Syntax and Morphology Multidimensional 9783110238754, 9783110238747

Citation preview

Syntax and Morphology Multidimensional

Interface Explorations 24

Editors

Artemis Alexiadou T. Alan Hall

De Gruyter Mouton

Syntax and Morphology Multidimensional edited by

Andreas Nolda Oliver Teuber

De Gruyter Mouton

ISBN 978-3-11-023874-7 e-ISBN 978-3-11-023875-4 ISSN 1861-4167 Library of Congress Cataloging-in-Publication Data Syntax and morphology multidimensional / edited by Andreas Nolda, Oliver Teuber. p. cm. — (Interface explorations; 24) Includes bibliographical references and index. ISBN 978-3-11-023874-7 (alk. paper) 1. Grammar, Comparative and general — Syntax. 2. Grammar, Comparative and general — Morphology. 3. Structural linguistics. I. Nolda, Andreas. II. Teuber, Oliver. P291.S95624 2011 415 — dc23 2011028427

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. © 2011 Walter de Gruyter GmbH & Co. KG, 10785 Berlin/Boston Cover image: iStockphoto/Thinkstock Typesetting: Asco Typesetters, Hong Kong Printing: Hubert & Co. GmbH & Co. KG, Göttingen f» Printed on acid-free paper Printed in Germany www.degruyter.com

Contents

Introduction: Multiple dimensions in syntax and morphology Andreas Nolda and Oliver Teuber The Parallel Architecture and its place in cognitive science RayJackendoff A phono-syntactic template for Turkish: Base-generating free word order Ash Goksel A coanalysis approach to Polish past tense agreement BertholdCrysmann Optimal specifications: On case marking in Polish Bernd Wiese Case competition in Russian: Genitive vs. accusative and nominative. An integrational account Hans-Heinrich Lieb and Svetlana Friedrich How can the polysemy of syntactic categories be conciliated with semantic coherence? Syntactic and lexical factors for the emergence of a global signification of the imparfait in French Marie-Helene Viguier On lexical and syntactic licensing of category mixing TatianaNikitina

1

17

45

77

101

129

187

217

AUTO- and INTER- versus (?) SE: Remarks on interaction and competition between word formation and syntax KatrinMutz

239

The interaction between morphology and syntax exhibited by the German writing system NannaFuhrhop

259

Subject index

289

Introduction: Multiple dimensions in syntax and morphology Andreas Nolda and Oliver Teuber

The present volume collects papers that approach theoretical and empirical problems in syntax and morphology from a multidimensional point of view. In such an approach to syntax or morphology, syntactic phrases, morphological words and the like are conceived of as syntactic or morphological constructs with multiple, interrelated components, each representing morphosyntactic properties of different kinds. Thereby one can describe, say, hierarchical structure and linear order, or morphosyntactic categories and functions in their own right, without neglecting their interrelations. The aim of this introductory article is to clarify what we mean by "multidimensionality" and to provide an overview of the volume. We proceed as follows. Section 1 discusses the distinction between multidimensionality and monodimensionality. Section 2 exemplifies monodimensional approaches by a Government and Binding analysis. Section 3 characterizes multidimensional frameworks insofar as they are represented in this volume. Section 4 finally shows how the papers of the volume relate to its general topic.

1. Multidimensionality vs. monodimensionality Morphosyntactic objects such as syntactic phrases or morphological words are conceived of in different ways by different theoretical approaches. They are modeled as constituent structures, dependency structures, feature structures. etc. or as combinations thereof, i Assume, for instance, that in a certain theoretical framework morphosyntactic objects are modeled jointly by a constituent structure - representing, say, hierarchical structure, linear order, and phonological form - and a feature structure - representing morphosyntactic categories and functions. Then we shall take the pair consisting of the constituent structure and the feature structure to be a two-dimensional morphosyntactic construct. If. 1. Note that constituent structures, dependency structures, and feature structures can themselves be formalized in different ways, e.g., by set-theoretical or graphtheoretical means.

2 Andreas Nolda and Oliver Teuber however, all of those properties are represented by a constituent structure only. the latter is a one-dimensional morphosyntactic construct. In more general terms, then, morphosyntactic constructs are multidimensional if they are composed of several components, each representing morphosyntactic properties of (at least partially) different kinds. Otherwise morphosyntactic constructs are mono dimensional. By extension, we shall say that a theoretical framework is multidimensional itself if it models morphosyntactic objects by multidimensional constructs. Similarly, a framework is monodimensional if it models morphosyntactic objects by monodimensional constructs. From a metatheoretical point of view, monodimensional frameworks may appeal because of the uniformity and ontological parsimony of their morphosyntactic constructs. The number of their components is reduced to the bare minimum: one. By contrast, morphosyntactic constructs in multidimensional frameworks are more complex. They are composed of several components, that do not only represent different kinds of morphosyntactic properties, but often also differ with respect to their type (constituent structure, dependency structure, feature structure, etc.). In addition, multidimensional frameworks have to provide means for relating corresponding parts in different components of a given morphosyntactic construct. On the other hand, monodimensional frameworks tend to posit rather intricate structures for their morphosyntactic constructs because the latter have to represent, and distinguish, properties of various kinds. In multidimensional frameworks, however, the individual components of morphosyntactic constructs can have relatively simple structures, which are specifically adapted to the one or few kinds of morphosyntactic properties they represent. Another advantage of multidimensional frameworks is that different kinds of morphosyntactic properties can be described in their own right, such that mismatches between them are accounted for easily. Assume, for instance, that in a given multidimensional framework hierarchical structure and linear order are represented by different components of morphosyntactic constructs. Then discontinuous constituents can be directly represented (instead of being 'simulated' by movement operations or other technical devices). As a matter of fact, monodimensional frameworks are often multistratal ones. In a multistratal framework morphosyntactic objects are modeled by sequences of denvationally related constructs. In such a sequence all constructs are of the same type (e.g. constituent structures) and represent morphosyntactic properties of (at least partially) identical kinds. Typically, though not necessarily, they are assigned to different 'descriptive levels' or 'derivational stages' (such as 'deep' and 'surface structures'). Multidimensional frameworks, however, normally happen to be monostratal ones. Due to the multidimensional

Introduction 3 nature of their morphosyntactic constructs there arises no need for assuming more than a single'descriptive level'.

2. A monodimensional example A well-known monodimensional, and multistratal, framework is the Government and Binding (or Principles and Parameters) incarnation of Generative Grammar (Chomsky 1981, 1986). Ouhalla (1999: 136-137), for example, models clause (1) in this framework by a sequence of constituent structures with (2) as the S-structure (roughly, 'surface structure') member: (1)

{that) this problem, I can solve

(2)

Spec

/

can

V

V

NP

solve

ti

4

Andreas Nolda and Oliver Teuber

(2) is a monodimensional syntactic construct, representing at the same time hierarchical structure, linear order^ syntactic categories, and - to a certain degree - syntactic functions. In addition, it contains a coindexed trace ?,, linking (2) with the D-structure (roughly, 'deep structure') member of the sequence (another constituent structure, not given here). As a rule, syntactic objects are modeled by constituent structures with a root constituent of a phrasal category XP, and morphological ones by constituent structures with a root constituent of a lexical category X According to X-bar theory, the syntactic functions head, complement, adjunct, and specifier are relations between positions in local tree configurations. In a local tree of category X', for example, the relation of a YP daughter to an X daughter counts as a complement relation, while its relation to a n X daughter counts as an adjunct relation. In a local tree of category XP, in turn, the relation of a YP daughter to a n X daughter counts as a specifier relation, while its relation to anXP daughter counts again as an adjunct relation. In order to represent syntactic functions in an unambiguous way,3 X-bar theory has to assume relatively complex constituent structures. The nonbranching VP configuration in (2) is a direct consequence of representing syntactic functions by means of X-bar theoretic relations between constituent structure positions.* Regarding hierarchical structure (in terms of part-whole relations), the non-branching configuration is redundant: both the VP node and the V node stand for the same verbal constituent. But without the intervening V node between the VP and the contained NP, the syntactic function of the latter could not be determined by reference to X-bar theoretic principles. The joint representation of hierarchical structure, linear order, and syntactic functions by the same constituent structure can lead to further complexity, since Chomskyan Generative Grammar does not allow for discontinuous constituents. In (1) this problem is a complement (the direct object) of solve, that is 2. In the Minimalist Program of current Generative Grammar, syntactic structures do not directly represent linear order. According to Chomsky's (1995) 'bare phrase structure' conception, syntactic structures are unordered sets. Their linearization is delegated to the phonological component. 3. As a matter of fact, functional ambiguities can occur mX-bar theoretic tree structures despite their relative complexity. This is the case in adjunction configurations where mother and daughters are of the same category XP. Unless further principles are stipulated for the distinction between head and adjunct (such as then relative linear order; cf Kayne 1994), the direction of the adjunct relation cannot be umquely determined. 4. By giving upX-bar theory, the Minimalist Program avoids non-branching structures (cf. Chomsky 1995).

Introduction 5 topicalized for mformation-structural reasons. In order to represent the complement function of this problem in accordance wrth X-bar theory, solve and this problem would have to form a n X constituent wrth solve as an X daughter and this problem as a YP daughter. Due to the totalization of this problem, t h a t X constituent would be discontinuous, though. In the Government and Binding framework one resolves this sort of dilemma by representing syntactic functions on D-structure but not necessarily on S-structure; surface linear order, in turn. is represented on S-structure, but not necessarily on D-structure.

3. Multidimensional frameworks We shall now characterize selected multidimensional frameworks - those frameworks that are presupposed in papers of this volume. These are LexicalFunctional Grammar, Head-driven Phrase Structure Grammar, the Parallel Architecture, and Integrational Linguistics.^

3.1. Lexical-Functional Grammar Lexical Functional Grammar (LFG) in the original version of Kaplan and Bresnan (1982) models morphosyntactic objects by constructs containing a constituent structure (c-structure) and a functional structure (f-structure). Newer versions assume additional components such as argument structure (a-structure) (cl Bresnan 2001). Kaplan (1995: 10-11), for instance, models sentence (3) jointly by the c-structure (4a) and the f-structure (4b): (3)

I saw the girl.

5. Multidimensional frameworks that are not represented withm the present volume include Construction Grammar approaches such as Berkeley Construction Grammar (cl, inter alia, Fillmore 1999; Kay 2002).

6 (4)

Andreas Nolda and Oliver Teuber a.

S

the girl b. SUBJ

PRED 'pro' PERS 1

LNUM

SG J

TENSE PAST PRED 'see XM LDOM

LDOM

(gssg

|T]g

Sr^-llJ JJ J

7. In his version of linearization-based HPSG, Reape uses the features "SYN" and "ARG-DTRS" instead of "SYNSEM" and "COMP-DTRS", respectively. Boxed numerals indicate token-identical values.

Introduction

9

The PHON value of (10) now results from concatenating the PHON values of the domain objects in the DOM list, to which LP constraints apply in linearization-based HPSG. Note that domain objects may be immediate constituents or non-immediate ones. In the latter case (as in the top-level DOM list in (10) with the domain objects versucht and das Buck zu lesen), constituents can be discontinuously linearized. The classical approach to morphology in HPSG is an item-and-process one: feature structures modeling morphological objects (e.g. stems) are mapped onto feature structures modeling morphological or syntactic ones (stems or words) by lexical rules (Pollard and Sag 1987: chap. 8). For a recent overview. including also alternative item-and-arrangement approaches, cf Muller (2008! chap. 19).

3.3. The Parallel Architecture In the Parallel Architecture (PA) framework (Jackendoff 1997,2002; Culicover andJackendoff2005)thegrammaticalstructureofasyntactic(ormorphological) object is conceived as a triple containing a (morphojphonologwal structure, a (morpho)syntactic structure, and a semantic structure. Each structure can in turn be composed of several tiers. Example (12) outlines the grammatical structure of sentence (11) (cf. Culicover and Jackendoff 2005: 193): (11)

Pat gave Dan a book

(12)

a. b. c. d.

Pat2 gave, Dan, a book, [NP^vpVxNPsNPJJ, GF2>GF3>GF4 [GIVE (PAT2, DAN 3 , [BOOK; INDEFWh

(12a) gives the phonological structure of (11) in orthographic terms, neglecting its articulation into several tiers. (12b) represents the hierarchical structure of (11) in form of a headed syntactic constituent structure. The grammatical function tier (12c) ranks unspecified grammatical functions. (12d) formulates the conceptual structure of (11) (further semantic tiers are ignored). Each structure in (12) is constrained by a separate combinatorial component of the grammar with its own primitives and principles of combination. The

10 Andreas Nolda and Oliver Teuber primitives of the syntactic combinatorial component, which constrains syntactic constituent structures, are syntactic categories and features. Its principles of combination are principles of constituency and principles of linear order.* Correspondences between parts of different structures - expressed by numerical indices in (12) - are constrained by interface components of the grammar. The multidimensionality of the grammar architecture allows Culicover and Jackendoff (2005) to design a 'simpler syntax'. Syntactic constituent structures can be as flat as possible, since they do not represent, e.g., any syntactic functions beyond the head function.

3.4. Integrations! Linguistics Ever since the seminal work of Lieb (1983),' Integration^ Linguistics (IL) conceives structured syntactic objects as triples composed of * syntactic unit, a syntactic structure, and a lexical interpretation. Syntactic structures in turn are taken to consist of a constituent structure, a categonal marking structure, and ^intonation structure. (14) gives, in informal notations, the syntactic unit (14a), the syntactic intonation structure (14b), the syntactic constituent structure (14c), the syntactic marking structure (14d), and the lexical interpretation (14e) of sentence (13) (cf Nolda 2007: 153-183): (13)

Ichhabenur franzosischen Rotwein. I have only French red wine 'I only have French red wine.'

(14)

a.

1 ich

2 habe

3 nur

4 franzosischen

5 Rotwein

8. In addition to GPSG-style LP rules, Cuticover and Jackendoff (2005) make use of 'default' LP rules, specifying the default order of sister constituents, and 'edge' LP rules, linearizing constituents at the left or right edge of the mother constituent. 9. For recent introductions to the framework cf. Nolda (2007: chap. 7) and Sackmann (2008). The framework is presupposed, inter alia, by Eisenberg's (2006) German reference grammar.

Introduction 11 b. 1 L c.

3 H

Vf 2

Ptf 3

5 Ld L

Nf

Nf

4

5

1 PERS-PRON, 1PERS,. . . Nom, Smg Nf ,. . .

3 QUAL-W,. . . Unmptf

e.

4 LHrLL

VGr

Nf 1 d.

2 HL

1 •I-

2 haver

2 MAIN-V, NOM+ACC,. . . IPers, Smg vf , Pres,. . .

4 ADJ,. . . Ace, SmgNf, Masc,. . .

3 c°

4 Frenehr

5 MASS-N, MASC, . . . Ace, Smg Nf ,. . .

5 redwing-

From a formal point of view, all of those components are set-theoretical functions. Their domains consist of position numbers, or sets of such numbers, representing linear order and linking corresponding parts of different components. The syntactic unit (14a) is a function from positions to phonological words (notated here as orthographic words). The intonation structure (14b) associates the positions with one or several sets of auditive values (one set per syllable: only pitches are considered above). The constituent structure (14c) maps sets of positions to constituent categories such as Nf ('noun form', including substantival as well as adjectival word forms) or NGr ('noun group'). Note that the VGr ('verb group') ich habe franzosischen Rotwem is a discontinuous constituent in (14c), interrupted by the Ptf ('particle form') nur. For those position sets that are assigned basic constituent categories, the marking structure (14d) supplies further categorizations in terms of lexical word categories (e.g. ADJ) and

12 Andreas Nolda and Oliver Teuber word form categories (Ace etc.), while the lexical interpretation (14e) assign lexical meanings to them.io Putting the components in (14) together yields a multidimensional construct, that serves as a basis for specifying occurrences of syntactic functions (represented by arrows in (15)): ^VGr

(15) VGr-«—— _

\\ Nf 1 ich L •IPERS-PRON, 1PERS, Nom, Sing Nf ,...

\ \

\ '

tffi>^

Vf 2 habe HL •haver MAIN-V,NOM+ACQ. IPers, Singyf, Pres,...

'NGT

mod Ptf 3 nur H

c° QUAL-W,. Unmptf

Nf-' 4 fremzosischen LULL •French] • ADJ,... Ace, SingNf, Masc,

\

5 Rotwein LdL •redwiner MASS-N.MASC,. Ace, Sing Nf ,...

In the classical IL conception structured morphological objects were conceived in a strictly analogous way, effectively combining an item-andarrangement approach with elements of the word-and-paradigm tradition (for details, the reader is referred to Lieb 1983; for an application to German, cf Eisenberg 2006: vol. 1). Recently, it has been proposed to replace it by an itemand-process approach, though (Lieb 2006, 2010).

4. The papers in this volume The contributions in this volume partly derive from papers read at the workshop on "Syntax and morphology multidimensional" at the annual meeting of the Deutsche Gesellschaft fur Sprachwissenschaft in Bielefeld in 2006. All of them subscribe to a multidimensional point of view of syntax and morphology presupposing, for the most part, one of the multidimensional frameworks presented in Section 3. In "The Parallel Architecture and its place in cognitive science", Ray Jackendoff argues that multidimensional parallel architecture approaches in general.

10. In (14e), "c°" denotes the empty concept. In IL, the meaning of 'particle forms' such as nur is accounted for in sentence semantics, not lexical semantics.

Introduction

13

and his own Parallel Architecture framework in particular, integrate better with cognitive science than mainstream Generative Grammar. This claim is justified with reference to PA's conception of the lexicon, conceptual semantics, and syntax. A PA lexicon includes 'stored' linguistic entities of all kinds, including idioms, constructions, and phrase structure rules. Given the independently motivated complexity of PAs mentalistic conceptual semantics, syntax can be cut down to the bare minimum that is needed to map between phonology and meaning. Ash Goksel advances a monostratal, multidimensional approach to Turkish syntax that includes prosodic structure as a crucial part. In "A phono-syntactic template for Turkish: Base-generating free word order", she proposes a tripartite template for capturing general properties of linear order in Turkish sentences. The domains in this template are defined independently from hierarchical structure solely by reference to the positions of sentence stress and the predicative verb. It is shown that this proposal can be successfully applied to phenomena such as functional ambiguity, quantificational dependency, and ellipsis. In "A coanalysis approach to Polish past tense agreement", Berthold Crysmann discusses past tense and conditional agreement markers in Polish, that can syntactically 'float off' to hosts on the left of the verb. He argues that the former, but not the latter, are best analyzed as verbal affixes that are 'visible' for syntactic linearization. This idea is implemented in a generalized version of linearization-based HPSG that allows words to introduce more than a single domain object into order domains. Bernd Wiese shows in his paper "Optimal specifications: On case marking in Polish" that the traditionally assumed multiplication of paradigms and declensions for Polish nominal inflection can be avoided by hierarchical, 'multi-level' classification systems for gender and case. On their basis, Wiese identifies largely unambiguous categonal specifications of noun endings. The use of the latter is controlled by associated application conditions that specify restrictions in formal or functional dimensions. In their article on "Case competition in Russian: Genitive vs. accusative and nominative. An integrational account", Hans-Heinrich Lieb and Svetlana Friedrich investigate the semantics of the genitive case and its syntactic basis in Russian. Presupposing the framework of Integrational Linguistics, they identify semantic effects of the genitive for negation-independent competition with the accusative as well as for negation-dependent competition with both the accusative and the nominative. The semantic effects are derived by interpreting syntactic functions in terms of semantic functions. The application of the latter is controlled by application conditions (of different nature than those in Wiese's paper), which make reference to various components of multidimensional

14

Andreas Nolda and Oliver Teuber

syntactic constructs, including, but not limited to, the appearance of case categories in the marking structure. Semantic functions and their applications conditions also play a crucial role mMarie-Helene Viguier\ paper "How can the polysemy of syntactic categories be conciliated with semantic coherence? Syntactic and lexical factors for the emergence of a global signification of the imparfait in French". Viguier argues that the polysemy of the French imparfait is at least partly semantic, not pragmatic. She shows on the example of the non-past interpretation of the imparfait in conditional ./-clauses how such an account can be implemented in Integrational Linguistics by the formulation of a 'multidimensional' application condition for the corresponding semantic function, which here directly interprets a syntactic category. In her paper "On lexical and syntactic licensing of category mixing" Tatiana Nikitina provides an LFG account of conditions on grammatical function realization in deverbal noun constructions in Italian and Wan (a Mande language. spoken in Cote d'lvoire). Nikitina proposes to explain the (non-)blocking of grammatical functions in such 'mixed-category' constructions by distinguishing two independent, and possibly conflicting, licensing mechanisms for grammatical functions (encoded in f-structure) and for their syntactic realization (encoded in c-structure). The paper on "AWV- and INTER- versus (?) SE: Remarks on interaction and competition between word formation and syntax" of Katrin Mutz studies the 'division of labor' between syntax and morphology on the example of syntactic reflexive markers and their morphological counterparts in Romance languages. The comparison of their functions is cast in a multidimensional representation that includes 'levels' such as conceptual semantic structure, argument structure, and morphosyntactic realization structure. In her paper on "The interaction between morphology and syntax exhibited by the German writing system", Nanna Fuhrhop shows how the dimensions of morphology and syntax have independent impact on the spelling system of German. The discussion is led along two phenomena, capitalization of nouns and juxtaposition of compound words. It is shown that the graphematic system falls back on morphological principles as well as on syntactic principles. Cases of doubt arise from conflicts between these independent sets of principles.

References BresnatUoan 2001 Lexical-Functional Syntax. Oxford: Blackwell.

Introduction

15

Chomsky, Noam 1981 Lectures on Government and Binding: The Pisa Lectures. Dordrecht: Foris. Chomsky, Noam 1986 Barriers. (Linguistic Inquiry Monographs 13.) Cambridge, Mass, MIT Press. Chomsky, Noam 1995 Bare phrase structure. In Evolution and Revolution in Linguistic Theory. Hector Campos and Paula Kempchmsky (eds.), 51-109. Washington: D.C, Georgetown University Press. Cuhcover, Peter W., and Ray Jackendoff 2005 Simpler Syntax. Oxford: Oxford University Press. Eisenberg, Peter 2006 Grundriss der deutschen Grammatik. 3rd ed. 2 vols. Stuttgart: Metzler Fillmore, Charles 1999 Inversion and constructional inheritance. In Lexical and Constructional Aspects of Linguistic Explanation, Gert Webelhuth, Jean-Pierre Koemg, and Andreas Kathol (eds.), 113-128. Stanford: CSLI Publications. Gazdar, Gerald, Ewan H. Klein, Geoffrey K. Pullum, and Ivan A. Sag 1985 Generalized Phrase Structure Grammar. Oxford: Blackwell Jackendoff, Ray 1997 The Architecture of the Language Faculty. (Linguistic Inquiry Monographs 28.) Cambridge, Mass, MIT Press. Jackendoff, Ray 2002 Foundations of Language: Brain, Meaning, Grammar, Evolution. Oxford: Oxford University Press. Kaplan, Ronald M 1995 The formal architecture of Lexical-Functional Grammar. In Formal Issues in Lexical-Functional Grammar, Mary Dalrymple, Ronald M. Kaplan, John T. Maxwell, and Annie Zaenen (eds.), 7-27. (CSLI Lecture Notes 47.) Stanford: CSLI Publications. Kaplan, Ronald M , and Joan Bresnan 1982 Lexical-Functional Grammar: A formal system for grammatical representation. In The Mental Representation of Grammatical Relations, Joan Bresnan (ed.), 173-281. Cambridge, Mass, MIT Press. Kay, Paul 2002 An informal sketch of a formal architecture for Construction Grammar. Grammars 5: 1-19. Kayne, Richard S. 1994 The Antisymmetry of Syntax. (Linguistic Inquiry Monographs 25.) Cambridge, Mass, MIT Press. Lieb,Hans-Heinrich 1983 Integration* Linguistics. Vol. 1. (Current Issues in Linguistic Theory 17.) Amsterdam: Benjamins.

16

Andreas Nolda and Oliver Teuber

Lieb,Hans-Hemnch 2006 Wortbildung auf morphologischer und Wortbildung auf syntaktischer Grundlage. Paper presented at the 28th Jahrestagung der Deutschen Gesellschaft filr Sprachwissenschaft in Bielefeld on February 23. Lieb,Hans-Hemnch 2010 Word formation: morphological and syntactic: A unified process view based on a Word and Paradigm model. Ms., Freie University Berlin. Miller, Stefan 2008 Head-driven Phrase Structure Grammar: Fine Einfuhrung. 2nd ed. (StauffenburgEmfilhrungen 17.) Tubingen: Stauffenburg. Nolda, Andreas 2007 Die Thema-Integration: Syntax und Semantik der ,gespaltenen Topikalisierung' im Deutschen. (Studien zur deutschen Grammatik 72.) Tubingen: Stauffenburg. Ouhalla,Jamal 1999 Introducing Transformational Grammar: From Principles and Parameters to Minimalism. 2nd ed. London: Arnold. Pollard, Carl, and Ivan Sag 1987 Information-based Syntax and Semantics. Vol. 1: Fundamentals. (CSLI Lecture Notes 13.) Stanford: CSLI Publications. Pollard, Carl, and Ivan Sag 1994 Head-driven Phrase Structure Grammar. Stanford: CSLI Publications and Chicago: The University of Chicago Press. Reape,Mike 1992 A formal theory of word order: A case study in West Germanic. Ph. D. Diss., University of Edinburgh. Reape,Mike 1996 Getting things in order. In Discontinuous Constituency, Harry Bunt and Arthur van Horck (eds.), 209-253. (Natural Language Processing 6.) Berlin: Mouton de Gruyter. SackmannRobm 2008 An introduction to Integration^ Linguistics. In Explorations in Integrational Linguistics: Four Essays on German, French, andGuarani, Robin Sackmann (ed.), 1-20. (Current Issues in Linguistic Theory 285.) Amsterdam: Benjamins.

The Parallel Architecture and its place in cognitive science* Ray Jackendoff

It has become fashionable recently to speak of linguistic inquiry as brolmgwstics, an attempt to frame questions of linguistic theory in a biological context. The Minimalist Program (MP; Chomsky 1995, 2001) is of course the most prominent stream of research in this paradigm. However, an alternative stream within the paradigm, the Parallel Architecture, has been developing in my own work over the past 3 0 years; it includes two important subcomponents, Conceptual Structure and Simpler Syntax (Jackendoff 2002, 2007b; Culicover and Jackendoff 2005). The present article will show how the Parallel Architecture is in many ways a more promising realization of biolinguistic goals than the Minimalist Program, and that it is more conducive to integration with both the rest of linguistic theory and the rest of cognitive science. 1. Parallel architectures, broadly conceived The Parallel Architecture can be explored at two levels: First, what is a parallel architecture in general? Second, what distinguishes "the" Parallel Architecture from other theories within this genre? In both cases, the basic question is: (1)

What is the best way to allocate the generative capacity of language, so as to account for the observed relations between sound and meaning?

Traditional generative grammar, from Syntactic Structures (Chomsky 1957) through the Minimalist Program, has answered: (2)

*

Syntactocentric architecture: The recursive rules of the syntactic component provide the generative capacity of language. The relation between sound and meaning arises through mapping syntactic structures into phonetic form (PF) (or the "sensorimotor interface") on one hand and logical form (LF) (or the "conceptual-intentional interface") on the other. This chapter was published previously in The Oxford Handbook of Linguistic Analysis, Bernd Heme and Heiko Narrog (eds.), 583-605, Oxford: Oxford University Press. We thank OUP for permission to include this chapter here. (The editors.)

18

RayJackendoff

However, theoretical developments as early as the 1970s showed that phonological structures have then own primitives and principles of combination that cannot be reduced to syntactic terms. For instance, rules of syllabification, prosody, and morphophonology are stated in terms of units that are thoroughly comfortable in phonological terms but often quite uncomfortable in syntactic terms. The same is true of meanings: semantic notions like event, manner, quantification, and focus cut across syntactic categories but are charactenzable in independent semantic terms. In particular, it has been a staple of linguistic theory and psycholinguists to distinguish semantic ill-formedness ^Colorless green ideas sleep furiously) from syntactic ill-form edness (M good ideas am rare), which depends on the distinction between semantic and syntactic combinatorial^ Within the syntactocentnc approach, mismatches between syntax and phonology have been either incorporated into syntax (e.g. "Spell-Out") or ignored. at least by syntacticians. Moreover, there has been a constant tendency to account for apparent mismatches between syntax and semantics by building more and more aspects of semantics into syntactic structure - as is required by an architecture in which all combinatonality is a consequence of syntax. The inevitable result is a syntactic component rich with invisible structure, in which many constituents move multiple times. Differences of opinion arise as to whether this is a Good Thing or not. A parallel architecture answers question (1) like this: (3)

Parallel architecture: a. The generative capacity of language is invested in multiple components - at the very least, autonomous generative components for phonological. syntactic, and semantic structure. Each component has its own distinctive primitives and principles of combination, and generates its own structure. b. The relation between sound and meaning is mediated by a set of interface components, which establish optimal linkings among the various structures and their parish c. The structure of a sentence is therefore an «-tuple of structures, one for each appropriate component, plus the linkages established among them by the interface components.

1. "Syntactic" is sometimes used to mean 'combinatorial in any sense', including music, phonology, and semantics. I am using the term here in the narrow sense of 'combinatorial^ whose units are things like Noun and Verb.' 2. Note that in this theory, an interface is not a level of structure, but rather a connection between/wo levels of structure.

The Parallel Architecture and its place in cognitive science

19

The main points of (3) can be illustrated by the following "W-model" of the language faculty. (4)

Phonological Formation Rules

1

1

Phonological Structures

\

Syntactic Formation Rules

Syntactic Structures

Semantic Formation Rules

1 Semantic Structures

*• Interface

Interface A prion, the traditional approach (2) seems simpler, since it has only one "generative engine" and fewer components overall. But, to paraphrase Chomsky (1972, 125-129) (who is responding to Postal's (1972) "The Best Theory" without citing it), architectures must be judged not only on their formal elegance, but also on their affordance for describing the data of language in full detail (descriptive adequacy), in explaining language variation and the possibility of language acquisition (explanatory adequacy), and in explaining how the system can arise from more general cognitive and biological principles ("beyond explanatory adequacy", to use the term of Chomsky 2001). In particular, formal elegance must not be conflated with biological or cognitive elegance. which might be quite different. Pursuing the goal of going beyond explanatory adequacy, consider which sort of architecture conforms more closely to what is known about the brain. For instance, the visual system is known to contain numerous areas specialized to different aspects of visual perception: detection of contour, detection of motion, detection of color, several independent mechanisms for constructing the perception of depth, possibly face perception, and many many others. Each of them accounts for a relatively limited aspect of visual understanding; the totality of visual understanding arises from their combined efforts. In order for their efforts to combine, they must communicate, linking their respective representations in an optimal fashion (Koch 2004). At the moment, we don't know a lot about the formal details of representations computed by various visual areas,

20

RayJackendoff

and there is still much dispute about what brain areas are responsible for different aspects of linguistic understanding. Nevertheless, the overall flavor of the visual system is far more compatible with a parallel architecture, with its multiple independent but linked components, than with a syntactocentnc one. There seems to be no part of visual perception/cognition from which all other parts can be derived; rather each part has its own special character. There is one cognitive capacity other than language for which formal details of the representations have been explored in some detail: music. Here it proves impossible to generate musical structures from a single component. Lerdahl and Jackendoff (1983) (see also Jackendoff and Lerdahl 2005) develop a parallel architecture for music containing four components linked by interface rules. One of these structures, grouping, is a general-purpose cognitive capacity that also plays an important role in vision. Another, metrical structure, bears strong similarities to the metrical systems that determine stress and prosody in language. The other two structures are, so far as we know at the moment, particular to music. One of my original motivations for a parallel architecture in language (Jackendoff 1997, 2002) was the existence of multiple independent tiers in phonology, such as syllabic structure, metrical structure, prosody, and tone, also linked by correspondence or interface rules. Similarly, it is now fairly clear that semantic structure can be dissected into semi-independent structures - at least propositions structure (who did what to whom) and information structure (topic vs. focus vs. common ground). Finally, the relation of language to vision. such that we can talk about what we see, has to be mediated by a set of principles that link linguistic representations of some level to visual representations of some level - it cannot be accounted for through further derivation from syntactic structure (Jackendoff 1987). Thus a parallel architecture can easily be scaled down to the internal structure of individual components, and scaled up to the relation of language to other faculties. A parallel architecture for language and other cognitive faculties displays a version of modularity. This is not modularity in Fodor's (1983) sense, which seals off various capacities from each other, but what could be called representational or structure-based modularity. Each separate form of representation has its own particular autonomous (i.e. domain-specific) structure, and its own interfaces to other structures. One form of representation is relatively informationally encapsulated from another to the degree that they can influence each other only through a series of interfaces, or through a narrowly specialized interface. For example, phonological structure is relatively encapsulated from visual representations, because in order to speak about what one sees, one has to pass from high-level visual understanding through linguistic semantic

The Parallel Architecture and its place in cognitive science

21

structure and syntactic structure in order to influence phonology - i.e. through a series of interfaces. However, there is also a narrowly circumscribed visionto-phonology interface that subserves reading, and this aspect of vision is rather tightly yoked to phonology. (For more detail, see Jackendoff 1987. chapter 12; 2002, section 7.5.) In short, the spirit of parallel architectures is in overall accord with what is known about (a) the brain, (b) the structure of other cognitive capacities, (c) the interior structure of linguistic components, and (d) the interaction of language with other cognitive capacities. The syntactocentnc architecture, including the Minimalist Program as one realization, is not* Many different theories of grammar employ parallel architectures in this broad sense. As noted above, phonological theory since the mid-1970s has been thoroughly parallel in conception. Among syntactic theories, the most prominent parallel architecture is Lexical-Functional Grammar (Bresnan 2001), where the work of syntax is divided between f-structure, c-structure, and the interface between them. Autolexical Syntax (Sadock 1991) has parallel components for morphosyntactic structure and phrasal syntactic structure, with the possibility of further subdivision. Role and Reference Grammar (Van Valin and LaPolla 1997) subdivides syntax into morphosyntax and phrasal syntax, and semantics into propositions and information structures, with interfaces running in all directions. Construction Grammar (Fillmore 1988, Goldberg 1995) is not formally laid out as a parallel architecture, but it acknowledges the independence of semantics from syntactic form, in that it emphasizes the many-many mapping between syntactic form and meaning, possible only if semantics is autonomous. And the granddaddy of them all is Stratificational Grammar (Lamb 1966), which decomposes the entire grammar into a long sequence of autonomous levels linked by interface components. Another fundamental question in the architecture of grammar is this: (5)

What formal operations are employed in building linguistic structure?

The mainstream architecture - along with Tree-Adjoining Grammar (TAG. Joshi 1987) - gives the following answer: (6)

Derivation-based generation: Syntactic trees are built algonthmically, either from the top down (as in pre-Minimalist theories) or from the bottom up (as in MP and TAG), and

3. An advocate of Minimalism might respond that this issue is one of performance or implementation, and so this sort of evidence is not pertinent to Minimalist inquiry. I would consider such a response simply a rhetorical avoidance of the evidence.

22

RayJackendoff

they undergo a sequence of distortions (movements and deletions) to derive sound and meaning. In parallel architectures, the interface relation between different components cannot be a sequenced derivation, since structures in different components often stand in a many-to-many relation. Rather, the interface components must be treated as constraints (possibly violable), which establish (or license) well-formed links among different kinds of structure. In principle, the rules responsible for each individual component of structure could be algorithmic. But in practice, almost all parallel architectures I have encountered have utilized a constraint-based formalism, in which each independent structure is licensed by simultaneously applied constraints.* To sum up, the answer to question (5) is (7). (7)

Constraint-based generation: The structures of each component are licensed by simultaneously applied component-internal constraints. The relationships among structures in different components are licensed by interface constraints.

Thus a parallel derivation has no notion of logical sequence, as is essential in a syntactocentric derivation. This has consequences for the relation of linguistic theory to theories of processing, as we will see in the next section.

2. The Parallel Architecture: The lexicon Having settled on a parallel architecture, the more specific question is: What are the autonomous representational formats, and what are the interfaces among them? What I have been calling "the" Parallel Architecture (in capitals. or PA) incorporates specific proposals about semantics, phrasal syntax, and the interface between them, plus less specific proposals about morphology and phonology. A leading question in the Parallel Architecture is the structure of the lexicon. The question is stated in essentially psycholinguists terms: (8)

What linguistic material does a speaker have to store in memory - i.e. What is in the lexicon? What structures can be built online in the course of speaking and understanding?

4. An exception is Synchronous TAG (Shieber and Schabes 1991).

The Parallel Architecture and its place in cognitive science

23

Traditionally, the lexicon is thought of as consisting of words (or morphemes), a distinct component of the language from the rules of grammar. Thinking in terms of question (8) leads to quite a different conception, as we will now see. A typical word - in any theory - is a triple of phonological, syntactic, and semantic information. In syntactocentric theories, a word is inserted into a syntactic derivation (by lexical insertion or Merge), and it is carried through the derivation to the points where its phonological and semantic properties are "read off." In the Parallel Architecture, the picture is quite different. The structure of a word suits it perfectly to function as a part of the interface components: It establishes a correspondence between small chunks of phonological, syntactic, and semantic structures. The correspondences among larger chunks are accomplished by other interface rules. There is no "point in the derivation" where a word is inserted. Rather, one can think of the word being "inserted" into all three structures at the same time. along with the links among them. Or one can think of the word as licensing the connection among preexisting structures. Alternatively, one can think in terms of processing. Given a perceived phonological structure, the word licenses the building of a connection to corresponding pieces of syntactic and semantic structure; given a piece of meaning to be expressed, the word licenses connecting it to appropriate pieces of syntactic and phonological structures. This last view suits PA to serve directly as a component of a theory of sentence processing (Jackendoff 2002, chapter 7; 2007a). PA itself is nondirectional, but its constraints can be implemented in an order suited to particular processing tasks. One aspect of the information coded in a lexical item is its contextual restrictions. Syntactic contextual restrictions include subcategonzation features on syntactic arguments; semantic contextual restrictions include selectional restrictions on semantic arguments. Often these two are partly redundant with each other, but not always (see Jackendoff 2002, section 5.9). Not every word has to connect all three components. English contains a small collection of "defective" words such as (9a). These have phonology and meaning, but no syntactic properties that allow them to combine into larger phrases (aside from within direct quotes, where anything at all is allowed). There are also a few words that have phonological and syntactic properties but no meaning, such as (9b). (9)

a. Phonology and meaning, no syntax: hello, ouch, upsy-daisy, allakazam, wow, shhh, gee whiz,. . . b. Phonology and syntax, no meaning: do (Jo-support), it (pleonastic), o/(N o/NP)

24

RayJackendoff

A lexicon conceived in terms of question (8) must contain more than single words. Most obviously, it must contain the thousands of idioms and other fixed expressions in the language such as (10), all of which are units known by native speakers. (10)

a. Idioms: kick the bucket, a breath of fresh air, right on the money, the jig is up. day in day out, clean as a Whistle,pie in the sky,... b. Fixed expressions (cliches, etc.): baby-blue eyes, home sweet home, take it from me, weapons of mass destruction, no money down, leave a message at the tone,. . .

Including these items in the lexicon (as they must be - where else would they be in the language?) leads to two important conclusions. First, lexical items cannot be conceived of as syntactic atoms, since many items in (10) have internal syntactic structure. Kick the bucket is a transitive VP. clean as a whistle is an AP with a comparative complement, weapons of mass destruction is a complex NP, and so on. Thus they cannot be inserted by a process like MP's Merge, which builds structure out of syntactic atoms. However. treated as interface constraints, they pose no problem: they simply link a complex syntactic structure with an idiosyncratic meaning. (This approach is shared withHPSG.) Second, the lexicon cannot be conceived of as a nonredundant list of exceptions, as Chomsky has often asserted (citing Bloomfield). The lexical item weapons of mass destruction contains four independently attested words, meaning exactly what they ought to mean. It adds the information that these four form a known unit, and adds some extra meaning or connotation. It is impossible to extract the redundant information, leaving only the extra information. and end up with something that is formally coherent. The conclusion is that the lexicon is full of redundancy. In terms of formal elegance this is less than satisfactory, but it is where the facts urge us. In terms of "brain elegance", though. it seems entirely in line with the rest of the brain, which seems to favor redundancy where possible, in the interests of more reliable memory and processing. In addition to items such as (10) that are larger than a word, the PAs lexicon also contains items that are smaller than a word. For example, the regular plural suffix -z/-s/-3z in English establishes a correspondence between a piece of phonology, a syntactic feature, and a piece of meaning. Its contextual restrictions state that it is to be affixed to a noun (syntactic context) that is count (semantic context); the conditions for its allomorphy depend on its phonological context. It can be affixed to a noun of any phonological shape, including novel ones (as

The Parallel Architecture and its place in cognitive science

25

in the wugs test). Thus its manner of combining with its host is formally no different from the way a transitive verb combines with its object, except that it combines below the word level rather than at the phrasal level. On the other hand, irregular plurals (oxen, women, axes, etc.) have to be learned individually and therefore have to be stored in the lexicon. Formally. they are semantically and syntactically composite, but phonologically unitary: They are therefore parallel in structure to idioms, which are phonologically and syntactically composite but not semantically composed of the meanings of their morphemes. We can therefore think of these cases as "morphological idioms.'* The treatment of regular inflectional morphology as lexical items extends easily to other regular morphological phenomena, including unusual ones. For instance, English expletive inflation (manu-fuckm-facturer) is a stored morpheme with a distinct (non-truth-conditional) meaning, and can be affixed to any syntactic category. Its main contextual restriction is prosodic. Similarly. reduplicative morphemes have meanings and syntactic contextual restrictions just like any other affix, but their phonological shape is listed in the lexicon as a sort of binding: "Copy such-and-such a part of the word I'm attached to" (Ghomeshi et al. 2004). PA's treatment of regular morphology parts company here with "lexicalist" theories such as LFG and HPSG (Pollard and Sag 1994), which derive morphologically complex words "in the lexicon", "prior to" inserting them into sentences. In PA, both phrasal grammar and morphology contain processes of free combination that can be used online, and both also include lexically listed "prefabs" (idioms and irregular morphological combinations respectively). The difference between phrasal grammar and morphology is only that the units and principles of combination for phrases are in part different from those for words. In this framework, LFG's notion of Lexical Integrity amounts to the claim that the two sets of principles do not interact, except through inflectional morphology. PA's lexicon also incorporates the insight of Construction Grammar that certain pieces of syntax can carry idiomatic meaning, with or without overt morphemes that mark the constructional meaning. Some of these constructional idioms have ordinary syntax, for instance the VP constructions in (11); others, such as (12), have unusual syntax ("syntactic nuts" in the sense of Culicover 1999). (11)

a. joke your way into the meeting (V Pro's way PP = 'go PP while/by V-ing')

5. There may of course be subregulanties among irregular forms, but we set this aside for purposes of the present article; see Jackendoff (2002, sections 6.2-6.4).

26

RayJackendoff

b. rumble around the corner (V PP = 'go PP in such a way as to make a V-ing sound') c. knit the afternoon away (V NP away = 'spend NP[time] V-ing') d. paint me a picture (V NPX NP 2 = 'V NP 2 for the benefit of M Y ) (12)

a. b. c. d.

The more you eat, the fatter you get One more beer and I'm leaving How about some lunch? student after student

(the more S, the more S) (one more X and S) (How about HP?) (NPN)

Each of these constructions is listed in the lexicon as a linking between a syntactic complex and a meaning; some parts of the syntactic complex may be linked also to phonology (e.g. way). The syntactic variables in these constructions correspond to semantic variables in the usual way, and the constructions can therefore be combined with other items to form a sentence in exactly the same way as words and other idioms are. However, notice that the verbs in (11 a-c), though they are syntactic heads, serve semantical^/ as manner or means modifiers. Since the lexicon contains linked phonological, syntactic and semantic complexes, nothing in principle prevents it from also containing phonological and syntactic complexes that are not inherently linked to anything. For example, a "generative" phrase structure rule such as (13a) - which, as part of one's knowledge of English, must be stored in memory somehow - can also be stated as a "treelet" (13b), a syntactic complex that constrains possible syntactic structures. PA treats it as a stored piece of structure; it can therefore be localized in the lexicon alongside semantically and phonologically linked VPs such as kick the bucket. (13)

a. V P - V - N P b. [ V pVNP]

Thus autonomous principles of syntax - fixed head position, the availability of ditransitive constructions, the means for forming relative clauses, and so on are stated in precisely the same format as constructional idioms, and they therefore belong in the lexicon as well. In phonology, one can view syllable structure rules as lexical entries that specify pieces of autonomous phonology. The upshot is that there is no principled distinction between words and rules of grammar. Both are stored pieces of structure, lying at opposite ends of a

The Parallel Architecture and its place in cognitive science

27

multidimensional continuum of idiosyncrasy and regularity. This conclusion has been arrived at within HPSG, Cognitive Grammar (Langacker 1987), and Construction Grammar as well as PA, in each case through attention to a multitude of intermediate cases such as idioms and constructions. Mainstream generative grammar, partly because of its algorithmic formulation, has followed traditional grammar in making a strong lexicon/grammar distinction. This has made it difficult to assimilate idioms and constructions into the theory, resulting in loss of descriptive adequacy. In pursuit of explanatory adequacy, the MP has arrived at the conjecture that there is actually only one rule of grammar, Merge, and that all differences among languages are localized in the lexicon (Chomsky 2001); this conjecture has not proven as simple in execution as in principle (particularly since MP has no theory of the organization of the lexicon!). Within PA, HPSG, and Construction Grammar, the counterpart of this conjecture is quite straighforward. All words and all rules of grammar are pieces of structure stored in the lexicon. The only "procedural" part of language is the fundamental operation of Unification (Shieber 1986), which assembles pieces of structure. Merge proves to be a special case of Unification: it combines two given elements with a piece of tree structure. Unification can be generalized to combinatorial cognitive capacities other than language, thus better satisfying the goal of "beyond explanatory adequacy." For example, in vision it can be used to integrate evidence for depth perception from disparate sources. It can also be used to weld lyrics to music in building songs. Merge cannot perform either of these functions. If Unification is a general brain mechanism for achieving combinatonality, it should be no surprise that language uses it too. (See Jackendoff 2008 for discussion of Merge vs. Unification.)

3. Conceptual Semantics To work out any version of a parallel architecture, it is necessary to have theories of the individual components and the interfaces among them. Unlike other parallel architectures in the literature, and unlike mainstream linguistic theory. PA is grounded in a highly articulated theory of semantics, Conceptual Semantics, that answers to the concerns of the biolinguistic perspective and that also offers considerable (and continually increasing) empirical coverage. There is space here only to list some of the highlights of the theory. First, Conceptual Semantics (like Cognitive Grammar) is thoroughly mentalistic: it is a theory of the information in a language user's mind/brain that is

28

RayJackendoff

involved in understanding utterances, connecting them to perceptual evidence. and making inferences. It recasts the traditional philosophical concerns with reference and truth in mentalistic terms: (14)

a. Traditional formulation: i. A phrase P refers to an entity E in the world (or in a possible world). ii. A sentence S is true if it meets conditions Ch . . . , Cn in the world. b. Mentalistic formulation: i. A language user LU understands a phrase P to refer to an entity E in the world as LUconceptualizes it. n. LU judges a sentence S true if S meets conditions Q , . . . , C„ in the world as LU conceptualizes it.

The seeming objectivity of language, stressed by traditional philosophy of language, is a consequence of language users sharing a common (or near-common) conceptualization of the world, so that agreement can largely be taken for granted (Jackendoff 1983; 2002, chapters 9 and 10). Second, Conceptual Semantics recognizes that many aspects of one's conceptualization of the world are independent of language. For instance, one can understand much of the behavior of physical objects ("naive physics") without any language at all. Decades of research on child development, linguistic and nonlinguistic, have shown that prelinguistic children bring a rich toolkit to the task of understanding the physical world, and that this understanding serves as a foundation for learning word meanings (e.g. solving Quine's gavagai problem). Thus the view of meaning espoused by Conceptual Semantics offers the potential of explanatory adequacy, i.e. helping to explain the innate basis from which children acquire lexicons (now including rules of grammar). It also appears that other primates - especially apes - negotiate the physical world in much the same way we do; humans differ only in being able to talk about it. This provides an evolutionary underpinning for the semantic system of language: our ancestors had thoughts - as it were, things to talk about - before they could talk. This view of meaning, then, helps satisfy the goal of "beyond explanatory" adequacy: it helps explain why (some part of) the semantic system of language is the way it is, because it is built upon pre-existing primate cognition. Recall that within the mainstream architecture, the combinatorial properties of the "conceptual-intentional interface" arise through derivation from the syntactic component. On the face of it, this amounts to the claim that babies and apes cannot think combinatorial^. It is possible to read certain passages of

The Parallel Architecture and its place in cognitive science

29

Chomsky as endorsing such a claim, for example, "unbounded Merge provides only a language of thought, and the basis for ancillary processes of externalization" (Chomsky 2009). This acknowledges the combinatorial character of thought, but it still does so in syntactocentnc terms. The basic units of Chomsky's 'language of thought' are NPs and VPs; and Merge, the capacity for combinatonality, is said to have arisen in the course of human evolutionary divergence from other primates. To my knowledge, this position has not been defended against the copious literature on primate intelligence. In PA, by contrast, the 'language of thought' is the combinatorial system in terms of which one understands the world. Its units are entities such as objects, events, properties, and trajectories. NPs and VPs are part of the combinatorial system of (narrow) syntax, which plays a role in the mediation between thought and sound, that is, as part of what Chomsky calls "processes of externalization." PA takes the combinatorial system of meanings to be universal (though use of the system can be biased by the means of expression if "Whorfian" effects prove to be genuine). It is just that meanings are not made of syntactic units. This approach is possible precisely because of the fundamental assumption of PA that language - and the mind in general - utilizes multiple sources of combinatonality. A third important aspect of Conceptual Semantics, again drawing on the Parallel Architecture, is that the system of meaning or 'language of thought' is itself bifurcated into at least two linked combinatorial systems, Spatial Structure and Conceptual Structure, each of which is responsible for part of the conceptualization of the world (with or without language). The representations of Spatial Structure are quasi-geometnc or topological in character. For a first approximation, Spatial Structure might be thought of as the highest level of the visual system, representing objects in terms of their detailed shape. However, it must be able to encode shapes in a perspective-independent fashion, so that they can be recognized from any distance and angle (i.e. accomplish object constancy). It must be possible also to represent objects schematically, so that, say, the action of sitting can be represented in terms of a generic or schematic human figure rather than a specific person. In fact, though, Spatial Structure is not exclusively visual: it can also code shape and configuration that has been derived haptically (sense of touch) and propnoceptively (body sense), and both of these can be compared and integrated with visual input. Thus Spatial Structure is more abstract and general than a visual image - it is conceived of as a central level of cognition that codes the physical world in a relatively modality-independent fashion. It plays a role in word and sentence meanings, in that it encodes what things and events look like, and it enables us to talk about what we see.

30

RayJackendoff

The second major division of meaning, Conceptual Structure, is an algebraic structure built up in terms of discrete features and functions. It encodes distinctions that cannot be represented in the geometnc/topological format of Spatial Structure, such as those in (15), which are not part of what an object or event looks like. (15)

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

the type-token distinction, distinguishing categories from individuals taxonomic relations: X i s an instance/subtype of 7 ' temporal relations: X i s past/future' causal relations: Xcauses 7 ' , Xenables F , Ximpedes F , . . . modal notions: X i s hypothetical/nonspecific/potential/nctional. . .' social notions: X i s the name of F , X i s dominant to F , X i s kin to/ friend of F , X i s member of group Z\ X o w n s F , X i s obligated to perform act F , 'action 7 is of normative value Z \ . . . g. theory of mind notions: Xbelieves 7 ' , Ximagines 7 ' , Xintends 7 ' . X i s committed to norm 7 ' , . . .

The overall architecture looks like this: (16)

Haptic ^

Phonological ^ Structure

w Syntactic ^ Structure

w Conceptual ^ Structure

^ Spatial ^ Structure w

System ^

Visual System

N Proprioceptive Systems LANGUAGE

PROPER

CENTRAL

COGNITION

Conceptual Semantics takes it that word meanings must be potentially composite in order to encode relations among word meanings and in order to state properly general rules of inference. On the other hand, it differs from classical views of word meaning in admitting conditions other than necessary and sufficient. For instance, the conditions for color words must be encoded in terms of relative distance from central instances. In judging a hue between focal red and focal orange, two such conditions come into competition, and the judgment is therefore variable and to some degree context-dependent. In addition, many word meanings contain multiple conditions interacting in "preference rule" fashion. For instance, stereotypical climbing involves moving (i) upward, (ii) in a clambering fashion. But one can climb down a tree (clambering but not moving upward), and an airplane can climb into the clouds

The Parallel Architecture and its place in cognitive science

31

(moving upward but not clambering). On the other hand, an airplane cannot climb down out of the clouds, because such motion is neither upward nor clambering. In other words, neither condition is necessary, either may be sufficient, and stereotypical cases satisfy both. This type of rule interaction produces socalled "cluster concepts", of which Wittgenstein's (1953) example of game is the most famous. These characteristics of word meanings, even if unusual according to standard philosophical preconceptions, are totally normal within the context of brain computation. As been observed since the gestalt psychologists of the 1920s (Wertheimer 1923), conditions based on central instances and rule interactions with the characteristics of preference rules are standard in vision. They also appear in phonetic perception and in musical cognition, and essentially anyplace that multiple factors can either combine or conflict in producing a judgment. Conceptual Semantics differs from most theories of semantics (but again. not from Cognitive Grammar) in that it denies a sharp division between linguistic meaning and encyclopedic meaning (or "knowledge of the world"). Every division that has been proposed turns out to eviscerate linguistic meaning to the point where it cannot serve as a basis for inference (see Jackendoff 2002, sections 9.6-9.7; Bolinger 1965; Langacker 1987; Levinson 2000). A related point is that "semantics" and "pragmatics" do not involve distinct representations. Rather, there is a pair of mental representations, Conceptual Structure and Spatial Structure, that are the locus of sentence understanding. Some aspects of these representations may come from the words in the sentence and their grammatical configuration; we may call these parts "semantic". Other aspects come from nonlinguistic sources such as perception, inference. and "world knowledge"; we may call these parts "pragmatic." But these parts are often intricately interwoven in the representation in such a way that one cannot do the "semantics" first and paste in "pragmatics" afterward. In Conceptual Semantics, the taxonomy of concepts ('a poodle is a kind of dog', 'a dog is a kind of animal', etc.) grounds out in a fundamental ontology of concepts - the basic types of things that humans can conceptualize in the world. Traditional philosophy of language and formal semantics attempt to make do with an absolutely minimal ontology such as individuals and truthvalues. Perhaps this makes sense if one thinks semantics is about the nature of reality and should ground out elegantly in fundamental physics. But if semantics is about the human conceptualization of the world, its fundamental units arise from evolution's building a brain that is equipped to guide an organism successfully through its life. Again "brain elegance" takes precedence over formal elegance.

32

RayJackendoff

One piece of evidence for the basic ontology comes from deictic expressions that pick out units in the visual field. Just as it is possible to point out objects for the hearer to identify, as in (17a), it is possible to pick out a wide range of other entities. (17)

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

Please pick that [pointing] up. [object] Please put your hat here [pointing]. [location] He went that way [pointing]. [path/trajectory] Please don't do that [pointing] around here anymore. [action] Did you hear that? [sound] / hope that [pointing] doesn 't happen again. [event] The fish I caught was this long [demonstrating]. [distance] There were about Ms many [gesturing] [amount/number] people here last night. i. Can you walk like this [demonstrating]? [manner]

Each of these ontological categories has its own conditions of individuation: many of them (but not all) allow a type-token distinction; many permit quantification. Adopting this relatively rich system from the start affords Conceptual Semantics a broad descriptive capacity and, to some extent, a better constrained relation between semantic and syntactic categories. Note also that (17) lists only ontological categories observable in the physical world; there are clearly others, such as information and value. Once the ontological system is laid out, it becomes possible to recognize entities that subsist simultaneously in more than one ontological domain (the "dot-objects" of Pustejovsky 1995). For instance, a book is simultaneously a physical object and a body of information. These two characterizations, moreover, are in a preference rule relation, since there are blank (i.e. informationless) books and books stored on a computer (i.e. not laid out on paper pages). Reading is a "dot-action", in that it involves both the physical act of scanning the page with one's eyes and the informational act of receiving information off the page. Dot-objects are therefore multidimensional entities within Conceptual Structure. Perhaps the most important case of a dot-object is a human being, who is conceptualized simultaneously as an animate physical object and as a person - an entity in the social domain. The two domains correspond to the (apparently) universal cultural conceptualization of people as composed of body and mind (or soul or spirit). The fact that people have faces and hands and livers falls into the physical domain; the social notions and theory-of-mind notions in (15f,g) above are predicated in the social domain. Again, in traditional beliefs

The Parallel Architecture and its place in cognitive science

33

at least, these two characterizations stand in a preference rule relation. For instance, * zombie is an animate physical object lacking conscious personhood; a ghost is a mind (or soul) lacking a physical body. Reincarnation and bodyswitching (both amply attested in human narratives) are one mind inhabiting different bodies in succession; multiple personality disorder is experienced as different personalities (i.e. different individuals) inhabiting the same body in succession (Jackendoff 2007b, chapter 5). The combinatorial possibilities of Conceptual Structure arise from (at least) three principles of combination: argument satisfaction, modification, and binding. In the default case, argument satisfaction is expressed by syntactic complementation, and modification by syntactic adjuncts. For instance, in (18), John expresses an argument of sleep, and beside the rtver expresses a place modifier. (18)

John slept beside the river.

However, there are exceptions to this typical configuration. For instance, in the sound + motion construction illustrated in (lib) and (19), the subject is a semantic argument not only of the verb, but also of an unexpressed predicate of motion. The PP is also an argument of the predicate of motion, and the verb expresses a modifier of this predicate, i.e. 'move while rumbling'. (19)

The trolley rumbled along the nver

A mismatch in the opposite direction is illustrated by (20). Here cheap margarine is syntactically an adjunct, but semantically it is an argument: it is what Bill put on the bread. (20)

Bill buttered the bread with cheap margarine.

Such mismatches are common. Binding, a direct connection between one conceptual constituent and another, comes in two varieties: identity of reference and identity of sense. This is reflected in two kinds of anaphoric elements in language. Identity of reference binding is expressed by definite pronouns and also by anaphoric epithets, such as in (21) (which does not display identity of sense). (21)

John wants to win, but the voor Suv never wtll.

Identity of sense binding is expressed by 0 «e-anaphora and also by VP anaphora with expressions like do so. These two types of binding must be distinguished in Conceptual Structure, as they give rise to different inferences.

34

RayJackendoff

Using argument satisfaction to create semantic combinations requires functions whose arguments are to be satisfied. A number of broad families of functions have been investigated within Conceptual Semantics. - Functions that encode spatial location, motion, and orientation. They all take two arguments: a Theme (the object being located or in motion) and a Location or Path: BE(Theme, Loc), GO(Theme, Path), STAY(Theme, Loc), ORIENT(Theme, Path), EXTEND(Theme, Path) - Functions that encode Locations and Paths relative to a reference object: IN(X), ON(X), TO(X), FROM(X), TOWARD(X), NEAR(X), etc. Some of these involve imposing a reference frame on the reference object; e.g. BEHIND(X) must be specified as to whether one is speaking of the intrinsic back of Xor its other side relative to the speaker. (This family has been heavily investigated within Cognitive Grammar as well.) - Causative functions that encode a Causer (an Agent or Event) being causally connected to anEffect(anotherEvent): CAUSE(Causer,Effect),LET(Causer. Effect), HELP(Causer, Effect), ENABLE (Causer, Effect), and others. - Mereological functions that encode part-whole relations: PART-OF (legs. handles, noses), BOUNDARY-OF (edges, surfaces, ends, etc.), MEMBEROF (members of aggregations), COMPOSED-OF (ingredients of mixtures) A founding insight of Conceptual Semantics (due to Gruber 1965) is that all of these functions can be applied to semantic fields other than physical space. For instance, an object being owned by someone (a social relation) is often expressed crosslinguistically as the object 'being at' the owner, and changes of possession are often expressed as the object 'going' 'from' the previous owner 'to' the new owner. Similarly, just as we talk about the end of a rope, we can talk about the end of a speech, a relationship, or a genealogical line. This suggests that these Conceptual functions can be decoupled from their physical context (where they connect with Spatial Structure) so as to apply to more abstract domains as well. In addition to possession, they also extend to such fields as time, event structure (such as aspectuality and telicity), ascription of properties. and (in the case of causation) social coercion and logical entailment Further functions that have been investigated (Jackendoff 2007b) involve the personal domain. They include: 6. This insight is treated somewhat differently in Cognitive Grammar (Lakoff 1987). where it is taken to show that underlying linguistic expression is an extensive and powerful system of conceptual metaphor.

The Parallel Architecture and its place in cognitive science

35

- Theory-of-mind predicates, e.g. Xperceives Y (in various modalities)', X i s committed to proposition /»' (belief), X i s committed to action A' (intention), X i s committed to norm N' (adherence to norms) - Value predicates in various domains (affective, normative, quality, etc.): X is of v a l u e P , X i s of value F to person F - Predicates of exchange: X d o e s action^ in exchange/return/retaliation for Y doing action^' - Obligations, rights, and authority: X i s obligated to 7 to perform action^', X has anghtto perform action^', X h a s authority over 7's performing action^' All of these functions are involved in constructing the prepositional tier of Conceptual Structure. In addition, sentence meaning involves an information structure tier, which designates certain semantic constituents as topic, certain as focus, and the rest as common ground. Further differentiation of the propositional tier has also been proposed, for which there is no space here: a referential tier in Jackendoff (2002) (involved for instance in identity-of-reference anaphora, specificity, referential opacity, and quantification) and an action tier or macrorole tier in Jackendoff (1990, 2007b). In short, Conceptual Semantics aspires to the formal richness necessary to encode the character of human concepts and their inferential affordances. It integrates comfortably with the Parallel Architecture, in that although it is a combinatorial system, its units and principles of combination - as well as the resulting structures - are quite different from those of syntax. In particular, it is a multidimensional formal system, in that it involves both Spatial Structure and Conceptual Structure, the latter itself split into multiple tiers connected by interface components. Only through looking at semantics on its own terms. grounded in the character of nonlinguistic cognition, can the independence of these structures from language - and their psychological and biological grounding - be revealed. If meanings have this sort of structure, they certainly cannot be derived from the syntax of NPs and VPs. 4. Simpler Syntax and the syntax-semantics interface An advantage of a parallel architecture over a "single-engine" architecture is that no single level of structure has to carry the entire informational load. In a syntactocentnc architecture, all semantic combinatonality has to be derived from syntactic combinatonality. Thus syntax is forced to be combinatonally at least as complex as semantics - if not more so, since it also has to answer to its own internal imperatives such as word order and agreement. And indeed this

36

RayJackendoff

outcome has been achieved twice in the history of generative grammar: the first time, in the Generative Semantics movement of the late 1960s and early 1970s (Lakoff 1971), and the second time, in Government-Binding Theory of the 1990s and the Minimalist Program. In MP, the rules of grammar and the contents of UG have been reduced to a minimum (allegedly - though only through drastic cuts in empirical coverage), but the structures and derivations have increased steadily in size and complexity (see Culicover and Jackendoff 2005. chapters 2 and 3). In PA, the combinatorial properties of meaning are a property of autonomous conceptual combinatonality. From this perspective, syntax functions in the grammar not as the fundamental generative mechanism, but rather as an intermediate stage in the mapping between meaning and sound (in either direction). The words in a sentence are interface rules that provide small-scale mappings between meaning and sound. What remains to complete the mapping is the relationships among the words: the function-argument, function-modifier. and binding relations. Syntax can be thought of as a way of encoding the semantic relationships among the words in a sentence in terms that are visible to phonology, such as linear order, inflectional morphology, and anaphoric elements - as well as coding the overall semantic force of a clause, such as declarative vs. interrogative. However, there is no need for syntax to encode any more of semantic structure than is necessary to map between phonology and meaning. In fact, many aspects of meaning are not supported by syntactic or lexical expression. For instance, in (22), the underlined constituents are understood as suggested in parentheses. (22)

a. Ellipsis: It seems we stood and talked like this before. We looked at each other in the same way then. But I can't remember where or when. (= 'where or when we stood and talked like this before and looked at each other in the same way as we're looking at each other now') [Spoken to someone about to jump off a building] Don't!!! (= 'Don't jump!') b. Constructional meaning: The trolley rumbled alonZ the river. (= 'the trolley went along the river rumbling') (cf (19)) c. Coercion: The ham sandwich over in the corner wants more coffee. (= 'guy with ham sandwich')

The Parallel Architecture and its place in cognitive science

37

Plato is on the top shelf. (= 'book by/bust of Plato') Joe jumped until the bell rang. (= 'jumped repeatedly') d. Implicature: Are vou SomS to be SomS near a mailbox? (= 'Will you mad some letters for me?') Some of these are treated in mamstream theory in terms of syntactic (or PF) deletion of unexpressed elements; others are not treated in mainstream theory at all. Culicover and Jackendoff (2005) show that they are all best treated in terms of elements of semantics that have no syntactic realization. Carrying this outlook consistently through the syntactic component leads to the approach of Simpler Syntax (Culicover and Jackendoff 2005): an attempt to cut syntactic structure down to the bare minimum necessary to accomplish the sound-meaning mapping. This is a "minimalist" approach to language, but with different premises about what is to be minimized than the Minimalist Program. The basic stance of Simpler Syntax is that the complexity of semantics is independently necessary in order to explain inference and the relation to perception. Therefore semantics and the syntax-semantics interface should play as large a role as possible in constraining the grammatical structure of sentences. and syntax as small a role as possible. On the other hand, the "generative engines" of syntax and morphosyntax are still necessary to account for differences among languages in word order, case marking, agreement, handling of long-distance dependencies, and the existence of special constructions. The resulting syntactic theory is by no means simple, but it is far simpler than mainstream models (whatever 'simple' means in this context). The Simpler Syntax lexicon is as described in Section 3: it contains words. regular affixes, idioms, constructions, and independent principles of phrase structure. Syntactic structures are as flat (i.e. as undifferentiated) as possible. Aside from linear order, there is no syntactic distinction between specifiers. arguments and adjuncts, as this is already provided for in the semantics. The result is predominantly two-layer X-bar skeleta, as in (23a-c). The exception is S, which is a three-layer projection of V, as in (23d). (23)

a.

NP

b.

AP

c.

PP

d.

S

One price of this structural simplification is the need for trees with multiply branching nodes rather than strictly binary branching as in MP. Culicover and

38 RayJackendoff Jackendoff (2005) give arguments why strictly binary branching is not an advantage, and in fact is often a disadvantage. Another price of this simplification is that some rules of grammar have to be sensitive to linear order as well as dominance. This is too often taken to be a disadvantage. But from a larger perspective it is actually an advantage. Linear order comes for free in the signal and hierarchical structure does not. So rules that depend in part on linear order ought actually to be easier for the child to learn. Simpler Syntax makes use of almost no empty nodes in syntactic structure. This is desirable in principle, because empty nodes make heavier demands both on the learner and on processing. Most empty nodes in the classical theory are posited either for semantic reasons or to promote syntactic uniformity. For instance, the phonologically empty element PRO is posited to fill in a semantic subject of an infinitival VP where there is none at the surface, thereby giving all verbs a syntactic subject. Simpler Syntax instead allows infinitival VPs without syntactic subjects, and it uses the interface to identify their "understood" subjects in Conceptual Structure. (24)

a. Mainstream: b. Simpler Syntax:

[John Med [s PRO [VP to leave]]] {John Med [VP to leave]]

Similarly, ellipsis is not derived through empty nodes or deletion. Rather. elliptical configurations, especially when they are syntactically unusual (as in Gapping), are treated as meaningful constructions listed in the lexicon. The interpretation of an elliptical construction is derived from the Conceptual Structure of its antecedent - or from the Conceptual Structure of the context - not from a deleted syntactic structure. Culicover and Jackendoff show many cases of ellipsis for which there is no plausible syntactic antecedent, such as those in (22a). A standard argument for syntactically derived ellipsis is that elliptical constructions often display syntactic properties that normally can arise only through syntactic licensing (so-called connectivity). For instance, in the dialogues in (25), the difference in the prepositions in the replies can be traced directly to the difference between the syntactic licensing of proud vs. pride. (25)

a. A: Bill is very proud. B: Yes, especially of his stamp collection. {ct proud of I* in] b. A: Bill has a lot of pride. B: Yes, especially in his stamp collection, [cf pride in/* of]

The Parallel Architecture and its place in cognitive science

39

However, similarly licensed syntactic properties appear even in sentences where there is no relevant linguistic context, such as (26). (26)

Do you like these? [pointing at a pair of pants],

Simpler Syntax proposes a relation of indirect licensing that accounts for these effects. Like other constraint-based theories, Simpler Syntax has no movement and no covert level of syntactic structure such as Logical Form. The effects ascribed to movement in mainstream theory are accounted for with a variety of mechanisms, most of them shared with other constraint-based theories, especially HPSG. These mechanisms include: - Free phrase order (e.g. among adjuncts in VP, where the order is constrained only by prosody and focus) - Alternative argument realizations (e.g. dative alternation) - For long-distance dependencies, operator-trace relations along the lines of HPSG (trace is the only kind of empty node in Simpler Syntax). The constraints on long-distance dependencies arise from multiple sources, only some of which are syntactic. Others arise from processing complexity and from semantics, especially information structure and referential structure. - Binding and control are relations over Conceptual Structure, not over syntactic structure, though they may involve syntactic conditions on the relation between anaphoric elements and antecedents. In order to account for so-called A-movements, in particular passive and raising, it is unfortunately necessary to introduce extra machinery. Simpler Syntax proposes a grammatical function (GF-) tier that modulates the syntactic realization of semantic arguments expressed as NPs, that is, subjects, objects. and indirect objects. We are not too dismayed by this extra mechanism, as the principles behind it appear in every substantive syntactic theory: as f-structure in LFG, as essentially all of Relational Grammar, as the complement hierarchy in HPSG, and as abstract case in GB/MP The analysis is closest to that in LFG and HPSG. However, in these two theories, passive is a rule that converts active verbs into passive verbs in the lexicon, altering their argument structure. As mentioned earlier, this is not an option in PA, in which the lexicon is where items are stored, and working memory is where structures are built online. Hence, in Simpler Syntax, passive is treated as a construction in the GF-tier that alters argument realization

40

RayJackendoff

online, without altenng the verb itself. The GF-tier is of course another piece of parallel architecture, this time a partial mediator of the syntax-semantics interface. (27) illustrates the linking between the various structures in an example involving raising. The linking relations are notated as subscripts; for visual clarity, some of them are also notated redundantly by vertical association lines. (27)

John seems to like scotch. [SEEM ([LIKE (JOHN 3 , SCOTCH 4 )] 2 )], [GF3L

[GF3

I

>

GF 4 ] 2

Grammatical Function Tier

I

[SNP3 [VP Vx [VP to 5 V 2 NP 4 ] 2 ]]! John,

Conceptual Structure

seems, to, like2 scotch,

Syntactic Structure Phonological Structure

In Conceptual Structure, JOHN is an argument of LIKE, just as in It seems that John likes scotch. It links to the GF array associated with the subordinate clause (bracketed expression subscripted 2). In turn, this GF (subscripted 3) is linked to a GF in the main clause array (subscript 1), which is then linked to the subject of the main clause and its phonology. The linking through the GF-tier is the Simpler Syntax counterpart of an A-chain in classical syntax. But it is not in syntax proper, as there is no syntactic subject at all in the subordinate clause, only a GF-subject. (See Culicover and Jackendoff 2005 for more motivation and detail.)

5. Concluding remarks An abiding issue between linguists and psycholinguists has been the competence-performance distinction. Mainstream linguistics tends to say that the grammar written by linguists is a description of competence, but it is somewhat obscure how this is utilized in performance. This has the effect of insulating linguistic theory from results in psycholinguists. By contrast, in the Parallel Architecture, language processing consists of assembling pieces of structure stored in the lexicon to form a triple of phonological, syntactic, and semantic structures in working memory. As a result, there is no mystery to the competence-performance distinction. Competence theory describes the pieces

The Parallel Architecture and its place in cognitive science

41

of structure and their affordances for assembly, while performance theory describes how these very pieces are assembled in real time, starting from either phonetic input (perception) or conceptual input (production). Details of a performance model in such a vein appear in Jackendoff (2002, chapter 7) and Jackendoff (2007a). The Parallel Architecture also offers an attractive vehicle for discussion of the evolution of the language capacity. It begins with the premise that some version of Conceptual Structure is present in apes, and therefore in our hominid ancestors. Bickerton (1990) and Givon (1979) have proposed that, prior to the development of modern language, there was a stage of "protolanguage", which persists in the human language capacity and emerges in situations such as pidgins and agrammatic aphasia. The denning characteristics of protolanguage are words concatenated into utterances, but lacking any syntactic organization beyond that afforded by linear order. A great deal of the informational load in such an utterance is earned by pragmatics. Within the Parallel Architecture, this form of language can be characterized in terms of a level of phonology linked to Conceptual Structure without the intervention of syntactic structure (Jackendoff 2002, chapter 8). From this stage, the evolution of a syntactic capacity can be seen as adaptive: it is a canonical coding of semantic relationships among words for greater accuracy and efficiency In any architecture, phonological and semantic structures have to be relatively rich, as they code the thousands of distinctions among words. In Simpler Syntax, syntactic structure is relatively lean: its elements comprise only a few parts of speech and phrasal categories, as might be expected of a relatively late evolutionary add-on. By contrast, in the mainstream architecture, an elaborate syntax would have had to evolve first before combinatorial phonology and semantics could be possible, a rather less enticing scenario. To sum up, this article has shown many ways in which the Parallel Architecture, with its components Conceptual Semantics and Simpler Syntax, instantiates the biolinguistic outlook better than does the Minimalist Program. In particular, it offers the prospect of integrating linguistics fully with cognitive science. There still remain, of course, many challenges to the approach, among which perhaps the most important are integrating phonology, morphology, language variation, and language change into the model - as well as a range of syntactic constructions from languages other than English - so that it covers a broader range of linguistic phenomena. In addition, a theory of language acquisition has been sketched (Jackendoff 2002, chapter 6), but it remains a promissory note. It is dearly to be hoped that some of these challenges can be undertaken by practitioners of the relevant subdisciplines.

42

RayJackendoff

References Bickerton, Derek 1990 Language and Species. Chicago: University of Chicago Press. Bolmger,Dwight 1965 Theatomizationofmeanmg.L««g M «g e 41: 555-573. Bresnan,JoanW. 2001 Lexical-Functional Syntax. Oxford: Blackwell. Chomsky, Noam 1957 Syntactic Structures. The Hague: Mouton. Chomsky, Noam 1972 Studies on Semantics in Generative Grammar. The Hague: Mouton. Chomsky, Noam 1995 The Minimalist Program. Cambridge, Mass, MIT Press. Chomsky, Noam 2001 Beyond Explanatory Adequacy. MIT Occasional Papers in Linguistics 20. Cambridge, Mass, MIT, Department of Linguistics. Chomsky, Noam 2009 Opening remarks. In Of Minds and Language: A Dialogue with Noam Chomsky in the Basque Country, Massimo Piattelh-Palmarim, Juan Unagereka,andPello Salaburu (eds.), 13-43. Oxford: Oxford University Press. Cuhcover, Peter W. 1999 Syntactic Nuts: Hard Cases in Syntax. Oxford: Oxford University Press. Cuhcover, Peter W., and Ray Jackendoff 2005 Simpler Syntax. Oxford: Oxford University Press. Fillmore, Charles 1988 The mechanisms of Construction Grammar. Proceedings of the 14th Annual Meeting of the Berkeley Linguistics Society, 35-55. Berkeley: Berkeley Linguistics Society. Fodor,JenyA. 1983 The Modularity of Mind. Cambridge, Mass, MIT Press. Ghomeshi, Jila, Ray Jackendoff, Nicole Rosen, and Kevin Russell 2004 Contrastive focus reduplication in English (The salad-salad paper). Natural Language and Linguistic Theory 22, 307-357. Giv6n,Talmy 1979 On Understanding Grammar. New York: Academic Press. Goldberg, Adele 1995 Constructions: A Construction Grammar Approach to Argument Structure. Chicago: University of Chicago Press. Gruber, Jeffrey 1965 StudiesinLexicalRelations.Ph.B.disMionMIT.PuUMmMfrey Gruber, Lexical Structures in Syntax and Semantics, Amsterdam: NorthHolland, 1976.

The Parallel Architecture and its place in cognitive science

43

Jackendoff, Ray 1983 Semantics and Cognition. Cambridge, Mass, MIT Press. Jackendoff, Ray 1987 Consciousness and the Computational Mind. Cambridge, Mass, MIT Press. Jackendoff, Ray 1990 Semantic Structures. Cambridge, Mass, MIT Press. Jackendoff,Ray 1997 The Architecture ofthe Language Faculty. Cambridge, Mass, MITPress. Jackendoff,Ray 2002 Foundations of Language. Oxford: Oxford University Press. Jackendoff,Ray 2007a A Parallel Architecture perspective on language processing, Brain Research 1146, 2-22. Jackendoff,Ray 2007b Language, Consciousness, Culture. Cambridge, Mass, MIT Press. Jackendoff,Ray 2008 Alternative minimalist visions of language. Papers from the 41st Annual Meeting ofthe Chicago Linguistic Society: The Panels, 189-226. Also to appear in Non-Transformational Syntax: Formal and Explicit Models of Grammar, Robert D. Borsley and Kersti Borjars (eds.). Oxford: Blackwell. Jackendoff, Ray, and Fred Lerdahl 2006 The capacity for music: What's special about it? Cognition 100, 3 3 72. Joshi,Aravmd 1987 An introduction to Tree-Adjoining Grammars. In Mathematics of Language, Alexis Manaster-Ramer (ed.), 87-114. Amsterdam: Benjamins. Koch,Chnstof 2004 The Quest for Consciousness. Englewood, Colo, Roberts. Lakoff, George 1971 On Generative Semantics. In Semantics: An Interdiscipliniary Reader in Philosophy, Linguistics, and Psychology, Danny Steinberg and Leon Jakobovits (eds.), 232-296. New York: Cambridge University Press. Lakoff, George 1987 Women, Fire, and Dangerous Things. Chicago: University of Chicago Press. Lamb, Sydney 1966 Outline of Stratificational Grammar. Washington: Georgetown University Press. Langacker, Ronald 1987 Foundations of Cognitive Grammar. Vol. 1. Stanford: Stanford University Press. Lerdahl, Fred, and Ray Jackendoff 1983 A Generative Theory ofTonalMusic. Cambridge, Mass, MIT Press.

44

RayJackendoff

Levmson, Stephen 2000 Presumptive Meanings: The Theory of Generalized Conversational IffivliccitiiTC C3xnbr1cl.se !Mtiss !M!IX Press Pollard, Carl, and Ivan Sag 1994 Head-Driven Phrase Structure Grammar. Chicago: University of Chicago Press. Postal, Paul 1972 The best theory. In Goals of Linguistic Theory, Stanley Peters (ed.), 131170. Englewood Cliffs, N . J , Prentice Hall. Pustejovsky, James 1995 The Generative Lexicon. Cambridge, Mass, MIT Press. Sadock,Jerxold 1991 Autolexical Syntax. Chicago: University of Chicago Press. Shieber, Stuart 1986 An Introduction to Unification-Based Approaches to Grammar. Stanford. CA:CSLI Publications. Shieber, Stuart, and Yves Schabes 1991 Generation and synchronous tree adjoining grammars. Journal of Computational Intelligence 1:200-22%. Van Valm, Robert, Jr., and Randy LaPolla 1997 Syntax: Structure, Meanings, and Function. Cambridge: Cambridge University Press. Wertheimer,Max 1923 Laws of organization in perceptual forms. Translated mA Source Book of Gestalt Psychology, Willis D. Ellis (ed.), 71-88. London: Routledge & KeganPaul. Wittgenstein, Ludwig 1953 Philosophical Investigations. Oxford: Blackwell.

A phono-syntactic template for Turkish: Base-generating free word order* Ash Goksel

0. Introduction With respect to the place of prosody in the syntactic description of sentences in Turkish, the general understanding has been that there is a neutral intonation pattern or a neutral position of focus, and that any other type of intonation caused by focal stress or other informational^ relevant prosodic contours results in a shift from this neutral state of affairs. If a sentence happens to be ungrammatical under such neutral intonation, the role of this intonation pattern as a possible source of syntactic ungrammatically has often not been taken into account, as such sentences have been considered to exemplify the unmarked case. This paper departs from this view in hypothesizing that prosodic structure itself can be the source of syntactic ungrammatically, and as a result, it is one of the contributors to syntactic well-formedness. The claim put forward here is a characterization of syntax in terms of a tripartite linear template combining intonational (H*) and syntactic (V) categories and which is made up of the following domains: (1)

*

H* V Pre-H*Area(Pre-H*) Focus Field (FF) Postverbal Area (PVP)'

I would Hke to thank the audiences at the Workshop on Word Order, Bogazici University 2001, the 38th Annual Meeting of the Societas Lmguistica Europaea, University of Valencia, 2005, the 28th Annual Meeting of the Deutsche Gesellschaft fur Sprachwissenschaft, Workshop on Morphology and Syntax, Multidimensional. 2006, Bielefeld University where earlier versions of this paper were presented: I am grateful to Meltem Kelepir, Wilfned Meyer-Viol, Ash Untak-Tarhan and Yuji Takano for discussing with me some the issues raised in this paper. Needless to say, I am the only person responsible for errors or misrepresentations. This work was supported by Bogazici University Research Foundation, BAP 07B401 and 5842. 1. "PVP" is the abbreviation for "postverbal position", the term used in the current literature on Turkish for the area following the verb. I will keep to this abbreviation for simplicity with the proviso that it refers to an area rather than a position, hence the usage of the term "postverbal area".

46

AshGoksel

Positing this template as the basis of a syntactic description paves the way for the re-examination of SOV as the 'basic' structure from which other word orders are derived. Under the present view, all word order permutations are base-generated, hence the template provides a basis not only for SOV sentences but for all of the other sentence types. Following from this, one of the claims will be that SOV is special not as the 'basic' underlying structure for word order permutations but as the pattern of presentational focus sentences, and then only with a particular type of intonation. Such a sentence type is not syntactically special as such, but is just one of the many sentence types that occur in the language. If (1) turns out to be an adequate basis for describing aspects of sentence structure in Turkish, discourse related factors will directly be expressed concurrently with syntactic and prosodic properties within a monostratal and multidimensional view of grammar. Hence the model proposed here simultaneously represents syntax, intonation and information structure, with none of these being the input to the other. This paper will present a preliminary sketch of a small portion of data along these lines.* The paper is organized as follows. Section 1 presents an introduction to the significance of prosody for syntax in Turkish and discusses the status of SOV in such a 'free word order' language. In Section 2 I give a brief critical analysis of previous accounts, starting with derivational analyses of word order variability in Section 2.1 and then turning to prosodic accounts in Section 2.2. In Section 3 I develop the model proposed here and apply it to a fragment of Turkish. In Section 4 I extend the investigation to other data to see whether the tripartite representation can be the basis for syntactic domainhood. The discussion concludes with implications of this model and some further issues.

1. Word order variation In a 'free word order' language such as Turkish the task of understanding the position of major constituents becomes one of the primary goals of syntactic de-

2. Surnlar analyses have been put forth in the literature, most notably Steedman (2000). A comparison of these approaches is not possible for reasons of space, however, it is worth noting that the present work conflates intonation and categonal information in a surface representation, while in Steedman mtonational categories are part of lexical items.

A phono-syntactic template for Turkish

47

scnption.3 With a few exceptions (Goksel 1993; Bozsahin undated) derivational models with mechanisms mediating between the derivational stages, be it movement or feature-checking, have dominated syntactic analyses of Turkish. A brief critical evaluation of these analyses will be discussed in Section 2.1. Most of these analyses predate the studies on prosody as a formal device and its relevance for syntax.* The main aim of the next two subsections will be to draw attention to the significance of prosody and to show how it interacts with word order.

1.1. The relevance of prosody to syntactic description While prosody has been accepted as a key feature in the interpretation of a sentence, its place in syntactic descriptions of Turkish has mostly been epiphenomenal. Much less attention has been paid to the syntactic aspects of sentences with identical word order but different prosodic patterns than to word order differences between sentences, irrespective of their prosodic patterns. Instead, where the link between prosody and word order has been investigated. the focus has mostly been on information structure. Here I shall look at several examples that highlight the syntactic relevance of prosody and turn to particular analyses in Section 2.2. It should be made clear that throughout this paper only one of the components of prosody will be taken into account. This is H*, the point of pitch accent (Pierrehumbert 1980). This is indicated in the following examples by setting a word in capital letters. It is clear that the model requires further development in other prosodic factors that influence syntax and interpretation, but this is a point I will leave to future work.5 3. Here it is not possible to give even a brief overview of the literature on word order and its relevance to syntactic structure, see Hale (1983); Boskovrc and Takahashi (1998); Baker (2003); Boeckx (2003) and the references in these works for a discussion of (non)configurationahty, free word order and scrambling. Publications after 2007 are not included due to the pubhcauon schedule of the present article. 4. Investigations of word order (Erkii 1983; Erguvanh 1984) and intonation (Tansu 1963; Demircan 1996, 2001, among others) existed side by side without much interaction. Earlier work of the present author (Goksel and Tsrplakou 1996) also suffers from not providing a finer description of differences in intonation where a sentence which is marked as ungrammatical is actually grammatical under an echo question reading. I am grateful to Yuji Takano for alerting me to the relevant example (Goksel andTsiplakoul996:(8b)). 5. More detailed analyses are given elsewhere (of. Ozge 2003 for declaratives; Goksel. Kelepir, and tintak-Tarhan 2008, 2009 for questions).

48 Ash Goksel 1.1.1. The scope of wh-phrases When a clause containing a ^-expression is embedded in a clause which has a verb of cognition as the main predicate, the intonation contour is the critical factor by means of which this sentence is identified as a declarative or as an interrogative construction :* (2)

a.

Sema-nm NE-YI iste-dig-in-i Sema-GEN what-ACC want-NOM-3POSS.SG-ACC diisun-ityor-uz. (falling intonation)? think-IMP-lPL 'We are thinking about what Sema wants/ #'What do we flunk Sema wants?" b. Semamn NEYI istedigini dusunuyoruz? (rising intonation) #'We are thinking about what Sema wants.' 'What do we think Sema wants?'

Hence, intonation is an integral part of the syntactic description of a sentence. given that 'declarative' and 'interrogative' are syntactic concepts (see Cheng and Rooryck 2000, Aboh and Pfau 2011, Goksel, Kelepir, and Untak-Tarhan 2008 forvanous characterizations of intonation as a morphological and syntactic category). 1.1.2. Scope reversal The scope of quantifiers in Turkish depends on the position of the quantifiers with respect to each other, the type of quantifier in question, and prosodyrelated factors.* The sentences below demonstrate the role of prosodic factors in inducing pairs with reverse scope: 6. Surnlar pairs are mentioned in Yarar (2006). 7. The difference between interrogates and declaratives is more than rising vs. falling intonation at the end, but starts at the beginning (Goksel, Kelepir, and Untak-Tarhan 2009). Here we overlook this additional difference for practical reasons. 8. The abbreviations in the glosses are as follows: "ACC": accusative, "AGR": (subject) agreement, "AOR": aorist, "DAT": dative, "EV": evidential, "FUT": future, "GEN": genitive, "IMP": imperfecta, "INS": instrumental, "NEG": negative: "NOM": nommahzer, "OBT': object, "P": past, "PF": perfective, "PL": plural "POS": possibility, "POSS": possessive, "PRES": present, "Q": question particle: "SG": smgular, "SUB": subject, "TOP": topic/topicahzer 9. One or more of these issues have been discussed in Rural (1994, 1997); Goksel (1998); Aygen-Tosun (1999); Kelepir (2000, 2001); Kennelly (2003, 2004), among others.

A phono-syntactic template for Turkish

(3)

49

a.

Biitiin ogrenci-ler ders-e GiR-ME-Di.«> all student-PL class-DAT enter-NEG-P 'It is not the case that all students went to class/ (adapted from Oztiirk 2005) NEG>V #V > NEG

b.

BUTUN ogrenciler derse girmedi. 'It is true of all of the students that they did not go to class.' #NEG>V V>NEG

As can be sunnised, the interpretation of the operators may not be related to their order with respect to each other and prosody alone can be the source of a particular interpretation. An account that aims at explaining scope effects in Turkish thus cannot eschew intonation, as this may be the only factor relevant for the description of minimal pairs.

1.1.3. The position of bare direct objects Apart from a handful of work (Sezer 1991, 2006; Goksel 1998; Goksel and Kerslake 2005; Uygun 2006) bare direct objects in Turkish have been described as obligatorily occurring in the immediately preverbal position, a description which has been the centerpiece of analyses of derivational accounts. However, the position of a bare object is intricately linked to the intonational contour of die sentence:" 10. (3a) is based on Oztiirk (2005) and is also analyzed by Arslan-Kechriotis (2006). Oztiirk describes it as having 'neutral intonation', which presumably means that it is an out-of-the-blue sentence. This, in turn, would mean that, being a negative sentence, the verb is stressed (as in (3a)). Under the same 'neutral intonation' as implied there, the reading associated with (3b) is not possible. However, this reading (i.e. V > NEG) is precisely what we get from the same order of constituents under a different prosodic pattern. 11. This is one of the areas where speakers are strongly divided with respect to the acceptability of bare objects occurring anywhere else than in the immediately preverbal position. Negative judgements may be due to the fact that such sentences never occur out-of-the-blue and require a particular context and intonation to be considered as acceptable.

50

AshGoksel

(4)

a.

Gazete hergiin ben OKU-YOR-UM. newspaper everyday I read-IMP-lSG 'I do read newspapers everyday'

b. 1GAZETE hergiin ben okuyorum.

(cf. (3 lb) and Note 27)

These data present another case where the presence of a bare direct object in the beginning of a sentence depends on whether it is stressed or not, indicating that intonation has to be analyzed as part of the syntactic description of that sentence.

1.1.4. Anaphoric

relation

A proper noun in a complex NP is more readily available as the referent of a following third person pronoun if the former is unstressed and the latter stressed, rather than the reverse stress pattern (i.e. where the proper noun is stressed and the pronoun is unstressed): (5)

a.

Sema-mn, anne-si-yle Sema-GEN mother-3POSS.SG-INS anlas-a-ma-ma-miz (bir tek) ON-U, agree-POS-NEG-NOM-3POSS.SG (only) s/he-ACC uz-uyor-du. upset-IMP-P 'The fact that we did not get along with Sema's, mother upset only her,,'

b. *SEMANIN, annesiyle anla^amamamiz (bir tek) oniij uzuyordu. (5b) is acceptable if Sema and the pronoun are not preferential. Again, a prosodic factor is crucial for the description of the structural properties of a sentence.

1.2. The status of SOV within free word order Another assumption that is prevalent in most syntactic analyses of Turkish is that free word order is restricted to the six construction types listed in (6) and exemplified in (7):

A phono-syntactic template for Turkish 51

(6) SOV OSV SVO OVS VSO VOS (7)

a. Semakurs-u bitir-di. Semacourse-ACC complete-P 'Sema (has) completed the course/ b. Kursu Sema bttird, c. Semabitirdikursu. d. Kursu bitirdi Sema. e. Bitirdi Sema kursu. f. Bitirdi kursu Sema.

(SOV)

(OSV) (SVO) (OVS) (VSO) (VOS)

However, the list of construction types in (6) is incomplete and incorrect. Once the position of stress is integrated into syntax, it turns out on the one hand that there is a larger number of permissible word orders and on the other hand that not all word orders are grammatical. Below the stressed constituents are underlined:

(8) a. SOV OSV SVO OVS VSO VOS b. SOV OSV SVO OVS *VSO *VOS c. SOV OSV *SVO *OVS *VSO *VOS Stress in Turkish can fall either on the verb (wherever it is located) or on any constituent to the left of the verb, but, crucially, not on a postverbal constituent. As can be seen, the grammaticality of some of the sentence patterns in (8) depends on at least one other factor, the location of H* indicated by the underlined constituent. The second assumption about Turkish word order is that although it is free. the language is SOV. The contradiction embodied in this description - the claim that Turkish imposes a certain ordering on its constituents and, at the same time, that it does not - is resolved only by analyses that invoke a link (such as movement or feature-checking) between SOV and non-SOV sentences. Thus the latter become derived structures while SOV represents the 'basic' structure. Even if not stated in explicit terms, traditional grammars of Turkish also take SOV as the 'basic' word order in Turkish, although they may not be clear on the mechanisms deriving the other orders.^ The next section will therefore look at whether the idea that there is a 'basic' structure can be abandoned. 12. See Ergm (1962: 377), Ediskun (1963: 361), Swift (1963: 179), Lewis (1967: 240). Banguoglu (1974: 532), Dizdaroglu (1976: 250), among others.

52

AshGoksel

2. Previous analyses of word order and prosody 2.1. Movement analyses The following will be a brief outline which highlights the main points and addresses some of the problems involved in derivational analyses. It is not possible to give a fair evaluation of movement analyses in Turkish within the space limitations of the present paper. For practical reasons, the analyses which assume derivational stages are treated without drawing the distinction between the mediators between the stages (such as movement and feature checking), and points which are not relevant to the present analysis (such as head movement) are left out. The assumption that there are derivational stages in syntax amounts to the idea that there cannot be a unique position which captures the various properties of a syntactic item hence constituents have to be relocated to satisfy various well-formedness conditions. The literature on word order variation in Turkish involves mainly two types of sentence where such relocation processes are assumed to take place: sentences where the object occurs before the subject and sentences where there is material in the postverbal position. I take each one in turn. 2.1.1. Object fronting: OSV The research on OSV in Turkish is centered around two issues: the position that the object ends up in as a result of movement and the motivation for this movement. * There are three competing analyses for the analysis of the position of the object: an adjunction site (A'-position) on the left, an argument site (A-position) on the left and A'-position on the right. 2.1.1.1. Movement to an adjunction site to a higher projection on the left The preposing of an object was first addressed in Kelepir (1996). For Temurcii (2005), this position is a higher functional category (denoted by "FP" in (9)) 13. Two recent works on this topic are Oztiirk (2005) and Arslan-Kechnotis (2006). Here I am not concerned wrth the differences between EPP motivated movement and movement for case-checking or strong versus weak case features associated with nominals, a point which is discussed in detail in these works. Nor am I concerned with whether a nominal is an NP or a DP. The insights relating to such distinctions may very well be relevant for the present analysis, although at this point they look like points associated with the particular models.

A phono-syntactic template for Turkish

53

where the object moves to for reasons of totalization, motivated by the EPP (following Miyagawa 1997). Similarly, for Ozsoy (2005) it is a projection associated with the topic ("TopicP" in (9)), again motivated by the EPP: (9)

[T0P1CP/FP NP(obj), [(XP). . . [IP NP(sub) ...t,

VP]]]

Leaving aside the theoretical implications of the motivation for this movement. these analyses fail to explain the conditions for a fronted object that is focalized and seem to imply that such fronting can only be associated with topicalization. In (10b), however, the fronted object is focused: (10)

a. DrlbrUm-r OGRENCl-LER seg-ryor-mus. linguistics-ACC student-PL choose-IMP-EV 'STUDENTS choose linguistics.' b.

DiLBlLiMlogrencilersegiyormus. 'Students choose LINGUISTICS.'

Oztiirk, although not dealing specifically with this issue, nevertheless addresses issues regarding the occurrence of a fronted object with or without a focused phrase. According to her analysis, if there is a focused phrase in a sentence the fronted object is in a topic position:" (ID

[T0 P 1 C pNP(0bj) ; [ T o p l c p N P ( s u b ) , . . . bocusP Lxpt ; t,]]]]

Again the description of a fronted object as topicalized hinges on the implicit assumption that the sentence has preverbal stress rather than stress on the fronted object. Moreover, the absence of a focused phrase in the sentence leads to a radically different positioning of the fronted object, as illustrated in the next point.

2.1.1.2. Clause-bound (TP internal) 'leftwardmovement'to position higher than the subject

an argument

This option for the positioning of a fronted object is suggested by Oztiirk for OSV sentences without a focus phrase, and as a general mechanism by ArslanKechnotis (2006): 14. Both Oztiirk (2005) and Ozsoy (2005) allow multiple 'topic' positions, which seems to be imposed by the technical requirements of the model they use rather than being empirically motivated by information structural considerations.

54 (12)

AshGoksel [TenseP N P ( o b j ) , [ A g e n t P N P ( s u b ) [

„

t, V P ] ] ]

The assumption that there are sentences without a focus phrase highlights the problem of eschewing prosody where it may be relevant, as discussed in Section 1.1. Furthermore, it is not clear how moving an object into that position. which would otherwise host a moved subject, is warranted. The problem is confounded by the fact that object-fronted sentences with focus phrases are different from those without focus phrases (cf (11) above). In both cases, the fronted object is non-focused and its interpretation is not altered by the other constituents. However, this fact does not seem to grant the object a unique position. Rather, the position of the object hinges on the presence or absence of another phrase, resulting in two positions different in nature: one an A'-position. the other an A-position. The ensuing movement/checking mechanisms therefore seem to be arbitrary."

2.1.1.3. Movement of the subject to an adjunction site at a higher projection on the right According to Takano (2005), OSV requires the object to remain /„ situ while the subject moves to a higher (TP) projection, with the difference that this position is on the right: (13)

[CP

[TP

[TP NP(sub) [vP NP(sub) NP(obj). . . ]] NP(sub)]]

Since the subject moves to the right (but remains in the preverbal position as the verb moves higher up), it occurs to the right of the (non-moved) object. Again, the motivation for such a movement is not clear, but what is more problematic is that this results in a constellation in which the subject should have scope over the object, which is not always the case (cf. (33a-b) below)." 15. Also in this analysis, the movement of the subject to the specifier of TP is barred only because there is no other focused phrase in the sentence. 16. A simpler objection might be raised against movement in OSV structures: that there is no evidence of a trace in the assumed preverbal position. Assigning a specific position to an object in Turkish can be traced back to earlier works on word order (Erkil 1983; Erguvanh 1984, see Section 3.1 for an evaluation) and on the presence of a VP (Kornfilt 1988). The former discuss 'displaced' objects, but are noncommittal with respect to the derivational history of these constructions. The latter work suggests that there is a VP in Turkish based on elliptical constructions targeting object + verb constructions, but the description is both too narrow (ellipsis can

A phono-syntactic template for Turkish

2.1.2. Postverbal scrambling: SVO, OVS, and initial

55

sentences

The occurrence of arguments in the postverbal position (PVP) was first analyzed as an instance of 'nghtward movement' by Kural (1994). (14)

[ C p[cp[iP...tJ][pvpX ; ]]

Kural bases his claim on data where postverbal constituents have unambiguous scope over preverbal ones, thereby presenting a counterexample to the Antisymmetry Hypothesis (Kayne 1994). The observation that postverbal constituents have unambiguous scope over preverbal ones is based on sentences where there is a preverbal temporal adverb which is stressed. The examples below illustrate this point with a postverbal object, but the same facts hold for postverbal subjects:" (15)

a.

Herkes dun ara-mis tic kisi-yi. everyone yesterday call-PF three person-ACC 'Everyone called three people yesterday' #V > num (# For every x there were three y such that x called y.) num > V (There were three y such that for every x, x called y.) (only non-distributive reading available)

also target a subject + verb, see Hankamer 1979: 82 for the absence of syntactic rules targeting the VP in Turkish) and too broad (what is said to be a VP contains more than a verb an object, e.g. tense, modality and agreement). There might, on the other hand, be arguments against movement, such as the lack of weak crossover effects. A stressed object wA-phrase can be coreferential with a pronominal when the former precedes the latter and not vice versa, see Goksel and Tsiplakou (1996) for details: (i) KtM-U anne-si,,, ara-di? who-ACCmother-3SG.POSScall-P 'Who, did his, mother call?' ®

Annesi,lhKlMUara-di? 'Who, did his,, mother call?'

If the direct object W/*-phrase kimi had originated left-adjacent to the verb and had been moved, there would have been an intervening pronoun (the possessive suffix -si) between it and its trace. Rather, when both W/*-phrase and pronoun are in the F(ocus)F(ield) (see Section 3), co-indexation is the only option. Otherwise (that is. when the pronoun is not in the FF), co-mdexation is not possible. 17. The glosses in some of the cited examples have been harmonized with the other examples used in this paper and thus do not reflect the cited authors' original choice for describing certain forms.

56 Ash Goksel b.

Of k¥yl dun arami^herkes. 'Everyone called three people yesterday/ V > num (For every x there were three y such that x called y) #num > V (#There were three y such that for every x, x called y) (only distributive leading available)

One of the problems with tins analysis was put forward by Kornfilt (1998) who presents data showing that not all postverbal constituents behave in this way. The internal argument of the verb can have ambiguous scope with respect to the subject: (16)

Herhes bu yd Mtap-lar-in-i rthafet-n^ flp everyone this year books-PL-3POSS.SG-ACC dedicate-PF three person-DAT 'This year everyone dedicated their books to three people/ (adapted from Kornfilt 1998: 110)

Hence a universally quantified constituent which is located preverbally can have scope of a postverbal one. The two claims are summarized below: ' (17)

Kural: Kornfilt:

#V>num num > V V>num num > V

(only interpretation) (primary interpretation) (secondary interpretation)

Another problem with Kural's analysis is that it does not take into account the interpretation of the constituents under different stress patterns (see Goksel 1998 for examples and discussion). Finally, the postverbal position can host multiple constituents, as discussed in Goksel (2001, see Section 4.1 below), Kornfilt (2005) and Takano (2005). In Takano (2005), this is a result of 'rightward movement' analyzed as 'complement formation' followed by 'tucking in' (in the sense of Richards 2001), rather than adjunction (as in Kornfilt 1998). and occurs before spell-out: (18)

[ C P X1[TPX2X3]]

According to Takano both constituents are in CP in a specifier-head configuration and are equidistant to each other. If this is the case, this is also similar to claiming that there is flat structure in the PVP Kornfilt (2005) makes this claim

A phono-syntactic template for Turkish

57

explicit, suggesting that this flat structure is the result of a linearization process with intonational effects (see also Ozge 2003 and Aydiner 2006 for downstepping in the postverbal area). The explanation that postverbal constituents must be the result of a PF process, a stance taken by Temurcii (2005) and an alternative entertained by Kornfilt (2005), has the disadvantage that both treat the postverbal area as being external to syntax. As such, PF-movement seems to cater for the requirements of the model which the explanation is set up in and is a mechanism resorted to just in those cases where the inter-constituent hierarchy breaks down. The view presented in the present work, on the other hand, takes all the visible locations of constituents as part of syntax proper. The main question to be raised with respect to movement analyses is what evidence there is for movement in Turkish in the first place. If a moved constituent cannot reconstruct (i.e. cannot be interpreted in its base position), then the onus is on the proponents of that account to explain why it is base-generated there. A similar criticism applies to proponents of PF movement. If a constituent is interpreted not in the postverbal position that it appears but elsewhere, an explanation is required as to how the postverbal position can be filled by this constituent at all.«

2.2. Prosodic analyses of Turkish and their relevance to syntax The idea that prosody can be integrated into a description of sentence structure in Turkish has as its predecessor the description of sentences in linear terms, an idea which goes back to Erguvanli (1984) who identifies three pragmatically distinct concepts corresponding to different parts of a sentence. According to her analysis, topic, focus, and background information are each realized in designated linear positions: topic in the initial position, focus in the immediately preverbal position and background information in the postverbal position. ^ Contrastively focused constituents are left out of the description as these are seen as part of marked sentences, the type that are not relevant to a description of unmarked sentences. These assumptions are taken up by Kilieaslan (2004)

18. Kilieaslan (2004) analyzes the presence of material in the postveibal position as extraposition from the coie domain (SOV) to the clause-external domain. Although not advocating the type of movement typical in Chomskyan approaches, Kilieaslan nevertheless assumes a basic SOV order, which is at odds with the proposal heie. 19. The exact description of Erguvanli (1984) is in terms of the constituents that host these positions, not the positions per se, but the diferences between these two concepts are not crucial for our purposes here.

58

Ash Goksel

and Issever (2003) who provide various accounts for the linear position of discourse functions. The integration of c o n t r a s t s focus into the description of structure in terms of linear positions in Turkish was first discussed in Goksel and Tsiplakou (1996), followed by Goksel (1998). These two works define the postverbal position as a linear domain (as opposed to hierarchical accounts described in Section 2.2 above). Different to what is proposed here, notions relating to prosody, more precisely focus, come into play in terms of where the stressed constituent occurs in the linear order of sentences.* The idea of defining a syntactic segment mprosodic terms, hence changing the role of prosody related concepts was first suggested in Goksel and Ozsoy (2000). There, one edge of a linear domain is defined in terms of stress. This domain hosts focus related constituents (focus-phrases and wA-phrases) without making reference to the type of syntactic constituents that marks this left edge. The right edge of this domain is demarcated by the verb, a syntactic category. Thus the description of this domain, the Focus Field, builds upon prosodic and syntactic categories concurrently: (19)

; V " Focus Field (FF)

The present work is concerned with the expression of syntax by way of using prosody. In the next section I build upon the implications of this claim and suggest that the partitioned picture in (19) is a fundamental part of the structure of sentences in Turkish.21

3. Proposal: A phono-syntactic template What would the characterization of Turkish syntax be like if prosody were integrated into it? (1), repeated below, is the skeletal structure of a sentence under this proposal: (20)

H* V Pre-H* Area (Pre-H*) Focus Field (FF) Postverbal Area (PVP)

20. The separation of precedence and dominance hierarchies was suggested for Turkish earlier in Goksel (1993). 21. This claun is supported by Untak-Tarhan (2006), who discusses the freedom of order between direct objects and adverbs in terms of prosody.

A phono-syntactic template for Turkish

59

This linearly denned representation takes into account the fact that H* does not occur to the right of V in Turkish, neither in declaratives nor in questions.* Taking H* and V as the basic points of demarcation yields three segments: (21)

i. the Pre-H* Area (Pre-H*): the domain before stress n. the Focus Field (FF): the domain delimited by stress and the predicative verb (Goksel and Ozsoy 2000) in. the Postverbal Area (PVP): the domain after the predicative verb*

Pre-H* is a prosodic domain which can be characterized using intonational primitives (Pierrehumbert 1980). The PVP however, does not lend itself to such a description since the intonation contour following H* may contain only the PVP, the verb + the PVP, or preverbally occurring constituents + the verb + the PVP. Hence the verb is of pivotal importance in the description of the domains. The representation in (20) is thus a 'template' which provides locations for all morphologically free constituents except for the verb form which is one of the markers of domain boundaries. The only domain which is obligatorily present is the FF. In the examples below, the domains are underlined and identified by the corresponding abbreviation: (22) HiCBIR oirenct buzun sinav-a Sir-me-vecek-miS.* FF student today exam-DAT enter-NEG-FUT-EV 'Apparently none of students will take the exam today.' The FF can be preceded by material in Pre-H* as in (23a) or followed by material in the PVP as in (23b). It can also be flanked by these domains as in (23c). The pattern in (23) is illustrated in (23'): (23)

a. [ p r e H - . - H p F . . . ] b.

[FF...][PVP...]

C

[preH*...][FF...][pVP...]

22. Sentences where the sequence V . . . H* is grammatical are given by Demircan (1996). However, these and other sentences which do not conform to (20) all seem to be instances of lexically formed sentences (in the sense of DiScmllo and Williams 1987; Cuhcover and Jackendoff 2005). 23. "Predicative verb" is a cover term for the predicates of all verbal sentences, and the hidden or overt copula in 'nominal' sentences. 24. It may be that the FF extends as far as the copula and that the lack of stress in the PVP is, in fact, an outcome of the lexical properties of the copula, an issue which I leave to future work.

60 (23')

AshGoksel a. Smava BUGUNhicbir Pre-H* FF

ogrenci zirmeyecekmis.

b. HiCBlRoirencizirmevecebnis FF

bumn smava. PVP

c. Smava HJCBIR o&renci zirmeyecebnis Pre-H* FF

bugim. PVP

Note that the order of constituents is irrelevant at tins point The sentences in (22) and (23') are naturally only a few of the permutations of the linear order of the constituents in these examples. The question is how these domains are structurally relevant. This is what I turn to next.

3.1. An application: The expression of grammatical functions Certain points follow from the proposed characterization of the template. Firstly there are no specific linear positions for arguments and adjuncts. A direct object does not have to be based-generated left-adjacent to the verb, nor does it have to occupy a designated position. NPs which have accusative markers are unambiguously interpreted as direct objects wherever they occur in the sentence. Hence case, in the proposed way of looking at Turkish syntax, is not 'assigned^ by a verb to a specific position, nor is it checked in a particular position, forcing an NP to have a connection with a specific location. Accusative case marking simply acts as an instruction for an NP to be interpreted as the internal argument of a two-place predicate. Similar conditions hold for other grammatical functions (see Goksel 1992 for details). The difficult cases are arguments with no overt case marking, the subject and the bare direct object, and this is where the template in (20) comes in. Consider the sentences in (24) and (25). When two bare NPs occur in the Pre-H* domain as in (24) or in the PVP as in (25) there are again no designated positions for objects and subjects:* 25. The relevance of Pre-H* in the 'free' ordering of subjects and objects correspond to the findings of Hoffman (1998) and Turan (1998), although the details have yet to be worked out. These works discuss the role of the surface positions of subjects and objects in Turkish in terms of discourse salience and centering, reaching the conclusion that both are equally accessible in these 'sentence-initial' and 'topic' positions. In the majority of the examples provided, the relevant constituents are in what corresponds to Pre-H* in this paper. Such examples might show that the similarity between subjects and objects is contingent to the properties of the domains, rather than their description as topics or sentence initial conatituents.

A phono-syntactic template for Turkish 61 (24)

a. Kediet Pre-H* cat meat 'Cats eat

YE.R. FF eat-AOR meat'

b. EtkecU YERPre-H* FF 'Cats eat meat/ (25)

(SOV)

(OSV; cf. (4a))

a. YE=R kecMet FF PVP eat-AOR cat meat 'Cats eat m e a t ' b. YER etkedi. FF PVP 'Cats eat m e a t '

Smnlarly, there are no designated positions when one bare NP is inside the FF and the other is in the PVP. The facts are the same whether the NP inside the FF is itself stressed or not: (26)

a. KEDl ye-r et FF PVP cat eat-AOR meat 'Cats eat m e a t ' b. AC kahnca kedi ye-r et FF PVP hungry remain cat eat-AOR meat 'Cats eat meat when hungry.'

(27)

a. ET ye-r kedi. FF PVP meat eat-AOR cat 'Cats eat m e a t ' b. AC kahnca et ye-r kecU. FF PVP hungry remain meat eat-AOR cat 'Cats eat meat when hungry.'

62

AshGoksel

However, when the subject is in the FF, the direct object is less acceptable in the Pre-H* than when it is in the PVP, although the acceptability increases with the distance from the edge of the FF: (28)

lEt brrtekKEDlve-r. Pre-H* " FF meat only cat eat-AOR 'Only cats eat m e a t '

This provides the initial positional restriction on the order of bare NPs: (29)

?NPIobj) NPisublV Pre-H* FF

Notice that in (29), the direct object not only precedes the subject, but is also located outside the FF. This characterization of facts is different from describing NP-NP-V sequences only in terms of the order of the constituents. One of the first examples of such a description is found in Erguvanli (1984) where the immediately preverbal position is claimed to be the basic position for the direct object. Her examples are the following: (30)

a. Huzur mutluluk getrr-m peace.ofmind happiness bnng-AOR 'Health brings happiness.' h. Mutluluk huzur getinr. 'Happiness brings health.'

Erguvanli claims that these sentences are unambiguous, taking this as a basis for describing Turkish as having SOV as the basic structure. However, describing these sentences as unambiguous is, crucially, contingent on their presentation as out-of-the-blue sentences. They are, thus, by definition, presentational focus sentences. In presentational focus sentences stress occurs left-adjacent to the verb (see Goksel and Ozsoy 2003), crucially rendering the direct object part of the FF, and the subject, part of the Pre-H*.26 Recall that NP-NP-V sequences are actually not unambiguous as demonstrated in (24a-b) by the SOV-OSV 26. Note that the FF does not necessarily coincide with the semantical^ focused part of a sentence. For example, in a presentational focus sentence, the whole proposition presents new information whereas the FF excludes the subject.

A phono-syntactic template for Turkish

63

pair. Hence the interpretation of the immediately preverbal NP as the direct object in (30a,b) comes not as a result of its position as the left-adjacent constituent to the verb, but as a result of it being an out-of-the-blue sentence plus the additional condition that it is located in the FF, as in (31 a), while the subject is in Pre-H*. This is, in fact, the structure of presentational focus sentences, as illustrated in (3 l b ) : * (31)

a. Kedi ET ye-r™ Pre-H* FF cat meat eat-AOR 'Cats eat m e a t ' b. NP(sub)NP(obi)V Pre-H* FF

Finally, we witness constraints when both bare NPs are inside the FF. In this case there is less acceptance by native speakers for the subject to follow the object as in (32b):» (32)

a.

KEDl et ye-r. FF cat meat eat-AOR 'Cats eat m e a t '

b. *lETkedi yer. FF 'Cats eat meat'

(cf (4b))

Hence the FF is singled out as the position where there is a dominance hierarchy between a subject and an object. The Pre-H* is less structured in this sense with respect to the FF (cf. (28)), but more so than the PVP (cf. (26)-(27)). A bare NP cannot be interpreted as the direct object when it precedes another bare NP in the FF (32b). These examples indicate that there is no reason to resort to 27. This may not be the only type of presentational focus sentence. See NakrpogluDemiialp (2002) for SOV sentences where V is the constituent in the FF yet the sentences have a presentational focus reading. 28. SOV sentences where the direct object is in the FF and adjacent to the verb are ambiguous between contrastive and presentational focus, see Goksel and Ozsoy (2003). Whether these two focus types may be distinguished by pitch is a matter unresolved. 29. Again, the distance between the object and the subject may positively contribute to acceptability see examples in Uygun (2006).

64

Ash Goksel

case assignment or case checking as mechanisms that link NPs to grammatical functions.^

4. Further applications of the template In this section I look at various syntactic phenomena in order to see whether a characterization in terms of the template is a possible way of analyzing them. The aim is to give a general idea about the application of the proposal here. rather than go into a detailed analysis of particular constructions.

4.1. Quantificational dependencies Ifirstlook at sentences with multiple quantifiers in order to understand whether the domains behave differently with respect to the construal of the scope of quantifiers. The two domains that will be compared are the Pre-H* and the PVR The scope interaction between two quantified NPs in the Pre-H* will be compared with identical NPs in the PVP." Note that in each case the verb is stressed, hence the FF in all of these constructions comprises just the verb form. As argued in Goksel (2001) the ordering of the quantifiers yields different results in the two domains.^ In the examples below, both dative and accusative marked NPs are used as non-subjects to show that there are no hierarchical differences between these two in terms of their relation with the other constituents.

30. Here I do not address some of the issues addressed in Oztiirk (2005, pseudo incorporation) and Arslan-Kechnotis (2006, word order restrictions in ECM clauses) which are relevant to the expression of argument structure. I leave these for future research. 31. The scope of quantifiers depends on a number of factors which affects then interpretation. For reasons mentioned above, the question whether the templatic model can characterize the data presented in these works will not be addressed. The linear positioning of the constituents, whether they are focused or not, dominance relations, and the type of quantifiers have a role in scope construal. One or more of the factors affecting interpretation have been addressed in various works, see Kural (1994, 1997); Goksel (1998, 2001); Aygen (1999); Goksel and Ozsoy (2000, 2003); Kelepir (2001); Kennelly (2003, 2004). 32. Kornfilt (2005) has similar observations but attributes the difference to that between topics and PF-movement, see Section 2.1.2 above.

A phono-syntactic template for Turkish

65

4.1.1. Quantifiers in Pre-H*: Verb-final sentences (33)

a.

Birhoca her asistan-i GOREVLENDiR-Mft a professor every assistant-ACC assign.a.task.to-PF 'A professor assigned a task to every assistant.' (SOV;sub:V,obj:3) 3>V (non-distributive reading available) ?V>3 (distributive reading marginal)

b. Her doktor bir hasta-yi KONTROL ED-IYOR. every doctor a patient-ACC examine-IMP 'Every doctor is treating a patient.' (SOV; sub: V, obj: 3) V>3 (only distributive reading available)

When a subject in the Pre-H* precedes the object, it either unambiguously has scope over it as in (33a) or the interpretation in which it is outscoped is marginal or not salient as in (33b). Conversely, when an object in the Pre-H* precedes the subject, the result is either ambiguous as in (34a), or the interpretation in which it is outscoped by a following subject is marginal, illustrated in (34b).

(34)

a.

Her hasta-ya btr doktor BAK-TI. every patient-DAT a doctor treat-PF 'A doctor treated every patient' (OSV;sub: 3,obj: V) 3>V (non-distributive reading available) V>3 (distributive reading available)

b.

Bir hasta-ya her doktor BAK-IYOR. a patient-DAT every doctor treat-IMP 'Every doctor is treating a patient.' (OSV; sub: V, obj: 3) 3>V (non-distributive reading available) ?V>3 (distributive reading marginal)

These examples show that for a subject in the Pre-H* to have unambiguous scope over the object, another condition has to be satisfied, namely, that the subject has to precede the object as in (33b). The only case where there is full ambiguity (that is, where both readings are equally salient) is when an object precedes the subject, as illustrated in (35). Note that the differences within each pair are partly due to the properties of the particular quantifiers.

66 (35)

AshGoksel Linear order: S...0 0...S

Interpretation: i. unambiguous: S > 0 ii. ambiguity questionable: S > 0 , ? 0 > S i. ambiguity questionable: 0 > S, ?S > 0 ii. ambiguous: 0 > S, S > 0

(33b) (33a) (34a) (34b)

This chart highlights the fact that the scope of quantifiers in Turkish is not read off directly from their surface positions, but that prosody plays a fundamental role in their interpretation. 4.1.2. Quantifiers in PVP: Verb-initial sentences These results contrast sharply with sentences where both quantified constituents are in the PVP. Here, whatever order the subject and the object appear in and irrespective of the particular quantifier, the sentences are ambiguous: (36)

a. BAK-TI bir doktor her hasta-ya treat-PF a doctor every patient-DAT 'A doctor treated every patient.' (VSO; sub: 3, obj: V) 3>V (non-distnbutive reading available) V>3 (distributive reading available) b. BAK-IYOR her doktor bir hasta-ya. treat-IMP every doctor a patient-DAT 'Every doctor is treating a patient.' (VSO; sub: V, obj: 3) 3>V (non-distnbutive reading available) V>3 (distributive reading available)

(37)

a. GOREVLENDlR-Mi$her asistan-i bir hoca assign.a.task.to-PF every assistant-ACC a professor 'A professor assigned a task to every assistant.' (VOS;sub:3,obj:V) 3>V (non-distributive reading available) V>3 (distributive reading available) b. GOREVLENDiR-Mi§ bir asistan-i her hoca. assign.a.task.to-PF a assistant-ACC every professor 'Every professor assigned a task to an assistant.' (VOS;sub:V,obj:3) 3>V (non-distributive reading available) V>3 (distributive reading available)

A phono-syntactic template for Turkish

67

These examples show that the PVP is much less structured that the Pre-H* and the FF.

4.2. Ellipsis Data on ellipsis show that the occurrence of a particular constituent in Pre-H* or FF affects the grammaticality of a sentence. Hankamer (1979) and Kornfilt (2000) point out that forward gapping is ungrammatical in the preverbal position but grammatical in the PVP: (38)

a. *Zeynep [[Hasan-in karides-i ye-dig-m-i], Zeynep Hasan-GEN shnmp-ACC eat-NOM-3POSS.SG-ACC [Mehmed-in de istiridye-yi ]] duy-du. Mehmet-GEN and oyster-ACC hear-P intended interpretation: 'Zeynep heard that Hasan ate the shrimp and Mehmet (ate) the oyster' b.

ZEYNEP duy-du Zeynep hear-P [[Hasan-in karides-i ye-dig-in-i], Hasan-GEN shnmp-ACC eat-NOM-3POSS.SG-ACC [Mehmed-in de istiridye-yi ]]. Mehmet-GEN and oyster-ACC 'Zeynep heard that Hasan ate the shrimp and Mehmet (ate) the oyster.' (adapted from Kornfilt 2000)

Seen from the perspective of the present work, the description that forward gapping is ungrammatical is not correct. The ungrammaticality of forward gapping is contingent on the position of stress in (38a) where presumably the preverbal constituent is stressed. If this is the case, the phrase Mehmet-m de istiridye-yi 'and Mehmet the oyster' falls within the FF, under the current account. Placing it in the Pre-H* makes (38a) grammatical: (39)

Zeynep [[Hasan-tn karides-i ye-dig-in-i], Zeynep Hasan-GEN shnmp-ACC eat-NOM-3POSS.SG-ACC [Mehmed-m de istiridye-yi ]] QOKXAN bihyordu. Mehmet-GEN and oyster-ACC for.some.time know-IMP-P 'Zeynep has known for some time now that Hasan ate the shnmp and Mehmet (ate) the oyster'

68

AshGoksel

As expected, the grammatical^ of a sentence depends on winch domain a particular constituent occurs in. In this case those domains are Pre-H* and FF.33

4.3. 'Dislocated'constituents Another piece of data which may be used for comparing Pre-H* and FF is the phenomenon referred to as "dislocated NPs". Kural (1994) notes that dislocation of a (non-subject) NP from an embedded clause to the postverbal position of the embedded clause is ungrammatical when the whole embedded clause is in the preverbal area of the matrix clause: (40) *Ayse [[Ahmet-in t, konu?-tug-un]-u ogrenciler-le-A Ayse Ahmet-GEN speak-NOM-3POSS.SG-ACC students-INS bil-iyor.

know-IMP intended interpretation: 'Ayse knows that Ahmet spoke with the students.' (adapted from Kural 1994: 6) As Kural does not mention the position of stress in (40) it is difficult to judge the acceptability of this sentence. However, his judgements are shared by Kornfilt (1988) and Kilicaslan (2004) who also analyze similar constructions and have similar judgements. One thing is clear, though. When the offending constituent is in the FF (which is presumably how those who have analyzed it construed this sentence), (40) is indeed unacceptable, as shown in (41a). However the important point is that (41b), where the constituent ogrencilerle 'with the students' is placed in the Pre-H*, is acceptable*

33. It is not clear at this stage to what extent the (un)grammaticality of forward gapping is dependent on the location of the constituents in the relevant domains. For example, the source of the ungrammatically of sentences whh 'non-nommahzed' embedded clauses (of. Kornfilt 2000: examples (11)-(12)) may be the result of the non-adjacency between the embedded verb and the mam verb rather than forward gapping. Hankamer (1972, 1979) does highlight the connection between gapping and focus, a point which is relevant to the model presented in the present work but an alternative explanation for many of the examples cited in these works remains yet to be explored. 34. Noted also by Aygen (2002), however, without the particulars of the stress pattern.

A phono-syntactic template for Turkish

(41)

69

a. *Avse Ahmetin konustuzunu OGRENCILERLE biliyor. Pre-H* " FF b.

AvseAhmetm konustu&unu o&rencilerle BJUYOR. Pre-H* " FF

This shows that there is a crucial deference in how the different domains behave with respect to dislocation. Further, as observed by Kornfilt (1988) and Kihcaslan (2004), the occurrence of the same constituent in the PVP is also acceptable. This is expected under the present analysis as the PVP is the domain where constraints apply the least.

4.4. Repetition of a constituent or of a part in part-whole relations Finally, in this section I look at repeated constituents which show that there is a significant difference between the PVP and the other domains. Constituents can be repeated in Turkish for discursive reasons, although the particular function of such constructions is not clear. In (42) below, the repetition expresses surprise. When repeated, a constituent can only occur in the PVP, showing that the verb plays an integral role in the partitioning of the domains: (42)

a.

Ahmet gel-di Ahmet. Ahmet come-PF Ahmet 'Ahmethas come!'

b. * Ahmet, Ahmet geldi. (ungrammatical irrespective of where domain boundaries occur) Similar constraints apply to constituents which express a concept that is semantically part of another concept expressed by another constituent in the sentence. Bugun 'today' and bu ak?am 'this evening' are two such concepts where the latter is a part of the former. Again, the only place that the part-expressing constituent can occur in is in the PVP: (43)

a.

Bugun gikahm mi, bu ak?am? today go.out Q this night 'Shall we go out tonight?'

b. *Bu ak?am bugun gikahm mi? (ungrammatical irrespective of where domain boundaries occur)

70

AshGoksel

The data in Sections 4.1-4.4 point out that the three domains which are proposed as a basis for sentence structure in Turkish vary with respect to their syntactic behaviour. The FF is more constrained in terms of the operations it allows and the Pre-H* is more structured than the PVR The PVP is unstructured with respect to dominance and precedence hierarchies and is furthermore a domain which cannot be denned without making reference to the verb.

5. Conclusion One of the points that have repeatedly been put forward in this paper is that the syntactic grammaticality of sentences cannot be understood without taking the effects of prosody into consideration. A concept such as 'neutral intonation sentence' is justifiable, given that such sentences are presentational focus sentences. But going from here to the conclusion that these sentences therefore embody the core structural relationships in a language is not. Rather, if the task is to understand structural relations across sentence types and not the properties of one such type, then sentences which deviate from this 'neutral intonation' become part of the crucial data, as neutral intonation itself can be the source of ungrammatically. Taking a further step and assuming that such sentences form the basis of all other sentences is equally unfounded. Presentational focus sentences are not only SOV, but they also come with a particular intonation pattern: the subject cannot be in the F F * It is the particular position of focus in these sentences that allow for the percolation of stress to the sentential projection, in other words to render these sentences 'presentational focus sentences' (see Goksel and Ozsoy 2003). Above we saw the effects of placing the constituents in different domains. It thus makes little sense to base the structural analysis of sentences with different word orders and intonation patterns on a sentence which itself has a particular order and prosodic pattern. The template suggested for capturing the structural properties of Turkish uses signals from different components by combining phonological and syntactic primitives as domain boundaries. This presents us with some questions: 1. What kind of grammatical architecture can support the parallel expression of phonological and syntactic building blocks in the same representation? 35. See Nakipoglu-Demiialp (2002) for details of presentational focus sentences.

A phono-syntactic template for Turkish

71

2. Where is the 'location' of the template if it cuts across components? 3. How can this model be applied to other languages? Starting from the last question, the template is the result of language-specific empirical considerations and, as such, cannot be expected to occur universally (in whichever form). Only studies into different languages, primarily but not exclusively free word order, will provide an answer to this. A related issue will provide the answer to the second question about the 'location' of the template. The most natural host for the phono-syntactic template as a linear object made up of a specific constellation of H* and V is the lexicon, as the template is a language specific construction which is probably learned as a unit. As for the first question, the present proposal is cast within the tradition of models where serial derivation is abandoned for empirical reasons. Serial models where the output of one component of grammar is the input to another have widely been discussed in the literature, especially with respect to the balance between maintaining the technical machinery that makes a model work, and the linguistic data that they set out to explain (see Culicover and Jackendoff 2005 for a detailed criticism). One such specific proposal which is relevant to the present work is Steedman (2000). There, the components of intonation in English, e.g. H*, are pre-syntactic lexical specifications, and tunes are associated with information structural constructs such as rheme and theme. Prosody and phrase structure are different aspects of a sentence since their boundaries may totally be unaligned. In the present work H* or any such other intonational category is not associated with lexical categories, and the syntactic model does not recognize rhemes or themes. The closest the present model gets to expressing information structure in prosodic terms is the domain FF, but this domain does not correspond to a specific informational structural unit (see Section 3.1 and Note 29). Rather, prosodic and syntactic categories define a construction in tandem, and it is such a 'hybrid' construction which is the basis for syntactic operations. It is possible that the differences between these two models are a direct result of the properties of two very different languages such as English and Turkish, especially when it comes to ordering restrictions on their constituents. If it turns out that free word order languages are constructed over domains of prosodic and syntactic primitives, as suggested here for Turkish, this will present another piece of evidence for replacing derivational models with a more simple device. The present work has looked at a fragment of Turkish with this understanding in mind. It remains to be seen whether the proposal here can capture other aspects of Turkish and whether it can form a basis for the analysis of other languages.

72

AshGoksel

References Aboh, Enoch, and Roland Pfau 2011 What's a wh-word got to do with it. InMapping the Left Periphery: The Cartography of Syntactic Structures, vol. 5, Paola Benmca and Nicola Munaro (eds.), 91-124. Oxford: Oxford University Press. Arslan-Kechnotis,Ceyda 2006 Case as an unmterpretable feature. Ph. D. diss., Department of Western Languages and Literatures, Bogazici University. Aydmer,Pola 2006 The properties of the postverbal area with flat intonation in spoken Turkish. Paper presented at the Thirteenth International Conference on Turkish Linguistics, Uppsala University. Aygen(-Tosun),Gulsat 1999 Specificity and subject object positions: Scope interactions in Turkish. Unpublished ms., Harvard University. Aygen(-Tosun),Gulsat 2002 Fmiteness, case and clausal architecture. Ph. D. diss., Department of Linguistics, Harvard University. Baker, Mark 2003 Three natures of nonconfigurationahty. In The Handbook of Contemporary Syntactic Theory, Mark Baltm and Chris Collins (eds.), 407-438. (Blackwell handbooks in Linguistics.) Oxford: Blackwell. Bangoglu,Tahsm 1974 Turkce'nin Grameri [The Grammar of Turkish]. Istanbul: Baha Matbaasi. Boeckx,Cednc 2003 Free word order in Minimalist syntax. Folia Linguistica 37: 77-102. Boskovic, Zeljko, and Daiko Takahashi 1998 Scrambling and last resort. Linguistic Inquiry 29: 347-366. BozsahmCem undated Word order, word order flexibility and the lexicon. Unpublished ms.. Middle East Technical University. Cheng, Lisa, and JohanRooryck 2000 Licensing wA-in-situ. Syntax*. 1-19. Cuhcover, Peter, and Ray Jackendoff 2005 Simpler Syntax. Oxford: Oxford University Press. DemircamOmer 1996 Turkce'nin Sesdizimi [Phonology of Turkish]. Istanbul: Der Publications. DemircamOmer 2001 Turkce'nin Ezgisi [Intonation of Turkish]. Istanbul: Yildiz Technical University Press. DiScmllo, Anna-Maria, and Edwm Williams. 1987 On the Definition of Word. Cambridge, Mass, MIT Press. Dizdaroglu,Hikmet 1976 Tumcebilgisi [Syntax]. Ankara: Turk Dil Kurumu Yaymlan.

A phono-syntactic template for Turkish

73

Ediskun,Haydar 1963 Yeni TurkDilbilgisi [New Turkish Grammar]. Istanbul: Remzi Kitabevi. ErgmMuharrem 1962 Turk Dil Bilgisi [Turkish Grammar]. Istanbul: Istanbul Umversitesi EdebiyatFakultesiYaymlan. Erguvanli,EmmeEser 1984 TheMctionofWordOrderinTurMsh.Berkdey.\JmyerS^ofCamnnaPreSS. Erku,Fende 1983 Discourse pragmatics and word order in Turkish. Ph. D. diss., University ofMmnesota. Goksel, Ash 1992 Case and configurationhty. SOAS Working Papers in Linguistics and Phonetics 2,147-172. Goksel, Ash 1993 Levels of representation and argument structure in Turkish. Ph. D. diss.. Department of Linguistics, School of Oriental and African Studies. Goksel, Ash 1998 Linearity, focus and the postverbal position in Turkish. In The Mainz Meeting, Lars Johanson, Eva Agnes Csato, Vanessa Locke, Astnd Menz, and Deborah Wmterlmg (eds.), 85-106. Wiesbaden: Harxassowitz. Goksel, Ash 2001 Asymmetries between the preverbal and postverbal positions in Turkish. Paper presented at Workshop on Word Order, Bogazici University. Goksel, Ash, Meltem Kelepir, and Ash Untak-Tarhan 2008 The morpho-syntactic implications of intonation as a clause-typer In Proceedings of the 4th Workshop on Altaic Formal Linguistics. Cambridge, Mass.: MIT Working Papers in Linguistics 55, 39-50. Goksel, Ash, Meltem Kelepir, and Ash Untak-Tarhan 2009 Decomposition of question intonation: The structure of response seeking utterances. In Phonological Domains: Universals and Deviations, Janet Gryzenhout and Bans Kabak (eds.), 249-286. (Interface Explorations 16.)MoutondeGruyter. Goksel, Ash, and Ceha Kerslake 2005 Turkish: A Comprehensive Grammar. London: Routledge. Goksel, Ash, and A. SumruOzsoy 2000 Is there a focus position in Turkish? In Studies on Turkish and Turkic Languages, Ash Goksel and Ceha Kerslake (eds.), 219-228. Wiesbaden: Harrassowitz. Goksel, Ash, and A. SumruOzsoy 2003 dA: A focus/topic associated clitic in Turkish. Lingua 113: 11431167. Goksel, Ash, and Stavroula Tsiplakou 1996 Linear domains: Variations between Greek and Turkish. In Proceedings oftheSiXteenthAnnualConferenceonGreekLinguistics,193-203.TheSsalomki: University of Thessalomki Publications.

74

AshGoksel

Hale, Kenneth 1983 Warlpiri and the grammar of non-configurational languages. Natural Language and Linguistic Theory 1: 5-47. Hankamer, Jorge 1972 On the nonexistence of mirror rmage rules in syntax. In Syntax and Semantics, John P. Kimball (ed), vol. 1,199-212. New York: Seminar Press. Hankamer, Jorge 1979 Deletion in Coordinate Structures. New York: Garland. Hoffman, Beryl 1998 Word order, information structure and centering in Turkish. In Centering Theory in Discourse, Marilyn A. Walker, Aravmd K. Joshi, and Ellen F. Prince (eds.), 253-271. Oxford: Clarendon Press. issever, Selcuk 2003 Information structure in Turkish: The word order-prosody interface. Lingua 113: 1025-1053. Kayne, Richard 1994 The Antisymmetry of Syntax. Cambridge, Mass, MIT Press. Kelepir,Meltem 1996 The implications of Turkish for the theory of antisymmetry of syntax. M. A. diss., Department of Western Languages and Literatures, Bogazici University. Kelepir,Meltem 2000 Scope of negation: Evidence from Turkish NPIs and quantifiers. Proceedings of GLOW Asiall, September 1999 at Nanzan University, Nagoya, Japan. Kelepir,Meltem 2001 Topics in Turkish syntax: Clausal structure and scope. Ph. D. diss., Department of Linguistics, Massachusetts Institute of Technology. Kennelly, Sarah 2003 The implications of quantification on the role of focus in discourse structure. Lingua 113: 1055-1088. Kennelly, Sarah 2004 Pragmatics and quantificational dependencies. Lingua 114: 367-388. KihcaslanYilmaz 2004 Syntax of information structure in Turkish. Linguistics 42: 717-65. KornfilUaklm 1988 m>-dehtionandoaSemarkmginTnrkiSh.InStudiesonTurkishLinguistics. Sabn Koc (ed.), 187-215. Ankara: Mddle East Technical University Pressi KornfilUaklm 1998 On rightward movement in Turkish. In The Mainz Meeting, Lars Johanson, Eva Agnes Csato, Vanessa Locke, Astrid Menz, and Deborah Winterlmg (eds.), 107-123. Wiesbaden: Harrassowitz. KornfilUaklm 2000 Directionality of identical verb deletion in Turkish. In An Electronic Festschrift for Jorge Hankamer, Sandra Chung and Jim McCloskey (eds.). http://lmg.ucsc.edu/Jorge/kornfilt.html.

A phono-syntactic template for Turkish

75

KornfilUaklm 2005 Asymmetnes between pre-verbal and post-verbal scrambling in Turkish. In The Free Word order Phenomenon, Its Syntactic Sources and Diversity, Joachim Sabel and Mamoru Saito (eds.), 163-179. Berlin: Mouton deGruyter. Kural,Murat 1994 Properties of scrambling in Turkish. Ms., University of California at Los Angeles. Kural,Murat 1997 Postverbal constituents in Turkish and the linear correspondence axiom. LingisticInquiry2Z:A9%-52\. Lewis, Geoffrey 1967 Turkish Grammar. Oxford: Clarendon Press. Miyagawa,Shigeru 1997 Against optional scrambling. Linguistic Inquiry 28: 1-25. Nakrpoglu-Demiralp,Mme 2002 Case and dehmitedness: Meeting the interpretive requirements of the interfaces. Paper presented at the 11th Conference on Turkish Linguistics. August 7-9, 2002, University of East Mediterranean, Famagusta/Gazi Magosa. Ozge,Umut 2003 A tune-based account of Turkish information-structure. MSc diss. Informatics Institute, Middle East Technical University. Ozsoy, Sumru 2005 Topic, focus, multiple specifiers, multiple spell-out. Paper presented at the Mediterranean Syntax Meeting, University of the Aegean, Rhodes. Ozturk,Balkiz 2005 Case, Referential^ and Phrase structure. (LmmSticAktue\\/LmmSticS Today 77.) Amsterdam: John Benjamins. Pierxehumbert, Janet 1980 The Phonology and Phonetics of English Intonation. Ph. D. diss.. Massachusetts Institute of Technology. Richards, Norvin 2001 Movement in Language: Interactions and Architectures. Oxford: Oxford University Press. Sezer,Engm 1991 Issues in Turkish syntax. Ph. D. diss., Department of Linguistics, Harvard University. Sezer,Engm 2006 Defocusmg and discourse linking in Turkish. Paper presented at the 13th International Turkish Linguistics Conference, Uppsala, 16-20 August 2006. Steedman,Mark 2000 Information structure and the syntax-phonology interface. Linguistic Review 31: 649-689.

76

AshGoksel

Swift, Lloyd B. 1963 A Reference Grammar of Modern Turkish. Indiana University Publications, Bloommgton. The Hague: Mouton. Takano,Yuji 2005 Making nghtward scrambling possible. Unpublished ms., Kinjo Gakum University. Tansu,Muzaffer 1963 DurgunGenelSesBilgisiveTurkceiArticuMoryPhoneticszndTurkish]. Ankara: TDK Yaymlan. Temurcu,Ceyhan 2005 The interaction of syntax and discourse in word order variability: Data from Turkish. In Dilbilim ve Uygulamalan, Giiray Komg, I. Ozyildirim, D. Aydm, and A. Altan (eds.), 123-159. Istanbul: Multilingual. Turan,Umit,D 1998 Ranking forward-looking centers in Turkish: Universal and languagespecific properties. In: Centering Theory in Discourse, Marilyn A. Walker, Aravmd K. Joshi, and Ellen F. Prince (eds.), 253-271. Oxford: Clarendon Press. Uygun,Dilek 2006 Scrambling bare singular nominal objects in Turkish. Paper presented at the 13th International Turkish Linguistics Conference, Uppsala, 16-20 August 2006. Untak-TarhanAsh 2006 Topics in syntax-phonology interface in Turkish: Sentential stress and phases. M. A. diss., Department of Western Languages and Literatures, Bogazici University. Yarar,Emme 2006 ^-complement clauses in postverbal position. Paper presented at the 13th International Conference on Turkish Linguistics, Uppsala University, Sweden.

A coanalysis approach to Polish past tense agreement* BertholdCrysmann

The morphosyntactic status of Polish past tense agreement markers has been a matter of considerable debate in recent years (Booij and Rubach 1987; Spencer 1991; Borsley and Rivero 1994; Borsley 1999; Bahski 2000; Kupsc 2000: Kupsc and Tseng 2005). Past tense agreement is expressed by a set of bound forms that either attach to the past participle, or else "float off" to a host further to the left. Despite this relative freedom of attachment, it is often noted in the literature (e.g., Borsley 1999; Kupsc and Tseng 2005) that the combination of verbal host and agreement marker forms a word-like unit. In this paper I will argue that these agreement markers are best analyzed as affixes uniformly introduced on the verb whose inflectional features they realize. Building on the linearization-based theory of morphology-syntax interaction proposed in Crysmann (2003), syntactic mobility of morphologically introduced material will be captured by mapping phonological contributions to multiple lexically introduced domain objects. It will be shown that this is sufficient to capture the relevant data, and connect the placement of floating "affixes" to the general treatment of Polish word order (Kupsc 2000).

1. Data In this section I shall provide an overview of the basic empirical pattern. Following an in-depth discussion of the phonology, morphology, and syntax of past tense agreement markers in Section 1.1,1 shall discuss agreement marking found with the closely related conditional marker by (Section 1.2). *

The research reported here has been partially supported by the DFKI project COLLATE IL funded by the German Mmstry of Education and Research (BMBF). The results reported here have been presented at the DGfS workshop on multidimensional approaches to syntax and morphology (February 2006, Bielefeld) and at the 2006 International Conference on Head-driven Phrase Structure Grammar (August 2006, Varna, Bulgaria). The present paper extends previously published work reported in Crysmann (2006).

78

BertholdCrysmann

Table 1.

Past tense paradigm of widziec Singular fern

Plural neut

fem/neut

1

widzia-l-em

widzia-l-a-m

-

widzie-l-i-smy

widzia-l-y-smy

2

widzia-l-es

widzia-l-a-s

-

widzie-l-i-scie

widzia-l-y-scie

3

widzia-l

widzia-l-a

widzia-l-o

widzie-l-i

widzia-l-y

1.1. Polish past tense agreement Past tense in Polish is marked using a combination of a participial ending /// on the verb, inflected for number and gender, plus a person/number agreement marker that realizes subject-verb agreement in first and second person (-(e)m, -(e)s, -smy, -scie). What is special about the agreement marker is that it may either attach directly right-adjacent to the verbal participle, or else float off to the left. (1)

(Ty) widzial-es te kiqzke. you see 2.SG this book 'You saw this book.'

(2)

Ty -s widziat t? Kqtk?. you2.SGsee this book 'You saw this book.'

The floating past tense agreement markers may attach to a wide range of preverbal hosts, including nouns, pronouns, adverbs, adjectives, conjunctions (Spencer 1991). (3)

Daleko-m poszla. far l.SG went 'I went a long w a y '

(4)

Wdomu -scie to zrobili? at home 2.PL that made 'Did you make it at home?'

However, realization in absolute clause-initial position is barred, a property shared with syntactic clitics in Polish (e.g., pronominal clitics, see Kupsc 2000),

A coanalysis approach to Polish past tense agreement

79

which is standardly interpreted as an instance of the Tobler-Mussafia Law (cf. Bahski 2000), a prosodic constramt that bans enclitics from absolute domaininitial position. (5)

*s widzial tf kiqzk? 2.SG see this book

In postverbal position (i.e., attached to the participle), past tense agreement markers display a good deal of interaction with lexical phonological rules, namely, assignment of primary lexical stress, word final vowel raising, and yer vocalization. However, in preverbal position, no systematic interactions can be observed (Bahski 2000). Yer vocalization is a systematic vowel/zero alternation in Polish, argued by Booij and Rubach (1987) to be a cyclic lexical phonological rule. Within the domain of the word, an underlying "yer" is realized as [e], if followed by another yer, or else deleted. Booij and Rubach (1987) relate the vowel/zero alternation observable with the past tense agreement markers to this well-attested rule. Since the domain of application is the word, it follows that vowel/zero alternation at the juncture between the past tense agreement marker and the verbal host suggests that these forms combine in the lexicon. (6)

a. robtl did.M.SG 'he did' b. robile -m did.M.SG l.SG 'I(masc.)did' c. robila -m did.F.SG l.SG 'I (fern.) did'

Another morphophonological rule that points in the same direction is raising of o to 6 (= [u]) in word final syllables before voiced consonants (Booij and Rubach 1987). Since attachment of past tense agreement apparently blocks the application of raising, Dogil (1987) concludes that these markers must already be attached when this lexical phonological rule applies. (7)

a.ja-m mogl I l.SG could 'I could'

80

BertholdCrysmann

b. ja moglem I could.l.SG 'I could' Finally, lexical stress in Polish regularly falls on the penultimate syllable of the prosodic word. If a singular past tense agreement marker is attached to the participle, lexical stress assignment to the penult takes the extra syllable resulting from yer vocalization into account. (8)

a. rob* did.M.SG 'he did' b. robil -em did.M.SG l.SG 'I did'

For plural markers, there is some variation amongst speakers: stress placement is either on the antepenult or the penultimate syllable, including the agreement marker: (9)

a. robilr did.M.PL 'they did' b. robitt -smy did.M.PL l.PL 'we did' c. robili-smy

If we turn to preverbal realization of said markers, we find that none of the above morphophonological effects can occur at the juncture between the floating agreement marker and its phonological host (Bahski 2000): neither yer vocalization, nor stress shift can be observed, i 1. As discussed by Booij and Rubach (1987) and Kupsc and Tseng (2005), there is a small set of hosts like, e.g.Jafc ' a s ' J « z 'already', chociaz 'although', sam 'alone' that do feature e -epenthesis when followed by a past tense marker. Although these forms are considered archaic by Kupsc and Tseng (2005), an account of Polish past tense agreement should nevertheless be able to provide an account of these forms: I would therefore suggest that these forms might be analyzed as modal verbs which

A coanalysis approach to Polish past tense agreement

(10)

Yer vocalization:* a. i. palc-a fingcr.GEN ii. palc-a fingcr.GEN palec fingerACC ii. *palece fingerACC

81

-mis 1/2.SG

b. i.

-mis 1/2.SG

Likewise, raising applies, as if the agreement marker were not there. (11)

Raising: a. i. krowy cows.NOM/ACC n. * W =icie cows.GEN 2.PL b. i.

krow COWS.GEN

ii. Vcrow =scie cows.GEN 2.PL Failure to undergo an expected and otherwise fairly regular morphophonological alternation constitutes evidence that, pre-verbally, these markers do not morphophonologically integrate with their host. The only phonological restrictions ("phonological friendliness") that do seem to hold between the floating agreement marker and its preverbal host concern the host's final segmental material, in particular sonority of final segments and complexity of the coda. In contrast to Kupsc and Tseng (2005), who regard this as a morphophonological idiosyncrasy, Bahski (2000) argues that the phonological selectivity can be explained in entirely prosodic terms, drawing on the sonority hierarchy. He argues further that the availability of phonologically less marked alternative attachment sites accounts for the low acceptability observable with suboptimal hosts. If we also consider further that non-local realization of agreement is a subcategory for an umnflected participle, akin to the conditional and future tense auxiliaries by and bedzie. This issue will be addressed in Section 3.4. 2. The vowel/zero alternation b e t w e e n ^ and palca suggests that palec is underlymgly yer-fmal. In contrast to verbal participles, attachment of the agreement marker does not make the stem-final yer surface as [e].

82

BertholdCrysmann

Table 2.

Conditional paradigm of widziec Singular

Plural

fern

neut

masc

fem/neut

1

widzia-l-by-m

widzia-la-by-m

-

widzie-l-iby-smy

widzia-l-y-by-smy

2

widzia-l-by-s

widzial-aby-s

-

widzie-l-iby-scie

widzia-l-y-by-scie

3

widzia-l-by

widzia-la-by

widzia-l-o-by

widzie-l-i-by

widzia-l-y-by

probably a marked option by itself- although cross-lmgmstically attested, it is not an option chosen by too many languages of the world - , unacceptability of cliticization to unfriendly hosts may well be accounted for by having two strikes against it: one prosodic, the other morphosyntactic. 1.2. The conditional auxiliary^ The Polish conditional marker by displays some striking parallelism to the past tense agreement marker: first, just like the past tense, the conditional is expressed by a combination of the participial form of a verb (inflected for number and gender) plus the auxiliary by, which is inflected for person and number. The form of the person/number markers is identical to past tense markers. Furthermore, the forms of the conditional marker by obey conditions on placement similar to those regulating the distribution of the past tense agreement marker: Postverbally, there is almost strict adjacency to the verb, the only exception being intervention of the particle -no (Kupsc, see Borsley 1999: fn. 12). (12)

Obejrzalno -bys ten film! NO COND.2.SG this film 'You would see this film!'

(13)

*Obejrzalno -s ten film! NO 2. SG this film

Preverbally, attachment is promiscuous, again with a ban on clause-initial position.

A coanalysis approach to Polish past tense agreement

83

With respect to morphophonology, however, the conditional marker does not display any of the expected properties of affixal attachment: forms oUy are entirely stress-neutral, regardless of their host. (14)

a. robtl did.M.SG 'he did' b. robtl-by did.M.SG-COND 'he would do' c. *robil-by did.M.SG-COND

(15)

a. robilt did.M.PL-COND 'they did' b. robili-by did.M.PL-COND 'they would do' c.

-robtli-by

Likewise, application of raising is entirely unaffected by the attachment of by (16)

a.

mag could.M.SG 'he could'

b.

moglby could.M.SGCOND 'he would be able to'

c. *moglby could.M.SGCOND Thus, I will follow Spencer (1991), Bahski (2000), and Kupsc and Tseng (2005) in that morphophonological evidence points towards their status as syntactic clitics. This difference in status is further corroborated by coordination data (cf Kupsc and Tseng 2005; Bahski 2000): while wide scope over a coordination of hosts is by-and-large impossible with past tense agreement attached to a verbal host (participle or copula), conditional markers easily take wide scope in this position.

84

BertholdCrysmann

(17)

Poszedl -em r zobaczyl *(-em). go.PAST l.SGand see.PAST.MASC l.SG 'I went and saw.'

(18)

Bylr -sorer jest *(-escre). be.PAST2.PLandbe.PRES2.PL 'You were and you are.'

(19)

Wlqczyl -bym sobre radio r posluchal turn.on COND.l.SG SELF radio and listen muzykr. music 'I would turn on the radio and listen to the music.'

(-bym) COND.l.SG

Preverbally, both markers may take wide scope (Kupsc and Tseng 2005). Another difference between past tense agreement and conditional markers concerns the degree of interaction with pronominal clitic placement. As observed by Kupsc (2000), Polish pronominal clitics either all precede or immediately follow the verb. Forms of clitic -by are always realized to the left of the pronominal clitics, regardless of whether by itself is realized in pre- or in postverbal position (Borsley 1999; Witkos 1997). (20)

Ty bys go wrdzraljutro. you COND.2.SG3.SGseen tomorrow 'You would see him tomorrow.'

(21)

?*7> go bys wrdzraljutro. you 3.SGCOND.2.SGseen tomorrow

(22)

Ty wrdzralbys go jutro. you seen COND.2.SG 3.SGtomorrow

(23)

l*Ty go wrdzralbys jutro. you 3.SG seen COND.2.SG tomorrow

Preverbal forms of the past tense agreement marker pattern with -by. Postverbal forms, however, show no interaction with pronominal clitic placement (Witkos 1997; Borsley 1999).

A coanalysis approach to Polish past tense agreement

(24)

Ty -s go widzialwczoraj. you 2.SG3.SG seen yesterday 'You saw Mm yesterday'

(25)

?*7> go -s wrdzralwczoraj. you 3.SG2.SGseen yesterday

(26)

Ty widziale-s go wczoraj. you seen COND.2.SG 3.SGyesterday

85

It seems thus that the difference in lexical status suggested by morphophonology between postverbal past tense agreement on the one side, and the conditional marker and preverbal past tense agreement on the other, is also reflected in terms of syntactic visibility.

1.3. Summary To summarize the empirical observations made above, I conclude that the status of Polish past tense agreement presents us with an analytical paradox: while postverbal realization of this marker suggests affixal status - as supported by their morphophonological properties, the strict adjacency requirement, the noninteraction with pronominal clitic placement, and the failure to take wide scope over a coordination of hosts - , preverbal realization, however, suggests syntactic clitic status - as witnessed by promiscuous attachment and the lack of morphophonological integration with the host. Nevertheless, pre- and postverbal realizations need to be systematically related in order to account for the identity of formatives and the unique marking of a verbal inflectional category. The forms of the conditional marker by, however, are probably best analyzed as syntactic clitics, regardless of position, since there is absolutely no evidence for morphophonological integration with their host, the adjacency requirement is not strict, they can take wide scope over a coordination of hosts, and they interact with pronominal clitic placement. Still, the inflected forms of the conditional marker should be related to the past tense agreement markers.

2. Previous analyses Probably the first generative approaches that do full justice to the apparent lexical properties of Polish floating agreement markers are the analyses by Dogil

86

BertholdCrysmann

(1987) and Booij and Rubach (1987). Based on the observation that the morphophonology of these markers displays sinking parallelism to other lexicalmorphological processes, Booij and Rubach (1987) suggest that attachment of the past tense agreement markers applies in the lexicon, regardless of the host. Thus, both postverbal suffixal realization and preverbal realization are derived by essentially the same morphological process. In order to rule out double realization*, Booij and Rubach (1987: 36) invoke a syntactic filter that bans all configurations in which two identical clitics surface. However, the way in which this filter is specified clearly runs counter to the Lexical Integrity Hypothesis, given that it makes full reference, in syntax, to sublexical structures. The model of Booij and Rubach (1987) also displays some shortcoming on the empirical level: as pointed out by Bahski (2000), a uniformly morphological theory of Polish past tense agreement, such as Booij and Rubach's, cannot derive the apparent lack of morphophonological effects with the great majority of nonverbal hosts, as witnessed e.g. by stress assignment, word-final raising, and absence of yer vocalization discussed above. It appears, thus, that the lexical theory presented by Booij and Rubach (1987) overemphasizes the parallelism observable with isolated lexicalized non-verbal hosts at the expense of a wide range of open-class host categories that show a different phonological pattern. In his recent in-depth study of Polish clitic agreement, Bahski (2000) observes that the phonological behaviour of the past tense markers differs depending on the host: while verbs and a handful of non-verbal hosts provide evidence for a tight morphophonological integration, this is not the case for the majority of non-verbal hosts. Instead of drawing a distinction between lexical and post-lexical attachment, he suggests instead to differentiate between cliticization at the X°-level vs. integration at the level of XP It follows, then, that, in Bahski's model, the phonology of Polish clitic agreement is uniformly associated with the level of post-syntactic phonology. Choice between X° and XP cliticization in Polish appears to be lexically governed. While Bahski's solution might make sense in a framework, such as Distributed Morphology, where late insertion of Vocabulary is assumed, it is pretty much incompatible with any other current theory of grammar, which tend to employ a more standard notion of the lexicon. Probably the first study of this set of phenomena in the framework of HPSG is Borsley (1999). In his paper, he focusses on the similarity in syntactic distribution between the past tense agreement marker and the conditional marker 3. Booij and Rubach (1987) also present data from substandard variants exhibiting double realization of the agreement markers.

A coanalysis approach to Polish past tense agreement

87

and develops an essentially parallel analysis of these markers in terms of weak auxiliaries. In order to capture the difference in syntactic mobility between preverbal and postverbal realization, he suggests that in preverbal position, these auxiliaries are syntactically independent signs, which take a participial syntactic complement, whereas postverbally, these auxiliaries are regarded as part of a morphologically derived verb-auxiliary complex. Syntactic realization in postverbal position is ruled out by a suitable LP constraint. Uninflected third person forms receive special attention: since an empty auxiliary analysis will give rise to spurious ambiguity, he suggests instead that third person finite past tense forms are derived from the non-finite participle by way of a unary conversion rule. There are, however, a few problems with this account in the light of the data discussed above: first, as pointed out by Kupsc and Tseng (2005), a uniform treatment of past and conditional cannot do full justice to the apparent differences in morphological status, as witnessed by morphophonological behaviour and the coordination facts. Second, deriving postverbal weak auxiliaries uniformly as a syntactically opaque daughter of a lexical compound cannot model the observable difference in interaction with pronominal clitic placement, which suggest that postverbal conditional markers must be syntactically visible, in contrast to postverbal past tense agreement. Third, the morphological analysis put forth in Borsley (1999) is inherently asymmetrical, postulating a lexical incorporation analysis for the conditional and nonthird person past tense auxiliaries on the one hand, and an analysis in terms of zero inflection on the other. Finally, it is far from obvious how the weak auxiliary analysis of the past tense agreement markers can be generalized to derive other inflected forms that draw on the same set of markers, including the conditional marker and the present (!) tense c o p u l a ^ . Identity of exponence across different paradigms therefore favors an analysis of the past tense agreement marker as an inflectional affix, realizing person and number specifications. In a recent paper, Kupsc and Tseng (2005) have argued for a non-uniform account of conditional auxiliaries and past tense agreement, according to which the former are considered to be syntactic clitics, whereas the latter are analyzed as morphologically derived agreement affixes. The authors, however, do not Table 3.

Present tense paradigm of jest Singular

Plural

~

jestem

jestesmy

2

jestes

jestescie

3

jest

sq

88

BertholdCrysmann

assign a difference in status to preverbal and postverbal occurences of the past tense agreement marker, but assume instead that the past tense agreement marker always attaches to its surface host as an inflectional affix. In order to relate the non-local realization of the agreement marker to the verbal inflectional features they are an exponent of, they suggest a special feature percolation mechanism using marker and trigger features. Essentially, the locally uninflected participle launches a trigger feature, inflection of a host for person/ number agreement launches a marking feature, and a unary clause-level schema discharges both features under unification. In essence, their approach can be regarded as a feature-based reincarnation of the syntactic filter suggested by Booij and Rubach (1987). Although I concur with Kupsc and Tseng (2005) in regarding postverbal past tense agreement markers as suffixes directly attached to their hosts, extending this perspective to their preverbal counterparts raises several issues. which I will briefly discuss: first, the feature percolation mechanism invoked by the authors does not connect past tense agreement to any well-understood subtheory of local or non-local phenomena in Polish or across languages. Likewise, past tense agreement appears as an isolated agreement process unrelated to other agreement processes in the language. Second, the syntactic similarity between preverbal past tense agreement markers and conditional auxiliaries remains unaccounted for. Third, and most importantly, Kupsc and Tseng (2005) do not provide evidence that preverbal past tense agreement markers show a similar degree of morphophonological integration with the host as their postverbal counterparts: in contrast to postverbal agreement, none of the expected lexical phonological rules may apply at the juncture between preverbal agreement markers and their hosts, like, e.g., stress shift or yer vocalization. Conversely, the observable conditions on phonological friendliness are probably best understood in prosodic terms (Bahski 2000). Finally, promiscuous attachment, i.e., the low degree of selection towards syntactic or morphological properties of the host (Criterion A of Zwicky and Pullum 1983) does not seem to support an analysis in terms of direct morphological attachment either.

3. A coanalysis approach In the analysis which I am going to propose I will try to synthesize the insights gained by Bahski (2000), Borsley (1999), and Kupsc and Tseng (2005) and assign the status of a syntactic clitic to the conditional marker regardless of position, yet treat the past tense agreement marker as a morphosyntactic hybrid:

A coanalysis approach to Polish past tense agreement

89

building on proposals by Kathol (1995) and Crysmann (2003), I suggest that Polish past tense verbs can contribute more than one domain object to linear domain structure, the level of representation relevant for word order in recent versions of HPSG. As a result, morphological rules of exponence will uniformly introduce exponents of agreement on the verbal host, yet the mapping of lexically introduced phonology to domain objects will permit the "affix" phonology to float off. The analysis of preverbal markers as syntactically visible floating affix phonology will prove to capture, in a straightforward way, the interaction with pronominal clitic placement, predict the lack of phonological integration with prosodic hosts, and account for uniqueness of exponence. Furthermore, this analysis not only connects the placement of floating past tense agreement to the standard HPSG approach to Polish word order (Kupsc 2000), but it also relates the phenomenon at hand to the strikingly similar case of floating subject agreement in Udi (Crysmann 2000).

3.1. Morphology As to the morphological status of the past tense markers -{e)m, -(e)s, -smy. -sate, I follow Kupsc and Tseng (2005) and assume that they are best regarded as exponents of person/number agreement rather than tense auxiliaries. This view is supported by a variety of considerations: first, the forms used in the conditional are identical to the ones used in the past, yet they do not select the participial form of by. The very same holds for the present tense c o p u l a ^ . Second, an analysis as tense auxiliaries would assign these forms the status of sign, which would make the wrong prediction concerning the interpretation of inflected forms of the present tense copula jest, which is clearly non-past.^ Third, zero marking of third person also favors an affixal treatment over a compound analysis. I therefore suggest to represent the person/number markers as an inventory of pure forms (exponents - not morphemes)^

4. Historically, the past tense markers are sard to derive from the cop^ jest 'to be'. See Booij and Rubach (1987: 42) and references cited there. 5. The paradigms generated by the r e l a t i o n a l schemata given here are all finite paradigms. As a consequence, we can localize the encoding of past tense with the constraint on /-forms given in the STEM dimension. Certainly, there is also a non-finite use of the /-form in periphrastic tenses such as future or conditional. This non-finite use may be licensed by a morphological schema of its own, which is simply the identity function.

90 BertholdCrysmann (27)

pst-agr PER pst-lst

NUM| pst-2nd

pst-sg

pst-pl

The forms are then selected by r e l a t i o n a l schemata: following previous work on type-based realization* morphology (Koenig 1999; Riehemann 1998: Crysmann 2003), I suggest to organize the realizational schemata into a twodimensional type hierarchy for affix and stem selection, where dimensions are conjunctively connected. This means that the range of well-formed fully resolved lexeme categories is obtained by combining subtypes from each dimension under unification, here, the subtypes in the STEM-SELECTION and AFFIXATION.

(28) M(

stem ) 0 list(pst-agr) HD verb

AFFIXATION

[M(/...

stem HD VFOl IMl

| STEM-SELECTION

\"

"7

[M(/...

stem PH (jest)

\1 [M(/...

-)]

stem PH (by)

\1

-)]

What is crucial for our analysis is that the relative order of stem and affix is not fixed on the supertype, as indicated by the shuffle operator (o), an orderindependent operation for combining lists, which states that the M(ORPH) list consists of the verbal stem and a (possibly empty) list of pst-agr affixes, in any order. In other words, the supertype merely constrains list membership, without actually restricting the order in which the members of the list may occur. Ordinary order-sensitive list concatenation, by contrast, is expressed by®.

A coanalysis approach to Polish past tense agreement

M(

stem HD verb

AFFIXATION

PER

91

) O Ust(pst-agr)

STEM-SELECTION

NUM

Under the natural assumption that the domain of application for morphophonological rules is the morphological structure, presence vs. absence of morphophonological effects can be simply related to the configuration found at this level: with suffixation, yer vocalization and stress shift will be triggered. With prefixation. the local condition for rule application is simply not met. Likewise, raising will be possible, if the stem is final, yet will be blocked by following affixal material. The possibility for affixes to be positionally non-fixed is quite common cross-linguistically: Morphologically conditioned positional alternation has been attested for French and Italian pronominal affixes (Miller 1992; Monachesi 1999), whereas morphosyntactically conditioned placement alternation of affixes has been observed for German separable particle verbs (Kathol 1995), European Portuguese pronominal affixes (Crysmann 2003; Luis and Spencer 2004), and Udi agreement (Harris 2000; Crysmann 2003).

3.2. Morphosyntactic mapping Having establishedhow agreement formatives are introduced into morphological structure, we can now proceed to the specification of the morphology-syntax interface: as already mentioned above, the key to our analysis of morphologically introduced, yet floating agreement markers is a natural extension of Linearization HPSG (Kathol 1995; Reape 1994), namely the possibility for lexical signs, just like phrasal signs, to introduce more than a single domain object,

92

BertholdCrysmann

an idea that has already been explored in the analysis of morphosyntactic paradoxa in German (Kathol 1995), European Portuguese, Fox, and Udi (Crysmann 2000, 2003). In order to preserve lexical integrity, morphological entities are not directly accessible to syntactic manipulation. Rather, it is only the phonological contribution of morphological entities that gets distributed over the lexically introduced domain objects. Interaction between surface syntax and morphotactics is limited to ordering: as guaranteed by the homomorphism constraints below, the sequence of PHON values on DOM must correspond to the sequence of PHON values in morphological structure. (30) DOM

const

(

M PH

hi

9

• *•J

PH

S

)]

El©... © S

PH

word

PH


/

I.e., in contrast to optimality-theoretic approaches, interaction between morphological constraints and syntactic constraints is monotonic, embodying a cooperative, rather than competitive view of the interface. All we need to do now to account for the difference in syntactic transparency between pre- and postverbal realizations of the agreement marker is to assume that Polish past tense verbs align their stem phonology with the right-most domain object. (31) DOM

list © ( PH \B © list 1 stem

M

( \

PH HD

ED verb

) 1

\ ) O Ust(pst-agr)

1

As a result of the interaction between the morphologically variable position of the agreement affix and stem alignment, we will obtain two different surfacesyntactic representations: - a pre-stem position, which is syntactically transparent,

A coanalysis approach to Polish past tense agreement

93

(32) DOM I

PH

( s \ , PH/widzian

- and a post-stem position, winch is syntactically opaque. (33) DOM I

PH (widziales)

A sample derivation of floating agreement will thus look as follows: (34)

DOM ( H

DOM ( U

PH

(i\

SS ILI CAT | HD E

LOC | CAT | HD H

SSE

The syntactic inseparability of the conditional auxiliary and the morphologically attached agreement markers can straightforwardly be captured by restricting the length of the lexical DOM list to 1, enforcing realization as an inseparable suffix. (35)

)

1

DOM

list © / PH [TJ © list

M

/ stem \ P H ED > 0 Ust(pst-agr) ( \ HD verb

1

A DOM(

)

\LJ/_

94 BertholdCrysmann Given that we analyze the conditional marker by, be it inflected or not, as a weak auxiliary, or, in other words, a syntactic clitic, how then is the agreement marking it bears related to the subject of the main verb participle? Following Borsley (1999), I shall assume that by selects the main verb participle as its complement. Along the lines of the standard HPSG treatment of auxiliaries (Pollard and Sag 1994), by is analyzed as a raising verb, inheriting the unsaturated subject valence from its verbal complement, as depicted by the lexical entry for by given below. As a result of raising, the main verb's unsaturated subject valence is locally accessible on the auxiliary's SUBJ list, where it can be targetted by the realization^ schemata for person/number agreement. (36)

stem PH

w

verb HD AUX +

SS L CAT

SUBJ

COMPS

E

L CAT

HD SUBJ

VFORM I H

3.3. Clitic order The final piece in our analysis of the data at hand concerns the syntactic placement of clitics. I assume that clitic status in Polish is probably best defined prosodically, e.g., in terms of prosodic extrametricality, an assumption that will directly predict the effects of the Tobler-Mussafia Law (cf Section 1.1). In the following, I will use the types nonclitic and clitic as mere short-cuts to refer to domain objects whose PHON starts with a prosodic word boundary, or not. In order to model the restrictions on clitic placement observed above (see Kupsc 2000 for a more in-depth study) a set of 3 LP constraints appears sufficient to derive the basic pattern: - Verbal clitics precede pronominal clitics.

A coanalysis approach to Polish past tense agreement

(37) DOM / ...

clitic HD noun

clitic HD verb...

95

>] /

\ - Clitics either all precede or follow the verb. (38)

nonclitic HD verb

DOM / ... clitic

clitic

)]

- Postverbal clitics must be verb-adjacent. (39)

nonclitic DOM

HD

nonclitic

verb

clitic

d v e n the constraint on verb adjacency for post-verbal clitics, formation of a clitic cluster turns out to be a mere corollary. Thus, the patterns of placement interaction between pronominal clitics on the one side and the conditional and past tense agreement markers on the other will be derived as follows: - Preverbal agreement marker must precede all other clitics. (40) a. r

clitic

nonclitic DOM

(tij

PH

b. nonclitic DOM

PH

(ty)

clitic

nonclitic

PH

U)

PH

^gO^

PH

/widzial)

HD

verb

HD

noun

HD

verb

clitic

clitic

nonclitic

PH

(so)

PH

U)

PH

/widzian

HD

noun

HD

verb

HD

verb

- Postverbal agreement marker may follow preverbal clitics.

(41) nonclitic DOM

PH

(ty^

clitic

nonclitic

PH

^go^

PH

/widziales)

HD

noun

HD

verb

96 BertholdCrysmann Conditional by must always precede all other clitics. (42) a. DOM

PH

(iy}

PH

(bys^

PH

HD

verb

HD noun

b.

clitic

nonclitic

PH

^gO^

(widzial)

HD verb nonclitic

clitic ^gO^

PH

(bys\

PH

/widzialy

HD noun

HD

verb

HD

verb

PH

DOM PH

nonclitic

clitic

clitic

nonclitic

(43) clitic

nonclitic DOM

/

PH

PH

(bys\

HD

verb

3.4. L e g a l i z e d non-verbal hosts As pointed out at various points in our discussion above, there is a circumscribed set of non-verbal hosts that pattern strikingly with postverbal attachment, as far as the morphophonology is concerned. Most notably, these hosts give rise to vowel/zero alternations, which have previously been analyzed as instances of yer vocalization (Booij and Rubach 1987). Although the set of hosts displaying such tight phonological integration is fairly limited and the forms are considered archaic by some authors (Kupsc and Tseng 2005), it is nevertheless a phenomenon that needs to be addressed by any theory of Polish floating agreement that claims for itself to be complete. Given the parallelism in morphophonology, I shall assume that forms, such wjakem ovjuzes, are indeed lexically derived, on a par with verb-clitic combinations, which display an identical behaviour, as far as morphophonology is concerned. While it would be straightforward to invoke a theory like the one suggested by Booij and Rubach (1987) or Kupsc and Tseng (2005) to take care of the problem, it should be evident that this would lead to a solution which is quite baroque. Worse, such a theory would face the same motivational problems as

A coanalysis approach to Polish past tense agreement

97

the ones we have raised above against the account provided by Kupsc and Tseng (2005). A first step to incorporate the treatment of lexicalized non-verbal hosts, in­ volves an extension of the realization* schemata, licensing inflection of these particles with person/number agreement markers. Thus, in addition to partici­ ples and the auxiliary by and c o p u l a ^ , we shall include particles such as jak and Jut into the set of stems that can serve as lexical hosts for the person/ number markers under discussion. In essence, what I am suggesting here is that these particles function as modal auxiliaries, adopting a type-raising approach akin to Kim and Sag (1996). In order to license inflection of these particles, all we need to do is to add appropriate morphological schemata to the STEM-SELECTION dimension, as shown below for the particle./** 'already'. Intersection of this schema with the realizations schemata for person/number agreement will provide the full set of inflected forms of the particle. Similarly to auxiliary by, these inflected parti­ cles subcategonze for a non-finite participle ([VFORM I\), thereby ruling out double inflection on both the particle and the verb. Still analogous to auxiliary by, the type-raised version of the particles are raising verbs, making the main verb's subject valence locally accessible for the inflectional schemata (indi­ cated by boxed "1"). Type-raising from a modifier-head relationship to a headcomplement one is performed by means of inheritance of the stem's MOD value onto the inflected particle's COMPS list (boxed "2"). (44) DOM

verb HD AUX + SS | L | CAT

SUBJ

1

COMPS

< \2\ L | CAT | HD | VFORM l

[stem PH ^juz˙^ M HD

MOD 2

SS L | CAT SUBJ

()

COMPS

()

L | CAT

HD

verb

SUBJ

1

98

BertholdCrysmann

Thus, by adopting aperspective of inflected particles as lexically type-raised auxiliaries we can integrate these archaic and isolated forms into our general theory of Polish floating agreement. Furthermore, the representation of the type-raising rule as l e g a l i z e d schemata straightforwardly accounts for the isolated, l e g a l i z e d nature of the phenomenon. Finally, since agreement markers and host are combined already at the level of morphology, we can model the fact that these inflected particles are subject to morphophonological rules such as yer vocalization. 4. Conclusion In the present paper, I have argued that the syntax and morphology of "floating" agreement markers in Polish can receive a unified treatment under the assumption that they are uniformly introduced as agreement affixes on the verb. Morphological introduction as exponents of person/number agreement naturally accounts for the paradigm-like properties, including zero exponence and crossparadigm parallelism. An analysis as morphologically introduced affixes also relates syntactic opacity and morphophonological properties, and derives the lexical-phonological effects (and lack thereof) by reference to the domain of application: morphological structure. The adoption of a lexically-controlled coanalysis approach has proven to reconcile the affixal properties of postverbal markers with the syntactic mobility of their preverbal counterparts, capturing uniformity of markers and uniqueness of exponence. The specific nature of the morphology-syntax interface in terms of multiple lexically introduced domain objects aligns the treatment of floating "affixes" with the general approach to Polish word order (Kupsc 2000). Finally, the account presented here for Polish floating affixes is highly reminiscent to the analysis of similar phenomena in Udi (Crysmann 2000, 2003). References Bansk^Rotr 2000 Morphological and prosodic analysis of auxiliary clitics in Polish and English. P h D . thesrs, Umwersytet Warszawskr Booij.Geert.andJerzyRubaoh 1987 Postcychc vs postlexrcal rules in lexical phonology. Linguistic Inquiry 18: 1-44. Borsley, Robert D. 1999 Weak auxiliaries, complex verbs and inflected complementizers in Polish. In Slavic in Head-Driven Phrase Structure Grammar, Robert D. Borsley

A coanalysis approach to Polish past tense agreement

99

and Adam Przepiorkowski, (eds.), 29-59. (Studies in Constraint-Based Lex 1 cal 1 sm2.) Stanford: CSLI Publications. Borsley, Robert, and Maria-Luisa Rivero 1994 Clitic auxiliaries and incorporation in Polish. Natural Langiuage and Linguistic Theory 12: 373-422. CrysmannBerthold 2000 On the placement and morphology of Udi subject agreement. Paper presented at 7th International Conference on Head-driven Phrase Structure Grammar, Berkeley. CrysmannBerthold 2003 Constraint-based Coanalysis: Portuguese Cliticisation and MorphologySyntax Interaction in HPSG. (Saarbrucken Dissertations in Computational Linguistics and Language Technology.) Saarbrucken: Computational Linguistics, Saarland University and DFKI LT Lab. CrysmannBerthold 2006 Floating affixes in Polish. In The Proceedings of the 13th International Conference on Head-driven Phrase Structure Grammar, Stefan Muller (ed.), 123-139. Stanford: CSLI Publications. Dogil,Grzegorz 1987 Lexical phonology and floating inflection in Polish. Phonologica 5: 39-47. Harris, Alice C. 2000 Where in the word is the Udi clitic? Language 76: 593-616. Kathol, Andreas 1995 Linearization-based German syntax. Ph.D. thesis, Ohio State University. Kim, Jong-Bok, and Ivan Sag 1996 French and English negation: A lexicahst alternative to head movement. Ms., Stanford University. Koemg, Jean-Pierre 1999 Lexical Relations. Stanford: CSLI Publications. Kupsc,Anna 2000 An HPSG grammar of Polish clitics. Ph.D. thesis, Polish Academy of Sciences and Umversitede Pans 7. Kupsc, Anna, and Jesse Tseng 2005 A new HPSG approach to Polish auxiliary constructions. In The Proceedings of the 12th International Conference on Head-Driven Phrase StructureGrammar, StefanMuller(ed.),253-273. Stanford: CSLI Publications. Luis, Ana, and Andrew Spencer 2004 A paradigm function account of 'mesochsis' in European Portuguese. Yearbook of Morphology 177-228. Miller, Philip 1992 Clitics and Constituents in Phrase Structure Grammar. (Outstanding Dissertations in Linguistics.) New York: Garland. Monachesi,Paola 1999 A Lexical Approach to Italian Cliticization. Stanford: CSLI Publications.

100

BertholdCrysmann

Pollard, Carl, and Ivan Sag 1994 Head-driven Phrase Structure Grammar. Chicago University Press. Reape,Mike 1994 Domain union and word order variation in German. In German in HeadDriven Phrase Structure Grammar, John Nerbonne, Klaus Netter, and Carl Pollard, (eds.), 151-197. (CSLI Lecture Notes 46.) Stanford: CSLI Publications. Riehemann, Susanne 1998 Type-based derivational morphology. Journal of Comparative Germanic Linguistics 2:49-77. Spencer, Andrew 1991 Morphological Theory. An Introduction to Word Structure in Generative Grammar. (Blackwell Textbooks in Linguistics 2.) Oxford: Blackwell. Witkos,Jacek 1997 Polish inflectional auxiliaries revisited. Paper presented at the 30th Poznan Linguistic Meeting, Adam Mickiewicz University, Poznan. Zwicky, Arnold M., and Geoffrey K. Pullum 1983 Chticization vs. inflection: English*'/. Linguistics 59: 502-513.

Optimal specifications: On case marking in Polish* BerndWiese

1. Introduction Polish noun inflection shows paradigms of case forms that combine features of the flexive type and of the agglutinative type of morphological formations. As I shall argue in Section 2 (which presents relevant data), this mixture provides a particular challenge for any approach that takes seriously the morphological forms (and their form-related properties) used in case marking. Section 3 starts from some well-known observations on differential case marking and case syncretism that turn out to be crucial when the interplay of gender and case is to be explained. Section 4 provides a detailed analysis of the inflectional system of Polish nouns that avoids the rampant multiplication of paradigms and declensions found so often in Polish grammars. On the basis of multi-level classification systems for gender and case that are supported by internal and external evidence, a limited number of noun endings are identified, which, for the most part, are given unambiguous categonal specifications and conditions of application that predict their distribution over inflectional forms or 'cells' of paradigms. Section 5 adds a short conclusion, i

*

This paper was presented at the Workshop on Theoretical Morphology (WoTM) 1. June 14, 2005, University Leipzig, and at the 28 Jahrestagung der Deutschen Gesellschaft fur Sprachwissenschaft (DGfS), February 22-24, 2006, University Bielefeld; I would hke to thank the audiences for stimulating feedback. The study presented is part of the project Grammatik des Deutschen im europaischen Vergleich (principal investigator Gisela Zifonun) at the Institut fur Deutsche Sprache (IDS), Mannheim, Germany. I am most grateful to the members of the project group for many productive discussions. Special thanks to Marek Konopka (IDS) for help with the Polish data. 1. The following abbreviations will be used: m. (masculine), f (feminine), n. (neuter): amm. (animate), man. (inanimate); pers. (personal), impers. (impersonal), hon (honorific); sg. (singular), pi. (plural); nom. (nominative), voc. (vocative), gen. (genitive), dat. (dative), ace. (accusative), abl. (ablative), loc. (locative), ins. (instrumental); dir. (direct), obi. (oblique); vel. (velar).

102

BerndWiese

Table 1.

singular plural

Case forms in Turkish

nom.

ace.

gen.

dat.

loc.

abl.

ev evler

evi evleri

evin evlerin

eve evlere

evde evlerde

evden evlerden

EV ('house')

2. Agglutination vs. inflection Case forms may be classified by two different types of criteria, viz. according to form and function, respectively (Comne 1986): (i) case forms are distinguished and are classified in terms of the occurrence or non-occurrence of pertinent morphological markers (or 'formatives' or 'exponents'), (ii) case forms are distinguished and are classified in terms of their syntactic potential as in traditional approaches. In agglutinative systems, formal and functional classifications may largely coincide as may be exemplified from Turkish. By standard analyses, Turkish (cf. Table 1) possesses six cases. The nominative (of the singular) exhibits the bare base form. As for the remaining cases, there are special endings each marking one and only one case. Depending on stem types, endings may show variants (primarily due to rules of vowel harmony) but variation is automatic and morphologically irrelevant. As usual, personal pronouns may show some irregularities. Otherwise, case suffixes remain unaltered and apply to arbitrary nouns both in the singular and the plural. Thus, Turkish case suffixes conform to the expectations raised by a classical morphemic model: here inflection realizes the ideal of a biunique relation between form and function favored by so many a linguistic theory. Flexive (or 'fusional') systems do not comply with this ideal as shown by paradigms of the Latin standard declensions (see Table 2 based on Risch 1977: the macron indicates vowel length). Again, case marking is realized by adding endings to stems. However, division of stems and endings is not trivial and endings are bound to numbers. Different from Turkish, the relation between form and function is non-unique in both directions. For instance, there are five distinct endings available for the genitive singular. What is more, these endings cannot be regarded as mere variants of a common basic pattern on account of their manifest formal dissimilarity. At the same time, one and the same ending may occur in apparently unrelated paradigmatic positions. Consider the ending -I It appears in the gen.sg., the abl.sg., the dat.sg., and the nom.pl. Obviously. the classical concept of morpheme where morphemes are conceived of as

Optimal specifications Table 2.

103

Latin declensions

m./f.

n.

singular

nom.

voc.

ace.

abl.

dat.

gen.

nom./voc./acc.

a-decl. o-decl. r'-decl. C-decl. w-decl. e-decl.

capra lupus ignis rex ictus dies

capra lupe ignis rex ictus dies

capram lupum ignem rigem ictum diem

capra lupo igm rige ictii dii

caprae lupo igni rig! ictuT diei

caprae lupi ignis rigis ictiis diei

iugum mare nomen genu

plural

nomVvoc

ace.

abL/dat.

gen.

nom./voc./acc.

a-decl. o-decl. i-decl. C-decl. w-decl. e-decl.

caprae lupi ignis regis ictus dies

caprds lupos ignis regis ictiis dies

capris lupis ignibus rigibus ictibus diibus

caprdrum luporum ignium rigum ictuum dierum

iuga maria nomina genua

CAPRA ('goat'), LUPUS ('wolf), IGNIS ('fire'), REX ('king'), ICTUS ('blow'), DIES ('day'); IUGUM ('yoke'), MARE ('sea'), NOMEN ('name'), GENU ('knee')

"roots and affixes winch serve as Saussurean signs" (Spencer 2006: 105) is not suited well to such a system. Wilkin the traditional word-and-paradigm model, there are two moves to be made in reaction. First, the inventory of noun lexemes is divided into classes of items which fit into a common pattern of building inflectional forms, called declensions (a-declension, o-declension, etc.). For each declension a separate set of case endings is established. (Neuter nouns deviate from the general sets of endings in the nominative, vocative and accusative. The remaining forms of neuters follow the pattern of masculines. They have not been listed in Table 2 for this reason.) The various sets of endings differ not only with respect to the make-up of forms, and different paradigms diverge not only by employing distinct sets of endings. Rather they also exhibit different patterns of syncretism, available endings being distributed differently over the range of relevant syntactic functions. For example, lexemes of the o-declension such as LUPUS show distinct forms in the nominative and the vocative, lupus and lupe, respectively. In the remaining declensions (and in the plural) this distinction is absent. In the plural. there are no distinct forms for ablative and dative. With neuters, nominative: vocative and accusative always coincide in the singular and in the plural as well.

104 BerndWiese From a syntactic point of view, this kind of variation between paradigmatic patterns may appear unfavorable. If maximally simple rules of agreement and government are desired, paradigms would be welcome that provide matching sets of case forms, hence, identical arrays of cells. For this to be achieved, all case distinctions that are formally drawn in some (sub-) paradigm are posited for all of the (sub-) paradigms (Lyons 1968: 293). This is the second move, crucial for the traditional model. Thus, as a matter of principle, formal differences in the structure of paradigms are made to disappear. While this contributes to the strength of the word-and-paradigm approach, it also constitutes a severe limitation of the model. Differences between patterns of syncretism (or patterns of differentiation) might well be arbitrary from a synchronic point of view. Often enough, however, there are systematic aspects that should not be neglected in a proper treatment of inflection. For instance, the luxury of allowing for specialized vocative forms seems to be particularly appropriate for the o-declension, that is, the declension that designations of (male) persons are preferably put into. The fact that there are less case distinctions in the plural than m the singular obeys a widely observed pattern, too. Turning to Polish, I shall consider first a representative set of singular paradigms of non-feminines, that is, of masculine and neuter nouns. The endings involved are displayed in Table 3. Notation of endings is orthographic using main variants in case of variation. The inventory of inflectional forms is assumed as given in Orzechowska (1999) and Swan (2002). These grammars also provide details and additional minor paradigms that are beyond the present discussion. For analyses of syncretisms in Polish noun inflection see Laskowski (1989), Menzel (2000), and Gunkel (2003). Noun stems may be 'hard' or 'soft'. Hard stems show off-sets in consonants from the basic (non-palatalized, non-affncated) sets of labials, dentals and Table 3. Polish noun endings in the singular: masculines and neuters nom. 11 voc. 11 ace.

loc.

dat.

| gen. |

ins.

STUDENI ,m. CUKIER, m. BIOLOG, m. BANK, m. BIURO, n. TANGO, n. POLE, a.

STUDENT ('student'), CUKIER ('sugar'), BIOLOG ('biologist'), BANK ('bank'), BIURO ('office'), TANGO ('tango'), POLE ('field')

Optimal specifications

105

velars or m M or /w/ (orthographically, m

,, . or ). Otherwxse stems are soft.* Remarkably, for some inflectional endings application depends on a stem alternation known as softening, i.e., a change from a hard stem alternant to a soft one. As usual this is indicated by putting a prime-sign (figuring as a 'soft sign') before the name of the ending (as in: '-e); cf studencie (< student + '-e), voc./loc.sg. of STUDENT. (For details see reference grammars; cf. also Cameron-Faulkner and Carstairs-McCarthy 2000. Stem alternations, which may be due to consonant or vowel shifts, will not be discussed in this paper.) To start with, it may be observed that the structure of the nominal inflectional system of Polish taken as a whole resembles the Latin one, as might be expected, but there are seven cases, not six. Again, casenumber-marking is cumulative and realized (primarily) by endings. Some familiar patterns of formal (non-)differentiation also reappear. For instance. as in related languages, neuters do not distinguish nominative, vocative and accusative. It may be noted here that, in the singular, masculines of the type POETA (base form in -a) are inflected like the corresponding feminines. Considering the form-function-relationship, we find non-unique relations in both directions. The ending -u provides the most noteworthy example. Within the seven exemplary paradigms shown in Table 3 this ending does not appear in the first paradigm. It appears in the genitive in the second paradigm; in the vocative and locative in the third paradigm; in the vocative, locative and genitive in the fourth paradigm; in the dative in the fifth paradigm; in the locative and dative in the sixth paradigm and in the seventh paradigm. Again, this distribution, arbitrary as it seems, does not fit well into a morphemic model. No particular problems seem to arise in a word-and-paradigm model on the Latin pattern: what one would need to do is establish seven declensions that differ with respect to the sets of endings employed. However, it should be observed that just about all endings appear in more than one declension. As a limit, in a particular case (here: the instrumental) all non-feminine nouns may share a common ending that is specific to this case. The same observation applies to the 2. Consonants that figure as off-sets of hard and soft stems may be ^^functionally hard and soft, respectively. Functionally hard consonants, e.g., /s/ and /z/, are also phonetically hard. Phonetically soft consonants (i.e., palatalized and palatal consonants) are also functionally soft, e.g., Isl and /*/. This group includes (in orthographical notation): , , and . However, the phonetic and the functional distinction are not co-extensive as there are also phonetically hard consonants that are functionally soft, e.g., /J7 and / 3 /. This group includes (in orthographical notation): . (The treatment of /U (), which is functionally soft, differs in the literature.) In the text of this paper 'soft' and 'hard' refer exclusively to functional notions.

106 BerndWiese Table 4. Polish noun endings in the singular: femmmes | nom. 11 voc. 11 ace. LAMPA, f. ZffiMIA, f. GOSPODYNI, NOC, f.

loc.

dat.

| gen.

ins.

f.

LAMPA ('lamp'), ZffiMIA ('land, earth'), GOSPODYNI ('landlady, hostess, housewife'), NOC ('night') feminine, winch possesses its own invariant ending for the instrumental. Table 4 displays four major feminine paradigms. For the most part, feminine endings differ from non-feminine ones. The four exemplary feminine singular declensions, however, diverge only occasionally Here, in distinct paradigms, endings reappear with partly overlapping distributions. Note that the ending -/ is rendered orthographically as or according to allophonic variation, cf. e.g., the forms nocy (< noc + -/) of NOC or lampy (< lamp + -/) of LAMPA vsi gO^O^/ofGOSPODYM.3

Table 5 presents a representative set of plural paradigms. As in the nonfeminine singular and in the feminine singular, in the plural, too, there is an invariant instrumental ending that is used in all regular paradigms (-ami, with an exceptional variant -mi). The same holds for the locative and the dative. Neglecting case-number-cumulation, these endings approach the Turkish pattern to a considerable degree: here, for one function (i.e., case-number combination) there is one and only one ending, which in its turn is restricted to just this function. The existence of competing case endings allows for a multiplicity of paradigms. This is in particular so if alternatives multiply each other's effects. In plural paradigms, there are instances of each and every combination of the three standard nominative/vocative/accusative formations (in -,, -e, and -a) and of the three genitive formations (in -6w, in -/, endingless). Thus nine types can be distinguished in addition to those that exhibit the special masculine endings 4 and -owie in the nominative (and vocative) plural. Moreover, the overall number of declensions is further increased if combinations of singular and 3. represents [i], which appears after phonetically hard consonants such as /S/ (), while represents [i], which appears elsewhere. Note that Ikl and /g/ 'automatically' soften before -i (Swan 2002: 15) and require, therefore, [i] (orthographically as in corki, gen.sg. of CORKA, 'daughter'). The same applies to /l/. Stem-final Ikl and /g/ also soften before -em (hence biologiem, bankiem, tangiem. mS.Sg. Of BIOLOG/BANK/TANGO).

Optimal specifications Table 5.

107

Polish noun endings in the plural nom./voc.

ace.

loc.

dat.

gen.

ins.

GENERAL, m. STUDENT, m.

owie '-i

-ow -ow

-ach -ach

-om -om

-ow -ow

-ami -ami

CUKIER, m. MYSZ, f. LAMPA, f.

-i -i -i

-i -i -i

-ach -ach -ach

-om -om -om

-ow -i -

-ami -ami -ami

PLAC, m. NOC, f. ZIEMIA, f.

-e -e -e

-e -e -e

-ach -ach -ach

-om -om -om

-ow -i -

-ami -ami -ami

MUZEUM, n. POPOLUDNIE, n. BR7RO, n.

-a -a -a

-a -a -a

-ach -ach -ach

-om -om -om

-ow -i -

-ami -ami -ami

GENERAL ('general, mil.'), STUDENT ('student'), CUKIER ('sugar'), MYSZ ('mouse'), LAMP ('lamp'), PLAC ('[town] square'), NOC ('night'), ZIEMIA ('land, earth'), MUZEUM ('museum'), POPOLUDNIE ('afternoon'), BIURO ('office')

plural paradigms are considered, these being not in biunique correspondence. All in all, Orzechowska (1999) assumes about 50 declensions. Throughout noun paradigms, there is massive overlap. There are considerable identities in the make-up of inflectional forms between different paradigms, and the relevant endings may be even identical from a functional point of view.* Thus it is reconfirmed that, in this system, most individual endings are not tied to a particular declension or paradigm. Consequently, there are no 'sets of endings' competing en bloc with other sets of endings (as in Latin). To a degree, the very notion of declension is, then, undermined in such a system. Instead of declension-specific sets of endings there are three subinventones that supply paradigms with endings, viz. (i) singular endings of non-feminines, (ii) singular endings of feminines, and (iii) plural endings. Paradigms are mainly distinguished by differences of choice between the items they select from these three subinventones. In sum, what we encounter in Polish noun inflection is neither a pure agglutinative system following the Turkish model nor a system of declensions according to the Latin pattern. 4.

Similar observations can be made wrth respect to other Slavonic languages. See. e.g., Miller (2004) on Russian, who emphasizes strongly the necessity to take care of both intra- and mterparadigmatic identities of form. Likewise, Baerman, Brown. and Corbett (2005: Sec. 5.4, also on Russian) point out the inadequacy of traditional accounts that treat paradigms as 'monolithic units'.

108

BerndWiese

Polish noun inflection holds a middle position between morpheme-centered and paradigm-centered morphology.

3. Differential case marking and case syncretism As in Latin, distinctions of case are sometimes marked on case forms in Polish. sometimes they are not. The distinction of nominative and accusative is a case in point. In languages that have it, this distinction is, as a rule, not always made formally explicit. In particular, direct objects may carry a pertinent morphological marking only under restricted conditions, a phenomenon termed differential object marking in Bossong (1985). Even Turkish, in spite of its nearly perfect biunique relation between case forms and case functions, exhibits such an asymmetry. Direct objects take the accusative if a specific or definite reading is intended, as in evi ('the house') or bir eVJ ('a house [specific]'). Otherwise the nominative (or, more appropriately put, the unmarked base form) usually takes over as in bir ev ('a house [non-specific]'), see Kornfilt (1997: Ch. 2.1.1). Similar procedures are adhered to in many languages. Markers appear when direct objects are to be distinguished that exhibit properties typically indicative of subjecthood otherwise (Lyons 1968: 294). This is true in particular if reference is made to animate beings (humans, in particular) or if the intended reading is specific or definite. Apparently, with other nominals used as direct objects, formal markings are more readily dispensable: if such nominals fill the object role, this agrees with expectations. In Latin, differential object marking separates genders. Neuter nouns never distinguish nominative and accusative, and, of course, neuters usually denote inanimates. As traditional treatments explain, a nominative-accusative distinction was not established in a class for which non-admittance of animate members was definitional. It is true that the nominative-accusative distinction, if marked, is not immune to fall victim to phonological erosion. However, it has been observed that in such cases various compensating strategies of repair may take effect if the need arises. In Latin (Table 2), it is exactly in the (masculine) o-declension and, subsequently following suit, also in the a-declension that a coincidence of nominative and accusative plural (expected by sound laws) has been avoided. The conspicuously deviant pattern of nominative formation (lupT, caprae) has been taken over from the pronominal declension (Brugmann 1904: 390, § 479). When formal markers are introduced or reintroduced that help tell direct objects from subjects, a path frequently taken is the adoption of morphological markers that are already in use for marking of objects. After the break-up of the

Optimal specifications

109

Latin case system, grammaticalized prepositions came into use as markers for syntactic relations. In Spanish, and similarly in other Romance languages. article-noun-groups (but not personal pronouns) normally remain unmarked as direct objects whereas the preposition a serves as a marker for indirect objects (as in Day el libra a Juan, '(I) give the book to Juan'). However, in Spanish, this type of marking, which is primarily of a 'dative' nature, is also used to flag direct objects in case these serve for specific or definite reference to humans (as in Veo a Juan, '(I) see Juan', vs. Veo el libra, '(I) see the book'). Thus, the formal differentiation of subjects and direct objects has been partly restored (for details see historical grammars, e.g., Hanssen 1910: 227, and, from a more theoretical point of view, Meillet 1921). In Slavonic, the nominative-accusative distinction had been lost in various paradigms, including the singular of the major declension type of masculine nouns. But this distinction has been renewed though not unvaryingly, rather, as might be expected, to different degrees (Thomson 1909/12; for details see Laskowski 1986). First and foremost, the new accusative formations apply to humans, then also to animate beings in general. The path taken to reestablish the distinction resembles the one adopted in Spanish: the missing formal marking of the accusative was taken over from another objective case, viz., in Slavonic, the genitive. Polish inanimate masculine nouns do not possess special forms for the accusative. As in the nominative, the base form applies; cf. dam, nom.-acc.sg. of DOM ('house') in Mam ladny dam, '(I) have (a) nice house': However, with designations of animates (except a-base nouns) it is the form otherwise used as agenitive that applies in the accusative, cf. sionia, accgen.sg. of SLON ('elephant') in Widze duzego sionia, '(I) see (a) big elephant'. This relation of substitution (which may be accounted for by a 'rule of referral') is encountered in the accusative masculine in general, i.e., it is not restricted to nouns. Agreeing items such as adjectives are subject to this rule as well. When used in construction with animate masculines (in the singular), they change to the genitive form wherever an accusative is required, as the examples show; cf. ladny, nom.-acc. of LADNY ('nice') vs. duzego, acc.-gen. of DUZY ('big'). With regard to this rule of agreement, a subclassification of the masculine gender into so-called subgenders, viz. inanimate masculines ('m.inan.') and animate masculines ('m.anim.'), has to be acknowledged in Polish (Meillet 1921: 208, "sous-genre"). These subgenders are true grammatical categories. Their extension does not coincide exactly with the corresponding semantic classes that lend them their names. What is, in semantic terms, inanimate may well fall into the class of animate masculine nouns grammatically. It should be stressed that the Polish (and Slavonic) rule of acc-gen.-referral does not present us with a case of a locally restricted adoption of markers

110

BerndWiese

Table 6.

Polish system of genders nouns l

I

non-f. i

I

i

m. i

I

m.anim. L

singular plural

I

f. n.

I

m.inan. i

m.pers.

m.impers.

+ +

+

acc.-gen.-referral: + (applicable),

(non-applicable)

(as exemplified by the special nommative plural formations in Latin referred to above). Rather, it applies to pronouns, numerals, adjectives and nouns in both singular and plural, which may exhibit quite different inflectional material. Whatever the relevant genitive form may look like and however it may be formed, it is taken over into the accusative if the conditions for applying the rule are fulfilled. With personal pronouns it even extends to inanimates. The overall gender system of Polish may be set up as in Table 6; cf Trubetzkoy (1934: 8) and Jakobson (1960) on Russian. In this system, a primary classification takes care of the distinction between non-feminine nouns and feminine nouns. The former class subdivides into two, masculine and neuter. Among masculines, animates and inanimates have to be distinguished. Finally, for Polish, a further subdivision is needed that distinguishes two subclasses of animate masculines: personal, viz. designations of male persons, and impersonal. This further subclassification is needed since it is only in the singular that the rule of acc,gen,referral holds for all of the animate masculines. By contrast, in the plural it is restricted to personal masculines. As remarked by Laskowski (1989: 220), the homonymy of accusative and genitive forms, which is brought about due to differential object marking, is, as it were, "communicatively costless", the two readings being readily distinguished in context. Actually, noun forms of the two cases may indeed stand in opposition when used as a non-partitive or a partitive direct object of a verb like KUPIC ('buy') as in kupii cukier vs. kupii cukru ('to buy sugar/some sugar'). Of course, only inanimate nouns are likely to occur in the singular partitive construction. But then, these nouns do not exhibit the acc-gen.-homonymy as they are outside the scope of the acc.-gen.-rule.

Optimal specifications

111

Another rule of referral is required to deal with vocative forms. In general. vocatives and nominatives may coincide (as, e.g., in most Latin declensions). especially in the plural of all genders and in the singular of neuters. In Polish, a distinction is called for only in the singular of masculines and of feminines. However, as it happens, part of the feminine nouns and most of the masculine nouns lack special endings for vocative marking. Again, as in the case of marked accusative forms that are missing, forms of another case stand in, here: locative forms.5 Locative forms that substitute for vocatives in their turn may show various suffixes, viz. '-e, -u (for masculines) and -/ (for feminines). This fact points to a systematic relation that is not bound to individual endings (Johnston 1997: 62) but may be accounted for by a rule of referral, such rules being blind, as it were, with respect to the make-up of substitutes (as is the rule ofacc,gen,referral).

4. Functions of Polish case endings 4.1. A two-level case system By the above considerations, the coincidence of nominative, vocative and accusative forms of the neuter in Polish (and related languages) would not appear to be due to 'arbitrary' homonymy Assume, then, that what we are dealing with in such cases are in fact unitary forms, not sets of homonyms. If so, what should a proper treatment of such forms look like? From a morphological point of view, case systems may be regarded as systems of classifications of forms of words. Polish has seven cases, thus, seven classes of case forms. On a most simple (and traditional) approach these classes would be given by a single classification on the basic set, i.e., by a 'flat' classification system. However, Trubetzkoy (1934), looking at Russian, set up a hierarchical system that starts from a primary division into two superordinate categories that he termed direct and oblique. Within these superordinate 5. Reference grammars make use of a mechanism of referral when slots in paradigms are filled not by forms but by pointers such as '= loc.'. The possibility of a rule of voc.-loc.-referral for Polish is also considered in Cameron-Faulkner and CarstairsMcCarthy (2000: 825 n. 10). (Apart from «-base nouns like POETA, masculine vocative forms that differ from locative forms are restricted to a small group of nouns ending in ec as, e.g., OJCIEC, 'father', cf ojcu, loc.sg., vs. ojcze, voc.sg., with an anomalous consonant shift c . cz. In addition there are isolated cases such as PAN, olpanu, loc.sg., vs. panie, voc.sg.)

112 Table?.

BerndWiese A two-level case system

caseforms direct nom.

voc.

oblique ace.

loc.

dat.

gen.

ins.

categories, traditional cases are identified as subcategories. Trubetzkoy's proposal may be adapted for Polish as shown in Table 7. If we assume such a hierarchical case system, the n o m . - v o c a c c f o r m s of neuters referred to above may be classified plainly as direct case forms, and they may thus be given a non-ambiguous characterization (cf also Williams 1994). Trubetzkoy's primary division, later on approved by Jakobson (1958). certainly plays a major role in a multitude of languages as may be gathered from the literature, and it fits into a general typology of syncretisms that has been elaborated in Baerman, Brown and Corbett (2005) and related publications on the basis of a representative sample of languages. As for Slavonic grammars, it is well established. 4.2. Singular endings of non-feminines While the quest for a hierarchical case system receives support from syncretism of direct cases, the gain is even higher if we turn to the oblique domain. Consider once more the endings of oblique cases in the singular of non-feminines. In Polish, there are five of them. Each of these endings is associated with a particular case except for the ending -u, which exhibits a seemingly arbitrary distribution. The ending '-e appears in the locative, -owi in the dative, -a in the genitive, and -em is an instrumental ending. It may be said that these endings specialize in a single case each. The ending -u appears in various oblique cases. Consequently, -u may be characterized provisionally as an unspecific (or 'plain') oblique ending. This can be done as the superordinate category oblique has been made available. Given this characterization, it is to be expected that -u appears whenever application of any more specialized ending is prevented for one reason or another.* Illustrative examples (to be discussed below) are 6. I assume a suitable version of the principle of specificity (or 'Elsewhere-prmcrple'. Krparsky 1973); see Wrese (2004: 331, wrth references). The utilization of superordmate (or 'aichi-') categories may be regarded as a more restricted analogue of the use of 'incomplete' specification or ' ^ s p e c i f i c a t i o n ' (cf. Halle and Marantz 1993).

Optimal specifications Table 8.

113

Polish noun endings: the non-feminine singular inventory caseforms

direct base

nom.

student m.anim., hard cukier m.inan., hard m.anim., hard, vel. biolog m.inan, hard, vel. bank biuro n., hard n., hard, vel. tango n., soft pole m.inan., hard, -a nos m.inan., soft, -a grosz m.anim., hard, *-owi hot minan., hard, -a, *-owi swiat

ace. voc. = loc. = gen. manim. m. '-e '-e -u -u

'-e -u '-e '-e

-a -a

-a

loc. '-e

dat. -owi m.

gen. -a *mjnan.

ins. -em

'-e '-e -u -u '-e -u -u '-e -u '-e '-e

-owi -owi -owi -owi -u -u -u -owi -owi -u -u

-a -u -a -u -a -a -a -a -a -a -a

-em -em -em -em -em -em -em -em -em -em -em

Conditions of application: m. (applicable to masculines only); m.anim. (applicable to animate masculines only); *m.inan. (not applicable to inanimate masculines) provided in Table 8. The table repeats the two-level case system introduced above. Names of case endings have been written into 'case boxes' where appropriate. In this way it is shown how endings are assigned their proper case specifications. The table also indicates pertinent conditions of application for endings. Restrictions on the use of case markers are form-based or functionbased. i. Form-based restrictions relate to the interplay of endings and stem alternations, for which the division of nouns by stem type into hard-stem nouns and soft-stem nouns is basic. As indicated by the prime-sign, the locative ending '-e always implies a stem alternation known as softening. Now, only a subset of noun stems allow for such an alternation. Soft-stem nouns in particular rule out softening. In the locative, non-feminine stems with velar offset, too, do not allow softening. Thus, applicability of this ending is heavily restricted. If '-e is inapplicable, -«takes over.

114

BerndWiese

11. Function-based restrictions basically relate to the role of the animacy hierarchy (or, more generally, the hierarchy of individuation) in case-marking. Among the oblique cases, dative and genitive are affected, these cases being used more frequently in reference to animates than are locative and instrumental Dative is a case that is used preferentially in reference to animates, humans in particular. Hence, it may be expected that formally distinguished dative forms are primarily needed for lexemes that denote animates whereas on forms of lexemes not so used dative marking may be less well developed or even absent. Given the type of gender system found in Polish, the class of nouns concerned is the neuter, i.e., the inanimate gender par excellence. As a matter of fact, neuters do not employ the special dative ending -owt that occurs on masculines.* To account for this observation I posit a condition of application that restricts the dative ending -owi to masculines (as indicated by the subscript 'm.' in Table 8). Accordingly, neuters show -u in the dative. In the genitive, too, nouns may or may not accept the pertinent specialized case ending (here: -a). In the latter case, the less specific ending -u will stand in. Once more, the split is related to the animacy hierarchy. Practically all nouns of the animate masculine subgender, i.e., nouns that employ ace,gen,referral. take -a in the genitive singular. Inanimate masculines for the most part do not form genitives in -a (for details see Orzechowska 1999: 306f.).9 Thus two patterns stand out: (i) nouns like STUDENT add -a in the genitive (and do not distinguish genitive and accusative forms), (n) nouns like CUKIER do not add -a (and distinguish genitive forms in -u from accusative forms, which are left without ending). As mentioned above, a formal distinction between accusative and genitive would be welcome for lexemes that may be expected to appear in both standard direct object constructions and partitive direct object constructions, and in fact, mass nouns form the hard core of this subclass. 7. Laskowski (1989: 212f), reporting on a (small) corpus of spoken Polish texts, provides the following frequencies for the oblique cases (in percentages of all occurrences of case forms in the corpus of animate and inanimate nommals, respectively). Animates: dat, 12.4, gen, 8.2, ins.: 3.8, loo.: 0.4. Inanimates: gen, 26.3, loo, 12.2. ms. 4.8, dat, 0.0(!). 8. As for cognate endings in Czech, Slovak and Ukrainian, the animacy hierarchy is also unquestionably identifiable as a major factor controlling their distribution (and diachromc spread), see Janda (1996: 170f). 9. A split in genitive marking as found in Polish is found also in related languages (Janda 1996: 145). Russian even developed a division of two separate cases, genitive I - a general genitive - and genitive II - a partitive (Trubetzkoy 1934: 10).

Optimal specifications

115

A minority class of inanimate masculines, mainly count nouns, may compromise both patterns. This includes various designations of instruments and implements, among others. In general, these nouns do not undergo acc.-gen-referral while at the same time their being used partitively (in the singular) is less likely. They show endingless accusatives along with genitives in -a. In a somewhat simplified analysis, such nouns may be handled as lexical exceptions. This said, the distribution of genitive formations will be accounted for by a condition of application that prevents the ending -a from being used with inanimate masculines (as indicated by the subscript '*m.inan.' in Table 8). Neuters invariably take -a in the genitive. Avoidance of genitive-dative syncretism has been adduced in order to explain this distribution (Schenker 1964: 50). In general, non-feminines may lack either the special dative ending (-owt) or the special genitive ending (-a), but not both. (This follows as the domains blocked by the associated conditions of application are disjunct.) As envisaged, the above analysis of non-feminine endings puts into relief systematic asymmetries in case marking. The neuter subsystem is patently less differentiated than the masculine one, which adds vocative and accusative marking (controlled by animacy) as well as a special dative marker and an animacy split in the genitive. Only animate masculines may tap the full potential of the system. Formal and functional restrictions take part in controlling the distribution of endings. Together with case specifications, they account for the variance between paradigms as may be gathered from an inspection of the sample nouns listed in Table 8. These exemplify eleven different distributions of endings. To begin with, consider the endings of oblique cases: -

('student') is an animate masculine from the class of hard-stem nouns (in the table: 'm.anim., hard'). With nouns of this type, we encounter maximally developed paradigms. As a matter of fact, all of the four specialized oblique endings are present. Thus, the ending -u does not get a chance to apply.

-

CUKIER ('sugar'), being inanimate, does not accept the genitive ending -a. As predicted the plain oblique ending -u stands in.

-

('biologist') is a noun that cannot undergo softening before '-e as its stem ends in a velar (indicated by 'vel.' in the table). Hence, the ending cannot apply and again, the plain oblique ending -u stands in.

-

BANK ('bank') does not accept either the locative ending '-e or the genitive ending -a according to the conditions discussed. Hence, in both positions it is -u that appears.

STUDENT

BIOLOG

116

BerndWiese

- BIURO ('office'), winch is neuter, exhibits the plain oblique ending -u in the dative since neuters do not accept the special dative marker -owt, which is restricted to masculines. -

('tango') and POLE ('field') are neuter and exhibit a hard velar stem and a soft stem, respectively. Aside from rejecting -owi (as do all neuters). they cannot add the ending '-e either (on account of their respective stem class membership). Thus, two of the oblique cases show the plain oblique ending -u. TANGO

- A few outliers deserve special mention. As noted above, the condition of application associated with the genitive ending -a, viz. '*m.inan.', may be violated with nouns from special groups, including various designations of body parts such as NOS ('nose'). Against the rule, such nouns may accept the ending -a. For this reason, they have to be treated as lexically marked (this is indicated in Table 8 by adding '-a' to the specification of the noun class). Significantly, there are considerable fluctuations to be found in this domain. The noun GROSZ ('penny') is another example of an inanimate masculine noun that does accept -a. Unlike NOS, GROSZ IS a soft-stem noun. Thus. it does not take the locative ending '-e. - The condition of application associated with the dative ending -owi, viz. 'm.', is rarely violated. Only a few nouns that should have it drop this ending and, for this reason, have to be considered as lexically marked; cfi, e.g., KOT. m. ('cat'). This is indicated in Table 8 by adding '* W to the specification of the noun class. Once more, the ending -u stands in. - The last example in Table 8, SWLAT ('world'), presents an isolated case. namely an inanimate masculine that does not accept the dative ending -owt but does add the genitive ending -a. It has to be treated as doubly marked in the lexicon. There are some more isolated cases and small groups including DOM ('house'), PAN ('mister, sir') and SYN ('son'). These three hard-stem nouns lack the locative ending '-e; again, -u stands in. I return to the direct cases. As a rule, base forms of masculines are endingless, base forms of hard-stem neuters show the ending -o, and base forms of soft-stem neuters show the ending -e. (For perspicuity, base-form endings are not represented in Table 8.) As discussed with reference to neuter nouns, base forms figure as unspecific (plain) direct forms. In fact, there are no regular nonfeminine endings that specialize in particular direct cases. Base forms are used throughout direct cases unless additional regularities intervene (as is the case with masculines only). Most importantly, this concerns the rule of referral for

Optimal specifications

111

'missing' marked accusative forms that are substituted by genitive forms, which has been described and motivated in Section 3, above. By another rule of referral, also discussed above, locatives may substitute for missing marked vocative forms. However, vocative marking is mostly optional and, if not applied, base forms stand in. As discussed in Wiese (2004, with reference to Russian), referrals may be taken care of by setting up equations between sets of forms, which are given in Table 8 (in the table: 'voc. = loc.' and 'ace. = gen.') together with their relevant conditions of application (noted as 'm.', i.e., applies to masculines, and as 'm.anim.', i.e., applies to animate masculines). Applying these 'rules of referral', we complete the derivation of case forms in the non-feminine singular. Remaining vacant positions (marked as ' - ' in Table 8) are filled by unaltered base forms. As may be read off the table, the distribution of endings is fully predictable given the specifications associated with the endings (including conditions of application), the classification of stems (including lexical markings where necessary, i.e., where needed to deal with exceptional cases), and the rules of referral (including their conditions of application).

4.3. A multi-level case system The adoption of hierarchical classification systems provides a natural way to deal with systematic syncretism, defined as "the suppression of a relevant opposition under certain determined conditions" (Kurylowicz 1964: 40). In fact. pursuing this idea further, it seems natural to adopt a multi-level system that replaces the two-level classification tentatively adopted above, as presented in Table 9. In the domain of direct cases, absence of a distinction of nominative and vocative forms, which may be more or less extensive, is a familiar phenomenon both in Polish and beyond (as noted above for Latin). In a multi-level case system, this syncretism may receive a straightforward account if a binary subclassification of the category direct is assumed, comprising non-accusative and accusative (where non-accusative is the union of the subcategories nominative and vocative the distinction of which may or may not be formally reflected in a paradigm). Trubetzkoy (1934: 8), in support of the direct-oblique-distinction, points out that, in Russian, the most simple, if anomalous, paradigms of words that have more than one inflectional form (viz., the numerals SOROK '40' and STO '100') possess just two forms, a direct one and an oblique one. Similarly, Polish numerals of the type exemplified by PDJ6 ('five') make a distinction between a

118

BerndWiese

Table 9.

A multi-level case system

caseforms direct non-acc. nom.

voc.

oblique ace.

non-ins. non-gen.

ins.

gen.

direct form prec and an oblique form precru, which is not further specialized and hence may occur in all of the four oblique cases. As a next step of differentiation within the oblique domain, a special instrumental form may be distinguished. Polish numerals do possess such forms (cf pifdama, instrumental of PIEC), their use being optional. (Essentially the same situation is found in Russian pronouns such as ETOT, 'this', cf. etoj, obl.sg.fem., vs. etoju, ins.sg.fem., obsolete or optional.) Of course, in as far as such special instrumental forms are used at all, appearance of the general oblique forms will be restricted to the remaining non-instrumental subdomain of oblique cases (viz. locative, dative and genitive), which, in these paradigms, undergoes no further subdivision. Consequently, in a multi-level case system, a binary subclassification of the category oblique may be assumed, comprising instrumental and non-instrumental (where non-instrumental is the union of locative, dative and genitive). Among Polish nouns, such an opposition of instrumental forms and forms covering all of the remaining oblique cases is encountered in soft-stem feminines, cf. ziemi, loc.-dat-gen.sg., vs. ziemiq. ins.sg. of ZIEMIA (as well as between feminine forms of adjectives, cf. bialej. loc.-dat.-gen.sg.fem., vs. biciq, ins.sg.fem. of BIALY, 'white'). Here, only the most marked oblique case in Polish (according to Laskowski 1989: 212) has a form of its own. The formal distinctions between the remaining oblique cases are 'suppressed'. As a further step towards a more elaborated partitioning of the oblique forms, hard-stem feminines show an additional distinction between forms that are restricted to the genitive (as, e.g., lampy of LAMPA) and forms covering both locative and dative (as in lampie). The latter may be characterized, in terms of the system presented in Table 9, as non-genitive forms (non-genitive being the union of locative and dative). A comparable pattern of distinctions is found with soft-stem neuters (and with velar hard-stem neuters as well). Oppositions

Optimal specifications

119

between genitive forms and less specific oblique forms may be found both inside and outside Slavonic. Compare, e.g., singular forms of Russian feminine nouns of the type GORA ('mountain'). Similarly, in Latin plural noun declensions, special genitive forms contrast with forms covering both ablative and dative (cf Table 2, above). Among Polish nouns, the final step of subdividing the oblique singular domain, viz. the differentiation of locative and dative, is taken only in (most) paradigms of hard-stem non-feminines. Masculines, in particular, may distinguish specialized dative forms (in -owt), as discussed in the previous section. The weakness of the locative-dative-distinction does not come as a surprise, of course, both from a system-internal and from a comparative point of view, the Indo-European dative being "nothing else than an offshoot of the loc. used with personal nouns", as Kurylowicz (1964: 190) put it. In Polish, the distribution of dative vs. locative is nearly complementary both semantically, i.e., with respect to (in-)animacy (cf. n. 7, above), and syntactically (the locative being a pure prepositional case while prepositions governing the dative are rare). Further subdivisions of the oblique domain are not present in Polish but are found in Russian (where locative and genitive each split into two subcategories). Overall, the degree of formal differentiation within different paradigms corresponds to relations of markedness between word classes (genders, in particular). I" Using a hierarchical system of classifications, categonal specifications of case endings may be optimized. As discussed above, a form such as pole (of POLE, n.), which shows the neuter base-form ending -e, may appear in the nominative, the vocative and the accusative. If higher-level case categories are available, no reference to single cases need be made. Thus, a categonal specification may be given that is not more specific than required. On the other hand. it may be expected that an adequate system of classification should provide for specifications that are, at the same time, not less specific than warranted by available evidence. As base-form endings are categorized as markers of the category direct, both requirements are accounted for. Such a specification is optimal as it is neither more nor less inclusive than can be justified by the data. A multi-level case system allows for further optimizations of case specifications. The ending -u has been treated above as an unspecific or plain oblique ending of the non-feminine singular. But, as noted, instrumental noun 10. Adjectives and other declmables show additional types of syncretism discussion of which is beyond the scope of this paper, but cf. Wrese (2004) for an integrated approach to nominal, pronominal, and adjectival inflection in Russian that may be applied also to Polish.

120

BemdWiese

Table 10. Polish noun endings: the non-feminine and feminine singular inventories caseforms oblique

non-ins. non-f./f. -ul-i nom.

voc. f. [-a] -o

gen. non-f. -a *m.inan. loc. non-f. '-e

f., hard, [-a] lampa f., soft, [-a] ziemia f., soft, [-/] gospodyni noc f-,soft,[-#l

-o -o loc. loc.

ins. non-f./f. -em/-q

dat. non-f. -owi m. -e -/ -i -i

-1 -Q.

-1

forms in Polish always show special instrumental endings (in all genders and numbers). Given the multi-level case system presented in Table 9, we may optimize the case specification for the noun ending -u, which never appears in the instrumental, by changing it to non-instrumental (given that non-ins. = loc. u dat. u gen.).

4.4. Singular endings of feminines Table 10 repeats the multi-level case system introduced in the preceding section. Again, names of case endings have been put into the 'case boxes' where appropriate. Both, non-feminine and feminine endings are given in order to facilitate a comparison of the two inventories. Membership in either set is indicated above the endings' names as ' n o n - f and T.\ respectively. Case specifications and conditions of application for non-feminine endings are unchanged excepting only -u, as justified above. Notational conventions are as explained

Optimal specifications

121

for Table 8. Feminine stems are classified by reference to the base-form endings they take. These classes are indicated by '[-a]', '[-,]', and '[-#]' (endingless) in the table. Compared to the non-feminine, the feminine singular shows less variation between paradigms and also a reduced degree of differentiation between oblique cases (see the standard example paradigms repeated in Table 10 for convenience). This is in accordance with the status of the feminine as a marked gender. Among feminines, too, there are those that distinguish nominative and accusative forms as well as others that do not (cf the paradigms of LAMPA, ZIEMIA, and GOSPODYNI vs. the paradigm of NOC). The latter type, which is clearly the minority type, comprises soft-stem nouns that exhibit endingless base forms. It is true, the division between those feminines that do syncretize nominative and accusative and those that do not is not drawn according to animacy in the way observed for the masculine. But significantly, the syncretizing type of feminines is known to comprise nouns that denote inanimate objects along with numerous abstract nouns in -osc but only a few animates (Damerau 1967: 38). Unlike masculines, feminines achieve nom.-accdifferentiation not by referral but by contrasting a special accusative ending (viz., -e) with a characteristic base-form ending, viz., -a for the predominating type and -/ for a smaller subclass. The fact that -e is used only with nouns that show a vocalic base-form ending is indicated in the table by the notation ' [ - K ] \ which precedes the name of the ending. In addition, feminines of the a-base type have at their command a special vocative ending -o, thus arriving at a fully differentiated set of direct case forms. In the table, the notation '[-a]' indicates that the ending -o is restricted to a-base nouns. (As mentioned, there are also masculine a-base nouns such as POETA, which follow the feminine pattern in the singular. but not in the plural.) Feminines that do not accept the vocative ending -o follow the by now familiar rule of referral and switch to the locative form, cf., e.g., nocy, voc.sg. of NOC. (AS it happens, for nouns of the type GOSPODYNI the target form is still homonymous with the nominative.) As for the oblique cases, there are three feminine endings (compared to five non-feminine ones). Once again, there is a rather unspecific ending that appears in three oblique cases (locative, dative, and genitive), viz. - , It may be addressed as the feminine counterpart of the non-feminine ending -u. Accordingly, it is also assigned the specification non-instrumental. Again, the instrumental has a typical 'nasal ending' of its own, viz. -q (where represents lol). Finally, in the feminine, too, the ending such that:/is a sequence of syntactic base forms (mostly, phonological words); s is a syntactic structure of this sequence; e is a lexical interpretation of/given s; and there is a sentence meaning u of/given s and e. An interpreted sentence is a quadruple < / s, e, u> such that u is a sentence meaning of/given s and e. - (Non-empty) sequences/are construed as sets of ordered pairs where / is a number 1, 2, . . . , „ ( „ > 1) and w is of the type of a phonological word. Abbreviations are used as m f= on, poslal2 materi, deneg4= {,, ,}. (Number subscripts that are part of variables, as in fh f2 etc., are unrelated to sequences.) Note that deneg4 = {}, and that on, = the unit sequence of on = {} is different from on. A syntactic structure s of a sequence/is itself a triple, , such that: * is a constituent structure of / m is a marking structure of/given *-, and / is an intonation structure of f. As an example, consider (8a): The first line orthographically specifies a sequence of phonological words, / = on,poslal2 materi, deneg4, and partly specifies its intonation structure / b y means of punctuation

142

Hans-HeinrichLieb

and Svetlana Friedrich

and the accent sign; the third line incompletely specifies the marking structure m by naming some relevant syntactic categories. For noun-form categories in Russian idiolect systems S we assume not only case, number, and gender categories but also categories of definiteness: Def(- S), Indeff- S), and Unsp D e f (- S), assigning all forms of substantives to Unsp Def ( - S), due to the lack of articles, where Unsp Def (- S) = the set of all Noun forms of S that are Unspecific for Definiteness. (These definiteness categories are syntactic not semantic.) There is no hint in (8a) of the constituent structure k, which we assume as follows (the phonological words do not belong to *-, the arrows represent grammatical relations, which are not part of k either): (9)

3'

^VGr comp3

\ ini Nf Vf 1 2 on postal

Nf ) A ~Nf j 3 * 4 materi deneg

i={,,,,}, N f = N f ( - 5")= Noun form in [Russian idiolect system] S; "Vf" for "Verb form"; "VGr" for "Verb Group". There are four primitive constituents, on,. poslal2, materia deneg,, where on, = {}, e t c , and the whole sequence / i s the only non-primitive constituent ( / i s a constituent of itself). Obviously Represents a surface-structure, non-binary analysis (only 'surface structures^ are allowed). A lexical interpretation e of /given s is a function that associates concepts b with the primitive constituents of/; for example, if e is the interpretation in (8) a n d / = {}, then we take the set {2} as an argument of e, and e({2})= send-; this way, the concept send- is associated w i t h / . The concept associated with a primitive constituent must be the meaning of a lexical word such that a form of the word is used by means of the constituent. The second lines in (8a) and (8b) name the concepts as assigned by the lexical interpretation, omitting the raised dots that are used in IL to form concept names. The arrows in (9) indicate that, given sentence (8a) as a triple < / s, e> in a Russian idiolect system S, with k as in (9), the following grammatical relations hold: 1, due to the fact that events etc. are represented separately but typically involve other entities that also ask for separate representation. Verb-meaning concepts can be classified by properties of the events etc. and of the entities involved in the events. In particular, we may have individual-friendly, quantity-friendly and groupfriendly concepts: These are «-place action concepts, n > 2, whose intension consists of an «-place relation-in-intension between actions xh agents x2, affected objects x3, and possibly other entities, such that x3 may be an individual/a quantity/a group. It is these types of verb meanings that will be of special importance.

B. Independent case competition: Genitive vs. accusative 4. The genitive effect and its location 4.1. The effect: Partitivity In Section 2, we started in (3) with the most basic question: What are the semantic effects of the genitive in case competition? It appears from our 4. Quantity concepts should not be construed as identity concepts, e.g., the intension of gold- should not consist of the property of being identical to the sum total of (all the) gold (in the world), the extension of gold- being the umt set of the sum total of (all the) gold (in the world); similarly, for money •. Such a proposal runs into trouble as soonas specific quantities are to be referred to, zsmEnglish the gold in Fort Knox. 5. For a recent axiomatic treatment of quantity mereology on the one hand and group ('plurality') mereology on the other, see Appendix B in Nolda (2007).

Case competition in Russian

147

overview of the literature that there has been growing consensus on the following answer when we restrict ourselves to negation-independent competition in present-day Russian: There is just one effect, and that is partitivity, i.e. expression of a part-whole relationship in the sentence meaning. We consider this as basically correct, and adopt the answer, speaking of partitivity as the semantic effect of the genitive in negation-independent case competition (of genitive vs. accusative) in present-day Russian. More precisely, we consider partitivity as the only 'positive' effect (an effect that does not consist in simply blocking another semantic effect). We next turn to the question of location (4), of where in the sentence meaning the effect is to be located, in the referential part of the meaning or in the proposition? The answer largely depends on the referential status of the genitive constituent: Is it - in cases of negation-independent competition with the accusative referential or non-referential? This question, recognized in the literature as an important problem in relation to Russian case competition generally, will be discussed in the present Section 4. Since the non-referentiality of the genitive constituent would contrast with the referentiality - usually assumed - of the accusative constituent, we begin by taking a closer look at the accusative.

4.2. Referentiality of the accusative constituent Consider, once again, sentence (8b), repeated as (12)

On postal materi den'gi. he send mother money PretPerfDat Ace P1N

This was a translation by Gladrow (1998: 55) of "Er schickte der Mutter das Geld." - "He sent his mother the money." The bare-noun occurrence den'gi, was taken to be a referential expression, with a definite referential meaning, 'the money'. The accusative constituent in (12) must indeed be referential but the referential meaning suggested by "das Geld" - "the money" is incorrect, given the sentence accent (not indicated by Gladrow) on den'gi, and a non-contrastive sentence meaning. If the speaker uses den'gi, in (12) with a definite referential meaning, then the hearer 'knows o f the referent of den'gh. However, (12), with sentence accent on den'gi, can be uttered appropriately only in a situation where the speaker does not know of the referent, the money sent, e.g., (12) may be uttered as a proper response to a preceding question by the hearer asking.

148

Hans-HeinrichLieb

andSvetlana

Friedrich

what did the o^-person send to the ^ e r / 3 - p e r s o n ? Only when the accent is moved away from den 'giA does a definite referential meaning become possible. as in On poslal mdteri den 'gi., which may be uttered as a proper response to! Whom did the o^-person send the (known) money to? a question that also allows for answers by, On poslal den'gi mdteri., and by Den'gi on poslal mdteri. In summary, the referential meaning oi den'gi A in sentence (12) must be indefinite not definite; a translation of den 'gi4, by "the money" is defensible only if it is meant to indicate not definiteness but specificity (as opposed to genencity), which is compatible with both definiteness and indefiniteness: In uttering (12): the speaker refers to a specific amount of money (specificity), but does not express a belief that the hearer knows of this money (definiteness). True, this is correct only in a situation where the amount of money was not previously introduced, either explicitly or implicitly, as a potential referent of den 'gi4 in (12): such introduction supersedes the role of sentence accentuation.*

4.3. The genitive constituent: Generic or non-referential? Now consider (8a), here repeated as: (13)

On poslal materi deneg. he send mother money Pret Perf Dat Gen P1N

Gladrow's German version (1998: 55) is, "Er schickte der Mutter Geld." - "He sent his mother money"; Gladrow appears to be claiming that deneg, has an indefinite meaning.? If deneg, is indeed referential, two possibilities can be considered for the referent of denegA: It is either the whole of the part-whole relationship, or the part. In Lieb (2007) and Friedrich (2009: Chapter 10) it is argued that the 6. The connection in Russian between intonation and the referential meanings of nommals is systematically studied in Friedrich (2009), whose relevant results are presupposed in the present essay. 7. Indefinite meanings for the genitive constituent in such sentences are more clearly assumed by Birkenmaier (1979), Leiss (2000); it is not entirely clear how Gladrow's Indeterminiertheit ("mdetermmateness") is related to indefiniteness. - In the following discussion we exclude from consideration so-called attributive referential meanings, as in cases like, "The murderer will be caught.", said by a policeman on the discovery of a slam person; such meanings occur as a subcase of definite and, possibly, indefinite meanings but are obviously irrelevant here.

Case competition in Russian

149

referent should be the whole but, for argument's sake, we will here allow both possibilities. On either reading of deneg4 in (13), definite referential meanings are excluded: There cannot be a felicitous utterance of (13) in a situation where the speaker wishes to express his belief that the hearer 'knows o f the money that was sent, or of which it is a part. This agrees with the relationship between accent placement and referentiality established in Fnednch (2009): deneg4 in (13) bears the sentence accent. and thus should permit only indefinite and generic referential meanings; as soon as the accent moves away, only definite or generic meanings should be possible. However, contrary to what is assumed in Lieb (2007) and Fnednch (2009: Chapter 10), indefinite meanings should be excluded for accented deneg4, and definite ones for unaccented deneg4: Given an indefinite meaning, the speaker who utters sentence (13) expresses his or her belief that some hearer does not know of the money that was sent, or the money of which it is a part; similarly that the hearer does have such knowledge if the meaning is definite. But either condition on speaker beliefs appears to be too strong; rather, no belief is expressed, be it in a negative or a positive direction. (Also, in either case there is a problem with specificity, if reference is to the whole: Sentence (13) with a specific meaning of deneg4, be it definite or indefinite, would then mean that part of a specific sum of money was sent, which appears to be wrong.) This leaves us with just two possibilities: Either the genitive constituent has a generic meaning, or it is non-referential. Distinguishing genencity from non-referentiality is notoriously difficult in many cases - this is one of them.

4.4. The genitive constituent: Can it be generic? Adopting the analyses presented in Lieb (2007) and Fnednch (2009: Chapter 10), there may be a (distributive) generic meaning of deneg4 in sentence (13). reference being to the whole not the part of the part-whole relationship. However, a more careful analysis of potential genenc meanings has led us to identify a serious problem, typical of sentences like (13). Characterizing the problem requires some technicality. The lexical meaning associated with deneg4 in (13) is a mass concept money. The extension of money- is the set of all quantities of money. Suppose, then, that deneg4 in (13) has a (distributive) genenc meaning (nondistnbutive meanings are hardly acceptable); that the part-whole relationship is accounted for in the proposition; and that reference by deneg4 is to the whole

150 Hans-HeinrichLieb andSvetlanaFriedrich not the part. The meaning of deneg4 may then be characterized by the following formulation: (14)

For any (entity) x, (utterance) V, and (speaker) V l : Vx refers by deneg4 in V to x if, and only if: a. x is in the extension of money. b. x is 'relevant' with respect to deneg4, V, V b and money•.»

The proposition is informally characterized as follows (cf Lieb 2007; Friedrich 2009: Chapter 10): (15)

If the speaker refers to x 2 by on, and to x3 by deneg4 and to x 4 by materh. then 'some part ofx 3 was sent by x 2 to x 4 \

We previously failed to notice that this analysis creates the following serious problem: Combining a proposition as in (15) with a referential meaning as in (14) results in a sentence meaning that implies: For every relevant quantity of money, some part of it was sent. But certainly, the sentence means that only some part of some relevant quantity was sent. There appear to be just two remedies that may be reasonably suggested. A. Assume that reference is not to the whole of the part-whole relationship but to the part, and change (15) by dropping "some part of". (14) is changed to read: "For any (entity) x, (utterance) V, and (speaker) V l : V, refers by deneg4 in V to x if, and only if, for some x': (a) x' is in the extension of money, (b) x' is 'relevant' with respect to deneg4, V, V b and money, (c) x is a part of x'." The part-whole relationship is now accounted for in the referential meaning of deneg4; (15) is changed to read: "If the speaker refers to x 2 by on, and to x3 by deneg4 and to x 4 by materi3, then x3 was sent by x 2 to x4". Combining the new versions of (14) and (15), we now obtain as a consequence that every x3 that is a part of some money quantity was sent, which is wrong again; i.e. proposal (A) must be rejected. B. Assume that reference is not to the whole of the part-whole relationship but to the part, and change (15) by dropping "and to x3 by Je«eg 4 " and substituting "some x3 to which the speaker refers by deneg4" for "some part of x3". 8. The notion of relevance in (b) is explicated in Integration^ Sentence Semantics through the notion of "reference basis", cf. Lieb (1983: 285).

Case competition in Russian

151

Generic meaning (14) is changed as before. No objectionable consequence as previously derived may be obtained by combining the new versions of (14) and (15). Solution (B) is, however, objectionable for a different reason. The proposition now is, informally: If the speaker refers to x 2 by on, and to x 4 by materh, then some x3 to which the speaker refers by deneg4 was sent by x 2 to x4. When sentence (13) is uttered, the proposition becomes the content of an assertion. However, it appears to be wrong to claim that the speaker asserts that something he or she is referring to by deneg4 was sent, rather than something that is a part of some quantity of money: The assertion is not on the existence or nonexistence of a referent for deneg4. Hence, proposal (B) should also be rejected. In summary, in order to save a generic meaning for the genitive constituent deneg4 in (13), we had to give up both assuming that the part-whole relationship is accounted for in the proposition and assuming that reference by deneg4 is to the whole not to the part; but even then no acceptable sentence meaning was obtained. We conclude that deneg4 does not have a generic meaning. Now generic meanings were the only candidates left for referential meanings. Therefore, deneg4 does not have a referential meaning in sentence (13), i.e., it is a non-referential expression in this sentence.' Obviously, this result generalizes to the genitive constituents in all comparable sentences.

4.5. The genitive effect: Proposition^ not referential The only plausible location for a genitive effect is in the referential part or in the proposition. We are now left with the proposition as a place for the single positive genitive effect. In summary: (16)

In sentences like (13), On postal materi deneg.. a. the genitive constituent is non-referential; b. semantically, the category Genitive is processed directly in the sitionofthe sentence meaning.

More generally, the following answer to the question of location (4) is implied. which is restricted to Russian but suggestive of wider application: (17)

In cases of negation-independent competition, a positive semantic effect of the genitive is to be accounted for in the proposition - not in the referential part - of the sentence meaning.

9. Cf. also "non-local-reference" in Durst-Andersen (1996) or "nereferentnoe upotrebleme" - "non-referential use" - in Paducheva (2005, 2006).

152

Hans-HeinrichLieb

andSvetlana

Friedrich

We can now see that there is also a negative effect: In negation-independent competition, the genitive (i.e. the category Genitive) together with other factors blocks the semantic functions (associated with the syntactic function of nucleus) that would bring about referential meanings of the genitive constituent. This will not be discussed any further. We next turn to the question of semantic mechanism (5) for the semantic effect that the genitive has in negation-independent case competition. As a preliminary step, we isolate the effect more precisely in two sample propositions.

5. The genitive effect in sample propositions 5.1. Informal version of a proposition The proposition for sentence (13), which is to include the genitive effect, can be informally characterized by modifying formulation (15): We must eliminate the referring by means of deneg4, and account for the lexical meaning moneyof deneg4 directly in the new formulation, as follows: (18)

If the speaker refers to x 2 by on, and to x4 by maten,, then 'some part x3 of some relevant quantity x5 in the extension of money was sent by x2 tox4\

For a more explicit formulation it is the part in inverted commas that has to be made more precise. We begin by making some general remarks. Formulation (18) contains the explicit requirement that some part of some relevant quantity of money was sent. Simply requiring that some relevant quantity of money was sent would be insufficient. For consider all money that the speaker considers relevant when uttering deneg4 in an utterance of sentence (13). This again is a quantity of money, hence, a relevant quantity of money. But if the speaker wanted to assert that this quantity was sent, he or she would have to use the accusative den'gi* i.e., would have to utter sentence (12). The all-comprehensive quantity must therefore be excluded in construing the proposition for the genitive sentence (13). Precisely this is achieved by the requirement in (18) of "some part (i.e., Q-part) x3 of some . . . x5": The Q-part relation is asymmetric, therefore x 3 ^ x 5 . The proposition, informally characterized in (18), embodies the semantic effect of the genitive. For isolating the effect, (18) has to be replaced by a more

Case competition in Russian

153

explicit formulation. As an essential step, the notion of sending, associated with the predicate c o n s t i t u e n t / ^ / , , must be made more precise.

5.2. Lexical meaning of the predicate constituent The lexical meaning o£poslal2 is a concept send-, not a concept have sent-. i.e., we consider postal, as a form of a verb that has both perfective and imperfective forms, not as a form of a perfective verb that has only perfective forms and whose meaning is a concept that introduces perfectivity as part of its intension. I" Adopting the conception of concepts characterized in Section 3.3, and obvious requirements for send- as a lexical meaning ofposM2 in sentence (13), we arrive at the following characterization of send-. Four entities x b x2, x3, and x4 are involved when send- is applied: an action Xl of sending by an actor x2 (e.g., the o^-person) of an object x3 (some money) to an object x 4 (the maten,person); naturally, this does not yet exhaust all aspects of an action of sending. send- must therefore be a four-place concept whose (four-place) extension is the set of quadruples < x b x2, x3, x4> such that the relation(-in-intension) SEND holds between x b x2, x3, and x4, the unit set {SEND} being the (four-place) intension of send-. The term " s e n d " is denned accordingly: (19)

send- = df the property of being a perception or conception z such that {SEND} is a subset of the content of z, where SEND = df lxx x2 x3 x4 [the relation-in-intension between any x b x2, x3. and x4 such that]: a. Xi is an action of sending by x2: b. x3 is an object affected by x b c. x4 is intended by x 2 in xx as a recipient of x3.

From the theory of concepts outlined in Section 3.3 we obtain the following consequences: (20)

a. send-is a four-place concept. b. "-send- [the four-place intension of send] = {SEND}. c. --send- [the four-place extension of send] = { | SEND relates x b x 2 , x 3 , and x 4 }.

10. Cf. Fnednch (2009: Section 9.5) for the underlying treatment of aspect in Russian.

154

Hans-HeinrichLieb

andSvetlana

Friedrich

The various aspects of the notion of sending that are essential to a reformulation of (18) have now been rendered explicit.

5.3. Formal version of the proposition Using the sentence-semantic framework outlined in Section 3.2 and replacing "part" by "Q-part" ("proper quantity-part"), we reformulate (18) as follows: (21)

Forallx 2 andx 4 , If i. [Speaker] V l refers by on, in [utterance] V to x2, AND n. Vxiefeis by materia in V to** THEN b. there are x b x 3 , and x5 such that: i. x* is an element of the one-place extension of money, AND n. x5 is an element of the one-place reference basis for deneg4, V b V. and money, AND in. x3 is a Q-part of x5, AND iv < x b x2, x3, x4> is an element of the four-place extension of send-. AND v. < x b x2, x3, XA> is an element of the four-place reference basis for ^/a/2,V,Vbandsend-,AND vi. for all x6, IF corresponds [Corr], for V b to theposlal2-part of V. THEN (3. Xj is earlier than x6, AND y. Xj is complete at x6.

Formulation (21) is a reading of the following formula, added for the sake of logical explicitness: (22)

(Vx2)(Vx4) [Ref V l 0 W l Vx 2 A Ref V 1 mater/ 3 Vx 4 -> (3xj) (3x3) (3x5) (x5 e e l -money A x5 e reb 1 (Je«eg 4 , V, V b -money) A e Partg A e e4-send- A e reb4(poslal2, V, V b -send-) A (Vx6) ( e Corr -^ < x b x6> e Earlier A < x b x6> e Complete))].

Case competition in Russian

155

In (21), conditions on reference bases - (b.n), (b.v) - restrict the entities in question to those which are 'relevant' for the speaker ( V l ) when he or she realizes a certain constituent (e.g., deneg4) in the utterance (V) together with a certain concept ( m o n e y ) that is the constituent's lexical meaning. Condition (vi) assumes a certain treatment for the indexicality of tenses (via) and represents the semantic effect of the categories Preterite (vi.(3) and Perfective (vi.y) that both markposlal,.

5.4. Partitivity in quantity sentences and in plurality sentences Intuitively, the semantic effect of the genitive marking of deneg, consists in the 'Q-part confining' - or 'Q-confining', for short - of poslal2 meaning send-, by a meaning of deneg4 (this meaning being represented essentially by (b.i) and (b.n) in (21)): The confining introduces some part of some money such that this part is being sent. Confining should in principle be possible in connection with any verb whose lexical meaning is a quantity-friendly concept (in the sense explained in Section 3.3), such as send-; put differently, in connection with any quantityfriendly verb as exemplified by the Russian verb for sending. It is exactly certain quantity-friendly verbs that permit one type of negationindependent competition m Russian. There is, however, a second major type: We may also have independent competition connected with group-friendly verbs - to which the Russian verb for sending also belongs - and the plural of count nouns." A sample pair of sentences is obtained from (8) by substituting papiros for deneg in (8a) and papirosy for den 'gi in (8b): (23)

a. On poslal maten papiros. he send mother cigarette PretPerfDat Gen P1N 'He sent his mother cigarettes.' b. On poslal materi papirosy. he send mother cigarette PretPerfDat Ace P1N 'He sent his mother specific cigarettes.'

(The translations - "cigarettes", "specific cigarettes" - are only approximate.) 11. deneg, and den'giA in sentences (8) are marked as Plural but are occurrences of forms of a substantive that is a mass noun having only Plural forms; the Plural category is semantically irrelevant here, as previously noted.

156

Hans-HeinrichLieb

andSvetlanaFriedrich

The analogies between (8) and (23) are striking, and the similarities between the two types of negation-independent competition are well known. Still, the two cases cannot be simply equated; a separate treatment for the second case will be proposed. We follow a recent tradition in the study of plural semantics by assuming that 'pluralities' - the potential referents of plural nominals - are groups (not sets). When a sentence meaning is constructed, then the concept associated with an occurrence of a plural form of a count noun - say, cigarette-, associated mthpapJros4 in (23a) and wthpaptrosy, in (23b) - is transformed into the corresponding 'group-concept', such as cigarette-group; and it is this concept that enters into further semantic processing.^ Our discussion of the 'quantity-sentences' (8), with all conclusions, carries over to the analysis of the 'plurality sentences' (23). In particular, the proposition of the genitive sentence (23a) is exactly like the proposition of the genitive sentence (8a), formulated in (21) and (22), requiring just the following changes in (21) (analogously, (22)): (24)

a. " m o n e y " is replaced by "cigarette-group": b. "Je«eg 4 " is replaced by "papirosf; c. "Q-part" is replaced by "G-part" ("proper group-part").

Because of the analogies between the two types of independent competition. our intuitive characterization of the genitive effect as 'confining' carries over from the quantity sentences to the plurality sentences; in the case of the genitive sentence (23a), the effect consists in the group-part confining (G-confining) of poslal2 meaning • send •, by a meaning oipapiros, represented essentially by the modified versions of(b.i) and (b.n) in (21).

6. The genitive effect: Semantic mechanism and syntactic basis 6.1. Introduction As stated in Section 3.2, a sentence meaning u is assigned to a given sentence by applying, typically, semantic functions that ultimately start from the primitive constituents o f / a n d the concepts associated with them by e, and 12. If b is an individualizing concept, then ft-group is the concept whose intension consists of the property of being a group whose members are elements of the extension of b, and the extension of ft-group is the set of all such groups.

Case competition in Russian

157

these functions are given either in the Syntactic Category Interpretation or the Syntactic Function Interpretation. We have informally identified 'confining^ functions that qualify as semantic functions for introducing the genitive effect when a proposition is constructed. These functions, together with their application conditions, each constitute a semantic mechanism connected with the genitive category in sentences of the relevant type. The mechanism must be anchored either in the Syntactic Function Interpretation or in the Syntactic Category Interpretation, i.e. the syntactic basis of the mechanism must be either a syntactic function or a syntactic category. From the very beginning, two different functions of confining, quantity confining or Q-confining and group-confining or G-confining, must be distinguished, due to the fact that two different part relations are involved, Q-part (proper quantity-part) and G-part (proper group-part). Because of the similarities between the two part relations it may be suggested that we should assume just a single type of confining functions, introducing a special relation component into the arguments of these functions that represents the two part relations. But such confining functions would no longer be semantic functions as permitted by the Integrational framework: A semantic function to be applied in sentence meaning construction may only have arguments that consist of constituents or their (lexical or constituent) meanings. Still, the two confining functions are analogous, and it should suffice to give a more detailed analysis of Q-confining, together with its application condition. using our standard example; the necessary changes for G-confining are then easily specified.

6.2. The genitive effect in quantity sentences: Example Consider, once again, the proposition for sentence (13), On poslal maten deneg., as formulated in (21). The consequent of (21) introduces three entities x b x3, and x5, by means of existential quantification: x, is an action of sending (b.iv), x5 is a quantity of money (b.i), and x3 is a Q-part of x5 (b.m) and is what is being sent (b.iv). Introduction of x, is an effect of the predicate relation holding between^oslal2 and the word sequence of the sentence, and is independent of the genitive effect. Similarly, the tense (Preterite) and aspect (Perfective) interpretations of poslal2, which involve only xh are not part of the genitive effect. However, introduction of x5, and of x3 as a part of x5, is directly or indirectly due to the genitive, given the two c o n s t i t u e n t s ^ / ^ and deneg, and their lexical meanings send- and money: It is x5 and x3 and their role in conditions (b.i) to (b.v)

158

Hans-HeinrichLieb

andSvetlanaFriedrich

that constitute the effect; xx (the sending), x 2 (the sender), x4 (the recipient), V (the utterance), and V, (the speaker) that also figure in these conditions must be treated as already given. It may now be suggested that we introduce Q-confining ("quantityconfining") as a function whose arguments are quadruples such as and that assigns to each argument a relation suitable to account for the genitive effect in the proposition. There is, however, a problem with this idea: Whereas poslal2, having complements, must be a primitive constituent and may therefore be associated with a concept like send- as its lexical meaning, the genitive constituent deneg4 is primitive only by accident; as soon as it is replaced by a complex constituent. the new constituent may no longer have a lexical but only a syntactic constituent meaning. In the cases under consideration, a constituent meaning is a relation-inintension d among entities x, [potential utterances] V, and [potential speakers] Vx. Even a primitive constituent may have a meaning of this kind. In the case of deneg4 in sentence (13), this is d*, defined as follows: (25)

d* = df Ix" V" Vx" [the relation-in-intension between any x", V", and Vf such that]: a. x" is an element of the one-place extension of money-: b. x" is an element of the one-place reference basis for deneg4 relative to V ' W i " , and money.

Both primitive and complex constituents may have such relations as their constituent meanings. For this reason, we do not include in the arguments of a confining function the genitive constituent and one of its lexical meanings but instead, simply a constituent meaning of the genitive constituent; i.e. the arguments of Q-confining will be construed as triples , where, as a matter of fact, / i s a predicate constituent, b a concept that is a lexical meaning off. and d a constituent meaning of the genitive constituent. We must also account for the number n of places of b, which leads us to introduce n functions Q-confining". These functions should be introduced in a way that makes the following statement come out true for our example: (26)

Q-confiningV 0 5 /a/ 2 , send-, d*) ["the four-place quantity-confining of ^/a/2andsendbyd*"] = lx{ x2' x4 V VY [the relation-in-intension between any x / , x 2 ', x 4 ', V . and V< such that]: There are x3 and x5 such that:

Case competition in Russian

a. b. c. d.

159

d* relates x5, V , and V{: x3 is a Q-part ["proper quantity-part"] of x5: is an element of the four-place extension of send-: is an element of the four-place reference basis for poslal2,N\N{, and send-.

Conditions (a) to (d) correspond to conditions (b.i) to (b.v) in (21). Given (25). the entire consequent (b) of (21) is logically equivalent to: (27)

There is an xx such that: a. has Q-confining4(poslal2, send-, d*) [i.e., the relation specified in (26) holds between x b x2, x4, V, and V J , AND b. [=(21b.vi)]forallx 6 , IF a. x6 corresponds, for Vi, to the poslaU-mxt of V. THEN (3. Xj is earlier than x6, AND y. Xj is complete at x6.

How, then, must the notion of Q-confining" be defined so as to make (26) and (27) come out true?

6.3. Semantic mechanism in quantity sentences (1): Quantity confining The arguments of the Q-confining" functions can now be characterized more precisely They are to be triples that satisfy the following conditions: i.

Ms an «-place concept like send, i.e. an «-place quantity-friendly concept as explained in Section 3.3; a concept that can be associated with an expression like poslal2 as its lexical meaning.

n. Because of (i), b is an action concept that involves an object affected by the action; b must therefore be at least 3-place. Moreover, since b is to be possible as the lexical meaning of a Russian verb, b can normally be at most (n* + l)-place, where n* is the maximal number such that some Russian verb has valency n*. Therefore, 3 < n < n* + 1.* 13. The valency of a lexical word is normally one less than the number of places of the concept that is the word's meaning, by the General Valency Hypothesis; see Lieb (1993: Section 5.6).

160

Hans-HeinrichLieb

andSvetlana

Friedrich

in. J i s to be a relation tike d*, more specifically, a quantity relation: a relationin-intension between entities x, [potential utterances] V, and [potential speakers] V, such that x is a quantity (winch may be a mass quantity or a substance quantity). Relations J e a n be associated as constituent meanings with certain types of primitive or complex constituents. The functions Q-confining" are formally introduced by a conditional definition schema that accounts for (i) to (in) in its antecedent (only (i) and (in) need be mentioned) and otherwise generalizes (26) by substituting V for "4"; y ' for "poslah"; "ft" for " s e n d " ; and " J " for "d*" We thus obtain: (28)

Suppose that ft is an «-place quantity-friendly concept; and d is a quantity relation. Then: Q-confimng\f, ft, d) = df W x2> x 4 ' . . . x„' V V^: There are x3 and x5 such that: b. x 3 i s a Q - p a r t o f x 5 ; " c. is an element of the «-place reference basis for/relative to V , V^, and ft."

It is easily seen that definition schema (28) makes (26) true for n = 4; / = poslal2; ft = send-; and d= d*; which in turn makes (27) come out true. We next establish when exactly the Q-confining" functions may apply in the construction of a proposition. Once again, we take our lead from the sample sentence (13).

6.4. Semantic mechanism in quantity sentences (2): Application conditions for quantity confining The following syntactic and semantic properties of sentence (13), construed as a specific triple of a suitable Russian idiolect system S, are relevant to applying Q-confining* in the construction of proposition (21) (for the syntactic properties, see Section 3.1, above): i. n.

is a sentence of S send- is afour-place quantity-friendly concept.

14. For n = 3 the schema is to be understood as not including x4' to x„'. Similarly in (34). below.

Case competition in Russian

161

111. d* is a quantity relation. iv.

send- is a lexical meaning ofposM2 in/, s, e, and S.

v.

d* is a constituent meaning of deneg, in/, *, e, and .9.

vi.

< 0 « b deneg4, materh,poslal2>

e comp3(/ s , e, S).

vn. Je«eg 4 is marked as Gen(- S) i n / s, e, and .9. vin. postal* is marked as NOM + AKK/GEN [+ DAT](-, - S) i n / *, e, and S Conditions (n) and (m) are needed in view of the antecedent in (28). Generalizing from (i) to (vm), we obtain a condition that is necessary and sufficient for applying the Q-confining" functions, to be called the QccondWon" ("the «-place quantity-confining condition", "c" for "confining") in Russian idiolect systems S. There is just one complication in formulating these conditions for quantity confining: In computing the valency of the predicate verb (3, in the example, see (vi)) from the number of places of the concept that is the predicate-verb meaning (4, in the example, see (n)), using the General Valency Hypothesis (Lieb 1993: Sections 5.5 and 5.6), we must subtract not only 1 but also the number of 'deictic' places of the concept, i.e. the places related to speakers or utterances (zero, in the example, which does not appear overtly). This will be taken into account. Formally, a value of Qc-condition" is a six-place relation (set of six-tuples) between certain syntactic and semantic entities; the relation is assigned to Russian idiolect systems S by means of the function denoted by the term "Qccondition"", for a suitable n, as specified in the following definition schema: (29)

For any Russian idiolect system S: Qc-condition\S) = df the set of all < / , b, d,f, s, e> such that: a. is a sentence of S;« b. Ms an «-place quantity-friendly concept; c. d is a quantity relation: d. Ms a lexical meaning o f / in/, s, e, and S: e. for m = n minus 1 minus the number of deictic places of b, there are fhf2,...,fm such that: i- e comp m (/ s, e, S): n. / 2 i s m a r k e d i n / , , e , a n d S a s G e n ( - , S ) : in. d is a constituent meaning of/2 i n / s, e, and S;

15. This unphes that there is a sentence meaning for < / ,, e>. ~ Generalizing (a) to 'sentence combinations' is a technical problem, here left undiscussed.

162

Hans-HeinrichLieb

and Svetlana Friedrich

f. f is marked i n / ., e, and S as an occurrence of a form of an ACC/ GEN-verbof^ Definition schema (29) is indeed a generalization of (i) to (vm): n = 4; m = 3: / ' = ^ / a / 2 ; ft = -send-; J = d*, / = on,, f2 = deneg4; f3 = materi3; < / s, e> = = Onposlalmatendeneg.

6.5. Syntactic basis for the semantic mechanism The syntactic basis for the semantic mechanism is either the genitive category itself, if the confining" functions are given in the Syntactic Category Interpretation; or else, is a syntactic function in case the confining" functions are supplied by the Syntactic Function Interpretation. Suppose that S is a Russian idiolect system and n is the number of places of at least one concept that is quantity-friendly and is the meaning of an ACC/ GEN-verbofS. We consider the pair (30)

.

This pair is an interpretation either of the syntactic category Genitive(- S) or of the restriction to S, g[S], of a syntactic function g. In the first case, we assume that the triple (31) is an element of the Syntactic Category Interpretation of S; in the second case. we assume the following triple as an element of the Syntactic Function Interpretation of S: the triple (32) . Either possibility is compatible with (30). There is, however, a reason to prefer (32) over (31). 16. An ACC/GEN-verb of S is any verb of S that is at least two-valued and allows as a second complement both an accusative constituent and a genitive constituent. By (iv), / ' is to be an occurrence of a form of such a verb such that the relevant government category of the verb appears in the marking structure, the second component of the syntactic structure s off, in connection with/'.

Case competition in Russian

163

It appears from (29e) that the function g that qualifies in the context of (32) is complementation. Now Q-confining" directly involves the constituent/ that has complements, and its lexical meaning b, whereas the constituent marked by Genitive is represented only indirectly, through its constituent meaning d. We therefore conclude that (32) is correct, i.e. we assume that (33)

is an element of the Syntactic Function Interpretation of S, for any n such that S is a Russian idiolect system and n is the number of places of at least one concept that is quantity-friendly and is the meaning of an ACC/GEN-verb of S such that m = n minus 1 minus the number of deictic places of the concept. Whenever in the construction of a proposition the Qc-condition"(S) is satisfied, the Q-confining" function applies. The Genitive category Gen(- S) figures only in the application condition Qc-condition"(S): Gen(- S) is not itself interpreted; its semantic effect is not direct, through an interpretation in the Syntactic Category Interpretation of S, but only indirect, by figuring in the application condition for a semantic function that belongs to an interpretation of the restriction of a syntactic function, comp™, for each admissible m. We next consider the situation for group confining.

6.6. The genitive effect in plurality sentences Sentence (23a), On postal materi papiros., may serve as a sample sentence. The group-confining functions G-confining" are analogous to the quantityconfining functions, with one major difference: The genitive constituent which may but need not be a primitive constituent like papiros, - is in the Plural, and this is semantically relevant; the constituent meaning d of the genitive constituent must be a group relation: a relation-in-intension between entities x, [potential utterances] V, and [potential speakers] Vi such that x is a group. (Allowing only for groups of 'small' objects in the present context, as suggested by a statement made by Gladrow in 1998: 55, appears to us unwarranted.) The reason is our conception of plural semantics by which the semantic effect of the category Plural (of nouns) consists in the determination of groups whose members are from a certain set, in the simplest case, from the extension of an individualizing concept that is the lexical meaning of a primitive constituent. such aspaptros, in sentence (23a). It is not enough, though, to require that d should be a group relation; the condition of the genitive constituent being marked as Plural is independently

164

Hans-HeinrichLieb

and Svetlana Friedrich

important: We must exclude the case of a genitive constituent by which a group noun is used in the singular. On the other hand, a genitive constituent that uses a group noun in the plural must be admitted. All these requirements are satisfied if the application conditions for the group-confining functions contain the conditions: (i) d is a group relation and a constituent meaning of the genitive constituent, and (11) the genitive constituent is in the Plural. The definition schema for the group-confining functions can now be obtained by making a few substitutions in (28), the definition of Q-confining": i.

"group-friendly concept" for "quantity-friendly concept":

n. "group relation" for "quantity relation"; in. "G-part"for"Q-part". This results in the following definition schema:" (34)

Suppose that ft is an «-place group-friendly concept; and d is a group relation. Then: G-confimng" (f, ft, d) = df ^ x2> x 4 ' . . . x„> V V^: There are x3 and x5 such that: b. x 3 i s a G - p a r t o f x 5 ; " c. is an element of the «-place reference basis for/relative to V , V x ', and ft.

Similarly, a definition schema for Gc-condition"(S) is obtained from (29) by means of the changes (i) and (n), plus: iv Add "and P1 N (- S)" in (29e.ii). This yields the following formulation: (35)

For any Russian idiolect system S: Gc-condition\S) = df the set of all 2* tl-confining" ["«-place type-1 c o n f i n i n g " ^ b, d) =xXx{ x 2 ' x 4 ' . . . x„' V V such that: a. is a sentence of S; b. M s anyplace concept; c. d is a singularity relation or a group relation: d. M s a lexical meaning o f / i n / . , e, and S; e. for m = n minus 1 minus the number of deictic places of b, there are A,f2,...,fm such that: i. e c o m p l y ; s, e, S): n. / 2 i s m a r k e d i n / W n d S a s G e n ( - , S ) : in. d is a constituent meaning off2 i n / s, e, and S: iv. if J is a group relation, then f2 is marked in f, s, e, and S as Sg N (- S); f. f is marked in /J s, e, and 5 as an occurrence of a form of a type-1 negation-friendly verb of S; g. f is negated in/, ,, e, and ^.24

We previously decided to associate the semantic functions of quantity-confining and group-confining with the syntactic complement functions, rather than directly with the category Genitive. The reasons given for this decision apply even more strongly in the case of the new confining functions, due to the fact that their application conditions impose an even stronger requirement on the government category of the predicate verb. True, it may be argued that we have also added another condition, that of a negated predicate constituent (42g): should we not take the syntactic function of negation as the syntactic basis of the semantic mechanism? However, the negation requirement is only a consequence of the strengthened government condition; the negation effect concerning tl-confining should therefore be construed as indirect, same as the effect of the category Genitive. - We therefore assume that (43)

24. This is to mean that/' is the scope of the (sentence) negator whose domain is the constituent that h a s / a s its nucleus.

Case competition in Russian

171

is an element of the Syntactic Function Interpretation of S, where Sis a Russian idiolect system and n is the number of places of at least one concept that is the meaning of a type-1 negation-friendly verb such that m = n minus 1 minus the number of deictic places of the concept.

7.5. Excluding lexical case competition. Intensional verbs Differences in verb government can also be adduced to solve the puzzle of apparent genitive/accusative competition that is just like type-1 competition but is not negation-dependent; such apparent competition has been discussed since Z a l i z n j a k ( 1 9 6 7 ) : ^ « a z ^ 0 ? v e t o ( G e n , Sg)/otvet (Ace, Sg). Rather than introducing anew case (zdatel'nyj padez, 'case of waiting', Zaliznjak 1967: 49), this may be explained by assuming not one verb but two verbs that differ in lexical meaning: One verb, governing the accusative, corresponds to German erWarten 'expect', in the sense of expecting something to arrive that exists; the accusative is referential: 'There is an answer that Anna is expecting.' The other verb. governing the genitive, corresponds to German warten aw/'wait for', in the sense of being involved in a waiting process whose aim is of a certain kind; the genitive is non-referential: 'Anna is involved in a waiting-for-an-answer'. Case competition, which is non-lexical and requires a single verb, is here mirrored by government differences between two different verbs whose lexical meanings are such that they jointly represent, in positive sentences, the effects that type-1 confining has in negative sentences. In the case of such verb pairs, we may loosely speak of lexical case competition, to be distinguished from syntactic case competition, or case competition, for short, which is the topic of the present essay. The proposed analysis may however be questioned, if not in this case then in other comparable cases. There is indeed an alternative: A single ACC/GENverb is assumed whose meaning is an 'intensional' concept of a certain type (such as the concept of searching), and a semantic function is associated with the relevant syntactic comp-function that is operative only when there is a genitive complement of such a verb. Details of sentence meaning composition could then be as suggested in Lieb (1983: Sections 26.5 and 26.7) for sentences with English look for. Given the first analysis, the use of the genitive under discussion is outside the competition area. Given the second analysis, it falls within the area as an additional instance; its treatment similar to other uses analyzed in this essay appears possible. (Borschev et al. 2008 argue that, adopting the 'property-type

172

Hans-HeinrichLieb

and Svetlana Friedrich

hypothesis', the genitive of negation and the 'genitive of intensionality' may be 'unified'. See also Partee 2008 for relevant discussion.) It remains for us to consider the much-discussed situation of the genitive competing under negation with the nominative, rather than the accusative.

8. Type 2: Genitive vs. nominative 8.1. Sample propositions For sample sentences, we choose the pair going back to Apresjan (1966: 193): (44)

a. Otvet iz polka ne prisel answer from regiment not arrive Norn Masc N Sg N 'The answer from the regiment has not arrived.' b. Otveta iz polka ne prislo. answer from regiment not arrive Gen Masc N SgN 'There was no answer from the regiment.'

In either case, we assume sentence accent on the predicate constituent, marked by the aspect category Perfective and the tense category Preterite in both sentences. The other verb-form categories that are relevant here remain to be established. As a lexical meaning of the predicate constituent, we take the partly deictic concept of 'arriving at speaker's place'. This concept is fourplace, involving an event x b an agent x2, an utterance x3, and a speaker x4; the places for x3 and x 4 are deictic. What are the propositions for the two sentences? They may be assumed to be strictly analogous to the propositions that were postulated in (39) and (40) for the sentence pair used in (37) to exemplify negation-dependent accusative/ genitive competition: (45)

a. Nominative sentence (44a): For all x2, if V l refers by otveh iz2polka3 in V to x2, then NOT: There are x b x 3 , and x4 such that: i. < x b x2, x3, x4> e e4 • arrive •; AND ii. < x b x2, x3, x4> E r e b V m e / 5 , V, V b -arrive-); AND in. x3 corresponds to V for V l ; AND

Case competition in Russian

173

iv. XA corresponds to Vi for V b AND v. [effects of tense and aspect]. b. Genitive sentence (44b): NOT: there are x b x 3 , and x 4 such that: i. for some x2: a. x2, V, and V\ are related by Lotvetax iz2polka^; AND B. e e4-arrive-; AND y. < x b x2, x3, x4> E r e b \ ^ m / 0 5 , V, V b -arrive-); AND ii. toiv[=(a.iii)to(a.v)]. (Conditions (B) and (y) correspond to (a.i) and (a.n). Condition (a) now takes the place of the antecedent in (a).) Informally, in uttering the nominative sentence the speaker is asserting that any x2 to which he or she is referring by the subject constituent has not reached the speaker, whereas the claim made by means of the genitive sentence is as follows: There is no x 2 such that (i) x 2 is related to the speaker by the constituent meaning of the subject constituent (the meaning is denoted above by enclosing the subject constituent in inverted commas), and (n) x 2 has reached the speaker. Moreover, the subject constituent in the nominative sentence must be used with a definite referential meaning, due to the fact that it is referential and does not bear the sentence accent (an attributive or a generic meaning may here be excluded). We therefore have it as part of the sentence meaning - though not the proposition - that the speaker is referring by the subject constituent to exactly one x2. In contradistinction, there is no such part of the sentence meaning when the genitive sentence is uttered, due to the fact that the subject constituent is non-referential; what is claimed, is the non-existence of the arriving of any x2 that is as required by the constituent meaning of the subject constituent.

8.2. Comments: "Perspectival Center" Partee and Borschev propose an analysis of type-2 case competition by which the nominative sentence and the genitive sentence in a sentence pair differ in 'Perspective Structure': through a specific choice of a 'Perspectival Center'; the authors oppose the nominative and the genitive sentence as follows (Partee and Borschev 2007b: 171): When the THING is chosen as Perspectival Center, its existence is presupposed. and the sentence speaks of its LOCation and potentially about other properties or states or actions.

174

Hans-HeinrichLieb

and Svetlana Friedrich

When we choose the LOCation as Perspectival Center, the sentence speaks about what THINGS there are (or not) in that situation and/or about what is happening m the situation.

Partee and Borschev further hypothesize (2007b: 172): [ . . . ] choice of Perspective Structure is [ . . . ] a "diathesis choice", a choice among two alternative argument structures for verbs that take both a "THING" and a "LOC" argument, analogous to the argument structure choices (diathetic alternations) for verbs like spray, load or like give, send.

On our analysis, 'Perspective Structure' can be construed as an automatic byproduct of word and sentence meanings, without involvement of a 'diathesis choice'. Consider the nominative sentence, where the subject constituent is referential. On an IL view, the referential part of a sentence meaning is separate from the proposition; hence, the potential referents of a referential expression are specified independently, and may therefore 'stand out' as the Perspectival Center, existence of a referent being 'presupposed' in the proposition. The subject constituent is non-referential in the genitive sentence. The same verb is used as a predicate in both the nominative sentence and the genitive sentence (in contrast to sentence pairs with English spray, where two verbs not one should be assumed: 'spray something on something', 'spray something with something'). On a defensible interpretation of the second quote from Partee and Borschev (2007b), the verb used as a predicate may be understood to have as its meaning a concept b such that: If is in the extension of b, then a location is assigned to the agent-place filler x 2 of event x b either implicitly (by requiring existence of a certain location through b 's intension) or as an explicit possibility (by requiring or allowing some x,, i > 2, as the location of x 2 ); in the second case, we have a verb that requires or allows a locational complement, possibly optional. When such a verb is used in a genitive sentence, no agent-place filler x 2 is referred to; existence of a filler becomes part of what is being predicated in the proposition: 'The sentence speaks about what THINGS there are (or not) in that situation'. At the same time, the locational aspects of the predicate's lexical meaning remain unaffected, and 'location' can stand out as the Perspectival Center. The difference may or may not be analogous to instances of diathesis, but it does not require any specific syntactic or semantic underpinning as its basis. In (38), consensus claims were characterized on genitive vs. accusative and nominative in negation-dependent case competition. These claims are sup-

Case competition in Russian

175

ported, in a precise form, by our analysis. We still have to identify the underlying semantic mechanism and its syntactic basis.

8.3. The semantic functions In the proposition formulated in (45b) for the genitive sentence in (44b), the following part must be introduced by a semantic confining function: (46)

For some x2: i. x2, V, and Vi are related by cotveta-\ iz2 polkaS; AND 11.

E^.arrlve.;AND

in. < x b x2, x3, x4> E r e b V ™ / 0 5 , V, V b -arrive-). This part of the proposition can be obtained by means of a semantic function t2-confining4 ("four-place type-2 confining") that is analogous to tl-confining but involves the constituent meaning not of an object constituent but of the subject, in addition to the predicate and its lexical meaning: If d* = 'otveta, iz2 polka,' is the constituent meaning of the subject, then t2-confiningV™7 05 , arrive-, d*) = Xx{ x 3 ' x 4 ' V V 1 .* t2-confiningn ["«-place type-2 confining"](f. b, d) = df Xx/ x 3 ' . . . x' V V i : There is an x2 such that: b.

E r e b » a V , V 1 ' , Z , ) . We obtain the type-2 confining functions by requiring existence of x2, rather than existence of x3 as we did for the type-1 functions. Definition schema (42), introducing the application conditions for type-1 confining, may serve as a starting-point for obtaining the conditions that must be satisfied for applying the type-2 confining functions. 25. For n = 2, the schema is to be understood as not including x3' to x„'.

176

Hans-HeinrichLieb

and Svetlana Friedrich

8.4. Towards application conditions We previously (Section 7.3) introduced type-1 negation-friendly verbs as a subclass of ACC/GEN-verbs. Similarly, we now need type-2 negation-friendly verbs; however, these are simply identical to the NOM/GEN-verbs, where "NOM/GEN-verb" is informally denned as "verb that can be used under negation, and only under negation, with a 'first' complement in the genitive, provided the verb form used is a 'permitted form'." (It is not yet entirely clear how the verb or verbs byf - lexical words one of which must be the copula that has the relational 'be' as its meaning - are related to the NOM/GEN class; see Partee and Borschev 2007a for discussion.) -Unpermittedforms comprise: (i) the so-called o-forms, i.e. the Singula^ Neuter v forms of the Preterite, all of them Unspecific for Person (Unsp Ps ); (n) the Plural forms of the Preterite, which are not only Unspecific for Person but also Unspecific for Verbal Gender (Unsp VG ); (iii) 3P Singular and 3P Plural forms of the Present and the Future. Typically, it is only forms (i) that are mentioned in discussions of genitival subjects. However, 3P Sg forms of the Present and the Future must clearly be included. Inclusion of Plural forms, Unspecific for Person in the Preterite and 3P in the Present and in the Future, is more debatable but may be defended in view of sentences with these forms that are analogous to 'negative existential sentences'with forms (i). Now suppose we do take the genitive constituent in relevant sentences, such as (44b), to be the subject. This raises a well-known problem: The requirements of subject/predicate agreement need not be satisfied either with respect to Number (Singular otveta, in (44b) could be replaced by Plural otvetov,) or Gender (otveta, is M a s c N , ^ m / 0 l is Neuter v ). When permitted Present or Future forms are used, there may be lack of number agreement. At least the two following solutions may be suggested: A. We retain the genitive constituent as a subject, and accept the violation of agreement requirements (the traditional solution). B. We reject the genitive constituent as a subject. As a matter of fact, a third solution may also be considered. Using categories of unspecificity: Unspecific for Person, for Verbal Number, and for Verbal Gender, an additional categorization is introduced for each permitted form that employs these categories, removing agreement violations when the genitive constituent is construed as the subject. We originally tried this out in detail. with uncertain success. We refrain from reproducing the discussion here because for independent reasons, it is Solution (B) - rejecting the genitive constituent as a subject - that may have to be adopted.

Case competition in Russian

111

For now we follow Solution (A), formulating the application conditions for t2-confining accordingly.

8.5. Application conditions. Syntactic basis The following formulation is analogous to (42): (48)

For any Russian idiolect system S: t2-condition\S) = df the set of all ; g. / i s negated i n / ,, e, and S.

(The three parts of the alternative in (48f) account for, respectively, permitted forms of type (i), type (n), and type (m), see Section 8.4.) In a marked deviation from previous application conditions, the type-2 conditions do not impose any particular requirements on the constituent meaning J of the genitive constituent / b which is in subject position: J may be a quantity relation, group relation, or singularity relation; a l s o , / may be in the Singular or in the Plural. Once again, it is the syntactic complement function that should be taken as the syntactic basis of the semantic mechanism: Both condition (f), requiring that the predicate constituent should be an occurrence of a permitted form, and condition (g), requiring that the predicate constituent should be negated, are in a sense implied by requirement (e), asking for NOM/GEN as a verbal government category. We therefore assume that (49)

etait? your name it be-iND.iMPF 'What was your name again?'

(13)

Quandest-ce, dejd, when is it again neme rappelle notmyselfremember 'When is it again that

qu' elle allait chez le me deem? Je that she go-iND.iMPF to the doctor I plus. more she was going to the doctor? I do not remember' (Moeschlerl993:64)

9. Finally in the hypocoristic use (observed in dealings with little children or pets), the event is present and an important pragmatic secondary effect has been noted: the speaker seems to "speak instead of the addressee" (see Bres 2003b: 118-122). However, the imparfait on its own is not responsible for this pragmatic effect since./?™*, other tenses can be used in association with it (see

192

Marie-HeleneViguier

Bres 2003b: 112-113), and second, the imparfait can also retam its standard (past) meaning despite the hypoconstic effect. But as Bres (2003b: 123-124) notes, the atypical temporal meaning of the rmparfart ("present") - in this use at least-leads to a hypoconstic reading. (14)

Commeil etait sage! how he be-iND.iMPF good Commeil aimait bien sa maman! how helike-iND.iMPFwellhismom 'How good he is! How much he likes his mom!' (Grevisse[1936]1986:1292)

To summarize, the imparfait has the following semantic features according to variation of use:

Uses Standard Absolute Narrative Unreal in si-clauses Counterfactual Preliminary Politeness Futurate Hypoconstic Figure 1.

Temporal past past past non-past past present present non-past present

Type of semantic effects Modal Aspectual Pragmatic real imperfective real real a type of perfective accentuation unreal unreal — accentuation unreal real imperfective politeness real real — indirect dialogue

Semantic properties of the different uses of the imparfait (a dash indicates that for a given use, there is no effect of the kind described)

According to a monosemic approach, the imparfait maintains the same singular meaning in all its uses. Faced with the functional plurality of the imparfait, two possible attitudes can be observed.* 1. The "purists" look for a minimal semantic basic value that would remain unchanged in all the occurrences of imparfait. They accept neither neutralization nor "filtrage" (selection of some properties of the basic meaning and omission of others depending on the co(n)-text, following the terminology in Bres 2003a: 5). Given fat, first, the imparfait is used to express past, present, and 4. These attitudes correspond to the "attitude A5" and "attitude A4" in Confais (1995: 63).

The polysemy of syntactic categories and semantic coherence

193

future events, second, it is not always imperfective, and third, it is used for both real and unreal sentences, one must either assume a series of tense metaphors (like Weinnch 1964) or abandon any reference to time and any aspectual or modal feature. These purists have no other choice than to assume an underspecification of the time, aspect, and modality parameters: "We ascribe to this tense an u n d e r s p e n d semantics that is able to receive multiple interpretations; we will make the case that it is the semantically inconsequent nature of this tense that explains this multiplicity of uses." (Caudal, Vetters, and Roussane 2003: 61, translation mine); In practice, this attitude is problematic because in numerous sentences, the imparfait can not be used precisely due to modal, temporal, or aspectual incompatibilities - or it can be used but it induces a pragmatic textual effect. Furthermore, it generates effects even with minimal elements in the context (see Bres 1999: 91 about the isolated single sentence Welle rentrait 'he/she was coming home'). Finally, even purist monosemists cannot forgo neutralization since even the authors cited above assume partial neutralization of the aspectual properties in the unreal uses (see Caudal, Vetters, and Roussane 2003: 69-70). 2. Most monosemists assume a basic meaning which appears differently depending on the context. Defenders of this thesis are, for example, Guillaume ([1929] 1965), Warnant (1966: 365), Waugh (1987: 2), Berthonneau and Kleiber (1993), Bres (1999, 2003a,b),« or Haillet (2003). According to such a conception, a general interpretation like "past, imperfective, and real" is, in principle, compatible with "present, neutral, and unreal" in ./-clauses, provided that 5. Original quote: "Nous attnbuons a ce tiroir une semantique sous-speficiee, susceptible de recevoir de multiples interpretations; nous avancerons l'hypothese que c'est a la nature semantiquement inconsequente de ce temps que l'on doit cette multiplie d des usages." 6. Although Bres (1999: 88, 2003b: 124) qualifies himself as purist monosemist, he assumes that due to the interaction between the core meaning of the imparfait and the context, the event is not necessarily understood as demanded by the core meaning, but that either nothing or only one feature of the core meaning remains visible. See his significant formulation: "Let us take, as an experiment, one and the same simple sentence in the imparfait: il/elle rentrait ['he/she was coming home']. What does it show? The time implicated by the process to come home crossing the timeline from a point localized in the past. This sentence, I 'dunk' it into two different cotexts: [ . . . ] " (Bres 1999: 91, translation mine, emphasis mine; M.-H.V.). Original quote: "Prenons, a titre d'expenence, un meme enonce tout simple a l'imparfait: W elk rentrait. Que me donne-t-il a voir? Le temps imphque par le proces rentrer en tram de s'msenre en traversee de la hgne du temps a partir d'un point de saisie situe dans le passe. Cet enonce, je le 'trempe' dans deux cotextes differents [ . . . ] . "

194

Marie-HeleneViguier

contextual elements can be held accountable for the selection or for the omission of some features of the basic meaning. But the functional diversity of the rmparfart can in no way be justified by the context. Let us take an example where the influence of the context on the final reading has been explicated in a precise way. Caudal, Vetters, and Roussane (2003: 68) ascribe to the imparfait the invariant meaning Nontranslnactuel, a mixture o£nan transUwnnel 'non transitional' and mactuel 'unactual'. Inactuel means that by using the imparfait, the speaker does not say anything about the validity of the event at the time of utterance - the imparfait brings out different facets of inactuel depending on the co(n)-text: a temporal one as "past" if temporality is given by the context, or a modal one as "non real" in non-assertive sentences (i.e. in hypothetical and counterfactual uses, see Caudal, Vetters, and Roussane 2003: 68-70). Among other things, non transitionnel has the effect that, at least in assertions, only the internal phase of the event is explicitly predicated, so that it is left unspoken whether or not the resulting phase is reached. In ./-clauses, the non-assertive component provided by the elements Hypothese, Consecution of the sentence structure explains that the event is understood as unreal and that not only the internal phase of the situation must be expressed - in other words, the imparfait is unmarked aspectually (see Caudal. Vetters, and Roussane 2003: 69). This analysis, however, is not tenable for the following reasons. First, these conditions for unreality (and, as a result, aspectual neutrality) are too vague. They cannot justify why the imparfait has no unreal interpretation in (16) and (17) as opposed to (15), although the co-textual ingredients (Hypothese, Consecution) are realized. This demonstrates in fact that it is not the non-assertive elements per se but the presence of ri that leads to unreality and aspectual neutrality. (15)

Si Yannig venait, on manger ait des crepes. if Yannig come-iND.iMPF we eat-coND.PRES crepes 'If Yannig came, we would eat crepes.' (Caudal, Vetters, and Roussane 2003: 68)

(16)

ISupposons que Yannig venait. Mors, on suppose-iMPERATivE.PRES that Yannig come-iMPF then we mangeraU des crepes. eat-coND.PRES crepes 'Let us suppose that Yannig came. Then we would eat crepes.'

(17)

Dans le cas oil Yannig venait, on mangerait in the case where Yannig come-iND.iMPF we eat-coND.PRES

The polysemy of syntactic categories and semantic coherence

195

des crepes. crepes ?'In case Yannig came, we would eat crepes/ Second, the imparfait making in (16) produces a past effect whereas nothing in the co(n)-text is said about a localization in time; in (18) an interpretation as unreal exists despite the temporal context provided by tout a Vheure ('earlier'). Thus, temporality does not specify the past vs. unreal meaning of the event. (18)

Tout a Vheure, vans faisiez un pas de plus, vous ettez earlier you make-iND.iMPF a step more you be-iND.iMPF dansle gouffre. in the abysm 'Earlier, one more step and you would have been in the abysm/

Third, this analysis does not correctly predict how inactuel will turn out temporally given a sentence - why in standard assertions the event is confined to the past (not to the past and future), why in ./-clauses it is located in the present or the future (not only in the present like in the hypoconstic use), and why in the counterfactual use, it is located in the past despite its being unreal. In principle, a monosemic analysis of the imparfait would be maintainable if at least the components of its meaning occurred systematically - for example, if in assertions it always led to "past" and if in non-assertive sentences it always led to "present"; in that case, one could say that inactuel results in "non present" in assertions (hence, would influence the sentence temporally), and in "non real" otherwise (hence, would influence the sentence modally). Yet as the counterfactual use illustrates, it is understood as "past" and "unreal" simultaneously. As we can see, the dimensions of time and modality do not compete with each other. Rather they coexist and vary independently^ 7. In analogy, Sthioul (1998) can be reproached for confusing temporality and modality whereas they should be combined: it can not be sufficient to interpret «P + S" (that is: the event is seen from a point of view that is outside of the time of utterance) for the preliminary and the hypoconstic imparfait in the sense of a passage in another speaker's world (instead of as an anteriority), because for the unreal uses of imparfait, we would have to speak about both a passage in another speaker's world and an anteriority. Moreover, sentences with the preliminary imparfait and those with hypoconstic imparfait have to be interpreted differently, even if one recognizes a passage in another speaker's world in both types of uses. If precise descriptions and application conditions for the respective interpretations are not given, vague instructions are unusable. 8. For an argument against a rigid theory of an overall meaning of tenses, see also Comne (1985: 18-23).

196

Marie-HeleneViguier

Remark. Cosenu (1976) argues for the existence of a Gesamtbedeutung, an 'overall meaning', of each functional umt, corresponding to the different possible uses. He underlines that the overall meaning justifies these uses because otherwise it would be impossible to explain "why the speakers of a language use the same forms - for different meanings and in new sentences and contexts which they have never heard before" (Cosenu 1976: 48, translation mine).' This idea is questionable because children do not use the imparfait in hypothetical ./-clauses on their own since they incorrectly employ the conditional in this situation. Moreover, this incorrectness also happens to many adults, sometimes even the most educated. On the other hand, it has been noticed by Sten (1952: 95) that native speakers never make mistakes concerning the aspect (i.e. putting the compound past instead of the imparfait or conversely). For this reason, I believe that in unreal ./-clauses, the speaker's only choice is between the imparfait and the pluperfect (see the clear opinion of Pollak 1960: 21-22 concerning this question). Rather, the existence of the imparfait in hypothetical ./-clauses seems to have a historical background (see Pollak 1960: 24).«> In my opinion, monosemists have made a mistake by concentrating on finding unity or at least an internal connection among the meanings of a syntactic category, although this should in fact be seen as a question of historical linguistics. On the contrary, it is crucial for theoretical as well as for didactical purposes to describe, as correctly and precisely as possible, both the different meanings of a syntactic category and the conditions in which these meanings are available. Analyses should aim to foresee which meaning(s) of a syntactic category are possible given a certain sentence. In this sense, Moeschler et al. (1998) conceive the meaning of a tense category as an algorithm (i.e. a set of ordered instructions) which leads to different concrete interpretations following the pragmatic principle of "optimal relevance," defined in Sperber and Wilson (1986) or Wilson and Sperber (1993). In this way, the variety of readings is already foreseen in the basic meaning of a tense: "To Guillaume's conceptual monoguism, we thus propose to counter a procedural monoguism: in a procedural schema, the various uses appear in the form of different possible outputs of a unique procedure that refers, in its entirety, to the signification of

9. Original quote: "warum die Sprecher emer Sprache dieselben Formen gebrauchen. - fur verschiedene Redebedeutungen und in neuen, mcht wiederholten, sondern von ihnen konstruierten Satzen und Zusammenhangen." 10. For a criticism of the monosemist analysis of Cosenu (1976), see also Blumenthal (1986: 15).

The polysemy of syntactic categories and semantic coherence

197

the given morpheme" (Sthioul 1998: 59-60, translation mine)." As we can see, this approach is rather polysemic in so far as the meaning of a tense is considered as the set of its several meanings connected to the respective conditions in the context. Similarly in the conception of Lieb (1983, 1993: 439), the semantic content of a category is the set of semantic functions that are associated with the category, with each function connected to precise conditions of application. Yet unlike the authors of Moeschler et al. (1998), these conditions are syntactic, lexical, or morphological, so that the variety of interpretations can be handled with in the syntactic semantics, and not only in the pragmatics - but the choice of the "right" interpretation in a concrete utterance is seen, of course, as pragmatic. This point shall be elaborated upon in the next section.

2. Category semantics within Integrational Linguistics In the Integrational conception, the semantics of a syntactic category does not consist of one meaning, but of a set of meanings which are each associated with a certain condition of application. Consequently, nothing goes against the fact that the imparfait is associated, outside of hypothetical ./-clauses, with a past and an imperfective function, and in hypothetical ./-clauses, with a basic non past- (and unreal-) function. I will illustrate this idea in the following example. Let us consider a modern part of Standard French, more precisely a system S of a Modern Standard French idiolectVI will use the following notations:

11. Original quote: "Au monoguisme conceptuel de Guillaume, nous proposons done d'opposer un monoguisme procedural: dans un schema procedural, les divers emplois apparaissent sous la forme de differentes sorties possibles d'une unique procedure qui renvoie, dans sa globalite, a la signification du morpheme donned 12. The part of a language held by the speaker at a particular time is not homogeneous. given that it is made up of various varieties (dialects, slang, technical languages: foreign languages), registers, or individual properties. As a result, Integrational Lim guistics introduced the concept of "idiolect" to represent the biggest homogeneous part of a speaker's language contribution. "Homogeneous" refers to the fact that this part can only belong to a variety of the language in question in its entirety (see Lieb 1980: 197). Note, however, that it can belong simultaneously to several varieties. Each idiolect has a system which contains a phonetic-phonological (or grapheticgraphemic in the case of a written idiolect), a morpho-syntactic and a semantic subsystem and that determines which texts belong to the idiolect.

198

Marie-Helene Viguier

- "Imparfait(- S)» for the category "Imparfait in S," i.e. the set of forms of the imparfait of S (for all 2-plaee relations R between entities/and idiolect systems S, «R(-, ST designates the set of all first components of pairs ofR, i.e. "{/I e R}")* - "past in S" for the semantic function whose role is to signify in a sentence that the event is currently over, - "non-past in S" for the semantic function whose role is to signify in a sentence that the event is currently not yet over, - "imperfective in S" for the semantic function whose role is to make the sentence understood in an imperfective way, especially to show that the event is in progress, - "unreal" for the semantic function which signifies that a statement is presented by the speaker as "not valid."" If the spectrum of imparfait meanings that we listed in the first section is correct, then the "semantic content" of Imparfait(- S) (that is, the set of the semantic functions applied if Imparfait(- S) is used) contains as its elements the functions past in S, non-past in S, imperfective in S, and unreal, among others. These functions are applied under certain conditions - for instance, non-past in S is applied only in hypothetical ./-clauses, whereas unreal can be applied in hypothetical ./-clauses, in sentence constructions which lead to the counterfactual use, and in children's role-playing games. Semantically relevant syntactic categories, semantic functions, and application conditions are connected to each other for all idiolect systems in the so-called "Syntactic Category Interpretation" (SCI) of the idiolect system, a three-place relation assumed in the semantic part of the system. The SCI of S contains, for instance, the triples "," "r "," and "," where Cl(S), c2(S), c3(S), and c4(S) are the conditions in which past in S, imperfective in S, non-past in S, and unreal occur if the imparfait is used.^ Note that Cl (S) and c2(S) are not identical since the imparfait is not always simultaneously understood as past and imperfective (see for example the absolute, narrative, and counterfactual meanings noted earlier). Similarly. o3(S) and c4(S) are different, since in the counterfactual use, the imparfait has an unreal but past meaning. The main question to be answered is the formulation of the application condition of a semantic function which is the interpretation of a syntactic category. Intuitively, given a verb form occurrence in a sentence, and given the condition. it must be possible to determine whether the function in question is applied or not. In other words, the condition must be the set of "cases," more specifically the set of tuples of a verb form occurrence and a sentence in which the function occurs. This is precisely the formulation I propose for this condition. The condition is a set of occurrences of word forms coupled to entities of the type of sentences. But what is a sentence? According to Integrational Linguistics, a sentence of an idiolect system is a triple of 1. a sequence of phonological words of the idiolect system, 2. a syntactic structure of the sequence in the idiolect system, and 3. a lexical interpretation of the sequence in the idiolect system given the syntactic structure." Again, 15. More precisely, the SCI of 5 is divided into two types. The second type contains the tuples where the semantic function modifies the concept obtained from the original verb meaning: for example, the function that interprets the passive, changing for instance the concept of "killing" into the concept of "being killed" (see Lieb 1993: 462, 2002). After many reflections which do not have their place here, I concluded in Viguier (2007: 332) that the imperfective functions in French actually belong to this last type of function. For the present demonstration, the distinction between type 1 and type 2 syntactic interpretations can be disregarded. 16. Strictly speaking, Lieb (1983: 272-273) foresees two other components: a morpholexical interpretation (which assigns a meaning to each of the morphological constituents of the phonological words of the sequence) and a morphological structure of the syntactic units of the sequence, especially required in order to take certain accentuation effects into account. The two components are also relevant with respect to syntactic semantic composition. For instance, if a syntactic unit like die deutsche sprachwissenschaft occurs in a sequence with a meaning of 'the science of the German language' (and not 'German language-science', i.e. 'German linguistics'), these two components are indispensable: the morpholexical interpretation assigns -language- to sprach3 and -science- to wissenschaft4, and on the basis of the morphological structure, it is then possible to combine -language- with -german- and •german language- with -science-. On a bare syntactic level, -german- could only be

200 Marie-HeleneViguier the syntactic structure has three components: the constituent structure, winch organrzes the constituents of the sequence into units and groups, and assigns a specific type to each, the marking structure, which establishes which syntactic categories occur in the sentence, and the intonation structure, which contains the intonational features of the sentence. Consequently, an application condition for a semantic function in any idiolect system S> is a set of quadruples of the type < / , / s, e>, where fhf

a sequence of phonological words of S' or a positional variant of such a sequence (for ex. in S the sequence of a and couru, formally: {, }, noted % couru?; the unit sequence of traversait, formally: {}, noted "traversait?; the sequence anmex traversal la, rueA, or the positional variants of sequences a2 couru, and

s:

a syntactic structure o f / i n S' (triple ) k: a syntactic constituent structure of f in S' (characterizes the constituents o f / a s particle, noun or verb forms or groups of ST. Formally, k is a set of pairs of 1. a set of natural numbers - the positive integers that correspond to the position of a constituent in the sentence sequence - called the "domain" of a constituent, and 2. a constituent category of S') m: a syntactic marking structure o f / i n S' relative to k (characterizes the primitive constituents of /syntactically as present, active, imparfait, plural, etc., and lexically as full verb, intransitive verb, etc. Formally, m is a set of triples of 1. a set of natural numbers - the domain of a constituent, 2. a set of marking categories of S>, and 3. a set of lexical categories of ST) I: an intonation structure of/in S' (a function that assigns a sequence of sets of auditory values to the domain of any primitive constituent of/) a lexical interpretation o f / i n S> relative to s (a function that assigns a concept to the domain of any primitive constituent of/ If a constituent has no lexical meaning - as is the case for auxiliaries - this concept is "the empty concept," noted "b 0 ")

traversal)

e:

combined with the concept -language-science- (this example is taken from Lieb 1983: 262-263). Since I will not be further concerned with such so-called mixed sentence meanings in this article, I will not mention morpholexical interpretations and morphological structures in the context of sentences and, more generally, syntactic meanings.

The polysemy of syntactic categories and semantic coherence

201

For example, consider the sentence, noted orthographically, "Annie traversal la rue," in S (S is a Modern Standard French idiolect system). This sentence is seen as the triple , where / = annie, traversal^ la, rue, s = , where k = {, , , }, where "Nf(-, 5 ) " = the set of forms of nouns of & "Vf(- ST = the set of forms of verbs of S,' "VGr(-,S)" = the set ofverb groups of S On the basis ofk, annie, and la, rue, are associated with the noun forms of S, traversal is associated with the verb forms of S, and/is associated with the verb groups of S. m = {, e>, S'> is t h e / ' that is completed by the constituent w h o s e / ' is a most internal uuc^-core. To conclude,/ must have an occurrence of si? as its most external introductory element. (32)

For all syntactic quadruples . most-external-introductory-element relative t o / ' , ,', e', and S' [notated: "me-inrroduct(/', s', e', Sy\ is the function a such that: i. the arguments of a are the constituents of/' relative to s', e', and S': n. for all arguments {,' of a, the value a ( / ' ) is t h e / ' such that a. or b , a. there is a n / " such that: for all/ 2 ' w h o s e / ' is a most internal nuc*core relative t o / ' , s',e', and S',f2' is a component of a complements-tuple f o r / " relative t o / ' , ,', e', and ST, a n d / ' is t h i s / " , b. there is n o / " as in a., a n d / ' is the empty set.

Now t h a t / ' is an occurrence of a certain word in a quadruple means that l.f' is a positional variant of one of the possible forms of that word and 2. the concept that is assigned to the domain of/' by e' is the lexical meaning of that word. In Integrational Linguistics, a word is seen as a pair of a "word paradigm" F and a "word meaning" V. F is a set of pairs of a word form and its syntactic categorization (i.e. a set J' of syntactic categories that contain the word form). T h a t / has an occurrence of si? as its most external introductory element therefore means: 1. that the most external introductory element h o f / is a positional variant of a word f o r m / such t h a t / is first component of a pair < / J> in the word paradigm of si?, and 2. that the concept assigned to the domain of/ 2 by e is the word meaning of si?. If we call "Pos" the function which assigns the set of its positional variants to any sequence, we obtain the condition upon / relative to / s, e, and S: "(V/2) (/ 2 = me-introduct(/ s, e, S ) ( / ) - ( 3 / ) (3P) (EL/) (/ 2 e P o s ( / ) A < / ./> E 19. "Be 5 an idiolect system and < / ,, e> a syntactic tnple of S. f" is a m o r f /»ter»«/ » MC i- CO r e of/' in/given ,, e, and 5 iff: a. / ' is a constituent of/given s, b. there is a biggest n > 0 and a sequence of constituents ^ of /gi such that for all," e {1, . . . , „ - 1 } : (!) snvia\f,S,e,S), (ii)

/"=/i,

(m)/'=/„." (Budde 2000: 355, translation mme)

208

Marie-HeleneViguier

P A = sitf)," to be read, "For all most external introductory element^ o f / relative t o / s, e, and S, there is a / , a /» and a J, such that: - / 2 i s a positional variant of/. - the pan o f / and J is an element of /», and - the pair of P and the lexical interpretation of the domain of/2 is identical with the word ,/ 1 w." 4. Lastly, the question of whether the main clause of a hypothetical ./-clause must be in the conditional arises. However, if we ignore indirect and free indirect discourse, the verb constituent can only be in the conditional, according to my analysis (see (33)). Moreover, a ./-clause can constitute its own sentence for instance optative or interrogative sentences like (34) and (35) respectively Finally, the presence of the conditional in the main clause would not be sufficient for the activation of the non-past meaning, as (36) shows. For these reasons, no condition should be made on the tense and mood in the main clause. (33)

Si Yannig venait, on mangeait des crepes. if Yannig come-iND.iMPF we eat-iND.iMPF crepes 'If Yannig came, we ate crepes.'

(34) Ah,sifetais riche! ah ifIbe-iND.iMPFnch 'Ah, if I were rich.'

(Roussane and Caudal 2002)

(35)

Sronprenait le tram? ifwetake-iND.iMPFtherrain 'What about taking the train?'

(36)

*Tu prenais le train, ca te reviendrait mains cher. you take-iND.iMPF the train it you be-coND.PRES cheaper - Y o u took the train, it would be cheaper.'

To summarize, non-past in S occurs when using Imparfait(- S) in the quadruples of c 3 (S), where C3(S) = the set of quadruples < / , / s, e> such that a. and b , a. the concept assigned t o / by e is not the empty concept b. i. / i s marked as verb form in the indicative imparfait relative t o /

s,e,mdS,

The polysemy of syntactic categories and semantic coherence 209 ii. the most external introductory element of/i relative to/, s, e, and Sis an occurrence of the hypothetical .,-subjunction. Thxs corresponds to the following formal expression (given the conventions set above):

c3(S) = { | e(Domain(f1)) 4- bo A (( e mark(/ s, e, S) A (V/2) (f2 = me-introducta s, e, S)(/i) - (3/ 3 , 3P, 3-/) (f2 e Pos(/ 3 ) A SPA = «!*))))}. The conditions for the remaining interpretation triples of the imparfait in S ( Cl (S), c2(S), c4(S) etc.) are formalized analogically. In my analysis, the wellknown use "past i m p e r f e c t a " corresponds to the cases where the semantic functions past in S and imperfecta in Sboth occur, that is in the quadruples of the intersection set C l ( S ) n c 2 ( S ) . Remark. It becomes clear that the theory of Integration^ Linguistics allows us to take care of the semantic variety of the imparfait, in that each meaning is associated with the co-text of the occurrence of the imparfait. Nevertheless, it is important to note that it does not mean that a new function must be assumed for each use of the imparfait. In Viguier (2007: 354-355, 383), I have attributed the politeness and hypoconstic uses to the meaning "past and imperfective" of the imparfait and I have analyzed them as conventionalized indirect speech acts. Furthermore, I have considered the futurate use as a case of indirect speech (Viguier 2007: 357). While it is possible to take into account that the speaker uses a tense metaphorically, this option is chosen only in justified cases (for instance, if the original meaning plays a part in the use) and not for a theoretical necessity as the case would be within a monosemic approach. In addition, this seems to me the only way to take diachronic changes of the semantics of a tense into account, namely if the use of a tense is first observed metaphorically in indirect speech acts and eventually as leading to a new meaning.

4. Summary and perspectives Faced with the fact that syntactic categories can have several uses, this essay argues that not all uses should be considered as pragmatically different uses of the same meaning. Rather, a syntactic category of an idiolect system can have a whole palette of meanings, formalized as semantic functions. In order to

210

Marie-Helene Viguier

avoid giving the procedure an ad-hoc character, these functions are associated with application conditions in the semantic part of the idiolect system. By specifying the application condition of the non-past meaning of the imparfait in a Modern Standard French idiolect system, we have assumed syntactic and lexical criteria. These are not confined to the verb form in question but can concern constituents in the direct environment of the verb form, like the hypothetical subjunction which introduces the whole verb group. Intonational as well as morphological criteria can also be mentioned as conditions for using a semantic function. For example, if one considers the hypoconstic imparfait as a meaning of the imparfait (and not as one of its indirect uses), the typical intonation observed in this case (see Wilmet 1997: 395; Bres 2003b: 119) should be seen as a condition. To summarize, syntactic polysemy can only be taken into account provided that 1. the syntactic theory foresees multidimensionality (especially integrating syntactic, lexical, and intonational elements in the sentence conception) and 2. it is possible to access these elements in order to build the sentence meaning. After accepting that syntactic categories can have more than one meaning. the difficulty consists only in describing the meanings and their application conditions as accurately as possible. I would like to give an example and make my own suggestions. Consider the narrative imparfait. The aspectual and speaker-specific effect can be described in a relatively satisfying way (see Viguier 2007: 358-362), but its application conditions turn out to be hard to identify First, in most cases this meaning concerns telic events (achievements in 8 1 % of the cases, according to Bres 1999: 95), but not always (for an example, see (37)). (37)

Claude Frollo avart repns precrprtamment la cle et Claude Frollo take-back-iND.PLUPERF quickly the key and un instant apres il etait surle sommet de la tour. one moment later he be-iND.iMPF on the top of the tower 'Claude Frollo had quickly taken back the key, and one moment later he was at the top of the tower.' (V. Hugo, Notre Dame de Paris, cit. in Bres 1999: 95)

Given that telicity does not only depend on lexical words (see Comne 1976: 45), it may be necessary to identify the precise syntactic conditions of telicity before formulating the condition of the narrative meaning. In Viguier (2007: 359), I chose to explain the narrative imparfait's correlation with telic events not through an application condition but rather by its aspectual property which is the most visible for telic events (so that the presence of the narrative meaning

The polysemy of syntactic categories and semantic coherence

211

and, consequently, of the textual effect, is made more obvious). Second, for the narrative imparfait, a logical textual relation is necessary; as Molendijk (1990) shows, the lack of a logical textual relation explains why the imparfait in (39). contrary to (38), cannot be understood as narrative. Hence, note that as soon as "M. Brown" is the secretary of "M. Chisnutt," the sentence (39) becomes acceptable. (38)

Le commandant sejeta sur Vinterphone et hurla the commandant pounce-iND.sP on the interphone and screech-imsP qu'il avait a parley a M. Chisnutt. Trois minutes plus tard, that he have-iND.iMPF to speak to Mr. Chisnutt three minutes later M. Chisnutt se presentait chez le commandant. Mr. Chisnutt appear-iND.iMPF before the commandant 'The commandant pounced on the interphone and screeched that he had to speak to Mr. Chisnutt. Three minutes later, Mr. Chisnutt appeared before the commandant.' (after a sentence from St. A. Steeman cit. in Tasmowski-De Ryck 1985: 66)

(39)

ILe commandant sejeta sur I' interphone et hurla the commandant pounce-iND.sP on the interphone and screech-iND.sP qu' il avait a parler a M. Chisnutt. Trots minutes plus that he have-iND.iMPF to speak to Mr. Chisnutt three minutes more tard, M. Brown se presentait chez le commandant. late Mr. Brown himself present-iND.iMPF before the commandant 'The commandant pounced on the interphone and screeched that he had to speak to Mr. Chisnutt. Three minutes later, Mr. Brown appeared before the commandant.'

It is again unclear how it can be expressed with an application condition. The oddness of (39) may be a consequence of the textual effect given the fact that in the narrative imparfait, the event is understood more or less as "surprisingly," "momentous" or even "tragic." In Viguier (2007: 361), I translated these ideas through a semantic function which makes the speaker signify that he believes that the existence of the event is "notable" (intuitively understood as "deserves the attention of the hearer"), in addition to his basic statement. Given a context, if the hearer does not understand in which regard it may be notable that the event takes place, he shall exclude the narrative meaning of imparfait and the sentence is not acceptable (see Viguier 2007: 362, rem. 3). Finally, the narrative imparfait occurs only in "closed" narrative segments (in (40) but not in (41), see Jayez 1998: 144). In (40) however, this observation cannot be

212

Marie-HeleneViguier

explained by the fact that the "notable" component of the event (that is: that the wife calls for aid) would not be understandable since the sentence remains odd even if an explanation is added (see (42)). (40)

Vers 20 heures, il eutun malaise. Sa femme llappelait a near 8 o'clock he black-out-iND.sP his wife call-imiMPF for Vaide. help 'Near 8 o'clock at night, he blacked out. His wife was calling for help.' (Jayez 1998: 144)

(41)

Vers 20 heures, il eutun malaise. Sa femme appelait a near 8 o'clock he black-out-imsP his wife call-iND.iMPF for Vaide et reussissait a faire venir un voisin. help and succeed-imiMPF to let come a neighbor 'Near 8 o'clock at night, he blacked out. His wife called for help and got a neighbor to come.' (Jayez 1998: 144)

(42)

Sa femme qui etait d'habitude toujours bien trapfiere pour his wife who be-iND.iMPF normally always all too proud to appeler de Vaide. Eh bien: Vers 20 heures, il eutun malaise. call for help well near 8 o'clock he black-out-iND.sP Elle ??appelait a Vaide. she call-iND.iMPF for help 'Normally his wife was always too proud to call for help. Near 8 o'clock at night, he blacked out. His wife was calling for help.'

Instead of symmetric schemata and unified overall meanings, I suggest concentrating on a precise description of 1. the different meanings of a category (and in this respect, we should proceed analytically instead of synthetically), and 2. the morphological, syntactic, lexical, and possibly intonational factors that. given a sentence, make one meaning possible as opposed to another.

References Berthonneau, Anne-Mane, and George Klerber 1993 Pour une nouvelle approche de l'rmparfart: l'imparfait, un temps anaphonque meronormque. Langages 112: 55-73. Blumenthal, Peter 1986 Vergangenheitstempora, Textstrukturierung und Zeitverstandnis in der franzosischenSprachgeschichte. Stuttgart: Sterner.

The polysemy of syntactic categories and semantic coherence

213

Bres, Jacques 1999 L'imparfait dit narratif tel qu'en lui-meme (le cotexte ne le change pas). Cahiersdepraxematique 32: 87-117. Bres, Jacques 2003a Presentation. Langue francaise 138: 3-7. Bres, Jacques 2003b Ma1S oui, il etait un joli temps du passe comme les autres, le petit rmparfzithypoconstique. Langue francaise 138: 111-125. Budde,Momka 2000 Wortarten: Definition und Identification. Ph.D. diss., Freie Umversitat Berlin. Bybee, Joan L., Revere D. Perkins, and William Paghuca 1994 The Evolution of Grammar Tense, Aspect, and Modality in the Languages of the World. Chicago: The University of Chicago Press. Caudal, Patrick, Carl Vetters, and Laurent Roussane 2003 L'imparfait, un temps inconsequent. Langue francaise 138: 61-74. Comne, Bernard 1976 Aspect. Cambridge: Cambridge University Press. Comne, Bernard 1985 Tense. Cambridge: Cambridge University Press. Confais,Paul 1995 Temps, mode, aspect: Les approches des morphemes verbaux et leurs problemes a I'exemple du francais et de I'allemand. Toulouse: Presses UmversitairesduMirail. Coseriu, Eugenic 1976 Das romanische Verbalsystem. Tubingen: Narx. Grevisse, Maurice 1986 Reprint. Le bon usage. 12th ed. Original edition 1936. Gembloux: Duculot. Guillaume,Gustave 1965 Temps et verbe: Theorie des aspects, des modes et des temps. Pans: Champion. Original edition, Pans: Champion, 1929. Haillet, Pierre P. 2003 Representations discursives, pomt(s) de vue et sigmfie unique du conditiormel. Langue francaise 138: 5-47. Jayez, Jacques 1998 DRT et imparfait: un exemple de traitement formel du temps. In Moeschleretal. (1998), 123-154. Lieb,Hans-Hemrich 1980 Zur semantischen Rechtfertigung syntaktischer Beschreibungen. In Empirische Rechtfertigung von Syntaxen: Beitrage zum Wuppertaler Kolloquium vom 25.-29. September 1978, 193-211. Bonn: Bouvier Verlag Herbert Grundmann. Lieb,Hans-Hemrich 1983 Integrational Linguistics. Vol. 1. Amsterdam: Benjamins.

214

Marie-HeleneViguier

Lieb,Hans-Hemnch 1993 Integrations Linguistics. In Syntax: Bin Internationales Handbuch Zeitgenossischer ForschunglAn International Handbook of Contemporary Research, Joachim Jacobs, Arnrm von Stechow, Wolfgang Sternefeld, and Theo Vennemann (eds.), vol. 1, 430-468. Berlin: de Gruyter. Lieb,Hans-Hemnch 2002 Lexical words and word paradigms. Paper presented at the conference "Operations on Argument Structure: Focus on Japanese and German" (Berlin, 7 * - 8 * March 2002). Moeschler, Jacques (ed.) 1993 Temps, reference et inference. Langages 112. Pans: Larousse. Moeschler, Jacques, Jacques Jayez, Momka Kozlowska, Jean-Marc Luscher, Louis de Saussure,andBertrandSthioul(eds.) 1998 Le temps des tenements: Pragmatique de la reference temporelle. Pans: Kime. Molendijk,Arie 1990 Le passe simple et I'imparfait: Une approche reichenbachienne. Amsterdam: Rodopi. Pollak, Wolfgang 1960 Studien zum Verbalaspekt'im Franzosischen. Sitzungsberichte der OsterreichischenAkademie der Wissenschaften, Philosophisch-historische Klasse 233,5. Wien: Rohrer. Roussane, Laurent, and Patrick Caudal 2002 Contributions semantique et discursive des constructions en si. Presented at the firth International Conference of Chronos (Gronmgen, 1