS, LF, and PF – or perhaps there are additional components involved. But the issue needs to be addressed (see, for example, Kallulli’s and Saleemi’s chapters). One (series) of many research questions surrounding this topic then, also addressed in some of the following chapters (though not necessarily in this context), can be formulated as follows: How many interface levels are there, what is the nature of these levels or components, and how are they accessed both before and after NS? The operation Spell-Out has also been introduced above. In Phase Theory its nature – and application, of course – has taken a slightly different form 6

On the biolinguistic program along these lines, see also Chomsky (2005, 2006, 2007a–b), although this aspect will not be developed here any further (but see many of the following contributions).

Exploring interfaces

9

from earlier instantiations of the Minimalist Program. Whereas in earliest incarnations the term designated a single operation that applied once per derivation, to mark the turning point from overt to covert syntax, it later became an operation that, like any other computational operation, may apply more than once – namely, whenever needed (Uriagereka 1999). In Phase Theory, Spell-Out is but one sub-operation of a more general operation called Transfer, as laid out above.7 While this terminological distinction is not always made in the chapters that follow the introduction, its ‘origin’ as sketched here may be borne in mind throughout the reading of this volume.8 The important notion relating to the present section header is that Transfer is the operation that leads to interpretation of a grammatical structure at the interfaces (ideally, dynamically) – and that such interface interpretation involves several modules beyond, after, or possibly simultaneously with NS. By hypothesis, Spell-Out (and presumably Transfer, more generally) applies at the phase-level, thus turning LF and PF into several ‘mini-interface components’ lf and pf (Lasnik & Uriagereka with Boeckx 2005), along the lines of (2). Specifically for Spell-Out, a recurring theme in this volume (addressed by virtually every contributing author one way or another), we can thus ask: When exactly, and how, does Spell-Out apply? But many more specific properties of derivations by phases, computational operations and conditions, and issues for interface interpretation arise. One major aspect of minimalist theorizing has been, from the beginning, the attempt to make computations as economical as possible. This started out in the GB-era with an intense discussion of derivational and/vs. representational economy (cf. Chomsky 1991). Phases are a new attempt to reduce so-called ‘computational complexity’ by narrowing the current computational load to the smaller window of a derivational chunk identified as phase (but see e.g. Epstein & Seely 2006 for critical remarks). Such economical concerns can also be applied to the inspection of syntactic operations, such as Merge and Move (as Toyoshima does). From a phase perspective, then, one could ask: How, if at all, do phases help achieve maximum economy in derivation and representation, i.e. interpretation at the interfaces? 7

8

For yet another way of (re)implementing Spell-Out within a dynamic approach to the grammar, see the references to my own work cited at the beginning of this section. The same goes for other finer distinctions – which is OK precisely because current phase-theoretic investigations form a constantly developing and evolving aspect of current syntactic theorizing with most, if not all, concepts or definitions not yet carved in stone.

10 Kleanthes K. Grohmann The last group of chapters not yet mentioned can be taken together to address other concerns for operations both in NS and at the LF and PF interface components. It is obvious that the above-mentioned ‘grammatical licensing mechanisms’ are a critical ingredient for syntactic operations. Take one such mechanism the checking of features, whether by (i.e. after) movement or through (long-distance) Agree – this is the focus of the sister volume. But the interplay of feature checking, overt vs. covert movement operations, and phase edges can also be examined from an interface perspective (see Geraci’s contribution, for example). And once the syntactic derivation has reached a pair <SEM, PHON> or <Σ, Φ> legible at the interfaces, at least on the sound side, there needs to be a mechanism, perhaps algorithm (cf. Kayne 1994) that linearizes the string of syntactic objects generated – again, a very typical interface property (as Tokizaki explores further). And last but not least, other long-standing two-way interaction between derivation and interpretation concerns binding and the so-called argument/adjunct asymmetry (addressed by Stroik & Putnam in an alternative minimalist fashion). Thus we can ask: How do interface properties interact with syntactic operations?

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. The present collection starts off with a chapter by Tobias Scheer, “Intermodular Argumentation and the Word-Spell-Out-Mystery”, in which the author presents an interesting new way of connecting syntactic theory and phonological theory. This is what Scheer calls intermodular and it adopts a general dynamic approach to the derivation (in the sense of Uriagereka 1999 and Chomsky 2000). In particular, the shipping back and forth of pieces between morphosyntax (NS) and interpretation (PF/LF) during the derivation of a sentence establishes a pipe between the concatenative and the interpretational devices that did not exist in GB or earlier versions of

Exploring interfaces

11

the inverted T-/Y-architecture (cf. (1) above). It creates, he points out, a situation where syntactic theories and analyses may have direct consequences on the phonological side, and the other way round. The new interactionist architecture that the minimalist orientation has installed allows to make intermodular arguments, which as Scheer argues is stronger evidence than what can be produced by modular-internal reasoning. It is therefore the right step towards bringing together syntacticians and phonologists again to talk about linguistic theory. The other goal is to make the conditions of intermodular argumentation explicit – what it can do, want it cannot do, and what it can do only if this or that question is settled beforehand. Martin Haiden discusses three PF-related problems for the minimalist program in his chapter, “On Bare Prosodic Structure and the Spell-Out of Features”, and argues that they all stem from common, yet unfounded assumptions about phonology: (i)

phonology is uniquely human, specific to language, and (by hypothesis) not recursive;

(ii)

a range of morpho-syntactic phenomena can only be explained if syntax can look ahead (to the phonology);

(iii) the derivation of interface representations from NS requires operations that appear to violate central minimalist guidelines (such as Inclusiveness). The first part of his contribution concentrates on phonology from the perspective of the look-ahead problem and its manifestation in the domain of non-concatenative morphology. Haiden concludes that phonology (PF) does not apply after syntax (NS), but in parallel with it. The second part of the paper proposes a formalism that derives interface representations as a necessary by-product of hierarchical structure. In that theory, the recursive core of grammar includes both prosody and syntax: Phonological and syntactic derivations apply in parallel, in a single cycle spanning over the two components. The phonetic/phonological and (morpho-) syntactic features of a given lexical entry are merged at the same stage of the derivation in their respective components. Linearization-patterns are derived by means of templates. Haiden exemplifies his theory with data from Classical Arabic and Standard German verbal morphology. He closes his chapter with an outlook on how generalized templates could replace syntactic parameters, and on their place in acquisition.

12 Kleanthes K. Grohmann “Spell Out before You Merge”, the chapter by Hisao Tokizaki, discusses the relation between building up syntactic structure with lexical items and linearizing it in order to send the phonetic features to PF. The standard assumption in the Minimalist Program is that two syntactic objects are merged to form a constituent, which is transferred to PF at each phase (Spell-Out). However, it has been claimed that this interface system faces the ‘assembly problem’ (Dobashi 2003), that is, the derivation proceeds from the bottom to the top in a tree, while linearization proceeds left-toright or top-down (at least in right-branching languages). An approach to solve this problem is to assume that structure building also proceeds in a top-down fashion, as, among others, argued by Phillips (1996, 2003), Kempson et al. (2001), and O’Grady (2005). However, such a ‘top-down building’ approach also faces a number of problems, as Tokizaki shows (see also Shiobara 2005). He proposes a new view of Merge, which is done with the help of a pair of syntactic brackets contained in the numeration. A left bracket is introduced into the derivation before a syntactic object, which is merged with another syntactic object when a right bracket is introduced after it. It is also assumed that Spell-Out strips away the phonetic features of a lexical item and sends them to PF when the item is introduced into derivation. Thus, as the title of the chapter suggests, a lexical item is spelled out before it is merged with another item or constituent. This organization of the grammar does not have the assembly problem just mentioned: Linearization proceeds left to right in a top-down fashion (Linearize Down), and Merge builds up structure in a bottom-up fashion (Merge Up). The Spell-Out-before-Merge model does not assume specific phases such as vP and CP, and thus can rather be regarded as a type of strong derivational approach such as that proposed by Epstein et al. (1998). A further interesting consequence of ‘Spell-Out before Merge’ is that it provides a simple model for parsing. Tokizaki argues that a syntactic bracket is spelled out as a silent demibeat (cf. Selkirk 1984), which a hearer interprets as a syntactic bracket to obtain the intended structure. Dalina Kallulli contributes the chapter “On the Derivation of the Relation between Givenness and Deaccentuation”. Starting from the wellknown observation that ‘given’ or ‘old’ information correlates with lack of phonetic prominence, or deaccentuation, the primary goal of this contribution is to provide a syntactic account of deaccentuation. Assuming minimalist theorizing, Kallulli examines the correlation between givenness/presupposition of clausal complements and deaccentuation, which, she shows, holds across several languages. Fitting the theme of the present collection,

Exploring interfaces

13

this chapter is also an attempt to shed light on the nature of the different components of grammar, and what has been called the ‘division of labour’ among them. Kallulli puts some doubt on the widely held view that the nature of operations taking place in NS is entirely different from those occurring at the interpretive interfaces, for example, that various tasks and even principles that have long been argued to constitute core syntactic operations are often simply put into the interfaces, especially PF. One would thus like to know where exactly the boundary is between NS and everything else. Of course, different components deal with different objects and representations – what is missing, however, is a clear model of how exactly the computational system feeds both PF and LF to produce the systematic correlations that obtain between the ‘sound’ part and the ‘meaning’ part for many linguistic phenomena. Starting out with Chomsky’s (2004) ‘best case’ scenario, according to which the components of the derivation of proceed cyclically in parallel, the central claim that Kallulli puts forth in this chapter is that information structure is encoded in NS, and in particular, that at least some information which leads to particular Spell-Out forms is part of the numeration/lexical array. Specifically, Kallulli argues for the following points: (i) the [+presupposed] and/or [+given] status of an embedded CP must be expressed, (ii) this is achieved syntactically via functional structure, and (iii) the head hosting the relevant syntactic feature, which is instantiated overtly either by some expletive-like element is a probe for the goal with an OCC feature, yielding the various patterns that are found and that are described in some detail in this chapter. The aim of Carlo Geraci’s chapter “Phase Theory, Linearization and Zig-Zag Movement” is to show how the grammar of human languages can capture the fact that certain elements may move from one edge of the vP phase to the opposite edge of the CP phase. This kind of movement is mainly instantiated by Italian Sign Language. In this language, wh-phrases move from a left peripheral position in the vP area to a right-peripheral position in the CP area, instantiating what the author calls zig-zag movement. The current development of Phase Theory as proposed in Chomsky (2000 et seq.) and the theory of successive-cyclic linearization proposed by Fox & Pesetsky (2005) interact with two general and plausible restrictions on overt movement: (i) the ban on movement from the complement of a head to the specifier of the same projection (Abels 2003), and (ii) the ban on movement from the specifier of a projection to an outer specifier of the same projection (Ko 2005). When combined, these ingredients generate a series of restrictions that are at odds with empirical data from Italian Sign

14 Kleanthes K. Grohmann Language as well as other signed and spoken languages. In particular, zigzag movement is predicted to be impossible, since it would produce a conflicting ordering among the lexical elements at the syntax-PF interface. The solution proposed to this puzzle refines the definition of movement (Internal Merge) and relies on the fact that those elements that are still active for the syntactic component (i.e. wh-phrases at intermediate [Spec,vP] or [Spec,CP] positions) are simply ignored by the operation that sends the structure to the interfaces. However, instead of postulating a specific structural position where wh-elements can hide, Geraci proposes that it is the fact that they bear an uninterpretable feature that makes them invisible to the linearization algorithm. In the chapter “Surviving Reconstruction”, Thomas Stroik & Michael Putnam offer a reconstructive analysis of constructions with Condition C argument/adjunct asymmetries that doesn’t require either the non-cyclical operations advocated by Lebeaux (1991), Chomsky (1993), and Fox (2003) or the syntactic de-cloaking operations advocated by Chomsky (2001). They propose that the Condition C binding asymmetries involving arguments and adjuncts can best be explained within Stroik’s (forthcoming) version of minimalism (called Survive) in which an optimal minimalist syntax will have only strictly local syntactic operations that map elements from the numeration onto the derivation. In Survive-minimalism, there are only two types of syntactic operations: (i) Merge, which introduces new lexical material from the numeration into the derivation, and (ii) Remerge, which remerges into the derivation syntactic objects in the numeration that have already been merged into the derivation but still have features that must be checked (these features have “survived” previous mergings). The authors demonstrate that the argument/ adjunct asymmetries result from the nature of this Remerge operation. Anjum Saleemi’s chapter, “On the Interface(s) between Syntax and Meaning”, is not directly concerned with any particular empirical problems. Instead, it addresses the foundational issue of the tension between form and content; more specifically, between morpho-syntactic form and meaning. Saleemi takes a rather unconventional view of the issue, as his approach argues for a conception of (broad) syntax, or linguistic derivations in general, that encompasses the pragmatic-illocutionary as well as the conceptual-lexical aspects of meaning as essential to the syntax-meaning interface(s), thus going well beyond LF and much else that is standardly held to be part of minimalist derivations. Saleemi also suggests entirely giving up the bifurcated derivational model, which is to say, the familiar Y-/T-model and what is claimed to be its residual equivalent within the MP (cf. (1)

Exploring interfaces

15

above), and the ‘neo-Saussurean’ reconceptualization of the minimalist linguistic theory undertaken by him involves the deployment of linear or horizontal derivations wherein linguistic form and meaning are connected with each other by means of a series of operations involving both narrow and broad syntax. An attempt is made to demonstrate that the framework in question is sufficient to successfully explain a range of phenomena, such as the conceptual basis of argument and event structure within v*P and the relationship between CP (or ForceP, as part of it is termed under the split-CP hypothesis) and illocutionary force. Some of the other empirical matters discussed and re-analyzed from Saleemi’s perspective include rightdislocation, polarity, the scope of negative and conditional operators, and binding. Saleemi’s chapter is both promising and promissory in note; it is undoubtedly revisionist, but it is far from clear to what extent it is also innovative. Whether the revised version of the MP he outlines is indeed executable seems less than certain, but it probably would be unfair to deny the basic force of some of Saleemi’s arguments, particularly the view that linguistic meaning is not just limited to propositional content, but should also incorporate its possible illocutionary extensions, which, it is suggested, are generative in nature, much like the other components of the human linguistic system. On the whole Saleemi’s take on the issues under consideration appears to be not to try to reduce meaning to form, or vice versa, and to deemphasize some of the day-to-day, arguably often merely notational, concerns of execution. On his view the syntax-meaning relationship needs to be redefined and reformulated in such a fashion that derivations do not lose sight of their overall purpose, namely, that of connecting form and meaning. In “Dynamic Economy of Derivation”, Takashi Toyoshima extends the notion of derivational economy to include limited (one-step) look-ahead, and further argues that contrary to standard assumptions, Move may in fact be more economical than Merge, at least in certain contexts, throwing a fresh light on the issue of the computational complexity and the nature of economy conditions in the minimalist program (see also the brief discussion around (6) above). Toyoshima sees that the crux of the issue is misplaced and has not yet been settled. The issues at stake are dealt with in dichotomous terms of representational vs. derivational economy, local vs. global economy, static vs. dynamic economy, deterministic vs. non-deterministic economy, Shortest Move vs. Shortest Derivation, and so forth. Drawing from the mathematical theory of computational complexity, Toyoshima takes on Fukui’s (1996) view on the nature of economy in language

16 Kleanthes K. Grohmann as discrete optimization. He then argues that the issue of economy in syntax is not an either/or-question, but rather whether a local solution can be found for the fundamentally global nature of derivation. He claims that the derivationally local one-step look-ahead is not a problem but rather the very essence of derivational economy, and what is to be avoided are global comparisons of derivations entailed in derivationally non-local look-farahead. Theoretical arguments are interwoven with some empirical ones from raising constructions and expletive insertion – the perennial EPP problem. Observing that the standard assumption of Merge over Move faces misgeneration problems in raising constructions, Toyoshima demonstrates that neither local economy in the sense of Collins (1997) nor Chomsky’s (2000) lexical (sub)array solve the problems, the latter without inducing a massive computational complexity. As a solution, Toyoshima proposes the Principle of Minimum Feature Retention that dynamically makes a derivationally local, deterministic choice of operations between Merge and Move. Even though it may still remain to be seen how the mathematical theory of computational complexity relates to the human linguistic competence as a biological system, perhaps the most important part of this contribution is to combine minimalist syntax with the mathematical theory of computational complexity, which may be the first attempt that forces the economy metric involved to be explicit. The final chapter of this volume, “The Conceptual Necessity of Phases: Some Remarks on the Minimalist Enterprise” by Dennis Ott, addresses some of the conceptual underpinnings of Phase Theory. He takes a phase to be essentially a syntactic structure that can be mapped onto a ‘thought’. From this view, it follows that the phases will correspond to the basic semantic categories that exist independently in the C-I system: presumably, at least propositions (CP), thematic structures (vP), and referential expressions (DP). Ott argues that phases, understood in this sense, are conceptually necessary once generation is reduced to a single cycle, emphasizing at the same time that the syntactic repercussions (impenetrability, etc.) are all but clear. Conceivably, phasal mapping to C-I implies no NS-effects whatsoever. The closing chapter is, in a way, the contributions that looks at the modeling from modular to linguistic interface, in the sense described above, at least for mapping NS to LF and subsequently to C-I.

Exploring interfaces

17

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. 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 2007 Agree, phases, and intervention effects. Linguistic Analysis 33 (1–2): 54–96. 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. 1981 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. 1991 Some notes on economy of derivation and representation. In Principles and Parameters in Generative Grammar, Robert Freidin (ed.) , 417–454. Cambridge, MA: MIT Press [Reprinted in Chomsky 1995: 129–166.]

18 Kleanthes K. Grohmann 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. 2005 Three factors in language design. Linguistic Inquiry 36: 1–22. 2006 Turing’s thesis. Keynote address presented at InterPhases: A Conference on Interfaces in Current Syntactic Theory, Castelliotissa Hall, Nicosia (18–20 May 2006). 2007a Approaching UG from below. In Interfaces + Recursion = Language? Chomsky’s Minimalism and the View from Syntax-Semantics, Uli Sauerland and Hans-Martin Gärtner (eds.), 1–29, Berlin /New York: Mouton de Gruyter. 2007b Of minds and language. Biolinguistics 1: 9–27. 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. Collins, Christopher T. 1997 Local Economy. Cambridge, MA: MIT Press. den Dikken, Marcel 2007 Phase extension: Contours of a theory of the role of head movement in phrasal extraction. Theoretical Linguistics 33: 133–163. Dobashi, Yoshihito 2003 Phonological phrasing and syntactic derivation. Ph.D. dissertation, Cornell University, Ithaca, NY. Epstein, Samuel David, Erich M. Groat, Ruriko Kawashima and Hisatsugu Kitahara 1998 A Derivational Approach to Syntactic Relations. New York: Oxford University Press. Epstein, Samuel David and T. Daniel Seely 2006 Derivations in Minimalism. Cambridge: Cambridge University Press. Fox, Danny 2003 On logical form. In Minimalist Syntax, Randall Hendrick (ed.), 82– 123. Malden, MA: Blackwell.

Exploring interfaces

19

Fox, Danny and David Pesetsky 2005 Cyclic Linearization of Syntactic Structure. Theoretical Linguistics 31: 1–45. Fukui, Naoki 1996 On the nature of economy in language. Cognitive Studies: Bulletin of the Japanese Cognitive Science Society 3: 51–71. 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. 2007a Deriving dynamic interfaces. Linguistic Analysis 33 (1–2): 3–19. 2007b Transfer vs. spell-out and the road to PF. Linguistic Analysis 33 (1– 2): 176–194. 2007c Spelling out dynamic interfaces. Linguistic Analysis 33 (3–4): 197– 208. 2008 Spell-out rules: Ranked competition of copy modification. Invited lecture presented at Interface Theories: The Filtering of the Output of the Generator (DEAL II), Universiteit Leiden (22–23 February 2008). 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. 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, Jairo Nunes and Kleanthes K. Grohmann 2005 Understanding Minimalism. Cambridge: Cambridge University Press. Kempson, Ruth, Wilfried Meyer-Viol and Dov Gabbay 2001 Dynamic Syntax: The Flow of Language Understanding. Oxford: Blackwell.

20 Kleanthes K. Grohmann 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.] Kayne, Richard S. 1994 The Antisymmetry of Syntax. Cambridge, MA: MIT Press. Ko, Heejeong 2005 Syntactic edges and linearization. Ph.D. dissertation, MIT, Cambridge. Lasnik, Howard and Juan Uriagereka with Cedric Boeckx 2005 A Course in Minimalist Syntax: Foundations and Prospects. Malden, MA: Blackwell. Lebeaux, David 1991 Relative clauses, licensing, and the nature of the derivation. In Perspectives on Phrase Structure: Heads and Licensing, Susan Rothstein (ed.), 209–239. San Diego, CA: Academic Press. Marušič, Franc Lanko 2005 On non-simultaneous phases. Ph.D. dissertation, Stony Brook University, NY. Nissenbaum, Jonathan W. 2000 Investigation of covert phrase movement. Ph.D. dissertation, MIT, Cambridge. O’Grady, William 2005 Syntactic Carpentry: An Emergentist Approach to Syntax. Mahwah, NJ: Lawrence Erlbaum. Phillips, Colin 1996 Order and structure. Ph.D. dissertation, MIT, Cambridge. 2003 Linear Order and Constituency. Linguistic Inquiry 34: 37–90. Radford, Andrew 2004 Minimalist Syntax: Exploring the Structure of English. Cambridge: Cambridge University Press. Ramchand, Gillian and Charles Reiss (eds.) 2007 The Oxford Handbook of Linguistic Interfaces. Oxford: Oxford 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.

Exploring interfaces

21

van Riemsdijk, Henk 1978 A Case Study in Syntactic Markedness: The Binding Nature of Prepositional Phrases. Dordrecht: Foris. 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). Selkirk, Elisabeth 1984 Phonology and Syntax. Cambridge, MA: MIT Press. Shiobara, Kayono 2005 Linearization: A Derivational Approach to the Syntax-Prosody Interface. Ph.D. dissertation, University of British Columbia, Vancouver. 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. 2002 Derivations. London: Routledge. 2008 Syntactic Anchors: On Semantic Structuring. Cambridge: Cambridge University Press. Uriagereka, Juan and Roger Martin 1999 Lectures on dynamic syntax. Lecture series given at the LSA Summer Institute, University of Illinois, Urbana-Champaign (June 21– July 30, 1999). 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.

Intermodular argumentation and the word-spell-out-mystery Tobias Scheer

1. Intermodular argumentation and its conditions This contribution calls attention to the intermodular potential of the interactionist architecture that multiple spell-out (Uriagereka 1999) and derivation by phase (Chomsky 2000, et seq.) have introduced. The shipping back and forth of pieces between (morpho-)syntax and the PF/LF interfaces during the derivation of a sentence establishes a pipe between the concatenative and the interpretational devices that did not exist in GB or earlier versions of the inverted T-/Y-architecture. It creates a situation where syntactic theories and analyses may have direct consequences on the phonological side, and vice versa. I submit that intermodular argumentation provides stronger evidence than what can be produced by modular-internal reasoning: it offers the maximal degree of independent assessment that linguists can expect without leaving their discipline. Be it only for that reason, the new interactionist architecture that the minimalist orientation has installed is a good thing to have: after a long period of quasi-silence, syntacticians and phonologists can talk again about things not concerning the weather or job openings. This is one goal of the paper, which will be briefly illustrated by two case studies where syntax can be argued to act as a referee for competing phonological theories (the PIC and the phase edge). The other goal is to make the conditions of intermodular argumentation explicit: what it can do, want it cannot do, and what it can do only if this or that question is settled beforehand. As far as I can see, much of the intermodular refereeing potential unfortunately hinges on a question that is subject to ongoing debate: whether morphology is just the lower part of syntax, or whether it is a computational system in its own right that builds hierarchical structure which is distinct from syntactic structure. Depending on this question is the number of spell-out mechanisms that exist in grammar, which, as we will see, is critical for the comparison of phonological and syntactic effects of cyclic spell-out. That is, if it turns out that morphology and syntax are two distinct computational systems, it could be argued with

24 Tobias Scheer some right that each system comes with its own spell-out mechanism. Note, however, that this is not a necessary conclusion: two distinct structures may as well be ‘harvested’ by the same mechanism. On the other hand, in case morphology and syntax are found to be emanations of one and the same computational system, there is of course no room for two distinct spell-out mechanisms. The reason why the number of spell-out mechanisms matters is a massive empirical asymmetry that has been encoded in phonological theories of the interface without, however, having been made explicit as such. I call this phenomenon the word-spell-out-mystery which – in a nutshell – is about the absence of phonological effects of the cyclic spell-out of words and larger chunks (against the plethoric phonological traces of the cyclic spellout of morphemes). There is no good reason for this asymmetry: nobody doubts that spell-out itself is cyclic – whatever the size of the chunks in question. Why, then, should an interpretational system (phonology) be sometimes impacted by the cyclic (wave-)structure of what it receives, but at other times simply ignore it? The null hypothesis is certainly that cyclic chunk-submission will leave traces in whatever (system) is exposed to its piecemeal fire. On this count, sub-word phonology is normal, but word phonology is weird. We start out with three examples of what intermodular argumentation can look like, before, alas, discovering that this argumentation hinges on a number of conditions that involve a major question of linguistic theory which is unlikely to be settled before some time to come (whether morphology and syntax are one or two computational systems), and precisely the empirical validity of the word-spell-out-mystery mentioned (which might turn out to be no mystery at all). The reader will have to fight through a number of additional issues down a decision tree where every ‘yes’ opens space for additional questions (be prepared for three levels in the tree). Unfortunately, indecision will be greater at the end of this chapter than after a couple of pages.1

1

The present article is a piece of Scheer (forthcoming), and includes (sections 2 and 3) most of the material of two other papers that originate in this book (Scheer, in press a,b).

Intermodular argumentation and the word-spell-out-mystery 25

2.

Intermodular argumentation

2.1. Interactionism, selective spell-out and the PIC have a phonological ancestor The minimalist focus on the interface has afforded a radical change in generative interface architecture. Since the 1960s (Chomsky 1965: 15ff.), the inverted T-model stands unchallenged (the generative semantics interlude lain aside): a concatenative device (morpho-syntax) feeds two interpretative devices (PF and LF). This architecture was supplemented with a proviso which requires that all concatenation be done before all interpretation. That is, the morpho-syntactic derivation is completed, and the result (S-structure) is then sent to PF and LF in one go. An alternative view of the communication between morpho-syntax and LF/PF was formulated in phonology in the early 1980s: the backbone of Lexical Phonology (Pesetsky 1979, Kiparsky 1982), so-called interactionism, holds that concatenation and interpretation are intertwined. That is, first some pieces are merged, the result is interpreted, then some more pieces are concatenated, the result is again interpreted, and so on. While GB-syntax of that time hardly produced any echo, generative orthodoxy in phonology reacted on this violation of “all concatenation before all interpretation”: Halle & Vergnaud (1987) proposed a non-interactionist version of Lexical Phonology that restores the interface landscape of SPE to a large extent. Halle & Vergnaud (1987) also promote a new idea: selective spell-out. Since cyclic derivation was introduced by Chomsky et al. (1956: 75) and formalized in Chomsky & Halle (1968: 15ff.), interpretation was held to run through the bracketed string (that is inherited from S-structure) from inside out; (roughly 2) every morpheme break defined a cycle. Halle & Vergnaud dispense with this definition of what an interpretational unit is: they propose to grant cyclic status only to a subset of morpho-syntactic divisions. That is, some nodes trigger interpretation, others do not.3 2

3

In actual fact, SPE holds that all morphemic and syntactic divisions are cycles, except for sequences of morphemes that belong to the same major category, which belong to the same cycle (hence [[[theatr]N ic + al]A i + ty] N, Chomsky & Halle 1968: 88f.). An important question is how to decide whether a given node is spelled out or not. The phonological take of Halle & Vergnaud (as much as of all other phonologists who practise selective spell-out) is that this depends on a lexical property of the piece (the affix) that is merged. In Halle & Vergnaud’s terminology, there

26 Tobias Scheer The reader will have understood that selective spell-out is exactly what modern (syntactic) Phase Theory is about: in more familiar terminology, nodes may or may not be phase heads, hence their material may or may not be an interpretational unit. As far as I can see, the phonological heritage is left unmentioned in the syntactic literature since derivation by phase was introduced by Epstein et al. (1998), Uriagereka (1999) and Chomsky (2000, et seq.). This is also true for interactionism: Epstein et al’s (1998) and Uriagareka’s multiple spell-out as well as Chomsky’s derivation by phase make the generative interface architecture interactionist, exactly along the lines that Lexical Phonology had laid out: first you do some concatenation, then some interpretation, then some more concatenation etc. For (extra-linguistic) reasons of computational economy regarding the limited availability of active memory, a costly cognitive resource (e.g. Chomsky 2000: 101, 2001: 15), modern Phase Theory applies the interactionist world view. Here again, the phonological origin of the idea has gone unnoticed as far as I can see (let alone the anti-interactionist reaction of generative orthodoxy in the 1980s). On the pages below we will also come across a question that is closely related to selective spell-out and interactionism: critical for current syntactic Phase Theory is a device which guarantees that previously interpreted strings do not burden further computation – in Chomsky’s terms, strings that are returned from interpretation are ‘frozen’ and ‘forgotten’ when concatenation resumes. No look-back devices are around in generative theory since Chomsky’s (1973) Conditions on Transformations, and their offspring – until its recent revival in the coat of the Phase Impenetrability Condition (PIC) – was essentially phonological (e.g. Mascaró’s 1976 and Kiparsky’s 1982 Strict Cycle Condition). No look-back devices are designed in order to prevent computation to consider ‘old’ strings. Depending on their precise formulation, however, they may have very different empirical effects, which correspond to the thing that the analyst wants the computation to be unable to do. We will see in section 3.2.2 that here again, are cyclic (interpretation-triggering) and non-cyclic (interpretation-neutral) affixes. Under the header of phasehood, this is also an important question discussed in current syntactic Phase Theory (more on this in section 3.3). Unlike in phonology where phasehood depends on a lexical property of affixes, the syntactic take is that it depends on the label of nodes (which of course is also a projection of a lexical property, but in a different sense): I call the two options node-driven vs. piece-driven phase (the contrast is further discussed in Scheer, forthcoming, in press b).

Intermodular argumentation and the word-spell-out-mystery 27

Chomsky’s PIC has a phonological precedent: unlike all other no look-back devices that the literature has accumulated since 1973, Kaye’s (1992, 1995) mechanism inhibits the modification of previously interpreted strings – which are thus ‘frozen’.

2.2. A syntactic referee for phonological theories In contrast to GB, where the completed morpho-syntactic derivation was merely dumped into PF (and LF) with a “good bye and don’t come back”, Phase Theory establishes a two-way pipe between the morpho-syntactic and the phonological (and semantic) modules. Actors on both ends are not free anymore to do what they want: their theories and analyses may make predictions on the other end. The intermodular potential of Phase Theory, however, has not received much attention thus far. Syntacticians use Phase Impenetrability for syntax-internal purposes, and Phase Theory evolves at high speed without taking into account what happens when the parcel spends time on the phonological side. On the other hand, phonologists have barely acknowledged the existence of Phase Theory, let alone taken into account the predictions that it makes on the phonological side. Below I first expose three intermodular arguments that I have made elsewhere: in each case, the existence of a device in current syntactic theory is taken to evaluate competing phonological theories according to whether they provide for this device or not. That is, since derivation by phase is based on selective spell-out, the PIC and the phase edge, phonological effects of cyclic spell-out must also feature these devices. Phonological theories that require all nodes to be spelled out, where no look-back devices play no role, or which do not spell out the sister of the phase head (cf. the phase edge), do not qualify. Once this is completed, I step back in order look at the conditions that need to be met for intermodular arguments to bite. The baseline is the fact that the chunks which are designated by the spell-out mechanism for interpretation must be the same on the syntactic side and at PF/LF. This presupposes that we have the same spell-out mechanism – something that seems to go without saying, but which deserves a second thought. This second thought is induced by the word-spell-out-mystery, which will be examined.

28 Tobias Scheer 3.

How syntax can referee phonological theories

3.1. Morpheme-specific phonologies (two engines) and non-selective spell-out Let us look at three cases where phonological theories may be evaluated by syntactic theory. The arguments that are reviewed in this section are originally made in Scheer (in press a,b). All concern phonological theories of affix class-based phenomena. We therefore start by introducing relevant evidence, as well as analyses thereof that the phonological literature has produced over the past 30 years. Both have been extensively discussed in the literature since the 1980s; I therefore only introduce aspects that are critical for the demonstration. Affix classes are best studied in English (see Booij 2000: 297 for a literature overview regarding other languages). Their existence was identified in SPE (Chomsky & Halle 1968: 84ff.); since then, the basic diagnostic for class membership is the behaviour of affixes with respect to stress: they may be stress-shifting (class 1) or stress-neutral (class 2). While the former roughly correspond to the Romance stock of lexical material (e.g. -ity, -ic, -ion, -ary, -aladj), the latter typically are of Germanic origin (e.g. -ness, -less, -hood, -ship, -ful). Relevant overview literature includes, Giegerich (1999), McMahon (2000), and Bermúdez-Otero (forthcoming). For example, a root such as párent appears with regular penultimate stress when it occurs in isolation; adding the stress-shifting affix -al produces parént-al, while the stress-neutral item -hood yields párent-hood. Another way of looking at these facts is that both párent and parént-al bear transparent penultimate stress, while párent-hood illustrates an opaque nonpenultimate pattern where stress behaves as if the suffix were not there. In other words, stress has been reassigned when -al was added (stress-shifting), but reassignment was blocked upon the merger of -hood. The task for the analyst is thus to organize underapplication of the stress rule, which must somehow be prevented from reapplying to strings that are headed by class 2 affixes. The table in (1) shows the solution that is proposed by Lexical Phonology.

Intermodular argumentation and the word-spell-out-mystery 29

(1) párent – parént-al vs. párent-hood in Lexical Phonology lexicon level 1 level 2

concatenation stress assignment concatenation rule application

parent parent – párent – –

parént-al parent parent-al parént-al – –

párent-hood parent – párent párent-hood –

The spine of Lexical Phonology is its stratal architecture: the lexicon contains underived roots, all class 1 affixes are concatenated at stratum 1 (level 1), while class 2 affixes join in at stratum 2 (level 2). After the concatenation is complete at each stratum, a stratum-specific phonology applies to the string as it stands. Rules are assigned to specific strata: in our example, the stress-assigning rule is a level 1 rule, which means that it is active at level 1, but absent from level 2. Another ground rule is that the derivation is strictly serial: given the order lexicon → level 1 → level 2, strings that are present at some level must run through all subsequent levels on their way to the surface. This means that they experience the computation that these levels. Under (1), then, /parent/ in isolation receives stress at level 1, where stress assignment is active. This is also true for /parent-al/, since -al has been concatenated in time. Stress assignment to /parent-hood/, however, concerns only /parent/ since -hood has not yet joined in. After its concatenation at level 2, stress does not move, since the stress rule is absent from this stratum. Note that this is critical: otherwise *parént-hood would be produced. Underapplication of stress assignment at level 2 is thus achieved by the split of phonological computation into two morpheme-specific mini-grammars: one that assesses class 1 strings (where the stress rule is present), another that takes care of class 2 strings (where the stress rule is absent). The set of rules that applies at level 1 is thus necessarily distinct from the set of rules that applies at level 2 – both phonologies specifically apply to a certain class of morphemes. Morpheme-specific phonologies have been carried over from serial Lexical Phonology into the constraint-based environment of OT, where the two-engine approach falls into two varieties – serial and parallel. On the one hand, Stratal OT (Kiparsky 2000; Bermúdez-Otero, forthcoming) and DOT (Rubach 1997 et seq.) faithfully continuate the stratal architecture of Lexical Phonology: strata are serially ordered, and any string that was present at stratum n–1 must run through stratum n and all subsequent strata. In OT,

30 Tobias Scheer differences among mini-grammars (the two engines) are expressed by means of different rankings of the same universal constraint set. Morpheme-specific phonologies therefore incarnate as different constraint rankings. That is, constraints are re-ranked between strata. The alternative implementation of the two engine-approach is parallel: class 1-strings are assessed by two distinct computational systems X and Y; the former applies to class 1, the latter to class 2-strings. In contrast to the serial solution, however, class 1-strings never meet class 2-computation, and vice versa: nothing is serially ordered, and hence strings that are headed by a class-specific affix do not run through other ‘strata’ (there are no strata in this approach) on their way to the surface. There are two competing representatives of this solution, co-phonologies (e.g. Inkelas 1998) and indexed constraints (e.g. Pater 2000). Further detail would lead too far afield. It suffices here to bear in mind that all versions of Lexical Phonology, past and present, share two essential properties: (i) phonology is made of two distinct computational systems and (ii) spell-out is non-selective. The latter is a consequence of the former: interpretation occurs upon every concatenation of an affix, only is the string sent to distinct mini-phonologies according to its morphological properties.4

4

It was mentioned earlier that the classical take of Lexical Phonology is to spell out only when two successive morphemes belong to different affix classes (“at the end of every stratum”), rather than at every morpheme break. This option is reminiscent of SPE (see note 2) and empirically indistinguishable from a situation where literally every boundary triggers interpretation (which is what some versions of Lexical Phonology actually practise).

Intermodular argumentation and the word-spell-out-mystery 31

3.2.

Selective spell-out and only one computational system (one engine)

3.2.1. Halle & Vergnaud (1987): Selective spell-out Halle & Vergnaud (1987) have introduced an alternative that works with only one computational system.5 The heart of their mechanism is selective spell-out. The idea has already been introduced in section 1: only some nodes of the morpho-syntactic tree trigger spell-out. Whether or not a node dominates an interpretational unit (i.e., is a phase head or not) is decided by its head: affixes are lexically specified for being interpretation-triggering (cyclic affixes in Halle & Vergnaud’s terms) or interpretation-neutral (noncyclic). This property is then inherited by the node that they project, and the spell-out mechanism does or does not send nodes to PF/LF according to it. Under (2) below, β triggers spell-out because it is projected by the class 1 affix -al; by contrast under (2), the stress-neutral class 2 affix -hood does not provoke the interpretation of its node. (2)

Halle & Vergnaud (1987): analysis of affix class-based stress a. parént-al

b. párent-hood

β α

al n

phon parent

spell-out [[parent] al]

5

β

phon

α

hood n

phon parent spell-out

[parent] hood

Halle & Vergnaud (1987) is a book about stress, not about the interface. The interface theory that it contains has only really emerged in subsequent work: Halle et al. (1991), Halle & Kenstowicz (1991), and Odden (1993). Modern offspring includes Halle & Matushansky (2006) and Halle & Nevins (forthcoming). I use Halle & Vergnaud (1987) in order to refer to the entire line of thought in recognition of the fact that this book appears to be the first source in print (except a 1986 unpublished manuscript of Halle’s which to date I was unable to hunt down).

32 Tobias Scheer An additional proviso is that all roots are interpretational units by themselves (Halle & Vergnaud 1987: 78). This is integrated into (2) by the fact that the root node α is always spelled out. The difference between parént-al and párent-hood, then, is one of cyclic structure: in addition to the root, the former is subject to interpretation as a whole, while the latter is not. The input that enters phonology is thus /[[parent] al]/ vs. /[parent] hood/.6 Penultimate stress assignment then applies to each cycle: while the derivation ends for the latter item when [párent] has received stress (there is no further cycle), it reapplies to [párent al]; that is, stress is shifted to the right, and the result is parént-al vs. párent-hood ([parent] in isolation of course comes out as párent). This analysis achieves underapplication by selective spell-out: class 2 affixes do not trigger interpretation, which prevents the stress rule from reapplying. Two more ingredients, however, make crucial contributions to the result: it was already mentioned that roots are always spelled out by themselves – this is nothing that selective spell-out enforces per se. Also, class 1, rather than class 2 affixes, are interpretation-triggering – this choice is not determined by any property of the theory either. In sum, then, Halle & Vergnaud achieve the same affix class-based effect as Lexical Phonology (and modern incarnations thereof), but without recurring to morpheme-specific phonologies: there is only one computational system that assesses all strings.7

3.2.2. Kaye (1995): A different implementation of selective spell-out Kaye (1992, 1995) adopts selective spell-out and, like Halle & Vergnaud, rejects morpheme-specific phonologies. The implementation of selective 6

7

Recall that Halle & Vergnaud are anti-interactionist, i.e. need to complete the morpho-syntactic derivation before the full string, augmented with cycle-defining brackets, is sent to PF for interpretation. For the sake of completeness, it needs to be mentioned that the single computational system at hand only refers to the contrast with morpheme-specific multiple phonologies. It disregards chunk-specific phonologies, which apply only to a certain size of pieces. Chunk-specific phonologies have been proposed for the word level (SPE’s word-level rules, adapted basically in all subsequent theories) and for the contrast between sequences of morphemes and sequences of words (in the familiar vocabulary of Lexical Phonology, the former are lexical, the latter post-lexical rules). Chunk-specific phonologies are a separate issue (see Scheer, forthcoming, for further discussion).

Intermodular argumentation and the word-spell-out-mystery 33

spell-out, however, is significantly different given the ‘secondary’ choices that Kaye makes. A comparison appears under (3) below.8 (3) differences between Halle & Vergnaud (1987) and Kaye (1995)

a. the root is an interpretational unit b. the word is an interpretational unit c. interpretation-triggering affixes trigger the spell-out of d. type of English affix-class that triggers interpretation e. underapplication is achieved by

Halle & Vergnaud

Kaye

yes no their own node class 1

no yes their sister

cycles

cycles and no look-back

class 2

Unlike in Halle & Vergnaud’s approach, the root is not an interpretational unit (i.e. a cycle) per se in Kaye’s system. By contrast, the word is always an interpretational unit (while, recall, it is not with Halle & Vergnaud: /[parent] hood/). A third contrast is that in Kaye’s system, it is class 2 affixes that are interpretation-triggering, while this privilege was granted to class 1 affixes by Halle & Vergnaud. Finally, the critical difference for the global purpose of this chapter is that the sister of the interpretationtriggering affix, rather than the node that dominates the affix itself, is spelled out in Kaye’s system. Table (4) below depicts this difference.

8

Space restrictions only allow for a digest version of the comparison, and also of the presentation of Kaye’s system. Further discussion is provided in Scheer (forthcoming, in press a).

34 Tobias Scheer (4)

interpretation-triggering affixes: what exactly is spelled out a. Halle & Vergnaud (1987): cyclic affixes trigger the spell-out of their own constituent β β

β

phon α

Xtrigg.

b. Kaye (1995): cyclic affixes trigger the spell-out of their sister α

α

Xtrigg.

x

root

phon

x

spell-out [root X]

root spell-out

[root] X

Given an interpretation-triggering (i.e. cyclic) affix X and a root, two significantly distinct results are produced: /[root X]/ vs. /[root] X/. Note that this is only the isolated result of the action of the affix, which needs to be supplemented by the computation-independent provisos under (3): the root is always a cycle with Halle & Vergnaud, the word is always an interpretational unit in Kaye’s system. This leaves us with identical structures: /[[root] X]/ is produced on both sides. This does not mean, however, that the two different spell-out strategies return identical results. The contrast is shown under (5) below where strings with interpretation-neutral (Y) affixes are opposed to strings with interpretation-triggering (X) affixes. (5) a. b.

root-Xtriggering root-Yneutral

Halle & Vergnaud

Kaye

[[root] X] [root] Y

[[root] X] [root Y]

The contrast between Halle & Vergnaud and Kaye thus concerns strings that bear an interpretation-neutral affix, and it is the result of the combined choices under (3). These are hard-wired in the two systems, i.e. independent of the situation in particular languages. By contrast, the analyst must still identify which are the interpretation-triggering and which are the interpretation-neutral affixes in every language studied – interface theory will

Intermodular argumentation and the word-spell-out-mystery 35

not help. It was already mentioned in (3) that Halle & Vergnaud and Kaye make opposite choices for English. Finally, Kaye’s analysis of the párent – parént-al vs. párent-hood pattern crucially relies on an ingredient that is absent from Halle & Vergnaud’s system, i.e. a no look-back device – modification-inhibiting no look-back to be precise (see (3)). Indeed, Kaye (1995) holds that previously interpreted strings cannot be modified by computation at later cycles. The striking parallel between Kaye’s no look-back in phonology and current ‘freezing’ Phase Impenetrability in syntax is the subject matter of Scheer (in press a), where it is argued that the absence of a PIC device disqualifies phonological theories. In Kaye’s system, then, párent and parént-al identify as [parent] and [parent al], respectively, while párent-hood comes down to phonology as the complex [[parent] hood]. Application of penultimate stress to the two former produces the correct result right away. The latter first receives penultimate stress on the inner cycle, i.e. [párent], but then Kaye’s PIC inhibits modification of this string (and hence stress shift) on the outer cycle. The English situation is further discussed in Scheer (in press a), namely, cases where underapplication must be achieved for class 1 strings (level 2 rules in Lexical Phonology, e.g. (sign, sign-ing2 vs. si[g]n-ature1) are examined. For the purpose of the present contribution, however, we have now taken stock of everything we need in order to compare the competing phonological theories of the interface that have been developed on the grounds of phonological effects with their cousins that are based on syntactic evidence.

3.3. Making sure that the PIC, the phase edge and selective spell-out are necessary properties of Phase Theory The logic of the intermodular argument is that some property of spell-out is firmly established on one side of the phasal pipe, and therefore must also be present on the other side. In our case, syntactic Phase Theory evaluates competing phonological theories. The logic of the argument thus requires to make sure that the evaluating syntactic mechanisms are firmly established as an indispensable property of the theory. This is certainly true of the PIC and the phase edge. These are absolutely critical for derivation by phase to work out at all. The PIC is the condition on computation that achieves the goal which motivates derivation by phase in the first place: it unburdens active memory. The phase edge is also a critical ingredient of Phase Theory: derivation by phase, supplemented with the ‘freezing’ PIC, cannot work in its absence, since it is undisputed that some material of phase heads

36 Tobias Scheer may continue to be available for further computation. Spelling out the complement of an interpretation-triggering XP is thus a necessary property of the spell-out mechanism. Finally, establishing selective spell-out as a necessary property of syntactic Phase Theory requires a few more words. Chomsky’s original take on phasehood identifies CP and vP, maybe DP (Chomsky 2005: 17f.), as phase heads. The DP track has been followed, and also DP-internal phases are argued for (Matushansky 2005). Den Dikken (2007: 33) provides an overview and argues that “[t]here is good reason to believe that DP and CP are each other’s counterparts in their respective domains (the noun phrase and the clause).” TP is also under debate: while Chomsky (e.g. 2000: 106, 2004: 124) is explicit on the fact that TP does not qualify as a phase head (because it is not propositional), den Dikken (2007) points out that according to Chomsky’s own criteria, this conclusion is far from being obvious. Indeed, TP is assumed to act as a phase head in a growing body of literature, and nodes below TP such as Voice0 (Baltin 2007, Aelbrecht 2008) and AspP (Hinterhölzl 2006) are also granted phasehood. It does not take a lot to predict the vanishing point of the atomization of phasehood: taken to the extreme, all nodes will trigger interpretation; or, in other words, interpretation occurs upon every application of Merge. As a matter of fact, this view is not utopia, but has actually been proposed in work by Samuel Epstein and colleagues: Epstein et al. (1998), Epstein & Seely (2002, 2006) practice Spell-out-as-you-Merge. Two things may be said in defense of selective spell-out. For one thing, it needs to be noted that it does not matter how fine- or coarse-grained interpretational units turn out to be – the only claim that selective spell-out makes is that there are nodes which are not spelled out. The distribution of phasehood over nodes is currently done according to quite distinct principles in syntax and in phonology (node-driven vs. piece-driven phase, see note 3) and therefore represents a separate field of investigation. The other thing is that there may be principled reasons that prevent the atomizing trend to reach Spell-out-as-you-Merge. Boeckx & Grohmann (2007) argue that the atomization of phasehood is marshaled by the antilocality of movement (see Grohmann 2003): if phase heads are too finegrained, escape-hatch movement through the phase edge out of the complement will not be possible anymore since it will be too local. I thus take it that at least for the time being Spell-out-as-you-Merge is marginal, and perhaps for principled reasons, syntactic spell-out is necessarily selective.

Intermodular argumentation and the word-spell-out-mystery 37

3.4. Phonology must provide for selective spell-out, the phase edge and the PIC We are now in a position to make the intermodular argument, which appears to be simple at first sight: if selective spell-out, the PIC and the phase edge are necessary properties of Phase Theory, phonological theories of the effects of cyclic spell-out must also have them. Recall from sections 3.1 and 3.2 that selective spell-out divides phonological theories in two camps, one where all nodes are spelled out (Lexical Phonology) and another where spell-out is selective (Halle & Vergnaud 1987, Kaye 1995). The former can thus be dismissed on intermodular grounds, while the latter qualifies. The PIC further filters phonological theories: Lexical Phonology and Halle & Vergnaud (1987) do not use any no lookback device,9 while modification inhibiting no look-back is the mechanism that achieves underapplication in Kaye’s system. Syntactic theory thus selects Kaye (1995), which is the only theory to pass both filters. Finally, let us look at the phase edge. Current syntactic Phase Theory holds that in case XP is a phase head, the spell-out of XP only triggers the interpretation of the complement; the head and its specifier – the edge of the phase – are spelled out only at the next higher phase (Chomsky 2000: 108 et seq.). Kaye’s (1995) version of interpretation-triggering affixes and Chomsky’s phase edge are contrasted under (6) below. (6)

The phase edge in syntax and phonology: spell out your sister! a. Chomksy (2000, 2001)

b. Kaye (1995)

phase head β

XP Spec

X’ X°

9

α

Xtrigg. comp

PF/LF

x

PF/LF root

Of course, (classical versions of) Lexical Phonology feature Kiparsky’s (1982) Strict Cycle Condition (SCC), which is a no look-back device. The SCC, however, is devised for derived environment effects – it plays no role in affix classbased phenomena. Further discussion of the history of no look-back devices in general, and of the SCC in particular, can be found in Scheer (forthc., in press a).

38 Tobias Scheer The parallel is quite striking: in both cases, given a phase head, the sister of its head is spelled out. In syntax, this is the complement (i.e. the sister of the head of the phase head XP); in Kaye’s system, this is the sister of the interpretation-triggering affix (whose projection is the phase head). Now in sections 3.1 and 3.2, we have come across three competing theories of cyclicity-induced phonological effects: one where all nodes are spelled out (Lexical Phonology), one where only the node that dominates interpretation-triggering affixes is spelled out (Halle & Vergnaud 1987), and one where only the sister of interpretation-triggering affixes is sent to interpretation (Kaye 1995, and more recently Ziková & Scheer 2007, Ziková 2008). In an intermodular prespective, then, if the spell-out mechanism spells out the complement of phase heads – their sister –, the latter is selected, while the two former must be dismissed.

4. The headstone of the argument: All effects are due to the same spell-out mechanism This is where we stand after the first round of intermodular argumentation. There is a second round (alas) because the premises of the argument turn out to be less obvious than they may appear to be at first sight. The headstone of the entire argumentation – which is tacitly assumed above – is that the syntactic and the phonological effects of cyclic spell-out that we are talking about are the two ends of the same pipe. That is, they are produced by the same spell-out mechanism. Were they not, there would be no reason for ‘phonological’ spell-out to mimic the properties of ‘syntactic’ spell-out: the latter could spell out the sister of the phase head, while the former spells out something else; the latter could also be selective and implement the PIC, while the former could ignore these devices. The existence of just one spell-out mechanism is probably intuitive, and taken for granted by most readers, who will never have thought of an option where the grammatical architecture accommodates more than one parcel-shipping company. We approach this question in two steps: first we have a second thought on the interpretation of the phase edge (in the present section), which will then set us on the track of a massive empirical generalization – the word-spell-out-mystery –, which has always remained unspoken in the phonological literature. Reconsider (6). Saying that Kaye’s way of doing spell-out is the phonological version of the phase edge is imprecise: it is certainly the result of phonological evidence, but it concerns morphology. That is, where

Intermodular argumentation and the word-spell-out-mystery 39

Chomsky’s mechanism spells out words and larger chunks, Kaye’s evidence is based on the spell-out of morphemes. What the parallel really is about, then, is syntax and morphology: chunks of whatever size seem to be spelled out by the same mechanism: spell out your sister! This result thus certainly contributes to the ongoing debate whether morphology and syntax are instances of the same computational system or not. Showing that they are is the goal of Distributed Morphology (see Embick 2007, for example, on this issue), while Lexical Phonology-affiliated morphological approaches (‘autonomous’ morphology, e.g. Booij et al. 2007) and syntactically oriented theories such as Ackema & Neeleman‘s (2005, 2007) argue for a distinct morphological device. Julien (2007) and Lieber & Scalise (2007) provide a survey of the issues. But the validity of this result as an intermodular argument (against phonological theories that do not spell out the sister of phase heads) precisely hinges on the debate at hand: if it turns out that morphology and syntax are two distinct computational systems, it could probably be argued that each system comes with its own spell-out mechanism (but note that this is not a necessary conclusion: two distinct structures may as well be ‘harvested’ by the same mechanism). In this case, the intermodular arguments that were made above do not bite since the phonological evidence for the phase edge concerns the spell-out of morphemes, whereas the syntactic evidence for the same device is based on the spell-out of words and larger pieces. If on the other hand it is found that morphological structure is just the lower part of syntactic structure, there can be only one spell-out mechanism, and the intermodular arguments made go through without further discussion. We have thus reached a point of indecision: the prima facie arguments only hold if we can be sure that there is only one single spell-out mechanism, and this question depends on the ever-lasting debate whether morphology and syntax are one. In this situation, an obvious thing to do is to try to circumvent this ‘technical’ difficulty by simply looking at phonological effects of the cyclic spell-out of words (rather than of morphemes). The striking fact is that there are none! This is at least what the phonological literature says – and what I call the word-spell-out mystery. Whatever its nature and eventual solution, it is clear that its existence is somehow related to our indecision problem – it can hardly be accidental that the critical piece of evidence that you need to look at in order to sort out two competing hypotheses does not exist.

40 Tobias Scheer 5.

The word-spell-out mystery

5.1. The absence of cyclicity-induced external sandhi: A consensual fact that theories build on, but do not talk about Traditional terminology distinguishes between internal and external sandhi. The former refers to phonological effects of morpheme boundaries, while the latter describes phonological effects of word boundaries. This distinction comes in handy for the purpose of this section, which is about something that the literature does not talk about: the cyclic spell-out of words. While the procedural (i.e. cyclic) management of morphemes has spilled a whole lot of ink (this is essentially what Lexical Phonology is about), I have not come across either a phonological phenomenon that requires, or an analysis that proposes, a procedural treatment of words or bigger chunks – except for intonation, on which more in section 5.4 below.10 The absence of procedural activity above the word level – or rather: of phonological traces thereof – is admitted as the correct empirical record in the field. The literature therefore offers only representational treatments of syntactically conditioned phonological effects. In practice, this means that all external sandhi phenomena are ascribed to some variation in prosodic constituency. The exclusive ambition of representational management at and above the word level is rarely made explicit, though. The only cases that I am aware of are Selkirk (1984) and Inkelas (1990). These authors observe that while prosodic constituency can cover the full spectrum of units (morphemes and words alike), Lexical Phonology is confined to the Lexicon, i.e. to morphemes. Since there is no place for two devices (procedural and representational) that do the same job below the word level, Inkelas (1990) argues, prosodic constituency should be extended to the Lexicon. Lexical Phonology, then, is an empty shell at best.

10

Traditionally there are two ways for morpho-syntax to bear on phonology: procedurally (cyclic spell-out, called the transformational cycle in SPE) and representationally (boundaries in SPE, i.e. # and the like, the prosodic hierarchy since the 1980s). The two channels may be easily distinguished, the simplest way being the fact that on the representational side some object – a boundary, a prosodic constituent or whatever item is favoured by the particular interface theory – is inserted into phonology independently of lexical (vocabulary) insertion (classically by mapping rules as in Prosodic Phonology).

Intermodular argumentation and the word-spell-out-mystery 41

5.2.

How (phonological) interface theories behave: Claims for and against cyclic spell-out of words, for and against its cyclic interpretation

5.2.1. The baseline position of SPE: everything is cyclic Given this empirical situation and its reception in the literature, let us briefly review the spectrum of positions have been taken by phonological interface theories. Two things need to be carefully distinguished: the submission of pieces to interpretational modules, which may or may not be cyclic on the one hand; and the phonological (or semantic) interpretation thereof (i.e. of whatever is submitted), which may also be cyclic or not. The former is a matter of the spell-out mechanism, while it is reasonable a priori to think that the latter is a decision of the phonological computational system: phonology either ignores that it receives a string in pieces and acts only at the end of the submission process, or assesses pieces as they are submitted. The baseline position is represented by Chomsky et al. (1956) and SPE, where cyclic derivation was introduced: both the spell-out and the phonological interpretation of word sequences is cyclic, as shown by the following quote: “The principle of the transformational cycle […] appl[ies] to all surface structure whether internal or external to the word” (Chomsky & Halle 1968: 27).

5.2.2. Lexical Phonology: The interpretation of word sequences is not cyclic Lexical Phonology has a different take. Following an insight from the Prague school (Booij 1997: 264, note 3), all versions of this theory implement what I call Praguian segregation, i.e. the distinction between word- and sentencephonology. This means that phonology subdivides into two distinct chunkspecific computational systems which compute sequences of morphemes (lexical phonology) and sequences of words (post-lexical phonology), respectively.11

11

The two chunk-specific phonologies, lexical and post-lexical, add to the distinct morpheme-specific phonologies that we have come across in section 3.1. That is, the Lexicon (where words are constructed) accommodates lexical phonology (word phonology), which itself falls into distinct morpheme-specific phononlogies. Lexical phonology as a whole is then opposed to post-lexical phonology (sentence phonology), and syntax applies in the midst, i.e. on the result of word, but before sentence-phonology. The total of distinct computational systems in

42 Tobias Scheer The term ‘cyclic rule’ (which is still used today in phonological quarters in a kind of lingua franca-understanding) is indicative of the front line that is set in Lexical Phonology: early versions of the theory assumed that all phonological rules which contribute to word-formation are cyclic (i.e. rules which apply to sequences of morphemes), while all post-lexical rules are non-cyclic (i.e. those rules that apply to sequences of words). The cyclic condition on lexical rules has been called into question later on (Rubach & Booij 1984 introduced lexical post-cyclic – i.e. non-cyclic – rules), but the necessarily non-cyclic character of post-lexical rules stands unchallenged in all Lexical Phonology quarters up to the present day. This is the only thing that is of interest for the present discussion.

5.2.3. Lexical Phonology makes no claim about spell-out and installs noncyclic interpretation of word sequences without argument On the other hand, as far as I can see, Lexical Phonology makes no claim regarding the cyclic character of spell-out. The only thing that is central for this theory is the contrast between the cyclic interpretation of morpheme sequences (lexical phonology), against the non-cyclic interpretation of word sequences (post-lexical phonology). The reasons for this fundamental distinction, however, are not made explicit as far as I can see. One may suppose that post-lexical phonology is declared non-cyclic on the grounds of the observation that cyclicityinduced external sandhi is absent from the record. This, however, is no more than a supposition: Kiparsky (1982) simply decrees that there is no cyclic interpretation of words without argument. The former, the rules of lexical phonology, are intrinsically cyclic because they reapply after each step of word-formation at their morphological level. The latter, the rules of postlexical phonology, are intrinsically noncyclic. Kiparsky (1982: 131f., emphasis in original)

Lexical Phonology is thus three: two in the Lexicon (or actually more, depending on the language and the particular brand of Lexical Phonology), one after syntax (post-lexical).

Intermodular argumentation and the word-spell-out-mystery 43

5.2.4. Halle & Vergnaud and Kaye: Restoration of SPE – everything is cyclic Let us now turn to Halle & Vergnaud (1987), who are committed to the SPE heritage, but also to Lexical Phonology and the lexicalist environment of the 1980s, of which Praguian segregation is an expression. The result is a system where both morphemes and words are subject to cyclic spell-out; the concatenative process, however, takes place in two rounds, one where words are created, another where sentences are built (word-internal vs. word-sequence strata in Halle et al. 1991). Following SPE, word- and sentence-construction is separated by a specific word-level phonology.12 This much for spell-out. Within this architecture, then, all phonological interpretation is cyclic, no matter whether the input are morphemes or words. This follows Halle & Vergnaud’s general orientation, which is to restore SPE: (i) there are no morpheme-specific phonologies, (ii) there is no distinction between a phonology of morphemes and a phonology of words: both chunk-sizes are interpreted by the same computational system; (iii) all phonological interpretation is cyclic. Kaye’s (1995) position is the same as Halle & Vergnaud’s as far as I can see: Kaye rejects morpheme-specific phonologies, but has morpheme- and word-sequences interpreted by the same computational system, which carries out cyclic interpretation across the board.

5.2.5. Distributed Morphology is entirely agnostic in phonological matters Finally, Distributed Morphology is entirely agnostic in regard of the issue at hand – simply because it is not concerned with, and does not make any claim about, phonological interpretation. From the vantage point of DM, morpho-syntax cannot accommodate multiple computational systems, but PF may or may not accommodate morpheme-specific and/or chunk-specific mini-phonologies, whose interpretational action also may or may not be cyclic.

12

In order to avoid confusion, below I do not mention Halle & Vergnaud’s specific word-level phonology anymore, which is irrelevant for the discussion.

44 Tobias Scheer 5.3. What a “trace of cyclic spell-out” is: PIC à la carte? Chunk-specific PIC? 5.3.1. Cyclic spell-out of words but no phonological traces? From a global perspective, the situation seems paradoxical: cyclic spell-out of words and larger chunks – derivation by phase in modern terminology – is a central piece of current syntactic thinking, but it looks like it has no phonological consequences. By contrast, the cyclic spell-out of morphemes is just as undisputed, but – as expected – leaves ample traces in phonology. Having distinct chunk-specific phonologies that distinguish word- and sentence-phonology as proposed by Lexical Phonology does not solve the problem: it merely records the contrast between the area which produces phonological effects (internal sandhi) and the area which does not (external sandhi). What it does not, however, is to correlate the phonological noneffect for chunks at and above the word level with the other end of the interactionist pipe: we want to know how it could be that the same input to (an) interpretational system(s) – the piecemeal submission of a string hacked into pieces of growing size – in one case produces (opacity) effects, but in another leaves no trace at all. 5.3.2. Wrong data or an on/off switch for Phase Impenetrability There are only two ways in which I can make sense of this mystery: either the empirical generalization is simply wrong (phonologists have not worked hard enough, if they have a closer look, they will find cyclicity-induced external sandhi) or interpretational systems are able to ignore their input conditions. The latter option means that a phonological system (or a semantic system for that matter) has a switch that decides whether ‘old’ strings, i.e. those that have already undergone previous computation, are subject to a special treatment or not. Or rather, as will be suggested below, the switch at hand is borne by the spell-out-mechanism (rather than by the phonological computational system). In other words, word phonology would feature a no look-back device, while sentence phonology has no Phase Impenetrability Condition and hence treats all strings in the same way, old and new alike. A third option that is logically possible is certainly not a serious candidate and may be left unexplored: spell-out of words and larger chunks could be non-cyclic (while morphemes are submitted piecemeal to interpretation). This would mean that cyclic derivation in general and interactionist derivation by phase in particular are flat out wrong.

Intermodular argumentation and the word-spell-out-mystery 45

5.3.3. Phase Impenetrability requires a memory-keeper Let us pursue the option according to which interpretational systems are parameterized for subjecting or not subjecting ‘old’ strings to a special treatment. An interesting question is what “special treatment” actually means, and by whom it is organized. As far as I can see, the literature on no lookback devices in general, and on Phase Impenetrability in particular, does not really address this question: no look-back is declared to be a property of the system, but what it actually takes for the system to implement the desired effect remains vague. It was already mentioned that the (Chomsky’s) whole motivation for cutting sentences into pieces and sending them to PF/LF piecemeal is to be able to achieve computational economy regarding active memory by imposing the Phase Impenetrability Condition on ‘old’ pieces. The economy effect is achieved by allowing further computation to ‘forget’ these ‘old’ pieces and their internal structure. Chomsky is also explicit on the fact that the economy of active memory concerns phonological as much as syntactic computation (“the phonological component too can ‘forget’ earlier stages of derivation”, Chomsky 2001: 12f.). Which means that Chomsky takes Phase Impenetrability to be a general condition on computational systems (at least in grammar); the alternative that we are currently pursuing has the opposite take: every computational system ‘chooses’ to benefit from computational economy or not. Now this computational economy does not come for free: it needs to be organized. Somebody must keep track of which portion of the string has already been subject to interpretation, and which portion is new. Everything that we know about how modules work suggests that the modular computation itself is perfectly unable to do this labour. Indeed, modular computation is known to have the following properties (e.g. Fodor 1983; Pylyshyn 1989; Smith 2002): it is automatic, mandatory and ‘blind’ (also domainspecific, i.e. content-based, autonomous, stimulus-driven and insensitive to central cognitive goals); that is, modules are input-output devices that perform a calculus on an input and return an output. Given this description, they are perfectly unable to distinguish between portions of a string submitted which are in need of computation, and other portions which are not: modules do not make decisions.

46 Tobias Scheer 5.3.4. Phase Impenetrability is a property of the spell-out mechanism, not of concatenative or interpretational systems This means is that Phase Impenetrability is not a property of computational systems such as morpho-syntax, phonology or semantics for that matter. What is it then a property of? The spell-out mechanism appears to be the only candidate for the management of Phase Impenetrability that is left. This means that the system which is supposed by Phase Theory has three, rather than two individuatable units (as far as the derivation of sound is concerned): a concatenative system (morpho-syntax), a spell-out system and an interpretational system. While the former and the latter are modules, the status of the spell-out mechanism is unclear (to me at least): it is reminiscent of mapping rules in Prosodic Phonology and Jackendoff ’s (1997) correspondence rules (more recently called ‘interface processors’; see Jackendoff 2002), but does not share the basic property of these, i.e. the ability to understand the vocabulary and the structure of the sending as much as of the receiving module. Rather, the spell-out mechanism reads the result of morpho-syntax and manages chunk-submission to phonology, which includes the distinction of ‘old’ and ‘new’ portions of the string. Figure (7) below tries to depict what this system could look like on an example where all nodes are phase heads, where (following regular assumptions regarding the phase edge) only the complement of the head of a phase head is actually sent to interpretation, and where the highest node is a CP, i.e. the end of the syntactic derivation.

Intermodular argumentation and the word-spell-out-mystery 47

(7)

No look-back managed by the spell-out mechanism morpho-syntax

action of the spell-out mechanism

end of the derivation

restores the content of the memory: [Z Y X W]

…

phonology

Y γ

1. adds X to memory 2. reads γ, sends Y X β

Z

1. stores [W] 2. reads β, sends X α

Y

X

reads α, sends W

W

W

When spell-out occurs at α, W is sent to phonology, where it is computed and sent back to morpho-syntax. It is then excluded (‘forgotten’) from further morpho-syntactic computation, but needs to be stored in order to be restituted at the end of the derivation. When spell-out treats β, only X is sent to phonology since, just like for morpho-syntactic computation, previously interpreted strings are excluded from phonological computation. This procedure works through the string until the end of the derivation, where the memory-keeper, that is the spell-out mechanism, restores its memory, which is the linear sequence of all interpreted pieces. This of course is only a rough schematic picture that does not come any close to what is actually going on. For one thing, the labour that the spellout mechanism is supposed to do under (7) describes only a vanishingly small portion of its real action (unpublished work by Michal Starke studies the properties of the spell-out mechanism in detail). Also, it is probably obvious for every phonologist that phonological computation needs to be able to see previously interpreted strings, even if it cannot modify them. These questions must be left pending. From a global minimalist perspective, the obligation for the spell-out mechanism (or some other device which is not on my radar) to act as a memory-keeper raises the question whether the trade-off between this extra burden for active memory and the computational economy that is realized by morpho-syntax and LF/PF is all that positive.

48 Tobias Scheer 5.4.

Intonation requires cyclic spell-out of words for sure – but this does not appear to impact phonological computation

5.4.1. Intonation is governed by syntactic structure Let us now examine intonation, the only case that I am aware of where the cyclic spell-out of words leaves phonological traces.13 The reader must be warned beforehand that the excursion into empirical fields will be quite lengthy – it is the price to pay if the empirical validity of the word-spellout-mystery is to be seriously addressed. That is, intonation is a typical concern of syntacticians, who will call the existence of the word-spell-outmystery into question on the grounds of the obvious cyclic conditioning of intonational phenomena. At least since Bresnan (1971), it has been established that intonation (also called sentence or phrasal stress) directly depends on syntactic structure. The topic is covered by a rich syntactic literature, including Berman & Szamosi (1972), Cinque (1993), Kahnemuyipour (2004) and Adger (2007). A classical example appears under (8) below (words in small caps bear prominent sentence stress). (8)

syntax-sensitive intonation a. b.

Helen left DIRECTIONS for George to follow. Helen left directions for George to FOLLOW.

(8) means that Helen has left some directions that George should follow, while (8) is an invitation for George to follow Helen. Since both sentences are phonologically identical but have contrasting syntactic structure, the different intonation must be a consequence of the latter: under (8) follow is transitive and belongs to a relative clause whose head is directions, while under (8) it is intransitive and complements directions. The causal relationship between syntax and phonology has also been thought of in the opposite direction: Szendrői (2001, 2003, 2004) argues that syntactic properties such as focus and climbing can be controlled by intonation. 13

I am aware of one other isolated case where the cyclic spell-out of words is argued to produce a segmental effect. According to Kaye’s (1995) analysis, French vowel nasalization goes into effect in mon ami [mç)n ami] ‘my friend’, but not in bon ami [bçn ami] ‘good friend’ because mon, but not bon, is a cycle in its own right: the relevant structures are [[mon] ami] vs. [bon ami]. Space restrictions preclude further discussion, which may be found in a Scheer (forthcoming).

Intermodular argumentation and the word-spell-out-mystery 49

In sum, there can be no doubt that cyclic spell-out of words and larger units impacts the calculus of intonation. In order for this fact to qualify as a prove that the cyclic spell-out of words may bear on phonology, however, it needs to be shown that the calculus of intonation is the result of phonological computation. Against the intuition that intonation is a phonological phenomenon, there may be good reason to believe that this is not the case. Another relevant issue is that intonation is the only phonological phenomenon which has been argued to be recursive (or rather: to require recursive structure, which as we will see is not quite the same thing). A pervasive and largely undisputed empirical generalization, however, is that recursion is the privilege morpho-syntax, the only concatenative device(s) in grammar (Chomsky et al. 2002, Neeleman & van de Koot 2006). This is reason enough to doubt that intonation, should it require recursive structure, is any phonological at all: its putative recursive nature disqualifies intonation as a phonological phenomenon. Whether or not intonation requires recursive structure will not be decided here. In case it does, however, it is shown that the bare existence of recursive structure in phonology does not mean that phonology itself is recursive as long as the structure in question has been built by mechanisms that are foreign to phonological computation. We will see that this is indeed the case for all recursive prosodic structure that has been argued for in the literature. We begin by considering this question. 5.4.2. Prosodic structure may be recursive … Beyond the syntactic literature, intonation has been studied in the perspective of Janet Pierrehumbert where it is phonologically represented as tones (e.g., Pierrehumbert 1980, 2000; Beckman & Pierrehumbert 1986, 1988 and Ladd 2000 provide overviews). Also, this approach puts to use the tools of Prosodic Phonology (among many others, Liberman 1975; Ladd 1986, 1997; Gussenhoven 1992, 2004; Selkirk & Kratzer 2007). As far as I can see, Ladd (1986) was the first to argue that intonation is recursive, in the sense that it requires recursive prosodic structure. Formally speaking, recursion is defined as a structure where a node is dominated by another node of the same kind. Ladd works with two prosodic constituents, the Major Phrase (MP) and the Tone Group (TG). He aims at showing that intonation cannot be adequately described unless an MP may dominate other MPs, and a TG other TGs. Nested prosodic structure was ruled out by the original version of Selkirk’s Strict Layer Hypothesis, and this is what Ladd stands up against.

50 Tobias Scheer Under this pressure (among others), Selkirk (1996) abandons the ban on recursive prosodic structure: in the new constraint-based formulation, the non-recursion of prosodic structure is demoted to a violable constraint. Languages and analysts are thus free to use nested prosodic structure. Since then, numerous analyses have taken advantage of this option (among many others, Booij 1996; Peperkamp 1997; and Truckenbrodt 1999).

5.4.3. …but phonology is not This state of affairs faces the fact that no-one has ever heard of recursive phenomena elsewhere in phonology. Recursive structure in syntax and morphology exists because sentences and morphemes may be embedded. Regular phenomena that lead to this description are shown under (9) below. (9)

recursion in syntax and morphology a. b.

Peter thinks [that John says [that Amy believes [that…]]] Czech iterative -áv dĕlat ‘to do’ dĕl-áv-at ‘to do repeatedly/often’ dĕl-áv-áv-at ‘to do even more often’ dĕl-áv-áv-áv-…-at ‘to do really really often’

Recursive structure in natural language has the property of producing grammatically unbounded embedding: grammar happily generates and tolerates an infinite number of embedded clauses, and in the case of recursive morphology, an infinite number of embedded morphemes. The limits on recursive structure in actual production are imposed by performance (factors such as memory), not by competence. That is, speakers will get confused upon the third of fourth level of embedding. As far as I can see, no equivalent phonological phenomenon has ever been reported (on recursion in general and its absence in phonology in particular, see Pinker & Jackendoff 2005a,b and Neeleman & van de Koot 2006). This empirical situation is the reason why the absence of recursion is widely held to be a major property that sets phonology (and semantics) apart from morpho-syntax. Using the same criteria as for the establishment of recursive structure in morpho-syntax, the empirical situation in phonology thus appears to be in overt conflict with the alleged existence of recursive prosodic structure – which has been admitted only fairly recently under the pressure of data from intonation.

Intermodular argumentation and the word-spell-out-mystery 51

But even if it turns out to be true that intonation is based on recursive structure, the trouble is that there is no reason why recursion should be confined to just this particular area of phonology: beyond intonation, phenomena analogous to those under (9) would be expected to produce, say, syllabic or segmental effects of the same kind.

5.4.4. What it means to say that “phonology is recursive”: Alien-inserted structure vs. phonological computation All worries mentioned turn out to be spurious when a distinction is made between the existence of prosodic structure and the computational system that it was created by. According to the classical conception (Nespor & Vogel 1986),14 prosodic structure ends up being part of phonology, where it may be referred to by phonological computation. However, it has not been created by phonology or anything phonological: prosodic structure is the output of mapping, which is done by a mapping device that is necessarily located outside of the phonology in modular no man’s land. Mapping is an exclusively top-down operation that transforms relevant morpho-syntactic structure into phonologically legible items (and eventually introduces distortions that are not motivated by the morpho-syntactic input, so-called non-isomorphism). The classical architecture that is required for the management of the prosodic hierarchy thus leaves us with a situation where recursive prosodic structure may well exist in phonology, but with phonological computation being perfectly non-recursive. This is because the recursive structure at hand has not been created by phonological computation – it has been inserted by somebody else, the mapping system. The meaning of the statement “phonology is recursive” therefore needs to be clarified: if somebody wants to refer to the existence of recursive structure in phonology, this statement is certainly true – at least for those 14

Though not in OT, where – in violation of modularity – prosodic structure is created by ALIGN and W RAP constraints in the phonology, i.e. the constraints at hand being interspersed with purely phonological constraints in the same constraint hierarchy. Mapping between morpho-syntax and phonology, which is what A LIGN and WRAP do, is a process that needs to be able to interpret morpho-syntactic structure – something that is impossible on modular grounds when sitting in phonology. More on the problematic relationship of OT with modularity in Scheer (forthcoming).

52 Tobias Scheer who consider it an advantage to have recursive prosodic structure in phonology. If on the other hand the intended meaning is that phonological computation is recursive, the statement is wrong: no evidence to this end has been reported, and intonation does not provide any. The contrast between intonation on the one hand and all the rest of phonology on the other, then, is not contradictory at all: there are no equivalent recursive phenomena on the segmental or on the syllabic side because these areas are the result of non-recursive phonological computation, not of (possibly) recursive mapping, which is managed beyond phonology.

5.4.5. PF and LF only interpret: Merge must be absent, hence the result of phononlogical computation is flat The absence of recursion in phonological (and semantic) computation is also central for the generative architecture of grammar. The inverted T-model embodies the insight that there is only one concatenative system in grammar, morpho-syntax. Phonology and semantics are interpretational systems (modules) which do not glue any pieces together: they merely interpret a string whose size, composition and linearity has been decided elsewhere. The fact that phonology and semantics do not concatenate anything is also a headstone of Chomsky et al.’s (2002) argumentation, who hold that ultimately only recursion (Merge) and the ability to talk to other modules in a specific way (Phase) make human language unique and different from animal communication. If phonology and semantics only interpret strings but are unable to build them, it follows that they must not have access to the building device, that is Merge in current syntactic theory. This is pointed out by Neeleman & van de Koot (2006), who provide detailed argumentation to the end that allowing for Merge (and hence for trees) in phonology and semantics wrongly predicts the existence of recursive structure, and hence of recursive phenomena in these modules. This is a strong argument against Jackendoff’s (1997 et seq.) parallel approach to modular structure where all modules are granted access to concatenation. Now the distinction between structure that is present in phonology but has not been created by phonological computation on the one hand and phonological computation itself on the other leaves an escape-hatch for the existence of recursive structure in phonology. Neeleman & van de Koot’s (2006) argument bites only for phonological structure that is also the output of phonological computation. Structure such as prosodic constituency may

Intermodular argumentation and the word-spell-out-mystery 53

be recursive because it has been inserted by extra-phonological computation, i.e. mapping. It is perfectly compatible with the non-recursive nature of phonology – to be precise: of phonological computation – and does not predict the existence of recursive phenomena all over the place in phonology.15

5.4.6. Is intonation a phonological phenomenon at all? Let us now return to the initial question which, recall, was about the absence of phonological effects of the cyclic spell-out of words. We have seen that cyclic spell-out is a factor that influences intonation for sure. Whether through the mediation of the prosodic hierarchy or otherwise, intonation seems to militate against the generalisation that cyclic spell-out of words leaves no phonological traces. This, however, is only true if intonation is a phonological phenomenon in the first place. Wagner’s (2005) work shows that intonation is much more syntactically bound than the phonological literature may suggest. In actual fact, it raises the question whether intonation is phonological at all: it may as well be syntax and nothing else. This direction is counter-intuitive because intonation is somehow ‘made of sound’, and hence should be treated in the phonology. This may be true – but we have seen that it does not mean that phonology builds the structure that ultimately gets to the surface in the coat of intonation. Phonology may well be responsible of nothing at all in the process that decides where intonation falls: like elsewhere, its action is purely interpretative; in the case of intonation, the action of phonology could well reduce to making a particular portion of the string phonologically prominent, whereby the portion itself has been designated by extra-phonological devices. 15

Finally, it is worth mentioning that the research programme of Government Phonology in general (Kaye et al. 1990), and of so-called CVCV in particular (Lowenstamm 1996, Scheer 2004), roots in the claim that syllable structure is better understood in terms of lateral relations among segments than by using traditional arboreal structure. This is where the title of Scheer (2004) comes from: a lateral theory of phonology. The lateral perspective, whose motivation is purely phonology-internal, thus allows for a non-contradictory implementation of the architectural requirement that phonology has no access to Merge, and that trees must not be the output of phonological computation: phonology-created structure is flat and lateral (see Scheer 2004: xlix ff.).

54 Tobias Scheer A strong argument in favour of this perspective is the following fact: it appears that intonation may be calculated in complete absence of phonological material, that is possibly before lexical (vocabulary) insertion and hence before phonological computation has a chance to do anything at all. Féry & Ishihara (in press) make this generalisation: the only thing that is needed in order to know where intonation falls is the syntactic structure of the sentence – the particular words and their phonological properties that will ultimately incarnate this structure are entirely irrelevant. Intonation, then, is the result of an exclusively top-down computation that depends on syntactic properties, information structure and eventually mapping decisions. If it is true that intonation may be calculated in complete absence of phonological material, the logical conclusion is that no phonological structure participates in the entire process: the prominence of this or that item is calculated in syntax alone; phonology only provides a specific pronunciation for the portion that has been designated.

5.5.

Conclusion

5.5.1. Either the generalization is wrong, or we need Praguian segregation plus an additional device At the end of this inquiry, we are left with a number of open questions, and some vague indications. While the cyclic spell-out of both morphemes and words is undisputed, why is it that the literature abounds in phonological effects of the former, but does not seem to have met any phonological trace of the latter? One solution is that phonological effects of the cyclic spell-out of words have simply been overlooked: the generalization is just wrong. Intonation, however, which appears to be a massive counter-example at first sight, turns out not to harm the word-spell-out-mystery – on the contrary, it rather adds grist to the mill of the mystery since it relies on recursive structure that phonology has no handle on. If on the other hand the generalization turns out to be correct, just jumping into Praguian segregation will not do: the existence of distinct chunkspecific phonologies, one for morpheme-, another for word-sequences, is necessary, but not enough to explain what is going on. What we need on top of that is a means to make post-lexical phonology ignore cyclic structure. For the zero hypothesis is certainly that if an interpretational system is exposed to piecemeal chunk submission, traces of this fact will appear in its output.

Intermodular argumentation and the word-spell-out-mystery 55

It is certainly empirically adequate to decree that lexical phonology is, but post-lexical phonology is not cyclic – this is the reaction of Lexical Phonology. This does not tell us, however, how come and what it means that interpretational systems can ‘decide’ to react on cyclic structure, or to ignore it.

5.5.2. The additional device may be the chunk-specific parameterisation of the PIC One way to go about this question has been explored in section 5.3.2: “taking into account cyclic structure” can mean two things for interpretational systems. One perspective is the solution of Lexical Phonology where the cyclic structure that (cyclic) lexical phonology encounters is interpreted by two distinct mini-phonologies according to the morpho-syntactic makeup of the string in question (level 1 vs. level 2, see section 3.1). By contrast, the cyclic structure that (non-cyclic) post-lexical phonology encounters is interpreted by just one single phonology. Another way of encoding the same contrast is through the parameterisation of no look-back (i.e. Phase Impenetrability). On the account of Kaye (1995), the phonological effect of the cyclic spell-out of morphemes is due to modification-inhibiting no look-back (see section 3.2.2). If thus the phonological effect of the cyclic spell-out of morphemes is due to Phase Impenetrability, and if this effect is absent when it comes to the cyclic spell-out of words, the conclusion is that the PIC constrains the computation below, but not at and above the word level (section 5.3.2). The next question, then, is what Phase Impenetrability is a property of. It appears that the PIC is unlikely to be managed by concatenative/interpretational systems; rather, it is a property of the spell-out mechanism, which in addition to organising chunk-submission, also decides to make ‘old’ strings subject to a specific treatment or not (section 5.3.4). In any event, about all that was said is more or less speculative. This is worth what it’s worth: the thing is that the whole issue – the word-spell-outmystery – appears to be entirely unreflected in the literature. One reason is certainly the fact that the absence of cyclic impact on external sandhi is set in stone since Lexical Phonology and the correlated idea of a completely regular (‘automatic’) post-lexical phonology. The study of a larger body of empirical material will have to show whether the current state of affairs is due to this theoretical bias (which, recall, was installed without argument), or whether natural language has really different ways of treating morphemes and words in phonological

56 Tobias Scheer computation. Which is but another occasion to be thrown back to the question whether morphology and syntax are one: back to where we started.

6. Conclusion The reader who has made it until the conclusion was courageous enough to go through a nested structure of topics, which are a priori unrelated but pieces of the same puzzle, as I hope to have shown: the track could be something like this: [[intermodular argumentation] [wso-mystery [chunk-specific PIC or wrong data] [intonation]]] where the distribution of topic and comment is classical – the subject introduces the problem, which is clear and appears to allow for a conclusive treatment at first. Just as the reader may think that the issue is settled and the article can be closed, however, doubt creeps in and the comment starts: do we really compare things that are comparable (please stress compárable)? Can we take for granted that the spell-out mechanism which sends word sequences to PF is really the same as the shipping mechanism which submits morpheme sequences to phonological computation? The intuitive answer is yes – but if it turns out that morphological and syntactic structure are distinct and not just the lower and upper part of the same tree, a negative answer may be construed: two different structures may (but do not have to) come each with its private spell-out mechanism, which means that there is no guarantee for the settings to be the same regarding the PIC (present/absent), selective spell-out (yes/no) and spell out your sister (yes/spell out something else). Which is what the clean intermodular arguments depend upon. This is how we discovered that – unfortunately – there is a hurdle to take for those who want to make intermodular arguments, which suppose that: (10)

The spell-out mechanism that handles the pieces on which morphosyntactic analyses/theories are built is the same as the one that is responsible for the shipping of pieces that ground phonological analyses /theories.

The king’s road to this question is to make sure that there is only one spellout mechanism in grammar. In this case no further discussion is needed, the intermodular arguments go through and we can sleep well. Unfortunately

Intermodular argumentation and the word-spell-out-mystery 57

though, as was mentioned, the only way to make sure that there is only one spell-out mechanism is to be sure that morphology is just the lower part of syntax. Alas nobody can take this for granted as morphologists and syntacticians continue to quarrel about this question, and probably still will in 20 or 50 years. So please, if you could decide one way or the other before that, you’d do a great favour to intermodular argumentation. For the time being, we must thus live with the eventuality that there are two independent spell-out mechanisms, one handling words, the other morphemes.16 This is where the comment-part of the topic-comment structure resumes, and eventually starts meandering seriously. For there seems to be a simple way to get around (10), which is to look at ‘syntactic’ pieces, rather than at their ‘morphological’ cousins. This strategy, however, fails miserably because there is nothing to look at: we hit the word-spell-out-mystery. This leaves us with another choice, empirical this time: either the mystery is real and language really makes a hard-wired difference between morphemes and words in that it takes into account the cyclic structure of the former, but not of the latter upon interpretation. Or there is no mystery at all, that is the mystery is a fact about a specific theory, rather than about language: if phonologists look hard enough, they will find phonological traces of the cyclic spell-out of words. This track may be promising because it is suspicious that the non-cyclic interpretation of words was set in stone by Kiparsky (1982) without the slightest mention of an argument, which has certainly contributed to the fact that nobody has looked at word-level data through the cyclic prism: all external sandhi was dumped into representational treatments. This is even true for those who – against Kiparsky’s stone-set rule – continue to subscribe to the original position of SPE that interpretation of pieces of whatever size is cyclic: as far as I can see, Halle & Vergnaud (1987) have not produced any relevant data or analysis. Only Kaye (1995) presents a candidate, the case of French vowel nasalization that was mentioned in note 13, but which could not be presented in detail. 16

Note that ‘word’ is just a cover term for a chunk size that is larger than a morpheme. What this actually comes down to is a different question that does not impact the demonstration: clitics may or may not count in, and a few other decisions may also have to be made. The only thing that is needed for the argument is the existence of a front line that separates ‘small’ and ‘big’ chunks. But anyway, defenders of the view that morphology is but the lower part of syntax actually use the impossibility to define such a red line as an argument. This all lies beyond the scope of the present chapter.

58 Tobias Scheer The chapter has made one contribution to this empirical debate, against its own goal, as it were: intermodular argumentation would be on the safe side if it could be shown that the word-spell-out-mystery is theory-, rather than language-created. A massive counter-example that especially syntacticians are familiar with is intonation. Intonation, however, is shown not to unsettle the mystery: it is certainly impacted by the cyclic spell-out of words – but it may not be a phonological phenomenon in the first place. Its recursive character does not allow it to enter phonology, which is definitely non-recursive, and the designation of the intonationally prominent item can be done on purely syntactic grounds, that is before lexical insertion and hence in complete absence of phonological material. Once intonation-placement is decided, then, all that phonology does is to highlight the designated item. We thus move forth in our decision tree: if there are two distinct spellout mechanisms for morpheme- and word-sequences, there may also be two distinct interpretational systems, one for each chunk size. Whether morpheme interpretation is cyclic, while word interpretation is not, depends on the empirical question regarding the word-spell-out-mystery: if the mystery is real, Praguian segregation is true and linguists will have to live with this fact. If it turns out to be an artefact of Lexical Phonology, phonologists may continue to claim that cyclic interpretation is a property of phonological interpretation as such, whatever the chunk size (phonologists, would you please find out about that). The deepest embedded level of our decision tree is when both previous questions have got a yes: yes there is a specific spell-out system for morphemes, yes the word-spell-out-mystery is real, which means that phonological interpretation may (morpheme sequences) or may not (word sequences) be cyclic. In this case, Praguian segregation needs to be somehow organized: it is hardly enough to transcribe the observation into prose, as Lexical Phonology does: we want to know (i) who decides that phonological interpretation will be cyclic or not and (ii) how cyclic interpretation is implemented. I have tried to answer both questions: interpretation is cyclic iff it the relevant computation is subjected to the PIC (section 5.3.2), and whether or not the PIC is switched on for a given computation is decided by the spell-out mechanism (rather than by the computation itself, which as a module is unable to make decisions at all, see section 5.3.4). This, in turn, raises inelegant questions about what kind of animal the spell-out mechanism actually is (a module? something else?), and how the PIC actually works (it is in need of a memory-keeper).

Intermodular argumentation and the word-spell-out-mystery 59

At the end of the venture, we are thus left with a situation where definitely more questions are raised than are answered, and where a clean situation that offers a nice perspective for intermodular argumentation has fallen prey to a whole series of conditions, embodied in a nightmare-like decision tree. So please, morphologists and syntacticians, find out that morphology and syntax is one; and please, phonologists, find out that there is cyclicity-induced external sandhi. You will make a good deal of the trouble vanish, and you will render a service to fellow (future) readers, which will be able to go home after a couple of pages, and to forget about the already spelled-out decision-meander.

References Ackema, Peter and Ad Neeleman 2005 Beyond Morphology. Interface Conditions on Word Formation. Oxford: Oxford University Press. 2007 Morphology ≠ Syntax. In The Oxford Handbook of Linguistic Interfaces, Gillian Ramchand and Charles Reiss (eds.), 325–352. Oxford: Oxford University Press. Adger, David 2007 Stress and Phasal Syntax. Linguistic Analysis 33: 238–266. Aelbrecht, Lobke 2008 Licensing ellipsis as Agree. Paper presented at the 27th West Coast Conference on Formal Linguistics, UCLA 16–18 May. Baltin, Mark 2007 Deletion versus Pro-forms: a false dichotomy? Ms., New York University. Beckman, Mary and Janet Pierrehumbert 1986 Intonational Structure in Japanese and English. Phonology Yearbook 3: 15–70. 1988 Japanese Tone Structure. Cambridge, MA: MIT Press. Berman, Arlene and Michael Szamosi 1972 Observations on sentential stress. Language 48: 304–325. Bermúdez-Otero, Ricardo forthc. Stratal Optimality Theory. Oxford: Oxford University Press. Boeckx, Cedric and Kleanthes Grohmann 2007 Putting Phases in Perspective. Syntax 10: 204–222. Booij, Geert 1996 Cliticization as prosodic integration: The case of Dutch. The Linguistic Review 13: 219 –242.

60 Tobias Scheer 1997

Non-derivational phonology meets Lexical Phonology. In Derivations and Constraints in Phonology, Iggy Roca (ed.), 261–288. Oxford: Oxford University Press. 2000 [1996]. The phonology-morphology interface. In The First Glot International State-of-the-Article Book, Lisa Cheng and Rint Sybesma (eds.), 287–305. Berlin /New York: Mouton de Gruyter. Booij, Geert, Luca Ducceschi, Bernard Fradin, Emiliano Guevara, Angela Ralli and Sergio Scalise (eds.) 2007 Proceedings of the 5th Mediterranean Morphology Meeting. Bologna: Università degli Studi di Bologna. Bresnan, Joan 1971 Sentence stress and syntactic transformations. Language 47: 257–281. Chomsky, Noam 1965 Aspects of the Theory of Syntax. Cambridge, MA: MIT Press. 1973 Conditions on Transformations. In A Festschrift for Morris Halle, Stephen R. Anderson and Paul Kiparsky (eds.), 232–286. New York: Holt, Rinehart & Winston. 2000 Minimalist inquiries: the framework. In Step by Step. Essays on Minimalist Syntax in Honor of Howard Lasnik, Roger Martin, David Michaels & 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. Oxford: Oxford University Press. 2005 Thee factors in language design. Linguistic Inquiry 36: 1–22. Chomsky, Noam and Morris Halle 1968 The Sound Pattern of English. Cambridge, MA: MIT Press. Chomsky, Noam, Morris Halle and Fred Lukoff 1956 On Accent and Juncture in English. In For Roman Jakobson. Essays on the Occasion of his 60th birthday, Morris Halle, Horace Lunt, Hugh McLean and Cornelis van Schooneveld (eds.), 65–80. The Hague: Mouton. Chomsky, Noam, Marc D. Hauser and W. Tecumseh Fitch 2002 The faculty of language: what is it, who has it, and how did it evolve? Science 298: 1569–1579. Cinque, Guglielmo 1993 A null theory of phrasal and compound stress. Linguistic Inquiry 24: 239–297. den Dikken, Marcel 2007 Phase Extension: Contours of a theory of the role of head movement in phrasal extraction. Theoretical Linguistics 33: 1– 41.

Intermodular argumentation and the word-spell-out-mystery 61 Embick, David 2007 Distributed Morphology and the Syntax – Morphology Interface. In The Oxford Handbook of Linguistic Interfaces, Gillian Ramchand and Charles Reiss (eds.), 289–324. Oxford: Oxford University Press. Epstein, Samuel D., Erich M. Groat, Ruriko Kawashima and Hisatsugu Kitahara 1998 A Derivational Approach to Syntactic Relations. Oxford: Oxford University Press. Epstein, Samuel and T. Daniel Seely 2002 Rule Applications as Cycles in Level-Free Syntax. In Derivation and Explanation in the Minimalist Program, Samuel D. Epstein and T. Daniel Seely (eds.), 65–89. Oxford: Blackwell. 2006 Derivation in Minimalism. Cambridge: Cambridge University Press. Féry, Caroline and Shinichiro Ishihara in press How Information Structure Shapes Prosody. In Information Structure from Different Perspectives, Malte Zimmermann and Caroline Féry (eds.). Oxford: Oxford University Press. Fodor, Jerry 1983 The Modularity of the Mind. Cambridge, MA: MIT-Bradford. Giegerich, Heinz 1999 Lexical Strata in English. Cambridge: Cambridge University Press. Grohmann, Kleanthes K. 2003 Prolific Domains: On the Anti-Locality of Movement Dependencies. Amsterdam: Benjamins. Gussenhoven, Carlos 1992 Intonational phrasing and the prosodic hierarchy. In Phonologica 1988, Wolfgang Dressler, Hans Luschützky, Oskar Pfeiffer and John Rennison (eds.), 89–99. Cambridge: Cambridge University Press. 2004 The phonology of tone and intonation. Cambridge: Cambridge University Press. Halle, Morris 1986 On the Phonology-Morphology Interface. Ms., MIT. Halle, Morris, James Harris and Jean-Roger Vergnaud 1991 A reexamination of the stress erasure convention and Spanish stress. Linguistic Inquiry 22: 141–159. Halle, Morris and Michael Kenstowicz 1991 The Free Element Condition and cyclic versus non-cyclic stress. Linguistic Inquiry 22: 457–501. Halle, Morris and Ora Matushansky 2006 The Morphophonology of Russian Adjectival Inflection. Linguistic Inquiry 37: 351– 404. Halle, Morris and Andrew Nevins forthc. Rule Application in Phonology. Architecture and Representations in Phonological Theory, Eric Raimy and Charles Cairns (eds.). Cambridge, MA: MIT Press.

62 Tobias Scheer Halle, Morris and Jean-Roger Vergnaud 1987 An Essay on Stress. Cambridge, MA: MIT Press. Hinterhölzl, Roland 2006 The Phase Condition and cyclic Spell-out: Evidence from VP-topicalization. In Phases of Interpretation, Mara Frascarelli (ed.), 237–259. Berlin /New York: Mouton de Gruyter. Inkelas, Sharon 1990 Prosodic Constituency in the Lexicon. New York: Garland. 1998 The theoretical status of morphologically conditioned phonology: a case study of dominance effects. Yearbook of Morphology 1997: 121– 155. Jackendoff, Ray 1997 The Architecture of the Language Faculty. Cambridge, MA: MIT Press. 2002 Foundations of Language. Brain, Meaning, Grammar, Evolution. Oxford: Oxford University Press. Julien, Marit 2007 On the Relation between Morphology and Syntax. In The Oxford Handbook of Linguistic Interfaces, Gillian Ramchand and Charles Reiss (eds.), 209–238. Oxford: Oxford University Press. Kahnemuyipour, Arsalan 2004 The syntax of sentential stress. Ph.D thesis, University of Toronto. 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. 1995 Derivations and Interfaces. In Frontiers of Phonology, Jacques Durand and Francis Katamba (eds.), 289–332. London /New York: Longman. Also 1993 in SOAS Working Papers in Linguistics and Phonetics 3: 90 –126. Kaye, Jonathan, Jean Lowenstamm and Jean-Roger Vergnaud 1990 Constituent structure and government in phonology. Phonology 7: 193–231. Kiparsky, Paul 1982 From cyclic phonology to lexical phonology. In The Structure of Phonological Representations I, Harry van der Hulst and Norval Smith (eds.), 131–175. Dordrecht: Foris. 2000 Opacity and cyclicity. The Linguistic Review 17: 351–365. Ladd, Robert 1986 Intonational phrasing: the case for recursive prosodic structure. Phonology 3: 311–340. 1996 Intonational Phonology. Cambrdige: Cambridge University Press.

Intermodular argumentation and the word-spell-out-mystery 63 2000

Bruce, Pierrehumbert, and the elements of Intonational Phonology. In The Interaction of Constraints on Prosodic Phrasing. Prosody: Theory and Experiments, Merle Horne (ed.), 37–50. Dordrecht: Kluwer. Liberman, Mark 1975 The intonational system of English. Ph.D dissertation, MIT. Lieber, Rochelle and Sergio Scalise 2007 The Lexical Integrity Hypothesis in a new Theoretical Universe. In Proceedings of the 5th Mediteranean Morphology Meeting, G. Booij, L. Ducceschi, B. Fradin, E. Guevara, A. Ralli and S. Scalise (eds.), 1– 24. Bologna: Università degli Studi di Bologna. Lowenstamm, Jean 1996 CV as the only syllable type. Current trends in Phonology. In Models and Methods, Jacques Durand and Bernard Laks (eds.), 419–441. Salford, Manchester: ESRI. Mascaró, Joan 1976 Catalan phonology and the phonological cycle. Ph.D. dissertation, MIT. Matushansky, Ora 2005 Going through a phase. MIT Working Papers in Linguistics 49: 157– 181. McMahon, April 2000 Lexical Phonology and the History of English. Cambridge: Cambridge University Press. Neeleman, Ad and Hans van de Koot 2006 On syntactic and phonological representations. Lingua 116: 1524– 1552. Nespor, Marina and Irene Vogel 1986 Prosodic Phonology. Dordrecht: Foris. Odden, David 1993 Interaction between modules in lexical phonology. In Studies in Lexical Phonology, Sharon Hargus and Ellen Kaisse (eds.), 111–144. New York: Academic Press. Pater, Joe 2000 Nonuniformity in English stress: the role of ranked and lexically specific constraints. Phonology 17: 237–274. Peperkamp, Sharon 1997 Prosodic Words. The Hague: Holland Academic Graphics. Pesetsky, David 1979 Russian morphology and lexical theory. Ms., MIT. Available at http://web.mit.edu/linguistics/people/faculty/pesetsky/russmorph.pdf. Pierrehumbert, Janet 1980 The phonology and phonetics of English intonation. Ph.D dissertation, MIT.

64 Tobias Scheer 2000

Tonal Elements and their Alignment. The interaction of constraints on prosodic phrasing. In Prosody: Theory and Experiments, Merle Horne (ed.), 11–36. Dordrecht: Kluwer. Pinker, Steven and Ray Jackendoff 2005 The faculty of language: what’s special about it? Cognition 95: 201– 236. 2005 The nature of the language faculty and its implications for the evolution of language (Reply to Fitch, Hauser and Chomsky). Cognition 97: 211–225. Pylyshyn, Zenon 1989 Computing in Cognitive Science. In Foundations of Cognitive Science, William Poser (ed.), 51–91. Cambridge, MA: MIT Press. Rubach, Jerzy 1997 Extrasyllabic consonants in Polish: Derivational Optimality Theory. In Derivations and Constraints in Phonology, Iggy Roca (ed.), 551– 581. Oxford: Clarendon. Rubach, Jerzy and Geert Booij 1984 Morphological and prosodic domains in Lexical Phonology. Phonology Yearbook 1: 1–27. Scheer, Tobias 2004 A Lateral Theory of Phonology. Vol. 1: What is CVCV, and why should it be? Berlin /New York: Mouton de Gruyter. forthc. A Lateral Theory of Phonology. Vol. 2. Interface: How MorphoSyntax Talks to Phonology. A Survey of Extra-Phonological Information in Phonology since Trubetzkoy’s Grenzsignale. Berlin /New York: Mouton de Gruyter. in press a Intermodular argumentation: morpheme-specific phonologies are out of business in a phase-based architecture. In The Sound Pattern of Syntax, Nomi Shir and Lisa Rochman (eds.). Oxford: Oxford University Press. in press b Spell out your Sister! To appear in the Proceedings of WCCFL-27. Selkirk, Elisabeth 1984 Phonology and Syntax: The Relation between Sound and Structure. Cambridge, MA: MIT Press. 1996 The prosodic structure of function words. In Signal to Syntax: BootStrapping from Syntax to Grammar in Early Acquisition, James Morgan and Katherine Demuth (eds.), 187–213. Mahwah, NJ: Erlbaum. Selkirk, Elisabeth and Angelika Kratzer 2007 Phase Theory and prosodic spellout: The case of verbs. The Linguistic Review 24: 93–135. Smith, Neil 2002 Modules, modals, maths and the mind. Glot International 6 (8): 248– 250.

Intermodular argumentation and the word-spell-out-mystery 65 Szendrői, Kriszta 2001 Focus and the Syntax-Phonology Interface. Ph.D dissertation, University College London. 2003 A stress-based approach to the syntax of Hungarian focus. The Linguistic Review 20: 37–78. 2004 A stress-based approach to climbing. Verb clusters. In A Study of Hungarian, German and Dutch, Katalin É. Kiss and Henk van Riemsdijk (eds.), 205–233. Amsterdam: Benjamins. Truckenbrodt, Hubert 1999 On the relation between syntactic phrases and phonological phrases. Linguistic Inquiry 30: 219–255. Uriagereka, Juan 1999 Multiple spell-out. In Working Minimalism, Samuel Epstein and Norbert Hornstein (eds.), 251–282. Cambridge, MA: MIT Press. Wagner, Michael 2005 Prosody and Recursion. PhD. dissertation, MIT. Ziková, Markéta 2008 Alternace e-nula v současné češtině. Autosegmentální analýza. Ph.D dissertation, Masarykova Univerzita v Brně. Ziková, Markéta and Tobias Scheer 2007 (Slavic) vowel-zero alternations and Phase Theory. Paper presented at OCP 4, Rhodes 19–21 January.

On bare prosodic structure and the spell-out of features Martin Haiden

In a recent article, (Pinker and Jackendoff 2005) reject the recursion only hypothesis proposed by (Hauser et al. 2002) with reference, among other things, to the claim that phonology is uniquely human, specific to language, and not recursive. If Pinker and Jackendoff’s view of phonology is correct, it casts doubt on an important background assumption of the minimalist research agenda. I will refer to this challenge as the rogue PF problem. Phonology is at the core of a second problem. A growing literature on stress and movement-triggers suggests that the syntactic derivation might ‘respond to’ certain features of the phonological component (Fox 2000; Reinhart 2006; Richards 2006). On standard assumptions about the architecture of the language faculty there should be no look-ahead from syntax into phonology. Finally, the derivation of interface representations is remarkably redundant, as compared with the general austerity of the minimalist program. The semantic interface must recognize and delete phonological features; the phonological interface must recognize and delete semantic features. In the best case, such redundancy should be avoided without weakening the inclusiveness condition. In this paper I argue that the three problems stem from common, yet unfounded assumptions about phonology. The first part of the paper concentrates on phonology from the perspective of the look-ahead problem and its manifestation in the domain of non concatenative morphology. It concludes that phonology does not apply after syntax, but in parallel with it. The second part of the paper addresses the redundancy problem and proposes a formalism that derives interface representations as a necessary by-product of hierarchical structure. In this theory, the recursive core of grammar includes both prosody and syntax, and the two components operate cyclically in parallel. The phonological and syntactic features of a given lexical entry are merged at the same stage of the derivation in their respective components. Linearization patterns in morphology and syntax are derived by means of templates. Another, post-syntactic, operation of SPELL-OUT does

68 Martin Haiden not exist.1 The theory is exemplified in an analysis of templatic positions in Classical Arabic and Standard German verbal morphology. The paper concludes with an outlook on how generalized templates can replace parameters, and on their place in acquisition.

1.

Looking ahead to PF

1.1. What is PF? In the standard minimalist architecture, the computational system derives representations that are interpretable at the phonological and semantic interfaces, respectively. Putting semantics aside, we might want to ask whether the phonological interface component is a shallow system of phonetic output generation, or rather a full-fledged component of the computational system. The former option is most likely, if phonology is “natural” in the sense that its descriptive framework is “limited to a set associated directly with measurable acoustic, articulatory, and/or auditory properties”, its rules are “limited to formal statements of surface regularities that have a demonstrable phonetic basis”, and its representations are “limited to ones which can be directly induced from the corpus of forms and alternations available to the child, and which are thus “learnable” in some general sense” (Anderson 1981: 496). Anderson argues forcefully that phonology is not natural in any one of these respects. However, this does not yet mean that phonological facts should be modeled in a global grammatical theory that includes syntax. Indeed it is argued in (Bromberger and Halle 1989) that the nature of phonological facts calls for an independent module with mechanisms that are qualitatively different from syntax. They claim that [p]honology […] is primarily concerned with the connections between surface forms that can serve as input to our articulatory machinery (and to our auditory system) and the abstract underlying forms in which words are stored in memory. Whereas syntax is concerned with the relations among representations that encode different types of information requiring different types of notation, phonology is concerned with the relationship between representations that encode the same type of information – phonetic information – but do so in ways that serve distinct functions: articulation and audition, on the one hand, and memory, on the other. (Bromberger and Halle 1989: 53) 1

It should be noted that immediate spell-out entails that there cannot be overt movement of syntactic categories; cf. (Bródy 1995) and subsequent work for a minimalist theory of syntactic chains that is compatible with this scenario.

On bare prosodic structure and the spell-out of features 69

Bromberger and Halle correctly observe that phonology manipulates a specific “type of information – phonetic information”, which is absent from syntactic representations. On the other hand, recent work on the PF interface has made it increasingly obvious that the two components, phonology and syntax, interact in a much tighter way than it had been envisaged by Bromberger and Halle. For example, case assignment has long been known to be sensitive to phonological domains (Neeleman 1994; Siloni 2001; Stowell 1981), and it is not unlikely that phonological properties have an influence on syntactic structure in areas like quantification, focus, and wh-movement (Fox 2000; Reinhart 2006). In the face of such facts, the traditional model of grammar, which assumes a linear ordering of phonology after syntax, is confronted with a serious ‘look-ahead’ problem: how can the syntactic derivation ‘know about’ domain-boundaries and requirements at the phonological level? (Richards, 2006) suggests putting this problem aside “until we have a better sense of what type of phonological information the syntax is allowed to ‘respond’ to” (p. 56). His hunch is that syntax might ‘respond to’ information at the prosodic level, but not to information that involves the segmental identification of prosodic constituents (p. 56). I will argue in the following sections that this distinction is basically correct, and derive it from a general theory of concatenation and interpretation in a parallel, rather than a serial arrangement of phonology and syntax.

2. Cyclicity and inclusiveness: the CV model One way of addressing the ‘look-ahead’ problem is to reject the assumption that the phonological component applies after syntax. (Chomsky 2004) indeed proposes that the three components of a derivation (syntax, semantics and phonology) operate in parallel. Let us adopt this position as a working hypothesis. Assume that all three components are cyclic […]. In the worst case, the three cycles are independent […]; the best case is that there is a single cycle only […] and the three components of the derivation of proceed cyclically in parallel. L contains operations that transfer each unit to Φ and

to Σ. In the best case, these apply at the same stage of the cycle. (Chomsky 2004: 107)

A major obstacle for this scenario is the assumption, shared by (Chomsky 2004), that phonology does not satisfy a number of minimalist principles, in

70 Martin Haiden particular the inclusiveness condition. This is indeed true for most theories of phonology that posit derivationally created objects like syllables, syllabic constituents, and super-segmental objects like feet. However, the assumption of such objects is a theoretical choice, not an empirical necessity. Indeed, much work conducted during the 1980s and 90s in the framework of Government Phonology was dedicated at deriving particular prosodic construction types like CV templates from universal principles of syllabic structure (Guerssel and Lowenstamm 1990; Kaye and Lowenstamm 1984; Lowenstamm and Kaye 1986). Pushing this reasoning further, (Lowenstamm 1996) proposes that “for all languages closed syllables, geminate consonants, and long vowels must be reanalyzed in terms of sequences of light open syllables” (p.419). On this assumption, a branching onset is represented as a sequence of two light open syllables, where the first V-position is phonetically unrealized (1a).2 A closed syllable is represented as a sequence of two CV positions with an empty final V position (1b). A branching nucleus (1c) is represented as a sequence of two CV units with an empty C position between the two vocalic positions. A geminate consonant is linked to the C positions of two subsequent CV units (1d). (1)

a. branching onset

b. closed syllable

C V C V g g g f l a

C V C V g g g f a l

fla

c. branching nucleus C V C V a

fal

d. geminate C V C V

a:

f

ff

Since neither onsets, nor nuclei branch, there is a one-to-one correspondence between syllabic constituents and skeletal positions. In this sense, the skeletal level is fully reduced to the constituent level. The traditional representation in (2a) can be replaced by the one in (2b).

2

Notice that the correctness of the structure in (1a) depends on the status of a given branching onset with respect to the calculation of phonological length. If it introduces a mora of its own, as it does in some languages, then (1a) is the correct analysis. If it does not, then the branching onset must be analyzed as a contour segment, cf. Lowenstamm (2003b) and references cited there.

On bare prosodic structure and the spell-out of features 71

(2)

a. constituent level skeletal level segmenal level

O g x g b

N g x g a

b. C g b

V g a

In (2b), the skeletal level has been eliminated in favor of the terminal elements of syllable structure, C and V positions in strict alternation. The CV model of phonological representation (Lowenstamm 1996; Scheer 2004) no longer requires derivationally created units of phonological time. Let us now see how the CV model can derive complex prosodic configurations as bare structures. In syntax, merge applies to two objects α, β, it derives a third object γ, and it labels γ as either α or β. If α is the head, then γ = {α, {α, β}}. The same procedure can be made to apply to the terminal objects of syllabic structure. Take as an example the closed syllable [fal] in (1b). The derivation starts out with an initial array of four terminal elements, V1, V2, C1, C2. In a first step, merge forms two light open syllables by joining C1, V1, and C2, V2, respectively. Next, it joins the two light open syllables, and determines V1 as the syllable nucleus. The resulting bare structure is illustrated in (3). (3)

[fal]: {V1, {V1, {C1, V1}}, {V2, {C2, V2}}}

The syllabic constituents of (3) are nothing but specific configurations of consonants and vowels. The inclusiveness condition is thus satisfied. If higher prosodic structure is formalized in terms of occurrences of basic timing units, as it is done in (Vergnaud 2003), then we can conclude that inclusiveness does hold at PF.

1.3. Look-ahead in morphology 1: Classical Arabic medial gemination PF and syntax are most intimately related in the area of root and template morphology. As argued in (McCarthy 1979), morpho-syntactic categories can be spelled out by phonological configurations (i.e., templates), rather than discrete lexical entries, in such systems. According to McCarthy’s theory, the rules of autosegmental phonology are sufficient to derive the correct output forms, once a root has been concatenated with a template. McCarthy’s theory indeed works for most cases. It does not work smoothly, however, in the case of medial gemination (Lowenstamm and Kaye 1986). Consider as an example the formation of

72 Martin Haiden the causative/intensive (form II) and of the reciprocal (form III) in Classical Arabic, as discussed in Guerssel and Lowenstamm (1990): the root ktb ‘write’ appears as kattab ‘make write’ in form II, and as kātab ‘correspond’ in form III.3 The principles of autosegmental phonology stipulate that melodic elements are associated with skeletal positions one by one, from left to right. Therefore, there should only ever be final gemination, as in the ungrammatical form (4). The fact that the medial root consonant in kattab ‘make write’ is associated with two skeletal positions must be determined by extra-phonological, viz. morphological conditions. (4) *katbab root template vocalic melody

k t b g y gy C V C V C V C a

(Lowenstamm 2003a: 22e)

Based on a comparable argument for Tiberian Hebrew, (Kaye and Lowenstamm 1984) propose that templates may contain empty positions that are segmentally identified after root segments have been linked to root positions. (Guerssel and Lowenstamm 1990) apply this reasoning to Classical Arabic. In (5), the CV-position marked as DS (derivational syllable) is skipped by the association of root consonants. In a next step, a derivational head merges with V in morpho-syntax. At this point, the CV position marked as DS becomes visible for segmental association, and it is immediately identified. In form II, identification is from the right, yielding gemination of the medial consonant (5a). In form III, identification is from the left, and the result is vowel lengthening (5b).

3

The discussion will be based on perfective active forms, abstracting away from agreement affixes.

On bare prosodic structure and the spell-out of features 73

(5)

a. form II: kattab ‘make write’

b. form III: kātab ‘correspond’

k t b

k t b

C V [DS C V ] C V C V

C V [DS C V ] C V C V

a (Guerssel & Lowenstamm 1990: 5a,b)

a (Lowenstamm 2003a: 31a,b)

Data like these are strong evidence for a parallel model, where the phonological component does not have to wait until a syntactic derivation is finished: phonological operations apply at the same stage of the cycle as syntactic operations. The templates of Classical Arabic are empirical support for the best case envisaged by Chomsky.

1.4. Look-ahead in morphology 2: English and Hebrew derived nominals Taking up a remark in Grimshaw (1990), Borer (2003) argues that the syntactic properties of derived nominals in English and Hebrew can only be predicted, if the overtness of derivational affixes is taken into account. Their syntactic properties cannot be predicted from lexical-semantic and syntactic features alone. Consider the contrast between a complex event nominal as in (a), and a result nominal as in (b): (6)

a. The instructor’s (intentional) examination of the student. b. The instructor’s examination/exam.

One of the most important differences between complex event nominals and result nominals is that the former, but not the latter, project argument structure. Borer quotes Grimshaw (1990: 16) for the observation that “[o]nly nouns that are derived from verbs […] by means of overt affixation can be [complex event] nominals, while nouns which do not have a verbal or adjectival source never are”. Borer capitalizes on this observation, to develop an argument for the relevance of phonological shapes for the availability of certain syntactic derivations. Take a lexical item with the semantic properties of |transform|. On standard assumptions, this lexical item can be inserted in a verbal syntactic frame, to yield the verb /transform/.

74 Martin Haiden (7)

|transform| + V



[V transform]

Then, still without accessing any phonological information, the verb /transform/ can be inserted into a nominal frame, to yield a derived noun: (8)

[V transform] + N



[N [V transform]]

Finally, [N [V transform]] is spelled out phonologically. Here is where we encounter the problem. The semantic properties of [N transform] can be spelled out in various ways, among them the following: (9)

a. b. c. d.

[N [V transform]] [N [V transform]] [N [V transform]] [N [V transform]]

   

transformation metamorphosis shift turn

As all of these phonological spell-outs have the same core meaning and, by hypothesis, the same derivational history, they should have exactly the same syntactic properties. In particular, they should all be argument-taking nouns. But this is plainly false. Only transformation, as a morphologically derived complex event nominal, projects the argument structure of the verb ‘transform’; metamorphosis, shift, turn do not. Therefore, the spell-out operations (b–d) must be blocked. Metamorphosis, shift, turn must spell out a different derivation, one that inserts the concept into a nominal frame directly, as follows: (10) a. |transform| + N b. [N transform]

 

[N transform] metamorphosis/shift/turn

Thus, the phonological shape associated with a given concept constrains the derivational options available for that concept. It might be objected at this point that transformation differs semantically from metamorphosis, shift, turn, and that this difference in meaning, not the phonological shape of the respective lexical entries, is responsible for their different derivational options. Borer rejects this challenge, quoting an example from Hebrew. The native Hebrew noun šinui ‘change/transformation’ contrasts with the synonymous borrowed word transformacia ‘transformation’. Native šinui can be used both as an argument-taking event nominal (a), and as a referential noun (b) – Borer’s example (36):

On bare prosodic structure and the spell-out of features 75

(11) a. ha-šinui the-transformation/change b. ha-šinuy the-change/transformation

šel merkaz ha-’ir ‘al yedey ha-’iriya of center the-city by city hall haya madhim was amazing

By contrast, the borrowed noun transformacia can only be used as a referential noun (b). It does not license arguments (a) – Borer’s example (37): (12) a. *ha-transformacia šel merkaz ha-’ir ‘al yedey ha-’iriya the-transformation of center the-city by city hall b. ha-transformacia hayta madhima the-transformation was amazing The only difference between the two nouns is that šinui is overtly derived from the verb šina, while transformacia lacks a verbal counterpart. Borer argues that the difference in syntactic behavior can only be accounted for, if the syntax can look ahead and identify the morpho-phonological properties of the forms it creates. In a serial model of derivation, this look-ahead could take the following shape. Take π to be a “reference to an indexed phonological representation” (Borer 2003: 19), which is associated with every lexical item by lexical stipulation. Then, derivations like (13) can be ruled out on the basis that π refers to the phonological shape transformation, not to metamorphosis, shift, turn. (13) ([N transformation], π)



*metamorphosis/shift/turn

On the assumption of a linear arrangement of syntax before phonology, we are forced to assume representations like (13), allowing syntax to recognize phonological information. The empirical problem does not arise in a model of parallel derivation. If spell-out applies to the root immediately, i.e., the moment it merges, then later syntactic operations cannot replace it with another lexical item.

1.5. Look-ahead in Morphology 3: German strong verbs and causativization There is a set of German verbs that are often considered to be irregular, because they do not mark the past tense by an affix. Instead, they employ an apophonic stem alternation (14a). For this reason Grimm (1819) called these verbs strong. Although the nature of the vocalic alternations is subject to

76 Martin Haiden debate, they are in fact fairly regular (Grimm 1819; Halle 1953; Ségéral and Scheer 1998). The overwhelming majority of German verbs is weak. This class marks the past tense by a dental suffix (14b). (14)

inf.

past 3sg

past participle

a. dring-en fahr-en fließen saug-en

drang fuhr floss sog

ge-drung-en ge-fahr-en ge-floss-en ge-sog-en

‘penetrate’ ‘go/drive’ ‘flow’ ‘suck’

b. schauspieler-n

schauspieler-te

ge-schauspieler-t

‘play-act’

Strong verbs are mainly interesting because of the position affected by the apophonic changes. In contrast to weak verbs, which appear in a number of different prosodic configurations, strong verbs are templatic. Their stem vowel, which is affected by the apophonic alternation, is always linked to the pre-penultimate CV position (Bendjaballah and Haiden 2002). The stem vowel of weak/affixal verbs is not constrained in this way. In (15b), stress falls on the diphthong [aw]. (15) a. strong – stable template: C V C V C V

C V C V C V

z a w

z o

g

saug-(en) ‘suck.inf.’

g

sog ‘suck.past’

b. weak – no constraint: C V C V C V C V (CV) C V C V ʃ a w

ʃ

p i

l

r

schauspieler-(n) ‘play-act’ The CV position that hosts the tense alternation in a strong verb can host other alternations as well. Consider as an example umlaut-causatives (Bendjaballah and Haiden 2002; Haiden 2005b). The base of causativization can be adjectival (16a), nominal (16b), or verbal (16c, 17). The interpretation of the derived form is always causative, sometimes with an additional intensive flavor.

On bare prosodic structure and the spell-out of features 77

(16) a. b. c.

base

umlaut-causative

stark hammer fallen

stärken hämmern fällen

(17) base bitten dringen fahren fallen flieβen liegen saufen saugen schinden schwimmen schwinden schwingen sinken sitzen springen trinken zwingen

‘strong – strengthen’ ‘(a) hammer – (to) hammer’ ‘(to) fall – (to) fell’

umlaut-causative beten drängen führen fällen (ein)flöβen legen (er)säufen säugen schänden schwemmen (ver)schwenden schwenken senken setzen sprengen tränken zwängen

ask – pray penetrate – push drive/go – lead fall – fell flow – pour/force into lie – put down drink – drown (causative) suck – breast-feed maltreat – violate swim – rinse fade – waste swing – wave/toss fall – lower sit – put jump – detonate drink – make drink force – squeeze

Bendjaballah and Haiden observe that the base of a deverbal umlaut-causative must be strong (18a), and that the derived verb is always weak (18b). The past tense of an umlaut-causative is always formed by means of a suffix.4 4

There is one apparent counterexample to this generalization: wiegen ‘weigh’ (as in weigh 5 kilogramms)’ vs. wägen ‘weigh’ (as in weigh 5 apples)’. The causative member of this pair remains strong. However, this pair is exceptional in more than one respect. First, the causative reading of wägen can already be observed for the base wiegen, as in (i). i. Er wiegt die Äpfel. ii. Die Äpfel wiegen 1 Kilo. he weighs the apples the apples weigh 1 kg Second, the I/E alternation appears to depend on number agreement. A google search suggests that causative wiegen is common if its subject is plural, but rare if the subject is singular. The stem wägen is rare altogether, unless it appears with the particle ab in ab-wägen ‘to weigh’. I conclude that the pair wiegenwägen does not encode a regular causative alternation.

78 Martin Haiden (18) a. b.

inf

pres. 2sg.

past 2sg

sink-en senk-en

sink-st senk-st

sank-st senk-te-st

‘fall’ ‘lower’

This is strong evidence to the effect that strong verbs are templatic, not irregular. On the assumption that strong verbs come with a single infixal position, we predict (i) that only strong verbs have stem derived causatives, and (ii) that stem derived causatives are weak: if the infixal position is identified as a causativizer, it is no longer available to host a tense marker (Bendjaballah and Haiden 2003). The analysis is furthermore appealing, because it extends to pairs like the following, where a vocalic alternation in the infinitive is absent, but the generalization regarding inflectional class membership is still intact: the base is strong, the derived causative is weak. (19) base erschrecken (er/ver)löschen (20) a. b.

causative erschrecken löschen

inf erschrecken (er/ver)löschen erschrecken löschen

‘get frightened – frighten’ ‘go/die out – put out’

past erschrak (er/ver)losch erschreck-te lösch-te

‘get frightened’ ‘go/die out’ ‘frighten’ ‘put out’

(21) summarizes the behavior of umlaut causatives: (21)

umlaut causatives a. A verbal base can be umlaut-causativized, if it contains an available infixal position. b. Causativization renders the infixal position inaccessible; umlaut causatives are weak.

(21) is straightforward as a morpho-phonological generalization – in a framework that recognizes morphology as an autonomous component of the language faculty. The generalization cannot be maintained as it stands, if wordinternal structure is generated in syntax, while the morpho-phonological features of lexical items are only accessible much later in the derivation: the principles “syntactic hierarchical structure all the way down” and “late

On bare prosodic structure and the spell-out of features 79

lexical access” of Halle and Marantz (1993, 1994). Can (21) be made compatible with these two principles? Notice first that umlaut causativization affects the syntactic argument structure of the verb. For example, fallen ‘fall’ is an unaccusative verb, while fallen ‘fell’ is transitive. This fact is accounted for, if umlaut marks the merger in syntax of a causative head Caus. Causativization by umlaut cannot be explained away as a late morpho-phonological readjustment. Rather, we need to inquire how (21) can affect the merging behavior of the syntactic head Caus. In order to state the first part of the generalization (21a), we might want to invoke Borer’s suggestion again. Assume π, a reference to indexed phonological representations, identifies the morpho-phonological property ‘lacks an available infixal position’. We can then state the syntactic condition that Caus cannot merge with {V, π}: (22)

*{Caus, {V, π}}

This move constitutes a case of partial look-ahead. The syntax has access to a morpho-phonological property of the verbal base.5 It is not enough, though, to access the features of the base. The phonological shape of Caus must be made explicit as well. Consider unmarked causative alternations (Haiden 2005b), as exemplified in (23). The transitive syntax of (23a) should be accounted for in parallel with that of fellen ‘fell’, by the merger of a head Caus. (23) a. Er hat den Ball ins Tor gerollt. he has the ball in-the goal rolled ‘He rolled the ball into the goal.’ b. Der Ball ist ins Tor gerollt. the ball is in-the goal rolled ‘The ball rolled into the goal.’ As far as syntax is concerned, it should be irrelevant that Caus is silent in rollen ‘roll’, but marked in fellen ‘fell’. However, silent Caus is not constrained in the same way that its umlaut-marked counterpart is. In particular, silent Caus is fully compatible with infixal Tense: many strong verbs are causative transitives (24). 5

Notice that even Embick (2000) is forced to a comparable move, conceding that “the features of certain Roots are visible in the syntactic derivation” (p. 226).

80 Martin Haiden (24) inf ess-en geb-en heb-en sauf-en saugen schinden tret-en

past, 3sg aß gab hob soff sog schund trat

‘eat’ ‘give’ ‘lift’ ‘drink’ ‘suck’ ‘maltreat’ ‘kick’

Furthermore, some strong causatives can undergo further umlaut causativization (25). Since the syntactic distribution of silent Caus and umlaut-Caus is not complementary, they should probably be treated as different syntactic heads. The only feature, though, that distinguishes the two Caus heads, is their phonological shape. (25) saufen saugen schinden

(er)säufen säugen schänden

‘drink – drown (caus)’ ‘suck – breast-feed’ ‘maltreat – violate’

If we are to maintain late lexical access, then we need to generalize phonological indices in syntax allowing, in principle, the assignment of a phonological index to every syntactic head. (21) must be stated as something like (22’), where ρ means ‘is spelled out as a vocalic alternation’. (22’) *{{Caus, ρ}, {V, π}} (22’) artificially recreates an environment of early lexical access, as it is proposed in Chomsky’s version of the minimalist program. It appears to be empirically necessary that the derivation can access lexical items as a whole, including their phonological features. But this conclusion inevitably leads us back to the redundancy problem: is it really necessary to refer to phonological features in syntax? The second part of the paper proposes a parallel model in which this problem does not arise.

On bare prosodic structure and the spell-out of features 81

2. Concatenation and interpretation 6 Two interpretational operations are generally assumed to exist: one of them, call it Σ, maps objects from syntax into semantics. Another operation, call it Φ, maps syntactic objects into phonology. Chomsky observes that neither of these operations is trivial. Apart from transferring the output of the computational system to the respective interface, each operation must eliminate those features that belong to the other interface, respectively. Semantic features must be eliminated by Φ, phonological features by Σ. Although the elimination of un-interpretable features can be motivated as an interface condition, general conditions of computational efficiency favor a theory that delivers pairs (PHON, SEM) directly, without deletion of mutually un-interpretable features. How can this be accomplished?

2.1. The derivation of SEM: bare phrase structure Chomsky (1995) assumes that the concatenating operation merge takes two objects x, y, it forms a pair (or set), and it assigns a label (either x or y) to the newly formed object. In effect then, merge is a complex operation. It consists of at least two components: pair/set formation and labeling. The two components are defined independently in Bendjaballah and Haiden 2003) and Haiden (2005b). Bare merge takes two variables ranging over a discrete domain E, x,y ∈ E2 , and delivers the product xy ∈ E: (26) Merge

µ:

(Haiden 2005a: 162) 2

E !E (x, y )! xy

By definition, the output of µ is in E, and it can merge with another variable in E: µ(µ(x, y), z)) = µ(xy, z) = xyz. As a product, µ creates flat structures. The order of subsequent merging operations does not determine any hierarchical relation among variables. Asymmetrical relations are introduced by a filtering operation that designates the head by eliminating its dependent. This operation can be modeled as a sum. 6

Sections 2.1–2.3 have appeared as Haiden (2005a). They are slightly adapted here.

82 Martin Haiden (27) Head-of

(Haiden 2005a: 162)

#µ (x, y) = Cx y" E

In (27), summation applies to y, and y is thus eliminated. The variable is replaced by a constant C, which corresponds to the sum of the values of y. This operation models the designation of the head x of xy by assigning a value to the complement y. In standard minimalism, the interpretation of features requires an independent operation Σ. In the present formalism, Σ delivers interpretations as a by-product of the generation of asymmetrical relations: A given variable is designated as a complement by being interpreted. Only unvalued features (variables) remain derivationally active. The last variable to be interpreted corresponds to the root node. Its interpretation closes the derivation. Let us consider an initial array of lexical items, represented as variables x,y,z. Assume x merges with y to form xy, and then head-of interprets y as a dependent of x. Next, z merges with the head x of xy, to form zx. (28) z

x

by merge

x

by head-of

x

y

by merge

Now there are two options to proceed. Head-of can interpret x as a dependent of z. This yields a complementation structure (29). By contrast, if z is interpreted as a dependent of x, the result is a shell- or specifier-structure (30). (29)

complement-structure z z

by head-of x

by merge

x

by head-of

x

y

by merge

On bare prosodic structure and the spell-out of features 83

(30)

specifier/shell structure

z

x

by head-of

x

by merge

x

by head-of

x

y

by merge

Both configurations are bare structures. They do not incorporate derivationally introduced objects like bar levels, or light heads. This is particularly evident in the shell configuration. Multiple specifiers/shells can be merged as long as the interpreted configuration is valid at the semantic interface. Merge may apply successively without the intervention of head-of. If a variable is not interpreted immediately, it “incorporates” and “moves” as part of a complex head. Assume an initial array of variables x,y,z,g. For concreteness, let z be a verb, y a noun, and x a feature F of N that needs to be checked against g, a functional head G. First NF, and then VNF, are derived by merge. Then head-of interprets N as a complement, while leaving F unaffected. This designates VF (not V alone) as the head of VNF. Informally speaking, we would say that an index of feature F has percolated from N to V. (31)

VF V

NF

Next, VF merges with G, forming GVF. Head-of interprets V as a complement of GF. Informally, the index of feature F has percolated, or moved to the next higher head G. (32)

GF G

VF V

NF

Finally, head-of interprets F as a dependent of G. This is similar to the deletion of a checked feature. Elimination of checked features is thus a further consequence of the generation of hierarchical relations. It does not need to

84 Martin Haiden be stipulated as a separate operation, i.e., an imperfection in Chomsky’s terminology. (33)

G GF G

VF V

NF

2.2. The derivation of PHON: bare prosodic structure Three observations come to bear on the formalization of the phonological interface. First, phonology concatenates objects to form asymmetrical prosodic structures (Halle and Vergnaud 1987; Nespor and Vogel 1986; Vergnaud 2003). This is true even for Pinker and Jackendoff (2005), who claim that phonology is not recursive in any meaningful sense. Second, prosodic structures are not identical with syntactic structures, although they tend to be similar in certain respects (cf. Cinque 1993; Kiparsky 1966; Reinhart 2006; Zubizarreta 1998). Third, the concatenation of lexical items X, Y in phonology must be identifiable as a concatenation of X and Y in syntax, even though their hierarchical relation might well be the opposite in the two domains. Observations one and two indicate that phonology has access to the structure-building algorithm of the computational system. In the preceding section, it was argued that the mapping to SEM is a natural consequence of the structure building algorithm itself. The same is now attempted for the mapping to PHON: merge concatenates timing units, head-of introduces asymmetrical prosodic structure by assigning a segmental value to dependent units of time. Let us assume with Lowenstamm (1996) that the primitive unit of prosodic structure is a single consonant position C followed by a single vocalic position V. Now consider a complex syllabic constituent as exemplified by the English word aim. This constituent must be represented by three CV units, CV1, CV2, CV3. These CV units are manipulated by the structure building algorithm exactly like syntactic objects. In the particular example, merge applies twice to create the sequence CV1CV2CV3. Next, head-of interprets CV3 as the complement (i.e., coda) of the complex nucleus CV1CV2

On bare prosodic structure and the spell-out of features 85

and spells it out as a segment [m]. Then, head-of applies again. It interprets CV2 as the complement (i.e., off-glide) of the nucleus CV1 , and it spells it out as [j]. CV1 is thus the head, which can merge with other CV units to form complex prosodic structures. If it does not undergo any further merger, it is itself spelled out as [e] by head-of. (34) illustrates the derivation. (34) Bare prosodic structure: aim [ejm] C V C V C V C V C V C V e j

m

Notice that this is a bare prosodic structure. It does not contain any derivationally created object like coda, onset, or syllable. It is thus possible to generalize the inclusiveness condition to the derivation of PHON.

2.3. Mapping and structure preservation The postulation of a phonological component with access to the recursive engine of the language faculty might appear to constitute what Chomsky (2004) calls the worst case: the existence of independent cycles. The “best case”, according to Chomsky, “is that there is a single cycle only”, and that the “components of the derivation of proceed cyclically in parallel” (p. 107). The assumption of a single cycle spanning over multiple components requires a transparent mapping. Let us assume the transparency of the mapping is guaranteed by a generalized version of the Projection Principle, adapted from Haiden (2005a: 158): (35) Projection Principle Lexically stipulated pairings (PHON, SEM) remain unaltered throughout a derivation. The Projection Principle can only apply to complex objects, if the concatenation of two lexical items X, Y in phonology is recognizable as a concatenation of X, Y in syntax. This is the case, if the mapping from phonology into syntax is structure-preserving for merge. Let us assume this is the case.

86 Martin Haiden (36) Mapping PF → syntax a. f: A → B where A ∩ B = Ø b. f: A → B is structure preserving for µ Notice before we proceed that structure preservation for merge does not entail the visibility of hierarchical structure across components. By contrast, it forces transparency of concatenation across components. Special rules that blur this transparency (like fission and fusion in Distributed Morphology) must not exist. We may ask whether the mapping is surjective: Are all members of B images under f ? In other words, do all syntactic heads have a phonetic shape? The answer to this question is probably negative. There are silent categories in syntax. The PF-syntax map is not surjective. Is the mapping injective – do all objects in A have an image in B? The way interpretation is formalized, we expect that they do. Every phonologically overt lexical entry has syntactic features. Higher occurrences of prosodic constituents must not manipulate the lexical integrity of their source. They are mapped on the identity element for merge, the integer 1.7

2.4. Recursion in bare prosodic structure: the rogue PF problem We now have the tools to address the rogue PF problem mentioned in the introduction. Although phonology has certain specifically human properties, Pinker and Jackendoff (2005) claim it is not recursive. They lay out their point as follows: “We note that the segmental and syllabic aspect of phonological structure, though discretely infinite and hierarchically structured, is not technically recursive. Recursion consists of embedding a constituent in a constituent of the same type, […], which automatically confers the ability to do so ad libitum […]. This does not exist in phonological structure: a syllable, for instance, cannot be embedded in another syllable. Full syllables can only be concatenated, an operation that does not require a pointer stack or equivalent apparatus necessary to implement true recursion.” (p. 210 f.)

7

A higher occurrence of a prosodic constituent may have a syntactic interpretation other than 1 by lexical stipulation; cf. the discussion of German verbs below.

On bare prosodic structure and the spell-out of features 87 “[…] overall, major characteristics of phonology are […] discretely infinite, and not recursive. Thus phonology represents a major counterexample to the recursion-only hypothesis.” (p. 212)

This position is empirically questionable for two reasons. First, the comparative literature on the issue uses phonological cues in order to test whether humans and non-human primates are capable of processing recursive structures. Fitch and Hauser (2004) argue that human subjects can identify recursive rules in sequences of meaningless CV syllables. If phonological stimuli are assigned to a system whose rules are qualitatively different from those found in syntax, as Pinker and Jackendoff seem to be claiming, then it is a mystery why Fitch and Hauser’s subjects behave as they do. Second, Pinker and Jackendoff insist that phonology cannot be recursive, because “a syllable cannot be embedded in another syllable”. This claim is odd, as it is contingent entirely on conventions of notation, rather than empirical fact. It is certainly true that the prosodic hierarchy (Nespor and Vogel 1986; Selkirk 1984) distinguishes objects at different levels of structure, and that these levels are strictly ordered. For example, feet embed syllables, not feet. However, this does not mean that the notions foot or syllable could not be derived from more basic notions. In a grid notation, a prosodic phrase, say, is represented by the syllable that has been assigned the largest number of asterisks. Had Pinker and Jackendoff adopted a grid notation, their argument against recursion in phonology would already be much harder to state. In fact, I have argued above that prosodic configurations should be represented in terms of bare structure, without reference to derivationally created objects like syllables or feet. Bare prosodic structure is recursive. I thus conclude that the facts of phonology do not threaten the recursion only hypothesis put forth in (Hauser et al.). If anything, they support it.

2.5. Templates and linearization To illustrate the empirical coverage of the model, let us go back to some of the facts discussed above. An important theoretical tool in the description of those facts was the template. Although templates were formalized in terms of CV units, the question of how infixal positions can be recognized was left open. Turning to this issue now, we will identify two strategies of morphological segmentation: linear and hierarchical segmentation. Let us start with linear segmentation. (37) repeats forms II and III of Classical Arabic (kattab ‘make write’, kātab ‘correspond’), as adopted from Guerssel and Lowenstamm above:

88 Martin Haiden (37) a. form II: kattab ‘make write’

b. form III: kātab ‘correspond’

k t b

k t b

C V [DS C V ] C V C V

C V [DS C V ] C V C V

a

a

In these forms, the infixal position does not host any root segment. It is therefore linearly segmentable as follows. Assume two sets of objects: set A contains root CV positions, set B contains affixal CV positions. The infixal position can then be identified by the general rule AxBAy . This rule is slightly more complex than the finite state grammar (AB)n tested by Fitch and Hauser (2004), but it is not recursive, and thus very easy to learn. From this perspective, the persistence of such forms in the grammar of non-concatentive languages (Prunet et al. 2000) is not surprising at all. As a matter of fact, templatic rules constraining the number and position of affixes abound in natural language. They are not limited to the non-concatenative type. Linear segmentation, as in (37), depends on the absence of a root value associated with the infixal site. The mechanism is not available in cases, where a single position expresses both root and affixal features, as in German strong verbs. The stem vowel of these verbs is nearly always associated with a root value. For example, the stem of the verb fall-en ‘fall’ is /fal/. This form contrasts, among others, with /fol/ (voll, ‘full’), /fɛl/ (Fell, ‘fur’). The stem /li:g/ (lieg-en, ‘lie/be lying down’) contrasts with /ly:g/ (lüg-en, ‘lie/cheat’), etc. The past tense forms of /fal/ and /li:g/ are /fi:l/ and /la:g/, respectively. Clearly, the stem vowel of a German strong verb expresses both lexical and functional information. Bendjaballah and Haiden (2003) observe that the realization of multiple morphological features by a single templatic site implies prosodic prominence. For example, the stem vowel of a German strong verb is not only part of the root and a marker of tense. It is also the prosodic head of the verb: it bears stress. On the other hand, the stem-final consonant, which is prosodically weak, bears no independent morphological function. It is morphologically significant exclusively in conjunction with the entire stem. The parallel structure of prominence in prosody and morpho-syntax suggests that markers are identified hierarchically, not linearly, in German strong verbs. (38) illustrates the PF-syntax map for the past form sank of sink-en ‘fall’: the pre-penultimate CV unit is secondarily mapped on T.8 8

For expository reasons, syntactic merger has been ignored in (38).

On bare prosodic structure and the spell-out of features 89

(38)

sank ‘fall/sink.past’

In order to satisfy the Projection Principle, secondary mappings of this kind must be lexically listed. We therefore predict that apophonic markers of the Modern German type can never be fully productive. This seems indeed to be true. Assuming as above a distinction of root (A) and affix (B), the template of German strong verbs must be formalized as a configuration of embedding: [A (CV) [B CV] CVCV]. In this configuration we expect to find values of B other than tense. This is the case of umlaut causatives. Since the internal position [B ] is absent in weak (regular) verbs, we predict that only strong verbs can have an umlaut causative. Furthermore, we predict the weakness of umlaut causatives. Once B is interpreted as Caus, there is no more templatic space for the interpretation of tense. This configuration is depicted in (39). Like apophonic tense marking, umlaut causatives make use of a secondary mapping of their prosodic head. By the Projection Principle, every token of this type must be listed: umlaut causatives are unproductive. (39)

senk-te ‘lowered’

ɛ

ə

3. Conclusion and outlook By way of a conclusion, let us return to the three problems stated at the beginning of this article, the rogue PF problem, the look-ahead problem, and the redundancy of feature deletion operations at the interface. The rogue PF problem has been eliminated as an artefact of an inadequate phonological notation. Bare prosodic structure is recursive. The look-ahead problem was interpreted as a strong argument against late lexical access in the serial architecture of the components of grammar.

90 Martin Haiden If grammatical derivation proceeds in a single cycle spanning over multiple components, the problem of look-ahead into PF does not arise, because phonological and syntactic operations apply at the same time. The redundancy problem was solved by a formalism that delivers interface representations as a necessary by-product of hierarchical structure in prosody and syntax, respectively. Let us finally consider some implications of the theory for parameter setting and acquisition. In the preceding sections the empirical coverage of templates has been illustrated for morphology. However, the tool can be generalized to apply to genuinely syntactic configurations as well. In these cases, templates express generalizations over the linear order of overt constituents. Consider as an example the head parameter that states that a syntactic category of type X takes its complement at the left or right. As a template, the head-final parameter could look roughly as given in (40) – a recursive rule that requires overt dependents of category X to surface at its left-hand side. (40) head-final parameter [X ([Y CV*]) [X CVn]] The statement of parameters in the form of rules, rather than features of lexical entries, finds support in recent experimental studies on the psychological reality of grammatical rule systems. For example, Pléh et al. (2003) use data from Williams Syndrome (WMS) to argue for “a dissociation between the rules of grammar vs. the mental lexicon” (p. 377): “WMS people have an intact rule system and an impaired lexicon” (p. 381). If Pleh et al. are right, then parameters cannot be acquired as part of lexical entries, as it was assumed in early versions of the minimalist program. The dissociation of parameters from the lexicon is independently defended in a series of work on the optional infinitive stage of first language acquisition (cf. Wexler 2002 for an overview). There is strong evidence that (i) parameters play an important role in language acquisition, and (ii) that they are correctly set already, when the child enters the two-word stage, i.e., at around 18 months. At this age, the lexicon of the child is extremely limited, so it cannot possibly be the source of correct parameter settings. By contrast, it has been shown that infants possess a remarkable capacity of extracting abstract rules from acoustic stimuli (Gomez and Gerken 1999; Naigles 2002; Shady et al. 1995). It is therefore reasonable to assume that infants can acquire parameters in the form of generalized templates like (40), and that they can do so before they produce their first sentence.

On bare prosodic structure and the spell-out of features 91

References Anderson, Stephen R. 1981 Why phonology isn’t “natural”. Linguistic Inquiry 12: 493–539. Bendjaballah, Sabrina and Martin Haiden 2002 Templatic architecture and German strong verbs. In 4èmes journées internationales du GDR CNRS 1954 “Phonologies”. Grenoble. 2003 Templatic Architecture. Recherches Linguistiques de Vincennes 32: 157–168. Borer, Hagit 2003 Exo-skeletal vs. endo-skeletal explanations: Syntactic projections and the lexicon. In Explanation in Linguistic Theory, Maria Polinsky and John Moore (eds.). Stanford: CSLI. Bródy, Michael 1995 Lexico-Logical Form. Cambridge, MA: MIT Press. Bromberger, Sylvain and Morris Halle 1989 Why phonology is different. Linguistic Inquiry 20: 51–70. Chomsky, Noam 1995 The Minimalist Program. Cambridge, MA: MIT Press. 2004 Beyond explanatory adequacy. In Structures and Beyond. The Cartography of Syntactic Structures, Vol.3, Adriana Belletti (ed.), 104 – 131. Oxford: Oxford University Press. Cinque, Guglielmo 1993 A Null Theory of phrase and compound stress. Linguistic Inquiry 24: 239. Embick, David 2000 Features, syntax, and categories in the Latin perfect. Linguistic Inquiry 31: 185–230. Fitch, W. Tecumseh and Marc D. Hauser 2004 Computational constraints on syntactic processing in a nonhuman primate. Science 303: 377–380. Fox, Danny 2000 Economy and Semantic Interpretation. Cambridge, MA: MIT Press. Gomez, Rebecca L. and LouAnn Gerken 1999 Artificial grammar learning by one-year-olds leads to specific and abstract knowledge. Cognition 70: 109 –135. Grimm, Jakob 1819 Deutsche Grammatik. 4 Theile. Göttingen. [Reprinted in 1837]. Grimshaw, Jane 1990 Argument Structure. Cambridge, MA: MIT Press. Guerssel, Mohamed and Jean Lowenstamm 1990 The derivational morphology of the Classical Arabic verbal system. Ms. UQAM, Paris 7.

92 Martin Haiden Haiden, Martin 2005a Concatenation and interpretation. In Organizing Grammar: Linguistic Studies in Honor of Henk van Riemsdijk, Hans Broekhuis, Norbert Corver, Riny Huybregts, Ursula Kleinhenz and Jan Koster (eds.), 157–165. Berlin /New York: Mouton de Gruyter. 2005b Theta Theory. Studies in Generative Grammar 78. Berlin /New York: Mouton de Gruyter. Halle, Morris 1953 The German conjugation. Word 9: 45–53. 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.), 111–176. Cambridge, MA: MIT Press. 1994 Some key features of distributed morphology. MIT Working Papers in Linguistics 21: 275–288. Halle, Morris and Jean-Roger Vergnaud 1987 An Essay on Stress. Cambridge, MA: MIT Press. Hauser, Marc D., Noam Chomsky, and W. Tecumseh Fitch 2002 The faculty of language: What is it, who has it, and how did it evolve? Science 298: 1569 –1579. Kaye, Jonathan and Jean Lowenstamm 1984 De la syllabicité. In Forme sonore du langage, structure des représentations en phonologie, François Dell and Jean-Roger Vergnaud (eds.), 123–159. Paris: Hermann. Kiparsky, Paul 1966 Über den deutschen Akzent. Studia Grammatica 7: 69–98. Lowenstamm, Jean 1996 CV as the only syllable type. In Current Trends in Phonology, Jacques Durand and Bernard Laks (eds.), 419–441. Manchester: ESRI. 2003a À propos des gabarits. Recherches Linguistiques de Vincennes 32: 7– 30. 2003b Remarks on Mutae cum Liquida and branching onsets. In Living on the Edge, 28 Papers in Honor of Jonathan Kaye, Stefan Ploch (ed.), 339–363. Berlin /New York: Mouton de Gruyter. Lowenstamm, Jean and Jonathan Kaye 1986 Compensatory lengthening in Tiberian Hebrew. In Studies in Compensatory Lengthening, Leo Wetzels and Engin Sezer (eds.), 97–132. Dordrecht: Foris. McCarthy, John 1979 Formal Problems in Semitic Phonology and Morphology. New York: Garland.

On bare prosodic structure and the spell-out of features 93 Naigles, Letitia R. 2002 Form is easy, meaning is hard: resolving a paradox in early child language. Cognition: 157–199. Neeleman, Ad 1994 Complex predicates. Doctoral dissertation, Utrecht University. Nespor, Marina and Irene Vogel 1986 Prosodic Phonology. Dordrecht / Riverton, NJ: Foris. Pinker, Steven and Ray Jackendoff 2005 The faculty of language: what’s special about it? [2005/3]. Cognition 95: 201–236. Pléh, Csaba, Agnes Lukacs, and Mihaly Racsmany 2003 Morphological patterns in Hungarian children with Williams syndrome and the rule debates. Brain and Language 86: 377–383. Prunet, Jean-Francois, Renee Beland, and Ali Idrissi 2000 The mental representation of Semitic words. Linguistic Inquiry 31: 609–648. Reinhart, Tanya 2006 Interface Strategies. Cambridge, MA: MIT Press. Richards, Norwin 2006 Beyond strength and weakness. Lingbuzz 000325. Scheer, Tobias 2004 A Lateral Theory of Phonology. Vol. 1: What is CVCV, and Why Should It Be? Studies in Generative Grammar. Berlin /New York: Mouton de Gruyter. Ségéral, Philippe and Tobias Scheer 1998 A generalized theory of Ablaut: the case of Modern German strong verbs. In Models of Inflection, A. Ortmann, R. Fabri and T. Parodi (ed.), 28–59. Tübingen: Niemeyer. Selkirk, Elisabeth O. 1984 Phonology and Syntax: The Relation between Sound and Structure. Cambridge, MA: MIT Press. Shady, Michele, LouAnn Gerken, and Peter W. Jusczyk 1995 Some evidence of sensitivity to prosody and word order in tenmonth-olds. In Proceedings of the Boston University Conference on Language Development, D. MacLaughlin and S. McEwen (eds.), 553– 562. Somerville, MA: Cascadilla Press. Siloni, Tal 2001 Construct states at the PF interface. Linguistic Variation Yearbook 1: 229–266. Stowell, Tim 1981 Origins of phrase structure. Ph.D. thesis, MIT, Cambridge, MA.

94 Martin Haiden Vergnaud, Jean-Roger 2003 On a certain notion of ‘occurrence’: The source of metrical structure, and of much more. In Living on the Edge, 28 Papers in Honour of Johnathan Kaye, S. Ploch (ed.), 599–632. Berlin /New York: Mouton de Gruyter. Wexler, Ken 2002 Lenneberg’s dream: Learning, normal language development and specific language impairment. In Language Competence across Populations: Towards a Definition of Specific Language Impairment, J. Schaffer and Y. Levy (eds.), 11–61. Mahwah, NJ: Erlbaum. Zubizarreta, Maria Luisa 1998 Prosody, Focus, and Word Order: Linguistic Inquiry Monographs 33. Cambridge, MA: MIT Press.

Spell out before you merge Hisao Tokizaki

In this paper, I will investigate the relation between the syntactic derivation of sentences and the Spell-Out to PF (Phonetic Form). A problem in the interface between syntax and PF is how computation merges words or constituents and Spells Out them in a bottom-up and cyclic fashion, while PFrepresentation is linear from left to right (cf. Chomsky 1995; Phillips 2003). The aim of this paper is to show that the problem can be solved if words are Spelled Out before they are Merged. I will argue that lexical items and silent demibeats (cf. Selkirk 1984) are introduced to the working space from top to bottom, and that an extra silent demibeat triggers Merge, which proceeds from bottom to top. This model gives us a new view of the syntax-phonology interface and the architecture of grammar.

1. A problem: Bottom-up merge and top-down linearization In the minimalist program, it is assumed that lexical items selected from the lexicon are introduced in the working space and are merged with each other step by step. Consider the following sentence for example: (1)

Alice loves small hamsters.

The lexical items needed to derive (1) are {Alice, Infl, loves, hamsters}. The derivation proceeds as shown in (2). (2)

a. b. c. d.

[small hamsters] [loves [small hamsters]] [Infl [loves [small hamsters]]] [Alice [Infl [loves [small hamsters]]]]

First, the two words small and hamsters are introduced to the working space and are merged as in (2a). Next, another lexical item loves is introduced into the derivation and is merged with [small hamsters] as in (2b). 12

96 Hisao Tokizaki The same process applies to Infl and Alice as shown in (2c) and (2d). Merge proceeds in a bottom-up fashion. Let us call this kind of derivation Merge Up.1 From a phonological point of view, PF representation must be linear, running from left to right. PF-representation of the example sentence (1) extends through the following steps: (3) a. b. c. d.

Alice Alice loves Alice loves small Alice loves small hamsters.

Thus, in right-branching structure, the process of linearization starts from the top of a syntactic tree and proceeds down to the bottom of the tree. Let us call this process Linearize Down. The order of (2a–d) and the order (3a–d) show that there is a tension between syntactic derivation and linearization. Merge proceeds upward while Linearization proceeds downward. Any plausible theories of the syntaxphonology interface must explain this paradox in some way.2 Note that the paradox cannot be explained by phase theories of syntaxphonology interface. If we assume that the sister of a strong phase head (v and C), namely VP and IP (or TP), is Spelled Out to PF (Chomsky 2001), the sentence (1) is derived as shown in (4). (4) a. b. c. d. e.

[small hamsters] [loves [small hamsters]] [v [VP loves [small hamsters]] Spell-Out 1: [VP loves hamsters]  PF [Infl [v [VP… ]]] [IP Alice [Infl [v [VP…]]] Spell-Out 2: [IP Alice [Infl [v [VP…]]]  PF

First the VP is Spelled Out at the stage (4c), and then the IP is Spelled Out at the stage (4e). If this order is the one in which Linearization takes place, 1 2

I thank a reviewer for the term and comments on the earlier version. The paradox becomes apparent when Merge replaces Rewriting rules in syntax. For example, rewriting rules expand VP loves small hamsters into V loves and NP small hamsters. The NP is in turn expanded into small and hamsters. It is possible to argue that the order of derivation and lexical insertion is top-down. In this sense, the paradox of direction between derivation and linearization is a new problem in the theory of grammar.

Spell out before you merge

97

we wrongly predict the following steps in PF-representation resulting in the word order in (5b). (5) a. loves small hamsters b. loves small hamsters Alice To get the right word order in (1), we must assume that the first output of Spell-Out, VP, is stored somewhere in the PF component, and that the second output, IP, is placed at its left. (6) a. loves small hamsters b. Alice loves small hamsters However, as Dobashi (2003) argues, it is not clear how we can guarantee the procedure in the PF component.3 The problem of word order is more serious if we consider the derivation and Linearization of sentences with multiple embedded clauses such as (7). (7) [IP3 Mary [VP3 thinks that [IP2 John [VP2 believes that [IP1 Alice [VP1 loves hamsters]]]]]] This sentence has the following phase units: (8) a. b. c. d. e. f.

[VP1 loves small hamsters] [IP1 Alice v [VP1 …]] [VP2 believes that [IP1 …]] [IP2 John v [VP2 …]] [VP3 thinks that [IP2 …]] [IP3 Mary v [VP3 …]]

The result of multiple Spell-Out would be (9), which is a wrong prediction. (9) loves small hamsters Alice believes that John thinks that Mary This problem becomes even more serious if we assume a hierarchical structure for a discourse. Larson (1990) argues that Coordination structures fall 3

Dobashi (2003) calls this undecidability of word order “the assembly problem”, and proposes that the initial element in the linear string of each Spell-Out should be available to the next Spell-Out. I will not discuss his proposal here.

98 Hisao Tokizaki under X-bar theory and have conjunctions as their heads and that discourses are extended coordinations in their default form, as shown in (10). (10) a. [[He came in] [and [John was tired]]] b. [[He came in.] [& [John was tired.]]] Then, Merge can proceed upward infinitely, as shown in (11). (11) [[IP John [VP woke up] … [&2’ &2 [&P1 [IP He1 [VP washed his face] [&1’ &1 [IP He2 [VP went out]]]]] Again, multiple Spell-Out cyclically sends the phase units, IPs and VPs, in (11) to make a wrong PF-representation in (12). (12) went out he2 washed his face he1 … woke up John It is clear that Linearization must start with the topmost phase unit and then go down to the lowest cycle in order to give the right order in (11). The point here is that a sentence or a discourse can consist of an infinite number of phase units. In the standard assumption in the current syntactic theory, Merge Up starts from the lowest phase unit while Linearize Down starts from the highest phase units. Note that the same problems occur with other theories of phase such as Uriagereka (1999) and strong derivational theory such as Epstein et al. (1998). They propose different domains of phase unit, but they assume that Merge applies in a bottom-up fashion to build up a sentence. If the phase units are smaller than VPs and IPs, as in strong derivational theory, the problem is more serious. The word order in PF-representation will be almost the opposite of the actual sentence or discourse if each Merge Spells Out the resulting constituent, as shown in (13). (13) a. [Alice [loves [small hamsters]]] b. small hamsters loves Alice We have seen that bottom-up Merge raises a problem with top-down Linearization in the multiple Spell-Out model of grammar. In the following sections, we will consider how we can solve the problem in the minimalist framework.

Spell out before you merge

99

2. Branch Right and its problems The up-down paradox we have seen above does not occur in the incremental model of derivation proposed by Phillips (1996, 2003), who assumes Branch Right instead of Merge. For example, a VP with Ns as its specifier and complement is constructed in the order shown in (14) (cf. Richards 1999). (14) a. [Mary saw] b. [Mary [saw John]] It is argued that the right node V dominating saw in (14 a) branches into VP [saw John] as shown in (14b). The parser proceeds from left to right, and the structure is built in a top-down fashion at the same time.4 This theory avoids the up-down paradox that occurs in the standard minimalist syntax with Merge and Linearization. However, the status of Branch Right in the grammar is not clear. Phillips does not specify how Branch Right applies to a single lexical item and makes it branch into two nodes. In other words, we do not know what mechanism changes the verb saw in (14a) into the VP saw John in (15 b). Moreover, Branch Right has an empirical problem. As Shiobara (2005) points out, it cannot build left branching structure such as the subject in (15). (15) [[The girl] [saw John]] Branch Right wrongly builds [The [girl [saw John]]]. This problem is more serious in building recursive left-branching structures such as [[[Wadeck’s Mother’s] Friend’s] Son].5 It is not clear how we can guarantee that Branch Right sometimes applies not to the rightmost word but to the constituent dominating it. Though the idea of Branch Right is appealing if we want to resolve the contradiction of direction between derivation and linearization, it has both conceptual and empirical problems. We will try to find other ways to show that the paradox is not a real contradiction.

4

5

Kempson et al. (2001) and O’Grady (2005) also propose a similar system of incremental derivation. The arguments against Merge Right to be presented below apply to their derivational systems. This is the title of an American movie directed by Arnold Barkus (1992).

100 Hisao Tokizaki 3. Spell-Out before Merge One way to keep both Merge Up and Linearize Down without contradiction is to assume that Spell-Out sends lexical items with information about the constituent structure to be built before Merge combines two syntactic objects. I will explore this possibility in detail below. First, I will illustrate the basic idea with the example sentence (1) above. I argue that the computational system puts a lexical item with a syntactic bracket in the working space of derivation and Spells Out them to PF at the same time. Then the derivation proceeds in the following steps: (16) a. b. c. d. e.

[Alice [Alice [Infl [Alice [Infl [loves [Alice [Infl [loves [small [Alice [Infl [loves [small [hamsters

So far, no Merge applies, but Linearization has applied from top to bottom. In this model, Merge applies when a closing bracket is introduced in the derivation. First, a closing bracket is added to (16e) to make a non-branching constituent [hamsters] as shown in (17). (17) [Alice [Infl [loves [small [N hamsters] Then, another closing bracket is introduced into derivation to make the NP small hamsters as shown in (18). (18) [Alice [Infl [loves [NP small [hamsters]] This process, combining two lexical items, is what is called Merge in the standard assumption. Similarly, a closing bracket is introduced into the derivation to merge two constituents, making a larger constituent, VP, I’, and IP, as shown in (19a), (19b), and (19c). (19) a. [Alice [Infl [VP loves [small [hamsters]]] b. [Alice [I’ Infl [loves [small [hamsters]]]] c. [IP Alice [Infl [loves [small [hamsters]]]]] In the derivation shown in (16) to (19), Linearization applies from top to bottom, and Merge applies from bottom to top. Thus, there is no contradiction

Spell out before you merge

101

between Linearize Down and Merge Up in this derivational model. Each lexical item is Spelled Out before being Merged with another syntactic object. There is no phase unit for PF interface such as VP or IP in the sense of Chomsky (2001). Each lexical item is Spelled Out when introduced into the derivation. Note that Spell-Out strips only phonetic features from lexical items in the working space, as is assumed in the standard theory of derivation and interface. Semantic features remain in the derivation until Merge applies to make semantic units and send semantic representations to LF, perhaps at both the CP and vP phases. Syntactic features also remain in the derivation until they are checked and deleted when Merge makes checking domains. In the next section, I will show the details of derivation, the Spell-Out to PF, and semantic features.

4. Spell-Out of brackets as silent beats A number of questions arise. Are starting brackets and closing brackets syntactic objects? Does the speaker Spell Out brackets? If so, how? I argue that brackets are real objects in syntactic representation. We need brackets to show that two syntactic objects are merged into one. We may think of brackets as a special kind of lexical items. They have phonetic features in that they indicate silent pause duration, syntactic features in that they show the boundary of a constituent, and no semantic features. They are included in Numeration with the number of times they appear, and introduced into derivation when they are necessary for convergence. For example, the Numeration of the sentence (1) should be {Alice, Infl, loves, small, hamsters, [x5, ]x5} for the convergent derivation. Moreover, it is clear that a sentence has silences or pauses between two words. These silences or pauses should be represented in some form in PF, which must have some corresponding syntactic objects. In other words, Spell-Out sends phonetic features of syntactic objects to PF. PF has some representation for silences or pauses. Then, syntactic representation must have some syntactic objects which are interpreted as PF objects corresponding to silence or pauses. Thus, it is plausible to assume that syntactic brackets are syntactic objects, which have a phonetic (silence) feature and no semantic features. I also argue that a bracket is introduced into derivation at the same time that a lexical item is introduced. Consider what would happen if no bracket was introduced before the first word, as shown in (20).

102 Hisao Tokizaki (20) Alice This is not going to give any convergent derivation. Spell-Out strips phonetic features from Alice at the step (20). After this Spell-Out, there is no chance for the word to be included in a constituent even if any brackets are introduced, as shown in (21). (21) Alice [loves [small [hamsters]]]] The derivation (21) crashes at LF because Alice is a “stray” in syntactic structure. Alice cannot be assigned any thematic role under the standard assumption that theta is assigned by configuration in a syntactic tree (Baker 1988; Hale and Keyser 1993). Thus, when a lexical item is introduced into a derivation, a bracket must also be introduced, as we have shown in (16). Before considering an example derivation in detail, let us consider what kind of phonological features a bracket has. As I mentioned above, a sentence has pauses and silences between two words. A plausible object to represent pauses and silences in PF is Selkirk’s (1984) Silent Demibeat. Silent demibeats are assumed to be contained in the metrical grid of an utterance. Selkirk argues that the representation of the rhythmic structure of the sentence Abernathy gesticulated is as follows, where the underscored grids are silent demibeats: (22) x x x x x x x Abernathy

x x x x xx x x x x x xxx gesticulated

Selkirk argues that silent demibeats are added at the end (right extreme) of the metrical grid aligned with a word or certain constituents.6 Here I will

6

Selkirk (1984) argues that the sentence in (ia) contains a number of silent demibeats (x) as shown in (ib). (i) a. [S [NP [ N Mary]] [VP [ V finished] [NP her [ A Russian] [ N novel]]]] b. Mary xxx finished xx her Russian x novel xxxxx a, b, d a, b a a, b, c, d The silent demibeats in (ib) are assigned by Silent Demibeat Addition (ii).

Spell out before you merge

103

assume that a syntactic bracket is Spelled Out as a silent demibeat. We can formulate the idea into the following rule for Spell-Out (cf. Tokizaki 1999): (23)

& $ $ $ $ $ ' $ $ $ $ $ (

[ ]

" $ $ $ $ $ # $ $ $ $ $ %



x

!

Spell-Out interprets a syntactic bracket, either left or right (cf. Ferreira 1993; Watson and Gibson 2004), as a silent demibeat. The rule (23), in effect, encodes the syntactic structure (24a) into PF representation (24b). (24) a. [Alice [loves [small hamsters]]] b. x Alice x loves x small hamsters xxx Note that derivation proceeds with multiple Spell-Out. Every time a syntactic object is introduced into derivation, its phonetic features are sent to PF. If it is a syntactic bracket, a silent demibeat is sent to PF. Now let us consider each step of the derivation in detail. For example, consider the derivation of the sentence (1) again. In each of the derivational steps in (25) below, the first line shows syntactic features, the second semantic features, and the third phonological features.

(ii) Add a silent demibeat at the end (right extreme) of the metrical grid aligned with a. a word, b. a word that is the head of a nonadjunct constituent, c. a phrase, d. a daughter phrase of S. In (i), according to Selkirk, Mary is a word, and a word that is the head of a nonadjunct constituent, and a daughter phrase of S. Thus, three silent demibeats are added to the right of Mary by (ii a), (iib), and (iic). Here I will generalize Silent Demibeat Addition. Putting aside (iib) and (iid), a word gets a silent demibeat after it by (ii a). If the word is the final one in a phrase, it gets another silent demibeat by (iic). Instead of listing categories receiving a silent demibeat at its right edge as in (ii a–d), I assume the syntaxphonology correspondence rule as shown in (23).

104 Hisao Tokizaki (25) a. [Alice ‘Alice’ x ælɪs  PF: x ælɪs b. [Alice [Infl ‘Alice’ PF: x ælɪs c. [Alice [Infl [loves ‘Alice’ ‘loves’ lʌvz  PF: x ælɪs x lʌvz d. [Alice [Infl [loves [small ‘Alice’ ‘loves’ ‘small’ smɔːl  PF: x ælɪs x lʌvz x smɔːl e. [Alice [Infl [loves [small [hamsters ‘Alice’ ‘loves’ ‘small’ ‘hamsters’ x hæmstɚz  PF: x ælɪs x lʌvz x smɔːl x hæmstɚz f. [Alice [Infl [loves [small [N hamsters] ‘Alice’ ‘loves’ ‘small’ ‘hamsters’ x  PF: x ælɪs x lʌvz x smɔːl x hæmstɚz x g. [Alice [Infl [loves [NP small [hamsters]] ‘Alice’ ‘loves’ ‘small’ ‘hamsters’ x  PF: x ælɪs x lʌvz x smɔːl x hæmstɚz xx h. [Alice [Infl [VP loves [small [hamsters]]] ‘Alice’ ‘loves’ ‘small’ ‘hamsters’ x  PF: x ælɪs x lʌvz x smɔːl x hæmstɚz xxx i. [Alice [ I’ Infl [loves [small [hamsters]]]] ‘Alice’ ‘loves’ ‘small’ ‘hamsters’  LF: ‘loves small hamsters’ PF: x ælɪs x lʌvz x smɔːl x hæmstɚz xxx j. [IP Alice [Infl [loves [small [hamsters]]]]] ‘Alice’  LF: ‘Alice loves small hamsters’ x -> PF: x ælɪs x lʌvz x smɔːl x hæmstɚz xxxx In (25a), the first lexical item Alice and a bracket are introduced into the derivation and their phonetic features are Spelled Out at the same time. The syntactic and semantic features of Alice remain in the derivation. In (25b), Infl and a bracket are introduced, but I assume that a phonologically empty element (e.g. Infl and trace) and the brackets merging it with another syntactic object are invisible to Spell-Out (cf. Tokizaki 2006). In (25c), the

Spell out before you merge

105

third item loves and a bracket are introduced, and its phonetic features are Spelled Out. What are left in the working space in the derivation are the syntactic and semantic features of Alice and loves. Similarly, in (25d) and (25e), small and hamsters are introduced together with a bracket, and are Spelled Out. A first closing bracket is introduced in (25f) and is Spelled Out as a silent demibeat. In (25g) to (25j), a closing bracket allows Merge to make a larger constituent, NP, VP, I’, and IP. At a phase (25i), the semantic features of VP are sent to LF. LF is completed at the next phase (25j) where the semantic feature of Alice is added.7 So far, I have argued that brackets, a kind of syntactic objects with a phonetic pause feature, are Spelled Out as silent demibeats in PF representation. I showed each step of syntactic derivation, the Spell-Out of phonological features to PF, and semantic interpretation. This shows how Linearization proceeds from top to bottom while Merge proceeds from bottom to top.

5. Parsing: Reconstruction of syntax from PF In this section, I will argue that parser also uses silent demibeats to reconstruct phrase structure. I have argued that Spell-Out interprets a syntactic bracket (either left or right) as a silent demibeat. Here I propose that parser interprets a silent demibeat as a syntactic bracket to reconstruct the intended phrase structure. The interpretation of a silent demibeat might be formulated as follows: (26) x 

& $ $ $ $ $ ' $ $ $ $ $ (

[ ]

" $ $ $ $ $ # $ $ $ $ $ %

!

This rule is the reverse of the Spell-Out rule in (23). However, this time we face the problem of which bracket is the one intended: right or left. Parser has no information about the direction of a bracket. I propose that a silent demibeat immediately followed by a lexical item is interpreted as a left bracket. This parsing rule can be formulated as (27). 7

The reviewer raises the question how contraction phenomena such as wanna are explained in this system. A possible answer is to assume that contracted forms are listed as items in the lexicon and are introduced into derivation as such, as shown in (i) (cf. Goodall 2006 and the references cited therein). (i) [I [Infl [wanna [dance [tonight]]]]]

106 Hisao Tokizaki (27) x α  [α

(α: a lexical item)

For example, the parsing of the sentence (24b) proceeds as shown in (28). Each step in (28) corresponds to a derivational step in (25).8 (28) PF Parsing a. b. c. d. e.

x ælɪs x ælɪs x ælɪs x lʌvz x ælɪs x lʌvz x smɔːl x ælɪs x lʌvz x smɔːl x hæmstɚz

[Alice [Alice [Infl [Alice [Infl [loves [Alice [Infl [loves [small [Alice [Infl [loves [small [hamsters

The question is what happens next. I will use the serial alphabet from (28) to show the correspondence between Spell-Out (25) and the reconstruction. (29) f. x ælɪs x lʌvz x smɔːl x hæmstɚz x The rule in (27) cannot change the silent demibeat following hæmstɚz into a left bracket because the silent demibeat is not followed by a lexical item at this step. Thus, no reconstruction applies until next step shown in (30g). (30) g. x ælɪs x lʌvz x smɔːl x hæmstɚz xx Now we need another parsing rule to change silent demibeats into right brackets. Let us assume the following rule for right brackets: (31)

xx  ]x

Then, parsing proceeds as in (32).9

8

9

How Infl in (28b) is reconstructed from PF is an open question. See the discussion on the invisibility of phonologically empty elements below (25). In fact, we need one more silent demibeat as shown in (32k) than we have seen in (25j) in order to Merge Alice and the I’. I argue that the final silent demibeat is supplied by the next sentence in a discourse. The PF in (32k) is mapped from such discourse as (i), which has six brackets between hamsters and she. (i) [Alice [Infl [loves [small [hamsters]]]]] [She….

Spell out before you merge

107

(32) PF Parsing g. x ælɪs x lʌvz x smɔːl x hæmstɚz xx [Alice [Infl [loves [small [N hamsters] h. x ælɪs x lʌvz x smɔːl x hæmstɚz xxx [Alice [Infl [loves [NP small [hamsters]] i. x ælɪs x lʌvz x smɔːl x hæmstɚz xxx [Alice [Infl [loves [small [hamsters]] j. x ælɪs x lʌvz x smɔːl x hæmstɚz xxxx [Alice [ I’ Infl [VP loves [small [hamsters]]]] k. x ælɪs x lʌvz x smɔːl x hæmstɚz xxxxx [IP Alice [Infl [loves [small [hamsters]]]] ] Thus, we can explain how parser builds a syntactic tree from pause durations between words.10 Notice that in (32), as well as the structures we have seen in section 3 and section 4, only the rightmost lexical item hamsters by itself is contained in a pair of brackets. In a tree diagram, the lexical item would be dominated by a non-branching node. (33) Alice Infl loves small hamsters Interestingly, this tree is the same as Kayne’s (1994) tree based on the Linear Correspondence Axiom. Kayne argues that syntactic structure must be asymmetric and that the rightmost element in a tree must be immediately dominated by a non-branching node. For example, Kayne (1994: 10) shows the following structure, where N is immediately dominated by a nonbranching NP node.

10

How the right bracket of I’ is reconstructed from PF in (32j) is an open question. See note 9.

108 Hisao Tokizaki (34) [K J [VP [V see] [NP [N John]]]] The analysis of linearization and parsing presented here also needs the rightmost non-branching node in a tree, as we have seen in section 4 and this section. The similarity between Kayne’s idea of asymmetry and the analysis presented here may be rooted in a deep principle of language, but I will not pursue it here.

6. Derivation and parsing of left-branching structure Now let us turn to cases of left-branching structure. For example, consider the following sentence: (35) [[Alice [Walker]] [Infl [loves [hamsters]]]] Here the subject NP branches into two words. Let us focus on the initial part of the derivation, Spell-Out, and parsing of (35). (36) a. b.

syntax

PF

Parsing

[[Alice

xx ælɪs

]x Alice ] [Alice

Here, two brackets are introduced into the derivation with Alice in order to get the convergent derivation. The Spell-Out rule (23) applies to (35) to give a PF with silent demibeats. However, the parsing rule (31) wrongly changes the first silent demibeat to a right bracket at the first step (36a). However, the parsing in (36b) is a vacuous closing of a constituent because there is no lexical item to be enclosed by the right bracket and no left bracket to be paired with. Such a vacuous parsing of a silent demibeat should be banned in principle. Then, in this case, parser must interpret the first silent demibeat as a left bracket as shown in (37). (37)

syntax a. b.

[[Alice

PF 

xx ælɪs

Parsing 

[x Alice [[Alice

It has been pointed out that in right-branching languages such as English, left-branching structure is somewhat marked compared to right-branching structure (cf. Quirk et al. 1985). We can argue that this markedness of left-

Spell out before you merge

109

branching structure may come from the marked interpretation of double silent demibeats. The derivation, Spell-Out, and parsing following (37) are straightforward as shown in (38), where I omit Infl for the sake of simplicity of illustration. (38) a. [[Alice [Walker  x wɔːkɚ  [[Alice [Walker b. [[Alice [Walker]]  wɔːkɚ xx  [[Alice [Walker] c. [[Alice [Walker]] [loves  wɔːkɚ xxx loves  [[Alice [Walker]] [loves The rest is similar to (25d) to (25j). Thus, we conclude that left-branching structure can also be derived by the Spell-Out and Merge operation we have proposed for right-branching structure.11 For parsing, I argued that left-branching structure can be parsed by marked interpretation of silent demibeats as shown in (37a), but not by the unmarked interpretation (31).

7. Phonological evidence for the analysis Finally, let us look at some phonological evidence for the analysis presented above. First, let us consider the prosody of structurally ambiguous sentences. Cooper and Paccia-Cooper (1980) report that speakers put a longer pause between cop and with in (39a) than in (39b). (39) a. Jeffrey hit [the cop] [with a stick] [127.7 msec] (Jeffrey had the stick) b. Jeffrey hit [the cop with a stick] [97.1 msec] (The cop had the stick) The Spell-Out and Merge operation we have argued derives the following syntactic structures: (40) a. [Jeffrey [hit [the [cop]] [with [a [stick]]]]] b. [Jeffrey [hit [the [cop [with [a [stick]]]]]]] The brackets are Spelled Out as silent demibeats by the rule (23) as shown in (41).

11

For the possibility of different juncture between left-branching and right-branching structure, see Tokizaki (2008).

110 Hisao Tokizaki (41) a. x dʒefri x hɪt x ðə x kɑp xxx wɪð x ə x stɪk xxxxx b. x dʒefri x hɪt x ðə x kɑp x wɪð x ə x stɪk xxxxxxx The number of silent demibeats between cop and with is three in (41a) and one in (41b). A silent demibeat is interpreted as a certain length of pause at the articulatory-perceptual system (A-P). Then, we correctly expect a longer pause for the three silent demibeats between cop and with in (41a) than for the one silent demibeat at that position in (41b). The Spell-Out and Merge operation proposed here explains why the pause duration at that point is longer in (39a) than in (39b). Thus, the different pause lengths in structurally ambiguous sentences give support to the analysis presented here. For parsing, we also have some evidence for our analysis. It has been argued that hearers tend to interpret a prosodic break as a major constituent break. For example, according to Pynte and Prieur (1996), if the second prosodic break is put between NP and PP in (42), the interpretation (43b) is preferred to (43a). (42) The spies [VP informed # [NP the guards] (#) [PP of NP]] (43) a. The spies [informed [the [guards [of [the palace]]]]] b. The spies [informed [the [guards]] [of [the conspiracy]]]] We can explain this preference with the analysis presented above. I argue that a prosodic break in the A-P system corresponds to a certain number of silent demibeats in PF representation (cf. Tokizaki 2006). Then, it is plausible to assume that the second prosodic break between NP and PP is perceived by a hearer as two more silent demibeats as shown in (44b). (44)

A-P a. .. gɑːdz əv.. b. .. gɑːdz # əv..

PF .. gɑːdz x əv.. .. gɑːdz xx .. gɑːdz xxx .. gɑːdz xxx əv..

Parsing .. guards [of.. .. guards] .. guards]] .. guards]] [of..

(= (43a))

(= (43b))

A hearer interprets a prosodic break between guards and of as a series of silent demibeats as shown in (44b) rather than (44a). The silent demibeats are parsed by (31) and then by (27) as two right brackets and a left bracket, as shown in the steps in (44b). Thus, we explain why a prosodic break tends to be interpreted as a major constituent break in cases such as (42). This discussion gives support to the parsing system proposed above.

Spell out before you merge

111

In this section, I have argued that some production and perception data can be naturally explained with the analysis presented here. This fact gives support for the Spell-Out before Merge hypothesis.

8. Summary I have argued that we can resolve the contradiction in direction between top-down linearization and bottom-up tree structuring if we assume a Spell Out before Merge hypothesis. Computational system Spells Out a lexical item and a bracket to PF stepwise as a sound and a silent demibeat. Merge is triggered by inserting a right bracket, which encloses a constituent with a preceding left bracket. Parser interprets the silent demibeats in PF representation as syntactic brackets which build a syntactic tree. Thus, there is no paradox between Linearize Down and Merge Up. I hope that the analysis of the syntax-PF interface presented here will constitute a small but steady step toward the overall goal of the minimalist program.

Acknowledgements This is a revised and enlarged version of a chapter of Tokizaki (2006). I would like to thank Kleanthes Grohmann, Richard Kayne, Howard Lasnik, Satoshi Oku, Kayono Shiobara, Yoshihito Dobashi, and the reviewer of this paper for their valuable comments and suggestions. Thanks go to William Green for correcting stylistic errors. All remaining errors are, of course, my own. This work is supported by Grant-in-Aid for Scientific Research and Sapporo University.

References Baker, Mark 1988 Incorporation. Chicago: Chicago University Press. Chomsky, Noam 1995 The Minimalist Program. Cambridge, MA: MIT Press. 2001 Derivation by phase. In Ken Hale: A Life in Language, Michael Kenstowicz (ed.), 1–52. Cambridge, MA: MIT Press. Chomsky, Noam and Morris Halle 1968 The Sound Pattern of English. New York: Harper & Row.

112 Hisao Tokizaki Collins, Chris 2002 Eliminating labels. In Derivation and Explanation in the Minimalist Program, Samuel D. Epstein and T. Daniel Seely (eds.), 42–64. Oxford: Blackwell. Cooper, William E. and Jeanne Paccia-Cooper 1980 Syntax and Speech. Cambridge, MA: Harvard University Press. Dobashi, Yoshihito 2003 Phonological Phrasing and Syntactic Derivation. Doctoral dissertation, Cornell University, Ithaca, NY. Epstein, Samuel D., Erich M. Groat, Ruriko Kawashima, and Hisatsugu Kitahara 1998 A Derivational Approach to Syntactic Relations. New York /Oxford: Oxford University Press. Ferreira, Fernanda 1993 Creation of prosody in sentence production. Psychological Review 100: 233–253. Goodall, Grant 2006 Contraction. In The Blackwell Companion to Syntax, Vol. 1, Martin Everaert and Henk van Riemsdijk (eds.), 688–703. Oxford: Blackwell. Hale, Kenneth and Samuel J. Keyser 1993 On argument structure and the lexical expression of semantic relations. In The View from Building 20, Kenneth Hale and Samuel J. Keyser (eds.) , 53–109. Cambridge, MA: MIT Press. Hawkins, John A. 1994 A Performance Theory of Order and Constituency. Cambridge: Cambridge University Press. Kayne, Richard S. 1994 The Antisymmetry in Syntax. Cambridge, MA: MIT Press. Kempson, Ruth, Wilfried Meyer-Viol, and Dov Gabbay 2001 Dynamic Syntax: The Flow of Language Understanding. Oxford: Blackwell. Larson, Richard K. 1990 Double objects revisited: Reply to Jackendoff. Linguistic Inquiry 21: 589–632. Nespor, Marina and Mauro Scorretti 1984 Empty elements and phonological form. In Grammatical Representation, Jacqueline Guéron, Hans-Georg Obenauer, and Jean-Yves Pollock (eds.), 223–235. Dordrecht: Foris. O’Grady, William 2005 Syntactic Carpentry: An Emergentist Approach to Syntax. Mahwah, NJ: Lawrence Erlbaum. Phillips, Colin 1996 Order and structure. Doctoral dissertation, MIT, Cambridge, MA. 2003 Linear order and constituency. Linguistic Inquiry 34: 37–90.

Spell out before you merge

113

Pynte, Joel and Benedicte Prieur 1996 Prosodic breaks and attachment decisions in sentence parsing. Language and Cognitive Process 11: 165–191. Quirk, Randolph, Sidney Greenbaum, Geoffrey Leech, and Jan Svartvik 1985 A Comprehensive Grammar of the English Language. London: Longman. Richards, Norvin 1999 Dependency formation and directionality of tree construction. MIT Working Papers in Linguistics 34: 67–105. Selkirk, Elisabeth O. 1984 Phonology and Syntax. Cambridge, MA: MIT Press. Shiobara, Kayono 2005 Linearization: A derivational approach to the syntax-prosody interface. Doctoral dissertation, University of British Columbia. Tokizaki, Hisao 1999 Prosodic phrasing and bare phrase structure. NELS 29, Vol. 1, 381– 395. 2005 Prosody and phrase structure without labels. English Linguistics 22: 380–405. 2006 Linearizing syntactic structure with brackets and silence: A minimalist theory of syntax-phonology interface. Doctoral dissertation, the University of Tsukuba. [Revised version published 2008 as Syntactic Structure and Silence. Tokyo: Hitsuji Shobo.] 2008 Symmetry and asymmetry in the syntax-phonology interface. Phonological Studies 11: 123–130. The Phonological Society of Japan. Uriagereka, Juan 1998 Rhyme and Reason: An Introduction to Minimalist Syntax. Cambridge, MA: MIT Press. 1999 Multiple spell out. In Working Minimalism, Samuel David Epstein and Norbert Hornstein (eds.), 251–281. Cambridge, MA: MIT Press. Wasow, Thomas 2002 Postverbal Behavior. Stanford, CA: CSLI. Watson, Duane and Edward Gibson 2004 The relationship between intonational phrasing and syntactic structure in language production. Language and Cognitive Processes 19: 713–755.

On the derivation of the relation between givenness and deaccentuation: A best-case model Dalina Kallulli

1. Introduction It is well known that, across languages, so-called “given” information (as opposed to “new” information) systematically correlates with lack of phonetic prominence, or deaccentuation (Halliday 1967; Taglicht 1982; Ladd 1996; Selkirk 1995; Schwarzschild 1999, among others). The primary goal of this paper is to provide a minimalist syntactic account for a sub-case of this phenomenon, namely the correlation between givenness and/or presupposition of clausal complements on the one hand, and deaccentuation, on the other – a correlation that, as I will show, holds across several languages. From a more general perspective, this paper is an attempt to contribute to the ongoing debate on the nature of the different components of grammar, and the division of labour among them. A widely assumed view among minimalist syntacticians is that the nature of operations taking place in (narrow) syntax is entirely different from those occurring at the PF (and LF) interface. Not infrequently, however, various tasks and even principles that have long been argued to constitute core syntactic operations are indiscriminately shovelled onto the interfaces, especially the PF one.1 The question then arises where exactly the boundary between narrow syntax and everything else is. Of course different components deal with different objects and representations, what is missing however is a clear model of how exactly the computational system feeds both PF and LF to produce the systematic correlations that obtain between the ‘sound’ part and the ‘meaning’ part for many linguistic phenomena. Inspired by Chomsky’s (2004) “best case” scenario, according to which the components of the derivation of proceed cyclically in parallel, the central claim that I put forward in this paper is that information structure is encoded in the (narrow) syntax, and in particular, that (at least 1

One well-known case in point is the EPP (Ndayiragije 2000; Holmberg 2000; Miyagawa 2001; Bobaljik 2002; Holmberg and Hróarsdóttir 2003; Landau 2007, among others).

116 Dalina Kallulli some) information which leads to certain prosodic outcomes, or spell-out forms, is part of the numeration (for instance, in the sense of an abstract morpheme – Embick and Noyer 2007), or its analogue.2 Specifically, I contend that: (i) the [+presupposed] and/or [+given] status of an embedded CP must be expressed; (ii) this is achieved syntactically (via functional structure); and (iii) the head hosting the relevant syntactic feature, which is instantiated overtly either by some expletive-like element (such as a clitic, a pronoun, a modal element, etc.), or simply by destressing (or deaccentuation), is a probe for the goal (here: CP) with an OCC feature (Chomsky 2001), yielding the various patterns that are found and that are described in some detail in this paper. This paper thus deals with interfaces and some sort of clausal domain/part that is relevant in the course of the derivation. While phases may bear relevance for this, my focus in this paper is however on the interface part. This paper is organized as follows. In section 2 I introduce the phenomenon of factivity, show that it can be triggered even for so-called “nonfactive” verbs of it, and make a point of why the distinction factive/nonfactive may be thought of in terms of information structure. Then section 3 lays out a syntactic model of the correlation between factivity and deaccentuation, which crucially relies on Chomsky’s (2004) architectural best-case desideratum. Finally in section 4 I discuss certain extensions that push the research agenda introduced here one step further towards a more ‘syntactic’ account of prosody, and more generally of information structure.

2. Presuppositions, factivity, and factivity triggers Kiparsky and Kiparsky (1970) argue that factive verbs differ from nonfactive ones in that the truth of the clausal complements of the former is presupposed, whereas the truth of the clausal complements of the latter is asserted. Consequently, negating the complement clause of a factive verb yields a contradiction, as in (1), whereas negating the complement clause of a non-factive verb does not, as in (2). 2

Some theorists (e.g. Frampton and Gutman 1999) have questioned the usefulness of numeration. However, whatever the fate of numeration in syntactic theory, the idea here is that at least some information which is spelled out prosodically is pre-syntactic (i.e. lexically specified). As such, it must partake in syntactic computation.

On the derivation of the relation between givenness and deaccentuation

(1)

I regretted that John left (*but in fact he didn’t).

(2)

I believed that John left (but in fact he didn’t).

117

However, also for non-factive verbs factivity can be triggered, for instance, by a modal, as in (3b) and (3c) (as opposed to (3a)), and/or by a so-called pleonastic or correlative pronoun “doubling” the embedded clause, as in (4).3 (3)

a. I believed that John left (but in fact he didn’t). b. Can you believe that John left? *In fact, he didn’t. c. I can believe that John left (*but in fact he didn’t).

(4) I didn’t believe it that John left. *In fact he didn’t. The same pattern is replicable in German. Thus, the sentence in (5b) differs formally from the sentence in (5a) only in that the former contains a correlative pronoun, namely es ‘it’, which seems to be responsible for the factivity of the verb glauben ‘believe’ here. (5)

a. Er glaubte, dass Peter verstarb (aber tatsächlich lebt er noch). he believed that Peter died (but factually lives he still) ‘He believed that Peter died (but in fact he is still alive).’ b. Er glaubte es, dass Peter verstarb (*aber tatsächlich lebt er noch). he believed it that Peter died (but factually lives he still) ‘As for the fact that Peter died he believed it (*but actually he is still alive).’

Similarly, across several so-called clitic doubling languages such as Albanian and Modern Greek, factivity is triggered by clitic pronouns doubling the clausal complement, as shown in (6b) vs. (6a) and (7b) vs. (7a) for Albanian and Modern Greek, respectively.4 (6)

3

4

a. Besova se Beni shkoi (por në fakt nuk shkoi). believed-I that Ben left (but in fact not left) ‘I believed that Ben left (but in fact he didn’t).’

The latter fact is discussed by Kiparsky and Kiparsky (1970) themselves, albeit under a non-pleonastic treatment of the pronoun it. I thank Marika Lekakou for providing the Modern Greek data.

118 Dalina Kallulli b. E besova se Beni shkoi (*por në fakt nuk shkoi). 3S.CL.ACC believed-I that Ben left (but in fact not left) ‘I believed the fact that Ben left (*but he didn’t leave).’ (7)

a. Pistepsa oti o Janis efije believed-I that the Janis left (ala stin pragmatikotita den ejine kati tetio). (but in.the reality not happened something such) ‘I believed that Janis left (but in fact he didn’t).’ b. To pistepsa oti o Janis efije it.CL.ACC believed-I that the Janis left (*ala stin pragmatikotita den ejine kati tetio). (but in.the reality not happened something such) ‘I believed the fact that Janis left (*but he didn’t leave).’

In sum, pronouns seem to be a preferred means of triggering factivity across several languages. The question why this is the case is legitimate. I will start approaching this question by highlighting the semantic contribution of doubling clitics and/or so-called pleonastic pronouns in languages that have them.

2.1. Clitic doubling and information structure Research on the semantic contribution of clitic doubling especially in Balkan languages where this phenomenon is prevalent has shown that clitic doubling systematically produces information structure in that it marks doubled material as [+topic] (in the sense: [+given]) – see Kallulli (2000, 2001) for Albanian and Greek, Franks and King (2000) for Bulgarian, and the references therein.5 Consider the Albanian examples in (8).

5

I will not attempt a formal definition of topichood/givenness here. Roughly, I take it to be the complement of focus – see Schwarzschild (1999). For my purposes here it suffices to state the core intuition around what it means for an utterance to express given information, namely that the utterance is already entailed by the discourse.

On the derivation of the relation between givenness and deaccentuation

(8)

119

a. Ana lexoi librin. Anna.NOM read book.the.ACC ‘Anna read the book.’ b. Ana e lexoi librin. Anna.NOM 3S.CL.ACC read book.the.ACC ‘Anna read the book.’

The examples in (8a) and (8b) constitute a minimal pair. They differ only with respect to the clitic element doubling the direct object in (8b). However, the way the examples in (8) stand, it seems as if the doubling clitic in (8b) is optional. But, as I have shown in Kallulli (1999, 2000), the felicity conditions for the sentences in (8a) and (8b), and more generally for sentences with and without clitic doubled direct objects are complementary, as elicited through the question-answer pairs in (9) through (12). Specifically, when the object is focus or part of the focus domain – as brought out by the contexts provided in (9A) and (10A) – a doubling clitic is not tolerated. Crucially, the doubling construction (in (11B) and (12B)), may only be a felicitous reply to the questions in (11A) and (12A), but not to (9A) and (10A). Moreover, note that the presence of the doubling clitic in these cases (i.e., in the contexts provided by (11A) and (12A)) is not only sufficient, but indeed necessary.6 Thus, direct objects in Albanian need to be clitic doubled in order to be interpreted as topics/given. (9)

A: What did Anna do?

B: Ana (*e) lexoi libr-in.

(10) A: What did Ana read?

B: Ana (*e) lexoi librin.

(11) A: Who read the book?

B: Ana *(e) lexoi librin.

(12) A: What did Anna do with the book? B: Ana *(e) lexoi librin. Secondly, Reinhart (1995: 85) remarks that “even in view of the massive varieties of opinions regarding what topics are, [there] is one context all studies agree upon: the NP in there-sentences can never be topic”. If my claim in Kallulli (1999, 2000) that direct object clitics license topichood of the DPs they double is correct, we expect that the object of the verb ‘to 6

The situation is slightly different in Modern Greek, in that a doubling clitic is a sufficient, though not a necessary condition for marking the direct object as topic – see Agouraki (1993) and Kallulli (1999, 2000).

120 Dalina Kallulli have’ may not be clitic doubled in Albanian existential constructions. As the example in (13) shows, this prediction is borne out.7 (13) (*I) kishte minj në gjithë apartamentin. 3PL.CL.ACC had mice.ACC in all apartment.the ‘There were mice all over the apartment’ Thus, strictly speaking, clitic doubling is not an optional phenomenon; it produces information structure in a systematic way. If doubling clitics mark their doubled associates as [+topic]/ [+given], as shown in this section, and in view of the fact that clitic doubling triggers factivity, which is in turn defined in terms of presupposition, an important question that arises concerns the connection between givenness and presupposition. A crucial idea that I would like to highlight is that a proposition can shift from being contextually given to being presupposed. That is, to say that a sentence is presupposed can mean one of two things: either it is assumed to be true, or the proposition expressed by the sentence (“der Gedanke” in the sense of Frege) has been mentioned before.8 For instance, it seems that the correlate es ‘it’ in German is not satisfied with a situation in which the proposition is just given in context; it must also be true, as the example in (14) shows. (14) Context: Hans has certainly heard in his geography class that Sydney is not the capital of Australia and that Toronto is not the capital of Canada. Dennoch GLAUBT er (??es), dass Sydney die Hauptstadt still believes he it that Sydney the capital von Australien ist. of Australia is ‘Still, he believes that Sydney is the capital of Australia.’ But in spite of this, this distinction mostly seems to be blurred, in the sense that propositions that are presupposed (i.e., assumed to be true) are given – either in the immediate context, or via world knowledge – and that contextually given propositions are most often taken to be true. In view of this, the 7 8

This also holds for Modern Greek (Anagnostopoulou 1994). I thank Manfred Krifka (personal communication) for helping me articulate this idea in the present form and for providing the example in (14).

On the derivation of the relation between givenness and deaccentuation

121

difference between the factive and the non-factive uses of ‘believe’ across the languages discussed so far may be reasonably stated in terms of information structure. The question then is at what point in the derivation information structure is encoded. Depending on core assumptions about the architecture of grammar and how its various components interact, this question can be answered in various ways. However, under common minimalist assumptions, since the pronouns – which as I have shown trigger givenness and/or factivity – are lexical items with their own properties, the view that givenness and/or factivity (i.e., information structure) is encoded in the narrow syntax is hard to dismiss. With this interim conclusion in mind, I now turn to yet another set of data, whose relevance bears on the status of prosody as syntactically relevant. Specifically, I look at the prosodic realization of the factive and nonfactive versions of the example sentences from Albanian, English and German. This should provide a picture of whether there are clear correlates of factivity in the prosodic structure, and whether and how the three languages compared here differ in this respect.

2.2. Prosody A look at the prosodic realisation of the factive and the non-factive versions of the example sentences introduced in section 2.2 across Albanian, English and German reveals clear correlates of factivity in prosodic structure. Strikingly, as Fig.1 through Fig. 6 clearly show, the prosodic structuring and the prosodic differentiation of the sentences with a verb used once as factive and once as non-factive is comparable in all the three languages.9 Crucially, all the sentences with factive believe/glauben/besoj have a nuclear pitch accent on the matrix verb.10 In contrast, the nuclear pitch accent (i.e., stress) in the sentences with non-factive believe/glauben/besoj in all three languages is not on the matrix verb, but on the embedded one.

9

10

Note that the point with the prosodic structures given here is not to provide a detailed tonal analysis, but to show the essential pitch accents in the utterances. Here I have left out the prosodic structures of (3a,b), but these are also rather similar to the one in Fig. 2.

122 Dalina Kallulli 550

500 450 400 350

300 250 200

150 100 50

left.

John

that

believed

I

H*L

L

H

1.39175

0

Time (s)

Figure 1.

English: non-factive, e.g. (2) 550 500 450

400 350 300 250

200 150 100

50

H*

I

didn't

believe

L

it

that

John

left.

0

2.35002

Time (s)

Figure 2.

English: factive, e.g. (4) 450 400 350

300 250 200 150

100 50

L

Er

glaubte

L

dass

Peter

0

verstarb. 1.62031

Time (s)

Figure 3.

L*

German: non-factive, e.g. (5a)

On the derivation of the relation between givenness and deaccentuation 450 400

350 300 250 200

150 100 50

H*

Ich

glaubte

es

dass

Peter

verstarb.

0

1.9463 Time (s)

Figure 4.

German: factive, e.g. (5b) 550 500

450 400 350 300

250 200 150 100

50

H

Besova

H*L

se

Beni

shkoi.

0

1.57513 Time (s)

Figure 5.

Albanian: non-factive, e.g. (6a) 550

500 450 400 350

300 250 200 150

100 50

H*L

E

besova

se

Beni

0

1.72123 Time (s)

Figure 6.

shkoi.

Albanian: factive, e.g. (6b)

123

124 Dalina Kallulli Moreover, while correlative pronouns and/or modals are sufficient to trigger factivity in English and German, they are not necessary. Factivity in both languages can also be induced in the absence of these elements, provided that the matrix verb (here: believe/glauben/besoj) carries nuclear stress. To show this, consider first the contexts in (15) and (16), which were provided to the test subjects in order to elicit the factive reading of the verb in the underlined sentences in them. (15) I didn’t see John leave my party, but then he called me from his home phone. Now it was obvious. I believed that John left. (16) Ich gab bekannt (die Tatsache), dass Peter verstarb. Zuerst wollte Hans nichts davon wissen. Dann zeigte ich ihm die Todesanzeige, und nun sah er die Sache anders. Er glaubte, dass Peter verstarb. ‘I made known (the fact) that Peter died. At first Hans didn’t want to hear of it. Then I showed him the death certificate and now he saw the matter differently. He believed that Peter died.’ The prosodic structure of the underlined sentences in (15) and (16) is shown in Fig. 7 and Fig. 8, respectively. Again, the factive believe/glauben here clearly carry a nuclear pitch accent. In other words, the prosodic structures of the relevant sentences in (15) and (16), in which believe/glauben is factive, are rather similar to the prosodic structures in Fig.2 and Fig.4, respectively. This fact is of course not surprising, since correlative and/or pleonastic pronouns, like clitics, are phonologically light elements. It is obvious from the Figures 2, 4, 6, 7 and 8 that the (factive) embedded clauses are deaccented, and as is well-known, deaccenting is one means of expressing presupposition, or discourse binding – see for instance Krifka (2001) on the role of deaccenting in determining the restrictor of an adverbial quantifier. Of course accent on the verb can also come about because the verb is focused, or f-marked (Selkirk 1995). Crucially, however, deaccenting of the CP and focus on the verb do not exclude each other. In other words, the implication is only one way: in order to get a factive reading, the (factive) verb must carry nuclear pitch accent, but nuclear pitch accent on the verb does not entail factivity. More generally, there seems to be an issue of verum focus (i.e., focus on the verb, or VP) interacting with information structure even in the contexts that I have provided. However, these contexts still do a good job in at least not excluding factive readings.

On the derivation of the relation between givenness and deaccentuation

125

550 500

450 400 350 300

250 200 150 100

50

H*

I

L

believed

that

John

0

left. 1.44771

Time (s)

Figure 7.

English: factive, e.g. (15) 450 400 350 300 250 200 150 100 50

H*

Er

glaubte

dass

Peter

0

verstarb. 1.56081

Time (s)

Figure 8.

German: factive, e.g. (16)

2.3. Interim summary The data presented in this section show that the prosodic structuring and the prosodic differentiation of the sentences with comparable information structure (i.e., with a verb used once as factive and once as non-factive) is comparable in all three languages. Only in Albanian (and Modern Greek) there is a mechanism in the overt syntax that restricts the information structure and the prosodic structure, namely the clitic (pronoun). In English and German there is no pronoun always, but in view of factivity/non-factivity and prosodic structure correspondences, the relevant features are obviously manifested prosodically. Crucially, this is an overt manifestation. The question then is whether and

126 Dalina Kallulli how the correlation between deaccenting and factivity described here should be derived (i.e., modelled syntactically). I turn to this question in the following section.

3. A “best case” model The systematicity of the PF/LF correlation with respect to the phenomenon of (induced) factivity that was described in section 2 is best captured by conceiving of prosodic information as encoded in the syntax, or as part of the numeration itself, which is what syntax manipulates. Thus, adopting a realizational framework of morphology, prosodic information (in the case at hand, deaccentuation), may be viewed as instantiating (or realizing) an abstract morpheme (Embick and Noyer 2007). In this sense, (the relevant chunk of) prosody has the status of a morpheme. This conception enables us to derive the correlation between factivity and deaccentuation in terms of Chomsky (2004: 107): [T]he best case is that there is a single cycle only […] and the […] components of the derivation of proceed cyclically in parallel. L contains operations that transfer each unit to Φ and to Σ. At the best case, these apply at the same stage of the cycle. Assume so. (Chomsky 2004: 107)

I contend that the status of an embedded clause as [+presupposed] must be expressed. This is achieved syntactically by way of functional structure. Specifically, the head hosting the relevant syntactic feature (abstract morpheme), which is realized overtly by some expletive-like element (such as a clitic, a pronoun, a modal element, etc.), or simply by deaccentuation, is a probe for the goal (here: CP) with an OCC feature. In the remainder of this section, I first introduce the essentials of Chomsky’s (2000) probe-goal mechanics, and then provide a structural implementation of factivity in terms of ‘Agree’ in the probe-goal relation.

3.1. Agree and the probe-goal relation According to Chomsky (2000, 2001), a probe is a set of uninterpretable φfeatures that are valued and deleted by establishing an Agree relation with a goal containing matching interpretable φ -features. Further, Agree is constrained by standard locality conditions, as in (17).

On the derivation of the relation between givenness and deaccentuation

127

(17) Locality Condition (Chomsky 2000) Agree holds between P and G just in case G is the closest set of features in the domain D(P) of P that match those of P. The domain D(P) of P is the sister of D, and G is closest to P if there is no G’ matching P such that G is in D(G’). The category with the probe-feature can also have an EPP- or an OCC-feature. OCC in turn is the condition that an XP is an occurrence of a probe and licenses information structure. For Chomsky (2001), the OCC-feature of a head gets saturated by moving the required category in the domain of H(ead). Crucially, however, Collins (1997) and Bowers (2002) argue that the OCC-feature of a head can in addition get saturated by merging an expletive with H. Building on this idea, I now turn to the implementation of clitic doubling as ‘Agree’ in the probe-goal relation.

3.2. Clitic doubling as ‘Agree’ in the probe-goal relation Adopting the structural configuration put forth in Sportiche (1996), whereby clitics head their own maxiaml projections, I submit that a clitic is a probe for the goal (i.e., DP or CP) with an OCC-feature, as shown in (18).11 (18)

Further, I submit that (clitic) doubling is a universal strategy, as formulated in (19). 11

The OCC-feature might be further specified as [+Topic] and/or [+Deaccentuate].

128 Dalina Kallulli (19) The (Clitic) Doubling Principle – A Universal Strategy Cl0 must be filled by information on prosodic realization (where ‘prosodic realization’ means ‘deaccent (goal)’). Thus, while some languages (e.g. Albanian) need an overt expletive-like element (such as a clitic), in order to obey this principle, others (e.g. English, German) do not need such an element. This difference can be captured in terms of the two strategies by which the OCC feature can get saturated, as mentioned earlier. Turning to the question of why (clitic) pronouns are some of the seemingly preferred means that languages systematically choose to trigger factivity, I assume – in line with previous work (Kallulli 2000, 2001) – that a (definite) pronoun naturally has the characteristic [+topic]/[+given]. Therefore, it can mark a constituent as such. The question arises however what the position of modal verbs – which as I showed in section 2 may also induce factivity – is. In particular, do modals also occupy Cl0 in (18)? While this might indeed be the case, whether or not modals occupy Cl0 or some other position is a matter of secondary importance. What is important is the by now obvious generalization that factives involve more structure than non-factives, as initially argued for by Kiparsky and Kiparsky (1970), and contra de Cuba (2006). Also important is the fact that this dependency can be captured syntactically through the same operation, namely ‘Agree’ in the probe-goal relation. However, further scrutiny notwithstanding since doubling (clitic) pronouns may co-occur with modal elements, it seems reasonable to state that these elements may occupy different positions in the structure of the clause. Yet, as an investigation of the interaction between modal elements and clitics is beyond the scope of this paper, nothing conclusive can be stated on this issue at present.

3.3. Language comparison – the locus of parametric variation Unlike in English and German, in Albanian and other languages givenness of object DPs is achieved through doubling clitics, which in turn entail deaccenting of their doubled associate. I suggest that this state of affairs is due to the fact that, while English and German allow free deaccenting, Albanian does not. Nonetheless, however, English and German are similar to Albanian with respect to ‘doubling’ of object CPs. As was discussed in section 2, both English and German display the ‘clitic’ strategy – recall the use of the pleonastic it and correlate es, respectively. This is potentially due to

On the derivation of the relation between givenness and deaccentuation

129

greater sentence length, or heavier phonological weight – see also Féry and Samek-Lodovici (2006), who have independently argued that pitch accents are also related to phonological weight and not necessarily to f-marking.

4. Extensions Incidentally, one of the central claims of this paper, namely that the [+presupposed] status of the embedded clause must be expressed via deaccenting and/or a doubling clitic / (correlative) pronoun is reminiscent of the use of definite articles in Spanish and other languages to mark what in English is expressed by deaccentuation, as in (20) vs. (21). (20) a. Los vaqueros mascan tabaco. (Laca 1990) the cowboys chew tobacco ‘Cowboys usually chew tobacco.’ b. Los vaqueros mascan el tabaco. the cowboys chew the tobacco ‘What cowboys usually do with tobacco is: they chew it.’ (21) a. Cowboys CHEW tobacco. b. Cowboys chew TOBACCO. Extending the structural implementation that I have proposed for the correlation between givenness and deaccentuation also to this (language-specific) instantiation of the same phenomenon is a trivial matter; basically, the determiner plays the role of the (clitic) pronoun and/or modal element discussed in section 2.12

5. Conclusion The main conclusion reached in this paper is that information structure is encoded in (core) syntax. I have presented novel evidence for a truly syntactic treatment of certain traditionally so-called PF phenomena, such as deaccentuation of so-called ‘given’ discourse material. In particular, I have 12

Note in this context also the well-known morphological affinity between clitics (and more generally, pronouns) on the one hand, and determiners, on the other (Postal 1969 and subsequent literature).

130 Dalina Kallulli argued that at least some prosodic information is encoded in the narrow syntax. Though the ramifications of such a view are far-reaching, it is worth noting that its implementation is perfectly in tune with other basic tenets of the minimalist agenda, such as lack of optionality and the non-creationist nature of syntax. Finally, the view that prosodic information is encoded in the narrow syntax opens up an entirely new avenue in approaching operations that have increasingly been argued to have an effect on information structure, or be motivated by information structure considerations, such as Germanic scrambling, so-called clitic left dislocation constructions, and others.

Acknowledgements I am grateful to Kleanthes Grohmann, Manfred Krifka, Georg Niklfeld and an anonymous reviewer for detailed comments. Thanks are also due to my Spring semester 2006 students, and the audiences of WCCFL 25 and InterPhases for their questions. Research for this paper was funded by the Hertha Firnberg fellowship T173-G03 from the Austrian Science Fund (FWF).

References Agouraki, Yoria 1993 Spec-Head Licensing: The Scope of the Theory. Doctoral dissertation, University College London. Anagnostopoulou, Elena 1994 Clitic Dependencies in Modern Greek. Doctoral dissertation, University of Salzburg. Bobaljik, Jonathan 2002 A-chains at the PF-interface: Copies and “covert” movement. Natural Language and Linguistic Theory 20 (2): 197–267. Bowers, John 2002 Transitivity. Linguistic Inquiry 33: 183–224. Chomsky, Noam 2000 Minimalist inquiries: The framework. In Step by Step: Essays on Minimalist Syntax in Honor of Howard Lasnik, R. Martin, D. Michaels and J. Uriagareka (eds.), 89–155. Cambridge, MA: MIT Press. 2004 Beyond explanatory adequacy. In A. Belletti (ed.) Structures and Beyond 104–131. Oxford: Oxford University Press. Collins, Chris 1997 Local Economy. Cambridge, MA: MIT Press.

On the derivation of the relation between givenness and deaccentuation

131

de Cuba, Carlos 2006 The adjunction prohibition and extraction from non-factive CPs. In Proceedings of the 25th West Coast Conference on Formal Linguistics, D. Baumer, D. Montero, and M. Scanlon (eds.), 123 –131. Somerville, MA: Cascadilla Proceedings Project. Embick, David and Rolf Noyer 2007 Distributed morphology and the syntax/morphology interface. In The Oxford Handbook of Linguistic Interfaces, G. Ramchand and C. Reis (eds.). Oxford: Oxford University Press. Féry, Caroline and Vieri Samek-Lodovici 2006 Focus projection and prosodic prominence in nested foci. Language 82 (1): 131–150. Frampton, John and Sam Gutman 1999 Cyclic computation. Syntax 2 (1): 1–27 Franks, Steven and Tracy H. King 2000 A Handbook of Slavic Clitics. New York: Oxford University Press. Halliday, Michael 1967 Notes on transitivity and theme in English. Part 1 and 2. Journal of Linguistics 3: 37–81 /199–244. Holmberg, Anders 2000 Scandinavian stylistic fronting: how any category can become an expletive. Linguistic Inquiry 31: 445–484. Holmberg, Anders and Thorbjörg Hróarsdóttir 2003 Agreement and movement in Icelandic raising constructions. Lingua 113: 997–1019. Kallulli, Dalina 1999 The comparative syntax of Albanian: On the contribution of syntactic types to propositional interpretation. Doctoral dissertation, University of Durham. 2000 Direct object clitic doubling in Albanian and Greek. In Clitic Phenomena in European Languages, F. Beukema and M. den Dikken (eds.), 209–248. Amsterdam: Benjamins. 2001 Direct object clitic doubling in Albanian and Greek. In Comparative Syntax of the Balkan Languages, A. Ralli and M. Rivero (eds.), 127– 160. Oxford: Oxford University Press. Kiparsky, Paul and Carol Kiparsky 1970 Fact. In Progress in Linguistics, M. Bierwisch and K. Heidolph (eds.). The Hague: Mouton. Krifka, Manfred 2001 Non-novel indefinites in adverbial quantification. In Logical Perspectives on Language and Information, C. Condoravdi and G. Renardel de Lavalette (eds.) 1– 40. Stanford, CA: CSLI Publications.

132 Dalina Kallulli Laca, Brenda 1990 Generic objects: Some more pieces of the puzzle. Lingua 81: 25–46. Ladd, D. Robert 1996 Intonational Phonology. Cambridge: Cambridge University Press. Landau, Idan 2007 EPP extensions. Linguistic Inquiry 38: 485–523. 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. Ndayiragije, Juvénal 2000 Strengthening PF. Linguistic Inquiry 31(3): 485–512. Postal, Paul 1969 On so-called “pronouns” in English. In Modern Studies in English, David A. Reibel and Sanford A. Schane (eds.), 201–244. Englewoods Cliffs, NJ: Prentice Hall. Reinhart, Tanya 1995 Interface Strategies. OTS Working Papers in Linguistics. Schwarzschild, Roger 1999 GIVENness, AVOIDF and other constraints on the placement of accent. Natural Language Semantics 7: 141–177. Selkirk, Elisabeth 1995 Sentence prosody: Intonation, stress and phrasing. In Handbook of Phonological Theory, John Goldsmith (ed.), 550 –569. Cambridge: Blackwell. Sportiche, Dominique 2006 Clitic constructions. In Phrase Structure and the Lexicon, Laurie Zaring and Johan Rooryck (eds.), 213–276. Dordrecht: Kluwer. Taglicht, Josef 1982 Intonation and the assessment of information. Journal of Linguistics 18 (2): 213–230.

Phase theory, linearization and zig-zag movement Carlo Geraci

The aim of this paper is to show how the grammar of human languages can capture the fact that certain elements move from one edge of the vP to the opposite edge of the CP, instantiating what I will call zig-zag movement. The theory of phases developed by Chomsky (2000) and subsequent works, and the theory of successive cyclic linearization proposed by Fox and Pesetsky (2005) interact with two general and plausible restrictions on overt movement: the ban on movement from the complement of a head to the specifier of the same projection (Abels 2003), and the ban on movement from the specifier of a projection to the specifier of the same projection (Ko 2005). When combined together, these three ingredients generate a series of restrictions that are at odds with empirical data from Italian Sign Language (LIS) and other signed and spoken languages. In particular, zig-zag movement is predicted to be ungrammatical, because it would produce a conflicting ordering among the lexical elements, which in turn would result in the crash of the derivation. The solution I propose to this puzzle will refine the definition of movement (internal merge).

1. Introduction In the current theory of phases (see Chomsky 2000 and subsequent works), syntactic structures are created by several application of merge and re-merge operations. To avoid working memory overload there are specific steps in the syntactic derivation of a sentence, where the information about the syntactic structure is sent to the interfaces (PF and LF interfaces); the operation that sends syntactic structures to the interfaces is named spell-out, and applies every time a vP, a CP and (possibly) a DP level has completed its operations. According to the theory of successive cyclic linearization (CL) developed by Fox and Pesetsky (2005, henceforth F& P), the structural information carried by the syntactic objects at the vP and CP levels is mapped into a linear order at the PF interface. Within this picture, two general and plausible restrictions on overt movement (internal merge) have been proposed in the recent literature: the first restriction bans the movement of the complement of a head to the specifier of that projection (Abels 2003). The

134 Carlo Geraci second restriction bans movements from a specifier position to an outer specifier in the same projection (Ko 2005). When combined with the theory of cyclic linearization proposed in F&P, a framework emerges that is at odds with empirical data from Italian Sign Language (LIS)1 and other signed and spoken languages. The puzzle is represented by movements from the edge of a phase to the opposite edge of the next phase. I will call these movements zig-zag movements. In order to provide a solution to the puzzle raised by zig-zag movement, it is necessary to spell out in detail how movement is derived. In particular, three aspects of the complex operation of movement are under scrutiny: the operations of Agree and pied piping, and the EPP feature, which is ultimately responsible for re-merge. The paper is composed as follows: I will briefly introduce the operations under investigation, the two restrictions on movement and the theory of cyclic linearization, in section 2. Since most of the data come from LIS, I will present the general properties of LIS that are relevant for the discussion, in section 3. Section 4 will be devoted to show that zig-zag movements are part of the grammar of both spoken and signed languages and that these movements are problematic for the theory of CL. I will provide a solution to the problem in section 5. Finally, section 6 will conclude the paper.

2. Phase Theory, Movement and Linearization In the phase theory developed by Chomsky (2000) and following works, syntactic structures are built by several application of merge, the syntactic operation that recursively takes two elements from the numeration (the set of syntactic objects that will constitute the sentence) and builds a two member set. According to Chomsky (2005), merge is driven by a feature hosted in the head of one of the two syntactic objects undergoing merge: the edge feature. It is to satisfy the requirement of the edge feature of the verb that merge applies to the verb and its complement, creating a set of two members, as shown in (1): (1)

1

2

Merge applies between the verb kiss and the DP the frog2 Merge (kiss, {the frog})  {kiss, {the frog}}

LIS is the sign language used by Italian Deaf Community and by Deaf Community of Ticino, a region in the south of Switzerland. The DP the frog would be better represented as the result of merge between the determiner the and the NP frog. Unless directly relevant for the discussion, I will not consider the internal structure of the DP.

Phase theory, linearization and zig-zag movement

135

However, merge is not limited to elements that are in the numeration. It is possible, under specific conditions, that merge applies also between the head of a syntactic object and an element that has already undergone merge. This operation of re-merge, also called internal merge, replace what in the generative tradition was known as movement. For the sake of clarity, I will use the terms internal merge and movement as synonymous. According to the most recent development of minimalism, movement is seen as a complex operation made up by three independent operations: agree, pied-piping and (re)merge of an element already introduced in the derivation. The operation of (re)merge ultimately results in the satisfaction of the EPP feature of the head that hosts the moved element in its specifier.3 The basic idea is that an unvalued feature of a head acts as a probe searching its c-command domain for a corresponding valued feature (the goal). Once the probe finds its goal, agreement is established between the probe and the goal (the result of the agree operation) and the two features share the same value. The operation of pied-piping selects the goal as eligible for (re)merge, and the presence of an EPP feature on the probe requires that the goal (re)merges to the specifier of the probe, as shown in (2): (2)

a. The probe agrees with the goal and pied piping selects the goal for re-merge. IP vP probe will EPP uφ[ ] goal the princess iφ[1] agree v0

vP VP kiss

3

the frog

In a sense, the role of EPP in the operation of movement is not crucially distinct from the role played by the edge feature with external merge.

136 Carlo Geraci b. The EPP feature of the probe requires the goal to be remerged in its specifier. IP IP re-merge

probe will uφ[1] EPP

vP

vP goal the princess iφ[1]

v0

VP kiss

the frog

c. Internal merge of the subject from the specifier of vP to the specifier of IP Merge({the princess}, {will, {{the princess}, {kiss, {the frog}}})  {{the princess}, {will, {{the princess}, {kiss, {the frog}}}} d. the princess will kiss the frog The definition of agree I will adopt here is the one proposed by Pesetsky and Torrego (2004): (3)

Agree (proposed by Pesetsky and Torrego) a. An unvalued feature F (a probe) on a head H at syntactic location α (Fα) scans its c-command domain for another instance of F (a goal) at location β (Fβ) with which to agree. b. Replace Fα with Fβ, so that the same feature is present in both locations.

According to their proposal, lexical features display two properties: the value, indicated within square brackets and the interpretability, indicated with i and u to the left of the feature name. The combination of these two properties yields the following range of possibilities:

Phase theory, linearization and zig-zag movement

(4)

137

Types of features – interpretable and valued, like the φ-features on nouns (iφ [1]); – uninterpretable and unvalued, like the φ -features on the auxiliary (uφ[ ]); – interpretable and unvalued like the Q feature on C (iQ[ ] ); – uninterpretable and valued, like the Q feature on wh-elements (uQ[2])

As a consequence of the definitions of movement and agree just sketched, an element that merges to the complement of a head cannot move to the specifier position of the same head. In fact, a constituent may only be merged (or remerged) if the operation leads to the immediate satisfaction of a previously unsatisfiable feature (agree must precede re-merge). However, as clearly shown by Abels (2003), moving a constituent from the complement position of a head to its specifier does not satisfy any previously unsatisfied feature. Hence, the movement of the complement of a head cannot target the specifier of that head as a possible landing site: (5)

1st Restriction on movement:

(Abels 2003)

[XP YP X tYP]

XP

XP X

YP

A complement cannot move to the edge of the projection that selects it

YP

X X

YP

More recently, another restriction on movement was pointed out by Ko (2005). She observed that an element that merges to the specifier of a head cannot move further up within the same projection, because the specifier position is not in the search domain of the head-probe (the search domain of a head-probe is its sister):

138 Carlo Geraci (6)

2nd Restriction on movement:

(Ko 2005)

[XP YP ZP tYP X KP]

no movement from a specifier to an outer specifier of the same projection

XP YP

XP YP

XP KP

X X

XP YP

ZP

X X

ZP

These two restrictions interact with the theory of successive-cyclic linearization proposed by F&P. CL is a theory of the mapping between syntax and PF. It states that, at the end of each phase, the overt elements in the syntactic object created by several applications of merge are ordered and linearized. Movement of an element may revise the order within a phase, namely before the phase is sent to spell-out, but not across phases. Therefore, movement from either edge and non-edge positions of a relevant spell-out domain is possible only if previously established orders are not disrupted. Elements that show cyclic movement must target the edge of the phase before the derivation is sent to spell-out. To have an idea of how elements are linearized phase by phase, consider the example in (7), and its derivation along the line of CL: 4 (7)

a. to whom did Mary give a book b. [vP to whom Mary give a book __ ]]]]

4

Following Chomsky (2000), I consider vP and CP to be the relevant spell out domains, at least for English.

139

Phase theory, linearization and zig-zag movement

Once the vP phase has completed, the syntactic object is sent to PF. Its elements are linearized in a table of orders that states the linear order among the elements of the phase. Since the wh- phrase has moved to the specifier of the vP, it will be linearized before the subject, the verb and its direct object: (7)

c. Table of orders: to whom < Mary, Mary < give, give < a book

After the CP is built and to whom has moved to the specifier of the CP, spell-out applies once again, producing the orders in (7e): (7)

d. [CP to whom did Mary [vP __ tMary give a book __ ]]]] e. Table of orders: to whom < did, did < Mary, Mary < give

Since the table of orders of CP does not contradict the orders established at the previous phase, the derivation ends without any trouble. If any of the order in (7c) were contradicted by one of the orders in (7e), the derivation would have crashed. CL together with the two restrictions on movement in (5) and (6) predicts movement from one edge of a phase to the opposite edge of the next phase to generate order contradictions, leading the derivation to crash. (8)

Movement predicted to be impossible according to CL a. BOOK BUY WHO ‘who bought the book’

(LIS)

b. [CP [IP ___ [vP __ BOOK BUY]]] WHO]

c. Table of orders at vP: WHO < BOOK, BOOK < BUY d. Table of orders at CP: BOOK < BUY, BUY < WHO The transitive closure produced by the table of order generated at the level of vP (WHO < BUY) is contradicted by the table of orders generated once CP is spelled out. Movements like the one in (8a) are instances of zig-zag movements.

140 Carlo Geraci (9)

Definition of zig-zag movements Movement of an element from the edge of a phase to the opposite edge of the next phase.

Before going on through the details of zig-zag movement, I will present some general properties of LIS. From these properties it will emerge that the movement presented in (8) is a genuine example of zig-zag movement.

3. General properties of LIS The data from LIS come from four native signers (Deaf people with Deaf parents),5 who regularly cooperate as informants with a broad research project whose purpose is to arrive at a detailed description of the properties of LIS.6 Signers’ production has been video recorded and the videos glossed with the SignStream software.7 This section aims to provide the reader with some general properties of the variety of LIS signed by our informants. The glosses for LIS sentences follow the standard convention of using capitalized words for signs. LIS basic word order is SOV, as sentence in (10) shows: (10) GIANNI MARIA LOVE ‘Gianni loves Maria’ Lexical elements that plausibly sit in the head of functional projections, such as modals or the aspectual marker DONE,8 appear in post-verbal position, as shown in (11–12). Adverbs like on time are found in post-verbal position, as shown in (13): (11) GIANNI METER 80 JUMP CAN ‘Gianni can jump 1.80 mt.’ 5

6

7

8

I will follow the general use of the upper case to indicate those deaf people who are members of the Deaf community and who use their signed language as the primary system of communication. Information about the project is available at: http://www.filosofia.unimi.it/~zucchi/ricerca.html. SignSream is a software specifically developed to the purposes of sign language research. Information about the software is available from: www.bu.edu/asllrp/SignStream (see also Neidle et al. 2001). See Zucchi (2003) for more exhaustive data and discussion on this marker.

Phase theory, linearization and zig-zag movement

141

(12) GIANNI HOUSE BUY DONE ‘Gianni bought a house’ (13) GIANNI ARRIVE ON-TIME ‘Gianni arrived on time’ Also negation and n-words (negative words like nobody and nothing) appear postverbally, as sentences in (14 –15) show: (14) GIANNI CONTRACT SIGN NOT ‘Gianni did not sign the contract’ (15) a. CONTRACT SIGN NOBODY ‘Nobody signed the contract’ b. GIANNI SIGN NOTHING ‘Gianni signed nothing’ An interesting property that LIS shares with many other sign languages, American Sign Language (ASL)9 and Indo-Pakistani Sign Language (IPSL) to mention two of them, is the position of wh-phrases. When they move overtly, wh-elements must appear at the right periphery of the sentence, as (16–17) show: (16) GIANNI BUY WHAT ‘What did Gianni buy?’ (17) HOUSE BUY WHO ‘Who bought a house?’ Putting all these things together, I take LIS to be a head-final language: the verb follows the object and the functional heads that host modals and the aspectual marker, DONE, both follow the main verb (see Cecchetto, Geraci and Zucchi (2006) for a more complete presentation of data supporting the hypothesis that LIS is SOV). A property of LIS that deserves careful attention for the purpose of this paper is the position of wh-phrases and n-words: in the examples considered here, they are verb arguments and appear postverbally, as shown in 9

Data from ASL are controversial, see Petronio and Lillo-Martin (1997) and Neidle et al. (2000) for two different perspectives.

142 Carlo Geraci (15–17). When both a wh-element and an n-word are in the same clause, however, the wh- element must rigidly follow the n-word, as in the examples in (18 –19): (18) a. BUY NOTHING WHO b. *BUY WHO NOTHING ‘who bought nothing?’ (19) a. BUY NOBODY WHAT b. *BUY WHAT NOBODY ‘what did nobody buy?’ The pattern of wh-elements is even more intriguing when the wh-phrase includes also a restriction, as in (20 –21): 10, 11 (20) a. STUDENT BUY BOOK WHICH b. STUDENT BOOK BUY WHICH c. STUDENT BOOK WHICH BUY ‘Which book did a student buy’ (21) a. BOOK BUY STUDENT WHO b. STUDENT BOOK BUY WHO c. STUDENT WHO BOOK BUY ‘Which student bought a book?’ The options displayed by LIS allow the wh-phrase either to be on the right periphery of the sentence, or to be splitted (the restriction in situ and the wh-element in the right periphery), or the whole wh-phrase in situ. These facts strongly suggest that wh-elements are structurally higher than n-words and support Cecchetto and Zucchi (2004) who argue that wh10

11

The picture needs to be further enriched with the non manual markings, namely special facial expressions that co-occur with wh-elements (eyebrow lowering) and n-words (headshake) and that may spread over specific syntactic domains. Given that the distribution of the non manual markings is not strictly relevant for the purpose of this talk, I will omit them (but see Cecchetto Geraci and Zucchi (in prep.) for an analysis of the spreading of non manual markings and the problems it raises for the theory of linearization proposed by Kayne 1994). The examples in (20a–b) and (21a–b) are equally fine and, apparently, do not display any difference in meaning. The examples in (20c) and (21c) can be used only in strict d-linked contexts, as observed in Cecchetto and Zucchi (2004).

Phase theory, linearization and zig-zag movement

143

phrases are in the specifier of CP, which is located to the right of C.12 To be more explicit, I assume in this paper the analysis proposed in Cecchetto and Zucchi (2004) that wh-elements in LIS instantiate rightward movement. The same analysis extends to n-words: following Geraci (2006a, 2006b), nwords move from their argument position to the specifier of a negative projection, which is to the right of the negative head.13 Taken together the data presented here are compatible with the structure in (22): (22) Structure for LIS sentences CP 3 CP wh-elements 3 NegP C0 3 NegP n-words 3 IP Neg0 3 Spec,IP IP 3 AspP MODALS 3 vP DONE 3 SUBJECT vP 3 VP v0 3 OBJECT

12

13

VERB

The same proposal for ASL has been previously made in the seminal work by Neidle et al. (2000). However, for a different proposal for ASL see Wilbur and Patschke (1999) and Petronio and Lillo Martin (1997). Sign languages make also a great use of facial expressions to vehicle syntactic, semantic and prosodic contents. For instance, wh-features are indicated by lowered eyebrows and negation by head shake, in LIS. These non manual markers generally co-occur with wh- and negative manual markers, but they may spread over specific syntactic domains. I will not deal with a detailed analysis of the Negative and wh- NMMs. Their interesting patterns are investigated in Cecchetto and Zucchi (2004), Geraci (2006a, 2006b), and Cecchetto, Geraci and Zucchi (in prep.).

144 Carlo Geraci 4. Zigzag movements and CL: the problem According to the Phase Impenetrability Condition (PIC), first proposed in Chomsky (2000), at the end of each phase, those elements that still bear uninterpretable, unvalued features must target the edge of the phase in order to be “syntactically visible” for further operations. Although, F& P do not discuss the PIC in detail, they assume that the elements that move cyclically must target the edge of each phase. Following F& P, I will assume there are elements that cyclically target the specifier of the phase they are escaping, such as wh-elements, and more generally, those elements that exhibit cyclic A-bar movement, including movement of n-words to the specifier of the negative phrase. However, these movements raise legitimate questions about what features are involved in the intermediate steps of the derivation. I will address the question in the next section. As shown in section 3, LIS exhibits overt movement of n-words and whelements from their argument position in preverbal position to the right periphery of the sentence. Since these movements affect the order of elements within vP and CP, one might wonder whether order contradictions arise.14 Let’s consider first the case of bare wh-elements in subject position: (23) a. BOOK BUY WHO ‘who bought the book?’ According to the structure in (22) above, the wh-element in (23a) starts from the preverbal subject position and lands in the specifier of the CP node, after having targeted the specifier of the IP node: (23) b. [CP [IP tWHO [vP tWHO [VP BOOK BUY]]] WHO]

c. Table of orders at vP: WHO < BOOK, BOOK < BUY d. Table of orders at CP: BOOK < BUY, BUY < WHO The kind of movement instantiated in (23a) and represented in (23b) is an example of zig-zag movement, as defined in (9), above: an element moves from the left edge of the vP to the right edge of the CP. This movement 14

I will develop the argument using examples of wh-constructions, but the same pattern can be obtained with n-words.

Phase theory, linearization and zig-zag movement

145

generates an order contradiction between the order stated at the vP level and the one established at the CP level, because the transitive closure of the ordering at vP predicts the interrogative pronoun WHO to precede the verb BUY, while the order at CP states that BUY precedes WHO. On the base of the example in (23a), one might try to accommodate the derivation in two ways: – assuming some movement of the subject within the vP; – arguing that in the case of wh-questions in LIS, the wh-element in sub-

ject position starts from the specifier of vP on the right and target the specifier of IP and then the specifier of CP both on the right, instantiating a two step rightward movement. I will disprove the two claims in turn. Given that the subject is generated in the specifier of vP, one might assume some movement to an outer specifier of the vP on the right, as in (24): (24)

vP 3

vP WHO 3 tWHO vP 3 VP v 3 BOOK V g BUY

However, this sort of movement is banned because the inner specifier of the vP is not in the search domain of the head of the phase, as stated in the restriction on overt movement presented in (6) above. According to the second option, the wh-element, starting from spec,vP on right, would produce a sequence of two rightward movements, as in (25a): (25) a. [CP [IP [vP [VP BOOK BUY] t WHO] t WHO] WHO] b. Table of orders at vP: BOOK < BUY, BUY < WHO c. Table of orders at CP: BOOK < BUY, BUY < WHO The table of orders of vP and CP will not raise an order contradiction. However, the derivation sketched in (25) is not a possible derivation for two rea-

146 Carlo Geraci sons, one conceptual the other empirical. The conceptual argument against the derivation in (25) is that it is not clear what is responsible for having the specifier of vP and IP on the right, given that non wh-subjects are found in preverbal position. Assuming that there is something special in the whfeature of LIS wouldn’t make the things better, first because there are other sign languages where the wh-elements exhibits the same pattern (see below), and second because also n-words in LIS show rightward movement. The empirical argument comes from sentences with complex wh-phrases, like the one in (26): (26) a. BOOK BUY STUDENT WHO b. STUDENT BOOK BUY WHO c. STUDENT WHO BOOK BUY ‘Which student bought a book?’ The range of possibilities displayed by LIS shows that d-linked wh-phrases may pied pipe (preferred option) the whole phrase to spec,CP (26a), or strand the restriction, plausibly in spec,IP (26b), or remain in situ (26c). These possibilities show that the specifiers of vP and IP are on the left also with wh-subjects, much like with declarative sentences. As in the case of (23a), also the sentences in (26a–b) are predicted to be ungrammatical, because they would yield order contradiction, as shown in the derivation for (26b), sketched in (27): (27) a. BOY BOOK BUY WHO ‘which boy bought a book’ b. build vP [vP BOY WHO [VP BOOK BUY]] Table of orders will contain: BOY < WHO, WHO < BOOK, BOOK
Phase theory, linearization and zig-zag movement

147

is that if we want to preserve F& P’s approach to linearization, then vP cannot be a phase in LIS. Following what F& P proposed for Mainland Scandinavian languages, we may take VP and not vP to be a phase in LIS. This move would save sentences with wh-elements and n-words in subject position. However, also with this amendment, problems arise for the Theory of Cyclic Lineazation, when wh-phrases and n-words in object position are considered: (28) JOHN BUY WHAT ‘what did John buy?’ The derivation would proceed along the lines sketched in (29): (29) a. build VP [VP WHAT BUY] b. move the wh-element to the edge of VP *[VP tWHAT BUY] WHAT VP 3 VP WHAT 3 WHAT V g BUY

Cyclic movement of the wh-element requires WHAT to move to the edge of VP as in (29b). This movement violates the first restriction on movement introduced above: moving the complement of a head to the specifier position of the same head is not allowed (Abels 2003). This fact would force us to go on with the alternative in (29b′ = 29a): (29) b′. [VP WHAT BUY] Table of orders at vP: WHAT < BUY c. build CP and remerge what needs to be remerged [CP [TP JOHN [vP tNP [VP tWHAT BUY]]] WHAT] Table of orders will contain: Previously ordered elements WHAT < BUY New orders JOHN < BUY, BUY < WHAT

148 Carlo Geraci The table of orders at the CP level contradicts the table of orders previously established at VP, predicting the sentence in (28) to be ungrammatical, contrary to facts. It is worth noticing that the problem is more general: assuming VP as the relevant phase raises problems also for VO languages with overt whmovement. The relevant example is given in (30): (30) a. what did John buy b. [VP what [VP buy what] In order for the wh-element to target the left periphery at the CP level, it must target the VP node instantiating a case of movement from the complement position of V to the specifier projection of the VP, violating the first restriction on movement.15 To sum up, the combination of the three ingredients (phase theory, CL and the two restrictions on movements) predicts zig-zag movement, instan15

In principle, one could claim that this fact is a good reason to abandon the first restriction on movement, while maintaining VP as the relevant phase. However, this move does not produce much benefit. Consider the case of a question with multiple wh-phrases as in (i): (i) a. who bought what b. [VP what [bought what] At the VP level, only the wh-complement has been merged in the derivation. At that stage, there is no chance to let the wh-phrase in situ without looking ahead in the derivation, where another wh-phrase is to come. Thus, the table of orders would require what to precede bought in (ib), and this ordering has to be maintained throughout the derivation, ruling out (ia), contrary to facts. Related to this is the contrast in (ii–iii) discussed in Frazier and Clifton (2002), and brought to my attention by an anonymous reviewer: (ii) a. Mary asked [whoi [ei read what] ]? b. *Mary asked [whati [who read ei] ]? (iii) a. Mary asked which mani [ei read which book]? b. Mary asked which booki [which man read ei]? Under the CL approach, the derivations for (ii) and (iii) exhibit the same problem as (i) shows. According to the proposal I will develop in section 5, the derivations for (ii) and (iii) does not raise order contradictions, since at the vP phase no order is established. This means that the ungrammaticality of (iib), and the contrast between (iib) and (iiib) has to be explained on other grounds, as suggested in Frazier and Clifton (2002). For an extension of my proposal to languages that do not display superiority effects (such as Slavic languages) see Geraci (2006b).

Phase theory, linearization and zig-zag movement

149

tiated by wh-sentences in LIS to be ungrammatical. Furthermore, it seems that neither vP nor VP are possible phases in LIS. Before proposing a solution to this puzzle, I will devote the rest of this section to show that zig-zag movement is not limited to wh-sentences in LIS but that it is found also with n-words in LIS and wh-sentences of other signed and spoken languages. N-words exhibit the same pattern shown by wh-elements in LIS. Although they are arguments of the verb, they do not appear in the canonical argument position, but are displaced in postverbal position, instantiating movement from the left edge of the vP phase, to the right edge of the CP phase: (31) a. tNOBODY BOOK BUY NOBODY ‘nobody bought a book’ b. JOHN tNOTHING BUY NOTHING ‘John didn’t buy anything’ (32) zig-zag movement of n-words in LIS a. [NegP [IP ___ [vP ___ [VP BOOK BUY] ] ]NOBODY ]

Other cases of zig-zag movement involve wh-movement in American Sign Language (ASL; but see footnote 11) and Indo-Pakistani Sign Language (IPSL):16 (33) a. JOHN BUY BOOK ‘John bought a book’ b. BUY BOOK WHO ‘who bought a book?’

(Aarons 1994)

(34) a. FATHER INDEX BOOK SEARCH ‘Father is/was searching for a book.’ (Pfau and Zeshan 2004) b. FATHER INDEX SEARCH G-WH ‘What is/was father searching?’ 16

ASL displays several possibilities for wh-elements, most of them correlated with a specific spreading of the wh-NMM (see Neidle et al. 2000). As in the case of LIS, I will not consider here the spreading of the wh-NMM in ASL and IPSL. What is relevant here is that the wh-phrase in preverbal position at the vP level has moved to a postverbal position at the CP level.

150 Carlo Geraci Sentences in (33a) and (34a) show the basic word order in ASL and IPSL, while the sentences in (33b) and (34b) show the dislocation of the whsubject in ASL and of the wh-object in IPSL to the right periphery of the sentence, respectively. In both instances, the wh-element moves from a preverbal position at the vP level to a postverbal one at the CP level. Let’s now turn to the case of zig-zag movement in spoken languages, and consider the English sentence in (35): (35) ?? What will you wear tomorrow that I bought for you last week? Following Lebeaux’s (1988) idea, extended later to the cases of covert quantifier raising by Fox & Nissenbaum (1999) and Nissenbaum (2000), the derivation for the sentence in (35) is sketched in (36): (36) a. [vP John [VP wear what] tomorrow] Rightward movement of the wh-phrase and late merge of the relative clause b. [vP John [VP read ___ ] tomorrow [what [RC that I bought for you last week]]] Movement of the wh-phrase to the specifier of CP, stranding the relative clause c. [CP what will [IP John [vP [VP read ___ ] tomorrow [ ___ [RC that I bought last week]]]]]

What is crucial here is the fact that the wh-element overtly moves to the right periphery of the vP projection (see Nissenbaum 2000; and Fox & Nissenbaum 1999, for arguments that the relative clause is not merged in the argument position within the VP node), and then it moves further to the left periphery of the sentence (spec,CP), instantiating a case of zig-zag movement. In this section, I showed that zig-zag movement is a fact documented in the grammar of both signed and spoken languages. However, this sort of movement is predicted to be ungrammatical when the phase theory is enriched with CL and the two restrictions on overt movement proposed in section 2. In the next section I’ll provide the solution to this puzzle.

Phase theory, linearization and zig-zag movement

151

5. Zig-zag movements, Pied-piping, EPP and cyclic movements CL imposes that the whole syntactic object is shipped to PF at the end of each phase. PF rules linearize the elements of the syntactic object in a table of orders without postulating any escape hatch. However, if an element is ordered on the edge of a phase, it cannot be re-ordered on the opposite edge in the next phase, because order contradiction between the table of orders stated at the two phases would lead the derivation to crash. In other words, CL predicts movement from the edge of a phase to the opposite edge of the following phase to be impossible: zig-zag movements are predicted to be ungrammatical, contrary to facts. Before entering into the details of my proposal and see how zig-zag movements are captured, something more needs to be said about how cyclic movements in general are derived in the current framework. This short digression will be relevant to the solution of the zig-zag puzzle. The PIC requires that elements that bear unvalued features (wh-, focus, topic features etc.) must target the edge of the intermediate phases. Given that internal merge requires a previous probe-goal agreement one might ask what the features involved in the intermediate steps of cyclic movement are. Let’s consider for instance a wh-sentence like (37): (37) what do you think that they think that I think? The wh-element moves, phase by phase, from the complement position of the most embedded vP to the specifier of the complementizer of the matrix clause, targeting every vP and CP edge on its way up, as shown in (38): (38) [CP what do you [vP __ think [CP __ that they [vP __ think [CP __ that I [vP __ think]]]]]] According to Pesetsky and Torrego (2004), every phase head must be endowed with an uninterpretable, unvalued Q feature (uQ[ ]) that forces the wh-element to move to the edge of the phase. In order to allow a whelement to move to the edge of an intermediate phase, three features are needed: – An uninterpretable, unvalued Q feature on the head of the intermediate phase that acts as a probe; – An uniterpretable, valued Q on the wh-element that act as a goal; – EPP feature on the phase head that requires the wh-element to be remerged to the specifier of the phase.

152 Carlo Geraci These assumptions may seem well motivated for intermediate CPs, especially if one thinks of the case of Irish, where wh-movement determines the morphological shape of the intermediate CPs (McCloskey 2001). However, if one considers the intermediate vPs, the presence of an uQ[ ] seems a stipulation motivated only by theory internal reasons. Moreover, this solution would not resolve the problem of the intermediate steps with other kinds of cyclic movements like focus, topic and n-word movements. To further complicate the situation, there is the EPP feature. EPPs are features hosted by the head of the phases, and whose number is determined by the lexical array (the subpart of the numeration that provides the lexical items for each phase). The role of EPP is to require that the element that has entertained agree with the head of the phase is merged to its specifier. It is hard to say that EPP can be interpretable or uninterpretable, valued or unvalued in the technical sense specified above in (4). Furthermore, EPP is tacitly assumed to be only in the head of a probe but never in goals. Thus, EPP cannot be considered a feature properly, but it is rather a subpart of a feature in the sense that its satisfaction is preceded by agree of another feature on the probe with its correspondent on the goal (Pesetsky and Torrego 2004). However, its parasitic status is questionable when other cases are considered where EPP seems to have much more independence. With expletive constructions, for example, EPP is not satisfied by the element that has entertained agree with the probe, but by an element of the lexical array that has never been considered in the previous stages of the derivation. The same is true for every case where merge pre-empts move, and more generally when external merge is at work. In these cases, there is not any sort of Agree at all, and EPP seems to play the same role of the Edge feature proposed by Chomsky (2005). The fact that EPP can be satisfied without a previous agree relation established between a different pair of features indicates that EPP works as a normal feature and that it is not parasitic on the presence of other features. If this were true, one expects that EPP will behave like a normal feature not only in cases of external merge, but also in cases of internal merge. My proposal is to treat the intermediate steps of cyclic movement as cases of movement where only EPP is at work. As a consequence of this, we can assume that EPP has a counterpart in the element that merges in the specifier of a projection, satisfying the EPP requirement of a head. Let’s consider for instance the case of cyclic wh-movement in (39a). At the end of the embedded vP phase, the elements that are still syntactically active (namely those bearing uninterpretable features that haven’t matched with the correspondent interpretable ones) must target the edge of the phase. The EPP feature on the phase head acts as a probe looking for a

Phase theory, linearization and zig-zag movement

153

syntactically active element, the wh-phrase in (39b). Once agree is established between EPP on the probe and its counterpart of the goal, re-merge applies, placing the wh-phrase at the edge of the vP phase, as shown in (39c). There is no need for other features to be involved in the process.17 (39) a. what do you think that I think? b.

vP 3 what vP 3 I vP 3 uEPP[ ] v VP 3 V what uQ[int] g iwh-[1] think iEPP[XP] agree

c.

vP 3 what vP 3 I vP 3 uEPP[XP] v VP 3 V twhat uQ[int] re-merge g iwh-[1] think iEPP[XP]

At this point one could ask what sort of value can be assigned to EPP. I propose that the value of EPP determines the size of the constituent that is 17

The case of intermediate CPs in Irish, where the morphology of the complementizer shows the passage of the wh-element, can be accounted for by considering the possibility that other features can be checked during the process of cyclic movement to the edge of the phase. Much like agree between Infl and subject is driven by the φ -features of the verb, and nominative case is checked as a free rider, I take the wh- morphology on the complementizers of intermediate CPs in Irish to be the reflex of the checking of a Q feature on C as a free raider.

154 Carlo Geraci undergoing movement. As indicated by the subscripts, the value of the EPP on the goal in (39b) is shared with the EPP on the probe in (39c), requiring the maximal projection represented by the goal to be re-merged. My proposal has two advantages: the first one is that EPP is now treated as a real feature and not as a parasitic one, vindicating the fact that it really works as a real feature both with cases of external merge (expletives and other kinds of merge from the numeration), and internal merge (intermediate steps of cyclic movements); the second is that the operation of movement can be simplified dispensing with the operation of pied piping (the size of the element that is undergoing movement is determined by the value of the EPP feature on the goal).18 We have seen that it is possible to derive intermediate cyclic movements without the postulation of extra features in the head of the phase. A crucial fact is that the elements that target the specifier position of a phase as the result of an intermediate cyclic movement are still syntactically active (they have at least one uninterpretable feature that needs to be matched with a corresponding interpretable one). I assume that these elements are not sent to spell out. This assumption is quite plausible given that the contribute of these elements to the syntactic derivation is not completed yet. This means that an element that is found at the edge of a phase counts for linearization only if it doesn’t have an uninterpretable feature. This last assumption makes possible to derive zig-zag movements. Let’s see how this happens considering in turn the case of LIS wh-elements both in object position and in subject position: (40) a. GIANNI SIGN WHAT Once vP has been completed, the subject GIANNI and the verb SIGN are shipped to PF, and ordered in the table of orders, as showed in (36b): 19 (40) b. [vP WHAT GIANNI [VP tWHAT SIGN]] Table of orders: GIANNI < SIGN 18

19

Notice that pied piping cannot be entirely excluded from the grammar. Some mechanism is still needed to determine the value of the EPP on the goal. In the case of PP movement, for instance, one could imagine some mechanism of feature percolation from the complement to the preposition making the whole PP eligible for re-merge (see Heck 2004 for a different theory of pied piping). The sentence in (40a) poses the question whether subjects in spec,vP are visible to spell-out operations. Since subjects bear at least an unvalued Case feature, in principle they should be linearized at the next phase. I leave the question open. In any case, postponing the linearization of subjects wouldn’t affect my proposal.

Phase theory, linearization and zig-zag movement

155

Once the CP phase has been completed, all the features of the wh- element have been valued, so it is now visible to the spell-out operations. The table of orders of CP is given in (40c): (40) c. [CP [IP GIANNI b. [vP tWHAT tGIANNI [VP tWHAT SIGN]]] WHAT] Table of orders: GIANNI < SIGN, SIGN < WHAT Since the orders established at the vP phase are not contradicted by the table of orders at the CP phase, the derivation concludes without problems. Let us now turn quickly to the case of zig-zag movement from the specifier of vP: (41) a. BOOK BUY WHO b. build vP [vP WHO [VP BOOK BUY]] Table of orders will contain (the wh-element is invisible for linearization): BOOK < BUY c. build IP and remerge what needs to be remerged [IP WHO [vP tDP [VP BOOK BUY]]] d. build CP and remerge what needs to be remerged [CP [TP tWHO [vP tDP [VP BOOK BUY]]] WHO] Table of orders will contain: elements ordered at vP: BOOK < BUY elements ordered at CP: BUY < WHO The orders established at the vP phase are not contradicted at the CP level. Similarly, other instances of zig-zag movements in other languages can be easily derived. In section 4, we saw that zig-zag movements are problematic both if we take vP as a possible phase in LIS, and if we take VP as a possible phase in LIS. This latter option with some further complication (see footnote 14). However, under my proposal zig-zag movements are correctly predicted to be possible both in LIS and in other languages, without postulating the existence of other phase levels. Furthermore, my proposal shares with the CL theory the idea that specifiers and heads do not differ with respect to linearization issues. The fact that an element is ignored by the interfaces is not due to its syntactic position but to the fact that it is still syntactically active. Once all the uninterpretable features are checked, an element becomes visible to the interfaces and thus it is linearized.

156 Carlo Geraci However, an element that is not at the edge of a phase can still serve as a goal for a probe in higher phases. This movement will not result in the satisfaction of any feature of the goal.20 Before concluding the paragraph one final consideration deserves discussion. Although the solution of the puzzle presented by zig-zag movement captures the derivations of LIS sentences with wh-phrases and n-words, it does not provide an explanation to the fact that wh-elements and n-words exhibit rightward movement in LIS,21 as opposed to parallel leftward movement exhibited by many spoken languages (for a proposal that provides an explanation to rightward movement in LIS see Cecchetto, Geraci and Zucchi, in prep.). However, when the case of late merge in English is considered one might ask why phrases that are late merged can only appear to the right periphery of the vP node and not also to its left periphery. I don’t have a specific answer to this question but a suggestive speculation can be made when the case of heavy NP shift (HNPS) is considered. The common explanation to HNPS is that an NP in object position is shifted because it is too big compared to the other phrase that follows and it would create difficulties for pronunciation, as shown in (42): (42) I gave to Mary the red book about King’s Arthur life. Now, the element that targets the vP phrase and is later merged with other syntactic structure (late merge of the relative clause, for instance) is somehow heavier that it was before. Thus, the speculation is that whatever is responsible for HNPS is also responsible for the fact that late merged elements appear to the right periphery of the vP and not to the left one (see Fox and Pesetsky 2003, for a different view).

20

21

A consequence of this fact is Holmberg generalization and Inverse Holmberg generalization are still captured in the same way they were captured with the original formulation of the CL. Notice that even adopting a rigid spec-head-complement order, and richer IP and CP layers, the fact that wh-phrases in subject position are preverbal at the vP phase and postverbal at the CP phase produces order contradiction for CL (of course, under that view the name “zig-zag movement” would be inappropriate).

Phase theory, linearization and zig-zag movement

157

6. Conclusion This paper showed that the existence of zig-zag movement is documented both in signed and spoken languages. Movements of this sort create a problem for CL when two plausible restrictions on movement are considered: ban on movement from a specifier to an outer specifier of the same head (Ko 2005), and ban on movement of a complement to the specifier of the same head (Abels 2003). The solution that I proposed relay on the fact that those elements that are still active for the syntactic component are simply ignored by the operation that sends the structure to the interfaces. However, instead of postulating a specific structural position where the elements can hide, I assumed that it is the fact that they bear an uninterpretable feature that makes them invisible to the linearization algorithm. Furthermore, the account I provided refines the mechanism that derives cyclic movements to the edge of a phase and makes the definition of movement simpler. The intermediate steps of a cyclic movement are now derived only via EPP, without the postulation of a specific feature on the head of each phase. EPP is seen as a real feature whose value indicates the portion of the element that is undergoing re-merge, and can be satisfied by both external merge and internal merge. Finally, the definition of movement now contemplates only the notion of agree and re-merge.

References Aarons, Debra 1994 Aspects of the syntax of American sign language. Doctoral dissertation, Boston University, Boston, MA. Abels, Klaus 2003 Successive cycliclity, anti-locality, and adposition stranding. Doctoral dissertation, University of Connecticut, Storrs. Cecchetto, Carlo, Carlo Geraci, and Sandro Zucchi 2006 Strategies of relativization in LIS. Natural Language and Linguistic Theory 24: 945–975. in prep. Another way to mark chains. Ms., Università degli Studi di MilanoBicocca. Cecchetto, Carlo and Sandro Zucchi 2004 Why is Spec,CP on the right in sign languages? Paper presented at GLOW, Thessaloniki, Greece.

158 Carlo Geraci Chomsky, Noam 2000 Minimalist inquiries. In Step by Step: Essays on Minimalist Syntax in Honor of Howard Lasnik, Roger Martin, David Michales and Juan Uriagereka (eds.), 89–156. Cambridge, MA: MIT Press. 2005 On phases. Ms. MIT Cambridge, MA. Fox, Danny and Jon Nissenbaum 1999 Extraposition and scope: A case for overt QR. In Proceedings of the 18th West Coast Conference on Formal Linguistics, S. Bird, A. Carnie, J. D. Haugen, and P. Norquest (eds.), 132–144. Fox, Danny and David Pesetsky 2003 Covert Merge and Later Merge. Class note: Topix in Syntax, MIT, Fall 2003. Cambridge, MA. 2005 Cyclic linearization of syntactic structure. Theoretical Linguistics 31: 1–45. Frazier, Lyn and Charles Clifton Jr. 2002 Processing “d-linked” phrases. Journal of Psycholinguistic Research 33: 633–660. Geraci, Carlo 2006a Negation in LIS (Italian Sign Language). In Proceedings of NELS 35, L. Bateman and C. Ussery (eds.), 217–229. Amherst, MA: GLSA. 2006b LIS tra ricerca e divulgazione. Doctoral dissertation, Università degli Studi Milano-Bicocca, Milano, Italy. Ko, Heejeong 2005 Syntactic edges and linearization. Doctoral dissertation, MIT, Cambridge. Kayne, Richard S. 1994 The Antisymmetry of Syntax. Cambridge, MA: MIT Press. Lebeaux, David 1988 Language acquisition and the form of the grammar. Ph.D. thesis, University of Massachusetts, Amherst. McCloskey, James 2001 The morphosyntax of WH-extraction in Irish. Journal of Linguistics 37: 67–100. Neidle,Carol, Judy Kegl, Dawn MacLaughlin, Benjamin Bahan, and Robert G. Lee 2000 The Syntax of American Sign Language: Functional Categories and Hierarchical Structure. Cambridge, MA: MIT Press. Neidle, Carol, Stan Sclaroff, and Vassilis Athitsos 2001 SignStream™: A tool for linguistic and computer vision research on visual-gestural language data. Behavior Research Methods, Instruments & Computers 33: 311–320. Nissenbaum, Jon 2000 Investigation of covert phrasl movement. Doctoral dissertation, MIT, Cambridge.

Phase theory, linearization and zig-zag movement

159

Pesetsky, David and Esther Torrego 2004 The syntax of valuation and the interpretability of features. Ms., MIT, Cambridge. Petronio, Karen and Diane Lillo-Martin 1997 Wh-movement and the position of Spec-CP: Evidence from American Sign Language. Language 73 (1): 18–57. Pfau, Roland and Ulrike Zeshan 2004 Wh-movement and wh-split in Indo-Pakistani Sign Language. Paper presented at SALA 23, Austin, TX. Wilbur, Ronnie B. and Cynthia Patschke 1999 Syntactic correlates of brow raise in ASL. Sign Language and Linguistics 2: 3–41. Zucchi, Sandro 2003 The semantics of FATTO: A case of grammaticalization. Paper presented at the 14th Amsterdam Colloquium.

Surviving reconstruction Thomas S. Stroik and Michael T. Putnam

1. Introduction Constructions with Condition C asymmetries (see (1) and (2)) pose the problem of explaining why some, but not all, the DPs contained in the bracketed wh-phrases can be co-referential with the pronoun he. (1)

[which picture of Bill that John likes] did he buy? *Bill…he/ Ok John…he

(2)

a. he likes [everything that John writes] b. [everything that John writes] he likes

*John…he Ok John…he

To account for asymmetries such as those in (1) and (2), Lebeaux (1991, 1995), Chomsky (1993), Epstein et al. (1997), Rubin (2003), and Fox (2003, 2004) propose two different applications of the Merge operation – one that applies cyclically and one that applies non-cyclically (i.e., after the application of Move or Remerge operation). Besides positing two applicational platforms of Merge, these theorists also stipulate that all arguments must be merged cyclically, while adjuncts can be merged cyclically or non-cyclically. These assumptions, together with Chomsky’s (1993) copy theory of movement, will generate the following derivation for (1). (3)

a. he did buy [which picture of Bill] – wh-movement  b. [which picture of Bill] did he buy [which picture of Bill] – adjunct Merge  c. [which picture of Bill [that John likes]] did he buy [which picture of Bill]

Importantly, this derivation allows us to explain the co-referential relations in (1): the DP Bill in (3c) cannot be co-referential with the pronoun he because the pronoun c-commands the most embedded DP copy Bill, whereas the DP John can be co-referential with the pronoun because the late merger of the relative clause prevents the pronoun from ever c-commanding the DP. This sort of analysis can account for the data in (1) and (2), but it

162 Thomas S. Stroik and Michael T. Putnam comes at a cost. In particular, having non-cyclical operations dramatically complicates processing since such operations force derivations to return to, and re-compute, structures already built (cf. Frampton and Gutmann 2002). In an attempt to simplify processing, Chomsky (2001) eliminates all (expensive) non-cyclical Merge operations. However, purging non-cyclical Merge leaves Chomsky with the problem of how to account for the reconstruction asymmetries in (1). To solve this problem, Chomsky proposes that the narrow syntax must have two separate, cyclical Merge operations: argument-Merge (Set Merge) and adjunct-Merge (Pair Merge). Although these two Merge operations are both cyclical, they differ in their structural properties – the output of the adjunct-Merge operation is not structurally visible (or “simple”) until it is converted to a Set Merge output via an operation called Simpl. Hence, for Chomsky, the relative clause in (1) is merged into the derivation, as in (4a), but it remains invisible until the derivation reaches (4b). (In (4) the italicized clause is syntactically invisible.) (4)

a. he did buy [which picture of Bill that John likes] – wh-movement  b. [which picture of Bill that John likes] did he buy [which picture of Bill that John likes] – Simple-operation  c. [which picture of Bill that John likes] did he buy [which picture of Bill that John likes]

It is the structural invisibility of the most embedded relative clause in (4a) and (4b) that keeps the DP John from being c-commanded by the pronoun. Though Chomsky explains the coreferential relations in (1), we should note the fact that to account for these coreferential data, without appealing to non-cyclical Merge operations, Chomsky must complicate his system of operations, in a rather bizarre way, by adding a non-simple adjunct operation (Pair Merge) and a Simpl operation that later undoes (makes simple) the output of the adjunct operation. In this paper, we offer a re-analysis of the reconstruction data in (1) and (2) that avoids both non-cyclical operations and Chomsky’s undoing operations. We propose, in particular, that the Condition C asymmetries in (1) and (2) can be explained within Stroik’s (1999, to appear) version of minimalism (called Survive). According to Stroik, an optimal minimalist syntax will have only strictly local Merge operations that map elements from the Numeration N onto the Derivation D. To maximally simplify processing, these operations will not have look-back or look-forward properties (this criterion rules out Attract and Agree operations, economy conditions, and Internal Merge operations). For Stroik, there are only two such types of

Surviving reconstruction

163

syntactic operations: Merge, which concatenates, in D, two elements from N (what actually gets merged are copies of elements of N – the originals of which continue to be contained in N); and Remerge, which remerges into D syntactic objects SOs in N that have already been merged but still have concatenative features that must be checked (these features have “survived” previous applications of Merge). An abbreviated Survive-derivation can be seen in (5) (note that lower copies of constituents are marked in boldface type). (5)

a. b. c. d.

Merge {hired, Sam}  hired Sam Survive {Sam} Merge {was, {hired, Sam}}  was hired Sam Remerge {Sam, {was, {hired, Sam}}}  Sam was hired Sam

The DP Sam, which is merged with the verb hired in (5a), must remerge, from N, later in the derivation (see (5d)); this is necessary to ensure that its Case and agreement features, which have survived the verb-merge, are appropriately checked. It is worth emphasizing here that (5d) has been derived without any non-cyclical operations and without any Internal Merge operations. As we will show in this paper, we can give derivations for (2a) and (2b), similar to the one in (5), if we adopt a version of Fox’s (2004) late adjunct Merge – i.e., by allowing adjuncts to merge with an SO prior to Remerge (such a merger should be permissible because the SO is in the Numeration, not in the Derivation; hence the merger will not be non-cyclical). Under this version of late Merge, (2b) could be derived as follows: (6)

a. b. c. d. e.

Merge {likes, everything}  likes everything Survive {everything} Merge {he, {likes, everything}  he likes everything Merge {everything, {John writes}}  everything John writes Remerge {everything John writes, {he, {likes, everything}}}  everything John writes he likes everything

Given that the pronoun in (6a–e) never c-commands the DP John, it is possible for the DP and the pronoun to be coreferential. In this paper, we will provide equivalent derivations, and explanations, for the coreferential relations in (1) and in (2a).

164 Thomas S. Stroik and Michael T. Putnam 2. On syntactic operations The ultimate goal of this paper is to investigate the sorts of syntactic operations that are conceptually necessary to account for the reconstructions effects in constructions such as those in (1) and (2). According to Chomsky (1993, 1995, 2001), the computational system of Human Language (HL) requires two structure-building operations – External Merge and Internal Merge (a variant of the Move operation). External Merge (EM) is a mapping from the Numeration (N) – a lexical array of items taken from the lexicon – to the syntactic Derivation (D), as stated in (7). (7)

EM: N  D

This operation builds syntactic structure from lexical input: it takes a lexical item and concatentates it in D with another lexical item or a syntactic object (as already formed in the derivation D). That is, for lexical item α ε N and for β ε N or D, EM syntactically connects α and β (see (8)). (8)

EM {α,β}  αβ

It is the EM operation that will derive the phrase the syntactic object that child from the lexical items that and child: EM {that, child}  that child. Some version of the EM operation is conceptually necessary if we are to account for Frege’s (1884) observations about the compositional (concatenative) nature of semantics. However, the simple version of the EM operation formulated in (8) is too strong for HL. As given in (8), EM would not place any constraints on the ability of the computational system to introduce lexical items into any derivation. In other words, this EM would allow any, and every, concatenation. The EM operation could, for example, concatenate a Determiner with a Determiner Phrase (DP), as in (9), producing an ungrammatical syntactic object. (9)

EM {the, {this, child}}  the this child

That the EM operation in (8) overgenerates permissible syntactic objects in a Language suggests that this operation cannot be freely concatenative; rather, it must impose constraints on licit concatenations.1 1

Collins (1997) also proposes a similar constraint on the primitive operation Merge for similar reasons.

Surviving reconstruction

165

Adger (2003) and many others have reduced the expressive power of EM (his Merge) by positing this operation not as a concatenating operation, but as a linking operation (Putnam (2006b, 2007) formalizes this as Link!). Noting that lexical items, and the syntactic objects formed from them, consist of sets of phonological, morphological, syntactic, and semantic features F, Adger proposes that lexical item α will merge with syntactic object β if and only if α,β share, and check, features – this is stated in (10). (10) EM {α, β}  αβ Under this analysis of Merge, the basic structure-building operation in HL compiles syntactic structure by linking together lexical items with matching syntactic features. We can see this in (11a) and (11b). In (11a), the verb give has object features for both a nominal object and a prepositional

object; hence the verb can merge with both the nominal constituent that book and the propositional constituent to Pat. On the other hand, in (11b), the verb admire has an object feature, but not a

feature; therefore this verb can merge with that book, but not with to Pat. (11) a. Chris gave that book to Pat b. Chris admired that book (*to Pat) Relatedly, the determiner that has a singular Number feature <SING>, so it will not be able to merge with any nouns that have an un-matching plural Number feature , as in the determiner phrase in (12). (12) *that books By placing a feature-matching constraint on concatenation, EM (10) radically reduces the generative capacity of the merge operation and it makes syntactic structure a linked structure and not merely a concatenated structure. Chomsky (1965, 1980) has long argued that syntactic structures are not only built – by phrase structure rules or by the Merge operation – but they are also modified (or transformed). Motivation for “transformed” structures comes from the data in (13) and (14), where the bracketed DP object of the verb like in (13a) can appear in a modified position in (13b) and where the wh-constituent in (14), which is merged as the object of the verb elected, must be transformed into the subject of the embedded sentence and must also be placed into the fronted interrogative position in the matrix sentence.

166 Thomas S. Stroik and Michael T. Putnam (13) a. I like [that book] b. [that book] I like (14) Which politicians does Chris think will be elected. To explain how a constituent can appear in a syntactic position distal from its merged position, Chomsky (2001) claims there is a second syntactic operation, called Internal Merge (IM), which can re-locate constituents within a Derivation. This operation differs from EM in one essential way – it builds structure not by linking elements from N into D, but by re-positioning elements already in D somewhere else in D. That is, IM is a D-to-D mapping, as stated in (15). (15) IM: D  D What IM does, in particular, is take an element a ε D and remerge it in D. So, if D consists of [d [ba]], then IM can select the element a and remerge it (see (16)). (16) IM {a, [d[ba]]}  [a[d [b a]]] where a is a phonetically inert copy of a. As with our first formulation of EM (8), IM (16) is much too powerful. It would allow any element in D to be remerged, thereby permitting the verb have in (17a) to remerge as in (17b). (17) a. Pat shouldn’t have left  IM b. *Have Pat shouldn’t have left. The fact that (17b) is ungrammatical demonstrates that IM (16) overgenerates syntactic structure and that IM must reduce its derivational capacity. Chomsky (2001) lessens the computational power of IM in two ways. First, rather than letting any element remerge at any point in the derivation, Chomsky limits remerge to cases involving feature-match. For remerge to occur, a head (say d) must have a concatenative feature that must be linked to a paired feature on another syntactic constituent. The head d will “probe” the derivation D searching for a constituent with an active feature, i.e., a feature not previously checked for feature-match by another head. If constituent a has such an feature, then the constituent will be able to be remerged, in accordance with IM (18). (18) IM {a, [d [b a]]}  [a [d [ba]]]

Surviving reconstruction

167

Notice that both EM and IM build syntactic structure by locally linking constituents with matching features. The second way Chomsky constrains IM is by delimiting the search that d can undertake to find a constituent with a counterpart feature. Chomsky seems keenly aware that having unbounded searches (searches through an entire derivation D) would greatly increase the processing burden on the computational system. So, he proposes that these searches be top-down searches that are domain-restricted within D by phase boundaries (where a phase is a semantically complete phrase – a CP or a vP). A head d can look down through the derivation D until it encounters a phase boundary; at this point, it can still look at the Spec of this phase. However, it can’t look any deeper in D. Should d not find any constituent with an active feature in its search-restricted domain, it must abort its search. Simplified even further, the search will also terminate once its goal – a constituent with an active feature--is found. The search will not continue to look for subsequent goals. To see how these operations are implemented, consider the derivation in (19). We will pick up the derivation once it has merged its C head, which has both a feature and a <WH>. (19) a. [C [Pat will read [what<WH> to whom<WH>]]]  IM b. [will [C [Pat will read [what<WH> to whom<WH>]]]]  IM c. [what<WH> [will [C [Pat will read [what to whom <WH>]]]]] ‘What will Pat read to whom?’ In (19a), the C head will look for a constituent with a feature within its search domain. Once it locates the constituent will, the search will terminate and IM will apply, remerging will in the matrix CP. Since the C head also has a <WH> feature, the head must begin another search – one that looks for a <WH> counterpart feature. The search will terminate when the first active <WH> is found on the constituent what<WH>. IM will apply again, this time remerging what<WH> in the matrix CP. Despite the seeming ease with which (19c) is derived, this derivation is actually extremely complex. If we look closely at how the derivation proceeds from (19b) to (19c), we will see some of this complexity. Of note here is the fact that IM can’t apply directly to (19b). Rather, before IM can apply to (19b), the C<WH> head must initiate a search for a constituent

168 Thomas S. Stroik and Michael T. Putnam with a <WH> feature – that is, a FIND<WH> operation must apply to D. The FIND operation itself is a complex (composite) operation that must LOOK-AT-FEATURES of every constituent K in the Derivation, that must DISREAGARD all features of K that are not a <WH> feature, and that must GO-TO-NEXT constituent should a <WH> feature not be found on a given K; and of course, all these operation will have to apply to each successive K that is searched. Furthermore, should the search successfully FIND<WH> in D (the what<WH> constituent), then the search ends, but another set of operations must now apply. These operations include a MAKE-COPY operation, an operation that, according to Chomsky (2001), will create a copy without phonetic features 2, and the REMOVE-ORIGINAL and INSERT-COPY operations, which will remove the constituent located by FIND and replace it with the copy formed by the non-phonetically realized MAKE-COPY operation. The removed constituent can now undergo IM with D, deriving (17c) from (17b) after the application of all the intermediate operations identified above.3 Needless to say, using IM to build syntactic structure is extremely costly, in processing terms, because of all the ancillary operations required to license the IM operation.4 For derivations that must apply IM several times (as would be the case for (14)), the expensive processing needs of IM are all the more daunting. (14) Which politicians does Chris think will be elected. The computational complexity surrounding the IM operation raises significant questions – similar to those raised by Brody (1998, 2002) and Stroik (to appear) – about the conceptual necessity of this operation.

2

3

4

This is a rather problematic operation in that it is unclear where this operation could take place. Another problem with current minimalism is that in the derivation illustrated in (19) the C-head possessing the [Q]-feature that will peer down into its ccommand domain to find a relevant goal is previously blind to the wh-items that have existed in the derivation for quite some time. That these wh-items are not found and recognized until the very end of the derivation is problematic for a true minimalist analysis of syntax. For example, Internal Merge requires ancillary operations such as FIND PHASE EDGE that burden the processing load of the language faculty.

Surviving reconstruction

169

3. Internal merge and reconstruction effects In this section, we look all the more closely at the IM operation by considering how an IM analysis can account for the Reconstructions Effects in (20). (20) a. [which picture of Bill that John likes] did he buy? *Bill…he/OK John…he b. He bought [a picture of Bill that John likes]. *he…Bill/*he…John What needs to be explained in (20) is why the coreferentiality relationship between the pronoun he and the DP Bill is not affected by whether the bracketed DP undergoes displacement (as in (20a)) or not (as in (20b)), while the coreferentiality relationship between the pronoun and the DP John is affected by the displacement of the bracketed DP. We need to determine what role, if any, IM plays in these differing coreferentiality relationships. The data in (20) led Chomsky (1993) to conclude that arguments contained within displaced constituents, such as the DP Bill in (20a), behave, in terms of Principle C relations, as if they are in their pre-displacement positions (see (20b)). In other words, the constituent [which picture of Bill] in (20a) acts as if it shows up twice in a syntactic derivation – once in its displaced (internally merged) position and once, in reconstructed copy form, in its (externally) merged position, as in (21). (21) [which picture of Bill that John likes] did he buy [which picture of Bill] For Chomsky (1993), Lebeaux (1995) Epstein et al. (1997), Fox (2003), and many others, the data in (20) provide important support for some version of a copy theory of movement (variants of Internal Merge). This above analysis, however, does not hold for adjuncts contained in displaced constituents. As we can see in (20a), the adjunct (the relative clause) contained within the displaced constituent does not appear to show up twice syntactically. If it did show up in the same places that the argument DP Bill does, then we would expect that (20a) would have (22) as its syntactic representation at some point in the derivation. (22) [which picture of Bill that John likes] did he buy [which picture of Bill that John likes]

170 Thomas S. Stroik and Michael T. Putnam But (22) is not a possible representation for (20a) because in (22) the pronoun he c-commands (and should not be co-referential with) the adjunct contained DP John. The fact that the pronoun and the DP can be coreferential argues against (22). What this means is that although the adjunct can show up once the bracketed constituent is displaced in (20a), it cannot also show up syntactically in the pre-displaced bracketed constituent. Within IM-style analyses, there are only two ways of explaining why the adjunct in (20a) shows up syntactically after the wh-constituent has been displaced. Either the adjunct is merged into the derivation after the wh-displacement (the Late Merge Hypothesis) or the adjunct is merged into the derivation prior to the displacement, but becomes syntactically visible only after the displacement (the make Simple Hypothesis). Late Merge analyses of data such as (20a) have been advanced by Lebeaux (1991), Chomsky (1993), Fox (2003), and others. Under the Late Merge analysis, (20a) is derived as follows: (23) a. he did buy [which picture of Bill] – wh-movement  b. [which picture of Bill] did he buy [which picture of Bill] – adjunct Merge  c. [which picture of Bill [that John likes]] did he buy [which picture of Bill] Even though this derivation expresses all the structural constraints on coreferential relations between the pronoun he and both the DP Bill and the DP John, it is an untenable analysis. Behind the apparent simplicity of (23) lurks a significant problem. As Chomsky (2001) notes, no Merge operation should “tuck” elements into a derivation D because of the processing complications that arise when structure is re-modeled from within (the structure would have to be, in essence, dismantled and then rebuilt). Merge operations, according to Chomsky, must be cyclic operations that build structure at the edges of D, not within D. If we look at derivation (23), we will see that the Late Merge of the adjunct is a case of a non-cyclic Merging of the relative clause into the wh-constituent; therefore this Merge should be disallowed. It might be possible, however, to circumvent the “tucking in” problem by allowing the IM operation to re-situate the wh-constituent in (23a) in some work-space outside the derivation (where it could Merge the adjunct) prior to remerging the constituent back into D; unfortunately, doing this would add several complexities to the IM operation beyond those we discussed in the previous section. These complexities would include chang-

Surviving reconstruction

171

ing IM from a D-to-D mapping to a D-to-Workspace mapping and then requiring a subsequent operation that could remerge elements from the Workspace to D, among other complexities. At this point, the IM operation loses any semblance of simplicity and, in fact, loses any claim to being an operation that involves the internal merge of constituents. It would seem, then, that if we are to have the IM operation in our Narrow Syntax, we must follow Chomsky in assuming that the adjunct in (20a) is not merged late into the derivation; rather, it is merged in its externally Merged position shown in (20b), though it is Merged differently than an argument would be. The adjunct is Pair Merged into D, which means, essentially, that it is merged, but it is not syntactically visible; and it will remain syntactically invisible and inert until the Simpl operation makes the adjunct syntactically present. Under these assumptions, (20a) will have partial derivation (24). (Recall that the italicized constituent is syntactically invisible.) (24) a. he did buy [which picture of Bill [that John likes]] –wh-movement  b. [which picture of Bill [that John likes]] did he buy [which picture of Bill [that John likes]] – Simpl  c. [which picture of bill [that John likes]] did he buy [which picture of Bill [that John likes]] Attractive as this analysis may be for the data in (20), it does not hold up if we look at some other adjunct constructions. Consider the data in (25). (25) a. [after Pat wakes up] I want her to leave b. I want her to leave [after Pat wakes up] In (25a), the bracketed constituent is a temporal adjunct that modifies the embedded verb leave and, under Chomsky’s analysis of adjuncts, this adjunct will have to be externally Merged in its verb-modifying position, as in (25b). Since the adjunct is Paired Merged in the embedded VP, it will be syntactically invisible there. But now we face a quandary. We need to move the adjunct from its embedded position in (25b) into its displaced position in the matrix sentence (as in (25a)). However, this adjunct can’t get a free ride the way the adjunct in (20a) does. That is, the adjunct in (20a) is contained within a wh-constituent that gets displaced, so the adjunct gets displaced as a by-product of the wh-movement. Since the adjunct in (25b) can’t get a free ride, it will have to move on its own accord. This requires, however, that the adjunct be syntactically visible for the IM operation. In

172 Thomas S. Stroik and Michael T. Putnam other words, before the adjunct can be moved, it will have to undergo Simpl. The derivation for (25a) will have to proceed as in (26). (26) a. I want her to leave [after Pat wakes up] – Simpl  b. I want her to leave [after Pat wakes up] – IM  c. [after Pat wakes up] I want her to leave [after Pat wakes up] This derivation leads us to a problematic conclusion; that is, given derivation (26), it should be impossible to have co-referential relations between the pronoun her and the DP Pat because the pronoun c-commands the DP in structures where the adjunct is syntactically visible – (26b,c). The fact that the pronoun and the DP can be co-referential suggests that the Simplbased derivation (26) is not a viable derivation for (25a). Of the two IM analyses we’ve considered for adjunct-displacement constructions, neither of them is feasible. Since it seems at present that there are only two possible IM analyses for these constructions, we are left as were at the end of the last section with the nagging sense that IM (or any movement-related operation) lacks both conceptual and empirical motivation.

4. Surviving reconstruction After exposing the conceptual faults that we encountered with the two IManalyses in accounting for adjunct-displacement constructions and the potential added burden that they add to the computational system with regards to language processing, in this section we will present an alternative analysis along the lines of Stroik (1999, to appear) and Putnam and Stroik (in preparation). Under this view, we interpret the displacement of syntactic objects from their based position not necessitated by Attract or Move, but enacted by means of survival. Stroik defines this grammatical primitive as the Survive Principle: (27) The Revised Survive Principle (based on Stroik 1999: 286) If Y is a syntactic object (SO) in an XP headed by X, and Y has an unchecked feature [+F] which is incompatible with the feature X, Y remains active in the Numeration. To provide an illustration of the Survive Principle in action, consider the following sentence in (28) with its derivational history following in (29) (data taken from Stroik (to appear): 79–80).

Surviving reconstruction

173

(28) Who snores? (29) a. b. c. d. e. f. g.

Merge {who, snores} → who snores Survive {who} Merge {T, {who, snores}} → T who snores Remerge {who, {T, {who, snores}}} → who T who snores Survive {who} Merge {C, {who, {T, {who, snores}}}} → C who T who snores Remerge {who, {C, {who, {T, {who, snores}}}}} → who C who T who snores

Upon the concatenation of a syntactic object with a head both bearing the matching feature δ through the operation Merge the syntactic object will survive and remain active in the lexicon if the syntactic object bears any additional features not present on the immediately governing head. In the derivation above the wh-item who will be mapped into the vP to check its θ-feature. At this point in the derivation a link is established signaling to the external interfaces (e.g. LF, PF) the thematic identity associated with this concatenate structure (cf. Putnam 2006b, 2007). Immediately after the concatenation of <who, snores> (29a) who survives from this position due to the additional features it possesses that must be properly licensed through iterative applications of Merge and Remerge in the course of the derivation. In steps (29d) and (29g) who remerges from the lexicon in order to properly discharge its agreement and Q-features. Perhaps the term ‘discharge’ is a bit of a misnomer, because the true motivation behind the sequence of MergeSurvive-Remerge is to generate concatenate structures that are interface interpretable. As explained earlier in Section 2 of this paper, the mapping of copies of lexical items into the narrow syntax rather than the objects themselves eliminates the need for Copy Theory and “movement” a priori from the theory, thus providing a purely derivational account of syntactic operations rather than a view of mixed theory that is weakly representational (cf. Brody 1998, 2002). The iterative application of Merge-Survive-Remerge also provides a straight-forward account to long-distance wh-movement previously unattainable in minimalism. (30) Whati do [TP ti you think [CP ti John likes ti]]? (31) *Whati do [TP ti you think [CP that John likes ti]]?

174 Thomas S. Stroik and Michael T. Putnam Sentences similar to (30) and (31) serve as canonical examples in the generative tradition to illustrate the reality of cyclic movement.5 In (30) the wh-item what must move to the left periphery of the embedded CP, TP and then to its final destination in the matrix CP. Example (31) shows that what must strictly adhere to cyclic movement or else the system will ultimately crash. Any theory of syntax employing either a Move or Attract model of constituent construal must delay the final feature-evaluation and subsequent checking or valuing process until the final C enters the derivation. Furthermore, successive cyclicity is an unsubstantiated formative in these models, i.e. it is a necessary component of the theory although we have little if any proof why it exists. Our version of XP-displacement under the MergeSurvive-Remerge mechanism forces the evaluation of the feature identity of all lexical items upon the merger of every head into the narrow syntax. In example (30), after concatenating with V, the wh-item what immediately survives (due to its remaining [Q] feature) and remains active in the lexicon for further operations. This syntactic object is an eligible candidate to remerge into the syntax at any time; however, it can only do so when a head with a matching feature appears. Upon every application of head merger an evaluation process takes place within the computation system.6 Returning to the focus of this paper, the remainder of this section will illustrate the conceptual advantages our approach has in properly deriving reconstruction structures in avoiding the aforementioned pitfalls of IManalyses. First, let’s return to example (20) from the previous section. (20) a. [which picture of Bill that John likes] did he buy? *Bill…he/OK John…he b. He bought [a picture of Bill that John likes]. *he…Bill/*he…John

5

6

Bear in mind that the derivational histories in (30) and (31) represent a movement-based analysis akin to former instantiations of generative theory. The proposal put forward in this paper does not support the theoretical approach that constituent displacement takes place by means of Move but rather Survive. David Pesetsky (personal communication) points out that the application of evaluation processes upon every iterative head merger could also potentially be very costly from a processing standpoint. Be that as it may, it is far more economical to envision a system which immediately evaluations candidates rather than one that makes use of look-ahead and look-behind operations.

Surviving reconstruction

175

Both the application of Late Merge and the Simpl operation are untenable options in explaining these Condition C inconsistencies. Late Merge (as currently formulated) requires the “tucking in” of the adjunct [that John likes] into the wh-items which in itself is an undesirable result, while the “peek-a-boo” effects of Simpl cloak the adjunct through some of the derivation and make it visible for syntactic operations and effects later on. Although it is an attractive alternative solution for (20), it does not hold up when we consider other adjunct constructions such as (25). (25) a. [after Pat wakes up] I want her to leave b. I want her to leave [after Pat wakes up] The crux of the matter is determining how and when the adjunct [that John likes] enters the syntax. Fox’s (2004) current version of Late Merge faces the unwanted “tucking in” problem since it is in the Derivation rather than in the Numeration. In a Survive-based model of syntactic derivation, we can avoid tucking the adjunct (qua Merge) to the complex wh-item by arguing that the adjunct resides in the Numeration and adjoins to the wh-item [what picture of Bill] prior to its remerging into the syntax since it survives and returns back to the lexicon due to its [Q]-features which must be checked in CP. Call this operation Late Num Merge.7 Two points must be clarified at this point to understand the conceptual advantages of our approach to a minimalist, derivational approach to generate syntax: First, the adjunction of [that John likes] is a syntactic object in the Numeration, therefore its concatenation with [what picture of Bill] will not be non-cyclic and therefore does not fall victim to “tucking in”. Second, the cyclic application of our reformulation of Late Merge forces the adjunct [that John likes] to be visible in the syntax for all operations. This fact allows us to abandon the now unnecessary Simpl operation on the grounds of virtual conceptual necessity. Since the DP John was not a part of the original complex wh-item [what picture of Bill] that merged into the VP prior to its repulsion there is no point in the derivation during which the pronoun he could potential ccommand John, thus explaining how John and he can be co-referential in (20a). The derivational history in (32) below highlights the pivotal steps in the composition of (20a).

7

We take an agnostic view at this time as to whether adjunction is motivated by some sort of feature or feature-like entity active in the Numeration or Derivation (also see Putnam 2006a: ch. 4 and Rubin 2003).

176 Thomas S. Stroik and Michael T. Putnam (32) a. Merge {buy, [which picture of Bill] } → buy which picture of Bill b. Survive [which picture of Bill] ([Q]-feature) → c. Merge {he, buy [which pictures of Bill]} → he buy which picture of Bill d. Merge {did, {he, {T, {buy [which picture of Bill] }}} → did he buy which picture of Bill e. Merge {C, {did, {he, {T, {buy [which picture of Bill] }}}} → C did he T buy which picture of Bill f. Late Num Merge { [which pictures of Bill], [that John likes] } g. Remerge { [which pictures of Bill that John likes], {C, {did, {he, {T, { buy [which picture of Bill] }}}}} → which pictures of Bill that John likes C did he T buy which picture of Bill ‘Which pictures of Bill that John likes did he buy?’ The non-cyclic application of Late Num Merge (32f) in the Numeration rather than in the course of the Derivation provides a straightforward explanation of Condition C asymmetries within core minimalist desiderata.

5. Theoretical consequences The removal of Internal Merge (Move) would undoubtedly have widesweeping effects on the entire generative enterprise. A closer look at these changes shows that they would be a welcome adjustment to the minimalist program. First, by dismissing Internal Merge in favor of a Survive approach, the theory needn’t construe and enforce economy constraints upon the language faculty anymore; the fact that local evaluation processes take place at ever step of Merge and Remerge throughout the course of the derivation mitigates the necessity of the existence of such constraints on the grammar. Economy is a natural by-product of the Survive Principle. Second, Chomsky’s formulation of phases (vP and CP) and multiple Spell-Out (cf. Uriagareka 1999 and a host of others) can be removed from the system. In current minimalist models of syntax supporting either a Move or Attract stance on XP-construal, not only must some version of look-ahead or lookback feature evaluation exist, but some sort of evaluating property that recognizes the shape of phases must also be a component of the theory. In a minimalist syntax that is argued to be label-free in the spirit of Collins (2002), it is conceptually puzzling and taxing from a processing standpoint why the human language faculty should/must be responsible for both fea-

Surviving reconstruction

177

ture evaluation and the recognition of larger units such as phases. A derivational model based on the Survive Principle has the distinct advantage of destroying the theory’s reliance on such rich ontological commitments. Before moving to the conclusion it must be pointed out that through the Survive Principle we seek to remove Internal Merge without abandoning a derivational view of syntax. Taking into account Brody’s (1998, 2002) excellent and accurate criticism of minimalism being a mixed, weakly derivational theory, we needn’t surrender minimalism to a purely representational line of thought. On the contrary, we seek to remove Internal Merge which, according to Stroik (to appear), is the culprit of minimalism’s classification as being partly-derivational and partly-representation. The elimination of Internal Merge in favor of Survive creates a pure derivational view of minimalism. Most importantly, our revision of the minimalism program does not increase the computational workload of the language faculty, but rather significantly reduces constraints and ontological internal interfaces in the narrow syntax as well as what materials appear at the interfaces. These adjustments bring us one step closer to Frampton and Gutmann’s (2002) vision of a crash-proof syntax and how it should operate.

6. Conclusion In this paper we sought out to address the Principle C asymmetries connected with adjunction reconstruction. We showed where both proposed solutions affiliated with Internal Merge (or anything theory of constituent displacement by means of Move for that matter) face significant shortcomings in the analysis presented here. Chomsky’s (2001) Simpl taxes the computational language faculty with an operation that forces it to reconfigure previous ‘invisible’ structure for syntactic considerations, while the application of Fox’s Late Merge (2004) forces the allowance of “tucking in” of structures to previously merged constituents. Based on the work of Stroik (1999, to appear) and Putnam and Stroik (in preparation), we proposed a version of constituent displacement perpetuated by the repulsion of objects rather than attraction (Survive). Furthermore we adopted a version of Fox’s Late Merge (2004) that applies in the Numeration rather than in the narrow syntax which we labeled Late Num Merge. Two conceptual advantages to our approach immediately come to the forefront: First, we do not have nor require any non-cyclic applications of Merge (or any other operation for that matter) in our system. Second, Late Num Merge obviates the “tucking in” issue associated with Fox’s (2004) current formulation of Late Merge.

178 Thomas S. Stroik and Michael T. Putnam The implementation of Survive has far-reaching effects on a view of minimalist syntax. By removing Internal Merge from the system, economy constraints, the concept of phases and multiple Spell-Out are also deemed disposable due to the fact they are no longer conceptually necessary.

Acknowledgements Portions of this paper have been presented at LASSO 2005, DEAL 2005 and the Syntax Support Group Colloquy Series at the University of Michigan. In connection with these presentations, this paper has benefited from comments and discussions with Sam Epstein, Kleanthes Grohmann, Dennis Ott, David Pesetsky and Peter Sells. All remaining shortcomings and errors remain our own.

References Adger, David 2003 Core Syntax. Oxford: Oxford University Press. Brody, Michael 1998 Projection and phrase structure. Linguistic Inquiry 29: 367–398. 2002 On the status of representations and derivations. In Derivation and Explanation in the Minimalist Program, Samuel D. Epstein and T. Daniel Seely (eds.), 90–105. Oxford: Blackwell. Chomsky, Noam 1965 Aspects of the Theory of Syntax. Cambridge, MA: MIT Press. 1980 Lectures on Government and Binding. Dordrecht: Foris. 1993 A minimalist program for linguistic 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. 2001 Beyond explanatory adequacy. MIT Occasional Papers in Linguistics 20. Cambridge, MA: MITWPL. Collins, Chris 1997 Local Economy. Cambridge, MA: MIT Press. 2002 Eliminating labels. MIT Occasional Papers in Linguistics 20. Cambridge, MA: MIT Press. Epstein, Samuel D., Erich M. Groat, Ruriko Kawashima, and Hisatsugu Kitahara 1997 A Derivational Approach to Syntactic Relations. Oxford: Oxford University Press.

Surviving reconstruction

179

Fox, Danny 2003 On logical form. In Minimalist Syntax, Randall Hendrick (ed.), 82– 123. Oxford: Blackwell. 2004 Condition A and scope reconstruction. Linguistic Inquiry 35: 475– 485. Frampton, John and Sam Gutmann 2002 Crash-proof syntax. In Derivation and Explanation in the Minimalist Program, Samuel D. Epstein and T. Daniel Seely (eds.), 90 –105. Oxford: Blackwell. Frege, Gottlob 1884 Die Grundlagen der Arithemtik. Eine logisch mathematische Untersuchung über den Begriff der Zahl. Breslau: Koebner. Lebeaux, David 1991 Relative clauses, licensing, and the nature of the derivation. In Perspectives on Phrase Structure: Heads and Licensing, S. Rothstein (ed.), 209–239. San Diego, CA: Academic Press. 1995 Where does the Binding Theory apply? In University of Maryland Working Papers in Linguistics 3: 63–88. Department of Linguistics, University of Maryland. Putnam, Michael 2006a Scrambling in West Germanic as XP-adjunction: A critical analysis of prolific domains. Ph.D. Dissertation, University of Kansas. 2006b Eliminating Agree. Ms. University of Michigan. 2007 Scrambling and the Survive Principle. Amsterdam: John Benjamins. Putnam, Michael and Thomas Stroik in prep. The Survive Manifesto: A Derivational Guide to Syntactic Theory [tentative title]. Michigan State University and the University of Missouri/Kansas City. Rubin, Edward 2003 Determining Pair-Merge. Linguistic Inquiry 34: 660–668. Stroik, Thomas 1999 The survive principle. Linguistic Analysis 29: 278–303. to appear Locality in Minimalist Syntax. Cambridge, MA: MIT Press. Uriagereka, Juan 1999 Multiple spell-out. In Working Minimalism, Samuel D. Epstein and Norbert Hornstein (eds.), 251–282. Cambridge, MA: MIT Press.

On the interface(s) between syntax and meaning Anjum P. Saleemi

This paper addresses the well-known foundational issue of the tension between form and content (meaning in a wider sense of the term); more specifically, between morphosyntactic form and meaning. We assume syntax to be a gradual, combinatorial interface between the two broad interface systems, namely, the Articulatory-Perceptual interface (which is not dealt with here) and the Conceptual-Intentional interface. The latter is conceived of as something considerably more complex than is traditionally assumed, encompassing the derivational level known as LF (Logical Form) and a number of other aspects of meaning, in particular the conceptual-lexical one and the pragmatic-illocutionary one. So conceived, the semantic component contains most of the necessary elements required for basic interpretation, which are transformed into increasingly syntactic (and ultimately phonetic) representations as a result of adaptation to a language-specific lexicon, phrase structure, and various transformations. As part of the reconceptualization of linguistic theory outlined above, one of our major aims is to demonstrate that the framework in question is sufficient to successfully explain a range of specific phenomena, such as the conceptual basis of argument and event structure within v*P and the relationship between CP and illocutionary force. Our investigation reinforces the view that a major function of syntactic Phases is to repackage universal semantic representations into speech signals in a step-by-step, but at the same time parallel, fashion. In addition, it reveals that Phases represent the intersection between the purely syntactic and the broadly semantic categories of the human linguistic system, functioning as hybrid categories constituting the interfaces between form and meaning. Finally, adopting a more coherent idealization, a ‘neo-Saussurean’ version of minimalism, deploying linear or horizontal derivations, rather than the usual bifurcated ones, is proposed.

1. Introduction What follows is an attempt to make some contribution to the theory of interfaces and Phase theory, both part of the Minimalist Program (e.g., Chomksy

182 Anjum P. Saleemi 1993, 1995a, 1995b, 2000, 2001). We have two related types of goals in mind: the programmatic-conceptual ones and the empirical ones. The accuracy of the former clearly must remain subject to further support and verification from data, as in most cases of scientific inquiry. Almost entirely focusing on the syntax-meaning interface, often known as the ConceptualIntentional interface, we do not have much to say regarding the Articulatory-Perceptual interface, which involves the phonological-phonetic system as well as the actual sensori-motor mechanisms of speech. A relatively recent addition to the Minimalist Program (henceforth MP), is Phase theory – PT for short (Chomsky 2001, 2004, 2005, to appear),1 which is an attempt to bring the issues bearing on the ‘psychological’ validity2 of linguistic constructions to the forefront of inquiry, assuming that packets or quanta (i.e., Phases, or Domains of Phases) of derived material are transferred to the two interface systems, namely, the Conceptual-Intentional interface (C-I ≈ LF) and the Articulatory-Perceptual interface (A-P ≈ PF), as soon as they acquire legibility in the relevant technical sense. Thus computational complexity and load are important motivating considerations in MP and PT, as is a renewed search for the identity of real linguistic units at a level of abstraction that strives to reduce structural categories and relations to a bare minimum (Chomsky 1995b), a line of inquiry explicitly initiated in Chomsky (1993), though it roots can be traced at least as far back as Chomsky (1955 /1975). However, Phases, understood as a subset of ‘proposition-like’ syntactic categories (i.e., v*P and CP), are there not only to decrease the computational load on working memory, but also to ensure mutual intelligibility and connectivity between form and meaning, or, more appropriately in the present context, between syntax and meaning. Ever since the Extended Standard Theory (EST) and its Y-model of derivation started to undergo radical changes and the elimination of D-structure and S-structure was proposed, it has been commonly understood that while the interface between syntax and A-P is relatively well-defined, C-I cannot just be equated with LF; thus Chomsky states that while “The level A-P has 1

2

Phases are not without historical antecedents, as Boeckx and Grohmann (to appear) rightly point out; see also this work for some constructive critical remarks. The processing/functional considerations in regard to Phases are important, but it is most likely that they exist as a result of phylogenetic change, rather than in an online and ontogenetic sense. The phylogenetic structure of language, moreover, might have been determined by a range of factors in addition to those bearing directly on functional efficiency; see Chomsky (2005), and Hauser, Chomsky and Fitch (2002) for some specific remarks and a general discussion.

On the interface(s) between syntax and meaning

183

generally been taken to be PF; the status and character of C-I have been more controversial” (1993: 2; see Grohmann, to appear, for some suggestions regarding PF). Indeed, it has even been suggested recently that “…the final internal level LF is eliminated, if at various stages of computation there are Transfer operations”… that deliver the already constructed Phases “to the semantic component, which maps it to the C-I interface” (Chomsky, to appear: 9). Such a view of C-I resonates well with the viewpoint outlined in the present work, one of whose aims is to demonstrate that the syntactic system as a whole is a complex set of interfaces between the two other primary or ‘outer’ interfaces of meaning and spoken form, and to suggest that the term ‘interface’ should not be confined to a single derivational level, particularly insofar as C-I is concerned. The other issues addressed in the following pages range from some of those that pertain to the inner working of the derivational cycle, but more importantly, to those that belong to a certain larger project that has not been paid sufficient attention to in the research; namely, that of defining and explicating the ‘outer’ interfaces, in particular the intentional part of C-I, the illocutionary (or pragmatic) aspects of meaning not having received the attention that they in principle deserve. It must be pointed out that our major focus, strictly speaking, is on illocutionary force as a unit of meaning, not on illocutionary acts or speech acts3 which belong more to the A-P interface, even though they too must originate in the first instance as mental events. We emphasize the crucial role of the illocutionary aspects of meaning in addition to the conceptual-propositional ones, arguing that the semantic interface should in fact be treated as a many-layered and composite system. Further, it is suggested that the relevant elements from illocutionary logic and speech act theory (Searle and Vanderveken 1985, henceforth S&V; 4 see also Searle 1969, 1979) should be incorporated into C-I to function as the pragmatic-semantic counterparts of the higher level(s) of the CP structure. In other words, the periphery of the CP Phase parallels the illocu3

4

Speech acts are the units of human communication, and illocutionary acts are minimal units of this kind. Further, an illocutionary point, one of the seven components underlying an illocutionary act, is the “purpose which is internal to it being an act of that type” (S&V: 13). See S&V and the other related work for further explanation. We rely in particular on this work because it embodies an illocutionary logic that is thoroughly worked out, but we do not take for granted all that it offers, nor do we consider it definitive. Consequently, later in this paper (see section 5.2.) some modifications to the theory presented in it are suggested.

184 Anjum P. Saleemi tionary force intended by the speaker, much as conceptual-lexical structure serves as the foundation of the basic propositional content of the v*P Phase. Also emphasized is the need to alter the derivational model currently in use in MP, typically assumed to bifurcate at SPELL-OUT into LF and PF, by ‘linearizing’ it. In essence such a perspective is by no means a radical departure from the standard minimalist approach, but it is innovative to the extent that it (re)introduces a ‘neo-Saussurean’ (Saussure 1916 /1986) element in regard to the way the linguistic interfaces are interrelated. In general, our approach is meant to be closer to MP in spirit than much other work, which remains confined to the syntactic execution of things within pre-set limits. In a sense what we try to accomplish is to delineate a perspective based upon and conceptualized within the MP, though in part the latter is also viewed from without, as it were, with the intention of attempting to further those of its objectives which remain by and large unrealized.

2. The C-I interface: Some fundamental issues We believe it is important to recall and reconsider the background of the form-meaning relationship within a broader conceptual and philosophical perspective, in part since any mention of illocutionary force and speech acts immediately brings to mind various ill-fated attempts within generative semantics, in particular the performative deletion analysis, to bring speech acts into syntactic derivations, or in general to reduce meaning to syntax (see Ross 1970; Gordon and Lakoff 1971/1975)5, not to mention some linguistic approaches which believe that the opposite aim, i.e., that of reducing syntax to meaning, should be held to be paramount. Thus it should be clearly stated that, unlike generative semantics, our approach does not involve the reduction of the illocutionary level of meaning to technical devices traditionally characteristic of transformational-generative syntax, nor indeed do we wish to achieve the converse. Some preliminary general remarks seem to be in order here to establish the relevant conceptual and philosophical backdrop. Thus, an apparent analogy can be drawn between the familiar mind-body problem and the problem of form vs. content, with the crucial difference that the latter problem involves entities both of which are mental. At any rate, an enormous gap has existed within each of these seemingly antithetical pairs of entities. 5

See Searle (1979) for some critical remarks on this enterprise.

On the interface(s) between syntax and meaning

185

One of the prevalent opinions is that the relationships between the members of each pair do not involve eliminative reduction, but are very likely to eventually boil down to mutual co-existence, perhaps in the end turning into some sort of unification (see Saleemi 2005, and the references cited therein, for a general discussion). It is noteworthy that in the case of the mind-body problem historically it is the concept of matter, one of whose manifestations is the body, that has undergone radical changes, not that of the mind, such that the former has expanded to include energy, fields of force, and the like. The distinction between form and content in human sciences, like the philosophy of meaning and linguistics, appears not to have evolved to the point where the validity of a certain conception of form could be addressed without recourse to its reduction in the direction of meaning, as in the case of generative semantics, or vice versa, as still remains the case in various kinds of cognitive or functional approaches to linguistics. Although the concept of form (particularly morphological, syntactic, and phonological) has admittedly kept undergoing changes even within a single paradigm like the generative one, most of its manifestations continue to retain an overextended concept of form. At least in principle, a notion of syntactic form constrained by “virtual conceptual necessity” (Chomsky 1993), MP, PT, and the theory of interfaces together seem to be a step in the right direction, as they aim to minimize or contract the scope of syntactic form, while at the same time to expand the role of the interfaces, without any tacit or open attempt at absolute reduction. This should make it easier to tease out form, on the reasonable assumption that any sort of reduction is impossible, and that syntactic form is not meaning in some elaborate disguise, but an independent vehicle intended to make the expression of meaning possible, with the two held together in an intricate fashion by a host of mapping devices. Consider, for example, the following three simple logical possibilities in respect of the form-meaning relationship, partially reiterated here: (1)

a. Reduction of form to meaning. b. Reduction of meaning to form. c. Redefinition of the concept of form, resulting in an expansion of the interfaces.

It is the clearly the third possibility (1c) that should be considered to be more viable in MP. In contrast, neither (1a) nor (1b) can explain the relative grammaticality of an expression like (2), a sentence containing non-sense lexical items but retaining minimally correct morphosyntactic form, nor can

186 Anjum P. Saleemi it account for examples such as (3), for similar reasons, or for that matter for the oft-cited example in (4), in which the lexical items are individually meaningful but mutually incoherent. In should be mentioned here, though, that most of the functional morphemes employed (-ed, -s, it, is, that, were, by) are interpretable, serving as anchors maintaining basic links between syntax and meaning, and that even when semantically bleached uninterpretable items (pleaonastic it and there, for example) are used, as in (3), they could well be non-randomly related to their cognate interpretable items. (2)

Bleep tinged shoops.

(3)

It’s shoops that were tinged by bleep.

(4)

Colourless green ideas sleep furiously.

On the other hand, in (5), an instance of random scrambling employing meaningful English lexical items, the interpretation remains recoverable although the form (basically, the word order) is illegitimate, again indicating the relative independence of meaning from pure syntactic form. (5)

*crime is poverty that it breeds well-known.

Let us emphasize once more that none of these examples is totally ungrammatical or meaningless. Now look at the following contrast. (6)

It’s well-known that poverty breeds crime.

(7)

*It be well-known that poverty breeds crime.

Here (6) is the grammatical version of (5), and (7) an ill-formed one, as in it the tense is missing. It can be postulated that sentences like (6) result from a lexical array (LA) that is conceptually coherent and ordered in conformity with the canonical word order of a given language (modulo transformational intervention),6 that (7) exemplifies the non-application of proper V-to-T movement (or, equivalently, feature matching followed by deletion), and that (5) is simply the outcome of the unordered insertion of the lexical items.7 6

7

Formally speaking, an array is an ordered set of related elements, or an arrangement of quantities and symbols in rows and columns. The reasoning employed here, and the investigation of (un)grammaticality, can be carried out in a more fine-grained manner, for example to incorporate various

On the interface(s) between syntax and meaning

187

Note that nothing in this conceptualization bars the generation of deliberately unordered LAs. As we know, Phases (CP or v*P) are considered to be propositional (e.g., Chomsky 2004). However, it may be the case that Phases are not merely propositional, but also have an illocutionary force associated with them. That propositional expressions can exist without a typical illocutionary force associated with them is trivially illustrated by innumerable run-of-the-mill expressions cited as linguistic examples in “out-of-the-blue …contexts” (Rizzi 2004: 238), such as (8). (8)

The cat sat on the mat.

That is to say, the syntactic engine can be made to work completely or partially, leading to PF strings that are well-formed, or which crash for one or more reasons. Thus the degrees of grammaticality and semantic coherence can be more or less independent of each other, depending on the extent of the intersection between the two, which is what we really consider to be the meaning-syntax interface, of which what is known as C-I forms the basis. Clearly, under MP the concept of syntactic form must be reduced in order to better accommodate elements that reside in this intersection, which by definition should consist solely of such interface devices. Any changes along these lines must also affect the status of argument structure (see Levin and Rappaport Hovav 2005 for a survey) and event structure (Rothstein 2004), both a significant part of the interface between the conceptual-lexical system and syntax, as is evident, e.g., in Hale and Keyser (1993, 2002: see section 5.1. below). However, a notable desideratum in the currently prevalent MP model, as already implied, is that although it does take on board the conceptual aspects of C-I insofar as argument and event structure is concerned, the illocutionary part of the picture has received scant attention in it, in spite of the not infrequent complimentary references in the literature to entities like the language of thought – in the sense of Fodor (1975) – propositional attitudes, and “mentally constructed” speech acts (see Chomsky 2000: innovative deviations in literary style, as stated by Chomsky (to appear). To quote: “Merge can apply freely, yielding expressions interpreted at the interface in many different kinds of ways. They are sometimes called ‘deviant’, but that is only an informal notion… And expressions that are ‘deviant’ are not only often quite normal but even the best way to express some thought; metaphors, to take a standard example… That includes even expressions that crash, often used as literary devices and in informal discourse, with a precise and felicitous interpretation at the interfaces.” (Chomsky, to appear: 11.)

188 Anjum P. Saleemi 91, 94; 2005a: 4, and indeed elsewhere), testifying to the importance of the intentional and illocutionary factors. To sum up, since illocutionary force is as integral a part of C-I as conceptual structure, any reformulation of the form-meaning interface should include the interrelationships between syntax and both these components of meaning. Needless to add, we consider the philosophical underpinnings outlined above to be crucial to the proper execution of the theory of C-I. Generally speaking, they are not at all incompatible with MP, because minimizing the role of syntax entails concomitantly maximizing the role of the interface mechanisms.

3. Propositional and illocutionary meaning Conceptual-lexical structure builds up expressions primarily in a bottom-up manner, whereas illocutionary structure primarily serves as a top-down constraint on the generation of expressions, with syntax (and linguistic form in general) intervening between the two as a formal interface. The former generates skeletal v*P structure, primarily conceptual-propositional in nature; in contrast CP is essentially illocutionary-propositional in nature, designed to interface with the illocutionary basis of an expression. Moreover, CP functions through the mediation of TP, and possibly some other bridging structures, such as AspP, extending the basic proposition to incorporate tense, modality, aspect, and voice. The alliance between syntactic mechanisms (call them S) generating a sentence (CP) and the related combination of illocutionary force (F) and propositional content (P), may be expressed as the simple schema below (based on S&V), where ↔ represents the interface between the two levels, namely: (i) P and F per se, and (ii) their syntactic expression in the form of CP. (9)

F(P) ↔ S(CP)

The point is that it is F(P) that is largely responsible for the selection of appropriate syntactic mechanisms, as well as lexical items, that are available in a particular language. Each expression is to be understood as having a purely conceptual, intentionally decontexualized interpretation, besides an intentionally contextualized interpretation. The italicization here is meant to emphasize the difference between intrinsic (mental) and extrinsic (situational) discourse determinants, a distinction that is often ignored. The initial step in the generation of an expression is the correct selection of an LA, that is, the set of lexical items and features constitutive of that

On the interface(s) between syntax and meaning

189

expression. An LA may be directly grounded in a speaker intention, embodied in its illocutionary force (henceforth, IF) and conceptual structure (CS), as is the case when language is used in normal situations, such that the mapping to LA yields the intended expression. The IF of an expression may be related to it only vacuously, in the sense that no specific type of it is identifiable, as exemplified by expressions cited out of context. In the latter case, an LA embodies a preposition P whose truth or otherwise may be subjected to investigation regardless of any reference to IF, in order to ascertain if it hangs together, i.e., is inherently coherent as a preposition in respect of the overall conceptual relationships among the individual items contained in the LA in question. As pointed out above and will be discussed further below, in natural conditions the composition of an LA will also be contingent upon its illocutionary purpose, depending on what is intended by a particular speaker; for example, the speaker may want to make a statement, ask a question, or convey his surprise about the content of a P. Therefore, the IF of an LA, even its components (see S&V: 87ff.), may be a crucial factor in demarcating the membership of the LA, eventually determining the clause type of the resulting expression as well. To state the obvious, CS (as described, e.g., in Jackendoff 19838, but see Pustejovsky 1995 for an alternative, more generative view of the conceptual bases of the lexicon 9), typically realized through the lexicon, being one of the two major components of C-I, is responsible for expressing, among other things, the argument and event structure of an expression. It contains constraints which are internal to it, so that each concept is often tied up with several others (but note that these constraints have their origin in factors which are ultimately intentional, as explained below). To return to the question of lexical selection, our conception of it relies upon both the components of the C-I system, and is conceived as a two-fold but necessarily overlapping process, involving illocutionary selection (driven by IF) as well 8

9

A conceptual view of the lexicon, reaching down to the sublexical elements, appears to have implications for syntactic operations that are far from trivial, as for example demonstrated by Culicover and Jackendoff (1995) in regard to binding theory. An example of a conceptual-lexical generative mechanism proposed in Pustejovsky (1995) is lexical inheritance structure, which may be considered to incorporate relations between a given lexical item X and the set of nodes in a semantic lattice that X is related to up to the nth degree of remove from itself, thus defining the range over which (for instance) co-compositionality and type coercion can operate.

190 Anjum P. Saleemi as conceptual selection (based upon CS), on the assumption that both IF and CS are necessary for an LA to be selected and an expression to be generated. Let us assume that CS primarily selects a (sub)array LACS that is appropriate for the basic elements of P, and IF selects another (sub)array LAIF, not necessarily disjoint from LACS,10 that inserts items relevant to the illocutionary factors. In general the selection of items contained in LAIF, and indeed the related syntactic structures, is based upon a search for the Illocutionary Force Indicating Devices (S&V), i.e., the linguistic correlates (as opposed to the mere propositional-content expressing devices) best suited for the expression of IF. These may be clause types, illocutionary verbs, or a variety of other elements. Evidently, illocutionary features are not confined to any particular part of syntactic structure, let us say, to the Edges of Phases. While recognizing a certain amount of tension between conceptual and illocutionary meaning, it should be made clear that, strictly speaking, even propositional meaning is (ultimately) intentional, either intrinsically or derivatively (see Searle 1990 for a clarification of this distinction11), so we consider C-I to be basically intentional. Consequently, both CS and IF are major components of C-I (= intentional meaning), interrelated in the manner shown in (10). (10)

C-I

CS

IF

In other words, CS largely comprises the micro-elements of the propositional content of intentions, and IF typically consists of the macro-aspects of meaning, i.e., those related to types or classes of illocutionary intentions. Conceptual categories, however, have to be cognitively antecedent to the il10

11

For instance, illocutionary verbs (order, declare,…) have both conceptual-propositional and illocutionary content. Briefly, on Searle’s (1990, and much subsequent work) view the distinction between intrinsic or inherent and derived intentionality hinges on the former being an inalienable characteristic of the human mind, of which linguistic and most other kinds of representations are a derived and externalized reflection. In other words, a sentence is intentional only by virtue of the fact that it can be uttered by a human speaker in some context or other.

On the interface(s) between syntax and meaning

191

locutionary ones, since it is they which form the fundamental defining units of intentionality and thought. IF categories are thus inevitably couched in the terminology of CS, forming as they do a special subset of the latter – the major reason why CS and IF appear to be somewhat deceptively indistinguishable from each other. However, IF categories are special because they are particularly relevant to the expression of the illocutionary aspects of intentional thought.

4. Illocutionary and syntactic force Among the Illocutionary Force Indicating Devices, what are traditionally known as clause or mood types are especially important for our purposes (though illocutionary verbs and certain other types of lexical items also play a significant role here). As a result, it is significant to figure out the interface system between particular types of illocutionary points (see footnote 3) at the clause level and the syntactic clause types, in other words, between IF and syntactic force (henceforth, SF). Not unlike the number of θ-roles in the theory of argument structure, the range of fundamental types of either SF or illocutionary points of IF is strikingly limited. It is generally agreed that there are five or so basic clause or mood types (see Huddleston 2002, Allan 2006; cf. Chomsky 1955/75: 301ff.12). Comparably, S&V’s classification of the fundamental types of IF consists only of the following five categories: assertives (which state the way things are), commissives (which express one’s commitment to doing something), directives (that are meant to get other people to do things), declaratives (which bring about changes in the ‘world’ through one’s utterances), and expressives (aiming to convey one’s feelings and attitudes). Likewise, Allan (2006) identifies the following five categories of what he calls “typical primary illocution” of clauses: declarative, interrogative, imperative, hypothetical, and expressive. Note that according to the framework followed in this paper “Questions are always directive, for they are attempts to get the hearer to perform a speech act” (S &V: 1999). 12

Interestingly, Chomsky (1955/75: 301) remarks that, “… had we given a more complete grammar…, the first statement of this grammar would have had to be  Imperative Sentence   Interrogative Sentence  Sentence →  Declarative Sentence    :  .  …”

192 Anjum P. Saleemi A thorny issue that yet remains to be resolved is what the exact nature of the algorithms involved in IF ↔ SF is. As is well known, the relationship between the types of SF and IF is hardly ever one-to-one, but is usually oneto-many or many-to-one, as partially exemplified below, where (11) contains syntactically distinct sentences all expressing directives (e.g., requests, 11a–b; commands, 11c–e), and the examples in (12) show the same basic sentence or clause type, namely (Yes/No or Wh-) interrogative, used to express two different IF types, i.e, directive (12a–b) and expressive (12c–d); note that (11b), a directive, is an interrogative too. (See Lin and Saleemi 200513; also Saleemi 1998.) (11) a. b. c. d. e.

Please tell me the name of this animal. Would you tell me the name of this animal? Tell me the name of this animal. You WILL tell me the name of this animal. You WILL tell me what the name of this animal is.

(12) a. b. c. d.

Is this your house? Who lives next door? Is this a dance or what? Where in the world can you find such enemies?

As is evident, particular kinds of IF and SF are interrelated in a rather complex fashion; thus, while discussing types of illocutionary acts, it has been stated that “Often one can do more than one of these things in the same utterance” (S&V: 52, emphasis ours). It seems to us that this happens to the extent that changes in a speaker’s illocutionary perspective can be assumed to take place ‘online’. Although in this context it is customary to talk about direct and indirect speech acts, and derived illocutionary force, the fact is that illocutionary acts may shift from one type of IF to another, depending on the speaker intentions and the context of utterance, somewhat like the type shift that word meanings undergo when they cooccur with one set of words rather than another, what Pustejovsky (1995) refers to as “sense-incontext”. Therefore, we venture to propose that, beyond the fundamental IF types, illocutionary acts are as generative in nature as are syntax proper or the lexicon (see Pustejovsky 1995 on the nature of the lexicon). However, the IF ↔ SF relationship may not be entirely unsystematic. For one thing, some combinations tend to be more characteristic than others 13

See Lin and Saleemi (2005) for further elaboration of this point.

On the interface(s) between syntax and meaning

193

(cf. Allan 2006, Huddleston 2002: 853ff. for a discussion); see the following correlations. (13) a. b. c. d. e.

Assertive ↔ Indicative Commissive ↔ Indicative Declarative ↔ Indicative Directive ↔ Indicative ∨ Interrogative Expressive ↔ Indicative ∨ Exclamative

The recurrence of the indicative SF points to another reason why the IF ↔ SF mapping may not be random. Generally speaking, only a relatively small subset of the set of all logically possible linkages between the members of the two sets is actually attested. Moreover, on the illocutionary side, assertives are apparently the most general or basic class, perhaps because they are held to be truth-functional (at least according to S&V), in comparison with commissives, directives, declaratives, and expressives, whose truth value at best can be only taken for granted as it is tied to the speaker’s viewpoint. On the linguistic side, indicatives appear to be the most general or superordinate clausal form; arguably, they, too, are the most directly truthfunctional of the clause types. At the other end of the spectrum, the declarative IF and the exclamative SF happen to be the most impoverished types, having a rather limited range of uses. (See Lin and Saleemi 2005 for further detail.) Regardless of the enormous difficulties involved in systematically relating IF and SF, various parallels between syntax in general and IF are not too hard to locate: consider, for example, the following examples of negation (14–16) and conditionals (17–18), all taken or adapted from Lin and Saleemi (2005). (14) [I promised [ that I’d come ]]. (15) F(~P): I promised not to come. (16) ~F(P): I didn’t promise to come. As is well-known, negation is understood as having scope over a certain domain. Thus, (15–16), the two negative counterparts of (14), which contains the commissive verb promise, are supposed to differ in respectively exhibiting narrow and wide scope, a structural difference that can also be captured, in fact springs naturally from, the deeper intentional distinction between whether the speaker’s intention is to negate the embedded proposition

194 Anjum P. Saleemi (15) or the IF of the root clause (16), i.e., the distinction between propositional negation (15) and what S&V term denegation (16). As to the following conditionals used in conjunction with bet, a commissive verb (both 17–18 are slightly adapted from S&V: here C = conditional operator), scope is operative here much the same way that it is in negation, depending on whether a sentence expresses an IF whose propositional content is conditional (17), or whether it is the IF itself that is conditional (18): “a conditional illocutionary act is not performed categorically but only on the condition that the proposition P that is its antecedent is true” (S&V: 157). (17) F(C(P)): I bet you five dollars that if a candidate gets a majority of the electoral votes, then he’ll win. (A bet on a conditional proposition.) (18)

C

(F(P)): If Bush is the next Republican candidate, then I bet you five dollars that the Democrats will win. (A conditional bet.)

To return to the relationship between IF and SF, a unique syntactic projection headed by Force has been proposed as part of the CP system by Rizzi (1997; also see Chomsky 1995). Rizzi’s (1997) formulation of ForceP is illustrated below (cf. Rizzi 2004)14. (19) … [ForceP [ that [TP she gave [v*P [VP [VP the bottle to the baby] full.]]]]] As Rizzi states, “We can think of the complementizer system as the interface between a propositional content (expressed by the IP) and the superordinate structure (a higher clause or, possibly, the articulation of discourse, if we consider a root clause). As such, we expect the C system to express at least two kinds of information, one facing the outside and the other facing the inside.” (Rizzi 1997: 282, the emphases added). However, despite an explicit recognition of the role of “the articulation of discourse”, Rizzi makes no attempt whatsoever to consider in any detail the kind of information “facing the outside”, an issue which we feel must be addressed, and will therefore be taken up again in greater detail in section 5.2.

14

See Rizzi (2004) for an even more elaborate version of the split-CP hypothesis, under which the C system has a considerably greater number of components.

On the interface(s) between syntax and meaning

5.

195

Phases and interfaces

Let us now take a closer look at the semantic interface(s) and the derivational model, reconsidering their role and reconceptualizing them, especially in terms of the commonly recognized Phases (i.e., v*P and CP), and consolidating and introducing some supportive arguments and empirical evidence in the process.

5.1. Phase one: the verbal complex Some aspects of the syntax-meaning interface have been studied relatively well, though they are often not treated as such. For instance, lexical-conceptual structure (e.g., argument structure and event structure) can itself be taken to be an interface between the lexicon and syntactic derivations, or between the CS part of C-I and the syntactic structure of and the lexicon. Recall that CS provides the basic propositional input to derivations, in collaboration with that provided by the outer illocutionary interface in C-I, namely, IF. As Chomsky points out, “The phases have correlates at the interfaces: argument structure or full propositional structure at meaning side, relative independence at the sound side. But the correlation is less than perfect, which raises questions” (2005: 18). In any case, conceptual interdependence among lexical items, based on a generative view of the mapping between concepts and morphological entities, could be instrumental in defining the notion of Phase, a proposition-like construct that, it will be suggested, combines syntactic and semantic characteristics, as is partly apparent from the fact that Phases are distinct from purely syntactic categories such as VP, vP, v*P, TP and CP, or their bare counterparts in MP. For an illustration of this point at the level of the verbal Phase (v*P, to be precise), take, Hale and Keyser’s (1993, 2002) ‘bare’ view of argument structure. Commendable though such an enterprise devoted to finding the structural or templatic correlates of meaning is, it is far too obvious that it can explain the facts only within a limited range. As Hale and Keyser (2002) point out, contrasts such as the one exemplified in (20) – from Hale and Keyser (2002: 24) – compel one to look for factors other than those definable exclusively in terms of bare arguments structure. (20) a. b. c. b.

The kids splashed mud on the wall. Mud splashed on the wall. The kids smeared mud on the wall. *Mud smeared on the wall.

196 Anjum P. Saleemi Additional semantic features, in this case manner features of splash and smear, have to be invoked to help explain not only this particular contrast between verbs whose argument structure is otherwise identical. Patently, the situation in regard to figurative expressions must require even further modifications in the concept of selection. Even if delimited to nonfigurative meaning, selectional factors, we believe, often go well beyond simple configurational and thematic matters, to the extent that the items selected may alter each other’s meaning compositionally (cf. the notion of “sense extension” in Pustejovsky 199515). To reemphasize the propositional character of v*P, it should be fruitful to consider an example of the interaction between verbal prefixation and event structure. The meaning of the verbal prefix re- (as in repaint), it has been claimed, does not connote the exact repetition of an event on the basis of data like the following (from Marantz 2005). (21) a. b. c. d.

They repainted the walls. *John re-put the books on the table. *John re-smiled. John smiled again.

On this view, re-prefixation, contrary to claims to the effect (e.g., Lieber 2004), appears not to be semantically equivalent to the adverbial modification by again,16 and consequently Marantz (2005; cf. Keyser and Roeper 1992) claims that re- in fact originates as a nominal prefix, having scope over it, and appears when the nominal in question is of a certain change-ofstate, eventive type, for example cake, which does not admit a simple iteration of the action denoted by a verb like rebake; after rebaking, the outcome is never going to be the same cake. Such an approach is clearly both counterintuitive and problematic because re- does after all seem to be a verbal affix, and it can in fact mean simple repetition of an action or event, as for instance in (22), or have other connotations when used metaphorically or idiomatically (23). 15

16

More generally, see Pustejovsky’s (1995) distinction between semantically “monomorphic” and “polymorphic” languages, and his evidently correct characterization of natural languages as necessarily belonging to the latter class of languages in terms of their expressive power. But note that there are languages, such as Urdu(-Hindi), in which a prefixal equivalent of re- is not available, and therefore appropriate adverbial modification is the only option.

On the interface(s) between syntax and meaning

197

(22) a. John redialled the number. b. Please reenter your password. (23) a. I must repay your generosity at some point. [i.e., ‘pay back’ the generosity.] b. I hope the outcome will repay the efforts. [i.e., the efforts will be rewarded.] An alternative could be to consider the meaning of re- to range from mere iterative action of event to change of nominal state, and indeed well beyond, with the exact fixation of the sense to be determined in context, i.e., in composition with the nominal, verbalized through conflation (as in denominal verbs or adjectives) or otherwise, such that the referential nature of the nominal influences the interpretation of the verbal meaning. The relevant semantic conditions may be met simply by postulating that re- occupies a verbal head position immediately above the lexical verb, which the latter moves into. Thus: (24) [v*P… [V re- [V+N V [ N]]]] This will enable the prefix in question to have scope over both V and N. One consequence of this suggestion will be that the semantics of lexical items is compositional, but only in a complex, co-compositional sense (cf. Pustejovsky 1995). Further, systematic sense extension plays a significant role in trying to find an LACS that is as close as possible to LAIF. To conclude, the driving force behind many such phenomena is the intentional nature of meaning, which is not averse to finding ways to exploit a language as much as possible in order to achieve maximum approximation to the intended meaning. Given the fact that purely syntactic structures such as VP/vP/v*P, and the corresponding propositional complexes, can achieve a full interpretation only when the totality of an LA is taken into account, and that choices regarding the thematic and conceptual appropriateness of the subject, or the insertion of expletive elements, if required, can be made only after the conceptual meaning has been computed, we suggest a modification in PT. We postulate that Phases, especially their Edges (as distinct from their purely syntactic counterparts), mediating between meaning and syntactic structure as they do, are themselves interface devices; in other words, Phases are in fact interfaces, which in the ideal case must obey the conditions imposed by both syntax (or linguistic form in general) and C-I; let us call this view

198 Anjum P. Saleemi the hybrid thesis of Phases. A natural corollary of this thesis is that some syntactic constraints could be overridden by certain C-I constraints just in case (a given segment of) an expression is not a Phase, or the expression in question is fragmentary and, therefore, as before, does not constitute a Phase. This can help account for some annoying exceptions to syntactic principles without having to modify the latter in an increasingly (and often self-defeatingly) complex manner, or to treat such exceptions as mere discourse-related oddities. To summarize, the v*P Phase contains both structural and intentional information (pertaining to CS or IF) in the form of Edge features (EF, for short), so it may be thought of as an entity of the following type, with ↔, as before, representing the mapping mechanism(s), and AE standing for argument and event structure. (25) [v*P {EF [CS] ↔ [AE]}… ] It is hoped that an extension of this view of Phases to CP, to be undertaken in the following section and naturally assumed to explicitly incorporate the illocutionary component of meaning as well, will further support the assumptions underlying the hybrid thesis of Phases.

5.2. Phase two: The clausal complex Evidently, the clausal equivalent of (25) should look like (26). (26) [CP {EF [IF] ↔ [SF]} [TP…]] The usefulness of (26), or of (25) by extension, can be illustrated by citing some data from Urdu(-Hindi), a typically head-final language, except when the complement CP is finite (see Davison 2007 for a discussion of this split in the wider context of South Asian languages). Besides, Urdu allows a good deal of scrambling.17 The data in question contain ‘right-dislocation’ of a CP element, a process that is supposed to take place before the Domain of the CP Phase is sealed off, due to the Phase Impenetrability Condition of PT (Chomsky 2000), and its contents sent off to A-P, or even before that if the constituent involved is not a Phase; for instance, is a DP or a PP. Consider the contrast between (27) and (28): in (28) the DP appearing at the right Edge has been either moved or copied from inside the TP, and 17

We put aside any cases involving further scrambling in this language.

On the interface(s) between syntax and meaning

199

arguably right-adjoined to it, whereas it appears in situ in (27); here much structural detail at both the verbal and the clausal level is suppressed. (27) us nee kahaa [CP ki [TP aaj [DP doonoN umiidvaaroN ki s/he ERG said that today both candidates Particle darmian mukaabla] hoo gaa]] between competition be will ‘Today there’ll be a competition between the two candidates.’ (28) us nee kahaa [CP ki [TP [TP aaj hoo gaa] [DP doonoN umiidvaaroN ki darmian mukaabla]]] If we consider the basic structure of (28) to be (29), it seems reasonable to claim that the scrambling out of the TP is triggered by the presence of an intentional Focus feature at its right Edge, sitting with a corresponding rightdislocation syntactic feature, presumably both inherited from the CP. Thus, contrary to Rizzi (1997, 2004), it is claimed that Focus is an intentionalillocutionary, rather than a syntactic, feature because it is a manifestation of the degree of strength18, one of the essential components of IF (S&V: 15, 98 –100) (29) [CP [TP [TP…] {EF [IF Focus] ↔ [SF Right Adjuction]}]] This is not mere terminological shift or notational variation, but part of a general attempt to reduce redundancy in the study of various aspects of language. Its particular significance in the present context is that such reclassification of features will reduce the ever-expanding vocabulary and structure of syntax, whose mechanisms should merely specify the exact variety of the movement or copying operation available in a language (in this case, Internal Merge to a right-adjoined position), as dictated by the true spirit of MP. The alternative, of course, is to make syntax increasingly elaborate by introducing non-syntactic, even pragmatic, categories into in, as, for instance, is done in Rizzi’s (1997) split-CP hypothesis where the sentence or the clause (CP) is claimed to have an elaborate structure of the following sort, a move closer to generative semantics than to MP. (30) [ForceP [Force] [TopP [Top] [FocP [Foc] [FinP [Fin]]]]]

18

Compare the illocutionary verbs suggest and insist, both directives but differing in the degree of strength of their IF.

200 Anjum P. Saleemi To consider some further empirical evidence reinforcing our hypothesis, an interesting case can be made about the way syntactic and illocutionary factors interact, with empirical consequences for syntax, by considering the distribution of the English word any, which can appear both as a so-called Negative Polarity Item (NPI), licensed by an explicit negator like not, and as a lexical item with a ‘free-choice’ (FC) interpretation, almost like a positive-polarity universal quantifier; observe the contrasts between (31) and (32), and between (31a) and (31b), all assertive-indicative sentences: (31a) is grammatical, but (31b) is not. (31) a. He didn’t buy any books. b. *He bought any books. (32) a. Any man can write a book. b. Almost any man can write a book. c. He bought any books he wanted.

[NPI]

[FC]

This obviously can imply that any is in fact two semantically distinct but morphologically isomorphic lexical items. An alternative possibility is that any is a single lexical item, but that it inherits its polarity (positive or negative) in a given context from a negator like not and (ultimately) from the P of the IF. If the latter option is taken, it will mean, by assumption, that the basic representational schema of (31a) and (32) is as follows, where the illocutionary-propositional feature set (in IF) and the syntactic feature set (in SF) are both marked for polarity (±Pol). Here S = sentence, A = assertive, and I = indicative. (33) [CP {EF [IF ±Pol(A(P))] ↔ [SF ±Pol(I(S))]} [TP… ]] If the approach illustrated in (33) is adopted, well-formed examples of the type of (31a) and (32), as well as the ill-formed (31b), can be accounted for: both (31a) and (32) are marked for positive polarity by the speaker’s choice, as implied by (33) and shown more specifically in (34). (34) [CP {EF [IF +Pol(A(P))] ↔ [SF +Pol(I(S))]} [TP… ]] Accordingly, the representation for (31a), where any receives a negative interpretation, is given in (35). (35) [CP {EF [IF –Pol(A(P))] ↔ [SF –Pol(I(S))]} [TP… ]]

On the interface(s) between syntax and meaning

201

To return to the ungrammatical (31b), we suggest its underlying representation is not marked for either positive or negative polarity at all, as shown in (36), and its EF is therefore incomplete. As a result, since any is sensitive to polarity, but not to specific negative or positive polarity, and given the LA the sentence contains, any cannot receive any interpretation in (31b). (36) [CP {EF [IF (A(P))] ↔ [SF (I(S))]} [TP… ]] To sum up the observations made so far, any can be construed either positively or negatively, or not at all, depending on the relevant choices made in EF and in the selection of the LA correspondingly deployed. Comparably, directive-imperative (37), commissive-infinitive (38), and declarative-indicative (39) contexts permit either FC or NPI any, much like assertives (31a–32) under our analysis of them (see 33, above): here the (a) examples are FC, while the (b) examples are NPI. (37) a. Buy almost any book you want! b. Don’t buy any book sold here. (38) a. I promise to borrow any book on syntax. b. I promise not to borrow any book on syntax. (39) a. I declare you the president of almost any country you like. b. I don’t declare you the president of any country. Therefore the explanation of these contrasts, too, could fall under the schema in (33). However, many non-assertive environments only permit the NPI any, whether or not their polarity is marked to be present: see the following examples of directive-Yes/No interrogative (40a–b), directive-Wh interrogative (41a–b), and commissive-if-conditional (42a–b) contexts. (40) a. Do you have any book on syntax? b. Don’t you have any book on syntax? (41) a. Who buys any books here? b. Who doesn’t buy any books here? (42) a. If you have any book on syntax, I promise to borrow it. b. If you have any book on syntax, I promise not to borrow it.

202 Anjum P. Saleemi Thus, it is only these which truly exemplify the exclusive use of the ‘negative’ any. Our explanation for this is that any type of IF (call it X) may be either polar or non-polar (¬Pol). However, Yes/No and Wh-interrogatives, both obviously questions (Q), and if-conditionals (C), are indeterminate in regard to polarity,or are inherently non-polar (¬Poli) at the root-clause level, as opposed to being so by the speaker’s choice, (though clearly they can contain embedded polar propositions). This, we surmise, is due to certain fundamental conceptual factors, which eventually must be precisely articulated. What happens in the case of the particular combinations of SF and IF in the contexts under consideration (namely, 40–42) is that the non-polarity, and possibly the slight conceptual mismatch between the IF and SF polarity features, result in an indefinite interpretation (Indef) of EF, and consequently that of any. Thus (¬ = non-polar): (43) [CP Indef{EF [IF ¬Pol(X(P))] ↔ [SF (¬Poli(Q ∨ C(S))]} [TP… ]] Therefore, it appears feasible to formulate a unified account of the distribution of any, as an alternative to the various standard accounts (e.g., see the papers in Horn and Kato 2000, Hsiao 2006, and the other references cited in these works), which is able to explain a wider range of data, at least in part because it cuts across conceptual-illocutionary factors and the syntactic ones. Such an approach should also account for the ‘logophoric’ any, i.e., the appearance of any in the contexts where no c-commanding licenser is available (see Progovac 2000); at any rate, Chomsky points out that c-command may not be actually necessary in explaining various syntactic phenomena (to appear: 8), presumably especially if they are attested in nonPhases. To revise our previous summation, we propose that there is only one lexical item any which is construed negatively, positively, not at all, or as an indefinite word, depending on the C-I factors outlined above. Another case in point, likewise demonstrating the usefulness of the C-I categories, is binding theory, which Chomsky considers to lie “at the outer edge of the C-I interface” (to appear: 8). The data we seek to examine include non-core or logophoric reflexive anaphors (44–48) as well as the core cases of canonically bound anaphors (49–50) of this sort (see Büring 2005 for an overview). (44) There was a wall right [PP behind ourselves]. (45) [ConjP As to myself], I don’t know where I’ll be staying. (46) [DP Pictures of himselfi] please Johni a lot.

On the interface(s) between syntax and meaning

203

(47) [DP Nude pictures of herselfi] don’t offend Maryi. (48) [DP Stories about himselfi] excite Johni. (49) a. John saw himself. b. *Himself saw John. (50) a. He wants [PROi to hurt himselfi]. b. *He wanted [Maryi to hurt himselfi]. It is typical to assume that most of the problematic cases (44–48) do not submit to the usual sentence-internal binding conditions due to some discourse or other factors. We wish to argue that incorporation of certain illocutionary elements can help one achieve a unified, and therefore more elegant, explanation for the distribution of most if not all kinds of anaphors. Our account is based upon (i) PT, and (ii) the notion of “direction of fit”, one of the components of IF, (S&V: 92ff.). The direction of fit determines whether IF is designed to express the way the world is, or to make it the way intended by IF. S&V propose the following directions of fit: (i) the word-to-world direction of fit (assertives), (ii) the world-to-word direction of fit (commissives and directives) – which can be speaker-based (commissives) or hearer-based (directives), (iv) the double direction of fit (declaratives), and (iii) the null or empty direction of fit (expressives). It should be made explicit that the term “world” is used here in a relative sense, not necessarily in the literal one. First consider the role of PT, more specifically our modified version of it. It appears that in several contexts non-core unbound anaphors occur in non-Phase constituents, such as DP or PP (46–48, 44, respectively), or the sort of Conj(unction)P used at the start of (45). This means that they are not subject to the binding conditions, on the reasonable assumption that if an anaphor cannot find an explicit antecedent within the Phase immediately containing it, it is free to look for one elsewhere. This is unlike (49-50) above and the contrast in (51) below, where the binding must occur within the CP, clearly a Phase. (51) a. [CP For himself to win the contest], John decided to take all possible measures. b. *Himself to win the contest, John decided to take all possible measures. In regard to (44–48), instead of assuming the presence of phonetically null antecedents (a legitimate option for 50a), and regardless of the referential

204 Anjum P. Saleemi ambiguity of such antecedents (who took the picture? John /Mary or someone else?), an alternative is to consider the non-Phase constituents to be part of a sentence which expresses a new kind of direct of fit, which makes reflexivity possible in general. Although the original claim is that “There are four and only four directions of fit in language” (S&V: 52), we suggest that the range of the direction of fit be extended to include the world-to-world fit, which naturally subsumes the word-to-word fit, thereby exhausting all the logical possibilities of how various entities relate to each other in the relevant terms. Note that ‘entity’ could mean a variety of things: a person or an object about which something is stated to be the case or to which something is attributed (52), the speaker himself (53), the hearer (54), an abstract object (55), literally the whole world (56), even a linguistic expression (57). More importantly, this fifth kind of direction of fit, it is claimed, must co-occur with all the other kinds of direction of fit, as also partially and indirectly illustrated by the following examples: (52), (54), (56), and (57) are assertive (see also 49a and 50a above), while (53) is a declarative, and (55) a directive (exhibiting a sort of speaker-to-hearer-to-hearer fit!), with one of the four standard categories of the direction of fit applying to each type of IF as well. (52) This room can clean itself. (53) I declare myself to be the president of this country. (54) This system can destroy itself. (55) Go hang yourself! (56) The world will destroy itself. (57) This expression contradicts itself. Armed with the new category of the world-to-world fit, we can now claim that the IF in these non-binding, and indeed all other binding, cases is marked for this particular kind of fit, over and above containing one of the remaining four categories. So it is this property of IF that legitimizes the use of reflexives in the non-core examples under consideration (44–48), where reflexives, as we already know, are all part of non-Phase constituents. The implication is that the resulting overarching illocutionary factors not only (redundantly) apply to the core syntactic contexts in which anaphoric binding must occur, but also that the same factors can explain the appearance of reflexives in the non-core environments.

On the interface(s) between syntax and meaning

205

A problem remains to be resolved, which is why DPs containing picturestory type nouns are ill-formed in certain sentences (58–59), but not others (cf. 46–48). (58) *[DP Nude pictures of herselfi] absolved Maryi of the crime. (59) *[DP Stories about himselfi] don’t describe Johni very well. An obvious solution to this problem is that it is the CS-IF of these verbs that is responsible for these facts: the default meaning of please, offend, and excite (see 46–48) is experiencer-oriented (CS) and expressive (IF), but that of absolve and describe is agent-oriented (CS), and declarative and assertive, respectively (IF). This difference in meaning makes it impossible for the DPs in (58–59) to be licensed by any IF feature in EF that would also participate in the selection of the verbs they contain, as each of these DPs contains the theme of the verb in the form of reflexives. An expanded binding theory, as outlined in the foregoing pages, is better able to account for expressions, or parts thereof, whether or not they happen to be Phases, thereby showing the importance of the C-I interface, especially its illocutionary component, in human language. Presumably, it should not be too hard to incorporate into EF the features relevant to this approach to the distribution of reflexives, as exemplified in the previous part of this section for Focus and the distribution of any. Finally, it is hoped that the same kind of approach can be extended, mutatis mutandis, to reciprocal pronouns, also an object of inquiry in binding theory.

6. Phasing out the residue of the Y-model We begin to conclude this paper with a plea to straighten out an unnecessary and undesirable wrinkle in the derivational model currently in use in MP, which is in fact merely a diluted version of the more traditional EST/Ymodel and therefore carries over some residue from the older conceptualization. More precisely, it is suggested that the bifurcated derivational model retained in MP be realigned so that the syntax-meaning interface (C-I) constitutes one end of a linear or horizontal derivation, and A-P the other, a suggestion that is in keeping with the spirit of the arguments presented so far. The EST/Y-model, depicted in (60), is not appropriate under the fundamental assumptions of MP, and is therefore no longer required; the dotted triangle below depicts the step-by-step construction of expressions as a re-

206 Anjum P. Saleemi sult of the successive application of External Merge and the other syntactic operations, e.g., Internal Merge. (60) The bifurcated model Lexicon

A-P

SPELL-OUT

C-I

The revised model is graphically represented below (61) for the sake of explicit comparison. (61) The linear model CS ↓↑ Intentional Meaning LEXICON → C-I →… SYNTAX… (non-)PHASE → A-P ↓↑ IF It is argued that (61) is more desirable because: (a) It is conceptually more elegant, and there is really no need to set apart brute-force combinatorial mechanisms and the lexicon as they can be built into the modified model without any loss of explanatory force. (b) It is equally feasible to execute any derivations in (61) that the existing model is able to handle. (c) It can generate, among other things, expressions based on both ordered and unordered LAs, and is thus able to account for why some derivations crash but can still be interpreted with more or less effort (see examples 5 and 7 above), while others converge, but only to be understood as incoherent gibberish (2–4). (d) It incorporates Phases as part of the system, and provides for the generation of fragmentary or non-Phasic expressions. The two interfaces (A-P and C-I) must be treated as such, not as syntactic levels, and the lexicon should not be located orthogonally along a derivational axis different from the direct C-I/A-I axis, since it is the linguistic

On the interface(s) between syntax and meaning

207

starting point of any derivation, and therefore a significant part of the C-I interface. One major consequence of such a shift in the place of the lexicon is that there is no dual SPELL-OUT to A-P and C-I, but only one SPELLOUT towards the A-P end of the derivation. Another important shift is that derivations, since they do not have two distinct starting points anymore, do not need to be made separately legible to A-P and C-I. Instead, the main idea is to make meaning representations legible to linguistic form as a whole.

7. Some implications and consequences A major outcome of the perspective outlined above will be a truly minimal view of the core of the linguistic system, or “the faculty of language in the narrow sense,” perhaps one solely consisting of certain combinatorial and recursive properties; such a view, inevitably, will need to be complemented by a derivational system vastly enriched at the interfaces, constituting, as it were, “the faculty of language in the broad sense”(Hauser, Chomsky and Fitch 2002). Part of our effort is aimed at teasing apart pure or narrow syntactic mechanisms and those apparent properties of syntax which in fact reflect an assortment of conceptual and illocutionary factors. The syntactic and semantic entities are by no means completely autonomous, but part of the divergence between them is brought about by their specific properties and mechanisms coming into conflict with each other in the course of derivations. The concept of Phases advocated in this paper, as encapsulated in the hybrid thesis of Phases, is not only conceptually and empirically of a piece with their propositional-illocutionary character, but should also help redefine the relationship between form and meaning. Given all these factors, it would seem best to adopt the most adequate, yet essentially simpler, idealization, namely, that based on the one famously proposed by Saussure (1916 /1986), as any other versions promote the illusion that they are empirically the case, and as a result tend to be treated too narrowly and rigidly. Such derivational systems, in our view, are eventually bound to fail as the scope of linguistic theory and its database expand further. It is evident to us that considerably more empirical support is needed in support of our relatively innovative approach, and before that much existing empirical evidence and argumentation will have to be reconsolidated. However, we are convinced that the general direction outlined in the foregoing pages is fundamentally along the right lines.

208 Anjum P. Saleemi References Allan, Keith 2006 Clause-type, primary illocution, and mood-like operators in English. Language Sciences 28: 1–50. Boeckx, Cedric and Kleanthes K. Grohmann 2007 Putting phases in perspective. Syntax 10: 204 –222. Büring, Daniel 2005 Binding Theory. Cambridge: Cambridge University Press. Chomsky, Noam 1955/75 The Logical Structure of Linguistic Theory. New York /London: Plenum 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. 1995a The Minimalist Program. Cambridge, MA: MIT Press. 1995b Bare phrase structure. In Government and Binding Theory and the Minimalist Program, Gert Webelhuth (ed.), 383–439. Cambridge, MA /Oxford: Blackwell. 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 Structure, Vol. 3, Adriana Belletti (ed.), 104– 131. Oxford: Oxford University Press. 2005 Three factors in language design. Linguistic Inquiry 36 (1): 1–22. 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 1995 Something Else for the Binding Theory. Linguistic Inquiry 26 (2): 249 –275. Davison, Alice 2007 Word order, parameters, and the extended COMP projection. In Linguistic Theory and South Asian Languages, Josef Bayer, Tanmoy Bhattacharya, and M. T. Hany Babu (eds.), 175–198. Amsterdam/ Philadelphia: John Benjamins. Fodor, Jerry 1975 The Language of Thought. New York: Crowell.

On the interface(s) between syntax and meaning

209

Gordon, David and George Lakoff 1971/75 Conversational postulates. In Syntax and Semantics, Vol. 3: Speech Acts. CLS 7, Peter Cole and Jerry L. Morgan (eds.), 63–84. New York: Academic Press. Grohmann, Kleanthes K. to appear The road to PF. In Proceedings of the 17th International Symposium on Theoretical and Applied Linguistics, E. Agathopoulou, M. Demitrikapoulkou, and D. Papadopoulou (eds). Thessaloniki: Aristotle University of Thessaloniki. Hale, Kenneth and Samuel Jay Keyser 1993 On argument structure and the lexical expression of syntactic relations. In The View from Building 20: Essays in Linguistics in Honor of Sylvain Bromberger, Kenneth Hale and Samuel Jay Keyser (eds.), 53–109. Cambridge, MA: MIT Press. 2002 Prolegomenon to a Theory of Argument Structure. Cambridge, MA: MIT Press. Hauser, Marc D., Noam Chomsky, and W. Tecumseh Fitch 2002 The faculty of language: What is it, who has it, and how did it evolve? Science 298: 1569 –1579. Horn, Laurence and Yasuhiko Kato 2000 Negation and Polarity: Syntactic and Semantic Perspectives. Oxford: Oxford University Press. Hsiao, Katherine 2006 Polarity and ‘Bipolar’ Constructions: Epistemic Biases in Questions. MA thesis, National Chi Nan University. Huang, C.-T. James 1984 On the distribution and reference of empty pronouns. Linguistic Inquiry 15: 531–574. Huddleston, Rodney 2002 Illocutionary force and clause types. In The Cambridge Grammar of the English Language, Rodney Huddleston and Geoffrey Pullum (eds.). Cambridge: Cambridge University Press. Jackendoff, Ray 1990 Semantics Structures. Cambridge, MA: MIT Press. Keyser, Samuel Jay and Kenneth Hale 1992 Re: The abstract clitic hypothesis. Linguistic Inquiry 22: 89 –125. Levin, Beth and Malka Rappaport Hovav 2005 Argument Realization. Cambridge: Cambridge University Press. Lieber, Rochelle 2004 Morphology and Lexical Semantics. Cambridge: Cambridge University Press. Lin, Nelly and Anjum P. Saleemi 2008 Global linguistic derivations. Proceedings of CLS 41.

210 Anjum P. Saleemi Marantz, Alec 2005 Rederived generalizations. (Handout of) Talk delivered at National Tsing Hua University, Hsinchu. Progovac, Ljiljana 2000 Coordination, C-command, and ‘logophoric’ N-words. In Negation and Polarity: Syntactic and Semantic Perspectives, L. Horn and Y. Kato (eds.), 88–114. Oxford: Oxford University Press.. Pustejovsky, James 1995 The Generative Lexicon. Cambridge, MA: MIT Press. Rizzi, Luigi 1997 The fine structure of the Left Periphery. In Elements of Grammar, L. Haegeman (ed.), 281–337. Dordrecht: Kluwer. 2004 Locality and Left Periphery. In Structures and Beyond: The Cartography of Syntactic Structures, Vol. 3, Adriana Belleti (ed.), 223–251. Oxford: Oxford University Press. Ross, John R. 1970 On declarative sentences. In Readings in English Transformational Grammar, R. A. Jacobs and P. S. Rosenbaum (eds.), 222–272. Waltham, MA: Ginn & Co. Rothstein, Susan 2004 Structuring Events: A Study in the Semantics of Aspect. Malden, MA/ Oxford: Blackwell. Saleemi, Anjum P. 1998 Skinner’s razor and the varieties of mentalism. In Text in Education and Society, Desmond Allison, Lionel Wee, Bao Zhiming, and Sunita Anne Abraham (eds.). Singapore: Singapore University Press. 2005 Introduction: The Enigma of unification. In In Search of a Language for the Mind-Brain: Can the Multiple Perspectives be Unified? Anjum P. Saleemi, Ocke-Schwen Bohn and Albert Gjedde (eds.), Aarhus: Aarhus University Press. Saussure, Ferdinand de 1916 /86 Course in General Linguistics. La Salle, IL: Open Court. Searle, John R. 1969 Speech Acts: An Essay in the Philosophy of Language. Cambridge: Cambridge University Press. 1979 Expression and Meaning: Studies in the Theory of Speech Acts. Cambridge: Cambridge University Press. 1990 Consciousness, explanatory inversion and cognitive science. Behavioral and Brain Sciences 13 (4): 585–595. Searle, John R. and Daniel Vanderveken 1985 Foundations of Illocutionary Logic. Cambridge: Cambridge University Press.

Dynamic economy of derivation Takashi Toyoshima

1. Introduction Since the onset of generative grammar (Chomsky 1951, 1955), a kind of economy notions has been in the theoretical considerations as an evaluation metric for the choice among competing grammars, namely, the Simplicity Measure. As the theory has advanced, the original simplicity measure short lived, but it has also become increasingly clear that some kind of economy plays a significant role in the fundamental properties of language. Perhaps, the first explicit endeavor that put economy considerations into the research agenda is Chomsky (1992). There, the Least Effort Principle was suggested, unifying the two types of economy conditions, Economy of Derivation and Economy of Representation. Since then, the issue of economy has been a lively topic of research, and it crystallized in the Minimalist Program initiated in Chomsky (1993), in which linguistic expressions are taken to be nothing more than “optimal” realizations of interface conditions imposed from the external systems (Economy of Representation), and those linguistic expressions are generated in an “optimal” way (Economy of Derivation). One of the issues that have been hotly debated in the Minimalist Program is the nature of economy conditions, whether they can be “global” or should be “local” (Collins 1997; Johnson & Lappin 1997, 1999; Nakamura 1997; Yang 1997; among others). The Shortest Derivation Condition as informally stated in Chomsky (1992) is a “global” condition, in the sense that it requires a comparison of all the derivations that are completed (trans-derivational comparison; Groat & O’Neil 1996, among others). The Procrastinate Principle proposed in Chomsky (1993) is also “global”, as it “looks ahead” to determine whether or not a particular application of overt operations at a given point in a derivation would lead to convergence at the interfaces, potentially ending up with comparison of all the derivations that are completed, as well as the ones that are terminated in crash. They entailed a computational complexity of combinatorial explosion, and it has been commonly assumed that “look-ahead” is to be avoided in the formulation of derivational economy. Yet, it has not been made clear as to what the computational complexity is in the minimalist syntax, not to mention how it is to be measured. It has

212 Takashi Toyoshima become a buzzword that appears in numerous works, claiming that their proposals reduce it, but without making explicit what they mean by computational complexity and how it is reduced. In this paper, I will attempt to frame the issue of computational complexity of the minimalist syntax into the theory of computational complexity developed in mathematics and information science. Although human syntactic computation may turn out to require a distinct notion of complexity, it is certainly desirable if the complexity of human syntactic computation can be made amenable to such a mathematical theory of computational complexity, as the latter is not domain-specific. I believe that it is useful to formulate economy conditions of the minimalist syntax in terms of the mathematical theory of computational complexity, to make the preferential metric involved more explicit in comparing different proposals. Toward that end, I propose a dynamic economy principle of derivation, arguing that the notion of locality must be differentiated for Economy of Derivation and for Economy of Representation, and that “look-ahead” is the very nature of derivational theory. In particular, I argue that “look-ahead” does not necessarily entail “global” comparison, and what must be avoided is the non-local “look-far-ahead” of all the potential derivations, not the derivationally local “look-one-step-ahead” of only the actual derivation at a given stage. This conception of derivational economy allows the dynamic choice at every local step of derivations which operations to apply, (external) Merge or Move (internal Merge). In turn, the question, which motivated the concept of Chomsky’s (2000: 106ff.) Lexical Subarray (LSA), why Merge of an expletive does not always preempt Move, can be answered without introducing such a novel concept of LSA. Chomsky (ibid.) claims that LSA is “propositional,” determining a phase, and hence can be selected straightforwardly from the initial lexical array (LA); LSA is determined by a single choice of C or of v. I show that the concept of LSA is problematic, the determination of which is not so straightforward that complexity can be reduced as Chomsky envisages, entailing combinatorial explosion of an exponential order. Reviewing Fukui’s (1996) insightful overview of the nature of economy in language, sections 2 discusses the nature of economy conditions as optimization, and the issue of “global” vs. “local” optima, in relation to the theory of computational complexity. In section 3, I review the background problems of misgeneration that motivated the concept of Lexical Subarray (LSA) in Chomsky (2000), and in section 4, I demonstrate the determination problem that the notion of LSA entails. I also review the non-deterministic problem of Collins’ (1997)

Dynamic economy of derivation

213

formulation of Local Economy in section 5, and the general problem of static economy that predetermines some operation over another, scrutinizing Shima’s (2000) preference for Move over Merge in section 6. In section 7, I propose a dynamic economy principle of Minimum Feature Retention (MFR) that makes a locally deterministic choice without a predetermined preference for Merge or Move, which solves the misgeneration problem that motivated the concept of LSA. We see there that the computational complexity of Merge over Move with LSA is far greater than that of MFR. In section 8, I propose that the expletive there is selected as a kind of “External Argument” of v, which dissolves an apparent problem for MFR, and section 9 summarizes the main points with some concluding remarks.

2.

Economy in language

2.1. Fukui (1996) Fukui (1996) claims that economy in language is best captured as a discrete optimization problem, studied in combinatorial mathematics and theoretical computer science. An optimization problem in general can be formalized as a search for the value of the decision variable x that optimizes (minimizes or maximizes) the value of the objective function ƒ(x) under some constraints (Papadimitriou & Steiglitz 1982). This can be formulated as follows: (1)

Let the decision variable x be an n-dimensional vector x = <x1 , x2 , x3 , …, xn > Objective function: ƒ(x) Constraints: x ∈ R Problem: find x such that OPT (ƒ(x)) where the domain of ƒ is an appropriately chosen set D (⊇ R).

The set R is called the feasible region of the optimization. If x satisfies the constraint x ∈ R, it is called a feasible solution to the optimization problem (1), and if a feasible solution x* (x* ∈ R) satisfies the following condition: (2)

1

ƒ(x*) ≤ ƒ(x), where x ∈ R where x* is the global optimum.1

The condition (2) is for optimization problems as minimization. As maximization, the condition will be: (2´) ƒ(x*) ≥ ƒ(x), where x ∈ R

214 Takashi Toyoshima Given a neighborhood set U(x*) that includes a feasible solution x* (∈ R), x* is called a local optimum if the following relation holds: (3)

ƒ(x*) ≤ ƒ(x), where x ∈ R ∩ U(x*)

The global optimum is necessarily a local optimum, but the converse is not necessarily true. That is, there are optimization problems that can be solved globally, but not locally. It is not exceptional that complex problems get stuck into a local optimum that is not the global optimum. In the minimalist syntax, it means that a local economy principle does not necessarily yield the optimal derivation. We will see that in section 5, where Collins’ (1997) Local Economy principles over-generate ungrammatical derivations. Formulating both the Condition on Economy of Representation and the Condition on Economy of Derivation as discrete optimization problems, Fukui (1996) argues that the former is solvable in a (strictly) local fashion, as its satisfaction can be verified with(in a well defined domain of) a single representation at the interface. On the other hand, the derivational nature of language, i.e., the Condition on Economy of Derivation, is fundamentally global. Thus formulated, the issue of economy in language is therefore not an either/or-question (Collins 1997; Johnson & Lappin 1997, 1999), or the division of labor between global and local economy (Nakamura 1997). Rather, it is whether the local economy can “solve” the global optimization of language, i.e., the combinatorial recursive generative property of its derivational character.

2.2. Computational complexity of global optimization The most important problem for global optimization arises in the computational complexity that is induced. In the theory of computational complexity, problems, formalized as mathematical objects, are classified into complexity classes, in terms of computational power and resources that an algorithm requires to solve those problems (Johnson 1990, Papadimitriou 1994, among others). The computational power is modeled on formal automata, such as Turing machines (Turing 1937). The resource requirements are usually characterized by the growth rate of time and/or space, with respect to the “size” of the problem.2 Time is measured by the number of (discrete) 2

For other computational resources that can be used for complexity measures, see Blum (1967), Hartmanis & Hopcroft (1971), Seiferas (1990), among others.

Dynamic economy of derivation

215

operational steps of the algorithm to solve a problem (time complexity), and space by the memory capacity expended by the algorithm (space complexity). Any problem solving procedure that can be reasonably called an “algorithm” can be formalized as a Turing machine (Church’s Thesis).3 The complexity class of decision problems that are solvable by deterministic Turing machines in polynomial-bound time is called the class P. The complexity class of decision problems that can be solved by non-deterministic Turing machines in polynomial-bound time is called the class NP (Garey & Johnson 1979). Informally speaking, the class P problems are computationally tractable, i.e., “efficiently solvable” (Edmonds 1965), whereas the class NP problems are not.4 There are often tradeoffs between time and space, but the space classes usually include the time classes; it is known that the polynomial-space bound classes (N)PSPACE, whether solvable by a non-deterministic or deterministic Turing machine, properly include the polynomial-time bound classes (N)P, and the exponential-time bound classes (N)EXP is properly included in the exponential-space bound classes (N)EXPSPACE. It is also known that the exponential classes (N)EXP(SPACE) properly include the polynomial classes (N)P(SPACE), whether by deterministic or nondeterministic, or time-bound or space-bound. That is, for the reduction of over-all computational complexity, the reduction of space-complexity is fromally more effective than the reduction of time-complexity. Optimization problems can be reformulated into a roughly equivalent decision problem, by supplying a target value (reduction). Thus reformulated, optimization problems can be classified in terms of their computational complexity. A variety of optimization problems, e.g., the network flow problem, are known to be reducible to the class P. On the other hand, the traveling salesman problem is a classic example of optimization problems that have not been reducible to the class P. It asks to find the shortest tour of a given number of locations with specified dis3

4

Church’s Thesis (Church 1941) was independently reached in its essentials in Turing (1937), Kleene (1936), Post (1936), and Markov (1954). Although it has not yet been formally proven, it is generally believed that the class NP properly includes the class P, as non-finite state non-deterministic automata are generally more powerful than the corresponding deterministic automata. It does not mean that the class NP problems cannot be solved, but that there is no general algorithm known to solve arbitrary instances of their types. If a given instance of a problem is small enough, it may be solved without difficulty.

216 Takashi Toyoshima tances from one another. It can be solved by enumerating all the possible tours, computing the total distance of each, and picking the shortest. This is a brute-force “global” search, in which the solution is guaranteed to be found, but requires time factorial to the number n of locations ( n–1! 2 ). However, once a candidate solution is supplied, such an optimization problem reduces to a class P decision problem; for the traveling salesman problem, it asks whether there is an equal or shorter tour with respect to the supplied value. That is, once an instance is provided, it becomes easy to verify the answer.5 This can be done by a “guess-and-check” procedure, “guessing” a candidate value deterministically, and “checking” it in polynomial-time bound. Various approximation schemes and “heuristic” algorithms have been proposed for these purposes. The strategy shared by all those heuristic algorithms is the use of local optimization to approximate the global optimal solution. Conjecturing that the Condition on Economy of Derivation, formulated as a discrete optimization problem, probably does not belong to the class P, Fukui (1996) argues that language is equipped with local conditions as biological counterparts of a heuristic algorithm that reduces the computational complexity induced by its global nature, inherent to the derivational system.

2.3. Economy conditions as optimization Adapting Fukui (1996), I formulate the Economy Condition of Derivation as a discrete optimization problem as follows: search for a derivation d that minimizes the value of the cost function ƒc (d), which maps a set D of derivations {d1, d2, d3, …, dn } to a set I of non-negative integers (cost of d), under the constraint that d belongs to a set Dc of convergent derivations. (5)

The Economy Condition of Derivation Objective function: ƒc (d): D → I Constraints: d ∈ Dc Problem: find d such that MIN(ƒc (d))

5

The “easy-verifiability” of the supplied value does not mean that the optimization is solved. To get the optimal value, many trials are needed with different supplied values, and yet there is no guarantee that the optimal value can be found, unless all possible combinatorial permutations are tried out – which simply reverts back to the brute-force exhaustive search.

Dynamic economy of derivation

217

Fukui (1996) takes the decision variable to be a binary relation r such that r ∈ R = {〈(π, λ), D〉} where (π, λ) is a structural description of a linguistic expression, and D is a convergent derivation that forms it, leaving the constraints open, as premature to postulate. I take a derivation to be the decision variable, whether convergent or not. A derivation, in turn, can be described as a binary relation, but from a given Lexical Array (LA) to a set S of structural descriptions (SDs), that is, d: LA → S.6 This allows the separation of the convergence condition as the constraint for optimization, defining the reference set among which the optimal derivation can be determined. In this way, the problem of misgeneration that will be discussed in the following sections can be addressed (Chomsky 1995: 220 ff., 227ff.). Yet, non-convergent derivations do not compete for the purpose of economy of derivation (Chomsky 1993, et seq.). For a derivation to converge, the given LA must be empty, and the structural description SD derived by d must be fully interpretable at both the Articulatory-Perceptual (A-P) interface and the Conceptual-Intentional (C-I) interface. That is, SD must be a pair of output representations <π, λ> that conform to the Economy Condition of Representation, namely the Principle of Full Interpretation, containing all and only information interpretable at the respective interfaces. Thus, I suggest the following formulation of the Economy Condition of Representation: (6)

The Economy Condition of Representation Objective function: ƒc (SD): S → I Constraints: SD =| d Problem: MIN(ƒc (SD))

Following Fukui again, I take the Economy Condition of Representation to be a discrete optimization problem of a search for a structural description 6

I take a derivation not to be functional. This is because two distinct outputs can be derived from exactly the same choice of lexical items as well as their associated formal features. For example, the following two sentences are both perfectly grammatical: i) Mary thinks that John is intelligent. ii) John thinks that Mary is intelligent. If LA is somehow structured in a way to control the order of lexical choices, or dynamically formed as suggested in Chomsky (2000), a derivation may be understood as a function, the problem of which is the main topic discussed in Section 4.

218 Takashi Toyoshima SD = <π, λ> that minimizes the value of the cost function ƒc (SD), which maps a set S of structural descriptions {SD1, SD2, SD3 , …, SDn } to a set I of nonnegative integers (cost of SD), under the constraint that SD is a pair <π, λ> formed by a derivation d, which is denoted as SD =| d in (6). Again, Fukui does not specify the constraint, but the interface representations π and λ cannot just be an arbitrary pair of representations that converge at the A-P and the C-I interfaces independently (Chomsky 1995: 225); they must be properly paired by a legitimate derivation d. This is not a trivial matter, insofar as some kind of deletion operation exists. The atomic soundmeaning correspondence paired in lexical items alone cannot guarantee the proper association between the structured propositional meaning and the phonetic symbols. Formulating the constraint this way embodies the Principle of Recoverability (Chomsky 1964) as a part of Economy of Representation without conceptual conflict.7 Being formulated as an optimization (minimization) problem, the Economy Condition of Representation as a whole also embodies the Principle of Inclusiveness (Chomsky 1995), insofar as there is no operation that “adds” an element absent in LA.

2.4. Other economy conditions in the Minimalist Framework Other than the Shortest Derivation Condition, Chomsky’s (1992) Economy of Derivation included the Shortest Movement Condition and the Last Resort Condition. There was a tension, in fact, a conflict between the Shortest Derivation Condition and the Shortest Movement Condition. Derivations with a single long-distance movement in one fell swoop are shorter than derivations with many shortest possible movement steps to achieve the corresponding long-distance effect. In the Minimalist Program of Chomsky (1995), the operation Move is reformulated in terms of Attract, incorporating the Shortest Movement Condition (Minimal Link Condition) as one of its defining properties. The Last Resort Condition is also incorporated as feature-checking (or Agree in Chomsky 2000, et seq.), another defining property of the operation Move. 7

Obviously, the most economical representation is the one in which everything is deleted, but unless a deletion operation can wipe out multiple items in one step, deleting more items takes more steps, resulting in longer derivations (cf. Kitahara 1995, among others). Thus, there is a tension between Economy of Representation and Economy of Derivation that is balanced on the Principle of Recoverability.

Dynamic economy of derivation

219

It meant that neither the Shortest Movement Condition nor the Last Resort Condition is any longer an economy condition on derivation; there is simply no movement operation that is not shortest or last resort. The computational system of human language CHL just does not have such an operation. That leaves only the Shortest Derivation Condition as the economy condition of derivation, resolving the conflict with the Shortest Movement Condition. Yet, it remains to be seen how we can get beyond its informal idea and improve on its implementation. How to determine the shortest derivation is not easy; “global” search among all the possible derivations is guaranteed to succeed, but not computationally feasible. The task is, then, to formulate a local condition that is guaranteed to yield the shortest derivation. To that end, the notion of locality must be rethought, and I propose a derivationally local principle that yields the shortest derivation in section 7. In order to see the problems of the conventional notion of locality, let us turn to the problems of misgeneration.

3. Problems of misgeneration 3.1. The over-generation problem Since Chomsky (1995), it has widely been held that the operation Merge is “costless,” whereas the operation Move is less economical than Merge. This is because Merge is required for any symbolic systems, whereas Move is not usually used in well-designed artificial symbolic systems. Thus, Move is an “imperfection” of natural language. In addition, Move can be considered as being composed of suboperations of Feature-Checking (or Agree in Chomsky 2000) + Merge (+ Generalized “Pied-Piping”). In this view that Merge preempts Move, there is an over-generation problem when an expletive is involved. In Chomsky’s (1995: 295ff.) account, superraising (7) cannot be generated. (7)

*Johni seems (that) it was told ti that TP

Presumably, (7) would have been through the following stages in its derivation:

220 Takashi Toyoshima (8)

a. b. c. d. e.

[Tʹ T was told John [CP that TP]] [TP it [Tʹ T was told John [CP that TP]]] Merge (it) [CP (that) [TP it [Tʹ T was told John [CP that TP]]]] [v/VP seems [CP (that) [TP it [Tʹ T was told John [CP that TP]]]]] [Tʹ T [v/VP seems [CP (that) [TP it [Tʹ T was told John [CP that TP]]]]]]

At the stage (8a), there was an option of either Merge (it) or Move (John). As Merge (it) is assumed to be more economical than Move (John), the former is chosen (8b). When the derivation reaches the stage (8e), there is no more relevant lexical item to fill the matrix Spec(TP). Thus, the only operation available at this stage is Move. As it is closer than John is to the matrix T, it moves to become the matrix Spec(TP), yielding the following. (9)

*[TP Iti [Tʹ T [v/VP seems [CP (that) [TP ti [Tʹ T was told John [CP that TP]]]]]]]

Chomsky (ibid.) reasoned that although the matrix T satisfies its EPP requirement, its Case-feature cannot be checked, since the Case-feature of it has been checked and erased when it is Merged to the embedded Spec(TP) – hence the derivation crashes. Thus, superraising (7) is blocked by the crashing derivation (9). As has been noticed, however, there are two serious oversights in this account. The first is that crashing derivations should not be able to block other derivations. Otherwise, Move will never be applied insofar as Merge preempts it. By assumption, derivations involving only Merge is more economical than the ones involving Move. However, most, perhaps all, I suspect, derivations involving only Merge crashes. If crashing derivations can block other derivations, no derivation with Move survives. Thus, another convergent derivation must be sought to block superraising (7). The second is that covert raising of John’s Case-feature in (9) should be able to check the Case-feature of the matrix T, as John’s Case-feature has not been checked. Thus, (9) should converge, contrary to Chomsky’s (ibid.) claim. That is, although superraising (7) is correctly blocked, another ungrammatical derivation (9) is over-generated.

3.2. The under-generation problem In the Phase-Based Probe-Goal Theory of Chomsky (2000), superraising (7) and the over-generation of (9) are both blocked. When the derivation

Dynamic economy of derivation

221

reaches the stage (8e), the probe T in the matrix clause cannot access John or it in the embedded TP by Phase-Impenetrability Condition (PIC). (10) Phase-Impenetrability Condition (PIC)

(Chomsky 2000: 108 (21))

In 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. In (8e), the relevant phases are the matrix vP and the embedded CP, so the probe T in the matrix clause cannot access it in the embedded TP. Therefore, neither the superraising (7) nor the ungrammatical (9) can be over-generated. Even if PIC does not hold, the superraising (7) cannot be derived from (8e) by Defective Intervention Constraint (DIC). (11) Defective Intervention Constraint (DIC) α>β>γ

(Chomsky 2000: 123 (42))

where > is c-command, β and γ match the probe α, β is inactive.

In (8e), the probe α is the matrix T, β is it, and γ is John. Even though John is still active and is a potential goal for the probe α, the matrix T, it is inactive and intervenes between them so that Agree between the matrix T and John cannot be established. Thus, John is unable to move to the matrix Spec(TP), correctly blocking the superraising (7). Yet, the grammatical (12) cannot be derived either, insofar as Merge (it) preempts Move (John) at the stage (8a) – the under-generation problem. (12) [TP It [Tʹ T [v/VP seems [CP (that) [TP Johni [Tʹ T was told ti [CP that TP]]]]]]] Note in passing that in this account, superraising is doubly ruled out by PIC and DIC, while creating the under-generation problem. Minimalist intuitions lead us to cast some doubts whether this is the optimal solution.

4. The determination problem of Lexical Subarray Acknowledging the under-generation problem, Chomsky (2000: 106 ff.) introduces a novel concept of Lexical Subarray (LSA), in terms of which a phase is defined. At the start of a derivation, the initial lexical array (LA) is selected from the lexicon, and as the derivation proceeds, LSA, a subset of

222 Takashi Toyoshima lexical items in LA, is extracted, placed in active memory (the “workspace”), and submitted to the computation. When LSA is exhausted, the computation may proceed, or it may return to LA, extracting another LSA, and proceed as before. Thus, at a given stage of derivation, Move may take place if LSA in active memory does not contain an expletive.8 Chomsky (ibid.) asserts that this cyclic access to LA reduces “operative” complexity.9 Chomsky (ibid.) then characterizes LSA as “propositional,” and claims: LAi [= LSA: TT] can then be selected straightforwardly: LAi [= LSA: TT] contains an occurrence of C or v, determining clause or verb phrase … [emphasis added: TT]

Even if LSA can be selected straightforwardly as Chomsky (ibid.) claims, it cannot be a genuine solution; it merely replaces the problem of (narrow) syntactic computation with the problem of how LSA is extracted from LA. “Expletive insertion” is a matter of syntax proper; it is the heart of the problem of EPP, a perennial troublemaker throughout the history of generative grammar. It should not be relegated to some “exo-syntactic” process of LSA formation, but should be resolved in narrow syntactic computation per se.10 In other words, it should be accounted for in terms of how lexical items are combined to build a structure, not in terms of how lexical items are made available for narrow syntactic computation. Merely characterizing LSA as “propositional” cannot guarantee the exclusion of an expletive from LSA either, since expletives are presumably “meaningless,” and hence it should not matter whether one is included in LSA or not; it neither contributes any meaning to, nor hampers the content of, a “proposition.” Neither will it suffice to determine LSA by containment of a single C or v. Selection of a relevant LSA is not so straightforward as Chomsky (2000) 8

9

10

This is, in fact, an over-simplification. Move cannot take place if LSA in active memory contains any lexical items, or if any complex structures have already been built in parallel spaces, that satisfy the EPP requirement, presumably of DP category. Merge of these should preempt Move of the raising argument in question. I take what Chomsky (2000) calls “operative” complexity to correspond to “computational” complexity measured by the number of “operational” steps, i.e., timecomplexity, in information science and discrete mathematics. LSA extraction is “exo-syntactic” in the sense that is not itself a structure-building operation. Yet, it still must be internal to the computational system of human language CHL.

Dynamic economy of derivation

223

claims. In principle, there are nCm possible ways to form LSA, extracting a subset of m lexical items from LA of n remaining lexical items. Take, for instance, the following example. (13) it T1 seems [that friendsi T2 were told ti [CP ]]

(cf. op. cit.: 129 (48b))

At the stage where the complement CP of told has somehow been constructed successfully, LA contains at least 8 remaining lexical items {T1, seem, that, T2, were, told, friends, it}, assuming for simplicity, were = v and there is no matrix C.11 The relevant LSA in question must contain at least 1 lexical item, either that (= C) or were (= v), but not both. Thus, there are 6 remaining lexical items {T1, seem, T2 , told, friends, it}, from which none to all items can be extracted to form LSA, either with that or with were. That is, there are 6C0 + 6C1 + 6C2 + 6C3 + 6C4 + 6C5 + 6C6 = 64 ways of extracting a subset to form LSA, either with that or with were. Therefore, there are 64 2 = 128 ways in total to form LSA at the stage where the complement CP of told has been constructed. If the expletive it is to be excluded at this same stage, potential combinations to be considered are among the 5 remaining lexical items {T1, seem, T2, told, friends}, from which none to all items can be extracted to form LSA, either with that or with were. There are 5C0 + 5C1 + 5C2 + 5 C3 + 5C4 + 5C5 = 32 ways of extracting a subset that can form LSA, either with that or with were. Therefore, there are 32 2 = 64 ways in total to form LSA without the expletive it, at the stage where the complement CP of told has been constructed. See Appendix. In a nutshell, there are 128 potential LSA, and 64 candidate LSA that do not contain the expletive it (underlined in Appendix), out of which there is only 1 correct LSA (marked ❖ in Appendix) to yield the desired derivation (12). Thus, extraction of the correct LSA is not a trivial matter, requiring a search among candidates exponential (2n – m∙m) to the number n of the remaining lexical items in LA that contains m phase-defining lexical items, i.e., C or v. That is, extraction of the correct LSA is a problem of the exponential class, be it the time-class (N)EXP or space-class (N)EXPSPACE (see section 2.2). 11

In the structure (13), were is not raised to T2 for the ease of discussion to follow. As far as I can see, there is no relevant effect even if were is not v but is directly generated as T2. If a matrix C must always be present, the combinatorial problem to be discussed below will further be compounded.

224 Takashi Toyoshima In sum, the “operative” complexity in derivation is reduced at the cost of the added “computational” complexity of an exponential order in the “exo-syntactic” process of extracting the correct LSA. Taking the redundant nature into account that superraising is doubly ruled out by PIC and DIC, this does not strike me as a genuine minimalist solution.

5. The problem of non-deterministic economy: Collins (1997) Underneath the problems we have reviewed lies the idea that Merge preempts Move (whenever possible). This view seems to be persisting even after Chomsky (2004: 110) reinterprets Move as internal Merge (IM), which comes “free,” just as the conventional Merge does, which is also reinterpreted as external Merge (EM). Thus, Chomsky (2004: 114) still adverts, “Move = Agree + Pied-piping + (external) Merge.” The displacement property of language is a necessity, not an imperfection, so that Move (IM) is a freely available operation, just as (external) Merge (EM) is; IM and EM are isotopes of Merge, and they both come “free.” One may argue that IM conceptually presupposes EM, as IM can be applied only to syntactic objects constructed through EM. On the other hand, EM operates on two separate syntactic objects, whereas IM operates on a single syntactic object. Or one may argue that Pure Merge, Merge that is not part of Move (Chomsky 2000: 103), involves c-/s-selection and determination of whether the terms are drawn from LA or from the syntactic objects already built. These are on a par with Agree and Pied-Piping for Move. Conceptual arguments can be made both ways, and ultimately the question is empirical, whether either one has the priority over the other or they are on an equal footing. Mere reinterpretation of Move (internal Merge) and (external) Merge as different or not comparable does not solve the over-generation problem we reviewed in section 2. In fact, Collins (1997) argues that Move and Merge are not comparable, and makes a proposal which in effect renders Move and Merge equal in cost. He contends that economy conditions should be formulated in a local fashion as follows: (14) Local Economy

(op. cit.: 4 (3))

Given a set of syntactic objects Σ which is part of derivation D, the decision about whether an operation OP may apply to Σ (as part of optimal derivation) is made only on the basis of information available in Σ.

Dynamic economy of derivation

225

Collins (op. cit.: 90ff.) regards Move as Copy + Merge, and the operation Select as a Copy operation out of the lexicon. Thus, both Move and Select + Merge are both instances of Copy + Merge operations, so that they both need to be triggered as last resort operations. Then, he formulates Last Resort as an independent economy condition, rather than a defining property of an operation, as in the following: (15) Last Resort (op. cit.: 9 (12)) An operation OP involving α may apply only if some property of α is satisfied. For Move, Collins (op. cit.) adopts the contemporary minimalist assumption that the relevant property to be satisfied is the checking of uninterpretable formal features. For Merge, he proposes the following principle: (16) Integration (op. cit.: 66 (8)) Every category (except the root) must be contained in another category. Then, Collins (op. cit.) formulates Minimality as another independent economy condition. (17) Minimality (op. cit.: 9 (13)) An operation OP (satisfying Last Resort) may apply only if there is no smaller operation OPʹ (satisfying Last Resort). Collins (op. cit.) adopts the standard Minimalist conception of the Minimal Link Condition as the metric for “smallness” of the operation Move. For Merge, the number of merged objects is counted, introducing a new formulation of Merge, which he calls Unrestricted Merge (op. cit.: 75 ff.). Unrestricted Merge is a generalized grouping operation that applies to any number of constituents, but its vacuous application to no element does not satisfy Last Resort, so that it is not possible. The binary application is smaller than ternary, quadripartite, etc., applications involving more than two elements. Given Last Resort and Minimality, a unary application to a single element is the “smallest” application of Unrestricted Merge, but if such a unary application is allowed, it inevitably yields infinite recursion, so that it is stipulated to be impossible. Therefore, only the binary application involving two elements is chosen by Minimality. Given these, Collins (1997) claims that Move and (Unrestricted) Merge are not comparable, and that they are equally economical insofar as they

226 Takashi Toyoshima obey Minimality (21).12 Thus, neither one is favored over the other in his local economy. In effect, they are equal in cost, and hence both options can be pursued. This non-deterministic nature of Collins’ (op. cit.) local economy faces the same over-generation problem as Chomsky (1995) we have seen in section 2. Since Merge (it) and Move (John) are not comparable and hence equal in cost at the stage (8a), repeated below as (18), there are two possible continuations, (8b) repeated below as (19), and (20). (18) [Tʹ T was told John [CP that TP]] (19) [TP it [Tʹ T was told John [CP that TP]]]

Merge (it)

(20) [TP Johni [Tʹ T was told ti [CP that TP]]]

Move (John)

As desired, continuation from (19 = 8b) cannot lead to superraising (7), repeated below as (21), since it violates Minimality, and the continuation from (20) may correctly yield the grammatical (12), repeated below as (22). (21) *[TP Johni [Tʹ T [v/VP seems [CP (that) [TP it [Tʹ T was told ti [CP that TP]]]]]]] (22) [TP It [Tʹ T [v/VP seems [CP (that) [TP Johni [Tʹ T was told ti [CP that TP]]]]]]] Yet, another continuation from (19 = 8b) will lead to over-generation of the ungrammatical (9), repeated below as (23). (23) *[TP Iti [Tʹ T [v/VP seems [CP (that) [TP ti [Tʹ T was told John [CP that TP]]]]]]]

12

There is an oversight in this reasoning of Collins’ (1997). Merge is not necessarily of lexical items, e.g., a complex subject DP with a transitive verb-complement structure (whether Vʹ or vʹ), so that it need not be preconditioned by Select (as Copy out of the lexicon). Thus, Merge of two complex structures already built in parallel, which is without Copy (= Select of a lexical item), may still be more economical than Move (= Copy + Merge).

Dynamic economy of derivation

227

6. The problem of static economy: Move over Merge As it is clear by now that Move (John) needs to preempt Merge (it) at the stage (18 = 8a), in order to derive the grammatical (22 = 12) while blocking superraising (21 = 7) and over-generation of the ungrammatical (23 = 9). Yet, Move cannot just always take precedence over Merge. Consider the continuation (24) from (20). (24) a. [CP (that) [TP Johni [Tʹ T was told ti [CP that TP]]]] b. [v/VP seems [CP (that) [TP Johni [Tʹ T was told ti [CP that TP]]]]] c. [Tʹ T [v/VP seems [CP (that) [TP Johni [Tʹ T was told ti [CP that TP]]]]]] At this stage, Move (John) over Merge (it) will yield the following, leaving it in LA. (25) [TP Johni [Tʹ T [v/VP seems [CP (that) [TP tʹi [Tʹ T was told ti [CP that TP]]]]]]] Chomsky (1995: 226) has claimed that if LA is not exhausted, “no derivation is generated and no questions of convergence or economy arise.” At (25), LA still contains it, so that it is not a completed derivation from (24c). Thus, Merge (it) is the only choice, yielding the grammatical (22 = 12). Notice at this point that the derivation of the grammatical (22 = 12) is one-step shorter than the one of the ungrammatical (23 = 9); in order to satisfy the EPP-feature and the Case-feature of the matrix T, the former invokes one movement of John, whereas the latter two movements, overt movement of it for EPP and covert movement of John’s Case-feature, with the number of mergers and of all the other movements being equal. Thus, the Shortest Derivation Condition, imposing the fewest-step requirement, could have chosen the grammatical (22 = 12) while blocking the superraising (21 = 7) as well as the over-generation of the ungrammatical (23 = 9), but only if Move had been chosen over Merge at the stage (18 = 8a). This illustrates a case where the computation is stuck in a local optimum that is not the global optimum, not being able to derive the shortest derivation due to the fixed decision for Merge over Move applied locally. Yet, a simple-minded appeal to the necessity of LA exhaustion does not solve the problem unequivocally. Suppose the derivation reached the stage (25), and LA still contained not only it, but also another T, is, likely, and another that (C). Then, the derivation could have continued as (26a–d).13 13

The putative raising of is to T is ignored here for the ease of discussion.

228 Takashi Toyoshima (26) a. [CP that [TP Johni [Tʹ T [v/VP seems [CP (that) [TP tʹi [Tʹ T was told ti [CP that TP]]]]]]]] b. [v/VP is likely [CP that [TP Johni [Tʹ T [v/VP seems [CP (that) [TP tʹi [Tʹ T was told ti [CP that TP]]]]]]]]] c. [Tʹ T [v/VP is likely [CP that [TP Johni [Tʹ T [v/VP seems [CP (that) [TP tʹi [Tʹ T was told ti [CP that TP]]]]]]]]]] d. *[TP It [Tʹ T [v/VP is likely [CP that [TP Johni [Tʹ T [v/VP seems [CP (that) [TP tʹi [Tʹ T was told ti [CP that TP]]]]]]]]]]] At (26d), LA is exhausted, so that the derivation is complete, and yet it is ungrammatical. This is a classic case of a Tensed-S Condition violation (Chomsky 1973), and in the Phase-Based Probe-Goal Theory (Chomsky 2000, et seq.), it is attributed to the Activity Condition, which renders a goal inactive after its uninterpretable/unvalued features being deleted/valued.14 The point is, the necessity of LA exhaustion cannot force Merge (it) at the stage (24c). Stipulating the following ancillary condition, Shima (2000) maintains that Move is preferred over Merge whenever possible. (27) [Spec, TP] can be filled only by a DP with structural Case. (op. cit.: 377 (10)) By (27), Move (John) is not an option at the stage (24c), since John has already checked its Case-feature in the embedded Spec(TP). Therefore, it is merged as the only choice to become the matrix Spec(TP), yielding the grammatical (22 = 12). Thus, Shima’s (op. cit.) proposal that Move is preferred over Merge, together with the condition (27), correctly generates (22 = 12). At the same time, both superraising (21 = 7) and the overgenerated (23 = 9) are correctly blocked as desired, repeated as (28, 29) below, respectively. (28) *[TP Johni [Tʹ T [v/VP seems [CP (that) [TP it [Tʹ T was told ti [CP that TP]]]]]]] (29) *[TP Iti [Tʹ T [v/VP seems [CP (that) [TP ti [Tʹ T was told John [CP that TP]]]]]]]

14

For problems of the Activity Condition, see Nevins (2004).

Dynamic economy of derivation

229

Nevertheless, this cannot be the whole story, either. For the contrast in existential constructions (30), Shima (2000: 382ff.) offers a Case-based analysis on the assumptions (31), following Belleti (1988) and Lasnik (1995). (30) a. Therei seems [ ti to [be someone in the room]] b. *There seems [someonei to [be ti in the room]] (31) a. The expletive there has a [structural: TT] Case feature, and a postcopular DP is optionally assigned [an inherent: TT] partitive Case by a copular. b. The expletive there has a formal feature to be checked by that of a DP with partitive Case. [emphases in bold added: TT] The common intermediate stage for (30a,b) is the following. (32) [Tʹ to [be someone in the room]] If someone is assigned a partitive Case, it cannot fill Spec(TP) by the condition (27). Therefore, there will be merged as Spec(TP), and the derivation can continue as the following, yielding (33d = 30a). (33) a. b. c. d.

[TP there [Tʹ to [be someone in the room]]] [v/VP seems [TP there [Tʹ to [be someone in the room]]]] [Tʹ T [v/VP seems [TP there [Tʹ to [be someone in the room]]]]] [TP therei [Tʹ T [v/VP seems [TP ti [Tʹ to [be someone in the room]]]]]]

If someone is not assigned a partitive Case, say, nominative, it can fill Spec(TP). Then, the preference for Move over Merge dictates its movement over merger of there at the stage (32), and the derivation would continue as the following. (34) a. [TP someonei [Tʹ to [be ti in the room]]] b. [v/VP seems [TP someonei [Tʹ to [be ti in the room]]]] c. [Tʹ T [v/VP seems [TP someonei [Tʹ to [be ti in the room]]]]] At this point, there are two options: Move (someone) and Merge (there). As someone still has a structural Case and there is also assumed to have a structural Case, they both meet the condition (27). Then, the preference for Move over Merge demands someone to move to the matrix Spec(TP). (35) [TP someonei [Tʹ T [v/VP seems [TP tiʹ [Tʹ to [be ti in the room]]]]]]

{there}

230 Takashi Toyoshima But that leaves there in LA. Shima (ibid.) claims, “hence [it] does not produce a single syntactic object, which makes the derivation crash (Chomsky 1995: 226).” However, this deduction is inaccurate; incomplete derivations neither converge nor crash. They fail to yield outputs at the interface levels. To quote Chomsky (1995: 226) accurately: Note that no question arises about the motivation for application of Select or Merge in the course of application. If Select does not exhaust the numeration [≒ LA: TT], no derivation is generated and no questions of convergence or economy arise. …

In other words, if a choice made by an economy consideration does not yield a complete derivation, that economy consideration becomes irrelevant, and an alternative that completes the derivation needs to be sought. Thus, at the stage (34c), Move (someone) over Merge (there) is not really an option, and only Merge is the possible continuation leading to (30b), which is ungrammatical, so it should be a crashing derivation. In Shima’s (2000) analysis, what is wrong with (30b) must be the inability of someone to check some formal feature of there. By hypothesis, someone in (34–35) is not assigned a partitive Case, and hence by the assumption (31b), it cannot check whatever the formal feature of there. Then, the question boils down to what is the feature of there that needs to be checked, and why it cannot be checked by other than that of DP with partitive Case. Thus, Shima’s (op. cit.) account crucially relies on the stipulated condition (27) and the assumption (31) that there must in effect cooccur with a DP with partitive Case, which is dubious as it is morphophonologically undetectable.

7. Preemptive Move: Dynamic economy A desideratum that emerges from the observations thus far is that Move needs to preempt Merge sometimes but not always. That is, an economy consideration that chooses between Merge and Move had better not be fixed statically. This view meshes well with Chomsky’s (2000: 110) reinterpretation of Move as internal Merge, “an operation that is freely available,” so that neither (external) Merge nor Move (internal Merge) is more economical than the other. This amounts to Collins’ (1997) proposal, but he did not pursue the issue further, ending up with non-deterministic economy that did not solve the over-generation problem as we have seen in section 5.

Dynamic economy of derivation

231

What kind of conditions can allow such a shifting, yet deterministic choice of operations, and how can we formulate it? In order to approach the answers to these questions, let us review the decisive points in the derivation for the “superraising” example. At the stage (18 = 8a), repeated below as (36), we want Move (John) over Merge (it). (36) [Tʹ T was told John [CP that TP]] At the stage (24c), repeated below as (37), Merge (it) needs to preempt Move (John). (37) [Tʹ T [v/VP seems [CP (that) [TP Johni [Tʹ T was told ti [CP that TP]]]]]] What could motivate these choices? At the stage (36 = 18 = 8a), the Casefeature of John and the φ-set of T are unvalued, and the EPP-feature of T needs to be deleted. Move (John) will delete the EPP-feature of T, valuing both the Case-feature of John and the φ-set of T. Merge (it) will also delete the EPP-feature of T, valuing the φ-set of T and the Case-feature of it itself, but will leave John’ Case-feature unvalued. At the stage (37 = 24c), the φ-set of the matrix T is unvalued, and its EPPfeature needs to be deleted as well. Move (John) will delete the EPP-feature of the matrix T, but cannot value its φ-set, as the Case-feature of John has already been valued in the embedded Spec(TP). Furthermore, it will be left in LA, generating no derivation. On the other hand, Merge (it) will delete the EPP-feature of the matrix T and value its φ-set as well as the Case-feature of it itself. Notice that the undesired choice of operations leaves unvalued/uninterpretable features in the resulting stage of the derivation. Note also that the number of features deleted/checked or valued/matched is the same whether Move (John) or Merge (it) at the stage (36 =18 =8a). Thus, neither Total Checking (Poole 1998) nor Maximum Matching (McGinnis 1998, Chomsky 2001: 15 (14)) of the sort can make the right choice. Exploiting these facts, I propose the following economy principle of derivation. (38) Principle of Minimum Feature Retention (MFR) At every stage Σn of a derivation, choose the operation that leaves the fewest unvalued/uninterpretable features in the resulting stage Σn + 1 of the derivation.

232 Takashi Toyoshima The intuitive idea behind this principle is that unvalued/uninterpretable features must ultimately be valued or deleted, so that carrying more of them along the derivation is less economical than valuing/deleting them as soon as possible. Keeping track of which unvalued/uninterpretable features are still unvalued or not deleted demands more memory capacity than to forget about them as soon as they are deleted or valued. As we have reviewed in section 2.2, the space-complexity is more important than the time-complexity, in reducing the over-all computational complexity. That is, memory consideration is more important than the number of operational steps, in the reduction of computational complexity. Although MFR was conceived from an intuitive idea of memory reduction, it is formulated to reduce the number of operational steps. MFR dynamically makes a locally deterministic choice between (external) Merge and Move (internal Merge). MFR is dynamic in that it makes a choice whenever options arise, without any preference predetermined for either Merge or Move. MFR is deterministic in that it chooses either Merge or Move (unless a tie) at every point in the derivation where options arise, unlike Collins’ (1997) non-deterministic economy that allows different operations to proliferate different continuations whenever options arise.15 Furthermore, MFR is local in that it does not make any trans-derivationally global comparison of all the derivations to determine which operation to choose at each point of the derivations. It crucially involves the so-called “look-ahead” of only one-step in the derivation Σn → Σn + 1, but it does not “look far ahead” in the derivation Σn → Σn + m (m > 1). This transitional onestep “look-ahead” is essential for the notion of locality in the derivational economy, and it does not impose any significant computational load, as the choices are, at worst, time-bound, to a linear function f(x) = n + m + l of the number n of the remaining lexical items in LA + the number m of the complex structures already constructed in parallel + the number l candidate constituents that can be moved.16

15

16

Determinism of MFR is limited to the choice between Merge and Move, and I leave open for now the choice among Merge (x) or the choice among Move (y). For Move, see also fn.16. Given the “attract-the-closest” formulation of Move with feature-checking and the fact that Move is contingent on unvalued/uninterpretable features of the probe/target that have not been valued or deleted, the number l of candidate constituents that can be moved in most cases is limited to 1. Also, n + m + l never exceeds the number k of lexical items in the initial LA at the start of derivations.

Dynamic economy of derivation

233

The infamous “look-ahead” is the “look-far-ahead” in the derivation Σn → Σn + m (m > 1) of the sort, which proliferates exponentially all the potential derivations to be compared, just as in Collins’ (1997) non-deterministic economy. In fact, if absolutely no “look-ahead” is allowed as in Collins’ (op. cit.: 4 (3)) local economy (14), the economy conditions will end up being equivalent to representational filters as argued in Johnson & Lappin (1997, 1999). I argue that the notion of locality needs to be differentiated for the Economy of Derivation from the one for the Economy of Representation. Derivations, by nature, map one representation to potentially more than one representation in a sequential manner. Thus, the locality in derivational economy should be defined in terms of one step in derivations as illustrated in (39), rather than on a single representation, which is still the basis for the locality notion in representational economy.17 (39)

LA q tgy p Σ1a Σ1b Σ1c Σ1d Σ1e … ty fgh fgh fgh ty Σ2a Σ2b fh fh Σ3a Σ3b g g

← derivationally local one step ← derivationally local one step ← derivationally local one step . : :

Let us now go through in some detail how MFR contrasts with Merge over Move escorted by LSA in Chomsky’s (2000, et seq.) Phase Theory, in deriving (13), repeated below with a little more detailed structure as (40). (40) [TP It T1 [vP v [VP seems [CP that [TP friendsi T2 [vP were [VP ti told [ CP ]]]]]]]

17

We may reformulate MFR to incorporate a minimality notion of representational economy as follows: (i) Principle of Minimum Feature Retention (reformulated) At every stage Σn of a derivation, choose an operation that yields the minimum representation in the resulting stage Σn + 1 of the derivation. How we define the minimum representation is an issue, however, which I have to leave aside for another occasion. The number of unvalued/uniterpretable features is certainly relevant, but so is the “size” of the structures. With the standard assumption of the Extension Requirement (Chomsky 1993: 22ff.), Merge always grows the structure “taller,” whereas Move need not, as in the case of head-tohead adjunction or of covert movement, in the relevant sense.

234 Takashi Toyoshima To converge as (40), the derivation had been through the following stage:18 (41) [Tʹ T2 [vP were [VP friends told [ CP ]]]] At this stage, there are n1 = 5 lexical items left in LA, {that, T1, seems, v, it}, assuming, again for simplicity, were = v and there is no matrix C (see fn.11). Presumably, there is m1 = 0 complex structure constructed separately, and there is l1 = 1 constituent that can be moved, namely, friends. Therefore, there are (n + m + l) = (5 + 0 + 1) = 6 options at this stage: Merge (that), Merge (T1), Merge (v), Merge (seems), Merge (it), and Move (friends). (42) a. b. c. d. e. f.

[TP that [Tʹ T2 [vP were [VP friends told [ CP ]]]] [TP T1 [Tʹ T2 [vP were [VP friends told [ CP ]]]] [vP v [Tʹ T2 [vP were [VP friends told [ CP ]]]] [TP seems [Tʹ T2 [vP were [VP friends told [ CP ]]]] [TP it [Tʹ T2 [vP were [VP friends told [ CP ]]]] [TP friendsi [Tʹ T2 [vP were [VP ti told [ CP ]]]]]

The computational system of human language CHL performs all the 6 possible operations, in order for MFR to compare the resulting stages. Merge (that) may delete the EPP-feature of T2, but cannot value the φ-set of T2 or the Case-feature of friends, leaving 2 uninterpretable/unvalued feature(set)s in the resulting stage. Merge (T1) may also delete the EPP-feature of T1, as well as the EPP-feature of T2, but does not value the φ-set of T2 or the φ-set of T1, as they can be conflicting. Consequently, the Case-feature of friends, cannot be valued, either, leaving 3 uninterpretable/unvalued feature(set)s in the resulting stage.19 Merge (v) or Merge (seems) may also be able to delete the EPP-feature of T2, but cannot value the φ-set of T2 or the Case-feature of friends, leaving 2 uninterpretable/unvalued feature(set)s in the resulting stage. Merge (it) will delete the EPP-feature of T2, valuing the Case-feature of it itself and perhaps incorrectly the φ-set of T2 as singular, but leaves the Case-feature of friends unvalued in the resulting stage. 18

19

I assume that the verb tell takes a complement CP as its direct object in its complement position and an indirect object in its specifier position as their base argument positions. I assume the φ-set of T is valued by the interpretable φ-set of DP, and the Casefeature of DP by the interpretable tense-feature of T.

Dynamic economy of derivation

235

Move (friends) will delete the EPP-feature of T2, and value the φ-set of T2 and the Case-feature of friends, leaving no uninterpretable/unvalued feature in the resulting stage. Therefore, MFR chooses Move (friends) as the most economical operation at the stage (41), and discards all the other options. Or one may think of MFR choosing the resulting representation (42e) that is “minimum,” the representation with the fewest unvalued/ uninterpretable feature(set)s (see fn.17). It may appear as a complex process, but any system has to decide which lexical item to Merge at a given point of derivations, even if Merge always preempts Move. Verification of remaining uninterpretable/ unvalued feature(set)s is not implied in other systems, but any system needs to keep track of uninterpretable/unvalued feature(set)s. CHL performs all the possible operations, either in parallel, or in random sequence one by one, while keeping each result in memory until a new winner appears. In either case, MFR does not really need to count the exact cardinality of them. MFR only needs to tell which resulting stage has fewer uninterpretable/unvalued feature(set)s. Collins (1997: 132, n.10) also broaches an alternative to counting, and in fact, counting may not be such a strange bedfellow of grammar after all – if the core property of the Faculty of Language in Narrow sense (NFL) is recursion (Hauser, Chomsky & Fitch 2002; Fitch, Hauser & Chomsky 2005) and the recursive application of Merge yields the successor function (Chomsky 2008, Hinzen 2008), it will be no surprise that grammar turns out to be able to count. In Chomsky’s (2000, et seq.) Phase Theory with Merge over Move, on the other hand, before the derivation reaches the stage (41), two LSA must be formed after the embedded CP is successfully constructed. At the stage where the embedded CP was completed, there were n = 9 lexical items left in LA, {it, T1, v, seems, that, T2, were, friends, told}, which contained m = 3 phase-defining lexical items, {v, that, were}. Thus, there were 29 – 3∙3 = 192 ways to form LSA, out of which only one correct LSA, {were, friends, told}, had to be searched for the vP-phase. (43) [vP were [VP friends told [ CP ]]] Furthermore, in order to derive (43), those three lexical items had to be merged in the correct order: Merge (told), Merge (friends), and Merge (were). There are 3! = 6 possible permutations, out of which the correct order had to be searched. Even if a wrong LSA is formed, CHL would not know it until all the lexical items in that LSA are merged in all the possible orders. The largest

236 Takashi Toyoshima possible LSA contains 7 lexical items: a single phase-defining lexical item, and all the rest of 6 non-phase defining lexical items {it, T1, seems, T2, friends, told}. As there are 3 phase-defining lexical items, there are 3 such largest LSA, for each of which there are 7! = 5040 permutations. The second largest LSA contains 6 lexical items, and there are 3 6 = 18 such LSA, for each of which there are 6! = 720 permutations. The third largest LSA contains 5 lexical items, and there are 3 15 = 45 such LSA, for each of which there are 5! = 120 permutations. The fourth largest LSA contains 4 lexical items, and there are 3 20 = 60 such LSA, for each of which there are 4! = 24 permutations. The fifth largest LSA contains 3 lexical items, and there are 3 15 = 45 such LSA, for each of which there are 3! = 6 permutations, and the sixth largest LSA contains 2 lexical items, and there are 3 6 = 18 such LSA, for each of which there are 2! = 2 permutations. The smallest possible LSA is the ones with only a phase-defining lexical item, v, that, or were. Thus, there are (3 5040) + (18 720) + (45 120) + (60 24) + (45 6) + (18 2) + 3 = 35,229 orders of Merge, out of which only one order, namely Merge (told) – Merge (friends) – Merge (were), could derive (43) from the stage where the complement CP had been completed. After such a massive search, another LSA had to be extracted for the next CP-phase, in order for the derivation to reach (41). When the vP-phase (43) was completed, there were n = 6 lexical items left in LA, {it, T1 , v, seems, that, T2}, which contained m = 2 phase-defining lexical items, {v, that}. The largest possible LSA contains 5 lexical items, and there are 2 such largest LSA, for each of which there are 5! = 120 permutations. The second largest LSA contains 4 lexical items, and there are 2 5 = 10 such largest LSA, for each of which there are 4! = 24 permutations. The third largest LSA contains 3 lexical items, and there are 2 10 = 20 such largest LSA, for each of which there are 3! = 6 permutations. The fourth largest LSA contains 2 lexical items, and there are 2 10 = 20 such largest LSA, for each of which there are 2! = 2 permutations. The smallest LSA contains just 1 phase-defining lexical item, and there are 2 such smallest LSA. Thus, for this CP-phase LSA, there are (2 120) + (10 24) + (20 6) + (20 2) + 2 = 562 possible orders of Merge. Therefore, in the worst case scenario, Chomsky’s (2000, et seq.) Phase Theory with Merge over Move by LSA needs to try out whopping 35,229 + 562 = 35,791 orders of Merge, in order for the derivation to reach (41) correctly. On the other hand, if there is no LSA formation, there are at most 9P4 = 3,024 orders of Merge that need to be tried out, in order for the derivation to reach (41) from the stage where the most embedded CP was completed. It must be by now obvious, that the time-complexity involved in LSA is far

Dynamic economy of derivation

237

greater than without it, by an order of magnitude. MFR requires just 6 more options to be considered after the stage (41). Note, in passing, that the exponential problem of extracting the correct SLA can be resolved if the lexical items are ordered in LA for extraction, contrary to the standard assumption. Furthermore, the factorial problem of Merge orders can be resolved as well, by ordering the lexical items in LA or in SLA. Nevertheless, ordering of the lexical items nullifies the raison d’être of LSA itself; LSA is not needed if the ordering can control when the expletive is merged. As MFR makes a deterministic choice, the derivational paths are pruned down logarithmically as derivations proceed. Likewise, Chomsky’s (2000, et seq.) Phase Theory presumably discards all the derivational paths that are not chosen at every derivational stage, and yet LSA needs to be formed when the derivation starts and the previous LSA is exhausted. Thus, while the derivational paths are reduced logarithmically, there are stages where an exponential search of the desired LSA is involved, and consequently a factorial search of the correct derivational path as we have seen. (44)

LA q ty p LSA1a LSA1b LSA1c LSA1d … ← exponential search ty fgh fgh ty Σ1a Σ1b fh fh Σ2a Σ2b fgh g g g Σnx q ty p LSA2a LSA2b LSA2c LSA2d … ← exponential search ty fgh fgh ty Σ(n + 1)a Σ(n + 1)b fgh g

8. There as external argument An immediate question may arise with respect to the there-existential constructions we have reviewed for Shima’s (2000) static economy of Move over Merge in section 6. MFR appears to face a problem with the contrast (30), repeated below as (45).

238 Takashi Toyoshima (45) a. Therei seems [ ti to [be someone in the room]] b. *There seems [someonei to [be ti in the room]] Presumably, both (45a,b = 30a,b) are derived from the common intermediate stage (32), repeated below as (46). (46) [Tʹ to [be someone in the room]] MFR does not seem to be able to decide between Move (someone) and Merge (there) at this stage because of a tie; Move (someone) will delete the EPP-feature of T (to) but leave the Case-feature of someone unvalued. Merge (there) will also delete the EPP-feature of T (to) and leave the Casefeature of someone unvalued. Assuming with Chomsky (1995, et seq.) but contra Shima (2000) that there does not have any Case-feature, either Move (someone) or Merge (there) will leave the same number of features unvalued, namely the Case-feature of someone. We want Merge (there) for (45a,b = 30a,b), but we cannot exclude Move (someone) at the stage (46 = 32) for the ECM paradigm, such as the following: (47) a. *Mary believes [ to [be someone in the room]] b. Mary believes [someonei to [be ti in the room]] Do we need to fall back on Shima’s (2000) static economy of Move over Merge with the Case-theoretical assumptions? Rather than a Case-based approach as Shima (op. cit.) takes, I would pursue another possibility hinted by Lasnik (1995: 624ff., fn.14): “… However, if it could be stipulated in some such fashion that there can be introduced only as the subject of unaccusative verbs and (certain instances of) be, several of the phenomena that have come up in the present discussion would reduce to this one property, as observed by the reviewer. For example, the following would all be excluded, along with (16): (i) (ii) (iii) (iv)

* There seems/is likely [that John is tall]. (12) * There strikes John/someone that Mary is intelligent. (13) * I want [there someone here at 6:00]. (14b) * There seems/is likely [there to be a man here]. (from footnote 10)

(16) * There is likely [someonei to be [ti here]]. [(16) appended for reference: TT] The selectional restriction possibility thus deserves further investigation, which I put aside for another occasion.”

Dynamic economy of derivation

239

Kidwai (2002) independently argues that merger of the expletive there is restricted to Spec(v*P), a specifier of v with full argument structure: transitive v or v for experiencer verbs (Chomsky 2001: 9, 43, n. 8), surveying the observation by Levin & Rappaport-Hovav (1995) as follows (Kidwai 2002: 4 (10)): Verb type

there-existential

inherently directed motion manner of motion existence appearance occurrence spatial configuration

(arrive) (fall) (live) (appear) (ensue) (lie)

yes yes yes yes yes yes

disappearance change of state change of state change of state change of color change of state

(disappear) (break verbs) (bend verbs) (cook verbs) (-en/-ify verbs) (deadjectivals)

no no no no no no

From Levin & Rappaport-Hovav’s (1995) study, Kidwai (op. cit.) draws the generalization that the there-existential construction is possible only with unaccusatives that are incompatible with an interpretation of external causation and takes both a Locative and a Theme argument. Remarking further that passives (48) and some ergatives (verbs of internally caused change of state (49a), verbs of sound/light emission (49b), as well as agentive manner of motion verbs (49c)) participate in the there-existential construction, Kidwai (op. cit.: 14) proposes a configurational licensing condition of the EPP-specifier (50). (48) There was a building demolished.

(ibid. (41))

(49) a. There bloomed a rose in the garden. b. There boomed a cannon in the distance. c. There ran a man into the room.

(ibid. (42)) (ibid. (43)) (ibid. (44))

(50) v* can be assigned an EPP-feature iff v is merged with [VP DP V X(P)]. (ibid. (40))

240 Takashi Toyoshima The configuration depicted as [VP DP V X(P)] in (50) is instantiated as follows: for unaccusatives (51a), passives (51b), and unergatives (51c). (51) a. Unaccusatives VP wo DP Vʹ (Theme) 3 Vunacc PP (Locative) b. Passives VP wo DP Vʹ (Direct Object) 3 Vpass Prt (PARTICIPLE) c. Unergatives VP wo DP Vʹ (External 3 Argument) Vunerg PP (Locative/ Directional) Kidwai (2002) argues that External Argument is merged as Spec(vP) for externally caused predicates, that is, transitive and some unergative verbs, while (unergative) verbs of internally caused change of state, select “External Argument” as Spec(VP) as in (51c). Abstracting away from Kidwai’s proposal that the expletive there is merged due to the EPP-feature assigned to v* in the above configurations, I would propose the following. (52) The expletive there can optionally be selected as a kind of “External Argument” of v that selects an intransitive VP that is saturated. (53) An intransitive VP is saturated if its head verb V is either: i) an unaccusative verb that is incompatible with an interpretation of external causation and selects a Locative and a Theme argument;

Dynamic economy of derivation

241

ii) an unergative verb that selects a Locative/Directional argument and an External Argument of internally caused change of state, of sound/light emission, or of manner of motion; or iii) passivized participle. Then, I modify the intransitive structures as follows: (54) a. Unaccusatives vP wo there vʹ wo v VP wo Vʹ PP 3 (Locative) Vunacc DP (Theme) b. Passives vP wo there vʹ wo v VP (be) wo (XP) Vʹ (Indirect 3 Object) Vpass DP (PARTICIPLE) (Theme) c. Unergatives vP wo there vʹ wo v VP wo DP Vʹ (External 3 Argument) Vunerg PP (Locative/ Directional)

242 Takashi Toyoshima I take the Theme argument, i.e., direct object, to be the complement of both unaccusative and passivized verbs,20 and following Kidwai (2002), External Argument of internally caused change of state, of sound/light emission, of manner of motion, is located at Spec(VP). The idea that the expletive there is an “External Argument” is not anything new, as Chomsky (2004: 126, n.37) acknowledges; … Note that nothing requires that EA [External Argument: TT] be an actual argument: it could be a there-type expletive with φ-features …

Given these, the contrast (45 = 30), repeated below as (55), ceases to be a problem for MFR. (55) a. Therei seems [ ti to [be someone in the room]] b. *There seems [someonei to [be ti in the room]] The common intermediate stage for (55 a,b = 45a,b = 30 a,b) was not (46 = 32), repeated below as (56), but (57). (56) [Tʹ to [be someone in the room]] (57) [Tʹ to [there be someone in the room]] With the structure at the stage (57), the relevant options to be considered are: Move (there) and Merge of whatever DP, available in LA or constructed already in parallel. MFR dictates Move (there) here; Move (there) will delete the EPP-feature of T (to) and leaves the Case-feature of someone unvalued, whereas Merge (DP) can also delete the EPP-feature of T (to) but will “add” the Case-feature of that DP, which will be left unvalued, on top of leaving the unvalued Case-feature of someone. That is, Move (there) leaves fewer unvalued/uninterpretable features at the next stage in the derivation than Merge (DP), which adds an unvalued Case-feature of DP that is left unvalued. Notice that the number of unvalued/uninterpretable features is the same in the structure at the stage (57), i.e., the EPP-feature of T (to) and the Case-feature of someone, and either Move (there) or Merge (DP) deletes only the EPP-feature of T (to) and leaves someone’s Case-feature for the next stage of the derivation. 20

Although I do not exclude the possibility that the Locative PP in unaccusatives is a right-branching Spec(VP), I take it to be an adjunct internal to VP. In passives, the External θ-role is suppressed, but an “External Argument” can still be optionally realized as there in Spec(vP), I assume.

Dynamic economy of derivation

243

This is where the derivational notion of locality in MFR becomes crucial, taking a one-step “look-ahead,” and the economy principle I propose is formulated in terms of minimal feature retention, instead of “maximum feature elimination.” Accordingly, MFR chooses Move (there) and the derivation can proceed as follows: (59) a. b. c. d.

[TP therei [Tʹ to [ ti be someone in the room]]] [v/VP seems [TP therei [Tʹ to [ ti be someone in the room]]]] [Tʹ T [v/VP seems [TP therei [Tʹ to [ ti be someone in the room]]]]] [TP Therei [Tʹ T [v/VP seems [TP ti [Tʹ to [ ti be someone in the room ]]]]]]

If there were in LA but not selected, no derivation would be generated as the LA would not be exhausted. (60) *Someonei seems [tiʹ to [be ti in the room]]

{there}

If there were not in LA, the following would result: (61) Someonei seems [tiʹ to [be ti in the room]] Again, Merge of some other DP, say, another expletive it, even if available in LA, is not an option, either at the embedded Spec(TP) or at the matrix Spec(TP) by MFR. In the embedded clause, Merge (it) at Spec(TP) will delete the EPP-feature of T (to) but leaves the Case-feature of someone unvalued, and further “adds” the Case-feature of it itself unvalued. Move (someone) to Spec(TP) will delete the EPP-feature of T (to) and leave the Case-feature of someone unvalued, but “add” no further unvalued/uninterpretable features. In the matrix clause, Merge (it) at Spec(TP) will delete the EPP-feature of T and value the Case-feature of it itself, but leave the Case-feature of someone unvalued. Move (someone) to Spec(TP) will delete the EPP-feature of T and value the Case-feature of someone, and “add” no further unvalued/ uninterpretable features. This entails that the expletive it can only be Merged at Spec(TP) when all the unvalued/uninterpretable features have been deleted or valued in its domain, i.e., vP. In other words, it is the only “pure EPP expletive” that is “meaningless,” whereas there is a selected “External Argument” with an existential import. Consequently, the followings are underivable by MFR.

244 Takashi Toyoshima (62) a. [TP Iti [Tʹ T [v/VP seems [TP ti [Tʹ to [be someone in the room]]]]]] b. [TP It [Tʹ T [v/VP seems [TP someonei [Tʹ to [be ti in the room]]]]]] Furthermore, the following contrast that Shima’s (2000: 384, fn.8) Casebased approach has troubles with, will follow from MFR with the assumption that there is selected as an “External Argument” of v. (63) a. It seems that there is someone in the room. b. *There seems that it is someone in the room. Shima (ibid.) admits: The preference for Move over Merge has nothing to do with the choice between it-insertion and there-insertion, and it, with no partitive Case, does not block association of someone, with partitive Case, to there. … I tentatively speculate that [Spec, T], which selects the copular be with partitive Case, must be filled by there rather than it, but otherwise I leave this problem open. [emphasis in bold added: TT]

That is, Shima (op. cit.) has to resort to “selection” of there after all, on top of his Case-theoretical assumptions (31), which was dubious as we have seen. Finally, for the ECM paradigm (47), repeated as (64) below, if there were in LA and selected, (65) would have been derived. (64) a. *Mary believes [ to [ be someone in the room]] b. Mary believes [someonei to [ be ti in the room]] (65) Mary believes [therei to [ ti be someone in the room]] If there were in LA but not selected, no derivation would be generated as the LA would not be exhausted. (66) *Mary believes [someonei to [be ti in the room]] If there were not in LA, (64b) would result.

{there}

Dynamic economy of derivation

245

9. Concluding remarks As I have shown, determination of Lexical Subarray (LSA) is not so straightforward that computational complexity can be reduced as Chomsky (2000) has envisaged, and therefore, it should be eliminated from the theory. The question, which motivated the concept of LSA, why Merge of an expletive does not always preempt Move, can be answered by MFR (38) that dynamically makes a derivationally local deterministic choice between (external) Merge and Move (internal Merge) without recourse to how lexical items are made available for narrow syntactic computation. (38) Principle of Minimum Feature Retention (MFR) At every stage Σn of a derivation, choose an operation that leaves the fewest unvalued/uninterpretable features in the resulting stage Σn + 1 of the derivation. MFR invokes one-step “look-ahead” Σn → Σn + 1, which is the crucial notion of “derivational locality,” without which the notion of locality will end up being equivalent to representational constraints. For there-existential constructions, MFR appears to face a problem of indeterminacy because of a tie, but it is only apparent since there is a selected “External Argument” of an intransitive VP that is saturated. (52) The expletive there can optionally be selected as a kind of “External Argument” of v that selects an intransitive VP that is saturated. (53) An intransitive VP is saturated if its head verb V is either: i) an unaccusative verb that is incompatible with an interpretation of external causation and selects a Locative and a Theme argument; ii) an unergative verb that selects a Locative/Directional argument and an External Argument of internally caused change of state, of sound/light emission, or of manner of motion; or iii) passivized participle. Insofar as I can see, there is no genuine case of a tie with regard to MFR. This implies that there is no real syntactic optionality. Alleged optionalities of word order, or constructional alternatives, must be due to lexical choices, distinct specifications of feature composition, not by the free choice of syntactic operations that economy conditions make available.

246 Takashi Toyoshima With regard to the notion of phase, I remain agnostic but open. Yet, I contend that a phase must be redefined in some other fashion than through Lexical Subarray, determination of which imposes a significant computational load to “exo-syntactic” processes in an exponential order as I have shown. My hunch is that a strong phase is to be defined, if it needs to be, in terms of completed formal licensing; that is, when XP has deleted/valued or moved away all the uninterpretable/ unvalued features, it is a strong phase. Otherwise, it is a weak phase, which I do not think needs to be defined. It remains to be seen whether or not Phase-Impenetrability Condition and Defective Intervention Constraint survive their functions after a new definition of phase is ever made.

Appendix: Possible LSAs 6C 0

{that}

6C 0

{were}

6C 1

{T1, that} {seem, that} {that, T2} {that, told} {that, friends} {that, it}

6C 1

{T1, were} {seem, were} {T2, were} {were, told} {were, friends} {were, it}

6C 2

{T1, seem, that} {T1, that, T2} {T1, that, told} {T1, that, friends} {T1, that, it} {seem, that, T2} {seem, that, told} {seem, that, friends} {seem, that, it} {that, T2, told} {that, T2, friends} {that, T2, it} {that, told, friends} {that, told, it} {that, friends, it}

6C 2

{T1, seem, were} {T1, T2, were} {T1, were, told} {T1, were, friends} {T1, were, it} {seem, T2, were} {seem, were, told} {seem, were, friends} {seem, were, it} {T2, were, told} {T2, were, friends} {T2, were, it} {were, told, friends} ❖ {were, told, it} {were, friends, it}

Dynamic economy of derivation

247

6C 3

{T1, seem, that, T2} {T1, seem, that, told} {T1, seem, that, friends} {T1, seem, that, it} {T1, that, T2, told} {T1, that, T2, friends} {T1, that, T2, it} {T1, that, told, friends} {T1, that, told, it} {T1, that, friends, it} {seem, that, T2, told} {seem, that, T2, friends} {seem, that, T2, it} {seem, that, told, it} {seem, that, told, friends} {seem, that, friends, it} {that, T2, told, friends} {that, T2, told, it} {that, T2, friends, it} {that, told, friends, it}

6C 3

{T1, seem, T2, were} {T1, seem, were, told} {T1, seem, were, friends} {T1, seem, were, it} {T1, T2, were, told} {T1, T2, were, friends} {T1, T2, were, it} {T1, were, told, friends} {T1, were, told, it} {T1, were, friends, it} {seem, T2, were, told} {seem, T2, were, friends} {seem, T2, were, it} {seem, were, told, it} {seem, were, told, friends} {seem, were, friends, it} {T2, were, told, friends} {T2, were, told, it} {T2, were, friends, it} {were, told, friends, it}

6C 4

{T1, seem, T2, were, told} {T1, seem, T2, were, friends} {T1, seem, told, were, friends} {T1, T2, were, told, friends} {seem, T2, were, told, friends} {T1, seem, T2, were, it} {T1, seem, were, told, it} {T1, T2, were, told, it} {seem, T2, were, told, it} {T1, seem, were, friends, it} {T1, T2, were, friends, it} {seem, T2, were, friends, it} {T1, were, told, friends, it} {seem, were, told, friends, it} {T2, were, told, friends, it}

6C 4

{T1, seem, that, T2, told} {T1, seem, that, T2, friends} {T1, seem, that, told, friends} {T1, that, T2, told, friends} {seem, that, T2, told, friends} {T1, seem, that, T2, it} {T1, seem, that, told, it} {T1, that, T2, told, it} {seem, that, T2, told, it} {T1, seem, that, friends, it} {T1, that, T2, friends, it} {seem, that, T2, friends, it} {T1, that, told, friends, it} {seem, that, told, friends, it} {that, T2, told, friends, it}

6C 5

{T1, seem, T2, were, told, friends} {T1, seem, T2, were, told, it} {T1, seem, T2, were, friends, it} {T1, seem, were, told, friends, it} {T1, T2, were, told, friends, it} {seem, T2, were, told, friends, it}

6C 5

{T1, seem, that, T2, told, friends} {T1, seem, that, T2, told, it} {T1, seem, that, T2, friends, it} {T1, seem, that, told, friends, it} {T1, that, T2, told, friends, it} {seem, that, T2, told, friends, it}

6C 6

{T1, seem, T2, were, told, friends, it}

6C 6

{T1, seem, that, T2, told, friends, it}

248 Takashi Toyoshima Acknowledgements This work is based on my earlier work, Toyoshima (1999, 2000), and is a revised version of the paper delivered at InterPhases: A Conference on Interfaces in Current Syntactic Theory, held in Nicosia, Cyprus, 18–20 May, 2006. I thank the audience for the questions and comments raised during the conference, and I am indebted to Kleanthes Grohmann for his patience in editing and an anonymous reviewer for detailed constructive criticisms, to which I hope not to have failed to do due justice. Any remaining errors or shortcomings are my own. This work is partially supported by a Grant-in-Aid for Scientific Research (C) #14510626 and a Grad-in-Aid for Exploratory Research #19652044 from the Japan Society for the Promotion of Science, which I gratefully acknowledge here.

References Belletti, Adriana 1998 The case of unaccusatives. Linguistic Inquiry 19: 1–34. Blum, Manuel 1967 A machine-independent theory of the complexity of recursive functions. Journal of the Association for Computing Machinery 14: 322– 336. Chomsky, Noam 1951 The Morphophonemics of Modern Hebrew. M.A. Thesis: University of Pennsylvania. [Published in 1979 from Garland Publishing: New York.] 1955 The logical structure of linguistic theory. Ms., Harvard University. [Published partially with revision in 1975 from Plenum Press: New York. Reprinted in 1985 from Chicago University Press: Chicago, IL.] 1973 Conditions on transformations. In A Festschrift for Morris Halle, Stephen R. Anderson and Paul Kiparsky (eds.). 232–286. New York: Holt, Rinehart and Winston. 1992 Some notes on economy of derivation and representation. In Principles and Parameters in Comparative Grammar, Robert Freidin (ed.), 417–454. 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. 1995 Categories and transformations. In The Minimalist Program, Noam Chomsky, 219–394. Cambridge, MA: MIT Press.

Dynamic economy of derivation 1998

249

Some observations on economy in generative grammar. In Is the Best Good Enough?: Optimality and Competition in Syntax, Pilar Barbosa, Danny Fox, Paul Hagstrom, Martha McGinnis and David Pesetsky (eds.), 115–127. Cambridge, MA: MIT Press and MITWPL. 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– 141. Oxford: 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. Church, Alonzo 1941 The Calculi of Lambda-Conversion. Princeton, NJ: Princeton University Press. Collins, Chris 1997 Local Economy. Cambridge, MA: MIT Press. Edmonds, Jack 1965 Paths, trees, and flowers. Canadian Journal of Mathematics XVII: 449–467. Fitch, W. Tecumseh, Marc D. Hauser, and Noam Chomsky 2005 The evolution of the language faculty: Clarifications and implications. Cognition 97: 179 –210. Fukui, Naoki 1996 On the nature of economy in language. Cognitive Studies: Bulletin of the Japanese Cognitive Science Society 3: 51–71. Garey, Michael R. and David S. Johnson 1979 Computers and Intractability: A Guide to the Theory of NP-Completeness. San Francisco, CA: W. H. Freeman. Groat, Erich and John O’Neil 1996 Spell-out at the LF interface: Achieving a unified syntactic computational system in the Minimalist Framework. In Minimal Ideas: Syntactic Studies in the Minimalist Framework, Werner Abraham, Samuel D. Epstein, Höskuldur Thráinsson and C. Jan-Wouter Zwart (eds.), 113–139. Amsterdam: John Benjamins. Hartmanis, Juris and John E. Hopcroft 1971 An overview of the theory of computational complexity. Journal of the Association for Computing Machinery 18: 444–475.

250 Takashi Toyoshima Hauser, Marc D., Noam Chomsky and W. Tecumseh Fitch 2002 The faculty of language: What is it, who has it, and how did it evolve? Science 298: 1569 –1579. Hinzen, Wolfram 2008 The successor function + LEX = human language? In InterPhases: Phase-Theoretic Investigations of Linguistic Interfaces, Kleanthes K. Grohmann (ed.). Oxford: Oxford University Press. Johnson, David S. 1990 A catalog of complexity classes. In Handbook of Theoretical Computer Science A: Algorithms and Complexity, Jan van Leeuwen (ed.), 67–161. Amsterdam: Elsevier / Cambridge, MA: MIT Press. Johnson, David E. and Shalom Lappin 1997 A critique of the Minimalist Program. Linguistics and Philosophy 20: 273–333. 1999 Local Constraints vs. Economy. Stanford, CA: CSLI Publications. Kidwai, Ayesha 2002 Unaccusatives, expletives, and the EPP-feature of v*. Ms., Jawaharlal Nehru University, New Delhi. Kitahara, Hisatsugu 1995 Target α: deducing strict cyclicity from derivational economy. Linguistic Inquiry 26: 47–77. Kleene, Stephen C. 1936 General recursive functions of natural numbers. Mathematische Annalen 112: 727–742. Lasnik, Howard 1995 Case and expletives revisited: On Greed and other human failings. Linguistic Inquiry 26: 615–633. Levin, Beth and Malka Rappaport-Hovav 1995 Unaccusativity: At the Syntax–Lexical Semantics Interface. Cambridge, MA: MIT Press. Markov, Andrey A. 1954 Teoriya Algorifmov. Akademii Nauk SSSR: Moskva. [Theory of Algorithms. The Academy of Sciences of the USSR: Moscow. English translation by Jacques J. Schorr-Kon and PST Staff (1961) Theory of Algorithms. The Israel Program for Scientific Translations: Jerusalem.] McGinnis, Martha 1998 Locality in A-movement. Doctoral dissertation, MIT, Cambridge. Nakamura, Masanori 1998 Global issues. Proceedings of the North East Linguistic Society 28 (1): 301–318. Nevins, Andrew 2004 “Derivations without the Activity Condition. In Perspectives on Phases: MIT Working Papers in Linguistics 49, Martha McGinnis & Norvin Richard (eds.), 287–310. Cambridge, MA: MITWPL.

Dynamic economy of derivation

251

Papadimitriou, Christos H. 1994 Computational Complexity. Reading, MA: Addison-Wesley. Papadimitriou, Christos H. and Kenneth Steiglitz 1982 Combinatorial Optimization: Algorithms and Complexity. Englewood Cliffs, NJ: Prentice-Hall. Poole, Geoffrey 1998 Constraints on local economy. In Is the Best Good Enough? Optimality and Competition in Syntax, Pilar Barbosa, Danny Fox, Paul Hagstrom, Martha McGinnis and David Pesetsky (eds.), 385–398. Cambridge, MA: MIT Press and MITWPL. Post, Emil L. 1936 Finite combinatory processes – formulation 1. The Journal of Symbolic Logic 1: 103–105. Seiferas, Joel I. 1990 Machine-Independent Complexity Theory. In Handbook of Theoretical Computer Science A: Algorithms and Complexity, Jan van Leeuwen (ed.), 163–186. Amsterdam: Elsevier / Cambridge, MA: MIT Press. Shima, Etsuro 2000 A preference for Move over Merge. Linguistic Inquiry 31: 375–385. Toyoshima, Takashi 1999 Move 1st: A Dynamic Economy Plan. In Proceedings of the North East Linguistic Society 29 (1): 409–425. 2000 Head-to-Spec movement and dynamic economy. Doctoral dissertation. Cornell University, Ithaca, NY. 2005 Preemptive move toward elimination of lexical subarray: Dynamic economy. Proceedings of the Israel Association for Theoretical Linguistics 21. (http://atar.mscc.huji.ac.il/~english/IATL/21/). Turing, Alan M. 1937 On computable numbers, with an application to the Entscheidungsproblem. Proceedings of the London Mathematics Society: Second Series 42: 230 –265. Yang, Charles D. 1997 Minimal Computation: Derivation of the Syntactic Structure. M.Sc. Thesis, MIT, Cambridge.

The conceptual necessity of phases: Some remarks on the minimalist enterprise Dennis Ott

1. Biolinguistics and Turing’s Thesis I would like to discuss some very general issues pertaining to recent proposals in linguistic theory. In order to do this, it is necessary to review some fundamental aspects of the Minimalist Program (MP). This first section introduces the framework, emphasizing that the MP is not a theory of language but a bet on what kind of biological object I-language is. One particular theoretical reflex of this ontological (actually, metaphysical) commitment of the program – the notion of syntactic phases – will be investigated in section 2. Section 3 is a brief comment on some conceptual problems, and section 4 concludes.1 The overall goal of this paper is very modest: to show that once certain plausible assumptions about the architecture of the Language Faculty are adopted, the notion of phase is a conceptual necessity. The question of what kind of object an I-language is is closely related to the evolutionary origin of natural language. Evidently, our tentative assumptions about the nature of the language organ depend to a significant extent on our equally tentative assumptions about the major factors that enter into its Entwicklungsgeschichte (evolution and development). As many authors have stressed, it seems reasonable to assume that an Ilanguage provides an evolutionary advantage, and that a linguistic organism will be favored by selection. While this vague assumption appears relatively innocuous, it does not ipso facto provide any insight into the structure of Ilanguage. The problem of morphogenesis, which arises in the study of the evolution of any biological object, is summarized by developmental biologist Gunther J. Eble, who notes that “form, and more generally, the phenotype, has always remained a fundamental problem demanding explanation in evolutionary biology. (…) While the structured entities of evolutionary dynamics certainly include genes, structure is not equivalent to genetic informa1

In what follows, MI = Chomsky (2000), DbP = Chomsky (2001), BEA = Chomsky (2004), OP = Chomsky (to appear), AUB = Chomsky (2007).

254 Dennis Ott tion; genes are necessary but not sufficient for the emergence of form” (Eble 2003: 34). Hence, “Darwin’s Dilemma” (Boeckx 2006a) remains in the age of the modern synthesis, calling for a general internalist research program in theoretical biology.2 The MP is one such program, appealing to structural principles of evolution and development. These observations are not without precursors: the gist of Alan Turing’s seminal work on morphogenesis is that “the forms we observe in organisms are often determined more by physics than by selection,” in the words of mathematician-biologist Peter Saunders (1994: 373). This is what we may call, with Chomsky (2006), Turing’s Thesis. Discussing the role of developmental morphospaces, Eble (2003: 42) notes that “[p]hysical models of form have been shown to be very suggestive of actual principles involved [in defining morphospaces]. (…) To the extent that such physical principles are involved in development, they become part and parcel of the developmental matrix that is potentially involved in structuring evolution in morphospace.” Unless we adopt some methodological dualism, this should apply to I-language as it does to other organs of the body; natural law is a “Third Factor” that enters into its design, as does (presumably) genetic information.3 For the domain of I-language, Darwin’s Dilemma is aggravated by the findings of theoretical linguistics: the entities postulated by linguistic theory (say, c-command or the CED) are unlikely to be results of successive, gradual adaptation of the language organ to adaptive pressures. More plausibly, they are epiphenomenal effects of a mutation that had deep consequences: 4 Many of the details of language (…) may represent by-products of [FLN, the Faculty of Language in the narrow sense], generated automatically by neural/computational constraints and the structure of [FLB, the Faculty of Language in the broad sense] (…). [The development of the Faculty of Language is] constrained by biophysical, developmental, and computational factors shared with other vertebrates. (…) [S]tructural details of FLN may result 2

3 4

For review see e.g. Webster & Goodwin (1996). See also Gould (2002) for an extensive outline of internalism in biology, reaching back to Goethe’s rationalist morphology. For general discussion of the Third Factor, see Chomsky (2005a,b). This conjecture is supported by the biological isolation of I-language, which is a species-specific human property, apparently without any relevant evolutionary precursors; notice e.g. Gould’s (1991: 59) assertion that “attributes unique to our species are likely to be exaptations,” hence not to be explained in terms of selection.

The conceptual necessity of phases

255

from such preexisting constraints, rather than from direct shaping by natural selection targeted specifically at communication. Insofar as this proves to be true, such structural details are not, strictly speaking, adaptations at all. (Hauser et al. 2002: 1574)

In general, the internal make-up of I-language does not appear to be specifically adjusted to communicative demands, and its essentials might to a significant extent be determined by the Third Factor (perhaps along with some minor co-adaptation of grammar and external systems, a possibility not contradicted by Hauser et al.’s suggestions).5 The MP assumes Turing’s Thesis to be on the right track, for I-language and beyond. As Boeckx (2006b) argues at length (see also Chomsky 2002: ch. 4), this speculative metaphysical thesis has proven fruitful for natural science in its quest for explanation beyond description: phenomenology is a hopeless chaos, but this chaos is merely the distorted expression of a higher nature, where basic laws and principles interact. To understand the chaos as peripheral and our models of the higher nature behind the chaos as real is perhaps the most important aspect of the Galilean scientific revolution. What does this mean for the study of I-language? Adopting Turing’s Thesis, the MP is an attempt to show that I-language approaches some significant degree of structural optimality, understood in terms of naturalness: its internal mechanisms and operations are not arbitrary, but a least-effort solution to minimal design specifications.6 Given that at least a subset of the body of knowledge that is given to the speaker by her I-language is usable, we can assume that (some of) the outputs of its computations are poised for access by external (performance) systems: to some extent, I-language meets interface conditions, sometimes termed “bare output conditions” (Chomsky 1995: 221; BEA: 106). If I-language is essentially an expression of the Third Factor, we expect it to meet interface conditions in a way that makes use of the minimal machinery necessary.7 Assuming that the system interfaces 5

6

7

For all we know, a human I-language might be the natural result of self-organizational principles of neural tissue, and the changes it underwent in the “great leap forward.” We can only speculate, but there is some suggestive work in neuroscience, most notably by Alessandro Treves, on the neural basis of infinite recursion (see Treves 2005). For specific proposals along these lines, see Chomsky (1995 et seq.), Fox (2000), and the papers in Epstein & Seely (2002), and much related work. Moreover, once an element in the design of I-language is identified as an interface requirement, it receives principled explanation: “Insofar as properties of [Ilanguage] can be accounted for in terms of [interface conditions] and general

256 Dennis Ott (minimally) with sensorimotor systems on the one hand and systems of conceptual organization and intentionality on the other, this claim amounts to saying that I-language is a perfect solution to the task of linking these kinds of systems. The Strong Minimalist Thesis (SMT) in its most rigorous form maintains exactly this assumption: (SMT) I-language comprises only the minimal mechanisms necessary to compute representations that can be accessed by the interfacing external systems.8 SMT is a guide line for theory construction, based on the bet that Turing captured some deep aspect of nature, and that I-language, due to its extraordinary evolutionary (non-)history, reflects this nature in a rather straightforward way. If SMT held fully, all principles of I-language would reduce to natural law or to interface conditions, hence in virtue of metaphysical necessity or the structure of external systems, not genetic encoding (that is, the principles would come “for free”). In this sense, the MP “approaches UG from below” (AUB) in that it attempts to shift the explanatory burden from the genetic endowment to natural principles of computation. A successful pursuit of the MP could eventually lend support to theories of I-language evolution along the lines of Hauser et al. (2002), making them more than a reasonable bet: “the less attributed to genetic information (in our case, the topic of UG) for determining the development of an organism, the more feasible the study of its evolution” (AUB: 3).

2.

The necessity of phases

2.1. Unconstrained merge and “overgeneration” As a first approximation, let us assume that I-language comprises a lexicon, a (narrowly) syntactic component, and (broadly) syntactic mapping components – “broadly” syntactic because they’re often labeled “semantics” and “phonology”, depite their presumably syntactic character. This architecture suffices in principle to generate an infinite number of (mind-internal) Ex-

8

properties of efficient computation and the like, they have a principled explanation: we will have validated the Galilean intuition of perfection of nature in this domain” (BEA: 2). For other formulations of SMT, see MI: 96, DbP: 1.

The conceptual necessity of phases

257

pressions, in the technical sense yet to be made precise. Minimally, UG makes available an operation of set-formation or “Merge”, which creates a syntactic object: Merge(X,Y) = {X,Y}. Recursive Merge yields digital infinity. Another way of saying this is that UG equips lexical items (LIs) with Edge Features (EFs), in the terminology of OP. EF on LI will result in LI being merged to some other LI. Iterated Merge yields complex structures as well as “dislocation”; merging an object that is already present in the structure yields multiple occurrences of the object in the tree. Following BEA (p. 110), we can descriptively distinguish external and internal Merge: if an element is merged from the lexicon, it is merged externally; if it is part of the already-formed structure, it is merged internally. But notice that the distinction is purely descriptive; no operation of movement or copying must be stipulated in addition to Merge. Dislocation follows logically as a subcase of unconstrained Merge (see Berwick 1998 for a very early formulation of this insight). In addition to elementary tree construction by Merge, there seem to be grammatical operations triggered by formal features of the elements in the derivation (surfacing in inflectional morphology). If some element bears a formal feature that is not valued in the lexicon, it reflexively probes the structure (naturally, its c-command domain) in order to find some valued non-distinct feature to agree with, its goal. Simplifying massively, it seems that Agree(X,Y) may trigger IM of the goal, but need not, depending on the probe’s EPP/EF status.9 Two mutually exclusive views of Merge have been advanced in the literature: 1. Merge is subject to Last Resort: each application of Merge(X,Y) must be licensed by a probe-goal relation between X and Y. 2. Merge applies freely (or, alternatively: EF on LI deletes freely). Most of the literature effectively takes the first route, assuming some kind of featural relation to ensure “correct” selection at each Merge step (see, e.g., Svenonius 1994; Collins 2002; Frampton & Gutmann 2002; Adger 2003; Pesetsky & Torrego 2004), and at least in some cases arguing explicitly that this makes syntax “crash-proof.” Chomsky has argued for the second view (see, in particular, BEA: 111f.). He argues that we can dispense with both c-selection (Pesetsky 1982) and s-selection as properties of syntax. Rather, selectional properties are “read 9

For details, see MI, DbP, BEA, OP, AUB, and references therein.

258 Dennis Ott off” the relevant heads at the semantic interface. This view implies that syntax is not crash-proof, and that Merge will freely generate an infinitude of structures that violate selectional (and other) requirements. At the interface, the interpretation of a head in a particular configuration will induce deviance if selectional properties are violated. Take, for instance, UTAH, as discussed by Baker (1997). If Baker is on the right track, his proposals raise the question whether UTAH is built into Merge, conditioning its application, or whether it is imposed by C-I systems. The minimalist working assumption (and Baker’s) is that UTAH is a descriptive statement of C-I interpretation principles, while the (undersirable) alternative would be to enrich narrow syntax accordingly. Notice that this would conflict with the “from below” approach of the MP, which seeks to minimize the amount of syntax-specific mechanisms. Free application of Merge entails the existence of an infinite array of structures that do not conform to externally imposed interface conditions (IC). In particular, “[elimination of s-selection] entails that derivations cannot be failure-proof (“crash-free”), thus undermining the strong version of IC (…). IC must be weakened” (BEA: 9f.), allowing for degrees of deviance. It seems to me that only this second view of Merge is adequate, given the flexibility of interpretation and the gradedness of deviance. Consider deviant cases like “John slept Mary”, “John kicked”, or “John ran Mary blue”, all of which are assigned some interpretation (and the same is true for many other violations in various domains; actual cases of true unintelligibility are hard to make up). As Chomsky (p.c.) puts the matter, “it’s important to remember that we do want to ensure that ‘deviant’ expressions are generated, even word salad. The I-language assigns it an interpretation, reflexively, with only limited variability” (my emphasis). In fact, the argument recapitulates a familiar truism, going back to Chomsky (1955), that natural language has no notion of well-formedness or “grammaticality” – a striking contrast with formal languages. Rephrasing this insight in modern terminology, Chomsky (p.c.) suggests that conditions on the application of Merge are actually empirically inadequate, and that a more plausible assumption is that degrees of deviance are interface effects (see also OP: 10, BEA: 3): We could add further conditions [on Merge] to block ‘overgeneration’ (e.g., no more SPECs after theta roles have been exhausted, keep to selectional requirements, etc.). But that seems to me wrong. First, such conditions duplicate what’s done independently at the C-I interface. Second, it imposes some ±grammatical distinction, and since LSLT in 1955, that’s seemed wrong to me.

The conceptual necessity of phases

259

I will assume that this view is correct, and that subjecting Merge to Last Resort is an empirically and conceptually dubious move. It is unclear how a crash-proof syntax could yield a substantial part of the linguistic knowledge that speakers undoubtedly have, and how the gradedness of deviance could arise. We thus shift the explanatory burden to interface conditions, while syntax is free to generate an infinite array of structures. It is plausible to assume that conceptual-intentional (C-I) and sensorimotor (SM) systems impose quite different conditions; call the syntactic objects mapped onto these interfaces SEM and PHON, respectively. Assume that in analogy to PHON, we can view SEM essentially as an instruction to construct a complex concept (see Pietroski to appear, forthcoming). The conceptual system will naturally impose conditions on what counts as a proper instruction; only syntactic objects of a certain kind will be able to interface with “thought” (whatever the relevant properties are). These syntactic objects are what Chomsky calls “phases,” and the empirical project is to determine what makes a syntactic object a phase.

2.2. Propositional cycles By definition, phases are those syntactic objects that can interface with “thought.” At the same time, the phase defines the syntactic cycle, a traditional notion in linguistic theory (see Lasnik 2006 for review). C-I interface conditions thus determine that the syntactic cycle must be “meaningful” (BEA: 108), corresponding to a proper instruction for concept construction. The question, then, is: What makes a syntactic object “meaningful” in the relevant sense?10 Chomsky suggests that phasal syntactic objects are defined as such by their “propositionality”:

10

Chomsky’s original formulation of the phase was mainly extensional: In MI and DbP, a phase is defined relative to a lexical subarray (LA). The LA, a collection of items from the lexicon, is exhausted by the computational system; the resulting structure (the phase) is transferred, and the next subarray is used. The argument for subarrays was mainly based on “Merge over Move” effects, an incoherent notion once the assumptions about internal/external Merge in BEA and later work are adopted. I will not discuss this view any further.

260 Dennis Ott Ideally, phases should have a natural characterization in terms of [interface conditions, IC]: they should be semantically and phonologically coherent and independent. At SEM, v*P and CP (but not TP) are propositional constructions: v*P has full argument structure, CP is the minimal construction that includes Tense and event structure [note omitted, DO], and (at the matrix at least) force. (BEA: 124; cf. also MI: 106, DbP: 12)

Let us summarize this statement as a first approximation: 1. CP is a phase because it specifies force/full event structure. 2. v*P is a phase because it represents full argument structure. CP and v*P thus correspond to what Chomsky calls the “duality of semantics”: local predicate-argument structure, and derived positions, which both enter into the semantic interpretation of a full expression. The particular phases we find in human syntax are thus not a matter of necessity; if the CI system were structured differently, different structures would be ‘picked out’ for concept construction. It is perfectly coherent to imagine a C-I system that can only deal with local theta-structure configurations but has no use for CP-like structures. The structure of Expressions is thus not determined by narrow syntax, but by C-I properties.11 Given the structural parallelisms of the verbal and nominal domains (cf. Chomsky 1970; Abney 1987, and much subsequent work), we can speculate that certain nominal structures will also meet Chomsky’s criterion of propositionality. Longobardi (1994, 2005) provides substantial evidence that there are at least two fundamentally different kinds of nominal phrases: referential, definite DP (“the picture of John”) and predicative, indefinite NP (“a picture of John”). Longobardi shows that only DP can be an argument, with either a definite determiner or the head noun raising covertly to D. Assuming that intentional systems allow the speaker to use certain linguistic expressions to refer to non-linguistic entities, Longobardi’s work can be interpreted as showing that it is D that encodes referentiality of DP, making it a “special” unit at the interface, hence that DP (but not NP) is a phase (cf. AUB: 25f.). Let us thus add a third hypothesis: 3. DP is a phase because it is referential.

11

Hinzen’s (2006) program is exactly the opposite of the view described here. Hinzen argues against independent C-I constraints, arguing that all semantic properties are fully determined by intrinsic properties of syntax.

The conceptual necessity of phases

261

The identification of CP, v*P and DP as phases is plausible, given the duality of semantics plus referential use of nominal expressions. Needless to say, hypotheses 1–3 require much clarification and empirical justification. However, notice that some notion of phase is conceptually necessary in any syntactic theory, in particular when Merge is assumed to be unconstrained. There must be some way of “picking out” those special structures that can instruct C-I systems; there is nothing in the structures themselves that marks them propositional (contra Hinzen 2006). In this sense, any generative model has to give substance to the notion of a “meaningful” cycle; the relevant C-I constraints being of course a matter of empirical discovery. Hence, it is of central importance to linguistic theory to develop a proper technical notion of propositionality that applies to those structures that instruct conceptual systems. Earlier theories took for granted that the basic unit of computation is essentially the clause (CP), i.e. that there is only a single phase. Phase theory denies that the clause is the only relevantly “special” structure: parallel propositional structures are expected in the verbal (v*P) and nominal (DP) domains as well. If true, the phases presumably reflect deep aspects of the C-I systems, perhaps divided along the following lines: – Information/discourse structure  CP – Thematic structure  v*P – Referentiality  DP

Notice that the multiple-phase approach pushes many global properties of (semantic and phonological) Expressions outside of grammar proper. Instances of long-distance binding and global aspects of interpretation beyond local compositionality (discussed in Chomsky 1977) are now a matter of computation beyond the interface. I-language provides an infinite array of propositional Expressions (clausal, verbal, and nominal), which are apt for concept construction. But the C-I system also interprets “full expressions” (something like sentences) that comprise several propositional units. Likewise, many prosodic and intonational properties are presumably computed over cumulative phonetic representations in the SM system.12 Hence, much 12

Notice in this regard that the basic intonation pattern of a sentence is at least in part determined by C, the functional head specifying force. However, C is fed into the phonological component only at the very end of the C phase; therefore, the computation of an intonation matrix requires a full expression, hence must take place outside of grammar.

262 Dennis Ott more computational work is ascribed to external systems in assembling the “full expression” that the speaker actually interprets in language use. In contrast to earlier models of locality (see Boeckx & Grohmann 2007 for review), phases are both propositional units of interpretation and derivational units. Hence, the phasal cycle should have syntactic repercussions.

2.3. Derivational cyclicity Notice that the phase-based model is a significant departure from the standard Y-model of syntax (Chomsky & Lasnik 1993). The previous independent cycles of X-bar theory, D-structure, S-structure and LF (as the output of narrow syntax and the input to covert operations) are reduced to a single cycle that maps a set of lexical items to the semantic interface, mapping to SM being an ancillary process (see DbP: 5, 15). Assuming that phases define syntactic cycles as well as units of interpretation, the size of the cycle is expected to restrict application of operations. Chomsky attempts to relate the notions of phasal derivation and minimal search space by assuming that phases are in some sense impenetrable for later operations. As Müller (2003: 6) phrases this desideratum, “minimality effects should emerge as epiphenomena of constraints that reduce the space in which the derivation can look for items that may participate in an operation (…).” The goal is to reduce phenomena of locality to the the necessary conditions on conceptual instructions. Concretely, Chomsky assumes that once a phase is completed, its domain (the complement of the phase head) is transferred: (1)

Phase Impenetrability Condition, version 1 (PIC1) In phase HP with head H, the domain of H is not accessible to operations outside HP; only H and its edge are accessible to such operations. (Cf. MI: 108)

Notice that the PIC as such does not state whether it holds for narrow syntax, for the mapping components, or both. Since the options are logically distinct, the domain of application of PIC is an empirical matter. Assume first that narrow syntax forgets transferred phases. As an illustration, consider the reduction of T’s search space under PIC1, printed boldface: 13 13

The notation is borrowed from Richards (2006), who in turn relies on Müller (2003).

The conceptual necessity of phases

(2)

263

T [v*P … v* … [VP … V … ] ]

As soon as the v*P phase is completed, VP is transferred, hence no longer accessible to T. Notice that as the above condition states, it is not exactly the phase itself that gets transferred; rather, the phase head’s complement domain is transferred. Successive-cyclic movement is a necessary consequence of (1), forcing elements to move through intermediate phase edges on their way to the final landing site. In addition to familiar empirical evidence for successive-cyclic movement (see Boeckx 2007 for extensive review), Fox (2000) and others have specifically argued for reconstruction effects at the v*P edge: (3)

a. [Which of the papers that hei wrote for Ms. Brownk] did every studenti get herk to grade twh b. *?Every student got herk to grade the first paper that he wrote for Ms. Brownk

The binding relations indicated in (3a) require a copy of the wh-phrase in a lower position. (3b) shows that the relevant reconstruction site cannot be the base position, which yields a Principle C violation. Thus, the reconstruction site must be in between, leaving SPEC-v* as the most plausible option: (3) a’. [TP every studenti did [v*P [which of the papers that hei wrote for Ms. Brownk] [v*P get herk to grade twh]]] A’-reconstruction thus provides some evidence for the special status of the v*P edge, as expected if v*P is a phase. Notice that Chomsky explicitly excludes defective v (passive and unaccusative) from his typology of phase heads. His motivation for restricting phasal status to v*P is that subextraction out of an agentive subject yields stronger deviance than subextraction from a passive subject (see OP: 14 for discussion). However, Legate (2003) casts doubt on Chomsky’s identification of v*P as the only relevant verbal structure. While confirming the phasal status of v*P (and CP), Legate argues on the basis of reconstruction effects, parasitic gaps and prosodic data that passive and unaccusative vP is equally phasal, exhibiting the same degree of ‘interface isolability.’14 As an 14

As a result of Legate’s caveat, Chomsky (in DbP) introduces the distinction of strong vs. weak phases. The distinction is entirely obscure: while abiding his original conception by referring to CP and v*P as strong phases, Chomsky acknowledges the phasal character of the relevant VPs, declaring them weak

264 Dennis Ott illustration, consider the pair in (4). In both the active (4a) and the passive (4b) case, the bracketed wh-phrase reconstructs below every student/every man, i.e. in the edge of the verbal phrase (marked “t’wh”), as indicated by binding relations.15 (4)

a. [Which of the papers that hei gave Maryk] did every studenti [v*P t’wh ask herk to read twh carefully] b. [At which of the parties that hei invited Maryk to] was every mani [vP t’wh introduced to herk twh]

Diagnostics like this suggest that the edge of passive vP serves as an intermediate landing site for movement, indicating that it is actually phasal (see Legate’s paper for discussion).16 In addition, Sauerland (2003) has provided compelling evidence that A-movement, too, can reconstruct in SPEC-v: (5)

[Every child]i doesn’t [vP ti seem to hisi father to be smart]

Assuming that the quantified phrase reconstructs below negation, this yields the “not every” reading of the sentence. It is thus still an open question whether Chomsky’s v*/v distinction in terms of phase heads is adequate; at least with regard to edge effects, v* and v appear to be on a par.17 Moreover, stated as in (1), PIC faces empirical challenges. Notice that according to PIC1, T in particular is unable to probe into VP. However, this incorrectly rules out instances of Agree(T, DP), as in English existential constructions or Icelandic DAT-NOM structures, where nominative is valued on DP by Agree(T, DP), with DP in situ (= in VP). In particular, in Icelandic a finite verb or auxiliary outside v*P can agree with a nominative DP inside VP across a transitive (hence, phasal) v*.18

15

16 17

18

phases. It remains unclear what the distinction, stipulated without further argument, actually expresses, since there seems to be no relevant difference between weak phases and non-phasal elements. It was consequently abandoned by Chomsky in class lectures (MIT, summer 2005) as well as in OP, AUB. As in the previously discussed cases, reconstruction in a lower position would presumably lead to a Principle C violation. But see Den Dikken (2006) for a thorough evaluation of Legate’s arguments. For some inconclusive evidence concerning the DP edge, see Matushansky (2005) and Heck & Zimmermann (2004). I’ll set the issue aside in what follows. The example is from Jonsson (1996: 157). Similar instances of T probing into VP and valuing nominative Case on DP in situ can be found in German, as discussed by Wurmbrand (2006). DP may not actually be in situ, strictly speaking:

The conceptual necessity of phases

(6)

265

Joni virdhast hafa likadh thessir sokkar Jon.DATi seem.PL [TP have [v*P ti like [VP these.NOM socks.NOM]]] ‘Jon seems to have liked these socks’

T apparently probes across the phase boundary, hence v*P is apparently not impenetrable in the sense of PIC1.19 There is also (less clear) evidence for agreement across the CP-phase boundary. In Tsez, some verbs that take clausal complements agree optionally with an argument embedded in the complement clause (see Polinsky & Potsdam 2001: 606). Assuming, as seems plausible, that these clausal complements are CPs, the data provide some further evidence against CP being impenetrable to agreement: (7)

enir uza magalu bac’ruli biyxo mother.DAT [CP boy.ERG bread.III.ABS III.eat] III.know ‘The mother knows the boy ate the bread’

In DbP, the PIC was revised accordingly: a phase domain is transferred only when the next higher phase head is merged. That is, VP remains within T’s search space, but is transferred upon merger of C. (8)

Phase Impenetrability Condition, version 2 (PIC2) For a phase HP with phase head H, the domain/complement of H is not accessible to operations at next-higher phase ZP; only H and its edge are accessible to such operations, the edge being the residue outside of H’ (specifiers of H or adjuncts to HP). (Cf. DbP: 13)

Under this conception, H and its complements are accessible until the nexthigher phase head is merged (but no longer). Consider again the example of T’s search space under the revised PIC: (9)

19

T [v*P … v* … [VP … V … ] ]  C [TP T [v*P … v* … [VP … V … ] ] ]

if feature inheritance (see below) extends to the v*-V complex, then v* (via V) may raise the object to SPEC-V. But even in this case, it would still be within the lower phase, hence inaccessible to T. However, Svenonius (2004) notes that Icelandic long-distance agreement is limited to Number agreement, whereas more local instances of Agree always involve Person.

266 Dennis Ott Other than under PIC1, the domain of the lower phase is fully accessible to T (in case no higher element intervenes), for it is only transferred when the next-higher phase head (C, in this case) is merged. Thus under PIC1, VP is accessible to v*, but not to any higher elements; under PIC2, it remains accessible to T, while C triggers its Transfer. In the light of problematic cases like those discussed above, this formulation of Phase Impenetrability appears to be empirically preferable over the earlier definition.

2.4. Feature Inheritance Chomsky (OP, AUB) argues that the relation between the phase heads C and v and their non-phasal complements is characterized by a mechanism of feature inheritance: T and V receive the uninterpretable features of C and v, respectively, at the point when the latter are merged.20 In this way, Chomsky attempts to capture the rather traditional observation that both T and V appear to be more complete (perhaps, phi-complete) when selected by C and v*, respectively, but defective in other contexts (in particular, ECM); see Fortuny (2006: section 3.1) for discussion. It follows that only phase heads are probes in a direct sense; if T or V probe, they do so only indirectly by virtue of the features inherited from the phase head. For example, the raising of a goal XP to SPEC-T is the result of C’s features probing down the tree, but these features appear on T (not C), after inheritance. The goal’s unvalued features are valued after raising to SPEC-T, hence movement stops there (in accord with the Activity Condition, which says that elements with all features valued are inert for further operations21) and does not proceed further to SPEC-C. T on its own is unable to probe; hence, structures that lack C (ECM/raising) also lack T-agreement effects (nominative Case on DP). If C is present, its EF and agreement features (the latter realized on T) probe simultaneously. Chomsky’s account raises the question whether feature inheritance is an ad-hoc solution or a conceptually necessary design feature. That is, why can the features on phase head P not remain on P, but instead must be passed on to the complement head H? 22 20

21

22

In AUB, Chomsky extends the reasoning to definite DP (n*P in his terminology), effectively declaring it a phase. See BEA, OP; also Nevins (2005) for an opposing view on activity as a precondition for movement. As Richards (2007) points out, feature inheritance cannot be a C-I requirement to encode the A/A’ distinction properly. (This is the rationale offered by Chomsky

The conceptual necessity of phases

267

Feature inheritance solves a problem concerning the valuation of uninterpretable features and the timing of Transfer, pointed out by Epstein & Seely (2002b). After valuation, there is no way for the system to tell which features had originally entered the derivation unvalued: once valued, all features ‘look the same,’ namely valued. This means that there is no way for the system to tease apart interpretable and uninterpretable features, hence feed the respective interface with those and only those features relevant to it (Full Interpretation; Chomsky 1986). The only way out seems to be back-tracking (reconstruction of the derivation), but this is clearly undesirable, as it dramatically adds to computational complexity. The solution proposed by Chomsky (OP, AUB) and discussed in detail by Richards (2007) is that valuation of unvalued features and Transfer coincide: features are not valued before Transfer, but at Transfer (that is, at the phase level). But this requires that the phase-head complements inside the phase domain not value any features prior to the phase level (viz., merger of the phase head). Hence, if valuation applies at phase level, it must be the case that all instances of valuation are triggered by the phase heads. But if the unvalued features are on the non-phasal complement head H from the start, it would be implausible to claim that H delays probing until the phase head comes in (as we were led to claim if valuation and Transfer were simultaneous). It is, far more natural to assume that the features are not there before the phase head comes in: only if a non-phasal head mediates between probe and goal can valuation occur at the point of Transfer. From this point of view, the phase head’s “passing on” its features indeed follows from good-design considerations: if valuation and Transfer must be one and the same operation for reasons of Full Interpretation, then SMT requires unvalued features to exist in phase heads only and be inherited by the complement head, which in turn must exist as a “receptacle” that receives these features in order to avoid timing problems.23

23

in OP: EF of P yields IM to A’ (= SPEC-P); EF of H yields IM or EM to A (= SPEC-H).) Assume that instead of T, C is merged to vP. C has, in this model, all the features familiar from T, and hence the A/A’ distinction could simply be encoded via first and second merge to C (multiple SPECs, in traditional terminology), without feature inheritance. Moreover, as Richards notes, the rationale only pertains to C-T, not to the v*-V complex. But, clearly, any attempt to trace feature inheritance to SMT must not discriminate between the different phasal complexes. The above considerations do not only provide a rationale for certain elements in terms of SMT, but also for a particular arrangement of these elements that has

268 Dennis Ott What this means in more concrete terms is that when T inherits C’s phifeatures, the subject in SPEC-v* is probed by both EF on C and phi-features on T. Agree(T,XP) assigns Case to XP and triggers raising to SPEC-T. To avoid countercyclicity, Chomsky proposes that all these operations apply simultaneously when the C-T complex is merged. Similarly, V inherits phifeatures of v*, raising the object XP to SPEC-V (V-to-v* raising restores word order). And some similar mechanism should exist in the nominal domain as well. Notice that the feature-inheritance variety of phase theory has an immediate empirical consequence if we assume it to hold for narrow syntax: it restores the original version of PIC (1) and is incompatible with the revised version (8). Hence, it does not straightforwardly capture those cases discussed above that led to the revision, on the assumption that it constrains agreement. If feature inheritance takes place in order to ensure valuation at Transfer, then Transfer of the complement must apply when the phase head is merged, and cannot be delayed until the next-higher phase (contrary to PIC2). Since feature inheritance seems to be supported on independent grounds, this invites the conclusion that PIC holds only for the mapping components, not for narrow syntax, considerably reducing computational load. This entails that Agree (as a narrow-syntactic operation) is not constrained by phase boundaries but only by intervention (cf. OP: 9), the data discussed in section 2.3 being instances of nonintervention. (See also Bošković 2007 for a similar conclusion.)

3. A Note on “I-functional Explanation” It has been pointed out (see Boeckx & Grohmann 2007 and references therein) that phase theory suffers from serious conceptual flaws. In particular, it is unclear why there is a discrepancy between the phases and the doso far simply been taken for granted. In the tree, each phase head P must dominate a receptacle head H: P … H … P … H … The basic structure of the phase (phase head, non-phase head) falls out immediately. This kind of structure is therefore a natural product of SMT: if we are on the right track so far, this is the clausal skeleton by the “perfect” nature of the system. Richards (2007) notes that this line of reasoning “might constrain the possible expansions of the core functional sequence into more richly articulated hierarchies” (Rizzi, Cinque). I will not dwell on the issue here; see Boeckx (to appear), Hinzen & Uriagereka (2006), Fortuny (2006) for some related proposals.

The conceptual necessity of phases

269

mains of Transfer, the phase-head complements. If CP, v*P, and DP are propositional objects, then it seems implausible to assume that what actually reaches the interface is not the phase, but only part of it (TP, VP, and NP, respectively – but set DP aside for now). Epstein (2007b) notes that “VP and TP are claimed to have NONE of the interface properties (…) that are alleged to make vP and CP natural candidates for phasehood.” Prima facie, Transfer of CP and v*P appears to be the optimal way, given their interface properties. But if CP and v*P were transferred, CP would not exist, because no derivation could go on beyond the level of the v*P phase. Epstein thus provides an I-functionalist rationale for the phase edge, echoed by Richards (2006: 9): “The disparity between CP/vP and TP/VP (…) follows as a way to ensure that Full Interpretation can be met – only a subpart of a phase can possibly be spelled out if language is to conform to SMT (…).” He goes on to argue that “no integrated, compositional, or convergent structure could be formed without some notion of phase edge (…), so that the latter is a requirement of good (let alone optimal) design.”24 Epstein (2007a,b) explicitly argues for the general validity of “I-functionalist” or “inter-organ explanation” in this sense: without property X, the system would not work, would not make use of all of its expressive potential, or would generate only deviant expressions; hence, X is a necessary design feature. This line of reasoning, however, is dubious with regard to explanatory adequacy. As Hinzen (2006: 213) points out, I-functional proposals have a “‘teleological’ flavor” to them: if property X is taken to exist because otherwise derivations would “crash”, one can ask: “why should it not crash? And how can the derivation, prior to reaching the putative interface, know what demands will be made there, and arrange itself accordingly?” 25 24

25

Similarly, Chomsky reasons: “applied to a phase PH, [Spell-Out] must be able to spell-out PH in full, or root clauses would never be spelled out [note omitted]. But we know that S-O cannot be required to spell-out PH in full, or displacement would never be possible” (BEA: 108). Hinzen illustrates the point with the case of dislocation. Prior to its reduction to Merge, externalist explanations for movement were sought to explain the system’s property of displacement on principled grounds: ‘movement is made for requirement X of an external system.’ This, indeed, has an obvious teleological flavor to it. Chomsky’s more recent explanation that displacement is merely a consequence of unbounded Merge, in contrast, is fully internalist, relying on ‘design’ factors internal to the narrow Language Faculty only.

270 Dennis Ott I agree with Hinzen: I-functional explanations of syntactic mechanisms are as empty as the engineering solutions that characterized a good deal of earlier P&P theories. This is true in particular for explanations that refer (explicitly or implicitly) to “richness of expression”; it is the system’s expressiveness itself that has to be explained.

4. Conclusions Let me briefly summarize the main points of this paper: 1. Free generation (unconstrained Merge) is empirically preferrable over Last Resort conditions on Merge. Abandoning the inadequate ±grammatical distinction, syntax cannot be “crash-proof” as a matter of empirical fact. 2. If syntax is “unprincipled” in this sense, it must be C-I constraints that determine which structures have the properties necessary for concept construction (“propositionality”). Those structures that can interface with “thought” are the phases. 3. Plausibly, “propositionality” is (at least) threefold: information/discourse semantics, thematic structure, and referentiality. The first two properties reflect the “duality of semantics”: surface vs. base properties that enter into interpretation; “referentiality” allows intentional (world-directed) language use. 4. The corresponding syntactic structures are CP, v*P (perhaps also vP), and DP (but not NP). C specifies force and comprises full event structure. v* contains information about thematic structure (the status of defective v remains unclear). The D position appears to be related to referentiality, and to argumenthood of nominal phrases. 5. The main empirical project is the identification of C-I properties to give flesh to the notion of “propositionality”. Moreover, many empirical questions concerning the syntactic repercussions of the phasal cycle (edge effects, feature inheritance) remain. It seems questionable that PIC holds for narrow syntax. 6. I-functional explanation is not valid, for “richness of expression” is a property to be explained. It remains to be seen whether any of the assumptions sketched here will resist empirical challenges. While I have not even attempted to advance the empirical debate concerning phase theory, my main goal in this paper was

The conceptual necessity of phases

271

to show that the notion of phase (as a proper format for instructions to C-I systems) is conceptually necessary, at least if current assumptions about the architecture of the broad Faculty of Language are roughly on the right track.

Acknowledgements I am indebted to Marc Richards, Cedric Boeckx, Wolfram Hinzen, Volker Struckmeier, Noam Chomsky, and Juan Uriagereka for valuable comments. All mistakes and misinterpretations are mine.

References Adger, David 2003 Core Syntax. New York: Oxford University Press. Baker, Mark C. 1997 Thematic roles and syntactic structure. In Elements of Grammar, Liliane Haegeman (ed.), 73–137. Dordrecht: Kluwer. 2001 The Atoms of Language. New York: Basic Books. Berwick, Robert C. 1998 Language evolution and the minimalist program. In Approaches to the Evolution of Language, James R. Hurford, Michael StuddertKennedy and Chris Knight (eds.), 320–340. Cambridge: Cambridge University Press. Boeckx, Cedric 2003 Islands and Chains: Resumption as Stranding. Amsterdam: John Benjamins. 2006a Darwin’s problem – or how did language get its spine, its spots, and its niche? Paper presented at Of Minds and Language, San Sebastian, Basque Country. 2006b Linguistic Minimalism. New York: Oxford University Press. 2007 Understanding Minimalist Syntax: Lessons From Locality. Oxford: Blackwell. 2008 Bare Syntax. New York: Oxford University Press. Boeckx, Cedric and Kleanthes K. Grohmann 2007 Remark: Putting Phases in Perspective. Syntax 10 (2): 204–222. Bošković, Željko 2007 Agree, phases, and intervention effects. Linguistic Analysis 33: 54–96.

272 Dennis Ott Cherniak, Christopher 2005 Innateness and brain-wiring optimization: Non-genomic nativism. In Evolution, Rationality and Cognition, A. Zilhao (ed.), 103–112. London: Routledge. Chomsky, Noam 1955 The Logical Structure of Linguistic Theory. Mimeographed, Harvard University/ MIT. 1977 Essays on Form and Interpretation. New York: North-Holland. 1986 Knowledge of Language. New York: Praeger. 1995 The Minimalist Program. Cambridge, MA: MIT Press. 2000 Minimalist inquiries. In Step by Step: Essays on Minimalist Syntax in Honor of Howard Lasnik, Roger Martin, David Michales and Juan Uriagereka, 89–156. Cambridge, MA: MIT Press. 2001 Derivation by phase. In Ken Hale: A Life in Language, Michael J. Kenstowicz (ed.), 1–52. Cambridge, MA: MIT Press. 2002 On Nature and Language. Cambridge: Cambridge University Press. 2004 Beyond explanatory adequacy. In Structures and Beyond, Adriana Belletti (ed.), 104–131. New York: Oxford University Press. 2005a Three factors in language design. Linguistic Inquiry 36 (1): 1–22. 2005b Some simple Evo-Devo theses: How true might they be for language? Ms., MIT. 2006 Turing’s Thesis. Talk delivered at InterPhases, Nicosia, Cyprus. 2007 Approaching UG from below. In Interfaces + Recursion = Language? Chomsky’s Minimalism and the View from Syntax-Semantics, U. Sauerland and H.-M. Gärtner (eds.), 1–30. 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. Chomsky, Noam and Howard Lasnik 1993 The theory of principles and parameters. In Syntax: An International Handbook of Contemporary Research, Vol. 1, Joachim Jacobs, Arnim von Stechow, Wolfgang Sternefeld, and Theo Vennemann (eds.). Berlin /New York: de Gruyter. (Reprinted in Chomsky 1995.) Collins, Chris 2002 Eliminating labels. In Epstein and Seely (2002a), 42–64. Den Dikken, Marcel 2006 A reappraisal of vP being phasal: A reply to Legate. Ms., CUNY. Eble, Gunter J. 2003 Developmental morphospaces and evolution. In Evolutionary Dynamics, James P. Crutchfield and Peter Schuster (eds.), 35–65. New York: Oxford University Press.

The conceptual necessity of phases

273

Epstein, Samuel D. 2007a Physiological linguistics, and some implications regarding disciplinary autonomy and unification. Mind & Language 22: 44–67. 2007b On I(nternalist)-Functional Explanation in Minimalism. Linguistic Analysis 33: 20–53. Epstein, Samuel D. and T. Daniel Seely (eds.) 2002a Derivation and Explanation in the Minimalist Program. Oxford: Blackwell. Epstein, Samuel D. and T. Daniel Seely 2002b Rule applications as cycles in a level-free syntax. In Epstein and Seely (2002a), 65–89. Fortuny, Jordi 2006 The emergence of order in syntax. Doctoral Dissertation, University of Barcelona. Fox, Danny 2000 Economy and Semantic Interpretation. Cambridge, MA: MIT Press. Frampton, John and Sam Guttmann 2002 Crash-proof syntax. In Epstein and Seely (2002a), 90–105. Gould, Stephen J. 1991 Exaptation: A crucial tool for evolutionary psychology. Journal of Social Issues 47: 43–65. 2002 The Structure of Evolutionary Theory. Cambridge, MA: Harvard University Press /Belknap Press. Hauser, Marc D., Noam Chomsky, and W. Tecumseh Fitch 2002 The faculty of language: What is it, who has it, and how did it evolve? Science 298: 1569 –1579. Heck, Fabian and Malte Zimmermann 2004 Phasenentwicklung in der Entwicklungsphase: Randphänomene der DP. Paper presented at GGS, Mannheim. Hinzen, Wolfram 2006 Mind Design and Minimal Syntax. Oxford: Oxford University Press. Hinzen, Wolfram and Juan Uriagereka 2006 On the metaphysics of linguistics. Erkenntnis 65 (1): 71–96. Jonsson, Jóhannes Gísli 1996 Clausal Architecture and Case in Icelandic. Doctoral Dissertation, University of Massachusetts at Amherst. Lasnik, Howard 2006 Conceptions of the Cycle. In Wh-movement Moving On, Lisa Cheng and Norbert Corver (eds.), 197–216. Cambridge, MA: MIT Press. Legate, Julie Anne 2003 Some interface properties of the phase. Linguistic Inquiry 34 (3): 506– 516.

274 Dennis Ott Longobardi, Guiseppe 1994 Reference and proper names: A theory of n-movement in syntax and logical form. Linguistic Inquiry 25 (4): 609–655. 2005 Toward a Unified grammar of reference. Zeitschrift für Sprachwissenschaft 24: 5–44. Matushansky, O. 2005 Going through a phase. In McGinnis and Richards (2005), 157–181. McGinnis, Martha and Norvin Richards (eds.) 2005 Perspectives on Phases (MIT Working Papers in Linguistics 49). MITWPL. Müller, Gereon 2003 Phrase impenetrability and wh-intervention. In Minimality Effects in Syntax, Artur Stepanov, Gisbert Fanselow and Ralf Vogel (eds.), 289–326. Berlin /New York: de Gruyter. Nevins, Andrew 2005 Derivations without the activity condition. In McGinnis and Richards (eds.), 287–310. Pesetsky, David 1982 Paths and categories. Doctoral Dissertation, MIT. Pesetsky, David and Esther Torrego 2004 Tense, case, and the nature of syntactic categories. In The Syntax of Time, Jacqueline Gueron and Jacqueline Lacarme (eds.), 495–538. Cambridge, MA: MIT Press. Pietroski, Paul M. to appear Systematicity via monadicity. Croatian Journal of Philosophy. forthc. Semantics Without Truth-values. New York: Oxford University Press. Polinsky, Maria and Eric Potsdam 2001 Long-distance agreement and topic in Tsez. Natural Language & Linguistic Theory 19: 583–646. Richards, Marc 2006 On phases, phase heads, and functional categories. Ms., University of Cambridge. 2007 On feature inheritance: An argument from the phase impenetrability Condition. Linguistic Inquiry 38 (3): 563–572. Sauerland, Uli 2003 Intermediate adjunction with A-movement. Linguistic Inquiry 34 (2): 308 –314. Saunders, Peter T. 1994 Evolution without natural selection: Further implications of the Daisyworld Parable. Journal of Theoretical Biology 166: 365–373. Svenonius, Peter 1994 C-selection as feature-checking. Studia Linguistica 48 (2): 133 –155.

The conceptual necessity of phases

275

2004

On the Edge. In Peripheries: Syntactic Edges and their Effects, David Adger, Cécile de Cat and George Tsoulas (eds.), 261–287. Dordrecht: Kluwer. Treves, Alessandro 2005 Frontal latching networks: A possible neural basis for infinite recursion. Cognitive Neuropsychology 22 (3/4): 276–291. Webster, Gerry and Brian Goodwin 1996 Form and Transformation: Generative and Relational Principles in Biology. Cambridge: Cambridge University Press. Wurmbrand, Susi 2006 Licensing case. Journal of Germanic Linguistics 18 (3): 175–235. Yang, Charles D. 2002 Knowledge and Learning in Natural Language. New York: Oxford University Press.

Contributors

Carlo Geraci

Dennis Ott

Università degli Studi di MilanoBicocca Piazza Dell’Ateneo Nuovo, 1 20126 Milano Italy

Harvard University Department of Linguistics Boylston Hall, 3rd floor Cambridge, MA 02138 USA

[email protected]

[email protected]

Kleanthes K. Grohmann

Michael T. Putnam

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

Carson-Newman College Department of Foreign Languages 1646 Russell Ave 341 Henderson Humanities Building Jefferson City, TN 37760 USA

[email protected]

[email protected]

Martin Haiden

Anjum P. Saleemi

Université François Rabelais de Tours Inserm U 930 “Brain and Imagery” 3, rue des Tanneurs 37041 Tours Cedex 1 France

GC University Higher Education Commission of Pakistan, Department of English Lahore 54000 Pakistan

[email protected]

saleemi_ [email protected]

Dalina Kallulli

Tobias Scheer

Universität Wien Insitut für Sprachwissenschaft Berggasse 11 1090 Vienna Austria

Université Nice Sophia-Antipolis Laboratoire BCL CNRS UMR 6039, MSH de Nice 98 Bd E. Herriot 06200 NICE France

[email protected]

[email protected]

278 Contributors Thomas Stroik

Takashi Toyoshima

University of Missouri – Kansas City Department of English Kansas City, MO 64110 USA

Kyushu Institute of Technology Department of Human Sciences Kawazu 860-4 Iizuka

[email protected]

Fukuoka 820-8502 Japan

Hisao Tokizaki

[email protected]

Sapporo University Department of English Nishioka 3–7 Sapporo 062-8520 Japan [email protected]

Index

accentuation (see also deaccentuation) adjunct, 10, 14, 161–163, 169–175, 242, 265 affix class-based phenomena, 28, 37 Agree, 6, 10, 119, 126–128, 134, 135–137, 152–153, 157, 162, 218–221, 224, 257, 264–270 Albanian, 117–121, 125, 128 assembly problem, 12, 97

144–156, 161–177, 181–184, 195, 205–207, 211–237, 242–245, 257– 258, 262, 267, 269 ~ by phase, 2, 23, 26–27, 35, 44 determinism, 15, 212, 215, 224, 226, 230–233 discourse, 97–98, 106, 118, 124, 129, 186–188, 194, 203, 261, 270 dynamic, 2–10, 15, 211–213, 230–232

bare prosodic structure, 11, 67, 84–89 Branch Right, 99

economy, 6–9, 15–16, 26, 45–47, 162, 174–178, 211–238, 243, 245 expletive, 16, 127, 152, 154, 197, 212– 213, 219, 222–223, 229, 237–245

causativization, 75–80 clausal complements, 12, 115–117, 265 clitic doubling, 117–120, 127 combinatorial, 181, 206–207, 211–216, 223 complexity, 9, 15–16, 167–168, 182, 211– 216, 222–224, 232, 245, 267 Condition C asymmetries, 14, 161–162, 175–176 contraction, 105 CV model (of phonological representation), 69, 71 cyclic, 6, 23–59, 69, 95, 133–134, 138, 144, 147, 151–157, 170, 174–175, 222 ~ derivation, 25, 41, 44 ~ spell-out (of words), 23–24, 27, 37– 44, 48–49, 53–58 Czech, 50 deaccentuation, 12, 115–116, 126–129 Defective Intervention Constraint (DIC), 221, 224, 246 derivation, 2–6, 9–16, 23, 25–27, 29, 32, 35, 44–47, 67–85, 90, 95–109, 115– 116, 121, 126, 133–135, 138–139,

factivity, 116–121, 124–128 ~ triggers, 116 feature, 1, 6, 10–14, 27, 37, 44, 67, 73, 78–83, 86– 90, 101–105, 116, 125– 128, 134–137, 144, 151–157, 163– 168, 172–177, 186–190, 196–202, 205, 217, 225, 228–235, 238, 242– 246, 257, 264–270 edge ~, 134–135, 198, 257 ~ inheritance, 264–270 EPP~, 127, 134–136, 151–154, 227, 231, 234–235, 238–243 f-marking, 129 generative lexicon, 189, 192, 195 generative semantics, 184, 185, 199 German, 11, 68, 75–76, 86–89, 117, 120–121, 124–125, 128, 264 givenness, 12, 115, 118–121, 128–129 global, 15–16, 33, 44, 47, 68, 211–216, 219, 227, 232, 261 Greek, Modern, 117–120, 125

280 Index head (definition), 4–5, 13, 27, 31, 36–38, 46–48, 71–72, 79–85, 88–90, 96, 102, 116, 126–127, 133–137, 140, 143– 147, 151–154, 157, 166–167, 173– 174, 197, 221, 240, 245, 258–268 Icelandic, 264–265 I-functionalism, 268 –270 I-language, 25–258, 261 illocutionary, 183, 187–195, 199–200, 202, 204–205, 207, ~ force, 15, 181, 183–184, 187–192, ~ logic, 183, ~ meaning (intentional), 14, 181, 183– 184, 188, 190, 198–199 Inclusiveness Condition (IC), 25, 67, 70– 71, 85, 258, 260 information structure, 8, 13, 54, 115–130 insertion, 16, 40, 54, 186, 197, 222 lexical ~, 58, 96 interactionism, 10–11, 23–26, 44, 128, 196 intermodular argumentation, 10 –11, 23– 27, 35–39, 56–59 intonation, 40, 48–58, 261 Italian Sign Language (LIS), 13, 133– 134, 139–149, 154–156 Late Merge, 150, 156, 161, 170, 175, 177 left-branching structure, 99, 108–109 level 1/2 rules, 29, 35 lexical (sub)array, 3–6, 13, 16, 152, 164, 186, 212, 217, 221, 245–246, 259 Linear Correspondence Axiom (LCA), 107 Linearize Down, 12, 96–101, 111 successive-cyclic linearization, 13, 138 top-down linearization, 95, 111 local (locally/locality), 4, 7, 14–16, 36, 126–127, 162, 167, 176, 211–216, 219, 224–227, 232–233, 243–245, 260–262, 265

look-ahead, 11, 15–16, 67–75, 79, 89–90, 174, 176, 211–212, 232–233, 243, 245 medial gemination, 71 Merge (definition) (see also Late Merge, Remerge), 4, 6, 9, 12–16, 28, 36, 52– 53, 71, 79–88, 95–96, 98–101, 105– 106, 109–111, 133–138, 151–154, 157, 161–178, 186, 199, 206, 212– 213, 219–239, 242–245, 256–261, 265–270 bottom-up ~, 98 External ~, 135, 152, 154, 157, 164, 206 Internal ~, 14, 40–41, 44–45, 89, 133– 136, 151–154, 157, 162–171, 176– 178, 183, 189, 199, 206, 212, 222, 224, 230, 232, 242, 245, 255–259, 269 Pair ~, 162, 171 Minimal Link Condition, 218, 225 Minimalism/Minimalist Program, 2, 5– 6, 9–15, 67, 80, 82, 90, 95, 111, 135, 162, 168, 173, 176–177, 181–182, 211, 218, 253 Minimum Feature Retention, Principle of (MFR), 16, 213, 231–238, 242–245 Model, 69, 73, 80, 83, 115, 174–176, 254–255, 261–262 bifurcated ~, 14, 181, 184, 205–206 derivational ~, 14, 75, 99, 101, 126, 177, 182, 184, 195, 205 linear ~ (‘neo-Saussurean’), 15, 181, 184, 206 T- ~, 2, 14, 25, 52, Y- ~, 2, 14, 205, 262, morpheme, 24–25, 29–32, 39–44, 50, 54–58, 126, 186 ~-specific phonologies, 28–32, 41–43 abstract ~, 116, 126 multiple computational systems in phonology, 13, 23–24, 30–32, 39–46, 51,

Index 281 68, 81, 84, 100, 111, 115, 164, 167, 172, 219, 222, 234, 259 no look-back mechanism, 4, 26–27, 33, 35, 37, 44–47, 55 non-branching constituent, 100, 107–108 non-concatenative, 11, 88 Numeration, 3, 4, 12–14, 101, 116, 126, 134–135, 152, 154, 162–164, 172, 175–177, 230 Optimality Theory (OT), 29, 51 stratal ~, 29 optimization, 14, 16, 162, 211–224, 269 overgeneration, 256, 258 parsing, 12, 105–110 vacuous ~, 108 partitive (Case), 229–230, 244 pause, 101–102, 105–110 performative deletion analysis, 184 phase, 1–16, 23–27, 31, 35–39, 44–46, 52, 55, 96–101, 105, 116, 133–134, 138–140, 144–157, 167–168, 176– 178, 181–184, 187, 190, 195–198, 203–207, 212, 221, 233–237, 246, 253, 256, 259–271 ~ edge, 10, 23, 27, 35–39, 46, 134, 139– 140, 151, 154–157, 168, 263, 269 ~ theory, 1–4, 7–9, 13, 16, 25–27, 35–37, 46, 96, 133–134, 148–150, 181–182, 233–237, 261, 268–270 ~ unit, 97–98, 101 Phase Impenetrability Condition (PIC), 4–7, 23–27, 35–38, 44–45, 55–58, 144, 151, 198, 216, 221, 224, 261– 265, 268, 270 ~ in phonology, 35, 37, 44 parameterisation of the ~, 55 phonologically empty element, 104, 106 phonology, 4, 11, 24–25, 28–32, 35–37, 40–58, 67–72, 75, 81, 84–87, 256 lexical ~, 25–32, 35–44, 55, 58

lexical vs. post-lexical ~, 42, 54–55 word- vs. sentence- ~ (Praguian segregation), 41–44, 54, 58 pitch accent, 121, 124, 129 polarity, 15, 200–202 presupposition, 12, 115–116, 120, 124 probe-goal, 6, 126–128, 151, 220, 228, 257 Projection Principle, 85, 89 pronoun, 116–118, 121, 124–129, 145, 161–163, 169–170, 172, 175, 205 clitic ~, 117 correlate ~, 120, 128 pleonastic ~, 118, 124 proposition(ality), 15–16, 36, 120, 165, 184, 187–188, 190, 193–197, 202, 212, 218, 222, 259–262, 269–270 propositional meaning, 188, 190, 218, 222 prosodic information, 126, 130 reconstruction, 14, 105–106, 161–164, 169, 172, 174, 177, 263–264, 267 recursion, 49–52, 67, 86–87, 225, 235, 255 ~ of prosodic structure in intonation, 49–52, 86 remerge, 14, 146–147, 155, 161–163, 166, 171–174, 176 representation(al), 8–9, 11, 13, 15, 40, 57, 67–71, 75, 79, 90, 96, 101–105, 110–111, 115, 169–173, 177, 181, 190, 200–201, 207, 211–214, 217– 218, 233, 235, 245, 256, 261 rewriting rules, 96 rightward movement, 143 –146, 150, 156 ~ of wh-phrases, 150 sandhi (external), 40, 44, 55, 57 cyclicity-induced ~ (absence of ~), 40, 42, 44, 59 sense extension, 196–197, 259 Shortest Derivation Condition, 211, 218– 219, 227

282 Index Shortest Movement Condition, 218–219 silent demibeat, 12, 95, 102–111 Simpl, 162, 171–177 Spanish, 129 speech act, 183–184, 187, 191–192 Spell-out, 2–13, 23–49, 53–58, 67–68, 74–75, 95–111, 116, 133, 138–139, 154–155, 176, 178, 184, 206–207, 269 ~ mechanism, 23–24, 27, 31, 36, 38– 41, 46–47, 55–58 ~ your sister, 37, 39, 56 ~-as-you-merge, 36 selective ~, 25–27, 31–32, 35–37, 56 word ~, 10, 23–24, 27, 38, 48, 54–58 ~-model, 4, 6 –7, 12, 98, 100 split-CP, 15, 194, 198–199 Strong Minimalist Thesis (SMT), 256, 266–269

Survive, 14, 162–163, 172–178, 246 syntactic bracket, 12, 100–105, 111 template, 11, 67–76, 87–90 third factor, 254–255 topichood, 118–119 Transfer, 3–4, 7, 9, 69, 126, 183, 266– 269 Tsez, 265 umlaut, 78–80, 89 unaccusativity, 79, 238–242, 245, 263 underapplication, 28–29, 32–33, 35, 37 Universal Theta Assignment Hypothesis (UTAH), 258 up-down paradox, 99 zig-zag movement, 13–14, 133–134, 139– 140, 144, 148–151, 154–157