Alternatives to Cartography 9783110217124, 9783110206036

Citation preview

Alternatives to Cartography

≥

Studies in Generative Grammar 100

Editors

Henk van Riemsdijk Jan Koster Harry van der Hulst

Mouton de Gruyter Berlin · New York

Alternatives to Cartography

Edited by

Jeroen van Craenenbroeck

Mouton de Gruyter Berlin · New York

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

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

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

Library of Congress Cataloging-in-Publication Data Alternatives to cartography / edited by Jeroen van Craenenbroeck. p. cm. ⫺ (Studies in generative grammar ; 100) Includes bibliographical references and index. ISBN 978-3-11-020603-6 (hardcover : alk. paper) 1. Grammar, Comparative and general ⫺ Word order. 2. Generative grammar. 3. Phrase structure grammar. I. Craenenbroeck, Jeroen van, 1976⫺ P295.A44 2009 415⫺dc22 2009043027

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 978-3-11-020603-6 ISSN 0167-4331 쑔 Copyright 2009 by Walter de Gruyter GmbH & Co. KG, D-10785 Berlin. All rights reserved, including those of translation into foreign languages. No part of this book may be reproduced in any form or by any means, electronic or mechanical, including photocopy, recording, or any information storage and retrieval system, without permission in writing from the publisher. Cover design: Christopher Schneider, Laufen. Printed in Germany.

Preface This volume contains papers that were presented at the second Brussels Conference on Generative Linguistics (BCGL2), held on June 25–27, 2007 in Brussels. BCGL2 was co-organized by CRISSP (Center for Research in Syntax, Semantics and Phonology) at the Hogeschool-Universiteit Brussel and the Grammatical Models Group of the University of Tilburg. I wish to thank not only the speakers, but also my fellow-organizers for making BCGL2 such an interesting and fruitful workshop: Lobke Aelbrecht, Marijke De Belder, Hans Broekhuis, Karen De Clercq, Dany Jaspers, Henk van Riemsdijk, and Guido Vanden Wyngaerd. Secondly, I wish to thank the reviewers for this volume, whose incisive comments have helped improve the quality of the papers and the thematic unity of the book: Boban Arsenijevi´c, John Bailyn, Hans den Besten, Cedric Boeckx, Guglielmo Cinque, Alexandra Cornilescu, Marcel den Dikken, Gisbert Fanselow, Jordy Fortuny, Alessandro Giorgi, Lisa Matthewson, Øystein Nilsen, Colin Phillips, Cecilia Poletto, Mark Richards, Norvin Richards, Maribel Romero, Vieri Samek-Lodovici, Kriszta Szendr˝oi, Daiko Takahashi, and Susi Wurmbrand. Finally, many thanks to Adrienn J´anosi for providing editorial assistance and to Ursula Kleinhenz and Julie Miess at Mouton for their support and their patience. Jeroen van Craenenbroeck Brussels, July 2009

Table of contents Preface

v

Alternatives to cartography: an introduction Jeroen van Craenenbroeck

1

A syntactic typology of topic, focus and contrast Ad Neeleman, Elena Titov, Hans van de Koot and Reiko Vermeulen

15

Focus, topic, and word order: A compositional view Michael Wagner

53

A focus-binding conspiracy. Left-to-right merge, scrambling and binary structure in European Portuguese Jo˜ao Costa

87

Phases and variation: Exploring the second factor of the faculty of language ´ Angel J. Gallego

109

Varieties of INFL: Tense, Location, and Person Elizabeth Ritter and Martina Wiltschko

153

CAT meets GO: Auxiliary inversion in German verb clusters Markus Bader and Tanja Schmid

203

A solution to the conceptual problem of cartography Denis Bouchard

245

Adjective placement and linearization Ion Giurgea

275

Some implications of improper movement for cartography Klaus Abels

325

There is no alternative to cartography Edwin Williams

361

Index

375

Alternatives to cartography: an introduction Jeroen van Craenenbroeck 1. A brief history of cartography In the 1980s generative grammar witnessed the birth and rise of the functional projection. Both in the nominal (Brame 1982; Abney 1983; Hellan 1986) and in the clausal domain (Stowell 1981; Safir 1982; Chomsky 1986), it was recognized that functional material is able to project syntactic structure in conformity with the X-bar-format. This insight, in combination with the principle in (1) (cf. Cinque and Rizzi 2009:2), soon led to a considerable increase in the inventory of known projections (cf. most famously Pollock 1989 on the split IP). (1)

One Feature One Head (OFOH) Each morphosyntactic feature corresponds to an independent syntactic head with a specific slot in the functional hierarchy.

A second influential factor was the assumption – essentially going back to Vergnaud (1982) – that all languages have the same set of morphosyntactic features, with cross-linguistic variation being mainly due to the overt or covert nature of the morphology spelling out those features. This led to a further proliferation of functional projections, so much so that by the mid to late 1990s each portion of clausal and phrasal structure was assumed to exhibit a richly articulated functional domain. For example, the domain above NP is argued to contain not only a rigidly ordered set of adjectival projections (Cinque 1994, 2007), but also a whole series of projections related to definiteness, number, specificity, possessive structures, and so on (see Matthewson 1998, Coene and D’hulst 2003a, 2003b and references mentioned there). Similarly, while projections such as CP, IP, VP or PP were traditionally considered to be unitary, they have now been split up into a whole string of projections (for CP see Rizzi 1997, 2001, 2004a, for IP see Cinque 1999, for VP see Ramchand 2008, and for PP see Koopman 2000, Holmberg 2002, Svenonius 2004, and the papers in Cinque and Rizzi to appear). The basic idea behind this line of reasoning is that sentence structure can be represented as a template of fixed positions, each of which can be filled by a limited set of syntactico-semantic elements. This template is taken to be a universal totally ordered set of functional projections, the specifiers of which

2

Jeroen van Craenenbroeck

serve as merger sites or as landing sites for XP-movement. Cross-linguistic word order variation is then reduced to (a) the absence or presence of such Merge and Move operations, and (b) the overt or covert nature of the heads and specifiers making up the functional sequence. Given that the general goal of this enterprise is to draw a detailed map of a particular portion of the clause, it often goes by the name of cartography (cf. in particular Rizzi 2004c, Cinque 2002, Belletti 2004, Cinque and Rizzi to appear). In recent years, however, a number of problems has been raised for this approach. I introduce some of them in the next section.

2. Challenges for cartography A first challenge targets the idea that the universal functional sequence can be described as a total linear ordering. First of all, this assumption seems to lead to word order paradoxes. For example, Bobaljik (1999) points out that while adverbs, verbal heads and floating quantifers are each strictly linearly ordered, it is not possible to combine all three groups into a single ‘macroordering’. Rather, the three groups of elements can be freely interspersed with one another, so long as their internal ‘micro-ordering’ is respected. Similarly, it is possible to detect transitivity failures in the functional sequence. Transivity is a cornerstone in cartographic reasoning, as it allows one to build a total linear sequence even if certain combinations of elements are missing for independent reasons. Thus, if A precedes B and B precedes C, then A is assumed to precede C even if the two do not actually co-occur in the language in question – or in natural language in general for that matter. As shown in (2) (Nilsen 2003:10– 11), however, transitivity is not always respected (see Van Craenenbroeck 2006 for comparable examples involving the CP-domain). (2)

a. muligens ‘possibly’ < ikke ‘not’ St˚ale har muligens spist hvetekakene sine. S. has not possibly not eaten the.wheaties his ‘Stanley possibly hasn’t eaten his wheaties.’ b. ikke ‘not’ < alltid ‘always’ St˚ale har ikke spist hvetekakene sine. S. has always not always eaten the.wheaties his ‘Stanley hadn’t always eaten his wheaties.’

Alternatives to cartography: an introduction

3

c. alltid ‘always’ < muligens ‘possibly’ Dette er et morsomt gratis spill hvor spillerne alltid this is a fun free game where the.players always mulligens er et klikk fra a˚ vine $1000! possibly are one click from to winn $1000 ‘This is a fun, free game where you’re always possibly a click away from winning $1000!’ What these types of examples suggest, is that it might not always be possible to provide a total linear ordering of functional projections. As such, these data threaten one of the basic underlying assumptions of cartography. A second problem related to the idea of a total linear ordering is the fact that it undergenerates. For example, as discussed by Neeleman and Van de Koot (2008), topics and foci in Dutch – both moved and non-moved ones – can occur in a wide variety of structural positions. There are two ways to incorporate such data into the cartographic framework. One would be to allow TopP and FocP to be freely merged anywhere along the functional spine, while the second approach would be to adopt a multitude of such projections, at regular intervals in the functional sequence. While the first approach is clearly in contradiction with one of cartography’s basic tenets, Neeleman and Van de Koot show in detail that the second one cannot account for the locality and intervention data they have uncovered (I refer to the original paper for details). A second challenge for cartography concerns its reliance on the LCA. While cartography and antisymmetry are logically independent, the latter has certain features that make it very suitable as a phrase-structural framework for the former. First of all, it offers a straightforward mechanism for converting precedence into hierarchical relations. Secondly, it does not allow for adjunction and limits the number of specifiers per projection to one. This meshes very well with the OFOH-principle formulated in (1) and the concomitant ‘criterial’ approach to movement a` la Rizzi (2004b). As shown by Abels and Neeleman (2009), however, a system which allows for more base generated orders and which does away with the LCA (and its host of word order salvaging movement operations) is no less restrictive than a strictly cartographic approach, and equally successful in deriving typological word order generalizations in the nominal domain. A third and final issue I want to raise here concerns movement. The picture painted by cartography in this respect is a very clear and simple one: each functional head is endowed with a specific morphosyntactic feature, and it attracts phrases with a matching feature to its specifier. Note that under such an approach it would be unexpected if the movement of a certain phrase X would have interpretive effects on another phrase Y not directly involved in the movement op-

4

Jeroen van Craenenbroeck

eration. As pointed out by Neeleman (1994), Lekakou (2000), Gill and Tsoulas (2004), Neeleman and Van de Koot (2008), however, such effects do occur. Summing up, while it is clear that the cartographic approach constitutes an important and very valuable development in generative grammar, in recent years a number of concerns have been raised regarding certain aspects of this program. In light of this it should not come as a surprise that a number of alternative proposals has been made. I introduce some of them in the next section.

3. Alternatives to cartography (ATC) Roughly speaking, two main lines can be discerned in the ATC-literature. The first – implicitly or explicitly – aims at reducing the number of functional heads assumed to make up the clausal or nominal spine. The second seeks alternatives for the idea that word order is the result of a universal base order with concomitant movement operations. I discuss each line in turn. Recent cartographic estimates of the number of functional projections present in natural language range up to four hundred (Cinque and Rizzi 2009:7). Add to this the assumption that all of these projections are present all of the time in all languages and it is clear why linguists have sought ways of reducing the number of FPs. One straightforward means of doing so is by giving up or weakening OFOH. In an approach that could perhaps be described as ‘cartography light’, some papers (e.g. Rizzi 1996; Thra´ınsson 1996; Giorgi and Pianesi 1997; Bobaljik and Thra´ınsson 1998) argue that languages differ as to which morphosyntactic features project their own functional projection and which ones are grouped together on a single functional head and do not project on their own. This approach keeps to the idea of a universal functional sequence, but allows parts of if to be ‘collapsed’ in certain languages or even certain constructions. Another way of reducing the number of functional heads is by rethinking the way movement works. As pointed out in the previous section, an important function of these functional heads is to drive movement operations: on the one hand they contain the relevant morphosyntactic feature that triggers the movement, while on the other they create the specifier that serves as landing site for the movement operation. There are – at least – two ways of bypassing the role played by functional heads in such a movement scenario. One is to assume that movement is triggered directly by interface conditions. In particular, syntactic movement takes place in order to facilitate the mapping onto or to serve as the input for a particular interface. This can be the interface with the phonologicalprosodic component (cf. Szendro¨ı 2001; Zubizarreta 1998; Vicente 2005), the information structural or pragmatic component (cf. Neeleman and Van de Koot

Alternatives to cartography: an introduction

5

2006; Lekakou 2000) or the semantic component (cf. Zwart 2004, Nilsen 2003). Another possible bypass operation starts from the idea that movement is triggered by the foot rather than the head of the movement chain (cf. Platzack 1996; Van Riemsdijk 1997; Nilsen 1997; Van Craenenbroeck 2006; Nash and Rouveret 1997; Koeneman 2000; Suranyi 2004; Ackema e.a. 1993). This once again eliminates the need for a wide array of functional heads – and matching specifiers – as triggers for the movement. The second main line in the ATC-literature seeks alternatives for the idea that all word order differences between languages arise from a universal base order combined with various movement operations (with the concomitant postulation of functional projections to host the moved phrases). Obviously, there are various ways of going about this. One is to assume that the flexibility in basegenerated word orders is larger than is assumed under the LCA (cf. Neeleman 1994; Neeleman and Weerman 1993; Abels and Neeleman 2009; Neeleman and Reinhart 1998). Others argue that word order is not dependent on or determined by (the hierarchical relations in) a single syntactic representation, but arises as a result of linear precedence rules active in particular word order domains (cf. Kathol 2000; Chung and Kim 2003; Jaeger 2003) or as a result of the interaction between different levels of syntactic representation (Williams 2003; Van Riemsdijk 2003). Alternatively, it has been claimed that word order is the result of the interaction between a limited number of violable output oriented constraints (cf. Costa 1998; Szendro¨ı 2001; Gutierrez-Bravo 2002; Broekhuis 2000, 2007). It is clear that theATC-accounts mentioned here by no means form a homogeneous group and that they display substantial differences in terms of empirical coverage and theoretical assumptions, but there is one thing they all have in common. In particular, they all give up the idea that there is a fixed, invariant structural position for each functional element. As such, they all diverge from the cartographic doctrine. It is against this background that the present volume should be situated.

4. This volume It is not my intention here to give a summary of the papers contained in this volume – that is what the abstracts are for. Rather, I want to situate the present book against the discussion of the previous two sections. In order to do so, I first highlight some of the challenges the papers collected here raise for cartography, and then point out some of the alternatives they propose. Finally, I give a brief outline of the entire book and indicate how it is structured.

6

4.1.

Jeroen van Craenenbroeck

Challenges for cartography

Some of the problems raised for cartography in this book show clear parallelisms with the cases discussed above, others less so. An example of the former type is Giurgea’s discussion of word order in the nominal domain. While the generalizations he points out do not constitute a word order paradox as such, they do not seem to be straightforwardly derivable under a cartographic system such as that of Cinque (2005). What Giurgea shows is that when postnominal adjecives and complements co-occur, they do so in that order (first adjectives, then complements), and while the adjectives display a left-branching order, the complements surface in a right-branching order. In a Cinquean-style analysis this would involve head movement of the noun around the arguments (to preserve their right-branching order) followed by remnant roll-up movement of the NP around the adjectives, but without taking along the arguments. As pointed out by Giurgea, arriving at the correct word order from an LCA-compliant base through movement would involve as an intermediate step a movement operation not allowed in Cinque’s system. The papers by Neeleman e.a. and Wagner show convincingly that topics and foci can surface in a wide variety of structural positions, limited only by their semantic combinatorial possibilities (Wagner) or their information-structural import (Neeleman e.a.). Both authors point out that a cartographic analysis of these facts would have to postulate either a flexible merger site for TopP and FocP or a multitude of such projections. The cartographic approach to movement is challenged in the contributions by Abels and Costa. Abels argues that the movement of a phrase X not only affects X itself but also phrases dominating and dominated by X. Crucial evidence comes from the ban on improper movement – extended to include more movement types than just A- and A-bar – which Abels shows to be active not only in cases of ‘regular’ movement, but also with remnant movement and subextraction out of moved phrases. He also explicitly points out that a phrase-structural account of these locality effects in terms of the cartographic hierarchy fails. Costa on the other hand discusses the interaction between focus placement and binding in European Portuguese (EP). Foci systematically target the right periphery of the clause in EP. This can be straightforwardly captured in a cartographic account by adopting movement to a VP-peripheral focus projection (cf. Belletti 2001), possibly followed by remnant VP-movement to ensure the focus is clause-final. What Costa shows, however, is that focus placement can be bled by binding. Specifically, while a focused direct object in a ditransitive normally obligatorily follows the indirect object, it precedes it when it binds a pronoun inside that indirect object. It is clear that this type of complex interaction between two at

Alternatives to cartography: an introduction

7

first sight unrelated phenomena is very hard to capture under a purely criteriondriven approach to focus placement. The final set of problems I want to discuss here involves OFOH. Neeleman e.a. argue that while the feature [contrast] can clearly be shown to play a syntactic role in Dutch and Russian (it drives movement), adopting a separate ContrastP – as OFOH would force us to do – is not an option. Rather, this feature seems dependent or parasitic on the features [topic] and [focus]. Ritter and Wiltschko on the other hand point out that applying OFOH to the features [tense], [location] and [person] (and hence adopting a separate T(ense)P, LocationP and PersonP) would fail to capture (a) the complementary distribution of these features across English, Halkomelem and Blackfoot, (b) the identical formal properties of these features in their respective languages, and (c) their functional equivalence. In other words, an analysis that treats each morphosyntactic feature as a separate syntactic entity misses an important cross-linguistic generalization. The list of challenges discussed here is not exhaustive, but it gives the reader an impression of the types of problems that are raised. In the next section I look at some of the alternatives that are being proposed in this volume.

4.2. Alternatives to cartography Just as in the previous section, my goal here is not to be exhaustive, but rather illustrative. That is, I want to highlight some of the main ATC-themes present in this volume. A first thing to note in this respect is that two of the papers adopt a ‘cartography light’ approach, albeit with a twist. In his discussion of the clausal left periphery, Gallego proceeds in two steps. He first considers the possibility of replacing Rizzi’s (1997) rich left periphery by a single functional projection, essentially following work by Juan Uriagereka. In the second half of the paper, however, he goes one step further and tries to derive left peripheral cartographic effects from a bare-bones Chomskyan phasal architecture, whereby the morphological richness of the phase head determines the number of phrases it can attract to its edge (and hence the cartographic richness of that edge). Ritter and Wiltschko on the other hand suggest a different way of limiting the inventory of possible functional projections. They propose that while a projection such as IP is universal, its substantive content is subject to parametric variation, constrained only by its universally determined core function. Thus, while the core function of IP is to link the eventuality to the utterance, there are various morphosyntactic means through which this can be established: English uses [tense], Halkomelem [location] and Blackfoot [person]. The upshot of this is

8

Jeroen van Craenenbroeck

that we do not need three separate functional projections, but can suffice with one ‘meta-projection’ that comes in three – or more – different flavors. A second noteworthy tendency in this volume is the fact that a large number of authors – either explicitly or implicitly – rejects the LCA in favor of a more flexible and liberal theory of base generation (see for example the papers by Giurgea, Bouchard, Bader and Schmid, and Costa). Accompanying – and constraining – this regained freedom in basic word orders are head parameters and explicit linearization rules or procedures. As pointed out by Bouchard and Bader and Schmid, this type of approach has the advantage of doing away with the many unmotivated movement triggers often found in cartographic analyses. The final approach I want to mention here is that of Wagner. He argues that certain cartographic word order restrictions follow from the semantic combinatorics of the elements involved. An interesting example of this concerns the order of contrastive topic and focus. Cross-linguistically it seems to be a fairly stable generalization that constrastive topics precede foci, or to put it in cartographic terms, that TopP dominates FocP. What Wagner shows is that a contrastive topic interpretation arises when a focus operator outscopes another focus operator. This straightforwardly explains why a contrastive topic can never be lower than a focus, without having to appeal to a universal hierarchy of functional projections. It is worth stressing that the type of approach put forward by Wagner is often assumed to be fully compatible with the basic tenets of cartography (cf. e.g. Cinque and Rizzi 2009:13–14), the idea being that the functional hierarchy is not a primitive of the grammar but can be derived from other principles, including semantics. As pointed out by Wagner, though, the compatibility is only partial. In particular, the claim that the order of functional elements can be derived from their compositional semantics does not entail that these elements yield a total linear ordering. A prime illustration of this is Nilsen’s (2003) analysis of the transitivity failure in (2) in terms of the polarity properties of the adverbs involved.

4.3.

Outline

The papers in this volume are thematically organized. The order in which they appear reflects the functional sequence, starting in the left periphery going all the way down to the nominal domain, with a special status for the final two papers. The first four papers (Neeleman e.a., Wagner, Costa, and Gallego) all deal with functional material traditionally associated with the left periphery (typically topic and focus). The IP-domain is covered by Ritter and Wiltschko, while Bader

Alternatives to cartography: an introduction

9

and Schmid focus on the verbal domain. Bouchard and Giurgea both concentrate on adjective ordering inside the nominal domain. The final two papers, Abels and Williams, approach the issue of (alternatives to) cartography from a more general perspective, focusing on its interaction with movement (Abels) or the transition between two cartographic domains (Williams). The interesting thing about these papers is that they both arrive at the conclusion – nicely reflected in the title of Williams’ paper – that the functional sequence has an existence and importance that is independent of its direct role in structure building and that cannot be straightforwardly derived from other principles or components of the grammar. It seems fitting to end a volume that presents challenges for a certain approach with a challenge for itself.

References Abels, Klaus and Ad Neeleman 2009 Universal 20 without the LCA. In: Jos´e M. Brucart, Anna Gavarr´o, and Jaume Sol`a (eds.), Merging features, 60–80. Oxford: Oxford University Press. Abney, Steven Paul 1983 The English noun phrase in its sentential aspect. Ph. D. diss., MIT. Belletti, Adriana 2001 ‘Inversion’ as focalization. In: Aafke Hulk and Jean-Yves Pollock (eds.), Inversion in Romance and the theory of Universal Grammar, 107–162. New York: Oxford University Press. Belletti, Adriana 2004 Structures and beyond. The cartography of syntactic structures, volume 3. New York: Oxford University Press. Bobaljik, Jonathan David 1999 Adverbs: the hierarchy paradox. Glot International 4:27–28. Boblajik, Jonathan David and H¨oskuldur Thra´ınsson 1998 Two heads aren’t always better than one. Syntax 1:37–71. Brame, Michael 1982 The head-selector theory of lexical specifications and the nonexistence of coarse categories. Linguistic Analysis 10:321–325. Broekhuis, Hans 2000 Against feature strength: the case of Scandinavian object shift. NLLT 18:673–721. Broekhuis, Hans 2007 Subject shift and object shift. Journal of Comparative Germanic Linguistics 10(2):109–141.

10

Jeroen van Craenenbroeck

Cinque, Guglielmo 1994 On the evidence for partial N movement in the Romance DP. In: Guglielmo Cinque, Jan Koster, Jean-Yves Pollock, Luigi Rizzi, and Rafaella Zanuttini (eds.), Paths towards Universal Grammar, 85–110. Georgetown: Georgetown University Press. Cinque, Guglielmo 1999 Adverbs and functional heads: a cross-linguistic perspective. New York: Oxford University Press. Cinque, Guglielmo 2002 Functional structure in DP and IP. The cartography of syntactic structures, volume 1. New York: Oxford University Press. Cinque, Guglielmo 2005 Deriving Greenberg’s Universal 20 and Its Exceptions. Linguistic Inquiry 36(3): 315–332. Cinque, Guglielmo 2007 The syntax of adjectives. A comparative study. Ms. University of Venice. Cinque, Guglielmo and Luigi Rizzi 2009 The cartography of syntactic structures. Ms. University of Venice and University of Siena. Cinque, Guglielmo and Luigi Rizzi to appear Mapping spatial PPs. The cartography of syntactic structures, vol.6. New York: Oxford University Press. Chomsky, Noam 1986 Barriers. Cambridge, MA: The MIT Press. Chung, Chan and Jong-Bok Kim 2003 Capturing word order asymmetries in English left peripheral constructions: a domain-based approach. In: Stefan M¨uller (ed.), Proceedings of the 10th international conference on Head-Driven Phrase Structure Grammar, 68–87. Stanford: CSLI Publications. Coene, Martine and Yves D’hulst 2003a From NP to DP. Volume 1: The syntax and semantics of noun phrases. Amsterdam: John Benjamins. Coene, Martine and Yves D’hulst 2003b From NP to DP. Volume 2: The expression of possession in noun phrases. Amsterdam: John Benjamins. Costa, Jo˜ao 1998 Word order variation: a constraint-based approach. Ph. D. diss., Leiden University. Craenenbroeck, Jeroen van 2006 Transitivity failures in the left periphery and foot-driven movement operations. In: Jeroen van de Weijer and Bettelou Los (eds.), Linguistics in the Netherlands 2006, 52–64. Amsterdam: Benjamins.

Alternatives to cartography: an introduction

11

Gill, Kook-Hee and George Tsoulas 2004 Peripheral effects without peripheral syntax: the left periphery in Korean. In: David Adger, Cecile De Cat, and George Tsoulas (eds.), Peripheries. Syntactic edges and their effects, 121–142. Amsterdam and New York: Kluwer Academic Publishers. Giorgi, Alessandra and Fabio Pianesi 1997 Tense and aspect: from semantics to morphosyntax. NewYork: Oxford University Press. Gutierrez-Bravo, Rodrigo 2002 Structural markedness and syntactic structure: a study of word order and the left periphery in Mexican Spanish. Ph. D. diss., UCSC. Hellan, Lars 1986 The headedness of NPs in Norwegian. In: Pieter Muysken and Henk van Riemsdijk (eds.), Features and projections, 89–123. Dordrecht: Foris. Holmberg, Anders 2002 Prepositions and PPs in Zina Kotoko. In: Bodil Kappel Schmidt, David Odden, and Anders Holmberg (eds.), Some Aspects of the Grammar of Zina Kotoko, 162–174. M¨unchen: Lincom Europa. Jaeger, Florian 2003 Topics First! In- and outside of Bulgarian wh-interrogatives. In: Stefan M¨uller (ed.), Proceedings of the 10th international conference on Head-Driven Phrase Structure Grammar, 181–202. Stanford: CSLI Publications. Kathol, Andreas 2000 Linear syntax. Oxford: Oxford University Press. Kayne, Richard 1994 The anti-symmetry of syntax. Cambridge, MA: The MIT Press. Koeneman, Olaf 2000 The flexible nature of verb movement. Ph. D. diss., Utrecht University. Koopman, Hilda 2000 Prepositions, postpositions, circumpositions, and particles. In: Hilda Koopman (ed.), The syntax of specifiers and heads, 204–260. Routledge, London. Lekakou, Marika 2000 Focus in modern Greek. MA-thesis, UCL. Matthewson, Lisa 1998 Determiner systems and quantificational strategies: evidence from Salish. The Hague: HAG. Neeleman, Ad 1994 Complex predicates. Ph. D. diss., Utrecht University.

12

Jeroen van Craenenbroeck

Neeleman, Ad and Hans Van de Koot 2008 The nature of discourse templates. Journal of Comparative Germanic Linguistics 11:137–189. Neeleman, Ad and Tanya Reinhart 1998 Scrambling and the PF Interface. In: Miriam Butt and Wilhelm Geuder (eds.). The Projection of Arguments, 309–353. Stanford: CSLI Publishers. Neeleman, Ad and Fred Weerman 1993 Flexible syntax. A theory of case and arguments. Dodrecht: Kluwer. Nilsen, Øystein 1997 Adverbs and A-shift. Working Papers in Scandinavian syntax 59:1–31. Nilsen, Øystein 2003 Eliminating positions: syntax and semantics of sentence modification. Ph. D. diss., Utrecht University. Platzack, Christer 1996 Germanic Verb Second Languages – Attract vs. Repel: On optionality, A-bar movement and the symmetrical/asymmetrical Verb Second hypothesis. In: Edwald Lang and Gisela Zifonun (eds.), Deutsch – typologisch, 92–120. Berlin: Walter de Gruyter. Poletto, Cecilia 2002 The left-periphery of V2-Rhaetoromance dialects: a new view on V2 and V3. In: Sjef Barbiers, Leonie Cornips and Susanne van der Kleij (eds.). Syntactic Microvariation. Published at http://www. meertens. knaw. nl/books/synmic/. Pollock, Jean-Yves 1989 Verb movement, Universal Grammar, and the structure of IP. Linguistic Inquiry 20:365–424. Ramchand, Gillian 2008 First phase syntax. Ms. University of Tromsø. Riemsdijk, Henk van 1997 Push chains and drag chains: complex predicate split in Dutch. In: Tonoike Shigeo (ed.), Scrambling, 7–33. Tokyo: Kurosio Publishers. Riemsdijk, Henk van 2003 Contour templates in syntax? A note on the spreading of (in)definiteness. In: Takeru Honma, Toshiyuki Tabata and Shin-ichi Tanaka (eds.), A new century of phonology and phonological theory: a festschrift for Professor Shosuke Haraguchi on the occasion of his sixtieth birthday, 559–570. Tokyo: Kaitakusha. Rizzi, Luigi 1996 Residual Verb Second and the Wh-criterion. In: Adriana Belletti and Luigi Rizzi (eds.), Parameters and functional heads, 63–90. NewYork: Oxford University Press.

Alternatives to cartography: an introduction Rizzi, Luigi 1997 Rizzi, Luigi 2001

Rizzi, Luigi 2004a

Rizzi, Luigi 2004b

13

The fine structure of the left periphery. In: Liliane Haegeman (ed.), Elements of grammar, 281–337. Dordrecht: Kluwer. On the position Int(errogative) in the left periphery of the clause. In: Guglielmo Cinque and Giampaolo Salvi (eds.), Current studies in Italian syntax. Essays offered to Lorenzo Renzi, 287–296. Amsterdam and New York: Elsevier. Locality and left periphery. In: Adriana Belletti (ed.), Structures and beyond. The cartography of syntactic structures, vol. 3, 223–251. New York: Oxford University Press. On the form of chains: criterial positions and ECP effects. Ms. University of Siena.

Rizzi, Luigi (ed.) 2004c The structure of CP and IP. The cartography of syntactic structures, volume 2. New York: Oxford University Press. Safir, Ken 1982 Inflection-government and inversion. TLR 1:417–467. Stowell, Tim 1981 Origins of phrase structure. Ph. D. diss., MIT. Svenonius, Peter 2004 Spatial P in English. http://ling.auf.net/lingBuzz/000001. Szendro¨ı, Kriszta 2001 Focus and the syntax-phonology interface. Ph. D. diss., UCL. Sur´anyi, Bal´azs 2004 The left periphery and cyclic spell-out: the case of Hungarian. In: David Adger, Cecile De Cat, and George Tsoulas (eds.), Peripheries. Syntactic edges and their effects, 49–73. Amsterdam and New York: Kluwer Academic Publishers. Thra´ınsson, H¨oskuldur 1996 On the (non-)universality of functional categories. In: Werner Abraham, Samuel David Epstein, H¨oskuldur Thrainsson, and Jan-Wouter Zwart (eds.), Minimal ideas: syntactic studies in the Minimalist framework, 252–281. Amsterdam: John Benjamins. Vergnaud, Jean-Roger 1982 D´ependances et niveaux de repr´esentations en syntaxe. Ph. D. diss., University of Paris VII. Vicente, Luis 2005 Word order variation in Basque as non-feature-driven movement. Ms. University of Leiden.

14

Jeroen van Craenenbroeck

Williams, Edwin 2003 Representation theory. Cambridge, MA: The MIT Press. Zubizarreta, Maria-Luisa 1998 Prosody, focus and word order. Cambridge, MA: The MIT Press. Zwart, Jan-Wouter 2004 Een dynamische structuur van de Nederlandse zin. TABU 33:55–71, 151–172.

A syntactic typology of topic, focus and contrast∗ Ad Neeleman, Elena Titov, Hans van de Koot and Reiko Vermeulen Abstract. In this paper we argue for a typology of various information-structural functions in terms of three privative features: [topic], [focus] and [contrast] (see also Vallduv´ı and Vilkuna 1998, Moln´ar 2002, McCoy 2003, and Giusti 2006). Aboutness topics and contrastive topics share the feature [topic], new-information foci and contrastive foci share the feature [focus], and contrastive topics and contrastive foci share the feature [contrast]. This typology is supported by data from Dutch (where only contrastive elements may undergo A’-scrambling), Japanese (where aboutness topics and contrastive topics must appear sentence-initially), and Russian (where the new-information foci and contrastive foci share the same underlying position). To the best of our knowledge, there are no generalizations over information-structural functions that do not share one of the features adopted here.

1. Introduction As is well-known, topics and foci have dedicated positions in a variety of languages. This paper is concerned with what this fact can tell us about the typology of information-structural notions and their mapping to the syntax. We argue that the data support two conclusions, both of which can be shown to clash with a cartographic outlook on sentence structure (for a general overview of the cartographic framework, see Cinque 2002, Rizzi 2003, and Belletti 2004). The first is that there are no fixed landing sites for topic and focus movement. The second is that there are cross-cutting generalizations over topics, over foci, and over contrastive elements. These jointly motivate the following four-way typology: (1) Non-contrastive Contrastive

Topic aboutness topic [topic] contrastive topic [topic, contrast]

Focus new information focus [focus] contrastive focus [focus, contrast]

What the table in (1) expresses is that topic and focus are basic notions in information structure that can be enriched to yield a contrastive interpretation.

16

Ad Neeleman, Elena Titov, Hans van de Koot and Reiko Vermeulen

In other words, a contrastive topic and a contrastive focus are an aboutness topic and a new information focus interpreted contrastively. We are not the first to make a suggestion along these lines; related ideas can be found in Vallduv´ı and Vilkuna (1998), Moln´ar (2002), McCoy (2003), and Giusti (2006). Cartography is based on two core tenets. First, it assumes a one-to-one correspondence between syntactic position and interpretive effect. Topic and focus will hence be realized in designated functional projections, which according to Rizzi (1997) are located in the left periphery of the clause. Second, cartography tacitly adopts a radically decompositional approach to syntactic categories. This second assumption is in fact implied by the first: if there are two interpretive effects, there must be two designated positions in which these are licensed. Consequently, every semantic feature that has syntactic import must head its own functional projection. This drives the cartographic research program to the highly articulated phrase structure found in much recent work. Our claim that topic and focus do not have designated positions in the clause obviously runs counter to the first tenet of cartography. The cross-cutting generalizations that we establish have the consequence that the presence of [contrast] is conditional upon the presence of either [topic] or [focus]. This conclusion does not mesh with the second tenet of cartography, that of radical syntactic decomposition. Specifically, an encoding of these features as heads of functional projections TopicP, FocusP and ContrastP cannot capture the data provided in the following sections. We will develop both arguments in more detail in sections 3 and 6. In order to make the comparison with cartography as explicit as possible, we will treat [topic], [focus] and [contrast] as privative syntactic features. This is a matter of presentational convenience; in fact, none of our arguments are adversely affected if [topic], [focus] and [contrast] are merely discourse notions targeted by mapping rules operating between syntax and information structure. The strongest evidence for the typology in (1) comes from the syntactic distribution of constituents with different information-structural functions. If we treat contrast, topic and focus as privative features, we expect to find rules that mention [topic] and therefore generalize over aboutness topics and contrastive topics, rules that mention [focus] and therefore generalize over new information focus and contrastive focus, and rules that mention [contrast] and therefore generalize over contrastive topic and contrastive focus. We do not expect to find rules that generalize over aboutness topics and new information foci, over contrastive topics and new information foci, or over aboutness topics and contrastive foci. None of these pairs share a feature. The import of these predictions of course depends on what we mean by notions like ‘focus’, ‘topic’ and ‘contrast’. We take ‘focus’ to be the information

A syntactic typology of topic, focus and contrast

17

highlighted in a proposition. For example, in the answer to a wh-question, the constituent that corresponds to the wh-expression is a focus. The rest of the sentence functions as the background to this focus; that is, the focus is highlighted with respect to this material (throughout we use small capitals to mark foci): (2)

a. b.

What did Rutger buy? Rutger bought A GUN.

We follow Reinhart (1981) in assuming that topics should be defined in terms of aboutness (the linguistic relevance of aboutness is motivated by several phenomena, including anaphora resolution). On Reinhart’s definition, a topic is the entity that the utterance is about. Thus, while ‘focus’ is notion operative at the level of propositions, ‘topic’ is primarily a discourse notion (see Tomioka 2009 for recent discussion). But of course some syntactic constituents are used to manipulate the topic of discourse (by introducing a new topic or narrowing down the current topic, and so on). These are often referred to as sentence topics or linguistic topics. We reserve the feature [topic] to distinguish them from constituents with other information-structural functions. Linguistic topics should be distinguished from expressions contained in the utterance that merely index the current topic of discourse (see Lambrecht 1994 for extensive discussion)1 . We can illustrate the distinction using the following discourse (throughout we use double underlining to mark topics): (3)

a. b.

Maxine was introduced to the queen on her birthday. She was wearing a special dress for the occasion.

In (3a), Maxine is a linguistic topic: it introduces a new topic of discourse. The initial comment made about Maxine is that she was introduced to the queen on her birthday. The pronoun her in this comment is not a linguistic topic, but a category that indexes the topic. We take the same to be true of the continuation of the discourse in (3b), which is what one might call an ‘all-comment’ sentence, linked to the topic Maxine through the pronoun she (see Vallduv´ı 1992, Lambrecht 1994, and Vallduv´ı and Engdahl 1996 for discussion).2 A consequence of this view is that the traditional notion of aboutness is a reliable indicator of discourse topics, but not of linguistic topics (i.e. constituents that bear the [topic] feature). In what follows, we will introduce tests for linguistic topichood where appropriate. The final notion we rely on is that of contrast. Constituents that are contrastive are understood to belong to a contextually given set out of which they are selected to the exclusion of at least some other members of the set. Both topics and foci

18

Ad Neeleman, Elena Titov, Hans van de Koot and Reiko Vermeulen

can be interpreted contrastively. In English, contrastive topics and foci are each marked by a special intonation. Contrastive foci typically carry what Jackendoff (1972) calls an A-accent: a plain high tone (H*), often followed by a default low tone (see B¨uring 2003 and references mentioned there). Contrastive topics carry a B-accent, maximally realized as L+H* followed by a default low tone and a high boundary tone (L H%). We will not indicate A- and B-accents in examples, but all constituents marked as contrastive topic or focus are taken to carry these accents in languages that distinguish them (like Dutch). Further tests that identify contrastive topics and contrastive foci will be introduced as we proceed. (Throughout we use boldface to mark contrast.) (4)

a. b.

Rutger bought A GUN. Maxine was introduced to the queen on her birthday.

On the appropriate intonational contours, (4a) conveys that Rutger bought a gun and not certain other relevant items, while (4b) highlights that the speaker knows that Maxine was introduced to the queen on her birthday, but could not make the same statement about other relevant individuals. (This could be because the speaker lacks knowledge about these other individuals or because he or she knows that a similar statement about them would not be true.) On these definitions of topic, focus and contrast, the kind of rules excluded by the table in (1) are indeed not attested. However, the kind of generalizations permitted by the table are indeed attested. In each of the following sections we discuss a language with a data pattern that can be captured by a rule that refers to [topic], [focus] or [contrast]. In section 2, we show that Dutch A’scrambling targets constituents with a contrastive interpretation. In section 4, we show that in Japanese both aboutness topics and contrastive topics must appear in the left periphery of the clause, a requirement that sets them apart from other constituents. In section 5, we show that in Russian new information foci and contrastive foci have the same underlying position. The presence or absence of contrastiveness, however, gives rise to very different surface syntax, as contrastive foci are fronted. There is one generalization whose effects can be found in all three languages: (5)

[Contrast] licenses A’-movement.

It is obvious that (5) holds in Dutch. In Japanese, contrastive topics can be shown to be associated with an A’-trace, while aboutness topics are associated with a null resumptive pronoun (see section 4). In Russian, contrastive foci undergo A’-fronting, while new information foci remain in clause-final position (see section 5).

A syntactic typology of topic, focus and contrast

19

The generalization in (5) is of course not new. It goes back to at least Kiss (1998), where evidence for it is provided from languages other the ones discussed here. An obvious question is why the generalization should hold. Kiss suggests that contrast is inherently quantificational, and licenses A’-movement because this operation in general is used to create structures of quantification (a similar idea can be found in Rizzi 1997). We find this an attractive suggestion, although it faces a number of complications (for example, quantification associated with contrast is different in various respects from quantifier scope and scope in wh-questions; compare Zubizarreta 1998). For reasons of space, we cannot address any of these complications here and must leave (5) as a descriptive generalization.

2. Dutch A’-scrambling3 There is general agreement that, in Germanic and beyond, there are two types of scrambling. A-scrambling feeds and bleeds binding and secondary predication, does not give rise to weak crossover effects, is clause-bounded, and does not give rise to scope-reconstruction. We cannot illustrate all these properties here, but for relevant discussion, see Vanden Wyngaerd (1989), Mahajan (1990), Zwart (1993), Neeleman (1994), and Neeleman and Van de Koot (2007). In contrast, A’-scrambling does not affect binding or secondary predication, gives rise to weak crossover effects, is not clause-bounded, and reconstructs (obligatorily) for scope. Again, we will not demonstrate these properties here, but refer the reader to Neeleman (1994), Jacobs (1997), Haider and Rosengren (1998), and Neeleman and Van de Koot (2007) for discussion. In Dutch, the two types of scrambling can be easily told apart, because in this language only A’-scrambling can alter the basic order of arguments (subject – indirect object – direct object).4 A-scrambling is restricted to the reordering of arguments and adjuncts (see Zwart 1993 and references cited there).5 The two types of scrambling are also associated with different interpretive effects. Ascrambling operations typically mark the scrambled DP as discourse-anaphoric (that is, the DP refers back to an entity introduced earlier in the discourse); see Reinhart (1995), Neeleman and Reinhart (1998), and Choi (1999), among others, for discussion. Abstracting away from pied-piping, A’-fronting operations typically require the moved DP to be interpreted as either a contrastive focus or a contrastive topic (see Neeleman 1994 and Frey 2001):

20

(6)

Ad Neeleman, Elena Titov, Hans van de Koot and Reiko Vermeulen

a.

b.

(7)

a.

b.

(8)

a.

b.

Ik geloof dat [alleen DIT boek]1 Jan Marie t 1 gegeven I believe that only this book John Mary given heeft. has ‘I believe that John has given only this book to Mary.’ Ik geloof dat [zo’n boek]1 alleen JAN Marie t 1 gegeven I believe that such-a book only John Mary given heeft. has ‘I believe that only John has given such a book to Mary.’ Ik geloof dat Jan [alleen DIT boek]1 Marie t 1 gegeven I believe that John only this book Mary given heeft. has ‘I believe that John has given Mary only this book’ Ik geloof dat Jan [zo’n boek]1 alleen MARIE t 1 gegeven I believe that John such-a book only Mary given heeft. has ‘I believe that John has given such a book only to Mary.’ [Alleen DIT boek]1 zou Jan Marie t 1 only this book would John Mary ‘John would give Mary only this book.’ alleen JAN Marie t 1 [Zo’n boek]1 zou such-a book would only John Mary ‘Only John would give Mary such a book.’

geven. give geven. give

The data demonstrate that A’-scrambling can target a variety of positions. Irrespective of whether the moving phrase is a topic or a focus, it can land in a position between the complementizer and the subject, as in (6), in a position between the subject and the indirect object, as in (7), or in the first position in main clauses, as in (8). Further landing sites are available in structures containing adverbs, as these are (usually) freely ordered with respect to moved topics and foci. DPs that are not interpreted contrastively cannot be scrambled across arguments, even if they are topic or focus. One might conjecture that A’-scrambling is licensed by a mapping rule that assigns a moved constituent an interpretation as contrastive. The drawback of this suggestion is that constituents that remain in situ can also be interpreted in

A syntactic typology of topic, focus and contrast

21

this way, so that it is difficult to see what effect the movement could have on the displaced category. This being so, we explore the possibility that A’-scrambling does not affect the interpretation of the moved category itself, but rather that of the constituent to which it adjoins. The idea, then, is that movement of a contrastive topic or focus marks that material in the sentence used to calculate the contrast (the statement made in relation to the contrasted category to the exclusion of alternatives). We will call this material the domain of contrast (DoC). Thus, if a contrastive topic or focus remains in situ, the domain of contrast need not be a constituent, as indicated in (9a). Therefore, in interpreting the sentence, the hearer must construe an appropriate domain of contrast based on contextual clues. This is different if the contrastive constituent in (9a) moves out of YP, so that an otherwise discontinuous domain of contrast is turned into a constituent, as shown in (9b). (9)

a.

b.

In sum, movements of contrastive topics and foci do not mark the discourse functions of these elements themselves, but rather their domain of contrast; that is, the material relevant to calculating the set of alternatives on which the contrast operates (see Wagner 2005 for related ideas). This proposal can be implemented through the mapping rule in (10). (For the purposes of this paper, we assume that (10) is part of the grammar of Dutch, but not necessarily of the grammars of Japanese and Russian. Of course future research should establish whether it holds more generally. The structure in (11), to which (10) refers, contains the diacritic M that we have used in previous work to encode A’-movement, on a par with the slash notation in HPSG. Nothing hinges on this.)6 (10)

(11)

DoC Marking In (11) N2 is interpreted as the domain of contrast for XP.

22

Ad Neeleman, Elena Titov, Hans van de Koot and Reiko Vermeulen

The proposal summarized in (10) and (11) entails that the examples in (6) and (7) differ as to whether or not the embedded subject is included in the domain of contrast. This is because both are movement structures and, given that the sister of the landing site of A’-scrambling is the domain of contrast, the status of the subjects in the examples in (6) differs from that of the subjects in the examples in (7). Note that in the absence of A’-scrambling the subject could of course be included in the domain of contrast. An immediate question that this proposal raises is how domains of contrast can apparently be of different semantic types (a proposition with a single λbound variable in (6) and a constituent with a λ-bound variable and an open slot for the subject in (7)). We assume, following Schwarzschild (1999) and others, that a contrast is always based on an expression containing a single λ-bound variable. This expression is used in information structure to generate the set of alternatives from which the contrastive constituent is chosen. If so, there must be an information-structural procedure of existential closure that transforms the domain of contrast in (7) into an expression of the right type. The interpretation of the existentially bound variable is then provided by the context (it must be as specific as the context allows). Thus, the contrast in (6a) is based on the expression λx [John has given Mary x], while the contrast in (7a) is based on λx ∃y [y has given Mary x]. If y is interpreted as ‘someone’, then (6a) and (7a) differ in the set of alternatives from which the contrastive focus is selected, namely the set of things that John would give Mary versus the set of things that someone would give Mary. The proposal put forward above makes a number of predictions. The first set of predictions has to do with the interaction between DoC marking and well-formedness constraints on information structure. It is usually assumed that the first partitioning of a sentence at information structure distinguishes a topic and a comment. The comment may then be further partitioned into a focus and a background. This order of partitioning is in line with the suggestion that the topic-comment partitioning applies to an utterance, while the focus-background partitioning applies to a proposition. Consequently it is possible to embed a focus in a comment (as in (12a)), but it is not possible to embed a topic in a background (as in (12b)). (For relevant discussion, see Prince 1981, Reinhart 1981, 1995, 2006, Vallduv´ı 1992, Lambrecht 1994, and Hajiˇcov´a et al. 1998.) (12)

a. b.

topic [COMMENT FOCUS [BACKGROUND . . . ]] ∗ FOCUS [ BACKGROUND topic [COMMENT . . . ]]

Given that these partitionings are exhaustive, the domain of contrast for a focus must consist of material chosen from the background, while the domain of con-

A syntactic typology of topic, focus and contrast

23

trast for a topic must consist of material chosen from the comment. This implies that DoC marking does not only identify material on which a contrast is based, but also forces this material to be construed as belonging to the background or the comment, depending on the interpretation of the contrastive constituent. We should emphasize that (12a,b) are information structures, and not syntactic configurations. Given that the mapping between syntax and information structure need not be isomorphic, the ban on the embedding of a topic-comment structure in a background will not directly restrict syntactic structure. For instance, it does not follow from (12b) that topics cannot be preceded by foci. Any impact of information-structural constraints on word order in Dutch must result from the application of DoC marking. If no movement takes place, no material is marked as belonging to a domain of contrast (and thus indirectly to a background or a comment). Hence the syntax does not impose any restrictions on the mapping to information structure. However, we expect that the effects of (12b) will be felt if focus movement takes place. The material contained in the domain of contrast marked by movement of a focus must be part of its background. Since by (12b) a background cannot contain a topic, it is predicted that focus movement out of a constituent containing a topic will be impossible. In other words, what we expect is that the placement of in situ topics and foci is free. However, while a topic can move out of a constituent containing a focus, a focus cannot move out of a constituent containing a topic. In order to demonstrate that these predictions are borne out, we must sharpen our criteria for classifying a constituent as topic or focus. As pointed out earlier, contrastive foci are associated with an A-accent and contrastive topics with a B-accent. In addition, there are contextual criteria. It is well known that in the answer to a WH-question, the constituent that corresponds to the WH-operator is (usually) a focus. If it is interpreted contrastively, it qualifies as a contrastive focus. By this criterion, de bonen ‘the beans’ in (13a,b) is a contrastive focus. As suggested in the introduction, a contrastive topic is a constituent used to shift the topic of discourse. Such a shift takes place if the hearer answers a question about an entity different from the entity the original question was about. Therefore, Wim in (13a,b) can be classified as a contrastive topic (the original question mentioning Fred).7 What the data in (13) show, then, is that an in-situ focus may follow a topic, but cannot move across it.8,9 (13)

Hoe zit het met Fred? Wat heeft hij gegeten? ‘What about Fred? What did he eat?’ Nou, dat weet ik niet, maar ik geloof ‘Well, I don’t know, but I believe’

24

Ad Neeleman, Elena Titov, Hans van de Koot and Reiko Vermeulen

a.

dat Wim van de BONEN meer gegeten heeft dan that Bill from the beans more eaten has than vorig jaar. last year ‘that Bill has eaten more from the beans than last year.’

b. # dat [van de BONEN]1 Wim t 1 meer gegeten heeft dan that from the beans Bill more eaten has than vorig jaar. last year There is a further test that can be used to corroborate the classification of topics and foci: negative quantifiers can function as foci, but not as topics (for obvious interpretive reasons: one cannot say something about nothing). This is corroborated by the fact that they cannot appear in the English as for construction, which marks topics: (14)

#As for no boy, I like him

Therefore, if in the relevant context a constituent can be replaced by a negative quantifier, it cannot be a topic. Indeed, when Wim in (13a) is replaced by niemand ‘nobody’, the result is decidedly odd, as shown in (15a).10 (The hearer is left to wonder which person is referred to as ‘nobody’.) However, the variant of (13a) in (15b), in which de bonen ‘the beans’ has been replaced by nergens ‘nothing’, is perfectly natural. (15)

a. # dat niemand van de BONEN meer gegeten heeft dan that nobody from the beans more eaten has than vorig jaar. last year ‘that nobody has eaten more from the beans than last year.’ b. dat Wim NERGENS van meer gegeten heeft dan vorig that Bill nothing of more eaten has than last jaar year ‘that Bill has not eaten more from anything than last year.’

The data in (16) show that, by contrast, an in-situ topic may follow an in-situ focus or move across it. This observation is corroborated by the fact that in neither (16a) nor (16b) can ‘the beans’ be replaced by a negative quantifier (see (17)), whereas replacing ‘Bill’ by ‘nobody’ is unproblematic in both of these examples (see (18)). The results of this test are consistent with a classification of ‘the beans’ as topic and of ‘Bill’ as focus.

A syntactic typology of topic, focus and contrast

(16)

25

Hoe zit het met de soep? Wie heeft die gegeten? ‘What about the soup? Who ate that?’ Nou, dat weet ik niet, maar ik geloof . . . ‘Well, I don’t know, but I believe . . . ’ a.

dat WIM van de bonen meer gegeten heeft dan vorig that Bill from the beans more eaten has than last jaar. year

b.

dat [van de bonen]1 WIM t 1 meer gegeten heeft dan that from the beans Bill more eaten has than vorig jaar. last year ‘that Bill has eaten more from the beans than last year.’

(17)

a.# dat WIM nergens van meer gegeten heeft dan vorig that Bill nothing of more eaten has than last jaar. year b.# dat [ nergens van]1 WIM t 1 meer gegeten heeft dan that nothing of Bill more eaten has than vorig jaar. last year ‘that Bill has not eaten more from anything than last year.’

(18)

a.

b.

dat NIEMAND van de bonen meer gegeten heeft dan that nobody has than from the beans more eaten vorig jaar last year dat [van de bonen]1 NIEMAND t 1 meer gegeten heeft that from the beans nobody more eaten has dan vorig jaar than last year ‘that nobody has eaten more from the beans than last year.’

The data considered so far could be captured by a linear constraint. However, as pointed out by an anonymous reviewer, the proposal based on the mapping rule in (10) makes a stronger prediction, namely that focus movement out of a constituent containing a topic will give rise to ungrammaticality whether the launching site of the relevant movement precedes or follows the topic. As we will now show, this stronger prediction is correct.

26

Ad Neeleman, Elena Titov, Hans van de Koot and Reiko Vermeulen

The answers in (19) involve a ditransitive verb. The context is set up in such a way as to favour a reading of the indirect object as focus (it answers the whquestion), and the direct object as topic (it switches the topic of the discourse from the antique sideboard to the clock). While the answer in (19a) is felicitous, the answer in (19b), where the focus has moved, cannot be produced with the intonation indicated. (19)

Hoe zit het met het dressoir? Wie heeft grootvader dat nagelaten? ‘How about the sideboard? To whom has granddad bequeathed that?’ Nou, dat weet ik niet, maar ik geloof ‘Well, I don’t know, but I believe . . . ’ a.

dat grootpapa zijn BUREN de klok heeft willen that granddad his neighbours the clock has want nalaten. bequeath

b. # dat [zijn BUREN]1 grootpapa t 1 de klok heeft willen that his neighbours granddad the klok has want nalaten. bequeath ‘that granddad wanted to bequeath the clock to his neighbours.’ As expected, it is possible to move a topic from a position preceding a focus, as shown in (20). Our classification of ‘neighbours’ as topic and ‘clock’ as focus in the examples in (20) receives support from the negative-quantifier test introduced above, but for reasons of space we will not demonstrate this here. (20)

Hoe zit het met tante Jo? Wat heeft grootpapa haar nagelaten? ‘How about auntie Jo? What has granddad bequeathed to her?’ Nou, dat weet ik niet, maar ik geloof ‘Well, I don’t know, but I believe . . . ’ a.

dat grootpapa zijn buren de KLOK heeft willen that granddad his neighbours the clock has want nalaten. bequeath

b.

dat [ zijn buren]1 grootpapa t 1 de KLOK heeft willen that his neighbours granddad the clock has want nalaten. bequeath ‘that granddad wanted to bequeath the clock to his neighbours’

A syntactic typology of topic, focus and contrast

27

Our proposal makes a final prediction concerning the distribution of contrastive topics and contrastive foci. This prediction is not about what cannot be contained in the domain of contrast, but rather about what must be contained in it. In particular, it can be shown to follow from our assumptions that a contrastive topic cannot move to a position below a contrastive focus. This is not because there is something inherent in the notion of DoC marking that requires a contrastive focus to be part of the DoC for a contrastive topic. However, if it is not included in this domain, the resulting discourse will be incoherent. We can demonstrate this using the following data: (21)

Hoe zit het met de nietmachine? Wie heeft Jan daarom gevraagd? ‘What about the stapler? Who has asked John for that?’ Nou, dat weet ik niet, maar . . . ‘Well, I don’t know, but . . . ’ a.

ik geloof dat PIET Jan om de liniaal heeft willen I believe that Peter John for the ruler has want vragen. ask

b. # ik geloof dat PIET [om de liniaal]1 Jan t 1 heeft willen I believe that Peter for the ruler John has want vragen. ask c.

ik geloof dat [om de liniaal]1 PIET Jan t 1 heeft willen I believe that for the ruler Peter John has want vragen. ask ‘I believe that Peter has wanted to ask John for the ruler.’

The leading sentence in (13) sets up a context in which we are discussing the various things that his fellow workers have asked John for. In all three answers there is a shift in topic from the stapler to the ruler, identifying the latter as a contrastive topic. The set of elements out of which this contrastive topic is selected are those things that Peter – as opposed to other fellow workers of John’s – might have asked John for (this is because Peter is a contrastive focus). (21a) is a felicitous answer because nothing moves, and hence the domain of contrast for both topic and focus can be construed freely so as to fit the context. The short topic movement in (21b) marks a domain of contrast that contains two open positions and that therefore must undergo existential closure, yielding λx ∃y [y has wanted to ask John for x] (see the beginning of this section for details).

28

Ad Neeleman, Elena Titov, Hans van de Koot and Reiko Vermeulen

Since by assumption the interpretation of y must be based on the immediately preceding context (here, the discussion about things his fellow workers have asked John for), we are forced to interpret the existentially bound variable as ‘his fellow workers’. In other words, the contrast is based on the set of office equipment that his fellow workers have wanted to ask John for, but it should be the set of office equipment that Peter has wanted to ask John for. Hence the degraded status of the answer in (21b). The answer in (21c) is fully grammatical because Peter is included in the domain of contrast and hence the contrastive topic is correctly selected out of the set of office equipment that Peter has wanted to ask John for. As expected, in a context in which the indirect object John is interpreted as contrastive focus, short movement of the topic is felicitous. We can see this if the leading question sets up a context in which we are discussing which pieces of office equipment Peter has asked his fellow workers for. In this case, the subject need not be included in the domain of contrast, as it is already made available by the context. Thus, in (22b), the open subject variable will correctly be specified as Peter: (22)

Hoe zit het met de nietmachine? Wie heeft Piet daarom gevraagd? ‘What about the stapler? Who has Pete asked for that?’ Nou, dat weet ik niet, maar . . . ‘Well, I don’t know, but . . . ’ a.

ik geloof dat Piet JAN om de liniaal heeft willen I believe that Peter John for the ruler has want vragen. ask

b.

ik geloof dat Piet [ om de liniaal]1 JAN t 1 heeft willen I believe that Peter for the ruler John has want vragen. ask ‘I believe that Peter has wanted to ask John for the ruler.’

Let us summarize the results of this section. We have argued thatA’-scrambling in Dutch is associated with the notion contrast. In particular, this type of movement marks the material relevant for calculating the set of alternatives on which the contrast operates. As a consequence, not all topics and foci can undergo A’movement, but only those that are interpreted contrastively. Despite the fact that topic and focus movement have an identical trigger, they behave differently in certain respects. This is because the domain of contrast of a focus is taken from its background, while the domain of contrast of a topic is taken from its comment.

A syntactic typology of topic, focus and contrast

29

As a consequence of the interaction between DoC marking and the informationstructural constraints in (12), movement of a topic out of a constituent containing a focus is acceptable, but not the other way around.11

3. Interim conclusion As explained in the introduction, cartography aims at a one-to-one relation between syntactic position and interpretive effect. For example, the interpretive functions combined in the C-node of Government and Binding Theory are distributed across several heads, which encode such things as subordination and sentence type. As a result of this kind of decomposition, a clause consists of a large number of functional projections generated in a fixed order. These should presumably include projections hosting topics and foci, as in Rizzi (1997). On this view of syntax, one would at first sight expect universal ordering restrictions between any two constituents with distinct syntactic functions. Of course, reality is more complex and the way this additional complexity is dealt with in cartography is by allowing languages to vary in the extent to which movement takes place in overt or covert syntax. However, it can be demonstrated that this is insufficient to deal with the distribution of topics and foci in Dutch. In other empirical domains that display word order variation, it has already been shown that an analysis in terms of a single hierarchy of functional projections is untenable, even under an extensive parameterization of movement operations. For example, in reaction to Cinque (1999), a number of authors have pointed to problems with the order of adverbials (see Bobaljik 1999, Haider 2000, Ernst 2001, and Nilsen 2003). Bobaljik observes that arguments, verbs and auxiliaries in Italian can be freely interspersed in Cinque’s (1999) adverbial sequence. This fact is incompatible with Cinque’s proposal that the adverbial sequence is given by the order of functional projections in the verbal domain. Mutatis mutandis, the same conclusion holds of the variable placement of topics and foci in Dutch (see section 6 for further details). Bobaljik argues that this hierarchy paradox may yet yield to a phrase-structural account. He suggests that there are independent argumental and adverbial hierarchies, which can be conceived of as separate tiers of the syntactic representation that are ultimately collapsed into a single structure. Bobaljik draws an analogy with the shuffling together of two decks of cards, a process that preserves the internal order of each deck while it intersperses the cards of one deck among those of the other. Although Bobaljik’s proposal provides an account of the Italian data in terms of what one might call ‘relativized cartography’, it cannot capture the syntax of

30

Ad Neeleman, Elena Titov, Hans van de Koot and Reiko Vermeulen

topic and focus movement in Dutch. One could imagine that there is a topic-focus hierarchy (TopP > FocP), whose positions can be freely interspersed with positions in Bobaljik’s argumental hierarchy (AgrSP > AgrIOP > AgrOP). However, this is not sufficient to explain the fact that the order of constituents interpreted as topic and focus is free when they remain in situ, but rigid as soon as one of them moves. As we have shown, a topic can move out of a constituent containing an in situ focus, but a focus cannot move out of a constituent containing an in situ topic. The problem this raises for a ‘relativized cartography’ approach is that moved topics and moved foci would occupy positions in the topic-focus hierarchy, while topics and foci that remain in situ would occupy positions in the argumental hierarchy. Hence, ordering restrictions must be formulated that involve positions in more than one hierarchy, showing that a tier-based account is insufficient.

4. Japanese topic fronting12 In this section we discuss an example of a rule that mentions the feature [topic] and hence generalizes over aboutness topics and contrastive topics. The rule in question is part of Japanese grammar and requires topics to occur in clauseinitial position: (23)

[Topic] is licensed in clause-initial position.

It may be surprising to some readers that the rule in question regulates word order rather than the distribution of the particle wa, which is often taken to be a topic marker. We will show, however, that there is only a one-way implication between topichood and wa-marking: topics are marked by wa, but constituents marked by wa need not be topics. We will show that non-clause-initial waphrases systematically fail to meet tests for topichood. This proposal differs substantially from what one might consider the standard analysis of Japanese topics. It was suggested early on by Kuno (1973) that there are two uses of wa: thematic and contrastive. Phrases marked with thematic wa typically appear in clause-initial position and are interpreted as what the rest of the sentence is about. Phrases marked with contrastive wa implicate contrast and may remain in-situ. Although this goes some way towards the proposal we defend here, thematic and contrastive wa-phrases are generally analyzed as aboutness topics and contrastive topics, respectively (see Heycock 2007 for an overview of the literature). (Some of) the latter, then, would be topics in non-clause-initial position, contra (23).

A syntactic typology of topic, focus and contrast

31

We will now provide some arguments for (23), based on well-established tests for topichood. To begin with, in the reply to a request such as tell me about X, X must be interpreted as an aboutness topic (see Reinhart 1981 for relevant discussion): (24)

a. b.

Tell me about John. Well, John is a PhD student enrolled at the University of Lund.

As an aside, we may note that there is something puzzling about this test. We have argued in the introduction that constituents that refer back to the current topic of discourse do not need to be linguistic topics themselves. Given that John seems to be introduced as the topic of discourse in (24a), why can’t John in (24b) simply be a discourse-anaphoric expression? The solution to this puzzle lies in the fact that any given discourse comes in chunks (units of discourse), and for each of these a topic must be established. The imperative in (24a) is an instruction to start a new unit of discourse, and hence the constituent used to fix the topic of discourse for this unit – John in (24b) – must itself be a linguistic topic. When we apply the ‘tell me about’ test to Japanese, it turns out that the item X in the response must be marked with wa and must appear in clauseinitial position, as predicted by (23). The examples in (25) illustrate the point for subjects; those in (26) do so for objects.13 (We will discuss the nature of the empty category in (26a) below.) (25)

(26)

Tell me about that dog. a. Sono inu-wa kinoo John-o that dog-WA yesterday John-ACC b. # John-o1 sono inu-wa kinoo t 1 John-ACC that dog-WA yesterday ‘The dog bit John yesterday.’

kande-simatta. bite-ended.up kande-simatta. bite-ended.up

Tell me about that hat. kinoo e1 kaimasita. a. Sono boosi-wa1 John-ga John-NOM yesterday bought that hat-WA sono boosi-wa kinoo kaimasita. b. # John-ga yesterday bought John-NOM that hat-WA ‘John bought that hat.’

Contrastive topics display comparable behavior. As explained in section 3, one function of a contrastive topic is to shift the current topic of discourse from one item to another. The exchange in (27) therefore forces Bill-wa in the answer to be a contrastive topic, as this constituent is used to shift the topic from John.

32

Ad Neeleman, Elena Titov, Hans van de Koot and Reiko Vermeulen

As predicted by (23), Bill-wa must appear in clause-initial position. (It differs from the simple aboutness topics in (25) and (26) in that it must bear stress.) The same observation holds when the constituent interpreted as contrastive topic is the object, as demonstrated by the examples in (28).14 (27)

(28)

What did John eat at the party yesterday? Hmm, John-wa doo-ka sira-nai-kedo, ‘Well, I don’t know about John, but . . . ’ MAME-O tabeteita a. Bill-wa 8-zi-goro Bill-WA 8 o’clock-around beans-ACC was.eating b. # MAME-O1 Bill-wa 8-zi-goro t 1 tabeteita beans-ACC Bill-WA 8 o’clock-around was.eating ‘As for Bill, he was eating beans around 8 o’clock.’

(yo) (PRT) (yo) (PRT)

Who ate the pasta at the party yesterday? Hmm, pasta-wa doo-ka sira-nai-kedo, ‘Well, I don’t know about the pasta, but . . . ’ tabeteita (yo) a. # BILL-GA mame-wa 8-zi-goro Bill-NOM beans-WA 8 o’clock-around was.eating (PRT) b. Mame-wa1 BILL-GA 8-zi-goro t 1 tabeteita (yo) beans-WA Bill-NOM 8 o’clock-around was.eating (PRT) ‘As for the beans, Bill was eating them around 8 o’clock.’

Thus, contrary to what is commonly assumed, wa-phrases that meet the interpretive criteria for contrastive topichood must appear clause-initially. They cannot remain in situ. Although the rule in (23) generalizes over aboutness topics and contrastive topics, the syntax of these two types of topics is not identical. As argued in the introduction, [contrast] licenses A’-movement (see (5)). This generalization seems to extend to Japanese: it has been argued that contrastive topics bind an A’-trace, while aboutness topics are base-generated in a left-peripheral position and can be associated with an (empty) resumptive pronoun (Saito 1985 and Hoji 1985, but see Kuroda 1988 and Sakai 1994 for an opposing view). Hoji demonstrates that this distinction explains a number of differences between aboutness topics and contrastive topics, including contrasts involving weak crossover, reconstruction for binding and sensitivity to island constraints. For example, an aboutness topic can be associated with a position inside a relative clause, but a contrastive topic cannot. This allows the former to appear in a non-thematic, dislocated position in the matrix clause. Moreover, the pronoun inside the relative

A syntactic typology of topic, focus and contrast

33

clause that is associated with an aboutness topic can be overtly realized, but positions associated with contrastive topics cannot contain a resumptive pronoun (the data below are slightly modified from Hoji 1985: 152, 161): (29)

Sono boosi-wa 1 [TP John-ga [NP [TP e2 e1 / sore-o1 kabutteita] John-NOM that hat-WA it-ACC was.wearing hito2 ]-o yoku sitteiru] person-ACC well know ‘As for that hat, John knows well the person who was wearing it.’

(30) ?*(Susan zya nakute) Mary-wa 1 [TP John-ga [NP [TP e2 (not Susan, but) Mary-WA John-NOM e1 /kanozyo1 -o butta] hito2 ]-o sagasite-iru] she-ACC hit person-ACC looking.for Lit.:‘(Not Susan, but) Mary, John is looking for a person who hit (her).’ In sum, contrastive topics in Japanese, like their Dutch counterparts, undergo A’-movement. Let us point out (perhaps superfluously) that the trigger for the movement is different in the two languages. The variation in landing site found in Dutch suggest that A’-scrambling marks the domain of contrast. The fact that topics must move to clause-initial position in Japanese is suggestive of a different trigger, possibly marking of comments. We will not elaborate on this here, leaving (23) as a descriptive generalization. As pointed out at the outset of this section, the rule in (23) mentions the notion [topic], rather than the morphological marker wa (which typically accompanies topics). This implies that if there are any wa-phrases not interpreted as topic, these need not appear in first position. The literature acknowledges that there are wa-phrases that need not be fronted, but at the same time it is often claimed that any constituent marked by wa is a topic (but see Kuroda 1988, 2005). However, as a matter of logic, the fact that topics are marked by wa does not warrant the conclusion that wa attaches only to topics. To elucidate the issue, let us consider the interpretation of non-clause-initial wa-phrases. If these systematically fail to be interpreted as topics, the rule in (23) would receive support. If they do meet tests for topichood, this would refute (23). (Rest assured: the data confirm the rule in (23).) We first turn to unstressed wa-phrases. Although an aboutness topic must appear in clause-initial position, an unstressed wa-phrase can follow a fronted focus in the response to a question like (31). The rule in (23) predicts that the wa-phrase in (31b) should not be a topic. This seems to be correct; in particular sono inu-wa ‘this dog-wa’ is best characterized as a simple discourse-anaphoric expression (on a par with the English pronoun in (3b)).

34

(31)

Ad Neeleman, Elena Titov, Hans van de Koot and Reiko Vermeulen

Who did the dog bite? a. sono inu-wa JOHN-O kinoo that dog-WA John-ACC yesterday b. JOHN-O1 sono inu-wa kinoo t 1 John-ACC that dog-WA yesterday ‘The dog bit John yesterday.’

kande-simatta bite-ended.up kande-simatta. bite-ended.up

Whereas the requests in (25) and (26) require that a new unit of discourse be opened, a wh-question and the utterance that answers it belong to the same unit of discouse. Therefore, no new topic of discourse needs to be established following the question in (31).15 This allows the topic of discourse (as previously introduced) to carry over to the reply, which in turn allows ‘the dog’to be a simple referring expression, rather than a linguistic topic. That discourse anaphoricity – rather than linguistic topichood – is the relevant notion for wa-marking of the subject in the above examples is confirmed by the fact that the reply in (31b) is infelicitous if uttered in response to a question that does not mention ‘the dog’, such as What happened? (Kuno 1973, Tomioka 2007).16 Moreover, the wa-phrases in (31) do not show the syntactic behavior of topics. Recall that an aboutness topic can be base-generated in a non-thematic position in the matrix clause and be associated with an (empty) pronoun inside a relative clause. However, as shown below, an unstressed wa-phrase licensed in a context like (31) does not display such behavior, regardless of its position: (32)

Who did the dog that the child bought yesterday bite? a. # [sono kodomowa]1 kooen-de [NP [TP pro1 e2 kinoo that child-WA park-at yesterday katta] inu2 ]-ga JOHN-O kande-simatta. bought dog-NOM John-ACC bite-ended.up b. # JOHN-O3 [sono kodomo-wa]1 kooen-de [NP [TP pro1 e2 John-ACC that child-WA park-at kinoo katta] inu2 ]-ga t3 kande-simatta. yesterday bought dog-NOM bite-ended.up ‘The dog that this child bought yesterday bit John in the park.’

If all unstressed wa-phrases were aboutness topics, the above examples should be grammatical, on a par with the example in (29). Thus, discourse-anaphoric wa-phrases have a syntax that differs from that of aboutness topics: only the latter are base-generated in a dislocated position, binding an empty pronominal in a thematic position.

A syntactic typology of topic, focus and contrast

35

We now turn to stressed wa-phrases that remain in situ, of which an example is given in (33) (see Kuno 1973, Hoji 1985, Saito 1985, Watanabe 2003, Hara 2006 and Tomioka 2009 for discussion). It is often assumed that such wa-phrases are contrastive topics. We believe that such a characterization is incorrect. Stressed in-situ wa-phrases do implicate contrast, but they fail to meet independently established criteria for topichood. (33)

Who was being helpful at the accident scene? JOHN-GA 3-nin-wa tasuketa. John-NOM 3-CL-WA helped ‘John helped at least three people.’

There has been much recent work on the interpretation of contrastive wa-phrases. Hara (2006), for instance, argues that a sentence containing a stressed wa-phrase induces the presupposition that a scalar alternative stronger than the assertion exists, as well as the implicature that this stronger alternative could be false. This seems to provide a correct characterization of the interpretation of stressed waphrases, as suggested by the use of ‘at least’ in the English translation of (33). However, nothing in Hara’s description of the semantics of contrastive wa-phrases forces these constituents to be topics. As Hara notes, stressed wa-phrases differ from contrastive topics in that they do not require the presence of a focus in the sentence, something that contrastive topics strongly favor (as observed for German in B¨uring 1997). Moreover, it is a well-known property of topics (both contrastive and non-contrastive) that they must be specific, simply because it is difficult to make a statement about something unspecific (see Reinhart 1981). This effectively rules out a topic interpretation for 3-nin ‘3-classifier’ in (33), which is a non-specific, quantified nominal. Indeed, an English translation of (33) that explicitly marks the object as a topic is decidedly odd: #As for at least three people, John helped them Finally, a theory that treats all contrastive wa-phrases as contrastive topics would have to say that fronting of topics is optional. However, the wa-phrase in (34) (which, to repeat the point, does not meet the criteria for topichood) resists fronting:

(34)

#3-nin-wa1 John-ga t 1 tasuketa 3-CL-WA John-NOM helped ‘John helped at least three people.’

Summarizing this section, the syntactic distribution of both aboutness topics and contrastive topics in Japanese can be captured by a single rule that refers to the notion [topic]. This correctly predicts the non-topical properties of wa-phrases that do not occupy clause-initial position.

36

Ad Neeleman, Elena Titov, Hans van de Koot and Reiko Vermeulen

5. The position of Russian foci17 In the previous sections we have discussed rules that mention [contrast] (Dutch A’-scrambling) and [topic] (Japanese topic fronting). We now consider a rule that mentions the third and final feature in (1), namely [focus]. The Russian data discussed below show that all foci in this language share an underlying clause-final position. In other words, the grammar of Russian has the following rule: (35)

[Focus] is licensed in clause-final position.

It is irrelevant to the argument we present whether foci are base-generated clausefinally or end up there via a derivation involving movement. The generalization in (35) does not hold on the surface. New information foci indeed show up clause-finally, as illustrated in (36)18 , but contrastive foci typically occupy positions further to the left, as we will see below (see Krylova and Khavronina 1988, King 1995, and Brun 2001).19 ˇ (36) a. Cto Saˇsa cˇ itaet? Saˇsa cˇ itajet KNIGU what-ACC Sasha reads? Sasha reads book-ACC ‘What does Sasha read?’ ‘Sasha reads a book.’ ˇ Knigu cˇ itaet SASA b. Kto cˇ itajet knigu? book-ACC reads Sasha who reads book-ACC ‘Who reads the book?’ ‘Sasha reads the book.’ Anja dala knigu KATE c. Komu Anja dala knigu? who-DAT Anna gave book-ACC Anna gave book-ACC Kate-DAT ‘Who did Anna give the book to?’ ‘Anna gave the book to Kate.’ ˇ KNIGU Anja dala Kate? Anja dala Kate d. Cto What-ACC Anna gave Kate-DAT Anna gave Kate-DAT book-ACC ‘What did Anna give Kate?’ ‘Anna gave Kate the book.’ What our claim amounts to, then, is that the launching site for the movement of contrastive foci is the position in which new information foci must surface. This follows if contrastive foci are a composite of the features [focus] and [contrast], while new information foci are characterized by the first feature only. Movement of contrastive foci to the left periphery would then be triggered by the feature [contrast], but the launching site of that movement would be dictated by (35): (37)

a. b.

[CP . . . [FOCUS] ] [CP [FOCUS, CONTRAST]1 . . . t1 ]

A syntactic typology of topic, focus and contrast

37

The reader may wonder how we can substantiate the generalization in (35), if contrastive foci move. A first argument is that it correctly follows from (35) that new information and contrastive foci are in complementary distribution: any sentence can contain only a single focus because all foci are licensed in clause-final position. For example, a locational PP like ‘in the concert hall’ can appear in final position on the condition that it is interpreted as new information focus (see (38a)). This is no longer possible when another category is fronted as a contrastive focus. The string in (38b) is only grammatical if ‘in the concert hall’ is interpreted as part of the background, in which case it remains unaccented and presumably precedes the trace of ‘the jazz pianist’. (38)

a.

Ja sluˇsala jazz-pianista v KONCERTNOM ZALE. I listened.to jazz-pianist-ACC in concert hall ‘I listened to the jazz pianist in the concert hall.’ b. *JAZZ-PIANISTA1 ja sluˇsala t1 v KONCERTNOM ZALE. jazz-pianist-ACC I listened.to in concert hall ‘I listened to the jazz pianist in the concert hall (and not the jazzguitarist)’

Our second argument is based on the scopal properties of contrastive foci, and in particular on the observation that they take scope in the same position as new information foci. The data fall out from (35), assuming that focus movement obligatorily reconstructs for scope.20 As a point of departure, consider the scopal properties of new information foci. In general, Russian exhibits surface scope (see Ionin 2001) and, in line with this, quantifiers that constitute a new information focus scope under other quantifiers. After all, new information foci occupy the clause-final position.21 (39)

a.

b.

ˇ ˇ Odin mal’ˇcik ljubit KAZDUJU DEVOCKU. one boy loves every girl-ACC ‘One boy loves every girl.’ ∃ > ∀; ∗ ∀ > ∃ ˇ V kaˇzdom klasse odnu devoˇcku ljubit KAZDYJ in every classroom one girl-ACC loves every ˇ MAL’CIK.

c.

boy ‘In every classroom every boy loves one girl.’ ∃ > ∀s ; ∗ ∀s > ∃ ˇ KAZDOMU Ty predstavil odnogo uˇcitelja you introduced one teacher-ACC every STUDENTU student-DAT

38

Ad Neeleman, Elena Titov, Hans van de Koot and Reiko Vermeulen

d.

‘You introduced one teacher to every student.’ ∃ > ∀s ; ∗ ∀s > ∃ ˇ KAZDOGO Ty predstavil odnomu studentu you introduced one student-DAT every ˇ UCITELJA. teacher-ACC

‘You introduced every teacher to one student.’

∃ > ∀; ∗ ∀ > ∃

The pattern of surface scope breaks down in the case of contrastive foci.22 Even though these are fronted, they systematically take lowest scope. That is, they reconstruct obligatorily to a position below other quantifiers. Thus, the fronted contrastive foci in (40) take scope in exactly the same position as the in situ new information foci in (39). (40)

a.

b.

c.

d.

ˇ ˇ KAZDUJU DEVOCKU ja xoˇcu cˇ toby odin mal’ˇcik ljubil loved every girl-ACC I want that one boy (a ne kaˇzduju babuˇsku). and not every grandma-ACC ‘I want one boy to love every girl ∃ > ∀; ∗ ∀ > ∃ (and not every grandma).’ ˇ ˇ KAZDYJ MAL’CIK ja xoˇcu cˇ toby v kaˇzdom klasse every boy I want that in every classroom odnu devoˇcku ljubil (a ne kaˇzdyj deduˇska). one girl-ACC loved and not every grandpa-NOM ‘In every class, I want one girl to be loved by ∃ > ∀; ∗ ∀ > ∃ every boy (and not by every grandpa).’ ˇ KAZDOMU STUDENTU ja xoˇcu cˇ toby ty predstavil every student-DAT I want that you introduced odnogo uˇcitelja, (a ne kaˇzdomu professoru) one teacher-ACC and not every professor-DAT ‘I want you to introduce every teacher to one ∃ > ∀; ∗ ∀ > ∃ student (and not every dean).’ ˇ ˇ KAZDOGO UCITELJA ja xoˇcu cˇ toby ty predstavil every teacher-ACC I want that you introduced odnomu studentu, (a ne kaˇzdogo dekana). one student-DAT and not every dean-ACC ‘I want that one students is introduced by you ∃ > ∀; ∗ ∀ > ∃ to every teacher (and not to every dean).’

It is not very surprising that Russian contrastive foci can move, given that contrast licenses A’-movement (see (5)). However, what is surprising is that the

A syntactic typology of topic, focus and contrast

39

position into which contrastive foci reconstruct should be as low as it seems to be. For example, subjects and indirect objects normally outscope direct objects in Russian, because their position c-commands (and precedes) the direct object position (see Dyakonova 2007). However, if these elements are fronted as contrastive foci, they must scope under the direct object, suggesting that the fronting operation is launched from the sentence-final position. A reviewer casts some doubt on the relevance of the example in (40b) in view of the preverbal position of the object, plausibly the result of a scrambling operation across the position from which the subject has moved. As a consequence, the fact that the object outscopes the subject may seem unsurprising. Three remarks are in order. First, this criticism does not extend to the data in (40c,d). Second, the preverbal position of the object is the result of it belonging to the background. Now, backgrounded objects of monotransitive verbs may marginally be placed in postverbal position. However, if we shift the preverbal object in (40b) to a postverbal position, this does not affect its scopal interpretation. Third, a backgrounded object of a ditransitive verb need not be fronted if the other object appears as a sentence-initial topic (for reasons unclear to us). In this situation, a displaced contrastively focused subject must still scope under the postverbal object: (41)

ˇ ˇ ]1 ja xoˇcu, cˇ toby dal dvum devoˇckam t1 (a Knigu [ KAZDYJ MAL’CIK ne kaˇzyj deduˇska). book-ACC every boy I want that gave two girls-DAT (and not every grandpa) ‘As for books, I want every boy to give them 2 > ∀; ∗ ∀ > 2 to two girls’

A final argument in support of (35) can be based on so-called split scrambling. Russian allows extraction of a contrastively focused element out of a larger constituent. There are many questions surrounding this phenomenon, but what is crucial here is that the material stranded by scrambling of the focused constituent provides an overt indication of the movement’s launching site. If (35) holds, what we predict is that the stranded material must always occupy a clause-final position. The data below bear this out. If extraction takes place out of the object, then the object itself must be clause-final. If it takes place out of the subject, it is this constituent that must appear in clause-final position. (Although not illustrated here, all other examples in which ‘performance’ is not clause-final are ungrammatical as well.)23

40

Ad Neeleman, Elena Titov, Hans van de Koot and Reiko Vermeulen

(42)

a.

JAZZ-PIANISTA1 mal’ˇciki slyshali [vystuplenie t1 ] (a heard performance-ACC (and jazz pianist-GEN boys ne jazz-gitarista) not jazz-guitarist-GEN) ‘The boys heard the performance of the jazz pianist (and not of the jazz guitarist).’ b. *JAZZ-PIANISTA1 mal’ˇciki [vystuplenie t1 ] slyshali (a performance-ACC heard (and jazz pianist-GEN boys ne jazz-gitarista) not jazz-guitarist-GEN)

(43)

a.

b.

JAZZ-PIANISTA1 devoˇcku potrjaslo [vystuplenie t1 ] (a (and jazz pianist-GEN girl-ACC amazed performance ne jazz-gitarista) not jazz-guitarist-GEN) ‘The performance of the jazz pianist amazed the girl (and not of the jazz guitarist).’ JAZZ-PIANISTA1 devoˇcku [vystuplenie t1 ] potrjaslo (a amazed (and jazz pianist-GEN girl-ACC performance ne jazz-gitarista) not jazz-guitarist-GEN)

The same point can be made on the basis of double object constructions. Split scrambling requires that the object containing the trace of the fronted constituent appear in final position: (44)

a.

JAZZ-PIANISTA1 ja podaril Saˇse [fotografiju t1 ] (a Sasha-DAT picture-ACC (and Jazz-pianist-GEN I gave ne jazz-gitarista). not jazz-guitarist-GEN) ‘I gave a picture of a jazz-pianist to Sasha.’ (a b. *JAZZ-PIANISTA1 ja podaril [fotografiju t1 ] Saˇse picture-ACC Sasha-DAT (and Jazz-pianist-GEN I gave ne jazz-gitarista). not jazz-guitarist-GEN) c. ?JAZZ-PIANISTA1 ja posvjatil etot prazdnik Jazz-pianist-GEN I dedicated this celebration-ACC [vystupleniju t1 (a ne jazz-gitarista). performance-DAT and not jazz-guitarist-GEN) ‘I dedicated this celebration to the performance of a jazz-pianist’

A syntactic typology of topic, focus and contrast

41

d. *JAZZ-PIANISTA1 ja posvjatil [ vystupleniju t1 ] etot Jazz-pianist-GEN I dedicated performance-DAT this prazdnik (a ne jazz-gitarista). celebration-ACC (and not jazz-guitarist-GEN) To summarize, if we assume, as stated in (35), that all foci in Russian are licensed in clause-final position, then both the scopal properties of fronted foci and the position of stranded material can be readily understood.

6. Decomposition and cartography The analytical depth of our analyses of Dutch, Japanese and Russian is not equal (only in the Dutch case do we have an explanation for the empirical generalization we argued for). Nevertheless, we may draw the following conclusions: (i) There is no designated landing site for moved topics and foci (based on the Dutch data); (ii) The features [topic], [focus] and [contrast] have independent syntactic effects (based on the Dutch, Japanese and Russian data). These conclusions allow us to complete our case against cartography. We have already argued that the variation in the landing site of A’-scrambling observed in Dutch is very hard to reconcile with the rigid phrase structure that characterizes cartography. We sketched the core of the argument in section 3, but we can be more precise. In earlier work we have demonstrated that the following assumptions must be made if moved topics and foci are licensed in the specifier of TopP and FocP, respectively (see Neeleman and Van de Koot 2008 for details): i.

The position of TopP and FocP is free (at least in Dutch). (This captures the observed variation in landing sites.) ii. Projection of either [contrast] or [topic] and [focus] is optional (at least in Dutch). (This allows in-situ topics and foci.) iii. Heads containing [contrast] mark their complements as the domain of contrast. (This, in conjunction with restrictions on information structure, captures the observed ordering restrictions.) However, each of these assumptions sacrifices an assumption central to the cartographic framework. Assumptions (i) and (ii) give up the idea that there is a fixed clausal skeleton. Assumption (iii) gives up the idea that movement is triggered by properties of the specifier. These sacrifices seem considerable. But the Dutch data are even more damaging in light of our second conclusion. Within the cartographic framework, the independent syntactic effects of [topic], [focus] and [contrast] require an account in terms of three separate functional projections: TopP, FocP and ContrastP. This is because cartography strives for a

42

Ad Neeleman, Elena Titov, Hans van de Koot and Reiko Vermeulen

one-to-one relation between syntactic position and interpretive effect, and the data motivate three separate interpretively relevant syntactic features. Crosslinguistic variation would be captured by the extent to which these projections trigger displacement. In Japanese, topics appear in the specifier of TopP. In Russian, foci appear in the specifier of FocP, while contrastive foci move on to the specifier of ContrastP. Finally, in Dutch, contrastive topics and foci move to the specifier of ContrastP. The addition of ContrastP to the topic-focus hierarchy requires an additional ordering statement. Given that TopP is generally assumed to dominate FocP, there are three logical possibilities: (45)

a. b. c.

ContrastP > TopP > FocP TopP > ContrastP > FocP TopP > FocP > ContrastP

The Russian data suggest that ContrastP dominates FocP, because contrastive foci in this language move from the position in which foci in general are licensed. This rules out (45c). We are thus left with (45a) and (45b), but it can be demonstrated that neither of these orders is compatible with the Dutch data. If ContrastP dominates TopP, the landing site for contrastive topics and foci is identical and it is therefore impossible to capture the ordering restrictions associated with topic and focus movement. In particular, this arrangement implies that moved foci always cross the position in which topics are licensed. As a result, it becomes hard, if not impossible, to explain why – at an observational level – topics may move across foci but foci may not move across topics. As far as we can see, this point is not affected by the possibility of a recursive ContrastP. This rules out (45a). If ContrastP occupies a position between TopP and FocusP, one might attempt to capture the Dutch data by saying that contrastive topics move on to the specifier of TopP, while contrastive foci surface in ContrastP. We should then require TopP to trigger movement, which in turn would lead to expect displacement of aboutness topics, contrary to fact. This rules out (45b). These considerations suggest that, although [contrast] has syntactic effects that can be distinguished from those of [topic] and [focus], a cartographic decomposition into three separate functional projections is not possible. But the logic of cartography requires such a decomposition.

A syntactic typology of topic, focus and contrast

43

Notes ∗

1.

2.

3. 4. 5.

6.

7.

8.

9.

This paper is part of the output of the AHRC-funded project ‘A Flexible Theory of Topic and Focus Movement’ (Grant nr. 119403). Our notion of (linguistic) topic corresponds to what Lambrecht calls ‘referenceoriented topic expressions’, while our notion of indexing elements corresponds to his ‘role-oriented topic expressions’. Backgrounds differ from comments in this respect: as focus is a notion associated with propositions, there is no such thing as an all-background sentence. Backgrounded information is backgrounded with respect to a focus. Therefore, if focus is a propositional notion, the background and focus must be part of the same proposition. This section is partly based on Neeleman and Van de Koot (2008). Note that A’-scrambling can also affect categories other than DPs, including adverbials. Some speakers of Dutch marginally allow A-scrambling of a direct object across an indirect object, a possibility more generally available in German. The judgments reported here are from speakers who reject such scrambling. The proposal outlined here constitutes a correction on Neeleman and Van de Koot (2008), where it is assumed that A’-scrambling marks a comment or a background rather than a domain of contrast. The judgments given here and below are based on a pronunciation of the examples in which the constituent marked as focus carries a plain high tone, and the constituent marked as topic carries a tune consisting of a high tone, a low tone and a high boundary tone (this intonation of Dutch topics is in line with the observations reported in Van Hoof 2003). As far as we can judge these matters, this pronunciation is very similar to what is found in English. As was pointed out to us by MichaelWagner (p.c.), the context in (13) and comparable ones below do not force the interpretations indicated, but merely favour them. All the contexts we use are based on implicit multiple WH-questions. In (13), this question is Who ate what?, while in (16) below, it is What was eaten by whom? Answers to multiple WH-questions tend to be constructed in such a way as to line up the topic with the fronted WH-phrase and the focus with the in situ WH-phrase. Thus, Who ate what? is most commonly answered by something like John ate the BEANS, Mary ate the CHEESE, etc. However, when there is reason to do so, it is also possible to swap the topic and focus functions, yielding answers like JOHN ate the beans, MARY ate the cheese, etc. (see Roberts 1996 and B¨uring 2003). Given that the contexts we use presuppose implicit multiple WH-questions, the possibility of a topic-focus swap also presents itself. Hence, in evaluating our empirical claims, one should not just rely on the effects of context, but also consider other indicators of topic- and focushood. Notice that overt topic movement cannot be used to rescue structures in which a topic is contained in the background of a moved focus. We speculate that this is

44

Ad Neeleman, Elena Titov, Hans van de Koot and Reiko Vermeulen because “repair by movement” would violate Relativized Minimality: focus movement creates an A’-position that blocks association of the topic with its trace. If so, comparable structures in which a topic is linked to a pronoun should be acceptable. Indeed, there is a sharp contrast between (i) and (ii). Moreover, topic fronting across a moved focus is widely attested in languages that have clitic doubling (such as Italian) or allow radical pro drop (such as Japanese). (i) *Marie 1 heeft [ DIT boek]2 Jan t 1 zeker t 2 niet gegeven. Mary has this book John certainly not given.

10.

11.

12. 13.

14.

15.

(ii) Wat Marie betreft, [DIT boek]1 heeft Jan haar zeker t 1 niet what Mary regards, this book has John her certainly not gegeven. given. For related discussion, see Rochemont (1989), Culicover (1991), Browning (1996), among others. The answer in (15a) is felicitous if the B-accent on the subject is omitted. Doing so allows the example to be construed as providing indirect information about Fred, namely that he did not eat more from the beans than he did last year. We assume throughout that in situ focus does not move at LF. The data can also be captured, however, if in situ foci do move, so as to create a representation in which focus and background are constituents (abstracting away from many details; see Krifka 2006). Such an analysis requires two additional assumptions. To begin with, when overt focus movement takes place, LF-movement of material marked as belonging to a background must not be allowed. Such movement would repair the ill-formed example in (13b). By contrast, when no overt focus movement takes place, LF-movement must be generally available. For instance, the topic in (13a) must move, as it would otherwise be part of a background created by LF focus movement. (Note, however, that parallel structures created by overt movement give rise to ungrammaticality; see footnote 9 for discussion.) The restriction that marking relations established in overt syntax cannot be undone at LF has a precedent in the literature on multiple WH-questions: if overt movement is used to mark the scope of a WH-operator, its scope cannot be extended through subsequent LF-procedures. Consequently, Who wonders what John bought? cannot be interpreted as a multiple WH-question, whereas Who said that John bought what? does allow such a reading (see van Riemsdijk 1978). This section is based on Vermeulen 2008. For reasons unknown to us, an object wa-phrase prefers not to surface adjacent to a verb. In order to circumvent this issue, adverbials are inserted between object and verb throughout this section. If movement of contrastive foci in Japanese marks a domain of contrast, then the example in (27b) would also be ruled out by the constraint in (12b). We will not explore to what extent this is true. The question in (31) establishes ‘the dog’ as a discourse topic if uttered discourseinitially. In this case, ‘the dog’ must appear in clause-initial position in the question,

A syntactic typology of topic, focus and contrast

45

as in (i), indicating its linguistic topical status. On the other hand, if there is prior discourse, in which ‘the dog’ is already established as topic, it need not occupy clause-initial position, as in (ii). (i) sono inu-wa dare-o kande-simatta no? that dog-wa who-acc bite-ended.up Q (ii) dare-o1 sono inu-wa t 1 kande-simatta no? who-acc that dog-wa bite-ended.up Q 16. The claim that an in-situ unstressed wa-phrase is discourse anaphoric, and not a topic, is illustrated here with subjects. The same observation holds for objects: (i)

Q: Did John borrow the book from the library? KATTA A: Iya, John-wa sono hon-wa kekkyoku honya-de No, John-wa that book-wa in.the.end bookshop-at bought

17. This section is based on Neeleman and Titov (2008). 18. The most neutral word order dictated by the context in (36) requires sentence-final focus (see also Krylova and Khavronina 1988). However, it is possible for the focused constituent to scramble to a preverbal position in a context that does not force contrastive interpretation on the focused constituent in case this constituent is interpreted emphatically. Such constructions do not require the presence of an explicit member of the set of alternatives in the context and can often be uttered out of the blue. The emphatically focused constituent, however, cannot be analyzed as new information focus, as it contains an additional interpretation, namely a conventional scalar implicature indicating that it is surprising or noteworthy in some way (see also Zanuttini and Portner 2003 for a similar analysis of exclamatives). In other words, although no explicit member of the set of alternatives is present in the context, the existence of such a set must be concluded and the emphatically focused constituent is interpreted as an unlikely member within this set: (i)

Kto cˇ itaet knigu? who reads book-ACC ‘Who reads the book?’

ˇ SASA cˇ itaet knigu Sasha reads book-ACC ‘(Out of all people) it is Sasha who reads the book!’

19. Space limitations do not permit us to explore why contrastive foci move in Russian. The null hypothesis from the current perspective is that Russian has DoC marking. However, the issue is orthogonal to the claim at the heart of this section, namely that new information foci and contrastive foci share an underlying clause-final position. 20. It has been claimed in linguistic literature that A’-fronting fixes scope relations in Russian (see, for example, Bailyn 2004). However, a careful examination of Russian sentences involving A’-scrambling reveals that a long-distance scrambled DP always reconstructs into the embedded clause: (i) Kaˇzduju devoˇcku odin mal’ˇcik xoˇcet cˇ toby ja poljubil I-nom loved every girl-ACC one boy-NOM wants that ‘One boy wants me to love every girl.’ ∃ > ∀; ∗ ∀ > ∃

46

Ad Neeleman, Elena Titov, Hans van de Koot and Reiko Vermeulen

The reason A’-fronting sometimes appears to result in ‘frozen’ scope (see (Ionin 2001) lies in the fact that a fronted quantifier not interpreted as a focus may have to reconstruct above the clause-final position, in case this position is occupied by a constituent interpreted as new information focus. This implies that the fronted quantifier must take scope over a quantified new information focus, even though it may take scope below quantifiers in other positions. 21. The examples in (39a) and (39b) do not form a minimal pair. This is because there is a tendency for clause-initial objects to be interpreted as topics, and hence as specific if they are indefinites. We control for this by adding a clause initial adverb that contains a universal quantifier on which the indefinite is dependent. This blocks a specific reading of the indefinite, as desired. The scope judgment given involves the indefinite and the universal subject (indicated by a subscript ‘S’). We leave out the adjunct in (39a) in order to avoid the slight artificiality of too many universal quantifiers in a single sentence. 22. In the Russian linguistic literature different types of focus are taken to receive different Intonational Contours (IKs). New information focus is marked by IK1 (a falling tone), whereas contrastive focus receives IK2 (emphatic stress similar to IK1, but higher in tone and more intense). For discussion, see Bryzgunova (1971), (1981), Yokoyama (1986), and Krylova and Khavronina (1988). The judgments in the main text presuppose that the constituents marked as contrastive foci bear IK2, while the rest of the sentence is destressed. 23. In (42) and (43) the DP out of which split-scrambling takes place must be sentencefinal and cannot be followed by any material including adverbs. An adverb can only be added as an afterthought after a fairly long pause.

References Bailyn, John 2004

Generalized Inversion. Natural Language and Linguistic Theory 22: 1–49. Belletti, Adriana (ed.) 2004 Structures and Beyond: The Cartography of Syntactic Structures. Vol. 3. Oxford: OUP. Bobaljik, Jonathan 1999 Adverbs: The hierarchy paradox. Glot International 4.9/10. Browning, Margaret 1996 CP recursion and that-t effects. Linguistic Inquiry 27: 237–256. Brun, Dina 2001 Information structure and the status of NP in Russian. Theoretical Linguistics: 109–135.

A syntactic typology of topic, focus and contrast

47

Bryzgunova, Elena 1971 O smyslorazliˇcitel’nyx vozmoˇznostjax russkoj intonatcii. Voprosy jazykoznanija 4:42–51. Bryzgunova, Elena 1981 Zvuk i Intonatsija Russkoj Rechi. Moscow: Russkij Yasyk. B¨uring, Daniel 1997 The Meaning of Topic and Focus. London: Routledge. B¨uring, Daniel 2003 On D-trees, beans and B-accents. Linguistics & Philosophy 26.5: 511– 545. Choi, Hye-Won 1999 Optimizing Structure in Context: Scrambling and Information Structure. Stanford: CSLI. Cinque, Guglielmo 1999 Adverbs and Functional Heads: A Cross-Linguistic Perspective. Oxford: OUP. Cinque, Guglielmo (ed.) 2002 Functional Structure in DP and IP: The Cartography of Syntactic Structures. Oxford: OUP. Culicover, Peter 1991 Topicalization, inversion, and complementizers in English. OTS Working Papers. Utrecht University. Dyakonova, Marina 2007 Russian Double Object Constructions. ACLC Working Papers 2(1): 3–30. Ernst, Thomas 2001 The Syntax of Adjuncts. Cambridge: CUP. Frey, Werner 2001 About the whereabouts of indefinites. Theoretical Linguistics 27: 137– 161. Giusti, Giuliana 2006 Parallels in clausal and nominal periphery. In: Mara Frascarelli (ed.), Phases of Interpretation. Berlin: Mouton de Gruyter. Haider, Hubert 2000 Adverb placement – Convergence of structure and licensing. Theoretical Linguistics 26: 95–134. Haider, Hubert and Inger Rosengren 1998 Scrambling. Sprache und Pragmatik 49. Lund. Hajiˇcov´a, Eva, Barbara Partee, and Petr Sgall 1998 Focus, topic and semantics. In: Elena Benedicto, Maribel Romero, and Satoshi Tomioka (eds.), Proceedings of the Workshop on Focus. UMass WPL 21, 101–124. Amherst: GLSA.

48

Ad Neeleman, Elena Titov, Hans van de Koot and Reiko Vermeulen

Hara, Yurie 2006

Grammar of knowledge representation: Japanese discourse items at interfaces. Ph. D. diss., University of Delaware. Heycock, Caroline 2007 Japanese -wa, -ga, and information structure. To appear in: Mamoru Saito, and Shigeru Miyagawa (eds.), Handbook of Japanese Linguistics. Oxford: OUP. Hoji, Hajime 1985 Logical Form constraints and configurational structures in Japanese. Ph. D. diss., University of Washington. Ionin, Tania 2001 Scope in Russian: Quantifier movement and discourse function. Ms. MIT. Jackendoff, Ray 1972 Semantic Interpretation in Generative Grammar. Cambridge: MIT Press. Jacobs, Joachim 1997 I-Topikalisierung. Linguistische Berichte 168: 91–133. King, Tracy H. 1995 Configuring Topic and Focus in Russian. Stanford: CSLI. Kiss, Katalin E´ 1998 Identificational focus versus information focus. Language 74: 245– 273. Krifka, Manfred 2006 Association with focus phrases. In: Val´eria Moln´ar and Susanne Winkler (eds.), Architecture of Focus. Berlin: Mouton de Gruyter. Krylova, Olga and Serafima Khavronina 1988 Word Order in Russian. Moscow: Russky Yazyk. Kuno, Susumu 1973 The Structure of the Japanese Language. Cambridge: MIT Press. Kuroda, Shigeyuki 1988 Whether we agree or not: A comparative syntax of English and Japanese. In: William J. Poser (ed.), Papers from the Second International Workshop on Japanese Syntax, 103–143. Stanford: CSLI. [Also in: Linguisticae Investigationes 12: 1–47. And reprinted in: Kuroda, Shigeyuki 1992. Japanese Syntax and Semantics: Collected Papers. Kluwer, Dordrecht.] Kuroda, Shigeyuki 2005 Focusing on the matter of topic: A study of wa and ga in Japanese. Journal of East Asian Linguistics 14:1–58. Lambrecht, Knud 1994 Information Structure and Sentence Form. Cambridge: CUP.

A syntactic typology of topic, focus and contrast

49

Mahajan, Anoop 1990 The A/A-bar distinction and movement theory. Ph. D. diss, MIT. McCoy, Svetlana 2003 Connecting information structure and discourse structure through “Kontrast”: The case of colloquial Russian particles -to, zˇ e, and ved’. Journal of Logic, Language and Information 12: 319–335. Moln´ar, Val´eria 2002 Contrast – from a contrastive perspective. In: H. Hallelgard, Stig Johansson, Bergljot Behrens, and Cathrine Fabricius-Hansen (eds.), Information Structure in a Cross-Linguistic Perspective, 147–161. Amsterdam/New York: Rodopi. Neeleman, Ad 1994 Complex predicates. Ph. D. diss., Utrecht University. Neeleman, Ad and Tanya Reinhart 1998 Scrambling and the PF interface. In: Miriam Butt and Wilhelm Gueder (eds.), The Projection of Arguments, 309–353. Stanford: CSLI. Neeleman, Ad and Elena Titov 2008 Focus, Contrast and Stress in Russian. Ms. UCL. Neeleman, Ad and Hans van de Koot 2008 Dutch scrambling and the nature of discourse templates. Journal of Comparative Germanic Linguistics 11. Nilsen, Øystein 2003 Eliminating positions: Syntax and semantics of sentential modification. Ph. D. diss., Utrecht University. Prince, Ellen 1981 Topicalization, focus-movement and Yiddish-movement: A pragmatic differentiation. Proceedings of the Berkeley Linguistic Society 7:249– 264. Reinhart, Tanya 1981 Pragmatics and Linguistics: An analysis of sentence topics. Philosophica 27:53–94. Reinhart, Tanya 1995 Interface strategies. Ms. Utrecht University. Reinhart, Tanya 1996 Interface economy – focus and markedness. In: Chris Wilder et al. (eds.), The Role of Economy Principles in Linguistic Theory. Berlin: Akademic Verlag. Reinhart, Tanya 2006 Interface Strategies. Cambridge: MIT Press. Rizzi, Luigi 1997 The fine structure of the left periphery. In: Liliane Haegeman (ed.), Elements of Grammar. Dordrecht: Kluwer.

50

Ad Neeleman, Elena Titov, Hans van de Koot and Reiko Vermeulen

Rizzi, Luigi (ed.) 2003 The Structure of IP and CP. The Cartography of Syntactic Structures, Vol. 2. Oxford: OUP. Roberts, Craige 1996 Information structure in discourse: Towards an integrated formal theory of pragmatics. In: James Yoon and Andreas Kathol (eds.), OSU Working Papers in Linguistics 49: Papers in Semantics, 91–136. Rochemont, Michael 1989 Topic islands and the subjacency parameter. Canadian Journal of Linguistics 34:145–170. Saito, Mamuro 1985 Some asymmetries in Japanese and their theoretical implications. Ph. D. diss., MIT. Sakai, Hiromu 1994 Complex NP constraint and case conversion in Japanese. In: Masaru Nakamura (ed.), CurrentTopics in English and Japanese: Proceedings of the Second English/Japanese Comparative Syntax Workshop, 179– 203. Tokyo: Hituzi Syoboo. Schwarzschild, Roger 1999 Givenness, AvoidF and other constraints on the placement of accents. Natural Language Semantics 7:41–177. Tomioka, Satoshi 2007 The Japanese existential possession: A case study of pragmatic disambiguation. Lingua 117:881–902. Tomioka, Satoshi 2009 Contrastive topics operate on speech acts. In: Malte Zimmermann and Caroline Féry (eds.), Information Structure: Theoretical, Typological, and Experimental Perspectives. Oxford: OUP. Vallduv´ı, Enric 1992 The Informational Component. New York: Garland. Vallduv´ı, Enric and Elisabet Engdahl 1996 The linguistic realization of information packaging. Linguistics 34:459–519. Vallduv´ı, Enric and Maria Vilkuna 1998 On rheme and konstrast. Syntax and Semantics 29:79–108. Vanden Wyngaerd, Guido 1989 Object shift as anA-movement rule. MITWorking Papers in Linguistics 11:256–271. Van Hoof, Hanneke 2003 The rise in the rise-fall contour: Does it evoke a contrastive topic or a contrastive focus? Linguistics 41:515–563.

A syntactic typology of topic, focus and contrast

51

Van Riemsdijk, Henk 1978 On the diagnosis of wh-movement. In: Samuel J. Keyser (ed.), Recent Transformational Studies in European Languages. Cambridge: MIT Press. Vermeulen, Reiko 2008 Topics in Japanese. Ms. UCL. Wagner, Michael 2005 Givenness and Locality. Ms. MIT. Watanabe, Akira 2003 Wh and operator constructions in Japanese. Lingua 113:519–558. Yokoyama, Olga 1986 Discourse and Word Order. Amsterdam: John Benjamins. Zanuttini, Rafaella and Paul Portner 2003 Exclamative clauses: At the syntax-semantics interface Language 79:39–81. Zubizarreta, Maria Luisa 1998 Prosody, Focus and Word Order. Cambridge: MIT Press. Zwart, C. Jan-Wouter 1993 Dutch syntax: A minimalist approach. Ph. D. diss., University of Groningen.

Focus, topic, and word order: A compositional view Michael Wagner Abstract. This paper looks at the cartographic approach to word order restrictions relating to focus and topic. The observed word order patterns are more flexible than is expected under the assumption of a totally ordered hierarchy of functional projections, suggesting that this standard assumption of the cartographic approach results in an alltoo-Procrustean bed for syntactic typology. The observed patterns are as expected under a compositional view that tries to derive the observed word order patterns from the semantics of the focus operators involved. The evidence suggests that a cartography of natural languages has to take into account a broader range of syntactic configurations. A closer look at the compositional meaning of sentences can help understand existing constraints on word order variation.

1. Constituent order and cartography Rizzi (1997) explores the syntax of interrogative and relative pronouns, topics, and focused constituents. Three of the many incisive observations of this agendasetting paper can be summarized as follows: (1)

a. b. c.

Interrogative and relative pronouns, topics, and focused constituents all occur in the left periphery of clauses. Their order relative to each other is fixed. They differ with respect to iterability: There can be only one focus or relative pronoun, but there can be multiple topics.

The proposed solution, henceforth called the cartographic approach, involves breaking up the earlier functional ‘CP’ projection into a sequence of functional projections:

54

Michael Wagner

(2)

ForceP

Force

TopP∗

Top◦

FocP

Foc◦

TopP*

Top◦

FinP Fin◦

IP

This hierarchy plays a role in accounting for the three observations listed above in the following way: (3)

a.

b. c.

Certain constituents occur in the left periphery because they move to the specifier of the relevant functional projection. These movements are induced by criteria: wh-criterion, focus-criterion, topiccriterion, . . . Their relative order is fixed because the functional sequence is ordered into a universal hierarchy. Due to their interpretative properties topics but not foci can be nested recursively.

The core assumption of the cartographic perspective as it is outlined in Rizzi (1997) and Cinque (1999) is that the universal functional sequence is a totally ordered sequence of functional heads. What’s universal is their hierarchical order, but not every sentence may have every head in its functional spine in every language. Rizzi (1997) hypothesizes for example that focus and topic projections in the left periphery are activated only when needed, and whenever they are both present their relative order is fixed. The cartographic approach to constituent ordering can be contrasted with what I want to call the compositional approach, which holds that the relative order of constituents follows from their semantics and the compositional interaction of the pieces involved in building up sentence meaning. Up to a point, the two approaches are compatible. Once the meaning of the pieces has been identified, considerations of compositionality can provide substantive reasons for why the observed hierarchy is what it is. An example

Focus, topic, and word order: A compositional view

55

of such a rationalization of an observed generalization based on interpretative properties is exemplified by Rizzi (1997) ’s account of why a sentence can have more than one topic but not more than one focus (3c). We may, however, find ordering patterns that are incompatible with the assumption of a universally fixed total ordering. An example of ordering patterns that are paradoxical from the point of view of cartography but can be rationalized on semantic terms are those of certain adverbs in Norwegian and other languages discussed in Nilsen (2002), who proposes a polarity-based account as an alternative to a fixed hierachy. Similarly, Van Craenenbroeck (2006) observes transitivity failures in the ordering of complementizers, wh-phrases, and topics in Venetian Italian which are incompatible with standard assumptions about what are possible triggers for movement in the cartographic approach. This paper explores the grammar of overt and covert focus and topic operators. As already observed in Cinque (1999), the syntactic distribution of overt focus particles such as only and even poses a problem for a cartographic analysis. I will first discuss the issues that arise with these overt focus operators, and then argue that similar issues arise with unpronounced focus operators such as focus and topic. If the analogy is correct, this speaks against the view of postulating a fixed hierarchy of focus and topic projections in the left periphery as it was proposed in Rizzi (1997). The more general point that can be made based on the data discussed is that some word order patterns found across languages might not fit into the rigid theory of a universally fixed total ordering. Furthermore, a better understanding of the interpretative properties of the operators involved and how they compose can provide the restrictive forces that constrain syntactic patterns without having to stipulate their hierarchical order. This is not incompatible with the essence of the cartographic project, which is to investigate cross-linguistic similarities in word-order patterns, and which has proven a very insightful project; but it may require changing some of the assumptions about what kinds of generalizations we should be looking for.

2. Focus and givenness 2.1.

Overt Focus Operators

The focus operator only takes two arguments, a focus constituent (marked by underlining), and a proposition in which focused constituents are replaced with variables, sometimes called the presuppositional skeleton (marked in (4) by hooks):

56

(4)

Michael Wagner

Maria read only Moby Dick.

We can give only the following semantic entry (Wagner 2006a, adapting the analysis in Horn 1969; Fintel 1999, i.a.): (5)

Meaning of only only = λ p .λ f .∀a ∈ C p(a) → (∧ p(f ) →∧ p(a)) Presupposition: p(f ) Prejacent

The compositional view sets out to explain the syntactic distribution of only by virtue of its semantic ‘needs.’ In particular, only as defined in (5) needs two arguments, and must be in the following configuration in order to be interpreted: (6)

only at LF

Focus Constituent only

λx. p(x)

In other words, only takes the open property as its complement, and its focus must move to its specifier. While the compositional approach would hold that the syntactic configuration is a result of the semantic denotation of only, a cartographic approach could postulate a functional head for only somewhere along the functional spine, with a specifier (XP) and complement (YP), with similar consequences for the configuration at LF: (7)

OnlyP XP only◦

YP

This is actually not the analysis proposed for only in cartographic work. E.g., Cinque (1999) notes that the syntax of focus particles is different and more complicated than such an analysis would predict. As we will see, the reason for this is that focus operators occur in a second syntactic configuration in which they attach to the focus constituent (which could be of various kinds and sizes) and then take a propositional argument as their second argument – a configuration for which the cartographic approach does not provide any template. Focus operators like only and even appear to able to occur in both configurations, suggesting a rather flexible syntactic patterning.

Focus, topic, and word order: A compositional view

57

Even if (overtly pronounced) focus adverbs were excluded from a cartographic treatment by Cinque (1999) already, it is still worthwhile exploring them further in order to find out why they are not amenable to a cartographic treatment. These insights can then be used to evaluate whether the approach to unpronounced focus and topic operators that Rizzi (1997) proposes avoids these problems. A first prediction of the compositional approach to only in (5) is that only should attach to predicates, which could be sentence-sized in the case in which an argument of a sentence is focused and abstracted over. In other words, it should be able to occur in the left periphery. There is indeed some evidence that in English, VP-only can attach to a sentence-sized node, and that even though, in the surface word-order, VP-only follows the subject and precedes the VP, the subject actually starts out below, and its surface position is just a surface quirk of syntax. There is a generalization about the associate of only in English which is that it must be c-commanded by only. But the subject of a sentence including VP-only can be contained in the focus of only, as was observed in McCawley (1970, 296). An illustration from Wagner (2006a): (8)

They promised to stage Macbeth in its entirety. But then Macbeth only gave his soliloquy.  So the witches didn’t give their dialog.

The position of only in the cartographic approach would not be predicted from its content but established empirically. A prediction of both approaches is that, in cases where the associate of only is smaller than the entire sentence, this focus constituent should move to a position above the attachment site of only, in order to derive the correct configuration for interpreting only. Evidence for association by movement in English is discussed in Wagner (2006a). In languages such as Hungarian, this movement can happen overtly. It seems, however, that there is variation as to which of the two arguments an operator takes first. This would not pose a problem to the compositional approach, but would be a conundrum for the fixed hierarchy approach. The syntax of VP-only seems compatible with a theory that posits a functional only projection as proposed above. The argument that semantically consists of an open property is the complement of the functional head, and the focus constituent moves to its specifier. However, not every sentence involving only fits this pattern, and there seems to be another construal for only in English: (9)

a. b.

John only played baseball. John played only baseball.

58

Michael Wagner

As discussed in Cinque (1999, 31), one obvious difference that sets DP-only apart from sentential adverbs such as often is that it can directly follow the verb: (10)

*He forgot often his name.

This can be taken as evidence that only forms a constituent with the DP that follows it.1 In fact, there is evidence that [only + focus constituent] can move as a constituent, deriving the following configuration: (11)

[ Only baseball [ λx. john played x. ] ]

λx. p(x) only

Focus Constituent

Evidence for this kind of movement comes from the following ambiguity, observed by Taglicht (1984): (12)

They were advised to learn only Spanish. a. They were advised not to learn any other language than Spanish. b. They were not advised to learn any other language than Spanish.

One interpretation is that only Spanish undergoes covert movement. An alternative is that the surface structure actually is such that only Spanish attaches high and takes scope overtly, as in a rightward movement analysis, or in a Kaynian analysis involving movement to the left and subsequent remnant movement (Kayne 1998). There is also an analysis using an unorthodox constituent structure, where as long as linear order is observed the syntactic parse can be rebracketed on the surface, as in the categorial grammar analysis in Blaszczak and G¨artner (2005). Both overt and covert movement approaches correctly capture that the ambiguity disappears with VP-only: (13)

They were advised to only learn Spanish. a. They were advised not to learn any other language than Spanish. b. *They were not advised to learn any other language than Spanish.

So there are two ways to construct the same meaning using only. It is not clear how an analysis that treats only as a functional head can account for this flexibility. The syntactic pattern observed for DP-only is the reason that Cinque (1999) exempts focus-sensitive adverbs from the restrictions imposed by the

Focus, topic, and word order: A compositional view

59

functional hierarchy and proposes that they are ‘heads taking their modifees as complements’ (p. 31). This account remains moot, however, with respect to any structural restrictions on the occurrence of focus adverbs of this type, of which there are quite a number. If only can freely attach to its focus, then why is it that it is subject to more general restrictions on movement? Consider the following example: (14)

She claimed that only her dad knows. a. ‘She claimed that her dad and no one else knows.’ b. *‘The only claim she made is that her dad knows, but she did not claim that anyone else knows.’

Also, at least for the case of English, this account is not quite sufficient, since it leaves open the question of why there is also the possibility of VP-only, which shows the kind of distributional pattern that the functional sequence does try to account for. Furthermore, from the semantic point of view, it is unclear which of the two arguments to call the ‘modifee’ in the first place, so the characterization of the distribution of these adverbs leaves much to be desired. Ultimately, it seems a fundamental shortcoming of the treatment in Cinque (1999) to invoke different kinds of explanations for focus adverbs and other kinds of adverbs. One difference between focus adverbs and other sentence adverbs such as often is that the former but not the latter take two arguments rather than one, and this difference might account for the difference in syntactic construals. It seems that only can be pronounced attaching to either its first or second argument. The compositional approach might offer a solution for explaining the distribution of the two types of construals of focus adverbs such as only, since both configurations are compatible with the semantic needs of the operator. We could either stipulate an inherent ambiguity in the lexical entry of only with respect to which argument it has to combine first, or try to reduce one occurrence of only to the other by invoking syntactic movement. Maybe even VP-only is in fact interpreted in the configuration in (11). Suppose association with only involves movement to the complement position of only, an idea discussed in Lee (2005) based on evidence from Korean and Wagner (2006a) for English. This movement analysis of focus association can be made compatible with the apparent island-insensitivity of focus association (Anderson 1972) by the claim that a constituent containing the focus can serve as the syntactic focus constituent (Drubig 1994; Krifka 1996; Wagner 2006a):

60

(15)

Michael Wagner

I don’t know anyone who grows bananas, I only know a guy who smokes them.

only

a guy who smokes them

λx I know

x

Rather than invoking movement to complement position one could also postulate two different entries for only, differing in the order in which they combine with the two arguments. For more discussion of syntactic restrictions on the distribution of only see also Jacob (1983), Bayer (1996) and Jaeger and Wagner (2003). One conclusion that we can draw based on the observed distribution of only is that it does not lend itself easily to a cartographic analysis involving a fixed hierarchy. The compositional view, on the other hand, provides a rationalization of the distribution.2 It is important to note that even in the compositional approach the particular syntactic configuration does not follow from the overall meaning, but rather it follows from assumptions about how that meaning is carved up into pieces and the order in which the pieces take their arguments. Natural language often picks out one out of many conceivable ways of divvying up complex meaning, and in the case of only any approach might have to stipulate the order in which the two arguments are taken. If all languages use the same decomposition, then we should indeed expect to be able to fit all languages into some sort of universal template. However, languages might simply differ in how they carve up complex meanings, in which case we might not find a consistent syntactic pattern across languages.

2.2.

Covert Focus operators

Covert focus operators negotiate the encoding of focus and givenness in English and other languages. Rizzi (1997) postulates functional Focus and Topic projections with a concrete semantic import in ways that mirror the configuration observed in the case of only in (6). The Focus Projection marks its specifier XP as the focus and its complement YP as the presupposition:

Focus, topic, and word order: A compositional view

(16)

61

Focus Projection FocP XP Foc◦

YP

We can mimic the cartographic postulation of a functional focus projection by postulating the following unpronounced focus operator, in analogy to only. Note that the antecedent for focus marking does not to be a true proposition and be part of the context set C. In fact, it is neither sufficient nor necessary for the antecedent to be true. Rather, it must be salient in the discourse. I assume that there is a set  which contains all constituents and entities salient in the discourse, and FOCUS introduces a presupposition about there being an antecedent in  (for more discussion Wagner (2009a).3 (17)

∀σ: FOCUS = λp . λxσ . ∃a ∈ ALT(x) ∩  ∧ p(x) →∧ p(a)).p(x)

The LF-configuration for the interpretation of sentences involving an unpronounced focus operator would then appear as follows: (18)

FOCUS at LF

Focus Constituent F OCUS

λx. p(x)

This two-place operator has the result that focus is interpreted in a configurational way, such that the second argument of the operator is the focus constituent and the first argument is the domain of the focus, not unlike the predictions of the functional projection in (16). It might be impossible to test whether this operator has the same flexibility, attaching to either the first or the second argument leaving the meaning of the result constant, as we observed for only. Since the meaning would be identical, there is no way to tell unless we identify some phonological reflex of the focus operator on the constituent it attaches to. However, the operator as formulated here may be too restrictive in another way. As we will see, this operator can applied very flexibly. We can even swap which of two sisters plays the role of first or second argument of the operator. This is also possible with only, but in a much more restricted way since only operates on propositions. Still, one can attach only to the subject or the VP,

62

Michael Wagner

leading to different truth conditions: Bill only plays baseball, vs. Bill plays only baseball. The FOCUS operator can apply to any node, taking either sister as its focused or given argument. This flexibility would be entirely unexpected if it was a associated with a functional head in a fixed hierarchy. An indication of the flexibility of this operator can be observed by looking at focus effects within a DP (cf. Rooth 1996), suggesting that maybe it need not always take a proposition-sized argument. I will henceforth mark accented words in small-caps, and prosodically subordinated material with underlining. (19)

A CANADIAN farmer met an AMERICAN farmer on the STREET.

This utterance could have wide focus or focus on street, and yet embedded within the DP there is an additional focus marking. So focus marking is even possible between sister constituents within DPs, or with respect to any other type of constituent. Moreover, one can swap what’s given and what’s focused, so at any given node either sister could be marked as given relative to the other. This reveals a flexibility that is unexpected if focus resides in a particular functional projection. The additional flexibility of the focus operator can be captured by the following semantic entry for a covert focus-sensitive operator, which operates on constituents of any type:4 (20)

∀σ, ∀δ FOCUS = λxσ .λyδ .∃ay ∈ (∩ALT(xy)) ay = xy.xy

This is a 2-place operator, and it introduces a presupposition that requires there to be a salient antecedent of a certain shape, just like the focus operator introduced in Rooth (1992). The only difference is that in this case it takes two arguments. Whenever this operator is used it takes two arguments, one constituent that is marked as given (which one could call the ‘background’ in Rizzi’s terminology), and one constituent that evokes alternatives, which we could call the ‘focus.’ The FOCUS operator has an effect on prosody: if the given constituent follows the focus, it is prosodically subordinated, which is at least reflected by a pitchrange reduction and can result in complete deaccenting Wagner (2005). I will not discuss why FOCUS has this effect. This operator predicts that it is not enough for a constituent to have been used before in order for it to be marked as given by FOCUS. In addition, there must be an alternative to the first argument, the focus, such that there is an antecedent involving that alternative. Conversely, it is not enough for a constituent to be focused; it must be focused relative to something else (although that something else can sometimes be elided). The intuition behind this can be brought out easily

Focus, topic, and word order: A compositional view

63

in the following type of example, a context in which the predicate convertible is given, and yet deaccenting it is dispreferred Wagner (2006b): (21)

A: Mary’s uncle, who produces high-end convertibles, is coming to her wedding. I wonder what he brought as a present. a. B: He brought a [CHEAP convertible]. b. #B: He brought [a RED convertible]. c. B: He brought [a red CONVERTIBLE]

Convertible is destressed, but the stress that falls on red is ‘loaded,’ in that it invokes alternatives. Red, however, is not an alternative to cheap, as can be illustrated by other instances where red invokes alternatives: (22)

high-end is not an alternative to red: Mary only likes RED convertibles.

Deaccenting convertible is only possible if there is a salient alternative Y to the sister of convertible such that [y convertible] is given in the context. Constituents are marked as given relative to another constituent: its sister. This analysis is similar to the one proposed in Williams (1997) in that it holds that every instance of deaccentuation involves marking a constituent anaphoric and another constituent dis-anaphoric. Williams (1997, 599) concludes: “The lesson is that it is impossible to destress one thing without stressing another, and the stress that falls on the other is loaded, not empty.” It differs, however, from Williams’s proposal in requiring that the corresponding constituent in the antecedent must be a true alternative. Sometimes, however, there does not seem to be any sister effect, and a constituent can be marked as given without a sense that alternatives for the sister constituent are evoked: (23)

A: Smith walked into a store. What happened next? a. #B: A detective arrested Smith. b. B: A detective arrested Sm ith.

Here it seems that the presupposition of FOCUS is too strong, since no alternative to arrested is necessarily invoked. That direct objects can be marked as given even with sentence-wide focus was already observed in Halliday (1967) and Ladd (1980). Some earlier accounts have interpreted these observations as evidence that apart from focus effects, there is a second and separate phenomenon of anaphoric destressing (Neeleman and Reinhart 1998; Reinhart 2006), which applies only

64

Michael Wagner

to certain types of arguments, such as DPs. Similarly, Selkirk (1995) proposed a focus projection mechanism that provided different conditions on givenness and focus. But not all given DPs can actually be marked as given that freely. Consider the following observation: (24)

A: Smith was suspected to have been involved in the burglary. Do you know what happened in the end? a. B: I’m not sure. The thing i heard is they were going to arrest JO NES or SMITH. b. ??B: . . . to arrest Jones O R Smith. c. #B: . . . to arrest J O N E S or Smith.

Marking Smith as given in a coordinate structure introduces a presupposition stronger than just that there is an antecedent for Smith in the context, in fact, it introduces precisely the presupposition of FOCUS. Consider a case where it can destress: (25)

A: Did they arrest only Smith? B: No, they arrested Jones AND Smith. (given constituent: only Smith)

Or consider the following example. Sentence (26) requires a contrasting individual for John, such that the entire constituent x or Smith is given: (26)

A: They want to arrest Lee or Smith? B: No, they want to arrest JONES or Smith. (Constituent given in context: Lee or Smith)

Why is it that in these cases a DP cannot be destressed just by virtue of referring to a discourse-given entity when they occur in a coordinate structure? The solution proposed by Wagner (2005) is that Smith is not actually the sister of the predicate. The structure based on which givenness is evaluated is rather the following: (27)

[ FOCUS([ λ x. [ The police arrested x. ] ]) Smith ]

This movement changes the sister relations and consequently weakens the presupposition introduced by FOCUS. Marking Smith as given relative to the lambda-abstracted proposition invokes the presupposition that there is an alternative open property that applies to Smith that is salient. This is precisely the weak presupposition that destressing a direct object intuitively encodes. Movement in the present example facilitates givenness marking: givenness marking in situ would not have been possible, since the presupposition introduced would

Focus, topic, and word order: A compositional view

65

have been too strong. And this movement is impossible out of the coordinate structure in (24) due to the coordinate structure constraint. The important lesson is that the presupposition introduced depends on where the operator attaches. It may be possible to postulate a flurry of focus positions at various points within DPs and CPs and account for the facts, but I think such an approach would miss the intrinsically relational nature of focus and givenness marking that the data suggests. Moving a constituent changes the sister relations, and as a consequence it changes the presupposition introduced by marking a constituent as given or focused. This relational approach makes interesting predictions for focus- and givenness movement that differ from the ones expected in a cartographic approach. Movement changes the syntactic configuration, and thus changes what can be marked as given relative to which other constituent. We observed that the moved constituent can be marked as given relative to the remnant, and the focus presupposition can thus be weakened. (28)

FOCUS([ λ x. [ The police ARRESTED x. ] ]) Smith

Suppose that the moved constituent can either be the first or second argument of FOCUS, in other words, that the moved constituent can either be the one that is marked as given or the one that its sister is marked as given relative to, as proposed in Wagner (2005). Consider for the example what would happen if we attached the FOCUS-operator to the moved constituent rather than the remnant: (29)

[λ x. [ The police arrested x. ] ] FOCUS(SMITH).

The predicted presupposition of (29) is: There is an alternative y to Smith such that The police arrested x is given in the context. A context in which this presupposition is fulfilled is the following: (30)

A: Who did the police arrest? B: The police arrested SMITH.

In other words, we can use the same givenness operator to analyze cases of givenness marking and cases of ‘narrow focus.’ There may be some evidence that such restructuring and focus marking is indeed possible even in surface structure. (cf. Rooth 2005) notes the following contrast: (31)

a. b.

A: What did Mary eat? B: I think Mary ate | the APPLE. A: What did Mary do? #? B: I think Mary ate | the APPLE.

66

Michael Wagner

According to Rooth (2005), it is possible to separate the direct object from the verb by a break, arguably a reflex of overt restructuring along the lines of (29), but only if the strong presupposition encoded by marking the entire remnant constituent as given, as in (b). It is not sufficient that Mary constitutes given information, as in (a). The same pattern holds for overt focus movement to the left in German, providing more evidence that focus/givenness movement is indeed restricted in the way that would be expected based on the presupposition encoded by FOCUS:5 (32)

a.

A: What did Mary eat? B: Ich glaube den APFEL I think the apple I think Mary ate the apple. b. A: What did Mary do? ?# B: Ich glaube den APFEL I think the apple

hat Maria gegessen. has Maria eaten

hat Maria gegessen. has Maria eaten

Moving the constituent the apple and subsequently marking the sister constituent as given results in the presupposition that there is an alternative x to Apfel such that Maria has eaten x.Alternatives of this sort are made available by the question in (32a) but not by the one in (32b), which does not include the predicate eat. Of course, in German, just as in English, it is also possible to move the constituent the apple and to mark it as given relative to the remnant. This is a case of givenness-related scrambling, which is in fact very common in German:6 (33)

A: What about the apple? B: Ich glaube den APFEL hat Maria GEGESSEN. I think the apple has Maria eaten

The compositional theory can thus rationalize givenness and focus movement: it adjusts the presuppositions encoded by marking one constituent given relative to its sister. This view of the interaction of movement and focus marking is intrinsically flexible: it is sister constituents that are marked as given relative to each other, and movement adjusts the syntactic relations of constituents.7 The cartographic view, on the other hand, fixes the syntactic configuration of focus marking by stipulating several focus projections in the spine of functional projections. This approach does not do justice to the relational nature of information structuring observed in the data discussed here. More evidence for the intrinsically relational nature of focus- and givenness-marking was discussed in Neeleman and van de Koot (2007).

Focus, topic, and word order: A compositional view

67

3. Recursive nesting of Focus operators A sentence can include more than one focus operator. In such situations, the two operators can take either scope with respect to each other. This presents a problem for the view that certain focus operators occupy a fixed position in a hierarchy. Again, first I will first illustrate the point with overt focus operators, and then extend the discussion to unpronounced operators.

3.1.

Nesting of overt operators

Two overt focus operators can occur in the same sentence Krifka (1992). I will use the focus operator even in addition to only to illustrate the point: (34)

Even John read Moby Dick. a. Asserted: John read Moby Dick. b. Conventional Implicature of even (Karttunen and Peters 1979, 25/ 26): There are other x under consideration besides John such that x read Moby Dick and for all x under consideration besides John, the likelihood of x reading Moby Dick is greater than or equal to the likelihood of John reading Moby Dick.

Here’s an example from (Wagner 2009b) with both focus operators in one sentence (I’m using underlining here to mark the focus of the operators): (35)

even > only: Except for Bill, the kids in this summer camp have no respect for animals and the potential dangers, which makes them take too many risks, including with poisonous snakes. a. Even the most poisonous snake only frightens Bill. b. Even the most poisonous snake frightens only Bill.

In this example, even outscopes only. The alternative propositions computed for even must include only, since otherwise the probabilities would not line up in the right way. Only Bill must be part of every alternative considered. It is clearly more likely that a more poisonous snake frightens Bill, but it is less likely that it would frighten only Bill: (36)

a. b. c.

least likely: The most poisonous snake frightens only Bill. more likely: Average poisonous snakes frighten only Bill. even more likely: Mildly poisonous snakes frighten only Bill.

The representation of the sentence at LF looks as follows:

68

(37)

Michael Wagner

λP1 . even ([the most poisonous snake])(P1) ([λx.( only (Bill) (λy. x frightens y)]).

In (35), word order matches the scope. The word order, however, can also be reversed: (38)

even > only: Except for Bill, the kids in this summer camp have no respect for animals and the potential dangers, which makes them take too many risks, including with poisonous snakes. a. Only Bill is afraid of even the most poisonous snake. b. #Only Bill is even afraid of the most poisonous snake.

When the focus operator attaches to the focus constituent, wide scope is possible. The analysis involves moving the focus operator together with the DP, i.e., even the most poisonous snake, to take the right scope. The fact that (38b) is bad constitutes evidence that movement is indeed involved. It is impossible to move even in (38b) such that it can outscope only. What’s puzzling about these data is that there is compelling evidence that there is an NPI version of even with a reversed presupposition requiring lowest probability Rullmann (1997). So why is movement necessary here? I think the reason may be that the NPI-version of even is not licensed by only, similar to other strong NPIs such as in weeks or the punctual reading of until (Zwarts 1998; Gajewski 2005). Compare the following example with (38b): (39)

No one is even afraid of the most poisonous snake.

Negation happily licenses the NPI-version of even and the sentence becomes acceptable. Note that if this was in fact the standard version of even, movement would be necessary here to yield the right truth condition, but that should be ruled out for the same reasons as in (38b). While the examples so far require that even outscope only, which in cartographic terms could be achieved by positing an even projection oustcoping an only projection, the inverse scope is also possible. Consider the following context: (40)

Context II: only > even: The kids in the summer camp are afraid of snakes to some degree, but it depends on how dangerous they are. Everyone is afraid of rattlesnakes, since they’re really poisonous, but almost everyone is ok with some less poisonous snake. a. Only Bill is afraid of even the least poisonous snake. b. Only Bill is even afraid of the least poisonous snake.

Focus, topic, and word order: A compositional view

69

In this context, only must outscope even for the probabilities to come out right. Once again, the reverse linear order is also possible, at least for some of my informants: (41)

Context II: only > even ? Even the least poisonous snake would frighten only Bill. # Even the least poisonous snake would only frighten Bill.

The reason these data raise a problem if one were to analyze it in cartographic terms is the following: suppose that there was an even and an only projection, then one should be higher in the hierarchy than the other. The fact that two different scopes are possible would have to be resolved by postulating multiple only and even projections, analogous to the multiple projections for topic and focus in (2) based on the observation that they can have different hierarchical arrangements. This is a possible but not very insightful move, and again does not do justice to the compositional and relational nature of the facts. The compositional view on the other hand can account for the data by virtue of the fact that adding one operator leads to an expression that is compatible with adding the other, and it is simply the syntactic scope that the two operators take with respect to each other that will decide on the ultimate truth conditions. This is not unlike the explanation in Rizzi (1997) for the iterability of topics, but in fact it does away with the need to postulate a place for either even or only in the functional hierarchy. Once again, we saw evidence that the syntax of overt focus operators such as only and even cannot be accounted for by the functional sequence – but that does not come as a surprise. As discussed before, it was already acknowledged in Cinque (1999) that focus operators are different. The compositional view might provide a better grip on the word order patterns observed for this class of adverb. It turns out, however, that an analogous kind of argument can be made for nested covert focus operators, in particular for the kind of topic-focus constructions that Rizzi (1997) discusses.

3.2.

Nested covert Focus operators

The notion of topic discussed in Rizzi (1997) and referred to in the functional hierarchy in (2) seems to be that of what is usually called a ‘contrastive topic.’ This becomes evident in the examples used as an illustration of topics, (e.g., Rizzi 1997, 285): (42)

Your book, you should give t to Paul (not to Bill).

70

Michael Wagner

The contrastive topic, your book, invokes alternatives, and constituents that generally do not allow for the invocation of alternatives are prohibited: (43) ?*It, you should give t to Paul. B¨uring (1997, 2003) observes that both the contrastive topic (CT) and the focus (FOC) of the sentence invoke alternatives. An example of a contrastive topic is the following:8 (44)

A: What did you buy on 59th street? B: On /FIFTYNINTH Street∨, I bought the SHOES\.

The answer in this dialog addresses the question under discussion, and the constituent shoes constitutes the focus. In addition to answering the question under discussion, this utterance also invokes a set of alternative questions about what the addressee bought at other locations. Invoking a set of a alternative questions is a typical use of contrastive topics. Interestingly, in English the contrastive topic can also follow the focus: (45)

A: What did you buy on 59th street? B: I bought the SHOES on /FiftyNINTH Street∨.

The analysis in B¨uring (1997, 2003) consists of marking one constituent as a contrastive topic using the diacritic CT and the other as a focus using FOC. A non-compositional two-step process then creates a topic-semantic value for the expression. The prediction is that there should be no ordering restrictions between contrastive topic and focus. Wagner (2009b), however, argues that there is a simpler, compositional way of accounting for contrastive topics, which does not involve positing two different features CT and FOC. The idea is that contrastive topics simply involve a recursive nesting of two focus operators, similar to the nesting of overt focus operators discussed in the previous section. The constituent that associates with the focus operator taking wider scope is what we call a contrastive topic. Compositional views of contrastive topics similar in spirit were discussed in Williams (1997); Van Hoof (2003); Sauerland (2005). I will not discuss the various parallels and differences between the analyses here. The basic idea put forward in Wagner (in press) is that a sentence involving a contrastive topic involves the same syntactic configuration as the one involved in a sentence containing both only and even: two focus operators each takes a focus constituent as its complement that then takes scope relative to the other. If this parallel is correct, then rather than deriving the syntactic distribution of topic and focus based on the functional sequence, we can simply derive it from

Focus, topic, and word order: A compositional view

71

the compositional structure. A semantic analysis of contrastive topic in those compositional terms is presented in Wagner (2009b). Suppose that contrastive topics are really just the associates of focus operators taking wide scope over another focus operator. Then we expect them to have an identical syntactic distribution to parallel cases with overt focus operators. We already saw that inverse scope between focus operators in English is possible, so it is no surprise that contrastive topic and focus can occur in either word order. But I would now like to turn to Italian, where we will see that (i) scope between focus particles is fixed at surface structure, and (ii) contrastive topics must precede foci. This is exactly as expected if covert focus particles are syntactically construed just like overt focus particles. Rizzi (1997, 289/290) observes that contrastive topics differ from foci in various ways. They are left-dislocated, and precede foci in general. Another difference, already noted in in Cinque (1990, 14, 56ff), is that contrastive topics can and sometimes must involve resumptive clitics while left-dislocated foci cannot. A clitic is obligatory in the case of direct objects: (46)

a.

Il tuo libro, lo ho comprato. the your book, it have.I bought ‘Your book, I bought.’ b. *Il tuo libro, ho comprato. the your book, have.I bought

Using a resumptive clitic in the case of foci, however, is reported to be impossible: (47)

a. *IL TUO LIBRO lo ho comprato (non il suo). the your book it have.I bought (not the his) b. IL TUO LIBRO ho comprato (non is suo). the your book have.I bought (not the his) ‘I bought your book, not his.’

This observation was used as evidence that contrastive topics and foci are substantively different. Within the cartographic approach, one involves movement to a topic projection and one involves movement to a focus projection, two different and hierarchically ordered projections in the functional spine. Under the compositional view, however, contrastive topics involve focus operators that simply outscope a lower focus operator. We would expect then that the clitic data reported here would generalize to examples with overt focus operators. This is indeed correct. If the direct object associates with a focus

72

Michael Wagner

operator that outscopes a focus operator associating with a higher argument, e.g., the subject, then clitic-left-dislocation is obligatory:9 (48)

There is a petting zoo, and for some reason they put some poisonous snakes in there, but people didn’t really want to pet them. In fact, most of of the snakes were not petted at all, except for one: the least poisonous snake. And even that got petted by only one visitor. So: a. *Soltanto uno dei visitatori (lo) ha toccato anche il only one of.the visitors (it) has touched even the meno velenoso dei serpenti. least poisonous of.the snakes ‘Only one of the visitors touched even the least poisonous snakes.’ b. Anche il meno velenoso dei serpenti, *(lo) ha even the least poisonous of.the snakes it has toccato soltanto uno dei visitatori. touched only one of.the visitors ‘Only one of the visitors touched even the least poisonous snakes.’

The reason for the linear order of constituents lies in their relative semantic scope, rather than in an inherently specified position in a fixed hierarchy. In Italian, but not in English, the relative order between them is fixed, and inverse scope is not possible. Arguments other than direct objects show similar restrictions with respect to linear order, but behave differently with respect to clitics. The surface word order must reflect the scope of the focus operators. The examples corresponding to the English ones discussed above in the context that motivates the reading in which only outscopes even are as follows: (49)

only > even a. Solo Paolo only Paolo velenosi. poisonous b. *Anche il even the Paolo. Paolo

ha paura anche dei serpenti meno has fear even of.the snakes least

serpente meno velenoso fa paura solo a snakes least poisonous make fear only to

If the intended meaning with only outscoping even is to be expressed using the predicate fa paura ‘make fear’ in (49b), the only way is to use a left-dislocated construction:

Focus, topic, and word order: A compositional view

(50)

73

only > even Solo a Paolo anche il serpente meno velenoso (gli) fa only to Paolo even the snakes least poisonous make fear paura.

If we switch to the context that motivates the reading in which even outscopes only, the judgments go exactly the other way, and only the sentence in which is the basic word order reflects the scope is grammatical: (51)

even > only a. *Solo Paolo only Paolo velenosi. poisonous b. Anche il even the Paolo. Paolo

ha paura anche dei serpenti piu’ has fear even of.the snakes most

serpente piu’ velenoso fa paura solo a snakes most poisonous make fear only to

If the verb ha paura ‘has fear’ is to be used, again the only way is to get the right scope is by left dislocation: (52)

even > only Anche dei serpenti piu’ velenosi, solo Paolo ha paura. even of.the snakes most poisonous, only Paolo has fear

The interesting difference between Italian and English is that overt focus operator take surface scope in Italian but not in English. We also saw that contrastive topics in English can either precede or follow the focus of a sentence. If overt and covert focus operators behave alike, we would then expect that the order between topics and foci is also fixed in Italian. This is indeed correct: (53)

A: Cosa hai comprato sulla cinquantanovesima strada? what have.2nd bought on-the 59th street? ‘What did you buy on 59th street?’

74

Michael Wagner

B: Sulla cinquantanovesima strada ho on-the 59th street have.1st scarpe. E sulla cinquantaduesima strada shoes. And on-the 52th street la giacca. the jacket.

comprato le bought the ho comprato have.1st bought

‘On 59th street, I bought the shoes. And on 52nd street I bought the new jacket.’ Inverting the two focus constituents is dispreferred in this context: (54)

A: Cosa hai comprato sulla cinquantanovesima strada? what have.2nd bought on-the 59th street? ‘What did you buy on 59th street?’ B: ?? Le scarpe, lo ho comprato the shoes them have.1st bought cinquantanovesima strada. E la giacca, 59th street. And the jacket comprato sulla cinquantaduesima strada. bought on-the 52th street

sulla on-the la ho it have.1st

‘I bought the shoes on 59th street. And I bought the new jacket on 52nd Street.’ This is converging evidence that overt and covert focus operators behave alike, and since the former are not amenable to a cartographic analysis, the same reasoning applies to covert ones.

3.3. Why is there a Clitic in Topicalization but not in Focus? Focus Operators in Italian take overt scope. In order to get a focus operator to take scope over a preceding argument, left-dislocation is necessary to change the word order. In the case of direct objects, the left-dislocated constituent must be resumed by a clitic. But why would it be that clitic resumption is necessary? The reason might be related to a restriction noted in Calabrese (1984), and discussed, more recently, in Stoyaanova (2008). Pair-list wh-questions are not grammatical in Italian. Pair-list-questions could be analyzed as questions involving two nested focus operators, each associating with a wh-word. Perhaps

Focus, topic, and word order: A compositional view

75

in Italian, there is a general constraint against having more than one focus operator within a clausal domain: (55)

Focus constraint on Italian: Each clause can contain only one focus operator.

If this is correct, and if contrastive topics involve nesting focus operators, then contrastive topics as we know them from English should actually be impossible in Italian. However, we might be able to add a focus/topic to a clause that already contains one if we add it outside of that clause. Take an analogous construction in English: (56)

a. As for John, Mary really likes him. b. *As for John, Mary really likes.

as for introduces a topic, but it’s clearly not part of the main clause itself, but seems to attach higher, perhaps taking a speech act as its argument. Similar to CLLD in Italian, we have to resume the direct object with a pronoun. The reason the pronoun is obligatory is because it’s an obligatory argument. Consider: (57)

As for Monday, we’ll go shopping (then).

If a left-dislocated constituent in Italian is indeed construed with an as for-like construction, this can explain various of the peculiar properties of CLLD. First of all, we expect obligatory clitics for all obligatory arguments (except the subject, since Italian is a pro-drop language). But other peculiarities can be explained as well. Rizzi (1997, 290–291) and Cinque (1990, 57–60) observe, for example, that certain constituents that Rizzi (1997) calls ‘bare quantificational elements like nessuno ‘no one’ and tutto ‘tutto’ cannot function as contrastive topics but can function as foci. But more generally, these elements cannot be pronominalized: (58)

a.

A: You saw no one? *B: Si, lo ho visto. Yes, it have.I seen. b. A: You did everything? ??B: Si, lo ho fatto. yes, it have.I done

The example in (58b) can only be used when ‘lo’ refers to an entire set of actions that are salient in the context. This is the same condition that allows for topicalization of everything Cinque (1990).

76

Michael Wagner

Negative quantifiers are also unacceptable in as for clauses, which also require a referential expression in order be able to link up to the main clause: (59)

*As for no one, I don’t like him.

The compositional analysis also sheds light on the question why it appears as if there can be more than one topic but only one focus. It’s simply because there is only one focus operator in a clause, and that’s what we call the focus. All higher ones must resort to a strategy that adds them outside of the main clause, but that can be done iteratively stacking more than one.

4. Another alternative: the templatic approach Neeleman and van de Koot (2007) discuss the relative order between focus and topic, and, similar argument in this paper, they present evidence that word order is more flexible than expected under the cartographic account of Rizzi (1997). The alternative explanation for the observed flexibility and also several restrictions proposed in Neeleman and van de Koot (2007) involves the postulation of so-called ‘discourse templates’ that are used to evaluate the well-formedness of certain syntactic configurations. I will henceforth call this the ‘templatic approach.’ The main difference from the functional hierarchy proposed in Rizzi (1997) – also a kind of discourse template – is that focus and topic can occur in multiple configurations, and only certain ones are ruled out. It can thus be characterized as a version of the cartographic approach with a more flexible hierarchy compared to the strict total ordering underlying the proposal in Rizzi (1997). In the following I will look at the particular discourse templates proposed in Neeleman and van de Koot (2007) and argue that they are based on incorrect assumptions about the actual distribution of topic and focus. A caveat should be made. The discussion here will only address aspects of Neeleman and van de Koot (2007) concerning the relative order of topic and focus; it does not address the substantial part of the paper on the linear order of discourse given constituents. It will hence not do justice to the full range of arguments put forward for the templatic approach in this work. Discourse templates have the effect that certain parts of the syntactic structure are mapped to either background or comment:

Focus, topic, and word order: A compositional view

(60)

77

Discourse Templates a.

b.

The treatment of focus is similar to Rizzi (1997)’s proposal that the complement of focus projections is part of the background of an utterance; it shares with the compositional approach in Wagner (2005) in that it is generally the sister of the constituent that evokes alternatives that is taken to be part of the background, i.e., part of an antecedent in the discourse. While in the templatic and cartographic approach this is a stipulated for the template or functional sequence respectively, in the compositional approach this is a result of the semantic denotation of the 2-place focus operator. The templatic approach is also similar to the compositional approach in Wagner (2005) in that the size of the background can be adjusted flexibly by moving the alternatives-invoking constituent more or less high in the phrase marker, which is not the case in the cartographic approach, where that position is fixed by the functional sequence. Once again, it is important to point out that, similar to the cartographic approach based on the functional sequence, the templatic approach is not in principle incompatible with a compositional semantic analysis. However, the actual proposal in Neeleman and van de Koot (2007) does not provide such a compositional theory, and in fact crucially assumes a non-compositional treatment of contrastive topics since focus and topic can stand in very different syntactic relations with no interpretive effect. One main prediction of the templatic approach is summarized in Neeleman and van de Koot (2007) as follows: (61)

Predictions of the Templatic Approach “The full range of predictions, then, is as follows. As long as we are dealing with in-situ topics and foci, their relative order is free. However, things are different when movement comes into play. While a topic can move across a focused constituent (whether in situ or not), a focused constituent cannot move across a topic (whether in situ or not).”

The starting point for this characterization of focus/topic distribution is the observation by Jackendoff (1972, 261) that contrastive topics can either precede or follow a focus in English, as discussed above:

78

(62)

Michael Wagner

a.

b.

Contrastive Topic ≺ Focus: A: Well, what about Fred, what did he eat? B: /FRED∨ate the beans. Focus ≺ Topic: A: What about the beans? Who ate them? B: FRED ate the /BEANS∨.

In the discourse-templatic approach, the contrastive topic can be analyzed here as being generated in situ, so this is just as expected. However, we saw above that languages that do not allow focus operators to take inverse scope cannot actually have contrastive topics following a focus. In fact, Dutch, the language Neeleman and van de Koot (2007) mostly focus on, patterns with Italian in this regard: in contrast to English, in Dutch it is impossible for contrastive topics to be ‘base-generated’ below foci and pronounced after them. As before, the context that we can use to control what’s the topic and what’s the focus is the one that B¨uring (1997) calls contrastive aboutness-topics. These are utterances that answer the question under discussion but in addition evoke a set of alternative questions. It turns out that in this kind of context, it is impossible in Dutch to let an accented constituent that evokes the alternatives for the topic set follow the focus of the question: (63)

A: Wat heb je gekocht op de Nieuwstraat? what have you bought on the Niewstraat? a. B: Op de NIEUWSTRAAT heb ik SCHOENEN gekocht. on the Nieuwstraat have I shoes bought. b. #B: Ik heb schoenen op de NIEUWSTRAAT gekocht. i have shoes on the Nieuwstraat bought

The only way to pronounce (63b) such that it is felicitous is to deaccent Nieuwstraat, in which case it is marked as given and is no longer acting as a contrastive topic. The templatic approach thus fails to explain why topics must precede foci in Dutch but not in English.10 The compositional approach explains the observed restriction straightforwardly: contrastive topics are constituents that associate with a focus operator that takes wide scope over another focus operator. Dutch belongs to the class of languages in which LF-movement of a focus operator plus its focus constituent is impossible, hence the word order restriction. The reason why a focus cannot move across a contrastive topic on the compositional view is simply that this would lead to the wrong interpretation, since it would mean that the ‘focus’ takes wide scope and ends up being the contrastive topic.

Focus, topic, and word order: A compositional view

79

There are a number of restrictions that Neeleman and van de Koot (2007) attribute to templatic restrictions that might receive a more insightful explanation in the compositional approach. For example, it seems to be impossible to move a constituent [only x] across another focus: (64)

A: Wie lezen er heden ten dage eigenlijk nog dichters? a. B: Piet leest veel dichters, maar ik geloof dat Fred Piet reads many poets, but I believe that Fred alleen Bloem leest. only Bloem reads ‘Piet reads many poets, but I believe that Fred reads only Bloem.’ b. *B: Piet leest veel dichters, maar ik geloof dat alleen Piet reads many poets, but I believe that only Bloem Fred leest. Bloem Fred reads

Under the compositional view, the focus operator attaching to alleen Bloem would take scope over the one attached to Fred, which would result in a reading in which alleen Bloem as a whole is a contrastive topic. However, as observed by Rizzi(1997), contrastive topics do not like to be ‘non-referential,’ and elements such as negative quantifiers are ruled out as contrastive topics. Clear topic contexts also reject constituents such as only x, including constructions that do not involve any movement at all: (65)

*As for only Bloem, Fred reads him.

Thus the compositional view also provides an explanation for this kind of restriction on word order, and in fact one that is linked to independently motivated restrictions on contrastive topics.

5. Conclusion This paper looked at the cartographic approach to word order restrictions relating to focus and topic, and compared it with an alternative compositional approach that tries to derive certain word order restrictions from the way the individual pieces are assembled to derive complex meaning. The observed word order patterns were argued to be incompatible with the strong assumption of a totally ordered universal functional spine, but were as expected under the compositional view.

80

Michael Wagner

What can we learn from the case of focus particles about the cartographic project? One important point that was also raised in the insightful discussion after Peter Svenonius’s talk at the cartography conference in Brussels is that it is not possible to derive word order from the overall meaning of a sentence, since the tools of semantics offer all sorts of ways to decompose complex meaning into separate pieces. So even if we embrace a more compositional approach, we have to first find out in what particular way languages break down complex meaning. The construction of expressions with identical truth conditions might be fundamentally different across languages, depending on their particular decomposition of meaning into pieces. This means that the very rigid expectation of an universal template for all languages such as the functional sequence may form an all-too-Procrustean bed for syntactic structure. Ultimately, however, this insight should simply lead to a more refined and semantically more informed set of questions compared with the ones already under investigation in the cartographic approach. The project of mapping out the composition of sentence structure across languages remains essential. Acknowledgements: Thanks to the audience at BCGL and to Jeroen van Craenenbroeck, Gisbert Fanselow, Jon Gajewski, Hotze Rullman and two very insightful reviewers for helpful comments.

Notes 1.

2.

3. 4.

B¨uring and Hartmann (2001) argue that at least in German, apparent DP-only in fact attaches to C. This analysis has not, however, been extended to English, and I think there is reason to be skeptical also for the case of German. A puzzle for the compositional view should be noted, however. The syntactic distribution of only and even differs in subtle ways, and it is not clear whether these differences can be reduced to their semantics. Jackendoff (1972, 251) observes that even but not only can associate with the subject across an intervening auxiliary (example adapted): (i.) a. John will even have given his daughter a new bycicle. b. *John will only have given his daughter a new bycicle. Note that the notion of alternative set ALT still needs to be defined. See Wagner (2005) for discussion and Wagner (2009a) for a formal definition. Cf. the ‘relative givenness’ operator in (Wagner 2006b). To make it more parallel to the entry of other focus operators such as only, the operator here takes the alternativeevoking argument first.

Focus, topic, and word order: A compositional view 5.

81

The example here contrasts with the case of a matrix clause, where fronting the direct object to the first position and deaccenting the rest is perfectly acceptable (Fanselow and Lenertov´a 2006, and references therein): (i.) A: What did Maria do? B: DEN APFEL hat Maria gegessen. the apple has Maria eaten ‘She ate the apple.’ I think the movement here adds a speech-act oriented exclamative meaning to this sentence. Alternatively, one can also insert an expletive here and deaccent the entire sentence, again with an exclamative impact, albeit a slightly different one: (ii.) A: What did Maria do? B: MANN hat die einen Apfel gegessen. man has she an apple eaten ‘Boy, what an apple she ate.’ Adding such an exclamative force and hence triggering this movement seems possible only in matrix clauses, however. For example, adding the expletive to the first position is out of the question in an embedded clause unless it’s a quote – which would be odd under a propositional attitude verb like believe: (iii.) *Peter glaubt MANN hat die einen Apfel gegessen. Peter believes man has she an apple eaten

6.

7.

8. 9.

Scrambled given constituents can but need not be deaccented, in fact they’re often not deaccented when placed in first position. It is only when the constituent marked as given occurs to the right of its sister (or some other constituent that it can become prosodically dependent on) that it must be deaccented (Wagner 2005). This also raises many questions, as a reviewer points out: if (32a) and (33) are really flip sides of the same type of givenness marking, then why do the two movements appear to differ in in their syntactic properties? This would require a discussion that goes beyond the scope of this paper. Notation: Accented elements are in capitals, some diacritics: fall: \; rise: /; fall– rise: ∨. A reviewer suggested that Spanish shows the same pattern, but if one of the two focus operators is not overt, the facts are different: (ii) There is a girl that only John visited. After discussing who that girl is and going back and forth about it possibly being Mary, the speaker, a bit impatiently, says: A MAR´IA ha visitado s´olo JUAN. to Maria has visited only Juan ’Only Juan visited MAR´IA’, or ’It is MAR´IA that only Juan visited.’ A difference between overt and covert focus particles would be unexpected here. Since I didn’t test the nested focus data in Spanish and know little about the parallels

82

Michael Wagner

and differences between Italian and Spanish CLLD I’m not sure how to address this issue. 10. At least in base-generated order. See Wagner (2009b) for some discussion of cases where in German in non-base generated orders both word orders are possible, although the order is fixed in the base-generated order. The reason might be that the derived word order allows for reconstruction.

References Anderson, Stephen R. 1972 How to get ‘even’. Language, 48(4): 893–906. Bayer, Josef 1996 Directionality and Logical Form: On the scope of focusing particles and wh-in-situ. Kluwer, Dordrecht. Blaszczak, Joanna and Hand-Martin G¨artner 2005 Intonational phrasing and the scope of negation. Syntax, 7: 1–22. B¨uring, Daniel 1997 59th Street Bridge Accent On the Meaning of Topic & Focus. Routledge, London. B¨uring, D. 2003 On D-trees, beans, and B-accents. Linguistics and Philosophy, 26(5): 511–545. B¨uring, Daniel and Katharina Hartmann 2001 The syntax and semantics of focus-sensitive particles in German. Natural Language and Linguistic Theory, 19: 229–281. Calabrese, Andrea 1984 Multiple questions and focus in Italian. In W. de Geest and Y. Putseys, (eds.), Sentential complementation: Proceedings of the International Conference Held at UFSAL, pp. 67–74. Cinque, Guigliemo 1990 Types of A’-Dependencies. MIT Press, Cambridge, Ma. Cinque, Guglielmo 1999 Adverbs and functional heads: A cross-linguistic perspective. Oxford University Press, Oxford. Drubig, Hans Bernhard 1994 Island constraints and the syntactic nature of focus and association with focus. In Arbeitspapiere des Sonderforschungsbereichs 340: Sprachtheoretische Grundlagen der Computerlinguistik, volume 51. Sonderforschungsbereich 430, T¨ubingen/Stuttgart. Fanselow, Gisbert and Denisa Lenertov´a 2006 Left peripheral focus: Mismatches between syntax and information structure. Ms., Universities of Potsdam and Leipzig.

Focus, topic, and word order: A compositional view

83

Gajewski, Jon R. 2005 Neg-Raising: Polarity and Presupposition. Ph.D. thesis, MIT. Halliday, M.A.K. 1967 Notes on transitivity and theme in English, part 2. Journal of Linguistics, 3: 199–244. Horn, Laurence R. 1969 A presuppositional analysis of only and even. In Robert I. Binnick, Alice Davison, Georgia Green, and Jerry Morgan, (eds.), Papers from the 5th annual meeting of the Chicago Linguistic Society, volume 4. Chicago Linguistic Society, Chicago. Jackendoff, Ray S. 1972 Semantic Interpretation in Generative Grammar. MIT Press, Cambridge, Ma. Jacobs, Joachim 1983 Fokus und Skalen. Zur Syntax und Semantik der Gradpartikel im Deutschen. Niemeyer, T¨ubingen. Jaeger, Florian and Michael Wagner 2003 Association with focus and linear order in German. Ms., Stanford University. semanticsarchive.net/Archive/DVkNDY4M. Karttunen, Lauri and Stanley Peters 1979 Conventional implicature. In D.A. Dinneen and C.-K. Oh, (eds.), Syntax and Semantics 11: Presupposition, pp. 1–56. Academic Press, New York. Kayne, Richard S. 1998 Overt vs. covert movement. Syntax, 2: 128–191. Krifka, Manfred 1992 A compositional semantics for multiple focus constructions. In Joachim Jacobs, (ed.), Informationsstruktur und Grammatik, volume Sonderheft 4 of Linguistische Berichte, pp. 17–53. Westdeutscher Verlag, Opladen. Krifka, Manfred 1996 Frameworks for the representation of focus. In Geert-Jan M. Kruijff, Richard T. Oehrle, and Glyn Morrill, (eds.), Formal Grammar Conference. Proceedings of the Conference on Formal Grammar at the 9th European Summer School in Logic, Language, and information, pp. 99–112. Follli, Prague. Ladd, D. Robert 1980 The Structure of Intonational Meaning. Indiana University Press, Bloomington. Lappin, Shalom, (ed.) 1996 The Handbook of Contemporary SemanticTheory. Blackwell, London.

84

Michael Wagner

Lee, Youngjoo 2005

Exhaustivity as agreement: The case of Korean man ‘only’. Natural Language Semantics, 13: 169–200. McCawley, James D. 1970 English as a VSO language. Language, 46(2): 286–299. Neeleman, Ad and Tania Reinhart 1998 Scrambling and the PF-interface. In Miriam Butt and Wilhelm Geuder, (eds.), The projection of arguments, pp. 309–353. CSLI. Neeleman, Ad and Hans van de Koot 2007 The nature of discourse templates. Ms. UCL. Nilsen, Øystein 2002 Eliminating Positions. Ph.D. thesis, University of Utrecht. Reinhart, Tanya 2006 Interface Strategies. MIT Press, Cambridge, Ma. Rizzi, Luigi 1997 The fine structure of the left periphery. In Liliane Haegeman, (ed.), Elements of Grammar. Kluwer. Rooth, Mats 1992 A theory of focus interpretation. Natural Language Semantics, 1: 75– 116. Rooth, Mats 1996 Focus. In Lappin (1996), pp. 271–297. Rooth, Mats 2005 Topic accents on quantifiers. Ms., Cornell University. Rullmann, Hotze 1997 Even, polarity, and scope. Papers in Experimental and Theoretical Linguistics, 4: 40–64. Sauerland, Uli 2005 Contrastive topic: A reductionist approach. Ms. ZAS Berlin. Selkirk, Elizabeth O. 1995 Sentence prosody: Intonation, stress, and phrasing. In John A. Goldsmith, (ed.), Handbook of Phonological Theory, pp. 550–569. Blackwell, London. Stoyanova, Marina 2008 Unique Focus. Languages without multiple wh-questions. John Benjamins. Taglicht, Josef 1984 Message and Emphasis. On focus and scope in English, volume 15 of English Language Series. Longman, London and New York. Van Craenenbroeck, J. 2006 Transitivity failures in the left periphery and foot-driven movement operations. Linguistics in the Netherlands, 23(1): 52–64.

Focus, topic, and word order: A compositional view

85

Van Hoof, Hanneke 2003 The rise in the rise-fall contour: does it evoke a contrastive topic or a contrastive focus? Analogy, Levelling, Markedness: Principles of Change in Phonology and Morphology, 41(3): 515–563. von Fintel, Kai 1999 NPI licensing, Strawson entailment, and context dependency. Journal of Semantics, 16: 97–148. Wagner, Michael 2005 Prosody and Recursion. Ph.D. thesis, MIT. Wagner, Michael 2006a Association by movement. Evidence from NPI-licensing. Natural Language Semantics, 14(4): 297–324. Wagner, Michael 2006 Givenness and locality. In Masayuki Gibson and Jonathan Howell, (eds.), Proceedings of SALT XVI, pp. 295–312. CLC Publications, Ithaca, NY. Wagner, Michael 2009a Relatively given, relatively focused. Ms. Cornell University/McGill University. Wagner, Michael 2009b A compositional theory of contrastive topics. In Muhammad Abdurrahman and Anisa Schardl and Martin Walkow (eds.), Proceedings of NELS 2007. Volume I, pp. 415–428. GLSA Publications. Williams, Edwin 1997 Blocking and anaphora. Linguistic Inquiry, 28: 577–628. Zwarts, Frans 1998 Three types of polarity. In Fritz Hamm and Erhard Hinrichs, (eds.), Plural Quantification, pp. 177–238. Kluwer, Dordrecht.

A focus-binding conspiracy. Left-to-right merge, scrambling and binary structure in European Portuguese Jo˜ao Costa Abstract. An interaction between binding and focus in ditransitive contexts is looked into in this paper. It is shown that when a focused constituent must bind into a nonfocused constituent its position is flexible, and it cannot stay in the rightmost position. In such contexts, anaphoric deaccenting is the preferred option. It is defended that this type of flexible word order is problematic for cartographic approaches to focus, in which the focus interpretation emerges from a rigid designated position. An alternative view is proposed based on Phillips (2003) incremental approach to phrase structure.

1. Introduction Cartographic approaches to clause structure propose that several semantic and discourse notions are syntactic primitives encoded as functional categories organized under a rigid universal hierarchy. According to these approaches, a certain reading emerges as a consequence of constituent movement to the specifier or the head of the functional category encoding that reading. For instance, Belletti (2001) proposes that focus at the right periphery is to be analyzed as a consequence of movement to a VP-peripheral focus projection. Under this analysis, focus is a syntactic primitive with the same status as other syntactic categories and notions. In this paper, the prediction that focus is a syntactic primitive is tested in ditransitive contexts. The structure of VP in ditransitive contexts has been subject to a long debate in the literature (Kayne 1984, Larson 1988, Pesetsky 1995, Phillips 1996, among others). As noted in Phillips (1996, 2003), part of the debated problems derives from the fact that constituency tests yield contradictory results, when applied to ditransitive VPs. For instance, NPI-licensing in (1) provides evidence for a shell structure like in (3), while the fronting tests, (2), provide evidence for a layered structure like in (4): (1)

a. John gave nothing to any of the children on his birthday. b. *John gave anything to none of the children on his birthday.

88

(2)

(3)

Jo˜ao Costa

John intended to give candy to children on his birthday . . . a. . . . and [give candy to children on his birthday] he did. b. . . . and [give candy to children] he did on his birthday. c. . . . and [give candy] he did to children on his birthday. (4)

VP

VP

V’ V

V’ VP

DO

V

IO DO

V’ V

IO

The goals of this paper are twofold and independent: i) In the first part of the paper (section 2), I discuss a problem from European Portuguese (EP) concerning binding possibilities by an object inside VP into another argument to its right when focus and binding interact. The problem to be presented, as will be argued, constitutes a problem for cartographic approaches to focus. ii) The second half of the paper (section 3) is dedicated to the exploration of an alternative to the cartographic approach to focus. Since the result of the first half of the paper reveals that assigning focus to a specific position may be problematic, it is our burden to provide a solution for the bindingfocus interactions described. The alternative to be proposed follows Phillips’ (1996, 2003) solution for the contradictory constituency tests. It will be argued that the data discussed can be understood in a model permitting different base-generated structures for similar grammatical functions. The analysis presented makes no use of pragmatic notions such as focus as syntactic primitives.

2. Problem: a binding asymmetry Before presenting the problem to be discussed, let me review some background information on word order in European Portuguese. As is shown by the following examples, the clause’s right periphery is the locus for information focus in this language (Costa 1998b):

A focus-binding conspiracy

(5)

Quem e´ que tu viste ontem? who is that you saw yesterday ‘Who did you see yesterday?’ a. Vi ontem a Maria. saw yesterday the Maria ‘I saw Maria yesterday.’ b. #Vi a Maria ontem. saw the Maria yesterday

(6)

Quando e´ que tu viste a Maria? when is that you saw the Maria ‘When did you see Maria?’ a. #Vi ontem a Maria. saw yesterday the Maria b. Vi a Maria ontem. saw the Maria yesterday ‘I saw Maria yesterday.’

(7)

Quem viu a Maria? who saw the Maria ‘Who saw Maria?’

89

Viu (a Maria) o Jo˜ao. saw the Maria the Jo˜ao. ‘Jo˜ao saw Maria.’ In Costa (1996, 1998b), it is argued that in SVO orders the subject is in Spec,IP. VSO and VOS orders instantiate subjects stranded in Spec,VP. According to the results in the works cited, the difference between VSO and VOS lies in the status of the object. Costa (1996, 1998b) claims that, in VOS orders, the objects scrambles out of VP, creating an adjunction configuration. Following this analysis, depicted in (8), scrambling is an instance of A-bar-movement: (8)

[IP V [VP DO [VP Subj t t ]]] Viu a Maria o Jo˜ao. saw the Maria the Jo˜ao ‘Jo˜ao saw Maria.’

The arguments put forward for the scrambling analysis are based on the similar behaviour of objects in VOS and scrambled objects in Germanic. Further arguments come from a comparison between a scrambling approach and a remnant

90

Jo˜ao Costa

movement approach to this word order (Costa 1998b, 2002). In table 1, a list of the similarities between scrambled objects in Germanic and EP is given: Table 1. (T1) apud (Costa 1998b)

Parasitic gap licensing Movement of NPs

Movement of PPs Floating quantifier licensing Order Adv-NP-Adv Interaction with V-movement

German/Dutch

Scandinavian

EP

yes1

yes yes

yes no (?)2

no yes (Icelandic and Faroese: NPs; the other languages: pronominal NPs) no yes

yes no

no yes

yes no

yes

yes no3

An additional argument for A-bar scrambling of the object comes from binding facts. As illustrated in (9), object scrambling to the left of the subject does not feed binding, unlike what happens in cases of A-movement (9c). (9)

a. *Viu o seu filho cada m˜ae. saw the her son each mother b. *Viu cada m˜ae o seu filho. saw each mother the her son c. Cada m˜ae foi vista pelo seu filho. each mother was seen by.the his son ‘Each mother was seen by her child.’

VSO VOS Passive

Sentences (9b) and (9a) are grammatical if interpreted as VSO and VOS, respectively, which confirms the idea thatVOS isA-bar scrambling, since it reconstructs into its base position for binding purposes. Ditransitives pose a challenge to the generalization that object scrambling to the left of an argument prevents the creation of a new binding relation. In Costa (1998a), it is shown that, at first sight, the ordering of arguments obeys the constraint forcing focused elements to surface on the right periphery, as shown in (10) and (11):4 (10)

A: A quem e´ que deste o livro? to who is that gave the book ‘To who did you give the book?’

A focus-binding conspiracy

91

B:

Dei o livro [F ao Paulo] gave the book to.the Paulo ‘I gave the book to Paulo.’ B’: #Dei ao Paulo o livro. gave to.the Paulo the book

(11)

A: O que e´ que deste ao Paulo? the what is that gave to.the Paulo B: Dei ao Paulo [F o livro]. gave to.the Paulo the book ‘I gave the book to Paulo.’ B’: #Dei o livro ao Paulo. gave the book to.the Paulo

However, if the focused argument is to bind an anaphor contained within the non-focused argument, the former cannot be rightmost:5 (12)

a.

b.

A: A quem e´ que deste os livros? to who is that gave the.POSS books ‘To who did you give his books?’ B: Dei [F a cada autor] o seu livro. gave to each author the his book ‘I gave to each author his book.’ B’: ?*Dei o seu livro a cada autor. gave the his book to each author A: A que aluno devolveste o seu primeiro to which student returned the his first ‘To which student did your return his first essay?’ B: Devolvi [F a cada autor] o seu primeiro returned to each author the his first ‘I returned to each author his first essay.’ B’: ?* Devolvi o seu ensaio a cada autor. returned the his essay to each author

(13)

a.

A: O que e´ que deste aos autores? the what is that gave to.the authors ‘What did you give to the authors?’ B: Dei [F cada livro] ao seu autor. gave each book to.the his author ‘I gave each book to its author.’

ensaio? essay ensaio. essay

92

Jo˜ao Costa

b.

B’: ?* Dei ao seu autor cada livro. gave to.the his author each book A: Que trabalho devolveste ao seu autor? which homework returned to.the his author ‘Which homework did you return to its author?’ seu autor. B: Devolvi [F cada ensaio] ao each essay to.the his author returned ‘I returned each essay to its author.’ B’: ?*Devolvi ao seu autor cada ensaio. returned to.the its author each essay

Interestingly, focus is still marked in sentences (12B) and (13B), since the focused constituent bears a heavier stress and the non-focused constituent undergoes anaphoric deaccenting (in the sense of Reinhart 1995, 2006).6 These data provide an interesting opportunity for evaluating a cartographic approach to focus, and for an explicit comparison with alternatives to it. Under a cartographic approach to focus, it is expected that a constituent is interpreted as focus as a consequence of the position it occupies. At first sight, ditransitive contexts provide interesting evidence for a one-to-one relation between focus and position. However, when a focused constituent must bind, it surfaces in an unexpected position for focus purposes. What is relevant is the fact that, in spite of this position, the constituent can still be interpreted as focus. Moreover, as predicted in Reinhart (1995, 2006) and Szendro˝ı (2001), heavy stress acts as a last-resort marking device for focus. Under a strictly syntactic approach, the emergence of stress would just be coincidental. Actually, any language or any construction in which positional and prosodic effects co-exist posit a serious challenge for a rigid cartographic approach to information structure, since it shows that focus marking is not a strict syntactic matter.7 Note that it cannot be the case that this configuration arises as a consequence of movement of a DP to a designated focus position, followed by subsequent movement of the other argument to a position c-commanding it, since these two operations should be instances of A-bar movement, which would not be able to create new binding configurations. As such, if focus and binding were both configurational, one would expect the conflict between binding and focus to yield ineffability and an ungrammatical result. The fact that focus can be marked post-syntactically provides the necessary evidence for a more distributed view on focus marking, which is incompatible with a cartographic approach, according to which focus is a syntactic primitive, and focus readings emerge as the consequence of movement to designated focus positions.

A focus-binding conspiracy

93

If cartography does not explain focus marking, an alternative must be found. I follow the view defended in Szendro˝ı (2001) and Costa (1998b), inspired by Reinhart (1995, 2006), according to which focus is marked at PF. If the focused constituent is rightmost, sentence nuclear stress will fall on it. Otherwise, stressshift and anaphoric deaccenting act as last-resort strategies for marking focus. This view predicts that the latter strategies will only be at play when, for syntactic reasons, a focused constituent cannot emerge at the position where nuclear sentence stress falls. The conflict presented in the ditransitive context, in which the indirect object should be rightmost for focus purposes, and simultaneously it must be to the left of the direct object for binding purposes thus provides evidence for the claim that focus is not necessarily encoded in the syntax. There is, however, a remaining problem. There is a binding asymmetry between scrambling across subjects and across internal arguments, in the sense that moving an object across a subject does not feed binding, while an indirect object scrambled across a direct object creates a new binding configuration. In the rest of this paper, I will be trying to answer the following two questions: A. What is the nature of the configuration in which the indirect object precedes the direct object: base-generation, A-movement, A-bar movement? B. Why doesn’t indirect object scrambling across a direct object yield the same results as direct object scrambling across a subject? Rephrasing the problem, the binding facts suggest that an object moved across a subject seems to be A-bar moved, while an indirect object moved across the direct object seems to occupy an A-position. The argumentation developed throughout this section reveals that the interaction between focus and binding described is problematic for a cartographic approach. An independent issue arises: how to account for this type of data with a non-cartographic type of analysis? In the next section, I will be presenting an alternative view on this problem, not resorting to rigid designated syntactic positions. It will be argued that an analysis based on left-to-right merge provides evidence for a loose relation between position and interpretation, contrasting with a rigid and universal cartographic structure.

3. Proposal: left-to-right merge, binary branching, the status of derived vs. basic positions In the previous section, one major point was made: a cartographic approach does not straightforwardly account for cases of conflict solving between binding and focus marking. This is because it is predicted that focus is marked at rigid desig-

94

Jo˜ao Costa

nated positions. The type of conflict between binding and focus and the strategy used in the language to solve it reveals that, in order to account for this type of data, one needs an approach in which proposals on constituent structure take into account not only the linear order, but also evidence for hierarchical relations. Cartographic approaches, sometimes, focus more on linear order effects than on the evidence that can be used for postulating a given hierarchy. Actually, when one considers linear order and hierarchical relations, there may be cases of conflict, and constituency tests may yield contradictory results. The example discussed in the previous section also illustrated an apparent contradiction for current assumptions on hierarchy and word order: on the one hand, we found evidence for postulating that VOS arises as a consequence of object scrambling across the subject, and hierarchy diagnostics suggest that scrambling is an instance of movement. On the other hand, similar data on indirect objects suggests that indirect objects occupy an A-position above the direct object, which cannot be accounted for in terms of A-bar movement. Reviewing this state of affairs, we get a complicated picture, according to which the application of the same diagnostics for constituency and structure lead to an interpretation of the word order facts in which similar configurations arise as the result of movement, disrupting the linearization established by base-generation (in the case of VOS), or as the result of a generation of an argument over another argument, suggesting that there are two potential base-generated orders (V-DO-IO and V-IO-DO). The puzzle to be solved is, then, why sometimes multiple base-generation is allowed, whereas, some other times, movement is required. Obviously, this is incompatible with a cartographic approach, which privileges movement operations to designated positions disfavoring more flexible views of the syntactic component according to which the same grammatical function can be generated in different positions, establishing one-to-one relations between meaning and placement. For this reason I will resort to a more flexible approach to constituency. In particular, the solution I will offer for the binding puzzle is based on the assumption that syntactic structure is strictly binary, and generated from left to right, as proposed in Phillips (1996, 2003). The option for this type of analysis over other flexible approaches to syntax (e.g. Neeleman and Weerman 1999) is motivated by the fact that the design of Phillips’ hypothesis permits accounting for contradictory constituency tests, and justifies why sometimes movement is the only option available. As such, it is superior to a radically flexible approach to syntax enabling generalized multiple base-generations.8 Phillips (1996, 2003) proposes that syntactic structure is generated from left to right in an incremental manner. According to this hypothesis, different steps of sentence construction may have different configurations. Consider the following example, taken from Phillips (2003), illustrating the generation of the sentence Wallace saw Gromit in the kitchen:

A focus-binding conspiracy

95

Figure 1. from Philips (2003)

As shown above, Phillips’ proposal predicts that, at different stages of the generation of the structure, certain strings are constituents, but that the same constituency relation is destroyed at a later stage (see Phillips (2003) for details). As such, it is predicted that constituency tests yield different results depending on the moment of the incremental derivation at which each test applies. An aspect that is crucial to our approach is that, according to Phillips (2003), the combinatorial options for grammatical heads depend on properties of the lexical items. Accordingly, legitimate objects can be created if lexical properties of the heads involved (such as thematic information) are satisfied. In what follows, I will try to show that these aspects of Phillips’ hypothesis give us the necessary ingredients for accounting for the puzzle identified in the previous section.

3.1. The unmarkedness of the order V-DO-IO I follow Belletti and Shlonsky (1995) in their claim that the order direct objectindirect object in Romance is unmarked. Evidence in favor of this comes from the emerging word order in sentence-focus contexts (14), and from the order found in idioms formed with ditransitive verbs (15–16):9 (14)

O que e´ que aconteceu? the what is that happened ‘What happened?’ a. O Jo˜ao deu uma prenda a` the Jo˜ao gave a gift to ‘Jo˜ao gave a gift to Maria.’ b. #O Jo˜ao deu a` Maria uma the Jo˜ao gave to Maria a

Maria. Maria prenda. gift

96

Jo˜ao Costa

(15)

a.

Dar p´erolas a porcos. give pearls to pigs ‘To cast pearls before swine.’ b. *Dar a porcos p´erolas. give to pigs pearls

(16)

a.

Pˆor mais lenha na fogueira. put more wood in.the fire ‘To add fuel to the fire.’

b. *Pˆor na fogueira mais lenha. put in.the fire more wood However, I deviate from Belletti and Shlonsky (1995), in assuming that the fact that this word order is unmarked does not necessarily mean that it corresponds to a base-generated word order. Note that constituency tests also yield contradictory results in EP: on the one hand, binding (18–19) and scopal facts (20) provide evidence for a shellstructure, indicating that the highest argument is base-generated: (18)

a.

Entreguei cadai livro ao seui autor. gave each book to.the its author ‘I gave each book to its author.’

a’. *Entreguei o seui livro a cadai autor. gave the his book to each author b.

Entreguei a cadai autor o seui livro. gave to each author the his book ‘I gave each author his book.’

b’. *Entreguei ao seui autor cadai livro. gave to.the its author each book (19)

(20)

a. *Apresentei a Mariai a` filha da Mariai . introduced the Maria to daughter of Maria b.

Apresentei a` filha da Mariai a Mariai . introduced to daughter of Maria the Maria ‘I introduced Maria to her daughter.’

a.

Apresentei uma mulher a todos os homens. introduced a woman to all the men ∃ > ∀ (preferred) ‘I introduced a woman to all the men.’

A focus-binding conspiracy

b.

97

Apresentei a todos os homens uma mulher. introduced to all the men a woman ∀ > ∃ (preferred) ‘I introduced a woman to all the men.’

On the other hand, gapping (21) and fronting (22) provide evidence in favour of a layered structure, contradicting the previous results. Assuming that both movement and ellipsis target constituents, the examples in (21) and (22) illustrate cases in which the elided and moved constituents include the verb and the direct object, but exclude the indirect object, suggesting that the verb and the direct object form a constituent, and that the indirect object is inserted at a different plane:10 (21)

(22)

a.

Eu dei livros a` Maria e o Pedro [ deu livros] a` I gave books to Maria and the Pedro gave books to Ana. Ana ‘I gave books to Maria and Pedro to Ana.’

b.

Eu pus os livros na prateleira e tu I put the books on.the shelf and you [puseste os livros] na gaveta. put the books in.the drawer ‘I put the books on the shelf and you in the drawer.’

O Pedro queria dar os livros ao Rui ontem. the Pedro wanted give the books to.the Rui yesterday a. e [dar os livros] ele deu ao Rui ontem. and give the books he gave to.the Rui yesterday b.

e [dar os livros ao Rui] ele deu ontem. and give the books to.the Rui he gave yesterday

c.

e [dar os livros ao Rui ontem] ele deu. and give the books to.the Rui yesterday he gave ‘Pedro wanted to give the books to Rui yesterday, and give the books to Rui yesterday, he did.’

It is important to note that the evidence for layered structures is not crystalclear, since with other types of ditransitives, stranding one of the arguments is not good, a problem I will not address in this paper:11

98

(23)

3.2.

Jo˜ao Costa

O Pedro queria the Pedro wanted a.??*e [pˆor os and put the

pˆor os livros put the books livros] ele pˆos books he put

na on.the na on.the

prateleira shelf prateleira shelf

ontem yesterday ontem. yesterday

b.

e [pˆor os livros na prateleira] ele pˆos ontem. and put the books on.the shelf he put yesterday

c.

e [pˆor os livros na prateleira ontem] ele pˆos. and put the books on.the shelf yesterday he put ‘Pedro wanted to put the books on the shelf yesterday and put the books on the shelf yesterday, he did.’

Base-generating V-DO-IO and V-IO-DO

The contradictory tests suggest that an analysis along the lines of Phillips (1996, 2003), as explained at the beginning of section 3, may be on the right track.12 The consequence of this type of analysis is that there are two basic ways of generating the two word orders, even if only one of them is unmarked. The conclusion from this type of approach is that there is no direct relation between unmarkedness and base-generation. Phillips’ analysis is based on the following two principles: (24)

MERGE RIGHT New items must be introduced at the right edge of a structure.

(25)

BRANCH RIGHT Where a terminal can be attached to more than one position in the existing structure with no effect on interpretation, the attachment that results in the more right-branching structure must be chosen.

The interaction between these two principles operates in the following way for the generation of V-DO-IO and V-IO-DO orders: (26)

Step1: merge V with Direct Object, discharging theta-role V

DO VP V

DO

A focus-binding conspiracy

99

Step2: create a copy of V (reanalyzing the direct object as a specifier), and merge a PP as its sister, discharging the goal theta-role. VP V

DO

a.

PP

VP V

b. VP

VP V

DO

V

VP DO

V’ V

PP

Phillips’ analysis predicts that constituency tests may target a step of the derivation in which V and DO form a VP, or the final structure, in which there is c-command of the DO over the PP. For generating V-IO-DO order, (24) and (25) operate in the following way: (27)

Step1: merge V with Indirect Object, discharging goal theta-role V

PP VP V

PP

Step2: create a copy of V (reanalyzing the indirect object as a specifier), and merge a DP as its sister, discharging the theme theta-role. VP V

PP

a.

DO

VP V

b. VP

PP

VP V

V

VP PP

V’ V

DO

Note that, as mentioned above, claiming that both orders are base-generated does not entail that they should be optional. The structure built in (27) is only generated if necessary for satisfying binding requirements (or any other constraint forcing the order IO-DO, such as heaviness).

100

Jo˜ao Costa

Suggesting that V-IO-DO can be base-generated for binding purposes makes two interesting predictions. First, as is also pointed out in Phillips (1996, 2003), it is expected that no case is found in which the verb and the indirect object may be fronted stranding the direct object. In other words, this word order cannot be created just to facilitate the movement of the V+IO complex, which is confirmed by the data in (28): (28) *O Pedro queria entregar the Pedro wanted give entregar aos vencedores give to.the winners

os pr´emios the prizes ele entregou he gave

aos vencedores e to.the winners and os pr´emios. the prizes

The only circumstances under which V-IO fronting is possible with DO stranding are cases in which binding is involved. Although the contrast is subtle, (29) is better than (28): (29) ?O the e and

Pedro queria entregar a cada Pedro wanted give to each entregar a cada vencedor ele give to each winner he

vencedor winner entregou gave

o the o the

seu his seu his

pr´emio prize pr´emio. prize

Second, this type of approach straightforwardly predicts well-known cases of indirect object control. In (30), PRO contained in the direct object is controlled by the indirect object, and it is not necessary to assume that the clause is rightdislocated, which would be problematic, since extraposed clauses are islands for extraction, which is not the case with complement clauses in indirect object control contexts. The analysis I am suggesting predicts that both DO and IO occupy A-positions, and the control pattern follows, because the IO c-commands the DO. (30)

Eui pedi aos meninosj para PROi/j dan¸car. I asked to.the children for dance ‘I asked the children to dance.’

3.3. VOS cannot be base-generated Let us finally turn to the solution of the puzzle presented in section 2: why it is not possible for a scrambled direct object in VOS to bind into the subject. Recall that the binding possibilities for V-IO-DO and V-IO-DO were derived from the fact that both word orders are base-generated. Therefore, no A-bar-movement

A focus-binding conspiracy

101

takes place. The problem must now be stated in different terms: the issue now is why VOS cannot be base generated. In both cases discussed above, the theta-criterion is satisfied: either the verb or its copy assigns the theta-roles to each one of the arguments. This entails that the theta-criterion is satisfied in the course of the clause’s generation (and not presyntactically). This consequence strongly clashes with Baker’s (1988) UTAH. In particular, there is no need to derive one of the orders in ditransitive contexts from the other, since the theta-criterion may be satisfied in both cases. There is, however, an important issue that is left unanswered: why is there a preference for the base-generation of V-DO-IO, and the other word order only arises for binding purposes? I have no clear answer to this question, other than suggesting that this some version of the thematic hierarchy is nevertheless needed, but it acts as an elsewhere condition, rather than as an absolute inviolable constraint. Obviously, this suggestion needs further elaboration. A complete answer to this question implies finding domains in which it can be established whether thematic hierarchies are categorical or a preference. Independent evidence for the claim that theta-roles may be assigned in the course of the derivation, and that the realization of the theta-role may depend on other factors comes from cases of complex predicate formation (Wurmbrand 1998, Gon¸calves 1999). As discussed in Gon¸calves (1999), the realization of an embedded agent as a DP or as a PP depends on the transitivity of the embedded verb, on the type of main verb, and on the process of complex predicate formation. This is illustrated in (31): (31)

a.

b.

Eu mandei correr os meninos. I made run the children ‘I made the children run.’ Eu mandei ler o livro aos meninos. I made read the book to.the children. ‘I made the children read the book.’

Crucially, if all agents were DPs generated in Spec,VP, the pattern in (31) would be unexpected, since the agent is a PP bearing dative case. The fact that the realization of the agent as a PP in (31b) is forced by constraints on complex predicate formation provides independent evidence for a theory of theta-role assignment that takes into consideration the sentence as a whole rather than each of the theta-assigning heads independently.13 The relevance of the type of main verb for determining the shape of the embedded agent becomes even more obvious in the cases discussed in Gon¸calves (1999) in which the matrix

102

Jo˜ao Costa

verb does not make available a PP position, and, as a consequence, the agent theta-role may not be discharged: (32)

a.

Eu vi correr os meninos. I saw run the children ‘I saw the children run.’ b. *Eu vi ler o livro aos meninos. I saw read the book to.the children ‘I saw the children read the book.’

Returning to the binding difference between V-IO-DO and V-O-S, recall that the object surfacing to the left of the subject does not create new binding possibilities, while the IO across DO is able to bind. As mentioned above, interpreting reconstruction as a consequence of A-bar movement, and the binding relation between IO and DO as a consequence of base-generation, the question to be answered is why the object cannot be base-generated above the subject in VOS. In what follows, I will show that it is not possible to satisfy left-to-right merge and the theta-criterion, creating a base-generated VOS. Step 1 in (33) represents the merging of the verb with the direct object, which like in the case of ditransitive contexts is unproblematic. (33)

V-O-S: Step1: merge V with Object, discharging theme theta-role V

DO VP V

DO

Step2: create a copy of V (reanalyzing the object as a specifier), and merge a DP as its sister, discharging the agent theta-role. VP V

DO

a.

Subj

VP V

b. VP

DO

VP V

V

VP DO

V’ V

Subj

A focus-binding conspiracy

103

As shown above, the problem comes about when the subject is merged with the copy of the verb, in compliance with MERGE RIGHT. The resulting configuration, illustrated in (33b) does not yield a proper context for assignment of the external theta-role. In other words, the problem is that the subject is not a sister to V’, and is therefore unable to receive theta-role, according to Koopman and Sportiche (1991).14 As a consequence, unlike V-IO-DO, VOS cannot be base-generated. Accordingly, the subject can only be base generated in a position where its theta-role is properly assigned. This will have to be a position hierarchically above the internal arguments. As such, the only option for the object to be linearized before the subject and to be in higher position is to perform scrambling under movement. Consequently, VOS must be an instance of movement. If it is A-bar movement, as the comparison with Germanic suggests (cf. table 1), the structure in (34) obtains and the binding possibilities follow. (34)

[IPV [VP O [VP S t t ]]]

It follows from the analysis proposed that scrambling across a subject is different from scrambling of an internal argument over another internal argument. Santorini (1990), Lee and Santorini (1994) and Frank, Lee and Rambow (1992) observe similar binding asymmetries in German and Korean. Their conclusions regarding the potential for reconstruction when different grammatical functions are considered are summarized in table 2 below: Table 2. (T2) Binding possibilities in German and Korean Weak Crossover Facts Subject bindee IO/DO bindee

Subject binder – reconstruction possible

IO/DO binder reconstruction impossible reconstruction impossible

The general conclusion to be drawn from these authors’ observations is that scrambling across subjects yields special results. It is important to compare the solutions offered by these authors for the asymmetry between subjects and internal arguments to the approach advocated in this paper. Santorini’s (1990) solution for the German facts starts from the assumption that pronominal INFL in German may act as a binder. Frank, Lee and Rambow (1992) note that such an approach is not tenable, since similar facts occur in Korean, where INFL is not pronominal. Instead, these authors suggest that subjects are not theta-marked, stipulating a β-marking relation that holds between subjects and the elements they c-command. Since c-command is a pre-requisite for binding, the problems of scrambling across a subject reduce to binding theory. The problem with this

104

Jo˜ao Costa

approach is that the German/Korean facts are not exactly the same as the EP facts; hence they should not follow from a general constraint on binding. As it stands, the solution presented above for the Portuguese data is advantageous. It maintains the idea that subjects are special, but in a different way, independently needed: they are special because they are assigned theta-role in a different way.

4. Conclusions: flexible syntax and cartography We draw the following conclusions, on the basis of the analysis made throughout this paper: a) The binding asymmetry between V-IO-DO and V-O-S follows from the different ways that may be used for generating these structures. Crucially, there is no base-generated option for VOS. b) The theta-criterion acts as a crucial factor to select different derivations. c) Although derivational mechanisms and base-generated options do not exclude each other globally, the latter are preferred, which explains the preference for base-generating V-IO-DO instead of deriving it in terms of IOscrambling across the DO. Coming back to the main issue of this paper – the comparison between cartographic approaches and alternatives to it – I hope to have shown that a cartographic analysis of focus leaves the interaction between binding and focus unexplained, since it predicts ineffability in the case of a conflict between the hierarchy for binding purposes and for focus marking. The alternative proposed, in terms of a more flexible syntax, which does not include discourse notions as primitives, was able to account for the facts. Crucially, the flexible view advocated here implies global comparison between converging derivations and the selection of outputs converging with discourse requirements taking place post-syntactically.

Notes 1.

2. 3.

Deprez (1989) presents some data suggesting that scrambled DPs in German license parasitic gaps. For relevant discussion, see Bennis and Hoekstra (1984), Webelhuth (1989), Neeleman (1994), among others. See Doetjes (1997) and Costa (1998b) for discussion of the status of floating quantifiers associated with objects. Note the ungrammaticality of (i) and (ii):

A focus-binding conspiracy

105

(i)

4.

*O Pedro leu os livros bem todos. the Pedro read the books well all (ii) *Leu os livros o Pedro todos. Read the books the Pedro all This is just another instantiation of the general behaviour of information focus in Romance, which tends to surface in rightmost position, as shown in (i) (Costa 1998b): (i) Quem tossiu? who coughed ‘Who coughed?’ a.

Tossiu o Jo˜ao. coughed the Jo˜ao ‘Jo˜ao coughed.’

b. #O Jo˜ao tossiu. the Jo˜ao coughed. 5.

The b. examples illustrate contexts in which the bound variable is contained in the question, to ensure that the information structure of the argument already referred to is the same in the question and in the answer. 6. This pattern is crucially different from what is discussed and analyzed in SamekLodovici (2006) for post-focal constituents in Italian. As shown by this author, in Italian, right-dislocated post-focal XPs are not c-commanded by the focused constituent. For this reason, his analysis does not extend to these cases. 7. For similar arguments on the basis of distribution of possessives, inverted subjects and processing costs in language acquisition, see Castro and Costa (2003), Costa and Figueiredo Silva (2006) and Costa and Szendrõi (2006), respectively. All these works converge in showing that the behaviour of focus marking is better understood if focus is not taken as a syntactic primitive. 8. An explicit comparison between different alternatives to cartographic approaches falls beyond the scope of this paper. See Phillips (2003) for a comparison between incremental structure-building and flexible approaches. 9. The only cases in which other orders are found in idioms involve heavy complements: (i) Dar a C´esar o que e´ de C´esar. give to Caesar the what is of Caesar ‘Render unto Caesar the things which are Caesar’s’ 10. As pointed out to me by Jeroen van Craenenbroeck, the argument made on the basis of the data in (21) is highly dependent on the idea that gapping affects constituents only, and not subconstituents or discontinuous constituents. Likewise, if Johnson’s (1994) analysis of gapping is on the right track, gapping is a remnant constituent signaling what underwent movement. Although I acknowledge the fact that gapping deserves further consideration, it is interesting to note the coincidence between gapping and fronting, by means of which it can be shown that the same “parts” of the VP that can be fronted can also appear in gapping configurations. For this reason,

106

11. 12.

13.

14.

Jo˜ao Costa I will adhere to the more conservative view, according to which gapping can be used as a constituency diagnostic. See Phillips (2003) for restrictions on argument stranding. An anonymous reviewer shows some skepticism towards the idea that the different syntactic tests apply at different points in the derivation, given that each relevant step in the derivation is not a separate level of representation. Actually, Phillips’approach is fully compatible with a model of syntax without levels of representation, such as the Minimalist Program. For this reason, there is no clear reason to identify the steps of an incremental derivation with traditional levels of representation. As noted by an anonymous reviewer, this is another case in which a flexible approach is clearly superior to a cartographic approach, since we see a dissociation between categories, forms, and meaning. Since this is not a restructuring context, isomorphism with a matrix verb will not play any role.

References Baker, Mark 1988 Incorporation. Chicago: Chicago University Press. Belletti, Adriana 2001 ‘Inversion’ as focalization. In: Aafke Hulk and Jean-Yves Pollock (eds.), Inversion in Romance and the theory of Universal Grammar, 107–162. New York: Oxford University Press. Belletti, Adriana and Ur Shlonsky 1995 The order of verbal complements: A comparative study. Natural Language and Linguistic Theory 13: 489–526. Bennis, Hans and Teun Hoekstra 1984 Gaps and Parasitic Gaps. The Linguistic Review, 4: 29–87. Castro, Ana and Jo˜ao Costa 2003 Weak forms as Xo : prenominal possessives and preverbal adverbs in Portuguese. In: Ana Teresa P´erez-Leroux and Yves Roberge (eds.), Romance Linguistics: theory and acquisition, 95–110. Amsterdam/Philadelphia: John Benjamins. Costa, Jo˜ao 1996 Positions for Subjects in European Portuguese. In: B. Agbayani et al. (eds.), Proceedings of WCCFL XV, 49–63. Stanford: CSLI. Costa, Jo˜ao 1998a Parameters vs. soft constraints in the analysis of discourse configurationality. In: M. Catherine Gruber, Derrick Higgins, Kenneth S. Olson , and Tamra Wysocki (eds.), Proceedings of Chicago Linguistic Society 36, 17:30. Chicago: University of Chicago

A focus-binding conspiracy Costa, Jo˜ao 1998b Costa, Jo˜ao 2002

107

Word order variation: a constraint-based approach. Ph. D. diss., HIL/Leiden University.

VOS in Portuguese: arguments against an analysis in terms of remnant movement. In : Artemis Alexiadou et al. (eds.), Dimensions of Movement, 69–89. Amsterdam/Phildadelphia: Johns Benjamins. Costa, Jo˜ao and Maria Cristina Figueiredo Silva 2006 On the (in)dependence relation between syntax and pragmatics. In: Val´eria Moln´ar and Susanne Winkler (eds.), The Architecture of Focus, 83–104. Berlin: Walter de Gruyter. Costa, Jo˜ao and Kriszta Szendro˝ı 2006 Acquisition of focus marking in European Portuguese. Evidence for a Unified Approach. In: V. Torrens and L. Escobar (eds.), The Acquisition of Syntax in Romance Languages, 319–329. Amsterdam: Johns Benjamins Deprez, Viviane 1989 On the typology of syntactic positions and the nature of chains: Moveα to the specifier of functional projections. Ph. D. diss., MIT. Doetjes, Jenny 1997 Quantifiers and selection: On the distribution of quantifying expressions in French, Dutch and English. Ph. D. diss., University of Leiden. Frank, R., Y-S. Lee and O. Rambow 1992 Scrambling as non-operator movement and the special status of subjects. In: S. Barbiers et al (eds.), Proceedings of LCJL 3, 135–154. Leiden: Leiden University. Gon¸calves, Anabela 1999 Predicados verbais complexos em contextos de infinitivo n˜ao preposicionado em portuguˆes europeu. Ph. D. diss, Universidade de Lisboa Johnson, Kyle 1994 Bridging the gap. Manuscript, University of Massachusetts atAmherst Kayne, Richard 1984 Connectedness and binary branching. Dordrecht: Foris. Koopman, Hilda and Dominique Sportiche 1991 The Position of Subjects. Lingua 85: 211–258. Larson, Richard 1988 On the double object construction. Linguistic Inquiry 19 335–391. Lee Y-S. and Beatrice Santorini 1994 Towards resolving Webelhuth’s paradox: Evidence from German and Korean. In: Norbert Corver and Henk van Riemsdijk (eds.), Studies on scrambling. Movement and non-movement approaches to free wordorder phenomena, 257–300. (Studies in generative grammar 41.) Berlin: Mouton de Gruyter.

108

Jo˜ao Costa

Neeleman, Ad 1994 Complex Predicates. Ph. D. diss., Utrecht University Neeleman, Ad and Fred Weerman 1999 Flexible Syntax; A Theory of Case and Arguments. Dordrecht: Kluwer. Pesetsky, David 1995 Zero syntax. Cambridge, MA: MIT Press Phillips, Colin 1996 Order and Structure. Ph. D. diss., MIT Phillips, Colin 2003 Linear order and Constituency. Linguistic Inquiry 34:1, 37–90. Reinhart, Tanya 1995 Interface strategies. OTS Working Papers in Linguistics. University of Utrecht. Reinhart, Tanya 2006 Interface strategies. Optimal and costly computations. Cambridge, MA: MIT Press Samek-Lodovici, Vieri 2006 When Right Dislocation Meets the Left-Periphery. Lingua 116: 836– 873. Santorini, Beatrice 1990 INFL and scrambling in German. Manuscript, University of Pennsylvania. Szendr˝oi, Kriszta 2001 Focus and the syntax-phonology interface. Ph. D. diss., UCL. Webelhuth, Gert 1989 Syntactic Saturation Phenomena and the Germanic Languages. Ph. D. diss., University of Massachusetts at Amherst. Wurmbrand, Susi 1998 Infinitives. Ph. D. diss., MIT.

Phases and variation: Exploring the second factor of the faculty of language∗ ´ Angel J. Gallego Abstract. This paper investigates the status of linguistic variation within Minimalism, providing a phase-based account of some parametric asymmetries within Western Romance languages. The discussion focuses on the observation that parametric (i.e. 2nd factor) concerns are typically ignored by minimalist theorizing, given the programmatic interest in the so-called 3rd factor (see Chomsky 2005), the only one dealing with interface and optimal design matters. I review the basics of Phase Theory (see Chomsky 2000 and subsequent work), paying special attention to the internal structure of phases. I compare this type of approach to cartographic ones, pointing out some empirical and conceptual shortcomings of the latter, and provide evidence from Western Romance that the same type of linguistic variation can be found in all phasal contexts (namely, CP, v*P, and DP), in a parallel fashion, within a given language, a fact I take to argue that variation cannot be captured by simply postulating some additional (set of) head(s). In line with Uriagereka (2002b; 2008), I pursue the thesis that linguistic variation associated with pragmatic (or discourse-oriented) data is better regarded as a phenomenon (not a specific projection), triggered by the morphological properties of the relevant languages.

1. Introduction This paper explores the interaction between linguistic variation (parameters, roughly in the GB sense) and Chomsky’s Phase Theory (see Chomsky 2000 and subsequent work), arguing for the existence of a parallel behavior across phase domains in a group of Romance languages. In particular, it will be claimed, following ideas of Boeckx (2008a), Poletto (2006), and Uriagereka (1988, 1995, 2002b) that Western Romance languages cluster with respect to the possibility of displaying extra (edge or left peripheral) fronting possibilities vis-`a-vis Central Romance languages. Notice that, in and of itself, such a proposal is nothing but a descriptive generalization: certain Romance languages allow for more word order combinations (in other words, they are less configurational). The interesting part of it has to do with whether a better understanding of language variation can be reached by exploring the facts from different perspectives, like that of the so-called Cartographic Project (see Belletti 2001, Cinque 1999, Rizzi 1997 and much related

110

´ Angel J. Gallego

work), which, to the best of my knowledge, has no agreed-upon formulation within minimalism (but see recent work by Boeckx 2008a and Fortuny 2008). There is no consensus as to why, say, a projection A must precede a projection B in Italian (but perhaps not in Catalan), or how cartographic projections emerge: do they follow from interface conditions, as Fortuny (2008) suggests? Do they involve some kind of efficient computation? Within minimalism, cartographies are alluded to in passing, at most noting that T or C should be taken “as surrogates for richer systems” (Chomsky 2000:143) or “cover terms for a richer array of functional categories” (Chomsky 2001:43). At the same time, and although cartographic findings cannot be empirically ignored, Chomsky seems to be skeptical about there being a proliferation of functional projections along the clausal spine: [C]onsider the target of verb raising. Evidence has accumulated that the verb can raise to a position higher than tense but lower than C, differentiating languages by the position of a functional category α, on current assumptions. But there is reason to doubt that such α can exist; or, to put it differently, if it does, then departures are needed from what appears to be the simplest and most principled form of phrase structure theory. Again questions arise as to whether there is some better way to conceive the matter. [from Chomsky 2000:93]

In what follows I will investigate these matters by adopting Uriagereka’s (2002b, 2008) claim that cartographic effects can be seen as an epiphenomenon – that is, a conglomerate of independent factors that cloud the nature of the facts.1 This idea can in principle be pursued from different viewpoints: one can flesh it out by invoking an extra category (as Juan Uriagereka did in the eighties and nineties, see Uriagereka 1988, 1995a, 1995b), an extra feature (some version of Chomsky’s EPP or P-features, see Chomsky 2000), or even an extra mechanism (Rizzi’s criteria, see Rizzi 2006a). Since it is not immediately obvious that the first and third options can shed some light on the data to be examined, I will try to take advantage of purely morphological facts, remaining agnostic as to whether these can be used to argue in favor of an extra category (e.g. Uriagereka’s FP): that may end up being the case, but I will not press the point here. For my purposes, all that matters is that one can relate the morphological evidence to some head. I claim that this head is ‘Ph’(a shorthand for ‘phase head’), which will be treated as a category-neutral morpheme displaying the same syntactic behavior (in terms of fronting to what Chomsky calls edge) within a given language or language family, regardless of its actual manifestation (C, v*, or D). If my contention is on track, then one expects that, for a language L, the different allomorphs of Ph manifest the same fronting (movement to the edge) effects. As we will see in section 4, this prediction borne out in Romance languages.

Phases and variation: Exploring the second factor of the faculty of language

111

It is worth noting that this overall discussion bears not only on cartographies in general, but on how these manifest themselves in specific languages. Hence, we are in fact dealing with the topic of (parametric) variation: what Chomsky (2005) calls second factor.2 I purposefully use parentheses on the adjective parametric, to signal the impression that there seems to be no clear idea of what parameters may look like within minimalism (see Boeckx 2008b; Kayne 2000, 2005; Uriagereka 2007 for different explorations). This, at least in part, is a consequence of minimalism being mostly devoted to investigate the third factor of the Faculty of Language, namely: [P]rinciples of structural architecture and developmental constraints that enter into canalization, organic form, and action over a wide range, including principles of efficient computation, which would be expected to be of particular significance for computational systems such as language. [from Chomsky 2005:6]

Since the focus of interest for minimalism has shifted from the connection between factors one and two (the ever present tension between descriptive and explanatory adequacy of the GB period) to factor three, trying to account for language variation has become a collateral topic on the minimalist agenda. In this paper, I will try to show that there are ways of incorporating second-factor effects into the agenda, by showing them to be connected to the hallmark of most recent minimalist theorizing: phases. The paper is divided up as follows. Section 2 discusses the notion of phase, focusing on Richards’ (2007) discussion about C-T ϕ-feature inheritance and its implications for the appearance of functional categories in the CP layer. Section 3 delves into Uriagereka’s (2002b) arguments for recruiting cartographic effects by means of Chomsky’s (2001) analysis of surface (i.e. discourse oriented) semantics arising at phase edges. Finally, in section 4, I establish a connection between the left peripheral fronting properties of v* and C in order to propose that these heads be regarded as contextual versions of a more abstract category, which I label ‘Ph.’ As we will see, the data that will be reviewed provide strong evidence in favor of the hypothesis of there being syntactic effects manifesting a parallel behavior across phases (see Poletto 2006).

2. Minimalism and cartographies: stretching/collapsing strategies in syntax In this section I want to consider some of the approaches to phases that one can find in the recent literature. I will compare three such approaches, arguing that two of them are problematic because they are either not restrictive enough or

112

´ Angel J. Gallego

too restrictive. In so doing, I will try to support a non interface-driven definition of phases (the one put forward in Chomsky 2001, 2004, 2007, 2008), and discuss how the skeletal make-up of these domains can be made compatible with cartographies.

2.1.

Defining phases

Capitalizing on work that goes back to the notion of phonological cycle (see Chomsky et al. 1956 and Chomsky and Halle 1968), Chomsky (2000) revamps the hypothesis that syntactic computation proceeds through small derivational periods. Chomsky (2000) calls the relevant units phases, a current alias for the previous bounding nodes (see Chomsky 1973, 1977, 1981) or barriers (see Chomsky 1986a). In their first formulation (see Chomsky 2000), phases were invoked largely to address questions of computational complexity that concern the way the lexicon and narrow syntax interact. In order to reduce computational load, Chomsky proposed to access the lexicon in a periodic fashion, and to select a lexical array in every access, a subpart of which is placed in ‘active memory:’ the phases. Ever since Chomsky (2000), there has been much controversy about how to define phases. Simplifying somewhat, three types of approaches can be identified: (1)

a. b.

c.

Standard: CP and v*P (see Chomsky 2000 through 2008) Dynamic: domains that are phonologically, morphologically, or semantically independent (see den Dikken 2007, Fox and Pesetsky 2005, Gallego 2007, and Uriagereka 1999a, 1999b) Radical: every application of Merge (see Epstein et al. 1998 and Epstein and Seely 2002)

Most of the literature I am familiar with takes option (b) as the right way to go about phases – plausibly because this option has a third-factor flavor. However, note that it is precisely the most difficult option to evaluate, for it depends on what one takes to be an object autonomous at the interfaces. If that route is chosen, then the criteria used to identify phases must deal with interface oriented notions: are phases propositions? Are they predications? Are they morphologically convergent domains? Are they command-units? Etc. I believe the motivation to look for interface tests is due to Chomsky’s claim that phases should have easily detectable semantic and phonetic properties that make them independent (see Chomsky 2004:124). Nevertheless, it must be clar-

Phases and variation: Exploring the second factor of the faculty of language

113

ified that Chomsky has never been arguing for these properties as being ‘the trigger’ for phases, as he notes through personal communication: In [Chomsky 2000] (and elsewhere) it’s suggested that phases should tend to have some reasonable characteristics at the interface levels, like being ‘propositional’ (a term that is very vague, and would have to be spelled out in some real theory of semantics, if we had one – or if there is indeed a true semantics for natural language, instead of just a general pragmatics involving language use, as I believe may be the case – rather as on the sound side). In what sense is v*P ‘propositional’? That’s actually a matter of discovery about what CI characteristics phases have. It surely won’t be anything like what formal semantics (in the sense of Frege, Tarksi, Carnap, etc.) call ‘propositional.’

To which he adds: [M]y feeling has been that phase theory should – and I think probably does – fall out from conditions of computational complexity, with interface motivation separate and ancillary, more a consequence than a cause . . . Phase theory makes LF unstateable, like D- and S-structure. That’s an obvious desideratum for optimal design, and in itself I think motivates phase theory. It leaves open what the phases are. Then comes the ancillary question of whether the phases have significant interpretations at the CI and (secondarily) SM level. That consideration has often been taken as the motivation for phase theory, but I think that has the matter backwards.

The various handicaps of defining phases on interface-based grounds are explicitly avoided from Chomsky (2001) onwards, for phasehood is then exclusively related to the Case-agreement systems – to be more precise, the phase heads are C and v*, and not others, because these are the loci of ϕ-features (see Chomsky 2005:17, 2008:154).3 In the same breath, Chomsky no longer emphasizes the role of lexical arrays (alternatively, numerations), focusing on the phase heads themselves. A consequence of this formal take on phases, as Chomsky (2007) underscores, is that it advocates a picture whereby uninterpretable morphology exists to reduce computational load by requiring quick transfer of certain units (the phases). Let us sharpen the idea that phases are motivated by computational efficiency in the sense just discussed. In his recent papers, Chomsky has argued that once a phase is completed, it is handed over to the interface levels (by means of an operation of Transfer). As Chomsky (2004:108) notes, phases cannot be transferred in full, for otherwise displacement would be impossible. He therefore restricts Transfer to the complement domain of phases, as shown in (2), where Ph and β stand for ‘phase head’ and ‘complement domain’ respectively: (2)

[ α [ Ph β ] ]

114

´ Angel J. Gallego

Chomsky (2004:108) dubs the remnant α-Ph area ‘edge,’ which includes the phase head itself and all its specifiers. The edge is what remains of the phase after Transfer applies, what is visible at subsequent derivational stages, all the rest being ‘forgotten.’ In order to formalize this idea, Chomsky (2000:108, 2001:14, 2004:108) formulates a Phase Impenetrability Condition (PIC).4 (3)

Phase Impenetrability Condition The domain of Ph is not accessible to operations (outside PhP), only the edge of PhP is

If numerations are dispensed with (as in Chomsky 2007, 2008), a question that immediately arises is how phases can ‘stretch’(i.e. have the structure we attribute to them) by simply picking C and v* up? Norbert Hornstein (p.c.) suggests this could be done by looking at the requirements of phase heads – their features. By hypothesis, C and v* contain a bundle of ϕ-features that must be valued. This implies that C and v* must at minimum select phase-mates that help them delete uninterpretable morphology.5 This issue brings us directly to the option (1c), and, more generally, to the structure of phases, something that will become most relevant below. In Chomsky (2001:14), it is argued that phases correspond to structures with the shape depicted in (4), where β must be interpreted as complex (a phrase), hence consisting of (at least) two heads. (4)

[ Ph β ]

Once stretched, (4) gives us the syntactic chassis in (5), where Ph is a phase-head (the Probe), N is a non-phase head (T or V), and G is a Goal (in Chomsky’s 2000, 2001 sense): (5)

[Ph – [N – G] ]

One can now ask whether the configuration in (5) is enough to characterize phases. I want to relate the answer to this question to the process of ϕ-feature inheritance recently put forward by Chomsky, which, as we will see, is not compatible with the logic behind Epstein and Seely’s (2002) claim that phases correspond to every application of Merge (see also Chomsky 2004:123–124 and 2007:17 for additional problems with defining phases as the external merger of two lexical items alone). To strengthen the role of phase heads, Chomsky (2008) argues that ϕ-features are generated in C and then downloaded to T, where they act as a Probe looking for a Goal. For consistency, the same idea is pushed to the v*-V relation. To minimize the search space of Probes, Chomsky (2007:22) further assumes that inheritance precedes probing:

Phases and variation: Exploring the second factor of the faculty of language

(6)

115

uF inheritance – Probe-Goal a. [ CP C b. [v*P v*

[TP T [v*P DP ]]]

[VP V DP ]]

-feature inheritance

[CP C [TP T

[v*P DP ]]]

Agree (T, DP)

-feature inheritance

[v*P v* [VP V

DP ]]]

Agree (V, DO)

Chomsky (2008) first speculates that the inheritance process may be motivated to yield the A / A-bar cut, but later on he adopts Richards’ (2007) suggestion that inheritance falls out from the conjunction of two background assumptions: (i) the PIC, and (ii) deletion of uninterpretable features. Since deletion is a part of Transfer (see Chomsky 2008), uninterpretable morphology must end up occupying the complement domain of phases, which is the only domain targeted for deletion. This can be expressed as in (7):6 (7)

Valuation by Transfer Value and Transfer of uninterpretable features must happen together

A crucial claim by Richards (2007) concerns the internal structure of phases. According to him, interface conditions (Chomsky’s 2000 Strong Minimalist Thesis) only require for there to be one non-phase head in every phase, as in (8): (8)

Ph – N – Ph – N (= C – T – v* – V)

Richards (2007) explicitly (and forcefully) argues against any of the patterns in (9) to be possible: (9)

a. b. c. d.

*Ph – Ph – Ph – Ph *Ph – Ph – N – Ph *Ph – N – N – Ph *N – N – N – N [from Richards 2007:570]

As Richards (2007) puts it: This logic implies that in any sequence of heads . . . only a single nonphase head should exist between any two phase heads; any additional nonphase heads would fail to be motivated by the SMT . . . and so their existence would not be sanctioned. What emerges, then, is a very narrowly constrained picture of ph(r)ase structure. [from Richards 2007:570]

As noted above, this Ph – N pattern has consequences for both Epstein and Seely’s (2002) take on phases (the radical option in (1c)) and the relation between cartographies and phases. Let us defer the latter issue to the next section. As for the former, notice that although (8) appears to favor Epstein and Seely’s (2002) thesis that two heads are enough to build up phases, this is not accurate: three

116

´ Angel J. Gallego

heads are actually needed, as we also need a Goal to value the ϕ-Probe. This point becomes clearer if we replace Ph by Probe in (8): (10)

a. Probe – N – Goal – Probe – N – Goal b. *Probe – N – Probe – N

Are there more grounds to support (10a) over (10b)? It is quite easy to see that some of the arguments that have been given in the literature to characterize phases will not be restrictive enough for (10a) to outrank (10b). For instance, Chomsky’s following of Marantz’s (1997, 2007) ideas on light heads (v, n, etc.) acting as tokenizers will fail, because (10b) suffices √for that. Differently put: if categorization is triggered by merging a so-called Root with a light category (`a la Marantz 1997), then the pattern in (10b) will be enough to define a phase. A better rationale emerges from Chomsky’s (2008) labeling algorithm, according to which only {X,YP} structures (where theYP element is to be regarded as a complex unit – say, {Y, Z}) ease computation, since X is the Probe by minimal search. Tacit here is the assumption that Probes/labels must be heads – simple elements. Therefore, in {X, Y} structures, minimal search will not be able to yield the correct outcome, because there are two heads.7 One other argument to prefer (10a) is provided by Cedric Boeckx (p.c.). As he observes (see Boeckx 2008c), in a structure like (11), it is impossible for the object to escape the VP under anti-locality (see Abels 2003). That is, if the object wants to abandon the v*P phase, an extra head is needed: v*. (11)

a.

b.

VP XP

XP

V’ V

v*P

t XP

v*’ v*

VP V

tXP

In sum, if one adopts either the conjunction of (3) and (7), or the notions of antilocality and minimal search, there seem to be good arguments to take phases to be at least of the form in (5), repeated here as (12) for convenience: (12)

[Ph – [N – G] ]

Once this much is assumed, one should – it would appear – agree with Richards (2007) that, because of the SMT, phases cannot expand beyond (12). However, there is empirical evidence suggesting otherwise. In the next subsection, I will try to sketch a possible way to overcome this tension.

Phases and variation: Exploring the second factor of the faculty of language

2.2.

117

On cartographic and anti-cartographic effects

The most relevant conclusion to be drawn from the preceding discussion is that Chomsky’s (2000, 2008) PIC forces one non-phase head to show up below every phase head, for inheritance to take place. This gives us a rigid ‘two-functionalheads-per-phase’ scenario that does not fit with cartographic findings. In fact, notice that it does not even fit with structures such as (13): (13)

John must have been drinking wine.

The worrying part of (13) is the must have been cluster. All other things being equal, the C-T backbone must be stretched to accommodate the additional (modal/auxiliary) heads – a single T will not be enough. There are various technical ways to proceed at this point. One is to argue that N in (12) can split in a three-way fashion, more or less along the lines of Boeckx’s (2008a) architecture (see (14) below).8 A second way to account for facts like (13) is to assume that there may be a potentially unbounded number of Ns, as long as these have some semantic content that is independent from Ph. Importantly, these Ns must not be able to probe, which means that they cannot have uninterpretable features (see 15, where I indicate the semantic content of N heads by the subscript i, a mnemonic for ‘interpretable,’ just to differentiate them from the non-phase head T):9 (14)

Ph – N

NAUX

–

NAUX

Ph – N

NAUX

NAUX NAUX

NAUX

(15)

Ph – N – Ni - Ni – Ni – Ni – . . .

It is reasonable to take (15) to embody the logic behind the Cartographic Project. The collection of N-heads could instantiate the bunch of categories that Belletti (2001), Cinque (1999), and Rizzi (1997, 2004, 2006) have identified around C, T, and v*. However, notice that from Chomsky’s streamlined take on pseudosemantic features (see Chomsky 2001, 2008), it follows that the Nis in (15) cannot act as (criterial) Probes (contra Rizzi 1997, 2004, 2006a, 2006b). In these cases, movement can be modelled by postulating a formal edge (formerly, EPP) feature requiring the creation of a specifier position (see Chomsky 2008).

118

´ Angel J. Gallego

(16)

CP -

C

ForceP InterrogativeP FocusP/TopicP ModifierP FinitenessP

Ni Ni Ni Ni Ni

T

One last option is that the additional Nis are Probes in and of themselves. But this would actually imply –under the assumptions made so far– that Ni is not an N, but a Ph, in which case we would need another N. (17)

Ph – N – Ni Ph

N

Consider (15) more closely. Given the claims made about the C-T connection in Chomsky (2007, 2008), it is worth asking whether some of the Ni heads can derive from C (or some other head, as suggested in (14)), or have an independent source. If they do not derive from some other head, then we have the purest manifestation of cartographic effects, roughly in the spirit of Pollock’s (1989) Split IP Hypothesis (i.e. we have a ‘one feature – one head’ correlation). However, on the other hand, if some of those features cluster in one head, we will have an instance of anti-cartographic effects. In a system such as Fortuny’s (2008), anti-cartographic effects would correspond to what he calls syncretic pattern, depicted in (18c) below. The options in (18) represent the different scenarios considered in Fortuny (2008): from the discontinuous pattern in (18a) to the syncretic pattern in (18c), going through the analytic pattern in (18b). Note that only (18c) corresponds to anti-cartographic effects, whereas (18a) and (18b) are both cartographic, although not in the same fashion: the former invokes a stretching strategy that resembles some featuresharing mechanism (see Giorgi and Pianesi 1997, Pesetsky and Torrego 2006, and references therein). (18)

a.

[ P]+C

b.

[ P]+Infl ...

P

c. P

P P

...

...

One question that arises is what the trigger for the stretching strategies in (18a) and (18b) could be. Descriptively, it resembles Chomsky’s inheritance, but it

Phases and variation: Exploring the second factor of the faculty of language

119

is fact different: nothing like the PIC will force cartographies to emerge, as they involve semantic information that need not be expunged. In this respect, Fortuny (2008) argues, partially building on Kratzer (1996), that there are two principles that regulate the spell-out of discontinuous (meaning ‘cartographic’) categories: (19)

Prohibition Against Tangled Structures (PATS) Two discontinuous syntactic templates [A1 . . . A2], [B1 . . . B2] cannot be broken off by yielding the tangled object [A1 . . . [B1 . . . A1] . . . B2] [from Fortuny 2008:107]

(20)

Prohibition Against Vacuous Quantification (PAVQ) For every occurrence of a variable x there must be a quantifier Q binding x, and for every quantifier Q there must be a variable x such that Q binds x [from Fortuny 2008:108]

We do not need to go into the details of Fortuny’s (2008) PATS and PAVQ here. What matters for my purposes is that this author regards them as manifestations of Chomsky’s (1986b) Principle of Full Interpretation (PFI). Fortuny takes this to indicate that studying the cartographies cannot tell us anything too deep about syntax, as we would actually be studying the Conceptual-Intentional systems: [T]he empirical results of the cartographic project do not lead us to a primitive element of the syntactic component, but rather to the study of the C-I system[s] . . . Devices such as the PATS and the PAVQ are clearly relative to the levels of interpretation of the C-I system[s], and hence one may be skeptical about coding them in the theory of grammar in the form of derivational devices or in the form of universal hierarchies . . . The theory of the syntactic component of the faculty of language need not – and therefore must not – encode devices that translate the kind of external requirements to be satisfied: the syntactic component does not have to be defined to avoid the generation of [structures violating vacuous quantification]; similarly, if cartographies derive from the Full Interpretation condition, cartographies do not reflect the knowledge of grammar. [from Fortuny 2008:112–113]

The PATS and the PAVQ are concerned with cartographic effects. What about their flipside, so-called anti-cartographic effects? Fortuny (2008) proposes that they emerge by a principle that tries to constrain feature expansion: the Maximize Matching Effects Principle (MMEP). (21)

Maximize Matching Effects Principle (MMEP) Match as many features as possible using the smallest span of structure [from Fortuny 2008:117]

120

´ Angel J. Gallego

As defined in (21), the MMEP looks like a third-factor-like principle favoring efficient (less costly) computation –not only derivationally, but also representationally, as Chomsky (1991) first argued. Fortuny (2008) invokes the MMEP when reviewing the well-known observation that T has A-bar properties in some Romance languages, a fact he reinterprets by saying that T and C can contract –T would inherit not only ϕ-features, but π-features (e.g. focus, topic, and wh) as well. The picture that emerges from Fortuny’s (2008) proposal can be summarized in (22): (22)

a.

b.

Cartographic effects: – 1 head / 1 feature (analytic pattern) – 2 heads / 1 feature (discontinuous pattern) [required by the C-I systems] Anti-cartographic effects: – 1 head / n features (syncretic pattern) [required by efficient computation principles]

Notice that, if I read him correctly, Fortuny (2008) is ultimately arguing for a theory of optionality.10 In fact, one can rephrase the intuition behind (22) by invoking Chomsky’s (2001:33–34) Fox/Reinhart approach to optional operations: (23)

Optionality of Operations Optional operations can apply only if they have an effect on the outcome: in the present case, v* may be assigned an EPP-feature to permit successive-cyclic A-movement or Int[erpretation] (under OS).

Adapted to the terms I am using, the logic in (23) holds if ‘optional operations’ are interpreted as ‘creating more structure.’ In standard cases, structures will tend to be as small as possible for convergence – only the Ph–N–G pattern, under Richard’s PIC-based argument. However, if an effect on the outcome is to emerge, structures will stretch, giving rise to cartographic hierarchies. I will assume this reinterpretation of Fortuny’s (2008) proposal in what follows. In the next section, I would like to be more explicit about how such a stretching can be implemented.

Phases and variation: Exploring the second factor of the faculty of language

121

3. Uriagereka’s F Parameter In Gallego (2007), I suggest in passing to reinterpret Rizzi’s (1997) Left Periphery along the lines of Hale and Keyser’s (1993, 1998, 2002) l-syntactic framework. Details aside, the gist was as follows: just like a DP is interpreted as, say, a theme because of the position it occupies by first-Merge, a DP should be interpreted as a focus or a topic because of the position it ocupies in its last-Merge position. To explore this possibility in more detail, here I would like to emphasize one key restriction put forward by Hale and Keyser. In their 1993 paper, these authors claim that argument structure is constrained by the X-bar schema in such a way that projections must be non-ambiguous – that is, there can be only one complement and one specifier per phrase (see Boeckx 2008a and references therein for similar bans proposed within the GB period): [E]ach lexical head X determines an unambiguous projection of its category – to a phrasal level, XP – and an unambiguous arrangement of its arguments, as specifier and complement [. . . ] We will speculate further that the unambiguous structure requirement will yield an additional limitation on the projection of categories to types: to wit, the requirement that ‘intermediate’ types (X’) be restricted to just one for any given projection [. . . ] The limitation on types follows, we wager, from the assumption that multiple ‘intermediate’ types would be linguistically (though perhaps not notationally) indistinct. [from Hale and Keyser 1993:66–67]

A simple way of interpreting this would be to subsume it under Boeckx’s (2003) Principle of Unambiguous Chains (PUC), which, once gain, must be regarded as nothing but a manifestation of Chomsky’s (1986) PFI. (24)

Principle of Unambiguous Chains Chains must be defined unambiguously [from Boeckx 2003:13]

In Boeckx (2003), chains are ambiguous if they contain more than one ‘strong’ (i.e. ‘checking’) position: a position that requires the creation of a specifier by means of an EPP feature. Among other things, this suffices to block movement from one left peripheral position to another left peripheral position (a species of A-bar hyper-raising). Although the intuition behind (24) is clear, it is not easy to put it into practice, especially when X-bar distinctions are no longer available (see Boeckx 2008a:40 and ff). Still, it is worth making sure that the same question Hale and Keyser raised forThetaTheory can be asked with respect to cartographies in the first place. That is to say, can we ask the question ‘why are there so few theta roles?’ in the case of the Left Periphery? The answer appears to be no, as cartographic projections are much richer than theta-roles (see, most

122

´ Angel J. Gallego

notably, Cinque’s 1999 typology of positions, and Boeckx 2008a as an attempt to narrow it down). This simple observation tells us that what is needed before addresing the parallelism with Hale and Keyser’s (1993, 1998, 2002) l-syntax and Rizzi’s (1997, 2004, 2006a, 2006b) Left Periphery is to narrow down the cartographic distinctions. The possibility that different cartographic units can collapse in one dedicated head was an original insight of Juan Uriagereka and his FP (F as a mnemonic for ‘further’), a syntax-pragmatics encoding projection that aimed at capturing the so-called ‘hot’ behavior of Romance languages. Consider the rationale for Uriagereka’s FP:11 [I]n my 1988 thesis . . . I was lead to postulate a category in the periphery of the clause which, to my surprise, seemed related to evidentiality. My argument was comparative, concerning Portuguese and Spanish clitics . . . I had to assume some functional element F . . . that was either morphologically there (Portuguese) or not (Spanish), which would then drive the verb past the clitics in root clauses, but not in embedded clauses where it could be hosted by governing elements . . . I took the facts to be even more radical in French, where the postulated F was not present in the (overt) syntax, with various consequences for the thus impoverished clitic system . . . That sort of phenomenology has been found in other languages . . . Once the formal need for F is postulated, one wants to know why it is there . . . In 1988, and then again in 1995, I suggested F may be somehow triggering the often called ‘hot’ behavior of languages: topics, foci, and the like . . . Needless to say, semantically, a focus and a topic are opposite extremes, which suggests two things. Whatever F is doing, it can’t be trivially semantic, as it is hosting both focus and topic material. Second, we should ask: do focus and topic share anything in common? . . . Topic and focus share the expression of an attitude. A topic or a focus are that for someone, the speaker or an embedded subject (with verbs that allow a perspective). In this respect, topics and foci are both evidential, and different from phrases with existential import. Importantly, also, ‘hot,’ attitudinal material is normally displaced to the left periphery, whereas ‘cool’ or declarative material never is. I took that to be important, and related to what I called F. [from Uriagereka 2002b:3]

The trees in (25) illustrate, quite precisely, what postulating some ‘broad spectrum’ functional category like Uriagereka’s F can buy us:

Phases and variation: Exploring the second factor of the faculty of language

(25)

a.

CP

b.

123

CP

ForceP C

C TopicP FurtherP FocusP FinitenessP

As Raposo and Uriagereka (2005) point out, empirical facts support the (25b) option. If the CP is to be split into projections that have their own ‘independent life’ (as one would be led to believe, see Rizzi 1997, 2004, 2006a, 2006b), then it is unexpected that, both diachronically and synchronically, the possibility of displaying focalization or topicalization is related to either clitic climbing or the presence of overt focus heads. Differently put: at first glance, it is not expected that the availability of some operations is parasitic on others. This species of ‘domino effect’ is straightforwardly captured if left peripheral operations target one and the same projection (with the restrictions just noted arising as a consequence of the morphological strenght of this projection).12 However, it is not so obvious how to do that if: “no free preposing and adjunction to IP is permissible, all kinds of movement to the left periphery . . . be[ing] motivated . . . by the presence of a head entering into the Spec-head configuration with the preposed phrase” (see Rizzi 1997:282, 287).13,14 This said, it must be mentioned that the cartographic studies do have ways of capturing these ‘domino effects.’ One such way is by widening the number of features that a given head can host (Fortuny’s 2008 syncretic pattern, see Bianchi 1999), and another one is by making a given left peripheral head H1 parasitic on a higher (or lower) left peripheral head H2 (see Haegeman 2002, Rizzi 1994,1995, and references therein). The first option is blatantly stipulative, and in fact goes against the spirit of the first cartographic paper: criteria satisfaction involves a peripheral head endowed with a dedicated feature (though see Rizzi 2006a:128 for the idea that, in some cases, criterial heads can cluster via head movement). The second possibility is usually dubbed truncation (or pruning), and has been explored in the context of language acquisition. Though much more interesting, the implementation of truncation is not entirely clear. Consider the gist of the idea: To account for the phenomenon of root infinitives in child language, for instance, Rizzi proposes that, in the early grammar, TP can be missing. The truncation of the projection TP entails that the projections AgrP and CP, projections dominating TP, must also be absent (see Rizzi 1994, 1995). The assumption is that the system does not allow one to selectively omit a projection, say FP, and then to project an extended higher projection of FP. [from Haegeman 2002:11]

124

´ Angel J. Gallego

At first glance, truncation looks like a counterpart of Chomsky’s (2001) defectiveness, but it is not quite. Unlike defective heads in Chomsky’s (2001, 2008) system, trunction works upwards, which just complicates any device designed to capture it: whatever that may be, it will have to assume unusual locality metrics, as somehow (higher) H1 depends on (lower) H2 being projected. Things are more natural under Chomsky’s option, since the relevant parasitic relationship obeys head-complement depedencies.15 Finally, observe the pair in (26), taken from Rizzi (2006b), who in turn builds on Grimshaw (1997) (see Stowell 1981 for original discussion of these and related data). As can be seen, the examples in (26) show that operations such as topicalization or Neg-fronting depend on the C head being present: (26)

a. b.

She thought (*that) this book, you should read. She thought (*that) never in her life would she accept this solution.

This, yet again, indicates that there is some parasitic relationship between different criterial heads and C. Although one can accommodate these facts as I have just indicated (through some kind of ‘feature-switching’ mechanism a` la Bianchi 1999 or through truncation a` la Rizzi 1994), I take all these facts to reinforce a non-cartographic approach, whereby one head subsumes different left-peripheral patterns.16 Let us go back to FP. Building on Uriagereka (1988, 1995a, 1995b), Raposo and Uriageraka (2005) propose (27) to trace the varying parametric behavior of this projection within Romance. The first cut is a syntactic one (whether F is projected or not), while the second one tries to express how rich F’s morphology is. As I did in Gallego (2007), I will dub this the F Parameter. (27)

The F Parameter F Overt F effects [+syntactic] Western Iberian [+morphological]

Covert F effects (French) [– syntactic]

Eastern/Central Iberian [– morphological]

[from Raposo and Uriagereka 2005: 644] If cartographic effects can be explored through a unique dedicated projection, it is tempting to go even further and try to entertain the idea that that option is also too much of a stretch. To make my point I want to capitalize on the fact that these effects are not restricted to CP, as in the seminal paper by Luigi Rizzi:

Phases and variation: Exploring the second factor of the faculty of language

125

they arise almost in every projection. If that is the case, then I think postulating a sole category (or a bunch of them, for that matter) is merely an indication that we are missing something.17 Uriagereka (2002b) is explicit about this matter: The problem is that pragmatic phenomena appear all over the place, even in unexpected domains . . . I wanted to discuss th[is] . . . because [it] force[s] us in one of two directions. The first posits elements like F every time there is a pragmatic effect. If we witness it internal to nominals, so be it; the same holds if we witness internal to various layers of VP. Of course, inasmuch as this sort of phenomenon is mainly peripheral to a clause-like domain, there should be something clausal to DPs, including adjectival modification, various layers of VPs, and possibly other contexts, aside from the obvious clausal periphery associated to the IP/CP juncture. I would thus call this approach ‘generative semantics,’ as it has to postulate hidden clausal structure all over, even when there is otherwise nothing clausal to relevant domains. Another approach is also possible: there is no F involved in these pragmatic instances. Rather, what is at issue is a PF prominence that phrases gain in a class of contexts, as a result of which they affect pragmatic interpretation. This view is close to the one sketched in Chomsky (2001). This characteristic, call it surface, interpretation arises at the border of phases. [from Uriagereka 2002b:8]

Uriagereka (2002b) makes a fair point: do we have to postulate an extra category every time we find a semantic effect? Surely, that is consistent with the SMT, as the SEM component will receive all it needs (or quite) from the syntax, as cartographic researchers often claim when they assert that “[t]he relation between syntax and the interpretive interface (LF) is expressed in an optimally simple way: the interpretation is read off the syntactic configuration” (Belletti 2004:17). But the thing is that syntactic configurations are so loaded with pseudo-semantic make-up that the question arises whether we really need an independent SEM component. If semantic specifications get wrapped around the syntactic backbone, as the Cartographic Project claims, then one could get rid of the SEM component altogether (see Hinzen 2006). Another way of going about these issues is to stick to the look-ahead-less view that stems from Chomsky’s autonomy of syntax thesis (see Chomsky 1955[1975], 1957, 1965), and take semantic effects to be a side-effect of syntactic computation. Under this approach, semantics is determined by computation, exactly “as if syntax carved the path interpretation must blindly follow” (from Uriagereka 2002a:275). As mentioned above, I want to explore the possibility that we can dispense even with something like F. I will do that by pursuing Boeckx’s (2008a) observation that some information is found – as Uriagereka notes for F – at different phrasal points:

126

´ Angel J. Gallego

[S]everal authors have already observed that various units such as Topic, Focus, and Agreement appear to be expressed at various clausal junctures (C-level, Infllevel, and V-level) . . . For example, what was once AGRP (Pollock 1989) was soon decomposed into AGRsP (for subject agreement), AGRoP (for object agreement), and ‘AGRcP’ (for complementizer agreement (Shlonsky 1994, Roberts and Cardinaletti 1991, 2002)). This amounts to treating AGRP as an iterated category, along the clausal skeleton (extended projection of V), in a way very reminiscent of the distribution of adverbs in the basic X-bar schema . . . A similar phenomenon of iteration may be at work in the context of negation. The most comprehensive cartographic study of negation I am aware of is Zanuttini 1997. Zanuttini examines the locus of negative expressions in Romance languages, and finds quite a bit of variation, ultimately leading her to postulate four distinct Negation heads in the clausal structure. Specifically, taking the inflected verb (in Inflo ) as her reference point, Zanuttini posits a pre-verbal NegP and three post-verbal NEGPs . . . [T]he similarity between NEGP and AGRP is very worth exploring some more. Both appear to act like adjoined categories, grafted onto the clausal skeleton . . . The framework developed here, which is nothing more than exploiting the X-bar schema in the context of groupings of projections, can make sense of the odd behavior of AGRP and NEGP. Both are ‘outside’ the clausal skeleton, on a parallel plane (like adjoined material typically is). Their somewhat erratic linearization pattern comes from the fact that they can be ‘collapsed’ onto the clausal skeleton at various junctures. [from Boeckx 2008a:104-110]

Let us recap so far. In this section I have expressed some problems that arise if one tries to push the basics of Hale and Keyser’s l-syntactic framework to the Left Periphery: there are too many peripheral distinctions, which would first need to be grouped somehow (something I have not done). Consequently, I have suggested, following Uriagereka’s work, that a more reasonable way to go about cartographic effects is to encode them in a unique ‘broad spectrum’ functional pojection responsible for discourse-semantics effects. But that also appears to be dispensable. In the next section, I would like to exploit phase-based technology to capture cartographic effects. In particular, I want to recast Boeckx’s (2008a) observation by taking agreement as the key to understanding the cross-linguistic properties of fronting. Therefore, instead of postulating an additional head right below C, v*, D, and possibly other heads, as depicted in (28), I want to rely on morphological richness as the key trigger for what I am calling cartographic effects: (28)

CP a. C

XP TP

v*P b. v*

XP VP

DP c. D

XP NP

Phases and variation: Exploring the second factor of the faculty of language

127

To be specific about it, I put forward (29): (29)

F-effects are triggered by rich overt agreement morphology.18

Under (29), one can exploit the traditional observation that non-configurational languages have richer Case-agreement systems. As I will argue, this is indeed the case for Romance family, which further splits into two groups: Western Romance (Galician, European Portuguese, and Spanish) and Central Romance (Catalan, French, and Italian) languages. If the account I have just sketched is on the right track, the parametric variation between these two groups must follow from them having richer Case-agreement systems, and not necessarily from such or such extra head being projected at some phrasal spot.Another advantage of (29) is that it allows us to recast Fortuny’s (2008) proposals precisely as in (23). Thus, Chomsky’s (2001) optionality need not deal with projections, but just with operations. This is a welcome result, because one might be tempted to blindly relate optionality of projections to parametric variation, a claim often made in the cartographic literature, when a given language/dialect is said to lack a given projection.

4. ‘Ph’ as an allomorph and cross-phasal effects Towards the end of the last section, I relied on both Uriagereka’s (2002b) and Boeckx’s (2008a) observation that the same type of information may ‘pop up’ at different structural points. It is worth pointing out that this idea is not new: it became popular in the early nineties, when Chomsky postulated a second Agr(eement) projection at the top of the VP in order to account for the participal agreement facts studied by Kayne (see Kayne 1993). The interest behind that claim was that it allowed one to analyze structural Case assignment in a uniform manner: subject and object DPs had to move to SPEC-AgrS and SPEC-AgrO respectively to get their Case feature checked.19 (30)

a.

AgrSP Subject

b.

AgrS’ Agr S

Object TP

T

AgrOP AgrO’ AgrO

...

VP Subject V

V’ tObject

Though appealing and theoretically elegant, the parallelism in (30) was abandoned the moment Agr-nodes were questioned on interface grounds by Chom-

128

´ Angel J. Gallego

sky (1995). Note, nonetheless, that Phase Theory opens this door again by burying ϕ-features within v* and C. In this paper, I want to explore the idea that phase heads can be regarded as contextual variants of the same element – that is, they are allomorphs. Call the relevant element, as above, Ph. Adopting (29), I also want to explore the idea that a morphologically richer Ph correlates with active (= ‘hot’) creation of specifiers (see Gallego 2007 and Uriagereka 1988, 1995a, 1995b, 2008). I will assume this richness is encoded in the lexicon, as it clearly cannot follow from C-I demands, or Chomsky’s SMT.20 (31)

C and v* are contextual versions of Ph.

If C and v* are nothing but the result of having Ph in a particular morphological guise, then we predict it could show the same syntactic behavior in an across-theboard fashion. This means that if C displays an active Left Periphery, v* will do so too, at least in principle. This is in fact the claim made in Poletto (2006), where cross-phasal parallelisms dealing with topicalization and focalization are found in Old Italian. I want to extend Poletto’s observation to present day Romance, not to argue that some projections happen to be projected in the Left Periphery of CP and v*P, but to argue that, since C and v* are the very same element, they should behave alike. Before moving on, let me sharpen (29) as follows: (32)

A Morphological Parameter If X is morphologically richer, it can exploit its edge (= Left Periphery)

I would like to start by considering some Spanish data noted by Esther Torrego in unpublished work back in the eighties (see Torrego 1980). At the time, Torrego was studying examples like the ones in (33a), where constituents appear to undergo focus-fronting to the CP layer. Importantly, the movement here is not analogous to contrastive focalization, in Rizzi’s (1997) sense, but has an affective/colloquial-like intonation: the contrast is (33) is clear to all speakers I have consulted.21 (33)

a.

b.

Mucha tonter´ıa dice el Gobierno. much crap say-3.SG the government ‘The Government says a lot of crap’ ´ dice MUCHA TONTERIA el Gobierno! much crap say-3.SG the government ‘The Government says a lot of crap’

(Spanish)

(Spanish)

Phases and variation: Exploring the second factor of the faculty of language

129

In (33b) I use capital letters to indicate that the focused constituent is pronounced in a way that the sequence is understood as involving a contrast (so, in the previous context, someone said something that is being qualified or denied). In Gallego (2007), I called the cases in (33a) and (33b) mild and contrastive focus, with no commitment whatsoever to these patterns requiring a different analysis, or suggesting the existence of two types of left peripheral focus – if that is the case, I do not know (see Irurtzun 2007 for discussion). What matters for my present concerns is that the availability of (33a) is severely parametrized. Similar examples are the following: (34)

a.

b.

c.

d.

Por algo ser´a. (Spanish) for something be-FUT.3.SG ‘It must be for some reason’ Para m´ı querr´ıa yo esos problemas. (Spanish) for me want-COND.1.SG I those problems ‘I wish all I had to worry about was that’ Poco dinero es ese, creo yo. (Spanish) little money be-3.SG that, think-1.SG I ‘That is not much money, I think’ Mucha tonter´ıa dices. (Spanish) much crap say-2.SG ‘You say a lot of crap’ [from Gallego 2007:157]

The mild focalization in (33a) and the paradigm in (34) are not available in all languages. Catalan, as (35) shows, systematically rejects it. (35)

a. *Per alguna cosa deu ser. (Catalan) for some thing must-3.SG be-INF ‘It must be for some reason’ b. *Per a mi voldria jo aquests problemes. (Catalan) for to me want-COND-1.SG I these problems ‘I wish all I had to worry about was these problems’ c. *Pocs diners s´on aquests, crec jo. (Catalan) little money be-3.SG those think-1.SG I ‘That is not much money, I think’ d. *Molta ximpleria dius. (Catalan) many crap say-2.SG ‘You say a lot of crap’ [from Gallego 2007:158]

130

´ Angel J. Gallego

I know of no account for the contrast in (34) and (35). From a cartographic point of view, one could account for it by saying that Catalan lacks a particular left peripheral position – say, MildFocusP. But one should be willing to look for an alternative, as this one would not give us much: it would only increase the typology of foci (a controversial issue, see Irurztun 2007:97 and ff. for arguments against there being more than one type of focus), and, more importantly, it would be nothing but a descriptive observation, similar to saying that Catalan lacksVSO declarative sentences. If it is finally concluded that Catalan lacks one position Spanish exploits, then the questions arise right at that point. Before trying to say something about the asymmetry above, let me incorporate one final piece into the puzzle. In Gallego (2007), I related the type of fronting (34) illustrates to the type of fronting of VOS sentences. In particular, and departing from static analyses, I argued that the different authors that have worked on this topic (see Belletti 2001, 2004, Cardinaletti 2001, Costa 2000, 2002, Ord´on˜ ez 1998, 2005, among others) were all right in arguing that there are two strategies to derive VOS: object shift (see Cardinaletti 2001, Costa 2000, 2002, and Ord´on˜ ez 1998, 2005) and VP fronting (see Belletti 2004):22 (36)

a.

v*P Object

b. v*’

Subject

XP

[VP V Object] v*’

v*

X’ X

VP V

VOS qua Object Shift

v*P

Subject tObject

v*’ v*

tVP

VOS qua VP-fronting

What I added to the picture in (36) is that both strategies are available, but appropriately parametrized. Western Romance languages (Galician, European Portuguese, and Spanish) display the object shift option, whereas Central Romance language (Catalan and Italian) choose the VP-fronting option. The binding data in (37) and (38) below support this conclusion: (37)

a.

Ayer visit´o a cada chicoi sui mentor. his mentor yesterday visit-PAST.3.SG to each boy ‘His mentor visited each boy yesterday’

(Spanish)

b.

Recogi´o cada cochei sui propietario. pick-up-PAST.3.SG each car its owner ‘Its owner picked each car up’

(Spanish)

Phases and variation: Exploring the second factor of the faculty of language c.

(38)

131

(Spanish) No rega˜no´ a ning´un ni˜noi sui madre. child his mother not scold-PAST.3.SG to no ‘His mother did not scold any child’ [from Gallego 2007:244,245]

a. ??Ahir va visitar cada yesterday go-3.SG visit-INF each professor. teacher ‘His teacher visited each student’ b. ??Va recollir cada cotxei go-3.SG pick-up-INF each car ‘His owner picked-up each car’ c. ??No va renyar cap neni not go-3.SG scold-INF any child ‘His mother did not scold any child’

estudianti el seui student the his

(Catalan)

el seui propietari. the his owner

(Catalan)

la sevai mare. the his mother

(Catalan)

[from Gallego 2007:245,246]

As predicted, Italian behaves like Catalan in that it disallows the derivation in (36a), barring binding as in (37). This has been argued for by Belletti (2004):23 (39)

A: Chi ha salutato Gianni? (Italian) who have-3.SG greeted Gianni ‘Who greeted Gianni?’ B: *Hanno salutato Giannii i proprii genitori. (Italian) have-3.PL greeted Gianni the own parents ‘His own parents have greeted Gianni’ [from Belletti 2004:36]

Gallego (2007) further relates the cut concerning the VOS strategy to another well-known observation made in the literature: Central Romance languages disallow VSO sentences, while Western Romance languages license them. (40)

*Fullejava en Joan el diari. (Catalan) browse-PAST.3.SG the Joan the newspaper ‘Joan was browsing the newspaper’ [from Picallo 1998:229]

(41)

*Ha comprato Maria il giornale. have-3.SG bought Maria the newspaper ‘Maria has bought the newspaper’

(Italian)

[from Belletti 2004:26]

132

´ Angel J. Gallego

Here I follow Ord´on˜ ez (2005) in that, in VSO sentences, the subject is not in its base position. In his analysis, Ord´on˜ ez (2005) argues that the subject undergoes displacement to SPEC-T (as he assumes an AgrS P above TP), but we can take his SPEC-T to be an outer-SPEC-v*. I will assume so, which forces me to adopt an analysis whereby the external argument (EA) is generated below v*, as has been proposed, for varying reasons, by different authors (see Gallego 2007, Mayr 2007, Sigurðsson 2006, etc.):24 (42)

v*P v*

VP

Subject (EA) V

V’ Object (IA)

The link between the strategy to derive VOS and the availability of VSO in turn brings us to Bobaljik and Jonas’ (1996) hypothesis that languages that display object shift license a third subject position (their SPEC-TP Parameter). In Gallego (2007) I phrase this connection as in (43):25 (43)

VOS – VSO Generalization If a language L generates VOS through object shift, then it licenses VSO [from Gallego 2007:265]

(43) falls into place if objects in VOS and subjects in VSO move to the same projection: v*. More generally, it must be noted that whatever is going on in (37) through (41) must have to do with this head. And, for consistency, whatever is going on in (34) and (35) must have to do with C. Granted, this parallelism could be a mere coincidence if (mild) focus fronting and object fronting were independent operations, targeting independent projections, but things are different from a phase-based perspective (see Poletto 2006). If something along the lines of (31) (repeated below as 44 for convenience) can be entertained, then the parallelism is not surprising at all, since C and v* are the very same element: Ph.26 (44)

C and v* are contextual versions of Ph.

We are now in a position where a key question must be addressed: why do C and v* behave like this, showing parallel edge-fronting properties? I would like to provide an answer based on the morphological parameter in (32), which I want to

Phases and variation: Exploring the second factor of the faculty of language

133

relate to tense morphology (see Biberauer and Roberts 2006 and Gallego 2007). If this is correct, then it must be the case that the tense paradigm of Catalan be weaker than that of Spanish. This prediction is borne out, as Catalan lacks the simple past tense (it is not used in present day Catalan). It has been replaced by a periphrastic version formed by the combination of the verb anar ‘go’ plus the relevant infinitive. (45) 1SG 2SG 3SG 1PL 2PL 3PL

Catalan tense paradigm Analytic simple past tense Jo cant´ı I sang-1.SG Tu cantares You sang-2.SG Ell/Ella cant`a He/She sang-3.SG Nosaltres cant`arem We sang-1.PL Vosaltres cant`areu You sang-2.PL Ells/Elles cantaren They-MASC/FEM sang-3.PL

Periphrastic simple past tense Jo vaig I go-1.SG Tu vas You go-2.SG Ell/Ella va He/She go-3.SG Nosatres vam We go-1.PL Vosaltres vau You go-2.PL Ells/Elles van They-MASC/FEM go-3.PL

cantar sing-INF cantar sing-INF cantar sing-INF cantar sing-INF cantar sing-INF cantar sing-INF

[from Gallego 2007:164] The proposal I have just sketched is reinforced by data concerning the indicative– subjunctive distinction within Romance languages. As is well-known, the latter has a weaker tense paradigm as well (see Gallego 2007:chapter 3 for details). Interestingly, subjunctive C appears to preclude the kind of focalization Torrego (1980) studied. This was first noticed by Torrego and Uriagereka (1992):27 (46)

a.

Juan dijo [CP C que [TP muchas cosasi pro (Spanish) Juan say-PAST.3.SG that many things hab´ıa visto ti ] ] ]! had-3.SG seen ‘Juan said that a lot of things he had seen!’ que [TP muchas cosasi pro (Spanish) b. *Juan quer´ıa [CP C Juan want-PAST.3.SG that many things viera ti ] ] ]! see-SUBJ-3.SG ‘Juan wanted a lot of things for him to see!’ [from Torrego and Uriagereka 1992: 17]

134

´ Angel J. Gallego

I take the data in (46) to indicate that, much like Central Romance C-v*, subjunctive C cannot exploit edge (left peripheral) fronting, due to tense defectiveness. Let us conclude. In this section I have argued that it is possible to capture parametric variation by taking morphological evidence seriously (under (32)). Therefore, it is not necessary to add or remove functional projections within the CP phase (be these pseudo-semantic, as in the Cartographic Project, or not); all that is needed is to consider the role played by phase heads and their morphological properties. An interesting corollary from the present study concerns the notion of phase, and the possibility that parallel effects (in the sense of Poletto 2006) are telling us something deep about the architecture of the system. I have followed that route, taking C and v* to be contextual manifestations of the same abstract category: Ph(ase). Pursuing this line, one final question is in order: what about other phases? In section 2.1. I largely kept to the strict scenario of Chomsky (2001, 2004, 2007, 2008), where only v*P and CP count as phases. The recent literature offers interesting arguments to extend the inventory to DP (see Svenonius 2004) and PP (see Abels 2003, Kayne 2004, and Raposo 2002). I would like to briefly focus on the DP case, taking into consideration the fact that adjectives can show up in both prenominal and postnominal position in Spanish. The examples in (47) show the (prenominal) D – Adj – N pattern: (47)

a.

b.

c.

d.

e.

Un elegante vestido. an elegant dress ‘An elegant dress’ Una interesante novela. an interesting novel ‘An interesting novel’ Un majestuoso edificio. a majestic building ‘An outstanding building’ Un famoso actor. a famous actor ‘A famous actor’ Un complicado art´ıculo. a complicated paper ‘A difficult paper’

(Spanish)

(Spanish)

(Spanish)

(Spanish)

(Spanish)

[from Bosque 2001:18,21]

Phases and variation: Exploring the second factor of the faculty of language

135

As can be seen in the previously unnoticed paradigm in (48), Catalan shows adjective placement restrictions similar to the ones we have been witnessing so far:28 (48)

a. *Un elegant vestit. an elegant dress ‘An elegant dress’ b. *Una interessant novella. an interesting novel ‘An interesting novel’ c. *Un majestu´os edifici. a majestic building ‘An outstanding building’ d. *Un fam´os actor. a famous actor ‘A famous actor’ e. *Un complicat article. a complicated paper ‘A difficult paper’

(Catalan)

(Catalan)

(Catalan)

(Catalan)

(Catalan)

Interestingly, Bosque (2001) analyzes the D – Adj – N pattern as involving Adjfronting from its base (rigthward) position in order to account for the interpretive effect that such an operation has (the DP is interpreted as specific, but we can ignore this now). If this analysis is correct, then we would have found yet another context where Ph’s edge (left peripheral) fronting is boosted due to morphological reasons. Putting all the pieces together, the final picture is as depicted in (49): (49)

a.

CP

b.

v*P

C’ C

c.

DP

v*’ TP

Active fronting to C’s edge

v*

D’ VP

Active fronting to v*’s edge

D

NP

Active fronting to D’s edge

This not only reinforces (31) and (32), but it in fact widens its coverage, as it should be modified as follows:

136

(50)

´ Angel J. Gallego

C, v*, and D are contextual versions of Ph.

It remains to be investigated to what extent the parallelism between C, v*, and D is accurate, and whether it can be pushed to other categories. This line of inquiry is interesting inasmuch as it forces us to think about the general architecture of the system, the way lexical items map into structures, and the role played by morphology in capturing language variation.

5. Concluding remarks In this paper I have discussed the relation among phases, cartographies, and linguistic variation. Assuming the Chomsky-Richards argument for limiting the span of structure that must be projected within a phase, I have considered the legitimacy of sandwiching additional projections somewhere in the basic Ph – N pattern. As I have argued, nothing in the framework of phases precludes insertion of additional non-phase heads, as long as these make a semantic contribution of their own (independent from that of phase heads). That would suffice to license patterns like (51), with the Ni formative being capable of cloning itself again and again: (51)

Ph – N – Ni1 – Ni2 – Ni3 – Ni4 – . . . Nin

Though legitimate in and of itself, the possibility in (51) is dubious at different levels. First, there is a series of issues that deal with restrictiveness matters: how many Ni can be projected? Do they cluster somehow? Can (pseudo-)semantic notions (e.g. topic, force) be given syntactic form? How is their order established? I have suggested some ways to go about these questions, but I have not provided any articulated answer (Boeckx 2008a and Fortuny 2008 do). A second concern has to do with the dependencies established between the extra heads themselves: is Ni32 parasitic on Ni31 (or Ni4 ), and, if so, what does it mean? With respect to this point, I have underscored Raposo and Uriagereka’s (2005) observation that some left peripheral operations go hand in hand: focalization and topicalization show similar restrictions, indicating that their trigger must not be as independent as their separate functional projections could lead us to believe. In the same vein, the possibility for Ni5 to be somehow dependent on Ni7 brings Rizzi’s truncation back to the fore, posing implementation doubts that are partially related to Chomsky’s defectiveness. A third (and perhaps more pressing) problem follows from the fact that notions such as focus, topic, and the like have been identified within virtually any projection, a claim that has been used to defend a full-blown cartographic

Phases and variation: Exploring the second factor of the faculty of language

137

analysis not only for CPs, but for v*Ps and DPs as well. It may be well the case that this is the correct way to approach the data, but I think one should be suspicious, especially when we see the same phenomenon emerge in different (in fact, in most) domains. Typically, this should be interpreted as a sign that we are dealing with something very general, as general as Lasnik and Saito’s (1992) reinterpretation of different transformational rules as Move-α. I insist: this is by no means to deny the particular properties that have been attributed to such and such left peripheral operations, but it should make us reflect on whether identifying those facts gives us an answer that goes beyond descriptive purposes. Uriagereka (2002b) is one of the most insightful works when addressing questions like the ones I have just pointed out, as it casts doubt not only on a rampant approach to cartographies, but also on a more restrictive one, embodied by the FP projection of Uriagereka (1995a, 1995b). This paper has followed the latter route, arguing that we can dispense with FP if we piggy back on the role played by morphology. This theoretical option has been illustrated by considering data from Romance. In particular, I have followed Poletto’s (2006) ideas about the parallel behavior that Old Italian manifests in both v*P and CP. As I have shown, the same scenario can be found in Western and Central Romance languages, whose edge (left peripheral) fronting can be boosted by morphological tense richness. I have exploited this to argue that v* and C are nothing but contextual manifestations of the very same element, namely ‘Ph(ase)’ –that is to say, to argue that v* and C are allomorphs of Ph.29 Does the same story apply to other languages? Clearly (and somewhat puzzlingly), it does not. English and languages that do not manifest active fronting (so-called ‘cold’ languages), will not be able to exploit Ph’s edge. What do they resort to, then? For more than ten years, Juan Uriagereka (see Uriagereka 1999a, 2002a, 2002b, 2008) has been arguing that the cut may be related to the Spell-Out strategies a given language has. Languages like Basque would resort to a radical Spell-Out strategy (see (52a)) allowing it to literally cash out chunks of structure (complex specifiers) to the interfaces, keeping a null resumptive element as a placeholder; crucially, Basque (and similar languages) will have a richer morphological inventory, which will be the trigger of active edge (left peripheral) fronting. On the other hand, languages like English will not entirely cash out the relevant chunk of structure: it will be kept, albeit in a diminished way (as a giant compound). It is precisely this latter strategy, dubbed conservative Spell-Out by Uriagereka (1999a) (see (52b)), that would be used as a cue to assign that very specifier two types of semantic interpretations: deep (theta-oriented) and surface (discourse-oriented) (see Chomsky 2001 for related discussion).

138

´ Angel J. Gallego

(52)

a.

XP YP [DEEP]

YP

b. X’

X

WP

XP [ … YP …] [DEEP] [SURFACE] X

X’ WP

Y’ Y [SURFACE]

What the structures in (52) intend to show is that complex specifiers manifest a different behavior depending on the Spell-Out version that applies to them. The consequences of the two strategies illustrated in (52) are the following: (53)

a. Radical Spell-Out languages

b. Conservative Spell-Out languages – trivial chains impossible – trivial chains possible – deep int. assigned to base – deep int. assigned to base copy copy – surface int. assigned to – surface int. read off of highest copy phrase marker [adapted from Uriagereka 2002b:14]

As Uriagereka (2002b) points out, in conservative languages, YP is not split in its way to the interpretive components. It should then be possible, in those languages, to have trivial chains with both deep and surface interpretation associated to them, since we can reconstruct deep and directly read surface characteristics from a syntactic object that is still unified. In radical languages, nothing along these lines is possible, for each spelled-out chunk follows its path to performance, and unification takes place only then, due to a dynamic system of Case-agreement. I would like to conclude by highlighting that, although Uriagereka’s MultipleSpell-Out proposal and Chomsky’s Phase Theory are ultimately saying different things, they are not incompatible. Actually, if what I have discussed here proves tenable, they can in fact converge: all we need is to study the morphological endowment of Ph, and its consequences for syntactic variation (or Spell-Out strategies). Hopefully, that will give us a way to give parameters (Chomsky’s 2005 second factor) an important role within minimalism (something that, at present, they do not have), a role that would ideally shed some light not only on how languages vary (a rather traditional concern of linguistics), but also on how the first factor (UG) may be deduced from third factor principles, as Chomsky (2004, 2005, 2007, 2008) invites us to consider.

Phases and variation: Exploring the second factor of the faculty of language

139

Notes ∗

1.

2.

3. 4. 5. 6.

7.

8.

This paper was presented at The 2nd Brussels Conference on Generative Linguistics (Alternatives to Cartography), whose audience I thank for questions and discussion. I also thank two anonymous reviewers and Jeroen van Craenenbroeck for comments and helpful observations on a previous version of this paper. I am especially indebted to Cedric Boeckx, Noam Chomsky, Aritz Irurtzun, Marc Richards, and Juan Uriagereka. Usual disclaimers apply. This research has been partially supported by grants from Ministerio de Educaci´on y Ciencia-FEDER (HUM2006-13295-C0202), and from Generalitat de Catalunya (2005SGR-00753). A reviewer is concerned about my use of the label cartographic effect, which (s)he thinks I use “to refer to the observation that languages differ as to (at least overtly) moving wh-elements.” The intended use is certainly not that. What I want to emphasize is the possibility for a given domain (e.g. the CP) to manifest a one to one correlation between features and projections. See section 2.2. for elaboration. A reviewer thinks it is not accurate to relate parametric variation to the second factor, for it is the data that the child will have access to in order to decide what is universal or parametric that is within the second factor. I agree that it is the input that a child has that will correspond to Chomsky’s second factor, but what should be clear is that variation cannot be, by hypothesis, either in the first factor or the third one – it must therefore be part of the second factor. The ultimate question that this raises, of course, is how to implement variation within minimalism. For additional problems both with phases and the properties typically attributed to them, see Boeckx (2007), Boeckx and Ghromann (2007), and Gallego (2007). There are two (non-trivially distinct) formulations of the PIC. See Richards (2004, 2006) for ample discussion. I am putting aside whether there are analogous semantic features (say, theta-roles) that require the merger of DPs with v* or C. Collapsing feature valuation and feature deletion at the Transfer point solves the problems noted by Epstein and Seely (2002). As these authors point out, there cannot be a delay between valuation and Transfer: that was a core property of Chomsky’s (2000, 2001) Probe-Cycle framework, where ϕ-features were generated in T. Note that if valuation takes place as soon as T (the ϕ-Probe) is introduced, then the system will not be able to distinguish interpretable from uninterpretable features by the time Transfer takes place. Many proposals have been made to solve this {X, Y} situation, which arises in the first step of every derivation (see Chomsky 1995, Fortuny 2008, and especially Guimar˜aes 2000), but none of them seems to be completely satisfactory. Note that a similar problem is posed by {XP, YP} structures, explored at length by Moro (2000, 2007). The way auxiliaries and modals combine suggests a more careful analysis. As is known, there are local restrictions indicating valuation processes between Aux2 and Aux3 in a sequence like Aux1-Aux2 -Aux3 (see Chomsky 1955[1975], 1957). These

´ Angel J. Gallego

140

possibilities may indicate that NAUX heads combine pretty much like clitics do, with movement restrictions and valuation effects that may result from the way they combine (in a cluster or split fashion, see Ord´onez ˜ 2001). I cannot investigate these issues here. 9. I am tacitly assuming that T has no semantic content of its own – it merely acts as a ϕ-feature container. See Chomsky (2007) and Pesetsky and Torrego (2001, 2004) for alternative views. 10. And Rizzi (1997) is too, at least when he argues that “the topic-focus system is present in a structure if ‘needed’ ” (Rizzi 1997:287–288). Unless empirical evidence indicates otherwise, Rizzi (1997:314) assumes the principle in (i) holds: (i)

Avoid structure.

As I see it, (i) plays – in Rizzi’s (1997) system – the role of Fortuny’s (2008) MMEP. 11. A reviewer asks what the feature composition of F is, and how it differs from C’s. In Raposo and Uriagereka (2005) it is suggested that F contains different types of features: (i) semantic features (related to F’s role in connecting syntax with discourse and contextual information), and (ii) formal features (both ϕ-features, visible in processes such as clitic climbing, and a feature they call ‘affective,’ adapting the terminology introduced by Klima 1964). As can be seen, the featural composition of F does not substantially differ from that of C, a situation that resembles the relationship between C and T in Chomsky (2007, 2008). However, notice that one more projection (apart from C) is needed, as Uriagereka (1988) argued on empirical grounds. The outcome is a system where the properties of C, F, and T are not too different: what remains to be investigated is whether the features attributed to them are inherent (say, tense in T) or not (perhaps copied/shared, or inherited). 12. A reviewer points out that “there is no empirical evidence for the idea that some languages ‘squeeze’ various features into one projection and others do not.” Of course, the data that will be provided here (Romance) will not give us overt evidence that features such as [topic] or [focus] can cluster or spread. What the data do show, though, is that the predictions made by a purely cartographic system are not met: otherwise, as Raposo and Uriagereka (2005) observe, one could expect languages (or stages within the same language) where a given operation is independent from the others, and this is not found in Romance. This same reviewer is not convinced by this reasoning, and suggetss that “one could say that you can access some (higher) projections only if others (located lower down) are already filled.” Although I see the gist of this type of approach, I fail to see how it could be implemented (does it mean we can have topicalization only if we have focalization, assuming some version of Rizzi’s cartography?), but, more importantly, I fail to see how that would fit with standard assumptions about phrase structure and checking mechanisms. 13. As the reader may see, the key (but not straightforward) assumption made by Rizzi here is that movement without feature checking (or, in his terms, criteria satisfaction) is free. If by free Rizzi understands ‘not having an effect on the output’ or ‘not being licensed by Last Resort,’ then movement to the Left Periphery is clearly not free. At the background of all of this is the hypothesis that operations must be triggered, and

Phases and variation: Exploring the second factor of the faculty of language

14.

15.

16.

17.

18.

141

the controversial issue is how the trigger must be encoded: with or without features. In this respect, a reviewer observes that Rizzi’s system, being “representational or configurational” (sic.), “is silent as to what triggers syntactic operations (or why X is inserted in SPEC-focus).” I disagree: Rizzi’s framework is representational (as it adopts a strong GB-like view), but it nonetheless resorts to feature-driven movement (qua criteria satisfaction). One important caveat is in order: what I am discussing now is the option that just one projection subsumes the role played by an articulated CP. Rizzi (2006a:126) also ¯ discusses the possibility of there being “only one, nonspecific formal A-feature” (akin to Chomsky’s 2008 “undiscriminate edge-Probe”) to capture left peripheral fronting, but discards it on morphological grounds. Rizzi’s empirical argument is well-taken, but since I am not assuming that Uriagereka’s F (or any other cartography collapsing projection) triggers movement through feature checking, it is irrelevant. This is not to say that the morphological effects discussed by Rizzi (verb-subject obligatory inversion, agreement marks, and so on) are not real: they are, but they can be a consequence of the nature of the element that is moving, and not of the would-be upstairs trigger. Haegeman (2002) investigates facts that deal with truncation within adverbial clauses. As this author argues, the availability of some left-peripheral (or egde) fronting in adverbal clauses is parasitic on these introducing an independent speech act, embodied by ForceP in her account (see Etxepare 1997 for related discussion). These facts appear to correlate with the ones I will discuss in section 4, but they are different, as the data explored by Haegeman (2002) are not subject to parametric variation (her central vs. peripheral adverbial clause distinction is cross-linguistically stable). Whatever the ultimate analysis of these facts turns out to be, I believe the parasitic relationship among heads is better seen as an instantiation of Grimshaw’s (1991) extended projections. See Boeckx (2008a) for novel discussion on extended paths being found behind all functional categories. One other parallelism can be drawn by considering Lasnik and Saito’s (1992) Move α. Just like wh-movement, NP-movement, and the constructions they are involved in (e.g. passive, interrogative, relative) ended up being regarded as the outcome of the very same phenomenon (movement), the different types of left peripheral frontings can be seen as the product of applying the same abstract process (see Chomsky 2002:94-95). I hasten to add that there is no sense in denying that operations such as topicalization, focalization, or relativization have properties of their own (as Rizzi 1997:289-295 has emphasized). My point here is that taking the (idiosincratic) morphological properties of left peripheral operations seriously is not incompatible with keeping in mind that we are talking about the very same subjacent phenomenon. A reviewer asks in what sense is morphology rich. As (s)he correctly points out, there are different ways to understand this, some of them related to the still not completely understood Null Subject Paramater. As I point out below, I am assuming that richness concentrates on the tense paradigm – roughly, the more tense distinctions a language

142

19.

20.

21. 22.

23.

24.

25.

´ Angel J. Gallego has, the more fronting it displays. Notice that, for this to be correct, the relevant head (say, Ph) must somehow be sensitive to the entire tense paradigm. As can be seen, object movement to SPEC-AgrO yields a minimality violation, since it involves raising over the in situ subject. Chomsky (1993, 1995) got around that by invoking a relation of equidistance between specifiers of the same relevant checking domain (typically, collapsed by means of head movement). A more recent way to overcome intervention effects is available through McGinnis (2004) leapfrogging. See Lasnik and Boeckx (2005) for more general discussion. A reviewer is skeptical about Ph (an abstract head) being morphologically rich, and thus triggering creation of specifiers. Although I speak of Ph (or Uriagereka’s F), this head must be understood as being part of the morphological span of the verb (including ϕ-features, tense features, and also mood morphology and complementizers). For additional discussion about this kind of fronting in Spanish, see Hernanz (2001, 2003), Leonetti and Escandell-Vidal (2008), and references therein. The structure in (36b) departs from Belletti’s (2004:34–38) implementation with respect to the amount of structure that is topicalized: for Belletti (2004), it is v*P (although she does not provide evidence from reconstruction, see Huang 1993), while the analysis in (36b) assumes it is the VP. Nothing crucially hinges on this, for what matters is that c-command between subject and object fails. Also, note that I am assuming that verb movement has not taken place yet. Some data due to Anna Cardinaletti (see Cardinaletti 2001), though, cast doubt on such a clear cut between Italian and Spanish. See Gallego (2007:chapter 3) for discussion. There are pre-decompositional analyses where the EA is treated as a VP adjunct/ specifier (see Hale and Keyser 1993, Kitagawa 1986, and Koopman and Sportiche 1991). The most compelling argument – to my mind – for having the external argument above v* was provided by Chomsky (1995), where presence/absence of v* was related to the presence/absence of the the EA and accusative Case. That motivation was gone in Chomsky (2001), where v* has a defective version (v), arguably present in passive and unaccusative structures. Building on Pylk¨annen’s (2008) observation that causative morphology can show up independently of the EA, Harley (2007) argues that the latter is related to Kratzer’s (1996) voice, which is above v*. Noam Chomsky suggests, through personal comunication, another reason why it is not obvious that the EA be generated in SPEC-v*: “The really difficult question is {EA, YP}. But there are so many problems with that construction that I suspect it’s wrong. Thus, why doesn’t it ever move as a unit? Why does something apparently have to extract from it (there are proposals, but none of them works, as far as I can see). I suspect that Ken Hale is right, and EA appears somewhere else – I don’t know where.” A reviewer casts doubt on the analysis in (42). First, (s)he does not see why I am forced to say that, in VSO sentences, the EA has raised from SPEC-V to SPEC-v*,

Phases and variation: Exploring the second factor of the faculty of language

143

and (s)he suggests the “possibility that the EA is generated in SPEC-v and remains there in VSO.” Though technically possible, that option is disfavored by the arguments provided in Ord´o˜nez (2005), which indicate S in VSO is in a derived position. The reviewer is also skeptical about my observation with respect to the v* / v cut (see previous footnote), which I take to weaken the analysis of the EA in Chomsky (1995); as (s)he notes: “[t]he presence/absence of EA could be related to the completeness or defectiveness of v (v*, vdef respectively), not to the presence/absence.” Again, this is possible, but what I was emphasizing is that the connection between EA and the light verb is weakened the moment there is a light verb even in unaccusative sentences. Differently put, and just to push this argument still further: what would then be wrong with saying that what vdef does is actually done by V? Finally, the reviewer is also confused by Chomsky’s quote, and in particular about the point where he mentions that something (i.e. something in the {EA, YP} configuration, where YP = v*P) must be extracted. What Chomsky tries to point out, as far as I understand, is that if EA and v*P formed a unit, they could move together, contrary to fact; then, there is the empirical observation that either EA or v*P must move (see Mayr 2007, Moro 2007). The reviewer concludes by pointing out that one gap in my argumentation is that I do not provide the feature that triggers VP movement (see 36 above): this is true, but since I am not assuming that movement is feature-driven (putting aside Chomsky’s 2008 edge feature), I can sidestep this issue. 26. As a reviewer notes, if C and v* are the same unit at some level, then they must have the same featural content. This is not so obvious. What we can reasonably conclude is that they (can) have ϕ-features (and, by assumption, also Chomsky’s 2008 edge features). Whether they contain other features (e.g. aspect, mood, tense) depends on their context. 27. The contrast can be approached by arguing subjunctive dependents lack a CP projection, as argued for by Kempchinsky (1998). However, since these clauses can spell-out the complementizer que ‘that,’ I will not adopt such an analysis. 28. A reviewer raises the question why the adjectives in (48) are placed between D and N: since DP’s periphery is above D, one would expect the order A D N. This is correct, but it may well depend on what the correct structure for the DP phases is. If the proposals by Boeckx (2009), Chomsky (2007), Hiraiwa (2005), and Leu (2008) are adopted (see below), then the DP is not as in (i), but actually as in (ii), which makes it possible for A to move below D. (i) [ D [ n [ N]]] (ii) [ [DP] n [N]] 29. A reviewer argues that my proposal is not compatible with the Distributed Morphology framework, since I am assuming that morphology drives syntactic operations. This is not accurate, because the features I attribute to phase heads belong to what is called ‘list A’ (i.e. morphosyntactic features) within DM (see Harley and Noyer 1999), which feeds syntactic operations (Merge, Agree, etc.).

144

´ Angel J. Gallego

References Abels, Klaus 2003

Successive Cycliclity, Anti-locality, and Adposition Stranding. Ph.D. diss., Universty of Connecticut

Belletti, Adriana 2001 ‘Inversion’ as Focalization. In: A. Hulk and J. I. Pollock (eds.), Subject Inversion in Romance and the Theory of Universal Grammar, 60–90. Oxford (New York): Oxford University Press. Belletti, Adriana 2004 Aspects of the Low IP Area. In: L. Rizzi (ed.), The Structure of CP and IP. The Cartography of Syntactic Structures (vol. 2), 16–51. Oxford (New York): Oxford University Press. Bianchi, Valentina 1999 Consequences of Antysimmetry. Headed Relative Clauses. Berlin: Mouton de Gruyter. Biberauer, Theresa, and Ian G. Roberts 2006 Subjects, Tense and Verb-movement in Germanic and Romance. Ms., University of Cambridge. Bobaljik, Jonathan David and Diane Jonas 1996 Subject Positions and the Roles of TP. Linguistic Inquiry 27: 195–236. Boeckx, Cedric 2003 Islands and Chains. Resumption as Stranding. Amsterdam: John Benjamins. Boeckx, Cedric 2007 Understanding Minimalist Syntax. Lessons from Locality in LongDistance Dependencies, Malden: Blackwell. Boeckx, Cedric 2008a Bare Syntax. Oxford (New York): Oxford University Press. Boeckx, Cedric 2008b Linguistic Invariance and Language Variation. A Minimalist Perspective on Parameters. Paper presented at the 9th AnnualTokyo Conference on Psycholinguistics (TCP), Keio University, Tokyo (Japan). Boeckx, Cedric 2008c Elementary Syntactic Structures. A Minimalist Inquiry. Ms., Harvard University. Boeckx, Cedric 2009 Aspects of a Theory of Phases. Ms., Harvard University. Boeckx, Cedric and Kleanthes Ghromann 2007 Putting Phases into Perspective. Syntax 10: 204–222. Bosque, Ignacio 2001 Adjective position and the interpretation of indefinites. In J. Guti´errezRexach and L.Silva-Villar (eds.), Current Issues in Spanish Syntax and Semantics, 17–63. Berlin: Mouton de Gruyter.

Phases and variation: Exploring the second factor of the faculty of language

145

Cardinaletti, Anna 2001 A second thought on emarginazione: Destressing vs ‘Right Dislocation.’ In: G. Cinque and G.P. Salvi (eds.), Current Studies in Italian Syntax. Essays offered to Lorenzo Renzi, 117–135. Amsterdam: North Holland. Chomsky, Noam 1955[1975] The Logical Structure of Linguistic Theory. New York: Plenum Press. Chomsky, Noam 1957 Syntactic Structures, The Hague: Mouton Chomsky, Noam 1965 Aspects of the Theory of Syntax. Cambridge (Mass.): MIT Press. Chomsky, Noam 1973 Conditions on Transformations. In: S. Anderson and P. Kiparsky (eds.), A festschrift for Morris Halle, 232–286. New York: Holt, Renehart and Winston. Chomsky, Noam 1977 On Wh-Movement. In: P. Culicover et al. (eds.), Formal Syntax, 71– 132. New York: Academic Press. Chomsky, Noam 1981 Lectures on Government and Binding. Dordrecht: Foris Publications. Chomsky, Noam 1986a Barriers. Cambridge (Mass.): MIT Press. Chomsky, Noam 1986b Knowledge of Language. Its Nature, Origin, and Use. New York: Praeger. Chomsky, Noam 1991 Some Notes on Economy of Representation and Derivation. In: Freidin, R. (ed.), Principles and Parameters in Comperative Grammar, 417–454. Cambridge (Mass.): MIT Press. Chomsky, Noam 1993 A Minimalist Program for Linguistic Theory. In: K. Hale and S. J. Keyser (eds.), The View from Building 20: Essays in Linguistics in Honor of Sylvain Bromberger, 1–52. Cambridge (Mass.): MIT Press. Chomsky, Noam 1995 Categories and Transformations. In: N. Chomsky, The Minimalist Program, 219–394. Cambridge (Mass.): MIT Press. Chomsky, Noam 2000 Minimalist Inquiries: The Framework. In: R. Martin et al. (eds.), Step by Step. Essays on Minimalist Syntax in Honor of Howard Lasnik, 89–155. Cambridge (Mass.): MIT Press. Chomsky, Noam 2001 Derivation by Phase. In: M. Kenstowicz (ed.), Ken Hale: A Life in Language, 1–52. Cambridge (Mass.): MIT Press.

146

´ Angel J. Gallego

Chomsky, Noam 2002 An Interview on Minimalism. In: A. Belletti and L. Rizzi (eds.), On Nature and Language, 92–161. Cambridge (Mass.): Cambridge University Press. Chomsky, Noam 2004 Beyond Explanatory Adequacy. In: A. Belletti (ed.), Structures and Beyond. The Cartography of Syntactic Structures (vol. 3), 104–131. Oxford (New York): Oxford University Press. Chomsky, Noam 2005 Three factors in language design. Linguistic Inquiry 36: 1–22. Chomsky, Noam 2007 Approaching UG from Below. In: U. Sauerland and H.-M. G¨artner (eds.), Interfaces + Recursion = Language? Chomsky’s minimanlism and the view from syntax-semantics, 1–30. Berlin: Mouton de Gruyter. Chomsky, Noam 2008 On Phases. In: C. Otero et al. (eds.), Foundational Issues in Linguistic Theory, 134–166. Cambridge (Mass.): MIT Press. Chomsky, Noam, Morris Halle and Fred Lukoff 1956 On accent and juncture in English. In: M. Halle et al. (eds.), For Roman Jakobson, 65–80. The Hague: Mouton. Chomsky, Noam and Morris Halle 1968 The Sound Pattern of English. New York: Harper Row. Cinque, Guglielmo 1999 Adverbs and Functional Heads. A Cross-Linguistic Perspective. Oxford (New York): Oxford University Press. Costa, Jo˜ao 2000 Word order and discourse-configurationality in European Portuguese. In: J. Costa (ed.), Portuguese syntax: New comparative studies, 94– 115. Oxford (New York): Oxford University Press. Costa, Jo˜ao 2002 VOS in Portuguese:Arguments against an analysis in terms of remnant movement. In: A. Alexiadou et al. (eds.), Dimensions of Movement, 69–89. Amsterdam: John Benjamins. Dikken, Marcel den 2007 Phase Extension. Contours of a theory of the role of head movement in phrasal extraction. Theoretical Linguistics 33: 1–41. Epstein, Samuel D., Erich M. Groat, Ruriko Kawashima, and Hisatsugu Kitahara 1998 A derivational approach to syntactic relations. Oxford: Oxford University Press. Epstein, Samuel D. and T. Daniel Seely 2002 Rule Applications as Cycles in a Level-free Syntax. In: S. Epstein and D. Seely (eds.), Derivation and Explanation in the Minimalist Program, 65–89. Malden (Mass.): Blackwell.

Phases and variation: Exploring the second factor of the faculty of language

147

Etxepare, Ricardo 1997 The Syntax of Illocutionary Force. Ph.D. dissertation, University of Maryland. Fortuny, Jordi 2008 The Emergence of Order in Syntax, Amsterdam: John Benjamins. Fox, Danny and David Pesetsky 2005 Cyclic Linearization of Syntactic Structure. Theoretical Linguistics 31: 1–45. ´ Gallego, Angel J. 2007 Phase Theory and Parametric Variation. Ph.D. dissertation, Universitat Aut`onoma de Barcelona. Giorgi, Alessandra and Fabio Pianesi 1997 Tense and Aspect: From Semantics to Morphosyntax. Oxford University Press, New York. Grimshaw, Jane 1991 Extended projections. Ms., Brandeis University Grimshaw, Jane 1997 Projection, Heads, and Optimality. Linguistic Inquiry 28: 373–422. Guimar˜aes, Maximiliano 2000 In Defense ofVacuous Projections in Bare Phrase Structure. University of Maryland Working Papers in Linguistics 9: 90–115. Haegeman, Liliane 2002 Speculations on adverbial fronting and the left periphery. Ms., Universit´e Charles de Gaulle – Lille III / CNRS. Hale, Ken and Samuel J. Keyser 1993 On the argument structure and the lexical expression of syntactic relations. In: K. Hale and S. Keyser (eds.), The View from Building 20, 53–109. Cambridge (Mass.): MIT Press. Hale, Ken and Samuel J. Keyser 1998 The basic elements of argument structure. MIT Working papers in linguistics 32: Papers from the Upenn/ MIT Roundtable on Argument Structure, 73–118. Cambridge (Mass.): MIT Press. Hale, Ken and Samuel J. Keyser 2002 Prolegomenon to a Theory ofArgument Structure. Cambridge (Mass.): MIT Press. Harley, Heidi 2007 External Arguments: On the Independence of Voiceo and vo . Paper presented at XXX GLOW Conference, University of Tromsø, Tromsø. Harley, Heidi and Rolf Noyer 1999 State-of-the-Article: Distributed Morphology. Ms., University of Pennsylvania. Hernanz, M. Llu¨ısa 2001 ¡En bonito l´ıo me he metido!: Notas sobre la afectividad en espa˜nol. Moenia 7: 93–109.

148

´ Angel J. Gallego

Hernanz, M. Llu¨ısa 2003 From polarity to modality: the case of bien. Ms., UniversitatAut`onoma de Barcelona. Hinzen, Wolfram 2006 Mind Design and Minimal Syntax. Oxford (New York): Oxford University Press. Hiraiwa, Ken 2005 Dimensions of Symmetry in Syntax: Agreement and Clausal Architecture. Ph.D. dissertation, MIT. Hornstein, Norbert and Juan Uriagereka 2002 Reprojections. In: S. Epstein and D. Seely (eds.), Derivation and Explanation in the Minimalist Program, 106–132, Malden (Mass.): Blackwell. Huang, James 1993 Reconstruction and the structure of VP: Some theoretical consequences. Linguistic Inquiry 24: 103–138. Irurtzun, Aritz 2007 The Grammar of Focus at the Interfaces. Ph.D. dissertation, Euskal Herriko Unibertsitatea. Kayne, Richard S. 1993 Toward a Modular Theory of Auxiliary Selection. Studia Linguistica 47: 3–31. Kayne, Richard S. 2000 Parameters and Universals. Oxford (New York): Oxford University Press. Kayne, Richard S. 2004 Prepositions as probes. In: A. Belletti (ed.), Structures and Beyond: The Cartography of Syntactic Structures (Vol. 3), 192–212. Oxford: Oxford University Press. Kayne, Richard S. 2005 On Paramaters and on Principles of Pronunciation. Ms., New York University. Kempchinsky, Paula 1998 Mood Phrase, Case Checking and Obviation. In: A. Schwegler et al. (eds.), Romance Linguistics: Theoretical Perspectives, 143–154. Amsterdam: John Benjamins. Kitagawa, Yoshihisa 1986 Subjects in Japanese and English. Ph.D. dissertation, University of Massachusetts (Amherst). Klima, Edward 1964 Negation in English, In J. Fodor and J. Katz (eds.), The structure of language, 264–323. Englewood Cliffs (NJ): Prentice Hall. Koopman, Hilda and Dominique Sportiche 1991 The Position of Subjects. Lingua 85: 211–258

Phases and variation: Exploring the second factor of the faculty of language

149

Kratzer, Angelika 1996 Severing the External Argument from its Verb. In: J. Rooryck and L. Zaring (eds.), Phrase Structure and the Lexicon, 109–137. Dordrecht: Kluwer. Lasnik, Howard and Cedric Boeckx 2005 Long NP-Movement. In: M. Everaert and H. van Riemsdijk (eds.),The Blackwell Companion to Syntax (vol. 3), 109–130. Oxford: Blackwell. Lasnik, Howard and Mamoru Saito 1992 Move α: conditions on its applications and outputs. Cambridge (Mass.): MIT Press. Leonetti, Manuel and Victoria Escandell-Vidal 2008 Fronting and Verum Focus in Spanish. Ms., Universidad de Alcal´a / UNED. Leu, Thomas 2008 The Internal Syntax of Determiners. Ph.D. dissertation, NYU. Marantz, Alec 1997 No Escape from Syntax: Don’t Try Morphological Analysis in the Privacy of Your Own Lexicon. U. Penn Working Papers in Linguistics 4.2.: 201–225. Marantz, Alec 2007 Phases and Words. In: S. H. Choe (ed.), Phases in the theory of grammar, 191–220. Seoul: Dong In. McGinnis, Marta 2004 Lethal Ambiguity. Linguistic Inquiry 35: 47–95. Mayr, Clemens 2007 Stylistic Inversion and VP-externalization. Ms., Harvard University. Moro, Andrea 2000 Dynamic antisymmetry. Movement as a symmetry breaking phenomenon. Cambridge (Mass.): MIT Press. Moro, Andrea 2007 Some Notes on Unstable Structures. Ms., Universit`a San Raffaele. Ord´o˜nez, Francisco 1998 Post-verbal asymmetries in Spanish. Natural Language and Linguistic Theory 16: 313–346. Ord´o˜nez, Francisco 2001 Some clitic combinations in the syntax of Romance. In: J. Mascar´o and G. Rigau (eds.), Catalan Journal of Linguistics. The Grammar of Clitics, 201–224. Ord´o˜nez, Francisco 2005 Two Specs for postverbal subjects: evidence from Spanish and Catalan. Ms., SUNY Stony Brook.

150

´ Angel J. Gallego

Pesetsky, David and Esther Torrego 2001 T-to-C Movement: Causes and Consequences. In: M. Kenstowicz (ed.), Ken Hale: A Life in Language, 355–426. Cambridge (Mass.): MIT Press. Pesetsky, David and Esther Torrego 2004 Tense, Case, and the Nature of Syntactic Categories. In: J. Gu´eron and J. Lecarme (eds.), The Syntax of Time, 495–537. Cambridge (Mass.): MIT Press. Pesetsky, David and Esther Torrego 2006 Probes, Goals, and the Nature of Syntactic Categories. In:Y. Otsu (ed.), Proceedings of the Seventh Tokyo Conference on Psycholinguistics, 25–60. Tokyo: Hituzi Syobo Publishing Company. Picallo, M. Carme 1998 On the EPP and Null Expletive Subjects. Probus 10: 219–241. Poletto, Cecilia 2006 Parallel Phases: a study on the high and low left periphery of Old Italian. In: M. Frascarelli (ed.), Phases of Interpretation, 261–293. Berlin: Mouton De Gruyter. Pollock, Jean-Yves 1989 Verb Movement, Universal Grammar, and the Structure of IP. Linguistic Inquiry 20: 365–424. Pylkk¨anen, Liina 2008 Introducing Arguments. Cambridge (Mass.): MIT Press. Raposo, Eduardo 2002 Nominal Gaps with Prepositional Modifiers in Portuguese and Spanish:A Case of Quick Spell-Out. Cuadernos de Ling¨u´ıstica del Instituto Universitario Ortega y Gasset 9: 127–144. Raposo, Eduardo and Juan Uriagereka 2005 Clitic Placement in Western Iberian: A Minimalist View. In: G. Cinque and R. Kayne (eds.), The Oxford Handbook of Comparative Syntax, 639–697. Oxford (New York): Oxford University Press. Richards, Marc D. 2004 Object Shift and Scrambling in North and West Germanic: A Case Study in Symmetrical Syntax. Ph.D. dissertation, University of Cambridge. Richards, Marc D. 2006 Deriving the Edge: What’s in a Phase? Ms., University of Cambridge. Richards, Marc D. 2007 On Feature Inheritance: An Argument from the Phase Impenetrability Condition, Linguistic Inquiry 38: 563–572. Rizzi, Luigi 1994 Early null subjects and root null subjects. In: T. Hoekstra and B. Schwartz. (eds.), Language acquisition studies in generative grammar, 151–177. Amsterdam: John Benjamins.

Phases and variation: Exploring the second factor of the faculty of language Rizzi, Luigi 1995 Rizzi, Luigi 1997

Rizzi, Luigi 2004

Rizzi, Luigi 2006a

Rizzi, Luigi 2006b

151

Some notes on linguistic theory and language development: the case of root infinitives. Language Acquisition 3: 371–393. The Fine Structure of The Left Periphery. In: L. Haegeman (ed.), Elements of Grammar. Handbook in Generative Syntax, 281–337. Dordrecht: Kluwer. Locality and Left Periphery. In: A. Belletti (ed.), Structures and Beyond. The Cartography of Syntactic Structures (vol. 3), 223–251. Oxford (New York): Oxford University Press,. On the Form of Chains: Criterial Positions and ECP Effects. In: L. Cheng and N. Corver (eds.), Wh-Movement: Moving on, 97–133. Cambridge, MIT Press.

Criterial Freezing, EPP and Asymmetries. Paper presented at Edges in Syntax, Cyprus College, Nicosia (Cyprus). Sigurðsson, Halld´or 2006 The Nominative Puzzle and the Low Nominative Hypothesis. Linguistic Inquiry 37: 289–308. Svenonius, Peter 2004 On the Edge. In: D. Adger et al. (eds.), Peripheries: Syntactic Edges and their Effects, 259–287. Dordrecht: Kluwer. Stowell, Timothy 1981 The origins of phrase structure. Ph.D. dissertation, MIT. Torrego, Esther 1980 Fronting in Spanish. Ms., University of Massachusetts (Boston). Torrego, Esther and Juan Uriagereka 1992 Indicative Dependents. Ms., University of Massachusetts (Boston) and University of Maryland. Uriagereka, Juan 1988 On Government. Ph.D. dissertation, University of Connecticut. Uriagereka, Juan 1995a An F Position in Western Romance. In: K. Kiss (ed.), Discourse configurational languages, 153–175. Oxford (NewYork): Oxford University Press. Uriagereka, Juan 1995b Aspects of the syntax of clitic placement in Western Romance. Linguistic Inquiry 26: 79–123 Uriagereka, Juan 1999a Multiple Spell-Out. In: N. Hornstein and S. Epstein (eds.), Working Minimalism, 251–282. Cambridge (Mass.): MIT Press, 251–282.

152

´ Angel J. Gallego

Uriagereka, Juan 1999b Minimal Restrictions on Basque Movements. Natural Language and Linguistic Theory 17: 403–444. Uriagereka, Juan 2002a Derivations. Exploring the Dynamics of Syntax. London: Routledge. Uriagereka, Juan 2002b Evidential Contexts. Ms., University of Maryland. Uriagereka, Juan 2007 Clarifying the Notion ‘Parameter.’ Biolinguistics 1: 99–113. Uriagereka, Juan 2008 Syntactic Anchors at the Semantic Interface. Cambridge: Cambridge University Press.

Varieties of INFL: T ENSE, L OCATION , and P ERSON Elizabeth Ritter and Martina Wiltschko Abstract. The main claim of this paper is that while the core function of the category INFL is universally fixed, its substantive feature content is not. The dissociation of INFL from its features immediately predicts that languages can display disparity in the substantive content of INFL – the only restriction being that the selected formal features must be compatible with the core function of INFL. We show that this prediction is borne out in Halkomelem (Salish) where the substantive content of INFL is Location, and in Blackfoot (Algonquian), where it is Person. We show that both Location and Person are functionally equivalent to Tense, formally identical to Tense, and in complementary distribution with it, and with each other. These facts provide compelling support for our hypothesis that the content of functional categories is subject to parametric variation, but they are problematic for the cartographic approach, which assumes that Universal Grammar provides a template of functional positions with fixed syntactico-semantic content.

1. Introduction The languages of the world differ in many obvious ways. But at the same time, they are also strikingly similar. Trying to understand this tension between language diversity and identity is one the most fruitful research agendas of modern linguistic theory. Our research into two languages indigenous to North America has allowed us to gain some new insight into this problem, specifically as it pertains to the nature of functional categories. Within the Chomskyan tradition, it is often taken for granted that Universal Grammar (UG) makes available a set of functional categories with fixed content. For example, adapting Pollock’s influential proposal (Pollock 1991), it is widely held that the head of the clause is a category – formerly known as INFL – which is universally associated with marking for tense. But, it has often been observed that many languages appear to lack tense marking. Thus, we find ourselves rediscovering the core observation of the American structuralists (Boas, Sapir, and Bloomfield): languages appear to differ in the categories they obligatorily need to express. We wish to show in this paper that languages do indeed differ in their categorial inventories, but at the same time we argue that this variation is constrained by a universally determined hierarchy of functional positions. That is, after careful examination of Halkomelem (Salish) and Blackfoot (Algonquian), we conclude that each of

154

Elizabeth Ritter and Martina Wiltschko

these languages appears to use different functional categories (Location and Person, respectively), but that these categories can be analyzed as instantiating the same universal category as Tense, namely INFL.1 To make this case, we introduce the Parametric Substantiation Hypothesis, according to which functional categories are not uniquely associated with the same substantive content (section 2). In section 3 we argue that in Halkomelem Location substantiates INFL, and in section 4 we argue that in Blackfoot Person substantiates INFL. We conclude in section 5 that the Parametric Substantiation Hypothesis is empirically more adequate than alternative Principles and Parameters based approaches (including the cartographic one) as an explanation for the tension between language variation and language universals.

2. Parametric substantiation as an alternative to cartography In accordance with the theme of this book, we start with a discussion of how the problem of language-specific categorial inventories presents itself within the cartographic approach (section 2.1). We point out a problem and proceed to introduce our proposal, the Parametric Substantiation Hypothesis (section 2.2). Assuming that the substantive content of INFL is not universally fixed, it is necessary to use formal diagnostics that are independent of content to identify INFL. In section 2.3, we introduce the diagnostics we have used in this research.

2.1. A problem with the cartographic approach In English, as in many other Indo-European languages, all finite clauses display an obligatory contrast in Tense. That is, English finite verbs are marked as [+past] by means of a suffix (-ed) or an auxiliary verb (was) as illustrated in (1), or they are marked as [-past] (i.e., present) by means of a third person singular agreement suffix (-s), or a different form of the auxiliary (is), as illustrated in (2). (1)

a. b.

He walked. He was walking.

[+past] [+past]

(2)

a. b.

He walks. He is walking.

[−past] [−past]

It is well-known, however, that not all languages display such an obligatory contrast in Tense. For example, in Halkomelem Salish, tense marking is not

Varieties of INFL: Tense, Location, and Person

155

required (Wiltschko 2003). Instead, finite clauses display an obligatory contrast in Location.2 This contrast is invariably expressed by means of locative auxiliaries: li marks a [+distal] specification as shown in (3), while i marks a [−distal] (i.e., proximate) specification as shown in (4).3 (3)

l´ı qw’ey´ılex t´u-tl’`o aux dance he ‘He is/was dancing (there).’

[+distal]

(4)

´ı qw’ey´ılex t´u-tl’`o aux dance he ‘He is/was dancing (here).’

[−distal]

Thus, an obligatory contrast in Tense is not a language universal. But given the data above, we might hypothesize that the presence of an obligatory contrast of some sort is in fact a language universal. If this is correct, then the substantive content of the obligatory contrast is subject to parametric variation: while in English the content of the contrast is Tense, in Halkomelem it is Location. This preliminary generalization defines an interesting research agenda: how can we account for this variation in the content of the obligatory contrast? Within the framework of cartography (Cinque 1999), the answer to this question is restricted by certain assumptions regarding the nature of UG on the one hand, and parametric variation on the other. Specifically, within this framework it is assumed that UG provides a template of functional positions with fixed syntactico-semantic content. As an illustration, consider a partial representation of a clausal structure: (5)

. . . [FinP Fin . . . [TnsP Tense . . . [AspP Aspect . . . ]]]

The positions shown in (5) are assumed to be universally available, universally organized in a hierarchical fashion, and universally associated with the same content. Accordingly, language variation is viewed as a matter of determining whether a given position is available for external or internal Merge, and whether it is overtly filled. Within this framework, how can we account for the variation in the content of the obligatory contrast observed in English and Halkomelem? We would have to assume that both Tense and Location are part of the universal template of fixed positions. Location has not yet been introduced into this universal template and thus it is not immediately obvious whether it would be situated above or below Tense. For the purpose of illustration, let us assume that Location is generated below Tense, as illustrated in (6). We could then assume that Tense,

156

Elizabeth Ritter and Martina Wiltschko

but not Location, is available for Merge in English, while the reverse is true in Halkomelem, that is, Location but not Tense is available for Merge. (6) English: Halkomelem:

[Tense Tense . . . [Loc Loc . . . ]]] [+/−past] – – [+/−distal]

The main goal of this paper is to argue that the cartographic approach to variation in the content of obligatory contrasts misses the essential characteristics of this type of variation. Specifically, we will show that Tense in English and Location in Halkomelem (as well as Person in Blackfoot) are functionally equivalent, formally identical, and in complementary distribution. We therefore propose an alternative approach, namely the Parametric Substantiation Hypothesis.

2.2. The Parametric Substantiation Hypothesis In line with the cartographic approach, we assume that UG provides a template of fixed functional positions. We depart from cartography, however, in proposing that these positions are only associated with abstract core functions rather than with fixed substantive content such as Tense or Location. We refer to this as the Parametric Substantiation Hypothesis: (7)

The Parametric Substantiation Hypothesis: The substantive content of a given functional category is subject to parametric variation, constrained only by the universally determined core function of that category.

Thus, the precise substantive content is not universally fixed. For the case at hand, we therefore make use of the original label for the head hosting Tense, namely INFL. We thus assume that UG provides the structure in (8). (8)

[CP COMP . . . [IP INFL . . . [AspP Aspect . . . ]]]

We will demonstrate that the choice of substantive content has far-reaching consequences that we can only begin to explore in this paper. Within the framework of the Parametric Substantiation Hypothesis, we propose to account for the difference between English and Halkomelem in the following way. Abstracting away from the contribution of aspect, we assume that the universal core function of INFL is to anchor the reported eventuality (Ev) to the utterance (Utt) (En¸c 1987). Specifically, we analyze INFL as a predicate of coincidence (+/−coin) (in the spirit of Hale 1986 and Demirdache and

Varieties of INFL: Tense, Location, and Person

157

Uribe-Etxebarria 1997, 2000) which orders the eventuality encoded by the VP with respect to the utterance encoded in SpecIP.4 As illustrated in (9), ordering proceeds via the relevant arguments in VP and IP. We assume that UG makes available participant and spatio-temporal arguments in SpecVP and in SpecIP (as in Demirdache and Uribe-Etxebarria 1997). (9)

IP Utt-arg

I'

INFL [+/-coin]

VP Ev-arg

V

We further propose that the substantive content of INFL varies: in English it is Tense, in Halkomelem it is Location, and in Blackfoot it is Person, resulting in different varieties of the category INFL (10). We represent these different subcategories as INFLTense , INFLLocation , and INFLPerson . We assume that concepts that are not mapped onto a functional head in a given language are still available and can be optionally merged as modifiers. (10)

[CP COMP . . . [IP INFL . . . . . . [v v ]]] English: Tense: [+/−past] Halkomelem: Location: [+/−distal] Blackfoot Person: [+/−local]

The substantive content of INFL determines the precise content of both the utterance argument and the event argument. Thus, in a Tense-based language, temporal arguments are ordered, while in a Location-based language spatial arguments are ordered, and finally in a Person-based language, participant arguments are ordered. (See Ritter and Wiltschko, in preparation, for a more detailed discussion.) In sum, we propose that languages differ substantially in the way eventualities are ordered relative to the utterance: English orders times, Halkomelem orders locations, and Blackfoot orders participants. The purpose of this paper is to motivate the Parametric Substantiation Hypothesis on the basis of a detailed contrastive investigation of English, Halkomelem, and Blackfoot. Before we can proceed, however, we need to establish how to identify the category INFL in any given language.

158

Elizabeth Ritter and Martina Wiltschko

2.3.

Formal diagnostics for INFL

In the context of the present research, there is a methodological issue arising. Once we abandon the assumption that each functional category is universally associated with distinct substantive content, we cannot use meaning as one of the criteria for category membership. But then, how do we know whether a given linguistic object (LO) maps onto a functional category, and if it does, which one it is? If a given LO expresses the concept of time, how do we know whether or not it maps onto a functional category Tense? Since the universalist framework adopted here generally assumes that the substantive content is fixed, this question does not arise in the same way, and thus this methodological issue is never addressed. This contrasts with earlier structuralist traditions that had precise criteria and procedures for identifying grammatical categories. In this section we introduce some diagnostics for identifying INFL. The Parametric Substantiation Hypothesis assumes that functional material can be merged either as a head or as a modifier. Accordingly, to identify INFL we need diagnostics that distinguish between heads and modifiers. This can be achieved as follows. First, heads differ from modifiers in whether or not they are unique: while heads are unique, modifiers are recursive.Another related property is that of obligatoriness: while heads are obligatory, modifiers are optional. The obligatoriness of heads has two immediate consequences. On the one hand, syntactic obligatoriness allows for LOs to remain semantically empty, that is, they can function as expletive placeholders. This contrasts with optional modifiers, which never remain uninterpreted (i.e., there are no expletive modifiers). Consequently, we expect that elements instantiating INFL may – in certain environments – lack substantive content. On the other hand, syntactic obligatoriness also allows for silent LOs. That is, it is also a general property of syntactically obligatory LOs that they can remain uninterpreted at PF (i.e., without phonetic content) in certain well-defined contexts. Again, this contrasts with optional modifiers, which are never silent (i.e., there are no phonetically empty modifiers). In addition to criteria that distinguish between heads and modifiers, we also need some formal diagnostics that set INFL apart from other functional heads. Here we follow standard practice and assume that a defining property of INFL is its ability to enter into a local relation with a particular functional head, namely COMP. In particular, INFL may undergo head-movement to COMP, and COMP may select for a particular instantiation of INFL (i.e., finite or non-finite). In section 3 we demonstrate that Halkomelem INFLLocation , as opposed to INFLTense , satisfies the formal criteria for INFL. And in section 4 we establish that Blackfoot INFLPerson satisfies the formal criteria for INFL. The fact that

Varieties of INFL: Tense, Location, and Person

159

INFLTense , INFLLocation and INFLPerson are in complementary distribution at least across the three languages under investigation provides evidence for the Parametric Substantiation Hypothesis as opposed to the cartographic approach to language universals and variation. We discuss the implications of these findings in section 5.

3. Location as an alternative to Tense To motivate the Parametric Substantiation Hypothesis, we start with a discussion of Halkomelem. We first show that INFLTense and INFLLocation are functionally equivalent: they both serve to anchor the eventuality to the utterance. We then show that INFLTense in English and INFLLocation in Halkomelem are formally identical in that they both satisfy the formal criteria for INFL laid out in section 2.3. Next we briefly show that temporal marking in Halkomelem does not satisfy the formal diagnostics for INFL. We conclude this section with a discussion of what can and what cannot serve as the substantive content of INFL.

3.1.

Functional equivalence

As stated in section 2.2, we assume that the core function of INFL is to anchor the reported eventuality to the utterance. Depending on the substantive content of INFL in a given language, anchoring proceeds in different ways. In English, INFL is substantiated by Tense, and thus if INFLTense asserts that the event time coincides with the utterance time ([+coin]), the result is a present tense interpretation. In contrast, if INFLTense asserts that the event time does not coincide with the utterance time ([−coin]), the result is a past tense interpretation. This is schematized in (11) (cf. Demirdache and Uribe-Etxebarria 1997, 2000).5 (11)

English: Tense substantiates INFL IP Utt-time

‘present’ ‘past’

INFL [+coin] [-coin]

I’ VP Ev-time

160

Elizabeth Ritter and Martina Wiltschko

The situation is different in a language such as Halkomelem where INFL is substantiated by Location. In such a language, INFLLocation does not assert when relative to the utterance the eventuality takes place, but rather where relative to the utterance the eventuality takes place. Thus, in such a language, [+coin] asserts that the event location coincides with the utterance location, resulting in a proximate location interpretation, while [−coin] asserts that the event location does not coincide with the utterance location, resulting in a distal location interpretation as shown in (12). (12)

Halkomelem: Location substantiates INFL IP Utt-location

‘proximate’ ‘distal’

INFL [+coin] [-coin]

I’ VP Ev-location

This analysis of Halkomelem captures the insight of two different traditional descriptions of the locative auxiliaries, which we analyze as occupying INFLLocation . On the one hand, Suttles (2004:35) states that “[t]he choice between /i and ni depends on the location of the speaker relative to whatever the predicate refers to.”6 On the other hand, Galloway (1993:359) states that “[t]he choice between /´ı and l´ı is governed by considerations having to do with the location of the eventuality. In particular, locative auxiliaries encode the semantic opposition of emplacement (‘here’ . . . ) and displacement (‘there’ . . . )”. While Suttles’ description focuses on the location of the speaker, Galloway’s description focuses on the location of the eventuality.7 According to the analysis sketched in (12), the auxiliaries in INFLLocation order the event location relative to the utterance location (i.e., speaker location), just as INFLTense orders the event time relative to the utterance time. Note in passing that location marking can have an indirect effect on the temporal interpretation of the clause. In the context of first (and second) person subjects, the use of the distal auxiliary will result in a past interpretation. Consider, for example, the sentences in (13) and their translations. The sentence in (13a) cannot receive a present time interpretation because the speaker cannot be in one location participating in the speech act and at the same time be elsewhere participating in the helping event; however, the speaker could have been at a distinct (distal) location at a time prior to the utterance time.8 In contrast,

Varieties of INFL: Tense, Location, and Person

161

a present time interpretation is possible in (13b), due to the use of a proximate auxiliary.9 (13)

a.

b.

ni c´n c´ e´ c´ ´w-´t aux 1sg help-trans ‘I was helping him.’ /i c´n c´ e´ c´ ´w-´t aux 1sg help-trans ‘I’m helping him.’

(Suttles 2004: 35)

Thus, if INFL is not substantiated by Tense but instead by Location, it can nevertheless have temporal effects. This much establishes that INFLLocation in Halkomelem is functionally equivalent to INFLTense in English, consistent with the Parametric Substantiation Hypothesis.

3.2.

Formal identity

We now turn our attention to the formal properties of INFLTense and INFLLocation . We demonstrate that they are formally identical, supporting our claim that the two are different instantiations of the same core category, namely INFL. Specifically, we use the following diagnostics developed in section 2.3: (14)

Formal diagnostics for INFL: i) uniqueness ii) lack of substantive semantic content (expletiveness) iii) movement to COMP iv) lack of phonetic content (silence) v) obligatoriness

3.2.1. Uniqueness of tense and location marking It is well-known that tense marking in English is unique. That is, every clause can only be marked for tense once. So even though tense marking can occur on either the main verb (15a) or an auxiliary (15b), in the presence of an auxiliary, tense has to be marked on the auxiliary alone, and cannot simultaneously be marked on both the main verb and the auxiliary (15c).10 (15)

a. He walked. b. He did walk. c. *He did walked.

162

Elizabeth Ritter and Martina Wiltschko

We observe a similar pattern in Halkomelem, where INFL is substantiated by Location, which in turn is expressed by the locative auxiliaries i and li.11 Crucially, only one instance of these auxiliaries is permitted in Halkomelem, suggesting that location marking is indeed unique. (16)

a.

b. c. d. e.

li/i chexw t’ilem aux 2sg sing ‘You sang (here/there). *li chexw i t’ilem12 aux 2sg aux sing *li i chexw t’ilem aux aux 2sg sing *li chexw li t’ilem aux 2sg aux sing *li li chexw t’ilem aux aux 2sg sing

Note further that both i and li can in fact occur twice in the same clause. However, if they do, the two instances of i or li are not both auxiliaries, but rather the second instance functions as the main predicate of the clause (meaning to be here or to be there, respectively). This is illustrated by the data in (17). (17)

a.

b.

i i aux be.here ‘He’s here.’ li li aux be.there ‘He’s there.’

Similar facts obtain in English where do, have, and be can function both as an auxiliary (occupying INFLTense ) and as a main verb (occupying V), and consequently can occur twice in a single clause, as shown in (18). (18)

a. b. c.

He didn’t do it. He has had a cat before. He is being silly again.

This establishes that INFLLocation , like INFLTense , is unique. However, nothing prohibits the LOs occurring in INFL from occurring elsewhere in the clause (for example, as main verbs).

Varieties of INFL: Tense, Location, and Person

163

3.2.2. Absence of substantive semantic content in Tense and Location Another formal diagnostic for heads (as opposed to modifiers) is the possibility of being interpreted without substantive semantic content in certain syntactically defined environments. Consider, for example, English past tense morphology. In the context of conditionals, it is used without necessarily inducing a past time interpretation. This is shown in (19a) where past tense morphology on the auxiliary co-occurs with an adverb of temporal presence (now). This cooccurrence of past tense morphology and a present time adverb is only possible, however, in the context of conditionals. In indicative clauses this co-occurrence results in ill-formedness, as in (19b). In this case, present tense morphology must be used, as in (19c). (19)

a. If I had a car now, I’d give you a ride. b. *I had a car now. c. I have a car now.

Similar facts hold in Halkomelem, where yes/no questions are formed by means of the locative auxiliary li and an optional suffix –a. In this context the distal force is lost. The examples in (20) show that the auxiliary li can be used to ask a question about an individual who is at the utterance location at the utterance time.13 In (20a), for example, li is used to ask a question about the addressee, who is at the utterance location; and in (20b), li is used to ask a question about a third person who is also at the utterance location. (20)

a.

b.

li(-´a) chexw ´ıtet tel´ewe aux-q 2sg sleep you ‘Are you sleeping?’ li(-a) ´ıtet t´utl’`o aux-q sleep 3indep ‘Is he sleeping?’ (about someone in the same room)

As noted in section 3.2.1, two locative auxiliaries cannot co-occur in the same clause in Halkomelem. This is the case even if one functions as a question marker while the other indicates spatial location of the eventuality (see also note 12). In the next subsection, we argue that li undergoes movement from INFLLocation to COMP in yes/no questions. Thus, both past tense morphology in English and distal locative auxiliaries in Halkomelem can lose their substantive content in certain well-defined environments.

164

Elizabeth Ritter and Martina Wiltschko

3.2.3. Tense and Location can undergo head movement to COMP One of the diagnostics that specifically targets INFL (as opposed to simply determining head-status) is that of movement to COMP. That is, since INFL is the head immediately below COMP, we expect direct interactions between these two heads, and this is precisely what we find. In both English and Halkomelem, material that usually occupies INFL (tense and location marking, respectively) may be displaced to COMP in certain syntactically defined environments. In English root questions the tensed auxiliary occurs in a position linearly preceding the subject, as shown in (21). (21)

Did he walk?

Subject-auxiliary inversion of this kind has been analyzed as the result of INFLTense moving into COMP, as illustrated in (22).14 (22)

[CP did [ IP he did walk ]]

A similar effect can be observed in Halkomelem yes/no questions, which are optionally marked with a suffix -a attached to the first element in the clause. (23)

a.

b.

li-´a t’´ılem aux-q sing ‘Is he singing/Did he sing?’ ew-´a li-s y´ethes-t-`ol`e-m neg-q aux-3s tell-trans-2sg.o-pass ‘Weren’t you folks told?’

(Galloway 1993:186)

As shown in (23a), the auxiliary can host the question marker in the absence of a higher LO, such as negation (see Wiltschko 2002 for evidence that negation in Halkomelem is higher than INFL). We assume that the Halkomelem question marker occupies COMP. If so, then the example in (23a) can be analyzed as involving INFLLocation to COMP movement, as illustrated in (24). (24)

[CP li-´a [IP li [VP t’ilem]

In sum, in both English and Halkomelem we find evidence that the material occupying INFL (tense and location marking, respectively) can move to COMP in the context of question formation.

Varieties of INFL: Tense, Location, and Person

165

3.2.4. Absence of phonetic content in Tense and Location The next property we discuss is zero-marking. If a given syntactic position is obligatory, its presence does not entail phonetic content. As such, we expect that INFL can be occupied by a phonetically empty marker. Here we show that both tense marking in English and location marking in Halkomelem display this property. In English, present tense is not marked overtly in most contexts. The only exception to this generalization is in the context of a third person singular subject, where there is an overt agreement marker that is restricted to present tense, and thus is often considered to constitute tense marking. All other verb forms, however, are not overtly marked in the present tense. We assume that all these forms are actually marked with a zero tense exponent, that is, an LO without phonetic content, as shown in (25). (25)

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

I walk-∅present You walk-∅present He walk-spresent We walk-∅present You walk-∅present They walk-∅present

Location marking in Halkomelem is also possible with a phonetically empty LO. Consider the example in (26), which appears to lack a locative auxiliary. (26)

tsel qw’ey´ılex 1sg.s dance ‘I was dancing.’

The absence of a locative auxiliary in (26) may be analyzed in one of two ways. Either this structure lacks a locative auxiliary, as schematized in (27a), or else there is a locative auxiliary which lacks phonetic content, as in (27b). (27)

a. – tsel qw’ey´ılex → no locative auxiliary b. ∅aux tsel qw’eyilex → locative auxiliary without phonetic content15

In light of the present discussion, the choice between the two alternative analyses illustrated in (27) is not trivial. If (27a) were correct, then we would be forced to conclude that location marking is not obligatory. Such a conclusion would undermine the claim that Location substantiates INFL in Halkomelem, because one of the diagnostics for INFL is obligatoriness (see section 2.3).

166

Elizabeth Ritter and Martina Wiltschko

In order to show that the presence of location marking is indeed obligatory, it becomes crucial to demonstrate that (26) contains location marking despite appearances to the contrary. But how do we decide between the two analyses shown in (27)? In what follows we discuss three arguments for the presence of a phonetically empty auxiliary in such examples. The first argument stems from the temporal interpretation associated with examples such as (26), in which the clitic precedes the verb (henceforth cliticverb order).16 As observed in Galloway (1993:175), the clitic-verb order tends to result in a past interpretation. This effect receives a straightforward explanation if we assume the presence of a phonetically empty [+distal] auxiliary in INFLLocation .17 Recall that the overt [+distal] auxiliary triggers a past interpretation with first or second person subjects because the speech act participants (speaker and addressee) cannot be talking at the utterance location and at the same time be participating in an event elsewhere (see section 3.1). Under this analysis, the past interpretation of the clitic-verb order reduces to the same effect as the past interpretation of the [+distal] auxiliary in the context of first or second person clitics:18 the clitic-verb order is formally identical to sentences with the [+distal] auxiliary li, as illustrated in (28): (28)

a. [∅]+distal tsel qw’eyilex → past b. [li]+distal tsel qw’ey´ılex → past

In contrast, if we assume that the clitic-verb order in (26) lacks a locative auxiliary (as in (27a)), then the past interpretation remains to be explained. We wish to show here that this is not at all a trivial task because Halkomelem is a tenseless language (see Wiltschko 2003, and section 3.3 for discussion). The past interpretation of a sentence with a [+distal] auxiliary is not encoded as such in the grammar; instead, it arises due to world-knowledge. But if the clitic-verb order does not contain a [+distal] auxiliary, then this explanation is not available. Instead we would have to assume that the absence of an auxiliary results in a past interpretation. But this assumption does not adequately capture the facts either, as will become clear in our second argument in favour of the presence of a phonetically empty auxiliary. Upriver Halkomelem has another construction that lacks an overt locative auxiliary. It differs from the one in (26) in that the order of the verb and the clitic is reversed (henceforth verb-clitic order). Crucially, this construction is interpreted as referring to a future eventuality (see Bar-el et al. 2003 for a detailed discussion of this construction) or, when used with a second person subject, as an imperative.

Varieties of INFL: Tense, Location, and Person

(29)

a.

b.

167

a´ lhtel-tsel eat-1sg.s ‘I’m going to eat.’ (Bar-el et al. 2003: ex. 2d) t’´ılem-chexw (Upriver Halkomelem) sing-2sg.s ‘You sing.’ Speaker’s comment: “You are asking somebody to sing.” (Bar-el et al. 2003: ex. 8)

This semantic effect appears to be categorical in the sense that it cannot be overridden by the presence of a temporal adverbial, as shown in (30). (30)

*´alhtel-tsel kwi chel´a:qelh eat-1sg.s det yesterday

(Upriver Halkomelem) (Bar-el et al. 2003: ex.15)

We have now seen that a difference in order is associated with a difference in temporal interpretation, as summarized below: (31)

a. clitic-verb b. verb-clitic

→ past → future

From this pattern we can immediately conclude that the absence of a locative auxiliary is not always associated with a past interpretation. This means that the past interpretation of the clitic-verb order is still unaccounted for under the assumption that there is no locative auxiliary. In addition, we wish to show that the alternation schematized in (31) further helps us to decide between the two analyses for the clitic-verb order (i.e., whether or not there is a phonetically empty auxiliary present). Consider first how we can account for this pattern given the assumption that the clitic-verb order contains a phonetically empty [+distal] auxiliary, as illustrated in (32a). Suppose that the verb-clitic order is derived via head-movement of V-to-INFL as illustrated in (32b). (32)

a. [∅]+distal clitic verb b. verb clitic verb

→ past → future

The analysis in (32) leaves us with two questions: i) what triggers movement of the verb to INFLLocation , and ii) why does this construction receive a future interpretation? The first question receives a straightforward answer if we assume that INFLLocation must be filled. A locative auxiliary (with or without phonetic content) can satisfy this requirement, as we have seen above. If no locative auxiliary is available, then INFLLocation can be filled via head-movement of

168

Elizabeth Ritter and Martina Wiltschko

the verb. Thus, V-movement in (32b) can be analyzed as a result of the familiar requirement that INFL must be filled. As for the second question raised by the pattern schematized in (32), we argue that the future interpretation associated with (32b) is a consequence of the fact that INFLLocation is filled by an LO which lacks substantive content. V-movement serves to fill INFLLocation , but, unlike locative auxiliaries, V lacks a [±distal] specification. Consequently, INFLLocation in (32b) fails to specify whether or not the event location coincides with the utterance location. We propose that in this context the eventuality denoted by the predicate cannot be located in the actual world, and that this is precisely what is required in order to talk about future eventualities: Future eventualities cannot be located in the actual world since they have not yet occurred. Under this analysis, a future interpretation can arise even in the absence of a functional category INFLTense , contrary to recent claims in the literature that future always involves both a modal component and interpretable content (e.g., [±past, ±future]) in INFLTense (see, for example, Abusch 1988, Matthewson 2006). However, there is independent evidence that a future interpretation is possible even in the absence of interpretable content in INFLTense . For example, infinitives, which are often analyzed as untensed (Wurmbrand 2006), can receive a relative future interpretation as illustrated in (33): (33)

Leo decided a week ago [to go to the party (yesterday)]. (Wurmbrand 2006:3, ex. 4b)

The fact that in a Tense-based language like English, untensed clauses can receive a future interpretation demonstrates that the availability of a future interpretation does not imply the presence of interpretable tense features; and given the Parametric Substantiation Hypothesis, the availability of a future interpretation does not imply the presence of a functional category INFLTense (see Ritter and Wiltschko, in preparation, for more detailed discussion). If our account of the future interpretation as arising in the absence of anchoring is on the right track, then we have indirect support for the claim that there is a phonetically empty auxiliary present in the clitic-verb order, as shown in (32a). Note that on the alternative view, namely that the clitic-verb order lacks a locative auxiliary, the pattern in (31) remains unaccounted for. In particular, on this view, the contrast between verb-clitic and clitic-verb order cannot be reduced to the absence or presence of an auxiliary, respectively. A final argument for the presence of an empty locative auxiliary in the cliticverb construction stems from the patterning of subject clitics across the different dialects of Halkomelem (and across the Salish languages in general). The Sal-

Varieties of INFL: Tense, Location, and Person

169

ish subject clitics typically pattern as second-position clitics (see Davis 2000 for a detailed discussion of the cross-Salish patterns). But this second-position requirement appears to be violated by the clitic-verb order unless we assume that the initial clitic is in fact preceded by a phonetically empty auxiliary.19 This conclusion is supported by the fact that in the closely related Downriver and Island dialects of Halkomelem, overt auxiliaries are obligatory. Thus, under the present analysis the possibility for the clitic-verb order in Upriver Halkomelem is due to the innovation of a phonetically empty auxiliary.20 In sum, we have argued that Upriver Halkomelem has a phonetically empty auxiliary that appears to be the silent counterpart of the [+distal] auxiliary li. This analysis allows us to understand the temporal interpretations associated with three superficially distinct constructions, summarized below: (34)

a. li+distal clitic verb b. [∅+distal clitic verb c. verb clitic verb

→ past → past → future

The clitic-verb order (34b) is analyzed as being identical to sentences with the [+distal] locative auxiliary (li; (34a)) while the verb-clitic order arises in the absence of a locative auxiliary, which leads to V-to-INFL movement (34c). The temporal effects associated with each construction also fall out from this analysis. The future interpretation of (34c) arises because without a locative auxiliary, the reported eventuality cannot be located in the actual world. The past interpretation of (34a,b) is analyzed as a matter of world knowledge: speaker and addressee cannot simultaneously be at the utterance location and at a distinct event location.21 This analysis makes a prediction regarding the temporal interpretation of sentences with third person subjects. Since third person participants need not be at the utterance location, they can participate in an eventuality that occurs simultaneously with the utterance but at a distinct location. Consequently, we expect that with third person subjects, a sentence with a [+distal] locative auxiliary can receive a present and a past interpretation. This prediction is borne out: (35)

li a´ lhtel te sw´ıyeqe aux eat det man ‘The man is eating.’ ‘The man was eating.’

Observe that in (35) the third person subject is not marked by means of a (secondposition) subject clitic. (This is a general property across the Salish family.) Thus, in the absence of a locative auxiliary, we cannot tell whether the verb

170

Elizabeth Ritter and Martina Wiltschko

has moved; such sentences should be compatible with a present, past, or future interpretation.22 This is indeed the case, as shown in (36): (36)

a´ lhtel te sw´ıyeqe eat det man ‘The man is eating.’ ‘The man was eating.’ ‘The man will be eating.’

We take the interaction between linear order and temporal interpretation discussed in this subsection as evidence for the existence of an empty locative auxiliary, which meets the expectation that INFL Location need not be associated with overt phonetic content. We can now show that INFL Location in Halkomelem is obligatory, just as INFLTense is in English.

3.2.5. Obligatoriness of Tense and Location In languages where Tense substantiates INFL, specification for [±past] is obligatory in tensed clauses, as shown in (37): (37)

*He walk.

In section 3.2.4 we saw that in Halkomelem INFL must be filled by an LO that is specified for [± distal] in located clauses, and that otherwise it is filled by an element that lacks this substantive content. Since there is an empty auxiliary available, the effect of obligatory Location cannot be directly observed, but must be deduced from the different interpretations associated with contrasting word orders. We now discuss another indirect piece of evidence for the obligatoriness of Location in Halkomelem. This evidence has to do with contexts in which specification for [±distal] is obligatorily absent from INFL, that is, in clauses which are unlocated: We show that these are precisely the contexts in which specification for [±past] is obligatorily absent from INFL in English, that is, untensed clauses. We argue that this is not a coincidence but instead follows from the Parametric Substantiation Hypothesis: Since both INFL Location and INFLTense are instances of the same category INFL, it follows that the absence of substantive content, whether temporal or spatial, should give rise to similar effects. Consider first English: there are two constructions that are obligatorily untensed: (at least some) infinitives (Wurmbrand 2003) and imperatives (Zanuttini 1991). Neither of these constructions situates an eventuality on a time-line: While infinitives denote a set of eventualities, imperatives encode the requirement for an eventuality to occur. It is precisely in such environments that specification for [±past] is ruled out, as shown in (38) and (39).23

Varieties of INFL: Tense, Location, and Person

(38)

Infinitive: a. He wants to walk. b. *He wants to walked.

(39)

Imperative: a. Walk! b. *Walked!

171

Strikingly, we also find this effect in Halkomelem, but not with respect to tense marking, but rather with location marking. Locative auxiliaries are ruled out in the same contexts as tense marking in English. We start with a discussion of imperatives, which are optionally formed with a dedicated suffix (-lha). Crucially, regardless of whether the imperative suffix is present, locative auxiliaries are categorically ruled out in this environment. Consequently, in the presence of an imperative suffix, the locative auxiliary results in ill-formedness (40), while in the absence of the suffix, the imperative interpretation is not available (41) but instead the sentence is interpreted as a yes/no question. (40)

a.

qw’ey´ılex-lha dance-imp ‘Dance!’ b. *l´ı qw’ey´ılex-lha aux dance-imp

(41)

a.

b.

qw’ey´ılex chexw dance 2sg.s ‘You dance!’ l´ı chexw qw’ey´ılex aux 2sg.s dance =‘You dance!’ = ‘Did you dance/Are you dancing/Do you dance?’

Next we turn to infinitival environments. Observe that Halkomelem does not have specific infinitival forms. Rather, clauses that are translated as infinitives are nominalized clauses that contain subject agreement (in the form of possessive morphology), as shown in (42). (42)

l-stl’´ı kw’-el-s qw’ey´ılex 1sg.poss-want det-1sg.poss-nom dance ‘I want to dance.’

172

Elizabeth Ritter and Martina Wiltschko

Crucially, in this context, location marking is impossible. That is, if a locative auxiliary is added to the embedded clause in (42), the interpretation is fundamentally different. It must be translated as a finite (when) clause and cannot be translated as an infinitival clause. (43)

l-stl’´ı kw’-el-s l´ı qw’ey´ılex24 1sg.poss-want det-1sg.poss-nom aux dance ‘I like it when I used to dance.’ =‘I want to dance.’

This phenomenon can be observed with all predicates that would require a nonfinite complement in English, and thus is fully productive. We have included one more example in (44). It shows that in the absence of an auxiliary, the embedded clause is readily translated as an infinitive, but in the presence of a locative auxiliary the result is necessarily finite: (44)

a.

b.

lhq’´ellexw-es tl’ Strang kw-s know-3s det.obl Strang det-nom ‘Strang knows how to sing.’ lhq’´ellexw-es tl’ Strang kw-s know-3s det.obl Strang det-nom t’´ıt’elem sing.cont ‘Strang knows that he used to sing.’ =‘Strang knows how to sing.’

t’´ılem-s sing-3poss l´ı-s aux-3poss

We have now shown that Halkomelem does not allow for location marking in precisely the same contexts where English does not allow for tense marking: with imperatives and infinitives.25

3.3.

Complementarity

Up to this point we have seen that in Halkomelem location marking displays formal properties identical to those of tense marking in English. We take this to indicate that INFLLocation and INFLTense are instances of the same abstract functional category, namely INFL. Thus, we expect tense and location marking to be in complementary distribution. Specifically, we expect that a Tense-based language does not have obligatory location marking, and conversely, that a Location-based language does not have obligatory tense marking of the type found in English. Rather, if notions related to time are expressed in Halkomelem,

Varieties of INFL: Tense, Location, and Person

173

and similarly, if notions related to location are expressed in English, we expect them to display different formal properties – specifically, we expect them to function as optional modifiers. This prediction is borne out. In English, a Tense-based language, locational modifiers can always be added but are never obligatory.26 (45)

a. b.

I was dancing (here/there/on the dance floor). He was dancing (here/there/on the dance floor).

In Halkomelem, a Location-based language, temporal marking is not obligatory. Instead, unmarked forms are “the catch all tense, used to indicate present action (which must be continuing as the speaker speaks – continuative aspect), habitual action (which may be spread over past, present, and future), momentaneous action (which the speaker is about to perform – noncontinuative aspect), and past action (historical present in narratives, legends, etc.)” (Galloway 1993:315). The sentences in (46) illustrate this property: temporally unmarked clauses can receive either a present or a past interpretation (see Wiltschko 2003 for a detailed discussion). (46)

a.

b.

´ı qw’ey´ılex t´u-tl’`o aux dance he He is/was dancing (here).’ l´ı qw’ey´ılex t´u-tl’`o aux dance he ‘He is/was dancing (there).’

This suggests that in the absence of temporal marking, a Halkomelem clause is truly unmarked regarding the time of the eventuality relative to the utterance. This is in stark contrast to English, where what looks like the absence of marking is really an instance of a zero-marked present tense (which receives a habitual interpretation in the context of an eventive verb): (47)

I dance.

Further evidence to the effect that Halkomelem clauses without temporal marking are in fact truly unmarked (rather than zero-marked for a specific temporal interpretation) stems from the fact that an unmarked clause is compatible with a temporal adverb of present or past time, as illustrated in (48). (48)

a.

xwm´ekw`ath-et-es t´utl’`o tel´o way´el kiss-trans-3s det-3 this day ‘He kissed her today.’

174

Elizabeth Ritter and Martina Wiltschko

b.

c.

tsel a´ lhtel w´ayeles 1sg.s eat tomorrow ‘I’ll eat tomorrow.’ tsel ´ı:mex tsel´aqelh 1sg.s walk yesterday ‘I was walking yesterday.’

For reasons of space we cannot discuss further evidence for the absence of a functional category Tense in Halkomelem (see Wiltschko 2003 for a detailed discussion; but see Matthewson 2006 for a different view).

3.4.

Interim conclusion

We have now shown evidence to the effect that INFLTense in English and INFL Location in Halkomelem are varieties of the same category. Specifically, we have shown that the two categories have identical formal properties as summarized in Table 1: both are uniquely marked; both may lack content; both may remain semantically or phonetically empty (i.e., expletive or silent); both may undergo head-movement to C; and both are obligatory. Table 1. Formal properties of INFL Substantive content of INFL English: INFL T ENSE Halkomelem: INFL L OCATION is unique yes yes may lack content yes yes may move to C yes yes may have silent exponent yes yes is obligatory yes yes

We have also shown that the INFLTense and INFL Location are functionally equivalent: both relate the eventuality to the utterance. INFLTense asserts when the reported eventuality took place relative to the utterance time, while INFL Location asserts where the reported eventuality took place relative to the utterance location. From this we conclude that INFLTense and INFL Location are really instances of the same category. Given that complementarity is a criterion for identity, this conclusion is supported by the fact that INFLTense and INFL Location are in complementary distribution. Thus, we propose that INFL is a universal category, which can be substantiated by either Tense or Location: (49)

INFL = {Tense, Location}

Varieties of INFL: Tense, Location, and Person

175

This much supports the Parametric Substantiation Hypothesis. But it also raises the following question: If INFL can be substantiated by either Tense or Location, are there any other categories that can substantiate INFL? In other words, are there any restrictions on the substantive content of INFL, and if so, what are these restrictions? To answer this question it is useful to consider again the core function of INFL, as shown in (9), repeated here as (50). (50)

IP Utt-arg

I'

INFL [+/-coin]

VP Ev-arg

V

Given that the core function of INFL is to relate the eventuality to the utterance, it follows that whatever content substantiates INFL must be able to fulfill this function. Specifically, it must be able to relate the eventuality to the utterance. Only deictic categories, which are interpreted relative to the extralinguistic context of the utterance, have this property. Thus it follows that the substantive content of INFL must be deictic. Both Tense and Location fulfill this criterion: they are among the deictic categories. But if this characterization of the substantive content of INFL is correct, we predict that cross-linguistically INFL will not be restricted to being substantiated by either Tense or Location; rather, any deictic category should be possible. For example, we predict that Person, another deictic category, can also substantiate INFL. (51)

INFL = {Tense, Location, Person}

In what follows, we show that this prediction is indeed borne out, though a word of caution is in order. While the translation from a Tense-based system to a Location-based system is relatively straightforward, the translation to a Person-based system is more complicated. While we believe that we have thus far gathered evidence to make the case, we also acknowledge that more research needs to be done to fully understand this system.

4. P ERSON as an alternative to T ENSE We argue that in Blackfoot (Algonquian), INFL is substantiated by Person. Consequently, in this system, anchoring proceeds via participants: it asserts

176

Elizabeth Ritter and Martina Wiltschko

who relative to the utterance participants participated in the eventuality. In section 4.1, we show that the category Person (like the other deictic categories) can serve to anchor the reported eventuality to the utterance. As such, it is functionally equivalent to Tense and Location. We also show that tense marking in Blackfoot does not meet the criterial properties of INFL. In section 4.2, we show that person marking displays the formal characteristics we have used to diagnose INFL. Due to the fact that person marking is also found in Tense, Location, and Person languages in the form of agreement, we also demonstrate that as INFLPerson is essentially different from person agreement. In addition, INFLPerson differs from both INFLTense and INFLLocation in the kinds of arguments that are ordered. We assume that the utterance takes place over one time span and in one location, and thus that there is only one utterance time and only one utterance location to be ordered relative to the event time or location. However, there are two utterance participants, the speaker and the addressee. Thus, INFLPerson has a more complex argument structure than either INFLTense or INFLLocation . We discuss this issue in section 4.3.

4.1.

Person substantiates INFL in Blackfoot

We argue that in Blackfoot, Person substantiates INFL. Consequently, we analyze person marking in Blackfoot as the formal and functional equivalent of tense marking in English and location marking in Halkomelem. Specifically, we argue that the person prefixes of Blackfoot represent the substantive content of INFL in this language:27 (52)

a.

b.

c.

nitsik´akomimmawa nit´ana nit-iik-w´akomimm-a-wa n-itan-wa 1-very-love(ta)-dir-3sg 1-daughter-prox ‘I love my daughter.’ (Frantz 1991:51, ex. a) kitsik´akomimmawa nit´ana kit-iik-w´akomimm-a-wa n-itan-wa 2- very -love(ta)-dir-3sg 1-daughter-prox ‘You love my daughter.’ (Frantz 1991:51, ex. c) otsik´akomimmoka nohk´owa ot´ani ot-iik-w´akomimm-ok-wa n-ohk´o-wa w-itan-yi 3- very -love(ta)-inv-3sg 1-son-3sg 3-daughter-obv (Frantz 1991:56, ex. k) ‘Her daughter loves my son.’28

Varieties of INFL: Tense, Location, and Person

177

4.1.1. Functional equivalence We propose that person marking serves to anchor the eventuality to the utterance in Blackfoot, and therefore anchoring proceeds via participants, rather than times or locations. Thus, the substantive content of INFLPerson asserts who relative to the utterance participants participated in the eventuality. The person prefixes specify whether or not an eventuality participant coincides with one of the utterance participants. Thus, first and second person prefixes assert that the relevant eventuality participant coincides with an utterance participant:29 nitindicates coincidence with the speaker, while kit- indicates coincidence with the addressee. In contrast, the third person prefix (ot-), asserts that the relevant eventuality participant does not coincide with an utterance participant. (53)

Blackfoot: Person substantiates INFL IP Utt-participant

1/2 3

I'

INFLPERSON VP [+coin] [-coin] Ev-participant …

According to (53), Blackfoot person marking is the functional equivalent of tense marking in English.

4.1.2. Complementarity If Person does indeed substantiate INFL in Blackfoot, then the Parametric Substantiation Hypothesis predicts that Blackfoot should not have tense marking of the kind found in English, because INFLTense and INFL Person are predicted to be in complementary distribution. This prediction is indeed borne out. In Blackfoot, just like in Halkomelem, a clause that is unmarked for tense can be interpreted as either reporting a past or a present eventuality:30 (54)

kit-´ana aas´a´ı’ni-wa 2-daughter cry-3sg ‘Your daughter cried.’ OR ‘Your daughter is crying.’

( Frantz 1991:36, ex. v)

Thus, Blackfoot appears to lack an inflectional tense system. A similar conclusion is reached by Uhlenbeck, who states that “this language possesses neither a true tense system, nor a true aspect system” (1938:133). In particular, there is

178

Elizabeth Ritter and Martina Wiltschko

no (dedicated) past tense marker which contrasts with a present tense marker, which we take to be the hallmark of an inflectional tense system. The absence of a functional head INFLTense further predicts that all properties that are tied to the specific substantive content of this head should be absent as well (in contrast to the properties of INFL that are independent of the substantive content). Structural case fits this description: it has been argued to be a direct consequence of Tense (see, for example, Pesetsky and Torrego 2001 for the claim that case is Tense on D). If so, we predict that Blackfoot should lack the effects of structural case. This prediction is borne out, as argued by Ritter and Rosen (2005). The evidence discussed there includes absence of morphological case, absence of case-motivated A-movement, and absence of A-binding. We thus conclude that Blackfoot lacks a syntactic category INFLTense , in line with the Parametric Substantiation Hypothesis.31

4.2.

Formal identity, or How to distinguish INFL-Person from person agreement

We now turn to empirical evidence to the effect that person marking displays the diagnostic characteristics of INFL.

4.2.1. Obligatoriness As noted above, we analyze the Blackfoot person prefixes as expressing the substantive content of INFL in this language. Thus, we predict that these person prefixes should be obligatory in the Blackfoot counterparts of English tensed clauses and Halkomelem located clauses. In this section, we argue that this is indeed the case. Blackfoot has four different verb paradigms (called orders in the Algonquianist literature), which are distinguished in part by whether or not they have person prefixes. The two orders that have obligatory person prefixes are the independent and conjunctive (but see note 28 on the restricted use of third person prefixes in the independent order). Verbs in the independent order are primarily used in matrix clauses. We have seen examples of such clauses in (52) above. The conjunctive order, which is marked with a suffix –hs, is used in most types of embedded clauses, as shown in (55) (see Frantz 1991:111 for discussion). (55)

a.

a´ yo’kaawa nit´a´ı’to’toohsi a´ -Io’kaa-wa nit-´a’-it-o’too-hs-yi dur-sleep(ai)-3sg 1-inchoat-there-arrive(ai)-conj-conj ‘He was asleep when I got there.’ (Frantz 1991:111, ex. a)

Varieties of INFL: Tense, Location, and Person

b.

c.

179

nit´a´ısskskammawa kit´a’waaway´akiyssi nit-´a-sskskamm-a:-wa kit-´a’-wa:wayaki-yi-hs-yi 1-dur-watch(ta)-dir-3sg 2-inchoat-hit(ta)-inv-conj-conj ‘I was watching over her, when she hit you.’ (Frantz 1991:111, ex. b) ´ı´ıkssoka’piiwa ot´a´ısootaahsi iik-soka’pii-wa ot-´a-sootaa-hs-yi very-good(ai)-in.sg 3-dur-rain(ii)-conj-conj ‘It’s good that it’s raining.’ (Frantz 1991:111, ex. e)

We analyze clauses with independent or conjunctive order verbs as personed, that is, INFL in such clauses is obligatorily specified for person features [±1, ±2], (on analogy with tensed and located clauses, which are specified for [±past] and [±distal], respectively). Clauses with imperative or subjunctive order verbs, in contrast, are unpersoned, that is, INFL lacks these substantive features. We further predict that the contexts in which we find unpersoned clauses in Blackfoot should be similar to those in which we find untensed clauses in English and unlocated clauses in Halkomelem, that is, imperatives and infinitives. Imperative order verbs are used in imperative clauses in Blackfoot. We interpret the absence of person prefixes in imperatives as evidence that INFLPerson in these clauses lacks [±1, ±2], just as [±past] and [±distal] are absent in imperatives in languages with INFLTense and INFLLocation , respectively. (56)

a.

b.

ooy´ıt! ooyi-t eat(ai)-2sg(imp) ‘Eat!’ ooyik! ooyi-k eat(ai)-2pl(imp) ‘Eat!’

(Frantz 1991:114, ex. r)

Blackfoot has no infinitives. The second verbal paradigm in which person prefixes are obligatorily absent is the so-called subjunctive order used for if-clauses and generic when(ever) clauses (Frantz 1991:113). Note that in these examples we still find person marking in the form of suffixes, but as we argue immediately below, these suffixes are agreement markers and do not substantiate INFL.

180

(57)

Elizabeth Ritter and Martina Wiltschko

a.

b.

c.

ikkam´ınimmiinnaaniki, nit´aaksowatoo’pinnaana ikkam-Ini-mmiinnaaniki nit-y´aak-Iowatoo-‘p-innan-wa if-see(t1)-1pl(s) 1-fut-eat(t1)-theme-1pl-in.sg ‘If we see it, we’ll eat it.’ (Frantz 1991:113, ex. m) ikkam´ayo’kainoainiki, nit´aakahkayi ikkam-´a-yo’kaa-inoainiki nit-y´aak-wa:hkayi if-dur-sleep(ai)-2pl(s) 1-fut-go.home ‘If you (guys) are sleeping, I’ll go home.’ (Frantz 1991:113, ex .l) ai’s´oo´ taasi, a´ akitsipiimmiaawa a’-sootaa-si y´aak-it-IpiiM:-yi-aawa inchoat-rain(ii)-in.sg(s) fut-then-enter-3pl-pro ‘When it rains, they will go in.’ (Frantz 1991:113, ex. p)

We hypothesize that the Blackfoot subjunctive is similar to the infinitive in a Tense-based language, in that it lacks specific substantive content in INFL necessary to anchor the situation denoted by the predicate (directly) to the utterance. As a consequence of the obligatoriness of person marking in Blackfoot, we might also expect the possibility for a dedicated zero-marker instantiating a specific value of Person.32 This is indeed the case: third person marking is always overt in the conjunctive order (see section 4.2.2) but is silent in the independent order, except in a transitive clause with an obviative (a.k.a. fourth person) agent and a proximate (a.k.a. third person) patient or goal. (58)

ik´akomimmiiwa nohk´owa kit´ani ∅-iik-w´akomimm-yii-wa n-ohk´o-wa k-itan-yi 3-very-love-dir-3sg 1-son-prox 2-daughter-obv ‘My son loves your daughter.’ (Frantz 1991:53, ex. l)

At this point, however, we have not seen any properties of person marking in Blackfoot that would distinguish it from person agreement in English. That is, even in Tense-based languages like English, INFL is assumed to host person marking in the form of subject-verb agreement. Consider, for example, the sentences in (59). The suffix -s on the inflected verb indicates agreement with a third person singular subject in the present tense, as illustrated in (59a). It is obligatory in this finite context but banned from infinitival contexts, as shown by the contrast between (59b) and (59c). And finally, person agreement can be realized as a phonetically empty marker (i.e., first and second person subjects are not associated with overt agreement morphology), as shown in (59d,e).

Varieties of INFL: Tense, Location, and Person

(59)

a. b. c. d. e.

181

John walk-s John walk*(-s). John wants to walk(*-s). I walk. You walk.

As we have now seen, person agreement in English, a Tense-based language, appears to have precisely the same formal characteristics as person marking in Blackfoot. So how can we tell whether person marking is really functioning to anchor the event participant to the utterance participant, and not simply encoding an agreement relation between the verb and its subject or object? In other words, how can we tell the difference between person marking that substantiates INFL and person marking that functions as agreement? This is the question we address in the next subsection.

4.2.2. INFLPerson differs from person agreement The difference between person agreement of the familiar sort and person marking that functions as an anchoring category can be established on the basis of a careful investigation of Blackfoot person and number marking. In addition to the person prefixes already introduced, Blackfoot has two other affixal positions for person and number marking. Thus, Blackfoot is traditionally described as making available three distinct agreement positions, as in (60), filled by the paradigms summarized in Table 2 (Frantz 1991). (60)

[Prefix - [Verb Stem] - Suffix1] - Suffix2]

Table 2. Three affixes for person and number marking in Blackfoot Prefix kitnit(ot-)

2 1 3

Suffix 1 -(i)nnaan -oaa(wa) -oaa(wa)

1pl 2pl 3pl

Suffix 2 -wa 3sg -yi 3pl/obv pl -yini obv sg

We argue that the suffixal positions are agreement markers of the familiar type, but that the person prefixes function as the anchoring category INFLPerson. The first argument for this claim derives from comparing the feature content of the person prefixes with that of the agreement suffixes. While the former encode person features only (first, second, and third), the latter encode a bundle of person and number features or number and proximate/obviation marking. The fact that the person prefixes encode person only is consistent with the hypothesis that they

182

Elizabeth Ritter and Martina Wiltschko

substantiate INFL as a Person category. The agreement suffixes, on the other hand, establish an agreement relation with a given argument and as such can encode more than one feature. Note that feature bundling is typical of agreement markers. A second argument stems from the well-known fact that person prefixes do not establish a relation with a fixed argument. Instead they merely establish whether or not an utterance participant (first or second person) is involved in the event. This captures Frantz’ insight that “the person affixes do not signal the function of persons, but only that they are involved” (1971:18). Thus, in the examples below, the person prefixes nit- and kit- are used whenever an utterance participant is involved in the event, regardless of whether the participant functions as an AGENT or as a THEME.33 (61)

a.

b.

(62)

a.

b.

nitsik´akomimmawa nit´ana nit-iik-w´akomimm-a-wa n-itan-wa 1-very-love(ta)-dir-3sg 1-daughter-3sg ‘I love my daughter.’ (Frantz 1991:51, ex. a) nitsik´akomimmokinnaani kit´aniksi nit-iik-w´akomimm-ok-innan-yi k-itan-iksi 1-very-love(ta)-inv-1pl-3pl 2-daughter-pl ‘Your daughters love us.’ (Frantz 1991:56, ex. i) kitsik´akomimmawa kit-iik-w´akomimm-a-wa 2- very -love(ta)-dir-3sg ‘You love my daughter.’ kitsik´akomimmoka kit-iik-w´akomimm-ok-wa 2- very -love(ta)-inv-3sg ‘My daughter loves you.’

nit´ana n-itan-wa 1-daughter-3sg (Frantz 1991:51, ex. c) nit´ana n-itan-wa 1-daughter-3sg (Frantz 1991:55, ex. e)

A final argument stems from the interaction of person marking with clausetypes other than matrix indicative. Specifically, the presence of person prefixes is determined by the clause-type, whereas agreement suffixes are present across all clause-types (cf. Frantz 1991). As discussed in section 4.2.1, Blackfoot (like other Algonquian languages) has different clause-types, known as orders in the Algonquian tradition, and these orders differ in terms of their morphology and their distribution. Both person prefixes and agreement suffixes are present in clauses in the so-called independent order, which roughly corresponds to matrix indicative clauses, and in the conjunct order. However, person prefixes

Varieties of INFL: Tense, Location, and Person

183

cannot occur while the agreement suffixes are still expressed in the imperative and subjunctive orders:34

Table 3. The distribution of person markers across different clause-types Order Independent Conjunct Subjunctive Imperative

INFL P ERSON yes yes no no

Agreement P ERSON yes yes yes yes

We analyze these co-occurrence restrictions on the person prefixes as an instance of selection. Assuming that clause-typing is encoded in COMP (Cheng 1991), we propose that information about order in Blackfoot is also located in COMP. If so, we can understand the co-occurrence restriction on person prefixes as the effect of a selectional relation between COMP and INFL. Specifically, we propose that if COMP is in the independent or conjunct order, it selects for a [+Person] INFL, while a subjunctive or imperative COMP selects for INFL that is specified for [-Person], as schematized in (63). (63)

CP COMP IP [order] INFL … +/-PERSON

This analysis provides a principled reason for the different co-occurrence restrictions between clause-type (order) and person markers. Only the person prefixes are heads (namely INFLPerson ) and as such they can enter into a selectional relation with another head, such as COMP. In contrast, the person and number suffixes are not so affected because – according to our analysis – they are not heads but instead markers of agreement relations between the verb and its arguments. As such, they are not susceptible to selectional restrictions. Another diagnostic property of INFL is that it can undergo head-movement to COMP. Evidence that Blackfoot person prefixes undergo head-movement to COMP is provided by na-, an epistemic modal realized in COMP (Bliss and Ritter 2007).35 As illustrated in (64) and (65), na- is in complementary distribution with person prefixes. More specifically, na- only appears in contexts that otherwise lack an overt person prefix, that is, when there is an inclusive

184

Elizabeth Ritter and Martina Wiltschko

AGENT or THEME, and when there are only third person arguments, but ot- is not permitted (see note 28 for details). (64)

a.

b.

c.

(65)

nit´oksta’si nit-´oksta’si 1-run.ai ‘I ran.’ kit´oksta’si kit-´oksta’si 2.run.ai ‘You ran.’ (na)´oksta’siwa na-´oksta’si-wa na-run.ai-prox ‘S/he ran.’

a. *nanit´oksta’si na-nit-´oksta’si na-1-run.ai b. *kitna´oksta’si kit-na-´oksta’si 2-na-run.ai

Following Bliss and Ritter (2007), we assume that na- and the person prefixes are competing for the same structural position. The contrast between (64a,b) and (64c) shows that the person prefix, rather than na-, appears in this context. These facts are exactly what we expect if person prefixes are INFLPerson elements that raise to COMP, a position otherwise occupied by na-, as illustrated in (66). (66)

a. [CP NA [IP nit-/kit- . . . [VP V ]]] b. [CP nit-/kit- [IP nit-/kit- . . . [VP V ]]]

Note that if nit- and kit- were agreement morphemes, we would expect, contrary to fact, that either na- would appear instead of the agreement prefix, or that nawould co-occur with the prefix. This treatment of na- is reminiscent of the facts of English conditionals, illustrated in (67). All three examples have the same semantic content, and differ only in the choice of elements in INFL and COMP. In (67a) there is no movement: COMP is realized as the complementizer if, INFL is realized as the modal would, and the auxiliary have is in a position subordinate to INFL. In (67b), have has undergone head-movement to INFL. In this context the

Varieties of INFL: Tense, Location, and Person

185

auxiliary bears tense inflection, and would is not expressed. Similarly, in (67c), have undergoes subsequent head-movement to COMP, and now occupies the position of if. (67)

a. [CPif [IP he would have [VP run ]]] [IP he had had [VPrun ]]] (V-to-INFL movement) b. [CPif c. [CPhad [IP he had had [VP run ]]] (V-to-INFL and INFL-toCOMP movement)

This establishes that person prefixes are formally distinct from person agreement. We argue that the behaviour of person prefixes reflects their status as INFLPerson elements: they can enter into a local relation with COMP, the higher head. In particular, we have seen evidence that COMP can select for a specific instance of INFLPerson , and that INFLPerson can move to COMP.

4.2.3. Uniqueness The final characteristic of INFL we discuss here is that of uniqueness: given that INFL is a head, we expect its marking to be unique. Assuming that person prefixes instantiate INFLPerson in Blackfoot, we predict that there can only be one person prefix per clause. This prediction is borne out. For completeness, note that the agreement suffixes differ again in this respect: as discussed above, there are two slots available for agreement suffixes. Given the Blackfoot-specific properties, this is in fact a non-trivial prediction. Recall that the person prefixes do not establish a relation with a fixed argument, but instead they merely establish whether or not an utterance participant (first or second person) is involved in the event. So in the present context it is interesting to observe what happens if both the speaker and the addressee are involved in the event. Interestingly, even in this context, only one person prefix is allowed, as shown in (68). (68)

a.

b.

c.

kitsik´akomimmoki kit-iik-akomimm-oki 2-very-love-2:1 ‘You love me.’ kitsik´akomimmo kit-iik-akomimm-o 2-very-love:1:2 ‘I love you.’ *nitkitsik´akomimmoki nit-kit-iik-akomimm-oki 1-2-very-love-2:1

186

Elizabeth Ritter and Martina Wiltschko

d. *nitkitsik´akomimmo nit-kit-iik-akomimm-o 1-2-very-love:1:2 e. *kitnitsik´akomimmoki kit-nit-iik-akomimm-oki 2-1-very-love-2:1 f. *nitkitsik´akomimmo nit-kit-iik-akomimm-o 1-2-very-love:1:2 Even if both speaker and addressee are involved in the reported event, only one person prefix is allowed. This is consistent with the claim that the person prefixes instantiate INFL, which can only be marked once, as discussed in section 3.2.1.

4.3. Tying up some loose ends There are two outstanding issues that are specific to a system in which Person substantiates INFL. Specifically, a Person-based system differs in at least two respects from a Tense- or Location-based system. First, we assume that there is only one utterance time and only one utterance location, and that there is only one event time and one event location. Person differs, however, in that, on the one hand, there can be more than one utterance participant (the speaker and the addressee), and on the other hand that there are up to three participant arguments of the event. We refer to the latter as event participants. While we cannot fully explore the consequences of these two properties specific to a Person-based system, we wish to briefly discuss the hypotheses we are currently exploring.

4.3.1. How do we deal with two utterance participants? Recall from the previous section that INFL is always uniquely marked. This is also the case in a Person-based system, even if both utterance participants are involved in the reported event. Which of the utterance participants is the event participant anchored to, the speaker or the addressee? Given the data in (69)–(70), we observe that the person prefix is always the one corresponding to second person regardless of whether the addressee is the AGENT or THEME. The use of the first person prefix here yields ungrammaticality.

Varieties of INFL: Tense, Location, and Person

(69)

a.

b.

(70)

a.

b.

187

kitsik´akomimmoki kit-iik-akomimm-oki 2-very-love-2:1 ‘You love me.’ kitsik´akomimmo kit-iik-akomimm-o 2-very-love-1:2 ‘I love you.’ *nitsik´akomimmoki nit-iik-akomimm-oki 1-very-love-2:1 *nitsik´akomimmo nit-iik-akomimm-o 1-very-love-1:2

In the traditional Algonquian literature, this phenomenon is described as a person-hierarchy effect. That is, given a choice between two event participants, the choice of the person prefix is determined by the following stipulations (known as the person hierarchy): if there is a third person and a local person, the local person outranks third person; if there are two local persons, second person outranks first person. We suggest that this person hierarchy is structurally conditioned (see also Jelinek and Carnie 2003). In particular, we speculate that INFLPerson is in fact more complex than INFLTense or INFLLocation , precisely because there are two utterance participants. Following ideas by Speas and Tenny (2003) we assume that INFLPersoncan project a complex shell-structure, similar to Larson’s (1988) VP-shell structure for di-transitive predicates. Specifically, we will assume that the speaker occupies the specifier position of the higher inflPerson , while the addressee occupies the specifier position of the lower INFLPerson within the INFL-shell structure. Speas and Tenny argue that the shell structure permits a structural definition of speaker and addressee: the speaker is the agent (i.e. the external argument) of the utterance and the addressee is the goal (i.e. the indirect internal argument). This is schematized in (71).

188

Elizabeth Ritter and Martina Wiltschko

(71)

iP Utt-part1: i’ Speaker inflPERSON

IP

Utt-part2: I’ Addressee INFLPERSON

VP

EvParticipant

This shell structure allows us to account for the observed person-hierarchy effect in Blackfoot. Assuming that anchoring of the reported eventuality to the utterance is subject to a locality condition, Spec, INFLPerson will always be a closer anchoring position than Spec, inflPerson . Thus, anchoring to the addressee will always take precedence over anchoring to the speaker, resulting in the person hierarchy effects (2>1). But why can there not be two person prefixes? In other words, why does the uniqueness condition hold despite the fact that INFL contains two head positions? We assume that general considerations of economy derive this effect: once the reported event is anchored, it cannot be anchored again, even if a second anchoring position were available.

4.3.2. How do we deal with two event participants? The next problem that a Person-based system presents has to do with the number of participants involved in the event. In the case of an intransitive predicate, the single event participant can straightforwardly be ordered relative to the utterance participant. However, if there are two or three event participants, it is not immediately clear which one is to be ordered relative to the utterance participant. So the following question arises: What determines which of the event participants is in fact ordered, the AGENT or the THEME? The empirical facts are as follows: The event role of the utterance participant is marked on the verb stem by a special suffix, know as the theme sign, which immediately follows the verb stem. If the anchoring utterance participant functions as the AGENT of the event, the verb stem is suffixed by a so-called direct sign (-a). In contrast, if the anchoring utterance participant functions as the THEME of the event, the verb stem is suffixed by the so-called inverse sign (-ok). (72)

a.

nit´aa´ wayakiaa nit-(w)aawayaki-a-wa 1-hit-dir-3sg ‘I hit him.’

Utt-participant = AGENT

Varieties of INFL: Tense, Location, and Person

b.

nit´aa´ wayakioka nit-(w)aawayaki-ok-wa 1-hit-inv-3sg ‘He hit me.’

189

Utt-participant = THEME

Suppose the theme signs occupy a functional head (labeled X in (73)), which triggers the movement of an argument into its specifier position: the direct marker signals movement of the AGENT as in (73a), while the inverse marker signals movement of the THEME as in (73b).36 (73)

a.

IP

Utt-Part

I’

INFLPERSON

XP X’

AGENT

X -a

VP V

AGENT

V

b.

THEME

IP

Utt-Part

I’

INFLPERSON

XP X’

THEME

X -ok

VP V

AGENT

V

THEME

According to this analysis, INFLPerson does not directly order event participants relative to utterance participants. Rather, this ordering relation is mediated via an intervening phrase: whatever participant occupies the specifier position of XP is asserted to either coincide or not with the relevant utterance participant. But what is the function of XP? According to the Parametric Substantiation Hypothesis, UG provides a set of core abstract categories with specific functions, but the substantive content of these categories is not universally determined. This leads us to expect that the position of the theme signs (direct and inverse marking),

190

Elizabeth Ritter and Martina Wiltschko

which is specific to a Person-based system, might have a counterpart in the more familiar Tense-based systems. We speculate that this is indeed the case and that the relevant functional head corresponds to Aspect in a Tense-based system, as illustrated in (74). (74)

IP Utt-Part

I’

INFLPERSON

ASPECTP

AGENT

ASPECT’ ASPECT -a

VP V

AGENT

V

THEME

If this approach is on the right track, we need to establish that Person-based Aspect (i.e., direct/inverse marking) is functionally equivalent and formally identical to the more familiar Tense-based aspect.A proper treatment of Aspect within the Parametric Substantiation Hypothesis is beyond the scope of the present paper, and so we limit ourselves to a few speculative remarks. We have thus far assumed a simplified view of the role of Tense, namely that it serves to order the event time relative to the utterance time. However, as Klein (1994, 1995) notes, this view is too simplistic. Rather, the relation between event time (his situation time) and utterance time is better analyzed as an indirect relation mediated by a third time that Klein calls Assertion Time or Topic Time and which is the time about which an assertion is made. Thus, a more complete representation of a clause in a Tense-based system is shown in (75) (see Demirdache and Uribe-Etxebarria 1997, 2000). (74)

IP Utt-time

I’ INFLTENSE

ASPECTP

Assertion-Time

ASPECT’

ASPECTTENSE Ev-time

VP …

Varieties of INFL: Tense, Location, and Person

191

Observe that this clausal structure of a Tense-based system is remarkably similar to the clausal structure we have proposed for the Person-based system of Blackfoot. Instead of the assertion time introduced in a Tense-based system, we may talk about an assertion-participant, that is, the participant about which an assertion is made. This much is in line with the main idea of the Parametric Substantiation Hypothesis: one and the same abstract functional category can be associated with different substantive content across languages. So just like INFL, Aspect may be substantiated by Tense, Location, or Person. Furthermore, on principled grounds we expect that the content of INFL will influence the content of Aspect (or vice versa): for example, if the substantive content of INFL isTense, it orders times and therefore Aspect must make a time available as well. Consequently, we expect Aspect to be temporal in a Tense-based system but participantoriented in a Person-based system and spatial in a Location-based system. In this way, we suggest that the direct/inverse system of Blackfoot is functionally equivalent to a temporally based aspectual system such as the one found in English.37

4.4.

Summary

This concludes our discussion of Blackfoot, which we analyze as a system in which Person substantiates INFL. We have seen that INFLPerson (as it manifests itself in the form of Blackfoot person prefixes) is formally identical and functionally equivalent to INFLTense and INFLLocation . It is obligatory, uniquely marked, may remain silent, and may enter into a relation with the higher head, COMP. In addition, we have seen Blackfoot-internal evidence to the effect that person marking that instantiates INFLPerson is formally distinct from person marking that serves to signal an agreement relation. Blackfoot has both types of person marking, and these two types are formally distinct in a way that is consistent with the proposed analysis. Finally, we have seen that a system in which Person substantiates INFL is necessarily different from a system in which either Tense or Location substantiates INFL. This is because there are two utterance participants (as opposed to one utterance time or utterance location) and there can be two event participants. We have briefly sketched avenues of analysis for these particular properties of a Person-based INFL system. Specifically, we have proposed that the two utterance participants are accommodated in a complex INFL-shell structure (following work by Speas and Tenny 2003). And we have argued that the Blackfoot direct/inverse system that serves to order only

192

Elizabeth Ritter and Martina Wiltschko

one of two available event participants is the functional equivalent of temporal Aspect in a Person-based system.

5. Conclusion In this paper we have investigated the nature of functional categories in three unrelated languages: English, Halkomelem, and Blackfoot. Even a superficial look at the categories that need to be expressed obligatorily in each of these languages reveals significant cross-linguistic variation. This has in fact been one of the main claims within the tradition of the American structuralists: the inventories of grammatical categories differ across languages. The main goal of the present paper was to understand this observation within a universalist framework, namely the Principles and Parameters framework in its Minimalist version (Chomsky 1995 and subsequent work). We have argued that a cartographic approach to the problem is untenable. Consider again what a cartographer would say in light of the empirical findings reported here. The source of cross-linguistic variation reduces to a languagespecific choice in which of the functional categories provided by UG are overt and which ones remain silent (see, for example, Cinque 1999 for this view). Accordingly, the languages under investigation here would be analyzed as follows: in English only Tense is overt, in Halkomelem only Location is overt, and in Blackfoot only Person is overt. Alternatively, it might be the case that language variation reduces to a choice among different categories (see, for example, van Gelderen 1995, Thrainsson 1996, and Wiltschko 2002 for such an approach). This cartographic approach to variation is illustrated in Table 4.

Table 4. Language variation in categories within a cartographic approach Tense Location Person

English overt silent silent

Halkomelem silent overt silent

Blackfoot silent silent overt

However, the cartographic approach towards categorial variation faces problems. Specifically, this approach has nothing to say about: i) the complementary distribution of categories across the three languages; ii) the formal identity of the three categories (Tense, Location, and Person); or iii) the functional equivalence of these categories.

Varieties of INFL: Tense, Location, and Person

193

In this paper, we have introduced an alternative to cartography, namely the Parametric Substantiation Hypothesis. This hypothesis shares with cartographic approaches the assumption that UG makes available a hierarchically organized inventory of functional categories. At the same time, however, it denies one of the core tenets of cartography, namely the assumption that functional categories are universally associated with fixed substantive content. Rather, we have explored the hypothesis that functional categories are substantiated by different semantic concepts across languages. Specifically, we have argued that the functional category INFL, which universally serves to anchor the reported event to the utterance, can be associated with different kinds of substantive content. The only restriction on the content of INFL, which follows from its function, is that it be deictic.And in fact we have seen that the major deictic categories, Tense, Location, and Person, can indeed be analyzed as substantiating INFL. This is illustrated in Table 5. Table 5. Parametric Substantiation Hypothesis Language

INFL category

Function of INFL

English

Tense

anchors reported event to utterance

Halkomelem

Location

anchors reported event to utterance

Blackfoot

Person

anchors reported event to utterance

The Parametric Substantiation Hypothesis predicts precisely the state of affairs that we have seen to hold, and that proved problematic for a cartographic approach: the three categories Tense, Location, and Person are functionally equivalent, formally identical, and in complementary distribution. We have also seen evidence that the substantive content of INFL might determine other properties of the clause: for example, we have analyzed the direct/inverse system of Blackfoot as the functional equivalent of Aspect in a Person-based INFL system. And we would expect Aspect to have spatial content in a language with Location-based INFL. Furthermore, we expect that any property of clause structure that is dependent on the substantive content of INFL will be affected by this kind of variation. A case in point is the licensing of nominal arguments. On some accounts, structural case is directly linked to the substantive content of INFL (i.e., Tense; Pesetsky and Torrego 2001). If this is indeed so, we expect that nominal licensing in Location- and Person-based systems will differ significantly from that of Tense-based languages. Specifically, we might expect the absence of case marking altogether, and in fact we have previously argued that this is indeed the case (see Wiltschko 2003 for Halkomelem, and Ritter and

194

Elizabeth Ritter and Martina Wiltschko

Rosen 2005 for Blackfoot). Instead we might expect that nominals are pervasively marked for Location or Person. This appears to be the case, as discussed in detail in Wiltschko (to appear). These consequences of the present proposal are summarized in Table 6. However, a detailed investigation of these properties has to await future research. Table 6. Consequences of the Parametric Substantiation Hypothesis

INFLTense INFLLocation INFLPerson

Aspect

Nominal licensing

temporally based spatially based participant based

DPTense = case DPLocation = location marking DPPerson = participant marking

Finally, we note that the Parametric Substantiation Hypothesis has many more implications that we were not able to address in this paper. For example, we might expect that INFL can be substantiated by deictic categories other than the ones discussed here. Social deixis as manifested in the honorific system of Japanese comes to mind. Moreover, if the Parametric Substantiation Hypothesis is on the right track, we expect that other core categories will be affected as well. We have already seen that this might indeed be so for Aspect, but other functional categories such as COMP and DET need to be investigated from this angle as well. The Parametric Substantiation Hypothesis as outlined in this paper thus opens up an entire new line of research into the nature of categorization.

Notes 1.

2.

3.

In this paper we use small capital letters to indicate the morpho-syntactic categories of Tense, Location and Person, and lower case for tense, location and person marking, which may be realized either on the verb or in the functional heads, Tense, Location and Person, respectively. In Upriver Halkomelem, the obligatoriness of location marking is not straightforwardly obvious since this dialect has introduced a zero auxiliary serving as an alternative to the overt [+distal] auxiliary li (see section 3.2.4 for discussion). In the other dialects of Halkomelem, however, this zero auxiliary is not available, and thus the obligatoriness of the auxiliary is more straightforwardly observable. The following abbreviations are used in this paper: ai – animate intransitive; arg – argument; aux – auxiliary; C – comp; coin – coincidence; conj – conjunctive; cont – continuative; det – determiner; dir – direct; dur – durative; ev – event; I – INFL; imp – imperative; in – inanimate; inchoat – inchoative; indep – independent; int – intensive; inv – inverse; loc – location; neg – negative; nom – nominative; o –

Varieties of INFL: Tense, Location, and Person

195

object; obl – oblique; obv – obviative; pass – passive; pl – plural; poss – possessive; pro – pronoun; prox – proximate; Q – question; s – subject; sg – singular; ta – transitive animate; ti – transitive inanimate; tns – tense; trans – transitivizer; utt – utterance; V – verb; 1 – first person; 2 – second person; 3 – third person. 4. More precisely, we assume that INFL anchors a reference argument (introduced by Aspect) to the utterance, but for the core of the paper we can abstract away from this. See section 4.3.2 for some discussion. 5. We assume that future is not a Tense but instead contains a modal component (En¸c 1996; Copley 2004). Therefore, a [-coin] setting of INFL will only result in a past interpretation. See section 3.2.4 for a brief discussion of the future in Halkomelem. 6. /i and ni are the locative auxiliaries of the Downriver dialect of Halkomelem corresponding to the Upriver auxiliaries i and li, respectively. 7. We have confirmed Galloway’s and Suttles’generalizations with our consultant using video clips to elicit judgments. 8. Note that the impossibility for a future interpretation here implies that the future must contain a modal component (see note 5). However, see Matthewson (2006), who argues that the impossibility for a future interpretation in similar contexts shows that there must be a phonetically empty Tense which restricts the event time to either past or present. 9. In Upriver Halkomelem a past interpretation is equally acceptable in this context. Although Suttles (2004) only mentions the present interpretation for this particular example, we assume that this is an artifact of the description. (Unfortunately, no speakers of Downriver Halkomelem remain to confirm this prediction.) 10. If more than one auxiliary is used, only the highest one is inflected: (i) John has been dancing. (ii) *John has is dancing. (iii) *John have is dancing. 11. Whether INFL is instantiated by a bound form (inflectional suffixes in English) or a free form (locative auxiliaries in Halkomelem) is independent of the substantive content of INFL (cf. Borer 2005). 12. One might argue that the co-occurrence of li and i is independently ruled out for reasons of semantic compatibility. That is, the event location cannot simultaneously coincide and not coincide with the utterance location. However, the point is still valid because in the context of yes/no questions the auxiliary li loses its locative force (see section 3.2.2) and should thus be compatible with the [proximate] auxiliary i, contrary to fact. 13. These examples without the question particle -a are ambiguous between interrogative and declarative interpretations. In this section we focus on the significance of the interrogative interpretation. 14. On general properties and motivation forT–to-C movement, see Pesetsky and Torrego (2001). 15. See also Davis (2000) for the postulation of an empty auxiliary in clitic-initial constructions in Lillooet.

196

Elizabeth Ritter and Martina Wiltschko

16. The temporal interpretation of these sentences differs if the verb is marked as continuative. (See Bar-el et al. 2003 for a description of the facts.) 17. The fact that there is only one covert locative auxiliary in Upriver Halkomelem is not surprising: If both proximate and distal locative auxiliaries were covert, there would likely be an intolerable amount of ambiguity in the system. (English also has only one covert tense marker, i.e., present in context of first, second, and third non-singular subjects.) We have no insight to offer as to why the distal, rather than the proximate, locative auxiliary is covert in Upriver Halkomelem. Note, however, that a covert locative auxiliary is not required for this to be an INFL category, as evidenced by the fact that in Downriver Halkomelem both locative auxiliaries have phonetic content. 18. Note that the situation is different with third person subjects, which are not associated with a subject clitic. We discuss such examples below. 19. It suggests, however, that the distribution of this second-position clitic is syntactically, not phonologically determined. 20. Another Salish language that allows for this order is Squamish, a closely related Central Coast Salish language (see Bar-el et al. 2003 for discussion). 21. Jeroen van Craenenbroeck (p.c.) raises the very interesting question of whether it would be possible to get a past tense interpretation in dissociation contexts, e.g., someone listening to himself on tape, or looking at a picture or movie of himself. Unfortunately, we have no data that would answer this question, and speakers are no longer available for elicitation. 22. There are contexts, however, in which third person subjects are associated with a third person clitic, namely in so-called subjunctive clauses (Galloway 1993). As discussed in Bar-el et. al. (2003), the same pattern holds: in the absence of an auxiliary that triggers V-to-INFL movement, a future interpretation obtains. 23. We assume that the unacceptability of the past tense suffix -ed in both infinitives and imperatives has the same explanation, i.e., that INFL lacks [± past] in these two constructions, and consequently, tense marking is impossible. Note that an alternative account of (38) that asserts that the infinitive marker to blocks [± past] cannot be extended to (39), where there is nothing in INFL to block the insertion of past tense marking. 24. The distribution of the auxiliaries in embedded clauses is still subject to investigation. That is, there appear to hold certain restrictions on the occurrence of the proximate auxiliary i (Thompson, in preparation). 25. Note, however, that infinitives are different in Halkomelem: they contain agreement, making them more akin to inflected infinitives in Greek. The crucial point is that location marking is unavailable in these contexts. 26. Specifications of location are sometimes obligatory in English as well, but their obligatoriness is then dependent on the main predicate and is not a function of the clause, as in (i)–(ii): (i) He put the book *(on the shelf). (ii) He used to live *(on Vancouver Island).

Varieties of INFL: Tense, Location, and Person

197

27. Blackfoot verbs belong to different paradigms, called orders in the Algonquianist tradition, which are distinguished by the distribution of person prefixes and the form of order and agreement suffixes. In current research, we are exploring the hypothesis that it is the order suffixes, rather than the person prefixes that instantiate INFLPerson in Blackfoot. One argument in favour of this alternative is that the form of the order suffixes expresses a contrast between local (1st and 2nd person) and non-local (3rd person) participants, which is parallel to the temporal and spatial contrasts discussed above. 28. The third person prefix has a particularly complex distribution: In the independent paradigm, the prefix ot- is only used on transitive verbs with an obviative (a.k.a. fourth person) external argument and a proximate internal argument. Otherwise, the third person prefix remains silent in this context. In the conjunctive paradigm, ot- is used whenever there are only third person participants, and in the subjunctive and imperative paradigms, which are characterized by a lack of person prefixes, ot- is never used (see Frantz 1991 for details). 29. This leaves open the question as to what counts as the relevant event participant in case there are two (AGENT and THEME, for example). See section 4.3.2 for discussion. 30. Our Siksik´a Blackfoot consultant allows both a past and present time interpretation for such an example. But see Matthewson and Reis Silva (2007), who report that in the absence of a durative, only a past interpretation is available for their Kainaa Blackfoot consultant. At this point, we do not know whether this is due to a dialectal difference, but we have replicated these results with our Siksik´a consultant over a number of sessions and with different predicates. 31. In fact we predict that both Blackfoot and Halkomelem should lack the effects of structural case. Wiltschko (2003) shows that this prediction is also borne out in Halkomelem: like Blackfoot, this language lacks morphological case, case-motivated A-movement, and A-binding. Thus, the case facts also provide further support for the claim that Halkomelem does not have a functional category Tense, in line with the Parametric Substantiation Hypothesis. 32. We have not yet found any candidates for person marking that lack their substantive content. 33. Throughout this paper we use AGENT and THEME as the labels for these thematic roles. They correspond to the traditional Algonquian terms Actor and Goal, respectively. 34. Note that the presence of agreement in subjunctives and imperatives supports the view that the substantive content of INFL (Tense, Location, and Person) is independent of agreement. Thus, Blackfoot subjunctive clauses are formally similar to agreeing infinitives/subjunctives in Greek: both require a negative specification for the substantive content of INFL ([-Person] in Blackfoot and [-Tense] in Greek) while agreement is still present.

198

Elizabeth Ritter and Martina Wiltschko

35. The prefix na- is only used by speakers of the Siksik´a dialect. Frantz (1991) analyzes it as a past tense marker, but see Bliss and Ritter (2007) for arguments against this treatment. 36. Compare this to Bruening (2001), an analysis in which direct marking involves movement of the AGENT into SpecIP, and inverse marking involves movement of the THEME into SpecIP. This analysis would predict that there is a uniquely identifiable grammatical relation subject in Blackfoot, but this does not appear to be the case (see Ritter and Rosen 2005, Ritter and Wiltschko 2005). 37. If indeed the substantive content of the functional category Aspect is dependent on the substantive content of INFL in a given language, then we expect Aspect in Halkomelem (a Location-based system) to be spatial in nature. We hypothesize that this might be a fruitful line of research: Halkomelem has a set of directional auxiliaries that might be amenable to such an analysis, but we will have to leave a more thorough discussion for future research. In this context, we will also need to address another issue: how do we analyze the types of morphemes and constructions that appear to be aspectual in the standard temporal sense (i.e., imperfective, perfective, continuative, etc.) and which both Halkomelem and Blackfoot appear to have in their inventory?

References Abusch, Dorit 1988

Sequence of tense, intensionality and scope. Proceedings of WCCFL 7. Stanford: Stanford Linguistics Association, Stanford University. Pp. 1–14. Bar-el, Leora, Carrie Gillon, Peter Jacobs, Linda Watt and Martina Wiltschko 2003 The position of subject clitics and its effect on temporal interpretation in Skwxw´u7mesh and Upriver Halkomelem. In: Donna B. Gerdts and Lisa Matthewson (eds.), Studies in Salish Linguistics in Honor of M. Dale Kinkade, 8–29. University of Montana. Occasional Papers in Linguistics No.17. Bliss, Heather and Elizabeth Ritter 2007 Grammaticalizing information status in Siksik´a Blackfoot:A tenseless analysis. Paper presented at WSCLA 12. University of Lethbridge, Lethbridge, Alberta, Canada. Borer, Hagit 2005 Structuring Sense. Oxford: Oxford University Press. Bruening, Benjamin 2001 Syntax at the Edge: Cross-Clausal Phenomena and the Syntax of Passamaquoddy. Ph.D. diss., Department of Linguistics and Philosophy, MIT.

Varieties of INFL: Tense, Location, and Person Cheng, Lisa 1991

199

On the Typology of wh-questions. Ph.D. diss., Department of Linguistics and Philosophy, MIT.

Chomsky, Noam 1995 The Minimalist Program. Cambridge, MA: MIT Press. Cinque, Guglielmo 1999 Adverbs and Functional Heads. Oxford: Oxford University Press. Copley, Bridget 2004 The Semantics of the Future. London: Routledge. Davis, Henry 2000 Remarks on Proto-Salish Subject Inflection. International Journal of American Linguistics 66: 499–520. Demirdache, Hamida and Myriam Uribe-Etxebarria 1997 The Syntax of Temporal Relations: A Uniform Approach to Tense and Aspect. In: Emily Curtis, James Lyle, and Gabriel Webster (eds.), Proceedings of WCCFL 16, 145–159. Stanford: CSLI Publications. Demirdache, Hamida and Myriam Uribe-Etxebarria 2000 The Primitives of Temporal Relations. In: Roger Martin, David Michaels and Juan Uriagereka (eds.), Step by Step: Essays on Minimalist Syntax in Honor of Howard Lasnik, 157–186. Cambridge, MA: MIT Press. En¸c, M¨urvet 1987 Anchoring Conditions for Tense. Linguistic Inquiry 18: 633–657. En¸c, M¨urvet 1996 Tense and Modality. In: Shalom Lappin (ed.), The Handbook of Contemporary Semantic Theory, 345–358. Oxford: Blackwell. Frantz, Donald G. 1971 Toward a Generative Grammar of Blackfoot. Summer Institute of Linguistics Publications in Linguistics and Related Fields 34. Norman: Summer Institute of Linguistics of the University of Oklahoma. Frantz, Donald G. 1991 Blackfoot Grammar. Toronto: University of Toronto Press. Galloway, Brent 1993 A Grammar of Upriver Halkomelem. Berkeley/Los Angeles/London: University of California Press. Gelderen, Elly van 1995 The Rise of Functional Categories. Amsterdam/Philadelphia: John Benjamins. Hale, Kenneth L. 1986 Notes on world view and semantic categories: Some Warlpiri examples. In: Pieter Muysken and Henk van Riemsdijk (eds.), Features and Projections, 233–254. Dordrecht: Foris.

200

Elizabeth Ritter and Martina Wiltschko

Jelinek, Eloise and Andrew Carnie 2003 Argument hierarchies and the mapping principle. In: Andrew Carnie, Heidi Harley and Mary Willie (eds.), Formal Approaches to Function: Papers in honor of Eloise Jelinek, 265–296. Philadelphia: John Benjamins. Klein, Wolfgang 1994 Time in language. London: Routledge. Klein, Wolfgang 1995 A time-relational analysis of Russian aspect. Language 71: 669–695. Larson, Richard 1988 On the double object construction. Linguistic Inquiry 19: 335–391. Matthewson, Lisa 2005 On the absence of tense on determiners. Lingua 115: 1697–1735. Matthewson, Lisa 2006 Temporal semantics in a supposedly tenseless language. Linguistics and Philosophy 29: 673–713. Matthewson, Lisa and Amelia Reis Silva 2007 An instantaneous present tense in Blackfoot. Paper presented at SULA 4, University of S˜ao Paulo, S˜ao Paulo Brazil. Pesetsky, David and Esther Torrego 2001 T-to-C movement: Causes and consequences. In: Michael Kenstowicz (ed.), Ken Hale: A Life in Language, 355–426. Cambridge, MA: MIT Press. Pollock, Jean Yves 1991 Verb movement, universal grammar, and the structure of IP. Linguistic Inquiry 20: 365–424. Ritter, Elizabeth and Sara Thomas Rosen 2005 Agreement withoutA-positions:Another look atAlgonquian. Linguistic Inquiry 36: 648–660. Ritter, Elizabeth and Martina Wiltschko 2005 Anchoring Events to Utterances without Tense. Proceedings of WCCFL 24, 343–351. Somerville, MA: Cascadilla Proceedings Project, Cascadilla Press. Ritter, Elizabeth and Martina Wiltschko In prep. The Composition of INFL. Ms. University of Calgary and University of British Columbia. Speas, Margaret and Carol L. Tenny 2003 Configurational properties of point of view roles. In: Anna Maria Di Sciullo (ed.), Asymmetry in Grammar, 314–344. Amsterdam: John Benjamins. Suttles, Wayne 2004 Musqueam Reference Grammar. Vancouver: UBC Press.

Varieties of INFL: Tense, Location, and Person

201

Thompson, James In prep. Nominalized clauses in Upriver Halkomelem. Ph.D. diss., Department of Linguistics, University of British Columbia. Thrainsson, H. 1996 On the (non-)universality of functional categories. In: Werner Abraham, Samuel David Epstein, H. Thr´ainsson and Jan-Wouter Zwart (eds.), Minimal Ideas: Syntactic Studies in the Minimalist Framework, 253–281. Amsterdam/Philadelphia: John Benjamins. Uhlenbeck, C.C. 1938 A Concise Blackfoot Grammar. North Holland: Amsterdam. Wiltschko, Martina 2002 Sentential negation in Upriver Halkomelem. International Journal of American Linguistics 68: 253–286. Wiltschko, Martina 2003 On the interpretability of tense on D and its consequences for Case theory. Lingua 113: 659–696. Wiltschko, Martina To appear What’s in a determiner and how did it get there? In: Jila Ghomeshi, Ileana Paul and Martina Wiltschko (eds.), Determiners: Universals and Variation. Amsterdam/Philadelphia: John Benjamins. Wurmbrand, S. 2003 Infinitives: Restructuring and Clause Structure. Berlin: Mouton de Gruyter. Wurmbrand S. 2006 Infinitives are tenseless. Proceedings of the 30th Annual Penn Linguistics Colloquium, Penn Working Papers in Linguistics 13.1: 407–420.

Zanuttini, Raffaella 1991 Syntactic properties of sentential negation: A comparative study of Romance languages. Ph.D. diss., Department of Linguistics, University of Pennsylvania.

CAT meets GO: Auxiliary inversion in German verb clusters Markus Bader and Tanja Schmid Abstract. German verb clusters with a modal verb in the perfect tense are exceptional by requiring auxiliary inversion, thereby deviating from the order constraint “selected verb before selecting verb” that normally holds in German. In the first part of this paper, we summarize prior evidence showing that native speakers of German are less restrictive than prescriptive grammars: auxiliary inversion is required but it can target any position in front of the modal verb and not just the cluster-initial one. Based on this evidence, we argue that base-generating verb clusters by means of the formal language CAT (Williams, 2003) is superior to a cartographic account which derives auxiliary inversion by means of syntactic movement. In the second part of this paper, we present two experiments that have addressed a certain idiosyncrasy of auxiliary inversion in German, namely its restriction to verb clusters of size three and greater. Our results show that native speakers are again less restrictive than prescriptive grammar by allowing auxiliary inversion in two-verb clusters too. We present a syntactic analysis which accounts for the observed grammaticality data by combining a CAT-based analysis of verb-cluster formation with the empty GO hypothesis of Riemsdijk (2002).

1. Introduction Modal verbs as well as a small set of other so-called ‘functional’verbs, including perception verbs and the verb lassen ‘to let’, have a range of properties which set them apart from the main bulk of German verbs. The property which has received most attention in the syntactic literature concerns the order among verbs when they cluster together in clause-final position. In accordance with the OV nature of German, a verb selecting another verb normally appears after the selected verb. An exception to this general rule is found when a modal verb (or any of the other verbs listed above) is put into the perfect or future tense. Consider, for example, a verb cluster consisting of a tense auxiliary, a modal verb, and a main verb. In a 3-verb cluster of this kind, the main verb has to precede the modal verb, as expected for an OV language, but the auxiliary does not necessarily have to follow the modal verb. In fact, a perfect tense auxiliary must obligatorily be put at the front of the verbal cluster according to prescriptive grammars of German. For the future tense auxiliary, fronting is optional. This property of German modal verbs is illustrated in (1).

204

(1)

Markus Bader and Tanja Schmid

. . . dass Peter ein Buch HAT lessen • . . . that P. a book has read ‘. . . that Peter had to read a book.’

mussen ¨ • must

The only position of the auxiliary in (1) that is grammatical according to the grammar of Standard German is the cluster-initial position. The other two possible positions – marked by bullets in (1) – are deemed ungrammatical. A further peculiarity of verb clusters of this kind is also visible in (1). Despite being in the perfect tense, the modal verb m¨ussen occurs as a bare infinitival form instead of a past-participle form, as is normally required in the perfect tense. This property of modal verbs is called ‘Infinitivus Pro Participio’ (IPP). What is illustrated in (1) with the help of a 3-verb cluster holds in similar ways for clusters of greater size. Whenever a verb cluster consisting of three or more verbs is headed by a modal verb in the perfect tense, the perfect auxiliary must appear in cluster-initial position according to the rules of Standard German. What is actually accepted as grammatical by native speakers of German, however, diverges from the Standard German pattern in a precisely stateable way. In a broad-scale investigation of verb-cluster formation in German, Bader and Schmid (to appear) and Bader, Schmid, and H¨aussler (2009) obtained standard binary grammaticality judgments in an experimentally controlled way. A major conclusion stemming from this investigation is that native speakers of German indeed require the perfect auxiliary to invert in front of the modal verb, but that they accept any position preceding the modal verb and not just the cluster-initial position. Since the observed optionality with regard to auxiliary placement was found to be independent of participants’ regional background, we have used the term Colloquial German for the grammar underlying the observed grammaticality judgments. Furthermore, Bader and Schmid (to appear) have shown how the verb-cluster analysis of Williams (2003) can be extended in order to account for the experimentally obtained grammaticality judgments. What was said so far holds for clusters with at least three elements, that is, clusters which minimally contain a tense auxiliary, a modal verb, and a main verb. The complete picture is more complicated, however, due to a second puzzling property of modal verbs in German. This is the fact that they can be used without a main verb as complement, as illustrated in (2) for Standard German. (2)

. . . dass Peter fr¨uh nach Hause (muss | gemusst HAT). that P. early to home must must-PART has ‘. . . that Peter (has | had) to go home early.’

CAT meets GO: Auxiliary inversion in German verb clusters

205

In (2), the modal verb seems to select a directional PP instead of a main verb. Some modal verbs can also appear together with NP arguments, but the usage illustrated in (2) is the most widespread one in that it is possible with all German modal verbs. Therefore, we will focus on this usage in the present paper. A first question raised by examples like this is how it is possible at all to combine a modal verb with a non-verbal complement. As shown in (2), this question also arises for modal verbs in the present tense, and is thus independent of the question of how verb clusters are serialized. As also shown in (2), modal verbs in Standard German occur in the expected order (selected verb before selecting verb) and the expected morphological form (past participle) when used without a verbal complement. In other words, modal verbs in this special usage behave like normal main verbs and not like modal verbs in clusters of size three or greater. The obvious syntactic differences between modal verbs used with and without verbal complement has led to a debate as to whether we are dealing with a case of lexical ambiguity here – with a verb like m¨ussen acting either as modal verb or as main verb – or whether there is only a single verb m¨ussen behaving differently depending on its syntactic context. In this paper, we contribute to this debate by testing a prediction that our prior work on verb clusters makes for 2-verb clusters. As said above, the rules of Colloquial German are more liberal than the rules of Standard German in allowing the auxiliary in any position preceding the modal verb. For a 3-verb cluster as in sentence (3) with the main verb fahren included, this means that the order V-Aux-Mod is accepted in addition to the Standard-German order Aux-V-Mod. (3)

. . . dass Peter nach Paris (fahren) HAT that P. to Paris drive has ‘. . . that Peter had to drive to Paris.’

mussen. ¨ must

As illustrated in (3), this order contains the subsequence Aux-Mod. A Standard German verb cluster can never contain this sequence. However, given that Colloquial German allows this sequence in longer clusters, it might also occur on its own as a 2-verb cluster, that is, the main verb fahren might be optional in (3). We present experimentally gathered grammaticality judgments showing that there is indeed a tight connection between 3-verb clusters ((3) including fahren) and 2-verb clusters ((3) excluding fahren): either speakers accept both, or they reject both. In order to account for this finding, we will combine the analysis that we have presented in Bader and Schmid (to appear) with the empty light verb (GO) analysis that is proposed in van Riemsdijk (2002) for modal verbs without verbal complements.

206

Markus Bader and Tanja Schmid

The organization of this paper is as follows. The next section summarizes the syntactic analysis of verb clusters of length three or greater which we have developed in Bader and Schmid (to appear). In section 3, we argue that our own analysis is superior to cartographic approaches to the syntax of verb clusters. In the remainder of this paper, we take a closer look at 2-verb clusters. Section 4 presents two studies that have obtained grammaticality judgments for 2-verb clusters in an experimentally controlled way. We integrate the results of this test into our existing analysis in section 5. The final section 6 contains a summary and a conclusion.

2. Verb clusters and CAT Verbs in clause-final position are ordered in German according to the schema in (4). (4)

Vn Vn−1 . . . V2 V1

The order among verbs in (4) transparently reflects the directionality of selection which is to be expected from an OV language like German. When a verb Vn selects another verb Vn+1 , Vn follows its complement Vn+1 . While most verb clusters in German strictly adhere to the schema in (4), some well-known exceptions also exist. Chief among them are verbal clusters with a modal verb in a complex tense form. According to the grammar of Standard German – as defined in prescriptive grammars (cf. Fabricius-Hansen et al. 2005) – the perfect tense auxiliary must obligatorily appear in cluster-initial position when it selects a modal verb; the future tense auxiliary, in contrast, is allowed to appear in cluster-final position, thereby respecting (4), but it can also show up in cluster-initial position, thereby behaving like its perfect-tense counterpart. In the syntactic literature, additional auxiliary positions in verb clusters containing modal verbs have occasionally been reported to be grammatical (e.g., Meurers 2000; Haider 2003). Furthermore, it has long been known that different regional variants of German exhibit a large amount of variation with regard to the order in which the verbs of a verb cluster are serialized (see Schmid and Vogel 2004, and references cited there). In order to have a firmer base for syntactic analyses of verb-cluster formation in German, we have run a series of experiments investigating which verb orders in verb clusters with three, four, or five verbs are accepted as grammatical by native speakers of German (see Bader and Schmid to appear, for 3- and 4-verb clusters and Bader et al. 2009, for 5-verb clusters). The participants of these studies were all students of the

CAT meets GO: Auxiliary inversion in German verb clusters

207

80 100

4 verbs

5 verbs

20

40

60

3 verbs

0

% grammati ca l

University of Konstanz; their task was to judge sentences as either grammatical or ungrammatical as fast as possible (see section 4.1.1 for a detailed description of the experimental procedure). Figure 1 presents selected experimental results for perfect tense 3-, 4-, and 5-verb clusters in which all verbs except the hierarchically highest one – the finite auxiliary – appear in completely nested order according to the schema in (4). For clusters of each size, the finite auxiliary occurred in any possible serial position.

1

2

3

1

2

3

4

1

2

3

4

5

Aux Position

Figure 1. Mean percentages of judgments ‘grammatical’ for verb clusters with a modal verb in the perfect tense. Data for 3- and 4-verb clusters are from Bader and Schmid (to appear); data for 5-verb clusters are from Bader et al. (2009).

The data in Figure 1 can be summarized as follows. First, verb clusters with the finite auxiliary in initial position received the highest percentages of judgments ‘grammatical’. Second, verb clusters in which the finite auxiliary occurs in cluster-final position, thereby following the modal verb, were rejected as ungrammatical most of the time. These two findings are in accordance with the grammatical rules of Standard German. With regard to the remaining positions of the auxiliary – that is, positions in which the auxiliary precedes the modal verb without occupying the initial position – the findings revealed by Figure 1 diverge from the rules of Standard German in a striking way. In Standard German, these positions are all ungrammatical. Nevertheless, the participants of our experiments judged verb clusters of this kind as only moderately worse than clusters with the auxiliary in initial position.1 In summary, when native speakers of German have to spontaneously assess the grammaticality of verb clusters containing modal verbs in the perfect tense, they accept the auxiliary in any position inside the verb cluster except in the final position, that is, in any position which precedes the modal verb. Since this result was found to be independent of the regional background of the speak-

208

Markus Bader and Tanja Schmid

ers, we have called the grammar underlying this pattern ‘Colloquial German’. Examples illustrating the data in Figure 1 are given in (5). In each sentence, the finite auxiliary appears in cluster-initial position. This is the position which is grammatical in Standard and in Colloquial German. The bullets in (5) mark additional positions in which the auxiliary is allowed to occur in Colloquial German, where ‘allowed to occur’ is defined as being accepted 50% or more of the time. (5)

a.

b.

c.

¨ dass sie das Auto [HATTE waschen that she the car has wash ‘that she had to wash the car.’ ¨ gewaschen • dass das Auto [HATTE washed that the car has ‘that the car had to be washed.’ ¨ gewaschen • dass das Auto [HATTE washed that the car has mussen.] ¨ must ‘that the car should have been washed.’

•

mussen.] ¨ must werden • be

mussen.] ¨ must

worden • been

sein • be

The data shown in Figure 1 are only a subset of the data on verb-cluster formation that we have obtained so far. Additional data show, among other things, that future tense 3-verb clusters are accepted with the auxiliary in any of the three possible positions. Furthermore, while Bader and Schmid (to appear) and Bader et al. (2009) only looked at clusters containing modal verbs, unpublished experiments have found similar results for perception verbs and lassen ‘make, allow’.2 In summary, the canonical verb-cluster template in (4) must be supplemented by the templates in (6) in order to capture those cases of verb-cluster formation which involve auxiliary inversion. (6)

Verb clusters with auxiliary inversion a. Standard German (Aux1 ) Vn Vn−1 Vn−1 . . . V2 (Aux1 [Fut]) b. Colloquial German (Aux1 ) Vn (Aux1 ) Vn−1 (Aux1 ) Vn−1 . . . (Aux1 ) V2 (Aux1 [Fut]) Notes: – V2 ∈ {modal verb, perception verb, lassen} – Aux = perfect or future auxiliary, Aux[Fut] = future auxiliary

CAT meets GO: Auxiliary inversion in German verb clusters

209

Standard and Colloquial German differ crucially in that the auxiliary inverts only to the cluster-initial position in Standard German whereas it inverts to any position in front of the modal verb in Colloquial German. In both Standard and Colloquial German, auxiliary inversion is obligatory in the perfect tense and optional in the future tense. We next summarize the syntactic analysis that we have developed in Bader and Schmid (to appear) for the data discussed above. This analysis is an application and extension of Williams (2003), who suggests the formal language CAT as part of Representation Theory. Three important properties of a CAT-based approach to verb-cluster formation are the following: First, verb clusters are base-generated and not derived by movement. Second, most of the information relevant for ordering is put into the lexicon. Third, the optionality seen above follows without additional assumptions. Williams (2003) presents a verb cluster analysis which avoids derivational complexity by providing the grammar with a means to base-generate all observed verb-order patterns. To this end, Williams introduces the Rule of Combination which is the core of the CAT language: (7)

Rule of Combination (Williams 2003: 205) X Y + Y Z → [X + Y]X Z

This rule is to be read as follows. ‘X Y ’ is a syntactic unit of category X which subcategorizes for a syntactic unit of categoryY. ‘Y Z ’ is accordingly a syntactic unit of category Y which subcategorizes for a syntactic unit of category Z. If ‘X Y ’ and ‘Y Z ’ are combined by the Rule of Combination, the resulting unit is of category X and subcategorizes for Z. In other words, the complex category ‘X Z ’ has the same type as the head ‘X Y ’ and the same subcategorization as the complement ‘Y Z ’. The Rule of Combination subsumes two rules from categorial grammar: functional application when Z is empty and functional composition when Z is not empty. The Rule of Combination operates on lexical items which can be associated with the three types of subcategorization information listed in (8). Grammars of individual languages are obtained by associating lexical entries with the relevant subcategorization information. (8)

a. b. c.

Type of complement: N vs. V vs. . . . Order of selection: left vs. right Level of complement: X0 vs. XN

Building on Williams’ (2003) own verb-cluster analysis, Bader and Schmid (to appear) have provided a CAT-based account of the experimental verb-cluster data

210

Markus Bader and Tanja Schmid

summarized above. Some sample lexical entries are given in (9). The direction of selection is specified by an arrow that points in the direction of the complement. In (9), selection is always to the left; these entries are thus all canonical given the OV nature of German. Applying the Rule of Combination to these entries, the syntactic structures in (10) result. (9)

Sample lexical entries for verbs in German a. Main verbs – VMain : DP← | VMain : PP← | VMain : DP PP← | . . . b. Modal and auxiliary verbs – VMod|Aux : V←

(10)

a.

VAux: DP← VMain: DP← gelesen

b.

VMod: DP←

VAux: V←

VMain: DP←

hat

lesen

VMod: V← will

The ability of the Rule of Combination to pass on the subcategorization frame of a complement to the newly formed unit solves an issue that all theories which base-generate verb clusters have to deal with: The non-locality of headargument combination inside a verb cluster. Taking (10-a) as an example, the main verb does not immediately combine with its DP complement but combines first with the perfect tense auxiliary. The subcategorization information of the main verb is not lost, however, but transferred to the complex auxiliary VAux . This kind of argument passing is shared by a whole family of syntactic accounts inspired by the mechanism of functional composition which was introduced by Geach (1970) into syntactic theory (e.g., Steedman 1983, Johnson 1986, Hinrichs and Nakazawa 1994, and much subsequent work). The optionality of auxiliary placement in Colloquial German verb clusters of the kind discussed above follows directly from the possibility of combining syntactic units by means of functional composition, as we show next. For Colloquial German, we posit the subcategorization frames in (11) for the perfect tense auxiliary haben and the future tense auxiliary werden. (11)

Subcategorization frames for Colloquial German tense auxiliaries: a. Perfect tense: haben – VAux : a. →VMod b. V← b. Future tense: werden – VAux : →VMod or V←

The two subcategorization frames of the perfect tense auxiliary stand in an elsewhere relation. The perfect auxiliary selects modal verbs to the right and all

CAT meets GO: Auxiliary inversion in German verb clusters

211

other verbs to the left. The future tense auxiliary has the same subcategorization frames, but here they stand in a disjunctive relation. This implies that the two options do not exclude each other: Selection of modals is to the right and selection of verbs in general is to the left. Since the label ‘verb’ (V) is a cover term for all kinds of verbs including auxiliaries and modals, modals can also instantiate the general case and be selected to the left.3 The Rule of Combination in (7) and the subcategorization frames in (11) are everything we need to account for the optionality that characterizes Colloquial German verb clusters with modal verbs in a complex tense form. For the case of 3-verb clusters, the resulting structures are shown in (12). (12)

a.

VAux: DP← VAux: →VMod-[VV]

VMod: DP← VMain: DP←

hat/wird

lesen

b.

VMod: V← wollen

VAux: DP← VMain: DP←

VAux: V← VAux: →V

lesen

hat/wird

c.

VMod: V← wollen

VAux: DP← VAux: VMod←

VMod: DP← VMain: DP← lesen

VMod: V←

wird

wollen

For verb clusters containing more than three verbs, similar considerations show that the syntactic means introduced so far allow the generation of exactly those variants which are actually accepted as grammatical. For reasons of space, we abstain from showing the trees for these complex clusters.

212

Markus Bader and Tanja Schmid

As briefly mentioned in the introduction, German allows 2-verb clusters consisting of modal verb and auxiliary. In such clusters, the modal verb has a non-verbal complement. Our analysis for Colloquial German given so far predicts that auxiliary inversion is also possible in these cases. Thus, a sentence like (13) is predicted to be grammatical in Colloquial German. (13)

. . . dass er ein Buch HAT ← wollen. that he a book has want ‘. . . that he wanted a book.’

Since this prediction has not been tested so far, we present relevant experimental evidence on this issue in section 4. In Standard German – to which we turn now – a sentence like (13) is ungrammatical. This is a consequence of the fact that verb cluster formation in Standard German is sensitive to the complexity of the cluster. Auxiliary inversion is not allowed in Standard German when the auxiliary embeds a single verb. Instead, auxiliary inversion requires that the auxiliary embeds a verb cluster, that is, a syntactic unit consisting of at least two verbs. Thus, auxiliary inversion does not occur in 2-verb clusters, and instead of (13) we get (14-a). In contrast to 2-verb clusters, clusters of at least three verbs, as in (14-b), allow or even require inversion. (14)

a.

b.

. . . dass er ein Buch gewollt← HAT. that he a book wanted has ‘. . . that he wanted a book.’ . . . dass er ein Buch HAT → lesen wollen. that he a book has read want ‘. . . that he wanted to read a book.’

The complexity requirement for auxiliary inversion that is found in Standard German cannot be captured by the subcategorization specifications given in (8) – syntactic category, order, and level information. Furthermore, in cases where an auxiliary is placed in a position preceding the modal verb, only a single position is allowed in Standard German, namely the cluster-initial position. This contrasts with Colloquial German in which an auxiliary can occur in any position preceding the modal verb. What we therefore need is a means to force the auxiliary to occur in initial position in complex verb clusters. Although we cannot make reference to the initial position directly, it is still easy to encode the necessary feature within the subcategorization information of the auxiliary selecting the modal verb. To achieve the desired effect, we replace the non-stem

CAT meets GO: Auxiliary inversion in German verb clusters

213

feature proposed in Williams (2003: 226) by the feature [VV-max] which is defined in (16). (15)

Additional complexity feature for verb clusters a. Verb cluster: [VV-max] b. Simple verb: [V]

(16)

Definition: A verb cluster counts as [VV-max] if it contains at least two verbs and all subcategorization requirements for verbs have been saturated.

The maximality requirement [VV-max] is imposed on verb clusters by Standard German auxiliaries. Only in cases of full auxiliary inversion as in hat lesen wollen (‘has read want’) is the complexity requirement [VV-max] fulfilled since in lesen wollen all subcategorization requirements for verbs have been saturated, as reflected by the label of this verb cluster, namely VMod : NP←. In contrast, verb clusters which are not maximal in the sense of [VV-max] carry the label VMod : V Main ← (see the tree in (12-b)). They cannot be combined with Standard German auxiliaries. We thus arrive at the subcategorization frames for Standard German tense auxiliaries in (17). (17)

Subcategorization frames for Standard German tense auxiliaries: a. Perfect tense: haben – VAux : a. →VMod−VV−max b. V← b. Future tense: werden – VAux : → VMod−VV−max or V←

In summary, Bader and Schmid (to appear) have presented an analysis of auxiliary inversion in verb clusters containing three or more verbs in Standard and Colloquial German. Two of the other main phenomena with regard to verbcluster formation – namely VP topicalization and verb projection raising – are also shortly addressed in Bader and Schmid (to appear). For reasons of space, we do not include this part of our analysis here. Still outstanding is the behavior of the auxiliary in 2-verb clusters in Colloquial German. The analysis summarized in this section predicts that the inverted order Aux-Mod is grammatical in Colloquial German in addition to the Standard German order Mod-Aux. Before we put this prediction to an empirical test, we first compare our CAT-based account of verb-cluster formation to an account in terms of the cartographic framework that has been developed in work by Rizzi, Cinque and others (e.g., Rizzi 1997, Cinque 1999, and much subsequent work) and ask how an analysis based on cartographic principles fares in comparison to our CAT-based approach.

214

Markus Bader and Tanja Schmid

3. CAT versus cartography: a comparison Looking at verb cluster formation more generally, two main approaches can be distinguished (see Wurmbrand 2006, for a detailed overview): (i) The base generation approach which assumes that each verb order variant in the verb cluster is directly generated by the grammar (e.g., Haider 1993), and (ii) the derivational approach in the tradition of Evers (1975) which assumes that all verb order variants are derived from a single underlying structure. Derivational theories can be further subdivided with regard to the clausal status of the embedded infinitival complement. Bi-clausal theories assume that the infinitival forms its own clausal domain (e.g., Hinterh¨olzl 2006). According to mono-clausal theories, infinitival and matrix verb are part of a single clause (e.g., Wurmbrand 2007). A cartographic approach to verb-cluster formation will necessarily fall into the class of derivational theories because cartography is built on two key assumptions of Kayne (1994). According to the first assumption, all phrases adhere to the order Spec-Head-Complement. Consequently, a head-initial VP must be assumed even for languages for which traditionally a head-final VP had been assumed. The verb-cluster languages German and Dutch have thus to be reanalyzed as having an underlying VO base structure (see Zwart 1996). The second assumption is that all languages share a universal base structure which reflects a fixed hierarchy of functional projections. Surface orders which diverge from the underlying base order are derived by movement operations. Taken together, these two assumptions imply one of the major tenets of cartography, namely the existence of a universal head-initial base structure. The cartographic approach has been extensively applied to the left periphery of the clausal domain, initiated by the seminal work of Rizzi (1997), and the structure of DP (e.g., Cinque 2005). Cinque (2004) has extended this approach to “restructuring” phenomena that are found in the verbal domain, concentrating on restructuring in Romance languages. A crucial aspect of the analysis presented in Cinque (2004) is that the sequence of verbs exhibiting restructuring properties is analyzed as belonging to a single clausal domain. This is achieved by postulating a hierarchy of functional projections above VP; the heads of these functional projections can be filled be various ‘functional’ verbs, including all kinds of auxiliaries as well as modal verbs. Thus, a cartographic approach along the lines of Cinque (2004) postulates a mono-clausal structure for verbal clusters of the sort considered here, a property which it shares with our CAT-based analysis presented in the preceding section.4 What is different, however, is the way verb order variation is accounted

CAT meets GO: Auxiliary inversion in German verb clusters

215

for, by movement in the cartographic approach and by base-generation in the CAT-based approach. To our knowledge, a full-fledged cartographic analysis of verb-cluster formation in the West-Germanic languages is still wanting. A first step in this direction has been taken by Svenonius (2007), who sketches a cartographic analysis of German verb-cluster formation with a special focus on verb order. We next discuss how the variation that is empirically observed with regard to the linearization of verbs in 3-, 4- and 5-verb clusters can be derived in accordance with cartographic principles. The particular derivations that we show are in the spirit of the proposals found in Svenonius (2007) concerning German verb clusters and the analysis of DP-internal word-order variation presented in Cinque (2005), although we have made certain simplifications which do not affect our main arguments. The first simplification concerns the hierarchy of functional projections above VP. For ease of presentation, we operate with the simplified functional hierarchy given in (18). This hierarchy contains five positions and thus subsumes all clusters discussed in this paper. Furthermore, when discussing particular verb clusters, we only include as much elements of the hierarchy in (18) as are required to provide a place for each element of the verb cluster. (18)

T > Mod > Perf > Voice > V

We begin by showing how a canonical verb cluster – that is, one adhering to the scheme in (4) – is derived. The example that we use for this purpose is shown in (19). The syntactic tree for the cluster in (19) is given in (20). (19)

. . . dass das Auto repariert 3 worden2 ist 1 that the car repaired been is ‘. . . that the car has been repaired.’

(20)

TP VoiceP VP V repariert3

T’ Voice’

tO

Voice worden2

T tVP

ist1

tVoiceP

216

Markus Bader and Tanja Schmid

The order among the verbs in (19)/(20) is the mirror image of the base order. This is achieved by first moving VP to the left of its selecting head Voice into the specifier of VoiceP, and then moving the resulting VoiceP to the left of its selecting head T into the specifier of TP. Consider next the case of a 3-verb cluster with a modal verb in the perfect tense. A relevant example of this kind is shown in (21). (21)

. . . dass Peter das Auto HAT 1 reparieren3 • that P. the car has repair ‘. . . that Peter has been able to repair the car.’

k¨onnen2 can

The two syntactic trees corresponding to the two possible auxiliary positions in (21) are provided in (22). (22)

a.

TP T

ModP

hat1

VP V

Mod’ tO

Mod

reparieren3

b.

tVP

können2

TP VP V reparieren3

T’ tO

T hat1

ModP tVP

Mod’ Mod

tVP

können2

The two trees in (22) share with the tree in (20) the first step of the derivation, namely the movement of the most deeply embedded phrase (VP) into the specifier of the next higher phrase (VoiceP in (20), ModP in (22)). For the cluster

CAT meets GO: Auxiliary inversion in German verb clusters

217

with the finite auxiliary in initial position, that is, (22-a), we can already stop here. In order to get the cluster with partially inverted auxiliary, that is, (22-b), an additional movement step is necessary. In contrast to the completely inverted cluster in (19)/(20), this final step does not target the complement of the highest verb as a whole, but only its specifier, that is, the VP that has been moved there in the initial step of the derivation. Let us consider next the case of 4-verb clusters, as illustrated in (23). (23)

¨ . . . dass das Auto [HATTE repariert • that the car has repaired ‘. . . that the car had to be repaired.’

werden • be

mussen.] ¨ must

The three elements embedded below the auxiliary appear in inverted order. This is achieved in a now familiar way by first moving VP (gewaschen) past Voice (werden) and the resulting VoiceP past Mod (m¨ussen). The resulting tree is shown in (24). (24)

TP T

ModP

hätte1

VoiceP VP V repariert4

Mod’ Voice’

tO

Voice

Mod tVP

tVoiceP

müssen2

werden3

The tree in (24) represents a 4-verb cluster with the finite tense auxiliary in initial position. In order to derive the two remaining auxiliary positions 2 and 3 from (24), an additional movement step becomes necessary. When only the VP repariert is moved to the left of h¨atte, h¨atte ends up in the second position of the cluster. When the larger VoiceP repariert werden is moved, a cluster with h¨atte in third position results. The derivation of 5-verb clusters as in (5-c) proceeds analogously. The main difference to the 4-verb case is that the structure in which the four elements below T have been inverted by successive movement gives rise to three additional structures instead of two. In [T [[[VP Voice t VP ] Perf t VoiceP ] Mod t PerfP , VP, VoiceP and PerfP are all possible targets for movement around T.

218

Markus Bader and Tanja Schmid

Summarizing the discussion so far, we have seen that all cluster-internal orders that have been found to be grammatically licit in Colloquial German are derivable from an underlying head-initial base structure. Since Standard German is a subset of Colloquial German, all Standard German verb clusters can of course be derived too. This implies that a comparison between the cartographic approach and the CAT-based analysis cannot revolve around the question of generative capacity. We will instead base our comparison on the issue of optionality as it is found in Colloquial German, and on the relationship between Colloquial and Standard German. As we argue now, a CAT-based analysis fares better than a cartographic analysis on both counts. Consider first the finding that the position of the auxiliary within a larger verb cluster shows a high degree of optionality in Colloquial German. As we have shown in the preceding section, in our CAT-based analysis this optionality follows automatically from the way the Rule of Combination applies to a sequence of items which consistently select to the left with the exception of the hierarchically highest element, which selects to the right. In a cartographic analysis along the lines sketched above, the observable optionality is much harder to state. Two alternative formulations of the derivations that must come out as grammatical are given in (25). (25)

a.

b.

Invert anything below the finite auxiliary, and then move either nothing, or the specifier of the auxiliary’s complement, or the specifier of the specifier of the auxiliary’s complement, . . . Invert anything below the finite auxiliary, and then optionally move VP, optionally pied-piping any material except the complete complement of the auxiliary.

For a working cartographic analysis of the verb-cluster data under consideration, the generalizations in (25) would have to be translated into a system of features which have the effect that all and only those parts of the verb cluster are moved which result in a well-formed linearization of the verbs. Since it is not our aim to provide such an analysis, we do not even try to provide the relevant feature system. We strongly suspect, however, that the triggers for accomplishing the necessary movement steps lead to a grammar that is more complex than a CATbased approach to verb-cluster formation, in particular because the observed optionality with regard to the position of the inverted auxiliary follows for free in the latter kind of approach.5 This conclusion becomes even stronger when we consider the relationship between Colloquial and Standard German. Under a CAT-based approach to verbcluster formation, the grammar of Colloquial German is less complex than the

CAT meets GO: Auxiliary inversion in German verb clusters

219

grammar of Standard German because only the latter must include the feature [VV-max]. For the case at hand, a grammar with optionality is thus simpler than a grammar without optionality. In effect, a highly idiosyncratic feature [VV-max] is needed to prevent this kind of optionality from occurring. Given that Colloquial German reflects native speakers’ spontaneous judgments, we surmise that it represents a natural grammar, and that Standard German imposes an artificial restriction on this grammar. A cartographic analysis along the lines sketched above makes the opposite claim. Under such an analysis, it is easier to state a grammar without optionality. Instead of the complex generalization in (25), a much simpler statement would do. As shown by a comparison of the tree in (20) for a canonical nested verb cluster with the trees in (22-a) and (24) for inverted verb clusters, an auxiliary selecting a modal verb differs from canonical auxiliaries in that it does not trigger the final movement step which normally moves the auxiliary’s complement to the left of the auxiliary. If we are correct in assuming that Colloquial German represents a less complex system than Standard German, the cartographic analysis thus implies the wrong relationship between the two grammars. We therefore conclude that a CAT-based analysis of verb-cluster formation is superior to a cartographic analysis.

4. Experimental evidence The CAT-based verb-cluster analysis summarized in section 2 was based on evidence from clusters with 3 or more verbs. An obvious question is how 2verb clusters fit into the larger picture. That this question arises at all is due to the well-known property of Germanic languages – with the exception of English – that modals may appear without an embedded verb (for discussion and further references, cf. Barbiers 2006; van Riemsdijk 2002). For German, this was already shown in the introduction. A further example is provided in (26). (26)

. . . dass Peter in die Stadt gemusst HAT. that P. to the town must-PART has ‘. . . that Peter had to go to town.’

In (26), the modal verb appears as past participle and precedes the auxiliary. In other words, the 2-verb cluster in (26) conforms to the default rules for verb clusters in German, and does not show the deviations found for clusters with modal verbs of size three or greater. According to the grammar of Standard German, (26) is the only grammatical form for clusters of this type.

220

Markus Bader and Tanja Schmid

A different situation holds for Colloquial German. Consider the tree in (27) which can be derived given the lexical specifications for Colloquial German in (11). (27)

VAux: DP← VMain: DP←

VAux: V←

lesen

VAux: →V

VMod: V← wollen

hat/wird

An unusual feature of the tree in (27) is the fact that the modal verb does not directly combine with its main verb complement. Instead, first modal verb and auxiliary are combined with each other and only then is the resulting complex combined with the main verb. This feature leads to an interesting prediction concerning verb order in 2-verb clusters. Since Colloquial German allows the derivation of substrings of the form Aux-Mod, a sentence like (28) might also be grammatical. In Standard German, in contrast, substrings of this form cannot be derived, and a sentence like (28) is therefore also impossible to derive. The relevant part of the tree for (28) is shown in (29). (28)

. . . dass Peter in die Stadt HAT mussen. ¨ that P. to the town has must. ‘. . . that Peter had to go to town.’

(29)

VP PP P[+ dir] …

VAux: V← VAux: →VMod hat

VMod: V← müssen

Although Colloquial German allows one to combine an auxiliary with a modal verb in the order Aux-Mod, a tree like the one in (29) cannot be the final representation for 2-verb clusters because it raises two issues. First, what happens with the subcategorization feature of the modal verb which requires a verb as

CAT meets GO: Auxiliary inversion in German verb clusters

221

complement? Note that this subcategorization feature is still present in (29) because this tree has been derived under the assumption that modal verbs have a single lexical entry, whether appearing in a 2- or 3-verb cluster. Second, how does the resulting verb cluster license a directional PP? Before addressing these questions, we present two experiments that have investigated the predicted relationship between 2- and 3-verb clusters. Experiment 1 tests the prediction of a close connection between 2- and 3-verb clusters containing modal verbs. In particular, it is predicted that someone who accepts the string Aux-Mod should do so with and without an additional main verb, and conversely for someone who rejects such a string. Thus, speakers should either accept both 3-verb clusters of the form V-Aux-Mod and 2-verb clusters of the form Aux-Mod, or reject them both. The subsequent Experiment 2 takes a closer look at the morphological form of the modal verb in 2-verb clusters.

4.1.

Experiment 1: the relationship between 2-verb and 3-verb clusters

Experiment 1 has two objectives. The first one is to determine whether German speakers accept non-standard 2-verb clusters of the form Aux-Mod at all. If so, the second objective is to test whether judgments on sentences with V-Aux-Mod clusters correlate with judgments on sentences with Aux-Mod clusters. Experiment 1 investigates both the perfect and the future tense of modal verbs. For modals in the perfect tense, only the infinitival form is used in this experiment. Possible effects of the morphological form are the topic of Experiment 2.

4.1.1. Method Participants. 24 students of the University of Konstanz participated in Experiment 1. In this and the next experiment, all participants were native speakers of German and naive with respect to the purpose of the experiment. Participants were either paid or received course credits for participation in the experiment. Materials. The materials for Experiment 1 consisted of 30 sentences each appearing in six different versions. All sentences consisted of a main clause followed by an embedded clause introduced by the complementizer dass (‘that’). The verb cluster of interest was always the final element of the embedded clause. Two factors were manipulated. The factor Verbal Cluster varied the form of the cluster which could be one of Aux-V-Mod, V-Aux-Mod, or Aux-Mod. The second

222

Markus Bader and Tanja Schmid

factor Tense varied the tense of the modal verb by using either a finite form of haben (perfect tense) or a finite form of werden (future tense) as finite tense auxiliary. All embedded clauses had a directional meaning. Sentences with a 3-verb cluster therefore contained a motion verb like gehen (‘to go’), fahren (‘to drive), steigen (‘to climb’), etc., and a directional PP specifying the target of the motion. Sentences with a 2-verb cluster were derived from sentences with a 3-verb cluster by omitting the motion verb. A sample sentence with all of its six versions is shown in Table 1. Three different modal verbs were used in this experiment: wollen (‘to want’), d¨urfen (‘may’), and m¨ussen (‘must’). Each modal verb appeared in 10 sentences. From the experimental material, 6 lists were prepared. Each list contained exactly one version of each sentence, with an identical number of sentences in each of the six conditions that result from crossing the two factors Verb Cluster and Tense. Each participant saw only one list. The sentences on each list were randomized individually for each participant and were presented interspersed within 108 filler sentences representing a wide variety of syntactic structures. Table 1. A sample sentence for Experiment 1. Ich weiß, dass Klaus im Sommer ans Meer . . . I know that K. in-the summer to-the sea Perfect

Aux-V-Mod

Future

V-Aux-Mod Aux-Mod Aux-V-Mod V-Aux-Mod Aux-Mod

hat has fahren hat wird will fahren wird

fahren drive hat wollen fahren drive wird wollen

wollen want wollen wollen want wollen

Translation for all conditions: ‘I know that Klaus (wanted|will want) to drive to the sea during the summer.’

Procedure. The following two experiments were run using the DMDX software developed by K.I. Forster and J.C. Forster at Monash University and the University of Arizona. Each trial began with the presentation of the words Bitte Leertaste dr¨ucken ‘Please press the space-bar’. After the space bar was pushed, a fixation point appeared in the center of the screen for 1050 ms. Thereafter, the sentence was presented in a word-by-word fashion with each word successively appearing in the center of the screen. The presentation time for each word was

CAT meets GO: Auxiliary inversion in German verb clusters

223

225 ms plus an additional 25 ms per character. There was no interval between words. Immediately after the last word, three question marks appeared on the screen, indicating to participants that they now had to judge the grammaticality of the sentence. Participants had to give their answer by pressing the right shift key for judging a sentence as grammatical and the left shift key for judging a sentence as ungrammatical. Type of response and response time were recorded automatically. If a participant did not respond within 2000 ms, the words “zu langsam” (“too slow”) appeared on the screen and the trial was finished. Each participant received at least 10 practice items before the experimental session started.

4.1.2. Results The percentages of judgments ‘grammatical’ for Experiment 1 are shown in Table 2.6 The results were analyzed by three-way ANOVAs which included the individual modal verbs (wollen, d¨urfen, m¨ussen) as an additional factor Modal Verb in addition to the two factors of main interest (Verb Cluster and Tense). Results were analyzed with either participants (F1) or items (F2) as an additional random factor. Table 2. Percentages of judgments ‘grammatical’ for Experiment 1. Standard errors by participants are given in parentheses. Perfect Future

AuxVMod 87 (4.1) 79 (6.1)

VAuxMod 71 (6.3) 60 (6.5)

AuxMod 61 (7.4) 50 (7.8)

The main effect of the factor Modal Verb reached significance (F1(2, 46) = 7.20, p < .01; F2(2, 58) = 8.22, p < .01), but none of the interactions involving this factor did (all p’s > .1). The reason for a significant effect of Modal Verb is the fact that sentences with the modal verb m¨ussen were more often judged as grammatical than sentences with either d¨urfen or wollen (77% for m¨ussen versus 63% for d¨urfen and 65% for wollen). Since neither of the interactions involving Modal Verb was significant, this means that sentences containing the modal verb m¨ussen were judged as more grammatical than sentences with one of the other two modal verbs across the board. Given that each modal verb occurred in ten different sentences, this finding might have come about because the particular ten sentences for the modal verb m¨ussen were somewhat more plausible than the sentences for the remaining modal verbs. The factor Verb Cluster also resulted in a significant main effect (F1(2, 46) = 16.53, p < .001; F2(2, 58) = 23.36, p < .001).This reflects the finding that sen-

224

Markus Bader and Tanja Schmid

tences with a Standard German cluster Aux-V-Mod received higher percentages of judgments ‘grammatical’ than sentences with a Colloquial German cluster V-Aux-Mod (83% vs. 65%) which in turn received higher ratings than sentences with a 2-verb cluster Aux-Mod, that is, clusters not containing a main verb (65% vs. 55%). Furthermore, judgments for perfect-tense sentences were about 10% higher than judgments for future-tense sentences, resulting in a significant main effect of the factor Tense (F1(1, 23) = 6.41, p < .05; F2(1, 29) = 6.82, p < .05). In contrast to the main effects, the interaction between Verb Cluster and Tense was not significant (both F-values < 1). The main purpose of Experiment 1 was to test the hypothesis of a tight connection between speakers’judgments on 2-verb and 3-verb clusters. In particular, speakers are expected to either accept both 3-verb clusters of the form V-AuxMod and 2-verb clusters of the form Aux-Mod, or reject both and only accept Standard German 3-verb clusters Aux-V-Mod. Thus, judgments on sentences in the 3-verb condition V-Aux-Mod should correlate positively with judgments on sentences in the 2-verb condition Aux-Mod. The relevant correlations between speakers’mean percentages in the various conditions of Experiment 1 are shown in Table 3 under the heading “Raw correlations.” Table 3. Correlations for Experiment 1. Perfect r

p

Future r

p

Raw correlations VAuxMod BY AuxMod AuxVMod BY AuxMod AuxVMod BY VAuxMod

.82 .40 .55

< .001 < .1 < .01

.87 .62 .52

< .001 < .01 < .01

Partial correlations AuxVMod BY AuxMod

.79

< .001

.82

< .001

As can be seen in Table 3, judgments on sentences with 2-verb clusters Aux-Mod strongly correlate with judgments on sentences with 3-verb clusters V-Aux-Mod, for both the perfect and the future tense. Note, however, that these two sentence types also correlate significantly with judgments on sentences containing Standard German clusters Aux-V-Mod. Since in Standard German the order Aux-VMod is either the only licit order (perfect tense) or one of two alternative orders (future tense), and since all of our participants were students who should be acquainted with this order, we might have expected that sentences with Aux-V-Mod clusters are accepted at a 100%. There are several “performance” reasons for why judgments on Aux-V-Mod clusters were actually below 100%. For example,

CAT meets GO: Auxiliary inversion in German verb clusters

225

the judgment task by itself is somewhat demanding because of the fairly rapid presentation rate and the short time window for giving a response; this can easily lead to occasional errors which are independent of the particular syntactic structures under investigation. Because performance factors affect all experimental conditions, we have to worry that the strong correlations between V-Aux-Mod and Aux-Mod sentences might be a side effect of individual differences in overall performance. One piece of evidence arguing against this possibility comes from the fact that the correlations between V-Aux-Mod and Aux-Mod are much higher than the correlations between these orders and the Standard German order Aux-V-Mod. Furthermore, we computed partial correlations between V-Aux-Mod and Aux-Mod sentences with Standard German Aux-V-Mod sentences taken as common factor measuring participant’s overall performance level.7 The resulting partial correlation coefficients, which are shown in the last line of Table 3, are almost as high as the original coefficients. This strongly argues that the observed correlations between V-Aux-Mod and Aux-Mod cannot be attributed to unspecific performance factors. Future

60 20

40

Aux-Mod

60 40 0

0

20

Aux-Mo d

80

80

100

100

Perfect

0

20

40

60

V-Aux-Mod

80

100

0

20

40

60

80

100

V-Aux-Mod

Figure 2. Judgments for V-Aux-Mod sentences plotted against judgments for Aux-Mod sentences, for both perfect tense (left side) and future tense (right side). Points represent single participants; stars represent sets of participants, with each beam standing for one participant.

A graphical depiction of these correlations is provided in Figure 2. For both perfect and future tense, Figure 2 shows a clear positive linear relationship between judgments forV-Aux-Mod sentences andAux-Mod sentences. Before we discuss the results of Experiment 1 in more detail, we first present Experiment 2.

226

Markus Bader and Tanja Schmid

4.2.

Experiment 2: morphological form and 2-verb clusters

In Standard German, the morphological form of the modal verb in the perfect tense correlates with the number of elements in the cluster and with the order among auxiliary and modal verb, as shown in (30). (30)

Cluster size: 2 verbs Cluster size: ≥ 3 verbs Aux before Mod – infinitive Mod before Aux past participle –

The infinitival form occurs in clusters with three or more verbs; in clusters of this size, the auxiliary must precede the modal. In clusters with 2 verbs, in contrast, the auxiliary follows the modal verb and the participle form of the modal verb has to be used. Experiment 2 investigates which of the two dimensions shown in (30) is responsible for the IPP effect: is the morphological form of the modal verb determined by the order among modal verb and auxiliary or by the complexity of the cluster? As shown by Experiment 1, 2-verb clusters of the form Aux-Mod with the infinitival form of the modal verb – even in the perfect tense (IPP) – are accepted to a considerable degree by speakers of Colloquial German. On the basis of this finding, we hypothesize that it is in fact the order among auxiliary and modal verb which is responsible for the IPP effect, and not the size of the cluster. Experiment 2 thus tests the prediction that the infinitival form of the modal verb, but not the participle form, should be accepted in clusters of the form Aux-Mod whereas the reverse should be true for clusters of the form Mod-Aux.

4.2.1. Method Participants and Procedure. 24 students of the University of Konstanz participated in Experiment 2. The procedure was the same as the one used in Experiment 1. Materials. For Experiment 2, we transformed the 30 sentences from Experiment 1 in such a way that each sentence appeared in six versions according to the two factors Tense/Form and Order. The factor Tense/Form jointly varied the tense of the verb cluster and the morphological form of the modal verb. This factor could take one of three values: perfect tense with infinitival modal verb, perfect tense with past-participle modal verb, and future tense with infinitival modal verb. In contrast to the preceding experiment, the current experiment only tests sentences in which the modal verb does not have a main verb complement. All clusters therefore consist of a modal verb and an auxiliary. According to

CAT meets GO: Auxiliary inversion in German verb clusters

227

Table 4. A sample sentence set for Experiment 2. The auxiliary in parentheses occurred either before the modal verb (order Aux-Mod) or after it (order Mod-Aux). Ich weiß, dass Klaus im Sommer ans Meer . . . I know that K. in-the summer to-the sea Perfect/Infinitive (hat) has Perfect/Participle (hat) has Future/Infinitive (wird) will

wollen want gewollt wanted wollen want

(hat) has (hat) has (wird) will

Translation for all conditions: ‘I know that Klaus (wanted|will want) to go to the sea during the summer.’

the factor Order, the modal verb either preceded the auxiliary (Mod-Aux) or followed it (Aux-Mod). Table 4 shows a complete sample sentence. As before, 6 lists of experimental sentences were created, with each list containing only one version of each sentence and an equal number of sentences within each factor combination. Each participant saw only one list which was randomized on an individual basis and presented together with 138 filler sentences.

4.2.2. Results Table 5 shows the results obtained in Experiment 2. These results were again analyzed by three-way ANOVAS including the factor Modal Verb in addition to the two factors of main interest, Tense/Form and Order. As expected given that the sentence material for the current experiment was derived from the material for Experiment 1, sentences with m¨ussen were judged better than sentences with either d¨urfen or wollen (60% for m¨ussen versus 54% for d¨urfen versus 52% for wollen). However, the advantage of sentences with m¨ussen was smaller in this experiment, and the factor Modal Verb therefore did not reach significance (F1(2,46) = 2.38, p = .10; F2 < 1), nor did any interaction involving Modal Verb (all p’s > .1). Of the two remaining factors, the factor Tense/Form was significant (F1(1,23) = 11.31, p < .01; F2(1,29) = 32.68, p < .001) but the factor Order was not (F1 < 1, n.s.; F2(2,58) = 1.00, n.s.). Importantly, the interaction between Order and Tense/Form was also highly significant (F1(2,46) = 53.19, p < .001; F2(2,58) = 95.70, p < .001). This interaction reflects the following finding. In the con-

228

Markus Bader and Tanja Schmid

Table 5. Percentages of judgments ‘grammatical’ for Experiment 2. Standard errors by participants are given in parentheses. ModAux AuxMod

Perfect/Part 87 (5.6) 24 (6.0)

Perfect/Inf 37 (7.4) 67 (7.4)

Future/Inf 64 (7.8) 48 (6.2)

dition perfect tense/past participle, sentences with order Mod-Aux were judged much better than sentences with order Aux-Mod (87% vs. 24%); in the condition perfect tense/infinitive, the reverse relation was found (37% vs. 67%). For sentences in the condition Future/Infinitive finally, Mod-Aux sentences were again judged better than Aux-Mod sentences (64% vs. 48%). There is clear difference to the condition perfect tense/past participle, however. First of all, even sentences with the standard German order Mod-Aux were judged as grammatical only 64% of the time in the future tense, in contrast to corresponding perfect tense sentences which received 87% judgments ‘grammatical’. This confirms the results of Experiment 1 that the future tense of modal verbs is less acceptable than the perfect tense. The 48% for the order Aux-Mod must be interpreted in relation to the already low value for the Standard German order Mod-Aux. The drop of about 16% between the Standard and the Colloquial German order is quite similar to the difference we see in the perfect tense, from 87% (Mod-Aux, participle) to 67% (Aux-Mod, infinitive).

4.3.

Discussion

The two experiments presented in this section show that many native speakers of German accept auxiliary inversion even in sentences containing a modal verb without verbal complement. In other words, auxiliary inversion is not confined to clusters with three or more elements but is also found in 2-verb clusters. The CAT-based analysis of verb clusters presented in Bader and Schmid (to appear) led us to expect a positive correlation between judgments on 3verb clusters with auxiliary-second order (i.e., V-Aux-Mod) and judgments on 2-verb clusters with auxiliary first-order (i.e., Aux-Mod). A grammar in which the perfect auxiliary haben selects modal verbs to its right without any further restrictions (Colloquial German) allows all three of Aux-V-Mod,V-Aux-Mod, and Aux-Mod. A grammar with the additional complexity feature [VV-max] on the tense auxiliaries (Standard German), in contrast, allows Aux-V-Mod but neither V-Aux-Mod nor Aux-Mod. The predicted correlation between judgments on 2-

CAT meets GO: Auxiliary inversion in German verb clusters

229

and 3-verb clusters was clearly borne out in Experiment 1, for modal verbs in both the perfect and future tense. Experiment 2 also provides new evidence on the IPP effect. The clear interaction between order (Aux-Mod vs. Mod-Aux) and morphological form (past participle vs. infinitive) rules out verb-cluster complexity as the trigger for the IPP effect. Instead, the results are by and large compatible with a rule stating that a modal verb in the perfect tense must appear as a bare infinitive instead of a past participle just in case the perfect auxiliary precedes the modal verb. A closer look at the results reveals, however, that the IPP effect might not even be contingent on the auxiliary’s position. In particular, sentences with perfect tense Mod-Aux clusters and an infinitive instead of the normally required participle were not sharply rejected. This additional finding suggests an order-independent IPP effect, that is, a rule simply stating that modal verbs in the perfect tense occur as bare infinitives. An implementation of this idea is presented below.

5. CAT meets GO: the final analysis With regard to auxiliary inversion in verb-cluster formation, Experiment 1 has provided evidence for two grammars: the Standard German grammar imposes the complexity feature [VV-max] on tense auxiliaries selecting a modal verb; Colloquial German lacks this complexity feature. Standard German therefore requires complete inversion of the finite auxiliary in clusters with three or more elements while at the same time prohibiting inversion in 2-verb clusters. Colloquial German, in contrast, allows both complete and partial inversion in clusters of all sizes. Furthermore, Experiment 2 has shown that the morphological form of the modal verb in 2-verb clusters is largely a function of the order between auxiliary and modal verb, although there was also some indication that the use of the bare infinitive (the IPP effect) is not strictly confined to the order Aux-Mod. What remains to be shown is how the preceding experimental results can be integrated into the analysis proposed in Bader and Schmid (to appear). When we discussed the generation of 2-verb clusters consisting of a modal verb and an auxiliary within CAT, the question arised how a modal verb, which normally selects a verbal complement, can combine with a non-verbal complement like a PP, as it did in the current experiments. Note that this issue is independent of the question of verb order. As already shown by example (2), the same alternation also occurs in the present tense. There are two main classes of accounts for modal verbs taking non-verbal complements. According to the first class, each modal verb has only a single lexical entry specifying that it can combine with a verbal complement. Under

230

Markus Bader and Tanja Schmid

this assumption, sentences which do not contain a verbal complement for the modal verb on the surface are either derived from an underlying structure containing such a verb which is deleted at a later point in the derivation (cf. Vanden Wyngaerd 1994), or they contain a phonetically empty light verb GO, as proposed in van Riemsdijk (2002). According to a second class of accounts, modal verbs are associated with two lexical entries, one for the modal use and a second one for the main verb use (cf. Barbiers 1995, 2006, and references cited there). Based on Swiss German data, van Riemsdijk (2002) has adduced an intriguing argument for the first kind of account. His argument rests on two premises. First, as in all Germanic OV languages, directional PPs in Swiss German cannot occur after a clause-final verb; second, Swiss German has Aux-Mod-V as the preferred order within 3-verb clusters and freely allows verb-projection raising. These two properties are summarized in (31). (31)

a. *C . . . V PP+dir b. C . . . Aux [VP . . . Mod [VP . . . V]]

Despite the ban on extraposed directional PPs, sentences like the following are grammatical in certain Swiss German varieties (example (10a) from van Riemsdijk 2002). (32)

¨ wele . . . das si n¨ome HAT [PP i d schuel] that she no-longer has wanted in the school ‘. . . that she no longer wanted to go to school’ a. *C . . . Aux [VP Mod PP+dir ] b. C . . . Aux [VP Mod [VP PP+dir GO]]

In (32), the modal verb is not accompanied by a main verb. If modal verbs used in this way were main verbs, a sentence like (32) should be ungrammatical because – as a comparison between its schematic representation in (32a) and the general pattern in (31a) shows – it would exhibit the forbidden extraposedPP pattern. If however, as argued by van Riemsdijk, (32) actually contains a 3-verb cluster with an empty verb GO (an empty light-verb of motion), its grammaticality is no longer a mystery. As a comparison of (32b) with the general pattern in (31b) shows, a sentence like (32) simply instantiates a verb-projection structure under the empty-GO hypothesis. This is a strong argument in favor of analyses postulating that modal verbs uniformly take verbal complements. The data cited in van Riemsdijk (2002) must accordingly be considered a benchmark test for theories of verb cluster formation. Nevertheless, our data prevent us from importing the empty GO analysis of van Riemsdijk (2002) into our own account without any further ado.

CAT meets GO: Auxiliary inversion in German verb clusters

231

The reason is that the availability of an empty GO together with the Standard German order Aux-V-Mod results in clusters of the form Aux-GO-Mod. Since GO is phonetically empty, the phonetic form of a 3-verb cluster Aux-GO-Mod is indistinguishable from the phonetic form of a 2-verb cluster Aux-Mod. This is shown in (33). (33)

VAux : PP← VAux : →V[VV ] AUX

VMOD : PP← VMain : PP← GO

VMod : V← MOD

A tree like (33) has the unwelcome consequence that it makes a 2-verb cluster of the form Aux-Mod grammatical in Standard German because such a cluster is really a 3-verb cluster under the empty light verb analysis, namely a 3-verb cluster with a phonetically empty main-verb.Thus, contrary to our results, speakers of Standard German and speakers of Colloquial German should both accept surface clusters of the form Aux-Mod if an empty verb GO were freely available. This problem only arises if 2-verb clusters Aux-Mod are underlyingly 3verb clusters Aux-V-Mod with an empty V. It does not arise if modal verbs in German could be ordinary lexical verbs selecting non-verbal complements. In this case, they would be able to select a directional PP directly instead of a verbal complement, and the subcategorization mismatch would no longer arise. Furthermore, sentences as in (34) (repeated from above) would not be mysterious anymore. (34)

. . . dass Peter in die Stadt gemusst HAT . that P. to the town must-PART has ‘. . . that Peter had to go to town.’

If m¨ussen in (34) were a main verb, this sentence conformed to the general rule that the perfect tense is formed with a past participle preceding the perfect auxiliary. Thus, a modal verb without a verbal complement, as in in die Stadt gemusst hat (‘to the town wanted has’), behaves overtly like any main verb with NP or PP complement. However, postulating that modal verbs in 2-verb clusters are generally main verbs cannot be the correct solution either. As shown by our experimental results, many native speakers of German accept 2-verb clusters of the form Aux-Mod in which the modal verb appears as bare infinitive, as in (35) (repeated from above).

232

(35)

Markus Bader and Tanja Schmid

. . . dass Peter in die Stadt HAT mussen. ¨ that P. to the town has must-INF ‘. . . that Peter had to go to town.’

The IPP effect and auxiliary inversion in verb clusters are confined to modal verbs and a few other functional verbs. Purely lexical verbs, in contrast, never appear in the bare infinitive and to the right of the selecting auxiliary in the perfect tense (*. . . , dass sie das Buch hat lesen ‘that she the book has read-Inf’), neither in Standard nor in Colloquial German. The Swiss German data that were used by van Riemsdijk (2002) in order to argue for the empty GO analysis point to the same conclusion. As already discussed above, if wele (‘want’) in sentence (32) were a main verb, it could not precede its PP complement. In sum, both morphological and syntactic properties indicate the existence of two types of 2-verb clusters involving modal verbs: either the order is Mod-Aux and the modal verb appears as past participle, or the order is Aux-Mod and the modal verb appears as infinitive (IPP). The order Aux-Mod is only grammatical in Colloquial German. In fact, as already implicated by the lexical entries in (11), modal verbs in Colloquial German behave like true modal verbs in clusters of any size, thereby allowing auxiliary inversion and the IPP effect even when they are not accompanied by a main verb, that is, in 2-verb clusters. This leaves us with two open issues. First, how is the directional PP in (35) licensed, and second, what happens to the subcategorization requirements of m¨ussen? To address these issues, we integrate the empty light verb analysis of van Riemsdijk (2002) with the CAT-based analysis of Bader and Schmid (to appear). To do so, we first have to introduce another ‘ingredient’ into our analysis: the notion of a spanning vocabulary provided by CAT. By way of illustration, (36) shows how the hierarchy of verbal functional elements is realized in English (from Williams 2003, p. 214). (36)

T > AgrS > Asp > AgrO > V |- - was - -| |- - - seeing - - -|

As indicated in (36), the auxiliary was and the main verb seeing both realize more than a single element of the verbal hierarchy. The auxiliary was simultaneously realizes T and AgrS , and the main verb seeing realizes the three elements Asp, AgrO , and V. The way we adopt the notion of a spanning vocabulary is shown in (37). (37)

Aux

>

Mod > VGO > PP |- - wollen - -|

CAT meets GO: Auxiliary inversion in German verb clusters

233

The basic idea is that a modal verb like wollen is able to realize simultaneously both itself and the empty verb GO. In a sense, GO gets incorporated into the modal verb. We will assume that a micro parameter in the sense of Kayne (2005) determines whether modal verb and empty GO are combined in the lexicon or in the syntax.As we show next, in Standard and in Colloquial German the parameter value is set to ‘combination in the lexicon’. Thus, in addition to the two simple lexical entries in (38), both variants of German have the complex lexical entry [Mod>VGO ] shown in (39), which enters syntax as one single element. (38)

a. b.

GO – V: XP[+dir]← wollen – Mod: Vx ←

(39)

wollenGO – [Mod>VGO ]: XP[+dir]←

The lexical entry in (39) allows the derivation of the tree in (40). This is the tree for a sentence with a 2-verb cluster Aux-Mod. (40)

The tree in (40) answers the two questions raised above. First, the empty verb GO satisfies the subcategorization feature of the modal verb which requires a verb as complement. When modal and GO are combined in the way indicated, the resulting lexical item is of the complex category [Mod>V GO ] and inherits the subcategorization properties of the empty motion verb; that is, the modal now subcategorizes for a directional XP. This answers the second question, namely how is the directional XP licensed. This XP is ultimately licensed by empty GO which transmits its subcategorization properties first to the combination of modal and GO, and then to the verb cluster as a whole. The final verb cluster can therefore be combined with a directional PP even in the absence of a visible verb of motion. A tree as in (40) can be derived in Colloquial German but not in Standard German. In Standard German, tense auxiliaries select modal verbs to their right, subject to the complexity feature [VV-max]. The tree in (40) contains only a modal of the category [Mod>V GO ], but no cluster consisting of GO and modal verb. Such a tree is thus excluded in Standard German because the comple-

234

Markus Bader and Tanja Schmid

ment of the auxiliary is not of the required complexity. The perfect auxiliary of Colloquial German, in contrast, selects a modal to the right without any complexity requirements. It can thus combine with [Mod>V GO ], and the tree in (40) is accordingly allowed. This small difference implies the correlation between 2-and 3-verb clusters revealed by Experiment 1, namely that speakers who accept the order V-Aux-Mod also accept the order Aux-Mod (Colloquial German), and speakers who reject the order V-Aux-Mod also reject the order Aux-Mod (Standard German). What is still missing is an account of 2-verb clusters in which the modal verb behaves like a main verb. In Standard German, this is the only option for modal verbs in 2-verb clusters. Because all our experimental participants accepted Standard German Mod-Aux clusters, it must be assumed that this is a grammatical option in Colloquial German too, alongside of Aux-Mod clusters which are grammatical only in Colloquial German. The easiest way to solve this issue would be to postulate that modal verbs are in fact homophone, being either modal verbs proper or main verbs. In Standard German, only the main verb use would be allowed in 2-verb clusters; in Colloquial German, modal verbs could be used either way in 2-verb clusters, with the now well-known syntactic and morphological consequences. In our opinion, plain homophony would miss an important point, however. In the examples considered so far, the non-verbal complement of the modal verb was always a directional PP. Directional PPs can be used with all modal verbs. However, there are also more idiosyncratic usages. For example, wollen can take an NP complement, as illustrated in (41). (41)

. . . dass Peter ein Fahrrad (gewollt hat | hat wollen). that P. a bicycle wanted has has want . . . ’that Peter wanted to have a bicycle’.

Our experimental evidence only pertains to modal verbs used with directional PPs, but it seems to us that for speakers who acceptAux-Mod at all there is always free variation between Aux-Mod and Mod-Aux, without any noticeable meaning differences between the two variants.8 If correct, this intuition strongly argues that clusters of the form Aux-Mod and clusters of the form Mod-Aux contain one and the same modal verb at some level of representation. Otherwise, we should find examples in which an Aux-Mod cluster cannot be freely interchanged with a Mod-Aux cluster (and finding such examples would accordingly falsify the current analysis). Integrating the main-verb use of modal verbs into our analysis thus faces two conflicting demands. On the one hand, we see two different types of 2-

CAT meets GO: Auxiliary inversion in German verb clusters

235

verb clusters, one in which modal verbs exhibit the morphological and syntactic properties of main verbs (Mod-Aux: no auxiliary inversion, no IPP effect) and a second one in which modal verbs behave like modal verbs proper (Aux-Mod: auxiliary inversion, IPP effect). On the other hand, modal verbs in 2-verb clusters seem to have the same selectional restrictions irrespective of whether occurring in an Mod-Aux or in an Aux-Mod cluster. At the moment, we can only make a tentative proposal how a formal analysis that fulfills both demands might look like. What we propose is that a modal verb with incorporated GO allows an alternative use in which the modal part of the complex category becomes syntactically inactive, although it is still active for purposes of selection and interpretation. We will indicate this by crossing out the modal part. The resulting category of type [Mod>VGO ] acts as a lexical verb: it is selected to the left of the auxiliary and it appears as a past participle in the perfect tense, just as any other lexical verb. The resulting tree is shown in (42). (42)

The change of [Mod>VGO ] to [Mod>VGO ] can be seen as a kind of reanalysis which is caused by the fact that modal verbs used without verbal complement superficially look like main verbs, in particular in non-complex tense forms like the present tense. For the child acquiring language, the input will contain many examples in which there is no overt difference between a real main verb and a modal used as main verb, as shown in (43). The assumed reanalysis comes therefore at no surprise. (43)

a.

. . . dass Peter in die Stadt f¨ahrt / will. that P. in the town drives wants ‘. . . that Peter drives/wants (to drive) into town.’

b.

. . . dass Peter einen Apfel hat / will. that P. an apple has wants ‘. . . that Peter has/wants (to have) an apple.’

236

Markus Bader and Tanja Schmid

Note that in Standard German the distinction between modal and lexical use of the unit [Mod>VGO ] does not lead to any visible consequences, in contrast to Colloquial German where we get two different verb orders with different morphology. Even if the unit [Mod>VGO ] would oscillate between modal and lexical use also here, the complexity requirement in the subcategorization frame of Standard German auxiliaries would prevent the distinction between modal and lexical verb from surfacing. We nevertheless assume that [Mod>VGO ] can only be used as a main verb in Standard German, that is, in the form [Mod>VGO ]. By making this assumption, the differences between Standard and Colloquial German are kept to a mininum. Furthermore, this implies that modal verbs behave completely uniform with respect to auxiliary inversion and the IPP effect, allowing a maximally simple statement of the latter (see (46) below). The final set of lexical entries for auxiliaries, modals, and lexical verbs resulting from our analysis is shown in (44) and (45). (44) lists lexical entries that are shared by Standard and Colloquial German. VX stands for verbs of any category, that is, for VMain , VMod and VAux . (44)

Sample lexical entries common to Standard and Colloquial German a. Main verbs – VMain : DP ← | VMain : PP ← | VMain : DP PP ← | . . . b. Modal verbs i. Mod: VX ← ii. [Mod>VGO ]: XP[+dir]← c. Auxiliary verbs i. Perfect auxiliary: Vaux : VX ← ii. Future auxiliary: Vaux : VX ←

In addition to the shared entries in (44), we need to postulate lexical entries for auxiliaries selecting modal verbs. It is here where Standard and Colloquial German differ from each other. As shown in (45), Standard German imposes a complexity requirement on verb-cluster formation but Colloquial German does not. Note that (45) only shows the subcategorization part of each entry because the category label is always the same, namely Vaux . (45)

Subcategorization frames peculiar to auxiliaries when applied to modal verbs Perfect auxiliary Future auxiliary Standard German →VMod/VV−max →VMod/VV−max or V← Colloquial German →VMod →VMod or V←

CAT meets GO: Auxiliary inversion in German verb clusters

237

To complete our analysis, we need a morphological rule to account for the IPP effect, that is, the use of the bare infinitive instead of the past participle in the perfect tense. To achieve this effect, we postulate the rule in (46). (46)

Rule of morphological realization: Modal verbs appear as bare infinitives if selected by tense auxiliaries.

This rule is stated in the most general way by not making reference to any verbcluster specific properties. Cluster complexity cannot be a defining feature of the IPP effect because Experiment 2 revealed an IPP effect in two-verb clusters. That the order between auxiliary and modal verb is not crucial either is suggested by those Austrian and Bavarian dialects which strongly prefer the nested order V-Mod-Aux and nevertheless use the bare infinitive of the modal verb in the perfect tense as shown by the following example from Upper Austrian (Martina Wiltschko p.c.) (see also Wurmbrand 2004b). (47)

. . . wei a kumma miassn hot because he come must has ‘. . . because he had to come’

The final question to consider is how the Swiss German data from van Riemsdijk (2002) – which gave rise to the empty GO analysis (cf. (32)) – fit into the current account. The crucial Swiss German example is repeated in (48). (48)

¨ wele . . . das si n¨ome HAT [PP i d schuel] that she no-longer has wanted in the school ‘ . . . that she no longer wanted to go to school’

An obvious difference between Swiss German and the German data presented here is the direction in which modal verbs select their verbal complements: whereas selection by modals is uniformly to the left in Colloquial and Standard German, it is to the right in Swiss German. Thus, Swiss German modal and auxiliary verbs have subcategorization frames like VMod /VAux : → VX .9 Stating different subcategorization frames for Standard and Colloquial German on the one hand and the Swiss German variants reported in van Riemsdijk (2002) on the other hand is not sufficient, however, to account for the word order differences in sentences with modal verb and directional PP. In the Swiss German example (48), the directional PP appears at the right edge of the clause, a position which is excluded in both Standard and Colloquial German. A modal verb with incorporated GO does not allow the generation of such a sentence because it selects its complement in the same direction as GO itself, namely to the left.

238

Markus Bader and Tanja Schmid

In order to obtain (48), empty GO must be combined with its PP complement in syntax. In contrast to Standard/Colloquial German, the (micro) parameter responsible for the combination of modal verb and empty GO must therefore be set to ‘syntax’ in Swiss German. For the Swiss German sentence (48), this gives us the tree in (49). (49)

This tree is nothing else than a literal translation of van Riemsdijk’s analysis of Swiss German into the CAT framework of Williams (2003) as used here. (50) and (51) summarize the differences between Standard, Colloquial, and Swiss German. The differences concern either lexical subcategorization properties of auxiliary and modal verbs, or the setting of the micro parameter governing empty GO. (50)

Differences captured in subcategorization frames: Verbs select their complements to the [left/right] Standard German Modal verbs left Tense auxiliaries right / selecting modal verb [VV-max]

(51)

Colloquial German left right

Swiss German right right

Differences captured by parameter governing empty GO: Empty light verbs are active [in the lexicon/in the syntax] a. Standard/Colloquial German: lexicon b. Swiss German: syntax

These small differences together with the ability of the grammar to base-generate structures according to the CAT-language lead to the range of verb order variation discussed in this paper.

CAT meets GO: Auxiliary inversion in German verb clusters

239

6. Summary and conclusion We have linked two phenomena occurring with modal verbs in German: first, auxiliary inversion with complex tense forms of modal verbs, and second, the property of modals to appear with non-verbal complements. We have presented experimental evidence from native speakers of German showing the following connection between these two phenomena: speakers either accepted both the non-standard order V-Aux-Mod in 3-verb clusters and the non-standard order Aux-Mod in 2-verb clusters, or they rejected both. Standard German orders were accepted by all speakers. Based on these findings, we can identify two grammars – Standard and Colloquial German – with Standard German forming a subset of Colloquial German. To account for these two grammars, we extended the CAT-based analysis of verb cluster formation of Bader and Schmid (to appear) which is itself an extension of the analysis presented in Williams (2003), and combined it with the empty GO hypothesis of van Riemsdijk (2002). The most striking feature of auxiliary inversion in Colloquial German that comes out of our work is its generality: auxiliary inversion occurs in clusters of all sizes, and it can target any position in front of the modal verb. We have argued that this generality follows naturally from a CAT-based analysis as proposed in Bader and Schmid (to appear) but not from an analysis of verb-cluster formation phrased in cartographic terms.

Notes 1. For reasons of space, we have to gloss over all issues related to the gradience of judgment data; see Bader and H¨aussler (to appear) for detailed discussion. 2. We have also investigated verb clusters in which the modal verb was inverted. For the case of a 4-verb cluster, this is shown in (i). (i) . . . dass das Auto h¨atte m¨ussen gewaschen werden. that the car had must washed been ‘. . . that the car had to be washed.’ Although such sentences are often considered grammatical in the syntactic literature, they were judged as grammatical only 35% of the time in Experiment 4 of Bader and Schmid (to appear). 3. Since the future tense auxiliary can appear in all possible positions, is is also possible to state its subcategorization frame as “VMod ”, that is, by specifying the syntactic category of its complement without imposing any order requirements. The statement in (11) was chosen because it reveals the relationship to Standard German in a transparent way.

240

Markus Bader and Tanja Schmid

4. The assumption of mono-clausality is also shared by Wurmbrand (2001, 2007). Since verb order variation is not the focus of Wurmbrand’s work, we don’t discuss it here. 5. As pointed out by Jana H¨aussler (p.c.), an obvious candidate for a triggering feature is a feature related to focus. Indeed, Wurmbrand (2004a) has proposed that the verb in a 3-verb cluster V-Aux-Mod moves in front of the auxiliary if it is focused. However, Bader and Schmid (to appear) obtained judgment data to test this claim but did not find any evidence for focus effects on verb-cluster order. 6. We do not show reaction times (RTs) because they do not provide additional information. By and large, RTs mirror judgments: A condition with a high percentage value (either a high percentage of judgments ‘grammatical’ or a high percentage of judgments ‘ungrammatical’) is typically associated with fast RTs for the respective response type (judgment ‘grammatical’ or judgment ‘ungrammatical’), whereas a low percentage value is associated with long RTs. 7. Partial correlations were computed as follows. First, we computed linear regression equations with judgments on Aux-V-Mod sentences as predictor variable and either judgments for V-Aux-Mod sentences or Aux-Mod sentences as predicted variable. Second, we computed residual percentages by subtracting from the actually observed values the values which are predicted if judgments on Aux-V-Mod sentences would completely determine judgments on V-Aux-Mod sentences or Aux-Mod sentences. Finally, the residual percentages obtained in this way were correlated with each other. 8. This sets the two uses of modal verbs apart from verbs like versprechen (‘promise’) or drohen (‘to threat’) which have both a raising and a control verb use, but which show clear semantic differences depending on the particular use. 9. In the subcategorization frames given so far we did not distinguish between V0 and VP but used the unspecified label ‘V’ instead. This was done to allow for verb projection raising, i.e., combination with VPs. Verb projection raising is the unmarked option in Swiss German but it is also found in Standard German.

References Bader, Markus and Jana H¨aussler to appear Toward a model of grammaticality judgments. To appear in: Journal of Linguistics. Bader, Markus and Tanja Schmid to appear Verb clusters in Colloquial German. To appear in: Journal of Comparative Germanic Linguistics. Bader, Markus, Tanja Schmid and Jana H¨aussler 2009 Optionality in verb-cluster formation. In Susanne Winkler & Sam Featherston, (eds.) The Fruits of Empirical Linguistics. Volume 2: Product, pp. 37–58. Berlin: de Gruyter. Barbiers, Sjef 1995 The syntax of interpretation. Ph.D. diss., Leiden University.

CAT meets GO: Auxiliary inversion in German verb clusters Barbiers, Sjef 2006

241

The syntax of modal auxiliaries. In Martin Everaert and Henk van Riemsdijk, (eds.), The Blackwell companion to syntax, volume 5, pp. 1–22. Blackwell, Oxford. Cinque, Guglielmo 1999 Adverbs and functional heads. A cross-linguistic perspective. Oxford University Press, New York. Cinque, Guglielmo 2004 ‘Restructering’ and functional structure. In Adriana Belleti, (ed.), Structures and beyond: The cartography of syntactic structures, volume 3, pp. 132–191. Oxford University Press, New York. Cinque, Guglielmo 2005 Deriving greenberg’s universal 20 and its exceptions. Linguistic Inquiry, 36: 315–332. Evers, Arnold 1975 The transformational cycle in Dutch and German. Ph.D. diss., Rijksuniversiteit Utrecht. Fabricius-Hansen, Cathrine, Peter Gallmann, Peter Eisenberg, and Reinhard Fiehler 2005 Der Duden, Bd.4 : Die Grammatik. Dudenverlag, Mannheim. Geach, Peter Thomas 1970 A programm for syntax. Synthese, 22: 483–497. Haider, Hubert 1993 Deutsche Syntax - generativ. Vorstudien zu einer projektiven Theorie der Grammatik. Narr, T¨ubingen. Haider, Hubert 2003 V-clustering and clause union: Causes and effects. In Pieter A. M. Seuren and Gerard Kempen, (eds.), Verb constructions in German and Dutch, pp. 91–126. John Benjamins, Amsterdam. Hinrichs, Erhard and T. Nakazawa 1994 Linearizing AUXs in german verbal complexes. In John Nerbonne, Klaus Netter, and Carl Pollard, (eds.), German in Head-Driven Phrase Structure Grammar, pp. 11–37. CLSI Publications, Stanford. Hinterh¨olzl, Roland 2006 Scrambling, remnant movement, and restructuring in West Germanic. Oxford University Press, Oxford. Johnson, Mark 1986 A GPSG account of VP structure in German. Linguistics, 24: 871–882. Kayne, Richard 1994 The antisymmetry of syntax. MIT Press, Cambridge, MA. Kayne, Richard 2005 Some notes on comparative syntax, with special reference to English and French. In Gugliemo Cinque and Richard Kayne, (eds.), The Ox-

242

Markus Bader and Tanja Schmid ford handbook of comparative syntax, pp. 3–69. Oxford University Press.

Meurers, Detmar 2000 Lexical generalizations in the syntax of German non-finite constructions. Sonderforschungsbereichs 340, University of T¨ubingen. Rizzi, Luigi 1997 The fine structure of the left periphery. In Liliane Haegeman, (ed.), Elements of grammar, pp. 281–337. Kluwer, Dordrecht. Schmid, Tanja and Ralf Vogel 2004 Dialectal variation in German 3-verb clusters. Journal of Comparative Germanic Linguistics, 7: 235–274. Steedman, Mark 1983 On the generality of the nested-depedency constraint and the reason for an exception in Dutch. Linguistics, 21(1): 35–66. Svenonius, Peter 2007 1 . . . 3-2. In Gillian Catriona Ramchand, (ed.), The Oxford handbook of linguistic interfaces, pp. 239–288. Oxford University Press, Oxford. van Riemsdijk, Henk 2002 The unbearable lightness of GOing. The projection parameter as a pure parameter governing the distribution of elliptic motion verbs in Germanic. Journal of Comparative Germanic Linguistics, 5: 143–196. Vanden Wyngaerd, Guido 1994 Pro-Legomena. Distribution and reference of infinitival subjects. Mouton de Gruyter, Berlin. Williams, Edwin 2003 Representation Theory. MIT Press, Cambridge, MA. Wurmbrand, Susanne 2001 Infinitives. Restructuring and clause structure. Mouton de Gruyter, Berlin. Wurmbrand, Susanne 2004a Syntactic vs. post-syntactic movement. In Sophie Burelle and Stanca Somesfalean, (eds.), Proceedings of the 2003 Annual Meeting of the Canadian Linguistic Association (CLA), pp. 284–295. Wurmbrand, Susanne 2004b West Germanic verb clusters: The empirical domain. In Katalin E´ Kiss and Henk van Riemsdijk, (eds.), Verb Clusters: A study of Hungarian, German and Dutch, pp. 43–86. John Benjamins, Amsterdam. Wurmbrand, Susanne 2006 Verb clusters, verb raising, and restructuring. In Martin Everaert and Henk van Riemsdijk, (eds.), The Blackwell companion to syntax, volume 5, pp. 229–343. Blackwell, Oxford. Wurmbrand, Susanne 2007 How complex are complex predicates? Syntax, 10 (3): 243–288.

CAT meets GO: Auxiliary inversion in German verb clusters

243

Zwart, Jan-Wouter 1996 Verb clusters in Contintental West Germanic dialects. In James R. Black and Virginia Motapanyane, (eds.), Microparametric syntax and dialect variation, pp. 229–258. John Benjamins, Amsterdam.

A solution to the conceptual problem of cartography Denis Bouchard Abstract. Cartography is not simply an empirical approach that tries to establish the basic facts about the fine-grained structure of the sentence and its major phrases. Several of its descriptive tools raise problems. They end up listing properties of language in a format which uses notions that do not provide a basis for a principled explanation of the facts. The solution to this conceptual problem is to shift to a neo-saussurean approach with notions which are externally motivated, properties of the substances of language. This is illustrated for some basic facts about the distribution and interpretation of adjectives in nominal expressions in English and French.

1. What is Cartography? Cartography is a collective project which attempts to chart the syntactic structures of individual languages, and potentially of language as a whole.The attempt to describe the structures of syntax is not new of course, and is present in virtually all frameworks in linguistics, including very ancient ones. We find proposals about this from Panini to Port-Royal, from Saussure to American structuralists, from Greenberg to Chomsky. Cartography is more ambitious than most of these proposals in the breadth of languages and grammatical features it takes into consideration. This is understandable as more data is available about various languages and their features. Cartography also differs by putting its emphasis on functional properties rather than on substantive properties, i.e. on properties which are usual traits of all phrases of a certain category rather than on properties due to a particular choice of lexical item. The main concern is to determine how functional properties are encoded in the syntactic structure, i.e., to determine what exactly constitutes the set of functional projections and how they relate to each other syntactically. If all that Cartography did was describe traits of languages, if Cartography was simply an empirical approach that tried to establish the basic facts about the fine-grained structure of the sentence and its major phrases, there would probably be no talk about the limits of Cartography, nor alternative accounts: these facts have to be preliminarily established before any theorizing can begin. However, science does not operate in such a dichotomic way. There is no deep distinction between an empirical approach and a theoretical approach: observational

246

Denis Bouchard

propositions are part of a theory, they are not external to it and independent. The only difference with the “regular” theoretical propositions is that an observational proposition is considered self-evident, generally valid, axiomatic, across a community of scientists. So it is made unfalsifiable by fiat: it is assumed that “there exists at the time a ‘relevant technique’ such that ‘anyone who has learned it’ will be able to decide that the statement is acceptable” (Lakatos 1970: 106). The most successful theories of the sciences to which the propositions could be subjected are given an “observational” status by convention and are used as extensions of our senses, as unproblematic background knowledge. However, like any other element of a theory, an observational proposition may be questioned and modified or dropped in the face of an overwhelming problem.1 Cartography is no more strictly empirical than any other approach. Morever, several of its descritive tools are far from self-evident, and subject to serious questioning. The typical modus operandi of a cartographist is as follows. Step 1: Determine that a semantic/pragmatic functional property is present in a certain language. Step 2: Determine how this property is encoded in the syntactic structure: this is assumed to be done by means of functional features and functional categories which appear in functional projections. Step 3: Compare languages to figure out the universal set of functional projections. Step 4: Determine how the functional projections relate to each other in the syntactic structure. This is assumed to take place in a strictly rightbranching structure of the type determined by Kayne’s (1994) Linear Correspondence Axiom (LCA), in which a functional head occupies its “natural position of interpretation” and merges with the functional and/or lexical projections that it selects. Since languages often diverge with respect to the universal order resulting from step 4, the hypothesis of a universal hierarchy is saved by proposing a covert level of representation at which it holds. To account for variation, different tools are introduced: Step 5: Devise a movement rule (including remnant movement) which relates the postulated universal hierarchy to a different structure which can be directly related to the surface order; each instance of Move is triggered by an uninterpretable feature (uF) which is particular to the target of Move. Some functional properties are overtly expressed in some languages but not in others. There are two strategies in this respect:

A solution to the conceptual problem of cartography

247

Step 6a: Assuming that individual languages choose only a subpart of the full universal templates of functional categories, determine which functional categories the language has in its repertoire. Step 6b: Assuming that all languages have the full template but may have +/− overt morphological elements corresponding to each functional element, postulate a zero morpheme for a functional property not overtly expressed. Each of these steps may be repeated several times as more functional properties are considered in additional languages and constructions. What is particular to Cartography, and where it takes its name from, is its assumption that a large amount of semantic/pragmatic functional traits are encoded in the syntactic structure by means of functional features and categories. Cartography is not just aiming at a description of the basic facts, independently of any theoretical approach. This is clear from the descriptive tools it uses and the claims that are made: all are highly “involved” and raise deep conceptual and empirical questions. This becomes immediately clear if we look in detail at a typical cartographic analysis. I illustrate this with one of the first and most influential examples of the cartographic approach – Cinque’s (1994) account of the distribution and scope of adjectives in nominal expressions in English and Romance languages.

2. Cartography and adjectives I first introduce some of the most basic facts about the distribution and interpretation of adjectives in nominal expressions in English and Romance (taking French as an example), then look at how Cartography presents those facts, and review the problems with that view.

2.1.

Basic facts about adjectives in nominal expressions

Fact 1: Order with respect to N. French and English differ in the distribution of bare (complementless) Adjs: they are almost exclusively prenominal in English (1),2 but they may appear before or after the noun in French (2). (1)

a. b. b.’ c.

a red ball an alleged new miracle a new alleged miracle a big red ball

(A1 A2 N) (A1 A2 N) (A1 A2 N)

248

(2)

Denis Bouchard

d. e. e’. f. g. h. h.’ h.” i. i.’

a persistent bad reputation (A1 A2 N) an interesting new proposition (A1 A2 N) a new interesting proposition (A1 A2 N) The bright students present/responsible received... (A2 N A1 ) a blue Chinese vase (A1 A2 N) the alleged dishonest Chinese professors (A1 A2 A3 N) the dishonest alleged Chinese professors (A1 A2 A3 N) the dishonest Chinese alleged professors (A1 A2 A3 N) an old church a former church

a. b. b.’ c. d. e. f.

une balle rouge un suppos´e nouveau miracle (A1 A2 N) un nouveau suppos´e miracle (A1 A2 N) une grosse balle rouge (A1 N A2 ) une mauvaise r´eputation tenace (A2 N A1 ) une nouvelle proposition int´eressante (A1 N A2 ) OR (A2 N A1 ) Les brillants e´ tudiants pr´esents/responsables ont re¸cu . . . (A2 N A1 ) un vase chinois bleu (N A2 A1 ) les pr´esum´es professeurs chinois malhonnˆetes

g. h.

Fact 2: scope among prenominal adjectives. When there are two (or more) Adjs in a nominal constituent, one is typically interpreted as having “scope” over the other. Thus in (1b), alleged has scope over new: alleged picks out individuals out of the set of new miracles, whereas in (1b’), we get the reverse scope, in which new picks out individuals out of the set of alleged miracles. This relative scope is indicated by the subscripted numbers in parentheses, where 1 indicates widest scope of the Adj, 2 indicates scope under a 1-Adj, and 3 indicates yet lower scope. As the English examples show, the generalization is that the scope of prenominal Adjs is directly reflected in their order: an Adj has scope over any Adj to its right. This also holds for French prenominal Adjs, as shown in (2b-b’).3 Fact 3: scope between prenominal and postnominal adjectives. French easily allows Adjs on both sides of the noun. Sometimes the prenominal Adj has scope over the postnominal Adj as in (2c): the corresponding English example (1c) has the two Adjs in prenominal position in the same relative order as in French. In other cases, the postnominal Adj has scope over the prenominal Adj as in (2d): the corresponding English example (1d) has the two Adjs in prenominal position in a relative order opposite to French. The scope between the Adjs can also be

A solution to the conceptual problem of cartography

249

ambiguous as in (2e), in which case there are two different orders in English (1e-e’) to express these meanings. In both languages, the postnominal Adj with a stage-level interpretation has scope over the prenominal Adj, as shown in (1f) and (2f). In fact, the relative scope of a prenominal and a postnominal Adj is quite free: pragmatic factors make a reading more salient, but this can be fairly easily reversed. For instance, if I say Maintenant que tu m’as montr´e une grosse balle verte, montre-moi une grosse balle rouge (‘Now that you have shown me a big green ball, show me a big red ball.’) then rouge ‘red’ has wide scope. But if I say Maintenant que tu m’as montr´e une petite balle rouge, montre-moi une grosse balle rouge (‘Now that you have shown me a small red ball, show me a big red ball.’), grosse ‘big’ has wide scope. Fact 4: scope among postnominal adjectives. French easily allows more than one postnominal Adj. Their relative scope is also directly reflected in their order: a postnominal Adj has scope over any other postnominal Adj to its left, as show in (2g). This is the mirror order of the prenominal Adjs in the English corresponding example (1g). The scope among postnominal French Adjs is immutable. However, they may interact with a prenominal Adj as in example (2h). As expected from the discussion above, the prenominal Adj may have scope over one or both postnominal Adjs, hence (A1 N A3 A2 ) or (A2 N A3 A1 ) corresponding to the English examples (1h) and (1h’) respectively, or both postnominal Adjs may scope over the prenominal Adj (A3 N A2 A1 ), corresponding to (1h”). But (2h) cannot receive any interpretation in which the rightmost postnominal Adj malhonnˆetes falls under the scope of chinois; so the following scopes are excluded: (A1 N A2 A3 ), (A2 N A1 A3 ) and (A3 N A1 A2 ). Fact 5: Serialization of Adjs. It has long been observed that there is a preferred order between Adjs of different semantic classes. Thus, the unmarked case is for size Adjs to precede color Adjs: compare big red ball with ?red big ball. A more developed ordered list is as follows: (3)

Serialization of adjectives in object-denoting nominals (Cinque 1994): poss> cardinal> ordinal> quality> size> shape> color> nation

However, this preferred order seems to correspond to unmarked pragmatic situations and can vary, as seen in (1b, b’, e, e’) and (2b, b’).4 Fact 6: meaning difference between Adj-N and N-Adj. In French, the same Adj and the same noun combined in different orders results in different interpretations:

250

(4)

Denis Bouchard

a.

b.

(5)

a.

b.

c.

une e´ glise ancienne a church old ‘an old church’ une ancienne e´ glise a old church ‘a former church’ une femme seule a woman alone ‘a woman who is alone’ une seule femme a alone woman ‘a sole woman, only one woman’ La seule femme seule s’est isol´ee dans un coin. the alone woman alone se.is isolated in a corner ‘The only woman who was alone isolated herself in a corner.’

The meaning difference is quite clear in (4–5).As argued extensively in Bouchard (2002), chapter 2, there is always a meaning difference between the two orders AN and NA in French: though it is subtle at times, contexts can be provided which make the difference more salient. Moreover, the difference always has the same general property: the prenominal Adj is used intensionally, while the postnominal one is extensional. Thus seule can appear in both positions without being tautological, as we see in (5c).

2.2. A cartographic account of the facts Let us look at how Cartography can account for these facts, and how much it furthers our understanding of them. 2.2.1. Fact 1: bare Adjs are almost exclusively prenominal in English but prenominal or postnominal in French. Cinque (1994) presents a typical cartographic account of this contrast (the same approach is used for adverbs in Cinque 1999). Adj phrases are generated in the Spec of functional categories which correspond to meaning classes of Adjs (henceforth AdjFC). These functional categories are generated in an exclusively right-branching structure following Kayne’s (1994) LCA. For instance, (1c) and (2c) are given the same underlying structure as in (6) (omitting the determiner).

A solution to the conceptual problem of cartography

251

(6)

It is assumed that in French there are uninterpretable features on the N and on the functional heads Color and Size which trigger the movement of the N balle to a position which happens to be between the two Adjs. Under this view, the reason why bare Adjs occur on either side of the noun in French is that French has uFs which happen to make the N move to the left of some Adjs. On the other hand, bare Adjs are generally prenominal in English because English does not have these uFs (or alternatively, French has strong features with a surface effect, whereas English has weak features with only an LF effect and no visible movement). So Cartography answers by giving (i) a list of ordered positions, and (ii) a list of features that trigger N-movement (different in French and English). This just restates the facts. It provides no indication why languages differ in this way, nor in what ways they could possibly differ. Is it possible for a language to move N but to a position different from the one in French? Are the options for these features in French and English related to any other properties of the languages, or are they isolated facts that must be bluntly listed? How can a child learn which of the numerous uFs are present in its language without making mistakes? Other problems also arise. Since the Adjs are phrases in the specifier of functional categories, it is not clear why a prenominal AP containing a complement is ungrammatical, as in (7): (7)

a. *a proud of her daughter mother a.’ a mother proud of her daughter b. *une fi`ere de sa fille m`ere b.’ une m`ere fi`ere de sa fille

In the Cartography system, an AP containing a complement is to the right of the N because N has a uF when such an AP is its specifier, and there is an additional F-projection above that AP to which the N must move. This No Complement restriction is frequently dealt with by similar kinds of filtering devices. This type

252

Denis Bouchard

of account is unexplanatory: “the filtering mechanism and auxiliary assumptions surrounding it are ad hoc since they are proposed for the sole purpose of protecting the theory from a threatening falsification and provide no additional testable results” (Bouchard 2002: 140). Obligatory ad hoc triggers for N movement and F-projections must also be assumed for the postnominal stage-level Adjs in (1f) and (2f). 2.2.2. Fact 2: the scope among prenominal adjectives. This fact has long been recognized, as well as its mirror equivalent for potsnominal Adjs (Blinkenberg 1933, Vendler 1968, Dixon 1982, Sproat & Shih 1990, and many others): the leftmost Adj gets wide scope, in the sense that it modifies the block formed by the substantive and the second Adj. This interpretive embedding can be directly represented by the following bracketing: (8)

a. b.

a [ nice [ small dog ]] un[ beau [ petit chien ]]

However, the simple adjunction structure corresponding to this bracketing is not compatible with the N movement assumed for postnominal Adjs since there is no FP providing attracting features and a landing site for the N. So more complex structures as in (6) are proposed and a list of several functional categories corresponding to various semantic classes of Adjs. This introduces a considerable amount of redundancy: every Adj is attributed a feature indicating to what semantic class it belongs (Color, Size, Shape, Nation, etc.), and each of these semantic features is present again in the head of functional projections. 2.2.3. Fact 3: scope between prenominal and postnominal Adjs. The scope of the Adjs in French example (2c) is correctly represented in structure (6) under the usual assumption that an element in a c-commanding position has wider scope than the elements it c-commands. On the other hand, if the structure in (6) is directly used to account for other examples with prenominal and postnominal Adjs, it makes wrong predictions about scope. Thus in (2d), it is the postnominal Adj which has scope over the prenominal Adj. A first strategy used by Cinque to deal with such cases is to assume that the postnominal Adj is in a higher “predicative” position. But this is not compatible with the strictly right-branching LCA structure. His solution is to propose another movement, somewhat like the remnant movement introduced by Richard Kayne. We can obtain the correct scope for (2d) by assuming two things: first, the derivation is as in (9), in which the whole QualityP is moved above and to the left of the MannerP, deriving the surface order mauvaise r´eputation tenace; second, the scope of the Adjs is determined on the structure which is the input to Move rather than its output.

A solution to the conceptual problem of cartography

(9)

253

➤

MannerP

tenace

Manner

QualityP

mauvaise Quality NP N réputation

To be complete, the analysis would have to indicate what feature triggers this movement, why it attracts the QualityP to that position, why the N does not raise to the head of MannerP before the remnant movement (incorrectly deriving the order *mauvaise tenace r´eputation), why a similar remnant movement does not take place in examples like (2b-b’), giving the Adjs the opposite scope, and why QualityP raises in French but not English (i.e., what determines that a language has this feature or not). We also have to be told what is the functional category that provides the landing site here, as well as for all examples in which there are only postnominal Adjs: there is no overt indication that there is an AdjFC in that position, nor can the landing site be SpecDP since this predicts the wrong order with the Det. So there must be some mysterious WP whose sole motivation is to save the analysis. To account for the ambiguous scope between a prenominal and a postnominal Adj as in (2e), it must be assumed that the elements can get uFs which trigger movement either as in (6) or as in (9), whereas the elements in (2c) and (2d) can only get one of the two sets of uFs. Even more troublesome for this highly syntactic presentation of the data is the fact that the relative scope of the prenominal and postnominal Adjs depends on pragmatic factors, i.e., that the sets of uFs depend on these factors. Finally, since a postnominal Adj with a stage-level interpretation has scope over the prenominal Adj in both English and French, as shown in (1f) and (2f) this means that raising as in (9) is possible in English too, but only when the Adj has this particular interpretation.5 2.2.4. Fact 4: scope among postnominal Adjs. It gets more complicated in cases like (2g), where the two French Adjs are postnominal and in the mirror order of the English Adjs. Following (6), we could expect that the difference between English and French is simply that N raises to the left of both Adjs. But this derives the wrong order *un vase cher chinois for the intended interpretation

254

Denis Bouchard

in which cher has scope over chinois. Under cartographic assumptions, it is an accident that French postnominal Adjs are never in the same order as English prenominal Adjs, i.e., it is never the case that N simply raises all the way. To account for examples like (2g), Cinque uses Kayne’s remnant movement. The derivation starts off with the underlying order in (10a). The N vase is attracted by some uF to the left of chinois, deriving (10b). Then vase chinois is attracted by yet another uF and moves to the left of cher, deriving the surface order in (10c). (10)

a. b. c.

basic order: un cher chinois vase → N-Move intermediate: un cher vase chinois → remnant Move surface order: un vase chinois cher

However, we are given no clue why remnant movement of this particular element must take place in French, but not in English, nor why it is not possible to just raise N an additional time. Note that the impossibility to have an additional raising of N cannot be attributed to a lack of a functional category as a landing site since there is one in (2h) for the Adj pr´esum´es, and some FC must be postulated as a landing site for the remnant movement in (2g). So it must be stipulated in some way that N cannot move there. The stipulative nature of the account is even more obvious for an example like (2h), with one prenominal Adj and two postnominal Adjs. As we saw, of the six possible sets of scopal relations in (11), only the first three are grammatical. Yet all six can easily be derived by using the three cartographic tools: the bracketings can correspond to strictly right-branching LCA structures, and each of these structures can be related to the surface order pr´esum´es professeurs chinois malhonnˆetes by N movement and remnant movement. (11)

[pr´esum´es [malhonnˆetes [chinois professeurs]]] → N-Move pr´esum´es malhonnˆetes professeurs chinois → remnant Move pr´esum´es professeurs chinois malhonnˆetes b. [malhonnˆetes [pr´esum´es [chinois professeurs]]] → N-Move malhonnˆetes pr´esum´es professeurs chinois → remnant Move pr´esum´es professeurs chinois malhonnˆetes c. [malhonnˆetes [chinois [pr´esum´es professeurs]]] → remnant Move malhonnˆetes pr´esum´es professeurs chinois → remnant Move pr´esum´es professeurs chinois malhonnˆetes d. # [pr´esum´es [chinois [malhonnˆetes professeurs]]] → N-Move pr´esum´es chinois professeurs malhonnˆetes → N-Move pr´esum´es professeurs chinois malhonnˆetes

a.

A solution to the conceptual problem of cartography

255

e. # [chinois [pr´esum´es [malhonnˆetes professeurs]]] → N-Move chinois pr´esum´es professeurs malhonnˆetes → remnant Move chinois professeurs malhonnˆetes pr´esum´es → N-Move professeurs chinois malhonnˆetes pr´esum´es → remnant Move pr´esum´es professeurs chinois malhonnˆetes f. # [chinois [malhonnˆetes [pr´esum´es professeurs]]] →remnant Move pr´esum´es professeurs chinois malhonnˆetes The descriptive tools of Cartography can derive the surface order and scope of Adjs in English and French, but they can also derive any order or scope. The uFs are just a means to stipulate when operations apply to correct the predictions made by right-branching LCA structures: but we are given no indication concerning how much languages can differ with respect to these operations, nor why and when. 2.2.5. Fact 5: serialization of Adjs. Cartography accounts for the preferred order of Adjs by assuming a set of functional categories which correspond to meaning classes of Adjs: these AdjFCs are generated as complements of one another in an strictly right-branching LCA structure, and Adj phrases are generated in the Spec of the functional categories of their semantic class. The fixed order of Adjs before the noun in English follows from the hierarchical organization of the AdjFCs. This hierarchy is determined by selectional properties of the functional categories. Thus a Size Adj precedes a Color Adj because Size selects ColorP as its complement. However, we are not told why Size selects a ColorP, and what it means. Without answers to these questions, the proposal is equivalent to a descriptive template.6 Moreover, since the preferred order can vary according to pragmatic situations, it has to be assumed that the selectional restrictions of AdjFCs have the rather unusual property of fluctuating according to pragmatic factors. 2.2.6. Fact 6: meaning difference between Adj-N and N-Adj. To account for meaning differences like e´ glise ancienne and ancienne e´ glise, Cartography must assume that the Adj belongs to two different semantic classes and so can appear in the specifier of two different AdjFCs, one of which is above the raised N and the other below it. The N in French just happens to appear exactly in this divide between the two sets of AdjFCs. All Adjs alternate in this way, so they all must be assigned at least two different semantic classes. Given the generality of the phenomenon, “a doublet theory considerably complicates the lexicon” (Siegel 1980: 56). More generally, Bouchard (2002: 21) points out that if we frequently admit more than one entry for a lexical item, “any counterexample

256

Denis Bouchard

to the predicted behavior of the item can be attributed to yet another entry. Explanation is reduced to an uninformative list.”7

2.3. Assessment of the cartographic approach Now that we have seen how some of the descriptive tools particular to Cartography operate for the distribution and interpretation of Adjs, we can make a first assessment of the quality of their motivation, their formal precision and their descriptive accuracy. There are four main tools: (i) AdjFCs corresponding to semantic classes of Adjs, (ii) right-branching LCA structures, (iii) N-Move and (iv) remnant Move. Consider first AdjFCs. These are poorly motivated since they redundantly introduce semantic properties already present in the lexical entries of the Adjs. They also create a problem of framing which has been recurrent in generative grammar: it raises the general question of how much of the semantic/pragmatic material projects in syntactic structure. Cinque (1994) proposes functional categories such as Quality, Size, Shape, Color, Nation, Speaker-oriented. Where do we stop? For instance, Beghelli and Stowell (1997: 74–75) have Distributive Phrase, RefPhrase, Share Phrase; Munaro and Obenauer (2000) use an Evaluative CP (i.e., an Opinion phrase). This is highly reminiscent of the use of performative verbs in Generative Semantics, and it is subject to the same criticism (see Bouchard 2002: 333–336 for a brief review). The selectional restrictions among AdjFCs, and with DP and NP, are crucial to the analysis since they determine the positions of the AdjFCs in the structure. Yet the selectional restrictions among the various functional elements are left for “further considerations” in Cinque (1994): we are not told why a functional head selects a particular FP, nor how to interpret this selection, nor what it means for a purely functional category to actually have selectional properties. This lack of formal precision about the selectional properties of functional elements is quite frequent in cartographic works. As a result, the position of functional elements is actually determined by an ordered list of positions, which is equivalent to a descriptive template. Moreover, since selectional relations indirectly determine the position and relative scope of Adjs, this is descriptively troublesome since we don’t expect selection to vary according to pragmatic/contextual factors. Also, the description of meaning differences like ancienne e´ glise/´eglise ancienne is attained by means of doublets of Adjs: by dealing with it in the lexicon, it fails to capture what is a syntactic generalization, holding across all lexical entries of Adjs. Next, consider the value of the LCA as a descriptive tool. As I point out in Bouchard (2002: 379f), the LCA approach appears to be a step in the direction

A solution to the conceptual problem of cartography

257

of linking the analytical apparatus directly to interface elements, to motivate it on external grounds, since Kayne makes an explicit appeal to physiological properties of our articulatory apparatus which force strings of sounds forming words to be produced sequentially.8 Consequently, we can improve on phrase structure rules and head parameters because properties of our sensorimotor apparatus can explain part of the phenomena. However, the effects of the LCA actually depend on two stipulations and do not derive from the external system as claimed. First, it must be stipulated that every phrase marker contains an abstract node A which asymmetrically c-commands every other node. The head of A is an abstract terminal that precedes all the other terminals. This abstract element has no physiological reality in the phonetic output, and it is ad hoc: the only purpose of its existence and the stipulation that it precedes all other terminals is to make a correlation between precedence and asymmetric c-command. Kayne crucially bases the intuition of the LCA on the physiological need to linearize some phonetic elements. If the element used to incorporate this intuition in the theory is a phonetically empty element, there is a contradiction: you can’t base your theory of linearization on an element that has no property of linearization. The second stipulation concerns the notion of asymmetric c-command: there is no principled reason to choose it over another locally linear relation on nonterminals, namely the relation of dominance; in fact, Bouchard (2002) argues that there are reasons to consider dominance as a better option. On the whole therefore, Cartography rests on a rather fragile cornerstone. Moreover, we saw that the LCA right-branching structures give the wrong scope among postnominal Adjs and among some prenominal and postnominal Adjs. This descriptive inadequacy requires corrections by means of additional tools – movements. With the addition of N-Move and remnant Move, it is possible to derive the correct surface order and scope of Adjs: these tools are strong enough to save the LCA structure whenever it does not directly predict the correct scope of Adjs. However, they are such powerful tools that it is just as easy to derive all the wrong orders and wrong scopal relations, i.e., to relate any surface order to any underlying structure. So these tools fail to make correct empirical generalizations. For instance, they can’t explain why Adjs have free relative scope in Adj-N-Adj sequences, but rigid scope in Adj-Adj-N and N-Adj-Adj sequences. The system is so removed from empirical data that if we invent anti-English, a language in which the order of Adjs is the same as in English but their scope is systematically the opposite (or vice versa), the descriptive tools can easily derive these results. We are left with no explanation why similar derivations cannot take place in true English. Though the system is formally quite complex with intricate structures and operations, it is not precise and informative: we are not given an indication of what determines the presence of a uF or a WP, hence

258

Denis Bouchard

what determines that N-move or remnant Move applies, so anything is possible: the system does not predict what scope there is between two Adjs in a certain order, and if we know the order and scope of two Adjs, we cannot ascertain how things got that way as the system allows several ways to derive a result.9 The only way to prevent the wrong derivations and admit the correct ones is to list every single detail by means of uFs in a nonsubstantive lexicon. The system reaches extreme complexity in which each property appears to be unrelated to any other. That is why it cannot make adequate empirical generalizations, but only provides ad hoc stipulations listing which positions elements can occupy and with what scope. Lists describe, they do not explain. A good theory is based on a model that contains only a few arbitrary elements and it accurately describes many observations. In Cartography, the lists contain a very large amount of arbitrary elements and make no predictions. Lists are merely assertions of existence: they are inventories of facts, they tell us what is. Lists do not deal with modalities of existence: they do not tell us what is possible. In the current absence of indications about what are possible functional elements and the selectional restrictions among them, the lists of these elements do not explain anything. Similarly, without some indication of the conditions which license the presence of uFs that trigger movement, a list like the nonsubstantive part of the lexicon reveals nothing about what are possible language-internal variations, nor what limits there are to crosslinguistic variation.10 It could be objected that every science needs a list of basic elements. For instance, any linguistic analysis of any particular language must provide a list of its morphemes. My criticism is directed at ad hoc lists, not at the Saussurean (“substantive”) lexicon which lists the signs, i.e., the arbitrary mediations between perceptual subtance and conceptual subtance. The nonsubstantive lexicon is an ad hoc list because each uF is custom-designed for one particular order or scope of a given element and is not generalizable. The only motivation for each uF is to save the theory based on right-branching LCA structures, N-Move and remnant Move, to make sure that it does not produce all that it could, but only the observed facts, one at a time. The Saussurean lexicon is not a problematic list because it is not ad hoc: it is motivated by the lack of natural link between the two types of substances (see the discussion in Bouchard 2002: 360–361).11 A principled explanation rests on properties of logically prior systems, which are language or even organism-independent. The prior properties come from general physio-biological properties which canalize the way a trait like language or some physical property of any organism can develop. In Language, elements of two orders combine–from the perceptual substance and from the conceptual substance (Saussure 1916). These two substances have their own systems, and it is the linking of elements from these two systems with very different sub-

A solution to the conceptual problem of cartography

259

stances which forces arbitrariness, and hence a listing of the signs since they cannot be predicted: these two substances don’t have a “reason of nature” to be linked in any particular way. Saussure explains the arbitrariness of signs in this principled way when he relies on the properties of the two prior substances: the choice of a particular signifiant or a particular signifi´e is not due to natural causes, but rather is arbitrary because our phonatory articulators produce sounds which cannot have a meaningful, logical or iconic relation with meanings. The Saussurean lexicon is therefore motivated by the canalization effects of the two substances involved. We can establish the basic notions of linguistics by following a method used in other sciences: some propositions are considered as generally valid, axiomatic, because they are logically prior to linguistic theory, since its object of study presupposes them (cf. Lakatos 1970 quoted above). Some people question the arbitrariness of language and claim that some words show iconic relations with their denotations. I know of no convincing example. If there was some ‘reason of nature’ to link a particular signifiant with a signifi´e, we would find languages with the same “iconic” words. But this does not occur. A reviewer claims that “at the level of structure, the link between the significant and the signifi´e is far from arbitrary. Consider scope: the structural position of an element and its scope properties directly and systematically go together.” This is simply false. For instance, the scope of a quantifier phrase often does not correspond to its surface structural position, and of course, only a surface position can count as a signifiant, not a covert position with no perceptual content, since a signifiant is necessarily an overt element by definition. On the other hand, some aspects of information structure do exhibit some iconicity: thus, positions close to the periphery of the sentence and strong prosodic features, often express information-structural prominence. This does not contradict Saussure’s proposal. He provides an explanation for the fact that signs are overwhelmingly arbitrary, but this does not rule out the possibility for signs to be motivated in the (rare) cases in which our phonatory articulators produce sounds which can have a relation with meanings. Loudness expresses attentional prominence in general, not just in language.12 As for elements in peripheral positions, they can be prominent, but most of the time they are not, they just happen to be peripheral because a sentence must have a beginning and an end. In contrast, the nonsubstantive part of the lexicon is an ad hoc list of uninterpretable features/categories: their sole motivation is theory-internal. This is the problematic list. But someone could try to explain away this ad hoc list by saying that they are no more theory-internal than those in the list of particles in physics, or the list of elements in chemistry. But the lists of elements in chemistry only made some sense when people began to question its status as a list and gradually understood why it is organized as it is. Similarly, physicists do not

260

Denis Bouchard

simply sit on their hands in the face of haphazard particles. As Greene (1999) remarks, innumerable “whys” leap to the fore: Why that number of particles? Why three families? Why such a seemingly random spread of masses? Did all of this occur by chance, by some divine choice, or is there a comprehensible scientific explanation for these fundamental features of our universe? Cartographists typically do not raise questions like “Why that number of functional categories, and why in that hierarchical ordering?” “Why those uninterpretable features?” and so on. Their model cannot offer an answer to this kind of question since it takes the list of elements and their properties as input. Had the data revealed a different feature or category content, possibly interacting with different operations, these changes could have been fairly easily incorporated in the cartographic model, as we saw in the discussion of anti-English. The cartographic model is too flexible to explain the properties of the linguistic elements, as it can accommodate any range of possibilities. Actually, given the technical apparatus of cartography, a better comparison is with the tools used in the Ptolemaic model to establish the cartography of celestial bodies. With a few primitive elements like points, circles (the perfect, natural shape) epicycles, deferents and rates of rotation, the Ptolemean model accurately describes the positions and displacements of the known celestial bodies. Similarly, with primitives like words/morphemes, a basic “natural” position of interpretation, functional categories, uninterpretable features, and Move plus remnant Move, the cartographic model describes the surface positions of words in the known languages. The problem with the Ptolemean model is not one of accuracy: the astronomical charts it produces could still be used for everyday navigation. But why are the elements there at those moments? It is not because of the circles, epicycles, deferents and rates of rotation. Newton showed that it is due to physical, substantive matters. This is the essence of his revolution. True, there could not have been a Newtonian revolution without the data about the positions of the celestial bodies at various times. But Newton could not have discovered the forces responsible for the positions of celestial bodies based on circles, epicycles, deferents, etc. These descriptive tools of the Ptolemean system are not in the Newtonian system: they play no role in it and the only effect they had was to make it more difficult to accept the new system by those with the habits of thinking which the old system fostered. Though it is possible to tack various figures and movements on the sky, and various structures and movements on the sentences of languages, they do not offer explanations. Just like no causes are likely to be found for the Ptolemean notions, so no causes are likely to be found for the cartographic notions.

A solution to the conceptual problem of cartography

261

Summarizing, the tools of Cartography exhibit weaknesses in the quality of their motivation, their formal precision and their descriptive accuracy. This leads to situations where it is impossible to decide what is going on. For instance, since UG operations can reverse the scope of Adjs generated in the LCA structure, as they do with postnominal French Adjs for example, and since this theory allows that operations take place covertly, it does not seem possible for a child to determine that such a scope reversal cannot take place among English prenominal Adjs. In other words, a child cannot learn the uFs of English and the uFs of French. Yet children learn languages. So there must be an alternative to Cartography.

3. A substantial approach to language 3.1.

Principled explanation from external properties

The most basic initial condition in linguistics is that language is a system which links concepts and percepts. So language is determined by the conceptual system found in human brains, and the sensorimotor systems of human bodies. Properties of the conceptual and perceptual substances provide a strong basis of explanation because it is possible to relate the explanandum to an explanans which is independently motivated. We get principled explanations in linguistics if we appeal to the substantive properties of language. In Generative Grammar, analyses appeal to properties of the interface substances typically as very late filters on the computational system. GG assumes that language is foremost a computational system, and bases its analyses mainly on general principles of efficient computation. Because it does not sufficiently take into account the substantive properties which could offer an explanation of the facts, GG ends up listing the facts with various formal tools. This is not an isolated property of the Cartography approach: GG has been laden with lists before – each language corresponding to a taxonomy of phrase structure rules, of transformational rules, of triggers, of parameters – and now Cartography lists properties down to token constructions in the nonsubstantive part of the lexicon. The various GG models have missed some explanations, because they don’t sufficiently rely on the substantives properties where the explanations are to be found. This is a problem which has been recurrent from the earliest days: “The problem is that our approach to features, to rules, and to evaluation has been overly formal. Suppose, for example, that we were systematically to interchange features or to replace [aF] by [-aF] (where a is +, and F is a feature) throughout our description of English structure. There is nothing in our account of linguistic

262

Denis Bouchard

theory to indicate that the result would be the description of a system that violates certain principles governing human languages. To the extent that this is true, we have failed to formulate the principles of linguistic theory, of universal grammar, in a satisfactory manner. In particular, we have not made use of the fact that the features have intrinsic content” Chomsky and Halle (1968: 400).

Computational principles may not be a very good basis for explation in linguistics. Frampton and Gutmann (1999: 2) note that “[i]t is far from clear that computational efficiency is either required or even expected for theories which purport to explain certain aspects of cognitive systems [. . . ] It would be a naive view of evolution to assume, a priori, that the brain is organized in such a way that computational efficiency for language processing is ensured.” In the same vein, Kosslyn (1980: 123) remarks that “some of the conditions on cognitive processing may not be those that foster the most computationally efficient processing for the tasks at hand.” For example, analyses are frequently based on parsimony, a general principle of computational efficiency such that there be no redundancy, no superfluous operations or elements in a derivation. However, this principle is regularly violated in language, most likely because redundancy facilitates cognitive processing. For instance, in the French nominal expression la petite chatte (thefem smallfem catfem ), the feminine gender is expressed three times.13 So the cognitive efficiency required for the system of language may be quite different from computational efficiency. Even the principle of computational efficiency most used by scholars in syntax–locality of dependencies–does not hold for a large class of dependencies, such as those involving pronominals, as Ross (1986) originally observed: (12) (13)

*Which constraints do you admire the linguist who discovered [which constraints]? a. b. c. d.

These constraints, I admire the linguist who discovered them. Speaker A: You seem to like these constraints. Speaker B: Yes, I admire the linguist who discovered them. Johni said that Mary thinks that the city in which she met himi is beautiful. Hisi father said that Mary thinks that the city in which she met Johni is beautiful.

Whichever way we analyze the dependencies in (13), it is a fact that the parser can easily handle these nonlocal dependencies. This means that (12) is not ungrammatical because of absolute parsing limitations on the computational system. As indicated in Bouchard (2002: 344), “the parsing restrictions hold

A solution to the conceptual problem of cartography

263

for a certain mechanism, not for the system as a whole, and we cannot make the particular restrictions follow from general locality restrictions.” I also show there that we can explain in which cases locality plays a role: it is due to the nature of the substantive means used to establish the relation and does not have to be stipulated.

3.2. The substances of syntax If we appeal to the properties of the conceptual and perceptual substances of language, we have a good basis for principled answers to basic questions about various properties of language. The first step is to look at basic syntactic relations from the perspective of these substances, i.e., to look at syntax as a part of the system of signs: syntax provides a signifiant to a combination of signifi´es. Consider the simple combination of the two words in the expression pink poodle. Since each sign is complex by definition – a form resulting from the union of a concept and a percept, when two signs are combined by a relation R, R operates simultaneously on both their signifi´es and their signifiants as in (14). (14)

Conceptual-Intentional: {PINK} === R(CI) === {POODLE} | | Sensory-Motor: {pink} === R(SM) === {poodle}

So R is itself a sign. This raises two questions. First, what is the nature of the relation R in CI, i.e., what signifi´e is it? The signifi´e of R is a relation of predication, in the broad sense that it links two units so that one unit tells us something about the other, either in the usual sense of subject-predicate and topic-comment as in (15), or in the sense of saturation as in (16). (15)

a. b.

John is sick/left early. (property of VP attributed to the subject) That book, I really liked. (property of comment attributed to the topic)

(16)

a.

liked that book (the property of the direct object is attributed to ‘liking’, the V; i.e., it saturates the Verb) in the kitchen (the property of kitchen is attributed to ‘location’, the P)

b.

Consider now the second question concerning the relation R: what is the combinatorial means to express R(CI) in SM, i.e., what is the signifiant R(SM)? The answer to this question is already provided in Bouchard (1996): this signifiant will take whatever form a language arbitrarily selects among those which

264

Denis Bouchard

our physiology provides as a combinatorial percept in the modality of this language. In an oral language, the auditory-oral channel is restricted to the physical dimension of time. So the options are very limited: this signifiant can be the Juxtaposition of the two signs, the Superimposition of one as an intonation on the other, or marking one in a conventionalized way that indicates that a relation is being established with the other. Schema 1 Combinatorial signifiants in oral languages Syntagmatic associations: – Juxtaposition (1) – Superimposition (2) Paradigmatic associations – Dependent marking – Juxtaposition (3) – Superimposition (4) – Head marking – Juxtaposition (5) – Superimposition (6) 1–syntactic ordering: head and object, phrase and modifier adjunct, etc. 2–syntactic intonation: focus, question, exclamation intonation, etc. 3–morphological ordering: case affixes in Latin, German 4–morphological tone: case tones in Rendille 5–morphological ordering: object marking on verb in Mohawk 6–morphological tone: tone on the verb indicating tense or aspect in Ngiti Languages vary in their choices of signifiants among these combinatorial percepts, as expected under arbitrariness.14 For instance, the syntactic relation ’direct object’ can be expressed by any of these combinatorial signifiants. This follows from Saussure’s general principle of Arbitrariness: there is no “reason of nature” for a language–let alone all languages–to choose any particular combinatorial signifiant among those in Schema 1 in order to express a particular relation: any of the signifiants is a possible candidate, and indeed each of the possibilities seems to be instantiated in some language or other. Languages choose among the various possibilities of combinatorial signs, just as they arbitrarily choose among the various possibilities of unit signs. Which combinatorial percepts are possible is not stipulated in some Universal List, but is determined by prior properties of the perceptual substance of the modality of the particular language.15

A solution to the conceptual problem of cartography

265

4. Adjectival modification in Exaptive Grammar Under the view that the modification of a noun by an Adj is a combinatorial sign, we must determine what are the signifiant and the signifi´e for each case. In French, Juxtaposition is the combinatorial signifiant used to express adjectival modification.There are two possible orders to a juxtaposition, hence two possible signifiants here: Adj-N or N-Adj. Since the signifiants are different, we expect them to be linked to different signifi´es, i.e., that each order receives a different meaning: this is indeed the case as we saw in examples (4–5) (Fact 6). Concerning the signifi´e, we can determine precisely how the predication operates by looking at the semantic contribution of various Adjs and the elements of meaning of nouns. As shown in Bouchard (2002:59–60), the signifiant N-Adj has for signifi´e a predication of the properties of the Adj onto the denotation of the noun, which is determined by a network of interacting elements that Montague Semantics identifies as (i) a characteristic function f which provides the property that interprets the N, (ii) a specification for a time interval i, which tells us at what moment f holds, (iii) an indication of the possible world w which allows us to know whether f holds in the “actual” world or in some other imagined world, and (iv) a variable assignment function g, that allows us to determine the truth value of the final formula by associating each variable with a particular entity in the model. This network of elements determines the set of things that have the property of being a f in w at i, i.e., the extension of the Noun; the variable assignment function g determines the denotation of the expression. Depending on its semantics, an Adj may apply to the whole network of the noun, and/or any of its subparts. In (4a) and (5a), the postnominal Adj modifies the whole network of the N, its denotation: ancienne indicates that the referent denoted is aged, and seule indicates that there is only one referent in the relevant context. When the signifiant is the opposite order Adj-N, the signifi´e cannot be the same predication as in N-Adj since a change of signifiant necessarily brings about a change of signifi´e. This falls under a general principle of Other-Value which is inherent in a system where signifi´e and signifiant are values for one another.16 For the order Adj-N the signifi´e is a predication of the properties of the Adj onto a subpart of the network of the noun. The meaning of the Adj determines which one is targeted. The temporal Adj ancienne is compatible with the interval of time i, stating that the interval of application of the characteristic function of e´ glise is ancient, not the building denoted. The Adj seule applies to the g-function of femme and this indicates that there is a single value assignment to which the characteristic function applies, that only one individual has the characteristic properties of a woman in the model.

266

Denis Bouchard

The choice of which order is the signifiant for which predication – onto the whole head or onto one of its subparts – is arbitrary. Therefore, it must be conventionalized, just like it is for any sign, in order for the information to be consistently encoded. This convention is the value selected by the language for the central linearization parameter (i.e., the Head Parameter): (17)

Linearization Parameter: The predicator precedes/follows the element it applies to.

French and English are head-initial languages: the head precedes the element which saturates or modifies it, as indicated in (18): (18)

In French/English, the predicator follows the element it applies to.

By (18), the Adj that follows the N in e´ glise ancienne targets the whole network of e´ glise, whereas in ancienne e´ glise, the Adj targets only a subpart of the N. The Adj has the same semantics in the two constructions – ancienne has only one meaning, one entry: the prenominal Adj gets an “intensional” interpretation because it targets a subpart of N, and the postnominal Adj gets a classifying reading since it targets the whole network of N subfunctions. Because it is conventionalized, the relationship between the order of the Adj and what the Adj modifies holds very generally. If there are two postnominal Adjs (Fact 4) as in (2g), the mechanism of juxtaposition predicts that the scope will be N A2 A1 , since the Adj directly juxtaposed to the N is necessarily combined with it, and the next Adj is juxtaposed to the combination N-Adj, as indicated by the brackets in (19): (19)

[[vase chinois] cher]

The mechanism of juxtaposition similarly predicts that prenominal Adjs will have fixed scope, but in a mirror order of postnominal Adjs (Fact 2): new and nouveau are directly juxtaposed with the N in (20), whereas alleged and suppos´e are juxtaposed to these constituents. (20)

a. b.

[alleged [new miracle]] [suppos´e [nouveau miracle]]

As for Fact 1, French and English differ in the distribution of bare Adjs due to an independently motivated difference in the way these languages encode semantic Number in nominal expressions: its signifiant is different, as arbitrariness predicts should be possible (Bouchard 2002, chapter 3). There are three basic possibilities: (i) the Det bears semantic Number (French), (ii) the Noun bears se-

A solution to the conceptual problem of cartography

267

mantic Number (English), (iii) a special word bears semantic Number (Wallon). Bouchard (2002) shows that these differing ways of expressing Number account for other differences between these languages. Thus, Det can appear alone in French (as a clitic pronoun) because it bears Number, which provides it with the minimum content required to function as a referential argument. In English, Det does not bear Number, but rather the N does, and it can appear alone in bare NPs, whereas bare NPs are fairly rare and require additional conditions in French. Since English nouns are actually complex expressions of the form [N + Number], an Adj does not establish the same relation with [N + Number] as it does with a numberless N because the semantics of these elements is not the same. According to (18), a postnominal Adj must target this whole complex form, and the semantic properties of [N + Number] make it compatible with only a special kind of argument-taking Adj, a very restricted class of transitory, stage-level Adjs as in (1f).17 Other “regular” bare Adjs must relate to the N alone. Since N is a subpart of a complex [N + NUM] in English, these Adjs can only target the N subpart by being prenominal, as predicted by the Linearization Parameter and the principle of Other-Value. We could account for Fact 3 by assuming that prenominal and postnominal Adjs have fairly free scope because the bracketing is fairly free, with the two options in (21): (21)

a. b.

[[Adj N] Adj] [Adj [N Adj]

However, we saw in (2h) that in a sequence Adj-N-Adj-Adj, the postnominal Adjs do not have free scope relatively to each other. Moreover, only prenominal Adjs are subject to some sandhi effects and to the No-Complement condition (Bouchard 2002). This indicates that the prenominal Adj is actually structurally closer to the N than the postnominal Adj. This closeness is also revealed in the fact that a prenominal Adj applies to the inner elements of the N, hence that it is part of the definition of the intension of the N, and so it forms a complex head with N. Therefore, in (2h), the prenominal Adj pr´esum´es and the N professeurs jointly head the structure as in (22) (also following Chomsky (1995) that a head projects not only its categorial features, but all its features). (22)

pr´esum´es professeurs pr´esum´es professeurs pr´esum´es professeurs pr´esum´es professeurs chinois malhonnˆetes

268

Denis Bouchard

The idea is that when a head has multiple specifications, it could operate in two different ways: the specifications may operate jointly, or independently from one another with respect to other elements in the structure. If the specifications A and B of a head function as a unit, whatever is under the scope of A is also under the scope of B, and whatever has scope over A also has scope over B. If the specifications function independently, we have a portmanteau. Both logical possibilities are equally efficient modes to process a head with multiple functional specifications, therefore a parametric choice arises, and a language must choose between these strategies. French opts for the portmanteau strategy, as can be seen in the behavior of lexically complex heads like aux ‘PREPa` MASC-PLUR-DEF’: its prepositional feature PREPa` can independently have scope over its argumental complement in aux enfants ‘to the children’ in (23): (23)

X PREPa` -MASC-PLUR-DEF enfants

Under a portmanteau analysis, in (22), malhonnˆetes is under the highest projection, and it may have scope over its sister constituent and the elements it contains, in particular over pr´esum´es and chinois. Moreover, the complex head pr´esum´es professeurs has scope over malhonnˆetes because it projects entirely and is thus sister to malhonnˆetes; since this complex head contains the features of pr´esum´es, pr´esum´es may have scope over malhonnˆetes by a portmanteau effect. Similarly, pr´esum´es and chinois have two possibilities of scope interpretation. On the other hand, due to the structural relation between them, malhonnˆetes always scopes over chinois. Thus we account for Fact 3 about the relative scope of prenominal and postnominal Adjs.18 Finally, to account for the preferred order of Adjs (Fact 5), Bouchard (2006b) expands an idea of Ziff (1960) and proposes the principle in (24). (24)

Principle of Adjective Serialization: The more the property expressed by an Adj can form a relevant and usual Concept with the N, the more this Adj tends to be close to the N, i.e., to modify the N more directly than other Adjs.

This principle predicts a relatively free order of Adjs, depending on the context. This accords with the facts. There is a tendency to favor a certain order, which scholars have expressed in templates, but it is only a tendency and it follows from the relative likelihood of the Concepts which are built. Humans have a propensity to construct Concepts which can be linguistically permanent – they have an entry in the mental lexicon –, or provisional – they

A solution to the conceptual problem of cartography

269

are built contextually as need arises (Krifka 1995), as combinations of Nouns and Adjs in the case at hand. The order in which Adjs are combined with the N is dependent on the expectations we have according to our knowledge. For instance, Sproat and Shih (1988) remark that Adjs of Origin are typically used as indicators of a taxonomy, as expected since these Adjs express a property which is very frequently used to differentiate objects. Consequently, they tend to form a Concept with the N, to be combined directly with it as in (1g) and (2g). But in a context where we are discussing various techniques and traditions to obtain blue color in pottery, color becomes more relevant, and the order in (25) is then natural. (25)

a. b.

Chinese blue vases vases bleus chinois

Similarly, depending on the context, either the Color or the Shape expressed by the Adjs in (26) can be relevant: in (26a), they are looking for cylindrical containers and consider variously colored ones, whereas red containers are more relevant in (26b). (26)

a. b.

They did not find any red cylindrical containers in Bagdad. They did not find any cylindrical red containers in Bagdad.

Bouchard (2006b) discusses several different orders of Adjs and shows that the principle in (24) accounts both for the apparent rigidity of some orders as well as their contextual flexibility in general.

5. Concluding remarks Cartography and similar enterprises end up listing properties of language in a format which uses notions that do not provide a basis for a principled explanation of the facts. The solution to this conceptual problem is to shift the burden of analysis to notions which are externally motivated, properties of the substances of language. If we take the properties of these substances as initial conditions, we obtain strong canalization effects on language and its acquisition. Due to the analytical tools it uses, Cartography has to more or less arbitrarily insert these effects into the theory in language-specific lists of uninterpretable features or general lists of principles of Universal Grammar. I argue instead that these effects arise from the logically prior properties of the perceptual and conceptual substances. In any model, these properties necessarily impose boundaries within which a

270

Denis Bouchard

child charts a highly circumscribed course in language development. I make the strongest hypothesis, i.e., that they are sufficient and are the only canalization elements of language. We have seen that this indeed seems to be a path to principled explanation for the distribution and interpretation of Adjs in nominal expressions – a construction that has been central to the development of Cartography.

Notes 1.

2. 3. 4. 5. 6.

7.

8. 9.

For instance, the observational proposition that the sun moves across the sky is a valid observation in everyday life (and the time at which it rises and sets is published in media every day); this proposition was an axiom of cosmology for a long period of time, but is no longer considered a valid proposition in that science. Except for a small class of Adjs as in (1f) when they receive a transitory, stage-level interpretation, as noted by Bolinger (1967). A comma intonation can change the relative scope of Adjs, but for reasons of space, I cannot discuss the details here. See Bouchard (2002:189ff.). Contrastive intonation on an Adj can affect what position it appears in, as indicated in Bouchard (2002:187ff.). A similar assumption is required for Adjs with a complement. Since Size and Color are not adjacent in the serialization poss> cardinal> ordinal> quality> size> shape> color> nation, it must be stated that Size can select either a ShapeP or a ColorP, or a NationP; alternatively, all these functional categories are always present but do not all appear on the surface. A reviewer says that Fact 6 “is actually something Cartography can neatly explain by playing with what moves to SpecDP. With some simplification, if only the noun moves, leaving the Adj in the postnominal position, the referent is described as a church that is old; if both the noun and the Adj move, then the referent is described as a church related to the past, i.e. something that used to be a church.” (Incidentally, movement is not to SpecDP, as this would predict the wrong order with respect to the Det une in (4).) The reviewer confuses correlation with explanation, an unfortunately frequent mistake. Why does N-Move correlate with the interpretation ‘church that is old’ and not ‘former church’? He could just as happily make the opposite correlation. Moreover, we may wonder why in this case it is the surface position that determines the interpretation, whereas the interpretation of the relative scope of Adjs is determined by the underlying positions. See Tesni`ere (1959) for an earlier proposal along those lines. Not only does Cartography have the power to reverse any correlation, such as the one between position and scope of Adjs, it also has the power to reverse all words of a sentence. For instance, it is sometimes assumed that Question intonation is anchored in a structural position of the sentence and spreads over all the terminals. Instead

A solution to the conceptual problem of cartography

271

of intonation, or concomitantly, a Question uF could appear on every terminal and trigger movements that reverse the order of the sentence as in (i):

10.

11.

12. 13. 14.

15.

(i) John took the money → John took money the → John money the took → money the took John Some alternative proposals attempt to solve these problems (see for example the papers by Neeleman e.a., Costa or Wagner in this volume), but so far there is no general motivation for the proposed options: they are typically case-by-case solutions to some problem with Cartography, but do not address the fundamental question: why Generative Grammar has been driven to lists. Of course, it is not just any perceptual element which can function as a signifiant: there are constraints of usability which limit the linguistic elements of the perceptual substance to controlable elements. So for instance, the system for producing and perceiving sounds and the one for gestures are both quite controlable, whereas the system for touch require an impractical proximity, as do the systems of smell and taste, which additionally are extremely limited in their productions. A reviewer questions the distinction between these two substances: “perceptive and motoric elements also must involve a cognitive level of representation, and at this level, the representations of perceptive and motoric elements may plausibly be treated as concept, equal to other concepts within the conceptual substance.” Saussure noted this and used the term ‘image acoustique’ to refer to the mental representation of a signifiant. Nevertheless, the two substances are not equal: whereas conceptual substance is strictly brain-internal, perceptive and motoric elements additionally necessarily involve systems which are outside the brain, a crucial requirement in order for them to “symbolize”, to act as a signifiant. But sometimes murmuring softly into someone’s ear also expresses prominence, stressing the importance of the privacy of the message. Even worse, to get the agreement by a spec-head relation, one must also assume a covert FC which also bears a feature of femine gender. In GG, one of these is taken as basic. The dominating proposal is that fixed positions encode R, i.e., that Juxtaposition is the universal (often underlying) means to encode R. This contrasts with classical schools, which took functional markers as basic, since Latin was the language of prestige on which the models were based. Neeleman and Weerman (1999) make a similar assumption. Some may object that this is misrepresenting the view of GG, that syntax does not deal with linearity, but hierarchy (see for instance Breheny and Adger 2005). But as I point out in Bouchard (2006a), linearity directly determines the hierarchical choices. Thus, Chomsky (2000) discusses an example of active/passive alternation and says that “the surface phonetic relations are dissociated from the semantic ones.” This shows that there is an implicit assumption that some semantic relations are ‘normally’ associated with particular surface phonetic relations. This contrasts with the assumption in GG that ‘direct object’ is coded by a sister position to the head. A nominal expression marked with Accusative case is an apparent violation of the fixed-position hypothesis since case-marked phrases are typically

272

Denis Bouchard

quite free in their ordering: this is a classical instance of functional covariation (see Bouchard 2001, 2002:382–389, and references therein). To maintain the hypothesis, additional assumptions are introduced, such as recoding the case marking into a positional relation by a scrambling operation, for example. Since there is no “reason of nature” to make the hypothesis that Juxtaposition is the universal signifiant – a very odd idea – the hypothesis must be given the status of a universal principle in UG, i.e., it must be stipulated. 16. This principle of Other-Value follows from a requirement of distinctivity. To give an extreme example, a language with /a/ as the sole value for all its signifiants would not be viable as a human language. 17. Adjs with a complement also have a semantics that forces them to modify an argument-like element, so to appear in postnominal position (Bouchard 2002:140f). 18. Bouchard (1995:406ff, 2002:127ff) explores the details of a portmanteau approach to tense and the distributrion of adverbs without recourse to movement.

References Beghelli, Filippo, and Tim Stowell 1997 Distributivity and Negation: The syntax of each and every. In: Anna Szabolcsi (ed.), Ways of Scope Taking, 71–107. Dordrecht: Kluwer Academic Publishers. Blinkenberg, Andreas 1933 L’ordre des mots en fran¸cais moderne: Deuxi`eme partie. Copenhagen: Levin and Munksgaard. Bolinger, Dwight 1967 Adjectives in English: Attribution and Predication. Lingua (18): 1–34. Bouchard, Denis 1995 The Semantics of Syntax. Chicago: The University of Chicago Press. Bouchard, Denis 1996 Sign Languages and Language Universals: The Status of Order and Position in Grammar. Sign Language Studies (91): 101–160. Bouchard, Denis 2001 The concept of ‘universal’ and the Case of Japanese. Lingua (111): 247–277. Bouchard, Denis 2002 Adjectives, Number and Interfaces: Why Languages Vary. Oxford: Elsevier Science. Bouchard, Denis 2006a Good Intentions and Actual Deeds: A response to my critics, Lingua (116): 523–542.

A solution to the conceptual problem of cartography

273

Bouchard, Denis 2006b S´eriation des adjectifs dans le SN et formation de Concepts, Recherches linguistiques de Vincennes 34: URL: http://rlv.revues.org/document1383.html. Breheny, Richard and David Adger 2005 Commentary on ’exaptation and linguistic explanation’. Lingua (115): 1673–1677. Chomsky, Noam 1995 The Minimalist Program. Cambridge: MIT Press. Chomsky, Noam 2000 Minimalist inquiries: The framework. In: Roger Martin, David Michaels, Juan Uriagereka (eds.), Step by step: Essays on minimalist syntax in honor of Howard Lasnik, 89–155. Cambridge, MA: MIT Press. Chomsky, Noam and Halle, Morris 1968 The Sound Pattern of English. New York: Harper and Row. Cinque, Guglielmo 1994 On the Evidence for Partial N Movement in the Romance DP. In: Guglielmo Cinque, Jan Koster, Jean-Yves Pollock, Luigi Rizzi and Raffaella Zanuttini (eds.), Paths Towards Universal Grammar, 85– 110. Georgetown: Georgetown University Press. Cinque, Guglielmo 1999 Adverbs and Functional Heads. Oxford: Oxford University Press. Dixon, Robert 1982 Where Have All the Adjectives Gone? The Hague: Mouton. Frampton, John and Gutmann, Sam 1999 Cyclic Computation, a Computationally Efficient Minimalist Syntax. Syntax (2):1–27. Greene, Brian 1999 The Elegant Universe. New York: Random House. Kayne, Richard 1994 The Antisymmetry of Syntax. Cambridge, MA: MIT Press. Kosslyn, Stephen 1980 Image and Mind. Cambridge, MA: Harvard University Press. Krifka, Manfred 1995 Common Nouns: a contrastive analysis of Chinese and English. In: Gregory Carlson and F.J. Pelletier (eds.), The Generic Book, 398–411. Chicago: The University of Chicago Press. Lakatos, Imre 1970 Falsification and the Methodology of Scientific Research Programmes. In: Imre Lakatos and Alan Musgrave (eds.), Criticism and the Growth of Knowledge, 91–196. Cambridge: Cambridge University Press.

274

Denis Bouchard

Munaro, Nicola and Hans-Georg Obenauer 2000 On underspecified wh-elements in pseudo-interrogatives.Venice Working Papers in Linguistics 10(1): 121–151. Neeleman, Ad and Fred Weerman 1999 Flexible Syntax. Dordrecht: Kluwer. Ross, John Robert 1986 Infinite Syntax. Norwood: Ablex. Saussure, Ferdinand de 1916 [1967]. Cours de linguistique g´en´erale, T. de Mauro (ed.). Paris: Payot. Siegel, Muffy 1980 Capturing the Adjective. New York: Garland Publishing. Sproat, Richard, and Chilin Shih 1988 Prenominal Adjectival Ordering in English and Mandarin. Proceedings of NELS (18): 465–489. Sproat, Richard and Chilin Shih 1990 The Cross-linguistic Distribution of Adjective Ordering Restrictions. In: Carol Georgopoulos and Roberta Ishihara (eds.), Interdisciplinary Approaches to Language. Essays in Honor of S.Y. Kuroda, 565–593. Dordrecht: Kluwer. Tesni`ere, Lucien ´ ements de syntaxe structurale. Paris: Editions Klincksieck. 1959 El´ Vendler, Zeno 1968 Adjectives and Nominalizations. The Hague: Mouton. Ziff, Paul 1960 Semantic Analysis. Ithaca: Cornell University Press.

Adjective placement and linearization Ion Giurgea Abstract. This paper investigates a word order paradox encountered inside NPs in languages with postnominal adjectives: while the relative order of adjectives cannot be derived by head-movement of the noun, the order of adjectives with respect to the complements of the noun and the relative order of complements indicate head-raising of the noun under current assumptions about complement merge and binding. I propose to derive this word order pattern by linearization rules rather than by syntactic movement. I show that the generalizations observed can be formally implemented if we allow linearization rules to make reference to pairs of elements which are not sisters, but are immediately dominated by projections of the same head. I also provide evidence for an adjunction analysis of postnominal adjectives.

1. Introduction This paper discusses some word-order generalizations which, considered together, constitute a problem for any simple mapping from syntactic structure to linear order, for systems based on Kayne’s (1994) LCA, such as the cartographic program, as well as for its current competitors. These generalizations concern the NP-internal word order in languages with postnominal adjectives. I propose that this word-order paradox may be solved by linearization rules, without resorting to syntactic movement. Crucially, these rules involve linearization parameters. The linearization system I propose departs from the current parametric theories of linearization by allowing linearization instructions to apply to pairs of elements which are not sisters, but are immediately dominated by projections of the same head. The paper is organized as follows: in section 2 I first present and illustrate the generalizations: (I) the relative order of postnominal adjectives is left-branching; (II) adjectives, except for heavy APs, precede complements; (III) in the unmarked order, higher complements precede lower ones. The rest of the section discusses generalization (I). After briefly showing how it came to be recognized in linguistic literature (section 2.2), I discuss the criteria which allow us to establish the hierarchical relations among adjectives independently of their linear order (section 2.3). In sections 2.4 and 2.5 I argue that generalization (I) is better accounted for by analyzing postnominal adjectives as adjuncts linearized to the

276

Ion Giurgea

right, than by syntactic movement, as it is done in the antisymmetric framework. A crucial piece of evidence in favor of a ‘symmetric’ analysis comes from the category of classifying adjectives. In section 2.6 I show that the most recent cartographic analysis of adjective placement, Cinque’s (2005a, 2005b), cannot explain generalization (I) in conjunction with the other two generalizations. In section 2.7 I show that this word order pattern is equally problematic for current systems with linearization parameters. In section 3 I present my own proposal: a linearization system where linearization instructions may apply to pairs of elements which are not sisters, but are immediately dominated by projections of the same head. Beside the head parameter, I propose a linearization parameter concerning the placement of light modifiers. Section 3.2 provides additional evidence for the explanation based on PF-rules, illustrating the relevance of weight in adjective placement. Section 4 is devoted to generalization (III), which is not limited to the nominal domain, but extends to the verbal domain, and discusses the criteria proposed for establishing the hierarchical relations among co-arguments independently of the linear order.1

2. Word-order in NPs in languages with postnominal adjectives (Romance, Celtic, Semitic) 2.1.

Basic generalizations

The generalizations which I discuss are about NP-internal word order in languages with postnominal adjectives. By examining different languages of this type, belonging to several linguistic groups – Romance, Celtic and Albanian from the Indo-European family, Hebrew and Standard Arabic from the Semitic family – one is led to the following empirical generalizations : (I) The relative order of (postnominal) adjectives is left-branching (LB): (((N A1 )A2). . . )2 (II) Adjectives, with the exception of heavy APs, precede complements. Heavy APs may either precede or follow, depending on their weight and on the language: N A . . Arg . . . (III) In the unmarked order, higher complements precede lower ones, if we take binding relations between co-arguments as indicative for c-command: so we find right-branching order (RB): N . . Arg2 (Arg1 (

Adjective placement and linearization

277

Considered together, these generalizations lead to the following paradox: while the order in (I) may be explained either without movement (in a system with a head-parameter) or by phrasal movement with pied-piping (in an antisymmetric analysis), the order in (II) and (III) requires head-movement of the noun, under standard assumptions about linearization and level of attachment of complements. Before providing further evidence in favor of these generalizations, I give some examples which illustrate all of these rules applying in one and the same NP: (1)

a.

b.

c.

le d´ebat philosophique incessant du jeune hommei the dispute philosophical unresting of-the young man avec soni aˆ me (French) with his soul o dedicat¸ie muzical˘a emot¸ionant˘a a compozitoruluii a dedication musical touching art composer-the.G c˘atre iubita sai (Romanian) to beloved-the his ha-hatkafa ha-gar’init ha-bilti-cfuya sˇel Amerika ’et the-attack the-nuclear the-unexpected of America obj Iran (Hebrew) Iran

2.2. A brief history: the discovery of generalization (I) Most research on adjective placement in the early 90’s focused on generalization (II): based on examples such as (2), the postnominal position of the adjectives in N-A-languages3 was explained by head movement of the N to some functional head in the extended nominal projection (Num, or another one) (Cinque 1994, 1996, Valois 1991, Crisma 1993, 1996, Bernstein 1993, Giusti 1993, Rouveret 1994): (2)

l’invasione brutale dell’ Albania the invasion brutal of-the Albania

(Italian)

But, gradually, generalization (I) came to be noticed: Lamarche (1991), followed by Bouchard (1998) and Laenzlinger (2000) established it for French, Bosque and Picallo (1996) for Spanish, Giurgea (2005) and Cornilescu (2006) for Romanian, Willis (2006) for Welsh, Sichel (2000), Shlonsky (2004) for Hebrew, Fassi Fehri (1999) for Arabic.

278

(3)

Ion Giurgea

a.

un fruit orange e´ norme a fruit orange huge Cf. ’a huge orange fruit’

(French, Lamarche 1991)

b.

hafgaza ‘isra’elit masivit ‘efˇsarit bombardment Israeli massive possible

c.

une e´ ruption volcanique massive an eruption volcanic heavy

(Hebrew, Shlonsky 2004) (French)

c’. *une e´ ruption massive volcanique an eruption heavy volcanic Cf. ‘a (heavy) volcanic (*heavy) eruption’ A first reply to these counterexamples to N-movement was to say that the second adjective in the sequence [[N A1 ] A2 ] is a predicative adjective, in a reduced relative structure (Cinque 1994, Bernstein 1993, Sleeman 1996). But it was soon noticed that the second adjective is not always an adjective which may appear predicatively (it is not always intersective) (see Bosque and Picallo (1996) for Spanish, Bouchard (1998) for French): (4)

a.

la classe politique fran¸caise4 the class political French ‘the French political class/elite’

b.

reu¸sita economic˘a american˘a success-the economic American ‘the American economic success’

(French)

(Romanian)

Moreover, LB-order appears even before complements (a position where we do not expect to find reduced relative clauses): (5)

a.

b.

o expunere s¸tiint¸ific˘a am˘anunt¸it˘a a a report scientific detailed art fenomenelor phenomena-the.G un expos´e scientifique d´etaill´e des ph´enom`enes a report scientific detailed of-the phenomena ‘a detailed scientific report of the phenomena’

(Romanian) (French)

Eventually, generalization (I) was fully acknowledged, and a different syntactic derivation was consequently proposed in the antisymmetric framework, involving roll-up phrasal movement, as shown in (6) (Laenzlinger (2000) for French, Cinque (1996) for Basque, Cinque (2000), Sichel (2000) and Shlonsky (2004)

Adjective placement and linearization

279

for Semitic languages). Recently, Cinque applied this derivation to Romance languages too (see Cinque 2003, 2004).5 (6)

Agr2P Agr1P Agr2´ Agr2

FA2P FA2´

AP(2) FA 2

NP

Agr1P Agr1´ Agr1 AP(1)

FA1P FA1´ FA1

NP

2.3. Apparent counter-examples to (I). Towards independent criteria for the hierarchical order of adjectives In some languages with post-nominal adjectives, RB-order was also evinced: Rouveret (1994) shows that the order of adjectives expressing size, color and origin is the same in Welsh as in English, despite the postnominal position of the adjectives: (7)

a.

cwpan mawr gwyrrd Sieineaidd cup big green Chinese a big green Chinese cup

(Welsh, Rouveret 1994)

The comparison to English is not the only support for the idea of a RB-order here; another source of support is the observation that the stacking order Size > Color > Origin (where ‘>’means ‘higher than’, ‘stacked above’) is found in many languages, as shown by typological studies of relative order of adjectives.6 However, it is important to notice that this exception to LB-order involves adjectives of the same semantic type (quality, intersective): every case in which the difference in semantic type between two adjectives is really significant dis-

280

Ion Giurgea

plays the LB-order, as shown by Willis (2006). Labels such as ‘Age’, ‘Shape’, ‘Color’ reflect at best what we may call cognitive types, rather than semantic types. Hierarchies such as Hetzron’s (1978) or Scott’s (1998) (which has no less than 16 classes of adjectives) do not distinguish major classes of adjectives, which differ by denotation or by the type of semantic composition, from minor ones. A major semantic distinction is the one between classifying (or relational) and thematic adjectives on the one hand, which directly apply to the N function (or kind) (as arguments or modifiers of implicit arguments in the lexical-conceptual structure of the noun) (type ), and intersective adjectives on the other hand, which are predicates of individuals () and combine with the property denoted by the NP by predicate intersection.7 In this case, the hierarchical order can be directly inferred from semantics: we expect classifying adjectives to be merged lower than quality adjectives.8 And, in this case, Welsh displays LB-order: (8)

a.

acen Saesneg gref accent English strong

b.

anghenion addysgol arbennig needs educational special

(Welsh, Willis 2006)

In contrast, for adjectives belonging to the same semantic type, like those in (7) (which are all intersective), semantics does not impose a certain hierarchical order, so we expect to find variation. And, indeed, we find that adjectives belonging to different cognitive classes inside the intersective type can be coordinated in some languages (ex. (9a)) and allow any order (ex. (10)); as (9b–c) show, coordination is never possible between adjectives belonging to different semantic types: (9)

a.

o minge mare s¸i ro¸sie a ball big and red

(Romanian)

b. *un studiu literar s¸i am˘anunt¸it a study literary and detailed c. *a visible and invisible star (S-level and I-level) (10)

a.

pi¸ekna du˙za piłka beautiful red ball

b.

du˙za pi¸ekna piłka red beautiful ball

c.

o minge ro¸sie mare a ball red big

(Polish, Hetzron 1978)

(Romanian)

Adjective placement and linearization

d.

281

o minge mare ro¸sie a ball big red

Even finer semantic distinctions among adjectives, such as the distinction between evaluative (affective) and non-evaluative adjectives, correlate strictly with LB-order in Welsh: (11)

a.

athro ifanc hoffus teacher young likeable

b.

caneuon newyd gwych songs new great

(Welsh, Willis 2006)

Moreover, quality adjectives do not always respect the order in (7) even in Welsh. As Willis (2006) pointed out, with other lexical choices we may obtain the expected Origin < Color and Form < Size orders: (12)

a.

ryg Twrcaidd coch rug Turkish red

b.

to crwn uchel roof round high

(Willis 2006: 25c–d)

We may thus say that the facts in (7) do not prove the existence of RB-order in Welsh. The order among quality adjectives may be subject to cross-linguistic and language-internal variation. The order in (7a) may be the result of other principles or tendencies than those operating in (7b) and (12), perhaps a preference to place more frequent and light adjectives closer to the noun. We see now that the comparison with English, often used as a rule of thumb, is not sufficient for establishing the hierarchical relation between two (noncoordinated) adjectives: we need more deeply grounded criteria. One of those, which I’ve already mentioned, is the difference in semantic type. Besides the relations presented above, classifying subject-oriented > manner: (14)

the unfortunate likely silly heavy bombardments

282

Ion Giurgea

However – and this fact has been less noticed in the literature – there are ways to establish a hierarchical order (independently of the linear order) even between individual adjectives in concrete contexts, abstracting away from their semantic type or cognitive class. I know of two situations of that kind: one is when one of the adjectives (let’s say A1 ) forms a current topic along with the noun, a topic from which another adjective, A2 , makes a selection. In this case it is reasonable to assume that [N A1 ] forms a constituent, so that A2 > A1 . Here are some examples from N-A languages, where the LB-order is found, as expected: (15)

a.

literatura francez˘a medieval˘a literature-the French mediaeval (OK when speaking about French litterature)

b.

literatura medieval˘a francez˘a literature-the mediaeval French (OK when speaking about mediaeval literature)

c.

l’´ecriture hittite id´eographique the writing Hittite ideographic (OK when speaking about Hittite writing)

d.

l’´ecriture id´eographique hittite the writing ideographic hittite (OK when speaking about ideographic writing)

(Romanian)

The other case where hierarchical order can be established looking at the contextual meaning and not at the semantic type is when one of the adjectives has active scopal properties, so that the hierarchical orders A2 >A1 and A1 >A2 are truth-conditionally different. One such case, pointed out by Svenonius (1994), looks as follows: when A2 takes scope over A1 , it introduces a different time of evaluation of A1 , as illustrated in (16): (16)

a. a.’

a broken valuable pot

(valuable before being broken) (Svenonius 1994) un vas pret¸ios spart (Romanian) un vase pr´ecieux cass´e (French) a pot valuable broken

Adjective placement and linearization

b. b.’

283

a valuable broken pot (valuable at the present/reference time) un vas spart pret¸ios (Romanian) un vase cass´e pr´ecieux (French) a pot broken valuable

These situations are interesting because they show that hierarchical order does not always depend on the class to which an adjective belongs, contrary to the tenets of cartography. I take this as an argument for the traditional analysis of adjectives as adjuncts (for other arguments, see Matushansky 2002, Baker 2003, Giurgea 2005). Now if we consider the relative order of adjectives in N-A languages according to these criteria, there is only one exception I know about: in Albanian, adjectives with preposed Agr, which constitute the vast majority of quality adjectives, usually precede adjectives without Agr even if the latter are classifying or thematic: (17)

nj¨e luft¨e e madhe shoq¨erore a fight agr big social

But even here several classifying adjectives always show LB-order, even when they follow a quality adjective: (18)

a.

gjuh¨esore shqiptare studimet studies-the linguistic Albanian

b.

t¨e gjitha debatet politike amerikane (m¨e m¨erzisin) agr all debates-the political American (me.acc annoy) ‘All the American political debates annoy me’

c.

e vjet¨er gjuh¨esore gjermane (¨esht¨e ende ljet¨ersi literature-the agr old linguistic German is still e vlefshme n¨e dit¨et tona) agr valuable in days-the our ‘The old German linguistic literature is still valuable nowadays’

Moreover, among classifying and quality adjectives, both orders are actually found, without clear meaning differences: (19)

a.

partive politike t¨e reja parties-D. politic agr new

b.

ndonj¨e komedi muzikore zbavit¨ese any comedy musical diverting

284

Ion Giurgea

c.

ndonj¨e komedi zbavit¨ese muzikore any comedy diverting musical

We can conclude that the exception to the LB-order only concerns one adjective, never several of them, and this adjective, which is the first in the series, is always qualitative. This suggests that quality adjectives may occupy a specifier position of a functional projection while the noun raises to a higher head (perhaps Num).10 It is important to stress that RB-order is never found with classifying and thematic adjectives, which, given their dependency on the lexical meaning of the noun, are the most likely to be NP-internal. I conclude that the data coming from languages with apparent exceptions to the LB-order support the following generalization: (20)

In N-A-languages several classifying and thematic adjectives have only the LB-order, and always follow the noun.

If this is correct, we have a strong argument for an adjunction analysis of adnominal adjectives, against cartographic approaches: classifying and thematic adjectives are heavily dependent for their meaning on the lexical meaning of the noun; the possibility of having several of them and the semantic roles they realize depend on the lexical choice of the noun (see examples in (21)). This makes them good candidates for adjuncts inside the lexical projection. If we assume the existence of a head-parameter for adjuncts, which is set head-first in languages with postnominal adjectives, (20) is immediately derived. No kind of N movement can change the relative order of these adjectives, since they belong to the same projection. By treating all types of adjectives alike, as specifiers of various projections, the cartographic program cannot explain this generalization. (There is no reason why for certain types of adjectives, namely classifying and thematic ones, roll-up movement should be obligatory.) (21)

a. a’. b. b’.

le d´eveloppement technique agricole europ´een m´edi´eval (French) the mediaeval European agricultural technical development la musique dod´ecaphonique instrumentale (French) the instrumental dodecaphonic music

2.4. Adjective placement in a symmetric system. The status of prenominal adjectives The conclusion of the preceding section was that the easiest way to explain the relative order of adjectives in N-A languages is to assume that they are

Adjective placement and linearization

285

N(P)-adjuncts, and that the head-parameter is set headForm, something like: (24)

agrSize [Size [agrForm [Form . . .

Are these Agr heads also universal? Notice that they are only used in N-Alanguages. So, either in Cinque’s hierarchy some heads are universal and some are not, which is not desirable under the assumptions of the cartographic program, or some languages have heads completely devoid of effects at any interface – since in A-N-languages these heads not only have no interpretive content, but also trigger no overt operation, so they have no phonological effect either. Another drawback of the cartographic system is the mechanism used for movement. Cinque (2005a) proposes that the feature triggering various movements inside DP is simply the categorial feature +N. But this feature is unlikely to be a movement trigger because, since we are dealing with nominal functional projections, the feature N is presumably already satisfied by c(omplement)selection. The only case where a c-selectional feature may be accompanied by a

Adjective placement and linearization

289

movement-triggering feature is the case of head-movement, which is presumably due to a formal property – the affixal character of some inflectional heads. But in Cinque’s system movement is phrasal, and is not triggered by heads with affixal properties (all the heads around which phrases raise are completely silent). More importantly, in the antisymmetric system the highest position to which the N raises is arbitrary: since this system treats all adjectives alike, as specifiers, it has no resources to explain why the N(P) stops somewhere in between and does not raise all the way up, past all adjectives. On the other hand, as we have seen, the symmetric system defended here makes some interesting predictions for N-A languages: (a) prenominal adjectives and postnominal adjectives displaying RB-order should belong to specific semantic classes, classes which should, moreover, be compatible with any noun (or, at most, with very general types of nouns, such as count vs. mass); (b) stacked adjectives which do not belong to such classes, and first of all those whose presence is sensitive to the lexical meaning of the N, should always appear in LB-order if postnominal. Classifying adjectives are typical candidates for NP-adjuncts (their presence, number and meaning are dependent on the lexical meaning of the N). And, indeed, they only appear in LB-order in N-A languages, as far as I know. On the other hand, prenominal adjectives are indeed either independent of the lexical content of the noun, just like functional items – see, for instance, former, other, various, true – or have a particular interpretation – non-restrictive –, which is naturally correlated to a position outside the lexical domain. The antisymmetric framework cannot make such predictions. So, if these predictions are further corroborated, we will have a strong argument in favor of the symmetric system.

2.6. The problem While the symmetric system gives the simplest explanation for the relative order of adjectives, the same cannot be said when it comes to including the other generalizations in section 2.1 into the picture. If we wanted to derive generalization (II) in a simple symmetric system without resorting to movement, we would need to say that all adjectives intervening between the noun and its complement(s) are merged lower than complements. But this is highly improbable: even if for classifying adjectives one could assume a position lower than complements on the basis of their close semantic relationship with the noun (see 2.4. above), such a position is much less likely for quality adjectives, which may neverthe-

290

Ion Giurgea

less appear between the noun and its complements (see (1), as well as examples below). (25)

a.

atacul masiv nea¸steptat al japonezilor asupra attack-the massive unexpected art Japanese-the.G on insulei island-the ‘the unexpected massive attack of the Japanese on the island’ (Romanian)

b.

Cette d´eclaration a provoqu´e une r´eaction violente this declaration has provoked a reaction violent imm´ediate aux actions des USA immediate to.the actions of.the US ‘This declaration has provoked an immediate violent response to the actions of the US.’ (French, Bouchard 2002)

c.

addasiad Cymræg newydd o ddrama Thomas Middleton adaptation Welsh new of play Thomas Middleton ‘a new Welsh adaptation of a play by Thomas Middleton’ (Welsh, Willis 2006)

d.

la participation fran¸caise impressionnante a` la the attendance French impressive at the conf´erence conference ‘the impressive French attendance at the conference’ (French, Abeill´e and Godard 1999)

Moreover, a symmetric analysis without movement predicts that complements closer to the noun are lower than complements to their right, which is at odds with generalization (III), under standard assumptions about binding. I give here other examples of this generalization (see also (1)), with binding as well as with the relative order of subjects and objects with deverbal nouns: (26)

a.

disgrifiad (manwl) y gyrrwr o’r description detailed the driver of the ddamwain (Welsh, Rouveret 1994) accident

b. *disgrifiad o’r ddamwain y gyrrwr description of the accident the driver

Adjective placement and linearization

(27)

291

c.

Mae Wyni wedi prynu [lluniau Picassoj ohonoi/∗j ] is Wyn Perf. buy portraits Picasso of-his

d.

Mae Wyni wedi prynu [lluniau Picassoj ei hun∗i/j ] is Wyn Perf. buy portraits Picasso his self

a.

ha-tmuna sˇel kol cayari sˇel kalb-oi the-picture of every painter of dog-his (Hebrew, Shlonsky 1988)

b. *ha-tmuna sˇel kol kelevi sˇel ha-cayar sˇel-oi the-picture of every dog of the-painter of-his The antisymmetric analyses developed for adjective orders cannot give a straightforward explanation to generalizations (II)–(III) either. As can be seen in (6), the standard antisymmetric analysis of adjective LB-order uses phrasal movement of the NP and then of bigger constituents containing the NP. But generalizations (II)–(III) require head movement of the N. Phrasal movement derives the unattested orders in (28) (assuming that higher arguments are merged as specifiers) (the indices note the order of merge, so a lower index means a lower position): (28)

a. b.

N Arg A1 A2 Arg2 N Arg1 A1 A2

If the N first moves as a head and then the phrasal movement begins, we also derive an unattested order15 : (29)

N Arg2 Arg1 A1 A2

Cinque (2005a), based on extensive cross-linguistic data, proposes a restricted set of NP-internal movements which have the effect of deriving all and only the attested orders among Dem, Num, A and N. It can be shown that in this constrained system the word orders in NPs with several postnominal adjectives and complements, reflected in the generalizations I-III, cannot be derived. Cinque restricts movements inside the nominal projections to (i) movements of the NP and (ii) movements of bigger constituents containing the highest member of the N(P)-chain – and not a trace of N –, which are of two types: (a) pied-piping of the whose-picture type, i.e. movement of [NP[XP]], and (b) pied-piping of the picture of who type, i.e. movement of [XP[NP]]. Now, if we move the N together with the complements, we obtain the unattested orders in (28), as we have seen. To be fair, Cinque didn’t consider complements in this article. Perhaps one could assume, in his system, that complements are generated, or moved for

292

Ion Giurgea

checking reasons, in levels above the NP (let’s say nP). Then, we may extract an NP containing just the noun, from below these complements. We can derive like that the order N A1 Complements, but when another adjective is added, we obtain the wrong order (30), because the whole constituent [AgrP NP A1 Compls] should be moved above A2 (this is the only option allowed in this restricted system). (30)

N A1 Compl(s) A2

A derivation which can derive the right orders must, upon each merger of a new adjective, do the following moves (as can be seen in (31), from which I eliminated the empty heads, for expository reasons): (a) move the FP containing the complements and the trace of NP (b) merge A (c) move the remnant AgrP (31)

In (31), Agr2 P gets the order N A1 A2 Arg2 Arg1 . If we want to merge another adjective, the same movements must be repeated: first extract FP, then merge A3 and move the remnant Agr2 P. In this derivation, step (a) contravenes the postulates of Cinque’s system (which does not allow movement of a projection containing just an NP-trace, and not the overt N). Moreover, this movement is unmotivated, and involves a sort of look-ahead, since it only happens before an adjective is merged. Cinque (2005b) provides an unsatisfactory analysis of these orders. He discusses orders N A1 A2 PP where A1 is very low, and he takes it for granted that A1 is always generated lower than the PP:

Adjective placement and linearization

(32)

293

a.

possibile [[ politico personaggio ] della sua citt`a ] → possible political figure of-the his city

b.

[ politico personaggio ]i [possibile [ ti della sua citt`a]] →

c.

[ personaggio]j [ [ politico tj ] [ possibile [ ti della sua citt`a ] ] ]

This analysis doesn’t work for the cases in which both adjectives are generated higher than the XPs following them, which is presumably the case when the XPs are complements and the adjectives are not classifying – see ex. (25a) –, or the first adjective is thematic and has a higher thematic role than the XP (see (25d)). Moreover, it involves an unlikely type of pied-piping, with larger constituents being pied-piped lower (the FP [politico personaggio]), and just the NP in a higher position of the tree (step (c) above). Another analysis proposed in the antisymmetric framework is Pereltsvaig’s (2006). She assumes that adjectives between N and the complements are heads, and she derives the LB-order by successive head-movement of the N and the complex heads N+A1 . . , left-adjoining to each higher head. Such an analysis is impossible for many reasons: adjectives in this position can be coordinated and modified (33a-b), they can even receive complements (ex. (33c)), to which we may add the usual arguments against analyzing adjectives as heads of the NP: the projections we talk about (introduced by determiners, numerals etc.) require the presence of a noun, while adjectives may appear without an NP complement in other environments – in predicative position, and when they introduce a complement of their own – and cannot be analyzed as functional heads of a nominal extended projection since they form an open class, with a wider distribution. (33)

a.

Atacul (aerian) nea¸steptat s¸i impardonabil al attack-the air(adj) unexpected and unforgivable art asupra Americii japonezilor Japanese-the.G on America (Romanian)

b.

l’arriv´ee massive tr`es probable d’immigrants en the arrival massive very likely of immigrants into France France (French)

c.

p˘art¸ile pline de gunoaie ale ora¸sului parts-the full of garbage art town-the.G

(Romanian)

294

Ion Giurgea

2.7.

Current variants of the symmetric system don’t offer a satisfactory solution

What I have called before the ‘symmetric system’ means, largely speaking, using parameters for linearization. Most linearization proposals belonging to this class which have been put forward in recent years have in common the fact that order is established between each member of a sister pair of a tree, under the assumption of binary branching. These proposals differ only when it comes to the application of the parameters – for instance, Takano (2003) proposes that specifiers are always on the left, so that they are exempt for parametrization; the same result is obtained using Kremers’s (2003) system. Richards (2004) applies parametrization only to the relation head-complement.16 Abels and Neeleman (2006) have a slightly different view: all orders are possible, but movement is restricted to the leftward direction. Under all these proposals, the facts covered by generalizations (I)–(III) cannot be explained without resorting to movement, and, moreover, to unmotivated types of movement. Assuming that postnominal adjectives are successively merged to the right, as we proposed in the previous section, there are several ways to derive the position of complements, and their relative order: (a) movement to a checking position, starting with the higher argument and continuing with the others; this kind of movement is likely for the genitive, but less likely for other arguments, which are generally PPs with ordinary prepositions; (b) movement of the N out of the projection hosting arguments, then movement of the phrase [Arg2 tN Arg1 ], as in the previous proposal, following merger of the adjectives. For both (a) and (b) one can suppose either direct movement to the right, or movement to the left followed by remnant movement. The problem is that such movements are unmotivated. In principle, movement is motivated either by morphological reasons (typically for head-movement) or by interpretive reasons (the standard cases are information structure reasons and scope reasons). Among morphological reasons we may perhaps include movement to a specifier position to signal an (Agree) relation for which there is not enough morphological evidence, which would motivate the tendency towards fixed order in languages with poor inflection (see the position of the subject in English compared to most Romance languages). But the movements which we should use in order to derive the generalizations (I)–(III) belong to none of these categories. There is, however, a type of movement which comes closer to what we are looking for: stylistic movements, especially Heavy XP-shift. The crucial observation is that heavy APs are acceptable after complements:

Adjective placement and linearization

(34)

a lui art gen *(foarte) very

295

a.

Am citit o traducere have.1 read a translation poet extraordinar˘a / poet extraordinary(fem.) /

Faust de c˘atre un Faust by a frumoas˘a beautiful(fem.) (Romanian)

b.

jum˘atatea terenului plin˘a ?(de suporteri englezi) half-the field-the.G full of supporters English

c.

une lettre (tr`es aimable) de Paul a` son directeur a letter very kind/agreeable of Paul to his manager (tr`es aimable) very kind (French, Abeill´e and Godard 1999)

d. *une lettre de Paul a` son directeur a letter of Paul to his manager aimable (where aimable modifies lettre) kind e.

ha-psalim ha-yafim sˇel Matis ha-yoter govhim the-sculptures the-beautiful of Matisse the-more tall mi-David sˇel Mikelandzelo from-David of Michelangelo (Hebrew, Shlonsky 2006) ‘the beautiful sculptures of Matisse taller than Michelangelo’s David’

So one might try to explain generalization (II) by heavy XP-shift of complements past light adjectives. However, this is problematic, because, as already noticed (see Willis 2006), heavy XP-shift is optional, while the order N-Arg-A is impossible with light APs (see (35)), and, as we have seen (ex. (33), quite acceptable with complex APs (coordinated, modified or even with complements). (35)

a. *atacarea Albaniei brutal˘a attacking-the Albania.G brutal b. *la lib´eration du Congo r´ecente the liberation of Congo recent(fem.)

(Romanian) (French)

Moreover, generalization (III) is unexpected: why should lower complements be moved past higher ones? In fact, generalization (III) appears in the verbal domain too, so it is in need of a more general explanation. I will propose in the next section a new variant of the ‘symmetric system’, changing some of its current assumptions, which will offer a simpler formulation of generalizations (I)–(III) and also a way to capture the importance of weight for linearization.

296

Ion Giurgea

3. A linearization-based approach to the problem 3.1. A linearization system directly encoding the generalizations The system I propose retains some standard assumptions about linearization: (36)

Spell-Out is cyclic (Chomsky 2001)

(37)

Linearization applies at PF (Chomsky 1995, Embick and Noyer 2001, Bobaljik 2002)

(38)

(PF-)linearization can be represented using linearization instructions (as in Richards 2004, Bobaljik 2002, Fox and Pesetsky 2004, Embick and Noyer 2001)

To these, I will add the following novel assumption: (39)

Linearizable units (i.e. terms to which linearization instructions apply) are heads and maximal XPs. Inside the projection of one head (a maximal XP), linearization instructions apply to pairs of the type headargument, head-modifier, argument-argument, argument-modifier and not to intermediate projections of the head

Now, representing linearization instructions using the function Prec, defined as Prec(X,Y) = 1 if X precedes Y and 0 if Y precedes X, I will tentatively formulate four universal linearization principles, from which the observed word order generalizations can easily be derived. I will use the notations H for the head and ‘>‘ for asymmetric c-command: (P1)

The head parameter establishes a value for Prec(H,X), where X is an internal argument, and for Prec(H,M), where M is a modifier, with possible further specifications for different types of modifiers – light vs. heavy. It may also specify a value for X a specifier, or, if specifiers are universally on the left, it is always set Prec(X,H) for X a specifier.

(P2)

The universal co-arguments rule: For X,Y arguments, if X > Y then Prec(X,Y) = 1 as the unmarked order.17

(P3)

The light modifier parameter (here, for head-initial structures): For X argument and M light modifier, if Prec(H,X) = 1 and Prec(H,M) = 1, then Prec(M,X) = 1

(P4)

Elsewhere rule: For A, B non-heads such that A > B, if Prec(A,H) = Prec(B,H) then Prec(A,B) = Prec(A,H). (the mirror-principle: if both

Adjective placement and linearization

297

precede the head, the higher one precedes the lower one; if both follow the head, the higher one follows the lower one) By defining precedence as transitive, we can leave some Prec relations unspecified: for instance, if Prec(H,M) & Prec(H,X), no specification for Prec(M,X) will be necessary: Prec(M,H) & Prec(H,X) → Prec(M,X). Before discussing this system in detail, I would like to address the first major objection which comes to mind when observing these principles together with (39): it seems that constituency and hierarchical relations are set apart too much from linear order, which may create a problem of interface legibility. However, if we consider things in greater detail, we can see that everything important for interpretation is readily available. First, we are only dealing with the projection of one head (see (39)) – I consider only lexical heads, but this way of formulating constraints could also apply to functional projections, if it proves useful. So, information about the level of attachment of phrases in structures with several functional layers will always be available, in the traditional fashion. Then, inside this domain (the projection of one head), scope relations that are important for interpretation are also readable: in fact, they only concern relations among modifiers, and those relations are only governed by the elsewhere rule, which directly maps c-command into linear order (LB- or RB- according to the head parameter). There is not much sense in speaking of different scope possibilities of a modifier with regard to an argument inside the lexical projection. Argument scope is of two kinds: one is quantificational scope, which, as is well known, is not confined to the lexical projection, but involves overt or covert movement to peripheral positions, so we do not expect it to be directly legible from the order inside the lexical projection. The other kind of scope relation is manifested in the binding relations between coarguments or arguments and modifiers. The principles which guide these relations will be discussed in a separate section (see section 4), where the well-founded-ness of (P2) and of generalization (III), which it encodes, will be questioned. Apart from this, it is not the hierarchical position they occupy inside the lexical projection that distinguishes among coarguments: either their theta roles are marked by morphology (prepositions or inherent case), or they bear an abstract case which is licensed in a functional layer (either in situ by Agree or by movement). The main difference between this system and the linearization procedures in Bobaljik (2002), Kremers (2003) and Fox and Pesetsky (2004) is that precedence is not recursively established between sister nodes: as stated in (39), intermediate projections are excepted. On the other hand, (39) presupposes smaller cycles inside the large linearization cycle which is the phase, every such cycle defined by a maximal projection. For instance, in a phase [HP . . H [LP [YP [L XP]]],

298

Ion Giurgea

the procedure specifies a precedence relation between H and LP rather than separately between H and L, H and YP and H and XP. One may think of a more radical version, where the linearized terms would only be heads and already built phases – everything which is not decomposable at a given phase. (This would be a cyclic-Spell-Out equivalent of Kayne’s (1994) formalization in which precedence is established for every pair of terminal nodes.) For the time being I do not see any reason for enlarging that much the number of pairs to which linearization instructions refer. Note that by directly referring to nodes that are not sisters, these instructions may produce a loss of hierarchical information; (39) guarantees that this imperfect match between structure and order is limited to XPs immediately dominated by projections of the same head, where, as shown in the previous paragraph, arguably the mismatch generated by principles (P2) and (P3) is able to preserve all the important structural information. Let’s illustrate now how these principles derive the order H–M1 –M2 –C2 –C1 in (1) (where M stands for modifier, C for complement and the numbers indicate the order of merge): (i) Supposing that modifiers are merged after complements, we start from the hierarchical structure [[[[H C1] C2 ] M1 ] M2 ]. (ii) (P1) establishes that all Cs and Ms follow H (Prec(H,C1), Prec(H,C2), Prec(H,M1 ), Prec(H,M2 )). (iii) (P2) establishes the relation Prec(C2,C1 ). (iv) Since M1 and M2 are light modifiers, according to (P3) they precede complements (Prec(M1,C1), Prec(M1,C2 ), etc.). (v) The elsewhere rule (P4) gives the relation between M1 and M2 , which reflects c-command (Prec(M1,M2 ) given that Prec(H,M1 ) and Prec(H,M2 )). We obtain thus the final order H–M1 –M2 –C2 –C1 , which is the surface order.18 Note that (P1) and (P4) together give the same result as a system with perfect match, where precedence is directly established between sister nodes (as in Kremers (2003), Bobaljik (2002)). (P4) may thus be considered a UG-rule, and (P1) a UG parameter. It is (P3) that is responsible for the placement of light modifiers with respect to other constituents. As will be shown in the following sub-section, the rule of light modifiers in a head-initial phrase may either put them after the head but before complements or even put them before the head. We may thus replace (P3) by a rule referring to the placement of light modifiers (see (P3)’ below), and consider (P1) an elsewhere rule. (P3) has the effect of placing light constituents close to the head. I do not know whether any effects related to weight can be found in head-final phrases, so that I cannot tell whether a rule like (P3) may be formulated for a head-final setting of (P1) and what its form would be. We may conclude that the existence and the form of (P3) is

Adjective placement and linearization

299

another linearization parameter. Regarding (P1), if it is true that specifiers are universally on the left (as proposed by Takano 2003, Kremers 2003, M. Richards 2004, Johnson 2007, see above), the parametrization will be limited to phrases which are not specifiers; as mentioned in note 17, if we use the rule (P2) for the ordering of several arguments inside a lexical projections, we will have to limit the term ‘specifier’ to non-complement phrases selected by a functional head. Regarding (P2), we should note that its effects can be derived without resorting to linearization, if we make the following assumptions: (i) in order to combine with more than one argument a head projects shells (either in the form H[HP] as in Larson’s (1988) original proposal, or via light heads); (ii) higher arguments are linearized to the left, by the universal specifier rule; (iii) the head further moves to a position above the shells (let’s say, n for NPs) and modifiers are attached to that projection. Note that if modifiers are allowed to attach lower, we would derive orders such as C2 –H–M1 –M2 –C1 or H–C2 –M1 –M2 –C1 , which contravene generalization (II). For the time being I do not know whether this syntactic explanation or the linearization one is to be preferred.19 Note however that if we eliminate (P2) and introduce the aforementioned assumptions, the effects of the other principles may be described in a simpler way, using a (stylistic) rearrangement rule: direct mapping (as provided by (P1) and (P4) or by a cyclic procedure which applies to any sister nodes) derives the order H–C2 –C1–M1 – M2 . Then the linear segment containing light modifiers (M1 –M2 ) is displaced in the immediate proximity of the head so as to satisfy (P3), resulting in the order H–M1 –M2 –C2–C1 . For further issues related to generalization (III), see section 4 below.

3.2.

Generalization (II) and weight

One of the main reasons for giving generalization (I) a PF-treatment, using the linearization procedure proposed in the previous section, rather than explaining it by syntactic movement, is the fact that this generalization makes crucial reference to weight, and weight is not a purely syntactic, abstract notion, although it is not purely phonological either. The importance of weight for linearization is pervasive across categories. Here are some examples from the verbal domain: (40)

a.

Jean a r´eussi a` (tr`es bien) lire ce texte (tr`es J. has succeeded to very well read this text very bien) (French, Abeill´e and Godard 1999) well

300

Ion Giurgea

b.

Jean J. texte text

a r´eussi a` (*incroyablement bien) lire ce has succeeded to incredibly well read this (incroyablement bien) incredibly well

c.

Jean a r´eussi a` (bien) lire ( ?? bien) ce texte J. has succeeded to well read well this text ( ?? bien). well

The facts in (40) show that, as is well known, weight is not a purely syntactic notion: it involves internal complexity of constituents but also pure phonological length, and these two properties do not interfere with the external combinatorial properties of phrases (except for the word order phenomena we are discussing now). At the same time, weight is not a purely phonological notion. A 2 or 3-syllable adjective or adverb is lighter than a 2 syllable PP. So internal complexity too is important. (41)

a. b.

He has carefully put everything in order (manner interpretation possible for carefully) He has (*with care) put everything in order (with care)

Conjunctions are always heavy (even conjunctions of light modifiers): (42)

a.

Jean lira (bien) ce texte (?? bien) J. readfut well this text well (French, Abeill´e and Godard 1999)

b.

Jean lira ce texte bien ou mal J. read-fut this text well or badly

Frequency could also play a role, if the prenominal position of some very common French quality adjectives is also a weight phenomenon, see 2.4. above (grand, petit, beau, gros, jeune, long). But weight can be purely phonological sometimes: (43)

a.

Une vraiment belle jeune fille / une jeune fille a really beautiful young girl a young girl vraiment belle really beautiful (Abeill´e and Godard 1999)

Adjective placement and linearization

301

b. *Une v´eritablement belle jeune fille / une jeune fille a truly/verily beautiful young girl a young girl v´eritablement belle truly/verily beautiful All these properties make weight a good candidate for the PF-module, the only place where we can find both structural and (morpho-)phonological properties, as shown by work in the field of Distributed Morphology (see Embick and Noyer 2001 for the preservation of some structural properties in earlier stages of the PF-derivation). We have already seen, in (34), that heavy APs, which are allowed after complements, are generally structurally heavy but sometimes even purely phonologically heavy. On the other hand, structurally light prepositional modifiers may precede adjectives, which means that they are treated as light modifiers. A light prepositional modifier can be a P+NP-phrase, as opposed to a P+DP-phrase: (44)

a.

Une robe de soir superbe a dress of evening superb ‘a splendid evening dress’

(French)

b. ??Une robe de Marie superbe a dress of Marie superb It is worth noting that weight constraints on the pre-head field seem to be stronger. For instance, it is reasonable to assume that the ban on non-head-final prehead modifiers (Emonds 1976: The Head-Final-Filter) in head-initial phrases (Haider’s ‘edge effect’) is also a weight phenomenon ((45)c vs. d shows that this constraint depends on the head parameter of the projection: NPs, unlike VPs, are head-initial in German with respect to the relation head-complement; see Haider 2004) : (45)

a.

He has [(much more) carefully (*than anyone else)] analyzed it. (Haider 2004)

b.

He has [(much less) often (*than I (thought))] rehearsed it.

c.

Er hat es [sehr viel sorf¨altiger als jeder andere] he has it very much more.carefully than any other analysiert (German) analyzed ‘He has analyzed it much more carefully than anyone else.’

d.

eine [viel gr¨oßere (*als ich dachte)] Summe a much bigger (*than I thought) sum

302

Ion Giurgea

But in N-A languages there is no similar restriction on theAPs occurring between the noun and the complements or PP modifiers: (46)

a.

partea plin˘a de flori a gr˘adinii part-the full of flowers art garden-the.G b. *the full of flowers side of the garden

(Romanian)

Coordination, although possible in prenominal position, is more often found after the noun. For instance, (47)a is good in Romanian even if the towers of the castle are both high and red, while in English the conjunction is understood as implying a coordination of objects: (47)

a.

turnurile ˆınalte s¸i ro¸sii ale castelului towers-the high and red art castle-the.G (OK if the towers are both high and red)

b.

the castle’s high (??and) red towers (‘??’ if those towers are both high and red)

(Romanian)

Since the ‘edge effect’ only holds for head-initial phrases (see ex. (45)), it seems that our principles have to be slightly reformulated: P1 would be a general principle stating the head-initial-ness or head-final-ness of a domain, then P3 would apply to light modifiers relative to P1: (P1)

Default head parameter establishes a value for Prec(H,X) where X is any dependent of H20

(P3)

The light modifier parameter (general): For M a light modifier and X any other type of dependent, if Prec(H,X) = 1, either – Prec(H,M) = 1 and Prec(M,X) = 1, and M is ‘minimally light’, or – Prec(H,M) = 0 (→ Prec(M,X) = 1), and M must be ‘strongly light’ – must end in a lexical head

This new version of the light modifier parameter could perhaps apply to French quality adjectives: assuming that very frequent quality adjectives may be treated as lighter than ordinary adjectives, we can say that French uses both possibilities of (P3) for the nominal projection, including in the class of ‘strongly light modifiers’ some lexically specified adjectives (those where lightness is linked to frequency).21

Adjective placement and linearization

303

4. Problems raised by generalization (III): Binding relations among co-arguments and Pesetsky’s paradox In this section I will scrutinize the well-founded-ness of generalization (III). If among several internal arguments binding relations are used as a criterion for hierarchical order, it appears to be a cross-linguistic generalization that, at least in the unmarked order, higher arguments precede lower ones even in head-initial domains (see Larson 1988, 1990). In fact, we have to distinguish two types of phenomena: (i) Binding of an argument by another one – ‘reflexivity’. This can be governed by an ‘obliqueness hierarchy’, independently of order (see B¨uring 2005). (ii) Binding or co-reference relations involving an argument (or modifier) and a phrase embedded in another argument(/modifier), together with NPI licensing: here linear order seems to be the relevant factor. Binding relations which are independent of order and governed only by obliqueness can be seen in the following Romanian data: (48)

a.

readucerea sufletului la sine ˆınsu¸si bringing-back-the soul-the.G to it self

(Romanian)

b. *readucerea sai la sufleti bringing-back-the its to soul c.

(49)

readucerea la sine ˆınsu¸si a sufletului bringing-back-the to it self art soul-the.G

d. *readucerea la suflet a sa ˆınsu¸si bringing-back-the to soul art its self ˆ a¸ta˘ tura lui readuce a. Inv˘ (la sine) sufletul (la sine) teaching-the his brings-back to itself soul-the to itself ˆ a¸ta˘ tura lui ˆıl b. *Inv˘ readuce (pe sine ˆınsu¸si) la teaching-the his cl.ac brings-back obj it self to suflet (pe sine ˆınsu¸si) soul obj it self

For obliqueness, B¨uring (2005) has proposed the following hierarchy), based on crosslinguistic data: (50)

‘subject’ > ‘indirect object’ > ‘direct object’ > ‘prepositional object’ (B¨uring 2005)22

304

Ion Giurgea

Type (ii) is illustrated by the following data from Pesetsky 1995 (largely reproducing Larson 1989): (51)

a. *We gave to himi on Friday [John’si brand-new toy] a.’ We gave John’si brand-new toy to himi on Friday b. We gave to themi at the interviews [copies of reports on each other i ] b.’ *We gave copies of reports on each other i to all of themi . c. Bill heard from each committee member i on Friday [a report on hisi activities] c.’ (?)* Bill heard a report on hisi activities from each committee member d. Tom threw to none of these people on Tuesday [ any set of keys that had “Do Not Copy” stamped on them] d’. *Tom threw any set of keys to none of these people e. Rosa spoke about each defendant i [to hisi lawyer] e’. *Rosa spoke to hisi lawyer about each defendant i

In some cases, obliqueness and precedence seem to apply together: (52)

a. *John talked about the meni to each other i /to themi b. *John talked about each other to the men c. John talked to the men about each other

(53)

a. *I spoke about Rosai with herself i /her i b. *I spoke about herselfi /heri with Rosai c. I spoke to Rosa about herself

However, it is crucial to notice that the precedence condition applies not only to complements, but also to post-complement adjuncts (as noticed by Larson 1990 and Stroik 1990), and these adjuncts display LB-order properties with regard to standard constituency tests such as movement and ellipsis, as well as with regard to scope relations (as noticed by Jackendoff 1990, Ernst 1994, Andrews 1982). This paradox was extensively treated by Pesetsky (1995), who proposed a special theory of syntax to account for it (that’s why it is also known as Pesetsky’s paradox). I reproduce here some relevant examples from Pesetsky (1995): (54)

a. Sue showed the kids to the adults on each other’s birthday b. Sue sent her money to none of these officials on any of these days c. *We gave to him at [that time]i [a copy of the report we’ve finished on Friday]i

(55)

a. *[To John about herself] Mary spoke

Adjective placement and linearization

305

b. *[In the garden on Tuesday] Bill said he gave the book to Sue c. John said he would give the book to them in the garden on each other’s birthdays, and [give the book to them in the garden] he did on each other’s birthdays and [give the book to them] he did in the garden on each other’s birthdays * and [give on each other’s birthday] he did the book to them in the garden * and [give to them in the garden on each other’s birthdays] he did the book (56)

a. b. c. d.

[John plays quartets in foreign countries] on weekends (Andrews 1982) [John plays quartets on weekends] in foreign countries [Henry sold his car reluctantly] for a thousand bucks. [Henry sold his car for a thousand bucks] reluctantly.

Ernst (1994) notes that (57)b has only a reading many times>willingly, while (57)a is ambiguous, having both a reading where willingly takes scope over many times and one in which many times scopes above willingly. The explanation he gives is that many times can undergo QR, being a quantified expression, while adverbs do not move at LF. Therefore these data constitute evidence for a LBstructure: (57)

a. b.

She kissed him many times willingly. She kissed him willingly many times. (Ernst 1994: only many times>willingly)

There are cases where different criteria give opposite results for the same sentence: licensing of parasitic gaps by heavy XP-shift vs. anaphora in (58) (from Pesetsky 1995), variable binding vs. adverbial scope in (59) (from Phillips 1995): (58)

a.

Sue offended by [neglecting to bow]i [her favorite uncle who’d also [done so]i b. *Sue offended by also [doing so]i [her favorite uncle who’d [neglected to bow]i

(59)

a. b.

Kremer plays quartets in foreign countriesi on their i national holidays. Kremer plays quartets on new federal holidaysi in their i first 5 years of existence.

306

Ion Giurgea

Based on binding data, Larson proposed that post-complement modifiers are generated lower than complements. However, it is highly implausible, under a derivation-based system, that space and time modifiers, which are in principle independent of the lexical meaning of the V, and are independently predicated on the event, should be merged lower than arguments, which often define the type of the event (the meaning of the verb is often dependent on that of the arguments it takes), and satisfy selectional properties of the V. One can conceive, perhaps, that adjunction may apply anywhere in the lexical domain, before or after merging of arguments, but there is no reason why it should not also be able to apply after the merging of arguments. A-movement of arguments above adjuncts (proposed by Lechner 2003) is unlikely for PP-arguments, because they do not need formal licensing. But PP-arguments also normally precede (and bind into) adjuncts: (60)

?(?) He thinks for hours in the garden about his girlfriend

Phillips (2003) holds that the LB-scope of adverbials is due to the focal stress on the final constituent, which creates a Topic-Comment structure. He claims that when another Adv is added, the RB-reading is much easier to obtain: (61)

a. b.

Sue kissed him willingly many times in front of the boss. Kremer plays concertos in foreign countries on weekends at the height of the season.

However, first of all, the LB-reading is still available in these examples. More importantly, the information-structure-based explanation is implausible for cases where the entire clause is not in the nuclear-scope domain: (62)

Anyone who kisses willingly many times should not be trusted

Given all these data, it becomes unclear whether binding relations can give criteria for hierarchical order in a post-head field. It is undisputable that semantic binding requires c-command at LF23 , but the LF-structure is probably quite different from the surface structure anyway, due to massive application of QR. For instance, the possibility of a sloppy reading in (63) shows that referential expressions too must be QR-ed: if Mary does not scope above [every one of her relatives] at LF, the pronoun her cannot get a bound reading, but at most a co-referent one. But without a bound reading the sloppy interpretation in the second sentence is impossible. So, if we assume that [every one of her relatives] is obligatorily QR-ed, it follows that the referential expression Mary must be QR-ed too. (63)

Mary invited every one of her relatives, and John did too.

Adjective placement and linearization

307

Then, the representation for (51)a’ will be like (64)

[[we]] λx [ John λy [ x gave to y the new toy of y on Friday]]

We see that the c-command required for semantic binding (at LF) does not apply internally to the VP: there, we only have occurrences of the bound variable y, among which c-command relations have no relevance. We have to distinguish between indexation rules and interpretation as bound pronouns (semantic binding – see B¨uring 2005). The obliqueness and the precedence conditions on bound pronouns illustrated before can be considered simply as conditions on indexation. Inside some domain, which can be the VP, or the vP-phase, these conditions can be stated as follows: (65)

Given x and y co-indexed (inside some domain), (i) if both are arguments, if x is more oblique than y then x must be a pronoun/reflexive (ii) otherwise, if x follows y in linear order than x must be a pronoun/reflexive

The bound variable interpretation of co-indexed pronouns is contingent upon QR, and is presumably optional (as shown by the strict readings of VP-ellipsis). This solution to Pesetsky’s paradox is similar to Ernst’s (1994), who proposed that binding and NPI-licensing involve precedence plus m-command, rather than c-command. Barss and Lasnik (1986) and Jackendoff (1990) also included precedence under the conditions, but they added it to c-command, considering that VPs with more than one argument are flat structures, something which is unlikely given the scope phenomena and constituency tests illustrated above. A similar analysis might be used for NPI-licensing, since negation, like quantifiers, scopes above the VP at LF, where no presumably translates as no . . . any – so (54b) would receive one the following representations (I will not choose between them here): (66) ¬ (Sue gave her money to ANY of these officials on ANY of these days) → (i) ¬ ∃x (x∈[[these officials]]) (∃y (y∈[[these days]]) (Sue gave her money to x on y)) (ii) ∀x (x∈[[these officials]]) (∀y (y∈[[these days]]) (¬(Sue gave her money to x on y)))) Here too we may distinguish between semantic NPI-licensing, which, just like binding, obtains at LF and requires c-command, and marking of a dependency

308

Ion Giurgea

between an n-word and other NPIs licensed by the negation this word introduces. The facts in (67) provide evidence for this proposal: if c-command at LF had been all that was required for NPI-licensing, reconstruction would have to rescue (67b), just as it does for (67c), for instance (after reconstruction the NPI would appear in the same configuration as in (67a), where it is licensed). The fact that this is not the case suggests that linear order is also involved in NPI-licensing: (67)

a. I don’t know his opinion about anything b. *Whose opinion about anything don’t you know? c. Which opinion about himselfi does Fredi like most?

The main problem of an analysis relying on precedence is that co-indexation, which is a procedure expected to apply on the LF-branch, must make reference to linearized structures, which, on our hypothesis, are obtained in the PF-derivation. We can think of a possible answer, based on the notion of cyclicity: if access to the semantic module is cyclic, and its cycles correspond to those of the PF module, as proposed by Chomsky (2001, 2005), one can assume that the result of applying the linearization procedure on a phase-domain (a Spell-Out domain) may also be accessible to a semantic module, in this case the module dealing with co-indexing and NPI-licensing. The solution Pesetsky gives to his paradox requires more profound modifications of standard assumptions about syntax and the interfaces: he assumes that each sentence has two structures built in parallel, a strictly RB one (the ‘cascade structure’), which is used for binding and NPI-licensing and is binarybranching, and a ‘layered structure’, which allows LB and flat trees, and is used for movement, ellipsis and scopal relations between adverbials. In my analysis, there are no parallel structures, but the structures are derivationally linked, the linearized one deriving from the hierarchical one by the linearization procedure. The linearized structure is accessed by an indexation procedure (perhaps the sort of dependency between a NPI and an n-word which is sensitive to linear order also involves some kind of indexation), and all compositional semantics is reserved for the single hierarchical structure.24 Now, if we accept this analysis, the status of our generalization (III) and of the principle (P2) which encoded it is less sure, as binding relations are no longer an unconditioned criterion for c-command. However, we have seen that for co-arguments obliqueness is also involved in binding relations, and sometimes overrides precedence completely (see (48)–(49)). Assuming that obliqueness is paired with hierarchical order, with less oblique arguments being higher than more oblique ones, we can still retain (P2). However, if we look at (52), we see that sometimes precedence and obliqueness must both be satisfied. A possible

Adjective placement and linearization

309

explanation for the exception to precedence manifested in (48) is that the higher argument in final position (the genitive in (48c)) has moved outside the domain to which the precedence condition applies. A similar result I found examining the verbal domain in Romanian: (i) precedence must be checked in any event, the only apparent exceptions to precedence being in fact right-dislocated topics (which presumably are outside the VPdomain), which must be arguments and must be less oblique than the anaphors (see (68)), and (ii) obliqueness is added as an extra condition for argument binding only25 (and not for binding inside an argument) (compare (69) to (70)): (68)

a.

Ion a vorbit despre ea ˆıns˘as¸ii cu Mariai Ion has spoken about her self with Maria (possible with deaccented cu Maria)

b. ??Ion a vorbit cu ea ˆıns˘as¸ii despre Mariai Ion has spoken with her self about Maria (even with deaccented despre Maria) c. *Ion a vorbit despre eai (ˆıns˘as¸ii ) cu p˘arint¸ii Ion has spoken about her self with parents-the Marieii Maria.G (even with deaccented despre p˘arint¸ii M.) (69)

a. ??Profesoara a vorbit despre fiecare elevi teacher-the(fem) has spoken about every pupil(masc) cu el ˆınsu¸sii 26 with him self b. ??Detectivul i la descris pe fiecare detective-the cl.d cl.ac. has described obj every invitati lui ˆınsu¸sii guest him.D self c. *Mediumul i-a descris personalitatea fiec˘arui medium-the cl.d has described personality-the every.G vizitatori lui ˆınsu¸sii guest him.D self

(70)

a.

Profesoara a vorbit despre fiecare copili cu teacher-the(fem) has spoken about every child with p˘arint¸ii luii parents-the his

310

Ion Giurgea

b.

Profesoara l-a descris pe fiecare copili teacher-the(fem) cl.ac.has described obj every child p˘arint¸ilor luii parents-the.D his

The association between obliqueness and c-command, which gave us the formulation in (P2), should be retained in cases where it is supported by independent evidence such as superiority effects and weak cross-over. It is well known that in English these phenomena indicate that the indirect object is higher than the direct object in the double object construction (see Postal 1971, Wasow 1972, Chomsky 1973, Barss and Lasnik 1986) (see (71)–(72)). Even if in this case the structure may be more complex, given the fact that the objects are not introduced by prepositions, so that they are likely to be licensed by some functional structure, it is worth noticing that similar contrasts appear between PP arguments too, which I take to occupy the lexical domain, as shown in (73). (71)

a.

Whoi did you show hisi reflection in the mirror? (Postal 1971, Wasow 1972) b. *Which moviei did you show itsi fans?

(72)

a. Who did you give which book? (Chomsky 1973) b. *Which book did you give who?

(73)

Cu cinei ai vorbit despre p˘arint¸ii luii ? with who have.2sg spoken about parents-the his (Romanian) b. ??Despre cinei ai vorbit cu p˘arint¸ii luii ? about who have.2sg spoken with parents-the his a.

In sum, we may retain the traditional association between ‘less oblique’ and ‘higher (c-commanding)’ based on other arguments than binding facts. In this case, (P2) may still be used to cover generalization (II) (or, as an alternative analysis, Larson’s shell structure and its successors).

5. Conclusions Languages where the normal position of attributive adjectives is postnominal typically show (i) left-branching relative order of stacked postnominal adjectives, (ii) placement of the noun’s complements after the adjectives, with the exception of heavy APs and (iii) normal right-branching order among several complements of the noun. Together, these three word-order generalizations con-

Adjective placement and linearization

311

stitute a problem for both the antisymmetry-based system developed in Cinque’s cartographic program and its current symmetric competitors. I showed that the order of stacked postnominal adjectives is best described by analyzing them as adjuncts linearized to the right. This solution is preferable on several grounds: (i) it is more economical insofar as it dispenses us from a number of ad-hoc projections (AgrPs) and movement operations; (ii) with respect to the specific implementation of the antisymmetric program in the cartographic theory, it allows a more flexible hierarchical ordering of adjectives, which is sustained by various data showing that hierarchical relations may vary according to the context and are fixed only for general semantic types, rather than for cognitive classes; (iii) it explains why classifying and thematic adjectives are always linearized to the right and show left-branching order in languages with a basic N-A order: given the fact that their meaning is heavily dependent on the lexical meaning of the noun, they are expected to be adjuncts inside the lexical projection. A consequence of the analysis of postnominal adjectives as adjuncts linearized to the right has been that prenominal adjectives in languages with a basic N-A order are specifiers. This proposal is sustained by the special conditions prenominal adjectives must fulfil, which may be described as features checked in the functional domain of the noun phrase. The order of complements with respect to adjectives was derived using a linearization mechanism in which linearization principles apply to pairs of nodes immediately dominated by projections of the same head, and not to intermediate projections of the head. The system contains two parameters, the headparameter, which establishes the order between the head and its complements and modifiers, and the light-modifier parameter, which establishes the position of light modifiers in the immediate proximity of the head, setting the order of the pairs (Modifier, XP) as Modifier Physical Defect > Color > Origin > Composition > Destination, and Sproat and Shi (1988), who tried to derive such hierarchies from a more abstract principle: the more intrinsic and objective the property is, and the less evaluative it is, the closer the adjective is to the noun. Classifying adjectives are subsective: I can say about the atomic theory that it is a theory, but not that it is (something) atomic. By contrast, a red ball is both a ball and (something) red. The meaning of a classifying adjective such as atomic varies depending on the noun it combines with: atomic power/level/bomb. The key to these various meanings is the noun, and that’s why it is said that classifying adjectives, just like thematic adjectives, act as arguments or as modifiers of implicit arguments of the noun: atomic bomb is a bomb using the energy freed in transformations of atoms, atomic theory is a theory whose content is distinguished by the use of the notion of atoms, etc. If, as can be inferred from what I said before (see note 7), classifying adjectives are defined only for specific noun meanings, it follows that classifying adjectives will not be defined for conjunctions of properties of the type λx(N(x) ∧ A(x)) formed by predicate intersection of a noun and a quality adjective. On the other hand, the

Adjective placement and linearization

9.

313

result of applying a classifying adjective to a noun is also a noun meaning, and not a complex property, so that we expect to find stacked classifying adjectives – and they are indeed found, see (15c–d), (21). Another important semantic distinction is the one between restrictive and nonrestrictive adjectives. Non-restrictive adjectives are interpreted very high, being speaker comments about the entity denoted by the DP (therefore they must be interpreted outside the DP, and the DP must be referential). So we would expect the hierarchy to be restrictive relation of subscripts corresponds to height within the hierarchy, as shown in the tree in figure 5. Cartographic hierarchies define templates of bounded size. To generate unbounded structures, (pieces of) these templates are embedded in each other recursively. I call every token of (a maximal fragment of) a cartographic hierarchy a cyclic domain. It should be clear now that within any given cyclic domain c-command relations between the functional heads and > yield the same ordering. However, across cyclic domains an item which is ordered low in one cyclic domain may ccommand an element high in an embedded cyclic domain (matrix T c-commands the C head of an object clause, but C>T). There are two possible types of solution to the problem posed by the ordering of operations. We can either use the hierarchy of functional heads outside of the moving elements to derive the order of operations, or we may employ the hierarchy of functional heads within the moving element. I will discuss the two options in the order given. Wherever possible, I frame the discussion in terms of analyses actually found in the literature.

2.1.

Solutions based on external syntax

2.1.1. Landing Site > Launching Site (Sternefeld, 1992) If we assume that the landing site of movement c-commands the launching site, then it follows for movement within a cyclic domain that (the head projecting) the landing site of movement > (the head projecting) the launching site of movement. This means that for movement within a given cyclic domain, the ordering of operations falls out from the fact that movement is upward in the

Some implications of improper movement for cartography

341

HPa Y HPu X . . . [HPb Y. . . ] . . .

HPv X . . . [HP Y . . . ] . . . b

Figure 6. Indexed structure for surfing paths

HPu X . . . [HPb Y . . . ]. . .

HPa Y HPv X . . . [HPb Y . . . ]. . .

Figure 7. Indexed structure for diving paths

tree. Sternefeld (1992) suggested to generalize this to movement across cyclic domains and to demand that movement in general, that is, even across cycles, obey the constraint that (the head projecting) the landing site of movement > (the head projecting) the launching site of movement. This constraint now, of course, no longer follows from the direction of movement. Sternefeld’s condition derives the ban on improper movement for the identity case, but there are problems both with the case of surfing paths, figure 6, and with remnant movement, figure 7. The correct ordering of operations is not derived in the case of surfing paths, because Sternefeld’s condition only guarantees that HPa > HPb and HPu > HPv . From this it does not follow that HPa > HPu .

342

Klaus Abels

Similarly, for diving paths (figure 7) Sternefeld’s condition makes sure that HPa > HPb and HPu > HPv . This again does not guarantee HPa > HPu . Sternefeld’s is a classical theory of improper movement, dealing only with the identity case. While Sternefeld’s condition overgenerates with respect to surfing and diving paths, it also undergenerates elsewhere under certain assumptions. If we assume a version of the raising to object analysis of ECM in English (Postal (1974); Lasnik and Saito (1991); Davies and Dubinsky (2004)), the exceptionally casemarked argument almost certainly lands in a position in the matrix domain which is projected by a head (AgrO) < the head projecting the highest position occupied by that argument in the embedded domain (T). This is shown in (12). (12)

I want [AgrOP John . . . [TP tJohn to leave]]

I now turn to Williams’ (2002) theory, which is considerably stronger than Sternefeld’s. Anticipating the conclusion of the next subsection, we will see that Williams’proposal solves the overgeneration issues facing Sternefeld’s proposal, but the undergeneration issues mutliply.

2.1.2. For all Hn along the path of movement, the head projecting the landing site ≥ Hn (Williams, 2002, LEC) Williams’ (2002) Level Embedding Conjecture (LEC) together with the extension condition derives the following corollary: For all Hn along the path of movement, the head projecting the landing site ≥ Hn . A head H counts as along the path of movement iff some projection of H dominates the launching site but does not dominate the landing site of movement. In other words, every head that m-commands the launching site but does not c-command the landing site of a given movement counts as part of the path. This definition of movement paths ensures that the head projecting the launching site and the one projecting the landing site are both included in the path. This formulation entails Sternefeld’s condition from subsection 2.1.1 (modulo any difference that arises from the difference between > and ≥). Hence, Williams’ condition rules out a proper superset of the structures ruled out by Sternefeld’s. The LEC derives the ban on improper movement for the identity case for the same reason that Sternefeld’s condition does. The LEC also derives the ban on improper movement for the case of surfing paths. To see this consider again the structure in figure 6. The LEC guarantees not only that HPa >HPb and HPu >HPv , it crucially makes sure that HPa >HPu. Under the paraphrase of the LEC in the preceding paragraph this is true technically, because the head H0u m-commands the launching site of Y and thus counts as lying along the path of Y. Finally, the LEC derives the ordering of operations for the case of remnant

Some implications of improper movement for cartography

343

movement. A look at figure 7 reveals that, as before, HPa >HPb and HPu >HPv , but crucially the LEC also demands that HPa > HPu . This is so because under the LEC the head projecting the landing site must be higher in the functional sequence than all of the heads along the path including, in particular H0a . Thus, the overgeneration problems facing Sternefeld’s approach are solved by the LEC. Exceptional case marking in English again spells trouble for this analysis if the raising to object analysis is correct (Postal (1974); Lasnik and Saito (1991); Davies and Dubinsky (2004)). The next example illustrates the problem on the assumption that negation is in the same hierarchical position in the embedded infinitive and in the superordinate clause. Mary crosses negation in the infinitive but lands below it in the superordinate clause. For concreteness assume that Mary ends up in [Spec, AgrO P]. On the way from the θ -position in the embedded VP to matrix [Spec, AgrO P] Mary crosses negation in the embedded clause. Therefore, Agr0O > Neg0 . However, the linear order of superordinate negation and Mary indicates that negation is higher than AgrO , i.e., Neg0 > Agr0O . A similar contradiction arises with respect to the ordering of T0 , hosting to, and Agr0O . There does not seem to be a way of avoiding this contradiction.19 (13)

I expect John not to want Mary not to be told tMary the truth.

The same issue arises for hyperraising (Nevins (2004); Williams (to appear)) if the movement analysis of hyperraising is correct, for the other cases of Amovement out of CP mentioned in McCloskey (2000, p. 71 fn. 20), for quantifier movement across a complementizer in French (noted as a problem in Williams (2002)) and negative quantifiers in Scandinavian, for scrambling in Russian (Glushan, 2006), and for long wh-movement in the Bantu language Kˆıˆıtharaka (see Muriungi (2005); Abels and Muriungi (2008), also Bergvall (1987); Schwarz (2003) for a similar facts from Kikuyu), . . . The problem always has the same shape: an element high in the hierarchy, typically T0 or C0, is crossed on the path of movement but the landing site of the movement is below T0 or C0 . This is quite obvious in the case of quantifier movement in French, (14a-b), where the quantifiers rien and tout might originate as low as the object position of the embedded verb, cross T0 and C0 along the way and end up in a position transparently below T0 . Similarly for the Icelandic examples in (15). These cases are only a problem for Williams’ approach, Sternefeld’s theory sketched above allows them. (14)

French ((Kayne, 1998, p. 141–142 ex. 52–3)) a. Il n’ a rien fallu que je fasse. it neg has nothing been-necessary that I do It hasn’t been necessary for me to do anything.

344

Klaus Abels

b.

(15)

Il a tout fallu que je leur enl`eve. it has everything been-necessary that I themdat remove I had to take everything off them.

Icelandic ((Kayne, 1998, p. 142 ex. 55-6) attributed to Rognvaldsson (1987, p. 44) and Jonsson (1996, p.86)) a. Hann mun ekkert hafa getadh gert. they will nothing have could done They haven’t promised to do anything. b. Their hafa ekkert lofadh adh gera. they have nothing promised to do They haven’t promised to do anytying.

The same point is made again by the Kˆıˆıtharaka examples in (16). Kˆıˆıtharaka is a Bantu language spoken in Kenya. This language allows wh-questions to be formed by leaving the wh-word in situ, (16a), by moving it partially, (16b–d), and by moving it all the way to its scope position, (16e). The wh-word uˆ uˆ ‘what’ is boxed in the examples for easier orientation. There is apparently no semantic difference between these five examples (see Muriungi (2003, 2005); Abels and Muriungi (2006) for discussion), the translation in (16e) is appropriate for all of them. The crucial example in the current context is (16d). The wh-word moves across the underlined complementizer atˆı in the lower clause and ends up right after the circled complementizer atˆı in the next higher clause. This is then another example where an element crosses over a position which is higher in the template than the position where it ends up.20 (16)

Kˆıˆıtharaka (Peter Muriungi, p.c.) a. U- ri- thugania atˆı John a- ug- ir- e atˆı Pat 2sg- ps- think that John sm- say- pf - fv that Pat a- ug- ir- e Lucy a- ring- ir- e uˆ uˆ sm say ps fv Lucy sm- beat- pf - fv who b. U- ri- thugania atˆı John a- ug- ir- e atˆı Pat 2sg- ps- think that John sm- say- pf - fv that Pat a- ug- ir- e n- uˆ uˆ Lucy a- ring- ir- e twh sm say ps fv F- who Lucy sm- beat- pf - fv c. U- ri- thugania atˆı John a- ug- ir- e atˆı n2sg- ps- think that John sm- say- pf - fv that F uˆ uˆ Pat a- ug- ir- e Lucy n- a- ring- ir- e twh who Pat sm say ps fv Lucy F- sm- beat- pf - fv

Some implications of improper movement for cartography

d.

U2sgatˆı that twh

ripsPat Pat

e.

N- uˆ uˆ F who atˆı Pat that Pat twh

thugania atˆı think that n- a- ug- irF sm say ps

nFe fv

uku- thugania 2sg- ps- think n- a- ug- ir- e F sm say ps fv

uˆ uˆ who Lucy Lucy

345

John a- ug- ir- e John sm- say- pf - fv n- a- ring- ir- e F- sm- beat- pf - fv

atˆı John nthat John F Lucy n- aLucy F- sm-

a- ugsm- sayring- irbeat- pf -

ir- e pf - fv e fv

Who do you think that John said that Pat said that Lucy beat? In this section we have seen that Williams’ theory based on the LEC and the extension condition avoids the overgeneration problems that remained unresolved under Sternefeld’s approach, but in doing so Williams’ theory gives rise to quite severe undergeneration problems.

2.1.3. Conclusion: External syntax Both of these potential solutions attempt to state the generalization derived above by (11) using conditions over launching and landing sites and/or paths of movement, that is, material external to the moving item. In both cases the statement makes reference to the hierarchy as an independent object, independent of the actual tokens that form a given syntactic object; reference to the resulting syntactic object with its structural hierarchy (c-command, dominance) alone is insufficient. The independent hierarchy of labels is necessarily invoked in stating the restriction. Both approaches have serious empirical disadvantages not shared by the system adopting GenPIM. Independently of whether Sternefeld’s, Williams’, or the present approach are to be preferred eventually, all of them require the hierarchy to be accessible for the syntactic computation beyond the narrow realm of building trees whose local structure is isomorphic to the hierarchy. All cartographic work that I am familiar with, uses the hierarchy in this very narrow and, as we have now seen, insufficient sense.

346

Klaus Abels

2.2.

Solutions based on internal syntax

The next two subsections discuss the option of deriving the ban on improper movement from the internal structure of the moving element. Unlike the approaches based on the external syntax, possible approaches based on the internal syntax have not been proposed in the literature, though a version of the top-down condition on feature accessibility under the name of ‘peeling’ has been suggested to me as a possible account of the ban on improper movement by Michal Starke in personal communication. Given the lack of a more or less worked out proposal, which would make use of the internal syntax to deal with improper movement, the discussion is necessarily more schematic.

2.2.1. Bottom-up condition on feature accessibility Since at least Abney’s work (1987) it has often been assumed that there is a strong parallelism between the noun phrase/DP and the clause/CP. Such parallelism would suggest that θ -related functional structure occupies the lower stretches of the noun phrase, case is located higher up, and wh- and information structure related features are even more peripheral to the NP. The functional hierarchy within the noun phrase might then be topic > wh > scrambling > Case > θ . Since θ -features have to be satisfied first, followed by case, and then and only then by operator features, one might conjecture that the features of a head Hx become accessible for checking/attraction only after the features of all heads Hy that are c-commanded by Hx (alternatively: that are dominated by a projection of Hx , i.e., that are m-commanded by Hx ) have been checked/attracted. This condition will directly rule out improper movement for fairly simple instances of the identity condition.21 The bottom-up condition on feature accessibility does not derive the ban on improper movement for the case of diving paths. This can be shown by considering example (2cii) above, repeated here for convenience. wh (2cii) *[A picture of twhich king ] is known which king to have been sold A ta picture of which king Structures like this are not ruled out, because first the more deeply embedded feature is checked and the higher feature is checked later in conformity to the bottom-up condition. The problem is that the features reside in different cyclic domains, thus cartography imposes no ordering on them. The original problem that motivated the ban on improper movement, the problem of enforcing an ordering across cycles, comes back here. When the moving constituent is bigger

Some implications of improper movement for cartography

347

than a single cyclic domain, the bottom-up condition on feature accessibility does not guarantee an ordering any more. In addition, the bottom-up condition on feature accessibility categorically rules out surfing paths, (2bi) above. The Case feature in the larger DP is checked first. This (wrongly) makes the more deeply embedded wh-feature of the complement inaccessible. The fact that the bottom-up condition on feature accessibility rules out surfing paths (i.e., derives the freezing principle) might seem relatively harmless even in the face of the apparent counterexamples given above, since every derivation with surfing paths can, in a sufficiently unrestrictive theory of movement (e.g. Koopman and Szabolcsi (2000)), be emulated by a derivation with diving paths. This is illustrated in figure 8. The symbol  means can be emulated by. The idea in figure 8 is to move α out of XP before XP moves for the first time (‘stacking it’ in Koopman and Szabolcsi’s terms), then letting XP undergo remnant movement, and finally moving α past XP. The structures are identical modulo the additional intermediate trace of α, t2α . α



... XP ... tα ...

... tXP

α

... XP

...

... t1α ...

t2α . . . tXP

Figure 8. Emulating surfing through diving

However, this solution inherits the problem that no ordering is imposed on the various steps of remnant movement. As long as the features reside in different cyclic domains A-movement before wh-movement as well as wh-movement before A-movement ought to be allowed. And even if this problem could be overcome somehow, the putative unification of subextraction and remnant movement in figure 8 would miss the generalization from table 1, since the cases of real, surface true, remnant movement would require the opposite ordering of operations from the cases of apparent remnant movement, i.e., subextractions on the surface, suggesting that the unification of subextraction with remnant movement would be a step in the wrong direction.

348

Klaus Abels

2.2.2. Top-down condition on feature accessibility We can ask ourselves whether the serious problems of the bottom-up condition on feature accessibility might have stemmed from having assumed the wrong arrangement of features within the DP. We might then put the hypothesis that there is parallelism between CPs and DPs in terms of feature organization on its head and assume that the featural organization of CPs and DPs are mirror images of each other: DPs have their θ -features outermost, followed by a Caselayer lower down, followed by a layer where wh- and discourse related features reside. Such an organization of the DP could then be coupled by a top-down condition on feature accessibility: Features of a head Hx become accessible for checking/attraction only after the features of all heads Hy that c-command Hx (alternatively: whose projections dominate Hx , i.e., that m-command Hx ) have been checked/attracted.22 On the positive side, the top-down condition on feature accessibility derives the ban on improper movement for simple cases of the identity condition.23 However, the top-down condition on feature accessibility does not derive the ban on improper movement for surfing paths. Recall example (2bii), repeated here for convenience. (2bii)

*Oscar is known [how likely tA Oscar to win] it was wh thow . likely t to win Oscar

Checking is top-down. Since the wh- and the Case-feature reside in different cyclic domains, the cartographic hierarchy is also respected. The top-down condition on feature accessibility thus does not derive the ordering of operations observed with subextraction. Worse still, the top-down condition on feature accessibility disallows remnant movement (diving paths) categorically. To see this consider again example (2ci), repeated below. (2ci) It is known [how likely tA Oscar to win] Oscar is wh thow . likely t to win Oscar

Such structures cannot be derived since the more deeply embedded feature is checked/attracted first creating the remnant. Later the higher feature is checked/ attracted leading to movement of the remnant. This sequence violates the topdown condition on feature accessibility. In general, no remnant movement derivation is allowed under the top-down condition, unless the features are not in a c-command relation to begin with. But in that case no ordering of operations can be derived from the top-down condition, counter to fact.

Some implications of improper movement for cartography

349

The prohibition against remnant movement is damaging. We saw above the subextraction could be emulated through remnant movement plus ‘stacking.’ The opposite is not true, leaving defenders of the top-down condition without an account of remnant movement.

2.2.3. Conclusion: Internal syntax We have seen that both solutions based on internal syntax face immediate serious problems. Stated generally, the problem with both attempts is that they do not generalize across cyclic domains within the moving item. As hinted at in the text, the solutions apparently only work in the most trivial of cases. Thus compare (17a), which is just like (2aii) except with a relative clause, with (17b) and (17c). While in the trivial case of (17a) the correct templatic arrangement of features together with the top-down or bottom-up condition on feature accessibility provides a solution to the improper movement puzzle, the hierarchical relations between the features are arguably altered in (17b) and (17c). But if this is so, then the top-down or bottom-up condition on feature accessibility fail even for the identity condition as soon as more than the most trivial examples are taken into consideration. (17)

wh a. *the man who is known tA who (that) it was told twho that he would be fired b. *the man whose brother is known tA whose brother (that) it was told wh twhose brother that he would be fired c. *the man the brother of whom is known tA the brother of whom (that) it was told twh that he would be fired the brother of whom

Of course these examples with DPs inside of DPs, that is, cyclic domains embedded within each other, underscore the problem just noted. The putative solutions based on internal syntax fail because they do not generalize across cyclic domains within the moving item. At least for the top-down condition this generalization can be achieved by replacing the > relation in the formulation of the condition on feature accessibility by the c-command or m-command relation. Given the apparent isomorphism between > and , the top-down condition turns into a variant of (11) and the cartographic hierarchy is stated and accessible independently of the structural tokens implementing it. This alternative theory would be just like (11), but replace the definition of affectedness in (11c) by the following definition. (18)

A constituent α is affected by a movement operation iff α is contained in the constituent created by movement.

350

Klaus Abels

This alternative theory has the advantage of offering a more symmetrical definition of affectedness. It derives all three cases of improper movement and shares with the LEC solution that all specifiers and adjuncts along the movement path are affected by the movement of α. The additional predictions of this formulation are largely correct (and follow from the LEC, too). However, it runs into difficulties with quantifier movement in French and Scandinavian.

3. Conclusion It should have become clear throughout this paper that the ordering of operations provides a novel source of evidence for – and testing ground of – cartographic hierarchies. To achieve this generalization, cartographic hierarchies must be stated independently of templates; they are more abstract hierarchies. These hierarchies seem to regulate the order of application of operations (merge and move), but in slightly different ways: For external merge recursion is possible, for internal merge (move) no recursion is allowed in the affected domain. On the whole this view leads to a more integrated theory of structure and movement: they are regulated by the same global constraint, namely the hierarchy, which applies to them in slightly different ways. In addition, we end up with an improved theory of movement. Under the current standard view, movement operations affect only the moving constituent itself and the attractor. This opens the door to very stipulative accounts of movement since movement of one constituent is completely independent of movement of any other constituent. The idea that movement affects parts of the structure distinct from the moving constituent and the attractor itself adds constraints on this system. The system sketched in (11) cuts considerably into the expressive power of the theory (see Abels (2007) for discussion) in a way which is, apparently, empirically desirable.24 If the hierarchy is more abstract than often thought, there must be an interface between the hierarchy and the structure projected from it. In the case discussed here, that of internal merge, this interface is provided by GenPIM and affectedness. Similarly, some interface will be necessary between the hierarchy and the structures projected from it via external merge. This interface is usually construed either as an isomorphism or a homomorphism, but other interfaces are imaginable that might solve some of the open problems such as Bobaljik paradoxes and transitivity failures that arise for cartographic hierarchies at present.

Some implications of improper movement for cartography

351

Notes 1.

2.

3.

4.

5. 6.

7. 8.

The status of passive in this respect is not entirely clear. While Larson’s (1988) account of the double object construction assumes that the passive behaves like raising, Postal (1971); Jackendoff (1990) report a pattern of judgments along the lines of (1b). Speakers consulted find a degradation in examples with reflexives or reciprocals as the complement of by. While Madam-Tussauds-contexts allow binding of the complement of by, no context saves (1bii). (1bi) and (1bii) are discussed in Postal (1971); Reinhart (1983); Jackendoff (1990); Larson (1990). Pesetsky (1995, p. 266) and McGinnis (1998, p. 204) in addition also discuss cases like (1biii). Pesetsky and McGinnis claim that with the inverted order of the two prepositions, that is, about before to, “although the object of the first preposition may not bind an anaphor that is the object of the second preposition, it may bind an anaphor contained within the object of the second preposition” (Pesetsky, 1995, p. 266, emphasis as in original). Several speakers consulted agree with the contrast given in (1b), though an anonymous reviewer fails to perceive a contrast here. This assumption is not universally shared, of course. The main two alternatives are a base generation approach a` la Jackendoff (1990), which wrongly predicts both (1bii) and (1biii) to be grammatical, and light predicate raising a` la Larson (1990), which predicts both of them to be ungrammatical. See Saito (1985); Tada (1993); Miyagawa (1997); McGinnis (1998); Miyagawa and Tsujioka (2004) among many others for relevant discussion and various analyses of these patterns. To the best of my knowledge, a movement operation with the opposite binding profile of (1b) has not been attested. There is an assumption running through much of the literature that a number of further asymmetries neatly correlate with the binding distinction discussed here: licensing of parasitic gaps, reconstruction for binding, reconstruction for scope, type of landing site, type of feature checked. . . Even a superficial look at the literature on what is often called Webelhuth’s paradox in the scrambling literature (Webelhuth (1989); Mahajan (1990); Neeleman (1994)) shows that these correlations do not hold. At best there might be implicational relations. See Starke (2001); Rizzi (2001, 2004) for similar conclusions regarding locality. A reviewer mentions the possibility of casting the distinction between (18a) and (18b) either in terms of the category of the moving item (NP vs. PP) or in terms of the material crossed. A formulation in terms of the category of the moving element is unlikely to be successful. This is so because scrambling in the languages mentioned in the text (German, Hindi, Japanese, Korean, Persian, Slavic) may uncontroversially affect NPs. In addition, according to Pesetsky (1995, p. 266), the binding pattern in (18b) also characterizes heavy NP shift in English, an operation which moves the NP (the analysis of heavy NP shift with the shifted NP in situ (Larson, 1988; Kayne, 1994) has largely been abandoned in the literature (den Dikken, 1995; Rochemont

352

9. 10.

11.

12.

13. 14.

15. 16. 17. 18.

19. 20.

21. 22.

Klaus Abels and Culicover, 1997; Kayne, 2005)). An explanation in terms of the nature of the material crossed faces the same difficulties as does a putative account in terms of the direction of movement. I thank an anonymous reviewer for pointing out the relevance of Martin’s work. As noted, so-called ‘VP-scrambling’ in Japanese also shows the binding pattern in (18a). In this case the standard movement analysis has been called into serious question (Miyagawa, 1997; Miyagawa and Tsujioka, 2004). Notice also that the account of the ban on improper movement in terms of the binding properties of traces is no longer available under the copy theory of movement. Also, see McCloskey (2000, p. 71 fn. 20) for a number of examples that plausibly involve A-movement out of a CP in violation of the binding theoretic account of improper movement. See also Postal (2004, chapter 7). Similarly, one can imagine an account of (2aii) in terms of Rizzi’s 2006 criterial freezing, which prevents an element that has reached a criterial position, e.g., a whoperator position, from moving on. Again this factor could be controlled by turning (2aii) into a matrix question. Again the account faces independent problems. Longdistance scrambling in Japanese famously reconstructs for wh-operator scope. This is compatible with criterial freezing only if the wh-phrase never occupies the criterial position with the wrong consequence that Japanese ought to allow violations of the ban on improper movement. See also Reis and Rosengren (1992). This proposal builds on Chomsky’s (2004) account of subject-island effects. Examples where wh-phrases are scrambled typically involve either contrastive stress on the wh-phrase or require that the item scrambled across is a quantifier or contains a focus particle. Thanks to Tarald Taraldsen for pointing this out. See also footnote 12. Example (10b) is independently ruled out by the locality of scrambling. In light of Reis and Rosengren (1992), the empirical assessment can probably be revised. Wh-movement can feed topicalization in the identity case after all. I include θ relations here. If they are only ever established in the position of first merge, then no interesting feeding/bleeding-relations will be observed. However, if θ -role assignment to derived positions is possible, as claimed in Boˇskovi´c (1994); Hornstein (1999b); Boeckx and Hornstein (2003) a.o., then we can ask whether movement to θ -positions precedes or follows other movement types. The answer seems to be that movement to θ -positions precedes all other types of movement. Williams (2002) is aware of the issue and weakens the impact of the LEC by allowing what he calls mismappings between levels. Non-obvious glosses are as follows: ps present tense sm subject agreement marker pf perfective fv final vowel F marker of focus and cyclicty The hedge ‘fairly simple’ will be discussed in the conclusion to this section, 2.2.3 Such a condition might be derivable for the simplest of cases by assuming that once a feature has been satisfied the relevant layer of the DP is stranded and not taken

Some implications of improper movement for cartography

353

along under movement. Thus, the topmost θ -layer of the DP would be stranded in the θ position when movement to the Case-position takes place, etc. This is what Starke calls ‘peeling.’ The ‘peeling’ solution apparently makes the prediction that moved wh-phrases should be caseless, seeing as they peel off their case layer. This expectation is not borne out. The ‘peeling’-solution also inherits the problems to be discussed in the main text. These are shared by all variants of the top-down condition on feature accessibility. 23. The hedge ‘simple’ will be discussed in the conclusions to this section, 2.2.3 24. A reviewer points out that the system developed in Rackowski and Richards (2005) has a number of properties that are desirable from the current perspective. Movement of an item (surfing, diving, or otherwise) from a movable constituent – which thereby is a phase on Rackowski and Richards’ assumptions – always goes through the edge of that constituent and is feature driven. That allows a reframing of the joint of UCOOL and GenPIM for the surfing and diving cases in terms of a constraint on the order in which the features on a phase head are checked. The UCOOL hierarchy would in that case manifest itself as a hierarchical ordering on the feature checking operations undergone by a single phase head. Examples like (17b) and (17c) pose prima facie problems for this formulation.

References Abels, Klaus 2007 Towards a restrictive theory of (remnant) movement. Linguistic Variation Yearbook, 7: 53–120. Abels, Klaus and Peter Muriungi 2006 The focus particle in Kˆıˆıtharaka. In Ines Fiedler and Anne Schwarz, (eds.), Papers on Information Structure in African Languages, volume 42 of ZAS Papers in Linguistics, pp. 1–20. ZAS. Abels, Klaus and Peter Muriungi 2008 The focus particle in Kˆıˆıtharaka: Syntax and semantics. Lingua, 118(5): 687–731. Abney, Steve 1987 The English noun phrase in its sentential aspect. Doctoral dissertation, MIT. Beck, Sigrid 1996 Quantified structures as barriers for LF movement. Natural Language Semantics, 4: 1–56. Hardcopy. Bergvall, Victoria Lee 1987 Focus in Kikuyu and Universal Grammar. Ph.D. thesis, Harvard University, Cambridge, Mass. Boeckx, Cedric and Norbert Hornstein 2003 Reply to “Control is not movement”. Linguistic Inquiry, 34: 269–280. ˇ Boˇskovi´c, Zeljko 1994 D-structure, theta-criterion, and movement into theta-positions. Linguistic Analysis, 24: 247–286.

354

Klaus Abels

Browning, Marguerite and Ezat Karimi 1994 Scrambling to object position in Persian. In Corver and Riemsdijk (1994), pp. 61–100. Chomsky, Noam 1973 Conditions on transformations. In Stephen R Anderson and Paul Kiparsky, (eds.), A festschrift for Morris Halle, pp. 232–286. Holt, Rinehart and Winston, New York. Chomsky, Noam 1981 Lectures on government and binding. Foris, Dordrecht. Chomsky, Noam 1986 Barriers, volume 13 of Linguistic Inquiry Monograph. MIT Press, Cambridge, MA. Chomsky, Noam 2004 On Phases. Ms. Cinque, Guglielmo 1990 Types of A’-dependencies. MIT Press, Cambridge, MA. Corver, Norbert and Henk van Riemsdijk, (eds.) 1994 Studies on scrambling, volume 41 of Studies in Generative Grammar. De Gruyter, Berlin. Davies, William D. and Stanley Dubinsky 2004 The Grammar of Raising and Control. Blackwell. den Dikken, Marcel 1995 Extraposition as intraposition, and the syntax of english tag questions. Vrije Universiteit Amsterdam/HIL. Fanselow, Gisbert 1990 Scrambling as NP-movement. In G¨unther Grewendorf and Wolfgang Sternefeld, (eds.), Scrambling and Barriers, pp. 113–142. John Benjamins. Fanselow, Gisbert 1991 Minimale Syntax. Habilitationsschrift, Universit¨at Passau. Fanselow, Gisbert 2001 Features, θ -roles, and free constituent order. Linguistic Inquiry, 32(3): 405–437. Fanselow, Gisbert 2004 The MLC and derivational economy. In Arthur Stepanov, Gisbert Fanselow, and Ralf Vogel, (eds.), Minimality Effects in Syntax, volume 70 of Studies in Generative Grammar, pp. 73–124. deGruyter. Friederici, Angela D., Matthias Schlesewsky, and Christian J. Fiebach 2003 WH-movement versus scrambling: the brain makes a difference. In Simin Karimi, (ed.), Word Order and Scrambling, pp. 325–344. Blackwell Publishing, Malden, Mass. Glushan, Zhanna 2006 Japanese style scrambling in Russian: myth and reality. Master’s thesis, Universitetet i Tromsø. Grewendorf, G¨unther 2001 Multiple wh-fronting. Linguistic Inquiry, 32(1): 87–122.

Some implications of improper movement for cartography

355

Grewendorf, G¨unther 2003 Improper remnant movement. Gengo Kenkyu, 123: 47–94. Hiraiwa, Ken 2002 Movement and derivation: Eliminating the PBC. In Penn Linguistics Colloquium 26. Philadelphia, PA. Hornstein, Norbert 1999a Minimalism and quantifier raising. In Samuel David Epstein and Norbert Hornstein, (eds.), Working Minimalism, pp. 45–75. MIT Press, Cambridge, MA. Hornstein, Norbert 1999b Movement and control. Linguistic Inquiry, 30(1): 69–96. Huang, C.-T. James 1993 Reconstruction and the structure of vp: Some theoretical consequences. Linguistic Inquiry, 24: 103–138. Jackendoff, Ray 1990 On Larson’s treatment of the double object construction. Linguistic Inquiry, 21: 427–456. Jonsson, J. G. 1996 Clausal architecture and case in Icelandic. Ph.d. dissertation, University of Massachusetts, Amherst. Kayne, Richard 1994 The antisymmetry of syntax. MIT Press, Cambridge, MA. Kayne, Richard 1998 Overt vs. covert movement. Syntax, 1: 128–191. Kayne, Richard 2005 Some remarks on agreement and on heavy NP shift. In Movement and Silence, Oxford Studies in Comparative Syntax, chapter 11, pp. 261–276. Oxford University Press, Oxford and New York. Kobele, Gregory 2008 Deriving Reconstruction Asymmetries. Ms. Humboldt Universit¨at zu Berlin. Koopman, Hilda and Anna Szabolcsi 2000 Verbal Complexes. MIT Press, Cambridge, MA. Kuthy, K. de and Walt Detmar Meurers 1999 On partial constituent fronting in German. In V. Kordoni, (ed.), T¨ubingen Studies in Head-driven Phrase Structure Grammar, pp. 22–73. Larson, Richard K. 1988 On the double object construction. Linguistic Inquiry, 19: 335–391. Larson, Richard K. 1990 Double objects revisited: Reply to Jackendoff. Linguistic Inquiry, 21(4): 589–632. Lasnik, Howard and Mamoru Saito 1984 On the nature of proper government. Linguistic Inquiry, 15: 235–289. [Reprinted in Essays on restrictiveness and learnablity, Howard Lasnik, 198–255. Dordrecht: Kluwer, 1990].

356

Klaus Abels

Lasnik, Howard and Mamoru Saito 1991 On the subject of infinitives. In Lise M. Dobrin, Lynn Nichols, and Rosa M. Rodriguez, (eds.), Papers from the 27th regional Meeting of the Chicago Linguistic Society Part One: The General Session, pp. 324–343. Chicago Linguistic Society, University of Chicago, Chicago, Ill. Lasnik, Howard and Mamoru Saito 1992 Move α: Conditions on its application and output. MIT Press, Cambridge, MA. Lee, Young-Suk and Beatrice Santorini 1994 Towards resolving Webelhuth’s paradox: evidence from German and Korean. In Corver and Riemsdijk (1994). Mahajan, Anoop 1990 The A/A-bar distinction and movement theory. Doctoral dissertation, MIT. Martin, Roger 2001 Null Case and the distribution of PRO. Linguistic Inquiry, 32(1): 141– 166. May, Robert 1979 Must comp-to-comp movement be stipulated? Linguistic Inquiry, 10: 719–725. McCloskey, James 2000 Quantifier float and wh-movement in an Irish English. Linguistic Inquiry, 31(1): 57–84. McGinnis, Martha Jo 1998 Locality in A-movement. Ph.D. thesis, MIT. Miyagawa, Shigeru 1997 Against optional scrambling. Linguistic Inquiry, 28(1): 1–25. Miyagawa, Shigeru and Takae Tsujioka 2004 Argument structure and ditransitive verbs in Japanese. Journal of East Asian Linguistics, 13: 1–38. M¨uller, Gereon 1998 Incomplete Category Fronting: A Derivational Approach to Remnant Movement in German. Kluwer Academic Publishers, Dordrecht, Boston. M¨uller, Gereon and Wolfgang Sternefeld 1993 Improper movement and unambiguous binding. Linguistic Inquiry, 24: 461–507. M¨uller, Gereon and Wolfgang Sternefeld 1996 A’-chain formation and economy of derivation. Linguistic Inquiry, 27: 480–511. Muriungi, Peter 2003 Wh-Questions in Kiitharaka. Master’s thesis, University of the Witwatersrand, Johannesburg. Muriungi, Peter 2005 Wh-questions in Kiitharaka. Studies in African Linguistics, 34: 43–104. Neeleman, Ad 1994 Scrambling as a D-structure phenomenon. In Corver and Riemsdijk (1994), pp. 387–429.

Some implications of improper movement for cartography

357

Nevins, Andrew Ira 2004 Derivations without the activity condition. MIT Working Papers in Linguistics, 49: 287–310. Pesetsky, David 1995 Zero syntax. MIT Press, Cambridge, MA. Pesetsky, David 2000 Phrasal Movement and Its Kin. Linguistic Inquiry Monograph 37. MIT Press, Cambridge, MA. Postal, Paul M. 1971 Cross-over phenomena. Holt, Rinehart, and Winston, New York. Postal, Paul M. 1974 On raising: One rule of English grammar and its theoretical implications. MIT Press, Cambridge, MA. Postal, Paul M. 2004 Skeptical linguistic essays. Oxford University Press, Oxford and New York. Rackowski, Andrea and Norvin W. III Richards 2005 Phase edge and extraction: A Tagalog case study. Linguistic Inquiry, 36(4): 565–599. Reinhart, Tanya 1983 Anaphora and semantic interpretation. Croom Helm, London. Reis, M. and I. Rosengren 1992 What do wh-imperatives tell us about wh-movement. Natural Language and Linguistic Theory, 10(1):79–118. Rizzi, Luigi 2001 Relativized minimality effects. In Mark R. Baltin and Chris Collins, (eds.), The Handbook of Syntactic Theory. Blackwell, Oxford. Rizzi, Luigi 2004 Locality and left periphery. In Adriana Belletti, (ed.), Structures and Beyond: Cartography of Syntactic structures, volume volume 3, pp. 104– 131. Oxford University Press, New York. Rizzi, Luigi 2006 On the form of chains: Criterial positions and ECP effects. In Lisa Lai Shen Cheng, (ed.), Wh-movment: moving on, pp. 97–134. MIT Press, Cambridge, Mass. Ms. University of Siena. Rochemont, Michael S. and Peter W. Culicover 1997 Deriving dependent right adjuncts in English. In Dorothee Beerman, David Leblanc, and Henk van Riemsdijk, (eds.), Rightward Movement, volume 17 of Linguistik Aktuell, pp. 279–300. John Benjamins. Rognvaldsson, E 1987 OV word order in Icelandic. In Proceedings of the Seventh Biennial Conference of Teachers of Scandinavian in Great Britain and Northern Ireland, pp. 33–49. University College London. Saito, Mamoru 1985 Some asymmetries in Japanese and their theoretical implications. Doctoral dissertation, MIT.

358

Klaus Abels

Saito, Mamoru 2002 Towards the unification of scramblings. MIT Working Papers in Linguistics, 41: 287–308. Sakai, Hiromu 1994 Derivational economy in long distance scrambling. In Masatoshi Kozumi and Hiroyuki Ura, (eds.), Formal Approaches to Japanese Linguistics 1, volume 24 of MIT Working Papers in Linguistics, pp. 295–314. MITWPL. Sauerland, Uli 1996 The interpretability of scrambling. In Masatoshi Koizumi, Masayuki Oishi, and Uli Sauerland, (eds.), Formal Approaches to Japanese Linguistics 2, volume 29 of MITWorking Papers in Linguistics, pp. 213–234. MITWPL. Sauerland, Uli 1998 The Meaning of Chains. Ph.D. thesis, MIT, Cambridge, Mass. Schwarz, Florian 2003 Focus marking in Kikuyu. In Regine Eckardt, (ed.), Questions and Focus, volume 30 of ZAS Papers in Linguistics, pp. 41–118. Zentrum fur allgemeine Sprachwissenschaft, Berlin. Starke, Michal 2001 Move Dissolves into Merge: a Theory of Locality. Doctoral dissertation, University of Geneva. Sternefeld, Wolfgang 1992 Transformationstyplogie und strukturelle Hierarchie. Paper presented at GGS 1992, Z¨urich. Tada, Hiroaki 1993 A / A-bar partition in derivation. Ph.D. thesis, MIT, Cambridge, Mass. Ueyama, Ayumi 1994 Against the A/A’-movement dichotomy. In Corver and Riemsdijk (1994), pp. 459–486. von Stechow, Arnim and Wolfgang Sternefeld 1988 Bausteine Syntaktischen Wissens: ein Lehrbuch der Generativen Grammatik. Westdeutscher Verlag, Opladen. Webelhuth, Gert 1989 Syntactic saturation phenomena and the modern Germanic languages. Doctoral dissertation, Universite de Geneve. Williams, Edwin S. 1974 Rule Ordering in Syntax. Ph.D. thesis, MIT. Williams, Edwin S. 2002 Representation Theory. MIT Press, Cambridge, Mass. Williams, Edwin S. to appear Subjects of different heights. In Hana Filip, Steven Franks, James E. Lavine, and Mila Tasseva-Kurktchieve, (eds.), Proceedings of FASL 14 – The Princeton Meeting. Slavica.

Some implications of improper movement for cartography

359

Wiltschko, Martina 1997 D-linking, scrambling and superiority in German. Groninger Arbeiten zu germanistischen Linguistik, 41: 107–142. Wiltschko, Martina 1998 Superiority in German. In E. Curtis, J. Lyle, and G. Webster, (eds.), Proceedings of the Sixteenth West Coast Conference on Formal Linguistics, pp. 431–445. CSLI Publication, Stanford, CA.

There is no alternative to cartography Edwin Williams Abstract. According to traditional pedagogical grammars be, have, etc., are ‘helping verbs’, and a clause has only one main verb, so ‘was seen’ is one clause, but ‘tried to see’ is two. But what is a clause? Is the distinction real? In this note I will suggest that even if the intricate orderings that the F-structure of a clause exhibits are explained away in semantic or other terms, F-structure itself will nevertheless be indispensable in two different ways – first, the Fn/F0 boundary (that is, the boundary between a matrix and an embedded F-structure) plays a distinctive role in spell-out; and second, the F-structure hierarchy regiments derivations in a way that does not reduce to what is embedded in what in the particular structure or structures undergoing derivation. F-structure is thus best thought of as the ‘clock’ that governs what happens when in the workspace, and only derivatively as the structure of the clause. If these conclusions are correct, then there may be alternatives to cartography, but there are not alternatives to F-structure.

In modern grammatical thinking, a clause has a ‘Functional Structure’ (Fstructure), a spine consisting of some number of nonlexical heads, perhaps as many as 150, with the lexical verb at the bottom. If we number these heads 0, . . . n and reserve 0 for the lexical head, then a clause is a sequence Fn . . . F0. If we embed a clause beneath a lexical verb, as in (1) below, we have two Fstructures, and we have a boundary between them, the ‘Fn /F0 boundary’, circled here: (1)

Fn

For better or worse, F-structure constitutes what we know about the structure of the clause, about what a clause is. But instead of asking what a clause is, we might ask, what is special about the Fn/F0 boundary? After all, isn’t it all just embedding, inside the clause and between clauses, so why attach special significance to resetting the F-structure clock as we pass from a main clause to an embedded clause? Maybe we shouldn’t.

362

Edwin Williams

One candidate answer we can rule out from the start is the idea that the elements of a single F-structure are bound together by a shared feature (say, ‘+V’ for clauses), as in, for example, Grimshaw’s (2005) account of F-structure. Identifying F-structures in this way does not distinguish main verbs from auxiliary verbs – ‘was seen’ will have every node labeled ‘+V’, but so will ‘tried to see’, and so we cannot draw the distinction between the one clause and the two clause cases.

1. Cartography A general program in linguistics for some time now is to map the Fn . . . F0 sequence (‘cartography’) and to explain the sequence. There are two broad subprograms of explanation. One is to identify intrinsic properties of elements in the sequence that account for their positioning in the sequence in terms of laws not specifically designed to regulate the sequence. Nilsen (2004) has proposed that polarity is such a property. We know for example that in a single clause, not must precede yet: *‘John has yet seen nothing’/‘No one has yet seen John’. But this follows not from the theory of clause structure, but from a broader law, which holds across the clause boundary: ‘No/*some one thinks that John has yet arrived’. So the theory of clause structure, and in particular the theory of the derivation of the Fn -F0 sequence, is relieved of the duty to explain this ordering. And likewise for other cases that Nilsen discusses. The other program of explanation proceeds by identifying substructures of F-structure with elements of an ontology of some kind, consisting of ‘events’, ‘facts’, ‘speech acts’ etc., where the ontology has at least enough structure to say that events are parts of facts, and facts are part of speech acts, and not viceversa; then, given that completely modifies events, and actually modifies facts, and frankly modifies speech acts, we can predict the relative ordering of these adverbs: (2)

a. John frankly actually completely startled Bill b. *John completely frankly startled Bill c. *John completely actually startled Bill d. *John actually frankly startled Bill e. But: John actually said that frankly he completely startled Bill

A problem for this program is illustrated in (e) – if we have 2 clauses, then we can get a ‘speech act’ (modified by frankly) embedded under a ‘fact’ (modified by actually). So the ontology must allow this. But it remains that this ordering is

There is no alternative to cartography

363

not possible within a single clause. So it appears that we cannot use the structure of the ontology to explain the ordering within a single clause without saying something like the ordering in the ontology must hold within a single clause, but not across clauses. But with this qualification, we cannot really be said to be explaining something about language purely in terms of something outside of language, and we are still stuck with the clause, and the Fn/F0 boundary, as a mysterious entity. Further problems of the same kind are illustrated by probably S, which is one clause, versus It is probable that S which is two; and needn’tV which is one clause, versus need toV, which is two, and canV versus be able to V. These last examples also suggest that there is not an obvious semantic criterion for distinguishing functional verbs (Fi ) from lexical verbs (F0), though obviously there are limits. In what follows I will have no interest in either of these explanatory programs, nor in the program of cartography itself. I will rather be interested in two other questions – first, why is cartography possible at all? And second, what is special about the Fn/F0 boundary? I will try to show that even if these ‘alternatives to cartography’ fully succeed, the Fn/F0 boundary will remain fundamental to a class of explanations. Under what circumstances would cartography be impossible? Consider the phrases NP and PP – and ask, what order of embedding do they occur in? There is no answer of course – either of them can embed the other. Now imagine that the Fi ’s of clausal structure were the same – they could occur in every possible order. Then there would be no mapping, or explaining, their order. There would at most be the project of identifying them, just as there is the project of identifying the basic parts of speech. So we might begin to understand F-structure by identifying what properties of it allow cartography to be done on it. It seems to me that F-structure has two properties that enable cartography. First, it is nonrecursive, and therefore finite, and second, elements occur in a fixed linear order. In fact, it would be hard to have one of these properties without having the other, but I will treat them as separate properties. They enable cartography in that if there were recursion within a clause, the clause itself would be infinite in its structure, and could not be mapped; and if elements could occur in any order, there would be no cartography to do. But in fact it seems that F-structure actually lacks both of these crucial properties. First, there is recursion within a single F-structure: (3)

. . . [[[[[John eats black eyed peas] on the first day] in the second week] in every third month] in every other year] in every fifth century] . . . .

364

Edwin Williams

Time modifiers of ever-increasing size can be added to (3) ad libitum, and if we ever run out of named units we use made-up ones (‘every other unit of 100 million eons’). And (3) does have the stacking structure illustrated in (3), although it possibly also has a structure in which there is one very complex PP (on the first day [in the second week [. . . ]]). The existence of the stacking structure is confirmed by the fact that we can front a middle term of the sequence: (4)

In every other week John eats black eyed peas on Tuesday, in every third year.

This fronting would be impossible if the recursion occurred within a single PP. So recursion within a single F-structure exists, since on all accounts these stacked time modifiers occur in a single clause, and so F-structure is not finite. Likewise, linearity fails to hold: (5)

John could (not) have (not) been (not) swimming

not can occur in any position in the auxiliaries except before the tensed one. It makes a different contribution to the meaning of the sentence in each position, but of course – in general, any difference in hierarchical structure will correspond to a difference in meaning, quite apart from the particular properties of F-structure. The point is, negation has no single privileged position of occurrence, and likewise for other elements like modals, as has been often noted. The upshot is, the two properties that at first glance would seem to be characteristic of F-structure, and which are necessary to make cartography possible, aren’t even true of the F-structure of clauses. But of course they are roughly true, or else cartography would indeed be impossible. But if they aren’t completely true, then they are not definitional; that is, they do not tell us why the Fn/F0 boundary is a distinguished boundary, and they do not tell us what a clause is, as distinguished from other Fi . . . Fj sequences, including ones which might include the Fn/F0 boundary in the middle – these too will show rough but not complete finiteness and linearity. There are three further candidate properties which might distinguish the Fn/F0 boundary. One of them is branchingness – it is generally assumed that F-structure is strictly right-branching, in that any phrase that develops on a left branch must begin a new F-structure. I will assume that this is true, for simplicity. A second property of F-structure predicates is that they seem to take only a single complement. The modal can takes a VP complement, but no modal takes an NP +CP complement parallel to a verb like persuade. Again, taken by itself it is not a defining property, nor even a sufficient diagnostic, as nonfunctional verbs like

There is no alternative to cartography

365

try, want, think, etc., also take a single complement. F-structure predicates of course take complements smaller than themselves (that is, Fn takes only Fn−1 complements, but again there is good reason to think that lexical verbs take small complements as well. And thirdly, F-structure defines the locality of some kind of syntactic relations. But what kind of relations? The clause boundary has always been the locus of locality conditions, enshrined in the term ‘clause mate’ and more recently occurring in the definition of ‘phase’. But neither of these provides a grounding for the definition of clause. Reflexivization, for example, does not really obey the clausemate condition, as reflexivization in ECM cases shows (John wants [himself to win]). And phases defined in terms of CP and vP fail to do so as well, in that they do not bestow significance on the Fn /F boundary: TPs can be embedded independently as clauses, as in raising constructions, but crucially don’t thereby constitute independent phases but do nevertheless introduce an Fn/F0 boundary. Also phases divide the clause into two parts, the upper phase and the lower phase, but these constitute a single clause with respect to Fn/F0 bounding. Also, phases are not absolutely impenetrable, they simply set the conditions for penetration (the Phase Impenetrability Condition, van Riemsdijk (1998), Chomsky (1999)). Again, it remains a possibility that clauses, or complete F-structures, have no special standing, that only phases do, but we will explore further what might make the Fn/F0 boundary significant. I ask again, is there any process that treat the Fn/F0 boundary as an absolute barrier? When we examine the familiar classes of syntactic relations, we find that none do: (6)

a. b. c. d.

e.

f.

Not WH: Who do you think [ Fn t will win]? Not NP movement: John is expected [ Fn t to win] Not case assignment: John expects [ Fn him to win] Not agreement: enir uza magalu b-acruli r-/b-ixyo Mother [boy bread-ABS(III) III ate] IV/III-know ‘The mother knows that the boy ate the bread’ (Tsez) Polinsky and Potsdam (2001) via Bobaljik and Wurmbrand (2005) Not phonological dependence: 1. Who does he think[ ’s coming]? 2. Who did he say [ ’z coming]? Not Incorporation: 1. Ni +V [DP... ti . . . ] 2. Kanekwarunyu wa’-k-akyatawi’tsher-u:ni it.dotted.DIST PAST-I-dress-make ‘I dress-made a polka-dotted one.’ Baker (1988)

366

Edwin Williams

In (a–c) we see familiar examples Wh-movement, NP-movement and Case Assignment that cross clause boundaries on any account. In (d) we see the much rarer and perhaps more controversial case of agreement crossing a clause boundary – the embedded absolutive object in Tzez can agree with the matrix verb if it is a topic of the embedded clause; one is tempted to give the object a covert presence in the matrix (as in ‘The mother knows of the bread that . . . ’), but P&P argue specifically against this. In (e) we see that voicing assimilation crosses a tensed clause boundary. So we see that nothing respects the Fn/F0 boundary per se.

2. But F-structure is crucial to two explanatory enterprises Suppose F-structure orderings could be entirely predicted from the semantics of the elements that entered into it, despite the skepticism expressed in the previous section. At first blush, such an account would seem to eliminate Fstructure as an object in the theory, it would all be explained away in other terms, leaving no remainder. There would then be no distinction between Fstructure embedding and lexical embedding – it is all just ‘embedding’. And in particular, the Fn/F0 boundary would have no special significance. But in fact, even with a full accounting of the ordering of the F-structure elements, Fstructure, and in particular the Fn/F0 boundary, would still be required to carry out two other explanatory enterprises. First, the Fn/F0 boundary plays a role in bounding head movement that no other boundary plays. And second, F-structure governs the ‘timing’ of syntactic operations, including sentential embeddings, and is thereby crucial to understanding locality conditions, improper movement, and reconstruction effects. Rationalizing F-structure by ‘eliminating positions’ in Nilsen’s sense does not address either of these other two questions.

2.1.

Bounding Head Movement

Head Movement is understood to be bounded by the Head Movement Condition (HMC) (Travis (1982)) and so we don’t have: (7)

*Have John should t left?

Since Head Movement is the means of deriving spelled-out forms, the Head Movement Condition governs spell-out, and gives it its particularly local character.

There is no alternative to cartography

367

But some cases of head movement, called Long Head Movement (LHM), violate the strict HMC: (8)

a. b.

Procel Read Lennet Read

sum t knigata have-I the-book (Bulgarian, Rivero (1992)) en deus Yann t al levr 3sgM has Yann the book (Breton, Roberts (2002))

The auxiliary verbs themselves in these cases trigger movement of the participle; Rivero (1992) plausibly suggests that these auxiliaries are clitics, and the head movement serves to provide hosts for them. But such cases of LHM are nevertheless subject to an absolute bound at Functional junctures (i.e. Fn/F0). Roberts in fact takes pains to distinguish LHM from remnant VP topicalization exactly on these grounds: (9)

a.

long V-movement: *Desket am eus klevet he deus Anna he learnt 1Sg have heard 3SgF have Anna 3SgF c’hentelio`u. lesson ‘I have heard that Anna has learnt her lessons.’

b.

VP-topicalization O lenn al levr a ouian ema˜n Yann. Prog read the book Prt know.1Sg is Yann ‘I know Yann is reading the book.’ (Breton, Roberts (2002) example 77)

LHM, but not VP-topicalization, is clause-bounded, that is, Fn/F0 bounded. Furthermore, Long Head Movement is blocked by certain heads, whereas VP-topicalization is not: (10)

a.

b.

long V-movement: *Lennet n’en deus ket Tom al levr. Read Neg3SgM have Neg Tom the book ‘Tom hasn’t read the book.’ VP-topicalization: O lenn al levr n’ema˜n ket Yann. Prog read the book Neg.is Neg Yann ‘Yann isn’t reading the book.’ (Breton, Roberts (2002) example 78)

368

Edwin Williams

Following Rivero’s suggestion that certain auxiliaries are clitics, the blocking in (a) is presumably because the negative auxiliary n’en is not a clitic, and so does not trigger Head Movement of the participle to support it; but see Roberts’ discussion for a different view. The rarity of LHM must be due to the fact that in general adjacent heads are appropriate targets for Head Movement, and so block movement of lower heads. In any case, the contrast with Remnant VP-topicalization shows that there is a non-relativized, absolute barrier to Head Movement across a Fn/F0 boundary. The most general constraint is that Head Movement, in every variety, including Lowering (Affix Hopping) is bounded by any Fn/F0 boundary. This would include cases of small clause embedding (Where a clausal Fi P is embedded under F, and i is non-maximal), and cases of F-structures with other lexical F’s. The bounding must take a particular form, in the light of considerations to follow. It is not simply that different F-structures cannot be entangled by Head Movement, most broadly conceived. So, for example, in the domain of lexical derivation it is possible to have a derivation like the following: (11)

-er N > [driveV truck N ]v → [drive-er truck] → truck-driver

Only a very particular formulation will allow (11) and not (7). In lexical derivation, F-structures are trivial, as there are only the F0s present. Nevertheless, the derivation just given can be understood as an instance of Head Movement (Lowering in this case), and a nominal element intrudes into a verbal domain. The question is then why such lowering is possible, but not the following: (12)

John T think [that Mary see Paul] → John think [Mary T+see Paul]

The reason for the difference between (11) and (12) lies the in definition of Head Movement, from which the absolute barrierhood of Fn/F0 boundaries arises.i In the following, I will assume that raising and lowering are both Head Movements, the difference between them having only to do with how the result is spelled out, not with how the two relevant head positions that are related are identified. The definition includes a ‘search algorithm’ for finding an eligible target for the head that is being moved: (13)

Head Movement Search Algorithm: for H1> [H2 W], first check H2 for eligibility; if it fails, apply the same search to H1>W iff H1 and W are in the same F-structure.

The recursive part of the search in (13) takes place only in the presence of Fstructure, and the recursion stops when an Fn/F0 boundary is encountered. So (12) is impossible because, assuming for the purposes of the example that the

There is no alternative to cartography

369

matrix verb think is an ineligible target, at the point where (13) is operating on T > [think [that Mary see Paul]], it will determine that think is not an eligible target, but it can look no further, because the that-clause is not a Functional complement of think. On the other hand (11) is still a valid derivation, because drive is an eligible target for -er, and since it is the highest head in [drive truck] there is no need to invoke the recursive part of (13). (13) thus gives us both the Minimality style bounding of Head Movement (the HMC) but also an absolute bound at Fn /F boundaries. See Williams (2008) and Williams (forthcoming) for further discussion of the formulation of (13). The formulation in (13) makes a further prediction – in a limited kind of case, Head Movement will cross Fn/F0 boundaries: exactly when the top head of a Functional structure is an eligible lower target: (14)

[F [Fn [Fn−1 . . . ]] ⇒ [ [F+Fn [Fn−1. . . ]]

Exactly this derivation is found in Preposition+Article amalgams in languages like French, German, and Romanian: (15)

a. b.

[zu [dem Bahnhof]DP ]PP → zum Bahnhof (German) to the train-station P > D → P+D

The derivation is valid, despite bridging a non-functional boundary (DP/PP), exactly because no recourse is made to the recursive part of (13). Another sort of example has been brought to my attention by Aaron Silberstein (p.c.) – the V+C moˇzda in Serbian. It has the same meaning as English maybe except that maybe is truly an adverb now, where has moˇzda is fixed in sentence-initial position ((it) may be that . . . ) suggesting that it is a result of V > C ⇒ V + C, parallel to the preposition-determiner combinations just discussed. Of special interest is that moˇz is not a main verb but an auxiliary.

2.2. Timing derivations The second project that must accord F-structure hierarchy and the Fn/F0 boundary a distinguished role is a principled relation between F-structure and the regimentation of derivations. In Williams (2003) it is proposed that the traditional A/A’ distinction should be generalized to a many-way distinction, with F-structure providing the dimension of generalization. That is, for every pair of F-structure elements, there is an Fi /Fj distinction. There were several consequences of that generalization, some of which can be stated in a way independent

370

Edwin Williams

of the model of grammar presented in that book (Representation Theory (RT)). I will so state them here, but without repeating the evidence given in the book, because if these consequences are embraced then it is especially clear that the role of F-structure goes beyond ordering the nodes in the spine of the clause, and then even when that ordering is fully rationalized, however that takes place, F-structure will not have been eliminated from the system. It has long been recognized that A’ movements are more likely to reconstruct than A movements. Under the F-structure generalization, this becomes: a movement targeting Fi is more likely to reconstruct that a movement targeting Fj , i > j, a somewhat weak but not useless conclusion. In Williams (2003) I+went somewhat further. Every reconstruction is a reconstruction of X for Y – for example, Wh-movement reconstructs for Anaphor-binding, as in the following: (16)

Which picture of himself do you think that Bob saw t?

The RT model has as a consequence the following stronger conclusion: (17)

Reconstruction: any movement targeting Fi will reconstruct for any relation (including movement, binding, whatever) that targets Fj , i>j.

Law (17) in fact entails the weaker generalization about reconstruction. Some cases that lie outside the A/A’ distinction are the following. ‘Prior to’ A-movement, we find that NP movement ‘reconstructs’ for theta role assignment. ‘After’ A-bar movement we find long topicalization in Japanese which ‘reconstructs’ for WH-interpretation: (18)

?Dono hon-oI Masao-ga [CP Hanako-ga tI tosyokan-kara which book-acc Masao Hanako library-from karidasita ka] siritageatteiru checked.out Q wants-to-know ‘Masao wants to know which book Hanako checked out from the library.’ Saito (1991)

The Wh-phrase dono hon-o at the top of the matrix is interpreted at the top of the embedded clause, even though it has been entirely removed from the embedded clause, which means that the WH phrase is licensed (WH-interpreted) in its reconstructed position. The reader is referred to Williams (2003) for further cases of reconstruction that go beyond the familiar A/A’ distinction. Law (17) also entails a corollary about remnant movement. Remnant movement can be seen as a special case of reconstruction – one movement rule ‘reconstructs for’ another movement rule. The corollary is:

There is no alternative to cartography

(19)

371

Remnant creation: If movement rule A targets Fi , and B targets Fj , i > j, then B can create remnants that A can move, but not vice-versa.

Importantly, (17) and (19) hold regardless of whether the Fi and Fj in question are in the same functional projection or not; that is, (17) is an absolute law governing reconstruction and (19) is an absolute law governing remnant creation. Again, see Williams (2003) for discussion. Finally, F-structure governs the locality of relations via a principle called the Level Embedding Conjecture (LEC; Williams (2003)). It has long been observed that A’ movements were not as local as A movements, but no general account has been forthcoming. Under the F-structure generalization of the A/A’ distinction, we might expect that movements targeting Fi might be more local than movements targeting Fj , i < j. The LEC in fact enforces exactly this conclusion, by ‘timing’ the embedding of subordinate clauses. Assume that embeddable clauses come in all sizes, where size is measured by the F-structure index of the top node of the clause. We can then ask “at what point in the derivation are clauses embedded in their matrix context?”. The standard answer is, clauses are embedded exactly when the embedding verb and the built-up clause are merged. But the LEC gives a different answer: a clause is embedded exactly when the matrix is built up to exactly the same ‘size’ as the clause to be embedded. This arises naturally in the Representation Theory model – at any given level in that model, there are only the things that are built up to that level, where the levels are keyed to the F-structure hierarchy. The LEC has the consequence that very big clauses will not be embedded until very late (let’s say Fi ), which means that the opportunity to apply a rule targeting the matrix Fj , j < i, and some position in the embedded clause will have passed; only rules at least as late as the rules that target Fi or greater can apply. This means that rules targeting Fi are never able to penetrate clauses of size Fj , j > i, a quite general locality condition. The LEC also entails an ‘improper movement’ condition generalized to Fstructure: (20)

Generalized Improper Movement Condition: An element moved by a rule targeting Fi cannot subsequently be moved by a rule targeting Fj , i > j.

As with (17) and (19), it is important to the questions addressed here to understand that (20) holds (as a matter of empirical fact and also in the RT model) whether Fi and Fj are in the same F-structure or not. Again, see Williams (2003) for examples and evidence.

372

Edwin Williams

3. Conclusion The most important conclusion for the purposes of this volume is the conclusion about the relation of F-structure to such things as Locality, Reconstruction, and Improper Movement. Simply put, no matter how the ordering of F-structure elements is rationalized, the F-structure hierarchy must be referenced in the course of the derivation, and must be referenced in a way that does not reduce to what structures have been built up at various stages of the derivation. To understand Locality, Reconstruction, and Improper Movement, we must be able to refer to the ordering of F-structure elements in the abstract, not as they occur in a particular structure. ‘Fi > Fj ’ cannot simply mean that Fi is higher in the structure than Fj ; for the purposes of explaining improper movement for example, we need to say that the position targeted by the first movement must be to a position which is lower in its F-structure that the position targeted by the second movement is, in its F-structure. Thus we must be able to make absolute comparisons of two positions with respect to the F-structure ordering. To carry forward the derivation, one must have not just the structures themselves, but the F-structure yardstick independently available. This conclusion that F-structure is indispensable is reinforced by the role of F-structure in the bounding of Head Movement. It is not enough to have principled explanations of the ordering of all the functional embeddings, as those explanations cannot account for head bounding. Since the F-structure hierarchy must be consulted independent of what structures exist, its role in determining the shape of structures seems redundant. One way to eliminate the redundancy is in fact to use F-structure to govern everything: movement, reconstruction, and the proper construction of clauses. The Representation Theory Model does exactly that. In any case, there remains the unique role of the Fn/F0 boundary, which will not go away even in the presence of a complete account of the F-structure ordering, at least not without at least one further discovery of unknown character. Given these conclusions, it is perhaps best to regard F-structure not primarily as the structure or ‘spine’ of the clause, but rather as a ‘clock’ that governs the timing of events in the workspace. It will be indirectly reflected in the structure of the clause, but only imperfectly, due to the perturbations of morpohosyntax. See Williams (2008) for elaboration.

There is no alternative to cartography

373

References Baker, Mark 1988

Incorporation: a theory of grammatical function changing. Chicago: University of Chicago Press. Bobaljik, Jonathan and Susi Wurmbrand 2005 The Domain of Agreement. Natural Language and Linguistic Theory 23, 809–865. Chomsky, Noam 1999 Derivation by Phase. In: MIT Occasional Papers in Linguistics, no. 18, Cambridge, MA: MIT Working Papers in Linguistics, Department of Linguistic and Philosophy, 1999. (Unpublished revision 1999; reprinted in Festschrift for Kenneth Hale. Publication pending) Grimshaw, Jane 2005 Words and Structure Stanford, CA: CSLI Nilsen, Oystein 2004 Eliminating Positions: Syntax and semantics of sentence modification. Ph. D. diss., Utrecht University. Polinsky, Maria and Eric Potsdam 2001 Long distance agreement and topic in Tsez Natural Language and Linguistic Theory 19(3): 583–646 Riemsdijk, Henk van 1998 Head Movement and Adjacency. Natural Language and Linguistic Theory 16: 633–678. Rivero, Maria Luisa 1992 Adverb Incorporation and the syntax of Adverbs in Modern Greek. Linguistics and Philosophy 15(3): 289–331. Roberts, Ian 2002 Head Movement. In: Mark Baltin and Chris Collins (eds) Handbook of Syntactic Theory, 113–147. Oxford: Blackwell. Saito, Mamoru 1991 Long Distance Scrambling in Japanese. Ms, UConn. Travis, Lisa 1982 Parameters and Effects of Word Order Variation. Ph. D. diss., MIT. Williams Edwin 2003 Representation Theory. Cambridge, Mass: MIT Press Williams Edwin 2008 Merge and Mirrors. Ms., Princeton University (available on Lingbuzz). Williams Edwin forthcoming Merge and Mirrors. In: Tal Saloni (ed), Memorial to Tanya Reinhart

Index adjectives, 6, 134–135, 143n28, 247–257, 265–269, 275–295, 300–302, 310–314, 316 adverbs, 2, 29, 163, 272n18, 305–306, 362 auxiliary inversion, 164, 203–204, 208–219, 229–239

French, 122, 124, 127, 247–255, 266–269, 278, 283–285, 287, 290, 293, 295, 299–302, 312n4, 343, 369 functional projections, 1, 3–5, 7–8, 16, 29, 41–42, 53–54, 60–62, 66, 69, 71, 87, 110, 117–118, 124–126, 136–137, 214–215, 245–246, 256, 268

binding, 87, 90–94, 96, 100–104, 119, 130–131, 262, 290–291, 303–310, 316–317n22, 317n25, 317n26, 326–327, 331, 334–335, 370 Blackfoot, 7, 153–154, 156–158, 175–194, 197n27, 197n30, 197n31, 197n34, 198 Breton, 367 Bulgarian, 367

German, 66, 80n1, 82n10, 90, 103, 203–240, 301, 333–336, 369

CAT, 203, 209–213, 218–219, 229–239 Catalan, 110, 127, 129–131, 133, 135 compositionality, 8, 53–60, 66, 69–71, 77–80, 308 contrast, 7, 15–42, 64, 69–75, 77–79, 270n4, 352n14, defectiveness, 124, 134, 136, 142n24, 143n25 ditransitive, 6, 26, 39, 87–88, 90–100, 104 Dutch, 3, 7, 19–30, 41–44, 78–79, 90, 214 English, 7, 18, 24, 57–60, 63–65, 67–70, 80n2, 137, 154–156, 159, 161–165, 170–174, 180–181, 192–193, 232, 247–255, 266–267, 281, 327, European Portuguese, 6, 88–93, 95–98, 100–102, 105 focus, 6–8, 15–30, 36–46, 53–82, 87–93, 104–105, 120–123, 128–130, 140n10, 240n5, 352n14, freezing principle, 329, 347, 352n12

Halkomelem, 7, 153–175, 192–198 Hebrew, 276–278, 288, 291, 295 Icelandic, 90, 343–344 improper movement, 6, 325–353, 366, 371–372 INFL, 103, 153–159, 174–175, 192–194 Italian, 29, 55, 71–76, 105n6, 131, 277 Japanese, 15, 18, 21, 30–35, 41–42, 327, 332, 352n10, 352n12, 370 Kîîtharaka, 343–344 Last Resort, 140n13 LCA, see Linear Correspondence Axiom LF, see Logical Form Linear Correspondence Axiom, 3, 5–6, 8, 246, 250–261, 275, 312n2, 314n16, linearization, 8, 94, 103, 126, 218, 257, 266–267, 275–276, 296–299, 314n16, 315n18 location, 7, 153–175, 192–194, 196n26, 263 Logical Form, 44n11, 56, 61, 67, 113, 125, 251, 305, 307–308, 313n9, 326 merge, 93–95, 98–99, 102–103, 112, 114, 121, 143n29, 155–156, 292, 339, 350, 352n18, modal verb, 117, 139n8, 168, 183–184, 203–240, 364

376

Index

morphology, 1, 7, 109–110, 112–115, 122–128, 132–138, 141n14, 142n20, 143n29, 163, 226–229, 232, 237, 264, 294, 301, 313n13 movement, A’–movement, 6, 89–93, 100–103, 115, 120, 141n17, 326–337, 347, 365–366, 370 A–movement, 6, 90, 93, 115, 141n17, 178, 197n31, 306, 326–333, 337, 347, 352n11, 365–366, 370 Head Movement, 6, 123, 142n19, 164, 167, 174, 183–185, 275, 277, 289, 291, 293–294, 366–369 neo–saussurean, 245, 258–259, 264, 271n11 parameter, 8, 111, 121–128, 132, 138, 139n2, 141n18, 154–157, 233, 238, 257, 266, 268, 275–276, 285, 288, 294, 296–299, 302, 311, 314n16 Parametric Substantiation Hypothesis, 154, 156, 158–159, 161, 168, 175, 177–178, 189–191, 193–194, 197n27 Person, 7, 153–157, 175–194, 197n27, 197n28, 197n32, 197n34 Pesetsky’s paradox, 303–308, 317n24, 317n25 PF, see Phonological Form phase theory, 7, 109–117, 126–128, 132, 134–136, 138, 139n3, 297–298, 308, 332, 353n24, 365 Phonological Form, 93, 125, 158, 276, 296, 299, 301, 308, 315n19 Polish, 280 prosody, 4, 92, 259 remnant movement, 6, 58, 89–90, 105n10, 246, 252–254, 258, 292, 294, 330–339, 341–342, 347–349, 370–371

Representation Theory, 209, 370–372 Romanian, 277–278, 280, 282–283, 286, 290, 293, 295, 302–303, 309–310, 312n1, 313–314n13, 315–316n21, 369 Russian, 7, 36–42, 45–46 scope, 8, 19, 37–41, 44n11, 45–46n20, 58, 67–74, 78–79, 96–97, 248–249, 252–261, 266–268, 270n3, 282, 297, 304–308 scrambling, 15, 18–29, 39–41, 43n4–6, 45n18, 46n23, 66, 81n6, 87, 89–90, 93–94, 271–272n15, 315–316n21, 327–328, 332–337, 346, 351n6, 351n8, 352n10 second factor, 109, 111–112, 138, 139n2 Spanish, 81–82n9, 122, 127–130, 133–134, 142n21, 142n23, 287 Spell–Out, 137–138, 296, 298, 308, 361, 366 Swiss German, 230, 232, 237–238, 240n9 tense, 7, 133, 141–142n18, 153–165, 168, 170–176, 178–180, 190–194, 195n5, 196n23, 203–213, 272n18, tenselessness, 166 topic, 3, 6–8, 15–35, 41–42, 43n1, 43n7–8, 43–44n9, 44–45n15, 45n16, 46n21, 53–55, 69–71, 74–79, 122–124, 136, 140n10, 140n12, 263, 306, 309, 332–333, 337, 346, 352n17, 370 transitivity failures, 2, 8, 55, 350 UG, see Universal Grammar Universal Grammar, 153, 155–157, 189, 192–193, 272n15, 298, 311 Welsh, 279–281, 290, 313n12 word order paradox, 2, 6, 29, 55, 275, 277, 350